Full Eigenvectors

CHEMICAL PROCESS
DYNAMICS AND
CONTROLS
Peter Woolf et al.

University of Michigan
University of Michigan
Chemical Process Dynamics and Controls
Peter Woolf et al.

This text is disseminated via the Open Education Resource (OER) LibreTexts Project (https://LibreTexts.org) and like the hundreds of other
texts available within this powerful platform, it is freely available for reading, printing and "consuming." Most, but not all, pages in the
library have licenses that may allow individuals to make changes, save, and print this book. Carefully consult the applicable license(s)
before pursuing such effects.
Instructors can adopt existing LibreTexts texts or Remix them to quickly build course-specific resources to meet the needs of their students.
Unlike traditional textbooks, LibreTexts’ web based origins allow powerful integration of advanced features and new technologies to
support learning.
The LibreTexts mission is to unite students, faculty and scholars in a cooperative effort to develop an easy-to-use online platform for the
construction, customization, and dissemination of OER content to reduce the burdens of unreasonable textbook costs to our students and
society. The LibreTexts project is a multi-institutional collaborative venture to develop the next generation of open-access texts to improve
postsecondary education at all levels of higher learning by developing an Open Access Resource environment. The project currently
consists of 14 independently operating and interconnected libraries that are constantly being optimized by students, faculty, and outside
experts to supplant conventional paper-based books. These free textbook alternatives are organized within a central environment that is both
vertically (from advance to basic level) and horizontally (across different fields) integrated.
The LibreTexts libraries are Powered by MindTouch® and are supported by the Department of Education Open Textbook Pilot Project, the
UC Davis Office of the Provost, the UC Davis Library, the California State University Affordable Learning Solutions Program, and Merlot.
This material is based upon work supported by the National Science Foundation under Grant No. 1246120, 1525057, and 1413739. Unless
otherwise noted, LibreTexts content is licensed by CC BY-NC-SA 3.0.
Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily
reflect the views of the National Science Foundation nor the US Department of Education.
Have questions or comments? For information about adoptions or adaptions contact [email protected]. More information on our
activities can be found via Facebook (https://facebook.com/Libretexts), Twitter (https://twitter.com/libretexts), or our blog
(http://Blog.Libretexts.org).
This text was compiled on 07/04/2022

TABLE OF CONTENTS
This textbook is a student-contributed open-source text covering the materials used in a senior level controls course. The authors cover a
variety of approaches to reduce or manage this uncertainty through the use of robust designs, dynamic systems theory, nonlinear dynamics,
control theory, and statistics.
1: OVERVIEW
1.1: Introduction to Controls- Background and design methodology
1.2: Introduction to DCS- Control system hardware
1.3: Current Significance- Process controls and you
1.4: Failures in Process Control- Bhopal, Three Mile Island
1.5: Process Controls in Everyday Life- Applying process control thinking to everyday situations
2: MODELING BASICS
2.1: Verbal Modeling- process description, control specifications, and connections
2.2: Degrees of Freedom- importance, calculation procedure, and examples
2.3: Incidence Graphs -Interpretations, Consistency, and Inconsistency
2.4: Excel Modeling - logical models, optimization with solver for nonlinear regression, sampling random numbers
2.5: Noise Modeling - White, Pink, and Brown Noise, Pops and Crackles
2.6: Numerical ODE solving in Excel- Euler’s method, Runge Kutta, Dead time in ODE solving
2.7: Solving ODEs with Mathematica- How to find numerical and analytical solutions to ODEs with Mathematica
2.8: Fitting ODE parameters to data using Excel- Using regression to fit complex models in Excel
2.9: Helpful Mathematica Syntax- Hints on how to use Mathematica to model chemical processes
3: SENSORS AND ACTUATORS

3.1: Control Systems - Measurement Devices
3.2: Temperature Sensors
3.3: Pressure Sensors
3.4: Level Sensors
3.5: Flow Sensors
3.6: Composition Sensors
3.7: pH and Viscosity Sensors
3.8: Miscellaneous Sensors
3.9: Valves - Types, Kinds, and Selection
3.10: Valves - Modeling Dynamics
4: PIPING AND INSTRUMENTATION DIAGRAMS

4.1: PandID General Information
4.2: Piping and Instrumentation Diagram Standard Notation
4.3: Piping and Instrumentation Diagrams - Location of Controls and Standard Control Structures
4.4: Piping and Instrumentation Diagrams - Standard Pitfalls
4.5: Safety Features in Piping and Instrumentation Diagrams
4.6: Regulatory Agencies and Compliance
5: LOGICAL MODELING
5.1: Boolean Models - Truth Tables and State Transition Diagrams
5.2: Logical Control Programs - IF... THEN… WHILE…
1
6: MODELING CASE STUDIES
6.1: Surge Tank Model
6.2: Heated Surge Tank
6.3: Bacterial Chemostat
6.4: ODE and Excel CSTR model with heat exchange
6.5: ODE and Excel model of a Simple Distillation Column
6.6: ODE and Excel model of a Heat Exchanger
6.7: ODE and Excel Model of an Adiabatic PFR
6.8: Modeling and PID Controller Example - Cruise Control for an Electric Vehicle
6.9: Blood Glucose Control in Diabetic Patients
7: MATHEMATICS FOR CONTROL SYSTEMS

7.1: Dirac delta (impulse) function
7.2: First-order Differential Equations
7.3: Second-order Differential Equations
7.4: Taylor Series
7.5: Laplace Transforms
8: OPTIMIZATION
8.1: Introduction to Optimization
8.2: Linear Optimization
8.3: Non-linear Optimization
9: PROPORTIONAL-INTEGRAL-DERIVATIVE (PID) CONTROL

9.1: Constructing Block Diagrams- Visualizing control measurements
9.2: P, I, D, PI, PD, and PID control
9.3: PID Tuning via Classical Methods
9.4: PID tuning via Frequency Responses with Bode Plots
9.5: PID tuning via Optimization
9.6: PID Downsides and Solutions
10: DYNAMICAL SYSTEMS ANALYSIS

10.1: Finding fixed points in ODEs and Boolean models
10.2: Linearizing ODEs
10.3: Eigenvalues and Eigenvectors
10.4: Using Eigenvalues and Eigenvectors to Find Stability and Solve ODEs
10.5: Phase Plane Analysis - Attractors, Spirals, and Limit cycles
10.6: Root Locus Plots - Effect of Tuning
10.7: Routh Stability - Ranges of Parameter Values that are Stable
11: CONTROL ARCHITECTURES

11.1: Feedback Control
11.2: Feed Forward Control
11.3: Cascade Control
11.4: Ratio Control
11.5: Common control loops and model for liquid pressure and liquid level
11.6: Common Control Loops and Model for Temperature Control
11.7: Common control architectures and model for reactors
2
12: MULTIPLE INPUT, MULTIPLE OUTPUT (MIMO) CONTROL
12.1: Determining if a system can be decoupled
12.2: MIMO control using RGA
12.3: MIMO using Model Predictive Control
12.4: Neural Networks for automatic model construction
12.5: Understanding MIMO Control Through Two Tanks Interaction
13: STATISTICS AND PROBABILITY BACKGROUND

13.1: Basic statistics- mean, median, average, standard deviation, z-scores, and p-value
13.2: SPC- Basic Control Charts- Theory and Construction, Sample Size, X-Bar, R charts, S charts
13.3: Six Sigma- What is it and what does it mean?
13.4: Bayes Rule, Conditional Probability and Independence
13.5: Bayesian Network Theory
13.6: Learning and analyzing Bayesian networks with Genie
13.7: Occasionally Dishonest Casino? Markov Chains and Hidden Markov Models
13.8: Continuous Distributions- normal and exponential
13.9: Discrete Distributions - Hypergeometric, Binomial, and Poisson
13.10: Multinomial Distributions
13.11: Comparisons of two means
13.12: Factor analysis and ANOVA
13.13: Correlation and Mutual Information
13.14: Random sampling from a stationary Gaussian process
14: DESIGN OF EXPERIMENTS

14.1: Design of Experiments via Taguchi Methods - Orthogonal Arrays
14.2: Design of experiments via factorial designs
14.3: Design of Experiments via Random Design
14.4: Summary- Summary on Control Architectures’ philosophies, advantages, and disadvantages.
INDEX
Book: Chemical Process Dynamics and Controls (Woolf) is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et
al. via source content that was edited to conform to the style and standards of the LibreTexts platform; a detailed edit history is available upon request.
3
CHAPTER OVERVIEW
1: OVERVIEW
1.1: Introduction to Controls- Background and design methodology
1.2: Introduction to DCS- Control system hardware
1.3: Current Significance- Process controls and you
1.4: Failures in Process Control- Bhopal, Three Mile Island
1.5: Process Controls in Everyday Life- Applying process control thinking to everyday situations
1: Overview is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source content that was edited to
conform to the style and standards of the LibreTexts platform; a detailed edit history is available upon request.
1
1.1: INTRODUCTION TO CONTROLS- BACKGROUND AND DESIGN
METHODOLOGY
1.1.1: INTRODUCTION
Process controls is a mixture between the statistics and engineering discipline that deals with the mechanism, architectures, and algorithms
for controlling a process. Some examples of controlled processes are:
Controlling the temperature of a water stream by controlling the amount of steam added to the shell of a heat exchanger.
Operating a jacketed reactor isothermally by controlling the mixture of cold water and steam that flows through the jacket of a jacketed
reactor.
Maintaining a set ratio of reactants to be added to a reactor by controlling their flow rates.
Controlling the height of fluid in a tank to ensure that it does not overflow.
To truly understand or solve a design problem it is necessary to understand the key concepts and general terminology. The paragraphs below
provide a brief introduction to process controls as well as some terminology that will be useful in studying controls. As you begin to look at
specific examples contained here, as well as elsewhere on the wiki, you will begin to gain a better grasp on how controls operate and
function as well as their uses in industry.
1.1.2: PROCESS CONTROL BACKGROUND

The role of process control has changed throughout the years and is continuously shaped by technology. The traditional role of process
control in industrial operations was to contribute to safety, minimized environmental impact, and optimize processes by maintaining process
variable near the desired values (1). Generally, anything that requires continuous monitoring of an operation involve the role of a process
engineer. In years past the monitoring of these processes was done at the unit and were maintained locally by operator and engineers. Today
many chemical plants have gone to full automation, which means that engineers and operators are helped by DCS that communicates with
the instruments in the field.
What are the benefits of Process Control?
The benefits of controlling or automating process are in a number of distinct area in the operation of a unit or chemical plant. Safety of
workers and the community around a plant is probably concern number one or should be for most engineers as they begin to design their
processes. Chemical plants have a great potential to do severe damage if something goes wrong and it is inherent the setup of process
control to set boundaries on specific unit so that they don’t injure or kill workers or individuals in the community.
1.1.3: THE OBJECTIVES OF CONTROL

A control system is required to perform either one or both task:
1. Maintain the process at the operational conditions and set points
Many processes should work at steady state conditions or in a state in which it satisfies all the benefits for a company such as budget, yield,
safety, and other quality objectives. In many real-life situations, a process may not always remain static under these conditions and therefore
can cause substantial losses to the process. One of the ways a process can wander away from these conditions is by the system becoming
unstable, meaning process variables oscillate from its physical boundaries over a limited time span. An example of this would be a water
tank in a heating and cooling process without any drainage and is being constantly filled with water. The water level in the tank will
continue to rise and eventually overflow. This uncontrolled system can be controlled simply by adding control valves and level sensors in
the tank that can tell the engineer or technician the level of water in the tank. Another way a process can stray away from steady state
conditions can be due to various changes in the environmental conditions, such as composition of a feed, temperature conditions, or flow
rate.
2. Transition the process from one operational condition to another
In real-life situations, engineers may change the process operational conditions for a variety of different reasons, such as customer
specifications or environment specifications. Although, transitioning a process from one operational condition to another can be detrimental
to a process, it also can be beneficial depending on the company and consumer demands.
Examples of why a process may be moved from one operational set point to another:
1. Economics
2. Product specifications
3. Operational constraints
4. Environmental regulations
5. Consumer/Customer specifications
6. Environmental regulations
1.1.1 https://eng.libretexts.org/@go/page/22539
7. Safety precautions
1.1.4: DEFINITIONS AND TERMINOLOGY

In controlling a process there exist two types of classes of variables.
1. Input Variable – This variable shows the effect of the surroundings on the process. It normally refers to those factors that influence the
process. An example of this would be the flow rate of the steam through a heat exchanger that would change the amount of energy put into
the process. There are effects of the surrounding that are controllable and some that are not. These are broken down into two types of inputs.
a. Manipulated inputs: variable in the surroundings can be control by an operator or the control system in place.
b. Disturbances: inputs that can not be controlled by an operator or control system. There exist both measurable and immeasurable
disturbances.
2. Output variable- Also known as the control variable These are the variables that are process outputs that effect the surroundings. An
example of this would be the amount of CO2 gas that comes out of a combustion reaction. These variables may or may not be measured.
As we consider a controls problem. We are able to look at two major control structures.
1. Single input-Single Output (SISO)- for one control(output) variable there exist one manipulate (input) variable that is used to affect the
process
2. Multiple input-multiple output(MIMO)- There are several control (output) variable that are affected by several manipulated (input)
variables used in a given process.
Cascade: A control system with 2 or more controllers, a "Master" and "Slave" loop. The output of the "Master" controller is the setpoint
for the "Slave" controller.
Dead Time: The amount of time it takes for a process to start changing after a disturbance in the system.
Derivative Control: The "D" part of a PID controller. With derivative action the controller output is proportional to the rate of change
of the process variable or error.*
Error: In process controls, error is defined as: Error = setpoint - process variable.
Integral Control: The "I" part of a PID controller. With integral action the controller output is proportional to the amount and duration
of the error signal.
PID Controller: PID controllers are designed to eliminate the need for continuous operator attention. They are used to automatically
adjust system variables to hold a process variable at a setpoint. Error is defined above as the difference between setpoint and process
variable.
Proportional Control: The "P" part of a PID controller. With proportional action the controller output is proportional to the amount of
the error signal.
Setpoint: The setpoint is where you would like a controlled process variable to be.
1.1.5: DESIGN METHODOLOGY FOR PROCESS CONTROL

1. Understand the process: Before attempting to control a process it is necessary to understand how the process works and what it does.
2. Identify the operating parameters: Once the process is well understood, operating parameters such as temperatures, pressures, flow
rates, and other variables specific to the process must be identified for its control.
3. Identify the hazardous conditions: In order to maintain a safe and hazard-free facility, variables that may cause safety concerns must
be identified and may require additional control.
4. Identify the measurables: It is important to identify the measurables that correspond with the operating parameters in order to control
the process.
Measurables for process systems include:
Temperature
Pressure
Flow rate
pH
Humidity
Level
Concentration
Viscosity
Conductivity
Turbidity
Redox/potential
Electrical behavior
Flammability
5. Identify the points of measurement: Once the measurables are identified, it is important locate where they will be measured so that the
system can be accurately controlled.
6. Select measurement methods: Selecting the proper type of measurement device specific to the process will ensure that the most
accurate, stable, and cost-effective method is chosen. There are several different signal types that can detect different things.
These signal types include:
Electric
Pneumatic
Light
Radiowaves
Infrared (IR)
Nuclear
7. Select control method: In order to control the operating parameters, the proper control method is vital to control the process effectively.
On/off is one control method and the other is continuous control. Continuous control involves Proportional (P), Integral (I), and Derivative
(D) methods or some combination of those three.
8. Select control system: Choosing between a local or distributed control system that fits well with the process effects both the cost and
efficacy of the overall control.
9. Set control limits: Understanding the operating parameters allows the ability to define the limits of the measurable parameters in the
control system.
10. Define control logic: Choosing between feed-forward, feed-backward, cascade, ratio, or other control logic is a necessary decision
based on the specific design and safety parameters of the system.
11. Create a redundancy system: Even the best control system will have failure points; therefore it is important to design a redundancy
system to avoid catastrophic failures by having back-up controls in place.
12. Define a fail-safe: Fail-safes allow a system to return to a safe state after a breakdown of the control. This fail-safe allows the process to
avoid hazardous conditions that may otherwise occur.
13. Set lead/lag criteria: Depending on the control logic used in the process, there may be lag times associated with the measurement of the
operating parameters. Setting lead/lag times compensates for this effect and allow for accurate control.
14. Investigate effects of changes before/after: By investigating changes made by implementing the control system, unforeseen problems
can be identified and corrected before they create hazardous conditions in the facility.
15. Integrate and test with other systems: The proper integration of a new control system with existing process systems avoids conflicts
between multiple systems.
1.1.6: REFERENCES
1. Romagnoli, Jose A. Introduction to Process Control. s.l. : CRC press, 2006.
1.1: Introduction to Controls- Background and design methodology is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by
Peter Woolf et al. via source content that was edited to conform to the style and standards of the LibreTexts platform; a detailed edit history is available
upon request.
1.2: INTRODUCTION TO DCS- CONTROL SYSTEM HARDWARE
1.2.1: INTRODUCTION
Digital Control Systems(DCS) also known as Distributed Control System is the brain of the control system. It is used mainly for the
automation of a manufacturing process and manages the logic that exist for major unit operations. A DCS in the past was tailor made for the
process, plant or company that intended to use the structure to control and model it’s process. Before the beginning of the DCS era there
were pneumatic devices that controlled process and engineers manually turned valves on the site. Modeling of the systems was made
possible by DCS as it allowed the ability to record and manage process from comfort of a computer screen. Because of DCS we are able to
control processes remotely and gain a better understanding of how the process operate and how they can be improved to both increase safety
and increase profit possibilities.
Control Systems are collectively named as "ICSS" Integrated Control and Safety System. Distinctly identified as "BPCS" Basic Process
Control System. "SIS" Safety Instrumentation System. "F&G" Fire and Gas System.
DCS is employed in BPCS as well as used and prevalent control system. How does a DCS work?
In the field you have sensors and gauges that give and recieve information. They convert this information into a electric signal that is sent to
a control room somewhere in the field. This control room has programmed logic that is able to converts the signal into a pressure, flow rate,
concentration, temperature, or level. This logic also contains the information that controls the process and takes the signal compares it with
the set point sent from the operator may or may not be in the field and sends a signal to the manipulated variables in the field. The DCS
covers all of the computer logic from the operator screen to the field box that contain the logic.
Shutdown systems
Shutdown systems are the emergency setting of the logic to make sure the process can be contained and is environmentally safe. These
setting are important for emergency response of the system. It is the job of the DCS to contain the logic for the shutdown system and be able
to operate when a process exceed a certain limit.
1.2: Introduction to DCS- Control system hardware is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al.
via source content that was edited to conform to the style and standards of the LibreTexts platform; a detailed edit history is available upon request.
1.3: CURRENT SIGNIFICANCE- PROCESS CONTROLS AND YOU
Authors: (December 12, 2009) Steve Dzul, Steve Pankratz, Derrick Boroski
1.3.1: INTRODUCTION
Industrial processes are central to the chemical engineering discipline. Generally, processes are controlled in order to do things such as
maximize safety, minimize cost, or limit effects on the environment. This course aims to help undergraduate engineering students
understand the mechanisms used to moderate these processes, such as to control their output.
1.3.2: AUTOMATION
Generally, process controls are designed to be automated. This means that given a change in system response, the control system can act on
its own to account for it. In order to minimize cost, automated systems have become widespread throughout industry. Before automation, a
huge amount of labor would be required to run even the simplest processes. For example, a technician might be hired to monitor the
temperature in a reaction vessel, and operate a valve to manipulate the cooling water flow rate in the jacket. In a sense, this technician
operated as a control system. If the temperature reading is too high, the technician will manipulate the system in order to bring the
temperature down. Via automation, this simple, arduous labor can be done by an algorithm.
By designing an effective control system, even the most complicated of processes can be run with minimal worker supervision. Telephone
operators, for example, have largely been replaced by automated telephone switch boards. Removing the need for telephone operators
decreases operating cost for phone companies, thereby allowing the general consumer to pay less for phone service. Automated process
controls, therefore, are enormously important in the modern world.
1.3: Current Significance- Process controls and you is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al.
1.4: FAILURES IN PROCESS CONTROL- BHOPAL, THREE MILE ISLAND
Authors: (December 12, 2009) Steve Dzul, Steve Pankratz, Derrick Boroski
Process controls can have a huge impact on surrounding communities, as well as the environment. An engineer of a large-scale process,
therefore, has an important ethical responsibility to operate a process safely and properly. These responsibilities extend well beyond the
scope of merely the company for which they work. Catastrophic failures in process control remind us of the importance of control systems
and engineering in today's world.
1.4.1: BHOPAL, INDIA DISASTER

1.4.1.1: OVERVIEW
The Bhopal Gas Tragedy in Bhopal, India on December 3, 1984 was a large toxic gas leak that killed thousands of people in the surrounding
area. A tank with 42 tons of methyl isocyanate(MIC) was contaminated with water. This in turn caused a run away reaction that greatly
increased the pressure and temperatures in the tank, which forced the emergency venting of the toxic gases to the atmosphere.
This tragedy was largely due to the failure or lack of safety controls:
1. Runaway reaction as temperature and pressure increased without regulation
2. MIC was supposed to be cooled, however in the Bhopal plant the refrigeration system was not turned on. Temperature control on the
tank could have greatly hindered the runaway reaction that ensued with the addition of water.
3. Flare tower to handle the leakage of toxic gases was not functional
4. The plant also had vent scrubbers, which were also not functional
5. Water curtain, which would neutralize some escaping gas, not designed properly. It was not tall enough to reach the top of the flare
tower, making it essentially worthless.
6. Alarms that would have alerted to a malfunction in the tank had not been operational for 4 years
The figure below illustrates some of these failures:
Had at least some of these been functioning the amount of toxic gas released would have been substantially reduced.
1.4.1.2: RESULTS
From this tragedy we can see that if the plant had proper safety controls the effects of the disaster would have been greatly reduced.
Therefore as a chemical engineer it is our responsibility to society to provide sufficient safety controls to chemical processes in order to
prevent disasters such as the Bhopal Gas Tragedy from happening. Unfortunately, industrial negligence is still a problem in many third-
world countries.
1.4.1.3: REFERENCE
http://www.corrosion-doctors.org/Pollution/bhopal.htm
1.4.2: THREE MILE ISLAND DISASTER

1.4.2.1: OVERVIEW
One of the largest and most far reaching plant failures in United States history took place at a nuclear power plant on Three Mile Island in
March 1979. The event was caused by either a mechanical or electrical failure of the main feed water pumps causing the power plant to
begin to overheat. As the heat increased, the control scheme caused the turbine and reactor to shut down. This caused a pressure increase in
the primary system (nuclear portion of the plant) and a relief valve automatically opened to release some of the pressure to prevent the
reactor from blowing. All of these actions were well designed to prevent a significant event from happening. The problem was that the
release valve did not close properly when the pressure in the reactor was relieved. As a result, when the reactor started back up, coolant in
the core of the reactor was lost through the pressure relief valve. Because there was no control mechanism that measured the level of the
coolant in the reactor, the operators, who only judged the water level by the pressure in the reactor, actually decreased coolant flow to the
reactor.
The figure below is a simplified diagram of the TMI-2-plant:
The result of the control design failure that prevented the operators from cooling the reactor was that the rods that held the nuclear fuel
melted causing the fuel to also melt. This is the worst thing to have happen in a nuclear power plant and is what happened to cause the
disaster at Chernobyl. Thankfully, the accident was largely contained and although the entire nation watched for 3 days as the threat of an
explosion or breach of containment loomed, 0 deaths or injuries resulted. In fact, corrective steps were so successful that the average
increase in radiation to the surrounding population was around 1% and the maximum increase at the boundary to the site is estimated to be
less than 100% of the natural background radiation present in the region.
1.4.2.2: RESULTS
The accident at Three Mile Island showed the importance of proper design of control systems. As a result the US Nuclear Regulatory
Commission took steps to tighten their regulation and increase the safety requirements on Nuclear Power Plants. These included revamping
operator training as well as increasing the design and equipment requirements. This also brought the dangers of all industrial processes to
the forefront and reminded people of the importance of the safety of the communities surrounding chemical and power plants.
Unfortunately, the incident also inspired intense fear of nuclear power in the general population and is partially responsible for the reduced
build rate for new nuclear power plants since that time. Although control failures can be corrected fairly quickly, after one safety issue it is
difficult to convince the general public that engineers have fixed the problem and that it will not happen again.
1.4.2.3: REFERENCE
References: www.nrc.gov/reading-rm/doc-collections/fact-sheets/3mile-isle.html: The United States Nuclear Regulatory Commission
1.4.3: TEXAS CITY DISASTER

Authors: (December 14, 2009) Virgil Humes, Fred Garner
1.4.3.1: OVERVIEW
In March of 2005, 15 people were killed at the of a refinery explosion in Texas City, TX. The refinery is the third largest in the country with
the capability to process over 400,000 barrels of crude oil daily. On this particular day, the isomerization unit (whose purpose is to boost the
octane rating of fuels) was being started up. As part of the normal start-up sequence, operators began feeding hydrocarbon liquid into the
tower. However, a discharge valve that should have been opened was overlooked. A high level alarm was ignored and a second high-level
alarm had not yet been re-enabled from the various maintenance activities that had been going on while the unit was down. Upon realizing
that the tower was filling, a discharge valve was opened to release the hot fluid from the bottom of the vessel. This fluid then passed through
a heat exchanger, warming the fluid that was still being pumped into the tower. The boiling that resulted in the bottom of the tower caused
liquids to spill over the top of the stack and into a blow down drum with an atmospheric vent and no flare. The blow down drum could not
contain the volume and began ejecting liquid and vapor hydrocarbon from the atmospheric vent. The vapors were soon ignited by a diesel
truck with its ignition on. Fifteen people in a nearby trailer were killed in the resulting explosion. Inadequacies in the written start-up
procedures, operator training, and the design of the safety relief system led to tragic and unnecessary loss of life. But examples such as these
serve to make process environments even safer places to work in and around.
An animation detailing the circumstances surrounding the explosion was released by the U.S. Chemical Safety Board during the
investigation. Some accounts of the incident vary in a few details, but all are in agreement that the written procedures, operator training, and
some aspects of the safety system design were to blame.
1.4.3.2: RESULTS
After the incident at the refinery, representatives from BP said the company would eliminate all blow-down drums/vent stack systems in
flammable service. Also, the Chemical Safety Board (CSB) recommended that BP commission an independent panel to investigate the
safety culture and management systems of BP. The findings of the panel showed that BP management had not distinguished between
occupational safety and process safety. This led to new implementations of process safety throughout BP North America.
1.4.3.3: REFERENCE
CSB Final Report into March 2005 Texas City Disaster
Wikipedia - Texas City Refinery (BP)
1.4: Failures in Process Control- Bhopal, Three Mile Island is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf
et al. via source content that was edited to conform to the style and standards of the LibreTexts platform; a detailed edit history is available upon request.
1.5: PROCESS CONTROLS IN EVERYDAY LIFE- APPLYING PROCESS
CONTROL THINKING TO EVERYDAY SITUATIONS
Written By: Andrew Sereno, Randy Tin (FA09)
1.5.1: INTRODUCTION
Have you ever wondered if the abstract and arbitrary subject of chemical engineering process controls has any relevance outside of Dow
1013? Below are a few examples of how the concepts learned in process controls can be applied to situations in your everyday life.
1.5.2: EXAMPLES OF PROCESS CONTROL FOR THE COMMON MAN

1.5.2.1: CONSUMPTION
Spicing Up Soup
Pavlo LaBalle just finished scaling up a mixing tank for use in the manufacture of horse glue. Scaling up mixing tanks for horse glue is hard
work - enough to make any man crave a hearty bowl of spicy liver soup. Fortunately, Mrs. LaBalle has a batch of liver soup ready for
Pavlo's return home. Using his built-in composition sensor, the tongue, Pavlo realizes the liver soup spice concentration is below his
specified setpoint - the soup is not spicy enough. Having never spiced his own soup, but not wanting to offend Mrs. LaBalle, Pavlo attempts
to increase the spiciness of the soup using his favorite hot sauce, Sriracha. In order to prevent catastrophic overspicing, Pavlo adds a drop of
Sriracha (a differential amount) and immediately begins sampling the soup after addition. In this way, Pavlo is able to determine the effect
of a small disturbance (the introduction of Sriracha) on the process output (the soup's spiciness). Using his new knowledge of Sriracha's
effect on the soup, Pavlo is able to optimize his addition of Sriracha so as to not catastrophically overspice the soup. Pavlo then adds a new
amount of Sriracha estimated to bring the soup closer to the spice setpoint and samples the soup shortly after. By tasting the soup after the
addition of spice, Pavlo is able to bring the soup closer to his desired spice setpoint through the use of feed backward spice control. By
applying his education in chemical engineering process controls, Pavlo has enabled himself to enjoy his spicy liver soup.
Imbibing Alcohol
Jimmy Johnson just finished a presentation and a final report for Chemical Engineering 460. Unfortunately, during the presentation,
Jimmy's personal record of 80 straight hours without sleep came to an abrupt end as he blacked out mid-line, and he was subsequently
criticized for insufficiently displaying the iron resolve befitting of a Michigan engineer. Jimmy decided to drown his sorrows at Good Time
Charley's through the careful ingestion of mood-improving fluids. Jimmy planned on drinking until he felt a moderate buzz, then leaving to
finish his graduate school application to pursue a masters degree in English.
Jimmy, inexperienced with the process of satiating his thirst with such beverages, planned to continuously drink until he felt adequately
affected. However, Jimmy failed to take into account the dead time inherent in his body's response to alcohol. Jimmy was unable to control
his body's response as he had planned, resulting in a vast overshoot into drunken stupor.
Jimmy woke up with no pants the next morning. Fortunately, by applying his education in chemical engineering, he earned the necessary
experience for tuning the amount he should drink in the future.
Purchasing Food at U-Go's
Rachel Malta was busy typing her technical report on the design space of a distillation column in a pilot horse glue plant. It was 11:50 p.m.,
and Rachel was making good progress until she felt the odd sensation of bubbling acids corroding through her stomach wall, at which point
she decided to visit U-Go's to find something to eat.
Confronted with the shop's wide array of processed foods and saturated fats, a difficult problem lay before Rachel. If she purchased too
much food, her hunger set point would be reached, but she would be left with bags of uneaten popcorn and sour patch kids. If she purchased
too little, she would remain hungry and would be unable to return to purchase more food after the store closed. Rachel needed to make a
prediction of how much food her digestion process would require to reach her desired fullness. Based on her previous experiences with U-
Go's "food," her predictive control (her memory) decided to purchase a bag of barbecue chips, one apple, and a bagel. By applying her
education in chemical engineering, Rachel was able to successfully predict how much food would satisfy her hunger without being forced to
find more food.
1.5.2.2: HYGIENE
Filling a Bathtub
Lan Ri has been in the Duderstadt Center for the past 3 nights working on his ChE 487 project. Having not bathed for 60 hours, Mr. Ri
decides it is time to wash himself for the sake of his fellow Duderstadt dwellers, however Mr. Ri has forgotten how to stand up (he's been
sitting at a computer for 60 hours). The ever creative Mr. Ri decides to take a bath instead of his usual standing shower. At the bath, Mr. Ri
finds himself confronted with two flow rate controllers: one "hot water feed" controller and one "cold water feed" controller. Turning the
controls, he realizes that they control the flow rate of the bathtub feed streams. Yet Mr. Ri has a problem - the hot water feed temperature, as
measured by his built-in temperature sensor (his skin), is far too high. The cold water feed temperature, also measured using his skin-type
temperature sensor, is too low for a comfortable bath. Accordingly, Mr. Ri performs a quick energy balance to determine expected ratio of
the feeds necessary to achieve a bath temperature close to what his desired setpoint. However, because Mr. Ri has poor control of the actual
flow rate of each stream (hot and cold), the actual resulting temperature is still below that which Mr. Ri desires. Mr. Ri then readjusts the
hot-to-cold feed ratio by increasing the hot water feed flow rate. As the bathtub fills up, Mr. Ri uses his built-in level sensor (his eyeballs) to
turn off the valves controlling the feed stream flow when the bathtub level is at his calculated level. Mr. Ri was able to calculate the level to
which the bathtub should be filled by using his knowledge of static fluid systems. By applying his education in chemical engineering
process controls, Mr. Ri has enabled himself to wash away the smell of the Duderstadt Center.
[Can use some diagrams, pictures, and/or equations to describe these problems - RZ].
1.5: Process Controls in Everyday Life- Applying process control thinking to everyday situations is shared under a CC BY 3.0 license and was authored,
remixed, and/or curated by Peter Woolf et al. via source content that was edited to conform to the style and standards of the LibreTexts platform; a detailed
edit history is available upon request.
CHAPTER OVERVIEW
2: MODELING BASICS
2.1: Verbal Modeling- process description, control specifications, and connections
2.2: Degrees of Freedom- importance, calculation procedure, and examples
2.3: Incidence Graphs -Interpretations, Consistency, and Inconsistency
2.4: Excel Modeling - logical models, optimization with solver for nonlinear regression, sampling random numbers
2.5: Noise Modeling - White, Pink, and Brown Noise, Pops and Crackles
2.6: Numerical ODE solving in Excel- Euler’s method, Runge Kutta, Dead time in ODE solving
2.7: Solving ODEs with Mathematica- How to find numerical and analytical solutions to ODEs with Mathematica
2.8: Fitting ODE parameters to data using Excel- Using regression to fit complex models in Excel
2.9: Helpful Mathematica Syntax- Hints on how to use Mathematica to model chemical processes
2: Modeling Basics is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source content that was edited
to conform to the style and standards of the LibreTexts platform; a detailed edit history is available upon request.
1
2.1: VERBAL MODELING- PROCESS DESCRIPTION, CONTROL
SPECIFICATIONS, AND CONNECTIONS
Authors: (September 8, 2006) Brian McQuillan, Crystal Miranda, Brandon Quigley, and John Zhang
Stewards: (September 5, 2007) Kevin Luchi, Mike Nappo, Matt Neff, Lisa Schauman
2.1.1: INTRODUCTION
Every process requires a great deal of planning in order to successfully accomplish the goals laid out by its designers and operators. In order
to accomplish these goals, however, personnel that are not familiar with the design must fully understand the process and the functions of
the control systems. Control systems consist of equipment (measuring devices, valves, etc.) and human intervention (plant operators and
designers). Control systems are used to satisfy three basic needs of every process:
1. Reduce the influence of external disturbances
2. Promote the stability of the process
3. Enhance the performance of the process
Verbal modeling is used for creating and also understanding a process control system. Verbal modeling consists of first receiving and then
gathering information about the process. A step-by-step process is then used to describe the control systems used to satisfy constraints and
objectives that have been outlined. In the following sections you will read what requirements are generally outlined for process control and
the step-by-step method used to meet these requirements.
2.1.2: PREREQUISITE INFORMATION REGARDING A PROCESS

For the sake of this article it is assumed that a process has already been designed and that certain restraints and criteria are provided by
either a customer, management, or the government. The goal of this section is to classify the types of criteria that are usually given. These
criteria then become the conditions that the control systems employed must satisfy. In general, there will be five sets of criteria, often
coming from different people and institutions. By gathering all of these criteria you will be able to describe the control system. If you do not
have a complete list of these criteria you must research the process to determine these constraints before beginning the step-by-step process
below.
Safety
The safe operation of a process is the biggest concern of those working in the plant and those that live in the surrounding community. The
temperatures, pressures, and concentrations within the system should all fall within acceptable limits, and these limits can be dictated by
either government agencies or company policy.
Production Objectives
The production objectives usually include both the amount and purity of the desired product. This criterion is generally set by the company
or customer.
Environmental Regulations
These come in the form of restrictions on the temperature, concentration of chemicals, and flowrate of streams exiting a plant. State and
federal laws, for instance, may dictate the exit temperature of a cooling water stream into a lake in order to prevent harm to aquatic wildlife.
Operational Constraints
Equipment found in the plant may have their own unique limitations, such as temperature or pressure that require proper control and
monitoring. For instance, a thermocouple may be damaged at extremely high temperatures, thus the location of the thermocouple must be
accounted for.
Economics
In general, a company will operate so that its profits are maximized. The process conditions that maximize these profits are determined by
way of optimization. Many costs must be considered when optimizing process conditions. Some of these costs are fixed, or will not change
with process variables (i.e. equipment costs) and others are variable, or do depend on process variables (i.e. energy costs). The overall
process is usually limited by certain factors including availability of raw materials and market demand for the final product. Therefore, the
economics of a process must be well understood before process changes are enforced.
2.1.3: STEP-BY-STEP METHOD FOR DESCRIBING CONTROLS AND THEIR PURPOSE
1.) Describe the Process
A brief description of the general process is needed while not dwelling on the details and calculations involved. The major steps of the
process, as well as inputs and outputs of the process, should be stated. A simple diagram should be provided detailing the chemical process
to help visualize the process.
2.) Identify Process Objectives and Constraints
The objectives and constraints of the process must be identified before process control actions can be performed.
The process objectives include the type, quantity, and quality of the product that is to be produced from the process. The economic
objectives, such as the desired levels of raw material usage, costs of energy, costs of reactants, and price of products, should also be
identified.
The process constraints include three different categories: operational, safety, and environmental limitations. Operational constraints refer
to the limits of the equipment used in the process. For instance, a liquid storage tank can only hold a certain volume. Safety constraints
describe the limits when the people or the equipment may be in danger. An example would be a pressure limitation on a reactor, which if
exceeded, could result in an explosion. Environmental constraints limit how the process can affect the immediate surroundings. For
example the amount of harmful chemicals that can be released before damage is done to nearby water supplies. All of these constraints
should be mentioned to build a robust control system.
Careful reading of the information provided to you by the customer, management, and government is required in order to properly identify
each constraint and objective. Often times, the process objectives will be very clearly laid out by the needs of the customer or management.
Operational constraints, or the limitations of the equipment being used, must be researched for each piece of equipment used in the process.
Generally, by satisfying the operational constraints a good portion of safety constraints are satisfied as well, but additional safety constraints
may exist and must be investigated by researching company policy and governmental regulations. Environmental regulations also have to
be researched through resources such as the EPA and Clean Air Act. Satisfying the economic aspect is largely determined by manipulating
additional variables after all other constraints and objectives have been met.
3.) Identify Significant Disturbances
Disturbances, in the sense of process description, are defined as inputs or external conditions from the surrounding environment that have
certain properties that cannot be controlled by the plant personnel. Examples of disturbances include ambient air temperature, feed
temperature, feed flow rate, feed composition, steam pressure changes, and cooling water temperature changes. Disturbances can drastically
affect the operation of a unit. A control system should be able to effectively handle all process disturbances. As such, all possible
disturbances must be identified and these disturbances need to be accounted for by the development of contingency plans within the
process.
4.) Determine Type and Location of Sensors
A proper design must ensure that adequate measurements of the system are obtained to monitor the process. To meet this goal, sensors must
be chosen to accurately, reliably, and promptly measure system parameters. Such parameters include temperature, flow rate, composition,
and pressure. Placement of sensors is important both in the usefulness of measurements as well as the cost of the system. Sensors should be
placed such that the measured quantities are appropriate in addressing control objectives.
5.) Determine the Location of Control Valves
Valves must be placed in a location to control variables that impact the control objectives. For example, control of the temperature of a
reactor could be obtained by placing a valve on either the stream of heating / cooling fluids or by placing a valve on the feed stream to the
reactor. One must determine which streams should be manipulated to meet process objectives.
6.) Perform a Degree of Freedom Analysis
The degrees of freedom in a system are equal to the number of manipulated streams (determined in step 5) minus the number of control
objectives and control restraints (determined in step 2). A degree of freedom analysis is used to determine if a system is being under- or
over-specified by the process objectives. The degrees of freedom come from the number of knowns and unknowns that are specified within
the system. If there are extra degrees of freedom present in a system, unused manipulated variables can be used to optimize the process. If
there are negative degrees of freedom, a system is over-specified because more objectives and restraints exist than manipulated streams. In
this case, all objectives cannot necessarily be met simultaneously and the least important objectives must be neglected. A system with zero
degrees of freedom is fully specified. All objectives can be met, but there is no room for optimization.
7.) Energy Management
In any system with exothermic or endothermic reactions, distillation columns, or heat exchangers, energy management becomes a factor that
must be accounted for. Heat must be removed from exothermic reactions in order to prevent reactor runaway, and heat must be supplied to
endothermic reactions to ensure desired production rates. Strategies such as pre-heating feed streams with the excess heat from a product
stream are helpful in maintaining efficient usage of energy, however, they also result in more complex processes that may require more
intricate control systems.
8.) Control Process Production Rate and Other Operating Parameters
The production rate can be controlled by a variety of manipulated variables. One manipulated variable may be the feed rate. The plant feed
rate can be changed and each subsequent unit can use its controls to accommodate this change, ultimately resulting in a change in the final
production rate. Other manipulated variables may also include reactor conditions, such as temperature and pressure. Temperature and
pressure affect reaction rates and can be used to alter the final production rate. It is important to choose the most suitable manipulated
variable to control production rate.
In addition to the production rate, other control objectives must be effectively managed by manipulated variables. For example, temperature
of an exothermic reactor may be controlled by the flow of a coolant stream passing over it in order to avoid dangerous high temperatures.
The pressure of a reactor may be controlled by the flow of feed gas in order to comply with the pressure limitations of the vessel.
9.) Handle Disturbances and Process Constraints
The effects of disturbances should be minimized as much as possible, in order to maintain the system at desired conditions and meet all
process objectives and constraints. Feedback or feedforward are specific control techniques and are common ways to overcome
disturbances. A feedback control works by studying the downstream data and then altering the upstream process. The actions executed are
reactive. Feedback can be viewed as an if-then statement: if a feed's temperature is detected to be lower than desired, then steam can be used
to preheat the feed. Feedforward is a more proactive approach in that it adjusts a manipulated variable before the disturbance is felt in the
process. Hence, if a sensor indicates low temperatures upstream of the feed, the feedforward control will counteract the effect of the cooler
upstream temperatures by preheating the feed before the feed temperature is effected. Note that a disturbance must be detectable and
measurable in order for the feedforward control to fix the anticipated disturbance before the system is effected.
Additionally, if constraints are reached during the process, controls should be implemented to avoid safety, operational, or environmental
hazards. This can also be done with feedback and feedforward controls on manipulated variables.
10.) Monitor Component Balances
Every component within a process, whether it is inert or not, should be accounted for at every step of the system in order to prevent
accumulation. This step is more crucial in processes that involve recycle streams. If such a stream is present, a purge stream is often
necessary to remove unwanted components. In addition, component balances are used to monitor yield and conversion or reveal locations in
the process where loss may be occurring. In order to monitor component balances, composition sensors are used.
11.) Control Individual Unit Operations
Most systems used today in industry employ the use of multiple unit operations. Each of these unit operations, however, needs to be fully
controllable in the sense that it has a control system that can adjust manipulated variables in order to maintain other parameters. For
instance, if an absorber is present, the system must be able to control the liquid solvent feed as some ratio to the gas feed. Another example
is a crystallizer. The refrigeration load of the crystallizer must be controllable in order to control the temperature.
12.) Optimize the Process
In most cases, there will be certain aspects of a process that will not be dictated to a designer and can be changed to make the overall
process more economical for the company. These are referred to as "unaccounted for" degrees of freedom and can be implemented as new
control valves or adjustable controller setpoints.
2.1.4: ALTERNATIVE METHOD OF VERBAL MODELING

This alternative method is also described in Peter Woolf's Recorded_Lectures Lecture 3. For purposes of this description, it is scaled down
to a single unit process. However, this method can easily be applied to describe an entire system of unit processes.
1. Describe the process in words
Some of the important questions to answer before delving deeper into a model are:
What are the components entering the system?
How do they enter? Separately? Combined stream? What physical states are they in?
What happens inside the unit process and what comes out at each exit point?
Remember to keep this part simple. There is no need to include chemical formulations or equations of any sort. Just lay out the basic flow of
material.
2. Define the primary goal of the process
The primary goal should be simple. Often, it is to maintain a specific measured variable above a minimum or below a maximum. In this
step, the only thing that needs to be determined is what the main goal is, and a few supporting details about why this is an important goal to
achieve.
For example, a primary goal could be to minimize the concentration of Compound Y in orange juice because studies show Compound Y
gives the juice a bad aftertaste.
3. Identify secondary processes that influence the primary goal
In a typical unit process, the primary goal will be directly influenced by one or two other aspects of the system. These can include
temperature, pressure, inlet conditions, and more and can occur at various points in the process.
The goal of this step is to determine which of these other process variables will be most likely to influence the primary goal and to step
down from there.
For example, the temperature of the orange juice mixer could have the greatest influence on production of Compound Y.
4. Identify safety and environmental risks
Next, you need to identify all of the points in the process that represent any type of risk. This will be important later in determining which
system variables need to be monitored.
Step through your process and identify any points that pose a significant risk of the hazards shown in the following figure.
Examples include: Boilers represent fire and explosion risks. Any stream with a dangerous chemical can represent corrosive, poison,
environmental, or all three risks.
5. Identify major costs associated with the process
How much something costs to produce is obviously a big deal in manufacturing. Identifying the largest sources of cost is critical in finding
ways to reduce cost overall. Typical places to start identifying costs are at inlet streams (what is the cost of raw materials) and at any portion
of the process where heat is added or removed.
It is important to include the high costs that can be associated with the risks identified in Step 4. Often the high cost of failure and risk
exposure will determine what other seemingly costly steps must be taken to ensure the safety of the process.
6. Identify variables you can directly manipulate
The basics of the process have been laid out, and now it's important to determine what variables you can actually control. Typically, you
only have direct control over the simplest of variables: switches and valves. Essentially, this just means that you cannot, in fact, choose a
temperature for your system and implement it. What you can do, is control a valve or switch that activates heating or cooling to control the
temperature.
During this step, you must decide where it is important to place these valves and/or switches. Use the information acquired previously about
the primary goal and secondary effects to determine what variables are worth controlling. Remember that you don't need to put valves
EVERYWHERE! Control valves are not costless and can also add unwanted complexity to your system. If you need to isolate equipment,
you can install manual valves. Keep the control valves to the needed level.
7. Identify sources of variation
In order to write a control scheme, you need to know what values in your system will change and why. Some common causes of variation
include:
Environment: ambient temperature
Other processes upstream or downstream: variable inlet conditions or outlet demand
Economic forces: product worth, material costs
Operators
Identifying what aspects of your process can be affected by these forces will allow you to assemble a more complete control scheme.
8. Describe your control system in words
Before you start trying to write everything out in computer code and mathematical equations, take the time to lay out your controls in
words. This is much like preparing an outline before writing a paper. It can save you from many headaches later on.
One example of generic, simple syntax for verbal modeling is: Maintain [system variable] at specified level by adjusting [variable I can
control].
2.1.5: THE BARKEL METHOD OF VERBAL MODELING

This method is an elaboration upon the steps outlined in Mr. Barry Barkel's lecture on September 29, 2009.
"If you design a system, you have the ethical responsibility to control it."
1) Understand the Process
Before you can control anything, you have to understand the process and how different parts of the system interact with each other. Make
sure that the overall process is understood. This includes inputs and outputs as well as major steps, however specifics are not necessary. You
should also be able to construct a diagram to help explain the process.
2) Identify Operating Parameters
Operating parameters can include temperature, pressure, flow, level, etc. Choose the parameters to manipulate in your system that will
safely result in the desired output.
3) Identify hazardous conditions
Consider all possible dangerous aspects of your process when designing your system. This could include a chemical overheating or a vessel
overflowing. It is imperative that you ensure the safety of your operators and being aware of all hazardous conditions can aid in this.
4) Identify measurables
The main three measureables addressed in this class are: temperature, pressure, and flow. However, there are many more measureables,
some more common than others. Here are some more: pH, humidity, level, concentration, viscosity, conductivity, turbidity, redox/potential,
electrical behavior, and flammability.
5) Identify points of measurement
It is important to place sensors in locations so that efficient and reliable measurements are taken of the system to monitor the process. For
example, in a distillation column, temperature sensors will display different temperatures at different locations down the tower. The sensors
must be positioned so that accurate readings of the system are given. Also, it is necessary to place sensors in an area of constant phase.
6) Measurement methods (thermo couple? choose for range)
After identifying what is to be measured and where it will be measured, you have to decide how it will be measured. For example, a
thermocouple can be used to measure the temperature. When choosing the equipment, be sure to check that the conditions of use fall within
the recommended range of operation for the equipment.
7) Select control methods
Decide whether to use feedback, feed-forward, cascade, or other types of control methods.
8) Select control system
A control system is a set of devices that will manipulate the actions of other devices in the system. A control system ranges from having an
operator manually open and close a valve to running a system with feedback such as with PID controllers. Control systems can vary from
relatively cheap to expensive. When picking a system, it would be most economic to choose the cheapest one that gets the job done. An
example of a control system is a PIC, or programmable interface controller.
9) Select control limits
When choosing setpoints for controllers, you will also have to decide on a range that the values are allowed to fluctuate between before a
corrective change is made. When selecting these limits, keep in mind that "equal" and "zero" do not exist due to the infinite number of
decimals that electronics are now able to handle. So you must further define what "equal" means, when is it "close enough" to count as
"equal" And when is the number small enough to count as "zero", is 0.1 or 0.01 or 0.0000000001 count as "zero"?
10) Define control logic
As every process is different, a customized code for each process must be written to tell the system what to do. For example, when a level
control is a tank has reached a critically high point, the logic should spell out the necessary changes needed to bring the tank level back
down. For example, this could be partially closing a valve upstream of the tank or partially opening a valve downstream of the tank.
11) Create redundancy system
In the real world, you must balance cost and efficiency/safety. On one hand, you don't want an out-of-control system if one control fails. But
on the other hand, you can't afford to order two of everything. The critical point to keep in mind is to optimize the safety while minimizing
the cost.
12) Define "fail-safe"
A fail safe is a set up in the control logic to ensure that in the event of a failure of a control method, the system will automatically reach a
safe condition so that there is little to no harm done to other equipment or personnel.
13) Set lead/lag criteria
Valves and other equipment do not necessarily open/close or turn on/off at the exact instant a button is pressed or the control logic kicks in.
There is often lag time associated with each controller. You must determine how long this lag time is so that you can take it into account.
14) Investigate effects of change before/after
Be sure to investigate effects of changing each controller. For example, what are the effects of closing/opening this valve?
15) Integrate all systems
Ensure that all systems are working together and that there are no holes in the system. Make sure that information does not fall through any
cracks in the system.
[Note - we can use an example of the Barkel method - RZ]
2.1.6: COMMON ERRORS

1. Impossible direct manipulations e.g., Change the concentration of salt in a tank
2. Missing the forest for the trees e.g., Sacrificing product quality for tight level control on a tank
3. Excessive or insufficient control e.g., Control every variable because you can or ignore the possibility of significant disturbances
2.1.7: WORKED OUT EXAMPLE 1

PLEASE NOTE: ALL VALUES IN THE FOLLOWING PROBLEMS ARE FICTICIOUS BUT MEANT TO BE LOGICAL
A heat exchanger uses steam to heat a stream of water from 50ºF to 80ºF. The water enters at a flow rate of 20 gallons per minute
from a nearby lake. The process costs $65 per hour and yields a profit of $2 per gallon of product. The steam is provided by the
plant and is 1000ºF, which is a temperature that the pipes can sustain. For safety reasons, the exchanger may only run for 12
consecutive hours and requires 4 hours to cool down. Using more than 10,000 gallons of water per hour would cause an
environmental disturbance to the water source. The diagram is shown below. Verbally model this system.
== Solution ==
The purpose of the process is to heat an incoming stream of water from a temperature of 50ºF to a temperature of 80ºF. The main equipment
involved is a shell-and-tube heat exchanger.
The product specification of the process is water at a flow of 20 gallons per minute and a temperature of 80ºF.
Economically, the process costs $65 per hour to operate. There are no costs for the raw materials, as the only inputs to the system are water
and steam. The finished product produces a profit of $2 per gallon. The economic objective is to reduce process costs while producing
sufficient product.
The operational constraints and safety concerns are due to the pipes. The pipes can only sustain a temperature of 1000ºF. Safety is a concern
because attempting to heat the incoming water to a certain temperature may cause the heat exchanger to malfunction, leading to equipment
damage and possible burn injuries to nearby personnel. The system may only operate for 12 consecutive hours, after which the system will
need to be cooled down for 4 hours to avoid the aforementioned hazards. A simplified assumption is that there are no constraints on steam
because it is provided by the plant and causes no safety issues. The only environmental constraints involve the incoming water stream. The
incoming water is gathered from the nearby lake, and a stream of greater than 10000 gallons per hour would cause a disturbance in the
equilibrium of the lake.
Significant disturbances can be found in the ambient air temperature, variable flow rates of the feed, and the temperature of the steam.
4.) Determine the Type and Location of Sensors
A flow sensor (FM) is placed at the incoming water stream. A temperature sensor (TS) is located on the product water stream. A flow
sensor is not needed for the steam stream for this problem because this value is not needed for control. A sensor could be placed here but the
information is not needed for this problem.
A flow valve is placed at the entrance of the incoming water stream. A flow valve is placed at the entrance of the steam.
6.) Perform a Degree-of-Freedom Analysis
There are two manipulated variables: the flow of the water feed stream and the flow of the incoming steam. There are two control
objectives: the flow of the feed stream, monitored by the flow sensor, and the temperature of the product, monitored by the temperature
sensor. Therefore the system has zero degrees of freedom.
The incoming steam is used to transfer heat to the cool water feed. The temperature sensor on the product stream determines the applicable
setting on the steam flow valve.
The process production rate is controlled by the flow valve on the entering water stream. The water temperature is controlled by the flow
valve on the incoming steam.
Changes in the ambient air temperature can be detected by the temperature sensor, and can be corrected by the flow valve on the incoming
steam stream. Variable flow rates of the water feed stream can be detected by the flow sensor and compensated by adjustments on the flow
valve on the water feed stream. Changes in the temperature of the steam can be detected by the temperature sensor. The flow valve on the
steam stream can be adjusted to increase or decrease the flow of steam and subsequently the amount of heat exchanged.
A vent is located on the heat exchanger to release excess steam from the system. Aside from that, any accumulation is unlikely and can be
neglected.
The outlet temperature of the product stream is controlled by the flow valve on the steam feed stream. The flow of the incoming water
stream is controlled by the flow valve on the water feed stream.
One might notice that the process is only using 1,200 gal/hr of water, well below the 10,000 gal/hr environmental constraint. If the profit of
the process is linear with the flow-rate of water, then increasing the flow-rate of water will increase the profits for the company. (With the
constraints specified, this is a Linear Programming optimization problem. The optimal setpoint falls on a boundary condition.) However, the
flow-rate of water entering the system is already specified, which results in zero degrees of freedom. (Zero degrees of freedom implies there
are no further control valves or setpoints.) Further investigation should be conducted to determine the reason for the flow-rate specification.
When considering increasing the flow-rate of water into the system, one should also check that the other constraints are not violated.
A process converts phenol into salicylic acid through a series of two reactors. Phenol and NaOH are fed in the liquid phase into the
first reactor where it reacts with gaseous carbon dioxide that is pumped in. Assume constant fresh feed temperature and that the
feed flow rate is within operational constraints. Management has dictated that salicylic acid production must be 200 moles per hour.
Also, management would like the product stream to have a molar composition of 80% salicylic acid. Due to environmental
concerns, a maximum flow rate of 10000 gallons per hour of cold water can be used to cool the first reaction chamber. The valve
controlling the flow of cold water does not allow a flow rate in excess of 7500 gallons of water per hour. The salicylic acid product is
used in another process to produce aspirin, which has a market value of $10 per mole. The first reactor can be operated at pressures
up to 200 atm, while the second can be operated at pressures up to 10 atm. The first reaction is exothermic, while the second
reaction is assumed to generate negligible heat. A diagram of this process is shown below, as well as the reaction scheme. Verbally
model this system.
== Solution ==
The purpose of the process is to convert phenol into salicylic acid, the precursor for aspirin. First, the phenol is reacted with gaseous carbon
dioxide (CO2) and sodium hydroxide (NaOH) under high pressure and temperature in the first reactor. The product of this reaction is then
combined with sulfuric acid to form salicylic acid in the second reactor. The reaction scheme is shown above.
The process is expected to produce 200 moles per hour of salicylic acid. The product stream must contain at least 80% by moles of salicylic
acid. The equipment used in the process dictates the operational limitations. The first reactor vessel can be operated up to a pressure of 200
atm, while the second reactor vessel has a 10 atm upward pressure limit. As such, pressures in excess of these limits must be avoided. Since
the first reactor will generate a significant amount of heat, the heat must be removed to avoid damage to equipment and possible runaway
reactions. Therefore, a heat exchanger (in the form of a reactor jacket in this case) with cool water should be included to decrease the
temperature of the reactor. Economic concerns demand that phenol, sodium hydroxide, and sulfuric acid should not be used in extreme
excess. The costs of these materials and the energy costs required to process them affect the overall profitability, so these compounds should
not be wasted. Environmental regulations limit the use of water to cool the reactor at 10000 gallons per hour, however the valve constraints
limits the amount of water to only 7500 gallons per hour.
The amount of cold water available to cool the reactor can be considered a disturbance because it comes from a reservoir outside of our
control. The ambient temperature is also a disturbance. If it drastically increases, the amount of cold water needed to cool the reactor would
need to increase as well. Composition of the feed streams will be assumed to be constant in this example. Therefore, they are not considered
disturbances.
4.) Determine the Type and Location of Sensors
A temperature sensor (TS) and pressure sensor (P) are located on the stream exiting the first reactor vessel. A flow meter (FM) is located on
the product stream leaving the second reactor. A composition sensor (CS) will also be located on the product stream leaving the second
reactor. The pressure drop can be controlled through the decompressor and thus is a control.
Control valves are located on the feed stream containing the phenol and sodium hydroxide, the incoming cold water to the first heat
exchanger, and the sulfuric acid feed stream. There is also a pump located on the carbon dioxide stream that enters the reactor.
6.) Perform a Degree of Freedom Analysis
There are 3 valves, 1 pump, and 1 decompressor but 5 objectives. This results in zero degrees of freedom. The valve located on the sulfuric
acid feed stream is meant to meet the composition constraint placed on the product stream leaving the second reactor. The valve located on
the feed stream carrying the reactants is set to satisfy production requirements. The valve on the cold water stream is used to maintain
reactor temperature, which satisfies an operational constraint. The pump is to ensure the correct pressure is achieved in the reactor, also
satisfying an operational constraint. The decompresser is to maintain a pressure of less than 10 atm in the second reactor, thus satisfying
another operational constraint.
The heat from the exothermic reaction in the first reactor is transferred to the cold water stream. The hot water stream exiting the reactor
vessel jacket could be used to heat streams on other processes. The second reactor is assumed to generate negligible heat during the
reaction, thus any release of heat from the reactor will be considered safe to release into the environment surrounding the process.
The production rate is measured by the flow sensor on the product stream and this signals the control valve on the feed stream through a
feedback mechanism to change the production rate as necessary.
If the temperature sensor on the reactor exit stream exceeds a certain level due to a diminished cold water supply, the feed stream valve
would decrease the amount of reactants entering the reactor. The amount of feed would also be decreased if more than 7500 gallons per hour
of cooling water were needed, as this is an operational constraint. If the pressure gauge controlling the pump begins to read higher than
allowed pressures, the pump would decrease the flow of the carbon dioxide entering the reactor. Also, if the pressure gauge reads out a
pressure that will be too high for the second reactor, the decompresser will be allowed to disperse more pressure. If ambient air temperature
drastically increases, the temperature sensor would open the cold water valve allowing more cooling water to enter the reactor vessel jacket.
If the composition of the product stream falls below 80 mole percent of salicylic acid, then the valve controlling the sulfuric acid feed would
allow more sulfuric acid into the second reactor to increase the conversion of reactants.
The composition sensor and flow meter on the product stream leaving the second reactor will account for every species to ensure that there
is no accumulation or loss within the system.
The first reactor vessel's pressure is fully controlled by the pressure gauge and pump system and its temperature is fully controlled by the
temperature sensor which controls the reactant feed valve and the cool water valve. The second reactor's pressure is fully controlled by the
same pressure gauge and the decompresser system, and its temperature will be highly dependent on the amount of cooling water used to
cool the product exiting the first reactor.
Since there are no unaccounted degrees of freedom, there are no valves to adjust in order to optimize the process. It should be noted,
however, that if there was no constraint on the composition of the product stream, the sulfuric acid feed valve would have become an
unaccounted for degree of freedom. If this had been the case, the valve could be adjusted to maximize the profit of the process. In order to
maximize the profit, the benefits of having higher conversion and more product would have to be weighed against the increase costs of
using more sulfuric acid feed.
2.1.9: SAGE'S CORNER

An example of verbal modeling
http://www.youtube.com/watch?v=YTs_IWccnvw&feature=player_embedded
Powerpoint Slides from the example
2.1.10: REFERENCES
Luyben, William L., Tyreus, Bjorn D., and Luyben, Michael L. “Chapter 8: Eastman Process” in Plantwide Process Control, McGraw-Hill,
New York, pp. 251-272.
Luyben, Michael L., Tyreus, Bjorn D., and Luyben, William L., "Plantwide Control Design Procedure" in AIChE Journal Dec. 1997, Vol.
43, No. 12 pp. 3161-3174.
Riggs, James B. and Karim, M. Nazmul. “Chapter 17: Multiunit Controller Design” in Chemical and Bio-Process Control, Ferret
Publishing, pp. 491-504.
Stephanopoulos, George. Chemical Process Control: An Introduction to Theory and Practice, Prentice Hall, New Jersey, pp. 1-41.
2.1: Verbal Modeling- process description, control specifications, and connections is shared under a CC BY 3.0 license and was authored, remixed, and/or
curated by Peter Woolf et al. via source content that was edited to conform to the style and standards of the LibreTexts platform; a detailed edit history is
available upon request.
2.2: DEGREES OF FREEDOM- IMPORTANCE, CALCULATION PROCEDURE,
AND EXAMPLES
Authors: (13 December 2009) Jon Butler, Matthew J LaVelle
2.2.1: INTRODUCTION
In control engineering, a degree of freedom analysis is necessary to determine the regulatable variables within the chemical process. These
variables include descriptions of state such as pressure or temperature as well as compositions and flow rates of streams.
2.2.2: DEFINITION
The number of process variables over which the operator or designer may exert control. Specifically, control degrees of freedom include:
1. The number of process variables that may be manipulated once design specifications are set
2. The number of said manipulated variables used in control loops
3. The number of single-input, single-output control loops
4. The number of regulated variables contained in control loops
The following procedure identifies potential variables for manipulation.
2.2.3: THE PROCESS

The method we will discuss is the Kwauk method, developed by Kwauk and refined by Smith. The general equation follows:
Degrees of freedom = unknowns - equations
Unknowns are associated with mass or energy streams and include pressure, temperature, or composition. If a unit had Ni inlet streams, No
outlets, and C components, then for design degrees of freedom, C+2 unknowns can be associated with each stream. This means that the
designer would be manipulating the temperature, pressure, and stream composition.
This sums to an equation of
Total Unknowns = Ni*(C+2) + No*(C+2)
If the process involves an energy stream there is one unknown associated with it, which is added to this value.
Equations may be of several different types, including mass or energy balances and equations of state such as the Ideal Gas Law.
After Degrees of Freedom are determined, the operator assigns controls. Carrying out a DOF analysis allows planning and understanding
of the chemical process and is useful in systems design.
2.2.4: APPLICATIONS
Single phase systems
All outlet streams have the same composition, and can be assumed to have the same temperature and pressure
Multiple phase systems
An additional (C-1) composition variable exists for each phase
Complete Process
When connecting units which share streams, one degree of freedom is lost from the total of the individual units
2.2.5: EXAMPLE 1: BLENDER

This example (from the ECOSSE Control Course web page listed below) investigates degrees of freedom in a simple vapor mixing unit.
Two gaseous streams enter a vessel and exit as a single well-mixed stream (Figure 1). We will apply the above equation to determine
degrees of freedom.
Figure 1: Blender Schematic
Here, there are 3 streams, each with C+2 unknowns for a total of 3C+6 Unknowns.
We have C mass balances and 1 energy balance for a total of 2C+1 equations. We also know composition, pressure, and temperature for the
incoming streams. This adds 2C+2 to the equation. Putting everything together gives:
Degrees of freedom = 3C+6 - (2C+1 + 2C+2)
Hence, the system has 3 degrees of freedom. Therefore, we can fix outlet composition, pressure, and flow rate. Figure 2 shows an example
control scheme:
Figure 2: Blender with controls
2.2.6: REFERENCES
Ponton JW, 1994, Degrees of Freedom Analysis in Process Control, Chemical Engineering Science, Vol. 49, No. 13, pp 1089 - 1095.
eweb.chemeng.ed.ac.uk/courses/control/restricted/course/third/course/formal.html
2.2: Degrees of Freedom- importance, calculation procedure, and examples is shared under a CC BY 3.0 license and was authored, remixed, and/or curated
by Peter Woolf et al. via source content that was edited to conform to the style and standards of the LibreTexts platform; a detailed edit history is available
upon request.
2.3: INCIDENCE GRAPHS -INTERPRETATIONS, CONSISTENCY, AND
INCONSISTENCY
3.1 Introduction
Incidence graphs are useful tools for chemical engineers to model various relationships in a process. They are used to systematically map an
entire chain of processes and controllers to describe the effect, which each element has on the others. This helps to visualize the possible
process pathways or a chain of effects.
Below is an example of an incidence graph. The circles are nodes which represent a particular device or controller, e.g. a temperature
sensor. The arrows indicate the directional pathway of influence. For instance, a temperature sensor node would be connected to an adjacent
node (such as a heat exchanger). This heat exchanger node would in turn be connected to other nodes representing devices or controllers.
This type of diagram could be extremely useful in identifying the redundancies within a control system. A more elaborate application of
incidence graphs will be discussed in the Worked Out Examples section.
2.3.1: MONOTONICITY
Before proceeding to the next few sections, it is imperative to understand the basics of monotone systems. A monotonic function defines a
function that preserves a given order. Monotonic functions are classified as monotonic increasing or monotonic decreasing. Taking the
example of a function f(x), for a monotonic increasing function an increase in variable ‘x’ would result in an increase in f(x). On the other
hand, for a monotonic decreasing function an increase in variable ‘x’ would result in a decrease in f(x).
Monotonic Increase : If x≥y, then f(x)≥f(y)
Monotonic Decrease : If x≥y, then f(x)≤f(y)

As long as a function is either always monotone increasing or monotone decreasing, it is referred to as a monotone function.
Monotonic functions are ideal systems to control because changes in one variable/device lead to known changes in other variables/devices.
A non-monotonic function on the other hand, does not yield a straightforward output for a particular input. In these cases IF-THEN-ELSE
statements are used to illustrate the controls for the nodes. For instances, IF node 1 is activated, THEN node 2 is repressed and so on.
An example of a possible non-monotonic graph
2.3.2: INCIDENCE GRAPHS

Incidence graphs (also “Levi” or “Causation” graphs) consist of monotonic connections between nodes, whereby each connection indicates
a particular effect of one factor on another. The term “monotonic” means that a change in the initiating device influences only one aspect of
the target device. This simplification facilitates describing how control systems work through the use of incidence diagrams.
In reality, not all interactions are monotonic. Some control devices can simultaneously affect several aspects of other devices or variables.
This may result in a non-monotonic relationship, in which one factor increases another over one region, but diminishes it over another
(imagine a parabola, which is a non-monotonic curve). Temperature often has such a relationship with the rate of a reaction: for an
exothermic reversible reaction, increasing the temperature will increase yield up to a certain point; however, further increase will diminish
the yield.
Incidence graphs can only be used to represent monotonic relationships. Therefore, over simplification of a system can cause problems in
constructing an incidence graph. Similarly a non-monotonic relationship can be resolved by breaking down a factor into separate
components, each of which has only monotonic relationships with other factors.
When two factors are connected by multiple signal paths, incoherence between the results of different paths leads to inconsistency. In other
words, the outcome cannot be predicted because the incidence graph does not communicate quantitative information. For instance, in the
reaction rate example, reaction rate could be separated into the variables which affect it: concentrations, rate constant, and equilibrium
constant. An incidence graph can then be constructed to demonstrate that temperature and reaction rate have an inconsistent relationship.
A diagram is consistent (to be discussed later in detail) if all signal paths between two nodes, for every pair of nodes, are in agreement with
one another (i.e. they have the same result).
2.3.2.1: TRAVERSING
Every incidence graph establishes a particular way to depict the direct effects of one device on its neighbors. Typically, a line will connect a
node to its neighbors--these lines are terminated by either a bar or an arrowhead.
An arrowhead at the end indicates that the initiator (from which the line is coming) is either increasing the value of the target (where the
arrow is pointing) or simply activating it (turning it on). The context of the connection will determine whether something is being increased
or turned on. For example, it would be illogical to assume that a drop in temperature would cause a temperature controller to deactivate a
pressure controller. Instead, it would be more prudent to assume that the temperature controller would adjust the pressure sensor.
arrowhead bar
A perpendicular bar indicates that the initiating device is either decreasing the value of the target device or inhibiting it (turn off). Note that
these indications are not necessarily strictly followed. Once again, context of the situation determines which interpretation should be used.
perpendicular bar
Graphs consisted of the arrowhead bar and the perpendicular bar allow you to see the result of an increase/decrease in one aspect of the
initial device on a designated device in the system. Everything done after the initial device comes as a result of changing the initial device.
In addition, these graphs provide the possibility of simplifying complicated relationships between devices into one conceptual unit. The
relationship between the initiating device and target nodes can be described by the following set --> {initiator,target1,target2,etc}. It is read
as "the increase or decrease in the initiating device affects the target nodes in the following manner... (+ for increase/activate and - for
decrease/inhibit). The initiator can be any device or variable in the system as long as the arrows support the designation. In other words, you
can show the effects of any device on any other device, as long as the pathway is supported by the diagram. There are often multiple routes
that a signal can travel.
2 helpful tips to keep in mind as you interpret an incidence graph:
1. Always travel from tail to head (or source to target) when traversing the graph from the initiating device to the designated device. Keep
in mind that it is possible to have arrowhead and/or perpendicular bars pointing in both directions in between two devices.
2. Arrowhead bars don’t always mean increase as perpendicular bars don’t always mean decrease. Arrowhead bars indicate that the effect of
performing an action to the device at the tail of the bar would yield the same effect to the device at the head of the bar. Penpendicular bars
indicate that the effect of performing an action to the device at the tail of the bar would yield the reverse effect to the device at the head of
the bar.
The following is a simple example of an incidence graph. We will refer to the circles below as “nodes” from this point onward. Nodes are
the selected devices or variables in a control system.
This incidence graph indicates that:

a. A change in an aspect of 1 will change some aspect of device 2 in the same direction -- otherwise written as {+,+}
The ends indicate the direction of influence as well as the type of influence.

a. increase in an aspect of 2 (initiator) decreases some aspect of 1 (target)
b. indicated by {+,-}
This next example builds on the previous two...

Given an arbitrary 3,1,2 order, an increase in 3 (initiator) will make 1 (target) go up. 1’s increase will then make 2 decrease. The symbolic
representation for this would be {+,+,-}. One can see that this diagram shows how node 3 influences everything directly and indirectly
connected to it.
Now, given a 1,2,3 order, node one becomes the initiator. The symbolic representation then changes to {+,-,-}. NOTE: the last node is
negatively affected because 2’s impact on 3 is shown to increase when 2 increases. But since 1 decreases 2, then the opposite is true!
The main objective of these diagrams is to determine the effect that a change in the starting node has on another node in the system.
However, in some cases, you can follow a pathway that ends at the starting node in order to check for ambiguity. For example, the 1,2,3,1
order yields a result of {+,-,-,-}. Since increasing node 1 cannot cause a decrease in itself, it is ambiguous.
Incidence graphs for complex controls systems will usually indicate that there are several ways for a device to influence another device in
the system.
Multiple paths
Remember these are done with respect to increases in a node’s value. So, if a path leads you to a point where you decrease a node, then you
have to do the opposite of what the diagram is indicating for all steps after that (for that route).
The incidence graph indicates:
1. For an order {1,2,3}: you observe {+,-,-}... remember not to count beginning node twice…
2. For an order {1,3}: you observe {+,+}
2.3.3: CONSISTENT, INCONSISTENT, AND PARTIALLY CONSISTENT

When designing controls for systems it is always important to keep in mind the need for redundacy and feedback control. Redundancy in a
system will ensure that a system is monitored by multiple controllers. So, if one controller was to fail there would be another to ensure the
same desired outcome. Consistent graphs can be used to describe redundancy. Feedback control is also very important and it is not good to
have the system overreact to external stimuli. A controlled response is desired. Inconsistent graphs can be used to describe feedback control.
Additionally, systems exist that both need to be redundant and feedback controlled and these can be described with partially consistent
graphs.
2.3.3.1: CONSISTENT GRAPHS

Consistent pathways happen when all pathways from the same starting node, that lead to the same target node, give the same output.
For example,
All possible pathways from node 1 to target node 4 yield the same output. These are called consistent pathways.
Paths(s) Sign series

1,4 (+,-)
1,2,4 (+,+,-)
1,3,4 (+,+,-)
From the table and graph above, all pathways from node 1 to node 4 yield the same outcome. Node 4 is being decreased.
A consistent graph is made up of only consistent pathways for all possible pathways between any combination of two nodes. The graph
below is an example of a consistent graph.
1,2,3 (+,+,-)
1,3 (+,-)
3,1,2 (+,-,-)
3,2 (+,-)
1,3,1 (+,-,+)
1,2,3,1 (+,+,-,+)
3,2,3 (+,-,+)
3,1,3 (+,-,+)
3,1,2,3 (+,-,-,+)
All pathways leading from node 1 to node 3 -

1,2,3 : (+,+,-)
1,3 : (+,-)
Since the two pathways cause the same change to node 3, this sub-pathway is consistent.
All pathways leading from node 3 and to node 2 -
3,1,2 : (+,-,-)
3,2 : (+,-)
Since the two pathways cause same change to node 2, this sub-pathway is consistent.
All pathways leading from node 1 and back to node 1 -
1,3,1 : (+,-,+)
1,2,3,1: (+,+,-,+)
Since the two pathways result in the same change to node 1, this sub-pathway is consistent.
3,2,3 : (+,-,+)
3,1,3 : (+,-,+)
3,1,2,3: (+,-,-,+)
Since the three pathways give the same effects to node 3, this sub-pathway is consistent.
Because all sub-pathways of this incidence graph are consistent, the entire process is consistent.
The table shown below gives examples of many different incidence diagrams and shows whether or not they are consistent.
Another type of consistent graph is one in which the valves of the system are not dependent on one another for feedback. In other words,
there is no proportion of the output signal from a particular valve of a system is passed to the input of another valve. An example of this
would be to consider the following process diagram:
This process is of a distillation column in which the bottom columns collect high intensity liquids and the liquid is removed from a series of
evaporators. The evaporator is a single-effect evaporator equipped with a simple heat exchanger, a separation vessel and a steam jet ejector
to supply the vacuum. The product is then sent to an oven where it is dried and sent to the manufacturer. It is the chemical engineer’s task to
create a consistent incidence graph in which the valves in the diagram do not depend on each other’s feedback. An example of this is below:

V1, TI, PI, TI (+,+,+,+)
V2, TI, PI, TI (+,-,-,-)
V3, PI, TI, PI (+,-,-,-)
Based on the table above, the incidence graph is consistent because the valves are not dependent on one another for feedback.
2.3.3.2: INCONSISTENT GRAPHS

Inconsistent pathways happen when pathways from the same starting node, that lead to the same target node, give different output.
For example,
Pathways from node 1 to target node 4 do not yield the same output. These are called inconsistent pathways.

1,4 (+,+)
1,2,4 (+,+,-)
1,3,4 (+,-,+)
From the table and graph above, all pathways from node 1 to node 4 do not yield the same outcome. Node 4 is being decreased while
simultaneously being increased.
An inconsistent graph is made up of only inconsistent pathways for all possible pathways between any combinations of two nodes. The
graph below is an example of an inconsistent graph.

1,2 (+,+)
1,3,2 (+,-,-)
1,2,1 (+,+,-)
1,3,2,1 (+,-,-,+)

1,2 : (+,+)
1,3,2 : (+,-,-)
Since the two pathways cause different changes to node 2, this sub-pathway is inconsistent.
1,2,1 : (+,+,-)
1,3,2,1: (+,-,-,+)
Because all sub-pathways in this incidence graph are inconsistent, the incidence graph is inconsistent.
2.3.3.3: PARTIALLY CONSISTENT GRAPHS

Partially consistent graphs are made up of both consistent and inconsistent pathways. If you increase node 1, all paths leading to node 4
cause a decrease in the node which means it is a consistent pathway. If node 1 is increased, one path leading to node 2 causes an increase
and another path causes a decrease in the node. This is an inconsistent pathway. Since this model contains both at least one consistent and
one inconsistent pathway, it is a partially consistent model.

1,2,3,4 (+,+,+,-)
1,2,4 (+,+,-)
1,4 (+,-)
1,2 (+,+)
1,4,2 (+,-,-)

1,2,3,4: (+,+,+,-)
1,2,4 : (+,+,-)
1,4 : (+,-)
Since the three pathways cause the same change to node 4, this sub-pathway is consistent.
All pathways leading from node 1 and to node 2 -
1,2 : (+,+)
1,4,2 : (+,-,-)
In this graph there are sub-pathways that are consistent and inconsistent, therefore the incidence graph is partially consistent.
2.3.3.4: SUMMARY
Just because a process is inconsistent does not mean that the desired results cannot be achieved. All it means is that you cannot consistently
achieve the desired results. Take the example of the flow of a final product. In a consistent process, no matter which path is taken, the flow
of the final product will always respond the same, whether desirable or not. In an inconsistent process, changing the path will not always
cause the flow of the final product to change the same way, but rather depends on the path taken.
In all, inconsistent processes are not bad. Often, processes will be very complex, containing many variables, and inconsistent processes will
be unavoidable. Care must just be taken with complex, inconsistent processes to achieve desirable results.
You are a process engineer and you have been assigned to devise a process control for a new reactor. The optimum operating condition for
the reaction inside the reactor is 220 oC ± 10oC. The reaction is exothermic and often goes way past the optimum temperature. It is
imperative that the reaction should not go above 250oC. The reactor has a built in heater (H1) to keep temperature at its optimum
temperature and a cooling water feed to cool the reactor in case temperature goes past 250oC. The reactor also has temperature sensor
installed to monitor the temperature. If temperature goes above 250oC, the desired mechanism is to increase the flow of water in the cooling
jacket surrounding the reactor and to turn off the heating element inside the reactor until it reaches its optimum temperature. As a control
redundancy mechanism, a pressure sensor is also installed inside the reactor. The pressure control will send signal to the cold water feed if
pressure reaches 1.2 atm which is the critical pressure. Draw a causation graph governing the situation when the temperature gets too hot
inside the reactor. Base your graph on the following variables:
Note: (There may be more than one correct solution to the assigned problem)
a. temperature
b. temperature sensor (T1)
c. cooling water feed (V3)
d. heating element (H1)
e. pressure
f. pressure sensor (P1)
High temperature will 'activate' the temperature sensor, telling it that the temperature has exceeded the optimal temperature and an action
need to be done to bring it back to the desired temperature. The temperature sensor would then open the valve for the cooling water feed
which would cause the water to flow through the jacket. This results in the temperature inside the reactor to decrease. The reverse happens
for the heating element. An activation of the temperature sensor would translate to an inhibition of the heating element in which the heating
element would stop increasing the temperature of the reactor. Temperature and pressure are directly related by ideal gas law and thus affects
each other in both ways. Thus, for the control redundancy mechanism, an increase in pressure would activate the pressure sensor which
would then activate the cooling water feed to cool the temperature for the reactor.
Logic control:
IF T1 =< 220 oC, THEN H1 is on, ELSE H1 is off
IF T1 >= 250 oC, THEN V2 and V3 is open, ELSE V2 and V3 is close
IF P1 >= 1.2 atm, THEN V2 and V3 is open, ELSE V2 and V3 is close

After completing the assignment of devising the temperature control, your boss asked you to devise a mechanism to control the flow of raw
materials and products. Draw a causation graph showing negative feedback relations between L1, V1 and V4 when the level inside the
reactor is increasing (this is an example of an inconsistent pathway). The normal operating conditions are shown below:
V1 is 50 gal/min, V1 is usually open
V4 is 50 gal/min, V4 is usually open
L1 is 20 meters, L1 should not go above 30 meters and should not be below 10 meters
The level sensor monitors the fluid height inside the jacketed CSTR. Assuming the level inside the reactor is increasing steadily, the level
sensor will send a negative feedback signal to V1 to decrease the input flow.
Logic control:
IF L1 > 30 meters, THEN V1 is closed
IF L1 < 10 meters, THEN V4 is closed

Determine whether the following causation graph is consistent, inconsistent or partially consistent. Include your reasoning.
Solution:
All relationships are consistent. Each variable 1,2,3,4 has a simple relationship which is monotonic among the other variables in the set.
{1,2} (+,-)
{1,2,3} (+,-,-)
{1,3} (+,-)
{1,4,3} (+,+,-)
{1,4} (+,+)
{2,1} (+,-)
{2,3} (+,+)
{2,1,3} (+,-,+)
{2,1,4} (+,-,-)
{4,3} (+,-)
2.3.7: REFERENCES
Woolf, Peter (2006). A Mini-Guide to Consistent Graphs.
Woolf, Peter (2006). Terms Review.
Angeli, David & Sontag, Eduardo D. (2003). Monotone Control Systems.
2.3.8: CONTRIBUTORS AND ATTRIBUTIONS

Authors: (September 26, 2006) Christopher Garcia, Anwar Stephens, Winardi Kusumaatmaja, Meng Yang Ng
Stewards: (September 5, 2007) Alexander Voice, Andrew Wilkins, Ibrahim Oraiqat, Rohan Parambi
2.3: Incidence Graphs -Interpretations, Consistency, and Inconsistency is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by
upon request.
2.4: EXCEL MODELING - LOGICAL MODELS, OPTIMIZATION WITH SOLVER
FOR NONLINEAR REGRESSION, SAMPLING RANDOM NUMBERS
2.4.1: INTRODUCTION
Microsoft Excel program is one of the most popular and useful computer programs for a wide variety of numerical applications. Excel
features several different functions, interfaces and graphing tools which can be applied in many fields. The following functions might be
especially useful for logical programming in Excel:
Logical functions (ex: IF, OR and AND): These functions can be used in control analysis, particularly in safety regulation (for example,
if the temperature exceeds X degrees, shut down the reactor and send cooling water to the jacket)
Solver: This function can be used to maximize, minimize or try to obtain an input value in a cell by varying referenced cells.
Random Number Generator. This can be used as a tool for probability or randomly selecting data points for analysis from a large pool of
points.
While it might be less accurate than other modeling programs, the Excel model provides good estimates, and facilitates observation and
understanding of process behavior while requiring minimum computer knowledge. The following sections will provide you with a better
understanding of these common Excel functions, followed by examples in engineering applications.
2.4.1.1: LOGICAL METHODS

A logical test is any expression that can be evaluated as TRUE or FALSE with specified input conditions. Commonly used logical tests are
IF, OR and AND statements. All three statements can be used in conjunction with each other for more complex and complete modeling.
Note: all functions and formulas in Excel require an equal (=) sign in front of the syntax.
2.4.1.1.1: EXCEL'S IF STATEMENT
The IF function in Excel tests values and formulas in the target cells and returns one specified value "if" the condition input evaluates to be
TRUE and another value if it evaluates to be FALSE. The IF statement in Excel is called out with the following syntax:
IF(logical_test,value_if_true,value_if_false)
Logical_test: any expression or correlation of values that can be evaluated as TRUE or FALSE.
Value_if_true: the value or expression that is returned IF the condition in the logical test returns TRUE.
Value_if_false: the value or expression that is returned IF the condition in the logical test returns FALSE.
The value_if_true and the value_if_false part in the syntax may vary and can be numerical values, expressions, models, and formulas. If
either of the values is left blank, the test will return 0 (zero) correspondingly.
Sample coding:
IF(1+1=2, "Correct!", "Incorrect!")
For the logical test part, 1+1 is equal to 2, and the iteration will return TRUE. The value_if_true is called for the TRUE response, and
"Correct!" will be output in the corresponding cell.
It is possible to nest multiple IF functions (called Nested IF) within one Excel formula. Such a formula is useful in returning multiple
answers, compared to a a single IF function which can return only two possible answers.
A Nested IF statement has the following syntax:
IF( logical_test1, value_if_true1, IF(logical_test2, value_if_true2, value_if_false) )
The following example shows how a Nested IF function is used to make a spreadsheet that assigns a letter grade to different scores.
Suppose that grade distribution is:
@ A= Greater than or equal to 80
@ B= Greater than or equal to 60, but less than 80
@ C= Less than 60
This logic function can be written in several different ways. If the score is tested to be the letter grade C first, then checked to see if it is a B
or an A, the syntax could be written as follows:
IF(A5<60,"C",IF(A5<80,"B","A"))
Note that the Nested IF statement can be placed in either the value_if_false OR the value_if_true positions. In the example above, it was
placed in the value_if_false position because of how the conditional was written.
Additional letter grades could be added by including more nested IF functions inside the existing function. One formula can include up to 7
nested IFs, which means you can return up to 8 different results.
2.4.1.1.2: EXCEL'S OR STATEMENT
The OR statement in Excel is similar to the IF statement. It returns TRUE if any argument or expression of values is TRUE, and it returns
FALSE if all arguments or expression of values are FALSE. The syntax for OR is:
OR(logical1,logical2,...)
logical1: there can be up to 30 arguments for the logical expression to test and return either TRUE or FALSE. If all of the arguments are
false, the statement returns FALSE for the entire expression.
Sample coding:
OR(1+1=2, 2+2=5)
In the logical1 part, 1+1 is equal to 2 and will return TRUE. In the logical2 part, 2+2 is not equal to 5 and will return FALSE. However,
since one of the test returns TRUE (logical1), the function returns TRUE even if logical2 is FALSE. TRUE will be output in the
corresponding function cell.
2.4.1.1.3: EXCEL'S AND STATEMENT
The AND statement returns TRUE if all its arguments or values are TRUE and returns FALSE if one or more argument or value is FALSE.
The AND statement syntax is:
AND(logical1,logical2,...)
logical1: similar to OR statement with up to 30 arguments to test that could return either TRUE or FALSE.
Sample coding:
AND(1+1=2, 2+2=5)
Logical1 will return TRUE for the correct calculation while logical2 will return FALSE for the incorrect calculation. However, since one of
the expressions is FALSE, the AND logical test will return FALSE even if one of the tests is true.
It is common and useful to understand how to implement different logical functions into each other. A good example would be:
IF(AND(2+2=4,3+3=5),"Right Calculation","Wrong Calculation")
During iteration, the AND statement should return FALSE since one of the expression/calculation is incorrect. The FALSE return will
output "Wrong Calculation" for the IF statement iteration.
2.4.1.1.4: OTHER USEFUL EXCEL FUNCTIONS
Excel has many useful functions that can help any engineer who is using Excel to model or control a system. An engineer may want to
know the average flow rate, number of times the liquid in a tank reaches a certain level, or maybe the number of time steps that a valve is
completely open or completely closed. All of these situations and many more can be addressed by many of the simple built-in functions of
Excel. Below are some of these functions and a brief definition of what they compute and how to use them. They are followed by an image
of a sample excel file with an example of each function in use. To see a list of all the built-in functions Excel has to offer, click shift-enter
on the cell.
AVERAGE(number1,number2,...) - Returns the average or arithmetic mean of the arguments. The arguments can be numbers, arrays, or
references that contain numbers.
CEILING(number1, significance) - Rounds a number up to the nearest integer or multiple of significance.
CHIDIST(x, degrees_freedom) - Returns the probability of the chi-squared distribution.
COUNT(value1,value2,...) - Counts the number of cells that contains numbers.
COUNTIF(range, criteria) - Counts the number of cells within the range that meet the given criteria.
COUNTIFS(criteria_range1,criteria1,criteria_range2,criteria2,...) - Counts the number of cells within multiple ranges that meet multiple
criteria.
FLOOR(number1,significance) - Rounds a number down toward zero, to the nearest multiple of significance.
FREQUENCY(data_array,bins_array) - Calculates how often a value occurs within a range of values (bins). The function will return a
vertical array of numbers having one more element than the bins_array. Once the formula is entered into the target cell, highlight the target
cell making a vertical array one cell larger than bins_array. Once selected press F2 to open the formula of the target cell and then press
CTRL+SHIFT+ENTER to execute your function.
LARGE(array,k) - Returns the k-th largest value in a data set. For example, the 2nd largest number.
MAX(number1,number2,...) - Returns the largest number in a set of values.
MEDIAN(number1,number2,...) - Returns the median of a set of values.
MIN(number1,number2,...) - Returns the minimum value in a set of values.
MODE(number1,number2,...) - Returns the most frequently occurring value in an array or range of values.
PERCENTILE(array,k) - Returns the k-th percentile of values in a range
ROUND(number1, num_digits) - Rounds a number to the specified number of digits.
SMALL(array,k) - Returns the k-th smallest value in a data set. For example, the 7th smallest number.
STDEV(number1,number2,...) - Returns the standard deviation of a given set of values ignoring text and logical values.
SUM(number1,number2,...) - Adds all of the values in a range of cells.
SUMIF(range,criteria,range,...) - Adds values of cells in a range meeting criteria.
VAR(number1,number2,...) - Estimates the variance of a given set of values.
2.4.1.1.5: NESTING
Though not complex or difficult, placing logical statements within one another can be a powerful practice. This is referred to as nesting.
An example demonstrating the use of a simple IF statement nested within another IF statement is shown below:
If one were to ask three individuals whether or not they agree that energy has a velocity, these individual will hypothetically respond in one
of three ways: "Yes","No", or (other). In this case, if the individual were to respond "Yes", then the questioner would respond "I agree!". If
the individual were to respond "No", the questioner would bitterly respond with a "Bah". And if the individual were to respond with a
nonsensical answer, then the questioner would respond "what .." in confusion.
The important point to get out of this example is that nested statements gives us the potential to process more information in one cell (or
line of programming code).
2.4.1.2: EXCEL'S DATA ANALYSIS TOOLS
Excel has several useful built in add-ins that can make data analysis much easier. Of these we will cover the solver tool, as well as the
ANOVA analysis.
2.4.1.2.1: SOLVER TOOL
The solver in Excel is a built-in tool with multiple capabilities. It could be used to optimize (find the maximum or the minimum value) a
specific cell (the target cell) by varying the values in other cells (the adjustable cells). It may also be used to solve a system of non-linear
equations. Solver formulas embedded in the spreadsheet tie the value in the target cell to the values in the adjustable cells. The user can also
specify numerical range constraints for the cells involved.
Optimization Model
In optimization, one seeks to maximize or minimize the value of a real function. Just as a student in a calculus class might use optimization
techniques to find a local maxima or minima of a complex function, a CEO might use similar techniques to find how much product their
company should manufacture to maximize profits. There are numerous optimization techniques, many specialized for specific types
problems. For the most part, MS Excel uses gradient-based techniques.
The solver tool is located in the Tools menu. If it does not appear there, it must be added in by selecting Add-ins and selecting the
appropriate check box.
The solver window contains the following fields:
Set Target Cell: This is the cell containing the value you want to optimize.
Equal To: Choose whether to maximize the value, minimize the value, or set it to a specific number.
By Changing Cells: Here you specify all the cells that can be varied when optimizing the target cell value.
Subject to the Constraints: Constraints are optional, but can be added by clicking Add and defining each constraint.
List of Available Constraints
<= (Less than or equal to) Stipulates that a selected cell must be less than or equal to a certain value.
= (Equal to) Stipulates that a selected cell must be equal to a certain value.
>= (Greater than or equal to) Stipulates that a selected cell must be greater than or equal to a certain value.
int (Integer) Stipulates that a selected cell must be an integer
bin (Binary) Stipulates that a selected cell must be equal to 1 or 0.
The window where these constraints can be entered is shown below.
It is important to make sure that constraints do not violate each other otherwise solver will not work without reporting an error. For
example, if a cell must be great than 2 and less than 1 Solver will not change anything.
Once all the fields in the Solver window are completed appropriately, click Solve. If solver can find a solution a window will pop up telling
you so, and the values will appear in the target and adjustable cells. If solver cannot find a solution it will still display the values it found but
will state that it could not find a feasible solution. Be aware that Excel solver is not a powerful enough tool to find a solution to nonlinear
equations, where programs such as Matlab might be a better choice.
* Guided Example: Suppose you have an Excel spreadsheet set up as shown below, with cell C6 depending on cell B6 according to the
quadratic relationship shown, and you want to minimize the value in C6 by varying the value in B6.
1) Open up the solver window.
2) Input C6 in the set target cell field, check the min button, and input B6 in the by changing cells field:
3) Click solve to find the solution.

You can place further constraints or further specify how you want your task solved under "Options." Some of the most commonly used
specifications include:
Max Time: This is the maximum amount of time excel will spend trying to find a converging value before giving up.
Iterations: This is the number of iterations excel will perform to converge to a value.
Precision: This is related to how close excel has to get before a value is "acceptable." In general, making this value several orders of
magnitude smaller will yield a more accurate result, but will take more time to do so, and may have a harder time converging.
Tolerance: This is also related to how accurate a solution is, this is related to the percent error. Making this smaller should also yield a
more accurate result, but with more difficulty finding a solution.
Convergence: Controls the amount of relative change that occurs during the last five iterations that Solver performs. The smaller the
number of the convergence the less change that will occur.
Box Options Click on the box to select the following options:
Assume Linear Model: This will speed up the time to find a solution because it will make Solver assume that the model is linear.
Assume Non-Negative: Solver will make the lower limit of adjust zero for all the cells that are programmed to be adjusted in the model
that are not limited by a chosen constraint.
Use Automatic Scaling: Used for large differences in magnitude to scale the inputs and outputs.
Show Iteration Results: Solver will pause between each iteration to show the results of that iteration in the model.
Estimates Specifies to Solver on how to use initial estimates for iterations
Tangent: Uses linear extrapolation by utilizing a tanget vector.
Quadratic: Uses quadratic extrapolation. Best to use for nonlinear models.
Derivatives Specifies what type of differencing is used for partial derivatives for both the objective and constraints.
Forward: Constrained values change slowly between iterations.
2.4.1.3: NONLINEAR REGRESSION

Excel's solver function can also be used to find a solution for two-variable non-linear regression. The description of the data by a function is
carried out by the process of iterative nonlinear regression. This process minimizes the value of the squared sum of the difference between
the actual data and the predicted value (residual).
There are three basic requirements for using solver including:
1. Raw data
2. A model equation to find predicted values
3. Initial guesses for varying values
Before using the solver function, the raw data should be organized into separate columns for the independent variable (first column) and
dependent variable (second column). A third column will show the predicted value that is obtained by using the model equation mentioned
earlier, and must reference various single cells that have initial values that Excel will use as a starting point for the regression. A fourth
column should be created with the square of the difference between the predicted value and the actual experimental value (raw data); this is
also known as the square of the residual. A cell at the bottom of this fourth column should include the sum of all the squared difference
values above it. Two more cells anywhere else in the worksheet should be designated as the initial guesses for the values to be varied to find
the solution.
It is important to choose a reasonable value for the initial guess. This will not increase the chance that Excel arrives at a global optimization
rather than a local optimization, but also reduces the time it takes for Excel to solve the system of non-linear equations.
Now, solver can be applied to the spreadsheet to optimize the regression. Refer to section 3.1 for how to access and use solver. The bottom
most cell of the fourth column that has the sum of the squares of the residuals will be the target cell. Since the residuals are essentially the
error, the sum of the residuals squared should be as small as possible, so the min circle should be selected. The initial guesses should be
entered into the "by changing cells". Constraints can be entered if the values of the cells are limited to a certain range. Solver will produce
the values of the varying parameters that yield the best fit to the nonlinear data.
Guided Example: A screen shot of a simple population problem is shown below. The yellow highlighted cells are the initial guesses for
the variables. In this case, the initial population (P0, cell E2) is entered as a constraint.
2.4.1.3.1: ANOVA ANALYSIS

ANOVA is another tool on Excel that can be used for data analysis purposes. It is used primarily to compare continuous measurements to
determine if they are sampled from the same or different distributions. This is usually done on set of data that consists of numerous samples
and/or groups. Single factor ANOVA analyzes the difference between "groups" which can be different units, sensors, etc. On the other hand
2 way ANOVA analyzes the differences on the basis of both samples (trials) and groups, allowing you to compare, which one most affects
the system. Two-way ANOVA can be employed for two different cases: data with or data without replicates. Data without replicates is used
when collecting a single data point for a specified condition, while data with replicates is used when collecting multiple data points for a
specific condition.
The telling result of ANOVA analysis is a p-value telling you if there is a statistically significant difference between data sets. A more in
depth discussion of the uses and implementation of ANOVA can be seen here.[1]
2.4.1.4: RANDOM NUMBER SAMPLING

A useful numerical technique is the random number generator. This generator uses random number sampling to create a predictive model.
This can be especially useful in probability problems.
A typical random number generator produces any number between 0 and 1 with equal frequency. At each step a random number is
generated that picks a possible outcome. A typical Excel statement would appear as the following: IF (RAND()<X, value1, value2). This
means that X percent of the time, you would get a value1 and 1-X percent of the time, you would get a value2. For example, suppose a
situation exists with two possible outcomes, A and B, where A takes place 70% of the time. If a random number is generated that is greater
than 0.3, then the outcome is A. Inversely if the random number is less than or equal to 0.3, then B is the outcome. This would be
represented as IF(RAND()<0.7, A, B). Each step within the generator does not affect the next step, i.e. each step is determined individually.
A process in which previous outcomes do not affect later outcomes is known as a Stochastic method or "Monte Carlo" method. A Monte
Carlo simulation uses a model that takes random input that follows an assumed distribution (in this case, 70% A and 30% B) and produces
output that models some phenomenon. The following is a simple example of such a simulation:
The Excel is equipped with a random number generator which can be called out using the RAND function. A random number between 0
and 1 can be generated in a cell by the following command:
In Microsoft Excel 2007 the RAND function returns a new random number every time the worksheet is calculated or when F9 is pressed. If
the same random number is needed to be referred to more than once in an Excel IF, AND, OR statement, one can create a column of random
numbers using =RAND(), and refer back to these cells in an adjacent column.
There are many ways to manipulate the random numbers that Excel produces. A random number can be generated within a different range
by multiplying the RAND function by a constant, i.e. multiply by 100 to change a decimal to percent. Another option to generate numbers
within a different range is the RANDBETWEEN function. RANDBETWEEN generates a random integer within a specified range. The
following command generates a random integer between -6 and 21:
It is also possible to modify your random numbers by nesting the RAND function within another function. For example, one can square the
results of one distribution to create a different distribution of numbers.
Another way to utilize the RAND function is to take a random sampling of a data set. If one has a spreadsheet with several hundred data
points, and would like to analyze only 100 random points, another column can be created titled "Random Number" or something similar. In
the first cell of this column, type the =RAND() function. Then drag this function all the way to the end of the data set. At this point, the user
can go to the DATA drag-down menu and select Sort... In this menu, select your random number column in the SORT BY drag down menu.
Either ascending or descending can be selected, as desired. Once the spreadsheet is sorted by the random numbered cells, the first 100 data
points can be used as a statistically random cross-section of the data.
The sort tool can be useful in cases when you are trying to find out the probability of having certain values above a certain threshold. For
example, if the 100 data points that you have represent the temperature variations occurring within a system, you can use sort tool to
determine how many temperature values exceed your threshold. From that, you can calculate approximately the probability of the system
temperature exceeding the threshold. Two screenshots of the sort process are shown below. Microsoft Excel 2007 uses a different menu,
allowing for one to choose between sorting from Largest to Smallest or vice versa. A screenshot below shows this new menu.
Sorting Setup:
Results to Randomization:
Microsoft Excel 2007:
2.4.1.5: WORKED OUT EXAMPLE 1

The Arrhenius equation defines the relationship between a reaction rate constant k and temperature:
−E /RT
k(T ) = Ae
where T is the absolute temperature, A is the frequency factor, E is the activation energy, and R is the universal gas constant.
It is frequently desirable to be able to predict reaction rates at a given temperature, but first you need to know the values of A and E. One
way to obtain these values is to determine the reaction rate constant at a few temperatures and then perform a nonlinear regression to fit the
data to the Arrhenius equation. Given the following rate constant data for the decomposition of benzene diazonium chloride, determine the
frequency factor and activation energy for the reaction.
−1
(s ) (K)
0.00043 313.0
0.00103 319.0
0.00180 323.0
0.00355 328.0
0.00717 333.0
Solution: The following Excel file contains a solution to the problem that uses Excel's solver tool to perform the nonlinear regression and
determine the values of A and E:
Example 1
The spreadsheet is set up as described in the nonlinear regression section above.
1. Open solver (Tools / Solver)
2. Set the Sum of Residuals Squared value cell as the target cell
3. Click the Min radio button
4. Set Pre-exponential Factor and Activation Energy as the adjustable cells. For this problem, keep A between 1E+13 and 1E+14 s^-1 and E
between 9.5E+4 and 1.05E+5 J
5. Sometimes, depending on what version of excel you are using, a message will come up like the one below. If this happens, click OK. The
graph will update with the new curves.
6. Click Solve and observe the values of A and E calculated by Solver and the changes to the plot showing the rate constant as a function of
temperature.
NOTE: If solver is not fitting the data well, click the options button in the solver window and make the precision value smaller by several
orders of magnitude.
2.4.1.6: WORKED OUT EXAMPLE 2
Kinetics can be used to determine the probability of a reaction taking place between gas-phase molecules. Consider molecules A and B,
which have a 60% chance of reacting upon collision. Create an Excel spreadsheet that utilizes a random number generator and an if-
statement to model an individual collision.
Solution
In this Excel file, sample coding for Random number generator and if-statement is shown.
Collision_ex
Random number generator is used to output numbers between 0 and 1 for four trials, ten trials, and one hundred trials on sheets one, two,
and three, respectively. Formula for the random number generator is employed in the cell where you want the number to be output (in this
case, B6 is the cell). An IF logical test is then employed to test whether the number follows the requirement or not. The IF-statement cell
will then output the corresponding result based on the logic test. In this collision case, if the random number function generated a number
greater than 0.4, the logic test will return TRUE and then output "Reacts" as the result. Assuming that the random number generator
produces an even distribution between 0 and 1, the random number will be greater than 0.4 sixty percent of the time. Coding of the IF-
statement is shown:
On the other hand, if the number generated is less than 0.4, the logic test returns FALSE and "Does not React" will be shown as the result.
Different numbers are generated each time the spreadsheet is updated, thus different result will return for the IF function. You can also
modify the numbers and return statement to better understand the operation of the functions.
The number of reactions is summed up and divided by the total number of trials to compare the predicted reaction probability with the value
given in the problem. It can be seen by comparing sheets one, two, and three that increasing the number of trials decreases the variation in
the predicted reaction probability.
2.4.1.7: ACCESSING EXCEL TOOLS ON WINDOWS

On many computers, excel tools such as solver and the data analysis package are not already included and must be installed before used.
This section will go over how to access these add-ins for those using Windows opposed to Macs.
1) Click on the windows symbol in Excel 2007. At the bottom of the opened window, select the Excel options box.
2) On the left side panel you will see a section labeled "Add-ins". Select this.
3) Here you will see a list of "Inactive Applications" that are available for excel. Of these the ones we are interested in are the "Solver Add-
In" and the "Analysis ToolPak." Select one of these and click "Go" at the bottom of the window.
4) A new window will appear asking you which components of the package you would like to install. Select the needed tools (solver and
analysis toolpak). Click "Ok" and windows should install the corresponding tools into excel.
5) Once done, these tools will be found under the data section of Excel, on the far right of the window.
2.4.1.8: REFERENCES
1. =RAND()
=RANDBETWEEEN(-6,21)
Collision Example Excel File:
=IF(B6>=0.4,"Reacts","Does not React")
Bender, E.A. An Introduction to Mathematical Modeling, Mineola, NY: Dover Publications, Inc.
Fogler, H.S. (2006). Elements of Chemical Reaction Engineering, Upper Saddle River, NJ: Prentice Hall Professional Technical
Reference. ISBN 0-13-047394-4
Microsoft, Excel, Proprietary EULA, www.microsoft.com.

Authors: (12 September 2006 / Date Revised: 19 September 2006) Curt Longcore, Ben Van Kuiken, Jeffrey Carey, Angela Yeung
Stewards: (13 September 2007) So Hyun Ahn, Kyle Goszyk, Michael Peterson, Samuel Seo
2.4: Excel Modeling - logical models, optimization with solver for nonlinear regression, sampling random numbers is shared under a CC BY 3.0 license
and was authored, remixed, and/or curated by Peter Woolf et al. via source content that was edited to conform to the style and standards of the LibreTexts
platform; a detailed edit history is available upon request.
2.5: NOISE MODELING - WHITE, PINK, AND BROWN NOISE, POPS AND
CRACKLES
2.5.1: INTRODUCTION
Noise is all around us in all sorts of forms. In the common use of the word, noise refers to sound. However, noise can more accurately be
thought of as random variations that are always present in one or more parts of any entity such as voltage, current, or even data. Rather than
thinking of noise only as an acoustic term, it should be thought of more as a random signal. Noise can be the inherent fluctuations in some
part in a system (ie. temperature at a given point) or it can be the unavoidable interference on a measurement from outside sources (ie.
vibrations from a nearby generator blur measurements from a pressure transducer). The static interference on your radio, the ‘snow’ on your
television, and the unresolved peaks on an infrared spectroscopy report are all examples of noise.
Chemical engineers can use statistical properties to characterize noise so they can understand a current process or develop an optimal
process. By characterizing the noise and determining its source, the engineer can devise methods to account for the noise, control a process,
or predict the path of a system. For example, when chemical engineers design plants, they use characterizations of noise to determine the
best control scheme for each process. Mathematical modeling can be used to characterize and predict the noise of a given system. In
modeling, to simplify the representation of general trends that reoccur, noise is classified in two major categories: frequency based and non-
frequency based. Frequency based noise consists of the colors of noise, and non-frequency based noise includes pops, snaps and crackle.
The purpose of the following sections is to give you a qualitative understanding of the sources of noise and how it can be characterized and
handled. Examples will be given to help you gain a quantitative understanding of how noise relates to controlling and modeling chemical
engineering processes.
2.5.2: GENERAL TYPES OF NOISE

Three general types of noise can be categorized in describing a process, and are as follows:
1) Gaussian noise is usually not dependent on time, meaning that it is random and not systematically planned. The amplitude of the
frequency can vary, making a crackling notation or sound (see #Crackles below). Some examples of Gaussian noise can include splashing in
a tank or unplanned interruption in a sensing device.
2) Drift noise is correlated to time, and has random movement. Examples of drift noise can include fouling or catalyst decay, in which the
potency of the substance may decline over time as decaying occurs. Stock price fluctuations also model somewhat like drift noise.
3) Shot noise may be defined as sporadic and short bursts of noise, in which the amplitude is similar among the bursts of noise. Shot noise
can be correlated to pops (see #Pops below), in which at random times, the same shot noise is witnessed with the same amplitude. Examples
of shot noise include partial clogging or jamming in the process in which the same amplitude will be seen by the noise whenever the
clogging or jamming in the process occurs. Another example is customer demand of the product. If the control system (when trying to
optimize it) depends on the customer order, and customer demand is not consistent at all times (meaning downtimes for orders) but the order
amount is the same when orders are placed, then it will be effected by the shot noise described by the customer order (customer demand).
Microsoft Excel Formulas for 3 Types of Noise

Noise Type Excel Formula
Gaussian Noise = [Original value] + RAND() - RAND() + RAND() - RAND()1
Drift Noise = [Previous value] + RAND() - RAND() + RAND() - RAND()1
Shot Noise = if(RAND()>0.9, [Original value]+[single error], [Original value])
1
The more add/subtract cycles of RAND() you use, the closer approximation you'll get for a Gaussian distribution
of random errors.
As you would see these in an Excel file:
\ A B C D
1 Time Gaussian Noise Drift Noise Shot Noise
2 1 original value original value original value
3 2 = $B$2 + RAND() - RAND() + RAND() - RAND() = C2 + RAND() - RAND() + RAND() - RAND() = if(RAND()>0.9, $D$2+[single error value], $D$2)
4 3 = $B$2 + RAND() - RAND() + RAND() - RAND() = C3 + RAND() - RAND() + RAND() - RAND() = if(RAND()>0.9, $D$2+[single error value], $D$2)
Note: These formulas can be extended for any length of time. Longer periods of time are useful for visualizing drift noise.
2.5.2.1: COMBINED TYPES OF NOISE
Sensors, processes, demands, etc. often do not behave with simple noise models. However, most noise can be derived from the three general
types of noise. An example of a process with random and shot noise is shown in the thumbnail.
This is a graph of the total daily electricity usage for a plant during a normal operation period. As we can see, there are minor fluctuations
occurring every day, but some days there are large changes of electricity usage. These large changes can be seen as shot noise when the
consumption is viewed on this daily scale.
Possible reasons for the minor fluctuations could be due to electric heaters kicking on and off. Changes in the operation of smaller pumps
can cause these changes in electricity demand.
The large shots could be due to an energy-intensive part of the process that only operates when needed. The positive shots could be due to
this process kicking on and the negative shots could be due to a large process kicking off when not needed.
With this graph, we can say that there is shot noise present with the large changes in consumption, but there is also random noise present
with the minor daily fluctuations. We would not say there is drift noise present because the consumption remains around a constant mean
value and does not drift away from there.
However, the total consumption when viewed from a different scale may show other forms of noise. If we viewed the hourly usage of
electricity on one of the days with shot noise, it may resemble drift noise on the hourly scale.
2.5.3: COLORS OF NOISE

The first category of noise is frequency-based noise, classified by the colors of noise. In understanding the colors of noise, we need to
understand the process of converting a given set of data to a point where it can be classified as some color of noise. The section below,
“Modeling Noise,” goes into more detail as to how this is done. At this point, it is important to understand the main colors of noise which
are possible in a system. Noise is classified by the spectral density, which is proportional to the reciprocal of frequency (f ) raised to the
power of beta. The power spectral density (watts per hertz) illustrates how the power (watts) or intensity (watts per square meter) of a signal
varies with frequency (hertz).
1
P SD ∝
β
f
where β ≥ 0.
Just like colors of light are discriminated by their frequencies, there are different colors of noise. Each beta value corresponds to a different
color of noise. For example, when beta equals zero, noise is described as white and when beta equals two, noise is described as brown.
2.5.3.1: WHITE NOISE

One major type of noise is white noise. A defining characteristic of this type of noise is it has a flat power spectral density, meaning it has
equal power at any frequency. β equals zero for white noise.
White noise and white light have similar properties. White light consists of all visible colors in the spectrum while white noise is created by
combining sounds at all different frequencies. Pure white noise across all frequencies cannot physically exist. It requires an infinite amount
of energy and all known energy is finite. White noise can only be created within a specific and defined range of frequencies. Again making
an analogy to white light, for a small band of frequencies, visible white light has a flat frequency spectrum. Another property of white noise
is that it is independent of time. This factor also contributes to the idea that white noise cannot physically exist since it is impossible to have
noise be completely independent of time.
A sample power spectral density chart of white noise is shown below. As can be seen from the chart, the power over all displayed
frequencies is essentially the same, suggesting the PSD is equal to 1.
image from http://en.Wikipedia.org/wiki/White_noise on 9/18/2006

It is important to note that white noise is not always Gaussian noise. Gaussian noise means the probability density function of the noise has
a Gaussian distribution, which basically defines the probability of the signal having a certain value. Whereas white noise simply means that
the signal power is distributed equally over time. For Gaussian white noise one finds the number of measurements greater than the power
spectral line as measurements that are less than the power spectral line with majority falling at the spectral line. Gaussian white noise is
often used as a model for background noise in satellite communication. White noise can also come from other distributions, such at the
Poisson distribution. Poissonian white noise will look like a normal distribution that has been shifted to the left for small number of
measurements. For larger number of measurements Poissonian white noise will look like the normal distribution (also known as gaussian
white noise). A Poissonian white noise model is useful for systems where things happen in discrete amounts such as photons being
transmitted or the number of cars that go through an intersection.
2.5.3.2: PINK NOISE

Pink noise is a signal whose power spectral density decreases proportionally to the inverse of the frequency, where the β value is equal to
one. Because of its decrease in power at lower frequencies when compared to white noise, pink noise often sounds softer and damper than
white noise.
Pink noise is actually what is commonly considered audible white noise and is evident in electrical circuits. Pink noise becomes important
to measure in things such as carbon composition resistors, where it is found as the excess noise above the produced thermal noise. Below is
a power spectral density chart for Pink noise.
image from en.Wikipedia.org/wiki/Pink_noise on 9/18/2006
2.5.3.3: BROWN NOISE

When β equals 2, the noise is Brownian. Brown noise is signal noise created by Brownian, or random, motion. It is a form of unavoidable
interference in the transmission of information and can be compared to a random walk that does not have a clearly patterned path. Brown
noise has more energy at lower frequencies. Compared to white or pink noise, brown noise is even more soft and damped. It can be
generated by integrating white noise. Interference from outside sources, such as vibrations from nearby machinery or background light, in
instrument readings usually have a brown noise pattern.
Below is a power spectral density chart for brown noise. From the charts of brown noise and pink noise, it can be observed that brown noise
loses power as frequency increases at a much faster rate than that of pink noise.
image from http://en.Wikipedia.org/wiki/Brown_noise on 9/18/2006
2.5.3.4: BLUE NOISE

Blue noise is a signal whose power spectral density increases proportionally to the frequency, where the β value is equal to negative one. An
example of would be a reversible, exothermic, batch reaction where the products are being accumulated at an overall constant rate, but the
reverse reaction occurs when the temperature gets too high. Thus it is overall increasing in product, but has small fluctuations as it
increases. Below is a power spectral density chart for blue noise.
image from en.Wikipedia.org/wiki/Blue_noise#Blue_noise on 12/11/2008
2.5.3.5: PURPLE NOISE

When β equals -2, the noise is violet. Purple noise has more energy at higher frequencies. It can be generated by differentiating white noise.
Below is a power spectral density chart for purple noise. From the charts of purple noise and blue noise, it can be observed that purple noise
gains power as frequency increases at a much faster rate than that of blue noise.
2.5.3.6: MODELING COLORS OF NOISE

Characterizing noisy signals is important to a chemical engineer so that he or she can determine the sources of noise and how to account for
it. Sometimes, the noise will be a characteristic of the system. In these cases, the system can not have a very sensitive control because if the
sensor for the controlling device is too sensitive, it will respond to the noise of your system and end up controlling your system in a noisy
way. Other times, the noise will be from unavoidable disturbances to the system. In these cases, the noise can be damped or accounted for in
an appropriate manner.
In order to apply the knowledge of the colors of noise we need to understand the process of converting a given set of data to a point where it
can be classified as some color of noise. The general process of classifying noise follows.
Data Curve to Fit Data Fourier Transform Power Spectral Density Classification of Noise Color
The classification of noise color becomes important when we need to have an estimate of future data. In other words, given the trends in our
current data we can use it to estimate future data. This is done as follows,
Noise Color Power Spectral Density Fourier Transform Curve Generate Data From Curve
The power of a noise signal is detected at a certain frequency. Then a plot of the log(power) vs. the log(frequency) can be constructed, and
the slope of the line gives the beta value. Following a backward thought process, one can produce a certain color of noise by creating
frequency components which have a value generated by a Gaussian distribution and then scaling by the appropriate beta power of
frequency.
A general method to characterize and model noise is explained below.
1. DATA
Data is what the signal transmits. The signal is dependent on what you are measuring or modeling. The data can be collected, for example, if
you're measuring temperature in a reactor, then your data is the temperature readings from a thermocouple at a certain position in the reactor
over a period of time.
2. CURVES
After you collect the data, you plot the data and find a best fit equation, , for that set of data. A math program can be used to find a best
fit equation. Microsoft Excel can be used for a simple model and programs such as Polymath or Mathematica can be used for more complex
models. A simple equation could be in the form of . The coefficients A and B can be varied to fit the data. Additional terms can
be added for more accurate modeling of complex data. The equation you find can then be used to predict and model future signal noise.
3. FOURIER TRANSFORMS
A Fourier transform is a tool used to convert your data to a function of . In this form, the noise can be more easily characterized. Apply a
Fourier transform to the curve, , you fit to your data to generate a relation for the power spectral density. The Fourier transform can be
performed by a computer depending on the complexity of (or see "Simplifying the Fourier Transform" below). The transform is the
integral shown below.
∞
−jwt
X(w) = ∫ x(t)e dt
−∞
where;
is the equation of the curve to fit the data.
is the exponential form of writing the relation . The j in the second term here indicates that that term is
imaginary.
is the frequency
4. POWER SPECTRAL DENSITY
This value is attained by simplifying and squaring the Fourier Transform. Since the series of simplifications needed are fairly complex, the
resulting power spectral density equation is displayed below.
−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−
2 2
P S D = √ (∑ x(t) cos(wt)) + (∑ x(t) sin(wt))
At this point we attain a numerical value for the PSD at the particular frequency, . These numerical PSD values can be plotted versus
frequency to obtain the PSD chart.
5. CLASSIFICATION OF NOISE
The summation is repeated over and over for different frequencies, . A plot of the PSD vs. frequency, is made with these values. Once
this is done a best fit line is applied to the data points which will give you the characterization relation . Based on this we can
then classify the noise as a color of noise.
image from en.Wikipedia.org/wiki/Violet_noise#Violet_noise on 12/11/2008
THE REVERSE PROCESS
Knowing how to convert data to a color of noise is only half the problem. What if we know what type of noise is possible and we need data
from it for a given process? Knowing the noise color means that we know the power spectral density relation to the frequency. From here
onwards we follow the reverse route as that taken to get to the noise color by using the inverse Fourier transform instead.
Inverse Fourier Transform
∞
1
jwt
x(t) = ∫ X(w)e dt
2π −∞
where;
is the equation of the curve to produce future the data.
is the exponential form of writing the relation . The j in the second term here indicates that that term is
imaginary.
is the frequency
We can either use it to produce a curve or, by making a similar simplification as we did for the Fourier transform, to generate data directly.
This reverse process should be trivial to someone who worked through the forward process.
2.5.4: POPS, SNAPS & CRACKLES

The second category of noise is non-frequency based noise. Three examples of this non-frequency noise are pops, snaps and crackles. The
study of these noise types is fairly new, and not a lot is known about how to deal with them when they arise in our research or studies.
However, one can successfully distinguish and classify these noise types by noticing certain properties which are typical of them.
2.5.4.1: POPS
At one end of the extreme of non-frequency noise is what is defined as pops. Pops are infrequent random spikes in noise of approximately
the same amplitude. An example of this would be a loose wire that is usually completing the circuit. However, it randomly disconnects for a
split second and then recompletes the circuit for normal operation. Chemical engineering processes may inherently have pops that are
unpredictable. Safety features can be added to a process, such as pressure relief valves, to handle the pops when they do occur. The
following is an image of what pops might look like in a system:
2.5.4.2: SNAPS
On the other end of the non-frequency noise spectrum there are snaps. Snaps are single independent events that occur only once. When a
pencil is bent and it breaks under the force is an example of snapping. Another example would be the snapping of a piece of chalk or
exploding a bomb. The “popping” of one’s knuckles is really snapping since it is an independent event, unless you snap all your knuckles all
at once or one after the other really fast. Just like pops, snaps are unpredictable, and safety features should be added to the system to handle
the snaps. The following is an example of a snap:
2.5.4.3: CRACKLES
In between popping and snapping there is crackling. A very common example for crackling is the sound heard coming from a burning piece
of wood. Like popping, there is a non-frequency or irregularity in which the crackles occur. In addition, there is also an irregularity of the
amplitude of the crackle. In the case of the fire, not only can you not predict when the crackle sound would be heard, but you cannot predict
how loud it will be either. Furthermore, there is a universality condition associated with crackling. Regardless of the scale, the similar
randomness in repetition and amplitude should be observed.
In dealing with this universality condition the concept of a critical exponent arises. For example, if we are looking at the same crackling
effect, S , over a larger period of time the two would have to be equal after scaling the larger one.
T
⟨S ⟩small (T ) = A⟨S ⟩large ( )
B
for some scaling factors and .

where ⟨S ⟩ is S average.
If the time scale is expanded by a small factor B = ( 1
(1−δ)
) then the rescaling of the size will also be small, say 1 + aδ .
⟨S ⟩(T ) = (1 + aδ) ∗ ⟨S ⟩(1 + δT )
For small δ we have

d⟨S ⟩
a⟨S ⟩ = T
dT
Solving gives
a
⟨S ⟩(T ) = So T (2.5.1)
The exponent a is called the critical exponent and is a universal prediction of a given theory. There are other concepts which are used as a
check for universality for times when the critical exponent cannot be used but the most common one is the critical exponent. The following
is an example of what a crackle might look like in a system:
 EXAMPLE 2.5.1: CRACKLE
A lead engineer monitoring the instantaneous flow rate of a coolant used to cool an exothermic reactor gathers data set ‘a’ over a 30hr
period. Later on that week he gathers data set ‘b’ for the same coolant flow rate. For both sets of data he determined that the way the
instantaneous coolant flow rate is reacting to the exothermic reactor is optimal to the reaction process. For his records he would like to
represent the noise in the data in a compact form. He wants you to first characterize the noise and then provide any other information,
which will generalize the trend in the data. Once you have completed these tasks, report back to him.
Data set 'a'

Note that between 0 and 30hr for all intervals of time not represented in this table there is a zero assigned as the instantaneous flow rate.
Data set 'b'

Note that between 0 and 50hr for all intervals of time not represented in this table there is a zero assigned as the instantaneous flow rate.
Solution
Each set of data is plotted, Flow Rate vs. Time. The two graphs show crackling trends.
Graph for data set 'a'
Graph for data set 'b'

Upon observing this, the general noise trend is crackling. To provide a final proof we apply the universality relation (Equation 2.5.1 ).
We first determine the critical exponent, a , for data set a.
We see that;
< S >= 1.54
S0 = 2
this value can be the first value attained or an average of the first 2 or 3 values.
T = 30
Solving for the critical exponent using the relation given above we get;
a = −0.08
Once again we carry out the same calculation for data set 'b'.
< S >= 2.12
S0 = 3
this value can be the first value attained or an average of the first 2 or 3 values
T = 50
Solving for the critical exponent using the relation given above we get;
a = −0.08
The similarity of the two critical exponent gives further proof that the data the instantaneous flow rate of the coolant over a period of
time will show crackling.
 EXAMPLE 2.5.2: COLORS OF NOISE
A chemical engineer is reading flow rates from flow meter. Every 0.1 day for 8 days a reading was taken and the data is given here:
Colors of Noise Example. The data displays the fluctuations from the set flow rate of 3000 liters per hour at a wastewater treatment
plant. The specifications for the plant say that the max flow rate is 6000 liters per hour, or the pipes will burst. Also, the flow rate
cannot fall below 200 liters per hour, or the system will automatically shut down.
The chemical engineer notices that there were some readings close to these limits. It is the chemical engineer’s job to determine if the
readings are accurate flow rates or if there is an error with the flow meter. By characterizing the type of noise, the chemical engineer
can determine the source of the noise and so take appropriate preventative measures. What type of noise is present and what protective
measures should be taken?
Solution
1) Plot data. From the data presented, a flow rate vs. time chart was graphed to gauge degree of fluctuation in the flow rate.
2) Calculate the power spectral density data. Using the simplified integral derived to calculate the power spectral density, a table was
created with the resulting PSD values at varying frequencies. These frequencies were self defined and were chosen to encompass a
broad range. More detailed calculations can be found here: Colors of Noise Example under the worksheet titled “PSD CALC”.
3) Plot the power spectral density. The power spectral density for each frequency was then plotted against the frequency, creating the
power spectral density plot shown below.
4) Characterize the noise. To determine the β value for the data, a linear trend line was taken on the data. This trend line can be seen
in the power spectral density plot above. The slope of this trend line is the β value for the data. In this case, the β=0.023. Since this
value is not that of white noise (β=0) nor that of pink noise (β=1), we can say that this noise is somewhere between white and pink
noise.
5) Determine the source of noise.
There are two possible major sources of noise. They are the liquid motion in the pipe and noise in the flow meter caused by itself or
outside sources. Since it is found earlier that β=0.023, the source of noise is probably from the flow meter. It is not from the motion of
liquid in the pipe because liquid motion tends to produce brown noise (β=2).
6) Protective measures that can be taken. Knowing this, a correction can be made on the calculations to determine the actual flow
rate. A full step by step solution can also be found here: Colors of Noise Example under the worksheet titled “Solution”.
2.5.5: SUMMARY
Noise is all around us and is an unavoidable characteristic of monitoring signals. When noise is frequency based, like the colors of noise, it
can be characterized and modeled using mathematical tools. When noise is not frequency based, like pops, snaps, and crackles, it can be
recognized, but ways to model it are still being devised. Characterizing and modeling noise is important for chemical engineers to
accurately handle their data, control processes, and predict future trends.
2.5.6: REFERENCES
Kosko, Bart. “White Noise Ain’t So White”, Noise. ISBN 0670034959
Ziemer, Rodger E. Elements of Engineering Probability and Statistics, New Jersey: Prentice Hall. ISBN 0024316202
Papoulis, Athanasios. Probability, Random Variables, and Stochastic Processes, New York: McGraw – Hill Book Company. ISBN
0071199810
Peebles, Peyton Z. Jr. Probability, Random Variables, and Random Signal Principles, New York: McGraw – Hill, Inc. ISBN 0071181814
Sethna, James P. Crackling Noise, Nature 410, 242-250 (2001). Also available here.
en.Wikipedia.org

Authors: Danesh Deonarain, Georgina Mang, Teresa Misiti, Carolyn Ehrenberger
Date Presented: September 08, 2006 Date Revised: September 19, 2006
2.5: Noise Modeling - White, Pink, and Brown Noise, Pops and Crackles is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by
upon request.
2.6: NUMERICAL ODE SOLVING IN EXCEL- EULER’S METHOD, RUNGE KUTTA,
DEAD TIME IN ODE SOLVING
Authors: (Presented: 9/8/06 /Date Revised: 9/19/06) Aaron Bennick, Bradley Anderson, Michael Salciccioli
Stewards: (9/5/07) Sean Gant, Jay Lee, Lance Dehne, Kelly Martin
2.6.1: INTRODUCTION
This article focuses on the modeling of ordinary differential equations (ODEs) of the form:
dy
= f(x, y)
dx
In creating a model, a new value y i+1is generated using the old or initial value yi, the slope estimate φ, and the step size h. This general
formula can be applied in a stepwise fashion to model the solution. All stepwise models will take the following general form:
yi+1 = yi + ϕh
The modeling methods discussed in this article are Euler’s method and the Runge-Kutta methods. The difference between the two methods
is the way in which the slope φ is estimated.
2.6.1.1: CHOOSING THE RIGHT MODEL AND STEP SIZE

The proper numerical modeling method heavily depends on the situation, the available resources, and the desired accuracy of the result. If
only a quick estimate of a differential equation is required, the Euler method may provide the simplest solution. If much higher accuracy is
required, a fifth-order Runge-Kutta method may be used. Engineers today, with the aid of computers and excel, should be capable of
quickly and accurately estimating the solution to ODEs using higher-order Runge-Kutta methods.
In all numerical models, as the step size is decreased, the accuracy of the model is increased. The tradeoff here is that smaller step sizes
require more computation and therefore increase the amount of time to obtain a solution. A balance between desired accuracy and time
required for producing an answer can be achieved by selecting an appropriate step size. One suggested algorithm for selecting a suitable
step size is to produce models using two different methods (possibly a second and third order Runge-Kutta). The steps size can then be
systematically cut in half until the difference between both models is acceptably small (effectively creating an error tolerance).
2.6.2: EULER'S METHOD

Euler's method is a simple one-step method used for solving ODEs. In Euler’s method, the slope, φ, is estimated in the most basic manner
by using the first derivative at xi. This gives a direct estimate, and Euler’s method takes the form of
yi+1 = yi + f (xi , yi ) h
dy
For demonstration, we will use the basic differential equation dx
= 3x
2
+ 2x + 1 with the initial condition y(0) = 1. If a step size, h, is
taken to be 0.5 over the interval 0 to 2, the solutions can be calculated as follows:
x y_Euler y_Actual Error
y(0) 1 1 0
y(0.5) 1 + [3(0)^2 + 2(0) + 1](0.5) = 1.5 1.875 0.375
y(1) 1.5 + [3(0.5)^2 + 2(0.5) + 1](0.5) = 2.875 4 1.125
y(1.5) 2.875 + [3(1)^2 + 2(1) + 1](0.5) = 5.875 8.125 3.25
y(2.0) 5.875 + [3(1.5)^2 + 2(1.5) + 1](0.5) = 11.25 15 3.75
The y_actual values in this table were calculated by directly integrating the differential equation, giving the exact solution as:
2.6.2.1: USE OF MICROSOFT EXCEL

Calculating an ODE solution by hand with Euler's method can be a very tedious process. Fortunately, this process is greatly simplified
through the use of Microsoft Excel. Creating a spreadsheet similar to the one above, where the x values are specified and the y_Euler values
are recursively calculated from the previous value, makes the calculation rather simple. Any step size and interval can be used.
2.6.2.2: DRAWBACKS OF EULER'S METHOD

There are several drawbacks to using Euler’s method for solving ODEs that must be kept in mind. First of all, this method does not work
well on stiff ODEs. A stiff ODE is a differential equation whose solutions are numerically unstable when solved with certain numerical
methods. For stiff equations - which are frequently encountered in modeling chemical kinetics - explicit methods like Euler's are usually
quite inefficient because the region of stability is so small that the step size must be extremely small to get any accuracy. In a case like this,
an implicit method, such as the backwards Euler method, yields a more accurate solution. These implicit methods require more work per
step, but the stability region is larger. This allows for a larger step size, making the overall process more efficient than an explicit method. A
second drawback to using Euler's Method is that error is introduced into the solution. The error associated with the simple example above is
shown in the last column. This error can be seen visually in the graph below.
It can be seen that the two values are identical at the initial condition of y(0)=1, and then the error increases as the x value increases and the
error propagates through the solution to x = 2. The error can be decreased by choosing a smaller step size, which can be done quite easily in
Excel, or by opting to solve the ODE with the more accurate Runge-Kutta method.
2.6.3: RUNGE-KUTTA METHODS

The Runge-Kutta method for modeling differential equations builds upon the Euler method to achieve a greater accuracy. Multiple
derivative estimates are made and, depending on the specific form of the model, are combined in a weighted average over the step interval.
The order of the Runge-Kutta method can range from second to higher, depending on the amount of derivative estimates made. The second-
order Runge-Kutta method labeled Heun's technique estimates derivatives by averaging endpoint measurements of the step size along a
function. This averaged value is used as the slope estimate for xi + 1. Third and higher power Runge-Kutta methods make mid-point
derivative estimations, and deliver a weighted average for the end point derivative at xi + 1. As the Runge-Kutta order increases, so does the
accuracy of the model.
The general form of the Runge-Kutta method is
yi+1 = yi + ϕ (xi , yi , h) h
Where φ(x i, yi , h) now represents a weighted average slope over the interval h .
ϕ (xi , yi , h) = a1 k1 + a2 k2 + … + an Kn
where the a's are constants and the k's are
The constants a, p, and q are solved for with the use of Taylor series expansions once n is specified (see bottom of page for derivation). The
resulting set of equations have one or more degrees of freedom. This means that for every order of Runge-Kutta method, there is a family of
methods. Below are some of the more common Runge-Kutta choices.
2.6.3.1: SECOND-ORDER RUNGE-KUTTA METHODS (N = 2)
Every second order method described here will produce exactly the same result if the modeled differential equation is constant, linear, or
quadratic. Because this is typically not the case, and the differential equation is often more complicated, one method may be more suitable
than another.
2.6.3.1.1: HEUN'S TECHNIQUE
The second-order Runge-Kutta method with one iteration of the slope estimate , also known as Heun's technique, sets the constants
1
a1 = a2 =
2
and
p1 = q11 = 1.
Huen determined that defining a and a as 1/2 will take the average of the slopes of the tangent lines at either end of the desired interval,
1 2
accounting for the concavity of the function, creating a more accurate result. When substituted into the general form, we find
1 1
yi+1 = yi + ( k1 + k2 ) h
2 2
with k1 and k2 defined as

k1 = f (xi , yi )
k2 = f (xi + h, yi + hk1 )
dy
For demonstration of this second-order Runge-Kutta method, we will use the same basic differential equation dx
2
= 3x + 2x + 1 with the
initial condition y(0) = 1. If a step size, h, is taken to be 0.5 over the interval 0 to 2, the solutions can be calculated as follows:
Figure 1: second-order Runge-Kutta
x k_1 k_2 y_Heun y_Actual Error
y(0) 3(0)^2 + 2(0) + 1 = 1 3(0.5)^2 + 2(0.5) + 1 = 2.75 1 1 0
y(0.5) 3(0.5)^2 + 2(0.5) + 1 = 2.75 3(1)^2 + 2(1) + 1 = 6 1 + [0.5(1) + 0.5(2.75)](0.5) = 1.9375 1.875 -0.0625
y(1) 3(1)^2 + 2(1) + 1 = 6 3(1.5)^2 + 2(1.5) + 1 = 10.75 2.375 + [0.5(2.75) + 0.5(6)](0.5) = 4.125 4 -0.125
y(1.5) 3(1.5)^2 + 2(1.5) + 1 = 10.75 3(2)^2 + 2(2) + 1 = 17 5.375 + [0.5(6) + 0.5(10.75)](0.5) = 8.3125 8.125 -0.1875
y(2.0) 3(2)^2 + 2(2) + 1 = 17 3(2.5)^2 + 2(2.5) + 1 = 24.75 10.75 + [0.5(10.75) + 0.5(17)](0.5) = 15.25 15 -0.25
When compared to the Euler method demonstration above, it can be seen that the second-order Runge-Kutta Heun's Technique requires a
significant increase in effort in order to produce values, but also produces a significant reduction in error. Following Runge-Kutta methods
can be worked through a similar manner, adding columns for additional k values. Below is a graphical description of how slope is estimated
using both Euler's method, and Heun's technique. Observe the increase in accuracy when an average slope across an interval of 0.5 is used
instead of just an initial estimate.
2.6.3.1.2: IMPROVED POLYGON METHOD
Using the improved polygon method, a2 is taken to be 1, a1 as 0, and therefore . The general form then becomes
Ralston's Method
The Ralston method takes a2 to be . Therefore and . It has been determined by Ralston (1962) and Ralston and
Rabinowitz (1978) that defining a2 as will minimize truncation error in second-order Runge-Kutta methods. The general form becomes
2.6.3.2: THIRD-ORDER RUNGE-KUTTA METHODS (N = 3)

The third-order Runge-Kutta methods, when derived, produce a family of equations to solve for constants with two degrees of freedom.
This means an even more variable family of third-order Runge-Kutta methods can be produced. A commonly used general third-order form
is
with
2.6.3.3: FOURTH-ORDER RUNGE-KUTTA METHODS (N = 4)

The family of fourth-order Runge-Kutta methods have three degrees of freedmon and therefore infinite variability just as the second and
third order methods do. The fourth-order versions are most favored among all the Runge-Kutta methods. What is know as the classical
fourth-order Runge-Kutta method is
with
2.6.3.4: FIFTH-ORDER RUNGE-KUTTA METHODS (N = 5)

Where very accurate results are required, the fifth-order Runge-Kutta Butcher's (1964) fifth-order RK method should be employed:
with
The integration of k's within other k values suggests the use of a spreadsheet. As with all Runge-Kutta methods, the calculation of values for
the fifth-order version would be greatly assisted through the use of Microsoft Excel.
2.6.4: INTERACTIVE MODEL COMPARISON IN EXCEL

To give a better understanding of the impact between different Runge-Kutta methods, as well as the impact of step size, the interactive excel
sheet below will allow you to enter the step size h into a set of models and observe how the models contour to match an example differential
equation solution.
Observe the relationship between model type, step size, and relative error.
ODE model comparison interactive spreadsheet
2.6.5: DEAD TIME

Dead time, or delay differential equation, occurs when there is a delay or lag in the process of a real life function that is being modeled. This
means that the numerical model is not accurate until the delay is over. This concept can come into play for the start up of a reaction process.
For example, at the start of a reaction in a CSTR (Continuous Stirred Tank Reactor), there will be reagents at the top of the reactor that have
started the reaction, but it will take a given time for these reactant/products to be discharged from the reactor. If one was modeling the
concentration of reagents vs. time, time t=0 would have started when the tank was filled, but the concentrations being read would not follow
a standard model equation until the residence time was completed and the reactor was in continuous operational mode. The dead time is the
time it would take for the readings to start meeting a theoretical equation, or the time it takes for the reactor to be cleared once of the
original reagents. Dead time can be determined experimentally and then inserted into modeling equations.
The solution of such a delay differential equation becomes problematic because in order to solve the equation, information from past times
(during the delay) is needed in addition to the current time. For example if to is the lag time for the given scenario, then the value to use in
the ODE becomes (t-to) instead of t. When modeling this in excel, (x-to) is substituted in for the x value. With this substitution, Euler’s
method can be used again in the same way to approximate the solution. This substitution will be carried into the differential equation so that
the new dead time solution can be approximated using Euler's method. To model the reactor before the initial deadtime is completed, piece-
wise functions are often used. This way for the dead time, a given model is used not characteristic of the reactor at normal operating
conditions, then once the dead time is completed the modeling equation is taken into effect.
The link below will help to show how to include dead time in a numerical method approximation such as Euler's method. As seen in the
excel file, the dead time that is specified by the user in the yellow box will change the delay in the model. The more dead time, the further
shifted from the theoretical equation the new model is. To take dead time into account in excel, the x value is simply substituted out for (x-t)
where t is equal to the dead time. If you look at the equations entered in the Y cells, you will see that the x value inserted into the differential
equation is (x-t), where t is the user specified dead time. It can be seen through this example spreadsheet that the effect of dead time is a
simple horizontal shift in the model equation.
Dead Time Interactive Spreadsheet
2.6.6: ERROR
There are two types of error associated with solving ODEs using stepwise approximation methods in Excel. These errors are also present
using other methods or computer programs to solve ODEs. The first, discretization, is the result of the estimated y value that is inherent of
using a numerical method to approximate a solution. Discretization errors, also called truncation, occur proportionately over a single step
size. Truncation error will propagate over extended results because the approximation from previous steps is used to approximate the next.
In essence, a copy of a copy is being made. The accuracy of the next point is a direct result of the accuracy of the previous. Just as the
quality of a second copy is dependant on the quality of the first. This error can be reduced by reducing the step size.
Please see ODE model comparison interactive spreadsheet to better learn how step sizes can influence error.
The second types of errors are rounding errors. These errors are dependent on the computer’s capacity for retaining significant digits. The
more significant digits that a computer can hold, the smaller the rounding error will be.
2.6.6.0.1: ESTIMATING ERROR IN EULER'S METHOD
To mathematically represent the error associated with Euler's method, it is first helpful to make a comparison to an infinite Taylor series
expansion of the term yi + 1. The Taylor series expansion of this term is
2 3 n
′
h ′′
h n
h
yi+1 = yi + f (xi , yi ) ℏ + f (xi , yi ) +f (xi , yi ) +…+f (xi , yi )
2 3 !n
When this expansion is compared to the general form of Euler's method it can be seen that Euler's method lacks every term beyond
. These missing terms, the difference between the Euler approximation and an infinite Taylor series (taken to be the true
solution), is the error in the Euler approximation. Mathematically respresenting the error in higher order Runge-kutta methods is done in a
similar fashion.
2.6.6.0.1: ESTIMATING AND MINIMIZING ERROR IN RUNGE KUTTA METHOD
Uniform time steps are good for some cases but not always. Sometimes we deal with problems where varying time steps makes sense.
When should you change the step size? If we have an nth order scheme and and (n+1)th order scheme, we can take the difference between
these two to be the error in the scheme, and make the step size smaller if we prefer a smaller error, or larger if we can tolerate a larger error.
This is fairly simple with Runge Kutta, because we can take a fifth order method and a fourth order method using the same k's. Only a little
extra work at each step.
Another way of estimating error in the Runge-Kutta method is to reverse directions at each step of the advancing solution and recompute the
previous ordinate. By considering the difference between the newly computed previous ordinate and the originally computed value, you can
determine an estimate for the truncation error incurred in advancing the solution over that step. This is a bit more tedious, but does give a
good estimate of truncation error.
2.6.7: EULER'S METHOD FOR SYSTEMS OF ODES

In chemical engineering and other related fields, having a method for solving a differential equation is simply not enough. Many real world
problems require simultaneously solving systems of ODEs. For problems like these, any of the numerical methods described in this article
will still work. The only difference is that for n ODEs, n initial values of y are needed for the initial x value. Then the numerical method of
choice is just applied to every equation in each step before going on to the next. This further complicates the step-by-step problem solving
methodology, and would require the use of Excel in nearly every application. The ODE solved with Euler's method as an example before is
now expanded to include a system of two ODEs below:
dy1 2
= 3x + 2x + 1, y1 (0) = 1
dx
dy2
= 4y1 + x, y2 (0) = 2
dx
Step size is again 0.5, over an interval 0-2.

x y_1Euler y_2Euler y_1Actual y_2Actual y_1 Error y_2 Error
y(0) 1 2 1 2 0 0
y(0.5) 1 + [3(0.5)^2 + 2(0.5) + 1](0.5) = 2.375 2 + [4(1) + 0.5](0.5) = 4.25 1.875 5.875 -0.5 1.625
y(1) 2.375 + [3(1)^2 + 2(1) + 1](0.5) = 5.375 4.25 + [4(2.375) + 1](0.5) = 9.5 4 18.5 -1.375 9
y(1.5) 5.375 + [3(1.5)^2 + 2(1.5) + 1](0.5) = 10.75 9.5 + [4(5.375) + 1.5](0.5) = 21 8.125 51.875 -2.625 30.875
y(2.0) 10.75 + [3(2)^2 + 2(2) + 1](0.5) = 19.25 21 + [4(10.75) + 2](0.5) = 43.5 15 124 -4.25 80.5
It should be observed that there is more error in the Euler approximation of the second ODE solution. This is because the equation also has
y1 in it. So there is the error introduced by using the Euler approximation to solve the 2nd ODE, as well as the error from the Euler
approximation used to find y1 in the 1st ODE in the same step!
Exact solutions were again obtained from directly integrating the ODEs: and
2.6.7.1: EXAMPLE 1: REACTOR DESIGN

The elementary liquid-phase reaction A --> B is to be carried out in an isothermal, isobaric PFR at 30 degrees C. The feed enters at a
concentration of 0.25 mol/L and at a rate of 3 mol/min. The reaction constant is known experimentally to be 0.01 min-1 at this temperature.
You have been given the task of building a reactor that will be used to carry out this reaction. Using Euler’s method with a step size of 0.05,
determine how large the reactor must be if a conversion of 80% is desired.
Simplified Design Equation:
dV FA0
=
dX k CA0 (1 − X)
Lumping the given flow, concentration, and reaction constant together gives:
dV 1
= 1200
dX 1− X
Since no volume is required for a conversion of zero, the initial condition needed is V(0)=0.
Now the information simply has to be entered into Excel. A column for the conversion, X, going from 0 to 0.8 in 0.05 increments is used for
the step size, and the first value, V(0), is known to be zero. To get the volume, simply add the previous volume to the constants multiplied
by the step size and 1/(1-X), or:
1
Vi+1 = Vi + 1200 ∗ 0.05 ∗
1− X
Copying this formula down the column to the final conversion value of 0.8, gives the results shown in the table below:
X V
V(0) 0
V(0.05) 63
V(0.1) 130
V(0.15) 200
V(0.2) 275
V(0.25) 355
V(0.3) 441
V(0.35) 533
V(0.4) 633
V(0.45) 743
V(0.5) 863
V(0.55) 996
V(0.6) 1146
V(0.65) 1317
V(0.7) 1517
V(0.75) 1757
V(0.8) 2057
The final reactor volume obtained is approximately 2057 L. This compares reasonably well to the exact value of 1931 L. The Excel file
used to obtain this solution, along with the exact solution, can be downloaded below.
Example 1
2.6.8: SECOND-ORDER RUNGE-KUTTA DERIVATION

The following example will take you step by step through the derivation of the second-order Runge-Kutta methods. Setting n = 2 results in a
general form of
The constants in the general form must be defined. To do this we will employ a second-order Taylor series expansion for yi + 1 in terms of yi
and . This Taylor series is
2
h
′
yi+1 = yi + f (xi , yi ) h + f (xi , yi )
2
Expanding with the chain rule, and substituting it back into the previous Taylor series expansion gives
2
∂f ∂f ∂y h
yi+1 = yi + f (xi , yi ) h + ( + )
∂x ∂y ∂x 2
The next step is to apply a Taylor series expansion to the k2 equation. The applied Taylor series expansion rule is
∂g ∂g
g(x + r, y + s) = g(x, y) + r +s +…
∂x ∂y
and the resulting equation is

∂f ∂f 2
f (xi + p1 h, yi + q11 k1 h) = f (xi , yi ) + p1 h + q11 k1 h + O (h )
∂x ∂y
where O(h2) is a measure of the truncation error between model and true solution.
When this Taylor series expansion result of the k2 equation, along with the k1 equation, is substituted into the general, and a grouping of like
terms is performed, the following results:
∣ ∂f
∣a q ∂f
∣ h2 |O ( h3 )
yi+1 = yi [a1 f (xi , yi ) |a2 f (xi , yi )] h [a3 p f (xi , yi )
∣ 1 ∣ s 11 ∂y ∣
Setting this result equal to the substituted general form will allow us to determine that, for these two equations to be equivalent, the
following relationships between constants must be true.
It should be noticed that these three equations, relating necessary constants, have four unknowns. This means that to solve for three
constants, one must first be chosen. This results in a family of possible second-order Runge-Kutta methods.

Example of usage of Euler's Method:
video.google.com/googleplayer.swf?docId=1095449792523736442
Narrated example of using the Runge-Kutta Method:
video.google.com/googleplayer.swf?docId=-2281777106160743750
Unnarrated example of Using the Runge-Kutta Method:
File:Numerical Solving in Excel, Unnarrated.ppt
2.6.10: REFERENCES
"Delay Differential Equation", Wolfram MathWorld, Online: September 8, 2006. Available
http://mathworld.wolfram.com/DelayDifferentialEquation.html
Chapra, Steven C. and Canale, Raymond P. "Numerical Methods for Engineers", New York: McGraw-Hill.
Franklin, Gene F. et al. "Feedback Control of Dynamic Systems", Addison-Wesley Publishing Company.
R. England. "Error Estimates for Runge-Kutta Type Solutions to Systems of Ordinary Differential Equations", Research and
Development Department, Pressed Steel Fisher Ltd., Cowley, Oxford, UK. October 1968.
Call, Dickson H. and Reeves, Roy F. "Error Estimation in Runge Kutta Procedures", ACM, New York, NY, USA. September 1958.
2.6: Numerical ODE solving in Excel- Euler’s method, Runge Kutta, Dead time in ODE solving is shared under a CC BY 3.0 license and was authored,
2.7: SOLVING ODES WITH MATHEMATICA- HOW TO FIND NUMERICAL AND
ANALYTICAL SOLUTIONS TO ODES WITH MATHEMATICA
Authors: Matthew Baumgartner, Olyvia Dean, Viral Patel, Joel Schweitzer, and Eric Van Beek
Stewards: Brian Hickner, Lennard Gan, Addison Heather, Monique Hutcherson
Date Released: September 6, 2006 /Date Revised: September 8, 2007
2.7.1: INTRODUCTION
Mathematica is an advanced mathematics solution program created by Wolfram Research, Inc. One of the most powerful software packages
of its kind, Mathematica is able to easily and conveniently solve complicated mathematical formulae, including differential equations. This
article focuses on the modeling of first and higher order Ordinary Differential Equations (ODE) in the following forms:
dy
= f(x, y) (Basic ODE)
dx
z
dy
= f(x, y) (Higher Order ODE)
z
d x
Like all software, Mathematica uses a specific language in order to perform calculations. The names of all functions must be capitalized-
this applies to both mathematical functions (such as Sin and Cos) and built-in functions (such as Plot and DSolve). For this reason it is
common for users to write their own functions in minimized letters. This decreases the chance of overriding or redefining a Mathematica
function. Square brackets always follow function calls, and the function's parameters are always enclosed in curved brackets. For example,
if a user wished to plot sin(x) from x = 0 to x = 1, they would type: Plot[Sin[x],{x,0,1}]. The user must type "Shift"+"Enter" to input the
function. Typing only "Enter" will add a line to the formula. . For full PC keyboards, the “Enter” key on the far right is the equivalent of
“Shift”+”Enter”. If the user does not want to see the output line, it can be suppressed by typing a semi-colon at the end of the expression.
Mathematica also recognizes common mathematical constants such as pi (Pi), Euler's constant (E), and imaginary numbers (I). Note that
these constants must also be capitalized when entered.
Mathematica features two functions for solving ODEs: DSolve and NDSolve. DSolve is used when the user wishes to find the general
function or functions which solve the differential equation, and NDSolve is used when the user has an initial condition. The prompts for the
functions are quite similar. Note: Two equal signs must be used to denote equality in an equation. Using one equal sign assigns a value to a
variable.
Example:
f(x) = 5x2 + 7 This input creates a variable named f(x) that has a value of 5x2 + 7.
f(x) = = 5x2 + 7 This input creates a function f(x) which is defined by 5x2 + 7.
Mathematica will sometimes store user created functions or variables into its cache. This causes some (potentially) correct commands to
output errors. It is recommended to quit the local kernel and reinitialize the commands after any changes to your code.
Also, check that every command in the code is being executed. Clearing all the outputs in the cell may be helpful to figure out which
commands had not been executed. This can be done by going to the “Cell” option found on the top and choosing the “Delete All Output”
Example:
OUTPUT IS NOT CLEARED
CLEARING THE OUTPUTS
CODE WITH CLEARED OUTPUTS
As seen in the examples, it is easier to troubleshoot and debug your program when it looks less confusing. Clearing the extra outputs helps
you focus on just the code that you have written.
2.7.2: FIRST ORDER ODES

Notation: In the following examples, eqn represents the ODE, y represents the function being solved for, i represents the initial condition,
and x and t are independent variables.
2.7.2.1: ODES WITH INITIAL CONDITIONS

NDSolve will not display a numerical value, but rather an “Interpolating Function” which can be displayed graphically. The easiest way to
display this graph is to assign the solution to a variable (called Solution in the example) in the input line, and then use the “Plot” function to
display it.
Input for One ODE: Solution = NDSolve[{eqn,y[0] == i},y,{x,xmin,xmax}]
Input for Multiple ODEs: Solution = NDSolve[{eqn1,eqn2,...,y1[0] == i1,y2[0] == i2,...},{y1,y2,...},{x,xmin,xmax}]
Input for Partial Differential Equations: Solution = NDSolve[{eqn,y[0] == i},y,{x,xmin,xmax},{t,tmin,tmax}]
After Mathematica solves the ODE, plot the solution by typing: Plot[Evaluate[y[x] /. Solution],{x,xmin,xmax}], where “/.” is a notation
used by Mathematica meaning “following the rule of”. It basically recalls the function stored as the variable in the previous input line.
Depending on the needs of the user, the functions “ParametricPlot” or “Plot3D” may also be useful. The notation for these functions is the
same as “Plot”.
Here is a simple example of what you would type into Mathematica:
Solution = NDSolve[{y'[x]==y[x]*Cos[x+y[x]],y[0]==1},y,{x,0,30}]
Mathematica will output:
Output[1]= {{y->InterpolatingFunction[ { { 0.,30.} },<>] } }
To plot this function you would type:
Plot[Evaluate[y[x]/.Solution],{x,0,30}]
Note: Remember to type "Shift"+"Enter" to input the function
2.7.2.2: ODES WITHOUT INITIAL CONDITIONS
Input for One ODE: DSolve[eqn,y,x]
Input for Multiple ODEs: DSolve[{eqn1,eqn2,…},{y1,y2,…},x]
Input for a Partial Differential Equation: DSolve[eqn,y,{x1,x2,…}]
Solution = DSolve[y'[x] ==4*x-2*x*y[x],y[x],x]
Output[1]=
2.7.3: SECOND ORDER AND HIGHER ODES

Input for Higher Order ODEs: Solution = NDSolve[{eqn,y[0] == i1, y' [0] == i2,…},y,{ x,xmin,xmax}]
Plotting the solution: Plot[Evaluate[{y[x], y' [x],...} /. Solution],{x,xmin,xmax}]
Solution = NDSolve[{y[x] +Sin[y[x]]+y[x]==0,y[0]==1,y'[0]==0,y,{x,0,30}]
Output[1]= {{y->InterpolatingFunction[ { {0.,30.} },<>] } }
To plot this function you type:
Plot[Evaluate[{y[x],y'[x],y[x]}/.Solution],{x,0,30}]
Mathematica should output a graph.
2.7.4: ALGORITHMS USED BY MATHEMATICA

Mathematica uses two main algorithms in order to determine the solution to a differential equation. These algorithms are the Adams method
and the Gear method.
The Adams and Gear methods are forms of linear multistep methods. An example of these would be the following:
In the example above, h denotes the step size and the coefficients are determined by the method used. Multistep methods are
expansions of more familiar single-step methods used to solve differentials (i.e. Euler, Runge-Kutta, Taylor). Each of these methods
requires an initial point in order to calculate the following point. Similarily, multistep methods also require initial points in order to solve the
ODE. The number of initial points required depends on which method is used to solve the ODE. Multistep methods typically produce less
error than the single-step methods because of multiple initial points.
In order to determine what method to use one must first find the stiffness of the function.
2.7.4.1: ADAMS METHOD

Euler, Taylor and Runge-Kutta methods used points close to the solution value to evaluate derivative functions. The Adams-Bashforth
method looks at the derivative at old solution values and uses interpolation ideas along with the current solution and derivative to estimate
the new solution [4]. In order to solve an ODE using this method, f(t,y) must be continuous and satisfy Lipschitz condition for the y-variable
which states [5]:
for all | h | < ε where B and β are independent of h, β > 0 and α is an upper bound for all β for which a finite B exists
This is a basic form of the Adams-Bashforth method. Note that two initial points are required for this method.
There is another Adams method that requires three initial points. The method is solved the same way, however the equation varies a little bit
and is referred to as the Adams-Moulton Method.
The coefficients/constraints, β can be solved for using knowledge of ODE's and other math tools. It was stated earlier that .
We can let f(t,y) = λy, therefore . We can also let if there is a constant step size and σ represents a polynomial.
Through substitution we find [6]:
We can expand the quadratic using another math identity and ultimately solve for constraints β1 and β2. Another method for solving for
coefficients β1,β2 is mentioned below:
In order to find the coefficient βj one must first use polynomial interpolation to find the polynomial p of degree s − 1 such that:
for
From this the Lagrange formula for polynomial interpolation yields
Now the polynomial p is a locally good approximation of the right-hand side of the differential equation y' = f(t,y) that is to be solved. Now
we must consider the equation y' = p(t) instead. This equation can be solved exactly by simply taking the integral of p.
The Adams–Bashforth method arises when the formula for p is substituted. The coefficients bj turn out to be
s−j 1 s
(−1)
bet aj = ∫ ∏(u + i) du.
(j − 1)!(s − j)! 0 i=1
i≠j
The Adams-Bashforth method is typically used for Linear and Non-liner ODE's with dense systems.
2.7.4.2: GEAR METHOD

Taking
bet ai = 0
for i ≥ 1
The Gear method, also known as the backward differentiation formulae (BDF, a.k.a. Gear’s formulae) is another multi-step method but is
generally used for multiple equations. In order to use the gear method your function must have a stiffness greater than 500, meaning that the
function is stiff.
As a particular cases, taking for i ≥ 2 and optimizing the remaining coefficients to maximize the accuracy of the resulting scheme
recovers the Implicit Euler method. Taking for i ≥ 3 gives
We now focus on this case in particular. Applying this method to the scalar model problem and assuming constant h and a solution of the
form , we find the following quadratic equation for
the two roots of which are given by

− −−−−−− − −−−−−−
− α1 + √ γ(1 + ϵ) − α1 − √ γ(1 + ϵ)
sigma = or
2(1 − Σ) 2(1 − Σ)
2
gamma = α − 4α2
1
Applying the identities, we may expand both roots in terms of powers of h. By our assumed form of the solution, it follows that
.The leading-order term in the expansion in h of (a “spurious root”) is proportional to h. For small h, quickly
decays to zero, and thusmay be neglected. The leading-order terms in the expansion in h of (the “physical root”) resemble the Taylor-
series expansion of the exact solution over a single timestep.
Matching coefficients with the expansion of the exact solution as indicated by underbraces in the above
expression, and applying the definition , we arrive at three equations for to achieve the highest order of
accuracy possible with this form. It is easily verified that satisfy these three equations. The leading-
order error term of this method is proportional to . Thus, over a single timestep, the scheme is “locally third-order accurate”; more
significantly, over a fixed time interval [0,T], the scheme is globally second-order accurate. The resulting method,
is thus referred to as BDF2, and may be viewed as a implicit alternative to AM3 that, for the same number of steps into the past, p =
max(m,n), has reduced order of accuracy but greatly improved domain of stability. Higher-order BDFs may be derived in an analogous
fashion; BDF3, BDF4, BDF5, and BDF6 in particular are found to have excellent stability properties as compared with their AM
counterparts with the same number of steps.

You have been assigned a non-isothermal reactor with the following material and enthalpy balances.
2 2
fracdca dt = ca (t ) − cb (t) − T (t )
Species a Material balance

3 2
fracdcb dt = cb (t ) − ca (t) − T (t )
Species b Material balance

2
fracdT dt = ca (t) − cb (t) − T (t )
Enthalpy balance
When is the concentration of species a and species b equal? The initial conditions for the reactor are as follows:
T0 = 1
ca0 = 2.0
cb0 = 1.8
tf = 0.2
s = NDSolve[{x'[t] == x[t]^2 - y[t] - z[t]^2, y'[t] == y[t]^3 - x[t] - z[t]^2, z'[t] == x[t] - y[t] - z[t]^2, x[0] == 2, y[0] == 1.8, z[0] == 1},
{x, y, z}, {t, 0.2}]
Plot[Evaluate[{x[t],y[t]} /. s], {t, 0, 0.2}]
You have been asked to study a semi-batch reactor for the reaction . The rate law is , where k = 2.7. Other
parameters are: initial volume = 5, volumetric flow rate = 0.05, initial concentration of A = 2500. Use Mathematica to create a Conversion
versus Time plot for 100 time units.
1. Mole Balance:
ra V
fracdXdt = −
NA0
2. Rate Law:
3. Stoichiometry:
4. Combine:
If the semi-batch reactor runs for 100 time units, the conversion is about 0.8. This can be seen in the plot below or by running the two
commands that follow into Mathematica.
s = NDSolve[{y'[t] == 2.7 (1 - y[t]) (5 + .05 t)/(2500 y[t]), y[0] == 0.0001}, y, {t, 0, 100}];
Plot[Evaluate[y[t] /. s], {t, 0, 100}, PlotRange -> All]
This is the Mathematica notebook file for the example: Media:MathematicaEx2.nb

Note: The parameters used in this problem are fabricated and are intended to illustrate the use of Mathematica in solving ODEs.
2.7.7: FORMATTING PLOTS IN MATHEMATICA

Here are some useful tips to formatting plots in Mathematica:
Label Plot: e.g. Plot[Sin[x], {x,0,2Pi}, PlotLabel->"Volume of Tank vs. Time"]
Label Axes: e.g. Plot[Sin[x], {x,0,2Pi}, AxesLabel->{"time(min)","volume(L)"}]
Color Plot: e.g. Plot[Sin[x], {x,0,2Pi}, PlotStyle->Red]. You can also make the plot thick, dashed, etc. e.g. Plot[Sin[x], {x,0,2Pi},
PlotStyle->{Red,Thick,Dashed}]
Insert Legend: e.g. Needs["PlotLegends`"] <--MUST insert this BEFORE your plot command. Plot[{Sin[x],Cos[x]}, {x,0,2Pi},
PlotLegend->{"sine", "cosine"}]

Solving ODEs with Mathematica Video
slides for this talk
2.7.9: ADDITIONAL TIPS AND TRICKS FOR TROUBLESHOOTING IN MATHEMATICA
Mathematica is a powerful computing tool, however the syntax can be a bit difficult to understand. Here are some notes for troubleshooting
in Mathematica.
1. Check to make sure that your variable names and signs are consistent.
Ex) Make sure you use xI everywhere instead of xI and x1 or xl.
Ex) Functions, including the ones you create, are usually followed by brackets such as Sin[x] or y[x]. However, brackets are not
necessary when you are solving for a function in a set of differential equations such as NDSolve[eqns, {y}, {x, 0, 50}];
Ex) Check to see if your parentheses are aligned such that you are actually entering the function you think you're entering. Recall order
of operations and the distributive property. x*(a+b) is NOT equal (x*a) + b. This seems simple, but often gets overlooked when dealing
with several lines of code.
2. You may find it easier to define all of your variables together at the beginning of your code. This makes it easy to go back and change the
values assigned to each variable if trying to determine the impact one variable has on a system of equations. For instance, say you are trying
to determine the effects the flow rates into two different tanks (F1, F2) will have on the tank volumes over ten time steps. The differential
equations governing the situation are: and , where F1 = 2, F2 = 12, V1(0) = 0, V2(0) = 0.
If you write the ODEs in Mathematica by directly substituting in F1 = 2 and F2 = 12, you will have to change the ODEs each time you
change the values of F1 and F2. Below is an example of what the Mathematica equations would look like.
s = NDSolve[{V1’[t] == 8 – V1[t], V2’[t] == 12 – (1/3)*V2[t], V1[0] == 0, V2[0] == 0},{V1,V2},{t,0,10}]
Another option is to define F1 and F2 before your equations, and then call them when solving for your ODEs. This allows you to easily
substitute in new values for your variables without changing the differential equations, and reduces the odds of making simple
computational errors in your ODE in the process. The Mathematica code would look as shown below.
variables = {F1 -> 2, F2 -> 12};
s = NDSolve[{V1’[t] == 4*F1 – V1[t] , V2’[t] == F2 – (1/3)*V2[t], V1[0] == 0, V2[0] == 0} /. variables,{V1,V2},{t,0,10}]
3. Be Aware of the Kernel
The Mathematica Kernel stores all of the information about variable definitions. When you define a variable , the definition is stored there.
The kernel is automatically started when a new Mathematica session is started. You may also start it manually. To start the kernel manually
go to Evaluation -> Start Kernal -> Local. Once the kernel is started and you wish to go back and change a variable definition you must
"Quit the Kernal" before you see the change occur. In order to "Quit the Kernal" you must go to Evaluation -> Quit Kernal -> Local.
Terminating the Mathematica Kernal, erases all of the definitions previously entered. For this reason, after you "Quit the Kernal" and enter
in new definitions for your variables you must re-enter all of your code. The images below show how to "Start Kernal" and how to "Quit the
Kernel."
Example:
An example of when you would want to "Quit the Kernal" when using Mathematica for Controls is when you are finding a steady state
point using a PID controller. We start by defining the variables for vset, taui, taud, and Kc. Mathematica solves this equation using the
defined variables and shows that (x->3.9, V->10) is the steady state point. This is shown in the figure 1 below.
Figure 1 when you defined Kc even though you have not defined it here.
Figure 3 shows the output once you have "Quit the Kernel" and re-entered the Mathematica code.
Now you see that the definition for Kc has been deleted because the steady state points are in terms of Kc. To find the impact of Kc you can
use this solution.
4. Define functions or formulas that are used often
If you will use a function often and it is not already defined in Mathematica by default, define your own. This can be especially helpful if
you intend repeat the same function multiple times while only changing variables. A scenario where defining your own function could save
you time and error is when comparing P-Fisher's values while keeping the margins constant. Only the variables will change but the function
will remain the same. The new function will only last as long as each session in Mathematica.
To define a function, begin by naming the function followed by an open bracket. List all the variables with an underscore after each name
and separate each variable with a comma. Close the bracket when all the variables have been listed. To let Mathematica know that this is a
newly defined function, after closing the bracket place a semi-colon and equal sign. Now define what the formula is using the variable
names. When you have finished, click shift-enter simultaneously. To test the new formula, start with naming the function and open bracket,
then list all the variables numerical values separated by comma's in the same order that was listed when defining the function. Close the
bracket after all the variables have been listed and click shift-enter simultaneously. The output will be the next line in Mathematica with the
answer.
Example:
It is also possible to use this tool to put equations in terms of unknown variables. To do this begin by defining the function the same way as
before. When plugging in the numerical values of the variables, leave the unknown variables as their variable name. The Mathematica
output will provide the answer in terms of the unknown variables.
Example:
5. "Make it pretty"
Oftentimes when people program in any language, they tend to crowd lines, almost as if they were trying to save paper. Well, since this does
not apply when you are typing on a screen, don't worry about it.
Ex) You could write:
vars = {V -> 1000, Cao -> 0.2, UA -> 20000, F -> 2000, DE1 -> 30000, k1 -> 1.0 10^13, DH1 -> 50000, rcp -> 60, R -> 1.987, To -> 500,
Tf -> 500, Tjin -> 500, Fj -> 1000, Vj -> 100, Caf -> 0.2, DH2 -> -192000, DE2 -> 30000, k2 -> 1.0 10^13};
eqns = {Ca'[t] == (Caf - Ca[t]))/V - Ca[t]*k1*Exp[-DE1/(R*T[t])], Cb'[t] == (0 - Cb[t])/V +k1*Ca[t]*Exp[-DE1/(R*T[t])] -
k2*Cb[t]*Exp[-DE2/(R*T[t])], T'[t] == (Tf - T[t])/V + (-DH1/(rcp))*k1*Ca[t]*Exp[-DE1/(R*T[t])] + (-DH2/rcp )*k2*Cb[t]*Exp[-
DE2/(R*T[t])] - (UA (T[t] - Tj[t]))/(V *rcp), Tj[t] == (Fj (Tjin - Tj[t]))/Vj + (UA (T[t] - Tj[t]))/(Vj *rcp), Ca[0] == 0.2, Cb[0] == 0, T[0] ==
500, Tj[0] == 500};
sol = NDSolve[eqns /.vars, {Ca, Cb, T, Tj}, {t, 0, 50}]; Plot[{Ca[t]} /. sol, {t, 0, 50}, AxesLabel -> {t, Ca}, PlotRange -> Full];Plot[{Cb[t]}
/. sol, {t, 0, 50}, AxesLabel -> {t, Cb},PlotRange -> Full];Plot[{T[t]} /. sol, {t, 0, 50}, AxesLabel -> {t, T}, PlotRange -> Full];Plot[{Tj[t]}
/. sol, {t, 0, 50}, AxesLabel -> {t, Tj}, PlotRange -> Full]
but it looks much better if you do this:
These thoughts will hopefully help you later on and make it easier to read your code.
6.) "Check the colors!"
Here is a list of font color that Mathmatica will output for a specific reason.
Local Variables in a certain Font Color
Local variables of Module and With in Green
1. Example:
Function arguments and pattern names in Green (Italics)
1. Example:
Variables made special by use in arguments in Turquoise
Example:
Errors and Warnings in a certain Font Color
Syntax Errors in Purple
Example:
Emphasized Syntax Errors in Dark Red with Yellow Background
Example:
Missing arguments in Bright Red
1. Example:
Excess arguments in Bright Red
1. Example:
Possible unwanted assignments in Bright Red
Example:
Unrecognized option names in Bright Red
Example:
Local Scope conflicts in Dark Red with Yellow Background
Example:
Variables that will go out of scope before being used in Bright Red
Example:
Shadowing in multiple contexts in Bright Red
Example:
Other in a certain Font Color
Comments in Light Grey
Example:
Strings in Dark Grey
Example:
Global symbols that have no value assigned in Bright Blue
Example:
2.7.10: ACCESSING MATHEMATICA FROM YOUR PERSONAL COMPUTER

If you prefer to use Mathematica from your home computer rather than one on campus, you can remotely log into a CAEN computer
through U-M’s Virtual Sites service. Follow the steps outlined below:
2. Visit http://virtualsites.umich.edu/ . The U-M WEBLOGIN page will appear. Log in using your unique name and Kerberos password.
3. The Virtual Sites page should appear on your screen. Click on the large, orange Connect Now button.
4. Under the heading What software do you need?, select Engineering (CAEN) Legacy XP. Then click on the Request Connection
button toward the bottom of the page.
5. Follow the instructions under the heading Launch Virtual Sites by downloading and opening the file in the Internet Explorer
information bar. A window will pop up on your screen titled Remote Desktop Connection. Click the Connect button.
6. You will then see the screen that appears when logging into a CAEN computer. Your unique name should already be included. Type your
Kerberos password and click OK to log in to the CAEN computer.
7. Click OK on the Notice window that pops up and your personal settings will be applied. Also click OK if a System Requirements
Wizard window appears.
8. If you scroll down using the scroll bar on the right-hand side of the screen, you will see the Start button. Click Start -> All Programs -
> Math and Numerical Methods -> Wolfram Mathematica -> Mathematica 6. Once Mathematica opens on the computer screen,
you may use it as though you were actually sitting in front of a CAEN computer on campus.
9. To log out, click Start -> Log Off. Under the Log Off Windows window, click Log Off.
2.7.11: REFERENCES
"Adams Method", Wolfram MathWorld, Online: August 5, 2007. Available http://mathworld.wolfram.com/AdamsMethod.html [1]
"Gear Predictor-Corrector Method", Davidson University, Online August 5, 2007. Available
http://webphysics.davidson.edu/Projects/SuFischer/node47.html [2]
Fogler, H. Scott (2006), Elements of Chemical Reaction Engineering, New Jersey: Pretince Hall PTR. ISBN 0-13-047394-4 [3]
"Adams-Bashforth", Pitt Math, Online: September 8, 2007. Available:
www.math.pitt.edu/~sussmanm/2071Spring07/lab02/index.html#AdamsBashforthMethods [4]
"Lipschitz Condition", Wolfram MathWorld, Online: September 7, 2007. Available
http://mathworld.wolfram.com/LipschitzCondition.html [5]
"Adams-Bashforth (AB) Methods", University of California-San Diego, Online: September 7, 2007. Available:
renaissance.ucsd.edu/chapters/chap11.pdf [6]
"Gears Method", University of California-San Diego, Online: September 7, 2007. Available: renaissance.ucsd.edu/chapters/chap11.pdf
[7]
Use the Harvard Referencing style for references in the document.
For more information on when to reference see the following Wikipedia entry.
2.7: Solving ODEs with Mathematica- How to find numerical and analytical solutions to ODEs with Mathematica is shared under a CC BY 3.0 license and
was authored, remixed, and/or curated by Peter Woolf et al. via source content that was edited to conform to the style and standards of the LibreTexts
2.8: FITTING ODE PARAMETERS TO DATA USING EXCEL- USING
REGRESSION TO FIT COMPLEX MODELS IN EXCEL
2.8.1: SECTION 8. FITTING ODE PARAMETERS TO DATA USING EXCEL: USING REGRESSION TO FIT
COMPLEX MODELS IN EXCEL
Note: Video lecture available for this section!
Authors: Anthony Campbell, Victoria Cardine, David Hines, Stephen Kerns /Date Revised: 14th September 2007
Stewards: Jeff Byrd, Khek Ping Chia, John Cruz, Natalie Duchene, Samantha Lyu
2.8.1.1: INTRODUCTION
One of the most important abilities of an engineer is to model measured values of a system. Modeling of a system aids in optimizing
objectives, as well as predicting future operations. This can be especially helpful when changes cannot be immediately implemented. In
general modeling terms, a set of data can always fit a set of variables. However, an excessive number of degrees of freedom usually exist,
creating multiple solutions. Therefore, the physical significance of the variables is an important consideration.
As time proceeds, various values may need to be changed to accommodate variations within the system. The following sections will discuss
the use of Microsoft Excel© as a tool to adjust the values of variables given a set of experimental data. The models presented below take
into account the design equations of the system being modeled, and the rate laws, where applicable.
2.8.1.2: ADDING IN THE SOLVER APPLICATION IN EXCEL 2007

In order to use "Solver" in Excel 2007, one must first add the Excel tool-pack. To add the tool-pack, follow these instructions:
Click on the Microsoft Office symbol at the top left corner of the screen
Click on “Excel options” at the bottom right corner of the drop box
Click on “Add-Ins” in the left hand column
Click on the “Solver Add-In” option at the bottom of the list
Click on “GO” at the bottom right corner
Check the “Solver Add-in” box at the bottom of the list
Click on “OK” at the top of the right column
To access Solver, click on the “Data” tab at the top of the screen. The Solver tool is located in the "Analysis" box on the right side of the
screen.
The user interface of Excel 2007 differs from those of previous versions. The following images illustrate the process just described.
In order to use the Solver function, click on Data and then the Solver button in the Analysis box.
NOTE: If using Microsoft Office 2003, the Solver Application can be added by clicking on the "Tools" tab at the top of the screen. The
same directions can be followed after this step.
NOTE: If you receive an error message while trying to access Solver, it may already be added in Excel. Uninstall Solver (by following the
same steps, but unchecking "Solver Add-in") then reinstall it.
2.8.1.3: ADDING IN THE SOLVER APPLICATION IN EXCEL 2008
Excel 2008 for Macs does not come with the solver add-in. The software can be downloaded at Solver.com
This solver application requires Excel 12.1.2. If this is not the current version of Excel being used, it can be updated by running Microsoft
AutoUpdate or by opening excel, going to help and then clicking on "Check for Updates". Make sure when updating Excel to close all
Microsoft Applications.
This application allows the user to have a solver program that can be opened and used while running Excel. It works the same way as the
solver add-in.
2.8.1.4: USING EXCEL SOLVER TO FIT ODE PARAMETERS

Excel Solver is a tool that can be used to fit function variables to given experimental data. The following process can be used to model data:
Define the data set by entering values into an Excel spreadsheet
Define the model that you want to fit to the data
Define the sum of least squares in one of the spreadsheet blocks
In Solver, have Excel minimize the sum of least squares by varying the parameters in the model
Accurate modeling is dependent on two factors; initial values and verifying results.
Initial values: As stated in the introduction, many data fitting problems can have multiple “solutions”. In numerical methods, given a set of
initial parameters the data will converge to a solution. Different initial values can result in differing solutions. Therefore, if the initial values
are not set by the problem statement, multiple initial guesses should be entered to determine the “best” value for each variable.
Verification of curve fit: When fitting a curve to given data points, it is important to verify that the curve is appropriate. The best way to do
this is to view the data graphically. It is fairly simple to use the Chart tool to graph the data points. Adding a trend line will show the
mathematical relationship between the data points. Excel will generate a function for the trend line and both the function and the R-squared
value can be shown on the plot. This can be done by clicking “Options” next to “Type” of trend and checking the empty boxes that are
labeled “Display equation on chart” and “Display R-squared value on chart”.
There are various types of trendlines that can be chosen for your given data points. Depending on the trend that you may see, you might
want to try a few of the options to obtain the best fit. Sometimes the squared residual value will be better for a certain type of trend, but the
trend is not necessarily “correct”. When you have obtained sample data, be aware of the trend you should be seeing before fitting a certain
trend type to the data points (if this is possible). The polynomial trend type gives you the option to change the order of the equation.
Typically, you will not acquire a squared residual value of zero. This value is a simple analysis of the error between the trend line, and the
actual data. Once the function and squared residual values are generated, you can begin to evaluate the generated solution.
During your solution analysis, it is important to check for variables or parameters that may be unnecessarily influencing the data. Excel can
generate a "best" solution, which is not correct due to background calculations or Excel's misinterpretation of the modeled system. To assess
this possibility, you can graph data generated by using the solved for variables against the given data.
Additional information about the Excel Solver Tool can be found in Excel Modeling.
2.8.1.5: WORKED OUT EXAMPLE 1: MASS BALANCE ON A SURGE TANK
In this example we will walk through the scenario in which you want to model a non-heated surge tank in Microsoft Excel©. This model
would be very critical in a process where smooth levels of process flow are required to maintain product specifications. Therefore, an
engineer would like to know whether they will need to periodically refill or drain the tank to avoid accidents. Let's look at this simple
example:
Problem statement:
Suppose water is being pumped into a 10-ft diameter tank at the rate of 10 ft3/min. If the tank were initially empty, and water were to leave
the tank at a rate dependent on the liquid level according to the relationship Qout = 2h (where the units of the constant 2 are [ft2/min], of Q
are [ft3/min], and of h are [ft]), find the height of liquid in the tank as a function of time. Note that the problem may be, in a way, over
specified. The key is to determine the parameters that are important to fit to the experimental surge tank model.
Solution:
The solution with fitted parameters can be found here: Surge tank solution This Excel file contains this scenario, where given the actual data
(the height of the liquid at different times), the goal is to model the height of the liquid using an ODE. The equations used to model h(t) are
shown and captioned. In order to parameterize the model correctly, Solver was used to minimize the sum of the least square differences at
each time step between the actual and model-predicted heights by changing the parameters' values. The model predicts the actual height
extremely well, with a sum of the least squared differences on the order of 10^-5. A graph showing the modeled and actual height of the
tank as a function of time is included as well.
2.8.1.6: WORKED OUT EXAMPLE 2: FITTING A HEATED SURGE TANK MODEL(A HEATED CSTR) PARAMETERS TO
DATA IN EXCEL
Problem Introduction: In this example we will analyze how we can fit parameters used for modeling a heated CSTR to data stored in
Excel. The purpose of this problem is to learn how to vary ODE parameters in Excel in order to achieve a square difference between the
calculated and experimental data of zero. Some questions to consider are, “How do I model this? Is this a first order ODE? Are there any
coupled ODE's? How many variables need to be considered?" This example will use an Excel file to help you in this process.
Problem Statement: Download the heated CSTR problem file.CSTR Problem Take the experimental data produced from the heated CSTR
model in the Excel spreadsheet*. You are modeling a heated CSTR. It would be wise and is highly recommended to review both the heated
CSTR problem and a wiki on using Excel's solver.
You know this is a heated CSTR...
2. Numerically solve the ODEs through modifying the given wrong set of parameters.
3. While changing the parameters, use Excel to calculate the error between the calculated model and the experimental data.
4. Use the Excel solver function to solve for some of the parameters to minimize error to less than 5%
5. How would your approach be different if our heat transfer area needed to be as small as possible due to capital constraints? Explain.
6. How would your approach be different if our mass feed rate has to be in the range of 3-3.5 kg/s? Explain.
Problem Solution: The solution with fitted parameters can be found here.CSTR Solution
With the model and data given in the example the error must first be defined. Here there error is the squared residual or [(model observation
at time i)-(data observation at time i)]^2 over (all i).
Chemical engineering intuition must be used to reasonably approximate some variables. In real life we would double check that "parameters
normally known, but erroneous" are correct. (Due to problem constraints, we will assume these are accurate, but a true chemical engineer
would check regardless). We must now modify "parameters often needing fitting". For part 3, we can modify the parameters as shown in the
solution.
The original parameters have been modified to produce an error of zero.
If we have restrictions on heat exchange area then we should try to fit the "heat transfer coefficient and area" as low as possible. Likewise, if
we have restrictions on feed rate, we must keep this above 4 while fitting other parameters as best as possible to achieve a small error. These
have certain industrial applications as costing of heat exchangers and limitations on mass flow rates need to be modeled before a reactor
design is implemented.
Side Note: The example above assumes a first order rate law. This is not always the case. To change the equations to fit a new rate law, one
must derive a new dCa/dt equation and dT/dt using their knowledge of reactor engineering. Once the new equations are derived write the
formula in an excel format in our dCa/dt and dT/dt column and then drag it to the end of the column. All other columns should stay the
same.
2.8.1.7: SAGE'S CORNER

A brief narration for a better understanding of Example 1 - Surge Tank Model and tips for effectively picking a model equation can be found
here: video.google.com/googleplayer.swf?docId=70222262469581741
A copy of the slides can be found here:
Unnarrated Slides
An example of fitting an ODE parameter to data using excel - based on a simple radioactive decay rate ODE:
A copy of the slides can be found here: Unnarrated Slides
A copy of the excel file can be found here: Excel File
An example of modelling a reaction rate ODE with rate constant and rate order variable parameters:
video.google.com/googleplayer.swf?docId=-4220475052291185071
A copy of the slides can be found here: Unnarrated Slides
A copy of the excel file can be found here: Excel File
2.8.1.8: REFERENCES
Holbert, K.E. "Radioactive Decay", 2006, ASU Department of Electrical Engineering, Online: October 1, 2007. Available
www.eas.asu.edu/~holbert/eee460/RadioactiveDecay.pdf
2.8: Fitting ODE parameters to data using Excel- Using regression to fit complex models in Excel is shared under a CC BY 3.0 license and was authored,
2.9: HELPFUL MATHEMATICA SYNTAX- HINTS ON HOW TO USE
MATHEMATICA TO MODEL CHEMICAL PROCESSES
2.9.1: SECTION 9. HELPFUL MATHEMATICA SYNTAX: HINTS ON HOW TO USE MATHEMATICA TO
MODEL CHEMICAL PROCESSES
Title: Useful functions in Mathematica as pertaining to Process controls
2.9.1.1: INTRODUCTION
Mathematica has many different functions that are very useful for controlling chemical engineering processes. The Mathematica syntax of
these functions is often very complicated and confusing. One important resource is the Mathematica Documentation center, which is
accessed through the program help menu. The Documentation Center provides an extensive guide of every function in the program
including syntax and sample code. A "First Five Minutes with Mathematica" tutorial is also present in the menu, which provides a quick
guide to get started with Mathematica. This article will briefly outline several useful Mathematica functions and their syntax.
2.9.1.2: MATHEMATICA BASICS

Here are some basic rules for Mathematica.
Notebook, Cells and Evaluation
The standard file for Mathematica is called a notebook. The notebook consists of two kinds of cells: input and output. The inputs are typed
into one cell and the outputs are displayed in another (See Below).
In order to have Mathematica calculate your inputs, you must evaluate the cells. Do this by pressing Shift+Enter. This will only evaluate
the current cell that your cursor is located. To evaluate all the cells of a notebook press Ctrl+A, to select all, and then press Shift+Enter.
Parentheses vs. Brackets
Brackets, [], are used for all functions in Mathematica. The only thing parentheses, (), are used for is to indicate the order of operations.
Equality, = vs. ==
The single equal sign (=), is used to assign a value or a function to a variable. The double equal sign (==) is used to set two values equal to
each other, such as solving for a value, or to test equalities. When testing equalities, Mathematica will output 'True' or 'False.'
If you recieve 'True' or 'False' as an output when you are expecting it, you likely used a == somewhere that you should have used a =.
Semicolon Use
Placing a semicolon after a line of code will evaluate the expression; however, there will be no output shown.
Mathematica Functions
All Mathematica functions start with a capital letter, eg. Sin[x], Exp[x], Pi or Infinity.
Assigning and Inserting Variables or Parameters
To assign values to variables, use -> rather than an equals sign. For example,
To insert a set of parameters to a function use /. This symbol applies a rule or list of rules in an attempt to transform each subpart of an
expression.
For example, if you want to enter the above parameters into the expression y = Kc * x + Ca0, enter the following in Mathematica:
Variables are case sensitive. For example, X is different than x. Also, if you define a variable, it will remain defined until you redefine it as
something else or clear it. You can clear a variable, for example, x, by using the Clear[x] command. You can also quit the Kernel, which will
clear all variables.
Forcing a Numerical Expression
To force Mathematica into giving a numerical solution, use the function N[expression]. For example,
Another method to do this is to place a decimal place in any of your numbers (i.e. "5." instead of just "5")
2.9.1.3: INTEGRATION
Mathematica can do some very complex integrations with fairly little coding. Integration can be very helpful for many engineering
applications, but when the functions become very complex they become increasingly difficult to integrate. Mathematica will integrate any
function that is possible to integrate.
To integrate a function f(x) the Mathematica, the command is Integrate[].
For an indefinite integral the correct coding is Integrate[f(x),x] where x is the independent variable.
For a definite integral with limits of y and z the correct coding is Integrate[f(x),{x,y,z}]
For example:
We can integrate a function like the one below:
f(x) = Sin(5 * x / 8) * x
We can find the indefinite integral as seen here:
We can find the definite integral from 0 to π as seen here:
2.9.1.4: SOLVER
Mathematica's Solve function attempts to solve an equation or set of equations for the specified variables. The general syntax is Solve[eqns,
vars]
Equations are given in the form of left hand side == right hand side.
A single variable or a list of variables can be also be specified. When there are several variables, the solution is given in terms of a list of
rules. For example,
When there are several solutions, Solve gives a list of them. For example,
Solve will not always be able to get explicit solutions to equations. It will give the explicit solution if it can, then give a symbolic
representation of the remaining solutions in terms of root objects. If there are few symbolic parameters, you can the use NSolve to get
numerical approximations to the solutions.
2.9.1.5: PLOTTING
Mathematica contains a very powerful plotting interface and this overview will go over the basics of plotting. Using Mathematica's Plot[ ]
function there is a basic syntax to be used:
Plot[function, {variable, variable_min, variable_max}]
For example, the command "Plot[y=x, {x,0,10}]" will output a graph of the function y=x for a range of x from 0 to 10.
The function could also be previously defined and then referenced in the Plot [ ] command, this functionality provides a lot of simplicity
when referencing multiple functions and helps prevent a typo from affecting your Plots.
y[x]==x
Plot[y[x],{x,0,10}]
To plot multiple functions on the same plot, use an array of functions using curly brackets { }, in place of the single one:
Plot[{y=x,y=2*x}, {x,0,10}]
Another useful plotting option in Mathematica is the ListPlot function. To use this function, the following formatting should be used:
ListPlot[Table[f(x),{x,min,max}]
This is obviously only the basics, Mathematica provides many more options including colors, legends, line styles, etc. All of these extra
features are well documented at Wolfram's Reference Site.
An additional piece of information: Type the following command on the line after the Plot command to determine the max value plotted. It
will display the number associated with the absolute max of the function within the plotting range.
Max[Last/@Level[Cases[%,_Line,Infinity],{-2}]]
2.9.1.6: SOLVING ODES

Within Mathematica, you have the option of solving ODEs numerically or by having Mathematica plot the solution.
Explicit solution:
The built-in function DSolve solves a differential equation or a list of differential equations for either one indepedent variable or a list of
independent variables, which solves a partial differential equation. The most basic syntax for the function DSolve is DSolve[eqn,y,x]. For
example,
The syntax for solving a list of differential equations is DSolve[{eqn1,eqn2,...},{y1,y2,...},x], while the syntax for solving a partial
differential equation is DSolve[{eqn, y, {x1,x2,...}]. For example,
Note that differential equations must be stated in terms of derivatives, such as y'[x]. Boundary conditions can also be specified by giving
them as equations, such as y'[0]==b.
Plotted Solution:
Solutions given by DSolve sometimes include integrals that cannot be carried out explicitly, thus NDSolve should be used. NDSolve finds a
numerical solution to an ordinary differential equation within a specified range of the independent variable. The result will be given in terms
of InterpolatingFunction objects, and thus the solution should be plotted. For example,
Note that a place holder was defined to store the NDSolve solution in order for the solution to be plotted. The general syntax for plotting is
Plot[Evaluate[y[x] /. s], {x, xmin, xmax}], where s is said placeholder. The plot below shows the solution to this example.
2.9.1.7: ANOTHER EXAMPLE SOLVING ODES

In this example the water level of a tank is controlled by adjusting the feed rate of water into the tank. The system is modeled by the
following equation:
There is a PID controller on the valve but is not important in explaining the Mathematica syntax. Consider it just a different way to define
Fin.
Parameters are assumed to be Vset = 10, Kc = 10, tauI = 0.1, and tauD = 1
For this problem the differential equation needs to be modeled with the given parameters in Mathematica. The problem asks what the steady
state of the system is. This is solved for by substituting in a PID expression for Fin (not important in this context), setting the derivatives
equal to zero, and solving for V and Xi(a made up variable for the PID controller). The following Mathematica code does all said things.
The final line shows that the steady state values for the problem are V = Vset or 10 units of volume, and Xi (the made up variable for the
PID controller) has a steady state value of 3.9
In this example there is syntax for defining parameters, applying those parameters to equations, setting up equations, and solving equations
for steady state values. There is also syntax for plotting multiple equations on the same graph and how to evaluate multiple functions over
time periods. Overall very useful for solving Controls Engineering problems in Mathematica.
2.9.1.8: MATRIX
It is very simple to create a matrix in Mathematica. Simply pick a name for the matrix (for this example we will use a) then simply enter
each row inside of curly brackets {} with the individual values separated by commas, and the the row separated by a comma. The function
MatrixForm[] will display a matrix in its matrix form rather than as a list of numbers.
EX: a =
MatrixForm[a] displays:
523
7 9 11
8 13 9
You can refer to any specific position in the matrix by using the system:
matrix[[row],[column]]
EX: For the matrix above a[[1],[2]] a value of 2 would be returned.
2.9.1.9: JACOBIANS AND EIGENVALUES

The Jacobian is a helpful intermediate tool used to linearize systems of ODEs. Eigenvalues can be used with the Jacobian to evaluate the
stability of the system. Mathematica can be used to compute the Jacobian and Eigenvalues of a system by the following steps.
1. Enter the system as individual equations
entered as
eq1 = 3*x^2+2y
eq2 = x^2-3*y^2
2. Compute the Jacobian with Mathematica's Derivate function
computed by
Jac =
3. Use Mathematica's Eigenvalues function to evaluate the stability of the system
Begin by entering a desired value of {x,y} into the matrix (i.e. steady state values of x and y).
Jac =
Then call the Eigenvalue function.
Eigenvalues[Jac]
Result: λ={-1.82, 0.85}
The Jacobian can be computed as long as the matrix is square. Many other function types are acceptable when entering eq1 and eq2,
including trigonometric and exponential functions.
2.9.1.10: USER-DEFINED FUNCTIONS

Mathematica also supports user-defined functions. Defining your own functions helps to simplify code, especially when lengthy expressions
are used consistently throughout the code. The generic Mathematica syntax for defining a function is as follows:
f[a_,b_,c_]:=(a+b)/c
Here, the function name is "f" with three arguments: a, b, and c. The syntax ":=" assigns the expression on the right hand side to the function
"f." Make sure that each of the arguments in the brackets on the left hand side are followed by underscores; this identifies the variables
being called by the function as local variables (ones that are only used by the function itself). For this example, the function "f" will take
three arguments (a, b, and c) and calculate the sum of a and b, and then divide by c.
After defining the function, you can use the function throughout your code using either real numbers, as the first example shows, or any
variables. This is done by calling the function with the correct number of arguments. When referencing variables, if the variable already has
an assigned value the value will be used, if it does not, Mathematica will display your function's results with the variables displayed. A few
examples are shown below.
2.9.1.11: NONLINEAR REGRESSION

Mathematica can be used to solve multiple variable, nonlinear regression problems. This form of regression can be useful to create models
of how temperature or pressure sensors depend on different valves in a system. The NonLinearRegress function in Mathematica can be used
to find a best-fit solution to a user defined functional form.
The NonLinearRegress function is not part of the standard package in Mathematica. For this reason, this function must be loaded into
Mathematica before an attempt to use.
After this step, the NonlinearRegress is loaded and can be used to solve for a set of data. To use the this function, enter
NonlinearRegress[data, expression, parameters, variables]. The best fit values for the parameters will be returned along with statistical
information about how well this form fits the given data. An example of Nonlinear Regression in Mathematica is shown in the image below.
2.9.1.12: PROBABILITY DENSITY FUNCTION

Often in controls, it is necessary to calculate the probability that a certain event will occur, given a particular known distribution. In these
scenarios, it is helpful to know how to use the built-in probability density function in Mathematica. The syntax for using the probability
density function to calculate the probability of observing a value x for a generic distribution dist is shown below.
PDF[dist,x]
A variety of distributions can be used in conjunction with the probability density function. These distributions include the normal
distribution and the binomial distribution.
Normal Distribution
For a normal distribution with a mean μ and a standard deviation σ, use the following syntax to specify the distribution.
NormalDistribution[μ,σ]
To calculate the probability of observing a value x given this distribution, use the probability density function as described previously.
PDF[NormalDistribution[μ,σ],x]
To calculate the probability of observing a range of values from x1 to x2 given this distribution, use the following syntax.
NIntegrate[PDF[NormalDistribution[μ,σ],x],{x,x1,x2]
For the theory behind the probability density function for a normal distribution, click here.
Binomial Distribution
For a binomial distribution with a number of trials n and a success probability p, use the following syntax to specify the distribution.
BinomialDistribution[n,p]
To calculate the probability of observing a successful trial exactly k times given this distribution, use the probability density function as
described previously.
PDF[BinomialDistribution[n,p],k]
To calculate the probability of observing between a range of k1 number of successes and k2 number of successes given this distribution, use
the following syntax.
NIntegrate[PDF[BinomialDistribution[n,p],k],{k,k1,k2]
Alternatively, binomial distribution can be found using a user defined function. Define the equations as follow: binom[nn_,kk_,pp_]:=
nn!/(kk! (nn-kk)!) pp^kk (1-p)^(nn-kk)
where:
number of independent samples = nn
number of events = kk
probability of the event = pp
To find the odds of getting 5 heads out of 10 coin tosses,assuming the probability of head = 0.5, substitute the values for nn,kk and pp as
following: binom[10,5,0.5]
Using the same scenario, but instead for the odds of getting 5 or more heads in the 10 tosses, the function can be used as:
Sum[binom[10,i,0.5],{i,5,10}]
For the theory behind the probability density function for a binomial distribution, click here.
2.9.1.13: MANIPULATE FUNCTION

Often in controls, it is useful to manipulate certain variables dynamically and see the result in a plot or table. This is particularly useful in
setting the constants in tuning PID controllers. The Manipulate function in Mathematica lets you do this. You can use the manipulate
function in conjunction with the Plot command. The general syntax for using the manipulate function is as follows:
Manipulate[expr,{u,umin,umax}]
Where expr is the function that you want to add controls to and u is the variable you wish to manipulate.
Manipulate a function:
You can change the variable x in the function s. Moving the bar lets you change the variable x.
Manipulate a plot:
You can also manipulate a plot. The slider bar lets change the variable n and the plot changes accordingly.
The Mathematica file used to create this example, can be downloaded below.
Manipulate Command Example
For more information regarding manipulate, please see the Wolfram's Reference Site.
[Note - need to add manipulate for the Bode plots - R.Z.]
2.9.1.14: ADDING GRIDLINES
2.9.1.15: PLOTTING LOG-LOG GRAPHS
2.9.1.16: OTHER USEFUL MATHEMATICA TIPS
[It would be good to replace this images with actual text - R.Z.]
2.9: Helpful Mathematica Syntax- Hints on how to use Mathematica to model chemical processes is shared under a CC BY 3.0 license and was authored,
CHAPTER OVERVIEW
3: SENSORS AND ACTUATORS

3.1: Control Systems - Measurement Devices
3.2: Temperature Sensors
3.3: Pressure Sensors
3.4: Level Sensors
3.5: Flow Sensors
3.6: Composition Sensors
3.7: pH and Viscosity Sensors
3.8: Miscellaneous Sensors
3: Sensors and Actuators is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source content that was
edited to conform to the style and standards of the LibreTexts platform; a detailed edit history is available upon request.
1
3.1: CONTROL SYSTEMS - MEASUREMENT DEVICES
3.1.1: CONTROL SYSTEMS: INDUSTRIAL APPLICATIONS
Many control systems are used today in a large number of industries consisting of applications from all kinds. The common factor of all
control types is to sustain a desired outcome that may change during a chemical reaction or process. The most common control type used
today in industry is a PID controller (proportional, integral, derivative), which allows the operator to apply different control techniques that
can be used to achieve different settings in an experiment or process. A PID controller can be used in two main control mechanisms that
include feed back and feed forward. The purpose of this article is to provide examples of common industrial control systems that apply
different control architectures.
3.1.2: TEMPERATURE CONTROL: THERMOCOUPLE

A thermocouple is a device to measure and control temperature within a system. They are used in a wide variety of industrial applications
(gas turbines, chemical reactors, exhaust, chemical manufacturing etc) due to their low cost and portability. The fundamental working
principle for thermocouple operations is the Seebeck Effect. Mostly, thermocouples operate in a P or a PID control mode. In order to
measure temperature between two points, the thermocouple employs two metallic ends (made from different alloys). When the two
conductor ends are exposed to a thermal gradient, they generate a voltage between them. This voltage drop gives rise to the temperature
measurement output that a thermocouple provides. Depending on the types of alloys in both conductor ends, and the magnitude of the
thermal gradient, thermocouples can measure temperature differences between 1-23000C. They can operate in feedback loops or feed
forward loops. Thermocouples are mostly digital control units. Thermocouple prices started around $100 and cost up to $2500 for more
accurate, and self-calibrated models.
3.1.3: PRESSURE CONTROL: PRESSURE SWITCH

A pressure switch is a device that controls systems against pressure drops or pressure spikes. The most basic types of pressure switches
work on an ON-OFF basis, but can also be manufactured to work in PID mode. The fundamental method of operation is to set the “Set-
Pressure” to a given quantity. This deactivates the pressure switch from the circuit connecting it to the control valve upstream. If, at any
point during the process, the pressure rises past the set-point, the switch is activated and completes the circuit, thus shutting off the control
valve. Pressure switches can be hydraulic, or pneumatic based (air-based pressure). One common application of a pressure switch in the
industry is to protect PD (positive-displacement) pumps from over-pressurization. A PD pump can generate very high pressures if not
controlled by a pressure switch; thus setting a pressure switch inline with a PD pump will prevent over pressurization since it shuts off the
control valve. Pressure switches are common in any industry since all require pressurization of certain components during manufacturing,
processing or refining stages. They sell between $200-$2000 depending on the magnitude of the set-point required for protection.
3.1.4: COMPOSITION CONTROL: RATIO CONTROL

A ratio controller is used to ensure that two or more process variables such as material flows are kept at the same ratio even if they are
changing in value. Control modes can be operated in different types, but mostly feedback PI controller is used for ratio control. In industrial
control processes, ratio control is used in the following processes: burner/air ratio, mixing and blending two liquids, injecting modifiers and
pigments into resins before molding or extrusion, adjusting heat input in proportion to material flow. The prices for industrial ratio
controllers start around $500 and increase depending on the sensitivity of the unit to different magnitudes of compositional changes and size
of the equipment to be annexed to.
3.1.5: LEVEL CONTROL: LEVEL SWITCHES
Level controls are used to monitor and regulate the liquid level in industrial vessels. There are many various sensors used in level control
including ultrasonic, lasers, and floatation sensors. They all work on the same general principle. A sensor measures the distance from the
base of the vessel to the top of the liquid level, mainly by using sonar waves or a laser beam. Based on the time it takes for the wave or for
the laser beam to return to the emitting source, the controller sends information to change or maintain the level. The magnetic float control
reads on a sensor located on the wall of the vessel and sends that information to the controller. Some examples of level control in industry
are: maintaining the liquid level in a distillation column, protecting from overflow, and measuring the amount of product in storage tanks.
Level sensors/controls vary in price based on the type and the accuracy required. Below is a picture of a magnetic float level control.
3.1.6: FLOW CONTROL: FLOW METERS

Flow controls are used to regulate the flow of a liquid or gas through a system. The main form of flow control is a valve. There are many
different types of valves, but they all change flow rates by opening or closing based on what amount is needed. A flow sensor reads the flow
rate, and a controller will operate the valve to increase or decrease the flow. Some basic types of flow sensors are rotameters and digital
mass meters. These controls are used in all forms of industry to control flows including water treatment, product measurement, and fuel
flow to furnaces. The low price end for flow sensors is typically around $500, and depending on the size of the process, and the process
material, the price can range into the thousands. Below are two pictures. The first is a manual control valve that can be opened or shut to
regulate flow. The second is a valve that can be used in an automatic control system to regulate flow.

Christian Hatfield, Varun Kaushik, Alon Mandel
3.1: Control Systems - Measurement Devices is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via
source content that was edited to conform to the style and standards of the LibreTexts platform; a detailed edit history is available upon request.
3.2: TEMPERATURE SENSORS
3.2.1: INTRODUCTION
Temperature sensors are vital to a variety of everyday products. For example, household ovens, refrigerators, and thermostats all rely on
temperature maintenance and control in order to function properly. Temperature control also has applications in chemical engineering.
Examples of this include maintaining the temperature of a chemical reactor at the ideal set-point, monitoring the temperature of a possible
runaway reaction to ensure the safety of employees, and maintaining the temperature of streams released to the environment to minimize
harmful environmental impact.
While temperature is generally sensed by humans as “hot”, “neutral”, or “cold”, chemical engineering requires precise, quantitative
measurements of temperature in order to accurately control a process. This is achieved through the use of temperature sensors, and
temperature regulators which process the signals they receive from sensors.
From a thermodynamics perspective, temperature changes as a function of the average energy of molecular movement. As heat is added to a
system, molecular motion increases and the system experiences an increase in temperature. It is difficult, however, to directly measure the
energy of molecular movement, so temperature sensors are generally designed to measure a property which changes in response to
temperature. The devices are then calibrated to traditional temperature scales using a standard (i.e. the boiling point of water at known
pressure). The following sections discuss the various types of sensors and regulators.
3.2.2: TEMPERATURE SENSORS

Temperature sensors are devices used to measure the temperature of a medium. There are 2 kinds on temperature sensors: 1) contact sensors
and 2) noncontact sensors. However, the 3 main types are thermometers, resistance temperature detectors, and thermocouples. All three of
these sensors measure a physical property (i.e. volume of a liquid, current through a wire), which changes as a function of temperature. In
addition to the 3 main types of temperature sensors, there are numerous other temperature sensors available for use.
Contact Sensors
Contact temperature sensors measure the temperature of the object to which the sensor is in contact by assuming or knowing that the two
(sensor and the object) are in thermal equilibrium, in other words, there is no heat flow between them.
Examples (further description of each example provide below)
Thermocouples
Resistance Temperature Detectors (RTDs)
Full System Thermometers
Bimetallic Thermometers
Noncontact Sensors
Most commercial and scientific noncontact temperature sensors measure the thermal radiant power of the Infrared or Optical radiation
received from a known or calculated area on its surface or volume within it.
An example of noncontact temperature sensors is a pyrometer, which is described into further detail at the bottom of this section.
3.2.2.1: THERMOMETERS
Thermometers are the most common temperature sensors encountered in simple, everyday measurements of temperature. Two examples of
thermometers are the Filled System and Bimetal thermometers.
Filled System Thermometer
The familiar liquid thermometer consistsof a liquid enclosed in a tube. The volume of the fluid changes as a function of temperature.
Increased molecular movement with increasing temperature causes the fluid to expand and move along calibrated markings on the side of
the tube. The fluid should have a relatively large thermal expansion coefficient so that small changes in temperature will result in detectable
changes in volume. A common tube material is glass and a common fluid is alcohol. Mercury used to be a more common fluid until its
toxicity was realized. Although the filled-system thermometer is the simplest and cheapest way to measure temperature, its accuracy is
limited by the calibration marks along the tube length. Because filled system thermometers are read visually and don’t produce electrical
signals, it is difficult to implement them in process controls that rely heavily on electrical and computerized control.
Bimetal Thermometer
In the bimetal thermometer, two metals (commonly steel and copper) with different thermal expansion coefficients are fixed to one another
with rivets or by welding. As the temperature of the strip increases, the metal with the higher thermal expansion coefficients expands to a
greater degree, causing stress in the materials and a deflection in the strip. The amount of this deflection is a function of temperature. The
temperature ranges for which these thermometers can be used is limited by the range over which the metals have significantly different
thermal expansion coefficients. Bimetallic strips are often wound into coils and placed in thermostats. The moving end of the strip is an
electrical contact, which transmits the temperature thermostat.
3.2.2.2: RESISTANCE TEMPERATURE DETECTORS

A second commonly used temperature sensor is the resistance temperature detector (RTD, also known as resistance thermometer). Unlike
filled system thermometers, the RTD provides an electrical means of temperature measurement, thus making it more convenient for use
with a computerized system. An RTD utilizes the relationship between electrical resistance and temperature, which may either be linear or
nonlinear. RTDs are traditionally used for their high accuracy and precision. However, at high temperatures (above 700°C) they become
very inaccurate due to degradation of the outer sheath, which contains the thermometer. Therefore, RTD usage is preferred at lower
temperature ranges, where they are the most accurate.
There are two main types of RTDs, the traditional RTD and the thermistor. Traditional RTDs use metallic sensing elements that result in a
linear relationship between temperature and resistance. As the temperature of the metal increases, increased random molecular movement
impedes the flow of electrons. The increased resistance is measured as a reduced current through the metal for a fixed voltage applied. The
thermistor uses a semiconductor sensor, which gives a power function relationship between temperature and resistance.
RTD Structure
A schematic diagram of a typical RTD is shown in Figure 1.
Figure 1. Schematic Diagram of Resistance Temperature Structure

As shown in Figure 1, the RTD contains an outer sheath to prevent contamination from the surrounding medium. Ideally, this sheath is
composed of material that efficiently conducts heat to the resistor, but resists degradation from heat or the surrounding medium.
The resistance sensor itself is responsible for the temperature measurement, as shown in the diagram. Sensors are most commonly
composed of metals, such as platinum, nickel, or copper. The material chosen for the sensor determines the range of temperatures in which
the RTD could be used. For example, platinum sensors, the most common type of resistor, have a range of approximately -200°C – 800°C.
(A sample of the temperature ranges and resistances for the most common resistor metals is shown in Table 1). Connected to the sensor are
two insulated connection leads. These leads continue to complete the resistor circuit.
Table 1. Common Metal Temperature and Resistance Ranges
Element Metal Temperature Range Base Resistance TCR(Ω/Ω/°C)
Copper -100 – 260 °C 10 Ω at 0 °C 0.00427
Nickel -100 – 260 °C 120 Ω at 0 °C 0.00672
Platinum -260 – 800 °C 100 Ω at 0 °C 0.003916
There are 4 major categories of RTD sensors. There are carbon resistors, film thermometers, wire-wound thermometers and coil elements.
Carbon resisters are the most commonly used. They are inexpensive and are accurate for low temperatures. They also are not affected by
hysteresis or strain gauge effects. They are commonly used by researchers.
Film thermometers have a very thin layer of metal, often platinum, on a plate. This layer is very small, on the micrometer scale. These
thermometers have different strain gauge effects based on what the metal and plate are composed of. There are also stability problems
that are dependent on the components used.
In wire-wound thermometers the coil gives stability to the measurement. A larger diameter of the coil adds stability, but it also increases
the amount the wire can expand which increases strain and drift. They have very good accuracy over a large temperature range.
Coil elements are similar to wire-wound thermometers and have generally replaced them in all industrial applications. The coil is
allowed to expand over large temperature ranges while still giving support. This allows for a large temperature range while decreasing
the drift.
RTD Operation
Most traditional RTD operation is based upon a linear relationship between resistance and temperature, where the resistance increases with
temperature. For this reason, most RTDs are made of platinum, which is linear over a greater range of temperatures and is resistant to
corrosion. However, when determining a resistor material, factors such as temperature range, temperature sensitivity, response time, and
durability should all be taken into consideration. Different materials have different ranges for each of these characteristics.
The principle behind RTDs is based upon the Callendar – Van Dusen equation shown in Equation 3.2.1 , which relates the electrical
resistance to the temperature in °C. This equation is merely a generic polynomial that takes form based upon experimental data from the
specific RTD. This equation usually takes on a linear form since the coefficients of the higher-order variables (a2, a3, etc.) are relatively
small.
2 3 4 n
RT = R0 (1 + a1 T + a2 T + a3 T + a4 T + … + an T ) (3.2.1)
RT: Resistance at temperature T, in ohms

R0: Resistance at temperature = 0°C, in ohms
an: Material’s resistance constant, in °Cn − 1
Another type of RTD is the thermistor, which operates based upon an exponential relationship between electrical resistance and temperature.
Thermistors are primarily composed of semiconductors, and are usually used as fuses, or current-limiting devices. Thermistors have high
thermal sensitivity but low temperature measuring ranges and are extremely non-linear. Instead of the Callendar - Van Dusen equation, the
thermistor operates based upon the nonlinear equation, equation (2), shown in degrees K.
1 1
RT = R0 exp(b ( − )) (3.2.2)
T T0
T0: Initial temperature, usually set at 298K

b: Material's temperature coefficient of resistance, in K
Errors associated with resistance thermometers will occur due to the individual or collective efforts of: defective insulation, contamination
of the resistor, or insecure lead wire connections.
3.2.2.3: THERMOCOUPLES
Another temperature sensor often used in industry is the thermocouple. Among the various temperature sensors available, the thermocouple
is the most widely used sensor. Similar to the RTD, the thermocouple provides an electrical measurement of temperature.
Thermocouple Structure
The thermocouple has a long, slender, rod-like shape, which allows it to be conveniently placed in small, tight places that would otherwise
be difficult to reach. A schematic diagram of a typical thermocouple is shown in Figure 2.
Figure 2. Schematic Diagram of Thermocouple Structure

As illustrated in Figure 2, the thermocouple contains an outer sheath, or thermowell. The thermowell protects the contents of the
thermocouple from mechanical and chemical damage.
Within the thermowell lies two metal wires each consisting of different metals. Various combinations of materials are possible for these
metal wires. Three common thermocouple material combinations used for moderate temperature measurements are the Platinum-Rhodium,
Iron-Constantan, and Chromel-Alumel metal alloys. The metal alloys chosen for a thermocouple is based upon the emf value of the alloy
pair at a given temperature. Sample emf values for the most common materials at various temperatures are shown in Table 2. For a given
pair of materials, the two wires are connected at one end to form a junction. At the other end, the two wires are connected to a voltage
measuring device. These ends of the wires are held at a different reference temperature.
Table 2. Common Metal Temperature and Emf Values
Alloy Type Emf Value at 20 °C Emf Value at 50 °C Emf Value at 100 °C
Platinum-Rhodium 0.113 mV 0.299 mV 0.646 mV
Iron-Constantan 1.019 mV 2.585 mV 5.269 mV
Chromel-Alumel 0.798 mV 2.023 mV 4.096 mV
Various methods are used to maintain the reference temperature at a known, constant temperature. One method consists of placement of the
reference junction within either an ice bath or oven maintained at a constant temperature. More commonly, the reference temperature is
maintained electronically. Though not as stable as an ice bath, electronically controlled reference temperatures are more convenient for use.
Reference temperatures could also be maintained through temperature compensation and zone boxes, which are regions of uniform
temperature. The voltage difference across the reference junction is measured and sent to a computer, which then calculates the temperature
with this data.
3.2.2.3.1: THERMOCOUPLE OPERATION
The main principle upon which the thermocouple function is based on is the difference in the conductivities of the two wire materials that
the thermocouple is made of, at a given temperature. This conductivity difference increases at higher temperatures and conversely, the
conductivity difference decreases at lower temperatures. This disparity results in the thermocouples being more efficient and useful at
higher temperatures. Since the conductivity difference is small at lower temperatures and thus more difficult to detect, they are inefficient
and highly unreliable at low temperatures.
The conductivity difference between the two wires, along with a temperature difference between the two junctions, creates an electrical
current that flows through the thermocouple. The first junction point, which is the point at which the two wires are connected, is placed
within the medium whose temperature is being measured. The second junction point is constantly held at a known reference temperature.
When the temperature of the medium differs from the reference temperature, a current flows through the circuit. The strength of this current
is based upon the temperature of the medium, the reference temperature, and the materials of the metal wires. Since the reference
temperature and materials are known, the temperature of the medium can be determined from the current strength.
Error associated with the thermocouple occurs at lower temperatures due to the difficulty in detecting a difference in conductivities.
Therefore, thermocouples are more commonly used at higher temperatures (above -125°C) because it is easier to detect differences in
conductivities. Thermocouples are operable over a wide range of temperatures, from -200°C to 2320°C, which indicates its robustness and
vast applications. Thermocouples operate over this wide range of temperatures, without needing a battery as a power source. It should be
noted that, the wire insulation might wear out over time by heavy use, thus requiring periodical checks and maintenance to preserve the
accuracy of the thermocouple.
To determine the temperature of the medium from the current strength, the emf or voltage values of the current and of the wire materials at
the reference temperatures must be known. Often, the measured temperature can be found by using standard thermocouple tables. However,
these tables are often referenced at 0°C. To correct for this different reference temperature, equation (3) can be used to calculate the
temperature from a given current.
ξT1 ,T3 = ξT1 ,T2 + ξT2 ,T3 (3.2.3)
: emf of an alloy combination generated at two different temperatures

T1: temperature of the medium whose temperature is to be determined
T2: reference temperature of the thermocouple
T3: reference temperature of the standard thermocouple table, which in this case is 0°C
Once the emf between two alloys is calculated relative to a reference temperature when T is 0°C, the standard thermocouple table can be
3
used to determine the temperature T1 of the medium. This temperature is usually automatically displayed on the thermocouple.
Apart from the common occurrence of the thermocouples being placed in the fluid to measure temperature change, thermocouples can be
also embedded in solids with excellent results. This is highly effective while establishing the different thermal properties for a solid. The
heat transfer to the thermocouple will now be in the form of conductive heat transfer. As a result, this setup would be very similar to heat
conduction in series, since the thermocouple is almost always made from a different material then the actual solid. Such discrepancies
depend on the manner in which the thermocouple is embedded in the solid and should be taken into account when the thermal properties are
being calculated and analyzed. One example is shown in the photo below.
Figure 3. Schematic diagram of how the thermocouple function
Laws for thermocouples
Law of homogenous material: If all the wires and the thermocouple are made of the same material, temperature changes in the wiring do
not affect the output voltage. Thus, need different materials to adequately reflect the temperature.
Law of intermediate materials: The sum of all the thermoelectric forces in a circuit with a number of dissimilar materials at a uniform
temperature is zero. This implies that if a third material is added at the same temperature, no net voltage is generated by the new
material.
Law of successive or intermediate temperatures: If two dissimilar homogeneous materials produce thermal emf1 when the junctions are
at T1 and T2 and produce thermal emf2 when the junctions are at T2 and T3 , the emf generated when the junctions are at T1 and T3
will be emf1 + emf2 .
Application
Steel industry: Monitor temperature and chemistry throughout the steel making process
Heating appliance safety: Thermocouples in fail-safe mode are used in ovens and water heaters to detect if pilot flame is burning to
prevent fire and health hazard
Manufacturing: Used for testing prototype electrical and mechanical apparatus
Process plants: Chemical production plants and refineries use computer programs to view the temperature at various locations. For this
situation, a number of thermocouple leads are brought to a common reference block.
3.2.2.4: PYROMETERS
Unlike the thermometer, RTD and the thermocouple, pyrometers (non-contact temperature sensors) measures the amount of heat radiated,
rather than the amount of heat conducted and convected to the sensor. Various types of pyrometers, such as total radiation and photoelectric
pyrometers, exist. Below is a schematic of an optical pyrometer in Figure 4.
Figure 4. Schematic diagram of an optical pyrometer

These pyrometers differ in the type of radiation they measure. There are many factors that influence the amount of radiated heat detected,
thus there are many assumptions that must be made regarding the emissivity, or the measure of the manner in which heat is radiated, of the
object. These assumptions are based upon the manner in which heat is radiated as well as the geometry of the object. Because temperature is
dependent on the emissivity of a body, these assumptions regarding the emissivity introduce uncertainties and inaccuracies in the
temperature readings. Therefore, because of the error associated with them, pyrometers are not often used in industry.
Table 3. Summary of Temperature Sensors
There are a few different types of pyrometers. There are optical and radiation pyrometers.
How Optical Pyrometers Work:
Compares the color of visible light given off by the object with that of a electrically heated wire
The wire can be preset to a certain temperature
The wire can be manually adjusted to compare the two objects
How Radiation Pyrometers Work:
This sensor works by measuring the radiation (infrared or visible light) that an object gives off
The radiation heats a thermocouple in the pyrometer which in turn induces a current
The larger the current induced, the higher the temperature is
Pyrometers are usually used at very high temperatures, but can be used at colder temperatures as well. There are lots of industrial
applications to pyrometers. Plant operators can use pyrometers to get a sense of what temperature certain processes are running at. The
downside to pyrometers is that they are not very accurate as thermocouples or RTD sensors are. This is because they rely on quantifying
colors of light.
3.2.3: TEMPERATURE REGULATORS

Temperature regulators, also known as temperature control valves (TCVs), physically control, as well as measure, temperature. Temperature
regulators are not capable of directly maintaining a set value; instead, they relate the load (in this case the valve opening) with the control
(temperature measurement). These regulators are most useful when temperature is correlated to a flow of a substance. For example, a TCV
may be used to control the temperature of an exothermic reaction that requires constant cooling. The TCV measures the temperature of the
reaction and, based upon this temperature, either increases or decreases the flowrate of cooling fluid to adjust the temperature of the
reaction. Similarly, the regulator could be used to adjust the flow amount of steam, which is typically used to heat a substance. Therefore, by
adjusting flowrate, the regulator can indirectly adjust temperature of a given medium.
3.2.3.1: REGULATOR STRUCTURE

The structure of a typical thermal regulator consists of four main parts, as shown in Figure 3. The temperature detecting element, which in
most cases is a temperature sensor, as described above, sends either an electrical or mechanical signal through the connector to the actuator.
The actuator then uses this signal to act upon the power source, which determines the position of the valve. (This will be further described in
the next section.)
Figure 3. Schematic Diagram of Temperature Regulator Structure. Note: The actuator consists of the capillary tubing, the power source, and
the adjustment.
3.2.3.2: REGULATOR OPERATION

The temperature regulator operates based upon a mechanical means of temperature control. As previously mentioned, the bulb of the
regulator is typically filled with a heat conducting substance. Due to the thermal expansion properties of this substance, the substance
expands as the temperature increases. This expansion causes a change in the pressure of the actuator, which correlates to the temperature of
the medium. This pressure change repositions a valve on the regulator, which controls the flowrate of a coolant. The temperature of the
medium is then altered by the change in the flowrate of this coolant.
3.2.3.3: TYPES OF TEMPERATURE REGULATORS

Though all regulators have the same basic build and purpose, they exist in a variety of forms. In particular, these regulators vary in four
primary ways: temperature detecting elements, temperature detector placement, actuator type, and valve type.
3.2.3.3.1: TEMPERATURE DETECTING ELEMENTS
Most temperature regulation systems use thermocouples or RTDs as temperature sensing devices. (Described above) For these systems, the
connector is a computer. The sensors send an electric signal to the computer, which calculates the temperature. The computer then compares
the temperature measured by the sensor to a programmed set-point temperature, thus determining the required pressure in the actuator. The
pressure in the actuator changes position of the power source (diaphragm or bellows), which consequently changes the flowrate through the
valve.
Some temperature regulation systems use a filled bulb as a temperature sensor. Based on the thermal expansion properties of the material
within the bulb, the material expands as the temperature increases. This expansion causes a change in the pressure of the actuator. The
pressure change actuator then repositions power source. Again, the change in the power source changes the flowrate through the valve.
Temperature regulation systems using thermocouples or RTDs as temperature sensing devices are much more common than regulation
systems using filled bulbs.
Temperature Detector Placement: Internal and Remote Detection
Temperature detection can be done with internal or remote elements. For internal temperature detectors, the thermal actuator and
temperature detector are located entirely within the valve. For remote temperature detectors, the primary temperature detecting element is
separate from the actuator and valve, and is connected to the actuator with either electrical wiring or capillary tubing, depending on the
mechanism of the temperature sensor. Remote temperature detectors are more common, as internal temperature detectors are limited in use.
Internal temperature detectors can only measure the temperature of the fluid flowing through the valve and not the temperature of the
process.
Actuator Type: Thermal Systems
There are four main categories of thermal actuators used in temperature regulators. Thermal actuators produce power and work,
proportional to the measured temperature of the process, on the power source. Actuator types include the vapor-filled system, the liquid-
filled system, the hot chamber system, and the fusion-type or wax-filled system. Of all the thermal systems mentioned, liquid-filled systems
are the most common, because they relate temperature and pressure change in a linear fashion.
Vapor-Filled Systems
In the vapor-filled system, the thermal actuator is partially filled with a volatile liquid. As the temperature of the sensor increases, the vapor
pressure of the liquid also increases. This increases the pressure on the power source, and adjusts the flowrate through the valve.
Liquid-Filled Systems
In liquid-filled systems, the thermal actuator is filled with a chemically stable liquid, such as a hydrocarbon. As the temperature increases,
the liquid expands, which produces a force on the power source.
Hot Chamber Systems
In hot chamber systems, the thermal actuator is partially filled with a volatile fluid. An increase in temperature of the system forces some of
this fluid into the power unit, where the heat of the unit causes this liquid to turn into a superheated vapor. The pressure increase produces a
force on the power source.
Fusion-Type (Wax-Filled) Systems
Of all the systems mentioned, the fusion-type system is the least common. In the fusion-type system, the thermal actuator is filled with
special waxes such as hydrocarbons, silicones, and natural waxes. The wax contains large amounts of copper, which increases the heat-
transfer quality of the wax. As temperature increases, the wax expands, producing a force that repositions the power source.
Valve Type: Direct and Pilot Actuated
The two main types of valves used in thermal regulators are the direct and pilot actuated valves. In all such thermal regulators, there is a
power source (such as bellows and diaphragms) that provides the force required to reposition the valve to control the temperature. These
power sources rely on a change in the pressure of the actuator in order to properly regulate temperature. In direct-actuated TCVs, this power
unit is directly connected to the valve, which provides the force required to open and close the valve. In pilot-actuated TCVs, the thermal
actuator moves a pilot valve, which then transfers energy in the form of pressure to a piston, which then provides the work necessary to
reposition the main valve.
Direct-actuated TCVs are often much simpler in structure than pilot-actuated TCVs, and therefore they are also much cheaper. In addition,
they respond better to smaller changes in temperature and more accurately reflect the temperature of the medium. Thus, if the exact
temperature of the system is essential to ensure correct operation, a direct-actuated TCV should be used. Pilot-actuated TCVs usually have
much smaller temperature sensing devices, a faster response time, and the ability withstand much higher pressures through the regulating
valve. Therefore, at high pressures or rapid temperature changes, a pilot-actuated TCV should be used.
3.2.4: RESISTANCE TEMPERATURE DETECTOR EXAMPLE

Note: This example problem has been made up for demonstration purposes.
A newly hired chemical engineer at Hypothetical Industries is responsible for monitoring and maintaining temperatures for one of the
company’s exothermic reactions. The process uses a platinum resistor thermometer to measure the temperature of the process, the properties
of which are given below. The reaction’s ideal range is between 250°C – 350°C. Below 250°C, the catalyst ceases to function, and above
350 °C, it can be classified as a runaway reaction. The engineer can control steam and cooling water to regulate the process temperature.
Describe what changes the engineer should make to either the flowrate of the coolant or the steam to keep the system functioning optimally.
Case I: RT = 25ohms
Case II: RT = 13.9ohms
Case III: RT = 19.4ohms
GIVEN DATA:
RT = R0(1 + a1T + a2T2)
R0 = 10 ohms
a1 = 3.91x10 − 3(°C)
a2 = − 6.72x10 − 8(°C − 2)
Solution:
Substitute the given value of RT for each of the 3 cases into the mathematical calculations below to get the answers, we have worked out
Case I with the corresponding numbers below.
Case I:
1. Factor R0 throughout the right side of the equation.
2. Subtract RT from both sides of the equation.
3. Solve for T using the quadratic equation.
ALTERNATE Solution:
1. Since the constant a2 is so small (x10 − 7) we can just neglect that term.
2. Now the problem is just a simple linear equation, which can be solved by isolating T on one side of the equation.
AnswerS:
Case I. The engineer should start flowing cooling water into the reaction process because the temperature is ~ 500°C which is
above the appropriate range.
Case II. The engineer should increase steam feed to the reaction process because the temperature is ~ 125°C which is below
the appropriate range.
Case III. The engineer does not have to do anything because the temperature is in the appropriate range ~ 300°C.
 EXAMPLE 3.2.1: TEMPERATURE SENSOR
You are a chemical engineer at Hypothetical Industries and you are responsible for monitoring and regulating the temperature for one of
the company’s reactions. Determine which temperature sensor should be used to optimally measure and regulate the reaction’s
temperature in each of the following situations.
Case I. T=900°C
Case II. T=500°C but sensor location in a large continuous reactor makes it difficult to repair
Case III. T=50°C and you are estimating the current temperature of a lab scale reaction
Solution
Case I. We know that the reaction occurs at 900°C. Taking this optimal temperature as the only necessary parameters, the temperature
sensor would be a thermocouple. The temperature is outside of the range for resistance thermometers. Regular thermometers do not
send electric signals; therefore they can not be used for this process.
Case II. Although this temperature is within the operating ranges of both thermocouples and RTDs, since the sensor may not be readily
accesible for repair, we must use an RTD because thermocouples lose accuracy after extended usage.
Case III. Since the temperature is within the range of all three sensors, and we only need a rough estimate of temperature, we can use a
thermometer which will be much less costly than the alternative sensors.
3.2.5: REFERENCES
Dougherty, E.P. ,Temperature Control Principles for Process Engineers, Hanser Publishers, 1993 ISBN 0344615980
Liptak,B.G. ,Instrument Engineer's Handbook, 4th Edition,Volume 2. Taylor and Francis Group,2005 ISBN 0849310814
Perry,R.H. ,Perry's Chemical Engineer's Handbook, 7th Edition, McGraw-Hill,1997 ISBN 0070498415
Richardson,J.F. and Peacock,D.G. ,Chemical Engineering, 3rd Edition,Volume 3. Butterworth-Heinemann,1994 ISBN 0080410030
Moffat,R.J. ,Notes on Using Thermocouples, 1997
Error! Hyperlink reference not valid.
Pyrosales
Cole parmer Suppliers
Temperature.com
Authors: (14 September 2006) Ardemis Boghossian, James Brown, Sara Zak
Stewards: (11 September 2007) Arthur L. Edge III, Kathryn Meintel, Renu Rao, Kaveh Saba
3.2: Temperature Sensors is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source content that was
3.3: PRESSURE SENSORS
3.3.1: INTRODUCTION
Pressure must be considered when designing many chemical processes. Pressure is defined as force per unit area and is measured in English
units of psi and SI units of Pa. There are three types of pressure measurements:
1. Absolute pressure - atomospheric pressure plus gauge pressure.
2. Gauge Pressure - absolute pressure minus atmospheric pressure.
3. Differential Pressure - pressure difference between two locations.
There are various types of pressure sensors that are available in the market today for use in industry. Each functions best in a certain type of
situation.
3.3.2: SENSOR SELECTION CRITERIA

In order for a pressure controlled system to function properly and cost-effectively, it is important that the pressure sensor used be able to
give accurate and precise readings as needed for a long period of time without need for maintenance or replacement while enduring the
conditions of the system. Several factors influence the suitability of a particular pressure sensor for a given process: the characteristics of the
substances being used or formed during the process, the environmental conditions of the system, the pressure range of the process, and the
level of precision and sensitivity required in measurements made.
3.3.2.1: PROCESS
The pressure sensing element (elastic element) will be exposed to the materials used in the process, therefore materials which might react
with the process substances or degrade in corrosive media are unsuitable for use in the sensor. Diaphragms are optimal for very harsh
environments.
3.3.2.2: ENVIRONMENT
The environment (as in the system -- pipes, vibrations, temperature, etc.) in which the process is carried out also needs to be considered
when choosing a suitable pressure sensor. Corrosive environments, heavy vibrations in the piping units, or extreme temperatures would
mean that the sensors would have to have an added level of protection. Sealed, strong casing materials with interior liquid containing
glycerine or silicone are often used to encase the internal components of the sensor (not including the sensing element), protecting them
from very harsh, corrosive environments or frequent vibrations.
3.3.2.3: PRESSURE RANGE

Most processes operate within a certain pressure range. Because different pressure sensors work optimally in different pressure ranges, there
is a need to choose pressure gauges which are able to function well in the range dictated by the process.
3.3.2.4: SENSITIVITY
Different processes require different levels of precision and accuracy. In general, the more precise the sensor, the more expensive it is, thus
it is economically viable to choose sensors that are able to satisfy the precision desired. There is also a compromise between precision and
the ability to detect pressure changes quickly in sensors, hence in processes in which the pressure is highly variable over short periods of
time, it is unadvisable to use sensors which take a lot of time to give accurate pressure readings, although they might be able to give precise
pressure values.
3.3.3: PRESSURE MEASURING METHODS

Several pressure measuring methods have been developed and utilized; these methods include visual inspection of the height of liquid in a
column, elastic distortion, and electrical methods.
3.3.3.1: HEIGHT OF LIQUID IN COLUMN
The height of a liquid with known density is used to measure pressure. Using the equation P = ρgh, the gauge pressure can be easily
calculated. These types of pressure measuring devices are usually called manometers. Units of length may be used to measure the height of
the liquid in the column as well as calibrated pressure units. Typically water or mercury is used as the liquid within these columns. Water is
used when you desire greater sensitivity (its density is much less than liquid mercury, so its height will vary more with a pressure change).
Mercury is used when you desire higher pressure measurements and not as great sensitivity.
3.3.3.2: ELASTIC DISTORTION

This pressure measuring method is based on the idea that deformation of an elastic material is directly proportional to the pressure being
measured. There are mainly three sensor types that are used in this method of measuring pressure: Bourdon-tubes, diaphragms and bellows.
(refer to “Types of Sensors” Section)
3.3.3.3: ELECTRICAL METHODS
Electrical methods used for measuring pressure utilize the idea that dimensional changes in a wire affect the electrical resistance to the
conductor. These devices that use the change in resistance of the wire are called strain gauges. Other electrical sensors include capacitive
sensors, inductive pressure transducers, reluctive pressure transducers, potentiometric transducers, piezoresistive transducers and
piezoelectric sensors. (refer to “Types of Sensors” Section)
3.3.4: TYPES OF SENSORS

There are many different pressure sensors to choose from when considering which is most suitable for a given process, but they can
generally be placed into a few categories, namely elastic sensors, electrical transducers, differential pressure cells and vacuum pressure
sensors. Listed below each general category are specific internal components, each functioning best in a certain situation.
3.3.4.1: ELASTIC SENSORS

Most fluid pressure sensors are of the elastic type, where the fluid is enclosed in a small compartment with at least one elastic wall. The
pressure reading is thereby determined by measuring the deflection of this elastic wall, resulting in either a direct readout through suitable
linkages, or a transduced electrical signal. Elastic pressure sensors are sensitive; they are commonly fragile and susceptible to vibration,
however. In addition, they tend to be much more expensive than manometers, and are therefore preferentially used for transmitting
measured data and measuring pressure differences. A wide variety of flexible elements could conceivably be used for elastic pressure
sensors; the majority of devices use some form of a Bourdon tube, bellows, or diaphragm.
3.3.4.1.1: BOURDON TUBE GAUGES
The principle behind all Bourdon tubes is that an increase in pressure on the inside of the tube in comparison to the outside pressure causes
the oval or flat shaped cross-section of the tube to try to achieve a circular shape. This phenomenon causes the tube to either straighten itself
out in the c-type or spiral cases or to unwind itself for the twisted and helical varieties. This change can then be measured with an analog or
digital meter connected to the tube. Tube materials can be changed accordingly to suit the required process conditions. Bourdon tubes can
operate under a pressure range from 0.1-700 MPa. They are also portable and require little maintenance; however,they can only be used for
static measurements and have low accuracy.
Types of Bourdon tubes include C-type, spiral (a more coiled C-type tube), helical and straight tube Bourdon tubes. C-type gauges can be
used in pressures approaching 700MPa; they do have a minimum recommended pressure range, though -- 30 kPa (i.e., it is not sensitive
enough for pressure differences less than 30 kPa).
3.3.4.1.2: BELLOWS
Bellows elements are cylindrical in shape and contain many folds. They deform in the axial direction (compression or expansion) with
changes in pressure. The pressure that needs to be measured is applied to one side of the bellows (either inside or outside) while
atmospheric pressure is on the opposite side. Absolute pressure can be measured by evacuating either the exterior or interior space of the
bellows and then measuring the pressure at the opposite side. Bellows can only be connected to an on/off switch or potientiomenter and are
used at low pressures, <0.2 MPa with a sensitivity of 0.0012 MPa.
3.3.4.1.3: DIAPHRAGMS
Diaphragm elements are made of circular metal discs or flexible elements such as rubber, plastic or leather. The material from which the
diaphragm is made depends on whether it takes advantage of the elastic nature of the material, or is opposed by another element (such as a
spring). Diaphragms made of metal discs utilize elastic characteristics, while those made of flexible elements are opposed by another elastic
element. These diaphragm sensors are very sensitive to rapid pressure changes. The metal type can measure a maximum pressure of
approximately 7 MPa, while the elastic type is used for measuring extremely low pressures (.1 kPa - 2.2 MPa) when connected to
capacitative transducers or differential pressure sensors. Examples of diaphragms include flat, corrugated and capsule diaphragms. As
previously noted, diaphragms are very sensitive (0.01 MPa) . They can measure fractional pressure differences over a very minute range
(say, inches of water) (elastic type) or large pressure differences (approaching a maximum range of 207 kPa) (metal type).
Diaphragm elements are very versatile -- they are commonly used in very corrosive environments or with extreme over-pressure situations.
Examples of these elastic element pressure sensors are shown here.
3.3.4.2: ELECTRIC SENSORS
Sensors today are not neccessarily only connected to a gauge meter needle pointer to indicate pressure, but may also serve to convert the
process pressure into an electrical or pneumatic signal, which can be transmitted to a control room from which the pressure reading is
determined. Electric sensors take the given mechanics of an elastic sensor and incorporate an electrical component, thus heightening the
sensitivity and increasing the amount of instances in which you could utilize the sensor. The types of pressure transducers are capacitive,
inductive, reluctive, piezoelectric, strain gauge, vibrating element, and potentiometric.
Capacitive
A capacitive sensor consists of a parallel plate capacitors coupled with a diaphragm that is usually metal and exposed to the process pressure
on one side and the reference pressure on the other side. Electrodes are attached to the diaphragm and are charged by a high frequency
oscillator. The electrodes sense any movement of the diaphragm and this changes the capacitance. The change of the capacitance is detected
by an attached circuit which then outputs a voltage according to the pressure change. This type of sensor can be operated in the range of 2.5
Pa - 70MPa with a sensitivity of 0.07 MPa.
An example of a capacitive pressure sensor is shown on the right.
Inductive
Inductive pressure sensors are coupled with a diaphragm or a Bourdon tube. A ferromagnetic core is attached to the elastic element and has
a primary and 2 secondary windings. A current is charged to the primary winding. When the core is centered then the same voltage will be
induced to the two secondary windings. When the core moves with a pressure change, the voltage ratio between the two secondary windings
changes. The difference between the voltages is proportional to the change in pressure.
An example of an inductive pressure sensor utilizing a diaphragm is shown below. For this kind of pressure sensor, taking Chamber 1 as the
reference chamber with a reference pressure P1 coming into the chamber and the coil being charged with a reference current. When the
pressure in the other chamber changes, the diagphragm moves and induces a current in the other coil, which is measured and gives a
measure of the change in pressure.
These may be used with any elastic element (though, it is typically coupled with a diaphragm or a bourdon tube). The pressure reading
generated will be determined by voltage calibration. Thus, the range of pressure in which this sensor may be used is determined by an
associated elastic element but falls in the range of 250 Pa - 70 MPa .
Reluctive
Reluctive pressure sensors also charge a ferromagnetic core. When the pressure changes, the flexible element moves a ferromagnetic plate,
leading to a change in the magnetic flux of the circuit which can be measured. The situations in which one would use a reluctive electric
element is one in which the inductive sensor does not generate a precise enough measurement. The pressure range is 250 Pa - 70 MPa with a
sensitivity of 0.35 MPa.
An example of a reluctive pressure sensor can be seen on the right.

Piezoelectric
Piezoelectric sensors use a crystal sensor. When pressure is applied to the crystal, it deforms and a small electric charge is created. The
measurement of the electric charge corresponds to the change in pressure. This type of sensor has a very rapid response time to constant
pressure changes. Similar to reluctive electric element, the piezoelectric element is very sensitive, but responds much, much faster. Thus, if
time is of the essence, a piezoelectric sensor would be desired. The pressure range is .021 - 100 MPa with a sensitivity of 0.1 MPa.
On the right is an example of a piezoelectric pressure sensor.
Potentiometric
Potentiometric sensors have an arm mechanically attached to the elastic pressure sensing element. When pressure changes, the elastic
element deforms, causing the arm to move backwards or forwards across a potentiometer and a resistance measurement is taken. These
sensing elements do posess an optimum working range, but are seemingly limited in their resolution by many factors. As such, they are low
end sensors that aren't used for much. With a low sensitivity and working range, they may be best suited as a cheap detector evaluating a
coarse process.The pressure range is 0.035 - 70 MPa with a sensitivity of 0.07 -0.35 MPa.
An example of a potentiometric pressure sensor is shown on the right.
Strain Gauge
The strain gauge detects changes in pressure by measuring the change in resistance of a Wheatstone bridge circuit. In general, this circuit is
used to determine an unknown electrical resistance by balancing two sections of a bridge circuit such that the ratio of resistances in one
section ( ) is the same as that in the other section( ), resulting in a zero reading in the galvanometer in the center branch. One of the
sections contains the unknown component of which the resistance is to be determined, while the other section contains a resistor of known
resistance that can be varied. The Wheatstone bridge circuit is shown below:
The strain gauge places sensors on each of the resistors and measures the change in resistance of each individual resistor due to a change in
pressure. Resistance is governed by the equation where ρ = resistivity of the wire, L = length of the wire, and A = cross-sectional
area of the wire. A pressure change would either elongate or compress the wire, hence a compression sensor is needed on one resistor and
an elongation sensor on the other. To control the effects of temperature (a wire would also either elongate or compress with a change in
temperature), a blank sensor would be placed on the remaining two resistors. These gauges are frequently a type of semiconductor (N-type
or P-type). Thus, their sensitivity is much greater than their metal counterparts; however, with greater sensitivity comes a more narrow
functional range: the temperature must remain constant to obtain a valid reading. These gauges are affected greatly by variations in
temperature (unlike the other types of electrical components). The pressure range is 0 - 1400 MPa with a sensitivity of 1.4 - 3.5MPa.
An example of an unbonded strain gauge is shown below. This makes use of strain-sensitive wires one end fixed to an immobile frame and
the other end attached to a movable element, which moves with a change in pressure.
An example of a bonded strain gauge can be seen below. This is placed on top of a diaphragm, which deforms with change in pressure,
straining the wires attached to the diaphragm.
Vibrating Element
Vibrating element pressure sensors function by measuring a change in resonant frequency of a vibrating element. A current is passed
through a wire which induces an electromotive force within the wire. The force is then amplified and causes oscillation of the wire. Pressure
affects this mechanism by affecting the wire itself: an increase in pressure decreases the tension within the wire and thus lowers the angular
frequency of oscillation of the wire. The sensor is housed in a cylinder under vacuum when measuring absolute pressures. These absolute
pressure measuring sensors are very efficient: they produce repeatable results and are not affected by temperature greatly. They lack
sensitivity in meausurement, though, so they would not be ideal for a process in which minute pressures need monitoring. The pressure
range is 0.0035 - 0.3 MPa with a sensitivity of 1E-5 MPa.
A vibrating wire pressure sensor is shown below.
A vibrating cylinder pressure sensor (for absolute pressures) is shown below.
3.3.4.3: DIFFERENTIAL PRESSURE CELLS

Differential pressure cells are used with various kinds of sensors in which measurement of pressure is a result of a pressure differential such
as orifice plates, flow nozzles, or venturi meters. The differential pressure cell converts the pressure differential into a transmittable signal.
Where the differential pressure (DP) cell is placed depends on the nature of the fluid stream that is being measured. A typical DP cell is
minimally invasive (an external component attached across the point of measurement); it is commonly employed with a capacitive element
paired with a diaphragm that allows the capacitive body to separate or move together, generating a signal (via change in capacitance) that
can be interpreted to a pressure drop. They are often used to detect small differences in large pressure drops. Its placement is similar to
connecting a voltmeter in parallel to a resistor to measure its voltage "drop" (analgous to the pressure drop).
An example of a differential pressure cell using bellows can be found here: http://www.tpub.com/fluid/ch2s.h49.gif
The range of pressure measured and sensitivity of a differential cell depends upon the electric and elastic components used in the cell itself.
It is a great sensor to use when measuring a pressure drop; however, for all other applications, it is fairly useless.
3.3.4.4: VACUUM SENSORS

Such sensors are able to measure extremely low pressures or vacuum, referring to pressures below atmospheric pressure. Besides diaphragm
and electric sensors designed to measure low pressures, there are also thermal conductivity gauges and ionization sensors.
Thermal Conductivity Gauges
The principle involved here is the change in gas thermal conductivity with pressure. However, due to deviation from ideal gas behavior in
which the relationship between these two properties is linear, these kind of gauges, which are also called Pirani gauges, can only be used at
low pressures, in the range of (0.4E-3 to 1.3E-3) MPa. They are amazingly sensitive elements as well (can detect changes of 6E-13 MPa).
In these gauges, a coiled wire filament has a current flowing through it, which heats up the coil. A change in pressure changes the rate of
heat conduction away from the filament, thereby causing its temperature to vary. These changes in temperature can be detected by
thermocouples in the gauge, which are also connected to reference filaments in the gauge as part of a Wheatstone bridge circuit.
An example of a Pirani gauge can be found here: www.bama.ua.edu/~surfspec/vacbasics_files/image046.jpg
Ionization Gauges
There are two categories for these types of gauges: hot cathode and cold cathode. For hot cathode gauges, electrons are emitted by heated
filaments, while for the cold cathode gauges electrons are released from the cathode due to collision of ions. Electrons hit the gas molecules
entering the gauge forming positive ions, which are collected and cause an ion current to flow. The amount of cation formation is related to
the gas density and consequently the pressure to be measured, as well as the constant electron current used, hence the ion current flow is a
direct measure of the gas pressure. These both are highly sensitive instruments and thus most suited for fractional pressures. The hot
cathode gauges are even more sensitive than cold cathode gauges and are able to measure pressures near 10-8Pa. Their sensitivity ranges
from (1E-16 to 1E-13) MPa
An example of an ionization gauge can be found here: www.bama.ua.edu/~surfspec/vacbasics_files/image049.jpg
3.3.4.5: SUMMARY
 EXAMPLE 3.3.1: SEMI-BATCH REACTOR PROBLEM
Suppose you have a semi-batch reactor (1,000 L) with 50 kg of zinc within the reactor at 1 atm pressure and 25 C. 6M hydrochloric
acid is flowing in at 1 L / min to react with the zinc to produce zinc chloride for use in another process.
a. What factors should be considered?
b. Say the valve fails at an operating pressure of 4 atm (i.e., it will not close and the reactor will be flooded with HCl) At what pressure
would you (safely) set the shutdown point?
c. What type of sensor should be used?
Solution
3.5.1 The factors that should be considered are:
1. Process
1. Hydrochloric acid is very, very caustic (especially at such a high molarity), and thus whatever sensor you choose, it should be able
to withstand the corrosive nature of hte process.
2. Range of pressures
1. Initially the reactor is at 1 atm of pressure. Considering the reaction 2 HCl(aq) + Zn(s) --> H2(g) + ZnCl2(aq), you're producing one
mole of hydrogen gas in addition to the existing air pressure in the vessel. As the reaction proceeds, the pressure would increase
substantially within the vessel. Modeling the pressure of H2(g) ideally, P = nRT / V
2. = 0.073 atm / min.

3. After about 1 hour, the pressure of H2(g) would increase to 4.38 atm, putting the total pressure in the vessel at 5.38 atm (past failing
pressure for the valve).
3. Environment
1. There is no danger of high temperatures or excessive vibration due to high flow / reaction.
4. Sensitivity
1. Since this is a moderately dangerous process, we need to have the sensor output to a computer so an engineer can monitor the
process as a fail-safe. We assume the sensor will signal the HCl valve to close once the operating pressure becomes 3 atm, however
devices do fail from time to time. We also need high sensitivity, so electric components will be preferrable (i.e., we don't want the
process to runaway as it potentially could if the sensor isn't highly sensitive to incremental changes).
3.3.4.6: THE SHUTDOWN POINT

Considering the rapid increase in pressure, as evaluated in (2), and the failure of the valve at 4 atm, the shutdown point should be
approximately 3 atm.
3.3.4.7: THE TYPE OF SENSOR:

1. Considering the types of sensors discussed, we can automatically eliminate vacuum sensors as they operate at very low pressures
(almost vacuum, hence the name). We can also eliminate differential pressure sensors as we are not looking for a pressure drop
across the vessel.
2. Since we desire high sensitivity, we would like to use electric components. Considering the range of pressures (3 atm max ~ .3
MPa) a capacitive element would be optimal because it is robust and functions well in low pressure system.
3. Noting the corrosiveness of the HCl in the system, a diaphragm is likely to be used as the elastic element. Diaphragms are also
fairly robust and offer quick response times.
4. This combination is likely to be filled in a strong, sealed, glycerine/silicone filled case so as to protect the sensor from degradation.
So, overall, the sensor we choose will utilize a diaphragm as the elastic component, a capacitive element as the electrical component,
and an anti-corrosive body.
 EXAMPLE 3.3.2
Your supervisor told you to add a pressure sensor to a very expensive and important piece of equipment. You know that the piece of
equipment is operating at 1MPa and at a very high temperature. What sensor would you choose?
Solution
Because the piece of equipment that you are dealing with is very expensive, you need a sensor that has the greatest sensitivity. An
electric sensor would be desired because you could have it connected to a computer for quick and easy readings. Also, you need to pick
a sensor that will operate at 1 MPa and a sensor that will be able to withstand high temperatures. From the information given in this
article you know that there are many sensors that will operate at 1 MPa, so you need to decide on the other factors. One of the most
sensitive electric sensors is the capacitive sensor. It has a sensitivity of 0.07MPa. The capacitive sensor usually has a diaphragm as the
elastic element. Diaphragms have a rapid response time, are very accurate and work at 1 MPa.
3.3.5: REFERENCES
2005/2006 Cole-Parmer General Catalog, 2005
Liptak, Bela G.(2005). Instrument Engineers Handbook, 4th Edition, Volume 2. Taylor & Francis CRC Press. ISBN 0849310814
Perry, Robert H.(1997), Perry's Chemical Engineers' Handbook, 7th Edition. McGraw-Hill. ISBN 0070498415
Richardson, J.F. and Peacock, D.G. (1994), Chemical Engineering, 3rd Edition, Volume 3. Butterworth-Heinemann. ISBN 0080410030
Freeman, Lawrence A., Carpenter, Michael C., Rosenberry, Donald O., Rousseau, Joseph P., Unger, Randy and McLean, John S. Use of
submersible pressure transducers in water-resources investigations. http://pubs.usgs.gov/twri/twri8a3/#N10423
"Pressure Sensors," www.data-
acquisition.us/industrial_electronics/input_devices_sensors_transducers_transmitters_measurement/sensors/pressure-sensors.html
Wikipedia, en.Wikipedia.org/wiki/Ion_gauge

Authors: (09/14/06 / Date Revised: 09/20/06 ) Wen Chung Lim, James Bennett, Jamila Grant, Ajay Bhasin
3.3: Pressure Sensors is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source content that was
3.4: LEVEL SENSORS
3.4.1: INTRODUCTION
Level sensors allow for the level control of fluid in a vessel. Examples of where these sensors are installed include reactors, distillation
columns, evaporators, mixing tanks, etc. Level sensors provide operators with three important data for control: (1) the amount of materials
available for processing, (2) the amount of products in storage, (3) the operating condition. Installing the correct level sensor ensures the
safety of the operator and the surrounding environment by preventing materials in vessels from overflowing or running dry.
There are several different types of level monitors, including:
Visual
Float
Valve Controlled
Electronic
Radiation
These different types of sensors can also be grouped into categories of process contact and non-process contact. As the name suggests,
process contact sensors are within the tank, in physical contact with the material. Non-process contact sensors transmit various types of
signals to reflect off of the material and thus measure the level. This sensor design can maintain its integrity within a potentially corrosive
material and/or be positioned such that it can monitor the level changes from above the tank.
3.4.2: VISUAL LEVEL SENSORS

Visual level controls were the earliest developed level sensor. These types of monitoring devices can be something as simple as looking into
an open container or inserting a marked object such as a dipstick. This type of sensor is the simplest and possibly the most reliable. These
devices do not provide a way of connecting to a control device. They require human input with no way of automation; however, electronic
issues will not be a possible problem.
3.4.2.1: COMMON USES

Visual indicators can be found in many places and are not limited to chemical engineering applications. Examples of these are: dipsticks
found monitoring the oil levels in a car and measuring cups with markings indicating different volumes. In chemical plants, visual indicators
are used to measure changes in level as well as for high and low level alarms.
3.4.2.2: BENEFITS OF VISUAL LEVEL SENSORS

Visual sensors are generally less expensive than other types of sensors. They are more reliable because of the simplicity of design.
3.4.2.3: RESTRICTIONS
Visual indicators do not always accurately measure how much volume is in the tank. They do not allow for any digital connections to
process control systems. In the case of the sight tubes, they must be affixed directly to the tank and cannot be read remotely without a
camera or other transmission tool. Tanks placed in hard-to-reach places would also cause problems contributing to inaccuracies in
measurement, such as not reading markers on eye level or interpolating between marker intervals. Also, the connections between the tube
and the tank are subject to residue buildup which could prevent the readings from being accurate. Weather is also a concern with sight tubes.
External temperature changes could affect the fluid in the tube and thus measurement inaccuracy. For example, the fluid in the sensor could
freeze, or the sensor could become clogged. This type of indicator requires modification to the vessel wall and would therefore be optimally
installed at the time of initial construction. If considered at the time of the initial capital installation of the tank a sight tube will not add a
large cost to the project. Yet, later modification to include this in the design and associated tank changes are potentially very costly. Despite
these shortfalls, a sight tube is a reliable and appropriate choice in many common level reading applications.
3.4.2.4: SIGHT TUBE INDICATORS
Sight tube indicators allow operators to monitor levels with precision while keeping the tank sealed. This type of monitoring device is
comprised of a vertical tube equal in height to the actual vessel. This tube is connected in at least two places directly to the vessel so that its
contents can flow into the monitoring tube. This ensures that the height of liquid in the tank will be equal to the height of liquid in the sight
tube. Markings can be affixed to the indicator for both calibration and volume readings. There are many styles of this indicator. Some
models have a float ball contained in the tube so that it will float on top of the liquid. Other models of the indicator have many paddles on
fixed posts through the length of the tube. These paddles float horizontally when submerged and are vertical when not submerged. This
results in a color change allowing for an easy identification of the tank volume.
A sight tube. More pictures of this type of gauge can be found on Seetru Limited's website.
3.4.3: FLOAT TYPE LEVEL SENSORS

Float Types of level sensors are based on the principle of buoyancy which is the upward force produced on a submerged object by the
displaced fluid. This force is equal to the weight of the displaced fluid. Float sensors take their measurements at the interfaces of materials,
where the movement of the float and/or the force on the float are caused by the differing densities of the float and the fluid. There are two
broad categories of Float Type level sensors: Buoyancy and Static.
3.4.3.1: BUOYANCY TYPES

Buoyancy level sensors are less dense than the fluid and thus change position along with the fluid level. The movement of the float
transmits the level information through some mechanical linkage to an output such as a valve or operator observation. There are basic three
types of mechanical linkage - chain / tape sensors, lever / shaft mechanisms and magnetically coupled devices.
Chain / Tape sensors – The linkage is by a flexible chain or tape.
Lever / Shaft mechanisms – The linkage is a rigid shaft.
Magnetically coupled devices – These devices are similar to the Chain/Tape sensors, except a magnet is attached to the float and another
is acted upon by the floating magnet moving a tape like the chain/tape type devices. The moving magnet can be sequestered from the
float attached magnet for use in corrosive media.
Schematic of a float type level sensor is shown below. As the fluid level rises or falls, buoyant force is transferred through mechanical
linkages to your output device.
3.4.3.2: STATIC TYPES
Static level sensors are denser than the media being measured and thus do not move. As the level changes, the buoyant force acting on the
"float", which is actually a weight, changes. The change in weight is measured by a scale. The level of the tank is calculated by
measurements of the weight change in the float, not its actual change in position.
3.4.3.3: FORMULAS
Changes in the volumetric rate of material flow, resulting from changes in such things as the pressure of your pump will affect the operation
of a float type level sensor. The same movement of the float will have the same effect on the valve, but as valve is moderating a different
flow it has a different proportional sensitivity K .
c
q
Kc =
k
where q is the flow rate and k is the height the float moves to completely traverse the valves operating range.
This is important because the given change in fluid height used to previously design the system will still change the valve’s position in the
same manner, but won’t affect the same change in flow rate. If your pressure has dropped, your valve will have to be opened wider to match
the same flow rate. The amount wider that it has to be opened is called an offset. Manipulating the valve response to the float movement,
also known as the gain:
(valveresponse )
Gain =
float movement
can be used to alleviate this problem, but it won’t work exactly the same way it did before the change in load.
Operation power is the amount of power the float has available to perform actions to control the system. Changes in the specific gravity of
the fluid (or the float) will affect buoyancy of the float and thus will affect the amount of force exert on your system. A temperature is a
very common parameter that can change your fluid’s specific gravity. Warmer fluids tend to be less dense, and so your float will generate
less buoyant force for a given change in fluid level.
Zero buoyancy force level is the level at which the weight of the float is exactly as buoyant as the fluid it replaces. It can be approximated
by the following formula:
(S Gfloat )
Zero Buoyant Force Level = D
S Gfluid
where D is the diameter of the float and SG is the specific gravity.
3.4.3.4: COMMON USES

Float Type Level Sensors regulate how much water is in the reservoir of a flush toilet; the float is attached to a lever which has a rotating
axis that stops when the flow of water reaches a certain level.
Chain / Tape sensors - regulating the level on storage tanks at atmospheric pressures.
Lever / Shaft Mechanisms - regulating the level on vessels under pressure.
3.4.3.5: BENEFITS
Float Type level sensors do not require external energy sources to operate. Since they are simple robust machines it is easy to repair. The
cost of these units makes them on the economical side, ranging from approximately $20 (homedepot.com) for the apparatus in a standard
toilet to a few hundred dollars for a 3/4" cast iron valve with float (Liptak 2005). Prices would increase with higher quality materials and
valve size. Floats could be made out of plastics or metals (steel, stainless steel, etc.). Material selection will depend on the application; more
chemically resistant materials would be used for corrosive mediums.
3.4.3.6: RESTRICTIONS
Float type level sensors should only be used in clean fluids. Fluids that are a suspension of solids or slurries could foul the operation of the
machine. Anything that could increase friction on the mechanical linkages could increase the dead band, which is the delay the sensor
experiences due to the excess of force required to overcome the static friction of a non-moving system. An excess of force is required to be
built up, so a float type level sensor will not respond immediately to changes in level.
3.4.4: VALVE-BASED LEVEL SENSORS

Valve-based level sensors not only measure the fluid level, but also cause the fluid level to change accordingly. Two basic types of valves
that will be discussed are altitude valves and diverter valves.
3.4.4.1: ALTITUDE VALVES
A simple altitude valve uses a spring that opens and closes different ports and lines when pressure changes due to the changing fluid levels.
When fluid levels exceed the setting of the spring, a diaphragm connected to the spring lowers, closes the drain port, and opens the main
line pressure. This turns off the main valve and stops fluid from flowing into the tank. When fluid levels decrease, the diaphragm raises,
opens the drain port, and closes the main line pressure. This turns on the main valve and more fluid is supplied to the tank.
An example of an altitude valve is shown in the bottom left corner of the following website:
http://www.gaindustries.com/html/01a_SAVV.htm
3.4.4.1.1: COMMON USES
Altitude valves are on/off controls and can be found in supply lines connected to basins, tanks, and reservoirs. The main job of these valves
is to prevent overflow of the fluid and to hold fluid level constant.
3.4.4.1.2: BENEFITS
There is no external power source needed when operating altitude valves because they are controlled by the pressure of the process fluid.
More complicated altitude valves can also be used for other purposes other than an on/off control. These features include the following:
Open the valve when pressure drops to a predetermined point
Delayed opening so the valve only opens when fluid level drops a certain amount below the set point
Close the valve slowly to eliminate pressure build-up
Two-way flow to allow fluid to return when the level has dropped below the set point
Open the valve while maintaining constant inlet pressure and plant distribution pressure
Reduce pressure for outlet streams. This may be needed when working with equipment that has high supply pressures, such as aerator
basins. Aerator basins are typically used in wastewater treatment because they have the capability of holding large amounts of water.
Check valves can be installed to allow the valve to close if the pressure reaches a predetermined low point (which could be the cause of
equipment failure elsewhere in the plant)
3.4.4.1.3: RESTRICTIONS
Altitude valves should not be used in pressurized vessels. These valves are constructed from a limited number of materials, such as cast
iron. This restricts the number of fluids that can be used with the valves. Therefore, altitude valves are usually used for water service. Since
these valves are usually used to operate a large amount of fluid, they are subsequently large with multiple functions that make them very
expensive. A 30 inch altitude valve can cost over $50,000. In addition, frequent maintenance is required because of the many components
and moving parts. These valves are operated at ambient temperatures and should be checked for freezing of stationary sensing lines.
3.4.4.2: DIVERTER VALVES

Diverter valves are connected to a dip tube that is submerged in a controlled tank. When the fluid level of the tank drops, the dip tube is
exposed to atmospheric pressure. Once air enters the dip tube, it utilizes the Coanda effect. Coanda effects occur when a curved surface
exists. Gases tend to follow the nearest curved surface while pushing other fluids in a different direction. In the case of the diverter valve, air
from the dip tube follows the curved surface of the control port back to the storage tank and pushes fluid flowing out of the storage tank to
the other wall of the valve. This fluid then flows into the controlled tank to adjust the level.
3.4.4.2.1: COMMON USES

Diverter valves control fluid level by acting as an on/off switch. They can also be modified to be used for throttling. Diverter valves are
used as low level indicators so they should only been used when low levels of a tank are a concern.
3.4.4.2.2: BENEFITS
Diverter valves need no external power source for operation. They have no moving parts, plugs, or packing, so they require minimal
maintenance. Diverter valves can be operated at any temperature and are not affected by vibration. They can be constructed from many
materials such as metals, plastics, and ceramics. Therefore, there is limited restriction to the type of liquid used with the valves. However,
when using fluidized solids and slurries, the dip tube must remain open, so build-up does not occur. Any build-up near or in the tube should
be removed.
3.4.4.2.3: RESTRICTIONS
Diverter valves should not be used when operating under conditions other than atmospheric pressure. They are also not for use when dealing
with hard to handle processes. Outlet pressure cannot be greater than atmospheric pressure or back pressure will occur and the valve will not
operate correctly.
3.4.5: ELECTRICAL LEVEL SENSORS

3.4.5.1: CONDUCTIVE LEVEL SENSING
Conductive level sensors work by applying a low-voltage across two electrodes at different levels in a vessel. When both electrodes are
immersed in a conductive liquid, a current flows. This type of electrical conductivity setup is best applicable for point level detection (level
detection at a specific point in the material.) They are typically made of titanium, Hastelloy B, or stainless steel.
Example of a conductive level sensor.

Common Uses
Conductive level sensors are commonly used to measure both conductive and corrosive liquids. A common conductive liquid is water,
whereas some common corrosive liquids are nitric acid, ferric chloride, and hydrochloric acid.
Benefits
This method is considered extremely safe due to the low voltages and currents used. Conductive level sensors are also known for their easy
installation and use.
Restrictions
The greatest concern with conductive sensors is maintenance. The probe needs to be monitored for buildup on the sensor. Residues from
wet or sticky fluids that cause build up can be prevented by coating the sensors with Teflon or polyethylene-based materials.
3.4.5.2: CAPACITANCE LEVEL SENSING

A capacitor is made up of two conductors (electrodes/plates) that are electrically separated by a nonconductor (dielectric). In the case of
level sensing, one of the electrodes is typically a vertically positioned rod while the other is the metallic vessel wall. The dielectric between
them is the material being measured in the vessel. The principle of capacitance level sensing is based on the formula below:
Capacitance = Dielectric constant x (Area of the plates ÷ Distance between plates)
If the dielectric is a liquid, the capacitance probe can measure the combined capacitance of both the liquid and gas. When the liquid level
rises or falls, the total capacitance value will change. Since the dielectric constant and distance between the plates are constant (the rod and
the vessel are stable), the only value changing the capacitance is the area of the plates immersed in the liquid. Total capacitance changes
approximately proportionally to the liquid rise or fall in the column. Consequently, the liquid level can be calculated by the change in
capacitance.
Common Uses
The list below shows just a few applications of capacitance level sensors found in industry and the materials they sense.
Chemical/Petrochemical - Oil, clay, soda ash
Food - Flour, Powered Milk, Sugar
Charcoal - Wood, Charred sawdust
Pharmaceuticals - Various powders and liquids
Mining - Various minerals, metals, stone,
Benefits
Capacitance level sensing is useful in its ability to sense a wide variety of materials such as solids, organic and aqueous solutions, and
slurries. For example, materials with dielectric constants as low as 2.1 (petroleum oil) and as high as 88 (water) or more can be detected
with capacitance level sensing. The equipment, typically made of stainless steel, is simple to use, clean and can be specifically designed for
certain applications. For example, they can be made to withstand high temperatures and pressures or have built-in splashguards or stilling
wells for environments prone to turbulence.
Restrictions
There are limitations to using capacitance level sensors. One major limitation for capacitance probes is found when using tall bins that store
bulk solids. Probes are required to extend for the length of the vessel, so in a lengthy vessel the long cable probes can be subject to
mechanical tensions and breakage. Another limitation is from build up and discharge of a high-voltage static charge that can result from the
rubbing and movement of low dielectric materials, but this danger can be eliminated with proper design and grounding. Also, abrasion,
corrosion, and build up of material on the probe can cause variations in the dielectric constant of the material being measured. To reduce
this issue, capacitance probes can be coated with Teflon, Kynar, polyethylene or other materials.
3.4.6: RADIATION-BASED LEVEL SENSORS

Radiation-based level sensors are based on the principle of a material’s ability to absorb or reflect radiation. The common types of radiation
used in continuous level gages are ultrasonic, radar / microwave and nuclear.
3.4.6.1: ULTRASONIC (SONIC) LEVEL SENSORS

Ultrasonic level sensor transmitters emit high frequency ultrasonic acoustic waves which are reflected back by the medium to the receivers.
By measuring the time it takes for the reflected echo to be received, the sensor can calculate the actual distance between the receiver and the
fluid level. These sensors can be accurate from a distance of 5mm to 30m.
Common Uses
Ultrasonic level sensors are commonly used for point level detection. They are best used for viscous liquids, slurries and bulk solids.
Benefits
The combination of ultrasonic level sensors’ high functionality for relatively low prices makes them a popular choice for non-contacting
level sensing.
Restrictions
Since the speed of sound in air can fluctuate in different conditions, ultrasonic sensors are not suitable for use in all applications.
Environments that have varying moisture and temperatures will influence the sensors readings. Turbulence, steam, and foam prevent waves
from reflecting properly and distort readings. In addition, the level sensor must be mounted properly so that it correctly senses the distance
between the transmitter and the fluid level.
3.4.6.2: MICROWAVE / RADAR LEVEL SENSORS

Microwave / radar level sensors are similar to ultrasonic level sensors in that they require a transmitter and receiver. In addition to these
materials, radar sensors also need an antenna and operator interface to use electromagnetic waves to calculate level distance.
Common Uses
These sensors are frequently used for non-contacting situations that require level sensing in varying temperature and pressure environments.
Benefits
Microwave / radar sensors have an advantage over ultrasonic sensors in that they are able to operate in high pressure and high temperature
environments. They can also sense solids with grain sizes larger than 20mm.
Restrictions
When fluid level is under a thick layer of foam or dust, these sensors may not detect the fluid level, but instead detect the level of the dust or
foam. Microwave / radar level sensors are also more costly than ultrasonic level sensors.
3.4.6.3: NUCLEAR LEVEL SENSORS

Nuclear level sensors rely on gamma rays for detection. Although these gamma rays can penetrate even the most solid of mediums, the
intensity of the rays will reduce in passage. If the gamma ray emitter and the detector are placed on the top and bottom of a vessel, the
thickness of the medium (level) can be calculated by the change in intensity.
Common Uses
These sensors are usually used when the material being measured presents a risk to human life or the environment. These include, for
example, materials that are toxic, carcinogenic, or explosive.
Benefits
Nuclear radiation has the ability to detect level even through solid tank walls. Since they appear to “see” through walls, the nuclear gage
may be modified and/or installed while the process is running and avoid expensive down time.
Restrictions
Nuclear level sensors are typically the last resort when choosing a level sensor. They not only require a Nuclear Regulatory Commission
(NRC) license to install, but are extremely expensive in comparison to other level sensors.
3.4.7: SUMMARY OF BENEFITS AND RESTRICTIONS
 EXAMPLE 3.4.1: PLANT STORAGE VESSEL
Imagine that you are responsible for maintaining a certain amount of chemical in a storage vessel for use later on downstream in your
process. Your chemical has a specific gravity of 1.2. Assume that it is effectively independent of temperature. At maximum flow rate, a
valve is feeding 80 gallons per minute into a storage vessel. 2 inches of valve travel is required to completely shut off the flow through
the valve. The spherical float you are given to work with is 8 inches across, chemically compatible with your chemical, and its specific
gravity can be altered. Your process dictates that you have to maintain about 400 gallons in this tank for use downstream, and that
corresponds to a height of twenty inches from the bottom of the tank. At this height, the ball is 4 inches submerged. The ball has to
travel 5 inches to completely close your valve. Calculate:
a. the specific gravity the float should be
b. the gain of the regulator
c. the proportional sensitivity of your system
d. assume the specific gravity of your chemical increases dramatically. Explain qualitatively what that should do to the gain, zero
buoyant force level and dead band of the sensor
e. was a float type valve appropriate to be used in this problem?
Solutions
3.4.7.1: USING THE FORMULA FOR ZERO BUOYANT FORCE LEVEL
AND PLUGGING IN VALUES FROM THE PROBLEM
STATEMENT, WE COME UP WITH THE FOLLOWING WE USED 4 INCHES,

AND NOT 20 INCHES BECAUSE THE ZERO BUOYANT FORCE LEVEL INDICATES THE
LEVEL THE FLOAT IS SUBMERGED, NOT THE LEVEL OF THE FLUID. ISOLATING
THE VALUE FOR SGFWE ARRIVE AT A SPECIFIC GRAVITY OF 0.6 FOR THE FLOAT.
b) Using the formula for the gain of the regulator and values from the problem statement, we come up with
the following proceeding with the calculation we arrive at a value of 0.4 for the gain.
c) Using the formula for the proportional sensitivity of the system and values from the problem statement, we come up with
the following proceeding with the calculation we arrive at a value of

d) Float type level sensors work on the principal of buoyancy. If the density of the liquid being displaced increases, the buoyant force
on the float increases for the same change in liquid height. The gain of the sensor refers to the amount of travel in the valve versus the
amount of travel in the float, so for a given change in fluid height, the valve will still change the same amount as it did. However, since
the buoyant force is increased due to the increased process fluid density, the dead band should decrease because the force of an
incremental increase in height should enable the sensor to overcome friction more easily.
e) A float type valve was appropriate to be used in this problem, because the requirements were that "about 400 gallons" be maintained
and the fluid was chemically compatible with the float. Due to unavoidable dead band, or the time it takes the float to respond to a
given change in fluid, it won't maintain exactly 400 gallons, but it will stay close to that.
 EXAMPLE 3.4.2: SENSOR SELECTION

Please answer the questions for the four hypothetical scenarios regarding selection of an appropriate level sensor. Give reasons for your
selections.
a. A commonly used tank contains Dichloromethane which requires constant monitoring to ensure that no leaks exist. The precise
volume will be regulated to monitor any decrease in level when a known transfer is not taking place. This tank will be filled to
~5,000 gallons and not be refilled until the level is less than 1,000 gallons of dichloromethane. What type of sensor should you
install?
b. Under what circumstances would you select a radiation sensor?
c. Toilets have a Float Level Sensor to indicate when a flush has occurred and the bowl needs to be refilled. Which type of Float
Sensor does the toilet have?
d. When would be a good time to use a visual sensor?
Solutions
a. The most appropriate selection of level sensor for this application is the capacitance sensor or the ultrasonic sensor. These are both
electrical sensors that allow for measurements of changes in height, whereas other types offer a binary result: either the level is
above or below a set point. The capacitance and ultrasonic sensors differ only by the way they are installed. Capacitance sensors
require contact with the liquid while ultrasonic sensors do not. In this specific case for dichloromethane, which is non-corrosive, it
is not necessary to monitor the volume without contacting the liquid, and therefore the capacitance sensor is also an acceptable
selection.
b. A radiation-based level sensor is most appropriate when sensors cannot be brought into direct contact with the measured substance.
This is only a possible option and often an expensive one.
c. Toilets have a lever/shaft mechanism. This float is less dense than the water which fills the tank. It will drop quickly when the toilet
is flushed and will slowly rise as the tank fills with water. When the tank is full the float will rotate the shaft upwards enough to
close off the flow of water.
d. Visual sensors should only be used when an electrical sensor is not required or as a backup. This is because they cannot be
monitored automatically as visual sensors do not provide a digital response. It is also a good idea to provide a secondary way of
obtaining level information when installing a new tank, and manual checks are often a good idea.
3.4.8: REFERENCES
[Image Available] Seetru, www.seetru.com/index_main.html, 12 Sept 2006.
Liptak, Bela G., Instrument Engineers Handbook, 4th Edition, Volume 2. Taylor & Francis CRC Press, 2005.
Perry, Robert H., Perry's Chemical Engineers' Handbook, 7th Edition. McGraw-Hill, 1997.
Richardson, J.F. and Peacock, D.G., Chemical Engineering, 3rd Edition, Volume 3. Butterworth-Heinemann, 1994.
Transactions in Measurement & Control: Volume 4, 2006. Omega Engineering, Inc. Retrieved September 10, 2006, from
http://www.omega.com/literature/transactions/volume4/
Wikipedia. Retrieved September 9, 2007, from en.Wikipedia.org/wiki/Level_sensor
Considine, Douglas M., Process/Industrial Instruments & Controls Handbook, 4th Edition, McGRAW-HILL, INC., 1993.

Authors: (September 14, 2006) Andrew MacMillan, David Preston, Jessica Wolfe, Sandy Yu
Stewards: (September 10, 2007) Yoo Na Choi, Yuan Ma, Larry Mo, Julie Wesely
3.4: Level Sensors is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source content that was edited
3.5: FLOW SENSORS
3.5.1: INTRODUCTION
Flow is defined as the rate (volume or area per unit time) at which a substance travels through a given cross section and is characterized at
specific temperatures and pressures. The instruments used to measure flow are termed flow meters. The main components of a flow meter
include the sensor, signal processor and transmitter. Flow sensors use acoustic waves and electromagnetic fields to measure the flow
through a given area via physical quantities, such as acceleration, frequency, pressure and volume. As a result, many flow meters are named
with respect to the physical property that helps to measure the flow.
Flow measurement proves crucial in various industries including petroleum and chemical industries. Consequently, flow measurement
becomes a major component in the overall economic success or failure of any given process. Most importantly, accurate flow measurements
ensure the safety of the process and for those involved in its success.
Before reading about the intricate details of various flow meters it's a good idea to think about aspects other than the design, governing
equations, and the mechanism a flow meter uses in identifying a flow profile. To gain more from this section consider the following
questions when learning about each flow meter because when choosing a flow meter the two main things to think about are cost and
application:
1. What mechanism does this flow meter use to measure flow?
2. How expensive is it?
3. What impact does it have on the system? (How much power does it withdraw from the system and does the disturbance of flow cause
significant problems up or downstream?
4. What are the accuracy limits of the tool? (Increased accuracy and precision results in higher cost especially for those with automated
noise filtering and signal amplifying capabilities.
Sometimes you may be given an assignment to purchase a tool for your company's system and given with it a list of qualities to look for in
order of importance, sometimes you're going to have to make this list yourself. The following table is a component of the TRIZ method for
developing a list of qualities in descending order of importance on the left, perpendicular to the factor it has an impact on, to the right. For a
given system you figured out that the cheapest solution for accurately measuring the flow rates of various pipe lines is by placing multiple
flow meters in a series of positions. From a few calculations using propagation of error you find a moderate range of accuracy limits
necessary for your system to be considered well monitored. You know the accuracy limits are less important than the total cost of all the
flow meters because your boss told you not to waste any money. On the table you can see that Cost is at the top of the left column so Cost
is the main concern. Accuracy is in the farthest left column on the top row so Cost most dramatically exacerbates the Accuracy of the flow
meter when it's minimized. In the intersecting box you see that the lowest price flow meters are the ones with the least amount of accuracy,
generally speaking. The next most important quality of your assignment is to get flow meters with the appropriate accuracy, thus Accuracy
is below Cost in the left column. Looking to the top row you see that the Accuracy of the flow meter most greatly effects the Impact on
the system. If you have a low quality flow meter it may be due to the side of the mechanism used to measure flow or the power the meter
draws from the system (through mixing, frictional losses, increase in the turbulence of the flow, or buoyant effects caused from heat
transfer). Completing the rest of the table you can decide if there are contradictions to what you thought were the most important qualities
based on the inputs in the intersecting cells.
An example where this model is important is for a system containing suspended particles in a fluid. If you want to measure the flow rate it
might be cheapest to use a pitot tube yet the increased cost of maintaining the flow meter extends into down time of the system, more work
from the technicians for a relatively small aspect of the process and lower profits as a result. For this system, maintenance would be the
most important factor followed by accuracy, impact on the system and cost. If cost were least important factor you could afford to install a
couple doppler meters, gather accurate measurements and the impact of the measurements on the system would be relatively low.
3.5.2: COMMON TYPES OF FLOW METERS

The flow rate as determined by the flow sensor is derived from other physical properties. The relationship between the physical properties
and the flow rate is derived from fundamental fluid flow principles, such as Bernoulli’s equation.
3.5.2.1: DIFFERENTIAL PRESSURE

These sensors work according to Bernoulli’s principle which states that the pressure drop across the meter is proportional to the square of
the flow rate.
2
−Δp ∝ V
Using the pressure drop across a pipe’s cross section is one of the most common manners to determine a flow measurement. As a result, this
property has heavy implications for industrial applications. Flow meters characterized by differential pressure come in several different
varieties and can be divided into two categories, laminar and turbulent. Differential pressure sensors operate with respect to Bernoulli’s
principle. Bernoulli’s principle states that the pressure drop across the meter is proportional to the square of the flow rate.
3.5.2.1.1: ORIFICE METER
Orifice plates are installed in flow meters in order to calculate the material balances that will ultimately result in a fluid flow measurement
on the sensor. An orifice plate is placed in a pipe containing a fluid flow, which constricts the smooth flow of the fluid inside the pipe. By
restricting the flow, the orifice meter causes a pressure drop across the plate. By measuring the difference between the two pressures across
the plate, the orifice meter determines the flow rate through the pipe. The larger the pressure drop, the faster the flow rate would be. There
are two types of orifice meters that are usually used in industry, they are the orifice-square edge and the orifice-conic edge. The orifice-
square edge has insignificant friction at the interface between the fluid and the orifice plate. These types of orifice plates are recommended
for smooth fluid flows, particularly clean liquids and gases such as steam. Generally, drain holes are incorporated in the design so that
liquids and gases do not accumulate inside the pipes. Multi-phase fluids are not recommended for the orifice-squared edge because clogging
becomes a significant problem as time progresses. The orifice-conic edge is similar to the orifice-square edge, the primary difference being
that the orifice-conic edge has a gradually expanding cross-sectional area, and this cross-sectional area is circular in shape. A conic-edge
design is often a better choice for low velocity, high viscosity flows. Both types operates best under comparable temperature and pressure
conditions, pipe sizes and provide similar accuracies.
Orifice meters used in conjunction with DP (Differential Pressure) cells are one of the most common forms of flow measurement. In
addition, an orifice meter can be used to measure flows when there is a significant difference in pressure in the pipe, like between the
upstream and downstream sides of a partially obstructed pipe, which is exactly what the orifice meter does on its own. The plate offers a
precisely measured obstruction that essentially shrinks the pipe and forces the flowing substance to constrict. A DP cell allows the
comparison of pressure on the upstream (unobstructed) side and the downstream (constricted) side. A greater rate of fluid flow would
usually result in a larger pressure drop, since the size of the orifice remains constant and the fluid is held longer building potential energy on
the upstream side of the orifice. Some of the other types of orifice plates include concentric, eccentric and segmental plates, each having
different shapes and placements for measuring different processes. These plates are available in varied shapes so that the meter has the
optimum structure for different applications. Moreover, the density and viscosity of the fluid, and the the shape and width of the pipe also
influences the choice of plate shape to be used.
Such a pressure drop across the plate is then related to the flow rate using basic fluid mechanics principles that incorporate parameters such
as density of the fluid and size of the pipe. The flow rate Q, given by the orifice meter, is usually modeled with the following equation:
Where p1 − p2 is the pressure drop across the plate, ρ is the fluid density, A1 is the pipe cross-sectional area, A2 is the orifice cross-sectional
area, and Cd is the discharge coefficient (usually about 0.6). Cd is used to account for energy losses within the system.
The orifice meter is one of the most commonly used flow meters, since it is inexpensive to install and operate, it is uncomplicated and easy
to construct, and it is highly robust and long lasting. Orifice meters are not only simple and cheap, they can also be delivered for almost any
application and be made of any material. This simplicity of its design and function is one of its paramount advantages, with the meter
essentially consisting of just a modified plate. This not only reduces its initial price but also shrinks its operating costs, maintenance
expenses, and spare parts expenditure and availability. Lower flow rates reduces their accuracy, whereas higher flow rates combined with
high quality, unworn orifice plates increases it. The orifice plate is best when a sharp edge is present towards the upstream side of the meter.
Wear reduces the accuracy of orifice plates. The turndown rate of orifice plates are generally less than 5:1. More information about
turndown rates is mentioned in a later section.
3.5.2.1.2: VENTURI METER
Venturi meters can pass 25 – 50% more flow than an orifice meter. In a Venturi meter setup, a short, smaller diameter pipe is substituted into
an existing flow line. Because the Venturi meter is insensitive to changes in the velocity profile of the fluid flow, the pipe design does not
need to be straight like the orifice meter. Though initially expensive, the Venturi meter has relatively low maintenance and operation costs.
In the Venturi Tube the fluid flowrate is measured by reducing the cross sectional flow area in the flow path, generating a pressure
difference. After the constricted area, the fluid is passes through a pressure recovery exit section, where up to 80% of the differential
pressure generated at the constricted area, is recovered.
There are two main types of Venturi meters. The first one, known as the classical Herschel Venturi meter, is a very long meter characterized
below. Pressure readings at different points in the meter are combined to provide an average pressure reading. Cleaning the classical Venturi
meter is limited. The second type of Venturi meter is known as the short form Venturi meter. This differs from its longer counterpart by
reduced size and weight.
By Bernoulli’s principle the smaller cross-sectional area results in faster flow and therefore lower pressure. The Venturi meter measures the
pressure drop between this constricted section of pipe and the non-constricted section.
The discharge coefficient for the Venturi meter is generally higher than that used for the orifice, usually ranging from .94 to .99.
The Venturi meter is most commonly used for measuring very large flow rates where power losses could become significant. The Venturi
flow meter has a higher start up cost than an orifice, but is balanced by the reduced operating costs.
Due to simplicity and dependability, the Venturi tube flowmeter is often used in applications where higher turndown ratiosor lower pressure
drops than orifice plates can provide are necessary. With proper instrumentation and flow calibrating the venturi meter flowrate can be
reduced to about 10% of its full scale range with proper accuracy. This provies a turndown ratio of around 10:1.
3.5.2.1.3: FLOW NOZZLE
Another type of differential pressure flowmeter is the flow nozzle. Flow nozzles are often used as measuring elements for air and gas flow
in industrial applications. At high velocities, Flow Nozzles can handle approximately 60 percent greater liquid flow than orifice plates
having the same pressure drop. For measurements where high temperatures and velocities are present, the flow nozzle may provide a better
solution than an orifice plate. Its construction makes it substantially more rigid in adverse conditions and the flow coefficient data at high
Reynolds numbers is better documented than for orifice plates. Liquids with suspended solids can also be metered with flow nozzles.
However, the use of the flow nozzles is not recommended for highly viscous liquids or those containing large amounts of sticky solids. The
turndown rate of flow nozzles is similar to that of the orifice plate. The flow nozzle is relatively simple and cheap, and available for many
applications in many materials.
The Sonic Nozzle- Critical(Choked)Flow Nozzle
One type of flow nozzle is the sonic nozzle. The Sonic Nozzle is a converging-diverging flowmeter. It consists of a smooth rounded inlet
section converging to a minimum throat area and diverging along a pressure recovery section or exit cone.
The Sonic Nozzle is operated by either pressurizing the inlet (P1) or evacuating the exit (P3), to achieve a pressure ratio of 1.2 to 1 or
greater, inlet to outlet. When a gas accelerates through a nozzle, the velocity increase and the pressure and gas density decrease. The
maximum velocity is achieved at the throat, the minimum area, where it breaks Mach 1 or sonic. At this point it's not possible to increase
the flow by lowering the downstream pressure. The flow is choked. Pressure differences within a piping system travel at the speed of sound
and generate flow. Downstream differences or disturbances in pressure, traveling at the speed of sound, cannot move upstream past the
throat of the Nozzle because the throat velocity is higher and in the opposite direction.
Sonic Nozzles are used in many control systems to maintain fixed, accurate, repeatable gas flow rates unaffected by the downstream
pressure. If you have a system with changing or varying gas consumption downstream and you want to feed it a constant or locked flowrate,
a Sonic Nozzle is an excellent way to achieve this.
3.5.2.1.4: PITOT TUBES
Pitot tubes measure the local velocity due to the pressure difference between points 1 and 2 in the diagrams below. Unlike the other
differential flow meters, the pitot tubes only detect fluid flow at one point rather than an overall calculation. The first diagram shows a
simple pitot tube configuration while the second shows a compact pitot tube configuration.
Both tubes work in a similar manner. Each pitot tube has two openings, one perpendicular to the flow and one parallel to the flow. The
impact tube has its opening perpendicular to the fluid flow, allowing the fluid to enter the tube at point 2, and build up pressure until the
pressure remains constant. This point is known as the stagnation point. The static tube, with openings parallel to the fluid flow gives the
static pressure and causes a sealed fluid of known density to shift in the base of the tube. Pressure drop can be calculated using the height
change along with the fluid densities and the equation below.
Δp = Δh (ρA − ρ) g
with Δp as the pressure drop, ρA as the known fluid density, ρ as flowing fluid’s density, and g as the acceleration due to gravity.
This pressure drop can be related to the velocity after accounting for the losses throughout the piping in the system, given by Cp. This
dimensionless coefficient is found through accurate calibration of the pitot tube. The equation below describes this relationship.
−−−−−−−−−
2 (p1 − p2 )
v = Cp √
ρ
with v as the fluid velocity, Cp as the loss coefficient, p1 as the pressure at point 1, p2 as the pressure at point 2, and ρ as the flowing fluid’s
density.
By placing the tube at the exact center of the pipe, the maximum velocity can be measured and the average velocity can be calculated via
the Reynolds number. The governing equation and chart are below.
Dvmax ρ
Re =
μ
with Re as the Reynolds number, D as the pipe diameter, vmax as the maximum velocity, ρ as the flowing fluid’s density, and μ as the
flowing fluid’s viscosity.
Adapted from Geankoplis.

Finally, the flow rate can be found by accounting for the area of the pipe.
π 2
Q = vavg r
2
with Q as the volumetric flow rate, v avg as the average velocity, and r as the pipe’s radius.
It should be noted that all the equations apply to incompressible fluids only, but these can be used to approximate gas flows at moderate
velocities. This flow meter must also be placed at least 100 pipe diameters in distance, downstream of the nearest flow obstruction. This
ensures no unwanted pressure fluctuations and accurate pitot tube readings. Furthermore, fluids containing large particles can clog the pitot
tube and should be avoided.
3.5.2.2: DIRECT FORCE
These flow meters are governed by balancing forces within the system.
3.5.2.2.1: ROTAMETER
A rotameter is a vertically installed tube that increases in diameter with increasing height. The meter must be installed vertically so that
gravity effects are easily incorporated into the governing equations. Fluid flows in through the bottom of the tube and out through the top.
Inside the glass tube there is a float that changes position with the flow rate. When there is no liquid flow, the float rests in the bottom of the
meter.
The applied concept for the rotameter is differential area. As the flow rate of the fluid changes, the position of the float changes and annular
area change directly, keeping a constant pressure drop across the meter. Changes in float position and annular area are approximately linear
with changes in flow rate. Upon achieving a stable flow rate, the vertical forces are balanced and hence the position of the float remains
constant. The volumetric flow is given by the following equation:
−−−−−−−−−−−−−−−− −

 Vf (ρf −ρ)
 2g ( − ρhf )
 Af
Q = CAb 
 2
 Ab
ρ [1 − ( ) ]
⎷ Aa
with C being the discharge coefficient, Ab being the cross sectional area of the top of the float, Vf volume of the float, ρf the density of the
float, ρ the density of the fluid hf the height of the float, Aa the cross sectional area of the bottom of the float.
Generally, rotameters are inexpensive and simple to use. This allows them to be used in many plant applications.
3.5.2.2.2: TURBINE METER
A turbine wheel is placed in a pipe that holds the flowing fluid. As the fluid flows through the turbine, the turbine is forced to rotate at a
speed proportional to the fluid flow rate. A magnetic pick-up is mounted to the turbine wheel, and a sensor records the produced voltage
pulses. Voltage information can then be translated into the actual flow meter reading.
The following is the equation used to model the turbine meter:

2 2
ω(r̄ ) A
Q = VA =
r̄ A tan β − 0.037Re−0.2 n (Ro + Ri ) D sin β
with A the pipe area, the root mean squared radius, ω rotational speed,β the angle between the flow direction and the turbine blades,Ro the
outer blade radius, Ri the inner radius, and D the distance between blades.
There are two main advantages of the turbine meter over conventional differential head devices
1. The extended are more accurate registration of flow in the low flow range of process operation. This results from the registration being
proportional to the velocity rather than the velocity square
2. The comparatively low head loss across the meter
Another advantage to using this type of flow meter is reliability. Extensive testing has proven these results. Additionally, the turbine flow
meter does not have a high installation cost. However, due to the turbine wheel motion, a low to medium pressure drop can result. Turbine
wheel replacement may also be required due to abrasion caused by particles within the fluid.
3.5.2.2.3: PROPELLER FLOW METER
Propeller flow meters have a rotating element similar to the wheel in turbine meters. Again, rotation is caused by fluid flow through the
propeller, and voltage pulses are created as the propeller passes a magnetic or optical sensor. Similarly, the frequency of the pulses is
proportional to flow rate of the fluid and the voltages can be directly correlated with the fluid flow rate.
Propeller flow meters are often used specifically with water, though other fluids may also be used. Low cost coupled with high accuracy
make propeller flow meters a common choice in many applications.
3.5.2.2.4: PADDLE WHEEL SENSORS
A kind of propeller sensor is the paddle wheel sensor. Paddle wheel sensors are similar to turbine sensors, except for one thing. The shaft of
the paddle wheel sensor is perpendicular to the flow of the fluid while the turbine sensor’s shaft is parallel to the flow of the fluid. This adds
several advantages for the paddle wheel flow sensor. Due to the shaft being perpendicular to the flow, it sustains less axial from the fluid,
and thus less friction. Paddle wheel sensors also have a smaller number of blades, so there is less force needed to turn the paddle wheel.
This means that a paddle wheel can be accurate at lower flows, have a high efficiency, as well as a longer lifetime.
There are two kinds of paddle wheel sensors, insertion and inline sensors. There is more than one design for an insertion sensor, but one
popular design has the bearing built into the rotor and the shaft sliding through it as the center axis of spin. The blade sticks out and is
inserted into the pipe with the flowing fluid. An inline paddle wheel sensor is used for smaller pipes. It contains a rotor assembly with the
magnet sealed inside and a main body.
3.5.2.2.5: CORIOLIS MASS FLOW METER
A Coriolis flow meter harnesses the natural phenomenon wherein an object will begin to “drift” as it travels from or toward the center of a
rotation occurring in the surrounding environment. A merry-go-round serves as a simple analogy; a person traveling from the outer edge of
the circle to its center will find himself deviating from his straight-line path in the direction of the ride’s rotation.
Coriolis flow meters generate this effect by diverting the fluid flow through a pair of parallel U-tubes undergoing vibration perpendicular to
the flow. This vibration simulates a rotation of the pipe, and the resulting Coriolis “drift” in the fluid will cause the U-tubes to twist and
deviate from their parallel alignment. This Coriolis force producing this deviation is ultimately proportional to the mass flow rate through
the U-tubes.
Fc
Mass Flow =
2wx
where F is the Coriolis force observed, w is the angular velocity resulting from rotation, and x is the length of tubing in the flow meter.
c
Figure 3.5.1: (left) The vibration pattern during no-flow and (right) The vibration pattern with mass flow. (CC BY-SA 2.5; Cleonis via
Wikipedia)
Because the Coriolis flow meter measures the mass flow rate of the fluid, the reading will not be affected by fluctuations in the fluid density.
Furthermore, the absence of direct obstructions to flow makes the Coriolis flow meter a suitable choice for measuring the flow of corrosive
fluids. Its limitations include a significant pressure drop and diminished accuracy in the presence of low-flow gases.
3.5.2.3: FREQUENCY
These flow meters use frequency and electronic signals to calculate the flow rate.
Vortex Shedding Flow Meter
A blunt, non-streamline body is placed in the stream of the flow through a pipe. When the flow stream hits the body, a series of alternating
vortices are produced, which causes the fluid to swirl as it flows downstream. The number of vortices formed is directly proportional to the
flow velocity and hence the flow rate. The vortices are detected downstream from the blunt body using an ultrasonic beam that is
transmitted perpendicular to the direction of flow. As the vortices cross the beam, they alter the carrier wave as the signal is processed
electronically, using a frequency-to-voltage circuit. The following diagram shows the basic principle of the vortex-shedding flow meter:
Vortex-shedding flow meters are best used in turbulent flow with a Reynolds number greater than 10,000. One advantage of using this type
of flow meter is its insensitivity from temperature, pressure, and viscosity. The major disadvantage to using this method is the pressure drop
caused by the flow obstruction.
3.5.3: ULTRASONIC FLOW METERS

There are two types of Ultrasonic meters, the transit time/time of flight and Doppler models, both of which have unique equations
representing the principles behind them. The basis for these meters is monitoring ultrasonic waves in fluid passing through a pre-configured
acoustic field. These meters are based on the technique of sound waves that change.
3.5.3.1: TRANSIT TIME/TIME OF FLIGHT FLOW METERS

Transit time meters have two opposing transducers outside of the pipe to measure the time of a signal sent from a transducer upstream to a
transducer downstream and vice versa.
This allows the average velocity and hence the flow rate, Q, to be determined.
3
πd tan α 1 1
Q = K ( )( − )
8 tU D tDU
where d is the diameter of the pipe, α is the angle between direction of the flow and the pipe, tUD is the time for the signal to reach
downstream transducer from the upstream transducer, and tDU is the time for signal to reach upstream transducer from the downstream
transducer.
With the Time of Flight Ultrasonic Flowmeter the time for the sound to travel between a transmitter and a receiver is measured. This
method is not dependable on the particles in the fluid.
Two transmitters / receivers (transceivers) are located on each side of the pipe. The transmitters sends pulsating ultrasonic waves in a
predefined frequency from one side to the other. The difference in frequency is proportional to the average fluid velocity.
Downstream pulse transmit time can be expressed as
L
td =
c + v cos Φ
where t is the downstream pulse transmission time and L is the distance between transceivers.
d
Downstream pulse transmit time can be expressed as
L
tu =
c − v cos Φ
where t is the upstream pulse transmission time.

u
Since the sound travels faster downstream than upstream, the difference can be expressed as
\[t = \dfrac{t_d - t_u) t = \dfrac{ 2 v L \cos Φ}{c^2 - v^2 \cos 2Φ} t = \dfrac{2 v L \cos Φ}{c^2} \label{4} \]
(since v is very small compared to c )
3.5.3.2: DOPPLER METERS

Doppler meters use the frequency shift of an ultrasonic signal when it is reflected by suspended particles or gas bubbles (discontinuities) in
motion. The Doppler Effect Ultrasonic Flowmeter uses reflected ultrasonic sound to measure the fluid velocity. By measuring the frequency
shift between the ultrasonic frequency source, the receiver, and the fluid carrier, the relative motion are measured. The resulting frequency
shift is named the Doppler Effect.
These reflected signals travel at the velocity of light.

Vs Δf
V; =
2factual cos α
where f is the actual frequency and Δf is the change in frequency or frequency shift.
The fluid velocity can be expressed as
c(fr − ft )
v =
2ft cos Φ
where
fr is the received frequency
ft is the transmission frequency
v is the fluid flow velocity
Φ is the relative angle between the transmitted ultrasonic beam and the fluid flow
c is the velocity of sound in the fluid
This method requires that there are some reflecting particles in the fluid. The method is not suitable for clear liquids.
ADVANTAGES WITH THE DOPPLER EFFECT ULTRASONIC FLOWMETER
Doppler meters may be used where other meters don't work. This could be in liquid slurries, aerated liquids or liquids with some small or
large amount on suspended solids. The advantages can be summarized to:
Limitations with Doppler Effect Ultrasonic Flowmeters
3.5.3.3: BENEFITS WITH ULTRASONIC FLOWMETERS AS A WHOLE
Doppler flowmeters performance are highly dependent on physical properties of the fluid, such as the sonic conductivity, particle
density, and flow profile.
Non uniformity of particle distribution in the pipe cross section may result in a incorrectly computed mean velocity. The flowmeter
accuracy is sensitive to velocity profile variations and to the distribution of acoustic reflectors in the measurement section.
Unlike other acoustic flowmeters, Doppler meters are affected by changes in the liquid's sonic velocity. As a result, the meter is also
sensitive to changes in density and temperature. These problems make Doppler flowmeters unsuitable for highly accurate measurement
applications.
Obstruction less flow
Pressure drop equal to an equivalent length of straight pipe
Unaffected by changes in temperature, density or viscosity
Bi-directional flow capability
Low flow cutoff
Corrosion-resistant
Both meters are effective in measuring open channels and partially filled pipes but are very sensitive to flow conditions and hence should be
calibrated with care. Also, there is no pressure drop since there are no obstructions in the flow path.
3.5.3.4: OTHER TYPES

3.5.3.4.1: MAGNETIC FLOW METER
One magnetic model flow meter positions electric coils around the pipe of the flow to be measured. A pair of electrodes is set up across the
pipe wall. The fluid flowing has a minimum value of electrical conductivity, the movement of the fluid through the pipe acts as a conductor
moving across the magnetic field. There is an induced change in voltage between the electrodes, which is proportional to the flow velocity.
The flow velocity is found by measuring the changes of induced voltage of the conductive fluid passing through a controlled magnetic field
at right angles. According to Faraday’s Law, the magnitude of the voltage induced is directly proportional to the product of the magnetic
flux, distance between probes and the velocity of the medium (fluid).
dϕ dA dz
E = −N = −N B = −N BD = −N BDv
dt dt dt
where E is the voltage of induced current, N is the number of turns, B is the external magnetic field, φ is the magnetic flux, D is the
distance between electrodes and v is the velocity of the fluid.
Some of the advantages are minimum pressure drop because of minimum obstructions in flow path; low maintenance cost because of no
moving parts. One of the disadvantages is that it usually requires a fluid with an electrical conductivity of higher than 3 µS/cm.
3.5.3.4.2: CALORIMETRIC FLOW METER
This type of flow meter is suitable for monitoring the flow rates of liquid and gaseous mediums. The calorimetric principle is based on two
temperature sensors in close quarters of one another but thermally insulated from one another.
One of the two temperature sensors is constantly heated and the cooling effect of the flowing fluid is used to monitor the flow rate. In a
stationary phase fluid condition there is a constant temperature difference between the two temperature sensors. When the fluid flow
increases, heat energy is extracted from the heated sensor and the temperature difference between the sensors are reduced. The reduction is
proportional to the flow rate of the fluid. The calorimetric flow meter can achieve relatively high accuracy at low flow rates.
Common applications: air compression, argon compression, nitrogen compression, carbon dioxide compression and flow detection of all
fluids ( liquids and gases)
3.5.3.4.3: GEAR FLOW METER
This type of flow meter has oval shaped gears with fitting teeth which control the amount of fluid passing through. The flow rate is
calculated by number of times the gears are filled and emptied. These meters have high accuracy and are used for measuring low flow and
for high viscosity fluids. It is very easy to install these types of meters because it requires no pipe.
3.5.3.4.4: THERMAL FLOW METERS

These flow meters take advantage of the thermal properties of the fluid to measure the flow of the fluid in a pipe. In most thermal flow
meters, a measured amount of heat is applied to the heater of the sensor. Portions of this heat is lost to the fluid as it flows. Accordingly, as
the fluid flow increases more heat is lost. The amount of heat lost is tracked using a temperature measurement instrument in the sensor.
Then an electronic transmitter uses the heat input and temperature measurements to calculate the fluid flow, taking into account the thermal
properties of the fluid.
Common applications of thermal flow meters are measuring the flow rate of clean gases like air, nitrogen, hydrogen, helium, ammonia, and
argon. Most thermal flow meters are used to measure pure gases that would be used in laboratory experiments or semiconductor production.
Mixtures like flue stack flow can also be measured but the mixture compositions must be known to use the appropriate thermal properties of
each component in the mixture. The key advantage of this technology is its dependence on the thermal properties which are mostly
independent of the gas density.
However caution should be taken if this type of flow meter is used to measure a fluid of unknown or varying composition. Additionally,
thermal flow meters should not be used to measure abrasive fluids because they may damage the sensor. Some fluids can coat the sensor
causing it to be inoperable and must be regularly cleaned to be useful.
3.5.3.5: TABLE OF FLOW METERS

The following table outlines specifics for most of the flow meters. This may be helpful in answering example 1. Accuracy is given in terms
of FSD (full scale deflection) which is the deflection of meter's pointer to the farthest point on the scale. This implies the highest
measurement.
3.5.4: FLOW PROFILE DISTORTION
In the real world, the flow profile is not always symmetrical. Pipe fittings such as elbows, tee-pieces, and reducers can change the flow
profile. One example of a fitting that alternates the flow profile is a sharp elbow which causes pure swirls throughout the fluid. Some flow
meters are more sensitive to particular types of flow distortion. More complex flow conditions produce better velocity profile but there is a
trade off since they are more expensive and give higher pressure drops.
3.5.5: TURNDOWN RATIO

The turndown ratio is a term used to describe the range of accurate operability of a specific flow meter. This rangeability is critical when
selecting flow meters for specific applications. Typically, in a plant setting gas flow may not be constant, but accurate measurement of gas
flow is needed ranging from no flow to full flow.
For example if nitrogen gas is being used in a plant conducting multiple batch reactions, sometimes the little nitrogen (100 m3/min) will be
needed and other times full nitrogen flow (1000 m3/min) will be needed. For this system, the turndown ratio is 10:1. Accordingly, a flow
meter must have a turndown ratio of at least 10:1.
For each type of flow meter, the turndown ratio is limited by the theoretical and physical constraints. For example, for an orifice meter the
accuracy may be compromised near the limits of the rangeability. Orifice meters create a pressure drop in the measured fluid which is
proportional to the velocity squared. If the range of differential pressures becomes too large the overall accuracy of the flow meter at its
range limits may be inconsistent.
Turndown ratio can be expressed as:
T R = qmax/qmin (3.5.1)
where
The table below shows a list of typical turndown ratios for different flow meters.
Example - Turndown Ratio for an Orifice Meter

The turndown ratio - TR - for an orifice meter with maximum flow of 12 kg/s and a minimum flow of 3 kg/s can be calculated as:
TR = (12 kg/s) / (3 kg/s) = 4 - normally expressed as turndown ratio of 4:1
This is a typical turndown ratio for a orifice plate. In general a orifice plates has turndown ratio between 3:1 and 5:1.
Turndown Ratio and Measured Signal
The graph above shows how the turndown ratio effects the measured signal % and flowmeter %. With an increased turndown ratio there is a
larger range that the flowmeters can operate within. In a flow meter based on the orifice or venturi principle, the differential pressure
upstream and downstream of an obstruction in the flow is used to indicate the flow. According the Bernoulli Equation the differential
pressure increases with the square of flow velocity. A large turndown ratio will cramp the measurement signal at low flow rate and this is
why venturi and orifice meters are not accurate at low flowrates.
 EXAMPLE 3.5.1
a) Oil is flowing in a medium sized pipe which is insensitive to temperature and pressure. Pressure drop measured is relatively
low. Which type of flow meter is recommended?
Coriolis flow meter because it effectively measures flow rates of high viscous liquids at a low pressure drop. Coriolis flow meters are
also insensitive to temperature and pressure.
(b) Why is Doppler only recommended for liquids containing suspended solids or gases?
The Doppler sensor transmits sound into the flow stream, which must be reflected back to the sensor to indicate flow velocity. Gas
bubbles or suspended solids act as reflectors for the Doppler signal. Doppler flow meters are recommended for liquids containing solids
or bubbles with a minimum size of 100 microns and a minimum concentration of 100 ppm.
(c) When would you use Vortex flow meters?
Vortex is mainly used to measure steam flow rates. They are also used to measure gases and liquids with low viscosity that contains
hardly any solids.
(d) A conductive liquid is flowing in a pipe (1000 mm size) at the temperature of 430K. Which type of flow meter is
recommended to measure the flow rate?
Electromagnetic flow meter is used in this case as it measures flow rates of conductive liquids.
 EXAMPLE 3.5.1
Kerosene at 20 deg C flows at 18 m3 / h in a 5 cm diameter pipe. If a 2 cm diameter thin plate orifice with corner taps is installed, what
will the measured drop be in Pa? Why is orifice meter used in this case?
Data given: Cd=0.6030; μ = 192E-3 kg/ (m s) for kerosene; ρ = 804 kg/m^3
Solution
Orifice meter is used in this case because Kerosene is a clean and non viscous liquid.
Usually, pipe velocity is calculated using the velocity formula to calculate Re and Cd values. The pipe velocity is:
18
Q ( )
3600
v = = = 2.55m/s
π 2
A ( ( 0.05 )
4
In this case, we already have the Cd. So,using orifice pressure drop formula, calculate
2 −−−−
18 (2π)(0.02) 2Δp 3
Q = = 0.6030 √ = 0.005m /h
−−−−−−−−
3600 (0.04)
2
804
√1 − 2
(0.02)
Final step is to solve for Δp to give 273 kPa.
2
(0.04)
2
(0.005) ∗ (1 − 2
) (804)
(0.02)
Δp = = 273kP a
(0.6030)2 (2π)2 (0.02)4 (2)

Selection of an Industrial Flow Sensor
3.5.7: REFERENCES

TR = Turndown Ratio
qmax = maximum flow
qmin = minimum flow
Cole-Palmer Catalog, 2005-2006
[5]Engineering fundamentals' website
Perry, Robert H., Perry's Chemical Engineers' Handbook, 7th Edition. McGraw-Hill, 1997.
Richardson, J.F. and Peacock, D.G., Chemical Engineering, 3rd Edition, Volume 3. Butterworth- Heinemann, 1994.
[6]Process Measurement and control Products
[7]Universal Flowmeters Presents Flowmeters.com
[8]Transactions Volume 4: Flow and Level Measurement
[9] Wikipedia entry on Coriolis Mass Flowmeters
[10]Omega Literature: Flow and Level Measurement
Geankoplis, Christie John, Transport Processes and Separation Process Principles, 4th Edition, Pearson Education, 2003.
McKetta, John J., Cunningham, William A., Encyclopedia of Chemical Processing and Design: Volume 43 - Process Control: Feedback
Simulation to Process Optimization, CRC Press, 1993.
Authors: (September 14, 2006) Ashwini Miryala, Kyle Scarlett, Zachary Zell, Brandon Kountz
Stewards: (September 5, 2007) Jeremy Goldfarb, Daniel McClure, Lilian Ngobi, Daniel Wheeler
3.5: Flow Sensors is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source content that was edited to
3.6: COMPOSITION SENSORS
3.6.1: INTRODUCTION
There are many useful analytical tools, such as photometric analysis, electrometric analysis, chromatography, mass spectrometry, thermal
conductivity, and various physical property measurements (density and specific gravity), which can be used to determine the composition of
mixtures. A wide array of methods to measure composition are available, so it is important to choose the best method given a set of
conditions. The first step is to decide between measuring physical or chemical properties to determine the composition. If you choose to
measure a physical property, make sure that it is unique to the desired component of the mixture and will accurately allow you to determine
the composition. The goal of this article is to explain the various analytical methods and tools used to determine the composition of a given
sample. After reading this, you should be able to determine which method of composition measurement is most appropriate for a given
circumstance.
3.6.1.1: TYPES OF TESTING: ON-LINE V. OFF-LINE

On-line analysis is the continuous monitoring of the composition of a sample, which is under the influence of a control system and directed
by an actuator which can respond and regulate the operating conditions in real time such that the desired set points are maintained. On-line
testing can either be performed in-line or by slip stream testing. In on-line testing, the sensor is attached directly to the line and provides
feedback via a transmitter. On the other hand, in slip stream testing, a side stream of the process runs alongside the main line. Such an
apparatus closely resembles the set up of a bypass. The slip stream process conditions can be continuously manipulated to make the
measurements easier to obtain. Similar to in-line testing, the sensor is directly attached to the slip stream and provides feedback through a
transmitter. Advantages of on-line analysis include an immediate and continuous feedback responding to changes in process conditions. The
main disadvantage of on-line testing is that it is usually much more complicated and more expensive than off-line testing. Also, off-line
testing is more robust and has more varied applications, while on-line testing may not work in every situation. On-line testing prevents the
continued production of undesired product, with an immediate response and correction of the flawed material.
Off-line analysis involves the extraction of a sample from the process or reaction, and its subsequent testing in a machine that may be
situated at a location far away from the process line in a lab. In this case, a sample is manually removed which is later sent to the
composition analyzer. The results of the analysis are examined and then they are sent to the control system or actuator to make the
appropriate adjustments. On-line analysis is comparable to off-line testing, the primary difference being that in on-line testing, the samples
are analyzed on a machine that is next to the process line. This greatly reduces time lost in transporting the sample, though it still permits
the prospect of introducing contaminants into the sample. In off-line analysis, some of the disadvantages include sample dead-time, which is
the time lost during transportation, variability of sample testing locations, and lag time for adjustments to be made to the process. In off-line
analysis, the lag time between removing the sample from the system and receiving the results of the test could cause significant losses to the
company, since the defective product is produced and the process is not corrected until the results are received. This prolonged defective
processing could result in the accumulation of losses worth thousands of dollars, which could have been avoided if an on-line analysis
system was set up in place.
While both methods are widely used in industry, there is a push towards more on-line testing. A paper published in 2003 urging drug
companies to adopt more in-process analytical testing such as on-line testing, endorsed by the Food and Drug Administration reflects this
trend. This has been a part of a broader effort by the FDA to encourage companies to move towards better manufacturing practices. While
on-line testing maybe comparatively expensive to install at first, the savings from the process would not only recoup the costs associated
with installation, but also prevent significant losses. Many companies have adopted lean manufacturing techniques, and one of its hallmarks
is to correct a defect as early in the production process as possible, not allowing defective products to move on to the next processing step.
Allowing defective products to perpetuate through the system may even contaminate other non-defective materials in the production
process.
3.6.1.2: STANDARDS AND CALIBRATION CURVES

Before discussing the different ways to measure the composition of a sample, one needs to understand that composition sensors use
standards or calibration curves to measure an unknown composition against a known one(s). The standard can either be an internal standard
against which all other measurements are compared, such as a reference cell or a series of standards used to create a calibration curve. With
a calibration curve, a series of known standards are measured with the sensor and the signal produced by the sensor is graphed on the y-axis
of a plot; the known information about each sample is graphed on the x-axis. From this information, a relationship can be developed
between the signal output from the sensor and the known quantity that you are trying to measure. The data measured for the unknown
sample can then be compared to the calibration curve graphically or using the equation written to describe the curve. Using this information,
the measurement made by the sensor can be interpreted as a useful composition measurement of weight percent, volume percent, mole
percent, etc.
3.6.2: PHOTOMETRIC ANALYSIS
Photometric analysis is the measurement of the intensity of visible light and other electromagnetic (EM) waves. Through the measurement
of these values, the composition of samples or flows can be determined in various different ways. A wide range of photometric devices
based on many differing principles are used in the chemical engineering industry for this purpose. A few of the most common instruments
are covered in this article.
3.6.2.1: TYPES OF PHOTOMETERS

A photometer is any type of instrument used to measure characteristics of light. There are two broad categories of photometers: dispersive
and non-dispersive.
Dispersive Photometers
In dispersive photometers, the light from the source is dispersed and a narrow spectral band is selectively directed to the sample and
detector. A monochromator is usually the tool that performs this function. The dispersing element of the monochromator can be a diffraction
grating, prism, and interference filters. Diffraction gratings are the most common; for each position of the grating a narrow band of
dispersed light passes through an exit slit. Dispersive devices can scan across a spectrum and make measurements at several wavelengths.
This capability allows them to be used for the analysis of multiple components.
Non-dispersive Photometers
Non-dispersive photometers use a narrow-band-pass filter to block out a large amount of undesired radiation. They make measurements at
selected discrete wavelengths. The filter passes radiation at the selected reference and measurement wavelengths. The reference wavelength
filter selects a band where none of the components present in a process stream absorb radiation; the measurement wavelength filter is
selected to match the absorption band of the component being analyzed. The ratio of the transmitted light at the reference and measurement
wavelengths is measured by the photometer. Non-dispersive photometers are usually used to measure a single component in the process
stream.
An example of such a device is the Non-Dispersive Infrared (NDIR) analyzer that is frequently used to measure the concentrations of
certain gases in a mixed gas flow. The NDIR analyzer uses a reference non-absorbent gas such as O2, Cl2 and N2 and compares that to a
sample gas like CO, CO2, SO2, CH4, etc. Two beams of infrared radiation are used: a reference beam for the non-absorbent gas and an
analyzing beam for the sample.
3.6.2.2: SPECTROPHOTOMETERS
Spectrophotometers are widely used as composition sensors. Spectrophotometers are photometers that measure the intensity of light at
specific wavelengths. Many different types of spectrophotometers exist, and they are typically categorized based on the ranges of
wavelengths they monitor: ultraviolet (280-380 nm), visible light (400-700 nm), infrared (0.7-1000 µm), or microwave (1-300 mm). Other
important distinctions are the measurement technique, the method of obtaining the spectrum, and the source of intensity variation.
The emission spectrum recorded by the spectrophotometer typically consists of a series of narrow peaks of varying heights, also known as
spectral lines. The height or amplitude of those lines can be related to the concentration or abundance of the specific material through a
calibration curve and their position on the spectrum distinguishes the components of the material. While there are numerous methods for
obtaining a spectrum, the most common application of a spectrophotometer is for the measurement of light absorption.
3.6.2.2.1: ABSORPTION
The degree of absorption of electromagnetic waves exhibited by the components in the sample is a distinctive feature that can be used to
reveal the composition of the sample. Absorption is related to the inverse of transmittance by:
1
A = log( )
T
where A is the Absorbance (Absorbance unit, AU) and T is the Transmittance (as percentage).
Spectrophotometers based on absorption have a radiation source that will emit a specified range of EM waves to be absorbed by the sample
by measuring the intensity of light that passes through the sample. A useful relationship used to quantitatively determine the concentration
of an absorbing species in solution is the Beer-Lambert Law:
I
A = − log( ) = ϵcL
I0
where
I = Intensity of incident light at a particular wavelength
I0 = Intensity of transmitted light
epsilon = molar absorptivity or extinction coefficient (1 / M * cm)
c = concentration of the absorbing species
L = path length through the sample
According to the Beer-Lambert Law, the absorbance of a solution is directly proportional to its concentration. The constant epsilon is a
fundamental molecular property for a given solvent. It can be found from references tables or determined through the construction of a
calibration curve. The Beer-Lambert Law is useful for characterizing many compounds, but does not hold for all species. For example, very
large, complex molecules such as organic dyes often exhibit a 2nd order polynomial relationship between absorbance and concentration.
Two widely-used devices that measure the absorbance of samples are Ultraviolet and Infrared spectrophotometers.
3.6.2.3: ULTRAVIOLET (UV) SPECTROPHOTOMETERS
UV spectrophotometry is a useful method for detecting and quantifying substances in solution. Certain compounds in the gaseous or liquid
state absorb ultraviolet radiation, resulting in absorption peaks that can be correlated with specific functional groups. These species include
aromatic and carbonyl compounds in organic substances and transition metal ions in solution. This property is useful for categorizing and
measuring the composition of these types of samples, however this property cannot be used as a specific test for a given compound. The
nature of the solvent, the pH of the solution, temperature, high electrolyte concentrations, and the presence of interfering substances can
influence the absorption spectra of compounds, as can variations in slit width in the spectrophotometer. Many organic solvents have
significant UV absorption and therefore will add peaks to the species dissolved in them. Solvent polarity, high electrolyte concentration, and
pH can effect the absorption spectrum of an organic compound by causing protonation or deprotonation of the compound. Temperature can
also cause complex molecules to change in conformation and can also alter the properties of the solvent used. Variations in the slit width
will alter the effective bandwidth, distorting wavelength measurements.
The basic components of the overall analyzer are the UV light source, sample holder, optical filters, and detector (typically a phototube).
Optical filters (typically a monochromator or diffraction grating) separate the wavelengths of light going through the sample and ensure that
only the desired wavelengths reach the detector.
Most UV analyzers are dispersive photometers that also function in the visible light spectrum and are called UV-Vis spectrophotometers.
Since these instruments are generally found in laboratories, they are generally used for offline analysis. However, there are inline UV
machines and sensors available for real-time inline measurements of process flow streams. UV spectrophotometers used throughout many
industries for various applications.
3.6.2.4: INFRARED (IR) SPECTROPHOTOMETERS

Infrared spectroscopy is based on the fact that there are specific frequencies at which molecular bonds rotate or vibrate when exposed to
discrete energy levels of radiation. A beam of infrared light is passed through the sample, either one wavelength at a time with a
monochromatic beam or using a Fourier transform instrument that measures all wavelengths at once. Inspection of the transmitted light
reveals how much energy was absorbed at each wavelength, this data is then plotted (see example below). This plot allows the frequency of
the vibrations to be associated with a particular bond type. The characteristic IR absorption spectra obtained for a sample can be used to
identify the compound, while the amplitude of absorbance is proportional to the analyte concentration. IR analyzers are used to analyze
gaseous or liquid compounds that can absorb IR waves. The system components are generally similar to those of an UV spectrophotometer,
with the exception of the IR light source.
IR instruments have been used regularly in a fixed laboratory setting for decades now. Both dispersive and non-dispersive IR techniques
exist. More recently, manufacturers of instruments have significantly reduced their size and power requirements. In addition to inline IR
sensors for process streams, portable instrumentation is available for field analysis such as the monitoring of stack gases. Another relatively
modern technology, the Fourier Transform Infrared (FTIR) Spectrometer, is widely used in industry today because of its high sensitivity and
broad range of application. For example, FTIR analysis is used in the quantification of harmful organic peroxides in lubricants and fuels.
3.6.2.5: RADIATION
Instead of measuring the absorption of light, many spectrophotometers measure the radiation of EM waves. Samples may radiate EM waves
in many forms such as heat, luminescence, fluorescence, and phosphorescence. The measurement of such radiation is often employed in
biological and pharmaceutical areas because of its high sensitivity and specificity. For example, fluorescence spectrophotometers are used to
assay the presence and concentration of target biological compounds. Fluorescent probe molecules are introduced to the sample that will
attach selectively to the biological molecule of interest and the device will then detect and measure the amount of fluorescent radiation. In
these experiments, the appropriate standards must be measured to properly calibrate the radiation measured with sample concentration.
Additionally, contamination of the sample can affect the radiation patterns.
Examples of other techniques utilizing the radiation EM waves include flame emission spectroscopy, particle-induced X-ray emission, and
X-ray fluorescence. These methods are strictly used for laboratory analysis.
3.6.2.6: PHOTOMETRY USING VISIBLE LIGHT

The manner in which certain materials interact with visible light can be used to analyze the composition of the sample material.
Refractometers, turbidimeters, and opacity meters are all analyzers which utilize this principle.
3.6.2.6.1: REFRACTOMETRY
The theory behind refractometry is that visible light bends (or refracts) when it passes two opaque media of differing densities. Snell’s Law
relates the angle of incidence α to the refractive index η by:
n1 α2
=
n2 α1
α1, α2 = angles of incidence

η1, η2 = refractive indices of the two media (unitless)
Alternatively, the critical angle may be used to analyze the composition. The critical angle is defined as the angle of incidence that produces
a 90o refraction with respect to normal of the interface i.e. α2 = 0o. Thus, Snell’s Law becomes:
n1 1
=
n2 sin αc
αc = critical angle
Please refer to the diagram below.
Figure 1: Light refraction at a glass-medium interface

Different materials will have different refractive indices; the experimental refractive index measured can be compared to a list of refractive
index values at a data bank. Continuous process or online refractometers generally measure the changes in the critical angle at a glass-
sample interface caused by the variations in composition concentrations. Refractometers are used in numerous industries such as food,
chemical, metalworking industry because of its high accuracy and repeatability, low maintenance and easy installation. An example is the
use of a refractometer to monitor the amount of juice concentrate in juice.
3.6.2.6.2: TURBIDIMETERS
Turbidimeters measure the turbidity, or sample clarity, of fluids. A light beam passes through the analyzer and if the water, or other
chemical that is to be tested, is "pure," then the light would travel undisturbed. However if the sample contains suspended solids or particles,
the light would interact with the particles causing the particles to absorb the light and reflect it into different directions. An example is
smoke in air that gives the air a hazy look.
Most modern turbidimeters are nephelometers. Nephelometers or nephelometric turbidimeters measure the scattered light reflected off the
suspended particles in the sample. This instrument contains a light source, some sort of container to hold the sample, and photodetectors to
sense the scattered light. A tungsten filament lamp is most commonly used as a light source. If the suspended particles, typically measured
in nanometers, are smaller than the wavelength than the incident light, the reflected light is scattered equally in all directions. If the particles
are larger than the wavelength of the incident light, the pattern of scattering will be focused away from the incident light.
Nephelometers are usually used when solids are present in the sample in small concentrations. When there is high turbidity in a sample,
multiple scattering occurs and the turbidimeter may no longer be able to analyze the fluid. Disadvantages of nephelometers include being
prone to fouling and incapability of analyzing fluids that have high concentrations of solid particles.
Turbidimeters are important to industry, specifically for chemical plants, oil refineries, and waste water treatment plants. For example,
nephelometers are used in detecting the pollutants in water in water quality monitoring.
3.6.2.6.3: OPACITY MONITORS
Opacity monitors measure the attenuation of light due to scattering and absorption by the sample. Attenuation is defined as the decrease in
intensity and amplitude of the measured light beam when compared to the emitted light beam. Opacity monitors differ from turbidimeters in
that they analyze samples by measuring the percentage of transmission of the light source that passes through the sample and NOT the
scattering of the light source. A common application is the measurement of smoke density in chimney stacks. The density of the pollutant or
particle is expressed as percent opacity, percent transmittance, or optical density.
% opacity = 100 x opacity = 100 – % transmittance
3.6.3: ELECTROMETRIC ANALYSIS

Electrometric analysis uses the principles of electrochemistry to analyze the composition of different samples.
3.6.3.1: CONDUCTIVITY CELLS

Conductivity cells measure a liquid’s ability to conduct an electric current. Since liquids conduct current due to the presence of ions, the
composition can be determined by measuring the concentration of the ions present. Conductivity cells determine concentration from the
definition of molar conductivity (lambda). Molar conductivity is a property that can be influenced by temperature. It also varies with
concentration if the chemical being measured is a weak electrolyte. It is defined by the following equation:
Λ = molar conductivity [Sm2/kmol]

κ = conductivity [S]
C = concentration [kmol/m3]
Since it is difficult to directly measure conductivity, conductivity cells measure the resistance (R) of a sample liquid between two electrodes
when a small current is applied. Resistance is related to conductivity by the following equation:
κ = conductivity [S]
A = area of electrode [m2]
x = distance between electrodes [m]
R = resistance [ohm, Ω]
Note that the electrode area, A, and distance between the two electrodes, x, are fixed for a particular sensor and are often expressed as a cell
constant (kc).
The following flow chart shows how composition is determined in terms of concentration:
Conductivity cells can be used in flow system and installed directly into the system of pipelines. Conductivity cells can come in three types:
two electrode probe, multiple electrode probe, or a non-contact, electrode-less probe (toroidal). The two electrode probe is one of the most
common types and typically uses carbon or stainless steel as the electrodes. The multiple electrode probed has either two, four, or six
electrodes. This probe has the advantage of reducing errors due to polarization (electrochemical reaction on the electrode surface) and
fouling. Finally, the toroidal probe has no electrodes and instead measures the AC current in a closed loop of solution using two embedded
toroids (a magnetic core with wire wrapped around it to make an inductor). The electrode-less probe is good for solutions that tend to have
problems with polarization or oily fouling. In-line conductivity cells are suitable for temperatures up to 480K and pressures up to 1700
kN/m2.
Since concentration is found from molar conductivity, which is a property of the fluid that is strongly dependent on temperature, a
temperature sensor is often paired with a conductivity cell so temperature effects can be accounted for in concentration calculations.
Conductance cells are best used in applications where a single type of ion is present in a dilute solution. Higher concentrations lead to
complications that cannot be described by the simplistic linear relation between concentration and conductivity, since molar conductivity
starts to vary with higher concentrations of weak electrolytes.
3.6.3.2: ION SELECTIVE ELECTRODES

pH Electrode
The pH electrode is commonly used in both the lab and industrial setting. This electrode is sensitive to the concentration of the H + ions. The
pH electrode basically measures the potential difference between a reference solution and the sample. Please refer to the pH sensor section
for examples and more information on relevant theory.
Other Ion Selective Electrodes
Electrodes can also be specific to other ions such as: Na + , K + , Ag + , ClO4, BF4, SO2, and CO2. An electrode’s ion sensitivity depends
mainly on the method of separating ions in the solution. The biggest problem with ion selective electrodes is the problem of not enough
selectivity for the desired ion, and subsequent interference by other ions with similar chemical properties to the desired ion. Most electrodes
use a membrane to separate a specific ion, though other techniques exist. Selective ion electrodes can be divided into the following three
categories based on the mode of separation: solid membrane (glass or crystalline), liquid membrane (classical ion exchanger with a neutral
or charged carrier), and membrane in a special electrode (gas sensing or enzyme electrode). Examples of ion selective electrodes used in
industry include nitrate sensors for sewage treatment or river quality monitoring, hydrazine and sodium sensors for in boiler water, and
choride and fluorine sensors for in water treatment plants.
Oxidation-Reduction Potential (ORP) Sensors
ORP sensors, or also called Redox sensors, can determine concentration of ions that exist in two different oxidation states in solution. For
example, this type of electrode can be used to analyze the composition of a solution with Fe2+ and Fe3+ ions. This type of electrode
measures the potential caused by a solution’s reduction or oxidation ability.
This type of sensor directly measures electric potential which indicates the reducing or oxidizing strength of a solution. The reducing or
oxidizing power of a solution depends on whether the higher or lower oxidation state is preferred by the ion.
When an ion favors the higher oxidation state (or has reducing properties), the ions tend to give up electrons and the electrode becomes
more negatively charged.
When an ion favors the lower oxidation state (or has oxidizing properties), the ions tend to gain electrons and the electrode becomes more
positively charged.
When purchasing an ORP sensor from a vendor, many models come already packaged with a pH sensor, since these two measurements are
often both monitored to keep a process running at the right conditions. For instance, in the pool and spa industry, its important to monitor
both the concentration of sanitizers in the water with an ORP sensor, as well as measure the pH of the water.
Polarographic Sensors
Polagraphic sensors work by performing electrolysis with two types of electrodes: one which is polarizable and one which is not. They
operate very similarly to voltameters, but with voltameters the potential is changed in a controlled manner and the voltage measured. The
oldest system of polarographic measurement used dropping mercury electrodes. Here the principle is that that the surface tension of mercury
changes with how much charge is applied between the top and bottom of the electrode, and this surface tension change can be seen in how
the weight of the mercury drops varies with time.
More modern types of polarography use two different types of metals which are connected in a solution with electrolytes, and electrons will
move towards the more positively charged metal. This movement of electrons induces a current, which eventually stops once the charges
are balanced (polarized). If all the charges are balanced within the system, no current will flow. Therefore a small potential is applied to a
polarographic sensor so that a current exists for the sensor to measure.
The sample being analyzed needs to have the ability to depolarize the system, or tip the balance of charges. The ion of interest crosses a
selective membrane and reacts by an oxidation/reduction reaction with a metal surface (often called the measuring electrode). The reaction
between the ion and electrode causes an uneven distribution of charges. Once a system is depolarized, the potential of the system increases
since a more positive and a more negative region exists separately. The increase in potential causes current to flow which is the signal
measured by the sensor. This creates a graph showing the relationship between applied current versus measured potential. A disadvantage to
this type of measurement tool is that this graph requires a calibration curve in order to interpret the data, and so analysis of composition
using polarography can’t be done in-line. The advantage of this type of method is that it is highly accurate and very reproducible.
Below is a basic schematic of a polarographic sensor.
Polagraphic sensors are most commonly used assessing pollutants in air such as Cl2, HCl, HBr and HF. It an also be used to measure O2 and
other inert gases in flue.
High Temperature Ceramic Sensors
These types of sensors have a heated section of zirconium oxide (ZrO2), also known as zirconia, that is slightly doped with yttria (Y2O3).
This stable lattice structure is able to measure oxygen ion conduction at high temperatures. This sensor is exclusively used for measuring O2
and is often used to measure O2 composition in flue gases.
3.6.4: CHROMATOGRAPHY
Chromatography is a set of physical methods used to separate and thus analyze complex mixtures. Chromatography consists of a mobile
phase, usually a gas or a liquid, and a stationary phase, typically a solid or a liquid. These two phases are combined in a column of the
stationary phase based on the type of chromatography one wishes to complete (i.e. Gas-liquid, Gas-solid, Liquid) in order to separate the
mixture. The retention time of a component is the time before the component elutes from the column. The retention time and elution order
are both based on the relative interaction between each solute and each phase. The stronger the solute’s interaction with the mobile phase,
relative to the stationary phase, the sooner the solute will elute from the column. Solute-phase interaction is a function of the charges of the
molecules, the absorption of the stationary phase, and the relative solubility.
Chromatography by itself does not detect substances, but it is routinely coupled with a detection unit. Each substance is detected as leaves
the chromatograph. The retention time for each substance can then be compared to a standard to determine what substances were in the
mixture. For unknown substances, the elution order can determine some properties of the substance. For example, a substance that initially
elutes after ethanol in a nonpolar column is less polar than ethanol. However, other composition detectors are far better suited for analysis of
unknown substances.
Chromatography is the most widely used composition detecting process for on-line analysis of samples within a complex mixture. Due to its
precision, chromatography is commonly used to separate or purify samples within the chemical or biochemical industries. Depending on the
type of chromatography used, even delicate mixtures like proteins can be analyzed. However, chromatography is primarily useful when
there are some expectations for the components in a mixture; chromatography is not best for a completely unknown sample, but it is useful
for detecting the composition of a process stream where most components are known and a suitable standard exists.
3.6.4.1: TYPES OF CHROMATOGRAPHY

Gas-Liquid Chromatography (GLC)
GLC is one of the more common types of chromatography. It is commonly referred to as GC (gas chromatography). In a typical run, the
liquid mixture to be analyzed is added to the system and vaporized before entering the column. A carrier gas, often Helium, carries the
vaporized sample through a column. The column can either be packed with stationary phase, or the column can have a very small diameter
in which case it is lined with stationary phase (capillary tube). The gas mobile phase runs through the stationary phase within the column at
various rates determined by relative volatility and affinity to the stationary phase. The longer a sample molecule spends in the gas phase, the
faster it will elute from the column. For this reason, temperature and the chemical identity of the coating on the stationary phase are two
important variables which may need to be manipulated to receive clear separation of the components in the mixture. In general, lower
temperature will result in better separation but longer elution times and sample spreading due to flow effects. Efficient use of this apparatus
requires a balance between these competing effects.
GC is relatively simple for a technician to use; a calibrated gas chromatograph can run several samples without a large amount of sample
preparation. Gas chromatography works well on substances that vaporize below 300°C and are free of ions; however, it does not work well
on substances that decompose below 300°C.
Gas-solid Chromatography (GSC)

GSC uses a gas mobile phase which runs through an absorbent solid (e.g. silica gel or alumina) stationary phase within the glass column as
determined by absorption. This method operates along the same principles as gas-liquid chromatography. It is most commonly used with
gases (as opposed to a vaporized substance) and with low molecular weight hydrocarbons.
Liquid-Chromatography (LC)
LC analyzes low volatility liquids within a solid stationary phase. Separation occurs from adsorption/desorption due to the solutes having
different degrees of attraction to the stationary phase, typically due to the stationary phase having different polarity than the mobile phase. It
is common to use a mixture of solvents to accurately separate mixtures. Liquid chromatography is used in several industries, including the
pharmaceutical and petrochemical industries.
3.6.5: MASS SPECTROMETRY
Mass spectrometry is among the most precise compositional sensing tools, and, accordingly, one of the most expensive. The basic idea is
illustrated below:
A sample is ionized—this may be performed by a number of techniques—the ions are subjected to an electrical force and accelerated
through a tube. Because the electrical force applied to each molecule is the same, a molecules acceleration through the tube depends on its
mass (F=ma). The acceleration is also dependent on any charge on the molecule due to magnetic attractive or repulsive forces. At the end of
the tube, a detector calculates the time of flight for each of the molecules.
Mass spectrometry is essentially a tool to create a spectrum of distinct mass/charge ratios. It is very often used after chromatography
separation techniques to serve as a molecule identification technique.
3.6.5.1: MS COMPONENTS
There are 3 fundamental components for mass spectrometry—an ion source, a mass analyzer, and a detector.
The ion source is the component responsible for ionizing the sample. There are many different methods including chemical, electrical, or
laser ionization. Selection of the proper ion source depends on the characteristics of the sample (phase, biologically active, etc.).
The mass analyzer is the technique and tool used to differentiate the ions in a spectrum based on the mass/charge ratio. The technique
described earlier is the Time-of-Flight technique. Another common method is a sector field mass analyzer where the molecules are
accelerated through a tube and a magnetic force is applied perpendicular to the direction of flight of the molecules. The molecules will
deflect at different magnitudes depending on their size.
The detector generally amplifies changes in an electrical current that occur when molecules hit the detector.
3.6.6: THERMAL CONDUCTIVITY FOR GASES

Changes in thermal conductivity may be used to detect variations in a mixture of gases. When a heat source is present, gases and vapor will
conduct heat. The change in heat between the source and gases will stabilize to a common temperature. The temperature is mainly
dependent on the thermal conductivity and therefore the composition of the gas.
Typical equipment for thermal conductivity analyzers includes: a reference cell, a sample cell, a combined heat source (wire filaments or
thermistors), and detector. The sample cell and reference cell are usually placed in a holder where the detector may be mounted. The
reference is an identical cell of the sample cell, through which a known gas will flow. The reference-detector resistance will be constant and
the sample-detector resistance will vary depending on the composition. Resistance is a function of temperature and the output from the
detector bridge will be a function of sample composition.
Thermal conductivity, like mass spectrometry, is often used in combination with gas chromatography
3.6.7: PHYSICAL PROPERTY MEASUREMENTS

Physical properties can be used to indirectly measure the composition, but this technique is not as versatile as those described above, since it
only applies to a limited number of circumstances. The advantage to using physical properties to indirectly measure composition is that
these types of measures are often cheaper than the types of sensors described above.
3.6.7.1: DENSITY AND SPECIFIC GRAVITY

Density and specific gravity measurements require a binary mixture with at least one fluid phase to make accurate measurements. Given the
temperature and pressure, the density of a gas can be found using the ideal gas law or some other relationship, and since density is a
function of composition, the composition can be measured indirectly. See Thermodynamicsfor more information. Specific gravity is simply
the ratio of density of water over the density of the non-compressible component at the same physical conditions, which means that it is
simply a dimensionless measurement of density.
Four different measuring devices for finding density are described below.
Liquid Column
In a liquid column, the pressure measurements in a column are used to determine the density. The column may be open to the atmosphere,
where the gauge pressure at the bottom of the fixed-height column is measured. Or, the column may be closed, in which case the differential
pressure measurement is made between the bottom of the fixed-height column and at the top of the column immediately below the liquid
surface.
Displacement – Hydrometer
A hydrometer is usually made of glass and consists of a cylindrical stem and a bulb weighted with mercury or lead shot to make it float
upright. The liquid to be tested is poured into a tall jar, and the hydrometer is gently lowered into the liquid until it floats freely. The point at
which the surface of the liquid touches the stem of the hydrometer is noted. Hydrometers usually contain a paper scale inside the stem, so
that the specific gravity can be read directly.
The operation of the hydrometer is based on the Archimedes principle that a solid suspended in a liquid will be buoyed up by a force equal
to the weight of the liquid displaced. Thus, the lower the density of the substance, the lower the hydrometer will sink.
Direct Mass Measurement

With this technique, fluid is continuously passing through a U-shaped tube, where the tube is oscillating at its natural frequency. An
electrochemical device strikes the tube periodically at the curved portion of the tube. The change in frequency between the electrochemical
strike and the natural frequency is proportional to the changes in density which correlates to the composition of the substance.
Radiation-Density Gauges
Gamma radiation inside a pipe/vessel is used to determine the density of the liquid. Below is a diagram of the setup.
If the path-length for the radiation source is under 610 mm then cesium 137 is used. Above 610mm, cobalt 60 is used. The detector is
usually an ionization gauge. The density is a function of the absorption of the radiation being detected. Decay of the vessel must be taken
into consideration when calibrating
3.6.7.2: VISCOSITY
Please refer to the Viscosity Sensor section of the process control website for in-depth information.
3.6.8: COMPREHENSIVE CHARTS

For simple, easy to read charts summarizing the advantages and disadvantages of each type of sensor, please follow these links
for a PDF version of each chart.
Media:Photometric sensors.pdf
Media:Electroanalytical Measuring Techniques.pdf
Media:Other Types of Measuring Techniques.pdf
For complete summary charts including more sensors not discussed in the article, please follow the links here for a PDF version
of these summary charts.
Media:CompositionSensorsGeneral.pdf
Media:CompositionSensorsPhysical.pdf
Media:CompositionSensorsChemical.pdf
Media:CompositionSensorsElectrical.pdf
3.6.9: APPLIED EXAMPLE: MARS PATHFINDER SPECTROMETER

The Mars Pathfinder uses an Alpha Proton X-Ray spectrometer to identify the chemicals that comprise the Martian surface. The incredible
tool shoots X-Rays at the surface of the planet and uses a backscatter detector to collect the alpha particles, protons, and X-rays
characteristic of specific elements. These atom specific scattering events allow NASA scientists to determine the surface composition.
 EXAMPLE 3.6.2
Your supervisor put you in charge of monitoring the emissions from an old smoke stack that the plant just re-opened to keep up with
product demand. It is important to measure the amount of smoke and oxygen leaving the stack. Choose the proper types of sensors to
install and keep in mind that the contents of the stack are exiting at very high temperatures.
Solution
To monitor the smoke density one could install an opacity monitor that measures the absoprtion and scattering of light. Please refer to
Sensor Assmebly on this website Opacity Monitor Assembly for a basic schematic diagram.
To monitor the oxygen concentration one could install either a polarographic sensor or a high temperature ceramic sensor because both
types of sensors are specific to O2 particle. The high temperature ceramic sensor is more suitable for this application since the
emissions are exiting at high temperatures.
 EXAMPLE 3.6.2
You are in charge of calibrating a new conductivity cell acquired by the company. The conductivity cell will be used to measure the
concentration of NaCl in the beginning reagents. NaCl has a molar conductivity of 12.6 Sm2 / kmol. The cell dimensions are as
follows: area of the cells (A) is 100 m2 and the distance between the cells (x) is 0.25 m. Calculate the concentration needed to produce
a resistance of 10 ohms in the sensor.
Solution
Referring to the Conductivity Cell section, the equation can be arranged to find concentraton:
conductivity
concentration =
.olar conductivity
Unknown = conductivity
A 1
conductivity = ( )( )
x R
100 1
= ( )( )
0.25 10
= 40
Plug back into initial equation

concentration = conductivity / molar conductivity
concentration = 40/12.6
concentration = 3.17 kmol
3.6.10: REFERENCES
Richardson, JF. Coulson & Richardson's Chemical Engineering Vol. 3, Oxford: Butterworth-Heinemann. ISBN 0080410030
Perry, GH; Green, DW. Perry's Chemical Engineering Handbook 7th Edition, New York: Mcgraw-Hill ISBN 0070498415
So, Peter; Dong, CY. Fluorescence Spectrophotometry. Encyclopedia of Life Sciences, London: McMillan
OMEGA Engineering. Turbidity Measurement.
Analytical Specialities. Electrochemical Sensors
Shimadzu. Quantification of Organic Peroxides Using FTIR and ATR.
Merriam-Webster's Dictionary.
Wikipedia, Mass Spectrometry
Berg, Jeremy M.; Tymoczko, John L.; Stryer, Lubert. Biochemistry, Sixth Edition. W.H. Freeman and Company.
Wilks, Paul. Process Monitoring: In-line, At-line, or Slip-stream?.
Thayer, Ann M. KNOW THY PROCESS:Regulatory push for process analytics sets new goals for pharmaceutical manufacturing.
Randall, Willis C. PAT Pending.
ABB Instruments. Types of Industrial Composition Sensors available for purchase.
Topac Inc. Conductivity Cells.
Wroblewski, Wojciech. Ion Selective Electrodes.
Chemical Engineering Department at Case Western Reserve. Electrochemistry Encyclopedia.
Basic Turbidimeter Design and Concepts. www.epa.gov/ogwdw/mdbp/pdf/turbidity/chap_11.pdf
3.6.11: CONTRIBUOTORS
Authors: (September 14,2006) Nicole Blan, Jessica Nunn, Pamela Anne Roxas, Cynthia Sequerah
Stewards: (September 27, 2007) Eric Gauthier, Kevin Lin, Matt Naughton, Rebecca Rutishauser
Anchors: (September 26, 2007) Megan Boekeloo, Stacy Young
3.6: Composition Sensors is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source content that was
3.7: PH AND VISCOSITY SENSORS
3.7.1: INTRODUCTION TO PH
pH is a measure of concentration of H+ ions ([H+]) in solution, and can be defined as follows:
+
pH = − log([H ])
Typically, pH values range between 0 (strongly acidic) and 14 (strongly basic, or alkaline). However, that is not to say that 0 and 14 are the
limits of pH values. A solution that has a H+ concentration of 10 M, such as very strong HCl, would have a pH of -1. Pure water, which has
a pH of 7, is said to be neutral because its H+ concentration is equal to its OH- concentration. An acid solution is said to have a greater H+
concentration than its OH- concentration. Conversely, a base, or alkaline solution, is said to have a greater OH- concentration than its H+
concentration. See for reference:
For weak acids and bases, the above equation is appropriate to use when calculating pH. However, many solutions are not weak acids and
bases, which means that the other ions in solution play a large role in acidity and alkalinity, and thus a more accurate method is needed to
calculate the pH. pH sensors are used to determine the acidity or alkalinity of a particular process solution. When dealing with a strong
solutions, the sensors measure the effective concentration, or activity, of H+ ions in solution, and do not measure the actual concentration.
Hence, for the purposes of pH sensors, pH is properly defined as:
pH = − log(aH + )
with
+
aH + = (γH + ) [H ]
where γ H
+ is the activity coefficient.
The activity coefficient accounts for the interactions between all the ions present. These interactions may prevent some of the H+ ions from
contributing to the acidity of the solution, and thus decrease the effective concentration of H+ ions. This effective concentration is the
concentration used when calculating pH.
3.7.2: WHY IS PH RELEVANT FOR CHEMICAL ENGINEERING?

pH sensors, while being relatively simple pieces of equipment, have numerous uses. Knowing the pH of a solution is valuable to an
analytical chemist attempting to determine the contents of an unknown solution, as well as a farmer trying to determine the appropriate
applications to his fields and many people in between. In industry pH sensors can be used as a simple way to monitor reactions involving
acids or bases as either a reactant or product of a chemical process. They can also be used to monitor the conditions of a reactor to ensure
optimal conditions. This is especially important in biological systems where very minor changes in pH can result in significantly lowered
production levels or even a complete stoppage of production due to the death of the organisms on which the reaction is dependent.
An example of pH measurement's importance in industry is seen in food processing during the manufacture of fruit jelly. The gel formation
brought about by pectin occurs over a small pH range, which is further complicated by the concentration of sugar. Too high of a pH value
results in an unacceptably runny liquid. Too low of a pH causes the mixture to stiffen prematurely, resulting in a nearly-solid product.
Continuous pH adjustment using edible acids such as citric acid yields the optimal gel consistency.
It is also important to keep in mind that application requirements should be carefully considered a pH sensor is chosen. In the case above, it
is important that the pH sensor be made of a material that will not interfere with the quality of the product. For instance, the sensor should
not be made of glass in order to avoid broken glass entering the product. Accurate pH measurement and the precise control that it allows for
helps to optimize a process and can result in increased product quality and consistency. Continuous pH monitoring also controls chemical
usage (such as the edible acids above), resulting in minimized system maintenance (i.e. less cleaning of hardened gel from containers, less
corrosion of containers, etc.).
3.7.3: TYPICAL PH SENSOR CONSTRUCTION

Because it is difficult to directly measure the H+ concentration of a solution, pH is typically measured by comparing the potential of a
solution with a known [H+] (this is known as a reference solution) with the potential of a solution with an unknown [H+]. The potentials of
these solutions are measured using a reference half-cell and a sensing half-cell, respectively.
3.7.3.1: REFERENCE HALF-CELL
The reference half-cell generally consists of a chamber with a conductor submerged in a reference electrolyte. The conductor, or external
reference electrode wire, is typically silver coated with silver chloride (Ag/AgCl(s)) or mercury coated with mercurous chloride
(Hg/Hg2Cl2(s)). The reference electrolyte is a standard solution such as KCl. The last component of the reference half-cell is a porous plug,
which serves as a liquid-liquid interface between the standard solution and the process solution being analyzed. This plug allows standard
solution to travel from the chamber out into the process solution, but does not allow process solution into the chamber. The purpose of this
interface is to establish the electrical connection, which provides the reference potential.
3.7.3.2: SENSING HALF-CELL

The sensing half-cell is of similar construction to the reference half-cell. A glass chamber contains an electrode (Ag/AgCl(s)), known as the
internal reference electrode, which is submerged in a standard solution of constant pH. However, instead of a porous plug acting as the
liquid-liquid interface between the standard and process solutions, there exists a glass membrane. This membrane is coated on the inside and
outside with a hydrated gel. When the sensing half-cell is placed in solution, metal ions (Na+) in the gel on the outside of the membrane
diffuse out into the sample being analyzed and are replaced by H+ ions from the sample. The change in energy associated with this
substitution creates the change in electric potential that is detected by the electrode. The combination of the hydrated gel and the glass
membrane are what make the sensor specific only to H+ ions. The solution of constant pH ensures that the surface of the glass membrane
and the electrode are at the same pH before the cell is exposed to the process solution being analyzed.
The electrode wires from the reference and sensing half-cells are each connected to the voltage measuring device.
The diagram below shows the setup of a typical pH measurement system:
While this figure shows the essential components of a pH sensor, it is not an accurate depiction of what pH sensors actually look like. The
reference and sensing cells in most pH sensors are generally encased together in a sturdy housing (usually made of some sort of plastic).
This does not affect the operation of the sensor, but allows for ease of use, as there is only one major piece of equipment that the user needs
to carry around.
pH sensors can be used both online and offline. If the sensor is being used to carefully control the pH of a system or to provide an accurate
record of the pH of a solution it is often desirable to have an online pH sensor. In order to have an online pH sensor determine accurate
readings it must be in a position to take a representative sample. This is most often done by diverting a side stream from the main process in
which the pH sensor can be inserted. Often times continuous pH measurements are not necessary, but rather infrequent measurements to
make sure that the system is running correctly are of interest. In these cases it is much more efficient to use an offline pH sensor.
3.7.4: HOW DO PH SENSORS CALCULATE PH?

pH is calculated using the potential drop between the reference and sensing electrodes. The potential across the overall pH measurement
system can be expressed as:
E = Einternalreference + Emembrane + Eporousplug − Eexternalreference
The potential of the porous junction is usually negligible, and the potentials of the internal and external references can be lumped into a
constant value, E . Thus, we get:
0
E = E0 + Emembrane
Using the Nernst equation, the potential of the membrane in a half-cell reaction can be expressed as:
RT
E = E0 + ln(a)
nF
where
R is the Universal Gas Constant
F = 9.6490 × 10
7
coulombs/kmol\) is Faraday's Constant
n is the net number of negative charges transferred in the reaction
a is the activity of the ions involved in the reaction
This equation is the standard equation used in electrochemistry to find the potential with respect to a reference electrode. It can be viewed as
a calculation of the change in free energy at an electrode, due to changing ion concentrations at that electrode. Assuming that the electrode
is completely selective to H ions (a = 1) and that there is only one positive ion carrying one elementary charge in each half-cell reaction (
+
n = 1 ) the pH of the unknown solution can be expressed as:
F (E0 − E)
pH =
2.303RT
where T is the absolute temperature of solution being analyzed.
3.7.5: TEMPERATURE COMPENSATION

As one can see from the Nernst equation, the pH as determined using the potential difference between the reference and sensing half-cells
varies linearly with changes in temperature. That is, the voltage drop/pH unit across the cells increases with increasing temperature. Thus,
pH measuring system needs to have some way to compensate for temperature changes in the process solution. Many pH control kits come
equipped with automatic temperature compensation (ATC), which continuously measures and incorporates temperature changes into their
pH calculations. Other older pH control kits require manual temperature compensation, meaning that the operator or user must manually
input the process solution temperature into the control system.
3.7.6: ALTERNATIVE METHODS FOR DETERMINING PH

While glass membrane pH sensors are an effective method of measuring pH they are not always the best choice. Certain properties of the
system of interest make using a glass membrane pH sensor impractical. Glass membrane pH sensors can become clogged by viscous
solutions or solutions containing suspended solids. Harsh conditions can also scratch or break the glass membrane. There are also cases
where a glass membrane pH sensor would work but using one would be an unnecessary cost. In these situations there are several
alternatives that can be used to determine pH. One method is using pH paper. This paper changes color when submerged in a solution. The
color of the paper can be compared to a standard scale included with the paper to determine an approximate pH. pH paper is an excellent
choice when all that is needed is a "ballpark" pH value. Another method of determining pH is the use of indicators. These are substances
that change color depending on the pH of the solution. They can be very useful if the primary interest is determining if the pH of the system
is above or below a particular point. Numerous indicators exist that change colors at different pH values.
3.7.7: OTHER ION-SPECIFIC SENSORS

The same concepts that allow pH to be measured using a reference and measuring electrode can also be applied to measure the
concentration of other ions. The differences between the pH sensors and ion specific sensors are the composition of the measuring electrode
the membrane used in the measuring electrode. For some ions of interest glass membranes can still be used, however for other ions solid
state membranes or liquid ion exchange may be necessary. The choice of membrane depends on both the size and the charge of the ion of
interest.
While ion specific electrodes can be fairly accurate a number of problems can compromise their accuracy. Ion specific electrodes can suffer
from interference. This can be especially problematic in complex solutions, especially those containing many other ions similar in charge
and size to the ion of interest. The measurements reported by ion specific sensors are also functions of the ionic strength of the solution.
This can also result in significantly decreased accuracy of the sensor if the effect of the ionic strength of the solution is not accounted for. As
with pH the property that is actually being measured with an ion specific sensor is the activity of the ion, not its concentration. The activity
can no longer be considered negligable in solutions with high ionic strength. In order to obtain proper readings the user must consider what
effect the ionic strength of the solution has on the activity coefficient and adjust the measurement accordingly. Despite these limitations ion
specific sensors can still provide accurate and valuable information, provided an appropriate measuring electrode membrane is used and the
ionic strength of the solution is accounted for, and are widely used.
3.7.8: PROBLEMS WITH PH SENSORS

Several problems can occur with pH sensors that can negatively affect the accuracy of these devices. The majority of these problems
involve the reference electrode. One such problem is a partial clogging of the porous plug. Another error associated with the plug is due to a
junction potential which can vary depending on the electrode setup. This junction potential adds error to the pH measurement when the
ionic composition of the solution being tested is significantly different than the ionic composition of the standard buffer used to calibrate the
electrode. In general, a junction potential develops at the interface of two solutions with different ionic compositions because the various
ions in each solution have different mobilities, and as oppositely charged ions diffuse at different rates a charge separation develops at the
interface between the solutions. A different type of error can occur in measurements of solutions with very low hydrogen ion concentration
and very high sodium ion concentration; pH electrodes can "mistake" the sodium for hydrogen, leading to a lower pH reading than the true
value.
Other problems include the contamination of the electrolyte in the reference electrode of a complete depletion of the electrolyte. Any of
these problems will result in an incorrect potential in the reference electrode and ultimately inaccurate pH measurements.
Problems can also occur with the measuring electrode. Any damage to the glass membrane including cracks, etching, or abrasion, as well as
coating of the glass membrane will result in poor pH measurements. In addition, if the electrode is dry it will not measure pH correctly; it
must be soaked in some aqueous solution for a few hours. Problems with pH sensors can be fairly easily detected. Any problems such as
difficulty calibrating the sensor, unsteady pH readings, or significant drift in the readings usually indicate damage to the pH sensor. It is now
possible to test some forms of pH sensors online to quickly and easily determine which electrode is the cause of the problems and
potentially the source of these problems. In some cases these problems can be fixed by a careful cleaning of the sensor, however if problems
persist it may be necessary to replace the pH sensor.
 EXAMPLE 3.7.1: PH SENSOR APPLICATION
Would it be wise to use a pH electrode in the following situations? If not what method would you use to determine the pH of the system
of interest?
a. You are an engineer responsible for a wastewater treatment facility. It is important that you know the pH of the water at several
points in the treatment process to detect any potential problems.
b. You work for a pharmaceutical company who uses bacteria to produce a desired protein. The bacteria are very sensitive to changes
in pH and production falls off significantly if the pH varies from optimal conditions.
c. You work in a plant that uses hydrochloric acid as a catalyst. After the reaction has gone to completion and the product is extracted
it is desirable to separate the acid from the remaining solution to be reused in the plant and allow for easy disposal of the remaining
solution. The EPA requires accurate data on the pH (among other things) of all discharged solution.
d. You work in a plant that produces acetic acid. You want to ensure that the solution exiting the reactor contains at least 75 mol%
acetic acid.
Solution
a. NO. A pH electrode would not hold up well under the harsh conditions of a wastewater treatment plant. It is also not necessary in
this case to have constant measurements or exact pH values as the pH is being used simply as an early problem detection system.
Simple pH paper would be sufficient in this situation.
b. YES. The sensitivity of the bacteria, as well as the likely high value of the proteins being produced calls for the accuracy and
frequent measurements afforded the engineer by a glass membrane pH sensor.
c. YES. Here not only constant measurements are required, but records of these measurements must be kept to prove that
environmental standards are being met. The most cost effective way to measure and record this data would be with a pH sensor
connected to a computer.
d. NO. While pH could be used as an effective way to monitor the progress of the reaction a sensor is not necessary. Here you are only
interested in ensuring that a certain benchmark, in this case a reaction completion percentage measured using pH, is reached.
Choosing an appropriate dye indicator would be a much simpler and more cost effective way to make sure this standard is met.
3.7.9: INTRODUCTION TO VISCOSITY

Viscosity quantifies the systematic deformation of the surface of a fluid upon an applied shear stress. Fluid, while in motion, typically
travels at varying velocities due to the geometry of contact surfaces. It may be characterized either as a Newtonian fluid or a Non-
Newtonian fluid. A Newtonian fluid flows continuously in a uniform manner irrespective of the forces acting upon it (a common example,
water). Conversely, the viscosity of a non-Newtonian fluid changes upon differing flow rates (common examples, table ketchup or mud).
Also, the effect of temperature on fluid viscosity is imperative. From a thermodynamic perspectice, temperature measures the random
movement of molecules of a given substance, and as movement increases internal energy increases, and temperature also increases. For
most traditional substances, having higher internal energy/temperature also implies a lower viscosity, as the substance will be less resistant
to movement (i.e. will deform sooner to an applied shear stress). For example, consider honey: at room temperature it oozes out of the
squeeze bottle, but after a minute in the microwave it may be easily poured out, similar to a syrup.
Nearly all modern viscometers have built in controls to address temperature issues, and a competent operator or engineer may input
temperature settings to that the appropriate calculations are carried out correctly.
Viscosity may be further subdivided into two distinct forms: dynamic viscosity and kinematic viscosity. Dynamic viscosity (μ ), a figure
representing shear stress as proportional to the strain rate, has the SI unit Pa*s. Kinematic viscosity (v = μ/ρ) describes shifts in
momentum and has SI units m2s − 1, but is also commonly represented by the Stoke, cm2s − 1 (content adapted from
http://scienceworld.wolfram.com/physics/DynamicViscosity.html).
As noticed above, the typical Newtonian fluid experiences friction upon contacting a surface and resultantly develops a characteristic
velocity profile. This profile may be described by examining the known properties of the fluid and the surrounding structure through which
it travels, commonly piping for plant chemical engineering.
In the above diagram we notice a non- cylindrical surface featuring two boundary plates, one stationary to the observer, and one mobile.
Fluid contained within the two boundaries may behave according to the constraining conditions of the boundary plates, and subsequently
characterized by fluid mechanics:
∂u
τ = μ
∂y
Assuming μ as the coefficient of viscosity, the above equation describes the relationship between the shear (parallel) stress and velocity of
fluid travel per unit height above the stationary boundary plate. Fluids having a relatively high viscosity resultantly also require greater
force to induce motion—resulting in a larger shear stress.
3.7.10: WHY IS VISCOSITY RELEVANT FOR CHEMICAL ENGINEERING?

For engineers a thorough knowledge of the physical and chemical properties of products is essential to the successful implementation of any
design. Viscosity, similar to volatility, density, or any other physical property, offers significant insight into the potential behavior of a
substance, whether classified formally as a solid, liquid or gas. By taking viscosity into account, engineers may correctly select and place
instrumentation in an environmentally friendly, sustainable method.
Many professional assignments necessitate a genuine understanding of viscosity and its effects on process engineering. For a more detailed
and insightful glance into the daily upkeep and activities associated with plant maintenance, please consult Michael Pearce at
www.reynoldsmixers.com/whitepapers/030910_hotmelt.doc. He speaks at length regarding the cost- effective selection of specific types of
blades used in mixing processes.
As an example, a key parameter in food process monitoring is viscosity. The viscosity is directly related to the flow characteristics of the
product, which impact pumpability, pourability, and spreadability. In the food-processing environment there are a number of challenges to
viscosity measurement, such as harsh process conditions and the complex rheological properties (relationships between deformations and
stresses of materials). Offline measurements are often cumbersome, labor-intensive, and prone to operator error. Online viscometers must be
able to deliver continuous measurement day after day with minimal maintenance. Therefore, considerations when choosing viscosity
sensors for food processing should include ease of cleaning, minimal risk of fouling, and whether or not they meet sanitary requirements.
3.7.11: OFF-LINE INSTRUMENTS

For the measurement of viscosity in controlled settings (typically laboratory) where the majority of variables are maintained as constant,
engineers and scientists use off-line viscometers. Typical examples include: Capillary, Couette, Falling Ball, Cone and Plate and Oscillating
Cylinder. It is very important to keep these instruments in a regulated environment with a stable temperature (such as in a water bath)
because of viscosity's sensitivity to changes in temperature.
3.7.11.1: CAPILLARY
The Ostwald U- Tube viscometer functions by measuing the amount of time a specified quantity of fluid takes to move from its initial
position to a final position.
Typically, a suction device holds the bottom of the fluid meniscus at the start position (indicated by the solid horizontal red line), and upon
its release the fluid drops an approximate distance of 5 mm (to reach the solid horizontal green line). The time for this drop is recorded so
that the acquired data may then be used in Poiseuille’s Law to determine the outcome of kinematic viscosity.
Both the dynamic and kinematic viscosities may be found by appling these formulae (where K0 is the non-dimensionless viscometer
constant):
Dynamic Viscosity: μ = K0ρt
Kinematic Viscosity: v = K0t
3.7.11.2: COUETTE
This device may be used in both off-line and on-line applications with appropriate modifications in the output settings. A couette type
viscometer measures viscosity by spinning a cylinder encapsulated in fluid. This is accomplished through the synchronization (identical
frequency, no phase difference) of a motor with magnetic coupling to rotate a magnet which in turns forces the inner cylinder to revolve
within the fluid.
The torque reaction of the motor is resultant of the viscous drag on the rotating cylinder. This torque on the motor is effectively counteracted
by the torsion bar (a thin rod connecting the control/ measuring bandto the linear variable displacement transformer). The deflection of the
torsion bar, a function of the fluid viscosity, is then subsequently converted into local signal available for laboratory analysis.
Engineers use viscometers of the couette type in in-line or in-tank applications. Also, the meters are appropriate for both Newtonian and
non-Newtonian fluids. The acceptable range of viscosity spans from 10 − 3 to 5 * 103Ns / m2.
Information for this device has been adapted from Richardson's Chemical Engineering.
3.7.11.3: FALLING BALL
A substantially more direct approach to viscosity measurement, testers allow a ball to fall through a column containing liquid of unknown
viscosity and then measure the amount of time necessary for the ball to reach the bottom of the column.
A commonly used method involves a stainless steel ball which sinks through the liquid under scrutiny; researchers measure the time
necessary for the ball to drop from one preselected level to another. To attain the terminal velocity of the ball in the liquid, u0, timers write
the initial timing mark a minimum of six ball diameters below the ball release point.
Also important to note, the ratio of ball diameter to column tube diameter should not exceed 1:10 to account for the drag effects of the
column tube walls on the terminal velocity of the falling steel ball. A simple viscosity derivation from buoyancy principles is presented in
Richardson's text, and is shown again here:
Kv (ρs − ρ)
v = t
ρ
where
Kv is the dimensional viscometer constant
ρs is the density of the ball
ρ is the density of the liquid
t is the time required for the ball to sink the measured distance
3.7.11.4: CONE AND PLATE

Another popular method used employs a cone placed in a manner so that the tip touches the center of a stationary plate. Fluid encapsulates
the cone and researchers take a measurement of the amount of torque required to keep the plate stationary.
This process allows the dynamic viscosity to be quantitatively described as:
3
v = T
3
πγr
T = Torque required to keep the top plate in place

γ = Rate of shear
The rate of shear is the ratio of the angular velocity of the cone to angle between the cone and the plate (note: this angle must be small
enough so that the sin of the angle approximately equals the angle itself).
Important Additional Notes:
Edge effects of the cone or plate are neglected. The equation may also be used for Non-Newtonian fluids provided that γ is relatively
constant.
3.7.11.5: PARALLEL PLATES

Parallel plates or discs viscometer is similar to the cone and plate method. The cone in the case is replaced with another plate. In this
viscometer, the fluid flows in a gap between two parallel discs. One of the discs rotates with an angular velocity ω which creates the shear.
Torque is applied to the other plate so it stays stationary. A normal force, F, is created by fluid elasticity and acts as to separate the two
plates. No slip at boundaries is assumed.
Parallel Plates viscometer

ω, T and F are all experimental parameters. Shear rate is given by:
ωr
γ̇ =
d
Shear rate
r = The radius of the disc
d = The distance between the two plates
Shear stress, τ, is given by :
T dlnT
τ = (3 + )
3
2πr dlnγ̇
Viscosity is, therefore, the ratio of shear stress to shear rate. Some limitations associated with parallel plates include:
1. Sufficient data of T vs w and F vs w must be available
2. Uniform temperature at different points in the gap between the two plates is required
3. Error might be caused from edge fracture, wall slip, misalignment of the plates and viscous heating.
3.7.11.6: OSCILLATING CYLINDER

This instrument involves an arrangement requiring a rotating cylinder placed in a quantity of a viscous fluid. The level of disturbance
noticed in the fluid resulting from the rotation of the cylinder is then noted, and may be used if the following equation to determine
viscosity:
2 2
(r −r )
2 1 ′
v = T
2 2
4πωv r r
2 1
r2 = Outer cylinder radius
r1 = Inner cylinder radius
ωv = Angular velocity
The outer cylinder of radius r2 rotates with an angular velocity of ωv while the inner cylinder of r1 remains stationary thus allowing for the
torque (alternatively reffered to as 'viscous drag'), T', to be measured at predetermined values of ωv.
3.7.12: ON-LINE INSTRUMENTS

On-line instruments are those that are capable of giving a precise measurement of viscosity under plant conditions. This means that they will
be able to withstand the wide range of viscosities and the variable temperature, pressure, and flow rate that occur within a process plant
while still obtaining the accuracy that is required. In all on-line instruments automatic temperature compensation is always required because
of viscosity's temperature dependence.
3.7.12.1: CAPILLARY
This type of viscometer is derived from Poiseuille’s law:
dv 2
= vs ∗ π ∗ r
dt
vs = Fluid velocity
r = Internal radius
The instrument has many temperature and pressure sensors in the intricate system because there must be a precisely controlled pressure drop
and temperature in the stream. Response time is minimized by adding a fast loop sampling system. A capillary viscometer would be useful
when using Newtonian fluids such as lubricating oils or fuel oils. It can endure viscosities ranging from 2 * 10 − 3 to 4 Ns / m2.
3.7.12.2: COUETTE
Identical to the description provided earlier, except now with a change in the output data processing. The deflection of the torsion bar, a
function of the fluid viscosity, is now converted into a 4-20 mA output signal and is readily available for additional plant data
mangement/processing.
Information for this device has been adapted from Richardson's Chemical Engineering.
3.7.12.3: VIBRATING ROD OR CYLINDER
This type of viscometer can also be used in-line or in-tank. It is suitable for Newtonian, non-Newtonian fluids, and mineral slurries. The
vibrating rod or cylinder works by using an electromechanical system as the driving force, or by a pulse of current through a specific alloy.
The lowest viscosity that it can handle is 10 − 4Ns / m2 and it can extend to 2 * 103Ns / m2. Normally on lower viscosity ranges the cylinder
is used, while with higher ranges the rod is used. Given a plants’ natural vibration care must be taken when deciding on the location of an
oscillation viscometer because of their sensitivity.
3.7.13: TABLE OF VISCOMETER USES

This table found in Richardson's text offers some guidance on the device selection process.
One may notice that the viscosity ranges of some of the viscometers overlap. In a situation where this occurs and it is necessary to choose
between multiple viscometers, one needs to take into account process conditions such as changes in temperature, pressure, and solution flow
rate. The appropriate viscometer should be able to handle these conditions without disrupting the process.
3.7.14: PROBLEM SOLVING SCHEMATIC

The following schematic may be employed to select an appropriate viscometer. Additional considerations must also take place, as the usable
viscometer ranges occasionally overlap. This requires the competent operator or engineer to base his or her selection from a host of other
valuable criteria, including but not limited to: cost- effectiveness, difficulty of implementation, heat generation, availability of qualified
local maintenance personnel, environmental impact, and long term sustainability.
 EXAMPLE 3.7.1: VISCOSITY SENSOR
You are an engineer running a process that requires hot water at some times and cold water at others. To heat and cool your water, you
run it through a heat exchanger. When you want to heat the water, you use 100°F glycol stream to heat the water (viscosity = 0.002
Ns/m2). When you want to cool the water, you use 20°F glycol stream to cool the water (viscosity = 5 Ns/m2). To ensure that the glycol
does not clog the pumps in your process, you need to monitor its viscosity on-line. What type(s) of viscometers could you use on the
hot glycol stream? On the cool glycol stream?
Solution
After looking at the summary table given above, we can immediately narrow down our search to just the on-line instruments from the
information given in the problem statement. For the hot glycol stream we find which instruments have a viscosity range which allows
for a viscosity of 0.002 Ns/m2. We find that the following viscometers would be acceptable: Capillary, vibrating rod or cylinder
The cool glycol stream involves the same process, but it must have a different range of viscosity which includes 5 Ns/m2. Viscometers
which are acceptable for this stream include: Couette, vibrating rod or cylinder
3.7.15: PH/VISCOSITY SENSOR KEY LEARNINGS

As is apparent after reading this article, pH and viscosity measurements do not have much in common with each other. However, there is a
common thread between these two types of sensors that needs to be acknowledged, and that is the method that needs to be taken when
deciding what type of sensor to use. While there is only one single type of pH sensor available, before it is to be used the user needs to
evaluate whether there is an easier and cheaper way (eg. using pH paper or a pH activated dye) to accomplish the task at hand. With
viscometers, there are multiple types of viscometers to use over various ranges of viscosity. The user needs to evaluate process conditions
such as temperature, pressure, and flow rate, in order to decide which type of viscometer is optimal for the application.
The math associated with both types of sensors is good to know for background purposes and to gain a fundamental understanding of how a
sensor produces a measurement, however it is not information that greatly helps the user decided what type of pH indicator or viscometer to
use. Hence, this article has tried to give the background theory behind the measurements, but a thorough analysis of this theory is not
included and not necessary for the purposes of Process Control.
3.7.16: REFERENCES
Cole-Palmer Catalog, 2005/2006
Liptak, Bela. Instrument Engineers Handbook - Process Control and Optimization, Fourth Edition, Volume 2. Taylor & Francis, New
York, 2005.
Perry, Richard. Perry's Chemical Engineers' Handbook, Seventh Edition. McGraw-Hill, 1997.
Richardson, J.F. and Peacock, D.G. Chemical Engineering, 3rd Edition, Volume 3. Butterworth-Heinemann, 1994.
Rodriguez, Ferdinand; et al. Principles of Polymer Systems, 5th Edition. Taylor & Francis, 2003. ISBN 1560329394
Stephanopoulos, George. Chemical Process Control - An Introduction to Theory and Practice. Prentice Hall, New Jersey.
Hydramotion. Online Viscosity Control Increases Process Productivity and Efficiency. ScientistLive 17 November 2008. 8 December
2008 http://www.scientistlive.com/European-Food-
Scientist/Processing_Equipment/Online_viscosity_control_increases_process_productivity_and_efficiency/21292/

Authors: Navin Raj Bora, Dallas Burkholder, Nina Mohan, Sarah Tschirhart
3.7: pH and Viscosity Sensors is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source content that
was edited to conform to the style and standards of the LibreTexts platform; a detailed edit history is available upon request.
3.8: MISCELLANEOUS SENSORS
This page is a summary of different sensors not as commonly used as others in industry. This page details the methods of operation,
industrial uses, and other relevant information about the sensors.
3.8.1: HUMIDITY SENSORS

“Humidity sensors are used to control the amount of water vapor present in many industrial processes. Textile, wood, and chemical
processing is very sensitive to humidity” [1]
3.8.1.1: PSYCHROMETERS
Psychrometers use latent heat of vaporization to determine relative humidity in the system. This can be done by using a dry bulb
thermometer with a wet bulb thermometer. The two temperatures recorded can be used with a psychrometric chart to obtain the relative
humidity, water vapor pressure, heat content, and weight of water vapor in the air.
3.8.1.2: HYGROMETERS
Hygrometers are devices that sense the change in either their physical or electrical properties. Some materials such as hair or thin strips of
wood change length depending on water content. The change in length is directly related to the humidity.
3.8.1.3: DEW POINT MEASURING DEVICES

These devices measure humidity by cooling the air until water starts to condense on the object. The amount that the air needed to be cooled
before water started to form on the object can be used to determine the relative humidity.
3.8.1.4: OTHER HUMIDITY SENSORS

Microwave absorption by water vapor can be used to measure the humidity in a material
Infrared absorption can be used by hitting the object with infrared radiation and measuring the energy of the reflectance.
3.8.2: SOUND SENSORS

Sound sensors are important because they can be used in industrial applications such as detection of flaws in solids and location and linear
distance measurements. Sound pressure waves can also cause vibration and failure.
3.8.2.1: MICROPHONES
Microphones are pressure transducers, and are used to convert sound pressures into electrical signals. There are six types of microphones:
electromagnetic, capacitance, ribbon, crystal, carbon, and piezoelectric.
3.8.3: SMOKE SENSORS

Smoke sensors are useful for not only safety for workers, but also environmental concerns and purity issues of processes.
3.8.3.1: INFRARED SENSORS

These sensors detect changes a signal received from an LED due to smoke or other objects in the light path from the LED.
3.8.3.2: IONIZATION CHAMBERS

These devices can detect the difference in current between two plates that have a voltage between them. The difference is due to carbon
particles from smoke that provide a conductive path between the two plates.
3.8.3.3: TAGUCHI-TYPE
Taguchi-type sensors are used to detect hydrocarbon gases, such as carbon monoxide and carbon dioxide. The sensor is coated with an
oxidized element that when combined with a hydrocarbon creates a change in the electrical resistance of the sensor.
3.8.4: REFERENCES
[1] Dunn, William C. Introduction to Instrumentation, Sensors and Process Control. Boston: Artech House, 2006. Print.
3.8.5: CONTRIBUTORS
Authors: Michael Kennett, Conor O'Bryan, Ryan Weigold
3.8: Miscellaneous Sensors is shared under a CC BY license and was authored, remixed, and/or curated by LibreTexts.
3.9: VALVES - TYPES, KINDS, AND SELECTION
Authors: Daniel Katzman, Jessica Moreno, Jason Noelanders, and Mark Winston-Galant
Stewards: Jeff Byrd, Khek Ping Chia, John Cruz, Natalie Duchene, Samantha Lyu
3.9.1: INTRODUCTION
Control valves are imperative elements in any system where fluid flow must be monitored and manipulated. Selection of the proper valve
involves a thorough knowledge of the process for which it will be used. Involved in selecting the proper valve is not only which type of
valve to use, but the material of which it is made and the size it must be to perform its designated task.
The basic valve is used to permit or restrain the flow of fluid and/or adjust the pressure in a system. A complete control valve is made of the
valve itself, an actuator, and, if necessary, a valve control device. The actuator is what provides the required force to cause the closing part
of the valve to move. Valve control devices keep the valves in the proper operating conditions; they can ensure appropriate position,
interpret signals, and manipulate responses.
When implementing a valve into a process, one must consider the possible adverse occurrences in the system. This can include noise due to
the movement of the valve, which can ultimately produce shock waves and damage the construction of the system. Cavitation and flashing,
which involve the rapid expansion and collapse of vapor bubbles inside the pipe, can also damage the system and may corrode the valve
material and reduce the fluid flow.
There are four general types of valves.
1. Electronic, or electrical valves. The movement of the ball or flap that controls flow is controlled electronically through circuits or
digitally. These types of valves have very precise control but can also be very expensive.
2. Non-Return valves. These valves allow flow in only one direction, and are common in various industries. When pressure in the opposite
direction is applied, the valve closes.
3. Electromechanical valves. These valves have electro magnets controlling whether the valve is open or closed. These valves can only be
fully open or fully closed.
4. Mechanical Valves. These valves use mechanical energy in the process of opening and closing the actual valve. Larger valves can be
opened and closed using mechanical processes such as levers and pulleys, whereas smaller mechanical valves can be opened or closed
via a turning wheel or pulling a level by hand.
There are four major valve types: ball valves, butterfly valves, globe valves, and plug valves. There is also an array of many other types of
valves specific to certain processes. Selecting which type of valve to use depends on what task the valve must carry out and in what space
said valve can fit to carry out the task.
Some general features that one can take into consideration when selecting a valve are the following:
1. Pressure rating
2. Size and flow capacity
3. Desired flow condition
4. Temperature limits
5. Shutoff response to leakage
6. Equipments and pipes connected
7. Material compatibility and durability
8. Cost
3.9.2: VALVE TYPES

There is a vast abundance of valve types available for implementation into systems. The valves most commonly used in processes are ball
valves, butterfly valves, globe valves, and plug valves. A summary of these four valve types and their relevant applications is in the table
below.
Valve Type Application Other information
Ball Flow is on or off Easy to clean
Butterfly Good flow control at high capacities Economical
Globe Good flow control Difficult to clean
Plug Extreme on/off situations More rugged, costly than ball valve
Following is a detailed description of the four main valve types.
3.9.2.1: BALL VALVES
A ball valve is a valve with a spherical disc, the part of the valve which controls the flow through it. The sphere has a hole, or port, through
the middle so that when the port is in line with both ends of the valve, flow will occur. When the valve is closed, the hole is perpendicular to
the ends of the valve, and flow is blocked. There are four types of ball valves.
A full port ball valve has an over sized ball so that the hole in the ball is the same size as the pipeline resulting in lower friction loss. Flow is
unrestricted, but the valve is larger. This is not required for general industrial applications as all types of valves used in industry like gate
valves, plug valves, butterfly valves, etc have restrictions across the flow and does not permit full flow. This leads to excessive costs for full
bore ball valves and is generally an unnecessary cost.
In reduced port ball valves, flow through the valve is one pipe size smaller than the valve's pipe size resulting in flow area becoming lesser
than pipe. But the flow discharge remains constant as it is a multiplier factor of flow discharge (Q) is equal to area of flow (A) into velocity
(V). A1V1 = A2V2; the velocity increases with reduced area of flow and decreases with increased area of flow.
A V port ball valve has either a 'v' shaped ball or a 'v' shaped seat. This allows the orifice to be opened and closed in a more controlled
manner with a closer to linear flow characteristic. When the valve is in the closed position and opening is commenced the small end of the
'v' is opened first allowing stable flow control during this stage. This type of design requires a generally more robust construction due to
higher velocities of the fluids, which would quickly damage a standard valve.
A trunnion ball valve has a mechanical means of anchoring the ball at the top and the bottom, this design is usually applied on larger and
higher pressure valves (say, above 10 cm and 40 bars).
Ball valves are good for on/off situations. A common use for a ball valve is the emergency shut off for a sink.
3.9.2.2: BUTTERFLY VALVES
Butterfly valves consist of a disc attached to a shaft with bearings used to facilitate rotation. The characteristics of the flow can be controlled
by changing the design of the disk being used. For example, there are designs that can be used in order to reduce the noise caused by a fluid
as it flows through. Butterfly valves are good for situations with straight flow and where a small pressure drop is desired. There are also
high performance butterfly valves. They have the added benefit of reduced torque issues, tight shutoff, and excellent throttling. It is
necessary to consider the torque that will act on the valve. It will have water moving on both sides and when being used to throttle the flow
through the valve it becomes a big factor. These valves are good in situations with high desired pressure drops.They are desirable due to
their small size, which makes them a low cost control instrument. Some kind of seal is necessary in order for the valve to provide a leak free
seal. A common example would be the air intake on older model automobiles.
Top view, open configuration Top view, closed configuration
3.9.2.3: GLOBE VALVES

A globe valve is a type of valve used for regulating flow in a pipeline, consisting of a movable disk-type element and a stationary ring seat
in a generally spherical body. The valve can have a stem or a cage, similar to ball valves, that moves the plug into and out of the globe. The
fluid's flow characteristics can be controlled by the design of the plug being used in the valve. A seal is used to stop leakage through the
valve. Globe valves are designed to be easily maintained. They usually have a top that can be easily removed, exposing the plug and seal.
Globe valves are good for on, off, and accurate throttling purposes but especially for situations when noise and caviatation are factors. A
common example would be the valves that control the hot and cold water for a kitchen or bathroom sink.
3.9.2.4: PLUG VALVES
Plug valves are valves with cylindrical or conically-tapered "plugs" which can be rotated inside the valve body to control flow through the
valve. The plugs in plug valves have one or more hollow passageways going sideways through the plug, so that fluid can flow through the
plug when the valve is open. Plug valves are simple and often economical. There are two types of plug valves. One has a port through a
cylindrical plug that is perpendicular to the pipe and rotates to allow the fluid to proceed through the valve if in an open configuration. In
the closed configuration, the cylinder rotates about its axis so that its port is no longer open to the flow of fluid. An advantage of these types
of valves is that they are excellent for quick shutoff. The high friction resulting from the design, however, limits their use for accurate
modulating/throttling. Schematics of this type of plug valve are below.
Top view, closed configuration Side view

Top view, open configuration
The other type of plug valve is the eccentric plug valve. In this design, the plug rotates about a shaft in a fashion similar to a ball valve. To
permit fluid flow, the plug can rotate so that it is out of the way of the seat. To block fluid flow, it rotates to the closed position where it
impedes fluid flow and rests in the seat. A schematic of this valve is below.
Top view, near-closed configuration

A common example would be a spray nozzle at the end of a garden hose.
3.9.3: SAFETY VALVES

3.9.3.1: PRESSURE RELIEF VALVES
Pressure relief valves are used as a safety device to protect equipment from over-pressure occurrences in any fluid process. Loss of heating
and cooling, mechanical failure of valves, and poor draining and venting are some of the common causes of overpressure. The relieving
system depends on the process at hand; pressure relief valves either bypass a fluid to an auxiliary passage or open a port to relieve the
pressure to atmosphere. Some areas of common usage include reaction vessels and storage tanks. In the Petroleum Refining Industry, for
example, the Fluidized Catalytic Cracker (FCC) reactor has several pressure relief valves to follow saftey codes and procedures on such a
high pressure/high temperature process. Each of the pressure relief valves have different levels of pressure ratings to release different
amounts of material to atmosphere in order to minimize environmental impact.
Here are three examples of pressure relief valves:
Conventional Spring Loaded Safety Valve
As the pressure rises, this causes a force to be put on the valve disc. This force opposes the spring force until at the set pressure the forces
are balanced and the disc will start to lift. As the pressure continues to rise, the spring compresses more, further lifting the disc and
alleviating the higher pressure. As the pressure inside the vessel decreases, the disc returns to its normal closed state.
Advantages:
Most reliable type
versatile
Disadvantages:
Pressure relief is affected by back pressure
Susceptible to chatter
Bellows Spring Loaded Safety Relief Valve

The bellows spring loaded safety relief valve has the same principle as the conventional spring valve, with the exception of a vent located
on the side of the valve. This vent lets releases the contents of the valve out to the surrounding environment.
Advantages:
Pressure relief is not affected by back pressure
Can handle higher built-up back pressure
Spring is protected from corrosion
Disadvantages:
Bellows can be susceptible to fatigue
Not environmentally friendly (can release of toxics into atmosphere)
Requires a venting system
Pilot Assisted Safety Relief Valve
The pilot operated safety relief valve is also similar to the conventional safety relief valve except a pneumatic diaphragm or piston is
attached to the top. This piston can apply forces on the valve when the valve is closed to balance the spring force and applies additional
sealing pressure to prevent leakage.
Advantages:
Pressure relief is not affected by back pressure
Can operate at 98% of set pressure
Less susceptible to chatter
Disadvantages:
Pilot is susceptible to plugging
Has limited chemical use
Condensation can cause problems with the valve
Potential for back flow
3.9.3.2: STEAM TRAPS

Steam traps are devices that exist in low lying places within a pressurized steam line to release condensate and non-condensable gases from
the system. Steam lines in industry are used to open/close control valves, heat trace pipelines to prevent freezing, etc. These steam traps are
used in industry to save money on the prevention of corrosion and loss of steam. When these traps fail, it can mean a lot of money for the
industry. There can be several hundred to several thousand in one process unit, therefore it is important to maintain and check the condition
of each trap annually. The checks can be done by visual, thermal, or acoustic techniques. Many suppliers have equipment to read the flow
within the pipeline to see if it is: blocked, cold, leaking, working.
There are many types of steam traps that differ in the properties they operate on including Mechanical (density), Temperature (temperature),
and Thermodynamic (pressure). For more information on the purpose and function of steam traps please view [Steam Traps].
3.9.3.3: OTHER SAFETY VALVES

Rupture Disc - A rupture disc (also called a "bursting disc" or "safety disc") is a thin membrane of material (usually metal) that acts as a
one-time use pressure relief device. At a critical pressure, the disc will fail and rupture allowing flow and the release of pressure. Often,
rupture discs are used as a back-up to a conventional spring-controlled safety valve. Given primary safety valve failure (either no or
inadequate pressure relief) the rupture disc will burst open and relieve pressure. Rupture discs are commonly used in chemical process
plants and large air-cooled steam condensers.
Pressure Vacuum Valves - Pressure vacuum (P-V) valves protect against both over-pressure and under-pressure conditions. They are
commonly used in atmospheric storage tanks, to prevent the build up of excessive pressure or vacuum which can dangerously unbalance
the system or damage the storage vessel.
3.9.4: OTHER VALVES

In addition to the four main types of control valves, there are numerous other valves that may be necessary to manipulate fluid flow in a
certain system. Some valves important to chemical engineering processes are stated below with a brief description of their design and
application.
Angle valves include inlet and outlet ports that are oriented at a 90 degree angle. The fluid leaves at right angles to the direction in which
it enters the valve.
Bleed valves vent signal line pressure to atmosphere before removal of an instrument or to assist in calibration of control devices.
Common bleed valves include ball and plug bleed valves.
Check valves are one way valves. Check valves only allow fluid in one way and out the other. They are often placed on individual fluid
streams when mixing fluids so as to prevent the mixture from flowing back into the original streams. Also, the speed with which the
valve closes is significant to prevent reverse-flow velocity. There are different types of check valves which include lift check, swing
check, tilting disk and diaphragm.
Lift check valves are quick closing. This valve should only be used for low viscosity fluids because they can be slowed by viscous fluids.
Swing check valve has a disc like closing method from a hinge which may or may not be spring loaded.
Tilting disk check valve is spring loaded for quick response. These are often more expensive and harder to fix when broken.
Diaphragm valves have excellent shut-off characteristics and are used when there is particulate matter in the fluids. Not a good choice
for controlling flow. A diaphragm valve has both a flexible and a rigid section. One advantage is that there are no crevices that affect the
flow of the fluid when open. Used mostly in the pharmaceutical and food industry for their ability to be sanitary.
Piston valves have a closure member shaped like a piston. When the valve opens, no flow is observed until the piston is fully moved
from the seat bore, and when the valve closes, the piston removes any solids that might be deposited on the seat. Therefore, piston valves
are used with fluids that have solid particles in suspension.
Gate valves work by raising a gate or barrier out of the way of the flowing fluid. The water spigot outside a house works as a gate valve.
Have the positive quality that, when fully open, the fluid flow is totally unobstructed. Two major types of gate valves are used: parallel
and wedge gate valves. The wedge gate valve, in which the closure member has a wedge shape, provides the advantage of sealing
against low pressure, as well as high pressure, flow. Gate valves have the ability to open very quickly. Advantages of gate valves are that
they have a high capacity, have good seals, relatively inexpensive, and don't have very much resistance to flow. Some disadvantages of
gate valves are that they sometimes can have poor control, they can cavitate at lower pressures, and they cannot be used for throttling.
Needle valves are similar to gate valves; however, they are usually applied to a much smaller orifice. Needle valves are excellent for
precise control of fluid flow, typically at low flow rates.
Flush Bottom valves are usually at the lowest point of a tank or reactor and used to drain out contents. Unique because it leaves no dead
space in the valve when it is closed, this eliminates the problem of product buildup within the valve.
Pinch valves are mainly used in order to regulate the flow of slurries in certain processes and systems. Pinch valves have flexible bodies
that can be shut by pinching them. They are often used when it is necessary for the slurry to pass straight through when the valve is not
pinched. They are often used for sanitary processes such as medical and pharmaceutical applications. Pinch valves can be controlled
mechanically or with fluid pressure.
Knife valves are used in systems that deal with slurries or powders. They are primarily used for on and off purposes; whether or not the
slurry or powder flows or not. A knife gate valve can be used for fibrous material because it can cut through to close the valve.
Ballcock valves are used in controlling levels in tanks. The valve is connected to a float in the tank using a lever. When the level in the
tank rises, the float rises and forces the valve to be shut at the maximum level of the tank allowed. Ballcock valves are used mostly in
water tanks and other low-risk areas within a certain process.
Solenoid valves are used very frequently in industry. The valves have a loop of wire that is wrapped around a metal core. A current is
shot through the valve creating a magnetic field, which in turn opens or closes plungers in pipelines controlling flow in that pipe. There
are three types of solenoid valves:
1. Electromechanical solenoid valves use electrical energy
2. Pneumatic solenoid valves use compressed air
3. Hydraulic solenoid valves use energy from pressurized fluids
3.9.5: FLOW CHARACTERISTICS

For all valves, flow rates of fluid through the valve depend on the percentage of a full valve opening. In ball and butterfly valves, the valve
opening is based on rotation. In the case for butterfly valves, an open valve is a result of a 90 degree rotation. When the valve is at the open
position, the flow is parallel to the valve. The flow is uninterrupted, and therefore, no pressure is on the valve. When the valve is throttling,
as is common for globe valves, the flow rate is reduced or increased depending on the opening of the valve, and there is unequal pressure on
the ends of the valve.
The ball valve flow stream involves at least two orifices; one for inlet and one for outlet flow. While traveling through an open ball valve,
the fluid will continue in its flow straight through, with little pressure loss. When the ball valve is throttling, however, the fluid is subject to
shearing and a change in flow rate in accordance with the percentage of the valve that is open. With high velocity liquids, the valve is
susceptible to cavitation and erosion, and could produce noise. Also, at sonic velocities, the vena contracta will stop expanding, and create
choking.
Globe valves are dependent on shape of the plug of the valve, in addition to the opening size, for flow variations. Lifting the globe valve
will cause it to open. A flat plug is used for a quick opening; a conically shaped plug will create linear flow as the valve is raised; a
rectangular shaped plug, where the bottom converges to one point directly in the middle of the plug, creates an equal percentage of flow
through the valve.
3.9.6: SIZING
Once a specific valve type is chosen for a process, the next step is to figure out what size the valve will need to be. Topics to consider in
valve sizing are the pressure-flowrate relationship and the flow characteristics to ensure an appropriate size for the desired valve.
3.9.6.1: PRESSURE-FLOWRATE RELATIONSHIP
Control valves predominately regulate flow by varying an orifice size. As the fluid moves from the piping into the smaller diameter orifice
of the valve, the velocity of the fluid increases in order to move a given mass flow through the valve. The energy needed to increase the
velocity of the fluid comes at the expense of the pressure, so the point of highest velocity is also the point of lowest pressure (smallest cross
section). This occurs some distance after leaving the smallest cross section of the valve itself, in a localized area known as the vena
contracta.
Beyond the vena contracta, the fluid’s velocity will decrease as the diameter of piping increases. This allows for some pressure recovery as
the energy that was imparted as velocity is now partially converted back into pressure. There will be a net loss of pressure due to viscous
losses and turbulence in the fluid.
3.9.6.2: OVERVIEW OF SIZING FORMULAS

In order to determine the correct size of a valve for a specific system many factors must be considered. The most important factor is the
capacity parameter, C , or the flow coefficient. The flow coefficient is a way of measuring how efficient a valve is at allowing fluid to flow
v
through it and generally determined experimentally. Valve manufacturers can provide you with Cv charts for the valves they sell. To
determine the valve size needed for your system, you can estimate C with the following equations:
v
Liquid Flow
−−−−
Gf
√
Cv = 11.7Q
ΔP
where:
The next equation is also for calculating the flow coefficient. It should be used when you have a design flow rate, that will be the maximum
flow through the valve, and a maximum pressure drop that will occur across the valve, which will be ΔP. This equation is also for turbulent
flow and the units used in the calculation are important.
−−−−
Gt
Cv = Fmax √
ΔP
where:
Another important piece of information about sizing and specifications is what diameter pipe may be used with a certain flow, seen in the
following equation:
Note: The second and third equations are from Peter Woolf's Fall'08 lecture titled Modeling Case Study: Surge Tanks, Valves, Level sensors,
and modeling.
Air and Gaseous Flow
(when Po < 0.53Pi)
−−−−−−−−−−
√ Gf (T + 460)
Cv = Q
660Pi
(when Po > 0.53Pi)
−−−−−−−−−−
√ Gf (T + 460)
Cv = Q −−−−−−−
1360√ΔP (Po )
where:
The relationship between the inlet and outlet pressure is important, as seen above, in determining which equation to use for gaseous flow.
When the outlet pressure is less than 53% the inlet pressure, this is said to be a critical pressure drop. The gas will behave differently when
the pressure drop is critical, therefore it is necessary to use the correct equation depending on the extent of pressure drop. Once you have
computed the desired value of Cv, you can choose the valve. The chosen valve must have a valve coefficient greater than or equal to the
computed value.
Other sizing information
When sizing a valve it is important not to choose a valve that is too small. Take the following tips into consideration when choosing a valve.
Valves should not be less than half the pipe size.
Valves are easiest to control when they are open from 10 to 80% of maximum flow.
3.9.6.3: USING THE FLOW COEFFICIENT TO DETERMINE VALVE SIZES

The flow coefficient, Cv, is used as a standard sizing parameter. It is used to determine valve sizes and is also used in modeling programs for
large piping systems. The Cv is an essential and practical variable that is essentially required when using pipes and valves. Try searching for
a valve information and you will most likely find information requiring the C as well as the valve type and use.
v
Try looking at information for specific valves at the website of American Valve. [1]
You'll notice that every valve type and size has a listed Cv for correct sizing and function. Tables using this value are available for
practically every valve imaginable and should be used to prevent over-use or under-use of valves in engineering processes.
3.9.6.4: CV CHARTS
When sizing a valve, it is important to choose the correct Cv chart. Different Cv charts are used for different types of valves. Some different
valve types are explained below. Note that valve stroke means the amount of flow through a pipe, and valve travel means the amount in
which the valve is turned to achieve the valve stroke.
Equal Percentage Valves: Valve which produces equal valve stroke for equal increments in valve travel. This is the most common type
of valve.
Linear Valves: Valves stroke is directly proportional to valve travel.
Quick Opening Valves: A small amount of valve travel produces a large valve stroke. Example: Gate Valves
Equal percentage and linear valve types offer the best throttling control. Examples are globe and butterfly valves.
Valves vary from manufacturer to manufacturer. Valve manufacturers can provide you with Cv charts to size each type of valve they
manufacture. An example CV chart is shown below. VT stands for valve travel.
3.9.7: MATERIALS
It is usually not enough to simply select the type of valve suited to given process parameters. Selecting compatible materials of construction
helps ensure the lifespan of the valve, as well as the safety of the workforce, the environment and the public.
Selecting the most appropriate materials of construction of control valves is guided primarily by the service of the valve, then secondarily
by cost; the least expensive material that is compatible with the service will be chosen to be used. Proper material selection promotes safety
by avoiding materials of construction that may react with or be corroded by the process fluid.
The principle materials that need to be selected carefully are the wetted materials, that is, the materials that come into contact with the
process fluid. These generally include the ball (for ball valves), the disk (for butterfly valves), and the plug (for plug and globe valves). Also
included are the seats, which is the area where the plug or disk "sits" when closed to provide the actual shut off. The seals and the valve
body are usually wetted as well.
There are many resources that contain what resources are compatible with a wide variety of process fluids, such as the Cole Parmer
Chemical Resistance Database and the Cat Pumps’ Chemical Compatibility Guide
There are design parameters inherent in the valve designs themselves that increase safety. For high service pressures (or in case of fire)
some valves are designed with initial flexible seal rings that function as the primary seals. Behind these primary seals would be a backup
seal of a more durable material such as 316 stainless, inconel or hastelloy. These backup seals assist in handling the additional pressure and
heat.
In the highly specialized case where the process fluid is so dangerous or unsafe that any release of process fluid is unacceptable, the valve’s
packing can be slightly pressurized with a barrier fluid. As long as the pressure of the barrier fluid is higher than the process fluid, any
leakage between the valve and the process will leak barrier fluid into the process, and not process fluid into the environment. Though as a
side note, these applications usually require double containment piping and a whole host of other precautions beyond simply the safety of
the valve. The most common barrier fluid is water or a water/antifreeze mix for freeze protection.
Some other considerations when selecting a material for valve are longevity/reliability of the valve and the temperature range of usage. If
the valve is a control valve that gets constant use, it is important to select durable materials or to plan for replacement of the valve
frequently. Service temperature is also important; materials need to be selected so the mechanical integrity of the valve is maintained
throughout the entire service temperature.
3.9.8: CONCERNS
3.9.8.1: NOISE
Noise in a system is created by the turbulence from a throttling valve. Throttling is when a valve is not used as a simple OPEN/CLOSE
valve, but instead to control a flow rate.
It can be quite loud and damage people's hearing as well as the equipment. Two methods can be used to control noise. The first is to use a
quiet trim that hinders system noise before it begins and the second is to treat the path (piping) where the noise already occurs. To do this,
one can use thicker piping or insulation.
3.9.8.2: CAVITATION
As previously mentioned, at the point where the fluid’s velocity is at its highest, the pressure is at its lowest. Assuming the fluid is
incompressible (liquid), if the pressure falls down to the vapor pressure of the fluid, localized areas of the fluid will vaporize forming
bubbles that collapse into themselves as the pressure increases downstream. This leads to massive shock waves that are noisy and will
certainly ruin process equipment. This is a process called cavitation.
For a given flow rate, a valve that has the smallest minimum flow area through the valve will have the most problems with cavitation (and
choking, as mentioned in the next section).
3.9.8.3: CHOKING
If the fluid is compressible and the pressure drop is sufficient, the fluid could reach sonic velocity. At this point the static pressure
downstream of the valve grows above design limits. This is a process known as choking, since it slows down, essentially “choking,” the
flow through the valve.
For a given flow rate, a valve that has the smallest minimum flow area through the valve will have the most problems with choking.
Choking is important for a gas-using process since the mass flow rate only depends on the upstream pressure and the upstream temperature.
It becomes easier to control the flow rate since valves and orifice plates can be calibrated for this purpose. Choke flow occurs with gases
when the ratio of the upstream pressure to the downstream pressure is greater than or equal to [ ( k + 1 ) / 2 ]^( k / ( k - 1 )). If the upstream
pressure is lower than the downstream, it is not possible to achieve sonic flow.
When choking occurs in liquids, it is usually due to the Venturi effect. If the Venturi effect decreases the liquid pressure lower than that of
the liquid vapor pressure at the same temperature, it will cause the bubbles in the liquid. These bubble burst can cause enough turbulence
(known as cavitations) that can lead to physical damage to pipes, valves, controllers, gauges, and/or all other equipments linked to that
section of the flow.
As mentioned above the limiting case of the Venturi effect is choked flow as the smaller space in a pipe will but a cap on the total flow rate.
The Venturi effect is fluid pressure resulting from an incompressible fluid travelling through a tight or constricted section of a pipe. The
Venturi effect can be explained by principles such as Bernoulli’s principle and the continuity equation.
Bernoulli's equation :
Continuity equation:
for incompressible fluids:
To satisfy the continuity equation, the velocity of the fluid must increase in the space but the pressure must decrease to agree with
Bernoulli’s equation, which is related to the law of conservation of energy through a pressure drop.
Venturi tubes are used to control the fluid speed. If an increase in speed is wanted, then the tube will decrease in diameter, or projected
area, as the fluid moves away from the pump or energy source. If a decrease in fluid velocity is wanted, the tube will increase in diameter as
it moves away from the energy source (which could be a pump). Using a venturi tube sometimes introduces air into the system which is
mixed with the fluid in the pipe. This causes a pressure head at the end of the system.
Venturi tubes can be replaced by orifice plates in some cases. Orifice plates use the same principles as the venturi tubes but cause the
system to suffer more permanent energy loss. Orifice plates, however, are less expensive to build then venturi tubes.
3.9.8.4: UPSTREAM PRESSURE INCREASE
If a control valve is sized incorrectly and is too small, fluid velocity upstream of the valve will slow, causing an increase in pressure (much
like when the end of a garden hose is partially obstructed). This increase in upstream pressure can be detrimental to certain processes, i.e.
membrane filtration processes, where a large pressure difference across the membrane is desired. If the valve increases the pressure on the
permeate side of the membrane, the driving force for separation will be reduced.
Decreasing the upstream static temperature will cause an increase in upstream static pressure therefore increasing the mass flow rate, so it is
important to keep track of your temperature.
3.9.8.5: HYSTERESIS
Hysteresis, in general, is defined as the phenomenon where previous events can influence subsequent events in a certain process. This can
create a relationship in a process going one direction, however, when carrying out the same process in the exact opposite direction, the same
relationship does not occur. When processes exhibiting hysteresis are graphed there is an area between the two equally opposite curves that
is used to describe the extent of hysteresis seen in the process.
The most commonly described process exhibiting hysteresis involves elastics and can be the loading and unloading of force (weight) to a
stretchable rubber band. As you load more and more weight onto a rubber band, it will stretch in response to increased weight. You can
subsequently remove the weight in similar increments and observe the ability of the rubber band to compress itself. Rubber bands, however,
do not obey Hooke's Law perfectly therefore as you unload, the rubber band will be longer at the corresponding weights as you loaded.
Below is a graphical representation of this phenomenon.
As seen in the above graph, for any given force applied to the rubber band, the extension of unloading is larger than the extension when
loading. This is showing how the process of loading effected the outcome of the process of unloading. The area of the graph between the
two curves is the area of hysteresis and, in this case, represents the energy dissipated as heat. The larger this area the less efficient this
rubber band was at maintaining shape, and consequently the worse it obeys Hooke's Law.
Hysteresis, regarding control valves, is the principle that a control valve is dependent on the prior valve position. A common valve opening
may correlate to different flow rates depending on if the valve was opened or closed to get to that position. This shows that the direction a
valve is being changed may need to be accounted for in a control system to obtain a desired flow rate. If hysteresis becomes too large, it can
cause the control architecture to force a system to oscillate around a desired point.
3.9.8.6: CHATTER
Chatter is mainly a concern when selecting and sizing safety valves. Chatter is the rapid opening and closing of a safety valve or pressure
relief device. Commonly, chatter is caused due to a large pressure drop between the vessel being relieved and the valve relieving the
pressure.
If the pressure behind a closed safety valve builds up to some critical pressure, it will open. Once open, fluid will flow from the vessel being
relieved out through the safety valve in order to reduce pressure within the vessel. It is possible to have enough pipe length, fittings, or other
energy-reducing obstacles so that once the high pressure flow reaches the safety valve, the pressure in the fluid is again below the critical
pressure of the valve. Since there is not enough pressure in the fluid to keep the safety valve open, it closes and the vessel re-pressurizes,
causing the safety valve to pop open again, locally depressurize and close again. This opening and closing is called chatter. The vibrations
caused by chatter can be extremely damaging, causing unnecessary safety valve wear, possible seat-misalignment within the fittings, and
even full failure in the valve or surrounding piping.
Chatter can be avoided by making sure that at critical release rates the pressure drop across the fittings to the safety valve is not large
enough such that the valve will shut once fluid release is initiated.
3.9.9: ACTUATORS
Actuators are the mechanical equipment that supply the force necessary to open or close a valve. Actuators are, essentially, the alternative to
manual operation of a valve. The method of applying the opening/closing force to a valve is what differentiates the various types of
actuators. When selecting the actuator, the most important feature to specify is whether you want fail-safe open or closed. This is
determined entirely by a careful analysis of the process to decide which is safer. If all the power goes out or some other emergency occurs,
the decision on the fail-safe mode of the valve is a huge factor in saving lives. For example, if a reaction is exothermic, the fail-safe mode of
the cooling water in the jacket of the reactor should be fail-safe open. Pneumatic actuators have a delayed response which make them ideal
for being resilient against small upsets in pressure changes of the source. Hydraulic actuators on the other hand use an incompressible fluid,
so the response time is essentially instantaneous.
3.9.9.1: PNEUMATIC
Pneumatic actuators are the most popular type of actuators. The standard design of a pneumatic actuator consists of a pre-compressed spring
that applies force against a disk on a sealed flexible chamber. The disk is usually attached to the stem of the valve it is intended to control.
As the chamber is compressed with air, the chamber expands and compresses the spring further, allowing axial motion of the valve stem.
Knowing the relationship between the air pressure in the chamber and the distance the stem moves allows one to accurately control flow
through the valve.
The biggest advantage of the pneumatic actuators is their failsafe action. By design of the compressed spring, the engineer can determine if
the valve will fail closed or open, depending on the safety of the process. Other advantages include reliability, ease of maintenance, and
widespread use of such devices.
3.9.9.2: MOTION CONVERSION
Motion conversion actuators are generally used to adapt a common translational motion from the actuator’s output to a rotary valve. The rod
that moves axially from the translational motion actuator is connected to a disk and the connection is pivoted. The disk itself is also pivoted
about its center. This system of pivots allows the translational motion to be converted into the rotation of the disk, which would open or
close the rotary valve.
The main advantage of this setup is that an inexpensive translational motion actuator can be used with rotary valves. The key drawback is
that the applications in which this can be used is very limited. Specifically, this setup is useless in the common case where the rotary motion
required is greater than 90°.
3.9.9.3: HYDRAULIC
Hydraulic actuators work using essentially the same principal as pneumatic actuators, but the design is usually altered. Instead of a flexible
chamber, there is a sealed sliding piston. Also, instead of using a spring as the opposing force, hydraulic fluid is contained on both sides of
the piston. The differential pressure across the area of the piston head determines the net force.
Hydraulic actuators offer the advantages of being small and yet still providing immense force. Drawbacks of hydraulic actuators are
primarily the large capital cost and difficulty maintaining them.
3.9.9.4: ELECTRIC
Electric actuators typically use standard motors, powered by either AC induction, DC, or capacitor-start split-phase induction. The motor is
connected to a gear or thread that creates thrust to move the valve. As a failsafe, some motors are equipped with a lock in last position on its
gear. This means that the gear cannot move from forces outside of the electric motor. This helps prevent overshoot on the motor as well as
helps create better positioning for the gear.
Another type of motor that can be used is called a stepper motor. It uses increments on gear reduction to alleviate problems with positioning
and overshoot. The increments are in a range of 5,000 to 10,000 increments in a 90 degree rotation.
A problem with using electric actuators is that a battery operated back-up system is needed or else the system is useless during power
failure. Also, the actuator needs to be in an environment that is rendered safe, meaning a non-explosive environment.
3.9.9.5: MANUAL
Manual actuators are usually used for overrides of power actuators described above. This is an important safety measure in case the power
actuator fails. Manual actuators typically consist of either a lever or a wheel (used for larger valves) connected to a screw or thread that
turns the valve.
3.9.10: SUMMARY TABLES
The table below shows a list of typical icons for different valves found in industry.
3.9.11: VALVE AND ACTUATOR SELECTION EXAMPLE

Note: This problem is completely fictionalized. Any relation to a real occurrence is completely coincidental.
A chemical engineer wants to use a valve at a start/stop flow of liquid water to a CSTR. The determination of the start/stop flow is a level
sensor in the CSTR. Once the water reaches a certain level inside the CSTR, the flow of the water will stop. If the water level drops to a
certain point, water will need to quickly flow into the CSTR. Your job is to list possible valves, and determine if each valve type can be used
to start/stop the flow of water. Also, list all the potential actuators for the valves and determine which actuators could be used and how.
Some issues to consider are choking, throttling, longevity, and reliability.
3.9.11.1: SOLUTION
Valves:
Ball valve: A ball valve can work for this example. A 2-way ball valve may not be the most ideal type of valve depending on the flow rate.
If the water is moving at a high velocity, the 2-way ball valve will be susceptible to choking. If a ball valve is used, a cage valve would be
the best choice, due to its durability and its on/off mechanics.
Globe valve: Just like the caged ball valve, a globe valve is a conceivable solution. The plug that would best help prevent water from
flowing in the CSTR quickly would be the quick opening plug. This way, there will be no throttling when closing the valve, creating a quick
on/off.
Butterfly valve: If a butterfly valve is selected, it would need to be a high performance valve. The benefits of this valve include a tight
shutoff and improved torque over the non-high performance butterfly valves. This is important for the on/off capabilities for water flow the
example calls for. Using a regular butterfly valve could lead to water leaking into the CSTR or torque issues after constant use.
Plug valve: A plug valve could work, but would encounter similar problems as a 2-way ball valve with choking. Also, the friction created
when constantly rotating could cause a problem in the future with the valve not closing properly.
Actuators:
Pneumatic: A pneumatic actuator has a good potential to be used in this example. Its failsafe action is great in case there is a failure in the
system. Its low maintenance and high reliability means that the pneumatic actuator is an ideal actuator for the valve.
Motion Conversion: Motion conversion actuators would be ideal for any rotating valve, like the high performance butterfly valve.
Therefore, this actuator should be used with rotational valves for best reliability.
Hydraulic: Hydraulic actuators have similar advantages that the pneumatic actuators have. Hydraulic actuator, however, cost more than
pneumatic actuators. Therefore, if another option exists, hydraulic actuators should not be used for this example.
Electric: An electric actuator will work best with rotational valves. So, if a high performance butterfly valve was chosen, then an electric
actuator could possibly be used. As an effort to limit throttling issues, rotational models for ball valves would be less ideal. This is, of
course, on the assumption that a back up battery would be included with the electric actuator
Manual: A manual actuator could be used in conjunction with any other actuator, but would it not be a good idea to use a manual actuator
alone. The manual actuator would be a good backup method if any of the other actuators fail, but tedious if used as the primary actuator.
3.9.12: SELECTING CONTROL VALVE EXAMPLE

Note this example problem and the data contained within fictitious, with the exception of some of the Cv data, which is taken from
Clark's Product Specification Site
An engineer in a chemical factory is asked to decide what type of control valve to use in throttling the line from a heat exchanger to a
reactor. The chemical is part of a very sensitive reaction in which the precise addition of the chemical in question is crucial. If an imprecise
valve is utilized, the entire reactor's $5 million worth of product would need to go down the sewer. The process considerations are as
follows:
The process is designed to pump the liquid chemical at 80 psig and 250 GPM, though the flow occasionally needs to be varied from 125 to
400 GPM. The chemical has a density of 0.60 kg/L (S.G. = 0.6). The heat exchanger should take 10 psi of the pressure drop across it. Also,
the reactor is operated at 30 psig Assume turbulent non-choked flow.
3.9.12.1: SOLUTION:
First one must choose the type of valve. The four main valves of interest are ball, butterfly, plug and globe. Since the application is a
throttling application a ball valve probably should not be our first choice. Since the chemical needs to be throttled at an extremely accurate
rate, a plug valve would not be best suited here. Also, since there is significant variability of the desired flowrate, a butterfly valve probably
is not the best decision either. A globe valve seems to work well here.
For reference, a list of Cv’s by valve size (when fully opened) is listed below. The Cv's of each valve are listed also by % valve travel (as
can be commonly acquired from a vendor), where 100% means the valve is fully open.
Now one must size the globe valve. Recall the equation
−−−−
Gf
Cv = Q√
ΔP
We can see that the pressure drop across the valve is the pressure at the pump minus the pressure at the reactor minus the pressure drop
across the heat exchanger. Thus
Delta P = 80 − 30 − 10 = 40 ψ
So getting Cv,
−−−
0.6
Cv = 250√ = 30.62
40
Looking at the chart it would appear that the 2” Globe valve would work, but remember, the process fluid may sometimes need to be
controlled up to 400 GPM. And the Cv at 400 GPM is:
−−−
0.6
Cv = 400√ = 49.0
40
Looking at this it is clear that the 2” globe valve would simply be fully open at 400 GPM. It appears that the 2 ½” globe valve would be a
better choice since it has the capability to throttle up to the maximum design flowrate.
Checking the Cv at the minimum flow that will need to be throttled, 125 GPM:
−−−
0.6
Cv = 125√ = 15.3
40
We see that the 2 ½” globe valve will be between 25 and 50% open, so it will throttle well at these low flow rates at also. This reaffirms that
the 2 ½” globe valve will work well for this application.
 EXAMPLE 3.9.1
When designing a jacketed CSTR that will contain a highly exothermic reaction, what type of actuator should be used to operate the
valve controlling the cooling water?
a. pneumatic
b. motion conversion
c. hydraulic
d. electric
Solution
a
 EXERCISE 3.9.2
Which of the following is a good application for a diaphragm valve?

a. inlet to a CSTR where precise volumes of reactants are required
b. food industry where ease of cleaning is important
c. steam heating line for jacketed CSTR
d. vent on a signal line to release pressure to atmosphere
Answer
b
3.9.13: REFERENCES
Q = flow rate (ft3 / h)
Gf = specific gravity of the gaseous fluid (dimensionless)
ΔP = absolute pressure drop through the valve (psia)
Po= outlet pressure (psia)
Pi = inlet pressure (psia)
T = gas temperature (degrees F)
Edgar, Thomas F., Sale E. Seborg, and Duncan A. Mellichamp. Process Dynamics and Control. New York: John Wiley & Sons.
Liptak, Bela G., Process Control and Optimization. Vol. II. New York: Taylor & Francis.
Perry, Robert H.(1997), Perry's Chemical Engineers' Handbook, 7th Edition. McGraw-Hill. ISBN 0070498415
"Valves", Cole-Parmer, 2005
Considine, Douglas M. (1993), Process/Industrial Instruments & Controls Handbook, 4th Edition, McGRAW-HILL, INC. ISBN
0070124450
Smith, Peter; Zappe, R.W. (2004),Valve Selection Handbook. Elsevier
Matthews,Clifford . (2004),A Quick Guide to Pressure Relief Valves (PRVs). John Wiley and Sons. ISBN 1860584578, 9781860584572
Websites for steam traps: www.energysolutionscenter.org/BoilerBurner/Eff_Improve/Steam_Distribution/Steam_Traps.asp
http://en.Wikipedia.org/wiki/Steam_trap www.iklimnet.com/expert_hvac/steam_traps.html
−−−−−−
4Fmax
d = √
πv
Where:
3.9: Valves - Types, Kinds, and Selection is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source
content that was edited to conform to the style and standards of the LibreTexts platform; a detailed edit history is available upon request.
3.10: VALVES - MODELING DYNAMICS
Authors: Erin Knight, Matthew Russell, Dipti Sawalka, Spencer Yendell
3.10.1: INTRODUCTION
A valve acts as a control device in a larger system; it can be modeled to regulate the flow of material and energy within a process. There are
several different kinds of valves (butterfly, ball, globe etc.), selection of which depends on the application and chemical process in
consideration. The sizing of valves depends on the fluids processing unit (heat exchanger, pump etc.) which is in series with the valve.
Sizing and selection of valves is discussed in the other wiki article on Valve Selection. Valves need to be modeled to perform effectively
with respect to the process requirements. Important components for the modeling of control valves are:
1. Flow
2. Inherent Flow Characteristics
3. Valve Coefficient, Cv
4. Pressure Drop
5. Control Valve Gain
6. Rangeability
7. Installed Characteristics
Efficient modeling of the valves can optimize the performance and stability of a process as well as reduce development time and cost for
valve manufacturers.
In the following sections we briefly define the various variables and equations involved in modeling valves. The purpose of the following
sections is to give you an overview of the equations required to model the valves for a particular system. Example problems at the end of the
article have been provided to aid in the qualitative and quantitative understanding of how valves are modeled for chemical engineering
processes.
3.10.2: FLOW THROUGH A VALVE

The following equation is a general equation used to describe flow through a valve. This is the equation to start with when you want to
model a valve and it can be modified for different situations. The unfamiliar components such as valve coefficient and flow characteristics
will be explained further.
−−−−
ΔPv
F = Cv f(x)√
sg
with
F is the volumetric flow rate
Cv is the valve coefficient, the flow in gpm (gallons per minute) that flows through a valve that has a pressure drop of 1psi across the
valve.
ΔP is the pressure drop across the valve
v
sg is the specific gravity of fluid
x is the fraction of valve opening or valve "lift" (x=1 for max flow)
f(x) is the flow characteristic
3.10.2.1: FLOW CHARACTERISTICS

The inherent flow characteristic, f(x), is key to modeling the flow through a valve, and depends on the kind of valve you are using. A flow
characteristic is defined as the relationship between valve capacity and fluid travel through the valve.
There are three flow characteristics to choose from:
1. f(x) = x for linear valve control
2. f(x) = √−x for quick opening valve control
3. f(x) = R x−1
for equal percentage valve control
R= valve design parameter (between 20 and 50)
note these are for a fixed pressure drop across the valve
Whereas a valve TYPE (gate, globe or ball) describes the geometry and mechanical characteristics of the valve, the valve CONTROL refers
to how the flow relates to the "openness" of the valve or "x."
1. Linear: flow is directly proportional to the valve lift (used in steady state systems with constant pressure drops over the valve and in
liquid level or flow loops)
2. Equal Percentage - equal increments of valve lift (x) produce an equal percentage in flow change (used in processes where large drops in
pressure are expected and in temperature and pressure control loops)
3. Quick opening: large increase in flow with a small change in valve lift (used for valves that need to be turned either on or off frequently
or where instant maximum flow is required, for example, safety systems)
For the types of valves discussed in the valve selection article, the following valve characteristics are best suited:
1. Gate Valves - quick opening
2. Globe Valves - linear and equal percentage
3. Ball Valves - quick opening and linear
4. Butterfly Valves - linear and equal percentage
Cvmax depends on pipe characteristics and was chosen to be 110 gpm in this example. Constant pressure throughout the pipe line is assumed
and the curves are accurate when the valve position is between 5% and 95% open.
Comparing the slopes of the graphs for the quick opening and equal percentage valves, we can see that a quick opening valve would
experience greater change in flow with slight change in valve position in the lower flow range. The opposite is true for the higher range of
flow. The equal percentage valve experiences slighter change in flow with respect to valve position in the lower range of flow.
When selecting the appropriate control valve, it is often the goal of the engineer to choose a valve that will exhibit a linear relationship
between F and x over the normal operating position(s) of the valve. This linear relationship provides the most control for the operator. The
flow characteristic observed through an installed valve, and all process factors considered (i.e. total pressure drop, etc.), is termed the
installed flow characteristic. Therefore, it is not always the case that an inherently linear valve is desirable or even useful. An inherently
linear valve is appropriate when there is a linear relationship between the valve position and the actual flow rate; however, consider the case
where and of significant value. In this case a valve with inherent equal percentage flow characteristic would be appropriate. The
inherently non-linear valve would compensate for ΔPL and result in an installed linear flow characteristic.
3.10.2.2: VALVE COEFFICIENT, CV

The valve coefficient, C , is defined as the flow in gpm that flows through a valve with a pressure drop of 1psi across the valve (ΔPv =
v
1psi). Cv is an important parameter that comes up in other modeling equations. It is specific to the valve you are using.
2
29.9d
Cv = −
−
√K
d = internal diameter of the pipe in inches

K = resistance coefficient
K is specific to the pipe shape, diameter and material. Table of typical K values
3.10.3: PRESSURE DROP

The pressure drop in the pipe line (pressure drop due to the pipe line and any other equipment in series with the valve), ΔPL, is defined as:
2
ΔPL = kL × sg × f
f = flow through the pipe in gallons per minute [gpm]
kL = [ ] = constant friction coefficient for the pipe and any equipment in series with the valve
sg = specific gravity of the liquid
The pressure drop across the valve is defined as:
2
f
ΔPv = sg
2
(Cv )
So, the total pressure drop is described by the equation:

1 2
ΔPo = ΔPv + ΔPL = ( + kL ) sgf
2
Cv
If the line pressure drop is negligble (constant pressure in the pipe line) then ΔPL = 0 and ΔPo = ΔPv. When ΔPL = 0 a valve with a linear
flow characteristic will be desirable. When and of significant value, a valve with flow characteristics closer to an equal
percentage or quick opening valve will be more desirable.
3.10.4: CONTROL VALVE GAIN

The gain of a control valve (KL) is defined as the steady-state change in output (flow through a valve, f ) divided by the change in input
(controller signal, m). The flow through a valve, f, can have units of gallons per minute (gpm), pounds per hour (lb/hr) or standard cubic feet
per hour (scfh). The controller signal, m, usually has units of percent of controller output (%CO). The basic relationship for control valve
gain is shown below.
df
Kv =
dm
One objective when choosing a valve is to achieve "constant valve gain". The gain is a product of the dependence of valve position on
controller output, the dependence of the flow on Cv, and the dependence of Cv on the valve position. The change in valve coefficient, Cv,
with respect to valve position depends on the valve characteristics f(x).
For linear characteristics:
dCv
= C vmax
dx
For equal percentage:

dCv
= (ln R)Cv
dx
3.10.4.1: CONSTANT PRESSURE DROP

The dependence of flow on the Cv depends on the pressure drop, so the equation for gain is different when there is a constant pressure drop
or a variable pressure drop. If the inlet and outlet pressures do not vary with flow, the gain for either liquid or gas flow in mass units is:
%CO = percent controller output

W = mass flow rate
R = valve design parameter (usually between 20 and 50)
Note: the sign is positive if the valve fails closed (air-to-open) and negative if the valve fails open (air-to-close)
3.10.4.2: VARIABLE PRESSURE DROP

The valve gain for variable pressure drop is more complicated. As an example, the gain for an equal percentage is
ln α flow gpm
Kv = ±
100 2
(1 + kL Cv ) %CO
kL = constant friction coefficient for line, fittings, equipment, etc.

The flow term cancels some of the effect of the Cv term until the valve is fully opened, so this gain is less variable with valve opening.
Therefore the installed characteristics are much more linear when compared to the inherent characteristics of an equal percentage valve.
3.10.5: RANGEABILITY
Valve rangeability is defined as the ratio of the maximum to minimum controlable flow through the valve. Mathematically the maximum
and minimum flows are taken to be the values when 95% (max) and 5% (min) of the valve is open.
Flow at 95% valve position Rangeability = -------------------------- Flow at 5% valve position
A smaller rangeablilty correlates to a valve that has a small range of controllable flowrates. Valves that exhibit quick opening characteristics
have low rangeablilty values. Larger rangeability values correlate to valves that have a wider range of controllable flows. Linear and equal
percentage valves fall into this category.
Another case to consider is when the pressure drop across the valve is independent of the flow through the valve. If this is true then the flow
is proportional to CV and the rangeability can be calculated from the valve's flow characteristics equation.
3.10.6: MODELING INSTALLED VALVE CHARACTERISTICS

When a valve is installed in series with other pieces of equipment that produce a large pressure drop in the line compared to the pressure
drop across the valve, the actual valve characteristics deviate from the inherent characteristics. At large in-line pressure drops the pressure
drop, and consequently the valve coefficient, varies with flow through the valve. These changes can cause changes in the rangeabilty and
distorts inherent valve characteristics.
In the following Microsoft® Excel model, the variation from inherent valve characteristics is illustrated. A number of parameters can be
changed to match the flow conditions through a valve. This model simulates both linear and equal percentage valve characteristics. To more
clearly demonstrate the deviation from inherent characteristics; simply change the CVmax value of the valve. Notice how the installed valve
characteristics and valve rangeability change drastically.
Installed Valve Characteristcs Model
3.10.7: SPECIAL CONSIDERATIONS FOR THE EQUATION DESCRIBING FLOW THROUGH A VALVE
Compressible Fluids
Manufacturers such as Honeywell, DeZurik, Masoneilan and Fischer Controls have modified the flow equation to model compressible
flows. The equations are derived from the equation for flow through a valve but include unit conversion factors and corrections for
temperature and pressure, which affect the density of the gas. It is important to remember to account for these factors if you are working
with a compressible fluid such as steam.
Accuracy
This equation, and its modified forms, is most accurate for water, air or steam using conventional valves installed in straight pipes. If
you are dealing with non-Newtonian, viscous or two phase systems the calculations will be less accurate.
3.10.8: EXAMPLE 1: VERBAL MODEL OF A CONTROL VALVE

3.10.8.1: PROBLEM STATEMENT:
Verbally model a fail open control valve positioned as a safety measure on a reactor processing an exothermic reaction.
3.10.8.2: SOLUTION
1. Describe the Process: In the fail-open control valve a quick opening valve opens with a failure signal. Open is its default position once
the signal goes off.
2. Identify Process Objectives and Constraints: A fail-open control valve is a safety measure. For example, if your cooling heat
exchanger fails and the reactor starts to heat up and produce excess gases, the fail-open control valve would release excess gasses before
pressure builds up to produce an explosion. The size of the valve is a constraint because it limits how much fluid can escape. The valve
size determines the maximum flow rate. The shape and angles of the valve are modeling constraints. Sudden and gradual contraction or
enlargement of the pipe diameter inside the valve and the connecting piping, will change the resistance coefficient and therefore the
maximum velocity.
3. Identify Significant Disturbances: Significant internal disturbances include the escalating pressure and temperature as the exothermic
reaction gets out of control.
4. Determine the Type and Location of Sensors: A pressure sensor would be located in the tank with the control valve that would
provide the signal to the fail-open control valve. To achieve redundancy, a temperature sensor would be located on the heat exchanger to
signal failure of the cooling apparatus.
5. Determine the Location of Control Valves: A fail-open control valve (or multiple valves) would be placed on the top of the tank to
allow exit of the gasses in the processing unit.
6. Apply a Degree-of-Freedom Analysis: The only manipulated variable is the valve coefficient. This depends on the valve’s diameter
and resistance coefficient K. The control objective is the maximum flow rate. The pressure drop will vary according to the failure.
Therefore there is one degree of freedom.
7. Implement Energy Management: This doesn’t apply to our confined example, but in a larger system we could install a backup cooler
if this reaction were truly dangerous.
8. Control Process Production Rate and Other Operating Parameters: The exit flow rate can not exceed the maximum flow rate
through the control valve.
9. Handle Disturbances and Process Constraints: If our first control valve fails to sufficiently lower the tank pressure, a signal would be
sent to a second valve and depending on the reaction, a backup cooling system. A secondary cooling system would be too expensive for
many cases, but if you were dealing with a nuclear reactor or something highly explosive it may be worth the investment.
10. Check Component Balances: Does not apply. Preventing accumulation is the point of this control valve.
11. Apply Process Optimization: Our manipulatable variable is choosing a valve with a specific Cv. The valve should be able to withstand
extreme temperatures and high pressures. It would be a gate valve, which opens completely upon failure. For other sizing concerns refer
to “Valve Sizing.”
 EXAMPLE 3.10.1: MODELING INSTALLED VALVE CHARACTERISTICS
A new valve is being installed downstream from a water pump. The friction coefficient of the pump and associated piping that will be
in series with the new valve is
−4
psi
kL = 1.4 × 10 ( ).
2
gpm
The flow through the line from the pump is 300 gpm. The desired pressure drop across the valve is 4 psi. A high level of control is
desired for the flow through the new valve. Two valves are being considered, one has an inherent linear characteristic, the other is equal
percentage (α=50). From the manufacturer’s literature, both have a CVmax value of 200. Use the Installed Valve Characteristics Model
to determine which valve has a higher range of controllable flows values.
Solution
Note that the pressure drop across the pipe is 13.5psi, which is significantly larger than the pressure drop across the valve (4 psi). These
conditions indicate that the characteristic flow through the valves may not match the inherent characteristics. This is verified by the
plots and also by the calculated rangeability values shown in the valve model spreadsheet. The equal percentage valve has a higher
rangeabilty value, corresponding to a higher range of controllable flows.
3.10.9: REFERENCES
Bequette, B. Wayne. Process Control Modeling, Design, and Simulation, Upper Saddle River, New Jersey: Prentice Hall.
Crane Co. Flow of Fluids Through Valves, Fittings, and Pipe, Joliet, IL: CRANE.
"Friction Losses in Pipe Fittings"(PDF), Western Dynamics, LLC., retrieved September 11, 2006.
Perry, R. H., and D. Green (ed). Perry's Chemical Engineering Handbook, 7th ed. New York: McGraw-Hill.
Seborg, Dale E., Thomas F. Edgar, Duncan A Mellichamp. Process Dynamics and Control, New York: John Wiley & Sons.
Smith, Carlos A., Armando B. Corripio. Principles and Practice of Automatic Process Control, 3rd ed. New York: John Wiley & Sons.
"Valve Sizing and Selection." The Chemical Engineers' Resource Page. 1442 Goswick Ridge Road, Midlothian, VA 23114. retrieved
Sept 24, 2006.
3.10: Valves - Modeling Dynamics is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source content
that was edited to conform to the style and standards of the LibreTexts platform; a detailed edit history is available upon request.
CHAPTER OVERVIEW
4: PIPING AND INSTRUMENTATION DIAGRAMS

4.1: PandID General Information
4.2: Piping and Instrumentation Diagram Standard Notation
4.3: Piping and Instrumentation Diagrams - Location of Controls and Standard Control Structures
4.4: Piping and Instrumentation Diagrams - Standard Pitfalls
4.5: Safety Features in Piping and Instrumentation Diagrams
4.6: Regulatory Agencies and Compliance
4: Piping and Instrumentation Diagrams is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source
1
4.1: PANDID GENERAL INFORMATION
Piping and Instrumentation Diagrams (P&IDs) use specific symbols to show the connectivity of equipment, sensors, and valves in a control
system. The following sections will outline general information about P&IDs that is necessary to to know before trying to draw one.
4.1.1: P&ID VS. PFD

P&IDs may often be confused with PFDs, or process flow diagram. P&IDs and PFDs generally utilize the same notation for equipment.
However, they serve different purposes and provide different information. The purpose of a PFD is to show exactly what a process does
during operation, and a P&ID shows all controllers, valve types and the materials that are used in construction. A PFD shows the
connectivity and relationships between the major equipment and materials in a process. However, it also includes tabulated design values
such as normal, minimum, and maximum operating conditions that a P&ID does not contain. A PFD does not include minor piping systems
or other minor components that a P&ID typically includes. The difference between P&IDs and PFDs is that P&IDs typically include more
information regarding piping and safety relief valves than process flow diagrams. P&IDs do not contain operating specifications that PFDs
contain, such as stream flows and compositions. It is important to note that differences between PFDs and P&IDs will vary between
institutions. Most corporations maintain designated standards to create and modify the documents. Both PFDs and P&IDs are controlled
documents and need to be maintained with a document control procedure. To see an industrial example of PFD vs P&ID refer to, PFD/PID
industry example Wiki Page.
4.1.2: INFORMATION INCORPORATED IN P&IDS

The following information is given on a P&ID that is not explicit on a PFD:
ALL valves and valve types
Controllers present
Controller architectures
Pipe diameters, materials of construction, and insulation properties (including minor piping systems)
Equipment materials of construction
4.1.3: USES OF P&IDS

Develop operational methodology
Develop safety philosophy and safeguards
Develop control philosophy
Serve as a basis for control programming
Serve as a communication document for how the process works
Serve as a basis for equipment design, pipe design, estimating cost, purchasing
Use for evaluation of construction process
Train employees
Serve as a conceptual layout of a chemical plant
Provide a common language for discussing plant operations
4.1.4: CHARACTERISTICS OF P&IDS

Grouped by specific section of plant
Show the connections between all the sensors and actuators
A general schematic - NOT a layout and NOT to scale. It should be noted that P&IDs do not specifically imply the following: same
elevation of equipment, relative sizes, where valves are located on the equipment, how close equipment is to each other, and impeller
types/location. They are also not the same as control or incidence diagrams. This type of information can be seen in either a plant layout
drawing (can be either satellite view, showing distance between units, or a slice of building, showing height of units) or construction
drawings, such as plant blueprints.
Must be clear and uncluttered
Must be systematic and uniform. P&IDs are used extensively in industry to document plant information. These documents need to be
easily read by anyone working within the company, and easily explained to anyone else. OSHA audits can occur anytime and it is
imperative that operational information can be provided to the auditor when requested. Without standard notation, it would be very
difficult to go from plant to plant within your company and understand the P&IDs.
Are generally highly confidential and have restricted access, as they can be used to replicate a process
4.1.5: WHAT A P&ID IS NOT
Not an architectural diagram of a process (should show the flow of material across the plant floor between sensors and actuators, not
necessarily corresponding to a 3D location)
Does not need to be drawn perfectly to scale
Does not imply any relative elevations
Do not need manual switches
No actual temperature, pressure, or flow data
Leave out any extensive explanations
4.1.6: WHAT A P&ID SHOULD INCLUDE

Instrumentation and designations
Mechanical equipment with names and numbers, and their specifications such as material, insulation, maximum flow rate, working
pressure and temperature, maximum power etc.
All valves and their identifications
Process piping, sizes and identification
Miscellaneous - vents, drains, special fittings, sampling lines, reducers, increasers and swagers
Permanent start-up and flush lines
Flow directions
Interconnections references
Control inputs and outputs
Interfaces for class changes
Vendor and contractor interfaces
Identification of components and subsystems delivered by others
Intended physical sequence of the equipment
4.1.7: P&ID REVISIONS

Revisions should be clearly identified
Regularly issued to all related employees at each significant change to the process, as well as at benchmark points
If small changes are made that don't warrant a completely new revision, "red pencil" additions are generally accepted between issues
15-20 revisions are typical during process design
All revisions need to be communicated to EVERYONE so that only the latest revision is used. This is critical in order to avoid serious
(not to mention expensive) construction mistakes. (Typically outdated P&ID is discarded to avoid confusion)
4.1.8: HOW DO YOU GENERATE A P&ID?

P&IDs can be created by hand or computer. Common programs, for both PC and Mac, that create P&IDs include Microsoft Visio (PC) and
OmniGraffle (Mac). For creating a P&ID by hand or on a computer, please refer to the P&ID Standard Notation Wiki Page for standard
equipment notation as well as computer templates.
4.1: PandID General Information is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source content
4.2: PIPING AND INSTRUMENTATION DIAGRAM STANDARD NOTATION
Authors: Halley Crast, Andrew Laskowski, Maurice Telesford, Emily Yatc
By: Stephanus Oscar, Kaitlin Harrington , Suhendra Lie
4.2.1: INTRODUCTION
Piping and Instrumentation Diagrams (P&IDs) use specific symbols to show the connectivity of equipment, sensors, and valves in a control
system. These symbols can represent actuators, sensors, and controllers and may be apparent in most, if not all, system diagrams. P&IDs
provide more detail than a process flow diagram with the exception of the parameters, i.e. temperature, pressure, and flow values. "Process
equipment, valves, instruments and pipe lines are tagged with unique identification codes, set up according to their size, material fluid
contents, method of connection (screwed, flanged, etc.) and the status (Valves - Normally Closed, Normally Open)."[1] These two diagrams
can be used to connect the parameters with the control system to develop a complete working process. The standard notation, varying from
letters to figures, is important for engineers to understand because it a common language used for discussing plants in the industrial world.
P&IDs can be created by hand or computer. Common programs, for both PC and Mac, that create P&IDs include Microsoft Visio (PC) and
OmniGraffle (Mac). As with other P&IDs, these programs do not show the actual size and position of the equipment, sensors and valves,
but rather provide a relative positions. These programs are beneficial to produce clean and neat P&IDs that can be stored and viewed
electronically. See below for P&ID templates for these programs.
This section covers four main types of nomenclature. The first section describes the use of lines to describe process connectivity. The
second section describes letters used to identify control devices in a process. The third section describes actuators, which are devices that
directly control the process. The final section describes the sensors/transmitters that measure parameters in a system.
4.2.2: LINE SYMBOLS

Line symbols are used to describe connectivity between different units in a controlled system. The table describes the most common lines.
Table 1: Line Symbols
In Table 1, the "main process" refers to a pipe carrying a chemical. "Insulated" is straightforward, showing that the pipe has insulation.
"Trace heated" shows that the pipe has wiring wrapped around it to keep the contents heated. "Lagged" indicates on a P&ID that the pipe is
wrapped in a cloth or fiberglass wrap as an alternative to painting to improve the appearance of the pipe see here for more information. The
last column in Table 1 shows pipes that are controlled by a controller. "Electrical impulse" shows that the manner in which information is
sent from the controller to the the pipe is by an electrical signal, whereas "pneumatic impulse" indicates information sent by a gas.
In addition to line symbols, there are also line labels that are short codes that convey further properties of that line. These short codes consist
of: diameter of pipe, service, material, and insulation. The diameter of the pipe is presented in inches. The service is what is being carried in
the pipe, and is usually the major component in the stream. The material tells you what the that section of pipe is made out of. Examples are
CS for carbon steel or SS for stainless steel. Finally a 'Y' designates a line with insulation and an 'N' designates one without it. Examples of
line short codes on a P&ID are found below in Figure A.
Figure A: Line Labels
This is useful for providing you more practical information on a given pipe segment.
For example in stream 39 in Figure A, the pipe has a 4" diameter, services/carries the chemical denoted 'N', is made of carbon steel, and has
no insulation.
4.2.3: IDENTIFICATION LETTERS

The following letters are used to describe the control devices involved in a process. Each device is labeled with two letters. The first letter
describes the parameter the device is intended to control. The second letter describes the type of control device.
Table 2: First Identification Letter
Table 3: Second Identification Letter
For example, the symbol “PI,” is a “pressure indicator.”
4.2.4: VALVE SYMBOLS

The following symbols are used to represent valves and valve actuators in a chemical engineering process. Actuators are the mechanisms
that activate process control equipment.
Table 4: Valve Symbols
Table 5: Valve Actuator Symbols
4.2.5: GENERAL INSTRUMENT OR FUNCTION SYMBOLS
Instruments can have various locations, accessibilities, and functionalities in the field for certain processes. It is important to describe this
clearly in a P&ID. Below is a table of these symbols commonly used in P&IDs.
Discrete instruments are instruments separate or detached from other instruments in a process. Shared display, shared control instruments
share functions with other instruments. Instruments that are controlled by computers are under the "computer function" category.
Instruments that compute, relay, or convert information from data gathered from other instruments are under the "Programmable logic
control" section.
For example, a discrete instrument for a certain process measures the flow through a pipe. The discrete instrument, a flow transmitter,
transmits the flow to a shared display shared control instrument that indicates the flow to the operator. A computer function instrument
would tell the valve to close or open depending on the flow. An instrument under the "Programmable logic control" category would control
the valve in the field if it was pneumatically controlled, for instance. The instrument would gather information from discrete instruments
measuring the position of the actuator on the valve, and would then adjust the valve accordingly.
In the chart above, it is necessary to know where the instrument is located and its function in order to draw it correctly on a P&ID. A
primary instrument is an instrument that functions by itself and doesn't depend on another instrument. A field mounted instrument is an
instrument that is physically in the field, or the plant. Field mounted instruments are not accessible to an operator in a control room. An
auxiliary instrument is an instrument that aids another primary or auxiliary instrument. Primary and auxiliary instruments are accessible to
operators in a control room.
4.2.6: TRANSMITTER SYMBOLS

Transmitters play an important role in P&IDs by allowing the control objectives to be accomplished in a process. The following are
commonly used symbols to represent transmitters.
Below are three examples of flow transmitters. The first is using an orifice meter, the second is using a turbine meter, and the third is using
an undefined type of meter.
Table 6: Transmitter Symbols
The location of the transmitter depends on the application. The level transmitter in a storage tank is a good example. For instance, if a
company is interested in when a tank is full, it would be important for the level transmitter to be placed at the top of the tank rather than the
middle. If the transmitter was misplaced in the middle because a P&ID was misinterpreted then the tank would not be properly filled. If it is
necessary for the transmitter to be in a specific location, then it will be clearly labeled.
4.2.7: MISCELLANEOUS SYMBOLS

The following symbols are used to represent other miscellaneous pieces of process and piping equipment.
Table 7: Process Equipment
Table 8: Line Fittings
Table 9: Pipe Supports
4.2.8: CRAFTING A P&ID

In order to greatly simplify P&ID diagrams for the purposes of this class, a standard convention must be employed. This convention
simplifies the many control devices that need to be used. For the sake of brevity, sensors, transmitters, indicators, and controllers will all be
labeled on a P&ID as a controller. The type of controller specified (i.e. temperature or level) will depend on the variable one wished to
control and not on the action needed to control it.
For instance, consider if one must control the temperature of fluid leaving a heat exchanger by changing the flow rate of cooling water. The
actual variable to be controlled in this case is temperature, and the action taken to control this variable is changing a flow rate. In this case, a
temperature controller will be represented schematically on the P&ID, not a flow controller. Adding this temperature controller to the P&ID
also assumes that there is a temperature sensor, transmitter, and indicator also included in the process.
As you can see on the P&ID above, these controllers are represented as circles. Furthermore, each controller is defined by what it controls,
which is listed within arrow boxes next to each controller. This simplifies the P&ID by allowing everyone the ability to interpret what each
controller affects. Such P&IDs can be constructed in Microsoft Office Visio.
4.2.9: SAMPLE DIAGRAM

Below is a sample P&ID Diagram that is actually used in an industrial application. It is clearly more complicated than what has been
detailed above, however, the symbols used throughout remain the same.
Table 10: Sample P&ID Diagram
4.2.10: EXAMPLE 1
Describe the following controlled process in words:
Answer: Reactants enter a jacketed CSTR where a reaction takes place and the products exit. The reactor is cooled via a coolant water
stream. The temperature inside the reactor vessel is monitored with a temperature controller (also contained in the controller is a sensor,
indicator, and transmitter), which electrically controls a valve. The valve can alter the flowrate of the coolant water stream, thereby
controlling the temperature inside the reactor. A pressure controller is also present which feeds back to an inlet valve. Therefore, we can
deduce that this reaction is most likely gas phase and if the CSTR becomes too full (high pressure) the inlet valve will close.
4.2.11: EXAMPLE 2
Draw a proper P&ID diagram of the following process:
A storage tank is filled with condensed products formed via the CSTR in Example 1. The tank contains a level controller at a set point on
the top of the tank. If this tank were to fill, materials would get clogged up in the reactor. Therefore, if the tank reaches 90% of its total
capacity, the level controller will send an electric signal, which opens an emergency drainage line located at the bottom of the tank. The
level controller will also activate an alarm alerting plant engineers that there is a problem with the storage tank. Finally, the level controller
will also close the inlet valve to the storage tank.
4.2.12: EXAMPLE 3
Below is a P&ID diagram of the transesterification process to produce biodiesel. Soybean oil, methanol, and the sodium methoxide catalyst
are pumped in to the reactor. The temperature of the reactor is regulated by the circulation water. The resulting biodiesel is then pumped out
of the reactor and goes on to other processes so that it can be sold. Below is a P&ID of the process that is missing the valves, pumps, and
sensors. Add the pumps, sensors, and valves that are needed to successfully control the process.
Solution:
4.2.13: EXAMPLE 4
Below is a example problem of a typical P&ID problem. A is a liquid at Tamp but boils at Trx. B and P are high boiling point liquids and C
is a solid. The reaction for the process is 2A+B+C-->P at Trx. Ais fed in excess.
Below is the solution to the problem above.
A Brief Guide To Piping and Instrumentation Diagram Standard Notation
video.google.com/googleplayer...40157432698049
4.2.15: ADDITIONAL READING

1. Wikipedia ChemE P&ID symbols entry
2. Wikipedia P&ID entry
3. Control Engineering Online Article: how to read P&IDs
Here are templates for creating P&IDs in Visio (PCs) and OmniGraffle (Mac).
4.2.16: REFERENCES
Mineral Engineering Technical Services; "Process Engineering.". Last Updated 05/09/2007.
Solar Rating and Certification Corporation; "Example System Design.". Last Updated 05/09/2007.
Karim, Nazmul M.; Riggs, James B. "Chemical and Bio-Process Control." 3rd Edition. Ferret Publishing.
Ogunnaike, Babatunde A.; Ray, W. Harmon. "Process Dynamics, Modeling, and Control." New York Oxford: Oxford University Press,
1994.
Beardmore, Roy; "Flow Sheet Symbols.". Last Updated 12/02/2006.
4.2: Piping and Instrumentation Diagram Standard Notation is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf
4.3: PIPING AND INSTRUMENTATION DIAGRAMS - LOCATION OF CONTROLS
AND STANDARD CONTROL STRUCTURES
4.3.1: INTRODUCTION
A Piping & Instrumentation Diagram (P&ID) is a schematic layout of a plant that displays the units to be used, the pipes connecting these
units, and the sensors and control valves. Standard structures located on a P&ID include storage tanks, surge tanks, pumps, heat exchangers,
reactors, and distillation columns. The latter three comprise most of the chemical process industry single-unit control problems.
P&IDs have a number of important uses in the design and successful operation of chemical process plants. Once a process flow diagram is
made, P&IDs help engineers develop control strategies that ensure production targets are met while meeting all safety and environmental
standards. These diagrams are also very important for locating valves and process components during maintenance and troubleshooting.
Each page of a P&ID should be easy to read and correspond to a specific action of the plant. Also, symbols used in the P&ID are uniform
throughout. Diagrams for heat exchangers, continuously stirred tank reactors (CSTRs), and distillation columns shown in subsequent
sections of this wiki represent a small sample of standard symbols for these structures. For a list and explanation of standard controls-related
symbols, please refer to PID Standard Notation.
4.3.2: GENERAL STRATEGIES FOR IMPLEMENTATION OF GOOD CONTROL SYSTEMS

There are several guidelines to follow when designing P&IDs for a plant. The first consideration is stable operation of the plant so that all
safety and environmental standards are met. Maintaining product quality should also be a primary design objective. Additionally, systems
should be designed to respond quickly to rapid changes in rate and product quality. Usually control systems should be run in an automatic
mode so the system will correct itself, as opposed to a manual mode, which requires operator supervision.
When setting up a control system, it is useful to first focus on mass balance control, which can be monitored by level and pressure control
loops that use gas or liquid flow rates. Additionally, a product control structure should be set up to ensure efficient process operation. To
prevent controls from conflicting with one another, a final examination of the controlled streams should be performed. Further
instrumentation can then be installed to prevent conflicts between controllers. Finally, appropriate tolerances should be established for
controllers that directly affect the action of other controllers. Determination of the optimal control placement is essential for successful plant
operation.
4.3.3: STANDARD STRUCTURES AND LOCATION OF CONTROL FEATURES

As mentioned above, control of heat exchangers, reactors, and distillation columns represent the majority of single-unit control problems.
The table below summarizes the control schemes derived in the following examples. The most common control schemes are listed in the
table below.
Table 1: Summary of common P&ID structures
For most equipment, a degrees of freedom analysis is first performed and then control scheme is designed based on the degrees of freedom.
4.3.3.1: HEAT EXCHANGER

To monitor the performance of a heat exchanger, the product stream is important. Usually, the product stream must be within some
temperature range before it continues to downstream process units. The outlet temperature of this stream can be used to calculate the heat
transfer. The steam is controlled in order to obtain the desired product stream temperature. One way to influence the product temperature is
by controlling the flow of the heated steam. This flow-based control may take some time to implement and therefore cause fluctuations in
the process. Depending on the process, these fluctuations may or may not be acceptable. Instead of monitoring the flow-rate, the steam
pressure may be monitored, achieving tighter control of temperature. A change in pressure is much easier to monitor and correlates directly
with a change in temperature of the steam. This offers an effective way to control process temperatures. Pressure control also enables the
physical condition of the piping to be monitored, since pressure changes occur as fouling progresses. The pressure, combined with the flow-
rate and temperature of the heated steam, can be used to calculate this fouling occurring inside a piece of equipment. It is important to
remember that the steam should always be controlled on the inlet side of the heat exchanger for better pressure control and safety reasons, as
shown in figure 1 below.
Figure 1. Heat Exchanger with pressure control on steam inlet and temperature control on the product stream.
4.3.3.2: DISTILLATION COLUMN

Because the economic viability of an overall process is based significantly on product purity, it is important that distillation columns
maintain stable operation. Changes in composition and flow-rate of the feed stream are common disturbances in distillation column
operation. Improper functioning of controllers can undermine the effectiveness of the product composition. A degrees of freedom analysis
can help place sensors and actuators in appropriate places, while not including too many sensors and actuators, in order to obtain an efficient
control system.
Degrees of Freedom Analysis
A simple degrees of freedom analysis can make design of a control scheme easier as well as improve the control scheme overall. A process
control approach to degrees of freedom, adapted from ECOSSE Module 3.1, requires tabulation of streams and extra phases.
DOF = Streams − ExtraPhases + 1
A typical distillation column contains the following streams; feed (1), bottoms (2), distillate (3), reflux (4), product (5), vapor-liquid mix
(6), cooling water (7), and steam (8). There are also three locations where there are two phases, vapor and liquid, present in equilibrium.
These are denoted as one "extra phase", since temperature and pressure are not independent in a two-phase system.
DOF = 8 − 3 + 1 = 6
Figure 2. Degrees of freedom analysis on a typical distillation column (adapted from ECOSSE).
In typical practice, the condenser and reboiler pressure are specified as atmospheric pressure. Once the pressure is specified in the condenser
and reboiler (both two-phase systems), the temperature is specified and therefore the degrees of freedom are reduced by two. The total
degrees of freedom, in practice, is six.
DOF = 8 − 5 + 1 = 4
A simple degrees of freedom analysis, in any system, can help define the variables and reveal where the critical sensors for process control
should be located. This type of analysis also minimizes the design of too many sensors, actuators, and valves, which may in theory control
and regulate the same variable.
Sensor and Actuator Options
Product streams, reboiler steam, and the reflux stream should have flow sensor/transmitters so that each flow can be adjusted to meet the
column’s control objectives. A flow sensor should also be placed on the feed stream because disturbance to the feed flow rate are common.
Each flow sensor should be connected to a flow valve to control the corresponding stream.
Differential pressure level sensor/transmitters should also be used for the accumulator (the vessel that collects condensed distillate) and
bottom of the column because maintaining these levels is essential for reliable operation of the column. If flooding is an issue, a pressure
differential across the column should indicate the onset of flooding. A pressure indicator should be installed at the top of the accumulator to
monitor column pressure. Temperature in a distillation column is typically controlled by manipulating steam flow to the reboiler.
Control Schemes
There are a number of common control schemes for distillation columns. Optimally, a distillation column should be run with dual
composition control because it saves energy. In dual compostion control, the temperature of both chemicals in a binary distillation is
controlled. The system is more complex to setup and measurements required for control may be difficult. As a result, many distillation
columns use single composition control instead. Common control configurations for distillation columns include reflux-boilup and
distillate-boilup.
In the reflux-boilup configuration, the distillate composition and bottoms composition are the control variables. The reflux flow and the heat
input control (vapor boilup) are the manipulated variables, which allow control of the liquid and vapor flow-rates in the column. With this
control system, a quick response to changes in the feed composition is possible. In the distillate-boilup control configuration, the distillate
flow and the vapor boilup are used to control composition. This configuration is a better choice for columns where the reflux ratio is high.
Figure 3. Distillation column with reflux-boilup control scheme (adapted from Luyben).
4.3.3.3: REACTOR (EXOTHERMIC REACTION IN CSTR)

In implementing controls in reactive systems, temperature is a good indicator for unit performance. Temperature is often related to reaction
rate and varies with time in most reactors. However, a reactor exit stream with a constant temperature is often desired downstream in the
process.
In an exothermic reaction in a CSTR, lower initial temperatures result in lower reaction rates and low heat generation. As the reaction
progresses, heat generation increases rapidly due to higher reaction rates and high concentrations of reactants. As reactant concentration
decreases, the heat generation once again becomes low. Exothermic reactor temperatures must be controlled to assure stable reactor
operation.
To remove heat from an exothermic reaction, basic heat transfer principles are employed. A coolant is pumped through a shell outside of the
reactor. Since the heat removal is linear, the temperature of the coolant should be controlled. In doing this, it is possible to increase the
driving force for heat transfer to slow a reaction, or conversely, to allow the reaction to further progress by decreasing the driving force for
heat transfer.
4.3.3.4: DEGREES OF FREEDOM ANALYSIS

In order to determine where to place controls, sensors, and valves in a exothermic reaction operation, a degrees of freedom analysis, similar
to that carried out in the distillation section 3.2, can be helpful.
A typical CSTR contains the following streams; reactant A (1), reactant B (2), product (3), and coolant to the jacket (4). There are no
locations where there are two phases, vapor and liquid, present in equilibrium - assuming the reaction is liquid phase with no simultaneous
evaporation or sublimation.
DOF = Streams − ExtraPhases + 1
DOF = 4 − 0 + 1 = 5
Figure 4. Degrees of freedom analysis on a typical exothermic reaction in a CSTR.

Since in most practical applications the reactants (1 and 2) as well as the product (3) flow rates are defined by demand for the product, there
are only 2 degrees of freedom which allow for placement of a 2 valves or controllers on the coolant stream (4) as well as one of the reactant
streams (1 or 2). Only one of the reactant streams need to be specified as the other can be determined by a ratio controller using
stoichiometric coefficients.
4.3.3.5: CONTROL SCHEMES

As discussed previously, a valve can be placed on the coolant inlet stream to ensure proper temperature control of the reactor. By controlling
the coolant stream based on inlet conditions of the reactant streams, the control can respond quickly. But, if the coolant is controlled based
on outlet conditions of the product stream, there is a lag in response, but it is easy to monitor heat transfer performance.
Figure 5. Jacketed CSTR with coolant control based on the outlet temperature conditions.
To translate this example to an endothermic reaction, the coolant flow would simply be translated to steam flow in order to provide heat to
the reaction through the CSTR jacket.
4.3.3.6: OTHER COMMON PROCESS EQUIPMENT

Beyond the heat exchangers, reactors, and distillation columns, many other pieces of process equipment, including furnaces, compressors,
decanters, refrigerators, liquid-liquid extractors (LLEs), and evaporators, are subject to disturbances and require careful control.
4.3.3.7: FURNACES
For example, furnaces may be subject to frequent load changes as a process or customer requires more energy. To cope with these demands,
the temperature of the outlet stream must be monitored and manipulated. Information from a temperature controller at the outlet stream can
be used to effect changes in valves that control the flow-rate of air and fuel to the furnace. At this point, the best setup of the control system
must be considered in light of factors such as safety, lag time for changes to occur, and equipment wear. For a furnace, the controls should
generally be set up so that an excess of air is more likely than an excess of fuel. This reduces the risk of explosion due to the furnace filling
with uncombusted fuel.
4.3.3.8: LIQUID-LIQUID EXTRACTORS

In liquid-liquid extractors, the interface level and pressure are the controlled variables. Disturbances in flow rate of the entering stream can
affect interface level and prevent complete separation of the heavy and light components. From this, it is obvious that there should be valve
controls on both exit streams. The best control scheme depends on the operation of the process. When the heavy phase is continuous (light
phase flows upward through heavy phase), changes in interface level should be controlled by adjusting the flow-rate of the light product,
while the pressure is controlled by adjusting the flow-rate of the heavy product out of the column. Figure A is a representation of what
occurs in a single stage extractor. Generally, single stage extractors are using in chemical labs, where as multistage extractors are used in
industry. A multistage extractor uses the immiscible liquid stream from the previous stage as the feed in the following stage. Figure 6
depicts the control scheme previously described.
Figure 6. Liquid-liquid extractor control scheme with interface level and pressure control.
When the light phase is continuous, the control system must be set up in the opposite manner. Figure C is again a depiction of what occurs
in a single stage extractor. Figure D is a representation of a control scheme that could be implemented. This is the reverse of the control
scheme in Figure 6.
Figure 7. Liquid-liquid extractor control scheme with light-phase control.

These representations are only two possibilities for basic control schemes that can be implemented in a process as there are several
controllers and aspects of the specific processes that may need to be controlled or monitored.
4.3.3.9: COMPRESSORS
Compressors are another valuable component in process design. They allow for the reduction of the volume of an incoming stream by
increasing the pressure at which the stream is maintained. They can also be used to pump liquids through the process, as liquids are highly
incompressible compressors cannot be used for volume reduction. For this, there must be a specific control system as to prevent adverse
effects due to extremely high or low pressures. There are several types of compressors, including: dynamic, axial, and rotary [1] to name a
few. Because the increase in pressure is governed by the ideal gas law, there is most often and increase in temperature as well. This can be
left as is, or sent to a heat exchanger for temperature reduction. Heat exchangers were discussed above.
One such example is the use of a centrifugal compressor to reduce the volume of a fuel stream for storage. Using a compressor will allow
for volume reduction as gasses are easily compressed, this is also economically friendly as it reduces the size of tank necessary to store the
fuel stream. The tank should also be equipped with a pressure reducing valve, to bring the stream back to a desired pressure, depending on
the process. A diagram of this scheme is as follows:
Figure 8. Compressor control scheme with pressure control.
The pressure controller on the compressor controls the valve on the incoming fuel stream. This ensures that if there is a build up in pressure,
the flow into the system will be stopped in time. Also, a pressure controller should be placed on the storage tank. This is controlled by the
pressure reducing valve mentioned earlier.
4.3.3.10: DECANTERS
Decanters, much like Liquid-liquid extractors, use solubility as their principle of separation. Unlike Liquid-liquid extraction, these require
some time for the separation to occur. Generally, the separation is a liquid-liquid or liquid-solid separation. Decanters are widely used in the
wine industry to separate out sediment from the wine. Utilizing the wine separation example, a possible control scheme is as follows. Here,
there is only a level sensor on the decanter as it is a liquid-solid separation. Note that there is a release stream, used to remove collected
sediment. An analytical or pH sensor could also be used here to maintain the correct chemistry or pH, as wine is the final product. It is also
important to note that the exact placement of the level sensor would vary depending on the desired level in the tank. Here, the level sensor is
not shown in its exact placement.
Figure 9. Decanter control scheme with level control and pH control.

Another vessel that is very similar to decanters are knockout drums. These vessels are generally located after heat exchangers or other
pieces of equipment that result in a multiphase system. These vessels are used to separate the two phases, generally gas-and liquid
separation. A possible control scheme is depicted below. The incoming stream is a liquid-gas mixture coming from a heat exchanger. Thus,
there is a pressure sensor on the knockout drum. A level controller could be used, but this is effectively measuring the same thing, so it has
been omitted. Also, because it is coming from a heat exchanger, a temperature controller has been included to control the amount of cooling
taking place in the heat exchanger.
The pressure controller (PC) controls V2, the stream into the tank. The temperature controller (TC) controls V1, the valve on the coolant
stream to the heat exchanger.
Figure 10. Knock-out drum control scheme with pressure and temperature control on different streams.
These examples illustrate the typical method of locating control systems for process equipment. After choosing the location for valves based
on process constraints, there still remain a number of possibilities for the actual manner in which to control the valves. The lag-time for
changes may be longer for certain configurations, while the danger of damaging equipment is smaller. The controls configuration will
depend strongly on the safety concerns associated with a specific process.
4.3.4: SELECTING CONTROLS AND THEIR LOCATIONS FOR A MULTI-UNIT PROCESS
The following steps should be followed when setting up controls for multi-unit processes. For more in-depth descriptions, refer to the Step-
by-Step Method For Describing Controls and Their Purpose.
1. Determine process objectives, taking into consideration product specifications, economic constraints, environmental and safety
regulations, etc.
2. Identify boundaries for normal operation. These can be based on equipment limitations, safety concerns, environmental regulations,
and the economic objectives of the processes.
3. Identify units and streams in the process that are susceptible to significant disturbances. These disturbances commonly occur in
feed streams, product streams, and reactor vessels, but can be present anywhere that temperature or pressure or other variables are
changing.
4. Select the types and locations of sensors in order to properly measure and monitor critical process variables.
5. Determine the appropriate types and locations for control valves in order to appropriately adjust process variables so that they
remain within the normal operating boundaries. Controls should be set up to minimize response time between sensing a change and
taking corrective actions. The ideal location for any given control depends on the process unit or units that it affects.
6. Perform a degree of freedom analysis.
7. Energy Considerations. An energy balance should be performed for the process. This step involves transporting energy to and from
process units. This may include removing heat generated by a reactor and using it elsewhere in the process. Control valves will help
regulate the flow of such streams.
8. Control Process Production Rate and Other Operating Parameters. Adjusting process inputs, such as reactant feed rates, can alter
other variables in the process. Process controls must be able to respond to these adjustments to keep the system within operating
boundaries.
9. Set up control system to handle disturbances and minimize their effects. (See Chemical Process Controls > PID Control)
10. Monitor Component Balances. Accumulation of materials within a system is not desirable and can lead to inefficiency in the process
or catastrophic failure.
11. Control individual unit operations. Each unit of a multi-unit process needs to be individually controlled in order for control of the
entire system to be possible.
12. Optimize the process. If the system has degrees of freedom, process variables can be manipulated in order to more efficiently or
economically create product.
 EXAMPLE 4.3.1: COMPLETING A CONTROL SYSTEM
Consider the following reaction in a semi-batch reactor:
(CH CO) O + H O ⟶ 2 CH COOH

3 2 2 3
Figure E-1. Basic Flow Diagram of reaction of Acetic Anhydride with water to produce Acetic Acid.
The objective is to produce CH COOH at a given concentration, i.e. 15-20%. The tank should also be drained once the given
3
concentration is achieved and the reaction has gone to completion. The tank can only hold a specified volume and 80% is filled with
water. The reaction should be run in excess water. Once a given amount of acetic anhydride has entered the reactor, it should be left in
the tank until the reaction has gone to completion. The completion of the reaction can be determined by temperature, as the reaction is
exothermic.
Using the multi-unit controller design from above, determine the location of needed sensors and valves on the flow diagram.
Figure E-2. Location of sensors and valves.
Using Figure E-2 determine how to use the level and temperature sensors to limit the amount of acetic anhydride entering the reactor
and also how to drain the system once the reaction has gone to completion.
Figure E-3. Control system for production of acetic acid.

Using the above system, when the right amount of acetic anhydride has entered the reactor the level sensor will stop the flow into the
reactor. Once the reaction has gone to completion and the temperature sensor has been triggered, due to a peak in temperatures from the
exothermic reaction, the tank will drain.
 EXAMPLE 4.3.2: CONTROL OF A HEAT EXCHANGER
Examine the figure below, which shows two different control schemes for a heat exchanger. The outlet stream temperature must remain
constant. How is this goal accomplished using the control systems shown below? Which stream is the manipulated stream? What are
the advantages and disadvantages of each setup?
Figure E-4. Two different control strategies for a heat exchanger (adapted from Turton)
Solution
In both setups, the cooling water flow is manipulated to control the temperature of the outlet stream. The control strategy for each heat
exchanger is different.
Figure E-4 (a) shows a heat exchanger where the flow of cooling water is adjusted based on the temperature of the outlet stream. The
advantage to this setup is that it is quite straightforward and the cause of the change in the temperature of the outlet stream does not
have to be identified. The disadvantage is that a temperature change has already occurred before action is taken, and if the lag-time for
the additional cooling water to cool the process stream is long, the outlet stream temperature may be inconsistent.
Figure E-4 (b) shows a heat exchanger were the flow of cooling water is adjusted based on the flow-rate and temperature of the inlet
stream and the temperature of the coolant stream. Based on a set of calculations using the heat transfer coefficient of the heat
exchanger, the outlet stream temperature can be predicted. The advantage of this system is that action is taken before changes occur in
the outlet stream temperature. The disadvantage is that if the actual heat exchange capacity differs from the expected value due to
fouling or mechanical problems, the outlet temperature will not be correct.
 EXAMPLE 4.3.2B: AN ALTERNATE NOTATION, WITH EXPLANATION OF CONTROLLER LOCATION

AND FUNCTION
This presents an alternate notation of controls. The arrows above the controller indicate what parts of the process that the controller
controls.
A + B ⟶ AB
1. B is added slowly into a full charge of A

2. The temperature is maintained at TR (RXN temperature)
3. Upon completion of reaction, AB is cooled to TP
4. C is added to AB to make final product, which leaves the tank
Solution
LC1: This controls M1 and V1.
This controller is used to keep the level of A in the A tank above a preset minimum.
This controller is used to keep the level of B in the B tank above a preset minimum.
This controller is used to keep the level of C in the C tank above a preset minimum.
FC1: This also controls M1 and V1.
This controller totalizes the amount of A that has flowed out of the A tank, and turns off flow when the flow amount has reached a
preset value.
This controller totalizes the amount of B that has flowed out of the B tank, and turns off flow when the flow amount has reached a
preset value.
This controller totalizes the amount of C that has flowed out of the C tank, and turns off flow when the flow amount has reached a
preset value.
LC4: This controls V1, V4, M4, and M5.
V1 - If level of liquid in the reaction tank goes over a preset maximum, V1 is closed.
V4 - If level of liquid in the reaction tank goes under a preset minimum, V4 is closed.
M4 - If level of liquid in the reaction tank goes under a preset minimum, M4 is turned off.
M5 - If level level of liquid in the reaction tank goes under the level of the agitator, M5 is turned off.
TC1: This controls V3, V5, and M3.
M3, V3 - If the temperature in the reaction tank is at TP, M3 is turned on and V3 is opened.
V5 - To maintain the temperature in the reaction tank at TP, V5 is opened and closed accordingly.
4.3.5: REFERENCES
Bequette, W.B. Process Control Modeling Design and Simultion., New Jersey: Prentice Hall, pp 466-477
Luyben, W.L. Plantwide Process Control., New York: McGraw-Hill, pp 53-69, pp 201-203, pp 235-247
Module 3.1: Control of Distillation Columns. The ECOSSE IGDS Controls HyperCourse. University of Edinburgh, Scotland.
Ogunnaike, Babatunde A. Process dynamics, modeling, and control, New York: Oxford University Press 1994, pp 1149-1153
Riggs, James B. Chemical and Bio Process Control., Ferret Publishing, pp 491-504
Turton, Richard. Analysis, Synthesis, and Design of a Chemical Process, New Jersey: Prentice Hall, pp 408-409
U.S. Chemical Safety and Hazard Investigation Board; "Morton International Explosion."
www.csb.gov/completed_investigations/docs/mortondigest.pdf. Last Updated 07/17/2003.
Tompkins, Bill. "Reaction and action." www.ehponline.org/members/2004/112-6/focus.html. Last Updated 09/09/2007
Wikipedia. "Gas Compressors" http://en.Wikipedia.org/wiki/Gas_compressor. Last Updated 11/27/2008
Wikipedia. "Liquid-Liquid Extraction" en.Wikipedia.org/wiki/Liquid-liquid_extraction. Last Updated 11/20/2008

Authors: Christopher Bauman, Jennifer DeHeck, Merrick Miranda, Evan Leonard
Stewards: Robert Chockley, Chloe Sweet, Irene Brockman, Scott Dombrowski
Anchors: Megan Boekeloo, Stacy Young
4.3: Piping and Instrumentation Diagrams - Location of Controls and Standard Control Structures is shared under a CC BY 3.0 license and was authored,
4.4: PIPING AND INSTRUMENTATION DIAGRAMS - STANDARD PITFALLS
4.4.1: INTRODUCTION
Piping and Instrumentation Diagrams (P&ID) are standardized in many ways, and there are some fundamental safety features that are
absolute requirements for all P&IDs. Unfortunately, many people forget these features in their designs unintentionally. Lacking these safety
features could lead to serious engineering problems. It is important to eliminate these pitfalls when designing a P&ID.
In the following sections, different pitfalls of P&IDs will be discussed. The equipment design section will concentrate on how equipment
might be accidentally neglected and misplaced while designing a safe and functional process. The design process section will describe how
a lack of consideration of process conditions would lead to serious design pitfalls. Overspecification and underspecification of equipment
and design will be discussed. There are also some miscellanous pitfalls associated with interpretation and inherent problems of the P&ID
itself. Examples are introduced at the end to illustrate common errors and pitfalls of P&IDs.
4.4.2: P&ID PRODUCTION PITFALLS

Most of the common pitfalls of P&ID production result from the engineer forgetting that the design on paper represents a real physical
process and that there are practical and physical limitations that need to be considered in the process design. The following are some errors
to be particularly careful of when drawing up a P&ID:
4.4.2.1: EQUIPMENT DESIGN

Safety valves
Safety valves are part of the essential valves system for P&IDs. Together with isolation valves, they are an absolute requirement for
instrument design. Safety valves are required to install for all gas, steam, air and liquid tanks regardless of the tank's function for pressure
relief purposes. Engineers should be aware of their system's set pressure, relief pressure, percent overpressure, maximum allowable
pressures etc. when selecting a safety valve. The US law requires all tanks of pressure greater than 3 psig to have safety valves installed.
Different pressure tanks require different safety valves to best fit their safety design. Therefore, engineers must be very careful in selecting
the right safety valves for their systems. However, when constructing a P&ID, engineers sometimes forget adding safety valves to their
design, and this could cause serious problems. For example, if one forgets to add a pressure relief valve or safety valve on a reaction tank of
gas and liquid, the extra pressure accumulating would exceed the preset pressure limits for safety design. This could lead to a serious
explosion!
A pressure relief valve is symbolized by two triangles orientated at 90 degrees to each other, as shown below.
This eliminated excess pressure that might build up inside a reactor.

Isolation valves
The isolation valve is used to isolate a portion from the system when inspection, repair or maintenance is required. Isolation valves are
placed around the junctions in the distribution system. They are also part of the absolute requirement for P&ID construction. Engineers
should be aware of their system's pressure, voltage, process medium, pipe sizing and flow rates when selecting the right isolation valves that
work under the conditions for their system.
If engineers forget to add isolation valves on their P&ID design, serious problems would occur when the system needs to be partially shut
down for maintance or other reasons. There would be no way to control the unit's operation other than existing flow valves. Some upstream
production problems could affect downstream production since appropriate isolation valves are not installed for safety and production
purposes.
Notice that the automatic valve after the pump is a flow valve, while the manual valve before the pump is an isolation valve. One example
of this use is for cleaning. If some parts of the pump cannot be cleaned in place, the pump will need to be taken apart. The isolation valve
can cut off the flow to the pump, allowing it to be safely taken apart.
Valves and pumps
Some of the most commonly used pieces of process equipment that show up on P&IDs are valves. Valves control the flow of fluid through
pipes by opening to allow flow or closing to stop flow. One of the problems associated with valves on P&IDs is the sheer number of them
and deciding where to place them in the process.
A common mistake with valve placement has to do with the interaction between valves and pumps. It is important that valves be placed
after pumps in a given pipe. When a valve is placed before a pump, should the valve close and the pump has not been shut off, there will
not be a constant supply of fluid to pump. This is known as starving the pump. Starving the pump can create a disturbance known as
cavitation, and it is very hard on the equipment and can cause the pump to break. Placing valves after the pump ensure that even if the pump
does not shut off, it is still filled with the proper fluid.
These same principles apply to valve placement with respect to compressors. Placing a control valve upstream of a compressor eliminates
your ability to control pressure in the pipeline downstream of the compressor and introduces a risk of starving the compressor.
Agitators
A point that is very easy to miss, and very expensive if it is missed, is that if a vessel is equipped with an agitator, the vessel must be filled
enough to submerge the agitator before the motor is turned on. Agitators are designed to operate against the resistance of fluid. Without this
resistance, the agitator will rotate much faster than the equipment is designed for. This rotation may result in hitting the harmonic frequency
of the structure, compromising its integrity and causing the agitator to rip from its foundation. This creates not only a fiscal predicament
(not only ruining the agitator but also the vessel), but a safety nightmare. When designing a process, one must make sure he or she knows
and accounts for how much fluid must be in an agitated vessel to avoid this situation. This can easily be solved by adding a level sensor to
the tank that the agitator is being used in.
When placing the level sensor on the tank, make sure to place the sensor above the level of the agitator. This will ensure that the agitator is
submerged in the fluid. It would be incorrect to place the level sensor below the agitator.
Instrument selection and placement
Instruments are designed to operate properly under specific conditions. Every instrument has a range over which it functions properly, and
instruments must be selected that are appropriate for their applications. For example, a pressure gauge might have a working range of 5 psig
- 50 psig. You would not want to use this gauge for sensitive measurements in the range 3 - 6 psig. Instrument material must also be
considered during the selection process. If the substance being monitored is corrosive, for example, the instrument must be made of a
corrosion-resistant material.
Once an appropriate instrument has been selected, it must be appropriately placed. For example, a level control is not useful in a pipe
because there is no need to measure any water level inside of a pipe, much like a flow controller is not useful in a storage tank because there
is no flow. Similarly, a flow controller should not be placed on a valve, but instead downsteam from the valve. However, level controls are
useful in storage tanks and reactors, while flow controllers are useful in pipelines. Instruments must be selected and placed to reliably
provide useful information and to accurately control the process.
Equipment Selection
When creating a P&ID, the equipment that is selected to be used is very important, not only to maintain a smooth process but also for safety
purposes. Each and every piece of equipment from 100,000 liter storage tanks to temperature sensors has Operational Limitations. These
are the conditions under which a given piece of equipment operates as expected, with safe, consistent, and reproducible results. For
example, when storing a highly pressurized gas at 2,000 psig, one wouldn't want to use a storage tank that has been pressure tested up to 3
psig. The process conditions are way outside the operational limitations and would pose a serious safety hazard. The same goes for sensors
and gauges of all types. On a pressurized vessel at 2,000 psig, it would be no good to use a pressure control system that has a sensor that is
meant to measure up to 100 psig, the results would not be accurate.
Operational limitations can usually be found in the equipment manual sent by the manufacturer or possibly on the manufacturer's website. If
it is not found in either place, an engineer is obligated to contact the manufacturer and find the operational limitations before using a piece
of equipment in a process. More can be found on equipment selection for some (but by no means all) equipment below:
Temperature Sensors
Pressure Sensors
Level Sensors
Flow Sensors
Composition Sensors
pH and Viscosity Sensors
Biosensors
Valves
4.4.2.2: PROCESS DESIGN

Unit operation input/output
Providing clearly specified inputs and outputs to and from process units is vital. For the safety of the system, proper control and the
prevention of disaster, it is important to show where each substance came from and where it is going. The P&ID must show all material
streams to and from separation units, heat exchangers, and reactors.
For example, if the P&ID is not clearly denoted in a reaction to create construction explosive materials, then a large and perhaps fatal
calamity can occur at the chemical plant. The exiting streams of a plant may pour into a river which may violate environmental regulations.
Therefore, specifying inputs and outputs is imperative in a P&ID.
Pressure and flow
The movement of fluid is essential in many production lines. Transporting material from a tank to a reactor, a reactor to a distillation column
or from a column into a tanker truck all involve the movement of fluids. When designing P&IDs, a process engineer must decide how they
are going to attack this problem, namely whether or not a pump is needed to move the fluid.
The main issue here is pressure. Fluid, both gaseous and liquid, moves down the pressure gradient from high pressure to low pressure. The
rule of thumb is that if the source of the fluid has a much higher pressure than the destination of the fluid, a pump is not needed. Even if the
source has a pressure only a fraction higher than that of the destination, a pump may yet be needed for the fluid to flow through the pipes,
from the source to the destination. However, the flow and pressure from the source must also be compared to the needs of the process. For
example, if a inlet flow has highly fluctuating pressure, a pump or valve should be used to regulate the pressure for the process. Also, if the
diameter of the pipe does not remain constant and is fluctuating as well, perhaps due to outside pressure or force, a pump should be used to
control the pressure inside the system.
A common mistake young process engineers make is when charging a pressurized vessel, they do not use a pump powerful enough to
overcome that pressure. This causes backflow and can ruin process equipment. This problem is especially prevalent in recycle streams. In
some situations it may be appropriate to use a check valve (a valve allowing fluid to flow through into it from only one direction), to add an
additional barrier against backflow. An example of such a situation would be when a fuel is to be mixed with air for combustion: if any air
were to backflow to the fuel source, a dangerous, explosive situation would arise. In this case the use of a check valve would be appropriate.
An example of when a pump is never needed is when liquid from one vessel at ambient pressure is being transported to another vessel at
ambient pressure that is at a lower elevation. This is known as a gravity feed and utilizing gravity feeds where possible can significantly
decrease the cost of a process. As long as the pressure at the exit of the pipe is lower than the pressure created by gravity, a gravity feed can
be used. An example of such a gravity feed is the flushing of a toilet, by which the water from the cistern at a higher elevation falls to the
water closet at the lower elevation.
4.4.2.3: UNDERSPECIFICATION
For safety and control purposes, redundancy is desirable in control systems. A process lacking in suitable redundancy would be
underspecified. For example, a reactor containing an exothermic reaction would be underspecified if it only contained a temperature
control. The potential danger posed by an exothermic reaction warrents a high degree of safety: if one controller were to malfunction, a
backup must be in place. A pressure control should also be included, and perhaps a level control as well depending on the specific process.
4.4.2.4: OVERSPECIFICATION
On the flipside of underspecification is overspecification. Adding too many controllers and valves on vessels and lines is unnecessary and
costly. Since process control is highly dependent upon the situation, it is difficult to set specific limits on the necessary amount of
specification. However, a controller should not be used unless there is a specific need for it. Misused redundancy is a common example of
overspecification. It is unlikely that a water storage tank needs level, temperature, and pressure controllers. A degree of freedom analysis, as
shown in a distillation column example, can be helpful in determining the exact number of controllers, sensors, actuators, and valves that
will be necessary to adequately control a process.
4.4.3: PROBLEMS WHEN USING A P&ID

Other than equipment and production pitfalls, there are also some general P&ID interpration and inherent problems.
4.4.3.1: INTERPRETATION OF P&IDS
Although it is essential for P&IDs to represent the right instruments with the right references, many P&IDs do not support scaling or do not
require scaling as part of the system. P&IDs are drawn in a way that equipment and piping are displayed for ease of interpretation. They do
not show the physical placement and location of different systems or the actual sizes and length of equipment and pipelines.
Experience tells us that many engineers have overlooked certain pieces of equipment or over-estimated the size of equipment and piping,
leading to calculation and construction errors. For example, an engineer could be designing new equipment on the existing P&ID and not
realize that existing equipment is so densely populated that extra equipment will not fit. It is important to check the physical space of an
area before adding equipment to a process.
4.4.3.2: INHERENT PROBLEMS OF P&IDS

There are a few problems with P&IDs that are only solved by being conscious of them. One problem is that P&IDs are constantly being
updated, revised, changed, and added to. When reading a P&ID it is always important to check the date it was last revised and if there is a
later revision available. Making adjustments to equipment that isn't there anymore not only causes confusion and frustration, but is a waste
of everyone's time. Moreover, making sure you are using the most recent edition to the P&ID will ensure that you do not purchase
equipment to install that has already been installed.
Usually, companies will have a computer database with the most current P&IDs. Before modifying or working with an old paper copy of a
P&ID, check to make sure it is up to date with the most current revisions.
Another problem with P&IDs is that even if they are well made and technically perfect, complex processes often appear cluttered. There is
an inordinate amount of information contained in a P&ID, and as such P&IDs are next to impossible to take in at a glance. It is important to
carefully study each document to fully understand the process. The mistake of assuming you gained all the information off of a P&ID from
simply scanning it can leave you with mental holes that you don't even realize you have. This can be avoided by reading the P&ID carefully
on the first pass and constantly referring back to it when questions arise.
 EXAMPLE 4.4.1
A pipe connects a water tank (open to the atmosphere) and a vessel pressurized to 28psi. The open tank has 10 feet of water in it. A
level control system ensures that the 10 foot level is maintained in the open tank. The bottom of the pressurized vessel is 20 feet below
the bottom of the open tank and starts with no water in it. The goal is to fill the pressurized vessel up to 5 feet. Due to a poor design by
the project engineer, the water is fed into the bottom of the pressurized vessel. Given that the density of water is 62.4 lbm/ft^3 and the
gravitational constant is 32.2 ft/s^2, is a pump needed? If so, where should it be placed? Assume that there is no pressure drop due to
friction against the pipe and that the air pressure of the pressurized tank remains at a constant 15psi. The figure below may be helpful in
visualizing the process. (Hint: Remember in English units, a “fudge factor” is needed when using P = ρ*g*h)
Solution
First, we must determine the pressure at the stream exit (point B). When the vessel is full, we know that the pressure created by the
water is:
ρgh
P =
32.2
Plugging in 62.4 for the density, 32.2 for g and 5 for h, we find that the pressure caused by the water is 312 lbf/ft^2. Dividing by 144 to
convert to psi, we see this is equivalent to 2.2 psi. When we add the 28psi from the air pressure in the vessel, we get a grand total of
30.2 psi at the pipe exit.
The pressure in the feed pipe at the height of the exit to the pressurized vessel (point A) is again given by:
ρgh
P =
32.2
Only this time, the top of the water is 30 feet above the pipe exit instead of 5 feet. Plugging in this value, we find that the pressure is
1872 lbf/ft2. After converting, we find this is equivalent to 13psi. Since the water tank is open, we must also add 14.7psi for
atmospheric pressure. This gives us a total pressure of 27.7psi.
The pressure of the feed at point A in the diagram is LESS than the pressure at point B, so a pump is needed. The pump should be
placed BEFORE the valve labelled V-1 on the pipe as shown below. This is to prevent starving the pump when the valve closes.
 EXAMPLE 4.4.2
There are several errors in the following P&ID, which shows a vessel where a liquid is vaporized using heat transfered from steam
passing through a surrounding jacket. Identify the errors and explain why they are mistakes.
Solution
1. Control valve V-1 should be after the pump, not before. Placing a pump after a control valve takes away your control of the flow
rate and can also damage the pump if the valve is shut off while the pump is running.
2. Vessel E-1 is only being monitored with a pressure control. This is not adequate redundancy for process control. There should also
be a temperature control (to make sure the vessel is at an appropriate temperature for vaporization) and a level control (to make sure
the agitator is always submerged in liquid).
3. Although the vaporization takes place at ambient pressure, there should be a safety valve on the vessel. What if there is a
malfunction and too much vapor is produced?
4. The input and output streams for vessel E-1 are not sufficiently labeled. The source of input streams and destination of output
streams should be specified to ensure adequate control of a process.
5. There should be a flow meter and control valve on the vapor stream exiting the vaporization vessel so that you can monitor and
control the amount of vapor exiting.
4.4.4: REFERENCES
Center for Chemical Process Safety. (March 2006). Process Safety Beacon: Pressure Relief Systems - Do You See Any Hazards? CEP
Melhem, G.A. and Howell, P. (September 2005). Designing Emergency Relief Systems for Runaway Reactions. CEP
Ulrich, G.D. and Vasudevan, P.T. (July 2006). Predesign With Safety in Mind. CEP
Williams, J.P. (September 2004). Reliability for Safety Instrumented Systems. CEP

Authors: Angela Yeung, Curt Longcore, Ben Van Kuiken, Jeffrey Carey
Stewards: Eric Black, Megan Boekeloo, Daniel Carter, Stacy Young
4.4: Piping and Instrumentation Diagrams - Standard Pitfalls is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter
Woolf et al. via source content that was edited to conform to the style and standards of the LibreTexts platform; a detailed edit history is available upon
request.
4.5: SAFETY FEATURES IN PIPING AND INSTRUMENTATION DIAGRAMS
4.5.1: INTRODUCTION
Safety has become integral to the manufacturing world. The implementation of proper safety techniques and accident prevention can not
only save time and money, but prevent personal injury as well. P&IDs, when properly utilized, are a powerful resource to identify safety
hazards within the plant operations. The following sections provide an overview for the safety hazards that exist within a process, and
illustrate the importance of P&IDs in a chemical plant.
4.5.2: SAFETY IN DESIGN

During the early stages of plant design it is critical to determine important safety features that remove potential hazards from effecting the
facility environment. Regulations require that plant designers play a major role in minimizing the risks associated with these hazards.
However, in order to do so, designers need to be aware of the hazards that exist during plant activity. The facility design team must develop
a detailed drawing (P&IDs) including specifications of the plant process and environment to ensure that every aspect with regards to safety
is covered.
4.5.2.1: HAZARD VS, RISK

When discussing safety, the terms hazard and risk are often used interchangeably. However, the difference in definition between the two
terms is critical in utilizing the information they provide in increasing the safety within a plant. Hazard is defined as a potential source of
danger and risk is defined as the level of threat associated with the hazard. A risk of a hazard occurring can be represented mathematically
by the following equation:
Risk = Frequency × Consequences
Frequency represents the probability that a hazard will occur, and consequence represents the impact of that hazard. Values for each
parameter of the equation above are assigned by using either experimental information or educated judgment based on engineering models
and experience. Each plant design process will have specific safety hazards and risks associated with it. Therefore, there is no
predetermined value that can be assigned to each variable. For example, in one situation, a water tower may be well supported inside the
plant facility, and in another situation a water tower of similar structure may be hoisted against a rusted frame work outside of the plant
facility. Both of these situations have different levels of risk associated with them. Needless to say, the risks associated with a plant setup
can be reduced substantially by minimizing the probability that a hazard will occur.
4.5.2.2: HAZARD LOCATIONS AND RISK HOTSPOTS

Common hazard locations exist in any place containing large amounts of energy. The degree of danger is proportional to the amount of
energy stored at that location. Risk can be directly linked to kinetic energy, potential energy, work, heat, enthalpy and internal energy
sources. Kinetic energy, otherwise known as energy in motion, is present in any moving component. The component may be vibrating,
rotating, or translating, and this motion causes the kinetic energy of that part to greatly increase. Within industry, personal injuries and
fatalities are common hazards associated with moving parts. Reactors and cooling towers placed in high locations contain potential energy.
If a structural failure were to occur within a plant, these structural units and their contents could fall from a large elevation onto another
processing system or a human being, releasing all the chemical contents stored inside. Similarly, the stored work in springs and other
devices can cause fatigue and wear on the mechanical system over time, and result in eventual machine failure. Heat released from a
reaction within a chemical reactor can be rapid and fatal if not accounted for. The buildup of heat can cause serious consequences with
runaway reactions and boiler explosions. The enthalpy and internal energy of a reaction typically are the cause of most runaway reactions
and destructive fires in a plant.
Risk Hotspots
Uncontrolled chemical hotspots created within a chemical plant are a common source of hazard, besides the hazard locations pertaining to
energy sources such as kinetic and potential energy. Risk hotspots mostly occurs in the piping system and associated valves of the system,
joints, traps, and various other piping elements. Possible malfunction of the system due to structural corrosion can be triggered by the failure
to maintain the piping systems efficiently and periodically. Even if one of the valves in the system has corroded and is unable to function
properly, the fluid flowing through the pipe might get trapped, and the resultant buildup of pressure in the pipe may cause major safety
hazards including fatal ones such as an explosion. If the system is not shut off before the pressure gets out of control, the pressure buildup in
the pipe will cause it to burst, releasing all of its internal contents to the surrounding environment.
Storage vessels are other pieces of equipment that must be cared for properly. Since the plant operators and engineers do not usually interact
with the storage vessels, as compared to other parts of the plant process such as the piping system, they are considered to be of secondary
importance and commonly overlooked. Storage vessels have much more content inside them than pipes, so a leak or a burst would take
longer to get back under control than pipes, which can by plugged more easily. Problems which may arise from storage vessels are not
associated with their design, but in fact from not thoroughly and periodically maintaining them. Possible complications that can arise from
neglecting the storage tanks are over-pressurization, overfilling, heating element malfunction, or simply equipment malfunction.
Chemical reactors are another common location where risk hotspots occur. The nature and design of commercial chemical reactors is to
handle a controlled explosion. However, if the control element is removed, disaster is bound to occur, just as in any other part of the
chemical plant. The most common type of hazard is a runaway reaction inside a batch reactor. When the plant facility looses electricity or
cooling water, a runaway reaction will strike inside the reactor. Once a runaway reaction has spun up inside the reactor, many other hazards
may follow, such as flow reversal in the pipes, incorrect reagent charging, heat exchanger failures, and external fires. Other hazards may
perhaps be even more serious, such as engineering errors that could potentially cause a runaway reaction to occur, including inappropriate
material selection, inadequate equipment inspection or failure to fully understand the chemistry or exothermic nature of a reaction. A
runaway reaction originating inside the chemical reactor can easily cause a chain reaction across the rest of the equipment at the facility, and
can result in the entire system malfunctioning.
Other process equipment that may be hazardous and where risk hotspots commonly arise, are vacuum operators, furnaces, pumps, gas
movers, compressors, and heat exchangers. The location and type of specific piping and unit operations are available on the process P&ID.
A responsible process engineer should use the P&ID to identify all risk hotspots, and act accordingly to monitor and maintain a safe
working environment. In addition, a standardized plan should be constructed so that in the event of a malfunction, the correct steps can be
taken to bring the faulty part back under control. Supplementary precautions should be taken to prevent a comparatively minor malfunction
in the system from becoming a disaster which may violate environmental regulations and even endanger human lives.
4.5.2.3: SAFE DESIGN PRINCIPLES

The ISD or Inherently Safer Design movement was a doctrine striving for safer chemical processing procedures. This movement was
pioneered by Trevor Kletz in 1976, and promotes the design of processes so safe, that no catastrophic failure can occur within the plant. The
following principles apply to initial process design:
1. Use the fewest number of hazardous substances in the smallest quantities and still maintaining plant productivity
2. When possible, substitute hazardous chemicals with chemicals that are less dangerous
3. Practice moderate operating conditions in the plant
4. Use the simplest plant design possible
5. Design equipment in the plant to minimize the effects of a hazardous incident
The specifications determined by the process designers are communicated through the P&ID.
4.5.2.4: HAZARDS IN CONSTRUCTION

In order to eliminate hazards, a operations personnel must be able to identify that a hazard exists. Hazards that may be encountered on plant
sites may be categorized into three main types:
1. Hazards harmful to health: When workers are exposed to or come in contact with asbestos, corrosives, irritants, toxins, or noxious gases
try to avoid by specifying the processes, which lead to this exposure
2. Hazards likely to cause personal injury: Hazard awareness is increased when people have to work in situations likely to expose them to
the risk of personal injury, including moving plant machinery or working in areas where objects are likely to fall. Situations where there
are live electrical circuits overhead, buried power lines, and confined working conditions are likely to cause personal injury.
3. Hazards likely to lead to a catastrophic event: These hazards have consequences beyond the site boundary. They include fire outbreaks,
explosions, flooding, or premature collapse of structures, cranes, tunnels and excavations.
4.5.2.5: FAIL-SAFE DESIGN

The fail-safe design of a unit operation (such as a reaction vessel) requires a complete understanding of the operation at hand, and the
knowledge of all the worst-case conditions. A fail-safe system is a unit operation such that, if any or all of the worst-case conditions were to
occur, the operation would shut itself down automatically and in a safe fashion. In the case of a run-away reaction, if reagent feed limits,
interlocking controls, and integrated heat balances are all properly maintained, the reaction cannot "run away." Other precautions such as
purges, vents, dump tanks and quenches are available for reaction vessels, and should be visible on the P&ID.
4.5.2.6: INHERENT SAFER PREDESIGN

The table below provides a guideline for identifying and minimizing hazards partly based upon Kletz's rules for ISD in the Safe Design
Principles section above. The step/rule column describes the action taken. The tools column describes the mechanism by which the actions
are taken. The experimental and analytical resources column describes the knowledge by which the mechanisms are created. The Literature
References column describes where the knowledge can be found.
Step/Rule Tools Experimental and Analytical Resources Literature References
Fire triangle, Flammability-limit chart, Chemical reactivity Thermodynamic calculations, Reactor
Perry's Handbook (10) Table 26-
1. Identify chart, Safety compatibility chart, Safety stream chart, design equations, calorimetry,
10 and associated pages
Toxicity ratings flammability charts
2. Eradicate (ISD Rule 2) Inert-gas blanketing, Failsafe design see above see above
3. Minimize, Simplify, Merk Index, MSDSs, Chemical
Logical chemical engineering analytical
Moderate, Attenuate (ISD Moderate ignition sources, Keep reactive volumes small Engineering textbooks and
and computational skills
Rules 1, 3, 5) monographs
Separate hazardous operations - Surround in an impenetrable
4. Isolate (ISD Rule 4) Logical chemical engineering judgment
structure
4.5.3: RELIEF SYSTEMS

Organizations such as the American Society of Mechanical Engineers, American Petroleum Institute and National Fire Protection
Association layout recommendations and design standards so that most engineers with proper training can setup proper emergency relief
systems for single phase flow. Unfortunately, this is often not good enough for reactive systems. These systems are much more complex and
include multiple phase flow, runaway reaction potential and self reactant material. When designing an emergency relief system (ERS) it is
necessary to understand all aspects of the chemicals and processes that will be in play. This includes but is not limited to: kinetics of the
possible reactions, contamination, interactions with air, rust, piping, or water, phase changes and runaway reactions. The follwing topics
deal with the hazards that require designing relief systems and the prevention of runaway reactions.
4.5.3.1: PLANNING AND DESIGN

When designing an ERS, it is important to consider the worst-case scenario. This is based upon a thorough knowledge of the reactions,
materials, and environment of the process. Some of the most critical scenarios are over-pressurization in a reactor.
All disaster scenarios can be analyzed using a hazard and operability (HAZOP) study. This HAZOP study will analyze a process based upon
human, equipment, and environmental factors.
The HAZOP team must take the following steps to ensure that all potential scenarios are taken into account:
Review the potential hazards of all chemicals. This includes non-operational conditions and interactions with contaminants
Study the chemical process including all possible reactions, rearrangements, decompositions, etc.
Review the P&ID’s for the process
Study the specific reactor and storage vessels for material composition compatibility, size, surface area, instrument ranges, and set points
With this information, the HAZOP team can determine most of the potential disaster scenarios. Most likely the worst case scenario will
involve fire induced runaway reactions.
4.5.3.2: DESIGN STRATEGIES
There are various techniques often used to prevent failure from over-pressurization, fire, runaway reaction or other disaster scenarios. The
following are some design suggestions that will decrease this risk:
The use of insulation in case of fire. Unfortunately, insulation will also minimize heat loss from the reactor during a runaway reaction. It
is important to consider the ability of a vessel to drain when using insulation. If the contents can be drained and cooled before the
reaction starts to runaway, catastrophe can be prevented.
Design to avoid fire damage on sensitive equipment. The possibility of fire damage on electrical equipment or sensors will cause control
difficulties when trying to slow down a runaway reaction, or monitor another emergency situation.
Consider the structural integrity of the entire system due to fire damage.
Use multiple purge streams and valves to separate materials. Separating reaction components into a storage vessel with a large surface
area will allow for faster cooling.
Install two separate relief devices in case one has been compromised due to fouling or solid particle blockage.
Overall safety relief plans must be made for all possible scenarios.
4.5.3.3: REACTION KINETICS
Reaction kinetics usually determine the potential for safety disaster. For the average exothermic reaction, the reaction rate doubles with
every 10°C increase. This will lead to an exponential increase in energy which will force both the temperature and pressure in a system to
uncontrollable levels. Since reaction rates are sensitive not only to temperature, but also pressure, contaminants, concentrations and phases,
all possibilities must be adequately analyzed for plant operation to be deemed safe.
Relief devices should be designed to handle chemicals and/or mixtures in any phase. For example, if a reactor temperature increases
suddenly and a safety relief valve is activated the material must be able to flow through the valve quickly and safely. If the chemical has
changed phases, or has a higher pressure that the valve cannot accommodate, the chemical will not be able to escape and build up pressure
in a reactor. This would greatly increase the risk of a disaster. Because of the extent of possibilities when considering reaction rates and
kinetics, it is usually necessary to consult some sort of computer simulation or dynamic simulation tool to plan for every possible scenario.
4.5.3.4: RUNAWAY REACTIONS

Runaway reactions are caused when exothermic reactions are fed more energy due to malfunctioning cooling systems. This causes an
exponential increase in temperature, which in turn causes an increase in pressure, and finally damage to the reactor and/or plant. The
possibility of malfunctioning cooling systems must always be considered for flammable materials. If materials are being stored at
temperatures above their flash point, fire is always a possibility. Fire will cause a reaction to reach runaway conditions with very little
reactant. Relief systems for fire induced runaway reactions must be larger than conventional runaway reactions.
While relief systems are often designed for over-pressurization, this might not be enough when considering flammable materials. Constant
or prolonged exposure to flame will cause most normal reactant or storage vessels to fail causing chemical leaks or plant fires. Fire proof
insulation must be used together with normal pressure relief systems to prevent system failure. An important design variable when
considering fire induced runaway reactions is fire flux. This variable considers heat impact on a reactor due to fire. Formulas for calculating
fire flux can be obtained from the National Fire Protection Association. The elevation of flames must also be considered when designing
reactor vessels and safety relief systems. Pool fires can produce flames that are hundreds of feet high; using the P&ID will be important in
ensuring that process components sensitive to fire are sufficiently protected from something like pool fires.
4.5.3.5: TWO PHASE FLOW

For systems in which multiple phase flow is possible, all phases must be planned for. For vapor hybrid systems, all-vapor flow should be
considered. Likewise, when foam flow is possible, all-foam flow should be planned for. This is all part of the mindset of planning for the
worst case scenario. The most conservative design should always be used when faced with uncertainties of phase flow.
Often times, a runaway reaction will cause high-viscosity two phase flow. Relief valves and safety features must be ready for viscous flow.
Many resources suggest averaging the viscosities of the two phases to plan for two phase flow. It is important to keep in mind however, that
two phase flow discharge will separate in the discharge line. This will lead to higher pressure drop. Piping sizes are often underestimated
due to this unplanned scenario. Undersized piping will lead to valve rupture and back pressure and could cause venting disturbances during
the emergencies where venting is critical to the safe shutdown of a process.
4.5.3.6: REGULATIONS
There are often conflicts and contradictions in federal regulations and recommended practices for safety design guidelines. When faced with
such confusion the designer should at least design to the level of the Occupational Safety and Health Administration's (OSHA)
requirements, as well as meeting regulations set by the Environmental Protection Agency (EPA), the Process Safety Management(PSM),
and the Risk Management Program (RMP). While these guidelines and regulations produce a minimum standard to follow, when dealing
with reactive systems, this does not always meet process needs. For more details on the various regulatory agencies that govern chemical
processes, see Regulatory Agencies and Compliance.
4.5.4: ADDITIONAL SAFETY HAZARDS IN CHEMICAL PLANTS

This section outlines potential safety hazards commonly found in places where chemicals are stored or chemical processes are taking place.
While these hazards are rather easily prevented through attention to detail and general awareness, neglecting them can have catastrophic
consequences.
4.5.4.1: TIME SENSITIVE CHEMICALS

Some chemicals have a "shelf life," or an expiration date provided by the manufacturer. The chemical must be used by this date or properly
discarded. These are typically reactive chemicals, which can become unstable after a certain period of time, possibly rupturing the vessel in
which they are stored. An example of this is a monomer that begins to polymerize unless an inhibitor is present. This inhibitor is completely
consumed after a certain period of time, allowing polymerization to occur, and therefore must be used or discarded by that time. Another
example deals with the formation of peroxide, which can be a severe fire and explosion hazard. It may also be a health concern, causing
severe mucous membrane, respiratory tract, skin, and/or eye burns. Peroxide-forming materials should be stored carefully, labeled with the
date received and the date first opened on the container. Chemicals should be disposed of or checked for peroxide formation after six
months; do not open any container with solid formation around the lid. There are several ways to prevent this from happening. Material
Safety and Data Sheets (MSDS) should be available for any chemical the plant uses. These can inform you if a particular chemical becomes
unstable after a certain period of time. It is also important to make sure there are procedures in place for handling time-sensitive materials.
Lastly, investigating near-miss occurrences can help to ensure that future incidents do not occur.
4.5.4.2: PRESSURE RELIEF SYSTEMS
Any open pipe in a chemical plant is a potential discharge site. While operator convenience and maintenance remains a concern, safety takes
precedence when dealing with relief systems. It is essential to operator safety that relief valve discharge sites are located in areas that pose a
low risk of exposing personnel to chemical hazards and are directed away from all access platforms. It is common in industry to tie multiple
pressure relief devices to an emergency vent header that releases on the roof of the plant. Long stretches of unsupported pipes also pose a
potential threat. Force generated by material flow could bend or break the pipe impeding on plant operations, and more importantly, injuring
personnel.
4.5.4.3: DUST EXPLOSION HAZARDS

Most flammable solids can form an explosive dust cloud if the particles are small enough. Materials such as wood, grain, sugar, plastics, and
many metals can all form these explosive dust clouds. Dust explosions occur when a combustible material accumulates in a confined area
and is exposed to an ignition source. High risks areas are usually those that are neglected such as tops of vessels and tanks, on pipes, storage
bins, bucket elevators, and dust collectors.
To prevent dust explosions, it is essential to implement good housekeeping practices. Care must be taken so that the cleaning process, such
as sweeping, doesn't turn a dust layer into a more dangerous cloud. Operations and process engineers should be aware of all the mechanical
and safety control equipment associated with preventing fires. Equipment like electric vacuums must be appropriate for use in an area where
an explosive dust cloud could form. This entails no holes or cracks in the cord, sufficient grounding, and receiving site approval for use of
the equipment.
4.5.4.4: OVERFILLING TANKS
The overfilling of vessels has long been a leading cause of serious incidents in chemical and petroleum industries. When a level sensor or
high-level alarm fails, reactive material can spill over the tank and accumulate. If this material is exposed to an ignition source, there is the
potential for an explosion, leading to serious property damage, environmental issues, and injury to operations personnel. When filling or
draining a tank with material, operators should be aware of all relevant level, pressure and temperature controls in place, and watch for
abnormal trends. Also, all safety critical alarms surrounding a vessel should be tested regularly at frequencies recommended in plant
process-safety-management procedures. Conducting regular process maintenance on safety critical alarms have numerous benefits, which
include reducing the risk of operating a plant with faulty equipment, and increasing operator awareness with the location and function of
critical safety devices.
4.5.4.5: CONTAINING STORAGE TANK SPILLS

Engineering controls are implemented into plant designs to account for potential disasters, such as a spill, leak, or complete emptying of a
storage tank. Industrial-sized chemical plants store large amounts of raw materials, products, and byproducts on-site. The amount of each
varies, but is typically between three to thirty (or more) days of the required supply or amount produced. The stored chemicals can be
hazardous, flammable, explosive, and/or reactive with each other. In the event of a tank spilling, dikes are built around tanks to contain the
spill and protect the surrounding community from the spill. The regulations of the dikes include the following: dike volume must be 1.5X
the largest storage tank contained by the dike, reactive materials cannot be stored in the same dike, and scuba gear must be present on-site if
any dike is deeper than four feet. Since the dikes cannot contain any reactive materials, the implementation of dikes affects the overall plant
layout. Typically, dikes are designed to be like speed bumps and have a height less than one foot so fork-lifts and tankers can easily
maneuver through the plant. Dikes are an effective engineering control that greatly improve the safety of a chemical plant.
4.5.4.6: TEMPERATURE AND PRESSURE RATINGS

Before a vessel is put into plant operation, it is rated and stamped by the manufacturer with temperature and pressure limits. Problems arise
when personnel overlook the inverse relationship between temperature and pressure for gases. Be aware that equipment rated for a specific
temperature and pressure, cannot be operated at the same pressure if the temperature is increased. While it may seem extremely intuitive
(think ideal gas law), this relationship is too often overlooked, usually with serious consequences. When operating any process, pay
attention to the temperature and pressure ratings. If they aren't readily known, review equipment files or contact the manufacturer before
making any changes to the process. Also, operation and maintenance should always be performed according to strict standards laid out by a
plant standard operating procedure (SOP). Any changes must be reviewed and approved by a cross functional team.
The above hazards and preventive measures have been outlined to illustrate how important safety is in a chemical plant, and the importance
of being aware of your surroundings and all possible safety hazards. Many of the aforementioned topics may seem like common sense, but
it is very easy to overlook small details in the scope of a large-scale chemical process. Keeping safety in mind at all times as the paramount
of any process can ensure that people leave their shift the same way they arrived.
4.5.5: ALARMS IN PROCESSES

Alarm configuration and specification is an important part in the design and operation of any chemical process. Alarms are implemented in
a process design to aid in the control of the process. Federal and industrial documents only specifically reference alarms in the context of
processes exceeding regulatory compliance limits. In this sense, alarms are used to control safety and environmental hazards. Other
important uses of alarms are to control product yield, product quality, and operational limits of process equipment. This section will discuss
the steps taken to implement alarms in processes, common different levels of alarms, and common instances in which alarms are useful to
comply with regulations.
4.5.5.1: ALARM LIFECYCLE
When it has been determined that an alarm is needed to aid in the control of a process, the alarm must be specified. The following are basic
steps in implementing an alarm.
1. First, the process designer needs to know what category the alarm fits into. This is important because responses from different categories
of alarms are usually managed differently. They may be prioritized in case multiple alarms occur at once, so that the proper follow-up
reports and procedures may be taken accordingly. Common categories include product quality parameters, safety, environmental
considerations, and equipment protection.
2. After determining the specific use of the alarm, the limits must be set. If the process variable exceeds the set limits, the alarm will be
triggered. (More information on limits will be discussed in Alarm Levels.)
3. Next, the computer system for the alarm must be configured. The computer system may contain logic loops that automatically change
control parameters to offset the problem and merely inform the operator that it has done so. Alternatively, an alarm may trigger horns,
flashing lights, or send a page to an operator, alerting them that there is a problem that needs attention.
4. The correct user response and interaction must then be defined for the alarm. This includes providing proper training for the particular
process, procedures and operator manuals describing how the event should be investigated, guidelines on when action needs to be taken,
and guidelines on when to escalate the situation to a more serious event.
Proper communication from the automated alarm system is critical. Care needs to be taken that alarms only signify abnormal conditions that
require a response. For example, the successful completion of a batch operation is an important piece of information and should generate a
computer message so that the operator knows. However, this does not represent an abnormal situation and should therefore not show up on
the computer as an alarm. While this seems quite obvious, most industrial plants struggle with maintaining alarm systems free of "nuisance
alarms." It is also important that alarms have proper descriptions so that the reason an alarm appears is clear. For example, an alarm could
appear in a large plant that says, “LI-501 exceeds limits” by default. Although this might be useful to the engineer designing the alarm
system as an indication that tank 501 has a high level of material, an operator that sees the alarm or other engineers working on the system
may not know what the alarm is communicating. A more universally meaningful alarm indicator might say, “Material level in tank 501 is
high.” With this simple change in the computer system, the alarm would be more effective for personnel to locate the problem quickly and
act accordingly.
4.5.5.2: ALARM LEVELS

Alarms are available in a wide variety of types, with multiple levels of alarm. In all processes, disturbances occur that can shift a plant's
operation away from normal. When this happens, measures are usually taken by computers, such as with the use of P&ID control loops, to
keep the process under control. With these control systems, processes are designed to fall within a range of acceptable normal operating
limits. When a process deviates beyond these normal limits, an alarm should be triggered.
For most processes, the minimum for safe operation is two levels of alarms: warning and critical. The warning alarm tells plant operators
that the process has deviated beyond the acceptable limits and provides them with the time and ability to take corrective action so that the
product quality is not affected and environmental and safety regulations are not exceeded. If the right actions are not taken or are not taken
quickly enough to correct the problem, a critical alarm may then be triggered. The critical alarm tells the plant operators that conditions are
dangerously close to breaching what is allowed. In many cases, the critical alarm will call for a systematic shut-down of the operation until
the problems can be addressed.
The conditions at which warning and critical alarms are triggered are those conditions that exceed the limits determined for the process.
Measurement uncertainty must always be considered because all devices in the control system will be subject to some possible error, even if
it is small as +/- 1%. Setting an alarm at exactly the proven acceptable range for the process could allow a measured value to fall within this
range, even though the actual value lies outside. This is called a "false acceptance." By performing error analysis and statistical distribution
theory, the alarm limits can be adjusted as needed. This is a process called "guard banding," and it prevents real disturbances in the process
from being ignored by the alarm system. Information about alarm limits should be well documented so if changes to the system are
proposed, designers know how the limits were originally determined.
The figure below provides a visual representation of alarm ranges.
As seen in the center of the above figure, the control loop set point is the optimum point of control for the process (e.g. the optimum
temperature and concentrations of reactants for a reaction). It is impossible to maintain the process at exactly this point, so there is a range
of "normal" operation, inside which the process is still considered to be running in an acceptable way. The warning alarm would be
triggered when the process goes outside of the limits of this range (lower than A or higher than B), allowing time for the process to be
brought back under control. The critical alarm would then go off if the process goes beyond the alarm setting (lower than C or higher than
D). This setting is determined by guard banding the process acceptable range (PAR) for uncertainty, seen in the figure by the fact that the
alarm setting lies well inside the PAR. The spaces between E and C and between D and F are determined by the uncertainty. Lastly, this
PAR must be inside the total range of the sensor that determines the instrument failure.
Alarms must be analyzed based on their priority:

SAFETY STEPS TO HIGH RELIABILITY
A safe system involves many layers of responses when an incident occurs.
The center of the ring is the basic process control system.

The first layer of response is the alarm system which draws attention.
The second layer is the Safety Interlock System which can stop/start the equipment.
The third layer is the Relief system which leases pressure build-up in the sytem.
The fourth layer is containment which prevents material from reaching workers, community, or the environment.
The last layer to the ring is the emergency response system which involves evacuation, fire fighting, etc.

A reagent recovery unit for a chemical process plant is being designed. The goal is to recover tin from a tin-plating waste stream through
binary extraction with carbon tetrachloride. The equipment used in the process and P&ID are as follows:
Carbon tetrachloride storage tank
Mixing Vessel
Pumps
Heat Exchanger
Associated piping
With knowledge of hazardous locations and risk hotspots, and safe design principles, analyze the flow diagram. Identify areas of risk and
specific improvements that should be made to the process design before implementation begins. (note: no piping, valves, or controllers exist
on the P&ID for the sake of clarity. Ignore these in your analysis) .
4.5.6.1: AREAS OF RISK

Any piece of equipment could potentially become dangerous if the right situation were to arise. Each piece of equipment used in this
process is no exception.
Storage tank - These process units that are of secondary importance to the process goal, don’t draw as much attention from plant
operators and engineers, and tend to receive less maintenance.
Mixing Vessel - The potential for heat buildup due to heat of mixing and the kinetic energy of the spinning motor makes this vessel a
risk hotspot. Improper control of temperature within the vessel or motor speed (especially during periods when the tank may be empty)
can lead to equipment malfunction or even explosions
Pumps and Heat exchangers - The energy stored through pressurization by the pumps and potentially high temperature solutions handled
by the heat exchanger present the possibility for danger. Wherever energy is stored, danger is associated with its potential release.
Associated piping - Corrosion or failure to maintain pipes and associated elements are the main cause of the malfunction leading to
danger. Leaks or total failures can release reactive materials on other equipment causing corrosion or malfunction
4.5.6.2: SPECIFIC IMPROVEMENTS

This process could be improved and made safer in the following ways.
Simplification of the piping used to connect the storage tank to the mixing vessel will reduce the risk of leaks and malfunctions due to
piping complications.
The addition of a emergency relief valve and temperature controlling/insulating element to the mixing vessel will greatly reduce the risk
of equipment failure due to overheating or overpressureization due to heats of mixing or outside heating influences.
Use a reactor with a rounded top, as this type can withstand much higher pressures.
The substitution of carbon tetrachloride with a less hazardous solvent, if the specific solubility required to extract the tin still exists, like
cyclohexane, isopropyl alcohol, or 1,1,1-trichloroethane reduces the risk of health complications due to exposure, and possible
explosions.
 EXAMPLE 4.5.1
A P&ID appears below for the production of a solution containing a pharmaceutically active compound. The reaction taking place in
the CSTR is highly exothermic. After examining the P&ID for this part of the process, describe a possible alarm system.
Solution
The CSTR for the exothermic reaction is jacketed with a cooling water stream. An alarm should be in place to monitor the reactor
temperature. A warning alarm can notify the operator that the temperature is too high and corrective action needs to be taken. A critical
alarm should be in place to warn that the reactor is nearing runaway conditions and an immediate response is needed. If the necessary
action is not taken, systematic shutdown of the reactor could occur. This would involve closing the valves, flooding the jacket with
cooling water, and having the impeller on. Another possibility for an alarm, although we do not know how the products are being used
specifically, is in a composition measurement of the product containing the pharmaceutically active compound. Depending on where
this stream is going and how it is being used, too high a concentration could be dangerous if no other concentration-altering steps occur
before the finished product goes out to consumers.
4.5.7: ACTUAL CASE

The bulletin below is taken for the US Chemical Safety and Hazard Invesitgation Board
Washington, DC, July 15, 2004 - The U.S. Chemical Safety and Hazard Investigation Board (CSB) today released a Safety Bulletin calling
on chemical plant and refinery operators to exercise appropriate caution when performing work on piping and equipment that could
contain hazardous materials.
The bulletin, set to be considered by the Board at a public meeting today, results from the CSB's investigation into a January 13, 2004,
explosion and fire at the Huntsman Petrochemical facility in Port Neches, Texas, where two employees were seriously burned and
significant damage occurred to nearby equipment. The explosion and fire occurred as workers attempted to purge a thousand-foot-long
chemical process pipe in preparation for a cutting and welding operation.
Huntsman managers and workers were aware of the importance of completely removing hazardous material before cutting into the piping,
and relevant warnings were contained in the company's written operating procedures. In this incident, workers first purged the piping with
nitrogen to force out residual chemicals, including a hazardous mixture of peroxide and alcohol that reacts violently when heated. But
unknown to the workers, the piping included a 300-foot-long section that was three feet lower than the rest of the piping, and despite the
nitrogen purge, a significant amount of the hazardous mixture remained trapped. The next step in the operation was to use high-temperature
steam to purge the piping of what workers believed would be a small amount of residual flammable hydrocarbon vapor. But the steam
heated the peroxide that was trapped in the low section of piping. The peroxide then began to decompose, releasing heat and creating
intense pressure. The pressure blew out a valve gasket and violently ruptured the pipe. Flammable vapors shot out of the openings and
ignited into a large fireball, injuring plant workers.
After the accident, Huntsman found two drains in the low section of the pipe, which could have been used to remove the trapped liquid. Had
Huntsman's procedures called for reviewing plant pipe drawings and physically walking the entire line within the work boundaries, the
accident would likely have been avoided, CSB said.
Lessons Learned for this incident:
Physically examine all piping and components between isolation devices such as valves, and be sure piping drawings are current
Use the drawings to identify key components, such as low-point drains that can be used to remove dangerous chemicals
Prepare a specific written procedure for removing hazardous material and consider the consequences of working on piping that is not
completely purged.''
4.5.8: REFERENCES
AIChE (January 2006). Process Safety Beacon, CEP Magazine
AIChE (March 2006). Process Safety Beacon, CEP Magazine
AIChE (May 2006). Process Safety Beacon, CEP Magazine
AIChE (November 2003). Process Safety Beacon, CEP Magazine
AIChE (September 2006). Process Safety Beacon, CEP Magazine
Gael D. Ulrich, Palligarnai T. Vasudevan (2006). Predesign With Safety in Mind, Univ. Of New Hampshire
Georges A. Melhem, Peter Howell (2005). Designing Emergency Relief Systems for Runaway Reactions, CEP Magazine
Joseph S. Alford, John Kindervater, Robert Stankovich (April 2005). Alarm Management for Regulated Industries, CEP Magazine
U.S. Chemical Safety and Hazard Investigation Board, 15 July 2004. CSB News Release
www.csb.gov/index.cfm?folder=news_releases&page=news&NEWS_ID=171

Authors:Bradley Anderson, Aaron Bennick, Michael Salciccioli Stewards: Jocelyn Anleitner, Stephanie Combs, Diane Feldkamp,
Heeral Sheth
4.5: Safety Features in Piping and Instrumentation Diagrams is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter
request.
4.6: REGULATORY AGENCIES AND COMPLIANCE
Authors: Jayshree Mahajan, Bradley Anderson, Aaron Bennick, Michael Salciccioli
Stewards: Jocelyn Anleitner, Stephanie Combs, Diane Feldkamp, Heeral Sheth
4.6.1: INTRODUCTION
Regulatory agencies govern various sections of chemical plants from emissions and water treatment to personnel behaviors and company
policy. Understanding of the types of organizations that demand high performance from chemical plants on a national and local level are
extremely important for the safety of the employees, community and environment.
4.6.2: REGULATORY COMPLIANCE

Compliance is an integral part of ensuring the safety of all that work in close contact with the chemical plant as well as minimizing fines
and fees that come with violating regulations. Compliance can be organized into two main categories; plant safety and environmental safety.
Plant safety ensures employees and the surrounding community are adequately protected during full-operation of the plant. Environmental
safety comprises protecting and treating the environmental carefully when it comes to various contaminated streams in the plant.
4.6.2.1: PLANT SAFETY FOR EMPLOYEES AND THE COMMUNITY

Safety is paramount in any chemical process, and alarms are essential for compliance with safety constraints. OSHA has established
guidelines that must be followed in any plant when dealing with chemicals defined as highly hazardous. Limits for temperature, pressure,
flow rates, and compositions need to be regulated. Alarms should be in place to warn operators when a limit is near so that steps can be
taken to ensure the safety of people in the plant. An example of this is a safe temperature limit for a CSTR. The alarm signals that the
temperature is too high and action is needed to prevent a runaway scenario. If the corrective action is not taken, or not taken quickly
enough, a critical alarm can signal a computer program to automatically shut down the entire process or specific unit operation.
Some industries may expect product quality to be closely regulated by the FDA or other government agencies. Typically, this will be in a
process where the final product is directly used by people and the margins for error are small. These processes include food processing and
manufacturing consumer products, especially pharmaceuticals. These industries usually require systems in place that frequently validate
alarms, as well as documentation for all critical alarm events. Measurements such as the weight percent of a pharmaceutically active
compound in a solution must be carefully monitored, with recorded uncertainty analysis.
4.6.2.2: ENVIRONMENTAL SAFETY

Emissions of solids, liquids, and gases in a plant are heavily regulated by government agencies. Regulations apply for processes that emit
chemicals to the atmosphere (either directly or following a scrubber), processes that discharge material into a body of water, or processes
that require containment control devices like check valves and rupture disks. Alarms are frequently used to comply with these regulations
by measuring things such as pH and organic solvent concentration. Typically, a warning alarm will alert personnel that a threshold may be
breached if action is not taken, allowing time to avoid an incident requiring a formal report. Critical alarms can alert operators that a
threshold has been passed and automatically trigger the appropriate action, such as a systematic shutdown of the process.
4.6.3: REGULATORY AGENCIES

Federal and national agencies maintain smaller state subsections of various programs and administrations. In most cases, the state level
requires stricter compliance and lower limits. Solely state-controlled programs usually handle the air and water quality since any regulation
violation results in the consequences for the immediate community.
4.6.3.1: FEDERAL AND NATIONAL AGENCIES OR PROGRAMS

Plant safety and environmental safety programs regulated on the national or federal level are monitored by three main agencies; the
Environmental Protection Agency (EPA), the U.S. Department of Labor, the Food and Drug Administration, and the Department of
Homeland Security. These three governing bodies have created numerous acts, committees, administrations and policies that protect the
welfare of the employee, community, and environment.
The EPA at the federal level, provides acts, laws, and regulations, that help maintain and improve the air and water quality. The risk
management program (RMP) is a mandatory program that "require facilities that produce, handle, process, distribute, or store certain
chemicals to develop a Risk Management Program". A Risk Management Plan (RMP) must be submitted to the EPA for approval. Overall,
risk management is a large part of process control as control systems must adequately function and maintain compliance of an entire facility.
Failure logic for instrumentation, redundant sensors, and critical alarms are essential in maintaining compliance, but a RMP is crucial for
handling low-likelihood emergency situations.
The U.S. Department of Labor maintains the Occupational Safety and Health Administration (OSHA) which provides rules and regulations
for employers and employees on safe workplace practices. Although individual states may maintain their own occupational health and
safety plans, OSHA is the governing body and authority on those programs. Inspections are performed to ensure that all employees have a
clean and safe working environment that is hazard-free and risk-mediated. OSHA also maintains the Process Safety Management Program
(PSM) which regulates requirements for facilities that handle highly hazardous chemicals. A list of chemicals that qualify include chlorine,
formaldehyde, and hydroflouric acid. Requirements of PSM include frequent process hazard analysis (PHA), pre-startup safety review
(PSSR), and incident investigations. A cooperative program that OSHA maintains is The Voluntary Protection Programs (VPP). These
programs aim to bring together management and labor to provide comprehensive safety and health guidelines and regulations that keep all
employees safe while on the job. Each facility must submit application for entry into the program and upon acceptance will follow a set of
standards to ensure continued safety. The following is a detailed description of why PSM should be important to chemical engineers, and
also some highlights of the main aspects of PSM as required by OSHA.
4.6.3.2: PROCESS SAFETY MANAGEMENT (PSM) -- WHY IS THIS IMPORTANT TO CHEMICAL ENGINEERS?
See the following links to videos provided by the United States Chemical Safety Board describing a few catastrophic events that have
occurred in the chemical engineering industry:
2005 Texas City Refinery Explosion
Bhopal Disaster
Besides the catastrophic nature of events that can occur from neglecting Process Safety, large chemical facilities are granted a privilege and
license to operate by the different federal regulatory agencies. If these regulatory agencies perform an audit on a specific facility, and find
that their regulations are not being followed, then extremely large fines can be levied to the company, even to the extent of shutting the
facility down permanently by removing that facility's privilege to operate. In the case of PSM, it is OSHA who deals out these fines. For
example, in 2009 OSHA attempted to levy a record 87 million dollar fine to an integrated oil company, which has not been finalized in the
legal system yet, but gives a good example of how important it is for companies operating in the U.S., if they want to continue to operate
safely and economically, to follow all government regulations as closely as possible.
Unexpected releases of toxic, reactive, or flammable liquids and gases in processes involving highly hazardous chemicals have been
reported for many years in various industries that use chemicals with such properties. Regardless of the industry that uses these highly
hazardous chemicals, there is a potential for an accidental release any time they are not properly controlled, creating the possibility of
disaster. As a result of catastrophic incidents in the past, and to help ensure safe and healthful workplaces, OSHA has issued the Process
Safety Management of Highly Hazardous Chemicals standard (29 CFR 1910.119), which contains requirements for the management of
hazards associated with processes using highly hazardous chemicals. OSHA’s standard 29CFR 1910.119 emphasizes the management of
hazards associated with highly hazardous chemicals and establishes a comprehensive management program that integrates technologies,
procedures, and management practices. A detailed list of these standards can be found on the United States Department of Labor website:
http://www.osha.gov/SLTC/processsafetymanagement/standards.html An effective process safety management program requires a
systematic approach to evaluating the whole process. Using this approach the process design, process technology, operational and
maintenance activities and procedures, training programs, and other elements which impact the process are all considered in the evaluation.
Process safety management is the proactive identification, evaluation and mitigation or prevention of chemical releases that could occur as a
result of failures in process, procedures or equipment. OSHA prescribes essential tools to the success of process safety management
including:
Process Safety Information
Process Hazard Analysis
Operating Procedures and Practices
Employee Training
Pre-Startup Safety Review
Mechanical Integrity
Management of Change
Incident Investigation
Emergency Preparedness
Compliance Audits
The thought is, with the simultaneous implementation of all of these things at a facility dealing with large amounts of highly hazardous
chemicals, the risk of a catastrophic incident resulting from an unplanned release will be minimized. Following is a detailed discussion of
each of these tools prescribed by OSHA.
Process Safety Information (PSI)
Complete, accurate, and up-to-date written information concerning process chemicals, process technology, and process equipment is
essential to an effective process safety management program. The compiled information will be a necessary resource to a variety of users
including the team that will perform the process hazards analysis, those developing the training programs and operating procedures,
contractors whose employees will be working with the process, those conducting the pre-startup safety reviews, local emergency
preparedness planners, and insurance and enforcement officials. PSI includes, but is not limited to:
Material and safety data sheets (MSDS)
A block flow diagram showing the major process equipment and interconnecting process flow lines
Process Flow Diagrams (PFDs)
Piping and Instrument Diagrams (P&IDs)
Process design information, including the codes and standards relied on to establish good engineering design
4.6.3.3: PROCESS HAZARDS ANALYSIS (PHA)

A process hazards analysis (PHA) is one of the most important elements of the process safety management program. A PHA is an organized
and systematic effort to identify and analyze the significance of potential hazards associated with the processing and handling of highly
hazardous chemicals. A PHA is directed toward analyzing potential causes and consequences of fires, explosions, releases of toxic or
flammable chemicals, and major spills of hazardous chemicals. The PHA focuses on equipment, instrumentation, utilities, human actions,
and external factors that might impact the process. These considerations assist in determining the hazards and potential failure points or
failure modes in a process.
A team from each process unit in the facility will be tasked with conducting a PHA for their process unit at regularly scheduled intervals as
defined by OSHA. One example is in an oil refinery, where a PHA has to be conducted and documented for each process unit every five
calendar years. The competence of the team conducting the PHA is very important to its success. A PHA team can vary in size from two
people to a number of people with varied operational and technical backgrounds. The team leader needs to be fully knowledgeable in the
proper implementation of the PHA methodology that is to be used and should be impartial in the evaluation. The other full or part time team
members need to provide the team with expertise in areas such as process technology, process design, operating procedures and practices,
alarms, emergency procedures, instrumentation, maintenance procedures, safety and health, and any other relevant subject as the need
dictates. The ideal team will have an intimate knowledge of the standards, codes, specifications and regulations applicable to the process
being studied.
There are various methodologies for conducting a PHA. Choosing which one is right for each individual facility will be influenced by many
factors, including the amount of existing knowledge about the process. For more information on the different methodologies for conducting
a PHA, see Center for Chemical Process Safety of the American Institute of Chemical Engineers
4.6.3.4: OPERATING PROCEDURES

Operating procedures provide specific instructions or details on what steps are to be taken or followed in carrying out the task at hand. The
specific instructions should include the applicable safety precautions and appropriate information on safety implications. For example, the
operating procedures addressing operating parameters will contain operating instructions about pressure limits, temperature ranges, flow
rates, what to do when an upset condition occurs, what alarms and instruments are pertinent if an upset condition occurs, and other subjects.
Another example of using operating instructions to properly implement operating procedures is in starting up or shutting down the process.
Operating procedures and instructions are important for training operating personnel. The operating procedures are often viewed as the
standard operating practices (SOPs) for operations. Control room personnel and operating staff, in general, need to have a full understanding
of operating procedures. In addition, operating procedures need to be changed when there is a change in the process. The consequences of
operating procedure changes need to be fully evaluated and the information conveyed to the personnel. For example, mechanical changes to
the process made by the maintenance department (like changing a valve from steel to brass or other subtle changes) need to be evaluated to
determine whether operating procedures and practices also need to be changed. All management of change actions must be coordinated and
integrated with current operating procedures, and operating personnel must be alerted to the changes in procedures before the change is
made. When the process is shut down to make a change, the operating procedures must be updated before re-starting the process.
4.6.3.5: EMPLOYEE TRAINING

All employees, including maintenance and contractor employees involved with highly hazardous chemicals, need to fully understand the
safety and health hazards of the chemicals and processes they work with so they can protect themselves, their fellow employees, and the
citizens of nearby communities. Training conducted in compliance with the OSHA Hazard Communication standard (Title 29 Code of
Federal Regulations (CFR) Part 1910.1200) will inform employees about the chemicals they work with and familiarize them with reading
and understanding MSDSs. However, additional training in subjects such as operating procedures and safe work practices, emergency
evacuation and response, safety procedures, routine and non-routine work authorization activities, and other areas pertinent to process safety
and health need to be covered by the employer's training program.
In establishing their training programs, employers must clearly identify the employees to be trained, the subjects to be covered, and the
goals and objectives they wish to achieve. The learning goals or objectives should be written in clear measurable terms before the training
begins. These goals and objectives need to be tailored to each of the specific training modules or segments. Employers should describe the
important actions and conditions under which the employee will demonstrate competence or knowledge as well as what is acceptable
performance.
Careful consideration must be given to ensure that employees, including maintenance and contract employees, receive current and updated
training. For example, if changes are made to a process, affected employees must be trained in the changes and understand the effects of the
changes on their job tasks. Additionally, as already discussed, the evaluation of the employee's absorption of training will certainly
determine the need for further training.
4.6.3.6: PRE-STARTUP SAFETY REVIEW

For new processes, the employer will find a PHA helpful in improving the design and construction of the process from a reliability and
quality point of view. The safe operation of the new process is enhanced by making use of the PHA recommendations before final
installations are completed. P&IDs should be completed, the operating procedures put in place, and the operating staff trained to run the
process, before startup. The initial startup procedures and normal operating procedures must be fully evaluated as part of the pre-startup
review to ensure a safe transfer into the normal operating mode.
For existing processes that have been shut down for turnaround or modification, the employer must ensure that any changes other than
"replacement in kind" made to the process during shutdown go through the management of change procedures. P&IDs will need to be
updated, as necessary, as well as operating procedures and instructions. If the changes made to the process during shutdown are significant
and affect the training program, then operating personnel as well as employees engaged in routine and non-routine work in the process area
may need some refresher or additional training. Any incident investigation recommendations, compliance audits, or PHA recommendations
need to be reviewed to see what affect they may have on the process before beginning the startup.
4.6.3.7: MECHANICAL INTEGRITY
Employers must review their maintenance programs and schedules to see if there are areas where "breakdown" is used rather than the more
preferable on-going mechanical integrity program. Equipment used to process, store, or handle highly hazardous chemicals has to be
designed, constructed, installed, and maintained to minimize the risk of releases of such chemicals. This requires that a mechanical integrity
program be in place to ensure the continued integrity of process equipment.
Elements of a mechanical integrity program include identifying and categorizing equipment and instrumentation, inspections and tests and
their frequency; maintenance procedures; training of maintenance personnel; criteria for acceptable test results; documentation of test and
inspection results; and documentation of manufacturer recommendations for equipment and instrumentation.
4.6.3.8: MANAGEMENT OF CHANGE
To properly manage changes to process chemicals, technology, equipment and facilities, one must define what is meant by change. In the
process safety management standard, change includes all modifications to equipment, procedures, raw materials, and processing conditions
other than "replacement in kind." These changes must be properly managed by identifying and reviewing them prior to implementing them.
For example, the operating procedures contain the operating parameters (pressure limits, temperature ranges, flow rates, etc.) and the
importance of operating within these limits. While the operator must have the flexibility to maintain safe operation within the established
parameters, any operation outside of these parameters requires review and approval by a written management of change procedure.
Management of change also covers changes in process technology and changes to equipment and instrumentation. Changes in process
technology can result from changes in production rates, raw materials, experimentation, equipment unavailability, new equipment, new
product development, change in catalysts, and changes in operating conditions to improve yield or quality. Equipment changes can be in
materials of construction, equipment specifications, piping pre-arrangements, experimental equipment, computer program revisions, and
alarms and interlocks. Employers must establish means and methods to detect both technical and mechanical changes.
Temporary changes have caused a number of catastrophes over the years, and employers must establish ways to detect both temporary and
permanent changes. It is important that a time limit for temporary changes be established and monitored since otherwise, without control,
these changes may tend to become permanent. Temporary changes are subject to the management of change provisions. In addition, the
management of change procedures are used to ensure that the equipment and procedures are returned to their original or designed conditions
at the end of the temporary change. Proper documentation and review of these changes are invaluable in ensuring that safety and health
considerations are incorporated into operating procedures and processes. Employers may wish to develop a form or clearance sheet to
facilitate the processing of changes through the management of change procedures. A typical change form may include a description and the
purpose of the change, the technical basis for the change, safety and health considerations, documentation of changes for the operating
procedures, maintenance procedures, inspection and testing, P&IDs, electrical classification, training and communications, pre-startup
inspection, duration (if a temporary change), approvals, and authorization. Where the impact of the change is minor and well understood, a
check list reviewed by an authorized person, with proper communication to others who are affected, may suffice. For a more complex or
significant design change, however, a hazard evaluation procedure with approvals by operations, maintenance, and safety departments may
be appropriate. Changes in documents such as P&IDs, raw materials, operating procedures, mechanical integrity programs, and electrical
classifications should be noted so that these revisions can be made permanent when the drawings and procedure manuals are updated.
Copies of process changes must be kept in an accessible location to ensure that design changes are available to operating personnel as well
as to PHA team members when a PHA is being prepared or being updated.
Incident Investigation
Incident investigation is the process of identifying the underlying causes of incidents and implementing steps to prevent similar events from
occurring. The intent of an incident investigation is for employers to learn from past experiences and thus avoid repeating past mistakes.
The incidents OSHA expects employers to recognize and to investigate are the types of events that resulted in or could reasonably have
resulted in a catastrophic release. These events are sometimes referred to as "near misses," meaning that a serious consequence did not
occur, but could have.
Employers must develop in-house capability to investigate incidents that occur in their facilities. A team should be assembled by the
employer and trained in the techniques of investigation including how to conduct interviews of witnesses, assemble needed documentation,
and write reports. A multi-disciplinary team is better able to gather the facts of the event and to analyze them and develop plausible
scenarios as to what happened, and why. Team members should be selected on the basis of their training, knowledge and ability to
contribute to a team effort to fully investigate the incident.
Emergency Preparedness
Each employer must address what actions employees are to take when there is an unwanted release of highly hazardous chemicals.
Emergency preparedness is the employer's third line of defense that will be relied on along with the second line of defense, which is to
control the release of chemical. Control releases and emergency preparedness will take place when the first line of defense to operate and
maintain the process and contain the chemicals fails to stop the release.
Employers will need to select how many different emergency preparedness or third lines of defense they plan to have, develop the necessary
emergency plans and procedures, appropriately train employees in their emergency duties and responsibilities, and then implement these
lines of defense. Employers, at a minimum, must have an emergency action plan that will facilitate the prompt evacuation of employees
when there is an unwanted release of a highly hazardous chemical. This means that the employer's plan will be activated by an alarm system
to alert employees when to evacuate, and that employees who are physically impaired will have the necessary support and assistance to get
them to a safe zone. The intent of these requirements is to alert and move employees quickly to a safe zone. Delaying alarms or confusing
alarms are to be avoided. The use of process control centers or buildings as safe areas is discouraged. Recent catastrophes indicate that lives
are lost in these structures because of their location and because they are not necessarily designed to withstand overpressures from shock
waves resulting from explosions in the process area.
Preplanning for more serious releases is an important element in the employer's line of defense. When a serious release of a highly
hazardous chemical occurs, the employer, through preplanning, will have determined in advance what actions employees are to take. The
evacuation of the immediate release area and other areas, as necessary, would be accomplished under the emergency action plan. If the
employer wishes to use plant personnel-such as a fire brigade, spill control team, a hazardous materials team-or employees to render aid to
those in the immediate release area and to control or mitigate the incident, refer to OSHA's Hazardous Waste Operations and Emergency
Response (HAZWOPER) standard (Title 79 CFR Part 1910.1 20). If outside assistance is necessary, such as through mutual aid agreements
between employers and local government emergency response organizations, these emergency responders are also covered by
HAZWOPER. The safety and health protection required for emergency responders is the responsibility of their employers and of the on-
scene incident commander.
Compliance Audits
An audit is a technique used to gather sufficient facts and information, including statistical information, to verify compliance with standards.
Employers must select a trained individual or assemble a trained team to audit the process safety management system and program. A small
process or plant may need only one knowledgeable person to conduct an audit. The audit includes an evaluation of the design and
effectiveness of the process safety management system and a field inspection of the safety and health conditions and practices to verify that
the employer's systems are effectively implemented. The audit should be conducted or led by a person knowledgeable in audit techniques
who is impartial towards the facility or area being audited. The essential elements of an audit program include planning, staffing, conducting
the audit, evaluating hazards and deficiencies and taking corrective action, performing a follow-up review, and documenting actions taken.
4.6.3.9: OTHER FEDERAL ENTITIES

The Food and Drug Administration (FDA) is an agency of the United States Department of Health and Human Services. The FDA regulates
food, drugs, cosmetics, biologics, medical devices, radiation-emitting devices and vetenary products manufactured in the United States. The
main goal of the FDA is to maitain that the products they regulate are safe, effective, and secure. The FDA is also responsible that the
products are accurately represented to the public. State and local governments also help regulate these products in cooperation with the
FDA. The FDA does not regulate alcohol, illegal drugs, and meat and poultry.
The Department of Homeland Security has recently taken a role in regulating chemical plants because of 9/11. Chemical plants are seen by
the government as targets for terrorists and security in and around the plant is a major concern. Although the laws are typically state run, the
Department of Homeland Security has required mandatory national security standard to chemical plants throughout the nation. Although the
mandates were fought by legislative for years, the Department of Homeland Security has influence in the security in chemical plants. The
law requires the plant to prepare a vulnerability test and submit a site security plan. In order to validate these activities, audits and site visits
are/will be performed by government officials.
4.6.3.10: STATE REGULATIONS
State regulations vary greatly from state-to-state depending on the main concern. For instance, beach quality is more important in California
than Nebraska due to the geographical location. Similarly state legislative acts and administrations can expect high performance from the
chemical industry as it has the capability to affect all parts of a community for many generations.
4.6.4: REFERENCES
U.S. National Profile on the Management of Chemicals. www.epa.gov/oppfead1/profile/
U.S. Department of Labor. Occupational Safety & Health Administration. Process Safety Management.
http://www.osha.gov/SLTC/processsafetymanagement/index.html
U.S. Department of Labor. Occupational Safety & Health Administration. State Occupational Health and Safety Plans.
http://www.osha.gov/dcsp/osp/index.html
U.S. Department of Labor. Voluntary Protection Programs. http://www.osha.gov/dcsp/vpp/index.html
Environmental Protection Agency. Regulatory Information by Business Sector. www.epa.gov/lawsregs/bizsector/index.html
Environmental Protection Agency. Regulatory Information by Environmental Topic. www.epa.gov/lawsregs/envtopics/index.html#air
Environmental Protection Agency. State Specific Regulatory Information Search Engine. www.epa.gov/lawsregs/where/index.html
Environmental Protection Agency. Risk Management Plan. http://www.epa.gov/oem/content/rmp/index.htm
United States Chemical Safety and Hazard Regulation Board. Homepage. www.chemsafety.gov/
U.S. Department of Health & Human Services. Food and Drug Administration. http://www.fda.gov/
United States Chemical Safety Board (CSB). http://www.csb.gov
Center for Chemical Process Safety (CCPS). http://www.aiche.org/ccps
4.6: Regulatory Agencies and Compliance is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source
CHAPTER OVERVIEW
5: LOGICAL MODELING
5.1: Boolean Models - Truth Tables and State Transition Diagrams
5.2: Logical Control Programs - IF... THEN… WHILE…
5: Logical Modeling is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source content that was edited
1
5.1: BOOLEAN MODELS - TRUTH TABLES AND STATE TRANSITION DIAGRAMS
INTRODUCTION TO BOOLEAN MODELS
A Boolean is a variable that can only attain two values: True or False. In most applications, it is convenient to represent a True by the
number 1, and a False by the number 0. A Boolean model, or Boolean network, is a collection of Boolean variables that are related by
logical switching rules, or Boolean functions, that follow an If-Then format. This type of Boolean model is known as an autonomous model
and will be the primary type of model discussed in this article.
In chemical engineering, Boolean models can be used to model simple control systems. Boolean functions can be used to model switches on
pumps and valves that react to readings from sensors that help keep that system operating smoothly and safely. A simple application for
level control of a CSTR is included in worked-out example 1. In this example, Boolean function is used to close the inlet stream and open
the outlet stream when the level is higher than a specified point.
BOOLEAN FUNCTIONS
Boolean functions are logical operators that relate two or more Boolean variables within a system and return a true or false. A Boolean
expression is a group of Boolean functions, which will be described individually below. When computing the value of a Boolean
expression, Parentheses are used to indicate priority (working from inside out as in algebra). After that, LOGICAL INVERSION will
always be first and LOGICAL EQUIVALENCE will be last, but the order of operation for the AND, OR, and EXCLUSIVE OR functions
are specified with parenthesis.
Descriptions and examples of these functions are given below. A quick reference of each of the functions can be found after the examples.
LOGICAL INVERSION
LOGICAL INVERSION is a function that returns the opposite value of a variable. The function is denoted as a prime on the variable (e.g.
A' or B') For example, if we say that A is true (A=1), then the function A' will return a false (A'=0). Similarly, if we say that A is false
(A=0) then the function A' will return true (A'=1).
Example:
A=1, B=A' then B=0
AND
The AND function relates two or more Boolean variables and returns a true if-and-only-if both variables are true. A dot is used to denote the
AND function, or it is simply omitted. For example "A and B" can be written as "A•B" or as "AB." In this example, the AND function will
only return a true if-and-only-if both Boolean variables A and B have a value of 1.
Examples:
Variables Results
A=1, B=1 AB = 1
A=1, B=0 AB = 0
A=1, B=1, C=1 ABC = 1
A=1, B=0, C=1 ABC = 0
OR
The OR function relates two or more Boolean variables and returns a true if any referenced variables are true. A plus is used to denote the
OR function. For example "A or B" can be written as "A+B." In this example, the OR function will return true if either Boolean variable, A
or B, has a value of 1.
Examples:
Variables Results
A=1, B=1 A+B = 1
A=1, B=0 A+B = 1
A=0, B=0 A+B = 0
A=0, B=0, C=1 A+B+C = 1
A=0, B=0, C=0 A+B+C = 0
EXCLUSIVE OR
The EXCLUSIVE OR function relates two or more Boolean variables and returns true only when one of the variables is true and all other
variables are false. It returns false when more than one of the variables are true, or all the variables are false. A circumscribed plus is used
to denote the EXCLUSIVE OR function. For example "A EXCLUSIVE OR B" can be written as "A B."
Examples:
Variables Results
A=1, B=1 A B=0
A=1, B=0 A B=1
A=0, B=1 A B=1
A=0, B=0 A B=0
A=0, B=0, C=0 A B C=0
A=1, B=0, C=0 A B C=1
A=1, B=0, C=1 A B C=0
A=1, B=1, C=1 A B C=0
LOGICAL EQUIVALENCE
The LOGICAL EQUIVALENCE function equates two Boolean variables or expressions. The LOGICAL EQUIVALENCE function,
denoted as =, assigns a Boolean variable a true or false depending on the value of the variable or expression that it is being equated with.
For example, A LOGICAL EQUIVALENCE B can be written as A = B. In this example, the value of A will be assigned the value of B.
BOOLEAN NETWORKS
As stated in the introduction, a Boolean network is a system of boolean equations. In chemical engineering, Boolean networks are likely to
be dependant on external inputs as a means of controlling a physical system. However, the following sections pertain mostly to synchronous
autonomous systems. An autonomous system is one that is completely independent of external inputs. Every Boolean variable is dependent
on the state of other Boolean variables in the system and no variable is controlled by an external input. A synchronous system is one that
logical switching (the changing of Boolean variables) occurs simultaneously for all variables based on the values prior to the incidence of
change.
Here is an example of an autonomous boolean network:
Boolean Functions
A = B+C'
B = AC
C = A'
TRUTH TABLES
A truth table is a tabulation of all the possible states of a Boolean Model at different time frames. A simple truth table shows the potential
initial states at time, Ti, and the corresponding subsequent states at time Ti+1, of a Boolean network. Truth tables can provide one with a
clearer picture of how the rules apply and how they affect each situation. Hence, they help to ensure that each output only has one control
statement so that the Boolean rules do not conflict with each other.
CONSTRUCTING TRUTH TABLES
1. Draw up a table with the appropriate number of columns for each variable; one for each input and output.
2. The left side of the column should contain all possible permutations of the input variables at time Ti. One method to accomplish this
might be to list all possible combinations in ascending binary order.
3. The right side of the column should contain the corresponding outcome of the output variables at the subsequent time Ti+1. A generic
example of this with 2 variables can be seen below:
A quick way to check that you have all of the possible permutations is that there should be 2x possible permutations for X input variables.
EXAMPLE OF A TRUTH TABLE

The sample system we will be using is based on hydrogen fuel cell technology. The equation for the operation of hydrogen fuel cells is
H +O ⟶ H O.
2 2 2
One aspect of Proton Exchange Membrane (PEM) fuel cells (a type of fuel cell) is that the performance of the fuel cell is highly
dependent on the relative humidity of the system (if humidity rises too high, the fuel cell will flood and H2 and O2 will be unable to reach
the cell. If humidity falls too low, the fuel cell will dry up and the performance will drop.) The task is to create a Boolean model for this
simplified water management system.
The system produces steam within the system, and there is a vent to release steam if the system becomes too saturated. In our system, we
will assume that the inputs are stoichiometric and react completely. Also we will assume that pressure buildup from steam is negligible
compared to the change in relative humidity. The only variable in question is the %relative humidity in the system.
note: this is not how water management actually works in a fuel cell system, but it is a simple example.
A will represent the moisture controller response (0 indicates relative humidity or %RH < 80%, 1 indicates %RH >80%)
B will represent the valve status (0 is closed, 1 is open)
The corresponding Boolean functions for this model are given below (normally you would have to design these yourself to meet the criteria
you desire):
A=A
B=A
For this example with 2 input variables, there are 22 = 4 possible permutations and 22 = 4 rows. The resultant permutations for the outputs
are: For A where Y=1, the number of 0s and 1s are 2(Y-1)=2(1-1)=1. For B where Y=2, the number of 0s and 1s are 2(Y-1)=2(2-1)=2.
The resultant truth table is below:
STATE TRANSITION DIAGRAMS

A state transition diagram is a graphical way of viewing truth tables. This is accomplished by looking at each individual initial state and its
resultant state. The transition from one state to another is represented by an arrow. Then they are pieced together like a jigsaw puzzle until
they fit in place. When one state leads to itself it simply points to itself. The following example is based on the truth table in the previous
section.
In this example, there are two state cycles. A state cycle is a combination of states around which the system continually enters and reenters.
For a finite number of states, there will always exist at least one state cycle. A state cycle is also a pathway or a flowchart that shows the
"decision making process" of a Boolean network. This feature is a direct result from two attributes of Boolean networks:
1. Finite number of states
2. Deterministic (there is a certain set of rules that determines the next state that will be entered)
In the example presented in the previous section, there were two state cycles. One advantage of state cycles is it easily allows you to see
where your model will end up cycling and if there are any states that are not accounted for properly by your model. In the previous diagram,
if the moisture controller indicated the humidity was below the set value, it would close the valve or hold the valve closed. If the moisture
controller indicated that the humidity was above the set value, it would either open the valve or hold it open.
Consider this alternate system.
In this example, the state cycle says that if the meter says that the humidity is below the set point it would cycle the vent valve open and
closed. This would hurt the system and is not a desired outcome of the model.
For safety and functionality issues, a process control engineer would want to consider all possiblities in the design of any Boolean network
modeling a real system.
LIMITATIONS OF BOOLEAN NETWORKS

ADVANTAGES OF BOOLEAN NETWORKS
Unlike ordinary differential equations and most other models, Boolean networks do not require an input of parameters.
Boolean models are quick and easy to compute using computers.
Boolean networks can be used to model a wide variety of activities and events.
Boolean networks can be used to approximate ordinary differential equations when there are an infinite number of states.
DISADVANTAGES OF BOOLEAN NETWORKS

Boolean networks are restrained to computing very simple math. They cannot be used for calculus and to calculate large quantities.
Boolean models have relatively low resolution compared to other models.
It is very time consuming and complicated to build Boolean networks by hand.
 EXAMPLE 5.1.1
A hypothetical CSTR needs to have its liquid level maintained below a safety mark by means of a sensor, L1, on the corresponding
mark and a control valve placed on the inlet and outlet streams – V1 and V2 respectively. A typical application of the afore-mentioned
system could involve heterogeneously catalyzed liquid reaction(s) with liquid product(s).
Solution
Conventions
Water level sensor
L1 0 1
water level desirable too high
Valve
V 0 1
position closed open
Initial State
Assume that the CSTR is empty and being filled up. CSTR, being empty, sets the value of L1 to zero. Filling up the CSTR could be
done by opening valve 1 - V1 assuming a value of one - and closing valve 2 - V2 assuming a value of zero.
In coordinate form, the initial state is as such: (L1, V1, V2) = (0, 1, 0)
Problem Interpretation
Let h be the water level and WL1 be the safety mark defined in the CSTR. The system could assume one of the following states at any
one time:
1) h < WL1 : desirable water level
Maximizing production of the chemical prompts the system to remain in its current state - that is, its initial state. (L1, V1, V2)final = (0,
1, 0) final state
2) h > WL1 : water level too high
Prevention of flooding requires that the tank be emptied. As such, valve 1 (V1) should be closed to stop the input while valve 2 (V2)
should be open to empty the extra water above the safety water mark. (L1, V1, V2)' = (1, 1, 0) trigger to valve (L1, V1, V2)final = (1,
0, 1) final state
State Cycle
Physical Significance
REFERENCES
James E. Palmer and David E. Perlman (1993). Schaum's Outline of Theory and Problems of Introduction to Digital Systems, McGraw-
Hill Professional. ISBN 0070484392
Stuart A. Kauffman (1993). The Origins of Order Self-Organization and Selection in Evolution, Oxford University Press. ISBN
0195079515
CONTRIBUTORS AND ATTRIBUTIONS

Joseph Casler, Andry Haryanto, Seth Kahle and Weiyin Xu, Adhi Paisoseputra, Andrew Kim, Hillary Kast, Stephanie Cleto
5.1: Boolean Models - Truth Tables and State Transition Diagrams is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter
request.
5.2: LOGICAL CONTROL PROGRAMS - IF... THEN… WHILE…
5.2.1: INTRODUCTION
A logical control program is a set of conditional statements describing the response of a controller to different inputs. A controller is a
computer used to automate industrial processes. Process engineers use control logic to tell the controller in a process how to react to all
inputs from sensors with an appropriate response to maintain normal functioning of the process. Control logic (sometimes called process
logic) is based on simple logic principles governed by statements such as IF X, THEN Y, ELSE Z yet can be used to describe a wide range
of complex relationships in a process. Although different controllers and processes use different programming languages, the concepts of
control logic apply and the conditions expressed in a logical control program can be adapted to any language.
The concepts behind logical control programs are not only found in chemical processes; in fact, control logic is used in everyday life. For
example, a person may regulate his/her own body temperature and comfort level using the following conditional logic statements: IF the
temperature is slightly too warm, THEN turn on a fan; IF the temperature is way too warm, THEN turn on the air conditioning; IF the
temperature is slightly too cold, THEN put on a sweatshirt; IF the temperature is way too cold, THEN turn on the fireplace. The person
takes an input from the environment (temperature) and if it meets a certain prescribed condition, she executes an action to keep herself
comfortable. Similarly, chemical processes evaluate input values from the process against set values to determine the necessary actions to
keep the process running smoothly and safely (aforementioned example illustrated below).
The following sections elaborate on the construction of conditional logic statements and give examples of developing logical control
programs for chemical processes.
5.2.2: LOGIC CONTROLS

Logical controls (IF, THEN, ELSE, and WHILE) compare a value from a sensor to a set standard for the value to evaluate the variable as
True/False in order to dictate an appropriate response for the physical system. The control program for a chemical process contains many
statements describing the responses of valves, pumps, and other equipment to sensors such as flow and temperature sensors. The responses
described by the system can be discrete, such as an on/off switch, or can be continuous, such as opening a valve between 0 and 100%. The
goal of a control program is to maintain the values monitored by the sensors at an acceptable level for process operation considering factors
like product quality, safety, and physical limitations of the equipment. In addition to describing the normal activity of the process, a control
program also describes how the process will initialize at the start of each day and how the controller will respond to an emergency outside
of the normal operating conditions of the system. Unlike a linear computer program, logic programs are continuously monitoring and
responding without a specific order. Before constructing logical control programs, it is important to understand the conditional statements,
such as IF-THEN and WHIlE statements, that govern process logic.
The following logic controls found below are written in pseudocode. Pseudocode is a compact and informal way of writing computer
program algorithms. It is intended for human reading instead of machine reading, and does not require stringent syntax for people to
understand. Pseudocode is typically used for planning computer program development and to outline a program before actual coding occurs.
5.2.2.1: IF-THEN STATEMENTS

IF-THEN statements compare a value from a sensor to a set value and describe what should happen if the relationship holds. The IF-THEN
statement takes the form IF X, THEN Y where X and Y can be single variables or combinations of variables. For example, consider the
following statements 1 and 2. In statement 1, both X and Y are single variables whereas in statement 2, X is a combination of two variables.
The ability in conditional logic to combine different conditions makes it more flexible than incidence graphs, which can only describe
monotonic relationships between two variables (See incidence graphs). A monotonic relationship is one where if X is increasing, Y is
always decreasing or if X is increasing, Y is always increasing. For complex processes, it is important to be able to express non-monotonic
relationships.
1. IF T>200 C, THEN open V1
2. IF T> 200 C and P> 200 psi, THEN open V1.
Where T is Temperature, P is pressure, and V represents a valve.
In statement 1, if the temperature happens to be above 200 C, valve 1 will be opened.
In statement 2, if the temperature is above 200 C and the pressure is above 200 psi, then the valve will be opened.
Otherwise, no action will be taken on valve 1
If the conditions in the IF statement are met, the THEN statement is executed, and depending on the command, the physical system is acted
upon. Otherwise, no action is taken in response to the sensor input. In order to describe an alternate action if the IF condition does not hold
true, ELSE statements are necessary.
5.2.2.2: ELSE STATEMENTS

The simple form of an IF-THEN-ELSE statement is IF X, THEN Y, ELSE Z where again X, Y, and Z can be single variables or
combinations of variables (as explained in the IF-THEN section above). The variable(s) in the ELSE statement are executed if the
conditions in the IF statement are not true. This statement works similar to the IF-THEN statements, in that the statements are processed in
order. The ELSE statement is referred to last, and is a condition that is often specified to keep the program running. An example is the
following:
IF P>200 psi, THEN close V1
ELSE open V4
In this statement, if the pressure happens to be 200psi or less, the THEN statement will be skipped and the ELSE statement will be
executed, opening valve 4.
Sometimes, if X, Y or Z represent many variables and several AND or OR statements are used, a WHILE statement may be employed.
5.2.2.3: CASE STATEMENTS

CASE statement is an alternative syntax that can be cleaner than many IF..THEN and ELSE statements. The example shown in the table
below shows its importance.
CASE IF..THEN and ELSE
IF T>Tset+1:
T>Tset+2: IF T>Tset+2:
v2=v2+0.1 v2=v2+0.1
T>Tset+1: ELSE:
v2=v2+0.05 v2=v2+0.05
T<Tset-1: ELSE IF T<Tset-1:
v2=v2-0.05 IF T<Tset-2:
T<Tset-2: v2=v2-0.1
v2=v2-0.1 ELSE:
v2=v2-0.05
Thus CASE statements make the code easier to read for debugging.
5.2.2.4: WHILE STATEMENTS
The WHILE condition is used to compare a variable to a range of values. The WHILE statement is used in place of a statement of the form
(IF A>B AND IF A<C). WHILE statements simplify the control program by eliminating several IF-AND statements. It is often useful when
modeling systems that must operate within a certain range of temperatures or pressures. Using a WHILE statement can allow you to
incorporate an alarm or a shut down signal should the process reach unstable conditions, such as the limits of the range that the WHILE
statement operates under. A simple example illustrating the use of the WHILE statement is shown below.
Example:
A tank that is initially empty needs to be filled with 1000 gallons of water 500 seconds after the process has been started-up. The water flow
rate is exactly 1 gallon/second if V1 is completely open and V1 controls the flow of water into the tank.
Using a IF...THEN statement the program could be written as follows:

IF t > 500 and t < 1501 THEN set V1 to open
ELSE set V1 to close.
The WHILE statement used to describe this relationship is as follows:
WHILE 500 < t < 1501 set V1 to open

ELSE set V1 to close.
It may not seem like much of a change between the two forms of the code. However, this is a very simple model. If modeling a process with
multiple variables you could need many IF...THEN statements to write the code when a single WHILE condition could replace it.
Example:
V1 controls reactants entering a reactor that can only run safely if the temperature is under 500K.
The WHILE can be used to control the process as follows:

WHILE T < 500 set V1 to open
ELSE set V1 to close. ALARM.
This example shows how a WHILE condition can be used as a safety measure to prevent a process from becoming unstable or unsafe.
In addition to lists of IF-THEN-ELSE-WHILE statements, control logic can be alternately represented by truth tables and state transition
diagrams. Truth tables show all the possible states of a model governed by conditional statements and state transition diagrams represent
truth tables graphically. Oftentimes, these are used in conjunction with booleans, variables that can only have two values, TRUE OR
FALSE. A Boolean model or Boolean function follows the format of IF-THEN statements described here. A description of Boolean models,
truth tables, and state transition diagrams is given here.
5.2.2.5: GO TO STATEMENTS
The GO TO statement helps to break out of current run to go to a different configuration. It can be an important operator in logical
programming because a lot of common functions are accessed using GO TO. However, many programmers feel that GO TO statements
should not be used in programming since it adds an extra and often unnecessary level of complex that can make the code unreadable and
hard to analyze. Even though the GO TO operator has its downsides, it is still an important operator since it can make help to simplify basic
functions. It can simplify code by allowing for a function, such as a fail safe, be referenced multiple times with out having to rewrite the
function every time it is called. The GO TO operator is also important because even advanced languages that do not have a GO TO function
often have a different operator that functions in a similar manner but with limitations. For example, the C, C++ and java languages each
have functions break and continue which are similar to the GO TO operator. Break is a function that allows the program to exit a loop
before it reaches completion, while the continue function returns control to the loop without executing the code after the continue command.
A function is a part of code within a larger program, which performs a specific task and is relatively independent of the remaining code.
Some examples of functions are as follows:
FUNCTION INITIALIZE: It runs at the beginning of a process to make sure all the valves and motor are in correct position. The operation
of this function could be to close all valves, reset counters and timers, turn of motors and turn off heaters.
FUNCTION PROGRAM: It is the main run of the process.
FUNCTION FAIL SAFE: It runs only when an emergency situation arises. The operation of this function could be to open or close valves
to stop the system, quench reactions via cooling, dilution, mixing or other method.
FUNCTION SHUTDOWN: It is run at the end of the process in order to shutdown.
FUNCTION IDLE: It is run to power down process.
All the functions mentioned above except FUNCTION IDLE are used in all chemical processes.
5.2.2.6: ALARM STATEMENTS

The ALARM statement is used to caution the operators in case a problem arises in the process. Alarms may not be sufficient danger to shut
down the process, but requires outside attention. Some example when ALARM statements are used in the process are as follows:
If the storage tank of a reactant is low, then ALARM.
If the pressure of a reactor is low, then ALARM.
If no flow is detected even when the valve is open, then ALARM.
If the temperature of the reactor is low even after heating, then ALARM.
If redundant sensors disagree, then ALARM.
In conclusion ALARM functions are very important in order to run a process safely.
5.2.3: CONTROL LANGUAGE IN INDUSTRY

As stated before, these commands are all in pseudocode, and not specific to any programming language. Once the general structure of the
controller is determined, the pseudocode can be coded into a specific programming language. Although there are many proprietary
languages in industry, some popular ones are:
Visual Basic
C++
Database programming (ex. Structured Query Language/SQL)
Pascal
Fortran
Pascal and Fortran are older languages that many newer languages are based on, but they are still used with some controllers, especially in
older plants. Any experience with different computer languages is a definite plus in industry, and some chemical engineers make the
transition into advanced controls designing, writing, and implementing code to make sure a plant keeps running smoothly.
5.2.4: LOGICAL FUNCTIONS IN MICROSOFT EXCEL

Microsoft Excel has basic logical tools to help in constructing simple logical statements and if needed more complex logical systems. A list
of the functions is shown below.
TRUE().............................................Returns the logical value, TRUE.
FALSE()...........................................Returns the logical value, FALSE.
AND(logical_expression_A,B,C).........Returns TRUE if all the expressions are true.
OR(logical_expression_A,B,C)...........Returns TRUE if one of the expressions are true.
NOT(logical_expression)....................Returns the opposite of the expected logical value. If the expression is TRUE, it will return FALSE
IFERROR(value,value_if_error)............Returns the value unless there is an error in which it will return the value_if_error.
IF(logical_expression,value_if_true,value_if_false)..............Checks the validity of the expression and returns TRUE or FALSE likewise.
The IF() function will be most useful in logical programing. It is essentially an IF THEN or IF ELSE function returning one of two values
based on the logical expression it is testing. Excel also allows logical functions within functions. This allows for logical expressions
involving more than one IF statement within itself for example. Use of these tools will be practical in quickly setting up control programs
and other systems in Microsoft Excel.
5.2.5: CONSTRUCTING A LOGICAL CONTROL PROGRAM

Understanding the conditional statements used in control logic is the first step in constructing a logical control program. The second step is
developing a thorough understanding of the process to be controlled. Knowledge of the equipment, piping, and instrumentation (contained
on a P&ID diagram), operating conditions, chemical compounds used, and safety concerns is necessary. Particularly it is important to know
the measured and controlled variables. For example, the pressure limits of a tank must be known in order to develop a control plan to ensure
safety; ignoring this constraint could lead to explosion and injury. Once the necessary controls are known, one can develop a plan using the
logical statements described previously. The third step is constructing a logical control program is understanding that there is not always a
right answer, meaning there are many different ways to ensure the same desired outcome.
Worked out Examples 1, 2, and 3 demonstrate the construction of simple logical control programs. The more complex the situation, the
longer the logical control plan becomes yet the process is still the same. An example of a more complex logical control program is given
here.This example is from the 2005 ChE 466 class at the University of Michigan and describes an entire chemical process from the delivery
of raw materials to the output of the final product.
5.2.6: DETERMINING FAIL SAFE CONDITIONS

Fail safe is the practice of designing a system to default to safe conditions if anything or everything goes wrong. The goals of fail safe
conditions are to:
Protect plant personal
Protect the local community around the plant
Protect the environment
Protect plant equipment
In order to establish safe conditions, fail safe programs must specify the desired positions of all valves and status of all motors and
controlled equipment. For example, in an exothermic reactor, fail safe conditions would specify opening all cooling water valves, closing all
feed valves, shutting off feed pump motors, turning on agitator motor, and open all vent valves.
Control programs frequently define fail safe conditions at the beginning of the program. These conditions are then activated using a GO TO
command when process conditions exceed the maximum or fall below the minimum allowable values.
All processes must be evaluated for conditions that could cause hazards and fail safe procedure must be designed counteract the effects.
 EXAMPLE 5.2.1: REBOILER
Reboilers are used in industry to cool down process streams by creating steam from water. This chemical process involves a phase
change from liquid (water) to gas (steam) and it is important to monitor flowrates, temperatures, and pressures. Below is a diagram of a
reboiler. The controlled variables are F1, F2, and F3; these are controlled by manipulating the corresponding valves. The measured
variables are P1, T1, and T2. Considering the operating conditions and constraints given below, write a logical control program for the
reboiler.
The operating constraints are as follows:

T1 must not exceed 350 ˚C
T2 must be between 100 and 200 ˚C
P1 cannot exceed 150 psi
The normal operating conditions for the controlled variables are as follows:
F1 is 20 gal/min
F2 is 10 gal/min
F3 is closed
Solution
The first step is understanding the system to be controlled. Looking at the diagram, one can see that F1 controls the flow rate of the
service water, F2 controls the flow rate of the process stream, and F3 controls the vent to the atmosphere. T1 is the temperature of the
process stream entering the reboiler and T2 is the temperature of the process stream exiting the reboiler. P1 is the pressure inside the
reboiler.
Below is a possible control program to ensure the operating constraints are met; there may be other solutions to achieve the same
objective:
IF T2 > 200 ˚C, THEN F1 = 30 gal/min
IF T2< 100 ˚C, THEN F1 = 10 gal/min
IF P1 > 150 psi, THEN open F3
IF T1 > 350 ˚C, THEN F2 = 2 gal/min
In order to control the exit temperature of the process stream, one can increase or decrease the service water from the normal flow rate.
Opening the vent to the atmosphere reduces the pressure if it reaches an unsafe value. If the entering temperature is too high for the
process stream, reducing the entering flow rate will ensure the exit temperature is not too high for the remainder of the process. Note:
reducing the process stream may negatively impact the rest of the plant.
 EXAMPLE 5.2.2: THERMOSTAT
Chemical engineers are increasingly involved in biological applications. In many biological processes, temperature control can be very
critical to the process. Take, for example, a simulated cell on a computer chip. Let's say that it is imperative for the chip to remain at
97oF ± 1oF, very similar to the temperature of the human body. Some of the reactions and processes on the chip will not function at
temperatures outside this range and can become irreversibly damaged unless the power supply to the chip is turned off. The power is
turned on and off by a switch S1. There are flows of cool water with an automatic valve V1 and warm water with an automatic valve
V2 for cooling and heating purposes. The chip is attached to a thermostat to sense and control the temperature, T. Write out a controlled
logic scheme to maintain the chip temperature and to prevent damage.
Hint: Heat or cool the chip before reaching the limits (about half a degree)
Solution
To control the temperature, the chip must be heated or cooled depending on the environment
IF (T<96.5) THEN V2 is open
ELSE V2 is closed
IF (T>97.5) THEN V1 is open
ELSE V1 is closed
The control is executed before the temperature limit is reached to allow for lag time in the heating or cooling water flow.
Also, to ensure that the chip is not damaged, the power needs to be shut off if the temperature goes above 98oF or below 96oF
WHILE (96<T<98) THEN S1 is on
ELSE S1 is off
 EXAMPLE 5.2.3: CHEMICAL REACTOR
There is an exothermic chemical reaction occurring in a CSTR (Continuous Stirred Tank Reactor) that involves two reactants being fed
at a 1:1 ratio. All valves are set to be 50% open normally. Write a control program that keeps the level in the CSTR vessel less than 8
meters (the tank is 10 meters tall) and the reactor temperature below 450 degrees Celsius.
Solution
While L1 > 8 set V3 to 100% open and close V1 and V2
Else set V1 and V2 and V3 to 50% open
If T2 > 450 THEN set V5 and V4 to 100% open
Else set V5 and V4 to 50% open
This solution gives an example of using AND statements to control multiple valves with just one condition.
 EXAMPLE 5.2.4: PROGRAMMING AND ALARMS

There is a process that is run by the P&ID shown below.
Based on this process and the steps listed below, write out a detailed control program for the process. Use comments (denoted with #) if
necessary to explain the logic behind the steps.
1. Measure Qw units of water into a tank
2. Add Qc units of dried chickpeas
3. Let dried chickpeas soak for 20 hours without mixing
4. Drain off soaking water to waste (assume the filter in the tank will not allow whole chickpeas through the pump) 5) Add Qw units
of fresh water to the tank.
5. Heat the tank to Tcook and maintain the pressure at 4 atm. Note that your tank is rated to withstand pressures between 0.5 and 6
atm, while outside of that range the tank may implode or explode.
6. Cook chickpeas for 20 minutes.
7. After cooking, turn off heat and allow the system to return to ambient temperature (Tamb) and ambient pressure. Beware of a strong
vacuum forming in the tank as the water vapor condenses!
8. Drain cooking water to drain.
9. Pump in Qs units of the tahini spice mix
10. Blend the mixture for 10 minutes to produce a smooth hummus paste.
11. Pump out product to packaging.
12. Fill tank with clean water and agitate to clean the reactor.
13. Pump wash water to drain.
Solution
FUNCTION INITIALIZE
Turn off M1, M2, M3, M4
Close V1, V2, V3, V5, V6, V7, SV1
Set all timers to zero
Set all totalizers to zero
FUNCTION FAILSAFE
Turn off M1, M2, M3, M4
Close V1, V2, V3, V5, V7, SV1
Open V6
FUNCTION PROGRAM
#Step 1 – Measure Qw unites of water into a tank
Turn on M1
Open V1
WHILE FC1tot < Qw:
Adjust V1 to FC1set
IF LC1 < LC1min:
ALARM
Close V1
Turn off M1
#FC1tot is the total amount of fluid that has gone through the flow meter
#FC1set is the set point (amount the valve is open) for V1 that FC1 has already programmed into it
#LC1min is the minimum acceptable level of fluid in S001
#Step 2 – Add Qc units of dried chickpeas
Open SV1
WHILE FC4 < Qc:
Adjust SV1 to FC4set
IF LC4 < LC4min:
ALARM
Close SV1
#FC4set is the set point for SV1 that FC4 has already programmed into it
#LC4min is the minimum acceptable level of fluid in S003. LC4 is not on the P&ID, however, it makes sense to have one on it so the
level on the tank can be properly monitored
#Step 3 – Let dried chickpeas soak for 20 hours without mixing
WAIT 20 hours
#Step 4 – Drain off soaking water to waste
Open V7
Turn on M3
WHILE FC3tot < Qw:
Adjust V7 to FC3set2
Turn off M3
Close V7
#FC3set2 is the set point for V7 that FC3 has already programmed into it
#Step 5 – Add Qw units of fresh water to the tank
Clear FC1tot
Turn on M1
Open V1
WHILE FC1tot < Qw:
Adjust V1 to FC1set
IF LC1 < LC1min:
ALARM
Close V1
Turn off M1
#Step 6 – Heat the tank to Tcook and maintain the pressure at 4 atm.
WHILE TC1 < Tcook:
Adjust v5 to Tcook
IF OR (PC1 < 0.5, PC1 > 6):
GO TO FAILSAFE
IF PC1 < PC1set:
Close V6
ELSE:
Adjust V6 to PC1set
IF LC3 > LC3max:
GO TO FAILSAFE
#PC1set is the setting that V6 must be set to for the tank to have 4 atm of pressure in it
#LC3max is the maximum level that the contents of the tank are allowed to get to. Anything higher indicates a problem with one of the
flow meters.
#Step 7 – Cook chickpeas for 20 minutes#
WAIT 20 minutes
#Step 8 –After cooking, turn off heat and allow the system to return to ambient temperature and pressure
IF PC1 > PC1amb:
Adjust V6 to PC1set2
IF OR (PC1 < 0.5, PC1 > 6):
GO TO FAILSAFE
WHILE TC1 > Tamb:
Close V5
#PC1amb is the ambient pressure of 1 atm that the system needs to get to
#PC1set2 is the second setting on PC1 which effects how open V6 is
#Step 9 – Drain cooking water to drain
Clear FC3tot
Open V7
Turn on M3
WHILE FC3tot < Qw:
Adjust V7 to FC3set
Turn off M3
Close V7
#Step 10 – Pump in Qs units of the Tahini spice mix
Clear FC2tot
Turn on M2
Open V2
WHILE FC2tot < Qs:
Adjust V2 to FC2set
IF LC2 < LC2min:
ALARM
Close V2
Turn off M2
#FC2set is the set point for V2 that FC2 has already programmed into it
#Step 11 – Blend the mixture for 10 minutes to produce a smooth hummus paste
Turn on M4
WAIT 10 minutes
Turn off M4
#Step 12 – Pump out product to packaging
Open V3
Turn on M3
WHILE LC3 > 0:
Adjust V3 to FC3set
Turn off M3
Close V3
#FC3set is the set point for V3 that FC3 has already programmed into it
#Step 13 – Fill tank with clean water and agitate to clean the reactor
Turn on M1
Open V1
WHILE LC3 < LC3max:
Adjust V1 to FC1set
IF LC1 < LC1min:
ALARM
Close V1
Turn off M3
Turn on M4
WAIT 10 minutes
Turn off M4
#Step 14 – Pump was water to drain
Open V7
Turn on M3
WHILE LC3 > 0:
Adjust V7 to FC3set2
Turn off M3
Close V7
5.2.7: WORKED OUT EXAMPLE 4: PROGRAMMING AND ALARMS

5.2.8: WORKED OUT EXAMPLE 5: ANOTHER CHEMICAL REACTOR (TAKEN FROM PROF. BARKEL'S
9/29/09 LECTURE)
WRITING THE CONTROL PROGRAM

Nomenclature:
xA = input amount of A
yB = input amount of B
Readings:
LL = low level
LH = high level
Lmax = maximum high level
Lmax = maximum low level
Things to note:
Be sure to use If, Then statements.
Lines in italics are comments used to organize the program.
CONTROL PROGRAM
Equal means within +/- 0.05 % of the value being held equal.
Initialize This is used to set all control devices to the position you want them in.
1. Turn off all valves
2. Shut off all motors
Fail Safe This is used to shut down the process in case something goes wrong and must be terminated immediately.
1. Turn off all valves except V5, V7, V9, V6
2. Shut off all motors except M2
Universal Statements
If T1>Tmax, then go Fail Safe.
1. Initialize
2. Open V6, V9, V1. [so that systems doesn't build up pressure]
3. If L1>LL, then turn on M1. [fill with A]
4. If L2>=Lag, turn on M2. [start stirrer]
5. If F1>=xA, turn off V1, shut off M1. [stop flow of A]
6. Open V5, V7. [allow for cooling]
7. If P1>=PL, then open V2. [fill with B]
8. If L4>=L4min, open V8, turn on M4. [release product AB]
9. If L2<Lag, then shut off M2. [so motor doesn't run dry]
10. If L4<L4L, close V8. [stop outflow of product AB]
11. If T1><TH, then open V5. [cool reactor temperature]
12. If T2><TH, then open V5. [cool upper product stream]
13. If T1<<TL, then close V5. [stop cooling of reactor]
14. If F2>=<yB, then turn off V2. [stop flow of B]
15. Close V8, M4, leave V6, V9 open. [shut down process but allow it to vent]
16. Pump out BAB.
A + B ⟶ AB
The reaction is exothermic

A, B, AB are all liquids
It goes to 100% completion
XA = total amount of A used
YB = total amount of B used
B is added slowly into a full charge of A
The temperature is maintained at TR
Upon completion of the reaction, the AB is cooled to TP
ZC, an amount of solvent C is added to the AB to make the final product
Using only the equipment and instrumentation shown, write the control logic for this BATCH reaction. Use If, Then logic.
Use of subscripts:
Ag for agitator
L for low-bottom of control range
H for high-top of control range
Min for minimum - lowest level allowed
Max for maximum - highest level allowed
tot for totalizer
Process Control Logic

Equal means within +/- 0.05% of the value being held equal.
1) Initialize
Shut all valves
Turn off all motors
Set all FCtot = 0
2) Fail safe
Shut off V1, V2, V3, V4
Open V5
Turn off M1, M2, M3
Turn off M4, Open V4
Turn on M5 if LC4>Lag, else turn off M5
3) If T1>Tmax, then go Fail Safe
4) If L2>L2min, then turn on M2, open V2
5) If L2<L2min, then turn off M2, shut off V2
6) If FC2tot = XA, then turn off M2, shut off V2
7) If L4>Lag, then turn on M5
8) If L1>L1max, then go Fail Safe
10) If L1<L1min,then turn off M1, shut off V1
11) If FC1tot = YB, then turn off M1, shut off V1
12) If T1>T1H, then open V5
13) If T1<T1L, then close V5
14) If L4>L4H, then close V1, turn off M1
15) If T1>TP, then open V5
16) If T1=TP, then close V5
19) If FC3tot = ZC, then turn off M3, shut off V3
20) If LC4<Lag, then turn off M5
21) Open V4
22) Turn on M4
5.2.10: REFERENCES
Savitch, Walter. "Problem Solving with C++ The Object of Programing". Boston: Pearson Education, Inc. 2005.
Stanford Encyclopedia of Philosophy
Woolf, Peter. "Possible Useful Notes", Notes for Boolean Logic, September 22, 2005.
Woolf, Peter. "Project 1 Sample Solution", October 6, 2005.
Woolf, Peter. "Useful Definitions", Terms Review, October 11, 2005.

Authors: Stephanie Fraley, Michael Oom, Benjamin Terrien, John Zalewski
Stewards: Ross Bredeweg, Jessica Morga, Ryan Sekol, Ryan Wong
5.2: Logical Control Programs - IF... THEN… WHILE… is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et
CHAPTER OVERVIEW
6: MODELING CASE STUDIES

6.1: Surge Tank Model
6.4: ODE and Excel CSTR model with heat exchange
6.5: ODE and Excel model of a Simple Distillation Column
6.6: ODE and Excel model of a Heat Exchanger
6.8: Modeling and PID Controller Example - Cruise Control for an Electric Vehicle
6: Modeling Case Studies is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source content that was
1
6.1: SURGE TANK MODEL
6.1.1: INTRODUCTION
Used to regulate fluid levels in systems, surge tanks act as standpipe or storage reservoirs that store and supply excess fluid. In a system that
has experienced a surge of fluid, surge tanks can modify fluctuations in flow rate, composition, temperature, or pressure. Typically, these
tanks (or “surge drums”) are located downstream from closed aqueducts or feeders for water wheels. Depending upon its placement, a surge
tank can reduce the pressure and volume of liquid, thereby reducing velocity. Therefore, a surge tank acts as a level and pressure control
within the entire system.
Since the flow to the surge tank is unregulated and the fluid that is output from the surge tank is pumped out, the system can be labeled as
unsteady-state [MIT], but the approach to an approximate solution (below) utilizes techniques commonly adhered to when solving similar
steady-state problems.
The technology behind surge tanks has been used for decades, but researchers have had difficulty fully finding a solution due to the non-
linear nature of the governing equations. Early approximations involving surge tanks used graphical and arithmetical means to propose a
solution, but with the evolution of computerized solving techniques, complete solutions can be obtained. [Wiley InterScience].
6.1.2: DERIVATION OF ORDINARY DIFFERENTIAL EQUATION

To accurately model a surge tank, mass and energy balances need to be considered across the tank. From these balances, we will be able to
develop relationships for various characteristics of the surge tank
6.1.2.1: DIAGRAM OF SURGE TANK SYSTEM
A surge tank relies on the level sensor to determine whether or not fluid stored in the tank should be removed. This regulated outflow is
pumped out in response to calculations made by a controller that ultimately opens and closes the control valve that releases the fluid from
the tank.
6.1.2.2: GOVERNING EQUATIONS OF SURGE TANK MODEL

A surge tank's components must be divided up and evaluated individually at first, then as a whole. First, the expression for the inlet stream
must be obtained. The simplistic sine function will be used as the basis for the expression of a stream because it typically describes the tidal
surge pattern of a low-viscous fluid, like water. The flowrate, w, will be given in units of kg h-1.
wi = a + b sin(cπt)
where a and b are constants determined by the specific problem circumstance.

A mass balance must now be performed on the tank as a system. Using the concept that mass in = mass out, and assuming that the tank is a
perfect cylinder devoid of diversions, an expression can be derived:
rate mass in − rate mass out = accumulated fluid in tank
Rewriting,
dV
wi − wo = ρ
dt
dh wi − wo
=
dt ρA
h t
wi − wo
∫ dh = ∫ dt
0 0 ρA
t
1
h(t) − ho = ∫ wi − wo dt
ρA 0
Where, at time t = 0, the amount of fluid in the surge tank is constant; thus h(0) = h . 0
Substituting the original equation for the inlet stream, wi, into the expression for the height of the tank, h(t), the governing equation for the
height of the tank is obtained:
t
1
h(t) = ho + ∫ [(a + b sin(cπt) − w0 ] dt
ρA 0
Integrating by parts,
x
h(t) = ho + (1 − cos(cπt))
ρA
where x is formed from the constants a and b during integration.
6.1.3: SECONDARY USES OF SURGE TANKS

Surge tanks are most commonly used to protect systems from changes in fluid levels; they act as a reservoir that stores and supply excess
fluid. In addition, the tanks shield the systems from dramatic changes in pressure, temperature, and concentration. They can also allow one
unit to be shut down for maintenance without shutting down the entire plant.
6.1.3.1: PRESSURE
There can be moments of high pressure, called hammer shock, in a system when the liquid (incompressible) flow is stopped and started. The
energy that liquids possess while traveling through pipes can be categorized as potential or kinetic. If the liquid is stopped quickly, the
momentum that the liquid carries must be redirected elsewhere. As a result, the pipes vibrate from the reactive force and weight of the shock
waves within the liquid. In extreme cases, pipes can burst, joints can develop leaks, and valves and meters can be damaged.
The extreme amounts of pressure are dampened when the fluid enters the surge tanks. The surge tank acts as a buffer to the system,
dispersing the pressure across a greater area. These tanks make it possible for a system to more safely execute their tasks.
6.1.3.2: TEMPERATURE
The temperature of a fluid can either be controlled or changed through the use of a surge tank. The surge tank allows for a rapid change in
fluid temperature. This is exemplified by the process of pasteurization; the milk needs to be at a high temperature for just a short period of
time, so it is exposed to the high temperature and then moved to the surge tank where it can be stored and cooled (see heated surge tank).
A substance can enter the surge tank at room temperature, and it will instantaneously mix with the rest of the tank. Substances entering the
tank will also subsequently rise to meet the high temperature and then exit the surge tank quickly thereafter.
6.1.3.3: CONCENTRATION
Concentration inside the surge tank is kept relatively constant, thus the fluid exiting the surge tank is the same as the fluid in the tank. This
is favorable when there is a concentration gradient in the incoming fluid to the surge tank. The tank homogenizes the entering fluid, keeping
the concentration of the reactants the same throughout the system, therefore eliminating any concentration gradient.
 EXAMPLE 6.1.1: MODELING SURGE TANK
Suppose we are to design a surge tank that deals with flow swings of +/- 40% over a 10 minute period modeled by the following
equation:
πt
w1 = 500 + 200 sin (6.1.1)
10
where flow is m3h-1 and time in hours. The outlet flow rate, wo, is desired to be 500 m3h-1.
The surge tank is 30m tall and has a cross sectional area of 5m2. The initial height of the fluid (ρ = kg / m3) is 10m.
a. Model this hypothetical example.
b. The surge tank has upper and lower limit controls that open or close the valve if the level gets too high or low. What is the highest
maximum set point that can be used without the surge tank overflowing?
Solution
a.
dh
ρA = wi − wo (6.1.2)
dt
with h(0) = h .o
t
1
h(t) = ho + ∫ (wi (t) − wo (t)) dt
ρA 0
Substituting Equation 6.1.1 into 6.1.2 , we get

2000 2πt
h(t) = ho + (1 − cos ) (6.1.3)
ρπA 20
To complete the design, we must have the cross-sectional area A of the surge tank, this would be given.
If we were to apply the following condition,
wi − wo = C (6.1.4)
and were to substitute Equation 6.1.4 into Equation 6.1.2 , we discover

C
h(t) = ho + t (6.1.5)
ρA
b.
Using the Microsoft Excel model provided, the highest possible maximum set point is 23m. Media:Model Surge Tank.xls
 EXAMPLE 6.1.2
In this example, we simulate random fluctuations that might be observed in a real world process stream. A random number generating
function (in MS Excel) is used to generate pseudo-random fluctuations of 200 kg/h about an average value of 1000 kg/h for the inlet
stream, wi(t) over a period of 5 hours. The result is as follows:
Examine the effect that these fluctuations have on the fluid level in surge tanks of various volumes. To vary the volume, assume tanks
of constant height hmax=20m, and vary the cross sectional area from A=1m2 to A=5m2. Use the following parameters for the surge tank:
Initial fluid level h0=10m, ρ=1 kg / m3, w0=1000 kg / m3, t0=0 h and tf=5h.
Solution
In the solution to the first example problem, we used a trigonometric function to simulate fluctuations in the inlet stream, and we
obtained an analytical solution to the differential equation. Since we are using a pseudo-random fluctuating inlet stream in this
example, we will solve this problem via numerical methods. Using the pseudo-random data for wi(t), we perform a numerical
integration on the previously derived expression:
t
1
h(t) = ho + ∫ (wi (t) − wo (t)) dt (6.1.6)
ρA 0
This integration was performed with the trapezoid rule (in MS Excel, using a slight modification of the posted Excel model), using the
specified surge tank parameters, with A=1 m2, 2 m2, and 5 m2 l. The following results were obtained:
We see that increasing the volume of the surge tank by increasing the cross sectional area A reduces the magnitude of the fluctuations
of the fluid level h in the surge tank. For A=1 m2 and 2m2, the capacity of the surge tank is exceeded and it overflows.
 EXAMPLE 6.1.3
An operator quickly adds 50 gallons from a drum of water into a cylindrical surge tank with a diameter of 4 feet. The initial volume of
liquid in the tank is 40 cubic feet and the total height of the tank is 5 feet. The inflow and exit flow rates are initally 6 cubic feet per
minute. The resistance has a linear relationship with liquid height.
a. Derive a first order linear model for the tank height and determine the value of R for the outflow
b. Will the tank overflow when the drum is added?
c. Show the height h(t) after the drum of water is added; treat the change in h(t) as instantaneous
d. Does the tank height return to steady state? What is it?
Solution
(a)
Assume the operator adds the water at t = 0, so when t < 0, the tank is under steady state which can be described as:qIN − qOUT = 0 (1)
Assume: where R is the resistance and h is the height of the liquid. Initial volume of the tank is:
Thus:
Substitute and into equation 1 above:

0 = 6 − hSS / R
At time t > 0, the 50 gallons (6.685 cubic feet) of water is added to the system which disrupts the steady state and the corresponding
initial condition of the system is:
V0 = VSS + Voperator = 40 + 6.685 = 46.685ft3
The corresponding dynamic linear model:
A\left(\frac{d h}{d t}\right)=q_{I N}-q_{O U T}=q_{I N}-\left(\frac{h}{R}\right) \label{ex2:2}
with initial condition: h

0 = 3.7151 ft
(b)
After adding the drum, h0 = 3.7151 which is lower than 5 feet. The system will not overflow at t0. Because the liquid level is higher
than the steady state level and , the out flow rate will be larger than qIN which pushes the liquid level drop. As a result,
water will not over flow.
(c)
Define deviation variables. h' = h − hSS ; qIN' = qIN − qINSS = 0 and substitute into equation (2)
\\left ( \frac{dh'}{dt} \right )=q_{IN}'-\left ( \frac{h'}{R} \right )=-\left ( \frac{h'}{R} \right ) \nonumber
(3)
Solve the linear model above (Eq. 3): (Either by direct integration or laplace transform and inverse lapace transform)
(4)
Put h' = h − hSS into equation 4:
Then:
(d)
Yes. Since
lim h = 3.183 + 0.5321 exp(−0.1499t) = 3.183f t

t→∞
which was the original steady state

If you are a sage for this page, please link your narrated powerpoint presentation here.
6.1.5: REFERENCES
Kundur, Prabha. Power System Stability and Control, McGraw-Hill 1994.
Cheng-Nan Lin and John S. Gladwell. Non-dimensional Surge Tank Analysis, State of Washington Water Research Center, Washington
State University 1970.
Slocum, Stephen Elmer. Elements of Hydraulics, McGraw-Hill 1915.
MIT OpenCourseWare. www.ocw.cn/OcwWeb/Chemical-Engineering/10-450Process-Dynamics--Operations--and-
ControlSpring2003/LectureNotes/index.htm. Design of a surge tank to smooth out fluctuations in flow. Definition of important process
control terms

Authors: (September 6, 2007) Cara Canady, David Carpenter, Che Martinez, Jeremy Minty, Bradley Novak
6.1: Surge Tank Model is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source content that was
6.2: HEATED SURGE TANK
6.2.1: INTRODUCTION
A surge tank is an additional safety or storage tank that provides additional product or material storage in case it becomes needed. Heat
exchange can be added to surge tanks, which provides temperature control for the tank. Within a system these tanks can appear as
distillation columns, reboilers, heated CSTR’s, and heated storage. They can increase production rates by allowing a batch of product to
finish reacting while the initial tank is reloaded, provide constant system parameters during start up and shut down, or create additional
storage space for product overflow or backup material.
Uses for Heated Surge Tanks:
Fuel surges caused by motion of a vehicle: If fuel cannot be drawn from the primary tank, the engine resorts to a surge tank. The heat
maintains the fuel’s temperature.
Caramelization: During the formation of caramel, the mixture must be maintained at a specific temperature for a predetermined amount
of time. Once the ingredients are thoroughly dissolved, the mixture is transferred to a heated surge tank and maintained until the caramel
has thickened and is ready to be drawn out.
Mixing of gases: Bulk gas lines can be connected to a heated surge tank with a pressure sensor. The pressure sensor would control the
temperature. By heating the gas when it first enters the tank, there is no risk of explosion later due to expansion.
Heated pools: Surge tanks are used to catch and store displaced water from a pool. If the pool is heated, a heated surge tank should be
used to maintain the temperature of the water.
De-aeration: Heated surge tanks are often used with de-aerators. They heat the component that will enter the de-aerator, because if the
component is not preheated, the de-aerator must wait until the component reaches the correct temperature. This could waste a lot of time
and energy.
Chemical Baths: Often in industry, things need to be treated with a chemical bath. The chemicals usually need to be at a certain
temperature so that it will adhere to the object. A heated surge tank is perfect for this application.
Reboilers: Liquids coming off of a distillation column can be reheated to enter the column again at a higher temperature to drive the
separation process. Many industries use this tool to obtain a more efficient separation and produce a higher net profit.
Product or Material Backup: Heated surge tanks can also be used as simple storage in two ways. First, a surge tank can be used excess
product not yet sold or otherwise moved out of the production system. Second, heated surge tanks can serve as backup for chemical or
fuel supplies to a production plant, such as outdoor gasoline tanks for a backup generator in case of power failure.
6.2.2: BASIC DESIGN FOR HEATED SURGE TANKS

Above is a basic example of a heated surge tank. While surge tanks can have multiple inputs and outputs, for simplicity we have only
included one of each here.
Connected to the tank is a temperature control, which controls the heater. Depending on the temperature of the fluid, this control will
increase or decrease the heating to the tank. This will keep the fluid at the necessary temperature to meet the process requirements.
There is also a level control connected to the tank to indicate when the tank has neared maximum capacity. When this happens, the control
will open the valve at the bottom of the tank, allowing the product to flow further down the process. The control can also slow or stop the
flow coming through the input valve. These mechanisms will prevent the tank from overfilling. The position of the level control depends on
the type of material in the process, the phase of the material, the type of level control, and the requirements of the system.
6.2.3: USEFUL EQUATIONS

The basic equations that govern heated surge tanks are shown below. First, a simple mass balance is done on the system. Second, the energy
balance was simplified using the assumptions listed below. Most problems involving this type of tank can be described by these equations.
Additional considerations may require additional variables and equations.
Assumptions:
1. The substance coming into the tank is uniform.
2. No reaction is taking place.
3. The tank is well mixed, which means the temperature profile is constant throughout the tank.
6.2.3.1: MASS BALANCE
Since there is no generation from reactions inside the heated surge tank, we obtain the rate of accumulation or level inside the tank by
subtracting what is coming out from what is coming in.
RateofAccumulation = (Flow In) − (Flow Out)
6.2.3.2: ENERGY BALANCE

The temperature change with respect to time is essential for the purpose of configuring a system to reach steady state. When turning a
system on or off, there is a time period in which the system is in unsteady state. During this time, the system is difficult to model. In steady
state, the system is easier to model because once steady state is reached the left hand term will become zero.
dT vρCp (T0 − T ) + U A (TC − T )
=
dt V ρCp
6.2.4: CASE STUDY - WATER PURIFICATION AT IBM
At IBM’s manufacturing facility outside Burlington, Vermont, a heated surge tank is used in the de-ionized water system. In order to wash
semi-conductor wafers in manufacturing, the water has to be about 1,000,000 times cleaner than the incoming city water. All of this
purification is done on site.
The water comes in from the municipal water source at a constant flow rate, but manufacturing demand is not constant. In order to
compensate for this, when the demand in manufacturing is low, a surge tank is used to store extra water for high demand periods. Because
the large tank is located outside and the winter in Vermont is very cold, the tank is heated to prevent the water inside from freezing.
During normal operation of the system, the surge tank is bypassed. When a flow controller downstream has low demand, the inlet valve
opens, letting water into the surge tank. A level controller monitors the tank to make sure it doesn’t overfill and can shut off the inlet valve
and let water out. A temperature controller controls the heater jacket to maintain the water around 50°C. When the demand for water
increases, the flow controller near the outlet can shut off the inlet valve to the tank, and/or further open the outlet valve to access the extra
water supply in the tank.
 EXAMPLE 6.2.1: DETERMINING TEMPERATURE OF HEATING FLUID FOR TANK
A heated surge tank is being designed to hold paraffin wax coming from a distillation column at an oil refinery. High pressure steam
will be used as a heating fluid in the heat exchanger to heat up and maintain the paraffin at 51°C (to maintain high viscosity and prevent
solidification). The physical parameters of the tank (volume of 5 m3) and heat exchanger within it are given. The tank is originally
filled with paraffin at room temperature. At what temperature must the high pressure steam be to sufficiently heat the paraffin; will a
proposed temperature of 130°C be adequate?
The paraffin comes into the tank at 37°C at a volumetric flow rate of 0.0005 m3/s. The heat exchange coefficient is equal to 50 W/m2/K
and the area of the heat exchanger is 2.0 2. The heat capacity of the paraffin is 2900 J/Kg/K and the density is 800 Kg/m3.
Solution
POLYMATH CODE:
− vin ∗ ρ ∗ Cp ∗ (T − Tin + U ∗ A ∗ (Ta − T )
\dfrcad(T )d(t) =
Cp ∗ V ∗ ρ
vin= 0.0005 m3/s

U= 50 W/m2/K
A= 2 m2
Ta= 130°C= 403K
Cp= 2900 J/Kg/K
rho= 800 Kg/m3
Tin= 37°C= 310K
V= 5 m3
t(0)= 0 s
T(0)= 298K
t(f)= 36000 s
With a heating fluid at a temperature of 130°C, the fluid only reaches a temperature of 44°C (317K). A higher capacity heating fluid
must be used.
Trying a heating fluid at 277°C, we generate the plot below.
The plot shows that the tank will reach a temperature of about 55.5°C (328.5K) with a heating fluid at 277°C. This will be sufficient to
maintain the paraffin in the liquid phase.
 EXAMPLE 6.2.2: TIME TO REACH STEADY STATE AFTER HEATING FAILURE
For the same surge tank from problem 1, if the heater fails for 2 hours after 10 hours of operation and is then restarted, how long will it
take after it is restarted to reach steady state?
Solution
Approximately 20 hours
The first manipulation that must be done to your Polymath code is to create an "if-then-else" statement for the dT/dt line for the case
before hour 10, the case between hour 10 and hour 12, and the case after hour 12. For the time ranging between 10 and 12 hours,
because the heating element has failed, the differential equation over this period must reflect that. Thus, the section (U ∗ A ∗ (T a − T) )
from the differential equation is dropped and the equation appears as follows:
− vin ∗ ρ ∗ Cp ∗ (T − Tin )
d(T )/d(t) = .
Cp ∗ V ∗ ρ
For any other time in this simulation, the normal dT/dt equation is used. In order to determine where the surge tank reaches steady state
again, the final time is increased until dT/dt approximately reaches zero. The graph generated in Polymath will look like the following,
using the code below as a parameter:
\[ \dfrac{d(T)}{d(t)= if(36000<t and t<42500) \nonumber \]
then
(− vin ∗ ρ ∗ Cp ∗ (T − Tin )/(Cp ∗ V ∗ ρ))else((−vin ∗ rho ∗ Cp ∗ (T − Tin ) + (U ∗ A ∗ (Ta − T )))
/(Cp ∗ V ∗ rho))

Authors: Angela Antosiewicz, Christopher Kline, Peter Heisler, Paul Niezguski
Stewards: Karen Staubach, Soo Kim, Kerry Braxton-Andrew, Joshua Katzenstein
6.2: Heated Surge Tank is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source content that was
6.3: BACTERIAL CHEMOSTAT
6.3.1: INTRODUCTION
Bioreactors are used to grow, harvest, and maintain desired cells in a controlled manner. These cells grow and replicate in the presence of a
suitable environment with media supplying the essential nutrients for growth. Cells grown in these bioreactors are collected in order to
enzymatically catalyze the synthesis of valuable products or alter the existing structure of a substrate rendering it useful. Other bioreactors
are used to grow and maintain various types of tissue cultures. Process control systems must be used to optimize the product output while
sustaining the delicate conditions required for life. These include, but are not limited to, temperature, oxygen levels (for aerobic processes),
pH, substrate flowrate, and pressure. A bacterial chemostat is a specific type of bioreactor. One of the main benefits of a chemostat is that it
is a continuous process (a CSTR), therefore the rate of bacterial growth can be maintained at steady state by controlling the volumetric feed
rate. Bacterial chemostats have many applications, a few of which are listed below.
Pharmaceuticals: Used to study a number of different bacteria, a specific example being analyzing how bacteria respond to different
antibiotics. Bacteria are also used in the production of therapeutic proteins such as insulin for diabetics.
Manufacturing: Used to produce ethanol, the fermentation of sugar by bacteria takes place in a series of chemostats. Also, many
different antibiotics are produced in chemostats.
Food Industry: Used in the production of fermented foods such as cheese.
Research: Used to collect data to be used in the creation of a mathematical model of growth for specific cells or organisms.
The following sections cover the information that is needed to evaluate bacterial chemostats.
6.3.2: BACTERIAL CHEMOSTAT DESIGN

The bacterial chemostat is a continuous stirred-tank reactor (CSTR) used for the continuous production of microbial biomass.
6.3.2.1: CHEMOSTAT SETUP

The chemostat setup consists of a sterile fresh nutrient reservoir connected to a growth chamber or reactor. Fresh medium containing
nutrients essential for cell growth is pumped continuously to the chamber from the medium reservoir. The medium contains a specific
concentration of growth-limiting nutrient (Cs), which allows for a maximum concentration of cells within the growth chamber. Varying the
concentration of this growth-limiting nutrient will, in turn, change the steady state concentration of cells (Cc). Another means of controlling
the steady state cell concentration is manipulating the rate at which the medium flows into the growth chamber. The medium drips into
culture through the air break to prevent bacteria from traveling upstream and contaminating the sterile medium reservoir.
The well-mixed contents of the vessel, consisting of unused nutrients, metabolic wastes, and bacteria, are removed from the vessel and
monitored by a level indicator, in order to maintain a constant volume of fluid in the chemostat. This effluent flow can be controlled by
either a pump or a port in the side of the reactor that allows for removal of the excess reaction liquid. In either case, the effluent stream
needs to be capable of removing excess liquid faster than the feed stream can supply new medium in order to prevent the reactor from
overflowing.
Temperature and pressure must also be controlled within the chemostat in order to maintain optimum conditions for cell growth. Using a
jacketed CSTR for the growth chamber allows for easy temperature control. Some processes such as biological fermentation are quite
exothermic, so cooling water is used to keep the temperature at its optimum level. As for the reactor pressure, it is controlled by an exit air
stream that allows for the removal of excess gas.
For aerobic cultures, purified air is bubbled throughout the vessel's contents by a sparger. This ensures enough oxygen can dissolve into the
reaction medium. For anaerobic processes, there generally is not a need for an air inlet, but there must be a gas outlet in order to prevent a
build up in pressure within the reactor.
In order to prevent the reaction mixture from becoming too acidic (cell respiration causes the medium to become acidic) or too basic, which
could hinder cell growth, a pH controller is needed in order to bring pH balance to the system.
The stirrer ensures that the contents of the vessel are well mixed. If the stirring speed is too high, it could damage the cells in culture, but if
it is too low, gradients could build up in the system. Significant gradients of any kind (temperature, pH, concentration, etc.) can be a
detriment to cell production, and can prevent the reactor from reaching steady state operation.
Another concern in reactor design is fouling. Fouling is generally defined as the deposition and accumulation of unwanted materials on the
submerged surfaces or surfaces in contact with fluid flow. When the deposited material is biological in nature, it is called biofouling. The
fouling or biofouling in a system like this can cause a decrease in the efficiency of heat exchangers or decreased cross-sectional area in
pipes. Fouling on heat exchanger surfaces leads to the system not performing optimally, being outside the target range of temperature, or
spending excess energy to maintain optimum temperature. Fouling in pipes leads to an increase in pressure drop, which can cause
complications down the line. To minimize these effects, industrial chemostat reactors are commonly cylindrical, containing volumes of up
to 1300 cubic meters, and are often constructed from stainless steel. The cylindrical shape and smooth stainless steel surface allow for easy
cleaning.
6.3.2.2: DESIGN EQUATIONS

The design equations for contiuous stirred-tank reactors (CSTRs) are applicable to chemostats. Balances have to be made on both the cells
in culture and the medium (substrate).
Mass Balance
The mass balance on the microorganisms in a CSTR of constant volume is:
[Rate of accumulation of cells, g/s] = [Rate of cells entering, g/s] – [Rate of cells leaving, g/s] + [Net rate of generation of live cells, g/s]
The mass balance on the substrate in a CSTR of constant volume is:
[Rate of accumulation of substrate, g/s] = [Rate of substrate entering, g/s] – [Rate of substrate leaving, g/s] + [Net rate of consumption of
substrate, g/s]
Assuming no cells are entering the reactor from the feed stream, the cell mass balance can be reworked in the following manner:
dCC
(RateAccumulationCells) = V (6.3.1)
dt
(Flow Entering) − (Flow Leaving) = 0 − ν0 CC (6.3.2)
Rate Cell Generation = V (rg − rd ) (6.3.3)
Similarly, the substrate mass balance may be reworked in the following manner:
dCS
Rate Accumulation Substrate = V (6.3.4)
dt
(Flow Entering) − (Flow Leaving) = ν0 CS0 − ν0 CS (6.3.5)
Rate Substrate Consumption = V rS (6.3.6)
Putting Equations 6.3.1 -6.3.3 together gives the design equation for cells in a chemostat:
dCC
V = 0 − ν0 CC + V (rg − rd ) (6.3.7)
dt
Similarly, Equations 6.3.4 -6.3.6 together gives the design equation for substrate in a chemostat:
dCS
V = ν0 CS0 − ν0 CS + V (rg − rd ) (6.3.8)
dt
Assumptions made about the CSTR include perfect mixing, constant density of the contents of the reactor, isothermal conditions, and a
single, irreversible reaction.
Rate Laws
Many laws exist for the rate of new cell growth.
Monod Equation
The Monod equation is the most commonly used model for the growth rate response curve of bacteria.
rg = μ Cc (6.3.9)
where
rg = cell growth rate
Cc = cell cencentration
μ = specific growth rate
The specific cell growth rate, μ, can be expressed as
Cs
μ = μmax (6.3.10)
Ks + Cs
where
μmax = a maximum specific growth reaction rate
Ks = the Monod constant
Cs = substrate concentration
Tessier Equation and Moser Equation
Two additional equations are commonly used to describe cell growth rate. They are the Tessier and Moser Equations. These growth laws
would be used when they are found to better fit experimental data, specifically at the beginning or end of fermentation.
Tessier Equation:
Cs
rg = μmax [1 − exp(− )] Cc (6.3.11)
k
Moser Equation:
μmax Cs
rg = (6.3.12)
−λ
1 + kCs
where λ and k are empirical constants determined by measured data.

Death Rate
The death rate of cells, r , takes into account natural death, k , and death from toxic by-product, k , where C is the concentration of toxic
d d t t
by-product.
rd = (kd + kt Ct ) Cc (6.3.13)
Death Phase The death phase of bacteria cell growth is where a decrease in live cell concentration occurs. This decline could be a result of a
toxic by-product, harsh environments, or depletion of nutrients.
Stoichiometry
In order to model the amount of substrate and product being consumed/produced in following equations, yield coefficients are utilized. Ysc
and Ypc are the yield coefficients for substrate-to-cells and product-to-cells, respectively. Yield coefficients have the units of g variable/g
cells. Equation 6.3.14 represents the depletion rate of substrate:
− rs = Ysc rg + mCc (6.3.14)
Equation 6.3.15 represents the rate of product formation:

rp = Ypc rg (6.3.15)
6.3.3: CONTROL FACTORS

The growth and survival of bacteria depend on the close monitoring and control of many conditions within the chemostat such as the pH
level, temperature, dissolved oxygen level, dilution rate, and agitation speed. As expected with CSTRs, the pumps delivering the fresh
medium and removing the effluent are controlled such that the fluid volume in the vessel remains constant.
pH level
Different cells favor different pH environments. The operators need to determine an optimal pH and maintain the CSTR at it for efficient
operation. Controlling the pH at a desired value during the process is extremely important because there is a tendency towards a lower pH
associated with cell growth due to cell respiration (carbon dioxide is produced when cells respire and it forms carbonic acid which in turn
causes a lower pH). Under extreme pH conditions, cells cannot grow properly, therefore appropriate action needs to be taken to restore the
original pH (i.e. adding acid or base).
Temperature
Controlling the temperature is also crucial because cell growth can be significantly affected by environmental conditions. Choosing the
appropriate temperature can maximize the cell growth rate as many of the enzymatic activates function the best at its optimal temperature
due to the protein nature of enzymes.
Dilution rate
One of the important features of the chemostat is that it allows the operator to control the cell growth rate. The most common way is
controlling the dilution rate, although other methods such as controlling temperature, pH or oxygen transfer rate can be used. Dilution rate is
simply defined as the volumetric flow rate of nutrient supplied to the reactor divided by the volume of the culture (unit: time-1). While
using a chemostat, it is useful to keep in mind that the specific growth rate of bacteria equals the dilution rate at steady state. At this steady
state, the temperature, pH, flow rate, and feed substrate concentration will all remain stable. Similarly, the number of cells in the reactor, as
well as the concentration of reactant and product in the effluent stream will remain constant.
Negative consequences can occur if the dilution rate exceeds the specific growth rate. As can be seen in Equation 6.3.16 below, when the
dilution rate is greater than the specific growth rate (D > μ), the dCC/dt term becomes negative.
dCC
= (μ − D)CC (6.3.16)
dt
This shows that the concentration of cells in the reactor will decrease and eventually become zero. This is called wash-out, where cells can
no longer maintain themselves in the reactor. Equation 6.3.17 represents the dilution rate at which wash-out will occur.
μmax Cs0
Dmax = (6.3.17)
Ks + Cs0
In general, increasing the dilution rate will increase the growth of cells. However, the dilution rate still needs to be controlled relative to the
specific growth rate to prevent wash-out. The dilution rate should be regulated so as to maximize the cell production rate. Figure 1 below
shows how the dilution rate affects cell production rate(DCC), cell concentration (CC), and substrate concentration (CS).
Figure 1: Cell concentration, cell production, and substrate concentration as a function of dilution rate
Initially, the rate of cell production increases as dilution rate increases. When Dmaxprod is reached, the rate of cell production is at a
maximum. This is the point where cells will not grow any faster. D = μ (dilution rate = specific growth rate) is also established at this point,
where the steady-state equilibrium is reached. The concentration of cells (CC) starts to decrease once the dilution rate exceeds the Dmaxprod.
The cell concentration will continue to decrease until it reaches a point where all cells are washed out. At this stage, there will be a steep
increase in substrate concentration because fewer and fewer cells are present to consume the substrate.
Oxygen transfer rate
Since oxygen is an essential nutrient for all aerobic growth, maintaining an adequate supply of oxygen during aerobic processes is crucial.
Therefore, in order to maximize the cell growth, optimization of oxygen transfer between the air bubbles and the cells becomes extremely
important. The oxygen transfer rate (OTR) tells us how much oxygen is consumed per unit time when given concentrations of cells are
cultured in the bioreactor. This relationship is expressed in Equation 6.3.18 below.
Oxygen Transfer Rate (OTR) = QO CC (6.3.18)
2
Where CC is simply the concentration of cell in the reactor and QO2 is the microbial respiration rate or specific oxygen uptake rate. The
chemostat is a very convenient tool to study the growth of specific cells because it allows the operators to control the amount of oxygen
supplied to the reactor. Therefore it is essential that the oxygen level be maintained at an appropriate level because the cell growth can be
seriously limited if inadequate oxygen is supplied.
Agitation speed
A stirrer, usually automated and powered with a motor, mixes the contents of the chemostat to provide a homogeneous suspension. This
enables individual cells in the culture to come into contact with the growth-limiting nutrient and to achieve optimal distribution of oxygen
when aerobic cultures are present. Faster, more rigorous stirring expedites cell growth. Stirring may also be required to break agglutinations
of bacterial cells that may form.
6.3.4: Q&A
Q1: Why is a chemostat called a chemostat?
A1: Because the chemical environment is static, or at steady state. The fluid volume, concentration of nutrients, pH, cell density, and other
parameters all are assumed to remain constant throughout the operation of the vessel.
Q2: What are some concerns regarding chemostats?
A2: a) Foaming results in overflow so the liquid volume will not be constant. b) Changing pumping rate by turning the pump on/off over
short time periods may not work. Cells respond to these changes by altering rates. A very short interval is needed for it to respond correctly.
c) Fragile and vulnerable cells can be damaged/ruptured when they are caught between magnetic stirring bar and vessel glass. d)Bacteria
contamination occurs because bacteria travel upstream easily and contaminate the sterile medium. This can be solved by interrupting the
liquid path with an air break.
Q3: The Monod equation uses a Michaelis-Menten relationship which is based on a quasi-state assumption. (T/F)
A3: T
Q4: An important feature of chemostat is the dilution rate. Define dilution rate.
A4: Dilution Rate = volume of nutrient medium supplied per hour divided by the volume of the culture.
Q5: What are the advantages/disadvantages over choosing a chemostat instead of a batch reactor for bioreactions?
A5: Advantages: 1. A chemostat has better productivity than a batch reactor. There is a higher rate of product per time per volume. A batch
process wastes time. 2. A chemostat is operated at steady state, therefore it has better control maintaining the same conditions for all product
produced.
Disadvantages: 1. A chemostat is less flexible than a batch reactor. A batch reactor can be used to make more than one product. 2. It is
harder to maintain a sterile system in a chemostat. A batch reactor is easier to clean.
Q6: What is the physical meaning of the Monod constant?
A6: The Monod constant is a substrate concentration at which the growth rate of the biomass of microbial cells participating in the reaction
is half the maximum growth rate.
 EXAMPLE 6.3.1
Researchers at the University of Michigan are using a bacterial chemostat to model the intestinal tract of a pig in order to study the
metabolism of E. Coli bacteria in that particular environment. The growth chamber of the chemostat has a volume of 500 dm3.
The initial concentration of E. Coli bacteria inoculated in the chemostat growth chamber is 1 g/dm3. A 100g/dm3 substrate feed is fed to
the chemostat at a volumetric flow rate of 20 dm3/hr. How much time is required for this biochemical process to reach steady rate from
the point of startup? Assume the growth rate is the Monod equation for bacteria bacterial cell growth, shown above.
Additional data pertaining to the problem is given: μmax = 0.8; Ks = 1.7 g/dm3; Ys/c = 8; Yp/c = 5; m = 0; rd = 0;
Schematic Drawing
Answer = 3.7 hours

Solution
The Chemostat was modeled in Excel using the design equations above and Euler's Method. A graph of Cell Concentration (g/dm3) vs
Time(hr) was then plotted. When the Cell Concentration become stable, steady state has been reached and the time can be read off the
graph. Below is a screen shot of the model and the graph created.
Excel Model Screen Shot
Excel Graph
This graph clearly shows that steady state is reached 3.7 hours after start up.
 EXAMPLE 6.3.2
After calculating the time required to reach steady state, the researchers decide to start up the chemostat. While do so, the control valve
for the inlet substrate feed malfunctions. The flow rate of substrate into the chemostat is accelerating at 40 dm3/hr2. Determine how
long they have to correct the problem before wash-out occurs and all of the bacteria in the chemostat is lost.
Schematic Drawing
Modeling the Malfunction

fracdν0 dt = kvalve = 40
Answer = 20 hours
Solution
The Chemostat was modeled in Excel using the design equations above and Euler's Method. A graph of Cell Concentration (g/dm3) vs
Time(hr) was then plotted. When the Cell Concentration becomes zero wash-out of the bacteria took place. Below is a screen shot of
the model and the graph created.
Excel Model Screen Shot
Excel Graph
This graph clearly shows wash-out occurs 20 hours after start up. We can see in example that process controls are extremely important
for Bacterial Chemostats.
The template model used for both Worked Out Example 1 and 2 can be downloaded here Media: Bacterial Chemostat Template.xls
6.3.5: REFERENCES
"Chemostat." McGraw-Hill Dictionary of Scientific and Technical Terms. McGraw-Hill Companies, Inc., 2003. Accessed September 16,
2007. Available http://www.answers.com/topic/chemostat
Fogler, H. Scott (2006). Elements of Chemical Reaction Engineering. New Jersey: Prentice Hall PTR. ISBN 0-13-047394-4
Kelly, C. "Ch 9 - Reactors" 17 February, 2004. Retrieved 2007-09-24. Available
www.ecs.syr.edu/faculty/kelly/handouts/biochemical%20engineering/chapter%209%20reactors.ppt
Smith, H.L. (2006). "Bacterial Growth". Retrieved on 2007-09-15.
Strandberg, Per Erik (2004). "Mathematical models of bacteria population growth in bioreactors: formulation, phase space pictures,
optimisation and control". Retrieved on 2007-09-16.
Strandberg, Per Erik (2003). "The Chemostat". Retrieved on 2007-09-15.
"What is a Bioreactor?" Transgalactic Ltd., 25 May, 2005. Retrieved 2007-09-24. Available
http://www.bionewsonline.com/o/what_is_bioreactor.htm

Authors: Shoko Asei, Brian Byers, Alexander Eng, Nicholas James, Jeffrey Leto
Stewards: Jeffrey Falta, Taylor Lebeis, Shawn Mayfield, Marc Stewart, Thomas Welch
Stewards: Sarah Hebert, Valerie Lee, Matthew Morabito, Jamie Polan
6.3: Bacterial Chemostat is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source content that was
6.4: ODE AND EXCEL CSTR MODEL WITH HEAT EXCHANGE
6.4.1: INTRODUCTION
A Continuous Stirred-Tank Reactor (CSTR) is a chemical reaction vessel in which an impeller continuously stirs the contents ensuring
proper mixing of the reagents to achieve a specific output (Figure 6.4.1 ). Useful in most all chemical processes, it is a cornerstone to the
Chemical Engineering toolkit. Proper knowledge of how to manipulate the equations for control of the CSTR are tantamount to the
successful operation and production of desired products. The purpose of this section is to model dynamic conditions within a CSTR for
different process conditions. Simplicity within the model is used as the focus is to understand the dynamic control process.
Figure 6.4.1: Cross-sectional diagram of a CSTR. (CC BY-SA 3.0; Daniele Pugliesi via Wikipedia)
6.4.2: ASSUMPTIONS
For the purposes of this section, we have made the following assumptions to explain CSTR with heat exchange modeling.
Perfect mixing: The agitator within the CSTR will create an environment of perfect mixing within the vessel. Exit stream will have
same concentration and temperature as reactor fluid.
Single, 1st order reaction: To avoid confusion, complex kinetics are not considered in the following modeling. \[\-r_{a}=k C_{A}]
Parameters specified: We assume that the necessary parameters to solve the problem have been specified.
Volume specified: In a control environment, the size of the vessel is usually already specified.
Constant Properties: For this model, we have made the assumption that the properties of the species we are trying to model will remain
constant over the temperature range at which we will be looking at. It is important to make this assumption, as otherwise we will be
dealing with a much more complex problem that has a varying heat capacity, heat of reaction, etc.
6.4.3: ODE MODELING IN EXCEL

To setup the model, the mass and energy balances need to be considered across the reactor. From these energy balances, we will be able to
develop relationships for the temperature of the reactor and the concentration of the limiting reactant inside of it.
6.4.3.1: VARIABLE DEFINITIONS

The following table gives a summary of all of the variables that we used in our mathematical formulas.
Symbol Explanations
Symbol Meaning Symbol Meaning
V Volume of Reactor ρ Density of Stream
m Mass Flow Rate CA0 Original Concentration
CA Current Concentration T0 Original Temperature
T Current Temperature TC Coolant Temperature
ΔHrxn Heat of Reaction ΔCp Overall change in Heat Capacity
k0 Rate Law Constant E Activation Energy
R Ideal Gas Constant UA Overall Heat Transfer Coefficient
6.4.3.2: MASS BALANCE
From our energy and material balances coursework, we know that the general equation for a mass balance in any system is as follows:
Rate Accumulation = (Flow In) − (Flow Out)−(\mathrm{Rate } \text{Generation})
dCA
Rate Accumulation = V
dt
m
Flow In − Flow Out = (CA0 − CA )
ρ
−E /RT
Rate Generation = V (− ra ) = V CA k0 e
In the case of a CSTR, we know that the rate of accumulation will be equal to V . This comes from the fact that the overall number of
dCA
dt
moles in the CSTR is V C , so the accumulation of moles will just be the differential of this. Since is a constant, this can be pulled out of
A
the differential, and we are left with our earlier result. We also know that the flow of moles in versus the flow of moles out is equal to
− C ) , which is the mass flow rate, divided by the density of the flow, and then multiplied by the difference in the concentration of
m
(C A0 A
ρ
moles in the feed stream and the product stream. Finally, we can determine what the rate of generation of moles in the system is by using the
Arrhenius Equation. This will give us the rate of generation equal to
.
dCA
Combining all of these equations and then solving for dt
, we get that:
dCA m
−E /RT
= (CA0 − CA ) − k0 CA e
dt ρV

From our thermodynamics coursework, we know that the general equation for an energy balance in any system is as follows:
In the case of a CSTR, we know that the rate of energy accumulation within the reactor will be equal to . This equation is
basically the total number of moles (mass actually) in the reactor multiplied by the heat capacity and the change in temperature. We also
know that the heat generated by this reaction is , which is the rate of mass generation ( − Vra) times the specific heat of reaction (ΔHrxn).
The overall rate of heat transfer into and out of the system is given by . This equation is the flow rate multiplied by the
heat capacity and the temperature difference, which gives us the total amount of heat flow for the system. Finally, the amount of heat
transferred into the system is given by . Combining all of these equations and solving the energy balance for , we get that:
In a realistic situation in which many chemical processed deal with multiple reactions and heat effects slight changes to the modeled
equation must be done. The diagram below evaluates the heat exchanger under heat effects in which there is an inlet and outlet temperature
that is accounted for in the enthalpy term in the newly modeled equation.
To model a heat exchanger that accounts for multiple reactions simply take the δH and δC term and add the Greek letter sigma for
rxn p
summation in front of the terms. When considering a case with multiple reactions and heat effects, the enthalpy and heat capacity of each
reaction must be implemented in the energy balance, hence i and j represents the individual reaction species. The equation now looks
something like this:
6.4.3.4: EULER'S METHOD

In order to model an ODE in Excel, we will be using Euler's Method. The reason why we chose Euler's Method over other methods (such as
Runge-Kutta) was due to the simplicity associated with Euler's. Trying to model with methods such as Runge-Kutta are much more difficult
in Excel, while Euler's method is quite simple. We also checked the accuracy of our results using Euler's Method by comparing our answers
with that of Polymath, a highly accurate ODE modeling program. We found that our answers were the same as that of Polymath, so we are
quite comfortable using Euler's Method for our model. Here are the full details on Euler's Method. For this model, we will primarily be
interested in the change in concentration of a reactant and the temperature of the reactor. These are the two differential equations that we
were able to obtain from the mass and energy balances in the previous section. The following is an application of Euler's method for the
CSTR on the change in concentration:
dCA
CAi = CA(i−1) + Δt
dt
This same application can be made to the change in temperature with respect to time:
dT
Ti = Ti−1 + Δt
dt
Assuming that all values in the ODE's remain constant except for and , the new value is then found by taking the pervious value and
adding the differential change multiplied by the time step.
6.4.3.5: LIST OF EQUATIONS
The following are a summary list of all of the equations to be used when modeling CSTR with a heat exchange.
dCA m −E /RT
= (CA0 − CA ) − k0 CA e
dt ρV
dCA
CAi = CA(i−1) + Δt
dt
dT
Ti = Ti−1 + Δt
dt
6.4.3.6: COMBINING USING EXCEL

To model the CSTR using Excel you use Euler's Method, Energy Balance and Mass Balance together to solve for the concentration at time
t.
Screenshot of Excel Model for CSTR with Heat Exchange
6.4.4: HOW TO USE OUR MODEL
In order to help facilitate understanding of this process, we have developed an Excel spreadsheet specifically for looking at the changes in
concentration and temperature given some change in the input to the CSTR system. An example of a change to the system could be that the
temperature of the feed stream has dropped by a given number of degrees, or that the rate at which the feed stream is being delivered has
changed by some amount. By using our spreadsheet, you will be able to easily plug in your given parameters, and look at the trend of the
concentration and temperature over a wide time interval.
The way in which this spreadsheet works is quite simple. Boxes are provided for you to input all of the given information for your CSTR
problem. Various columns containing values for the temperature, concentration, etc., with respect to time have also been provided. There are
then more columns that contain the values for the various differential equations from above. With the time derivative in hand, we are then
able to predict the value of the temperature or concentration at the next given time interval.
Our easy-to-use Excel model is given here: CSTR Modeling Template. In our model, you will find a column of unknowns that must be
specified in order to solve for the optimal conversion temperature and optimal concentration of A. There are then two cells that will display
the optimal temperature and concentration. Graphs are also provided to look at the change in temperature and concentration over time. Most
of the variables in the model are self-explanatory. One important feature of our model is the option to have a change in the temperature of
the feed stream or the concentration of A after a given time t. You do not need to input a value for these cells if there is no change in the
feed; it just provides a convenient way to look at the change of temperature and concentration of A. You are also provided with a cell for the
time step, . Depending on what size time step you choose, you may need to choose a larger value if your graphs do not reach steady
state. If this is the case, the output cells will tell you to increase the time step.
6.4.5: APPLICATIONS TO PROCESS CONTROL

The model created can account for many different situations. However, in process control, it is very important for the following to be
considered.
1. Exothermic reaction - In industry, the exothermic reaction is typically of more importance to controls as there is a safety factor
involved. (i.e. explosive reactor conditions)
2. Volume constant - For CSTRs, liquids are mostly used.
The model developed can assist in controlling an actual process in many ways including the following, but not limited to:
Maintaining product concentration despite changes operating conditions
Predicting and preventing explosive reactor conditions
Flow rate optimization
Input concentration optimization
Operating temperature optimization
Coolant temperature optimization
This model is very useful for many simple reactor situations and will aid in the understanding of a process and how dynamic situations will
affect operation.
In real-life applications, the reaction may not be first order or irreversible or involve multiple reactions. In all these cases, we modify the
Energy and Mass Balance with additional rate law considerations.
 EXAMPLE 6.4.1
You are contracted by WOW Chemical to control the operation of their 3000 L jacketed CSTR. They desire to create chemical B from
chemical A at an optimal conversion. What is the temperature at which the optimal conversion is achieved?
A → B is a first order, irreversible reaction.
Some information about the process:
Feed stream temperature = 400 K
Coolant temperature = 350 K
Heat of reaction = -200 J/mol
Inlet concentration of A = 9 mol/L
Inlet flow rate = 4 kg/s
Density of A = 1000 g/L
UA of the heat exchanger = 7 kcal/s
Rate constant = 1.97x1020 s-1
Activation energy = 166 kJ/mol
Overall change in Heat capacity = 1 kcal/kg-K.

4.6.1 Answer to Example Problem #1
From using our handy-dandy worksheet, we see that the optimal conversion occurs at 368.4 K. This comes from inputting all of the given
information from the problem statement into the Excel sheet, and then reading off the value for the optimal temperature. Also, make sure
you are converting to proper units, such as from kJ to J, for use of our model (just to see if you were paying attention to units).
Example 1 Solution

1. In the previous example, the plant was operating in the daytime. At night, there is a drop of 5 K in feed stream temperature. In order to
maintain the same concentration do we need to...
A. Decrease the temperature of the cooling water in the heat exchanger.
B. Increase the temperature of the cooling water in the heat exchanger.
C. Maintain the same temperature of the cooling water in the heat exchanger.
2. At night, the inlet flow rate of the system also decreases due to the plant not operating at full capacity. What happens to the temperature
of the reaction if concentration of A in the feed stream stays the same?
A. Decrease
B. Increase
C. Stay the same
3. Finally, if the rate constant doubled, what effect will this have on the overall concentration if temperature is held constant?
A. Increase
B. Decrease
C. Stay the same
6.4.7.1: ANSWERS TO EXAMPLE PROBLEM #2

1: B. Since the temperature of the feed stream has decreased, we will need to increase the temperature of the cooling stream in order to
maintain the same temperature at which the CSTR was operating at before. Using the Excel spreadsheet from Example 1 and lowering the
feed stream temperature by 5 K, we can first see that the temperature of the CSTR has fallen to 366.5 K.
We can then adjust the temperature of the feed stream and see that it must be raised by 3 K in order to return to the original operating
temperature of 368.4 K.
2: A. Since there is less feed coming into the CSTR, the optimal temperature of the reactor will decrease as well. This can again be seen by
using our Excel model from Example 1. We know that originally the temperature of the reactor was at 368.4 K. By decreasing the flow rate
to 3 kg/s, we can see that the temperature of the reactor decreases by over 3 K down to 365.1 K.
3: B. We see in the equations for the differential equations that the rate constant has a negative effect on the conctration of A. By doubling
its value, we will then be decreasing CA as well. This can also be seen in the Excel model from Example 1. We see that the original
concentration of A is 6.3 mol/L. We can then double the value of the rate constant, and we see that the concentration of A drops down to 4.8
mol/L, which is what we had predicted would happen.
6.4.8: REFERENCES
Bequette, B. Wayne. Process Dynamics Modeling, Analysis, and Simulation, New Jersey: Prentice Hall.
Fogler, H. Scott. The Elements of Chemical Reaction Engineering, Prentice Hall. ISBN 0-13-047394-4
Kravaris, Costas. Chemical Process Control: A Time-Domain Approach, The University of Michigan.
Riggs, James B.; Karim, M. Nazmul. Chemical and Bio-Process Control, Texas Tech University, Ferret Publishing.
6.4.9: CONTRIBUTOR
Authors: Jason Bourgeois, Michael Kravchenko, Nicholas Parsons, Andrew Wang
6.4: ODE and Excel CSTR model with heat exchange is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al.
6.5: ODE AND EXCEL MODEL OF A SIMPLE DISTILLATION COLUMN
6.5.1: INTRODUCTION
Distillation is a commonly employed separation technique bases on difference in volatilities. The modern form of distillation as is is known
today may be credited to early Arab alchemist, Jabir ibn Hayyan and the development of one of his inventions, the alembic. The distillation
apparatus is commonly referred to as a still and consists of a minimum of a reboiler in which mixture to be separated is heated, a condenser
in which the vapor components are cooled back to liquid form, and a receiver in which the concentrated liquid component fractions are
collected. Ideally, distillation is governed by the principles of Raoult’s Law and Dalton’s Law. Dalton’s Law states that for a mixture, the
total vapor pressure is equal to the sum of the individual vapor pressures of the pure components which comprise this mixture. The
relationship giving the vapor pressure of a volatile component in a mixture, PA, is Raoult’s Law and is governed by the following equation:
o
PA = XA P
A
Where XA is the mole fraction of component A in the mixture and PA° is the vapor pressure of pure component A. This ideal model is based
on a binary mixture of benzene and toluene but for other mixtures severe deviations from Raoult’s Law may be observed due to molecular
interactions. For these aforementioned mixtures where the components are not similar the only accurate alternative is obtaining vapor-liquid
equilibrium by measurement.
In simple distillation, two liquids with differing boiling points are separated by immediately passing the vapors from the reboiler to a
condensing column which condenses the vapor components. As a result the distillate is not usually pure but its composition may be
determined by Raoult’s Law at the temperature and pressure at which the vapors were boiled off. Consequently, simple distillation is usually
used to separate binary mixtures where the boiling temperatures of the individual components are usually significantly different or to
separate volatile liquids from non-volatile solids.
A reference of terms used in distillation reference are included at the end of this article.
6.5.2: DISTILLATION CONTROL

Distillation columns comprise an enormous amount of the separation processes of chemical industries. Because of their wide range of uses
in these industries and because their proper operation contributes to product quality, production rates and other capital costs, it is clear that
their optimization and control is of great importance to the chemical engineer. Distillation control becomes problematic because of the wide
variety of thermodynamic factors stemming from the separation process. For example:
Separations deviate from linearity of equations as product purity increases.
Coupling of process variables occurs when compositions are controlled.
Disturbances occur due to feed and flow agitation
Efficiency changes in trays lead to non-steady state behavior.
To improve upon distillation control you must be able to characterize these potential problems and realize when they occur because they
lead to dynamic behavior of the column. Of key importance to control is the maintenance of material and energy balances and their due
effects on the column. Shown below is a schematic of a simple binary distillation column. Using the material balance formulas
D z−x
=
F y−x
where z , x, and y are the feed, bottoms and distillate concentrations respectively, you find that as D (Distillate) increases, its purity
decreases. This leads to the idea that purity level varies indirectly with the flow rate of that product. Energy input is also key because it
determines the vapor flow rate (V ) up the column which has direct effects on the L/D ratio (reflux ratio) and hence relates to an increase in
the amount of separation occurring. To summarize, energy input determines the amount of separation, while material flow relates the ratio
of separation in the products.
Vapor liquid dynamics within the column also contributes to the theory behind process control because of a few important relations.
Changing V (by changing the reboiler energy), causes an extremely rapid response in the overhead composition, while changing the reflux
ratio requires a longer response to its effect on the reboiler.
In lower pressure columns, a phenomena known as entrainment or flooding occurs in which liquid is blown up into trays instead of
dropping down into trays. This significantly decreases separation efficiency and therefore less product gain occurs. Using a packed column
in these low pressure applications provides greater efficiency over tray columns and also allows for the faster accomplishment of a steady
state profile. Controlling the occurrence of entrainment in either case is another crucial aspect which should be recognized when designing
control systems for columns.
6.5.2.1: REGULATORY CONTROLS

For the distillation process it is imperative that regulatory controls such as level, flow, and pressure controllers are functioning properly to
further ensure the effectiveness of the product composition controllers.
In terms of regulatory control, level controls are used to maintain specified levels in the reboiler, accumulator, and in the case of a
distillation column with two columns due to high tray numbers for a single column, also maintain the level in an intermediate accumulator.
Inept use of level controls may lead to problems elsewhere in the distillation process. For example, poor level control on the accumulator
and reboiler may lead to problems with composition control for material balance control configurations. Also if the reboiler duty is
maintained by one of these level controllers and the controller causes oscillation in the reboiler, consequently cycling may also occur in
column pressure.
Flow controllers are used to manipulate and maintain desired flow rates of the reflux, distillate and bottoms products, and the heating
medium employed in the reboiler. The setpoints of this specific type of controller is determined by the various composition and level
controllers in the process.
Pressure controls are located to the top of the distillation column in the distillate vicinity. Here the column overhead pressure, caused by the
accumulation of components in the vapor phase, acts as an integrator causing a change in the level of the accumulator. This pressure may be
controlled using a variety of methods.
Pressure Control through Condenser Operation
1. Maximize cooling water flow rate to the condenser (operate at minimum column pressure)
2. Adjust the rate of condensation of the overhead (for example by adjusting the flow rate of refrigerant to the condenser)
3. Adjust level of liquid in the condenser (changes heat transfer area)
Pressure Control through Accumulator Operation
1. Purge vapor from the overhead accumulator
2. Directly changing amount of material phase (for example by pumping in inerts)
6.5.2.2: COMPOSITION AND CONSTRAINT CONTROL

Because distillation requires a desired product concentration or flowrate, constraint control is used to ensure the desired operation
conditions by having setpoints designated for the requirements of the system. The constraint is usually a concentration and the control of
this concentration can vary depending on the application. In almost all industry applications only one product concentration is controlled,
while the other is allowed to vary. This is known as Single Composition control and is much easier to achieve and maintain than the Dual
composition control which specifies both product concentrations. The advantage to Dual Control however being increased energy efficiency
because of increased separation. P&ID of common control placement used for composition control are shown below. Control lines are not
included in this diagram because of the various number of control possibilites for a system.
Setting constraints on a column allows for proper control of the product as well as points to issues of safety and maintenance. The most
common constraint controls are in the maximum reboiler and condenser duty which results from a number of variables including fouling ,
improper valve sizing, and excessive increases in feed. Other common constraints are flooding and weeping points which indicate incorrect
L/V ratios and can be compensated for by adjusting the pressure drop across the column.
The most common adjustments for constraint control involve changing the reboiler duty to satisfy the constraints and almost always follow
with subsequent fall back mechanisms to ensure product quality.
1. Ensuring single composition control
2. Reducing feed rate
3. Increasing product purity setpoints
6.5.2.3: EFFECTIVE DISTILLATION CONTROL

Before doing an in depth analysis of distillation control, it is of paramount importance that the following basics are attended to.
Firstly, ensure that regulatory controls are indeed functioning congruously.
For changes in reflux temperature, employ the use of reflux controls.
Make sure to check and evaluate analyzer deadtime, accuracy, and reliability. This is required to account for the lagtime from the
product composition analyzer when using feedback control to control feed flow, reflux ratio and reboiler power. Refer to for deadtime.
To select proper analyzer refer to
Ensure that any thermistors or RTD’s employed to measure tray temperatures for composition inference are fully operational and
correctly situated. Care should be taken here to ensure that pressure corrected temperatures are used. Refer to the following for
temperature sensors
When streams such as D, B, L, and V are used as manipulated variables for composition control, they should be changed with respect to
the measured feed rate when column feed rate changes are a common disturbance.
Summary of Distillation Control Methods
Control Parameter Example of Control Method
Distillate flow rate (D) Flow controller (Setpoint controlled by accumulator level)
Bottoms flow rate (W) Flow controller (Setpoint controlled by level in bottom of column)
Reflux flow rate (LD) Flow controller (Setpoint controlled by top tray temperature)
Reboiler Steam flow rate Flow controller (Setpoint controlled by bottoms composition analyzer)
Feed flow rate Flow controller (Setpoint manually entered based on unit operations)
Distillate purity (xD) Reflux flow controller (Setpoint controlled by top tray temperature)
Bottoms purity (xW) Steam flow controller (Setpoint controlled by online analyzer)
Column pressure Purge flow controller (Setpoint controlled by column pressure)
6.5.2.4: CONTROL PROBLEMS AND DISTURBANCES

Feed Composition & Feed Flow Upsets
In order to properly determine the ultimate product purity and flow you must consider the impact of disturbances in the column system. The
most significant, however most easily remedial of these disturbances is a feed composition upset, in which there is a change in the feed
composition resulting in a major disturbance in the product composition. Hence, configuration of a distillation column must consider the
control regulations of such a feed upset. Feed flow upsets are regulated using ratio control of L/F, D/F, V/F and B/F through level and reflux
ratio sensors.
Feed and Reflux Enthalpy Disturbances
Feed Enthalpy upsets become an issue for columns operating at low reflux ratios and cause a large deviation from expected product
concentrations due to the changes in vapor and liquid flowrates in the column. The usual compensation for this is using a feed heat
exchanger to control the proper enthalpy to the column. In most cases, the feed is preheated before entering the column. By adjusting the
duty of the preheater (i.e. decreasing or increasing the heating medium flow rate), a constant vapor/liquid ratio in the feed can be
maintained. Rapid changes to the external conditions of a column, (especially large temperature deviations in rainstorms) can cause a
subcooled reflux, changing the composition of the products. Reflux control can amend this properly.
Steam and Column Pressure Upsets
The most severe disturbance occurring in a distillation column occurs when there is a loss of steam pressure in the reboiler. A sharp drop in
steam pressure results in a drop in reboiler effectiveness and therefore a huge increase in the impurity of the product. This can be avoided
using an override control loop for this particular occurrence. Due to the effect of pressure on the relative volatility of the components in the
system, a disturbance in column pressure leads to altered product quality. This can be effectively maintained by a composition controller to
compensate for these pressure differences.
6.5.3: ODE MODELING OF A DISTILLATION COLUMN

There are two methods used for modeling a distillation column: dynamic modeling and steady state modeling. The key difference between
the models is that dynamic modeling is used to monitor changes in the distillation column as a function of time, while the steady state model
looks at a given set of conditions at one particular time (i.e. when the column is at steady state).
6.5.3.1: DYNAMIC MODEL

Dynamic modeling of a distillation column may be used for a variety of different reasons: monitoring variations in the column as a result of
feed changes, to predict the effects of tray fouling, and to predict when flooding with occur. The dynamic model allows the user to improve
distillation control by being able to deal with disturbances that cause upsets to the column's normal operation.
The algorithm for developing a dynamic distillation column model is as follows:
6.5.3.2: STEP 1) WRITE COMPONENT MATERIAL BALANCE FOR EACH STAGE IN THE COLUMN.
Component material balance for all stages, except the feed tray, overhead condenser, and reboiler:
dMi xi
= Li−1 xi−1 + Vi+1 yi+1 − Li xi − Vi yi
dt
Assumption: For simplicity, accumulation in the each stage is constant;

dMi
= 0
dt
Simplified component material balance for each stage (only composition changes with time):
dxi
Mi = Li−1 xi−1 + Vi+1 yi+1 − Li xi − Vi yi
dt
The following are examples of equations used in the Excel Interactive ODE Distillation Column Model, which are provided to help the user
understand how the model works.
ODE used to solve for the liquid composition leaving tray 2 (rectifying section):
dx2 1
= [L1 x1 + V3 y3 − L2 x2 − V2 y2 ]
dt M2
ODE used to solve for the liquid composition leaving tray 5 (stripping section):
dx5 1
= [L4 x4 + V6 y6 − L5 x5 − V5 y5 ]
dt M5
Overhead condenser component balance:

dxD 1
= [V1 (y1 − xD )]
dt MD
Feed tray component balance:

dx3 1
= [L2 x2 + V4 y4 − L3 x3 − V3 y3 ]
dt M3
Reboiler component balance:

dxW 1
= [L6 x6 − W xW − V7 y7 ]
dt MW
6.5.3.3: STEP 2) WRITE TOTAL MATERIAL BALANCES AROUND CONDENSER AND REBOILER
Condenser material balance:
Assumption 1: Total condenser (all vapor from the top of the column is condensed into a liquid).
Assumption 2: Overhead accumulator liquid level remains constant.
D = [V1 + LD ]
Reboiler material balance:

W = [F − D]
For these equations to work, the user must specify:

reflux flow rate (mol/min)
bottoms flow rate (mol/min).
6.5.3.4: STEP 3) DEFINE ALL FLOW RATES

Vapor Leaving Feed Stage:
V3 = V4 + F (1 − qF )
Liquid Leaving Feed Stage:

L3 = L2 + F (qF )
Vapor flow rates in stripping section:
Assumption: Equimolal overflow for vapor in stripping section
V4 = V5 = V6 = (V7 )
Vapor flow rates in rectifying section:

Assumption: Equimolal overflow for vapor in rectifying section
V1 = V2 = (V3 )
Liquid flow rates in rectifying section:

Assumption: Equimolal overflow for liquid in rectifying section
L2 = L1 = (LD )
Liquid flow rates in stripping section:

Assumption: Equimolal overflow for liquid in stripping section
L6 = L5 = L4 = (L3 )
6.5.3.5: STEP 4) DEFINE EQUILIBRIUM CONDITIONS

The binary system used in the Excel ODE model is a benzene-toluene system. The equilibrium data for this system was put in the model
and the relative volatilities were calculated for various equilibrium compositions.
Relative Volatility (from equilibrium data):
ybenzene χtoluene
α =
χbenzene ytoluene
where α is defined as the relative volatility of the two components in the system.
These relative volatilities were plotted against temperature and linear regression was used to fit the data.
Relative volativity as a function of temperature:
α = [−0.009T + 3.3157]
This equation models how the separation changes on each tray as a function of tray temperature, which decreases up the column.
Equilibrium Vapor Composition for each stage:
Assumption: Trays in the Column are 100% efficient (vapor and liquid leaving a tray are in equilibrium)
αxi
yi =
1 + (α − 1)xi
Replacing alpha with the temperature dependent equation shows how tray temperature affects the amount of benzene in the vapor leaving
each tray.
6.5.3.6: STEP 5) WRITE COMPONENT ENERGY BALANCES FOR EACH STAGE.

The ODE energy balances are essential for the dynamic model to run properly. Mass transfer occurs within the column because the
temperature varies from the top of the column to the bottom thereby allowing separation of the components in the system.
Since the energy input into the column is added in the reboiler, the reboiler ODE is the first equation entered in the model. In our model, this
is given as:
dT7 1 qr
= [L6 x6 − W xW ] [T6 − T7 ] +
dt MW MW cp
The next step is adding energy balances for each subsequent stage in the distillation column. The only stage in the column which has a
slightly different energy ODE is the feed stage, given by:
The last energy balance is around the condenser.

Assumption: Reflux return temperature is constant (overhead condenser duty varies to compensate for this).
6.5.3.7: STEP 6) DETERMINE INPUTS INTO ODE MODEL
Once all of the equations have been put into the model, all remaining unknown variables must be placed in a section so that the user can
specify these input values when running the model. For the Excel ODE distillation model, the users inputs include:
1. Feed flow rate
2. Mole fraction of light key in the feed
3. Reflux flow rate
4. Condenser, reboiler, and tray levels
5. Phase of the feed (q-value)
6. Feed temperature
7. Integration step size
To model effects of disturbances, the user may also change these input values:
Feed flow after 200 time steps
Feed composition after 600 time steps
6.5.3.8: STEP 7) USE EULER'S METHOD TO SOLVE THE ODE'S
This step involves using Euler's method. to integrate each ODE over each timestep in the interval to solve for the parameter value at the next
time step. Creating a plot of these values versus time allows the user to see how changes in the input values effect parameters such as
distillate and bottoms composition or flowrates.
6.5.3.9: ADDITIONAL CONSIDERATIONS FOR DYNAMIC DISTILLATION MODELING

The Interactive Excel ODE Distillation Column Model does not take into account heat effects within or surrounding the column. It may be
beneficial in modeling an actual column to determine the optimal temperatures that feed and reflux should enter at to achieve the greatest
possible separation. The energy input into the reboiler is another consideration that may need to be modeled for economic purposes.
Also, to tune the dynamic distillation model with greater precision, additional parameters and equations may be added. One example of this
would be to add a KD or KU input value to control the levels in the bottom of the column or overhead condenser.
6.5.3.10: STEADY STATE MODEL
The steady state model which is described by using a McCabe-Thiele diagram shows the theoretical stages in a binary component
distillation column. An example of a McCabe-Thiele diagram is shown below.
The upper operating line is a graphical representation of the vapor/liquid dynamics in each stage in the rectifying section of the column
(above the feed stage), while the lower operating line represents the vapor/liquid dynamics in the stripping section of the column (below the
feed stage). The starting point of the upper operating line represents the distillate composition and the bottom point of the lower operating
line represents the bottoms compostion. The feed line shows the entering feed composition as well as whether the feed is vapor, liquid, or a
combination of the two. To develop this steady state model, one must know the components in the system, so that equilibrium data can be
obtained. Also, this model requires that the following parameters must be known:
reflux ratio
distillate composition
bottoms composition
feed composition
feed phase
The general equations used in the steady state model are given in the table below:
Media:EquationsforODEModeling.doc
The McCabe-Thiele diagrams are excellent for modeling steady state operation, but they do not describe how disturbances affect the column
operation. For each change in a particular parameter, a separate McCabe-Thiele diagram must be made. The dynamic model, although more
complex than the steady state model, shows how a column operates during start-ups, when disturbances occurs, and where steady state
conditions occur. Therefore, the dynamic ODE model of a distillation column can allow the user to how product purities and flow rates
change with time.
6.5.3.11: GLOSSARY OF TERMS

Mi = Molar holdup on tray i
Li − 1 = Liquid molar flowrate into tray i
Li = Liquid molar flowrate leaving tray i
Vi + 1 = Vapor molar flowrate entering tray i
Vi = Vapor molar flowrate leaving tray i
xi = mole fraction of light component in the Liquid phase of Tray i
yi = mole fraction of light component in the Gas phase of Tray i
B = Bottoms flowrate
D = Distillate flowrate
f = Feed flow rate
= Relative volatility of Benzene-Toluene system.
q = Vapor Liquid compostion value
6.5.4: OUR EXCEL DISTILLATION MODEL

Below you will find a link for our distillation model in Excel. It is an exceptionally large download so it may take some time if you are not
using a high speed internet connection. Because it takes some time for the iterations to run to completion, calculations are not continuous
within the spreadsheet. Be patient as it may take a few seconds for the calculations to run when operational parameters change. Also,
because of the nature of the iterative process Excel uses to determine the cell values, large deviations may cause the model to crash resulting
in #Error# for many cells. If this occurs, close the model and re-open it from the website to try different parameters.
In order to demonstrate dynamic changes occuring during operation, step changes are incorporated into the Excel model. To show the
effects of a feed change, input the change into cell C-18. This change will occur at t=2min for a length 2 minutes and then the feed flowrate
will revert back to the original value. To demonstrate a change in the feed compostion at t= 6min, input the change into cell C-19 on the
Excel model. This compostion change will be constant over the remainder of the time and will not revert to the previous value. Notice the
changes occuring in the graphs as these step changes occur. The Excel model can be found here => Media:ODEDistillationModel-Final.zip
This model has a specified control range because typical operation of distillation columns requires operation within a specific range of
values where input parameters marginally vary. It is recommended to use the following variables for the initial model.
Feed Flow to Column = 50.00
Feed Composition (xF) = 0.5
Reflux Flow = 15
Bottoms Flow @ T= 0 = 35.00
Mtray = 10
Mcondenser = 50
Mreboiler = 50
qF = 0.4
Feed Temperature = 80
Reflux Temp = 80
Initial Column Temp = 80
Qreboiler = 100
Avoid extreme changes in the initial values, otherwise a significant error may occur causing the model to crash. The principal cause of this
is due to the heat effects which are taken into account in the model and the fact that the number of trays on the column is set at 6.
6.5.5: TERMS COMMONLY USED IN DISTILLATION
Active tray area: The region of the tray where the upward moving vapor comes into contact with the downward flowing liquid (where
the mass transfer occurs)
Downcomer: Area on side of trays where liquid flows down through the distillation column
Disturbance (wrt distillation): Any minor change in the distillation column caused by an external or internal source that causes product
variability
Flooding: Liquid from the active tray area is carried up into the vapor stream (occurs at low L/V ratios)
Ratio Control: Controlled ratio of two manipulated variables
Tray Fouling: Active area of tray is deminished, thus reducing separation efficiency within the column
Upset: Any major change in the distillation caused by an external source that produces erratic column operation that requires manual
override to gain control
Weeping Points: Liquid from the active tray area seeps downward through the tray instead of flowing through the downcomer (occurs
at high L/V ratios)
 EXAMPLE 6.5.1
Using the Excel Model of a Distillation Column posted above, see how a change in the feed concentration upsets the steadystate
equilibration of the inital flow and how the column thereby adjusts to a change in feed composition at the time step indicated. Then
determine the effect of changing the feed flowrate and reflux flow at the indicated times during the time the column is running. Also,
determine the effect of making a larger step size for use in Eulers method.
Remember: an excessive change may cause the distillation column to blow up, so take care in determining reasonable changes. Input
the changes in the flowrate or concentration into C17-C20 of the spreadsheet. Input the new step size in cell D21.
Solutions
The normal concentration output for the distillation column is shown below. Feed flow rate is 50 mol/min, reflux flowrate is 15
mol/min, feed composition is 0.5 mol fraction of the light component. Step size is 0.01 minutes.
Students should see that an impulse change in flowrates does not really change the steady state profile of the column. However, a
change in feed concentration will cause the column to jump and then equilibrate for the new concentration values. Below is a graph
with a change in the feed flowrate to 100 mol/min, and a concentration change to 0.6 of the light component.
Larger step sizes will cause the model to crash because the error associated with the derivatives becomes larger with each derivative
taken. Below the step size is 0.05 minutes. At some point the derivatives become uselsess data and therefore the distillation column
model blows up with outrageous values as shown below.
 EXAMPLE 6.5.2
As a new hire at a specialty chemical company you are placed on project where you are asked to design and specify the controls on a
state of the art distillation column used to separate a binary system. Your supervisor gives you a few clues to the control mechanism
based on the chemistry occurring.
The salable component is the heavy component and the mixture has a low relative volatility.
The light component is not usable at all, however it is extremely toxic and must be controlled precisely.
This is a small operation which requires a low pressure separation.
The facility has had issues with the onsite steam that is typically used to heat the reboiler.
Draw the P&ID model of how you would model this column, mention all the items you must consider for the process and what you
would do to control them.
6.5.5.1: ANSWER
Since the heavy component control is desired you will want a feedback loop based on the bottoms composition controlling reboiler
duty. Also note how precise control of the vapor flowrate up the column can be best accomplished through reboiler control. Because it
is low pressure you must begin to worry about entrainment and the possibility of flooding the column. Therefore there should be
pressure sensors controlling the pressure in the column to maintain it at optimal levels using a packed column instead of a tray column
to produce a better separation of the components. Because onsite steam has been shown to be a problem you would probably want to
have a redundancy built into the reboiler steam or perhaps have its own autonomic steam delivery because this could potentially
jeopardize the quality of your product.
6.5.6: REFERENCES
Bequette, B. Wayne. Process Dynamics Modeling, Analysis, and Simulation, New Jersey: Prentice Hall
Perry, Robert H. Perry's Chemical Engineers' Handbook, 7th Ed. McGraw Hill. ISBN 0-07-049841-5
Riggs, James B.; Karim, M. Nazmul. Chemical and Bio-Process Control, 3rd Ed., Texas Tech University, Ferret Publishing

Authors: Jennifersue Bowker, George Cater, Kibwe John
6.5: ODE and Excel model of a Simple Distillation Column is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf
6.6: ODE AND EXCEL MODEL OF A HEAT EXCHANGER
6.6.1: INTRODUCTION
In process industries, heat exchangers are designed to transfer heat from one fluid to another. Heat exchangers have many different
applications, especially in chemical processes, air conditioning, and refrigeration. They are classified by their design and type of flow. In
some types of heat exchangers, the two fluids are separated by a wall or membrane and the heat transfer takes place by both convection and
conduction. In another, less common type of exchanger, the two fluids physically come into contact with each other as the heat transfer
occurs.
Since heat exchangers have a wide variety of applications and are commonly used in industry, control of the system is essential. A dynamic
model may be created to allow the chemical engineer to optimize and control the heat exchanger. By utilizing this model, predictions can be
made about how altering the independent variables of the system will change the outputs. There are many independent variables and
considerations to account for in the model. If done so correctly, accurate predictions can be made about the system.
6.6.2: TYPES OF HEAT EXCHANGERS

1. Double - Pipe Heat Exchanger
A double-pipe heat exchanger is the simplest type of heat exchanger and can operate with co-current (Figure 1) or counter-current (Figure 2)
flow. The design consists of a single small pipe (tube-side) inside of a larger one (shell-side). A co-current heat exchanger is most
commonly used when you want the exiting streams to leave the exchanger at the same temperature. A counter-current heat exchanger is
used more often than co-current because they allow for a more efficient transfer of energy.
2. Shell-and-Tube Exchanger
A shell-and-tube exchanger is used for larger flows, which are very common in chemical process industries. The design of this exchanger is
a shell with a bundle of tubes inside. The tubes are in parallel and a fluid flows around them in the shell. There are many different
arrangements such as straight (Figure3) or u-tube (Figure 4). Each arrangement allows for a different type of flow such as co-current,
counter-current and cross flow. The tube-side can have one or more passes to increase the energy exchange from the tube-side fluid. The
shell-side may contain baffles, or walls, that channel the fluid flow and induce turbulence, and thus, increase energy exchange. Correlations
can be developed to predict the increase in energy exchange. These correlations are out of the scope of this article.
3. Cross-flow Exchanger
The most common application for a cross-flow heat exchanger is when a gas is being heated or cooled. This device consists of multiple
tubes in parallel, usually containing a liquid, and the exterior gas flows across the tubes. In some cases the air is confined in separate flow
chambers, such as fins, and in others it is open to flow freely (Figure 5).
6.6.3: DYNAMIC MODELING OF HEAT EXCHANGERS

Since heat exchangers are so widely used in industry, it necessary for a chemical engineer to be able to optimize and control the system and
know how independent variables will affect the outputs from teh system. To do this, a dynamic model is developed and utilized.
A dynamic model of a heat exchanger may be used, for example, to predict how a change in the fluid flowrates or the addition of an
insulating jacket will affect the outlet temperature of the product stream. The model uses ordinary differential equations (ODEs) to describe
the process and, using a program like Microsoft EXCEL, gives plots of the variables vs. time for the entire process. There are many
independent variables in a heat exchanger, which can cause modeling to be very complex since multiple ODEs are required to define all of
the process variables.
Some of the independent variables in a heat exchanger system include:
Shell-side:
fluid
flowrate
temperature
number of baffles
Tube-side:
fluid
flowrate
temperature
number of passes
Flow configuration
Co-current
Counter-current
Cross flow
Insulating jacket
The primary dependent variable of concern is the outlet temperature of the product stream (usually the tube-side fluid.) The outlet
temperature, in a controlled system, is monitored by a sensor. The sensor then transmits a signal to an actuating device of one or more of the
independent variables (usually shell-side flow controller) to perform some desired response.
6.6.4: MODELING WITH ODES

The following section outlines the method for developing a dynamic model of a heat exchanger. The model is for a double-pipe heat
exchanger that has the ability to flow in co-current or counter-current configurations and the option of an external insulating jacket. It
assumes constant fluid properties and perfect heat transfer through the metal of the tubing. The outlet temperature of the tube-side fluid is
monitored by a temperature sensor, and the flow rate of the shell-side fluid is controlled by a actuated flow-controlling device.

An energy balance is first performed on the tube-side fluid.
Rate of accumulation of thermal energy in tube-side fluid = rate of energy in − Rate of energy out (6.6.1)
− Heat transferred from shell-side
The left-most term in the energy balance in Equation 6.6.1 is the amount of thermal energy that accumulates in the tube-side fluid and
causes a change in its outlet temperature. The terms on the right side of the above energy balance describe the thermal energy of the fluid
flowing in and the fluid flowing out and the amount of heat transfer from the shell side fluid. In the term for heat transfer from the shell-
side, the temperatures are the temperatures of the outlet streams. The outlet temperatures will change according to whether you are running
co-currently or counter-currently. The energy balance is written as:
dTt,out kAi
mcp,t = ρcp,t Ft,in Tt,in − ρcp,t Ft,out Tt,out − (Tt,out − Ts,out ) (6.6.2)
dt Δz
where,
mass of the fluid =
constant pressure heat capacity of the fluid
Temperature
time
conductive heat transfer coefficient
surface area of tube that fluid contacts
length of tube
density of the fluid
volumetric flowrate of the fluid
and subscripts denote
- tube-side fluid
- outlet
- inlet
- inside
An similar energy balance is next performed on the shell-side fluid.
Rate of accumulation of thermal energy in tube-side fluid = rate of energy in − Rate of energy out (6.6.3)
− Heat transferred to tube-side fluid − Rate of heat loss to the surroundings
The left-most term in the above energy balance is the amount of thermal energy that accumulates in the shell-side fluid and causes a change
in its outlet temperature. The terms on the right side of the above energy balance describe the thermal energy of the fluid flowing in and the
fluid flowing out, heat transfer to the tube-side fluid, and also the heat lost by convection to the surroundings. The energy is written as;
dTs,out k ⋅ Ao
mcr,s − ρcp,s Fs,sn Ts,in − ρcp,s Fs,cut Ts,out − (Ts,cut − Tt,out ) − h. As (Ts − T∞ ) (6.6.4)
dt Δz
where,
= coefficient of convective heat transfer for air
- conductive heat transfer coefficient.
and subscripts denote
- shell-side fluid
- air
- outside of tube
- outside of shell
6.6.4.2: CONSIDERATIONS
There are considerations and simplifications you can make in order to solve the differential energy balances. The validity of these
assumptions depends on how accurate of a model you require.
1. The heat capacity of the fluid may be temperature dependent. If this is the case, a polynomial equation could be written for the Cp
values of each of the fluids. It would take the form-
2 3
Cp = a + bT + cT + dT (6.6.5)
Values of a, b, c, and d are properties of the fluid and can be found in Perry's Chemical Engineers' Handbook.
It should also be noted that if the fluids in the process are gases, their fluid properties will be affected by changes in pressure and the use of
a constant Cp value would be inappropriate.
2. The density of the fluid may be temperature dependent. This would be likely if the fluid were a vapor, as in the case of using steam as
the shell-side fluid to heat the tube-side process fluid. If this is the case, a differential equation could be written for the value of the fluid
and would take the form-
dρ
∝ C (6.6.6)
dT
where C is the coefficient of cubic expansion (relates kinetic energy to temperature).

3. Heat loss to the surroundings may be neglected. This would be the case if the heat exchanger is well insulated or if the shell-side fluid
is about the same as ambient temperature. In this case-
hA (Ts − T∞ ) = 0 (6.6.7)
4. The temperature sensor may have an inherent time delay. This means the temperature output from the sensor lags the actual
temperature at the time of the reading. The temperature lag may be accounted for with the differential equation-
dToutlet,sensor 1
= (Toutlet,actual − Toutlet,sensor ) (6.6.8)
dt τT s
where = time constant for the temperature sensor. is a process parameter and usually ranges from 6 to 20 seconds depending on the
properties of the sensor. This value would either be given in literature from the manufacturer or would have to be determined through
experimentation.
5. The actuator system for the control valve may have a slow dynamic response. When the actuator system is based on a control valve,
the response to the pressure change is slower than the flow through the valve. The flow rate through a control valve can be modeled with the
differential equation-
dQactual 1
= (Qsetpoint − Qactual ) (6.6.9)
dt τv
where = time constant for the flow control valve. is a process parameter and usually ranges from 0.5 to 2 seconds depending on the
properties of the sensor. This value would have to be determined through experimentation.
6.6.5: USING EXCEL TO SOLVE ODES

Since temperature of the heat exchanger varies across the metal as well as along the length of the pipe, a partial derivative for how
temperature varies with length must be described. In order to solve this problem using excel, an approximation of length was used. Instead
of taking the partial derivative across the length with respect to temperature, the pipe was divided into differential segments, Δz. Ideally Δz
is an infinitesimally small cross section of the length of the heat exchanger. We will assume that through this differential segment, the
temperature of the liquid that leaves the segment is the same as the temperature of the liquid within the segment. Since we are assuming the
same temperature for the exiting streams as the inside of the segment, the choice of length for these Δz’s helps dictate the accuracy of the
solution. The combination of these differential units allows us to model a heat exchanger without the use of partial derivatives. The figure
below shows an example of the simplification where the heat exchanger is split into three segments (Figure 6).
To begin modeling, start by solving the appropriate energy balance (either Equation 6.6.2 or 6.6.4 ) for and make any simplifications
necessary. Then use a method to solve ODEs in Excel, such as Euler's or Heun's. Go here: ExcelModelingODE for further explanation on
how to set up an Excel spreadsheet using these methods. For each unit, Δz, and each fluid in that unit of the heat exhanger, you should have
an approximation equation. Because ideally Δz represents an infinitesimal section, it follows from the first assumption that we can take the
temperature in Δz as being the exit temperature of the hot and cold streams for the Δz ’s respectively.
The equation for Euler's Method takes the form:
dT
Th1 (ti+h ) = Th1 (ti ) + Δt [ (ti )] (6.6.10)
dt
The equation for Heun's Method takes the form:

1 dT 1 dT
Th1 (ti+1h ) = Th1 (ti ) + Δ [ (ti Ti ) + (ti+h Ti+1hi )] (6.6.11)
2 dt 2 dt
The temperatures can then be plotted versus time in order to model how the system inputs affect heat exchange. The following examples
walk you through a model of a simple heat exchanger in Excel.
 EXAMPLE 6.6.1: EULER'S METHOD
As a process engineer in a chemical processing company, you were assigned to scale up a process for your company’s latest and
greatest chemical. Between one of the steps of the reaction, it is necessary to quickly and efficiently cool down the solution from 330 K
before allowing it to enter the next step. From your previous knowledge, you know that a heat exchanger is necessary and the counter
current heat exchanger is the most efficient to achieve a desired temperature. You decide to use cold water entering in at 250 K as your
coolant since it is relatively inexpensive. In order to allow for the plant engineers in the future to have more control, create a model to
determine when this process will go to steady state and graph the temperature profiles of the outlet streams. You can assume the heat
exchanger is perfectly insulated and that there is perfect conduction through the metal from shell-side fluid to the tube-side fluid, and
this conduction is described by the heat transfer coefficient, k, between the two fluids.
General Relevant Information:
Tair=296.15 K (Temp. of the Air)
ri=0.1 m (Inner Diameter of Inner Tube)
ro=0.12 m (Outer Diameter of Inner Tube)
ro'=0.15 m (Outer Diameter of Outer Tube)
Δz=1 m (Increment of length)
k=450000 W/m2*K (heat transfer coefficient between shell and tube-side fluids)
Shell Side Fluid Information (Cold Water)
Cps=4185 J/kg*K (shell side fluid heat capacity)
T0s=250 K (input temp. of shell side fluid)
As=0.02543 m (cross sectional area where shell side fluid is present)
ρs=1000 kg/m^3 (density of shell side fluid)
Fs=0.1 m3/s (volumetric flow rate of shell side fluid)
Tube Side Fluid Information (Solution)
Cpt=1200 J/kg*K (tube side fluid heat capacity)
T0t=330 K (input temp. of tube side fluid)
At=0.0314 m (cross sectional area where tube side fluid is present)
ρt=1030 kg/m^3 (density of tube side fluid)
Ft=0.2 m3/s (volumetric flow rate of tube side fluid)
Solution
The general Algorithm can be followed to develop the Excel model.
1. Make a spreadsheet that has all relevant process information (ie: T inputs, ρ values, etc).
2. Perform an overall energy balance for the tube-side and shell-side fluids. Make simplifications and state assumptions. Solve for
3. Divide the heat exchanger into smaller segments of length Δz

4. In your spreadsheet, set up a column for time. In the first row, start at t=0 and then increase time by Δh down the column.
5. Set up a column for the incoming tube-side temperature, Tt0. The value of incoming temperature is inputted by the user. Do the
same for the shell-side temperature, Ts0.
6. Set up a column for the exiting tube-side temperature from each Δz unit. Do the same for the shell-side temperatures. For each of
these temperature columns:
a. In the first row, make an initial guess of the temperature. This is usually done by setting the temperature equal to the temperature
of the stream coming into the heat exchanger.
b. In subsequent rows, input the Euler’s equation (Eq 10) by referencing the appropriate cells for the process information and take
all temperatures from time, ti, which are the temperatures in the row above.
7. Make a graph of variables of interest (Temperatures) vs. time.
The graph below illustrates how the exiting hot and cold temperature changes over a short period of time.
For a more detailed view of this process please see the attached spreadsheet:
 EXAMPLE 6.6.2: HEUN'S METHOD
Again, you are working on the same process, but your supervisor requires you to have a more accurate description of the dynamics of
the system. Being as bright as you are, you decide to utilize the 2nd order Runga Kutta method, known as the Heun's method.
Solution
Follow the Algorithm in Example 1, except in step 6b, use the equation for Heun's method, Eq 11.
The graph below illustrates how the exiting hot and cold temperature changes over a short period of time.
 EXAMPLE 6.6.3
A month after you developed a model of the heat exchanger used in the process for your company's latest and greatest chemical, one of
the plant engineers calls you and says that at t=2s into the process, the temperature of the incoming fluid changes abruptly from 330K
to 300K. How do the temperatures of the outlet streams react to this change. Use the information and model from Example 1.
Solution
The Excel spreadsheet can be modified so that at time t=2s, the incoming temperature of the tube-side fluid is 300K. The graph below
illustrates how the exiting hot and cold temperature changes over a short period of time.
Tube-side Temp Change
6.6.6: REFERENCES
Riggs, James B.; Karim, M. Nazmul. Chemical and Bio-Process Control, Texas Tech University, Ferret Publishing.
Authors: Teresa Misiti, Carly Ehrenberger, Danesh Deonarain, Georgina Mang
6.6: ODE and Excel model of a Heat Exchanger is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via
6.7: ODE AND EXCEL MODEL OF AN ADIABATIC PFR
6.7.1: INTRODUCTION
A plug flow reactor (PFR) is a tubular reactor used in chemical reactions. Reactants enter one end of the PFR, while products exit from the
other end of the tube. A PFR is useful because of high volumetric conversion and good heat transfer. PFRs carry out power law reactions
faster than CSTRs and generally require less volume. Excel modeling for an adiabatic plug flow reactor is useful for estimating conversion
as a function of volume. By inputting the values of the constants and iteratively changing the volume, the conversion at a specific volume
can be determined. This removes the need for complicated hand calculations and helps the user visualize the reaction and notice trends
through the PFR. Alternatively, excel modeling can also determine the temperature inside the reactor as a function of volume. Euler’s
integration is used to estimate the conversion or temperature through the PFR. By ensuring that the volume intervals are small, the
conversion or temperature values generated should be fairly accurate. Refer to the Euler's method section for more information on its
implementation.
6.7.2: BASIC ALGORITHM TO MODEL ADIABATIC PFR
The model presented will include the following assumptions:

1. Negligible pressure drop
2. Single elementary reaction
3. Power law kinetics
4. Constant fluid properties (i.e. heat capacity)
To describe the basic algorithm for an adiabatic PFR, a simple case will be considered where pure A enters the reactor:
aA −
↽⇀
− aB
1. Reactor Type
PFR
2. Limiting Reagent
The design and rate equations should be expressed in terms of the limiting reagent. Here, A is the limiting reagent.
3. Design Equation
The design equation for a PFR can be expressed in terms of several variables, including conversion, moles, and concentration. It is
important to note that the design equation in terms of conversion may only be used if one reaction is occurring. Otherwise, it must be
expressed in terms of moles or concentration. The equation to describe conversion as a function of volume, as derived from a simple mole
balance, is shown below in Equation 6.7.1 .
dX −rA
= (6.7.1)
dV FA O
X= conversion
V= volume
rA= reaction rate of A
FAo= initial moles of A
4. Rate Law
The rate law is independent of reactor type. It is expressed in terms of a rate constant and concentration. If the reaction is irreversible, the
rate law is modeled according to Equation 6.7.2 . If the reaction is reversible, the rate law models Equation 6.7.3 in which the concentration
of B and the equilibrium constant must be accounted for.
a
−rA = kc (6.7.2)
A
b
c
a B
−rA = k (c − ) (6.7.3)
A
Kc
where
k= rate constant
= concentration of A with stoichiometric coefficient a
= concentration of B with stoichiometric coefficient b
KC= equilibrium constant
If the reaction is not isothermal, the rate constant and equilibrium constant should be written in terms of temperature as shown in Equation
6.7.4 and 6.7.5 .
E 1 1
k = k0 exp( ( − )) (6.7.4)
R T0 T
ΔHRX 1 1
KC = KC1 exp( ( − )) (6.7.5)
R T1 T
where
k0= rate constant at T 0
E = activation energy
R = ideal gas constant
T = temperature
K C1=equilbrium constant at T 1
ΔH RX = standard heat of reaction

5. Stoichiometry
Stoichiometry is used to express concentration of a species in terms of concentration of the limiting reactant. The stoichiometry equation
depends on whether the reaction occurs in the liquid or gas phase. If it occurs in the gas phase, Equations 6.7.6 and 6.7.7 are used. For
liquid phase, ε becomes zero and the correlation reduces to Equation 6.7.8 . There is no pressure term included because it is assumed that
pressure drop is negligible, and the initial pressure is equal to the final pressure.
(1 − X) T0
CA = CAo (6.7.6)
(1 + ϵX) T
b
( θB + X) T
a 0
CB = CAo (6.7.7)
(1 + ϵX) T
CA = CAO (1 − X) (6.7.8)
b
CB = CAo (θB + X) (6.7.9)
a
where
ε = y
∗
Ao
(stoichiometric coefficients of products - coefficients of reactants)
yAo = initial mole fraction of A
θi = ratio of initial moles (or concentration) of species i to initial moles (or concentration) of A
6. Combine
The design equation, rate law, and stoichiometric equations are combined. For example, assume the reaction is irreversible, liquid phase,
and isothermal, with pure A entering the reactor and has the combined equation as shown below.
a a
dX k ∗C (1 − X)
A0
= (6.7.10)
dV FA0
7. Energy Balance
The energy balance can be used to relate temperature and conversion. It assumes heat flow and shaft work are not present.
X (−ΔHRX ) + ΣΘi Cpi To + XΔCp Tr
T = (6.7.11)
ΣΘi Cpi + XΔCp
with
Cpi= heat capacity of species i
ΔCp= Cp products- Cp reactants
If $ΔC_p= 0$, the energy balances reduces to the following:
X (−ΔHRX ) + ΣΘi Cpi To
T = (6.7.12)
ΣΘi Cpi
8. Solve
Solve the system of equations in Excel. The ODE is approximated by Euler's method. See the attached spreadsheet for the ODE/Excel
modeling of the adiabatic PFR.
 EXAMPLE 6.7.1
A plant at the ETF (Equipment Testing Facility) has been acquired in a hostile takeover by Brown Industries, a company for which you
are a process engineer. The Engineering Manager, Phil Savage, has assigned you to work on a PFR located within this new plant. The
particular PFR in question is used to conduct the following reaction:
A ⟶ B
He wants you to provide him with two pieces of information. First, he wishes to know how far down the reactor must the reaction go to
achieve 80% conversion. Secondly, he wants to know the temperature at this point.
You decide that first it would be best to ascertain all the thermodynamic and kinetic properties that are inherent to this reaction. Instead
of looking them up in Perry's or your old ChemE books, however, you decide to ask the Lead Plant Engineer, Pablo LaValle. Pablo tells
you the following:
"Listen wisely to what I say, young grasshopper. This reaction is irreversible, liquid-phase, and elementary. We will feed the reactants
into the tube reactor of diameter 50cm at 273 K. We will begin with an initial concentration of A at 1.6 mol/L and an initial molar flow
rate of A at 2 mol/min. The heat of reaction at 300 K is -20 kJ/mol A. The rate constant at 300K is 0.2 mol / L * min. The activation
energy at 300K is 15.2 kJ/mol. The Cp values for the products and reactants are as follows.
Cp A = 200 J / mol K
Cp B = 100 J / mol K
"Go forth and use the Excel sheet that Group B has created and your job will be very easy."
Solution
The model for this example can be found here.
Pablo gave all the data needed to complete the model. After inputing the data, the final page of the spreadsheet was used to determine
the reactor conditions Savage desired.
X = 0.8 at a reactor length of 33.4 meters. The temperature at this point of the reactor is 388.4K
 EXAMPLE 6.7.2
You have just completed your initiation in the ultra-super secret elite society known as the League of Chemical Engineers which is
headquartered high in the Himalayans. They have an adiabatic PFR (diameter = 10cm) in their possession which they use to produce
two gases, known as B and C, from a gas-phase reactant known as A. The reversible gas reaction is shown below:
A < = ⟶ B+C
Your sensei wishes to know what the maximum conversion possible is and also what length of reactor would be needed to reach a
conversion of 15%. First, however, he orders you to climb K2 where the reaction data is stored by the society. After a perilous 5 day
journey you discover the following information:
Cp A = 40 kJ/mol K
Cp B = 25 kJ/mol K
Cp C = 15 kJ/mol K
CAO = 0.271 mol/L
FAO = 5 mol/min
Initial Temp = 450K
Heat of reaction at 300K = -20 kJ/mol A
Activation energy = 31.4 kJ/mol
Keq at 300K = 0.01 mol/L
k_forward at 300K = 0.133 mol /(L*min)
Solution
After climbing back down K2, and finding a computer, the data obtained can be input into the adiabatic model. The solution can be
found here.
Be sure to set the values for both gas phase reaction and reversible reaction to 1.
Using the data table found on the last page of the spreadsheet, the desired information can be extracted.
The maximum conversion that can be achieved is X = 0.187
The reactor requires 64.8 meters to achieve a conversion of 15%.
6.7.3: REFERENCES
Fogler, H. Scott. Elements of Chemical Reaction Engineering. Pearson Education, Inc., 2006.

Brian McQuillan, Crystal Miranda, Brandon Quigley, and John Zhang
6.7: ODE and Excel Model of an Adiabatic PFR is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via
6.8: MODELING AND PID CONTROLLER EXAMPLE - CRUISE CONTROL FOR
AN ELECTRIC VEHICLE
6.8.1: INTRODUCTION
Controls principles developed in this course can be applied to non-chemical engineering systems such as automobiles. Some companies,
such as NAVTEQ, are developing adaptive cruise control products that use information about the upcoming terrain to shift gears in a more
intelligent manner which improves speed regulation and fuel economy. This case study will examine the basics of develop a speed
controller for an electric vehicle.
An electric vehicle was chosen for the following reasons:
Electric vehicles are interesting from an engineering perspective and may become a reality for consumers in the future
Torque produced by an electric motor is instantaneous (for all practical purposes). Thus actuator lag can be ignored, simplifying the
development of said controller.
Some electric vehicles feature motors directly integrated into the hub of the drive wheel(s). This eliminates the need for a transmission
and simplifies vehicle dynamics models.
6.8.2: FORCES
As shown in the free body diagram below, there are six forces acting on the vehicle:
1. Rolling Resistance
2. Aerodynamic Drag
3. Aerodynamic Lift
4. Gravity
5. Normal
6. Motor
Figure 1: A free body diagram of the forces acting on the vehicle. −θ is used to denote the grade of the road such that a positive value of θ
corresponds to the vehicle traveling uphill.
6.8.2.1: ROLLING RESISTANCE
Rolling resistance is due the tires deforming when contacting the surface of a road and varies depending on the surface being driven on. It
can be model using the following equation:
FRR = Crr1 v + Crr2 FN
The two rolling resistance constants can be determined experimentally and may be provided by the tire manufacture. is the normal
force.
6.8.2.2: AERODYNAMIC DRAG
Aerodynamic drag is caused by the momentum loss of air particles as they flow over the hood of the vehicle. The aerodynamic drag of a
vehicle can be modeled using the following equation:
is the density of air. At 20°C and 101kPa, the density of air is 1.2041 kg/m3.
is the coefficient of drag for the vehicle times the reference area. Typical values for automobiles are listed here.
is the velocity of the vehicle.
6.8.2.3: AERODYNAMIC LIFT

Aerodynamic lift is caused by pressure difference between the roof and underside of the vehicle. Lift can be modeled using the following
equation:
is the density of air. At 20°C and 101kPa, the density of air is 1.2041 kg/m3.
is the coefficient of lift for the vehicle times the reference area.
is the velocity of the vehicle.
6.8.2.4: GRAVITY
In the diagram above, there is a component of gravity both in the dimension normal to the road and in the dimension the vehicle is traveling.
Using simple trigonometry, the component in the dimension of travel can be calculated as follows:
FG,travel = mg sin(−θ)
m is the mass of the vehicle.

g is the acceleration due to gravity.
6.8.2.5: NORMAL FORCE
The normal force is the force excerted by the road on the vehicle's tires. Because the vehicle is not moving up or down (relative to the road),
the magnitude of the normal forces equals the magnitude of the force due to gravity in the direction normal to the road.
m is the mass of the vehicle.

g is the acceleration due to gravity.
6.8.2.6: MOTOR
The torque produced by an electric motor is roughly proportional to the current flowing through the stater of the motor. In this case study,
the current applied to the motor will be controlled to regulate speed. Applying a negative current will cause the vehicle to regeneratively
brake.
τ = kmotor I
τ kmotor I
FM = =
r r
is the torque produced by the motor.

is the current flowing through the motor.
is the radius of the tire.
is a constant.
6.8.3: NEWTON'S SECOND LAW

Using Newton's Second Law, a differential for the vehicle's speed can be obtained.
ma = ∑ F
= FM − Fdrag − FRR + FG,travel
Substituting in the expressions for various forces detailed above yields the following:
Further substituting the expression for normal forces yields the following:
Substituting for results in the following:
Grouping like turns results in the following:
In order to simply the remaining analysis, several constants are defined as follows:
β = rCrr1
γ = r Crr2 mg cos(−θ) − rmg sin(−θ)
Substituting these into the differential equation results in the following expression:
2
dv τ − αv − βv − γ(θ)
=
dt mr
It is important that θ (and thus also γ) is a function of vehicle position.
6.8.4: PID CONTROLLER

One way to regulate vehicle speed is to control the torque generated by the electric motor. If the motor torque is greater than the resistive
torque acting on the vehicle (a summation of aerodynamic drag, rolling resistance, etc.) the vehicle will accelerate. If the motor torque is
less than the resistive torque, the vehicle will slow down.
6.8.4.1: EXPRESSION FOR PHASE CURRENT

For an electric motor, the phase current flowing through the motor is proportional to the torque produced. Thus one strategy for controlling
the vehicles speed is to controller the motor phase current.
Using a PID controller architecture, the expression for motor current is the following:
1 d (vset − v)
I = Kc (vset − v) + ∫ (v − vset )dt + τD + Coffset
τI dt
6.8.4.2: DIFFERENTIAL EQUATION FOR VELOCITY

Substituting this expression into the differential equation for vehicle position results in the following:
1 d(vset −v)
2
dv kmotor [Kc (vset − v) + ∫ (vset − v)dt + τD + Coffset ] − αv − βv − γ(θ)
τI dt
=
dt mr
Defining another variable x allows for the removal of the integral from the expression.
1
dx1
= vset − v
dt
x1 d(va=t −v)
2
kmotor [Kc (visct − v) + + τD + Coffset ] − αv − βv − γ(θ).
dv τl dt
=
dt mr
dvset
If all changes in vset are gradual then dt
≈ 0 . Applying this simplification results in the following expression:
x1 dv 2
kmotor [Kc (vset − v) + − τD + Coffset ] − αv − βv − γ(θ)
dv τ1 dt
=
dt mr
Using a little algebra, an expression for can be obtained as follows:
α
kmotor 2
kmotor Kc (v x1 |kmotor C ) (θ)
dv set cffses
=
dt ˙
kmotor lL + mr
6.8.4.3: FIND FIXED POINT

The fixed point can be obtained by setting the derivatives to zero and solving the system of equation.
System of Equations:
kmotor 2
dv kmeclor Kc (vsel − v) + x1 + kmelur Cv ∫ js≤ι
− αv − βv − γ(θ)
τI
= = 0
dt kmotor tD + mr
x1
= vset − v = 0
dt
Solution
2
− τI ( kmotor Coffset − αvset − βvset − γ(θ)])
x1 =
kmotor
v = vset
As expected, a fixed point exists when the set velocity equals the actual velocity of the vehicle.
6.8.4.4: LINEARIZE SYSTEM OF ODES

Before the stability of the system can be (easily) examined, the system must be linearized around a fixed point.
Note: To simply the matrix expressions in section the following notation will be used:
dyi
′
y =
i
dt
Overall a linearized system of ODEs has the following form:

′
y y1 k1
⎡ 1 ⎤ ⎡ ⎤ ⎡ ⎤
⎢ y′ ⎥ ⎢ y2 ⎥ ⎢ k2 ⎥
⎢ 2 ⎥
⎢ ⎥ = J⎢ ⎥+⎢ ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢ ⋮ ⎥ ⎢ ⋮ ⎥ ⎢ ⋮ ⎥
⎣ ′ ⎦ ⎣ ⎦ ⎣ ⎦
yn yn kn
The first step of linearizing any system of ODEs to calculate the Jacobian. For this particular system, the Jacobian can be calculated as
follows:
′ ′
∂v ∂v
− kmotor Kc −mr[2αv+β] kmotor
⎡ ∂v ∂ x1
⎤
mr+kmotor τD tl (mr+ kmotor τD )
J = ′ ′
= [ ]
∂ x1 ∂ x1
⎣ ⎦ −1 0
∂v ∂ x1
The Jacobian is then evaluated at the fixed point:

− kmotor Kc −mr[2α vset +β] kmotor
mr+kmotor τD tI (mr+ kmotor τD )

J = [ ]
−1 0
The next step if to calculate the vector of constants. For this particular system, said vector can be calculated as follows:
2
−(− kmotor Kc −2α vset −β])vset (kmotor Coffset −α v −βvset −γ (θ)])
set
kv +
[ ] = [ (mr+ kmotor τD ) (mr+ kmotor τD ) ]
kx
1
vset
Combining the Jacobian and vector of constants results in the following linearized system:
′
v v kv
[ ] = J[ ]+[ ]
′
x x1 kx1
1
6.8.4.5: STABILITY ANALYSIS

To assess the stability of the controller, the eigenvalues of the Jacobian in the linearized systems of ODEs can be examined. In general, an
eigenvalue (lambda) is the solution to the following equation:
|J − λI| = 0
Using a computer to solve said equation, the eigenvalues of this particular system can be found to be the following:
For the system to be stable, the real component of all eigenvalues must be non-positive. The following inequality must be true for a stable
controller:
For the system to not oscillate, the imaginary component of all eigenvalues must be zero. The following inequality must be true for a non-
oscillating controller:
2 2 2 2 2 2
τI (k Kc τI + 2kmotor Kc τI β + 4kmotor Kc τI αvset + β τI + 4βτI αvset + 4α v τI − 4kmotor mr − 4k tD )
motor set motor
Interestingly, neither of these criteria depend on the grade of the road (θ). However, during the analysis, it was assumed that θ is
constant. For most roads, this is not the case; θ is actually a function of vehicle position. In order to add this additional level of detail, the
original system of ODEs needs to be revised:
dv kmotor Kc (vset , v) |kmotor x1 | kmotor Coffset
=
dt kmotor lD + πur
dx1
= vset − v
dt
ds
= v
dt
θ = f(s)
Unfortunately for any normal road, the grade is not a simple (or even explicate) function of position (s). This prevents an in depth analytical
analysis of stability. However for a very smooth road with very gradual changes in grade, the stability of the controller should be unaffected.
6.8.4.6: EXAMPLE ELECTRIC VEHICLE
Simulating the system also for other properties of the controller to be examined. This section presents the results from simulating a specified
fictitious electric vehicle.
Parameters
For the fictitious electric vehicle simulated in this analysis, the following parameters were used. These parameters are roughly based on the
parameters one would expect to see in a typical electric automobile.
Root Locus Plots

Using the parameters above, Root Locus Plots of the system were constructed to numerically explore the stability. The following controller
constants were used when constructing the plots:
These Root Locus plots show that the controller with said constants is both stable and does not oscillate.
Phase Portrait
A phase portrait of the system with the following parameters was also constructed:
The phase portrait shows that the system is both stable and does not oscillate, as predicted by the Root Locus plots.
Driving Simulation on Level Terrain
The vehicle was simulated starting at 10 m/s and accelerating to 25 m/s via cruise control on level terrain ( ). For this simulation, the
following constants were used:
2
αv + βvset + γ(θ = 0)
set
Coffset = 146 ≈
kmotor
Interestingly this graph shows oscillation, despite the Root Locus plots and phase diagrams. It is important to remember, however, that the
Root Locus plot and stability methods involve linearizing the system. It is possible the linearized system is not a good approximation of the
system.
It is also important to remember that this particular example involves a large set point change, which can induce oscillations in certain
systems.
Driving Simulation on Unlevel Terrain
In order to explore how the controller behaves on a road with a non-zero grade, a route with hills was constructed from the following
equation, where h is elevation and s is position.:
s s
h = 100 sin( )+
250π 2000
The vehicle was simulated driving said road starting with the following initial conditions and controller constants:
Initial Conditions
Controller Constants
The current controller tuning is inadequate for this road. There are vary large variations in velocity. In order to reduce these variation, the
propartional gain Kc was increased by a factor of 5. Below are the results:
Using the Optimization Toolbox in MATLAB, the controller was optimized by minimizing the sum of the errors (difference between vehicle
velocity and set velocity) on this particular segment of road. Below are the optimized controller constants and a plot of the velocity profile:
This example shows the power of optimization and model predictive control.
6.8.5: SUMMARY
This example demonstrates the following:
1. Modeling a simple, non-chemical engineering system.
2. Develop a PID controller for said physical dynamical system.
3. Manipulate said system to develop an system of differential equations.
4. Find fixed point(s) for said system.
5. Linearize said system of differential equations.
6. Find the eigenvalues of said linearized system.
7. Construct Root Locus plots for said linearized system.
8. Construct a phase portrait for said dynamical system.
9. Simulate said dynamical system under various conditions.
10. Demonstrate the idea of model predictive control by optimize said controller under a specific scenario.
6.8: Modeling and PID Controller Example - Cruise Control for an Electric Vehicle is shared under a CC BY 3.0 license and was authored, remixed, and/or
curated by Alex Dowling via source content that was edited to conform to the style and standards of the LibreTexts platform; a detailed edit history is
6.9: BLOOD GLUCOSE CONTROL IN DIABETIC PATIENTS
6.9.1: BACKGROUND OF WHY INSULIN CONTROL IS IMPORTANT FOR DIABETIC PATIENTS
Diabetes mellitus is a disease of the endocrine system where the body cannot control blood glucose levels. There are two general
classifications of diabetes:
Type I (also known as juvenile diabetes)
Genetic predisposition and/or an autoimmune attack destroys T-cells of pancreas
Body cannot produce insulin to regulate blood glucose
Type II
Most common form of diabetes and has reached epidemic status in the United States
Usually caused by lifestyle
Obesity reduces body's responsiveness to insulin
Treatment for both types of diabetes may include exercise, dieting, oral medications, or insulin injections. Most insulin dependent diabetics
follow a management plan that requires frequent testing of blood glucose levels and then injection of a prescribed dose of insulin based on
the blood glucose level. However, the downside of this treatment method is that there is no predictive control. If blood glucose levels are
falling and insulin is administered, a hypoglycemic episode may occur. Recent biomedical advancements have resulted in continuous blood
glucose monitoring devices as well as insulin pumps. Continuous monitoring allows for finer blood glucose control and can help predict
fluctuations in the blood glucose level. Insulin pumps replace the need to administer insulin injections by automatically injecting a
prescribed dose, however it requires blood glucose level input from the patient. In the future, insulin pumps and continuous blood glucose
monitors may be integrated forming a closed loop control system which will can replace the body's own faulty control system.
Figure 1: Control Schematic of Insulin Infusion
6.9.2: MATHEMATICAL MODEL FOR A CLOSED LOOP INSULIN DELIVERY SYSTEM

The following set of differential equations are known as the Bergman "minimal model":
dG Gmeal
= − p1 G − X (G − Gb ) +
dt V1
dX
= − p2 X + p3 I
dt
dI U
= −n (I + Ib ) +
dt V1
where:
G = deviation variable for blood glucose concentration
X = deviation variable for insulin concentration in a "remote" compartment
I = deviation variable for blood insulin concentration
Gmeal = a meal disturbance input in glucose
U = the manipulation insulin infusion rate
Gb = steady state value of blood glucose concentration
Ib = steady state value of blood insulin concentration
Blood parameters include p1, p2, p3, n, V1(blood volume). These are specific to the blood specimen and must be predetermined.
A linear state space model can be used to express the Bergman equations seen above. The general form for a state space model can be seen
below:
ẋ1 a11 ⋯ a1n b11 ⋯ b1m
⎡ ⎤ ⎡ ⎤ ⎡ x1 ⎤ ⎡ ⎤ ⎡ u1 ⎤
⎢ ⎥ = ⎢ ⎥⎢ ⎥+⎢ ⎥⎢ ⎥
⎢ ⋮ ⎥ ⎢ ⋮ ⋱ ⋮ ⎥⎢ ⋮ ⎥ ⎢ ⋮ ⋱ ⋮ ⎥⎢ ⋮ ⎥
⎣ ⎦ ⎣ ⎦⎣ ⎦ ⎣ ⎦⎣ ⎦
ẋn an1 ⋯ ann xn bn1 ⋯ bnm um
and
y1 c11 ⋯ c1n d11 ⋯ d1m

⎡ ⎤ ⎡ ⎤ ⎡ x1 ⎤ ⎡ ⎤ ⎡ u1 ⎤
⎢ ⎥ ⎢ ⎥⎢ ⎥+⎢ ⎥⎢ ⎥
⎢ ⋮ ⎥ = ⎢ ⋮ ⋱ ⋮ ⎥⎢ ⋮ ⎥ ⎢ ⋮ ⋱ ⋮ ⎥⎢ ⋮ ⎥
⎣ ⎦ ⎣ ⎦⎣ ⎦ ⎣ ⎦⎣ ⎦
yr cr1 ⋯ crn xn dr1 ⋯ drm um
In general:
ẋ = Ax + Bu
and
y = Cx + Du
where:
x = states
u = inputs
y = outputs
−p1 −Gb 0
⎡ ⎤
A = ⎢ 0 −P2 P3 ⎥
⎣ ⎦
0 0 −n
1
0
⎡ V1 ⎤
B = ⎢0 0 ⎥
⎢ ⎥
1
⎣ 0 ⎦
V1
C = [1 0 0]
0
D = [ ]
0
Using this general formula, we can deconstruct the Bergman equations as a linear state space model. The first input is the insulin infusion
and the second input represents the meal glucose disturbance.
First Input:
1
˙
⎡G⎤ −p1 −Gb 0 G ⎡ 0 V1 ⎤
⎡ ⎤⎡ ⎤
u1
⎢ Ẋ ⎥ = ⎢
⎢ ⎥ 0 −P2 P3 ⎥ ⎢ X ⎥ + ⎢ 0 0 ⎥[ ]
⎢ ⎥
⎣ ⎦⎣ ⎦ u2
⎣ ˙ ⎦ ⎣ 1
I 0 0 −n I 0 ⎦
V1
where
= differential blood glucose concentration
= differential insulin concentration in a "remote" compartment
= differential blood insulin concentration
Second Input:
u1
y = [1 0 0 ]G + [0 0 ][ ]
u2
where
u1 U − Ub
[ ] = [ ]
u2 Gmeal − 0
6.9.3: EXAMPLE SET OF PARAMETERS

Gb = 4.5 mmol/liter
Ib = 4.5 mU/liter
V1 = 12 liters
p1 = 0/min
p2 = 0.025/min
p3 = 0.0000013 mU/liter
n = 5/54 min-1
It is important to keep track of units when using these parameters. The concentrations listed above are in mmol/liter, but the glucose
disturbance has units of grams. Therefore, it is necessary for us to apply a conversion factor of 5.5556 mmol/grams to the Gmeal term. Using
these variables solve one can solve for the steady states, and calculate the basal insulin infusion rate (Ub) such that is is equal to 16.1667
mU/min. For these parameters, the resulting state space model is
0 −4.5 0
⎡ ⎤
A = ⎢0 −0.025 0.000013 ⎥
−5
⎣0 0 ⎦
54
and
0 0.4630
⎡ ⎤
B = ⎢ 0 0 ⎥
⎣ 1 ⎦
0
12
It is common practice in the U.S. to describe glucose concentration in units of mg/deciliter as opposed to mmol/liter. Therefor, the units will
be converted from mmol/liter to mg/deciliter. The molecular weight of glucose is 180g/mol, and therefore one it is necessary to multiply the
glucose state (mmol/liter) by 18, so that the measured glucose output obtained will be in units of mg/deciliter. The following state-output
relationship will handle that:
C = [ 18 0 0]
D = [0 0]
Through Laplace transforms it will be found that the process transfer function is :
−3.79
Gp (s) =
(40s + 1)(10.8s + 1)s
The disturbance transfer function due to pole/zero cancelation is simply :

8.334
Gd (s) =
s
However, in reality glucose does not directly enter the blood stream. There is a "lag time" associated with the processing of glucose in the
gut. It must first be processed here before entering the blood. However, it can be modeled as a first-order function, with a 20-minute time
constant. This modifies the above equation for the disturbance transfer function to include the lag in the gut such that:
8.334
Gd (s) =
s(20s + 1)
6.9.4: DESIRED CONTROL PERFORMANCE

The steady-state glucose concentration of 4.5 mmol/liter corresponds to the glucose concentration of 81 mg/deciliter. A diabetic patient, in
order to stay healthy, must keep their blood glucose concentration above 70 mg\deciliter. If the blood glucose concentration falls below 70
mg/deciliter the patient may be likely to faint, as this is a very typical short-term symptom of hypoglycemia. The insulin infusion rate
(manipulated input) cannot fall below 0, and this constraint must be set in simulations. Numerous different control strategies can be used.
The equation given for the process function above can be simplified down to the following form and a PD or PID controller can be used.
kp
Gp (s) =
(τp + 1) s
A diabetic will know when they are consuming a meal, and therefore when their blood glucose concentration may rise. Therefore a feed-
forward control system may be desired.
6.9.5: REFERENCES
Bequette, B. W. Process Control: Modeling, Design, and Simulation New Jersey: Prentice Hall, 2003. pp 81-83, 694-697.

Written by: Robbert Appel, Jessica Rilly, Jordan Talia
6.9: Blood Glucose Control in Diabetic Patients is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via
CHAPTER OVERVIEW
7: MATHEMATICS FOR CONTROL SYSTEMS

7.1: Dirac delta (impulse) function
7.2: First-order Differential Equations
7.3: Second-order Differential Equations
7.4: Taylor Series
7.5: Laplace Transforms
7: Mathematics for Control Systems is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source content
1
7.1: DIRAC DELTA (IMPULSE) FUNCTION
7.1.1: INTRODUCTION
The Dirac delta function δ(t − t ) is a mathematical idealization of an impulse or a very fast burst of substance at t = t . (Here we are
0 0
considering time but the delta function can involve any variable.) The delta function is properly defined through a limiting process. One
such definition is as a thin, tall rectangle, of width ε:
1
δ (t − t0 ) =
ϵ
for
ϵ ϵ
t0 − < t < t0 +
2 2
and zero otherwise, in the limit that ϵ → 0 .

Then, we have
b
1
∫ δ (t − t0 ) dt = lim [ϵ ⋅ ( )] = 1
a
ϵ→0 ϵ
as long as a < t 0 < b . When t is outside the range of (a, b) , then the integral is zero.
0
Likewise, for any function f(t) that is continuous and differentiable (analytic) at t , 0
b t0 +ϵ/2
1
∫ δ (t − t0 ) f(t)dt = lim [ ∫ f(t)dt] = f (t0 )
a
ϵ→0 ϵ t0 −ϵ/2
where the quantity in the square brackets above is just the average value of f(t) in the interval t 0 −
ϵ
2
< t < t0 +
ϵ
2
. Thus, when ϵ → 0 , it
becomes just the value at t . For example:
0
∞
2 2
∫ δ(t − 2)t dt = 2 = 4
0
∞
2
∫ δ(t − 2)t dt = 0
3
7.1.2: DELTA FUNCTION AT THE INITIAL TIME

Note: if one of the limits of the integral coincides exactly with t , then the result is usually cut in half:
0
∫ δ (t − t0 ) f(t)dt = f (t0 ) /2
t0
for b > t . For example:

0
2 2
∫ δ(t − 2)t dt = (1/2)2 = 2
2
However, when we think of an impulse to a system at the initial time t , then we really consider that the entire delta function is added to the
0
system - that is, the actual time is an infinitesimal amount beyond t ; that is, t = t . In that case
0
+
+
∫ δ (t − t ) f(t)dt = f (t0 )
0
t0
7.1.3: DEFINITION IN TERMS OF A GAUSSIAN

Another equivalent definition of the delta functions are as a Gaussian function:
2
(t− t )
0
1 −
2
δ (t − t0 ) = lim −−e
2σ
σ→0 σ√2π
–
All the properties work out to be the same where a = √2σ is used.
7.1.4: PHYSICAL APPLICATIONS
In Control, the delta function is used an idealization of a very fast perturbation to the system. For example, if you dump a bucket of water
into a tank, then the "flow rate" is essentially a delta function - a very highly peaked function, but with a net integral (the total amount of
water in the bucket).
In mechanics, and example of the delta function is the force when hitting an object by a hammer. Say you hit a steel ball with a hammer. It
moves with a certain velocity representing the total momentum transferred by the hammer. Rather than talk about the force x time (the net
momentum transfer), one talks about an "impulse" which is the net momentum transferred in an infinitesimally short amount of time.
7.1.5: RELATION TO THE STEP FUNCTION

The step function, Θ(t − t ) , is the integral of the delta function or alternatively, the delta function is the derivative of the theta function,
0
where Θ(t − t ) is defined at 1 for t > t and 0 for t < t :

0 0 0
δ (t − t0 ) = (d/dt)Θ (t − t0 )
b b
∫ δ (t − t0 ) dt = ∫ (d/dt)Θ (t − t0 ) dt = Θ (a − t0 ) − Θ (b − t0 ) = 1 − 0 = 1 for a < t0 < b

a a
Here, the smooth or gaussian definition of the delta function corresponds to a smooth representation of the Θ function as the integral of a
gaussian or equivalently, the error function.
7.1: Dirac delta (impulse) function is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source content
7.2: FIRST-ORDER DIFFERENTIAL EQUATIONS
7.2.1: INTRODUCTION
We consider the general first-order differential equation:
dy(t)
τ + y(t) = x(t)
dt
The general solution is given by:

−(t− t0 )/τ t
e ′
−(t− t0 )/τ ′ (t − t0 )/τ ′
y(t) = y0 e + ∫ x (t ) e dt
τ t0
where y 0 = y (t − t0 ) . Note that t is used to be distinguished from the upper limit t of the integral.
′
To obtain the general solution, begin with the first order differential equation:
dy(t)
τ + y(t) = x(t)
dt
Divide both sides by τ :

dy(t) 1 1
+ y(t) = x(t)
dt τ τ
Rewrite the LHS in condensed form using the integrating factor e −t/τ
:
d 1
−t/τ t/τ
e [e y(t)] = x(t)
dt τ
Notice how a chain differentiation will return the LHS to the previous form
Simplify:
d t/τ
1 t/τ
[e y(t)] = x(t)e
dt τ
Integrate both sides:

t
t/τ t0 /τ
1 ′
′
t /τ ′
e y(t) − e y (t0 ) = ∫ x (t ) e dt
τ t0
Solve for y(t):

−t/τ t
e ′
−(t− t0 )/τ ′ t /τ ′
y(t) = y0 e + ∫ x (t ) e dt
τ t0
 EXAMPLE 7.2.1: SOLUTIONS OF FIRST ORDER DIFFERENTIAL EQUATIONS

Consider:
dy(t)
= x(t)
dt
Multiplying both sides by dt gives:

t t
dydt
′ ′
∫ = ∫ x (t ) dt
t0
dt t0
The general solution is given as:

t
′ ′
y(t) = y0 + ∫ x (t ) dt
t0
Now Consider:
dy(t)
= −ay(t)
dt
Dividing both sides by y(t) gives:
1 dy(t)
= −a
y(t) dt
which can be rewritten as:

d
[ln(y(t))] = −a
dt
Multiplying both sides by dt , integrating, and setting both sides of the equation as exponents to the exponential function gives the
general solution:
−a(t− t0 )
y(t) = y0 e
Now Consider:
dy(t)
= −ay(t) + x(t)
dt
The detailed solution steps are as follows:

1. Separate y(t) and x(t) terms
dy(t)
+ ay(t) = x(t)
dt
2. Rewrite the LHS in condensed form using the "integrating factor" e −at
at
d (e y(t))
−at
e = x(t)
dt
Notice how a chain differentiation will return the LHS to the form written in step 1
3. Divide both sides by e −at
at
d (e y(t))
at
= e x(t)
dt
4. Multiply both sides by dt and integrate

t
′
at at0 ′ at ′
e y(t) − e y0 = ∫ x (t ) e dt
t0
The general solution is given as:

t
′
−a(t− t0 ) −at ′ a(t − t0 ) ′
y(t) = y0 e +e ∫ x (t ) e dt
t0
7.2.2: STEP FUNCTION SIMPLIFICATION

dy(t)
τ + y(t) = x(t)
dt
For x(t) = Θ (t − t ), which is the step function and y

0 0 = 0
That is, x(t) = 1 for t 0 > to and x(t) = 0 otherwise:

The previously derived general solution:
−(t− t0 )/τ t
−(t− t0 )/τ
e ′
′
(t − t0 )/τ ′
y(t) = y0 e + ∫ x (t ) e dt
τ t0
reduces to:
−(t− t0 )/τ
y(t) = e
since the integral equals 1.

The quantity tau can be seen to be the time constant whereby y(t) drops to 1/e of its original value
 EXAMPLE 7.2.2
Consider the differential equation

dy(t)
= −0.5y(t) + x(t)
dt
where x(t) = 2 + 0.01t

Assuming y o = 0 , what is the behavior of y(t)?
Solution
General solution was derived previously as:
t
′
−a(t− t0 ) −at ′ a(t − t0 ) ′
y(t) = y0 e +e ∫ x (t ) e dt
t0
For a = 0.5 and y o = 0 , while setting t

o = 0 , the solution reduces to:
t
′
−0.5t ′ 0.5(t ) ′
y(t) = e ∫ (2 + 0.01t ) e dt
0
The following link may be referred to for integral tables: S.O.S. Math
Simplifying the solution gives the following:
−0.5t 0.5t
(e ) ((x + 198) ( e ) − 198))
y(t) =
50
Plotting this function displays the following behavior:
As can be seen clearly from the graph, initially the systemic response shows more exponential characteristics. However, as time
progresses, linear behavior dominates.
7.2: First-order Differential Equations is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source
7.3: SECOND-ORDER DIFFERENTIAL EQUATIONS
7.3.1: INTRODUCTION
We consider the general Second-order differential equation:
2
d Y (t) dY (t)
2
τ + 2ζτ + Y (t) = X(t)
2
dt dt
If you expand the previous Second-order differential equation:

2
d Y (t) dY (t)
τ1 τ2 + (τ1 + τ2 ) + Y (t) = X(t)
2
dt dt
d d
(τ1 + 1) (τ2 + 1) Y (t) = X(t)
dt dt
where:
−−−−
τ = √τ1 τ2
τ1 + τ2
ζ =
−−−−
2√τ1 τ2
Expansion of the differential equation allows you to guess what the shape of the solution (Y (t)) will look like when X(t) = 1.
The following rules apply when τ 1 = Re(τ1 ) + i ∗ Im(τ1 ) and τ
2 = Re(τ2 ) + i ∗ Im(τ2 ) :
If Re(τ1) and Re(τ2) are both positive, and there are NO imaginary parts, Y (t) will exponentially decay (overdamped).
If Re(τ1) and Re(τ2) are both positive, and there ARE imaginary parts, Y (t) will oscillate until it reaches steady state (underdamped).
If Re(τ1) and Re(τ2) are both negative, and there are NO imaginary parts, Y (t) will exponentially grow (unstable).
If Re(τ1) and Re(τ2) are both negative, and there ARE imaginary parts, Y (t) will oscillate and grow exponentially (unstable).
If Re(τ1) and Re(τ2) are both zero, and there ARE imaginary parts, Y (t) will oscillate and neither grow nor decay.
If τ1 and τ2 are both zero, Y (t) is equal to X(t).
7.3.2: SOLUTION OF THE GENERAL SECOND-ORDER SYSTEM (WHEN X(T)= Θ(T))

The solution for the output of the system, Y (t), can be found in the following section, if we assume that the input, X(t), is a step function
θ(t). The solution will depend on the value of ζ. If ζ is less than one, Y (t) will be underdamped. This means that the output will overshoot
and oscillate. If ζ is equal to one, Y (t) will be critically damped. This means that the output will reach the steady state value quickly,
without overshoot or oscillation. If ζ is greater than one, Y (t) will be overdamped. This means that the output will not reach the steady state
value as quickly as a critically damped system, but there will be no overshoot or oscillation.
7.3.2.1: UNDERDAMPED (Ζ<1)
If ζ < 1 , the solution is:
1 − −−− − t
−ζt/τ 2
Y (t) = 1 − −−−− −e sin(√ 1 − ζ + ϕ)
√1 − ζ 2 τ
where:
−−−− −
√1 − ζ 2
−1
ϕ = − tan ( )
ζ
The decay ratio (C/A) can be calculated using the following equation:
C −2πζ/√1−ζ
2
= e
A
The overshoot (A/B) can be calculated using the following equation:

A −πζ/√1−ζ
2
= e
B
The period (T ) and the frequency (ω) are the following:

2πτ
T = t2 − t1 = −−−− −
√1 − ζ 2
−−−− −
2π √1 − ζ 2
ω = =
T τ
7.3.2.2: CRITICALLY DAMPED (Ζ=1)

If ζ = 1 , the solution is:
t −t/τ
Y (t) = 1 − (1 + )e
τ
7.3.2.3: OVERDAMPED (Ζ>1)

If ζ > 1 , the solution is:
− −−−−
1 −ζt/τ 2
t
Y (t) = 1 − −−−−−e sinh(√ ζ − 1 + ϕ)
2 τ
√ζ − 1
where:
−−−− −
√ζ 2 − 1
−1
ϕ = − tanh ( )
ζ
 EXAMPLE 7.3.1
Given:
A1 = 1 m2
A2 = 1.5 m2
R1 = 0.25 s/m2
R2 = 0.75 s/m2
where:
A is the area of the tank
Q is the volumetric flowrate
R is the resistance to the flow of the stream exiting the tank
H is the height of liquid in the tank
develop an expression describing the response of H2 to Qin. Determine if the system is over, under or critically damped and determine
what the graph of the expression would look like using the complex τ plane above. A diagram of the system is shown below:
Solution
Performing a mass balance on each tank:
dH1 H1
A1 = Qin − (7.3.1)
dt R1
dH2 H1 H2
A2 = − (7.3.2)
dt R1 R2
where the left hand terms account for the accumulation in the tank and the right hand terms account for the flow in the entering and
exiting streams
Let τ1 = R1A1 and τ2 = R2A2
Equations 7.3.1 and 7.3.2 now become
dH1
τ1 = R1 Qin − H1 (7.3.3)
dt
dH2 R2
τ2 = H1 − H2 (7.3.4)
dt R1
Put like terms on the same side and factor
d
(τ1 + 1) H1 = R1 Qin (7.3.5)
dt
d R2
(τ2 + 1) H2 = H1 (7.3.6)
dt R1
Apply (τ 1
d
dt
+ 1) operator from Equation 7.3.5 to Equation 7.3.6
d d d R2
(τ1 + 1) (τ2 + 1) H2 = (τ1 + 1) H1 (7.3.7)
dt dt dt R1
The (τ 1
d
dt
+ 1) H1 term from the left hand portion of Equation 7.3.5 can be substituted into the right hand side of Equation 7.3.7
d d R2
(τ1 + 1) (τ2 + 1) H2 = R1 Qin
dt dt R1
d d
(τ1 + 1) (τ2 + 1) H2 = Qin R2
dt dt
This expression shows the response of H2 to Qin as a second order solution like those pictured above. Here Y(t)=H2 and X(t)=R2 Qin
τ1 + τ2 (0.25 ∗ 1) + (0.75 ∗ 1.5)
ζ = −−−− = − −−−−−−−−−−−−−− − = 1.296
2√τ1 τ2 2√ (0.25 ∗ 1)(0.75 ∗ 1.5)
Overdamped.
Both values of τ are positive real numbers, and the behavior of the graph of the equation can be found on the complex τ plane above.
 EXAMPLE 7.3.2: ANALOGY TO PHYSICS - SPRING SYSTEM
It might be helpful to use a spring system as an analogy for our second order systems.
From Newton's second law of motion,
F = ma
where:
F is Force
m is mass
a is acceleration
For the spring system, this equation can be written as:
′′
Fapplied − Ffriction − Frestoring = mx
where $x'$ is the acceleration of the car in the x-direction

′′
Fapplied − fx − kx = mx
where:
k is the spring constant, which relates displacement of the object to the force applied
f is the frequency of oscillation
m ′′
f ′
x + x + x = Fapplied
k k
As you can see, this equation resembles the form of a second order equation. The equation can be then thought of as:
2 ′′ ′
T X + 2ζTX + X = Fapplied
−
−−
m
τ = √
k
f
ζ = −−
−
2√mk
Because of this, the spring exhibits behavior like second order differential equations:
−−
−
If ζ > 1 or f > 2√mk it is overdamped
−−
−
If ζ = 1 or f = 2√mk it is critically damped
−−
−
If ζ < 1 or f < 2√mk it is underdamped
7.3: Second-order Differential Equations is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source
7.4: TAYLOR SERIES
A Taylor Series is a representation of a function in the form of an infinite sum. Each term is calculated from using a derivative of the
function as well as a factorial. The following equation is the defining equation of a Taylor series:
′ ′′′ (3) (n)
f (a) f (a) f (a) f (a)
2 3 n
f(x) = ∫ (a) − (x − a) + (x − a ) + (x − a ) +⋯− (x − a ) (7.4.1)
1! 2! 3! n!
where a is the point, around which the function is defined.

The Taylor approximation is most accurate at a , and becomes less precise as the distance from a increases. The following equation is the
same as Equation 7.4.1 mentioned above, in summation notation:
∞ (n)
f (a)
n
f(x) = ∑ (x − a )
n!
n=0
We will first consider a simple example, the function, e , defined around the point, 0:
x
x 0
1 1 2
1 3
1 n
e = e + (x − 0) + (x − 0) + (x − 0) +⋯+ (x − 0)
1! 2! 3! n!
The numerators of each equation are all 1, because any level derivative of e is e , and e , defined at 0 is always 1. The Taylor expansion
x x x
simplifies to the following equation:

2 3 n
x
x x x x
e = 1+ + + +⋯+
1! 2! 3! n!
Also note that a Taylor series centered around 0 is called a Maclaurin series. In addition to the ex Maclaurin series, some other common,
simplified Maclaurin series are listed here:
1 2
1 4
1 6
cos(x) = 1− x + x − x +⋯
2! 4! 6!
1 1 3
1 5
1 7
sin(x) = x− x + x − x +⋯
1! 3! 5! 7!
1 1 1
2 4 6
cosh(x) = 1+ x + x + x +⋯
2! 4! 6!
1 1 3
1 5
1 7
sinh(x) = x+ x + x + x +⋯
1! 3! 5! 7!
1
2 3 n
= 1+ x+ x +x +⋯+x
1− x
n+1
1 1 1 (−1)
2 3 4 n
ln(1 + x) = x− x + x − x +⋯+ x
2 3 4 n
These Maclaurin series can be modified to become Taylor series when the equation is not centered around zero.
Expansion of functions using Taylor series may be useful in derivations. Maclaurin series will be useful for when the function is being
approximated for small values of x. Using an infinite number of terms is generally unnecessary to model the function around the center
point. The first few terms of either a Taylor or Maclaurin series usually sufficiently approximate the function.
A stimulating discussion of Taylor series may be found in Comtet's "Calcul pratique des coefficients de Taylor d'une fonction algébrique"
(Enseign. Math. 10, 267-270, 1964) as well as Whittaker and Watson's landmark treatise, "Forms of the Remainder in Taylor's Series."
found in A Course in Modern Analysis, 4th ed.
7.4: Taylor Series is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source content that was edited to
7.5: LAPLACE TRANSFORMS
7.5.1: INTRODUCTION
Laplace Transform are frequently used in determining solutions of a wide class of partial diffferential equations. Like other transforms,
Laplace transforms are used to determine particular solutions. In solving partial differential equations, the general solutions are difficult, if
not impossible, to obtain. Thus, the transform techniques sometimes offer a useful tool for attaining particular solutions. The Laplace
transform is closely related to the complex Fourier transform, so the Fourier integral formula can be used to define the Laplace transform
and its inverse[3]. Integral transforms are one of many tools that are very useful for solving linear differential equations[1]. An integral
transform is a relation of the form:
b
F (s) = ∫ K (s, t)f(t)dt

a
where K (s, t) is the kernel of the transformation

The relation given above transforms the function f into another function F, which is called the transform of f. The general idea in using an
integral transform to solve a differential equation is as follows:
1. Use the relation to transform a problem for an unknown function f into a simpler problem for F
2. Solve this simpler problem to find F
3. Recover the desired function f from its transform F
Several integral transforms are known to be useful in applied mathematics, but in this chapter we consider only the Laplace transform. This
transform is defined in the following way Let f(t) be given for . Then the Laplace transform of f, which we will denote by ,
is defined by the equation:
∞
−st
L{f(t)} = ∫ e f(t)dt
0
Whenever this improper integral converges. The Laplace transform makes use of the kernel K(s,t) = e − (st). Since the solutions of linear
differential equations with constant coefficients are bsed on teh exponential function, the Laplace transform is particularly useful for such
equations.
7.5.1.1: TABLE 1 - TABLE OF LAPLACE TRANSFORMS
7.5.2: EXAMPLE PROBLEMS
7.5.2.1: ELEMENTARY FUNCTIONS USING LAPLACE TRANSFORMS
Example 1. Given , is a constant
∞
−st
mathcalL[c] = ∫ e c dt
0
Example 2. Given , is a constant

∞
at −st at
mathcalL[e ] = ∫ e e dt
0
, where
Example 3. Given , Then
∞
2 −st 2
L [t ] = ∫ e t dt
0
Integration by parts yields
2 −st ∞ ∞ −st
t e e
= [− ] +∫ 2tdt
s 0
s
0
Since as , we have
−st ∞ ∞ −st
2
2 e 2 e
mathcalL[t ] = [− t] + ∫ dt
s s 0
s 0 s
−st ∞ ∞ −st
2
2 e 2 e
L [t ] = [− t] + ∫ dt
s s s 0
s
0
2
=
3
s
Example 4. Given . Then

∞
−st
F (s) = L[sin ωt] = ∫ e sin ωtdt
0
Thus, solving for F (s) , we obtain

ω
L[sin ωt] =
(s2 + ω2 )
7.5.2.2: IN CLASS EXAMPLE USING LAPLACE TRANSFORMS

Solving a differential equation:
dy(t)
= −ay(t)
dt
Rearranging the equation to one side, we have:

dy(t)
+ ay(t) = 0
dt
Next, we take the Laplace transform

(sy(s) − yo ) + ay(s) = 0
where and Solving for y(s), we find:

yo
y(s) =
s+a
yo
−1 −1 −at
y(t) = L {y(s)} = L { } = yo e
s+a
7.5.3: REFERENCES
1. William E. Boyce, Elementary Differential Equations and Boundary Value Problems (2005) Chapter 6 pp 307-440.
2. Dr. Ali Muqaibel, EE207 Signals & Systems (2009) <http://tinyurl.com/ycq46qn>
3. Tyn Myint-U, Partial Differential Equations for Scientists and Engineers (2005) pp 337 -341.

Written By: Paul Rhee, Eric Chuang, and David Mui
7.5: Laplace Transforms is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source content that was
CHAPTER OVERVIEW
8: OPTIMIZATION
8.1: Introduction to Optimization
8.2: Linear Optimization
8.3: Non-linear Optimization
8: Optimization is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source content that was edited to
1
8.1: INTRODUCTION TO OPTIMIZATION
Optimization aims to obtain the best results in a given situation, or to minimize input to maximize benefit [1].
8.1.1: ENGINEERING APPLICATION FOR OPTIMIZATION

The following are examples of optimization across all engineering disciplines [1].
Design of aircraft and aerospace structures for minimum weight
Finding the optimal trajectories of space vehicles
Design of civil engineering structures such as frames, foundations, bridges, towers, chimneys, and dams for minimum cost
Minimum-weight design of structures for earthquake, wind, and other types of random loading
Design of water resources systems for maximum benefit
Optimal plastic design of structures
Optimum design of linkages, cams, gears, machine tools, and other
mechanical components
Selection of machining conditions in metal-cutting processes for minimum production cost
Design of material handling equipment such as conveyors, trucks, and cranes for minimum cost
Design of pumps, turbines, and heat transfer equipment for maximum efficiency
Optimum design of electrical machinery such as motors, generators,and transformers
Optimum design of electrical networks
Shortest route taken by a salesperson visiting various cities during one tour
Optimal production planning, controlling, and scheduling
Analysis of statistical data and building empirical models from experimental
results to obtain the most accurate representation of the physical phenomenon
Optimum design of chemical processing equipment and plants
Design of optimum pipeline networks for process industries
Selection of a site for an industry
Planning of maintenance and replacement of equipment to reduce operating costs
Inventory control
Allocation of resources or services among several activities to maximize the benefit
Controlling the waiting and idle times and queueing in production lines to reduce the costs
Planning the best strategy to obtain maximum profit in the presence of a competitor
Optimum design of control systems
8.1.2: OPTIMIZATION ACROSS AN ORGANIZATION

The highest level of optimization includes the optimization of the raw material supply chain and the optimization of the packaging and
product distribution chain. The plant wide optimization must consider documentation, maintenance, scheduling, and quality management
considerations. Plant wide optimization resolves the conflict of objectives between the unit operations and the envelope strategies required
to optimize the entire plant. Within the unit operations level multivariable optimization cannot be achieved when individual processing
equipment is defective or when the control loops are not properly tuned. It is important that measurements be sampled fast enough, that
controls loops be tuned for fast rates of recovery, and loop cycling be eliminated. When no mathematical model can describe a process, the
process can only be optimized experimentally and empirical optimization is required [2].
8.1.3: OPTIMIZATION DESIGN
8.1.4: CONSTRAINTS
Design constraints are physical limitations or restrictions that must be satisfied to produce an acceptable design [1].
Operation conditions – safety, environmental
Equipment constraints – e.g. pump rates
Storage capacities
Product quality and impurities
8.1.5: OPTIMIZATION SITUATIONS

The following list describes common reasons for optimization in an industrial plant.
Sales limited by production (e.g. reduce costs by minimizing downtime)
Sales limited by market (e.g. be the “low cost producer”)
Plants with high throughput
High raw material or energy consumption
Product quality better than specifications
Loss of valuable or hazardous components through waste stream
8.1.6: REAL-TIME OPTIMIZATION

The following describes the steps in order to optimize a chemical engineering process.
1. Identify process variables
2. Select objective function (e.g. profit $$$)
3. Develop process model and constraints
4. Simplify model to objective function (e.g linearization)
5. Compute the optimum
6. Perform sensitivity study
8.1.7: INDUSTRY EXPERIENCE

Optimization can be applied to every aspect of a process. For example, at a refinery there are operators that work out in the units.
Optimization can be applied to increase the operator/engineer communication by implementing "real-time" computer programs that allow
the process engineers to see what is actually happening in the plant. This will help optimize process by allowing engineers to see what the
conditions in the plant are in real time.
8.1.8: PILOT PLANT EXPERIENCE

Optimization of processes in a pilot plant will allow for more efficient scale-up to commercial size. In the ChE 460 course, four unit
operations are optimized to produce soybean biodiesel. In order to optimize the reaction conversion, the project engineers vary catalyst
concentration, agitation rate, and temperature. A design of experiment (DOE) is used to find the best set of input parameters.
8.1.9: REFERENCES
1. Rao, Singiresu S. Engineering Optimization - Theory and Practice (3rd Edition). (pp: 5). John Wiley & Sons. Online version available
at: http://knovel.com/web/portal/browse/display?_EXT_KNOVEL_DISPLAY_bookid=1100&VerticalID=0
2. Lipták, Béla G. Instrument Engineers' Handbook: Process control and optimization. Boca Raten, FL: CRC Press, 2006.
8.1: Introduction to Optimization is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Brittany Flaherty, Christine Curran,
Lauren Pakan, & Lauren Pakan via source content that was edited to conform to the style and standards of the LibreTexts platform; a detailed edit history is
8.2: LINEAR OPTIMIZATION
Linear optimization is a method applicable for the solution of problems in which the objective function and the constraints appear as linear
functions of the decision variables. The constraint equations may be in the form of equalities or inequalities[1]. In other words, linear
optimization determines the way to achieve the best outcome (for example, to maximize profit or to minimize cost) in a given mathematical
model and given some lists of requirements represented as linear equations [2].
8.2.1: APPLICATIONS
Linear optimization can be applied to numerous fields, in business or economics situations, and also in solving engineering problems. It is
useful in modeling diverse types of problems in planning, routing, scheduling, assignment and design [2].
8.2.1.1: SOME EXAMPLES OF APPLICATIONS IN DIFFERENT INDUSTRIES
Petroleum refineries
One of the early industrial applications of linear optimization has been made in the petroleum refineries. An oil refinery has a choice of
buying crude oil from different sources with different compositions at different prices. It can manufacture different products, such as diesel
fuel, gasoline and aviation fuel, in varying quantities. A mix of the purchased crude oil and the manufactured products is sought that gives
the maximum profit.
Manufacturing firms
The sales of a firm often fluctuate, therefore a company has various options. It can either build up an inventory of the manufactured
products to carry it through the period of peak sales, or to pay overtime rates to achieve higher production during periods of high demand.
Linear optimization takes into account the various cost and loss factors and arrive at the most profitable production plan.
Food-processing industry
Linear optimization has been used to determine the optimal shipping plan for the distribution of a particular product from different
manufacturing plants to various warehouses.
Telecommunications
The optimal routing of messages in a communication network and the routing of aircraft and ships can also be determined by linear
optimization method.
8.2.2: CHARACTERISTICS OF LINEAR OPTIMIZATION

The characteristics of a linear optimization problem are:
1. The objective function is of the minimization type
2. All the constraints are of the equality type
3. All the decision variables are non-negative
Any linear optimization problem can be expressed in the standard form by using the following transformation:
1) The maximization of a function f (x 1, x2 , … , xn ) is equivalent to the minimization of the negative of the same function.
For example:
Minimize
f = c1 x1 + c2 x2 + … + cn xn
is equivalent to
Maximize
′
f = −f = − c1 x1 − c2 x2 − … − cn xn
Consequently, the objective function can be stated in the minimization form in any linear optimization problem.
2) If a constraint appears in the form of a "less than or equal to" type of inequality as
ak1 x1 + ak2 x2 + … + akn xn ≤ bk
it can be converted into the equality form by adding a non-negative slack variable as follows:
ak1 x1 + ak2 x2 + … + akn xn + xn+1 = bk
Similarly, if the constraint is in the form of a "greater than or equal to" type of inequality, it can be converted into the equality form by
subtracting the surplus variable .
3) In most engineering optimization problems, the decision variables represent some physical dimensions, hence the variables will be
non-negative.
However, a variable may be unrestricted in sign in some problems. In such cases, an unrestricted variable (which can take a positive,
negative or zero value) can be written as the difference of two non-negative variables.
Thus, if is unrestricted in sign, it can be written as x
j = x
′
j
−x
′′
j
, where
0 ≤ x
′
j
and 0 ≤ x ′′
j
It can be seen that will be negative, zero or positive, depending on whether is greater than, equal to, or less than
 EXAMPLE 8.2.1
This example comes from Seborg, Edgar, and Mellinchamp (with some corrections). A chemical company is mixing chemical
component, A and B to produce product, E and F . The chemical reaction is listed below:
A+ B = E
A + 2B = F
The conditions of this production is listed as below:

Cost of Materials Maximum Available per day Fixed Cost
A $0.15/lbs 40,000lbs 350
B $0.20/lbs 30,000lbs 200
Value of Products Maximum Production Rate
E $0.40/lbs 30,000lbs
F $0.33/lbs 30,000lbs
The profit of this production can be simply described as the function below:
Profit
= ∑ FsVs − ∑ FrVr − C.P. − F.C.
s r
Flow Rates of Products

Flow Rates of Raw Materials
Values of Products
Values of Raw Materials
Cost of Production
Fixed Cost
Constraints
:
:
:
:
: ≥
Solution
1 1
FA = FE + FF
2 3
1 2
FB = FE + FF
2 3
Profit = ∑ Fs Vs − ∑ Fr Vr − C. P . −F . C.
s r
′ ′ ′
= (0.4ψFj + 0.33FF − ) − (0.15F − 0.2F ) − (0.15FFi + 0.05F ) − (350 + 200i
A H F
F_A0 " src="/@api/deki/files/18014/image-717.png">
\frac{1}{2} F_E + \frac{1}{3} F_F0 " src="/@api/deki/files/18016/image-718.png">
F_F-\frac{3}{2} F_E ...........(1) " src="/@api/deki/files/18018/image-719.png">
F_B0 " src="/@api/deki/files/18025/image-723.png">

\frac{1}{2} F_E + \frac{2}{3} F_F0 " src="/@api/deki/files/18027/image-724.png">
F_F-\frac{3}{4} F_E ...........(3) " src="/@api/deki/files/18029/image-725.png">
Solution using Mathematica

INPUT:
profit = 0.25 FE + 0.28 FF - 0.15 FA - 0.2 FB - 350 - 200
sol = Maximize[ profit, {FA > 0, FB > 0, FA < 40000, FB < 30000, FE > 0, FE < 30000, FF > 0, FF < 30000, FA == (1/2) FE + (1/3)
FF, FB == (1/2) FE + (2/3) FF}, {FE, FF}]
OUTPUT: {3875., {FE -> 30000., FF -> 22500.}}
Solution using "Solver" in Excel.
Result is FE = 30,000, FF = 22500.

If only process ! or process 2 were running a full capacity, the profit would be less.
8.2.2.1: LINEAR OPTIMIZATION
The above is an example of linear optimization. It is often used in oil refinery to figure out maximal profit in response to market
competition.
2.4.4 Graph
8.2.3: EXAMPLE 2
Example of Linear Optimization Problem in Excel
Written by: Jennifer Campbell, Katherine Koterba, MaryAnn Winsemius
8.2.3.1: PART 1: ORGANIZE GIVEN INFORMATION

As stated in the Linear Optimization section example above, there are three categories of information needed for solving an optimization
problem in Excel: an objective function, constraints, and decision variables.
We will use the following example to demonstrate another application of linear optimization. We will be optimizing the profit for Company
X’s trucking business.
To reach capacity, Company X must move 100 tons of cargo per day by truck. Company X’s trucking fee is $250/ton. Besides the weight
constraint, the company can only move 50,000 ft^3 of cargo per day due to limited volume trucking capacity. The following amounts of
cargo are available for shipping each day:
8.2.3.2: PART 2: SET UP THE PROBLEM USING EXCEL
Solver is an Add-in for Microsoft Excel. It will be used to optimize Company X’s profit. If ‘Solver’ is not on the ‘Tools’ menu in Excel,
then use the following steps to enable it:
For Windows 2007:
Click on the Office button at the top left corner of the screen. Click on the “Excel Options” button on the bottom right of the menu.
Select “Add-ins.” Make sure that “Excel Add-ins” is selected in the “Manage” drop down list. Click “Go.”
A new window will appear entitled “Add-ins.” Select “Solver Add-in” by checking the box. Click “Go.”
A Configuration window will appear. Allow Office to install the Add-in.
The solver has been successfully installed. (See Windows Help for more instruction.)
Use the figure below to set up your Excel worksheet.
Enter in the following formulas to the cells as shown below:
8.2.3.3: PART 3: RUNNING SOLVER
Click on the “Data” tab and select “Solver”. A dialog box will appear.
Enter the parameters as shown in the figure below.
Detailed steps are as follows:

In “Set Target Cell,” enter the cell corresponding to the company’s profit (B4).
Select “Max” under “Equal To.”
Click on the “Options” tab and check the “Assume Linear Model” box.
For “By Changing Cells:” select the cells in column B corresponding the cargo amounts (B9, B10, B11).
To add constraints, select “Add” under “Subject to the Constraints” A dialog box will open.
In the “Cell Reference:” field, enter the cell location of the decision value that is subject to constraint (i.e. B9).
Use the pull-down menu in the middle to select the appropriate inequality relation (i.e. <=)
In the “Constraint:” field, enter the cell location of the constraint value (i.e. D17).
Continue to click the “Add.”
Repeat above steps until all of the constraints are entered. Then click “OK.”
When all the proper settings have been entered, click “Solve.”
A “Solver Results” box will appear. Select “Keep Solver Solution” and click “OK.”.
The solved worksheet is below.
8.2.4: SENSITIVITY REPORT
Written by Michael Chisholm and Doug Sutherland, Dec. 2009
Excel's solver program allows us to analyze how our profit would change if we had an alteration in our constraint values. These values can
change due to a variety of reasons such as more readily available resources, technology advancements, natural disasters limiting resources,
etc.
First, it analyzes whether the constraints are binding or non-binding. The binding constraints limit the profit output where the non-binding
constraints do not limit the overall process. If the non-binding constraints were changed, the profit would not be effected as long the change
in these constraints lies within the allowable increase and decrease that is indicated within the sensitivity report. If the binding constraints
are changed, the profit will be directly affected. The affect on the profit is shown with shadow price values, also displayed in the sensitivity
report. The shadow price is the resulting increase or decrease in profit per unit increase or decrease in the constraint. This applies as long as
the change in constraint remains within the allowable increase or decrease where a linear relationship can be assumed.
The shadow price only analyzes the change in one variable at a time. In order to do two, you must plug the new constraint value for one of
the variables and solve using solver. Using the new sensitivity report, analyze the effect that changing the second variable would have with
the change being made in the first constraint.
Looking at Example 1 above, we will now walk through the steps on how to create a sensitivity report.
After clicking "solve" in excel, a solver results dialogue box appears as seen below.
There is a list of three options on the right; answer, sensitivity, and limits. Select the sensitivity option before clicking ok. A new tab will be
generated in the worksheet titled "sensitivity 1." A view of the sensitivity report within the tab is seen below. As you can see, two tables are
generated. For this example, resource A and product F are non-binding as shown with a shadow price of 0 and an infinite allowable
increase. The allowable decrease is the amount the capacity changes until the final value is reached. Past this point the constraint would
become a binding constraint. For the constraining variables (resource B and product E), their constraints are binding. Regarding resource B,
if its constraint was increased by up to 5000 or decreased by up to 15000, this would have a linear effect on profit within this range. For
each unit increase or decrease, the profit will change by 12 cents per unit, respectively. The same is true if our capacity for product E
changes with it's allowable values and shadow price on the table.
If the constraint on B increased by 5,000 lbs our new profit would be $8,600/day (8,000+.12*5,000). Instead, if our facility could increase
production of E by 30,000 lb/day the resulting profit would be $12,950/day (8,000+.165*30,000).
8.2.5: SOLVING LINEAR OPTIMIZATION PROBLEMS USING THE PRIMAL SIMPLEX ALGORITHM
Written by: Tejas Kapadia and Dan Hassing [Note: needs specific reference, and also solution to the preceding problem by this method
would be good -- R.Z.]
Instead of solving linear optimization problems using graphical or computer methods, we can also solve these problems using a process
called the Primal Simplex Algorithm. The Primal Simplex Algorithm starts at a Basic Feasible Solution (BFS), which is a solution that lies
on a vertex of the subspace contained by the constraints of the problem. In the Graph in Example 1, this subspace refers to the shaded region
of the plot. Essentially, after determining an initial BFS, the Primal Simplex Algorithm moves through the boundaries from vertex to vertex
until an optimal point is determined.
The basic procedure is the following:
1. Find a unit basis.
2. Set-up the problem in standard form using a canonical tableau.
3. Check optimality criterion.
1. If criterion passes, then stop, solution has been found.
4. Select an entering variable among the eligible variables.
5. Perform pivot step.
6. Go back to 1.
For simplicity, we will make the following assumptions:
1. The optimum lies on a vertex and is not unbounded on an extreme half-line.
2. The constraints are equations and not also inequalities.
1. In the case that the constraints are inequalities, slack variables will need to be introduced. Although the process is not very different in
this case, we will ignore this to make the algorithm slightly less confusing.
3. Decision variables are required to be nonnegative.
4. The problem is a minimization problem. To turn a maximization problem into a minimization problem, multiply the objective function
by -1 and follow the process to solve a minimization problem.
We will begin with the following example:
Objective Function: Minimize z = −x 5 − 8x6
Subject to the constraints:

x1 − x5 + x6 = 2 (8.2.1)
x2 + x5 + x6 = 1 (8.2.2)
x3 + 2x5 + x6 = 5 (8.2.3)
x4 + x6 = 0 (8.2.4)
xi ≥ 0 (8.2.5)
First we begin by finding a unit basis:

A shortcut method to finding this unit basis is putting numbers in for each variable so that every constraint equation is satisfied.
In this case, setting , , and will satisfy the final equation and also set the values for , , and to 2, 1, and 5,
respectively. Remember, these decision variables must be nonnegative.
Set up the canonical tableau in the following form:
As you can see, the first four rows correspond to the constraints, while the final row corresponds to the objective function. The “b” column
corresponds to the right hand side (RHS) of the constraints. As you can see, the “-z” column is on the left hand side (LHS) of the equation,
rather than the RHS.
First, we should perform pivot steps so that the tableau corresponds to the unit basis we found earlier. By performing pivot steps on , ,
, and , we will reach the feasible point where ( , , , , , and ) = . Because , , and all equal zero,
the pivot step on can actually be done on or , but in this example, we used . These pivot steps can be performed on any row as
long as they are all different rows. In this example, we performed pivot steps on , , , using the Pivot and
Gauss-Jordan Tool at people.hofstra.edu/Stefan_Waner/RealWorld/tutorialsf1/scriptpivot2.html. To use this tool, place the cursor on the cell
that you wish to pivot on, and press “pivot”.
After four pivot steps, the tableau will look like this:
As you can see, this is identical to the initial tableau, as , , , and were set up such that an initial feasible point was already
chosen.
The optimality criterion states that if the vector in the bottom left of the tableau is all positive, then an optimal solution exists in the “b”
column vector, with the value at the bottom of the “b” column vector as the negative of the value of the objective function at that optimal
solution. If this is not true, then a pivot step must be performed. In this example, clearly, a pivot step must be performed.
Next, we need to choose an entering variable. We want to choose an entering variable that has a negative element in the bottom row,
meaning that the objective value could be improved if that variable was nonzero in the solution. So, we will choose in this example.
Now, we must calculate ratios of each RHS coefficient divided by the coefficient of the entering variable in that row. In this case, the vector
corresponding to this calculation would equal . We cannot pivot on a zero element, so we cannot pivot on the
fourth row. We want to keep the RHS positive, so we cannot pivot on the first row. We must choose the minimum nonnegative ratio to
remain at a feasible solution, so we choose the second row in the column, which has a ratio of 1/1.
After the pivot step:
As we can see, has a negative coefficient in the bottom row, indicating the same step must be repeated on that column. We calculate
ratios for that column, and get: . Consequently, we choose to pivot on the fourth row because it corresponds to the
minimum nonnegative ratio of 0.
After another pivot step:
Because the bottom row is all positive, we are now at an optimal solution. To understand this final tableau, we look at each column for
variables that only have one “1” in the column. If the column has only one “1”, the RHS value in that row is the value of that variable. In
this case, , , and . Any variable that does not have just a single “1” in the column is equal to zero. So, the optimal
solution is ( , , , , , and )= , and the optimal value is (z was on the LHS in the tableau).
Now, we have successfully solved a linear optimization problem using the Primal Simplex Algorithm. Verification of the solution can be
easily performed in Microsoft Excel.
8.2.6: REFERENCES
1. D. E. Seborg, T. F. Edgar, D. A. Mellichamp: Process Dynamics and Control, 2nd Edition, John Wiley & Sons.
2. Rao, Singiresu S. Engineering Optimization - Theory and Practice, 3rd Edition, 129-135, John Wiley & Sons.
8.2: Linear Optimization is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Danny Hsiao, Jenny Ou, Huey Shann Sue, &
Huey Shann Sue via source content that was edited to conform to the style and standards of the LibreTexts platform; a detailed edit history is available upon
request.
8.3: NON-LINEAR OPTIMIZATION
Various conditions and situations are not adequately described using linear systems. In this case, nonlinear optimization may be applied.
Unlike linear optimization, the optimal operating condition does not exist at the boundaries.
8.3.1: QUADRATIC OPTIMIZATION

1
T T
f(x) = c − x b+ x Ax
2
To optimize, it is necessary to find when the gradient of f is equal to zero.
∇f(x) = 0
∇f(x) = b − Ax
−1
x∗ = A b
It may be possible to solve the optimal by a linear equation, approximated by a Taylor series.
′
1 ′
f (x∗ ) = f(x) + (x∗ − x) ∇f(x) + (x∗ − x) ∇∇f(x) (x∗ − x) + …
2
8.3.2: ITERATIVE METHODS

When direct methods cannot solve the equation (i.e. A is not symmetric positive definite), iterative methods are possible [1].
By starting with an initial guess of , an algorithm may lead to a that better satisfies the equation. Through iteration, theoretically,
.
8.3.3: APPLICATIONS
Finance: Portfolio optimization
Businesses: Optimize inventory
Engineering: Rigid body dynamics
Biochemistry: Kinetic modeling [2]
8.3.4: EXAMPLE: TYPICAL NONLINEAR 3D CURVES
(Image from [1])

As observed, the optimal condition does not necessarily exist at the boundary of the curve.
8.3.5: EXAMPLE: QUADRATIC OPTIMIZATION

T 1 T
f(x) = c ⃗ x⃗ + x⃗ Qx⃗
2
where
T
c ⃗ = (c1 , c2 , … , cn )
T
x⃗ = (x1 , x2 , … , xn )
For a quadratic system, n = 2 , thus, Q (the quadratic term constant) is defined as a symmetric matrix as follows.
Q1 Q3
Q = [ ]
Q3 Q2
Thus, multiplying out the f ,

1
2 2
f(x) = (c1 x1 + c2 x2 ) + ( Q1 x1 + 2Q3 x1 x2 + Q2 x2 )
2
8.3.6: REFERENCES
1. Lippert, Ross A. "Introduction to non-linear optimization." D.E. Shaw Research, February 25, 2008.
http://www.mit.edu/~9.520/spring08/Classes/optlecture.pdf
2. Mendes, Pedro and Kell, Douglas B. "Non-linear optimization of biochemical pathways: application to metabolic engineering and
parameter estimation." Journal of Bioinformatics, Volume 14, 869-883. 1998.
3. "Introduction to Non-linear optimization." Georgia Institute of Technology Systems Realization Laboratory.
www.srl.gatech.edu/education/ME6103/NLP-intro.ppt
8.3: Non-linear Optimization is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Matthew Goh, Andrew King, Edwin Yik,
& Edwin Yik via source content that was edited to conform to the style and standards of the LibreTexts platform; a detailed edit history is available upon
request.
CHAPTER OVERVIEW
9: PROPORTIONAL-INTEGRAL-DERIVATIVE (PID) CONTROL

9.1: Constructing Block Diagrams- Visualizing control measurements
9.3: PID Tuning via Classical Methods
9.4: PID tuning via Frequency Responses with Bode Plots
9.5: PID tuning via Optimization
9.6: PID Downsides and Solutions
9: Proportional-Integral-Derivative (PID) Control is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via
1
9.1: CONSTRUCTING BLOCK DIAGRAMS- VISUALIZING CONTROL
MEASUREMENTS
It is often convenient to express a control system with a block diagram. These diagrams help to visualize the relationships between each part
of the control system (Figure 9.1.1 ).
Figure 9.1.1: Block diagram representing a control system.

The control system shown is called a closed-loop system or a feedback system because the measured value of the controlled variable is
returned or "fed back" to the comparator. In the comparator the controlled variable is compared with the desired value or set point. If there is
any difference between the measured variable and the set point, an error is generated. This error enters a controller, which in turns adjusts
the final control element to return the controlled variable to the set point [1].
R(t) = "set point", is a synonym for the desired value of the controlled variable
ε(t) = process error or comparator
X(t) = process variable
M(t) = measurement variable
Y(t) = output variable
The "hat" above the G indicates that it is an operator (not a simple function) and may include integration and differentiation. Thus, you
cannot divide by a , but have to multiply both sides by its inverse (see below). This is a more general representation that the
standard Laplace transform form; here it is not necessary that the operators be linear.
From the block diagram the equations for a system can be constructed. Define the following:
=Controller operator
=Process operator
=Measurement delay operator
The equations for the system in Figure 9.1.1 are given below. Note that these equations are always written as "output is equal to operator
applied to input."
Process:
Controller:
Comparator: ε(t) = R(t) − M(t)
Measurement delay:
Typically to plot the response of the controller, Y (t), it is necessary to solve these equations to obtain a single equation for Y (t). This can
be done by first plugging the measurement delay equation into the compartor equation, or.
^
ϵ(t) = R(t) − Gm Y (t).
This equation can be put into the controller equation, or

^
X(t) = Gc [R(t) − Gm Y (t)] .
Finally, the equation for Y (t) can be obtained by putting the above equation into the equation for the process, or
^ ^ ^
Y (t) = Gp Gc [R(t) − Gm Y (t)] .
Frequently we have a differential equation for X(t) in terms of derivatives of Y (t) so we write
−1
^
X(t) = Gp Y (t)
which implies the form of the general equation

−1
^ ^
Gp Y (t) = Gc [R(t) − Gm Y (t)]
If the process, controller, and measurement delay operators are known, Y (t) can be plotted and the response of the system can be observed
(i.e. using Mathematica).
That is, we have
1. Y (t) = G
^
X(t)
p
2. X(t) = G ϵ(t)]
^
c
3. ϵ(t) = R(t) − M (t)

4. M (t) = G^
Y (t)
m
−1
5. ^
Gp
^
Y (t) = Gc [R(t) − Gm Y (t)]
Assume no measurement lag:

^
Gm = 1
Assume first-order system:

−1 1 d
^
Gp = (τp + 1)
Kp dt
Assume PI controller:
t
−1 Kc
^
Gc = Kc + ∫ dt
τI 0
Substituting equations (2) and (3) into (5) gives:

t t
1 d Ke Kc
(τp 1) Y (t) = (Kz , ∫ at) c(t) − (Kz | ∫ dt) {H(t) Y (t)
Kp dt η 0 71 0
1
Simplifying further gives:

Kc
′
τp Y (t) + Y (t) = Kc (R(t) − Y (t)) + (R(t) − Y (t))
τI
Differentiating the above equation gives:
′′ ′ ′ ′
Kc Kc
τp Y (t) + Y (t) = Kc R (t) − Kc Y (t) + R(t) − Y (t)
τI τI
Rearranging the above equation:

Kc Kc
′′ ′ ′
τp Y (t) + (1 + Kc ) Y (t) + Y (t) = Kc R (t) + R(t)
τI τI
Multiply the above equation by to eliminate the co-efficient of Y ′

(t) gives:
τI τp 1 + Kc
′′ ′ ′
Y (t) + τI Y (t) + Y (t) = τI R (t) + R(t)
Kc Kc
9.1.1: REFERENCES
1. Coughanowr, D.R. and S.E. LeBlanc (2009). Process Systems Analysis And Control. Third Edition. Mc-Graw Hill.
9.1: Constructing Block Diagrams- Visualizing control measurements is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by
upon request.
9.2: P, I, D, PI, PD, AND PID CONTROL
9.2.1: INTRODUCTION
Process controls are necessary for designing safe and productive plants. A variety of process controls are used to manipulate processes,
however the most simple and often most effective is the PID controller. The controller attempts to correct the error between a measured
process variable and desired setpoint by calculating the difference and then performing a corrective action to adjust the process accordingly.
A PID controller controls a process through three parameters: Proportional (P), Integral (I), and Derivative (D). These parameters can be
weighted, or tuned, to adjust their effect on the process.
Much more practical than the typical on/off controller, PID controllers allow for much better adjustments to be made in the system. While
this is true, there are some advantages to using an on/off controller including that they are (1) relatively simple to design and execute and (2)
binary sensors and actuators (such as an on/off controller) are generally more reliable and less expensive.
Although there are some advantages, there are large disadvantages to using an on/off controller scheme. They are (1) inefficient (using this
control is like driving with full gas and full breaks), (2) can generate noise when seeking stability (can dramatically overshoot or undershoot
a set-point), and (3) physically wearing on valves and switches (continuously turning valves/switches fully on and fully off causes them to
become worn out much quicker).
To allow for much better control and fine-tuning adjustments, most industrial processes use a PID controller scheme. This section will
provide a brief introduction on PID controllers as well as methods to model a controlled system in Excel.
The Process Gain (K ) is the ratio of change of the output variable (responding variable) to the change of the input variable (forcing
function). It specifically defines the sensitivity of the output variable to a given change in the input variable.
Δ Output
K =
Δinput
Gain can only be described as a steady state parameter and give no knowledge about the dynamics of the process and is independent of the
design and operating variables. A gain has three components that include the sign, the value, the units. The sign indicates how the output
responds to the process input. A positive sign shows that the output variable increases with an increase in the input variable and a negative
sign shows that the output variable decreases with an increase in the input variable. The units depend on the process considered that depend
on the variables mentioned.
For example, the pressure was increased from 21 psi to 29 psi. This change increased the valve position from 30% vp to 22% vp.
(29 − 21) psi
K = = −1.0 psi/(%vp)
(22 − 30)%vp
Dead Time (t ) is the between the change in an input variable and when the output variable begins. Dead time is important because it
0
effects the controllability of the control system. A change in set point is not immediate because of this parameter. Dead time must be
considered in tuning and modeling processes.
9.2.2: TYPES OF CONTROL

Process controls are instruments used to control a parameter, such as temperature, level, and pressure. PID controllers are a type of
continuous controller because they continually adjust the output vs. an on/off controller, when looking at feed forward or feed backward
conditions. An example of a temperature controller is shown in Figure 9.2.1 .
Figure 9.2.1: Temperature controller in a CSTR
As shown in Figure 1, the temperature controller controls the temperature of a fluid within a CSTR (Continuous Stirred Tank Reactor). A
temperature sensor first measures the temperature of the fluid. This measurement produces a measurement signal. The measurement signal
is then compared to the set point, or desired temperature setting, of the controller. The difference between the measured signal and set point
is the error. Based on this error, the controller sends an actuating signal to the heating coil, which adjusts the temperature accordingly. This
type of process control is known as error-based control because the actuating signal is determined from the error between the actual and
desired setting. The different types of error-based controls vary in the mathematical way they translate the error into an actuating signal, the
most common of which are the PID controllers. Additionally, it is critical to understand feed-forward and feed-back control before
exploring P, I, and D controls. Feed Forward Control Feedback Control
9.2.3: I, D, PI, PD, PID CONTROL

As previously mentioned, controllers vary in the way they correlate the controller input (error) to the controller output (actuating signal).
The most commonly used controllers are the proportional-integral-derivative (PID) controllers. PID controllers relate the error to the
actuating signal either in a proportional (P), integral (I), or derivative (D) manner. PID controllers can also relate the error to the actuating
signal using a combination of these controls.
9.2.3.1: PROPORTIONAL (P) CONTROL
One type of action used in PID controllers is the proportional control. Proportional control is a form of feedback control. It is the simplest
form of continuous control that can be used in a closed-looped system. P-only control minimizes the fluctuation in the process variable, but
it does not always bring the system to the desired set point. It provides a faster response than most other controllers, initially allowing the P-
only controller to respond a few seconds faster. However, as the system becomes more complex (i.e. more complex algorithm) the response
time difference could accumulate, allowing the P-controller to possibly respond even a few minutes faster. Although the P-only controller
does offer the advantage of faster response time, it produces deviation from the set point. This deviation is known as the offset, and it is
usually not desired in a process. The existence of an offset implies that the system could not be maintained at the desired set point at steady
state. It is analogous to the systematic error in a calibration curve, where there is always a set, constant error that prevents the line from
crossing the origin. The offset can be minimized by combining P-only control with another form of control, such as I- or D- control. It is
important to note, however, that it is impossible to completely eliminate the offset, which is implicitly included within each equation.
P-control linearly correlates the controller output (actuating signal) to the error (diference between measured signal and set point). This P-
control behavior is mathematically illustrated in Equation 9.2.1 (Scrcek, et. al).
c(t) = Kc e(t) + b (9.2.1)
where
c(t) = controller output
Kc = controller gain
e(t) = error
b = bias
In this equation, the bias and controller gain are constants specific to each controller. The bias is simply the controller output when the error
is zero. The controller gain is the change in the output of the controller per change in the input to the controller. In PID controllers, where
signals are usually electronically transmitted, controller gain relates the change in output voltage to the change in input voltage. These
voltage changes are then directly related to the property being changed (i.e. temperature, pressure, level, etc.). Therefore, the gain ultimately
relates the change in the input and output properties. If the output changes more than the input, Kc will be greater than 1. If the change in the
input is greater than the change in the output, Kc will be less than 1. Ideally, if Kc is equal to infinity, the error can be reduced to zero.
However, this infinitesimal nature of Kc increases the instability of the loop because zero error would imply that the the measured signal is
exactly equal to the set point. As mentioned in lecture, exact equality is never achieved in control logic; instead, in control logic, error is
allowed to vary within a certain range. Therefore, there are limits to the size of Kc, and these limits are defined by the system. Graphical
representations of the effects of these variables on the system is shown in PID Tuning via Classical Methods.
As can be seen from the above equation, P-only control provides a linear relationship between the error of a system and the controller output
of the system. This type of control provides a response, based on the signal that adjusts the system so that any oscillations are removed, and
the system returns to steady-state. The inputs to the controller are the set point, the signal, and the bias. The controller calculates the
difference between the set point and the signal, which is the error, and sends this value to an algorithm. Combined with the bias, this
algorithm determines the action that the controller should take. A graphical representation of the P-controller output for a step increase in
input at time t0 is shown below in Figure 2. This graph is exactly similar to the step input graph itself.
Figure 2. P-controller output for step input.
To illustrate this linear P-control relationship, consider the P-only control that controls the level of a fluid in a tank. Initially, the flow into
the tank is equal to the flow out of the tank. However, if the flow out of the tank decreases, the level in the tank will increase because more
fluid is entering than is leaving. The P-only control system will adjust the flow out of the tank so that it is again equal to the flow into the
tank, and the level will once again be constant. However, this level is no longer equal to the initial level in the tank. The system is at steady-
state, but there is a difference between the initial set point and the current position in the tank. This difference is the P-control offset.
9.2.3.2: INTEGRAL (I) CONTROL
Another type of action used in PID controllers is the integral control. Integral control is a second form of feedback control. It is often used
because it is able to remove any deviations that may exist. Thus, the system returns to both steady state and its original setting. A negative
error will cause the signal to the system to decrease, while a positive error will cause the signal to increase. However, I-only controllers are
much slower in their response time than P-only controllers because they are dependent on more parameters. If it is essential to have no
offset in the system, then an I-only controller should be used, but it will require a slower response time. This slower response time can be
reduced by combining I-only control with another form, such as P or PD control. I-only controls are often used when measured variables
need to remain within a very narrow range and require fine-tuning control. I controls affect the system by responding to accumulated past
error. The philosophy behind the integral control is that deviations will be affected in proportion to the cumulative sum of their magnitude.
The key advantage of adding a I-control to your controller is that it will eliminate the offset. The disadvantages are that it can destabilize the
controller, and there is an integrator windup, which increases the time it takes for the controller to make changes.
I-control correlates the controller output to the integral of the error. The integral of the error is taken with respect to time. It is the total error
associated over a specified amount of time. This I-control behavior is mathematically illustrated in Equation 9.2.2 (Scrcek, et. al).
1
c(t) = ∫ e(t)dt + c (t0 ) (9.2.2)
Ti
where
c(t) is the controller output
Ti is the integral time
e(t) is the error
c(t ) is the controller output before integration

0
In Equation 9.2.2 , the integral time is the amount of time that it takes for the controller to change its output by a value equal to the error.
The controller output before integration is equal to either the initial output at time t=0, or the controller output at the time one step before the
measurement. Graphical representations of the effects of these variables on the system is shown in PID Tuning via Classical Methods.
The rate of change in controller output for I-only control is determined by a number of parameters. While the P-only controller was
determined by e, the rate of change for I-only depends on both e and Ti. Because of the inverse relationship between c(t) and Ti, this
decreases the rate of change for an I-only controller.
The I-only controller operates in essentially the same way as a P-only controller. The inputs are again the set point, the signal, and the bias.
Once again, the error is calculated, and this value is sent to the algorithm. However, instead of just using a linear relationship to calculate
the response, the algorithm now uses an integral to determine the response that should be taken. Once the integral is evaluated, the response
is sent and the system adjusts accordingly. Because of the dependence on Ti, it takes longer for the algorithm to determine the proper
response. A graphical representation of the I-controller output for a step increase in input at time t0 is shown below in Figure 3. As
expected, this graph represents the area under the step input graph.
Figure 3. I-controller output for step input.
9.2.3.3: DERIVATIVE (D) CONTROL
Another type of action used in PID controllers is the derivative control. Unlike P-only and I-only controls, D-control is a form of feed
forward control. D-control anticipates the process conditions by analyzing the change in error. It functions to minimize the change of error,
thus keeping the system at a consistent setting. The primary benefit of D controllers is to resist change in the system, the most important of
these being oscillations. The control output is calculated based on the rate of change of the error with time. The larger the rate of the change
in error, the more pronounced the controller response will be.
Unlike proportional and integral controllers, derivative controllers do not guide the system to a steady state. Because of this property, D
controllers must be coupled with P, I or PI controllers to properly control the system.
D-control correlates the controller output to the derivative of the error. The derivative of the error is taken with respect to time. It is the
change in error associated with change in time. This D-control behavior is mathematically illustrated in Equation 9.2.3 (Scrcek, et. al).
de
c(t) = Td (9.2.3)
dt
where
c(t) is the controller output
Td is the derivative time constant
de is the differential change in error
dt is the differential change in time
Graphical representations of the effects of these variables on the system is shown in PID Tuning via Classical Methods.
Mathematically, derivative control is the opposite of integral control. Although I-only controls exist, D-only controls do not exist. D-
controls measure only the change in error. D-controls do not know where the setpoint is, so it is usually used in conjunction with another
method of control, such as P-only or a PI combination control. D-control is usually used for processes with rapidly changing process
outputs. However, like the I-control, the D control is mathematically more complex than the P-control. Since it will take a computer
algorithm longer to calculate a derivative or an integral than to simply linearly relate the input and output variables, adding a D-control
slows down the controller’s response time. A graphical representation of the D-controller output for a step increase in input at time t is 0
shown below in Figure 4. As expected, this graph represents the derivative of the step input graph.
Figure 4. D-controller output for step input.
9.2.4: CONTROLLER EFFECTS ON A SYSTEM

The following images are intended to give a visual representation of how P, I, and D controllers will affect a system.
Description
Figure 5. Stable data sample.
Figure 6. Data disturbance.
Figure 7. P-controller effect on data.
Figure 8. I-controller effect on data.
Figure 9. D-controller effect on data.

Continue reading to see the results of combining controllers.
9.2.4.1: PROPORTIONAL-INTEGRAL (PI) CONTROL
One combination is the PI-control, which lacks the D-control of the PID system. PI control is a form of feedback control. It provides a
faster response time than I-only control due to the addition of the proportional action. PI control stops the system from fluctuating, and it is
also able to return the system to its set point. Although the response time for PI-control is faster than I-only control, it is still up to 50%
slower than P-only control. Therefore, in order to increase response time, PI control is often combined with D-only control.
PI-control correlates the controller output to the error and the integral of the error. This PI-control behavior is mathematically illustrated in
Equation 9.2.4 (Scrcek, et. al).
1
c(t) = Kc (e(t) + ∫ e(t)dt) + C (9.2.4)
Ti
where
c(t) is the controller output,
Kc is the controller gain,
T is the integral time,
i
e(t) is the error, and
C is the initial value of controller
In this equation, the integral time is the time required for the I-only portion of the controller to match the control provided by the P-only part
of the controller.
The equation indicates that the PI-controller operates like a simplified PID-controller with a zero derivative term. Alternatively, the PI-
controller can also be seen as a combination of the P-only and I-only control equations. The bias term in the P-only control is equal to the
integral action of the I-only control. The P-only control is only in action when the system is not at the set point. When the system is at the
set point, the error is equal to zero, and the first term drops out of the equation. The system is then being controlled only by the I-only
portion of the controller. Should the system deviate from the set point again, P-only control will be enacted. A graphical representation of
the PI-controller output for a step increase in input at time t0 is shown below in Figure 5. As expected, this graph resembles the qualitative
combination of the P-only and I-only graphs.
Figure 10. PI-controller output for step input.
9.2.4.2: EFFECTS OF KC AND TI

With a PI control system, controller activity (aggressiveness) increases as Kc and Ti decreases, however they can act individually on the
aggressiveness of a controller’s response. Consider Figure 11 below with the center graph being a linear second order system base case.
Figure 11. Effects of Kc and Ti [2]

The plot depicts how Ti and Kc both affect the performance of a system, whether they are both affecting it or each one is independently
doing so. Regardless of integral time, increasing controller gain (moving form bottom to top on the plot) will increase controller activity.
Similarly, decreasing integral time (moving right to left on the plot) will increase controller activity independent of controller gain. As
expected, increasing Kc and decreasing Ti would compound sensitivity and create the most aggressive controller scenario.
With only two interacting parameters in PI control systems, similar performance plots can still cause confusion. For example, plots A and B
from the figure both look very similar despite different parameters being affected in each of them. This could cause further problems and
create a wildly aggressive system if the wrong parameter is being corrected. While trial and error may be feasible for a PI system, it
becomes cumbersome in PID where a third parameter is introduced and plots become increasingly similar.
Another noteworthy observation is the plot with a normal Kc and double Ti. The plot depicts how the proportional term is practical but the
integral is not receiving enough weight initially, causing the slight oscillation before the integral term can finally catch up and help the
system towards the set point.
9.2.4.3: PROPORTIONAL-DERIVATIVE (PD) CONTROL

Another combination of controls is the PD-control, which lacks the I-control of the PID system. PD-control is combination of feedforward
and feedback control, because it operates on both the current process conditions and predicted process conditions. In PD-control, the control
output is a linear combination of the error signal and its derivative. PD-control contains the proportional control’s damping of the
fluctuation and the derivative control’s prediction of process error.
As mentioned, PD-control correlates the controller output to the error and the derivative of the error. This PD-control behavior is
mathematically illustrated in Equation 5 (Scrcek, et. al).
de
c(t) = Kc (e(t) + Td )+C (9.2.5)
dt
where
Kc = proportional gain
e = error
C = initial value of controller
The equation indicates that the PD-controller operates like a simplified PID-controller with a zero integral term. Alternatively, the PD-
controller can also be seen as a combination of the P-only and D-only control equations. In this control, the purpose of the D-only control is
to predict the error in order to increase stability of the closed loop system. P-D control is not commonly used because of the lack of the
integral term. Without the integral term, the error in steady state operation is not minimized. P-D control is usually used in batch pH control
loops, where error in steady state operation does not need to be minimized. In this application, the error is related to the actuating signal
both through the proportional and derivative term. A graphical representation of the PD-controller output for a step increase in input at time
t0 is shown below in Figure 6. Again, this graph is a combination of the P-only and D-only graphs, as expected.
Figure 12. PD-controller output for step input.
9.2.4.4: PROPORTIONAL-INTEGRAL-DERIVATIVE (PID) CONTROL

Proportional-integral-derivative control is a combination of all three types of control methods. PID-control is most commonly used because
it combines the advantages of each type of control. This includes a quicker response time because of the P-only control, along with the
decreased/zero offset from the combined derivative and integral controllers. This offset was removed by additionally using the I-control.
The addition of D-control greatly increases the controller's response when used in combination because it predicts disturbances to the
system by measuring the change in error. On the contrary, as mentioned previously, when used individually, it has a slower response time
compared to the quicker P-only control. However, although the PID controller seems to be the most adequate controller, it is also the most
expensive controller. Therefore, it is not used unless the process requires the accuracy and stability provided by the PID controller.
PID-control correlates the controller output to the error, integral of the error, and derivative of the error. This PID-control behavior is
mathematically illustrated in Equation 6 (Scrcek, et. al).
1 de
c(t) = Kc (e(t) + ∫ e(t)dt + Td )+C (9.2.6)
Ti dt
where
e(t) = error
Ti = integral time
Td = derivative time constant
C = intitial value of controller
As shown in the above equation, PID control is the combination of all three types of control. In this equation, the gain is multiplied with the
integral and derivative terms, along with the proportional term, because in PID combination control, the gain affects the I and D actions as
well. Because of the use of derivative control, PID control cannot be used in processes where there is a lot of noise, since the noise would
interfere with the predictive, feedforward aspect. However, PID control is used when the process requires no offset and a fast response time.
A graphical representation of the PID-controller output for a step increase in input at time t0 is shown below in Figure 7. This graph
resembles the qualitative combination of the P-only, I-only, and D-only graphs.
Figure 7. PID-controller output for step input.
In addition to PID-control, the P-, I-, and D- controls can be combined in other ways. These alternative combinations are simplifications of
the PID-control.
 NOTE: ORDER OF E(T)

The order of the elements in the e(t) can vary depending on the situation. It could be the fixed element minus the varying element or the
other way around. To better illustrate the concept let's go to an example. Let's say you are creating a PID control to control the fluid
level in a tank by manipulating the outlet valve. When the fluid level in the tank exceeds your set value, you will want the valve to open
up more to allow more flow out of the tank. You are looking for a positive response. Therefore your e(t) should give a positive value
when the fluid level is higher than the set. In this case your e(t) will be (V-Vset). The same logic can be used for other systems to
determine what the e(t) should be in the PID controls.
9.2.5: MODELING PID CONTROLLERS WITH EULER IN EXCEL

As with many engineering systems, PID controllers can be modeled in Excel via numerical methods such as Euler's Method. First begin
with the initial value for a given parameter. Determine the change in that parameter at a certain time-step by summing the three controllers
P, I, and D at that step, which are found using the equations listed in the P, I, D, PI, PD, PID Control section above. Take this change,
multiply it by the chosen time-step and add that to the previous value of the parameter of interest. For more detailed information see
Numerical ODE Solving in Excel. An example of a chemical engineering problem that uses this method can be seen in Example 4 below.
9.2.5.1: TROUBLESHOOTING PID MODELING IN EXCEL
When setting up an Excel spreadsheet to model a PID controller, you may receive an error message saying that you have created a circular
reference. Say you are controlling the flow rate of one reactant (B ) to a reactor which is dependent upon the concentration of another
reactant (A ) already inside the reactor. Your PID equations look as follows:
1 d (A − Aset )
F B = 1 + Kc [(A − Aset ) + ∫ (A − Aset ) dt + τd ]
τi dt
and
d (A − Aset )
xi =
dt
After you have set up your columns for A − A , d(A − A )/dt , x , and the cells for your parameters like K , τ and τ , you will need
set set i c i d
to set up your PID column with your PID equation in it. After entering your equation into the first cell of the PID column, you may receive
the Circular Reference error message when trying to drag the equation down to the other cells in your column.
There are two things you can do:
1. It is likely that you need to start your PID equation in the second or third cell of your PID column. Enter reasonable values into the first
couple of cells before you start the PID equation, and you will find that these values shouldn’t affect the end result of your controller.
2. You can also try decreasing the step size (Δt ).
9.2.5.2: SUMMARY TABLES
A summary of the advantages and disadvantages of the three controls is shown below is shown in Table 1.
Table 1. Advantages and disadvantages of controls
A guide for the typical uses of the various controllers is shown below in Table 2.
Table 2. Typical uses of P, I, D, PI, and PID controllers
A summary of definitions of the terms and symbols are shown below in Table 3.
Table 3. Definitions of terms and symbols.
 EXAMPLE 9.2.1
Hypothetical Industries has just put you in charge of one of their batch reactors. Your task is to figure out a way to maintain a setpoint
level inside of the reactor. Your boss wants to use some type regulator controller, but he is not quite sure which to use. Help your boss
find the right type of controller. It is extremely important that the level inside the reactor is at the setpoint. Large fluctuation and error
cannot be tolerated.
Solution
You would want to use a PID controller. Because of the action of P control, the system will respond to a change very quickly. Due to
the action of I control, the system is able to be returned to the setpoint value. Finally, because it is so critical for the system to remain at
a constant setpoint, D control will measure the change in the error, and help to adjust the system accordingly.
 EXAMPLE 9.2.2
You go back to your high school and you notice an oven in your old chemistry class. The oven is used to remove water from solutions.
Using your knowledge from ChE 466, you begin to wonder what type of controller the oven uses to maintain its set temperature. You
notice some high school students eager to learn, and you decide to share your knowledge with them in hopes of inspiring them to
become Chemical Engineers. Explain to them the type of controller that is most likely located within the oven, and how that controller
works.
Solution
Since the oven is only used to remove water from a solution, fluctuation, error, and lag between the set point and the actual temperature
are all acceptable. Therefore, the easiest and simplest controller to use would be the On-Off controller. The On-Off controller turns on
the heating mechanism when the temperature in the oven is below the setpoint temperature. If the temperature of the oven increases
above the set temperature, the controller will turn the heating mechanism off.
 EXAMPLE 9.2.3
Having taken your advice, your boss at Hypothetical Industries decides to install a PID controller to control the level in the batch
reactor. When you first start up the reactor, the controller initially received a step input. As the reactor achieves steady state, the level in
the reactor tends to fluctuate, sending pulse inputs into the controller. For a pulse input, provide a grahical representation of the PID
controller output.
Figure 8. Pulse input.

Solution
The PID-controller output will be a combination of the P-only, I-only and D-only controller outputs. Analogous to the P-controller
output for the step input, the P-controller output for the pulse input will exactly resemble the input.
Figure 9. P-controller output for pulse input.

The I-controller output represents the area under the input graph. Unlike the step input, the area under the pulse input graph dropped
back down to zero once the pulse has passed. Therefore, rather than continually increase, the I-controller output graph will level off in
the end.
Figure 10. I-controller output for pulse input.

The D-controller output represents the derivative of the input graph. The derivative at the first discontinuity of the graph would be
positive infinity. The derivative of the second downward discontinuity is negative infinity.
Figure 11. D-controller output for pulse input.

Combining the qualitative characteristics of all three graphs we can determine the PID-controller output for a pulse input.
Figure 12. PID-controller output for pulse input.
 EXAMPLE 9.2.3
Different kinds of disturbances are possible when determining the PID controller robustness. These different disturbances are used to
simulate changes that might occur within your system. For a CSTR reactor, you decide to heat up your system to account for the cold
outside weather. The disturbance in the input temperature is a ramp disturbance, as shown in figure #. If the controller responds to the
input temperature, what will be the PID controller output?
Figure 13. Ramp input.

Solution
Using a controller with a p-only controller only, we will see a proportional change in the controller output corresponding to the input
variable change. See figure 14 below
Figure 14. P-controller output for ramp input.

Using an I-only controller, we will see the controller corresponding to the area under the graph, which in this case, seem to increase
exponentially with the ramp geometry.
Figure 15. I-controller output for ramp input.

Using a D-only controller, we will see a step response to the ramp disturbance. This is because the D-component corresponds to the
derivative, and a ramp input shows a constant slope (positive in this case) which is different than the starting condition slope (zero
usually). See figure 16.
Figure 16. D-controller output for ramp input.
Using a PID controller, the three components all come to play in the controller output. As we would expect, the result will be just a
simple addition of the three separate component graphs.
Figure 17. D-controller output for ramp input.
 EXAMPLE 9.2.6
Following is a P&ID of the process A+B-->C.
Figure 18. P&ID for a reaction process.

What is the PID controller expression on V3 controlling the volume in TK001 to a setpoint of 50 liters? Note: The PID controller uses
LC1 to measure the volume.
Solution
The general equation for a PID controller is:
1 de
c(t) = Kc [e(t) + ∫ e(t)dt + Td ]+C
Ti dt
where
e(t) = error
T = integral time
i
T = derivative time constant

d
C = initial value of controller
Therefore, for this example, the solution is:

1 d (V1 − Vvet )
Fout = offset + Kc [(V1 − Vset ) + ∫ (V1 − Vset ) dt + Td ].
Ti dt
 EXAMPLE 9.2.6
In this problem, the differential equations describing a particular first-order system with a PID controller will be determined. This is
designed to show how well-defined systems can be modeled or explained mathematically. As a supplement to this problem, visit
Constructing Block Diagrams. Note that this is an example of solution using integro-differential operators rather than Laplace
transforms. Here we have set K = 1 .
p
Consider a general first-order process:

′
τpY (t) + Y (t) = X(t)
Where Y(t) is the output of the system and X(t) is the input. Add a PID controller to the system and solve for a single, simple
differential equation. The operator or equation for a PID controller is below. Assume there is no dead time in the measurement.
t
1 d
G = Kc [1 + ∫ dτ + τD ]
τI 0
dt
Solution: Use Constructing Block Diagrams as a reference when solving this problem.
Equations defining system,
Process: τpY'(t) + Y(t) = X(t)
Controller: X(t) = Gε(t)
Comparator: ε(t) = R(t) − M(t)
Measurement: M(t) = Y(t)
When these equations are combined into one equation, the following equation results. This is achieved by adding the measurement to
the comparator to the controller to the process equation.
′
τp Y (t) + Y (t) = G(R(t) − Y (t))
Substituting the controller operator and then evaluating yields:

t
′
1 d
τp Y (t) + Y (t) = Kc [1 + ∫ dτ + τD (R(t) − Y (t))]
τI 0 dt
Because there is an integral in the differential equation, it is necessary to take the derivative with respect to time.
To put this in standard notation for solving a second order differential equation, the Y (t) need to be on one side, and the R(t) terms
need to be on the opposite side. Also, the coefficient of the Y (t) term needs to be one.
τI (τp − Kc τD ) τI (1 + Kc )
′′ ′ ′′ ′
Y (t) + Y (t) − Y (t) = τI τD R (t) + τI R (t) + R(t)
Kc Kc
The above equation can then be solved by hand or using a program such as Mathematica. If using a computer program, different values
for the control parameters Kc, τI, τD can be chosen, and the response to a change in the system can be evaluated graphically.
Exercise 9.2.1
What type of controller is displayed by the equation below?
1
c(t) = Kc [e(t) + ∫ e(t)dt]
Ti
a. Feedforward
b. PID
c. Derivative
d. Proportional Integral
Answer
d
Exercise 9.2.2
Which type of controller increases the stability of the system by keeping it at a consistent setting?
a. Derivative
b. Proportional
c. On-Off
d. Integral
Answer
a
Exercise 9.2.3
Which type of controller increases the speed of response to reach the desired set point the fastest while eliminating offset?
a. On-Off
b. Proportional
c. Integral
d. Proportional-Integral
Answer
d
A microbiology laboratory discovered a deadly new strain of bacteria, named P. Woolfi, in the city's water supply. In order to sterilize the
water, the bacteria has to be heat killed at a temperature of 105 degrees Celsius. However, this is above the boiling point of water at 1 atm
and the bacteria is only susceptible to heat when in liquid due to spore formation when in gas or vapor. To accomplish this sterilization it
was suggested that an auto-clave be used to keep the water in the liquid state by keeping it at a pressure of 5 atm while heated for 30
seconds. The auto-clave can only handle up to 7 atm of pressure before exploding, so to ensure that the process is running at the desired
specifications, a PID Controller Model must be created in Excel. See figure 18 for a visual of the system.
Figure 18.Auto-clave with PID Controls for Temperature and Pressure
Click on this link for the worked out Excel Solution
Explanation:
To simulate the real situation of pressure varying in the system, column B calls an equation to generate a random fluctuation in pressure.
Pset is simply the desired specification. Error is the difference between the set pressure and measured signal. du/dt is the sum of the P, I, and
D terms. The equations used to calculate each of these can be found in the article, these take into account the error associated with each
time-step. dU/dt is the parameter that is varied in order to correct for the difference between the measured pressure and desired pressure.
9.2.6: REFERENCES
Astrom, Karl J., Hagglund, Tore., "Advanced PID Control", The Instrumentation, Systems and Automation Society.
Cooper, Douglas J. "Practical Process Control E-Textbook " http://www.controlguru.com
Scrcek, William Y., Mahoney, Donald P., Young, Brent R. "A Real Time Approach to Process Control", 2nd Edition. John Wiley &
Sons, Ltd.
9.2: P, I, D, PI, PD, and PID control is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Ardemis Boghossian, James Brown,
Sara Zak, & Sara Zak via source content that was edited to conform to the style and standards of the LibreTexts platform; a detailed edit history is available
upon request.
9.3: PID TUNING VIA CLASSICAL METHODS
9.3.1: INTRODUCTION
Currently, more than half of the controllers used in industry are PID controllers. In the past, many of these controllers were analog; however,
many of today's controllers use digital signals and computers. When a mathematical model of a system is available, the parameters of the
controller can be explicitly determined. However, when a mathematical model is unavailable, the parameters must be determined
experimentally. Controller tuning is the process of determining the controller parameters which produce the desired output. Controller
tuning allows for optimization of a process and minimizes the error between the variable of the process and its set point.
Types of controller tuning methods include the trial and error method, and process reaction curve methods. The most common classical
controller tuning methods are the Ziegler-Nichols and Cohen-Coon methods. These methods are often used when the mathematical model of
the system is not available. The Ziegler-Nichols method can be used for both closed and open loop systems, while Cohen-Coon is typically
used for open loop systems. A closed-loop control system is a system which uses feedback control. In an open-loop system, the output is not
compared to the input.
The equation below shows the PID algorithm as discussed in the previous PID Control section.
t
1 dϵ(t)
′ ′
u(t) = Kc (ϵ(t) + ∫ ϵ (t ) dt + τd )+b
τi 0
dt
where
u is the control signal
ε is the difference between the current value and the set point.
Kc is the gain for a proportional controller.
τi is the parameter that scales the integral controller.
τd is the parameter that scales the derivative controller.
t is the time taken for error measurement.
b is the set point value of the signal, also known as bias or offset.
The experimentally obtained controller gain which gives stable and consistent oscillations for closed loop systems, or the ultimate gain, is
defined as Ku . Kc is the controller gain which has been corrected by the Ziegler-Nichols or Cohen-Coon methods, and can be input into the
above equation. Ku is found experimentally by starting from a small value of Kc and adjusting upwards until consistent oscillations are
obtained, as shown below. If the gain is too low, the output signal will be damped and attain equilibrium eventually after the disturbance
occurs as shown below.
On the other hand, if the gain is too high, the oscillations become unstable and grow larger and larger with time as shown below.
The process reaction curve method section shows the parameters required for open loop system calculations. The Ziegler-Nichols Method
section shows how to find Kc, Ti, and Td for open and closed loop systems, and the Cohen-Coon section shows an alternative way to find Kc,
Ti, and Td.
Open loop systems typically use the quarter decay ratio (QDR) for oscillation dampening. This means that the ratio of the amplitudes of the
first overshoot to the second overshoot is 4:1.
9.3.2: TRIAL AND ERROR

The trial and error tuning method is based on guess-and-check. In this method, the proportional action is the main control, while the integral
and derivative actions refine it. The controller gain, Kc, is adjusted with the integral and derivative actions held at a minimum, until a
desired output is obtained.
Below are some common values of Kc, Ti, and Td used in controlling flow, levels, pressure or temperature for trial and error calculations.
9.3.2.1: FLOW
P or PI control can be used with low controller gain. Use PI control for more accuracy with high integration activity. Derivative control is
not considered due to the rapid fluctuations in flow dynamics with lots of noise.
Kc = 0.4-0.65
Ti = 6s
9.3.2.2: LEVEL
P or PI control can be used, although PI control is more common due to inaccuracies incurred due to offsets in P-only control. Derivative
control is not considered due to the rapid fluctuations in flow dynamics with lots of noise.
The following P only setting is such that the control valve is fully open when the vessel is 75% full and fully closed when 25% full, being
half open when 50% filled.
Kc = 2
Bias b = 50%
Set point = 50%
For PI control:
Kc = 2-20
Ti = 1-5 min
9.3.2.3: PRESSURE
Tuning here has a large range of possible values of Kc and Ti for use in PI control, depending on if the pressure measurement is in liquid or
gas phase.
Liquid
Kc = 0.5-2
Ti = 6-15 s
Gas
Kc = 2-10
Ti = 2-10 min
9.3.2.4: TEMPERATURE
Due to the relatively slow response of temperature sensors to dynamic temperature changes, PID controllers are used.
Kc = 2-10
Ti = 2-10 min
Td = 0-5 min
9.3.3: PROCESS REACTION CURVE

In this method, the variables being measured are those of a system that is already in place. A disturbance is introduced into the system and
data can then be obtained from this curve. First the system is allowed to reach steady state, and then a disturbance, Xo, is introduced to it.
The percentage of disturbance to the system can be introduced by a change in either the set point or process variable. For example, if you
have a thermometer in which you can only turn it up or down by 10 degrees, then raising the temperature by 1 degree would be a 10%
disturbance to the system. These types of curves are obtained in open loop systems when there is no control of the system, allowing the
disturbance to be recorded. The process reaction curve method usually produces a response to a step function change for which several
parameters may be measured which include: transportation lag or dead time, τdead, the time for the response to change, τ, and the ultimate
value that the response reaches at steady-state, Mu.
τdead = transportation lag or dead time: the time taken from the moment the disturbance was introduced to the first sign of change in the
output signal
τ = the time for the response to occur
Xo = the size of the step change
Mu = the value that the response goes to as the system returns to steady-state
τdead
R =
τ
Xo τ
Ko =
Mu τdead
An example for determining these parameters for a typical process response curve to a step change is shown below.
In order to find the values for τdead and τ, a line is drawn at the point of inflection that is tangent to the response curve and then these values
are found from the graph.
To map these parameters to P,I, and D control constants, see Table 2 and 3 below in the Z-N and Cohen Coon sections.
9.3.4: ZIEGLER-NICHOLS METHOD

In the 1940's, Ziegler and Nichols devised two empirical methods for obtaining controller parameters. Their methods were used for non-first
order plus dead time situations, and involved intense manual calculations. With improved optimization software, most manual methods such
as these are no longer used. However, even with computer aids, the following two methods are still employed today, and are considered
among the most common:
9.3.4.1: ZIEGLER-NICHOLS CLOSED-LOOP TUNING METHOD

The Ziegler-Nichols closed-loop tuning method allows you to use the ultimate gain value, Ku, and the ultimate period of oscillation, Pu, to
calculate Kc . It is a simple method of tuning PID controllers and can be refined to give better approximations of the controller. You can
obtain the controller constants Kc , Ti , and Td in a system with feedback. The Ziegler-Nichols closed-loop tuning method is limited to
tuning processes that cannot run in an open-loop environment.
Determining the ultimate gain value, Ku, is accomplished by finding the value of the proportional-only gain that causes the control loop to
oscillate indefinitely at steady state. This means that the gains from the I and D controller are set to zero so that the influence of P can be
determined. It tests the robustness of the Kc value so that it is optimized for the controller. Another important value associated with this
proportional-only control tuning method is the ultimate period (Pu). The ultimate period is the time required to complete one full oscillation
while the system is at steady state. These two parameters, Ku and Pu, are used to find the loop-tuning constants of the controller (P, PI, or
PID). To find the values of these parameters, and to calculate the tuning constants, use the following procedure:
Closed Loop (Feedback Loop)
1. Remove integral and derivative action. Set integral time (Ti) to 999 or its largest value and set the derivative controller (Td) to zero.
2. Create a small disturbance in the loop by changing the set point. Adjust the proportional, increasing and/or decreasing, the gain until the
oscillations have constant amplitude.
3. Record the gain value (Ku) and period of oscillation (Pu).
Figure 1. System tuned using the Ziegler-Nichols closed-loop tuning method

4. Plug these values into the Ziegler-Nichols closed loop equations and determine the necessary settings for the controller.
Table 1. Closed-Loop Calculations of Kc, Ti, Td
Advantages
1. Easy experiment; only need to change the P controller
2. Includes dynamics of whole process, which gives a more accurate picture of how the system is behaving
Disadvantages
1. Experiment can be time consuming
2. Can venture into unstable regions while testing the P controller, which could cause the system to become out of control
9.3.4.2: ZIEGLER-NICHOLS OPEN-LOOP TUNING METHOD OR PROCESS REACTION METHOD:

This method remains a popular technique for tuning controllers that use proportional, integral, and derivative actions. The Ziegler-Nichols
open-loop method is also referred to as a process reaction method, because it tests the open-loop reaction of the process to a change in the
control variable output. This basic test requires that the response of the system be recorded, preferably by a plotter or computer. Once
certain process response values are found, they can be plugged into the Ziegler-Nichols equation with specific multiplier constants for the
gains of a controller with either P, PI, or PID actions.
Open Loop (Feed Forward Loop)
To use the Ziegler-Nichols open-loop tuning method, you must perform the following steps:
1. Make an open loop step test
2. From the process reaction curve determine the transportation lag or dead time, τdead, the time constant or time for the response to change,
τ, and the ultimate value that the response reaches at steady-state, Mu, for a step change of Xo.
3. Determine the loop tuning constants. Plug in the reaction rate and lag time values to the Ziegler-Nichols open-loop tuning equations for
the appropriate controller—P, PI, or PID—to calculate the controller constants. Use the table below.
Table 2. Open-Loop Calculations of Kc, Ti, Td
Advantages
1. Quick and easier to use than other methods
2. It is a robust and popular method
3. Of these two techniques, the Process Reaction Method is the easiest and least disruptive to implement
Disadvantages
1. It depends upon purely proportional measurement to estimate I and D controllers.
2. Approximations for the Kc, Ti, and Td values might not be entirely accurate for different systems.
3. It does not hold for I, D and PD controllers
9.3.5: COHEN-COON METHOD

The Cohen-Coon method of controller tuning corrects the slow, steady-state response given by the Ziegler-Nichols method when there is a
large dead time (process delay) relative to the open loop time constant; a large process delay is necessary to make this method practical
because otherwise unreasonably large controller gains will be predicted. This method is only used for first-order models with time delay,
due to the fact that the controller does not instantaneously respond to the disturbance (the step disturbance is progressive instead of
instantaneous).
The Cohen-Coon method is classified as an 'offline' method for tuning, meaning that a step change can be introduced to the input once it is
at steady-state. Then the output can be measured based on the time constant and the time delay and this response can be used to evaluate the
initial control parameters.
For the Cohen-Coon method, there are a set of pre-determined settings to get minimum offset and standard decay ratio of 1/4(QDR). A
1/4(QDR) decay ratio refers to a response that has decreasing oscillations in such a manner that the second oscillation will have 1/4 the
amplitude of the first oscillation . These settings are shown in Table 3..
Table 3. Standard recommended equations to optimize Cohen Coon predictions
where the variables P, N, and L are defined below.
Alternatively, K0 can be used instead of (P/NL). K0,τ, and τdead are defined in process reaction curve section. An example using these
parameters is shown here [1].
The process in Cohen-Coon turning method is the following:
1. Wait until the process reaches steady state.
2. Introduce a step change in the input.
3. Based on the output, obtain an approximate first order process with a time constant τ delayed by τdead units from when the input step was
introduced.
The values of τ and τdead can be obtained by first recording the following time instances:
t0 = time at input step start point t2 = time when reaches half point t3 = time when reaches 63.2% point
4. Using the measurements at t0, t2, t3, A and B, evaluate the process parameters τ, τdead, and Ko.
5. Find the controller parameters based on τ, τdead, and Ko.
9.3.5.1: ADVANTAGES
1. Used for systems with time delay.
2. Quicker closed loop response time.
9.3.5.2: DISADVANTAGES AND LIMITATIONS

1. Unstable closed loop systems.
2. Can only be used for first order models including large process delays.
3. Offline method.
4. Approximations for the Kc, τi, and τd values might not be entirely accurate for different systems.
9.3.6: OTHER METHODS

These are other common methods that are used, but they can be complicated and aren't considered classical methods, so they are only
briefly discussed.
9.3.6.1: INTERNAL MODEL CONTROL
The Internal Model Control (IMC) method was developed with robustness in mind. The Ziegler-Nichols open loop and Cohen-Coon
methods give large controller gain and short integral time, which isn't conducive to chemical engineering applications. The IMC method
relates to closed-loop control and doesn't have overshooting or oscillatory behavior. The IMC methods however are very complicated for
systems with first order dead time.
9.3.6.2: AUTO TUNE VARIATION

The auto-tune variation (ATV) technique is also a closed loop method and it is used to determine two important system constants (Pu and
Ku for example). These values can be determined without disturbing the system and tuning values for PID are obtained from these. The
ATV method will only work on systems that have significant dead time or the ultimate period, Pu, will be equal to the sampling period.
 EXAMPLE 9.3.1
You're a controls engineer working for Flawless Design company when your optimal controller breaks down. As a backup, you figure
that by using coarse knowledge of a classical method, you may be able to sustain development of the product. After adjusting the gain
to one set of data taken from a controller, you find that your ultimate gain is 4.3289.
From the adjusted plot below, determine the type of loop this graph represents; then, please calculate Kc, Ti, and Td for all three types of
controllers.
Solution
From the fact that this graph oscillates and is not a step function, we see that this is a closed loop. Thus, the values will be calculated
accordingly.
We're given the Ultimate gain, Ku = 4.3289. From the graph below, we see that the ultimate period at this gain is Pu = 6.28
From this, we can calculate the Kc, Ti, and Td for all three types of controllers. The results are tabulated below. (Results were calculated
from the Ziegler-Nichols closed-loop equations.)
 EXAMPLE 9.3.2
Your partner finds another set of data after the controller breaks down and decides to use the Cohen-Coon method because of the slow
response time for the system. They also noticed that the control dial, which goes from 0-8, was set at 3 instead of 1. Luckily the
response curve was obtained earlier and is illustrated below. From this data he wanted to calculate Kc, Ti and Td. Help him to determine
these values. Note that the y-axis is percent change in the process variable.
Solution
In order to solve for Kc, Ti and Td, you must first determine L, ΔCp, and T. All of these values may be calculated by the reaction curve
given.
From the process reaction curve we can find that:

L = 3
T = 11
$ΔC_p = 0.55\, (55%)$
Now that these three values have been found N and R may be calculated using the equations below.
N=
R=
Using these equations you find that
N = .05
R = 0.27
We also know that since the controller was moved from 1 to 3, so a 200% change.
P = 2.00
We use these values to calculate Kc, Ti, and Td, for the three types of controllers based on the equations found in Table 3.
9.3.6.3: TUNING THE PARAMETERS OF A PID CONTROLLER ON AN ACTUAL SYSTEM

There are several ways to tune the parameters of a PID controller. They involve the following procedures. For each, name the procedure and
explain how the given measured information is used to pick the parameters of the PID controller.
 EXAMPLE 9.3.1
a. The controller is set to P only, and the system is operated in "closed-loop", meaning that the controller is connected and working.
The gain is tuned up until a resonance is obtained. The amplitude and frequency of that resonance is measured.
b. The system is kept in "open-loop" mode, and a step-function change is manually made to the system (through a disturbance or
through the controller itself). The resulting response of the system is recorded as a function of time.
Solution
a. We will use the Ziegler-Nichols method.
Ki=0.5Ku
Ku is the ultimate gain when the system started oscillating.
b. We will use the Cohen-Coon method.

We can locate the inflection point of the step function and draw a tangent. Τdead is located at the crossing of that tangent with t, and Τ is
located at the cross of the tangent with M(t)
Exercise 9.3.1
Which of the following do you RECORD during the Ziegler-Nichols Method?
1. K c
2. τi
3. K o
4. τd
Answer
Answer:C
Exercise 9.3.2
For the Ziegler-Nichols Method, it is important to:
1. Find a gain that produces damped oscillation
2. Set P and I controllers to zero
3. Record the period of oscillation
4. Calculate Tc
Answer
Answer:A,C
9.3.7: REFERENCES
Svrcek, William Y., Mahoney, Donald P., Young, Brent R. A Real Time Approach to Process Control, 2nd Edition. John Wiley & Sons,
Ltd.
Astrom, Karl J., Hagglund, Tore., Advanced PID Control, ISA, The Instrumentation, Systems and Automation Society.
"ACT Ziegler-Nichols Tuning," ourworld.compuserve.com/homepages/ACTGMBH/zn.htm
Ogata, Katsuhiko. System Dynamics, 4th Edition. Pearson Education, Inc.
9.3: PID Tuning via Classical Methods is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by James Bennett, Ajay Bhasin,
Jamila Grant, Wen Chung Lim, & Wen Chung Lim via source content that was edited to conform to the style and standards of the LibreTexts platform; a
detailed edit history is available upon request.
9.4: PID TUNING VIA FREQUENCY RESPONSES WITH BODE PLOTS
9.4.1: DESCRIPTION OF AMPLITUDE RATIO, PHASE SHIFT AND FREQUENCY RESPONSE
A good way to understand the physical meaning of amplitude ratio and phase shift and how they change as a function of the frequency of
the input stream is through the analysis of a simple system. A well-insulated continuously-stirred tank reactor (CSTR) with an input flow
equal to its output flow is a good example (Figure 9.4.1 ). For this description we consider inlet and outlet temperature, but inlet
concentration or feed rate could also be used.
Motor Feed
Cooling jacket
Baffle
Agitator
Mixed product
Figure 9.4.1: Schematics of a jacketed agitated CSTR. Agitator geometry is a "Rushton type" (the most common in chemical industry), but
many other impeller geometries can be used. (CC BY-SA 3.0; Daniele Pugliesi via Wikipedia)
If the CSTR is initially at 55˚C, and the inlet flow has a temperature that fluctuates sinusoidally between 50˚C and 60˚C, the outlet
temperature will also vary sinusoidally. If the fluid has a large residence time in the tank, and the frequency of the temperature variation of
the input fluid is high relative to the residence time. Then the outlet temperature will also vary quickly in a sinusoidal fashion, but with a
significantly smaller amplitude than that of the inlet. The large hold up in the well mixed tank dampens the fluctuations in the inlet
temperature. In contrast, if the the frequency of the inlet temperature sine wave is low relative to the fluid residence time, the output will
also be a sine wave with temperatures ranging from 50˚C to 60˚C.
In both cases described above there will be a lag or phase shift between the input and the output of the system (see the figure immediately
below). In our example, the phase shift is controlled by the average residence time of fluid in the tank. The time constant of this system is
the average residence time of the tank. A system with a small residence time will respond very quickly to changes in input flow rate, and
large residence times will result in sizable lag time.
As frequency of the temperature variation increases, the period of the input decreases and the phase shift becomes larger. This can be seen in
Figure 9.4.2 :
Figure 9.4.2: Solid line = input, dashed line = output (graphs adapted from Bequette)
Imagine that we have a sensitive reaction that must be maintained with 3˚C of 55˚C, it now becomes of utmost importance that we
understand the frequency response of this CSTR system. Here in lies the utility of Bode plots; they depict a range of different frequency
responses in two plots. This allows a relatively rapid determination of system robustness. In addition, frequency response and Bode plot
analysis can be used to tune PID control systems.
9.4.2: FREQUENCY RESPONSE

For a given process described by Y (t) = G
^
X(t) , one considers a sinusoidal input
X(t) = sin(ωt)
After a long time, the solution will also be sinusoidal, with

Y (t) ∼ AR sin(ωt + ϕ)
where AR is the amplitude ratio (must be a positive number) and ϕ is the phase lag (ϕ < 0 ) or phase lead (ϕ > 0 ).
−1 −1
Sometimes, to solve for Y (t) using an ODE solver, it is useful to have the process equation in the form X(t) = G
^
Y (t) and define ^
G
instead of G.
^
9.4.2.1: EXPRESSIONS FOR AR AND ϕ

First-Order System
1
AR = −−−−−−−
2 2
√ 1 + ω τp
−1
ϕ = tan (−ω τp )
Second-Order System
1
AR = −−−−−−−−−−−−−−−−−−
2 2
2 2
√ (1 − ω τp ) + (2ζω τp )
−2ζp ω τp
−1
ϕ = tan ( )
2
1 − ω2 τp
Dead Time
\[A R=1]
ϕ = −ωτdead
Systems in series
AR = AR1 ⋅ AR2 ⋅ …
log AR = log AR1 + log AR2 + …
ϕ = ϕ1 + ϕ2 + …
P Controller
^
Gc = Kc
AR = Kc
ϕ = 0
PI Controller
t
1
^ ′
Gc = Kc (1 + ∫ dt )
τI 0
−1
1
ϕ = tan (− )
ωτI
PD Controller
−−−−−−−
2 2
AR = Kc √ 1 + ω τ
D
−1
ϕ = tan (ω τD )
PID Controller
−1
1
ϕ = tan (ω τD − )
ωτI
9.4.3: INTRODUCTION TO BODE PLOTS

Because it’s impossible to perfectly model a real chemical process, control engineers are interested in characterizing the robustness of the
system – that is, they want to tune controllers to withstand a reasonable range of change in process parameters while maintaining a stable
feedback system. Bode plots provide an effective means to quantify the system’s stability. Bode plots describe an open or closed-loop
system as a function of input frequency and give a picture of the system’s stability. If all inputs into a system were constant, it would be a
relatively simple task to control the system and its output. However, if an input value, such as temperature, varies sinusoidally, then the
output should also be describable as a sinusoidal function. In this case Bode plots are a useful tool for predicting the response of the system.
There are two Bode plots used to describe a system. The first shows amplitude ratio as a function of frequency. The amplitude ratio is the
amplitude of the output sine wave divided by the input sine wave. The second plot graphs phase shift as a function of frequency, where
phase shift is the time lag between the output and the input sine curves divided by the period of either of the sine curves and then multiplied
by -360° (or −2π radians) to express this shift as an angle.
9.4.4: CONSTRUCTING BODE PLOTS

Bode plots concisely display all relevant frequency input and output information on two plots: amplitude ratio as a functions of frequency
and phase shift as a function of frequency. The amplitude ratio plot is a log-log plot while the phase angle plot is a semilog (or log-linear)
plot.
To construct a Bode plot, an engineer would have empirical data showing input and output values that vary as sinusoidal functions of time.
For instance, there might be inlet temperature data that varies sinusoidally and the outlet temperature data that also varies sinusoidally. For
example, given a heat exchanger that is used to cool a hot stream, we can vary the flow rate of the hot stream and model it sinusodially.
(This can be done using ODEs & an Excel model of a heat exchanger.)
Figure 9.4.3 includes inlet flowrate (process fluid stream) varying as a sinusoidal function and the hot temperature exiting the heat
exchanger likewise varying as a sinusoidal function. Portions of this graph are not yet at steady-state.
Figure 9.4.3: Inlet Flowrate modeled by F s

= 0.1 + 0.1 sin(
π
4t
) .
Bode plots are constructed from steady-state data. Figure 9.4.4 shows part of the steady-state region of the same data used for Figure 9.4.3 .
Figure 9.4.4: Steady-state inlet flow rate and outlet temperatures varying as sunusoidal functions.
To collect a single data point for a Bode plot we will use the information from a single period of the inlet flow rate and the corresponding
temperature from the hot exiting stream.
The amplitude ratio, AR , is the ratio of the amplitude of the output sinusoidal curve divided by the amplitude of the input sinusoidal curve.
output amplitude
AR =
input amplitude
The value of the amplitude ratio should be unitless so if the units of the input frequency and the units of the output frequency are not the
same, the frequency data should first be normalized. For example, if the input frequency is in ˚C/min and the output frequency is also in
˚C/min, then AR is unitless and does not need to be normalized. If, instead, the input frequency is in L/min and the output frequency is in
˚C/min, then AR = ˚C/liters. In this case the inlet and outlet frequencies need to be normalized because the ratio AR = ˚C/L doesn’t say
anything about the physical meaning of the system. The value of AR would be completely different if the units of AR were Kelvin/gal.
To find the phase shift, the periods of the input and output sine curves need to be found. Recall that the period, P , is the length of time from
one peak to the next.
1 2π
P = =
f ω
with f is the frequency and ω is the angular frequency (rad/sec)

The phase shift is then found by
ΔP
ϕ = −2π ( )
P
Using these values found from multiple perturbations in feed flow rate it is possible to construct the following Bode plots:
Figure 9.4.5: (left) Bode Plot (Amplitude Ratio) [Log-Log Plot]. (right) Bode Plot (Phase Shift) [Semi-Log]
9.4.4.1: RULES OF THUMB WHEN ANALYZING BODE PLOTS

Generally speaking, a gain change shifts the amplitude ratio up or down, but does not affect the phase angle. A change in the time delay
affects the phase angle, but not the amplitude ratio. For example, an increase in the time delay makes the phase shift more negative for any
given frequency. A change in the time constant changes both the amplitude ratio and the phase angle. For example, an increase in the time
constant will decrease the amplitude ratio and make the phase lag more negative at any given frequency.
Table 9.4.1: Rules of Thumb
Change Effect
Gain Change Amplitude ratio is shifted up or down; no change in phase angle
Time Delay Phase angle changes, but no change in amplitude ratio
 HISTORICAL NOTE
Prior to the advent of powerful computer modeling tools, controls engineers would model systems using transfer functions. Readers
interested in learning more about how these were used to construct Bode plots should refer to Bequette's Process Control: Modeling,
Design, and Simulation (see References). This wiki assumes that the engineer already has data in Excel, etc, that shows the sinusoidal
behavior of input and outputs.
9.4.5: BODE STABILITY CRITERION

A control system is unstable if the open-loop frequency response exhibits an AR exceeding unity at the frequency for which the phase lag is
180 degrees. This frequency is called the crossover frequency. Adjusting the gain so that the AR < 1 for - phi < (180 minus phase margin)
results in acceptable stability if the phase margin is greater than 30 degrees.
As previously mentioned, the controls engineer uses Bode plots to characterize the stability of the system. If you were to find that the
amplitude ratio is greater than 1 the system would be unstable at that frequency. An important stability criterion is that AR < 1 when phase
shift = -180 degrees (or -pi radians). This is called the Bode stability criterion and if it holds for a closed-loop system, then that system will
be stable. If this criterion is not satisfied and you were to put a feedback controller on this process, as done in PID tuning via optimization,
the process would explode due to the response lag. The reason for this is that the system might think temperature is too low when it's
actually to high, so it would decrease cooling water rather than increase it. Also, with an AR < 1, an error in input is amplified in the output,
so the system will lose control due to amplification of error. This could obviously be a dangerous situation, for example in an exothermic
CSTR, which could explode.
9.4.6: FREQUENCY RESPONSE IN MATHEMATICA

Bode plot for PD controller, ΤD = 1:
AR = Sqrt[1 + omega^2]
phi = 180/Pi*ArcTan[omega]
LogLogPlot[AR,{omega,0.01,100}]
LogLinearPlot[phi,{omega,0.01,100}]
Bode plot for PID controller, using the \Manipulate" function to watch the effect of varying the values of both Tau_I and Tau_D:
Manipulate[LogLogPlot[Sqrt[1 + (omega*tauD - 1/(tauI*omega))^2],{omega,0.01,100}],{tauI,0.1,10},{tauD,0,10}]
Manipulate[LogLinearPlot[180/Pi*ArcTan[omega*tauD - 1/(tauI*omega)],{omega,0.01,100}],{tauI,0.1,10},{tauD,0,10}]
Bode plot for open-loop operation of third-order system with a P controller:
AR = Sqrt[1/(1+omega^2)]*Sqrt[1/(1+omega^2)]*Sqrt[1/(1+25*omega ^ 2)] phi = 180/Pi*(ArcTan[-omega] + ArcTan[-omega] +
ArcTan[-5*omega]) LogLogPlot[AR,{omega,0.01,100}] LogLinearPlot[phi,{omega,0.01,100}]
Find crossover frequency:

FindRoot[phi==-180,{omega,1}]
This yields omega = 1.18322 which is the crossover frequency, and can be used to determine the amplitude ratio (ask mathematica for AR)
the gain (Kc=1/AR)
To determine what the gain should be set to at a phase margin of 30 degrees (safe operating conditions) use the following in mathematica:
FindRoot[phi==-150,{omega,1}]
The phase margin of Cohen-Coon and other frequency analysis (such as Zielger-Nicholas) can be calculated by using the gain (Kc) value
gotten from the method a typing the following into mathematica:
NSolve[Kc*AR==1,omega]
9.4.7: DERIVATIONS
Assume a first-order process with input X(t), output Y (t), and operator
−1 δ
^
G = (τ + 1)
δt
where
−1 δ
^ ′
X(t) = G Y (t) = (τ + 1) Y (t) = τ Y (t) + Y (t) (9.4.1)
δt
For Bode Plot Analysis, we assume a periodic input to the system:

X(t) = sin(ωt)
Based on this input, we can hypothesize that:
Y (t) = AR ∗ sin(ωt + φ)
From this hypothesis, we can replace Equation 9.4.1 with our functions for X(t) and Y (t). This is shown below.
′
X(t) = τ Y (t) + Y (t) (9.4.2)
δ
sin(ωt) = τ (AR sin(ωt + ϕ)) + AR sin(ωt + ϕ) (9.4.3)
δt
= AR(τω cos(ωt + φ)) + sin(ωt + φ) (9.4.4)
Using trigonometric identities, we can replace cos(ωt + φ) with cos(ωt)cos(φ) − sin(ωt)sin(φ) and sin(ωt + φ) with
sin(ωt)cos(φ) + cos(ωt)sin(φ) to get:
sin(ωt) = AR(ωT (cos(ωt) cos(φ) − sin(ωt) sin(φ)) + sin(ωt) cos(φ) + cos(ωt) sin(φ)) (9.4.5)
To solve for AR and φ , we can match the coefficients for the sin(ωt) term and cos(ωt) term on the left and right side of Equation 9.4.5 .
Coefficient of cos(ωt):
0 = AR(ωτ cos(φ) + sin(φ)) (9.4.6)
Coefficient of sin(ωt):
1 = AR(−ωτ sin(φ) + cos(φ)) (9.4.7)
From Equation 9.4.6 , we can divide both sides by AR cos(φ) and rearranging to get:
tan(ϕ) = −ωτ (9.4.8)
or
−1
ϕ = tan (−ωτ) (9.4.9)
Based on this relationship, we can draw a triangle with the angle φ and the relationship shown in Equation 9.4.9 . A picture of the triangle is
shown below.
From Equation 9.4.7 and the triangle above, we can solve for AR:
1
AR = (9.4.10)
−ωτ(sin(ϕ)) + cos(ϕ)
1
= (9.4.11)
−ωτ 1
−ωτ ( )+
√1+ω2 τ 2
√1+ω2 τ 2
1
= (9.4.12)
− −−−−− −
√ 1 + ω2 τ 2
9.4.8: DETERMINING STABLE CONTROLLER GAIN AND PHASE MARGIN

Tuning via phase margin is a more precise method than tuning through Cohen-Coon Rule and Ziegler-Nichols Tuning Rule. We can find
the gain and phase margin by using bode plots or by an analytical method.
 GRAPHICAL METHOD
To find a safe controller gain, use the following steps:
1. Construct Bode Plot of log AR versus log ωτ and φ versus log ωτ
2. Use the above plot to find ωτ such that φ = −180 + φo
. PM
3. Use ωτ found from above step to find AR in log AR versus log ωτ plot.
4. K Csafe =
1
AR
To find the phase margin, use the following steps:
1. Construct Bode Plot of LogAR versus Log ωτ and φ versus Log ωτ
2. Use the above plot to find ωτ such that AR=1.
3. Use ωτ found from above step to find φ in φ versus Log ωτ plot.
4. φPM = 180 − | φAR = 1 |
 ANALYTICAL METHOD
Finding the gain with a phase margin:

1. Use your expression for Φ, set Φ to the desired value and find the root for ω.
2. Input the value obtained for ω in your expression for AR.
3. K =
C
1
AR
Finding the phase margin with a gain:
1. Use your expression for AR, set and solve for ω.

2. Input the value obtained for ω in your expression for Φ.
9.4.9: DETERMINING THE SLOPE OF A BODE PLOT

Sometimes it is necessary to determine the relative shape of a Bode Plot without the use of a computer or ODE equation solver. In sketching
the relative shape, there are three main parts of a basic Bode Plot that must be considered.
1. Crossing point
2. Starting value
3. Slope
9.4.9.1: CROSSING POINT

The crossing point of the Bode Plot is when the frequency, ω, is equal to 1 / τ . If the x-axis of the Bode Plot is ωτ, the crossing point will be
at x-axis = 1. If there are multiple values of τ, there may be multiple crossing points, each at the frequency where ω is equal to 1 / τ . If there
are multiple τ values, and the x-axis is ωτ, it must be specified which τ the x-axis is referring to.
9.4.9.2: STARTING VALUE
The starting value depends on the K and K values contributing to the process.
C P
9.4.9.3: SLOPE
The slope for an uncontrolled process is equal to the negative of the order of the process. For example, if there is a first-order process with
no controller the slope of the Bode Plot would be -1 (-1 for first order), after the crossing point has been reached. A picture of this Bode Plot
is shown below.
For a process controlled by a PD controller, the AR is essentially the inverse of the first-order system AR, meaning that the slope addition
from a PD controller is a +1, instead of a -1 as in a first-order process. For a second-order system with a PD controller the final slope will be
-1 (-2 from the second order process, +1 from PD). However, there may be multiple crossing points, whose location depends on the value of
τP and τD. Remember that the crossing point is where ωτ equals one. Since the slope contribution only comes after the crossing point, a
Bode Plot for τD less than τP would look like the picture below.
Crossing point 1 is where ωτP is 1 and crossing point 2 is where ωτD is 1.
Similarly, if τD is greater than τ, the slope would first go to +1, then to -1.
For a PI controller, the slope contribution from the controller comes before the crossing point, and then goes to zero after the crossing point.
A PID controller would therefore look like the picture shown below, assuming τD = τI .
 EXAMPLE 9.4.1: DETERMINE THE AMPLITUDE RATIO
You are a controls engineer and wish to characterize a heat exchanger used in a chemical process. One of the many things you are
interested in knowing about the system is how the hot outlet temperature responds to fluctuations in the inlet flow rate. Using data for a
particular inlet flow rate, you graphed normalized (why?) flow rate and normalized hot outlet temperature vs. frequency (rad/min). Use
this graph to determine the amplitude ratio.
Solution
Because flow rate units and temperature do not have the same units, these values needed to be normalized before calculating the
amplitude ratio. To normalize flow, use the following equation:
F − Fmin
Fnorm =
Fmax − Fmin
where F is the flow rate of a particular data point.

In this problem temperature, T , is in ˚C, so
T − 273.15
Tnorm =
100
To find the amplitude of both wave functions, first recall that the amplitude of a wave is the maximum height of a wave crest (or depth
of a trough). For one steady-state wave produced from a column of values in Excel, you could quickly calculate the amplitude by using
the max( ) and min( ) Excel functions. [This can be found using Excel help.] If you opted to find the amplitude this way, then the
amplitude for a single wave function would be
[max() − min()]
Amplitude =
2
Note that this is just one way to find the amplitude of a wave. It could also be found simply by reading the values off the graph of a
wave.
Once the amplitudes of the inlet and outlet waves are found, use the values to find the amplitude ratio, AR, as follows:
outlet streams amplitude 0.0486
AR = = = 0.0974
inlet streams amplitude 0.499
 EXAMPLE 9.4.1: DETERMINE THE PHASE SHIFT
The following graph shows how inlet flow and both the hot and cold outlet temperatures vary as sinusoidal functions of time. This
graph was generated using the same data for the heat exchanger of Example 1. Use this graph to find the phase shift between the inlet
flow and the hot outlet stream.
Solution
Determine the period (P) – This can be done by finding the time from one peak to the next of a given wave. In this case, we want to
know the period of the inlet flow rate, so P = 1.14 s .
Determine the lag (delta P) – This can be done by finding the time difference between the peak of the inlet flow rate and the peak of the
hot outlet stream. (Remember that the hot outlet wave lags the wave of the inlet flowrate).
δP = 0.87s
0.87 s
Phaseshift = ∗ −2π = −4.80
1.14 s
Since we're only concerned with time values for finding the phase shift, the data doesn't need to be normalized.
9.4.10: REFERENCES
Bequette, B. Wayne. Process Control: Modeling, Design, and Simulation. pp 172-178, 214-235. Upper Saddle River, N.J. Prentice Hall,
PTR 2003.
Liptak, Bela G. Process Control and Optimization. Volume 2. Instrument Engineers Handbook. Fourth Edition. CRC Press, Boca Raton
Fl. 2003. Insertformulahere
written by: Tony Martus, Kegan Lovelace, Daniel Patrick, Merrick Miranda, Jennifer DeHeck, Chris Bauman, Evan Leonard
Editted by: Alfred Chung (Derivations), Ran Li (Determining Stable Controller Gain), Nirala Singh (Determining Slope of Bode Plot),
Katy Kneiser and Ian Sebastian (synthesis with 2006 wiki "Bode Plots"), Robert Appel, Jessica Rilly and Jordan Talia (Formatting
9.4: PID tuning via Frequency Responses with Bode Plots is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et
9.5: PID TUNING VIA OPTIMIZATION
9.5.1: INTRODUCTION
Tuning a controller is a method used to modify the effect a process change will have on the piece of equipment being controlled. The goal of
tuning a system is to construct the most robust process possible. The method chosen for tuning a system varies depending on the parameter
being measured, the sensitivity of the materials, the scale of the process, and many other variables unique to each process. This chapter
discusses the basics of tuning a controller using predictive methods. To learn more about using an effect based method see the Classical
Tuning section.
9.5.2: OPTIMIZATION
When tuning a PID controller, constants K , T , and T need to be optimized. This can be accomplished using the following equation from
c i d
the Classical Tuning section:

t
1 de(t)
M V = Kc (e(t) + ∫ e(τ)dτ + Td ) (9.5.1)
Ti 0
dt
where
MV is the manipulated variable (i.e. valve position) that the controller will change
Kc accounts for the gain of the system
Ti accounts for integrated error of the system
Td accounts for the derivative error of the system
Tuning by optimization uses computer modeling programs, such as Microsoft Excel, to find the optimal values for the coefficients K , T , c i
and T to yield the minimum error (the Solver function in Excel can be used in this situation). For more information about the use of these
d
parameters and their overall effects on the control system see the P, I, D, PI, PD, and PID control section.
9.5.3: EXCEL MODELING

For instructions on how to install Solver in Excel, see the Adding in the Solver Application in Excel 2007 section. For instruction on how to
use Solver in Excel, see the Excel's Solver Tool section. Equation 9.5.1 can be used in Excel to optimize the values of Kc, Ti, and Td. The
following are the steps to optimize these constants:
Pick a set point. Examples of set points include the temperature at which a reaction is expected to stay and the flow rate of a cooling
stream.
Make a column for error. Create a function in each cell so that the error is equal to the set point minus the actual value.
Make a column that will calculate Equation 1.
At the end of the error column, use the sum function to add up all of the errors in the column.
Make three cells for Kc, Ti, and Td, and insert the initial values.
Open the Solver function in Excel.
In Solver, set the target cell to the summed errors cell. Set the target cell equal to the minimum value possible. Set the changing cells to
Kc, Ti, and Td.
Click 'Solve' to minimize the sum of the error and therefore optimize the Kc, Ti, and Td constants.
Tips for using Excel Solver to optimize PID controller parameters:

When choosing initial values for your parameters, try starting with Kc=1, Ti=1000, and Td=1. Since Solver only does a limited number
of iterations (usually set at 100), it may settle on an incorrect answer.
Change the values for Kc manually and optimize with Solver until you get the lowest value in the target cell. This is the most efficient as
Kc usually has the largest effect on a PID controller.
Make sure to set constraints on your parameters in the 'Subject to the Constraints:' box in Solver (e.g. Kc > 0).
Make sure you incorporate any physical limitations (e.g. maximum heater temperature) into your Excel model. Even though your system
may not be close to the limit initially, Solver may try and go past that limit if it is not programmed into your model.
9.5.4: INDUSTRIAL APPLICATION

The parameters (Kc, Ti, Td) found through optimization will not necessarily give the best control. To optimally tune a controller, a
technician must adjust equipment to tune the process. So why is it necessary to tune by optimization? The parameters found by optimization
give an accurate starting point for the technician.
9.5.5: EXAMPLE OF PID TUNING BY OPTIMIZATION

This example is based on the example in the CSTR Heat Exchange Model section. The goal is to optimize the PID controller for the coolant
temperature. Follow the fully worked out example at: [Media:Optimal PID.xls Optimal PID]
As shown in the Excel file, the coolant temperature changes in response to variations from the set point by a PID type controller. Changing
the Kc, Ti, and Td values using Solver, to minimize the total error, provided the values in the green box. On the second sheet, labeled
'disturbances', it can be seen that once optimized, the parameters should be fit to the system, regardless of changes in operating conditions.
It is important to note that the starting values you choose for the PID parameters will greatly affect your final results from using an excel
solver. So you should always use your intuition to judge whether your starting values are reasonable or not. According to ExperTune ,
starting PID settings for common PID control loops are:
Loop Type: Flow ; P = 50 to 100 ; I = 0.005 to 0.05 ; D = none ;
Loop Type: Liquid Pressure ; P = 50 to 100 ; I = 0.005 to 0.05 ; D = none
Loop Type: Gas Pressure ; P = 1 to 50 ; I = 0.005 to 0.05 ; D = 0.02 to 0.1 ;
Loop Type: Liquid Level ; P = 1 to 50 ; I = 1 to 100 ; D = 0.01 to 0.05 ;
Loop Type: Temperature ; P = 2 to 100 ; I = 0.2 to 50 ; D = 0.1 to 20 ;
The above values are rough, assume proper control loop design, ideal or series algorithm and do not apply to all controllers. These
information should only be used as a possible consideration and should not be taken as an absolute starting value for all PID controls. For
more information about the above values please go to www.expertune.com/tutor.html
As mentioned before, to optimize the parameters in a PID-controlled system the solver function can be used in excel. However the solver
function can sometimes run into its own problems because of the process it uses to solve for these “optimal” values. To ensure that solver
gives you the correct optimal values, some manual “optimization” is needed. Using your excel sheet, the controller on the system can be
separated into its different components P, I, and D. The behavior of each component can be monitored as the parameters are changed
manually. Theoretically you would choose values that have every component reach steady-state over time. The excel solver can then be
used as a precision tool. If your values are already closer to the optimal values, then solver should have no problem making them more
precise. It is recommended that the graphs of the components be checked after using solver with your initial guesses. You may not have to
go through this long and most of the time painful process. This is an example of how the P, I, and D plots change with different parameters.
Note that the combined PID plot will also reach steady-state if each of the components reach steady-state.
(Parameters are not optimized)
(Optimized parameters)
9.5.6: EXAMPLE: OPTIMIZATION OF A HEAT EXCHANGER

This example is based on the ODE Modeling in Excel of the Heat Exchange Model.
Find the best values of Kc, Ti, and Td in the given spreadsheet Media:PID_HeatExchange.xls.
Note the three different tabs:
the first tab shows the problem to be solved
the second is a step by step method for creating this sheet on your own
the third is the answer to the problem
After using Solver:
Click through the cells in one row to determine which parameters are affected by MV changes
Change the values of Kc, Ti, and Td to see how they will effect the overall temperature profile. Which parameter has the greatest effect?
Based on the equation, does this make sense?
Exercise 9.5.1
What do you want to minimize when optimizing the constants Kc, Ti, and Td?
Answer
time
9.5.7: REFERENCES
Svrcek, William Y., Mahoney, Donald P., Young, Brent R. "A Real Time Approach to Process Control", 2nd Edition. John Wiley &
Sons, Ltd.
9.5.8: CONTRIBUTOR
Authors: Andrew MacMillan, David Preston, Jessica Wolfe, Sandy Yu
Stewards: YooNa Choi, Yuan Ma, Larry Mo, Julie Wesely
9.5: PID tuning via Optimization is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source content
9.6: PID DOWNSIDES AND SOLUTIONS
9.6.1: INTRODUCTION
A proportional-integral-derivative (PID) controller is one of the most common algorithms used for control systems. It is widely used
because the algorithm does not involve higher order mathematics, but still contains many variables. The amount of variables that are used
allows the user to easily adjust the system to the desired settings. The algorithm for the PID uses a feedback loop to correct the difference
between some measured value and the setpoint. It does this by calculating and outputting some action that will correct this error in the
system. A PID controller has a proportional, integral and a derivative control which handles the current, past and predicted future of the
signal error. For more information about PID, please refer to PID Intro. The PID controller can operate systems that run in a linear or
nonlinear fashion. Tuning processes are done to the controller to tackle the possible nonlinear system. Limitations arise within the system
because tuning is limited to only three different parameters (proportional, integral, and derivative controls). Additional information on
tuning of PID can be found at [1] or [2]. The most common limitations that occur within the PID control specifically involve the integral
control. The following article addresses some of the common limitations faced by each control type, with an emphasis on the integral
control, and some solutions to overcome each of these limitations.
9.6.2: PROPORTIONAL CONTROL

The main purpose of the proportional control is minimize the fluctuations that occur within the system.
9.6.2.1: LIMITATIONS
The P-controller usually has steady-state errors (the difference in set point and actual outcome) unless the control gain is large. As the
control gain becomes larger, issues arise with the stability of the feedback loop. For instance, reducing the rise time implies a high
proportional gain, and reducing overshoot and oscillations implies a small proportional gain. This is not possible to achieve in all systems.
Below is a general process outcome diagram showing the terminology used above.
Below is a sample diagram of process output of proportional control.
9.6.2.2: SOLUTIONS
The way to eliminate these steady-state errors is by adding an integral action. The integral term in the equation drives the error to zero.
Higher Integral constant (1 / Tt) drives the error to zero sooner but also invites oscillations and instability. Read on the integral control
section below to know more about limitations associated with this integral term.
Below is a sample process output diagram when integral control is added.
The above picture shows the reduction of overshoots and oscillations compared to the picture before adding the integral action.
9.6.3: INTEGRAL CONTROL

The contribution from the integral term is proportional to both the magnitude of the error and the duration of the error. Summing the
instantaneous error over time (integrating the error) gives the accumulated offset that should have been corrected previously. The
accumulated error is then multiplied by the integral gain and added to the controller output. The magnitude of the contribution of the
integral term to the overall control action is determined by the integral gain, K .
i
The integral term is given by:

de
Dout = Kd
dt
where
Dout: is the Derivative output
Kd: is the Derivative Gain, a tuning parameter
e: is the Error = SP − PV
t: is the Time or instantaneous time (the present)
9.6.3.1: LIMITATIONS
Windup
A basic knowledge of the concept of windup is useful before describing a specific type. Windup is defined as the situation when the
feedback controller surpasses the saturation (i.e. maximum) limits of the system actuator and is not capable of instantly responding to the
changes in the control error. The concept of the control variable reaching the actuator’s operation limits is reasonable considering the wide
variety of operating conditions that are possible. When windup occurs the actuator constantly runs at its saturation limit despite any output
the system might have. This means that the system now runs with an open loop instead of a constant feedback loop.
Integrator Windup
The most common type of windup that occurs is integrator windup. This occurs when the input into the system receives a sudden positive
step command and causes a positive error when the system first responds to the actuator. If the rate of integration is larger than the actual
speed of the system the integrator’s output may exceed the saturation limit of the actuator. The actuator will then operate at its limit no
matter what the process outputs. The error will also continue to be integrated and the integrator will grow in size or “wind up”. When the
system output finally reaches the desired value, the sign of the error reverses (e.g. \ yy_{sp} " src="/@api/deki/files/18198/image-
870.png">) and causes the integrator to “wind down” until things return back to normal. Through the wind down the control signal is still
maximum for a long period of time and the response becomes delayed. The integrator takes a long period of time to fully recover to the
operating range of the actuator. Integrator windup may occur from large set point changes, significant disturbances, or equipment
malfunctions.
The illustration is a flowchart showing the specific steps that take place through the integrator controller. It shows the input ( ) and output
of the system ( ) , along with the integrator, the actuator, the system, and sensor involved in the process. The sigma used in each flowchart
is used to represent the summation of all variables inputed to it.
The graphs above are illustrations of integrator windup. The error ( ) is shown in the top graph as , where is the set point
and is the measured signal. The bottom graph displays the control signal ( ). In this specific case of integrator windup, a set point change
occurs and the control signal becomes saturated to its maximum amount . The system's error is still not eliminated though because the
control signal is too small to make the error go to zero. In turn, this causes the integral to increase, which causes the desired control signal to
increase. Therefore, there continues to be a difference between the desired control signal and the true control signal. After a period of time,
the set point is lowered to a level where the controller is able to eliminate the control error. At this point \ yy_{sp} "
src="/@api/deki/files/18198/image-870.png">, which causes a change in the sign of the error. This change in sign causes the control signal
to begin decreasing. The true signal ( ) is stuck at for a while due to desired control signal u being above the limit . The
changes in setpoints throughout this specific example occur because the input is changed in order to get a minimal error in the system.
File:Integrator windup output control integral5.JPG
The images above are another display of integrator windup. The top most shows the error, the middle shows the control signal, and the
bottom shows the integral portion. The integral term begins to decrease, but remains positive, as soon as the error becomes negative at .
The output stays saturated due to the large integral termal that developes. The output signal remains at this level until the error becomes
negative for a sufficiently long time ( ). The control signal then fluctuates several times from its maxmimum to its minimum value and
eventually settles down on a specific value. At every place where the control signal reaches its maximum the integral value has a large
overshoot, and it follows that every place the control signal reaches a minimum the integral has a dampened oscillation. The integral term
accounts for the removal of error in the oscillating system, which in turn causes the system's output to eventually reach a point that is so
close to the setpoint that the actuator is no longer saturated. At this point, each graph begins to behave linearly and settles down. This
example will be used later to show a solution to integral windup. The following section lists several ideas of how to prevent windup.
9.6.3.2: SOLUTIONS
The solutions involved in integral windup are termed as anti-windup schemes. Anti-windup is a method of counteracting the windup that
happens in the integration that occurs in the integral controller of the PID.
9.6.3.2.1: BACK-CALCULATION AND TRACKING
When the output is saturated, the integral term in the controller is recomputed so that it too is at the saturation limit. The integrator changes
over time with the constant . The difference between the output controller ( ) and the actuator output ( ) is found, and this is defined as
an error signal. This error signal is then fed back to the input of the integrator with a gain of . The controller output is proportinal to the
amount of time that the error is present. The error signal is only present when the actuator is saturated, therefore it will not have any effect
on normal operation. With this method, the normal feedback path around process is broken and a new feedback path around the integrator is
formed. The integrator input becomes:
K
frac1Tt es + e
Ti
Knowing that , the following equation can be made:
is the control variable when saturated. The signs for and are always the same; therefore is always bigger than . This in
turn prevents the windup in the integrator. The feedback gain is and is the rate at which the controller output is reset. is called the
tracking time constant and is used to tune the amount of anti-windup. The flowchart shown below shows a PID controller with anti-windup
implemented. The error signal in this process is only non-zero when the actuator is saturated. When this occurs the normal feedback path
around the process is broken because the process input remains constant. A new feedback path is then formed around the integrator and
causes the integrator output to be driven towards a value such that the integrator input becomes zero. Back-calculating never allows the
input to the actuator to reach its actual saturation level because it forecasts what will actually go into the actuator model beforehand.
The following plots describe what happens to the same system (second example) described in the introduction to Integrator Windup section,
only the controller now has anti-windup. The integrator output is at a value that causes the controller to be at the saturation limit, however,
the integral is intially negative when the actuator is saturated ( ), which contrasts with the original behavior of the integrator. This has a
positive effect on the process output, as it converges to the set point much quicker than the normal PI controller. There is only a slight
overshoot in the process output ( ).
9.6.3.2.2: CONDITIONAL INTEGRATION
Conditional integration operates similarly to back-calculating. In this method the integration is turned off when the control is far from
steady state. Integral action is then only used when certain conditions are met, otherwise the integral term is held constant. There are two
possible conditions that may cause the integral to be set as a constant value. First, if the controller output is saturated (integration is shut off
if error is positive but not negative), then the integrator input is set to a constant value. A second approach involves making the integrator
constant when there is a large error. There is a disadvantage to the second approach because the controller may get stuck at a nonzero
control error if the integral term has a large value at the time of switch off. For this reason, the first approach looks at the saturated
controller output, not the saturated actuator output because referring to the actuator output will generate the same disadvantage[1] .
There is very little difference performance wise between integration and tracking, but they move the proportional bands differently. Read on
to learn about proportional bands.
To demonstrate what it means to turn the integral term off, here is the equation (with logic) representing a control with conditional
integration.
1
u(t) = K (e(t) + ∫ e(t)dt) + u (t0 )
Ti
If , then
Else
9.6.3.2.3: SERIES IMPLEMENTATION
To avoid windup in an interacting controller a model of the saturation can be incorporated into the system. The model of saturation will
limit the control signal indirectly. Too hard a limitation will cause an unnecessary limitation of the control action. Too weak a limitation will
cause windup. More flexibility is provided if the saturation is positioned as follows:
With this setup it is possible to force the integral part to assume other preload values during saturation. The saturation function is replaced
by the nonlinearity shown. This anti windup procedure is sometimes called a “batch unit” and is considered a type of conditional
integration. It is mainly used for adjusting the overshoot during startup when there is a large set point change.
9.6.3.3: THE PROPORTIONAL BAND

The proportional band is useful to understand when dealing with windup and anti windup schemes. The proportional band is a range such
that if the instantaneous value of the process output or predicted value on the interval falls within the range, the actuator does not saturate.
The proportional band (defined as the area between ylow,yhigh) is given by the following equations:
Control signal for PID control without derivative gain limitation:
dy
u = K ∗ (b ∗ ysp ) + I − K ∗ Td ∗
dt
The predicted process output is given by

dy
yp = y + Td ∗
dt
Combining the above equations for a maximum and minumum control signal ( and )which corresponds to the points for which
the actuator saturates,
I − umax
ylow = b ∗ ysp +
K
I − umin
yhigh = b ∗ ysp +
K
If the predicted output is in the proportional band, the controller operates linearly. The control signal saturates when the predicted output is
outside of the proportional band. The following plot illustrates how the proportional band is useful in understanding windup for the same
example. At time equals zero, the proportional band is increased, indicating that the integral part is increasing. Immediately after , the
output is greater that the setpoint, which causes the integral part to start to decrease. As can be seen, the output value, does not reach the
proportional band until it is greater than the setpoint, . When the output finally reaches the proportional band at , it goes through the
band very quickly, due to the band changing at a high rate. The control signal decreases immediately and saturates in the opposite
direction, causing the output to decrease and approach the setpoint. This process repeats as the output value converges with the setpoint.
The tracking time constant has a major influence on the proportional band. The proportional band is moved closer to the output because of
tracking. The tracking time constant determines the rate that the proportional band moves at.
9.6.3.4: REAL LIFE SCENARIO

Imagine a situation where a water storage tank that is needed to store water and supply for use in a plant and it is necessary to keep the
water level constant. A level sensor is installed to keep track of the water level in the tank, and it sends this data to the controller. The
controller has a set point for instance 75% full. The rate of water entering the tank is increased by opening valve and decreased by closing
the valve. In the case of integral windup, the valve is all the way closed or open and the actuator sends the signal to close it more or open it
more which is not possible.
9.6.4: DERIVATIVE CONTROL

Derivative action can be thought of as making smaller and smaller changes as one gets close to the right value, and then stopping in the
correct region, rather than making further changes. Derivative control quantifies the need to apply more change by linking the amount of
change applied to the rate of change needed. For example, an accelerator would be applied more as the speed of the car continues to drop.
However, the actual speed drop is independent of this process. On its own, derivative control is not sufficient to restore the speed to a
specific value. Pairing the match in change with a proportionality constant is enough to properly control the speed.
The derivative control is usually used in conjunction with P and/or I controls because it generally is not effective by itself. The derivative
control alone does not know where the setpoint is located and is only used to increase precision within the system. This is only control type
that is open loop control (also known as feedforward loop). The derivative control operates in order to determine what will happen to the
process in the future by examining the rate of change of error within the system. When the derivative control is implemented the following
general equation is used:
1 d
u(t) = e(t) + u (t0 )
Td dt
where
= control signal
= derivative time
= error
= initial control signal
This equation shows the derivative control is proportional to the change in error within the system.
9.6.4.1: LIMITATIONS OF DERIVATIVE CONTROL

The major problem associated with this control is the noise problem. When the frequency within the system is high (change in error of the
system is large), taking the derivative of this signal may amplify the signal greatly. Therefore, small amounts of noise within the system will
then cause the output of system to change by a great amount. In these circumstances, it is often sensible use a PI controller or set the
derivative action of a PID controller to zero.
The following is a graphical represenation of some sinusoidal output signal function of a system.
9.6.4.2: SOLUTIONS
To eliminate/minimize this problem, an electronic signal filter can be included in the loop. Electronic signal filters are electronic circuits
which perform signal processing functions, specifically intended to remove unwanted signal components and/or enhance wanted ones.
Electronic filters can be: passive or active, analog or digital, discrete-time (sampled) or continuous-time, linear or non-linear, etc. The most
common types of electronic filters are linear filters, regardless of other aspects of their design.
For more information on electronic signal filters, reference the following website:
en.Wikipedia.org/wiki/Electronic_filter
The derivative term in the equation is usually changed by putting a first order filter on the term so that the derivative does not amplify the
high frequency noise that is attenuated. Below is a sample outcome figure of a possible derivative of the output signal shown above along
with the filtered signal.
As shown, it is possible for the amplitude to be magnified when the derivative is taken for a sinusoidal function. A filter is usually a set of
complicated equations that are added to the derivative that effect the function as shown.
9.6.5: ROBUSTNESS OF PID CONTROLLERS

When designing a PID controller it is important to not only create a controller that will correct the error in an efficient manor, but also a
robust controller. No matter how good the controller is if it cannot handle the magnitude of certain disturbances then it will not be able to
perform well and complete the job. The robustness of a controller is the ability of the controller to withstand disturbances and still operate
effectively. In order to calculate controller robustness, a weighted sum of the integral of the squared controller effort (ISC) and the integral
of the squared controller error (ISE) is calculated. The smaller the weighted sum the more robust the controller.
ISE = 0 ∫ ∞ e(t)2 dt
ISC = 0 ∫ ∞ [u(∞)-u(t)]2 dt
IT = w ISC + ISE
In these equations u is the input variable, w is the weight factor which is based on downstream processes and IT is the total error and
controller effort. The weight factor is used to assign a higher significance to the ISC or the ISE. The significance depends on the design and
whether it is more important to have minimal error and tight control or less controller effort with more error. Low values of w are used when
tight control is desired. Robustness calculations are determined experimentally. The robustness depends a lot on the type of control used,
whether it is feedback, feed- forward, cascade, or a combination.
9.6.5.1: NOTES ON ROBUSTNESS
It is important to realize that when calculating the robustness of a feed forward system, the system will reach a steady state, but it is unlikely
that the system will reach the original set point. As a result, the feed forward system will most likely be less robust than a feedback system.
The reason the feed-forward system doesn't return to the set point but finds a new steady state is because it uses open loop control and there
are no sensors to inform it that its adjustment to the disturbance did not work correctly. For this reason, feed-forward controllers are
sometimes coupled with a feedback system. When combined the feed-forward controller can now see if its corrections achieved the desired
result or not, resulting in a more robust controller.
9.6.6: EXAMPLE ROBUSTNESS CALCULATION
In this system, the controller is regulating the temperature of the feed to a certain downstream process. After collecting experimental data
from your controller, the following graph was made.
Using this graph the squared controller effort and the squared error can also be graphed and the ISE and ISC can be calculated. The graph of
the corresponding effort and error can be seen below.
Using the squared controller effort and the squared error graph, the robustness can be calculated. To find the integral of the squared error
and the squared controller effort the trapezoid rule can be used.
IT = wIS C + IS E
IS C = 21IS E = 94
Assuming a weight factor of 1, the robustness of the controller is 115.
 EXAMPLE 9.6.1
Consider the below output from using one of the controllers. The general equation of the output signal, U = K(e). What problems do
you see in the below picture and how can you correct them?
Solution
The response time is very high, and it has a high overshoot and oscillation problems as well as steady-state errors. The equation
confirms that it is a P-controller and
 EXAMPLE 9.6.2
A heat exchanger uses a PID control system to control the temperature of the outlet of tube side of the heat exchanger. It’s required to
keep water at 20 C for the cooling of an effluent from an exothermic reaction. The specific controller has a min output of 1 volt and a
max output of 6 volts. The experimental data shows a large lag time in the controller’s ability to cool the incoming fluid (the lower
portion of the sinusoidal curve). The following data shows the correlation between the controller output and the calculated signal.
Describe the type of problem and a possible solution to the problem.
Solution
The Problem shown is an example of integrator-windup. The controllers calculated signal is lower than the min output that it can
handle. This explains the lag time seen for the lower portion of the sinusoidal curve. To fix this problem the integral control can be
recomputed using back track calculation.
 EXAMPLE 9.6.3
(a) You are using a PID controller to maintain the feed to a reactor at a constant temperature. It is imperative that the feed remain within
1 °C of the set-point. You’ve tuned your PID control system according to the Ziegler-Nichols open loop system and obtained the
following values for Kc, τI, and τD: Kc = 0.218, 1/τI = 0.0536, τD = 4.67
You want to know if these are good control values or if they can be improved. Your control variable output looks like this:
Additionally, you have available the following plot of the voltage output from your control parameters:
Is this system in control? If so, explain why. If not, explain why and what (qualitative) improvements to the PID gains will bring the
system into better control.
Solution
The system is in control in that it appears to roughly stay within the defined acceptable boundaries of ±1°C. However, it does display
oscillation, indicating that the system is working a little harder than maybe it has to. Looking at the controller voltage output, we see
that the derivative gain is giving high oscillations. Thus maybe decreasing this gain would keep the system in control with a little less
wear and tear on the control valves.
(b) At another flow rate, you observe the following behavior in your controller:
Is the system under control in Trial 1? In Trial 2? Why or why not? What physically happened between trials to give the observed
change in output?
Solution: In trial 1, without even looking at the direct system output, we can guess that it is probably not in control. The controller
output voltage indicates that the integral gain is way too high, and its wild oscillations will doubtlessly be causing oscillations in the
system. Physically, this means that the system is opening and closing the valve constantly, creating a lot of wear and tear on the
machinery, and probably overcompensating for small changes and further destabilizing the system. In trial 2, we can see that the
integral voltage oscillations have damped, indicating that the gain has been reduced. As these oscillations level out, the oscillations in
the other control gains also damp down, as the system comes into better control and none of the terms are working as hard as they had
been before.
 EXAMPLE 9.6.4
Note this is an open-ended problem meant to give readers an intuition about PID controllers
Imagine the classic differential equation problem of two tanks in a series governed by the differential equations:
dh1 −−
= P ID − K1 √h1
dt
dh2 −− −−
= K1 √h1 − K2 √h2
dt
\[\P I D=1+K_{p} e(t)+\frac{1}{t_{i}} \int_{0}^{t} e(\tau) d \tau+t_{d} \frac{d e(t)}{d t}]

where
h1 = height of tank 1
h2 = height of tank 2
Ki = valve constant associated to tank 1 or 2
The valve constants K1 = 5 and K2 = 5 and the set point is 3. What values of Kp, ti, and td will cause the two tanks to reach steady state
the fastest? Also what values will give the most oscillations?
Solution
Click on this link for the worked out Mathematica solution
Other questions that can be explored with slight modifications of the code are:
1. Will the values of Kp, ti, and td yield similar results if the PID controller was placed on tank 2? What new values of Kp, ti, and td will
cause the two tanks to reach steady state the fastest?
2. For the values of Kp, ti, and td that causes the two tanks o reach steady state the fastest, what happens if you increase K1? What will
happen if you increase K2?
 EXAMPLE 9.6.5
Label the following general response charts as overdamped, critically damped, underdamped, undamped, and unstable.
(a)
(b)
(c)
(d)
(e)
Answers
a)Underdamped b)Undamped c)Critically Damped d)Unstable e)Overdamped
Exercise 9.6.1
What is a back-calculation used for?
a. To reduce noise problems
b. To reduce steady-state errors
c. To reduce errors in integral term
d. To reduce errors in proportional term
Exercise 9.6.1
What is a proportional band?
a. A way to eliminate errors in integral term
b. A way to eliminate errors in proportional term
c. A way to eliminate errors in derivative term
d. A way to understand anti-windup problem
9.6.7: REFERENCES
Bhattacharyya, Shankar P., Datta & Silva, PID Controllers for Time-Delay Systems, 2005 [1]
Astrom, Karl J., Hagglund, Tore., Advanced PID Control, ISA, The Instrumentation, Systems and Automation Society.
U of Texas chE
Wikipedia Article, PID Controller, en.Wikipedia.org/wiki/PID_control
9.6.8: CONTRIBUTORS
Authors: Ashwini Miryala, Kyle Scarlett, Zachary Zell, Brandon Kountz
Stewards: Brian Hickner, Lennard Gan, Addison Heather, Monique Hutcherson
9.6: PID Downsides and Solutions is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source content
CHAPTER OVERVIEW
10: DYNAMICAL SYSTEMS ANALYSIS

10.2: Linearizing ODEs
10.4: Using Eigenvalues and Eigenvectors to Find Stability and Solve ODEs
10.5: Phase Plane Analysis - Attractors, Spirals, and Limit cycles
10.6: Root Locus Plots - Effect of Tuning
10.7: Routh Stability - Ranges of Parameter Values that are Stable
10: Dynamical Systems Analysis is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source content
1
10.1: FINDING FIXED POINTS IN ODES AND BOOLEAN MODELS
Authors: Nicole Blan, Jessica Nunn, Pamela Anne Roxas, Cynthia Sequerah
Stewards: Matthew Kerry Braxton-Andrew, Josh Katzenstein, Soo Kim, Karen Staubach
Engineers can gain a better understanding of real world scenarios by using various modeling techniques to explain a system's behavior. Two
of these techniques are ODE modeling and Boolean modeling. An important feature of an accurate ODE model is its fixed point solutions.
A fixed point indicates where a steady state condition or equilibrum is reached. After locating these fixed points in a system, the stability of
each fixed point can be determined (see subsequent Wikis). This stability information enables engineers to ascertain how the system is
functioning and its responses to future conditions. It also gives information on how the process should be controlled and helps them to
choose the type of control that will work best in achieving this.
10.1.2: CONCEPT BEHIND FINDING FIXED POINT

A fixed point is a special system condition where the measured variables or outputs do not change with time. In chemical engineering, we
call this a steady state. Fixed points can be either stable or unstable. If disturbances are introduced to a system at steady state, two different
results may occur:
1. the system goes back to those original conditions (stable point)
2. the system deviates from those conditions rapidly (unstable point)
Subsequent wiki articles will discuss these different types of fixed points in more detail. The focus of this article will be simply finding
fixed points, not classifying them. We will discuss several methods of finding fixed points, depending on the type of model employed.
10.1.2.1: ODE MODEL

When a process or system is modeled by an ODE or a set of ODEs, the fixed points can be found using various mathematical techniques,
from basic hand calcuations to advanced mathematical computer programs. Independent of the method used, the basic principle remains the
same: The ODE or set of ODEs are set to zero and the independent variables are solved for. At the points where the differential
equations equal zero there is no change occurring. Thus, the solutions found by setting the ODEs equal to zero represent the numerical
values of independent variables (i.e. temperature, pressure, concentration) at steady state conditions. If a single ODE or set of ODEs
becomes too complicated to be solved by hand, a mathematical program such as Mathematica can be used to find fixed points. The latter
part of this article focuses on how to use Mathematica to find fixed points of complicated systems of ODEs.
Note that in some cases there may not be an analytical method to find a fixed point. This case commonly occurs when the solution to a fixed
point involves a high degree polynomial or another mathematical function that does not have an analytical inverse. In these cases, we can
still find fixed points numerically if we have the parameters.
10.1.2.2: BOOLEAN MODEL
A Boolean Model, as explained in “Boolean Models,” consists of a series of variables with two states: True (1) or False (0). A fixed point in
a Boolean model is a condition or set of conditions to which the modeled system converges. This is more clearly seen by drawing state
transition diagrams.
State Transition Diagram from Boolean Models

From the state transition diagram above, we can see that there are two fixed points in this system: 0,1,1 and 1,1,1. Starting in any state on
the diagram and following the arrows, one of these two states will be reached eventually, indicating that the system tends to achieve either
of these sets of operating conditions. If slight disturbances are introduced to the system while it is operating at one of these sets of
conditions, it will return to 0,1,1 or 1,1,1. Also noted in the state transition diagram are state cycles. The difference between a state cycle
and a fixed point is that a state cycle refers to the entire set of Boolean functions and transition points leading to the steady-state conditions,
whereas a fixed point merely refers to the one point in a state cycle where steady-state conditions are reached (such points are indicated by a
yellow circle in the diagram).
10.1.3: FINDING FIXED POINTS: FOUR POSSIBLE CASES

There are four possible scenarios when finding the fixed points of an ODE or system of ODEs:
1. One fixed point
2. Multiple fixed points
3. Infinite fixed points
4. No fixed points
10.1.3.1: ONE FIXED POINT

The first type of ODE has only one fixed point. An example of such an ODE is found in the Modeling of a Distillation Column. An ODE is
used to model the energy balance in the nth stage of the distillation column:
dTn 1 qr
= [Ln−1 xn−1 − W xW ] [Tn−1 − Tn ] +
dt MW MW cp
Which can also be written as:

dTn 1 qr 1
− [Ln−1 xn−1 − W xW ] [Tn−2 ] + + [Ln−1 xn−1 − W xW ] [− Tn ]
dt MW MW cp MW
If initial conditions i.e. Tn − 1,Ln − 1,xn − 1 are known, the equation above reduces to:
dTn
= a + bTn
dt
where a and b are constants since all the variables are now known.
1 qr
a = [Ln−1 xn−1 − W xW ] [Tn−1 ] +
MW MW cp
−1
b = [Ln−1 xn−1 − W xW ]
MW
By analyzing the equation

dTn
= 0 = a + bTn
dt
we can immediately deduce that at steady state T = −a/b . Clearly, there is only one fixed point in this system, only one temperature of
n
the distillation column which will be at steady-state conditions. We can use Mathematica to solve for the fixed point of this system and
check our results. In Mathematica, the Solve[] function can be used to solve complicated equations and systems of complicated equations.
There are some simple formatting rules that should be followed while using Mathematica:
1. Type your equation and let the differential be called an arbitrary variable (e.g. T[t])
2. Type Solve[T[t]==0,T] and hit Shift+Enter
3. This produces an output contained inside curly brackets
Please read the Solving ODEs with Mathematica section for more information on syntax and functions.
A sample of how the format in Mathematica looks like is shown below:
Maple can be used to visualize a single fixed point. Wherever the plot intersects the x-axis represents a fixed point, because the ODE is
equal to zero at that point.
The following Maple syntax was used to plot the ODE: plot(0.5+4t, t=-2..2,T=0..5,color=black);
The constant a = 0.5 and the constant b = 4 in the above example.
The resulting graph is below, the red point indicates at what T a fixed point occurs:
Solving a single fixed point for an ODE and a controller in Mathematica

1. Identify what type of controller it is (P, I , PI, or PID etc.)
2. Identify your ODE equations (Is the controller a function of the ODE?)
Example: Solve for the fixed points given the three differential equations and the two controllers (u1 and u2).
\left.\frac{d H}{d t}=(1 / A) (F(i n)-F_{out}\right)\right) \nonumber
\left.\frac{\left.d F_{in} \right)}{d t}=K(v 1)\left(u_{1}\right) \nonumber
dFout )
= K(v2 ) (u2 ) H
dt
Where H is the level in the tank, F is the flow in, F the flow out, and u and u are the signals to the valves v and v . K v1 and K v2
in out 1 2 1 2
are valve gains (assumed to be linear in this case, although this does not have to be). Note that the exit flow also depends on the depth of
fluid in the tank.
You next parameterize your model from experimental data to find values for the constants:
A=2.5 meters squared
K_(v1)=0.046 meters cubed/(minute mA)
K_(v2)=0.017 meters squared/(minute mA)
Next you want to add:
A full PID controller to regulate Fout via FC1 connected to v2.
A P-only controller to regulate H via LC1 connected to v1.
For this system you want to maintain the tank level at 3 meters and the exit flow (Fset) at 0.4 m3 /minute. The following Mathematica code
should look as follows:
10.1.3.2: MULTIPLE FIXED POINTS
Multiple fixed points for an ODE or system of ODEs indicate that several steady states exist for a process, which is a fairly common
situation in reactor kinetics and other applications. When multiple fixed points exist, the optimal steady-state conditions are chosen based on
the fixed point's stability and the desired operating conditions of the system.
The following is an example of a system of ODEs with multiple fixed points:
dCA
2
= 14CA − 2C − CA CB
A
dt
dCB
2
= 16CB − 2C − CA CB
B
dt
The above system of ODEs can be entered into Mathematica with the following syntax:
This system in particular has four fixed points. Maple can be used to visualize the fixed points by using the following syntax:
with(plots):
fieldplot([14*x-2*x^2-x*y,16*y-2*y^2-x*y],x=0..10,y=0..10,fieldstrength=log);
The first line initializes the plotting package within Maple that allows for plotting vector fields. The second line uses the command
“fieldplot” and inputs the two ODEs that make up the system. The scales of the x and y-axis are set to range from 0 to 10. The fieldstrength
command is mainly used for visual purposes, so that the direction of the arrows becomes more apparent. Below is the resulting plot:
The red dots indicate the fixed points of the system. On the plot, these points are where all the surrounding arrows converge or diverge.
Converging arrows indicate a stable fixed point, in this example the point at (4,6) is a stable fixed point. Diverging arrows indicate an
unstable fixed point, in this example (0,0), (0,8) and (7,0) are unstable fixed points.
10.1.3.3: INFINITE FIXED POINTS

An example of an ODE with infinite fixed points is an oscillating ODE such as:
where a is a constant.
Using Mathematica to solve for the fixed points by setting
If you click the "More" link on Mathematica it will basically state that there are other solutions possible according to the Help section
shown below:
The Maple syntax used to graph the solved differential equation is:
plot(cos(3t),t=0..10,T=-1..1,color=black);
The constant a = 3 in this case.
The infinite fixed points can be seen in the graph below, where anytime the function crosses the x axis, we have a fixed point:
10.1.3.4: NO FIXED POINTS

The fourth type of ODE does not contain fixed points. This occurs when a certain variable (such as temperature or pressure) has no effect on
a system regardless of how it changes. Generally, systems with this sort of behavior should be avoided because they are difficult to control
as they are always changing.
This can be modeled by vertical or horizontal lines due to the fact that no fixed points are found by setting the line equal to zero. An ODE is
used to model a line held constant at a:
Where, a can be any constant except 0.
Intuitively, trying to find a fixed point in this system is not possible, because a constant such as 3 can never equal zero. Solving this ODE is
not possible even by analyzing the system. Therefore, when inputting this into Mathematica, it yields {}. The notation {} means that there
are no fixed points within the system. The image below is how Mathematica solves the ODE.
By using Maple (version 10), one can visually see a lack of fixed points by using the following syntax:
plot(3, t = 0..10, T = 0..10, color = black);
The constant a = 3 in the above case.
This image shows that the line is horizontal and never crosses the x axis, indicating a lack of fixed points.
10.1.4: SUMMARY
A fixed point is a system condition where the measured variables or outputs do not change with time. These points can be stable or unstable;
refer to Using Eigenvalues to evaluate stability for an introduction to a common method for determining stability of fixed points.
There are four possible cases when determining fixed points for a system described by ODEs:
One fixed point
Multiple fixed points
Infitite fixed points
No fixed points
There are methods described above for using Mathematica or Maple to solve for the fixed points in each case. Fixed points can also be
determined for a Boolean model.
Knowing the fixed points of a system is very important when designing a control architecture for the system. These are the operating
conditions that the system will exhibit at steady-state. Controllers can have influence on the fixed points, so a thorough analysis of fixed
points using equations describing the system and the controllers should be conducted before implementation of the control scheme.
10.1.5: WORKED OUT EXAMPLE 1: MANIPULATING A SYSTEM OF EQUATIONS

Recall the example system of ODEs used in the Multiple Fixed Points Section:
Find how the fixed points change when m = 2 and n = 3.

Solution
10.1.6: WORKED OUT EXAMPLE 2: SYSTEM OF ODES
You are asked to estimate the fixed point(s)of a heat exchanger in your plant. The fixed points corresponds to the temperatures of your tube-
side (hot) and shell-side (cold) fluids respectively, Tt,out and Ts,out. Neglect heat loss to the surrounding area i.e. .
Given the two main ODEs used to model a heat exchanger, use Mathematica to solve for the fixed points of the system in terms of the
known variables.
(equation 1)
(equation 2)
The values for m, cp, ρ, Ft,in, Ft,out, k, A, δz, and Tt,in, Fs,in, Fs,out, Ts,in are given and fixed.
Please refer to the Wiki article on HeatExchangeModel for detailed explanation on the meaning of the variables and the derivation of the
ODEs above.
Hint: Lump up all known variables under one general variable
Solution
Simplify the 2 equations to the ODEs below:
(equation 1a)
(equation 2a)
where , , and , ,
Type these equation into Mathematica using the appropriate syntax and use the Solve[] function to find the fixed points.
We have found our fixed point. Just plug in the variables as defined earlier for a, b, c, d, e, f and you will obtain the temperatures in terms of
the useful parameters.
10.1.7: MULTIPLE CHOICE QUESTION 1

The solutions found by setting the ODE equal to zero represent:
a) independent variables not at steady state conditions
b) dependent variables not at steady state conditions
c) independent variables at steady state conditions
d) dependent variables at steady state conditions
Answer: C

How many fixed points are there when the following equation is solved by Mathematica?
a) none
b) 1
c) 2
d) 3
Answer: D

Stability of Fixed Points
10.1.10: REFERENCES
Edwards H., Penney D.(2003), Differential Equations: Computing and Modeling, Third Edition. Prentice-Hall. ISBN 0130673374
Strogatz, Steven H.(2001), Nonlinear Dynamic and Chaos: With Applications to Physics, Biology, Chemistry, and Engineering, 1st
Edition. Addison-Wesley. ISBN 0738204536
10.1: Finding fixed points in ODEs and Boolean models is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et
10.2: LINEARIZING ODES
Chemical engineering processes often operate in nonlinear and unsteady manners (i.e. not always at steady state), and are generally
governed by nonlinear ordinary differential equations (ODEs). The ODE is a relation that contains functions of only one independent
variable and derivatives with respect to that variable. Many studies have been devoted to developing solutions to these equations, and in
cases where the ODE is linear it can be solved easily using an analytical method. However, if the ODE is nonlinear and not all of the
operating parameters are available, it is frequently difficult or impossible to solve equations directly. Even when all the parameters are
known, powerful computational and mathematical tools are needed to completely solve the ODEs in order to model the process. In order to
simplify this modeling procedure and obtain approximate functions to describe the process, engineers often linearize the ODEs and employ
matrix math to solve the linearized equations.
A linear equation is an equation in which each term is either a constant or the product of a constant times the first power of a variable. These
equations are called "linear" because they represent straight lines in Cartesian coordinates. A common form of a linear equation in the two
variables x and y is y = mx + b . This is opposed to a nonlinear equation, such as
x 2
m = e +x + 2x + 5.
Even though 2x + 5 is a linear portion of the equation, e

x
and x
2
are not. Any nonlinear terms in an equation makes the whole system
nonlinear.
Non-linear system of equations:
dA 2 3
= 3A + 2B + C − 7D
dt
dB
2
= A+ C + 2D
dt
dC 2 2
= A + 4B −C
dt
dD
= 2C − D
dt
After linearization (around the steady state point {-0.47,-0.35,0.11,0.23}:

′
A −2.83 2 1 0 A k1
⎛ ⎞ ⎛ ⎞⎛ ⎞ ⎛ ⎞
′
⎜ B ⎟ ⎜ 1 0 0.23 0⎟⎜B ⎟ ⎜ k2 ⎟
⎜ ⎟ = ⎜ ⎟⎜ ⎟+⎜ ⎟
⎜ C′ ⎟ ⎜ 1 −2.78 −0.23 0 ⎟ ⎜ C ⎟ ⎜k ⎟
3
⎝ ′ ⎠ ⎝ ⎠⎝ ⎠ ⎝ ⎠
D 0 0 2 0 D k4
Note that each equation is comprised solely of first order variables.

Even though it is unlikely that the chemical engineering process to be modeled operates in a linear manner, all systems can be approximated
as linear at a point. This is preferred as linear systems are much easier to work with than nonlinear equations. Although linearization is not
an exact solution to ODEs, it does allow engineers to observe the behavior of a process. For example, linearized ODEs are often used to
indicate exactly how far from steady state a given process deviates over specified operating ranges. This wiki page discusses how to solve a
linearized ODE by hand and by using Mathematica, and proceeds to work out several examples of linearized ODEs commonly seen in
chemical engineering practice.
10.2.2: APPLICATIONS TO CHEMICAL ENGINEERING

As mentioned above, linearizing ODEs allows engineers to understand the behavior of their system at a given point. This is very important
because many ODEs are impossible to solve analytically. It will also lead to determining the local stability of that point. Most of the time a
system will be linearized around steady state, but this is not always the case. You may be interested in understanding the behavior of your
system at its operating point or equilibrium state (not necessarily steady state). The linearization approach can be used for any type of
nonlinear system; however, as a chemical engineer, linearizing will usually involve ODEs. Chemical engineers use ODEs in applications
such as CSTRs, heat exchangers, or biological cell growth.
It is also important to understand the advantages and disadvantages of linearizing a system of ODEs:
10.2.2.1: ADVANTAGES
Provides a simpler, more convenient way to solve the ODEs
The behavior of a process can be observed
Any type or order of ODE can be used
10.2.2.2: DISADVANTAGES
The solution is only an exact solution at the chosen point; otherwise it is an approximation and becomes less accurate away from the
point
Although linearizing is a quicker alternative, it takes time to initially learn the process (ex: using Mathematica)
Another use for linearization of the equations that govern chemical processes is to determine the stability and characteristics of the steady
states. Systems of linearized ODEs can be used to do this, and the methods of doing so can be found in Fixed Points, Eigenvectors and
Eigenvalues, and Using eigenvalues and eigenvectors to find stability and solve ODEs
10.2.3: GENERAL PROCEDURE FOR LINEARIZATION

Linearization is the process in which a nonlinear system is converted into a simpler linear system. This is performed due to the fact that
linear systems are typically easier to work with than nonlinear systems. For this course, the linearization process can be performed using
Mathematica. The specific instructions on how to do this can be found below.
1. Choose a relevant point for linear approximation, two options available are:
Steady state- points where system does not change
Current location- given where you are now
2. Calculate the Jacobian matrix at that point. The Jacobian is essentially a Taylor series expansion.
3. Solve to find unknown constants using algebraic methods.
10.2.4: LINEARIZATION BY HAND

In order to linearize an ordinary differential equation (ODE), the following procedure can be employed. A simple differential equation is
used to demonstrate how to implement this procedure, but it should be noted that any type or order of ODE can be linearized using this
procedure.
1. Use a Taylor series expansion (truncating after the linear terms) to approximate the right-hand side of the ODE.
Let’s say we start with the following ODE:
dx 2
= f(x) = 3x .
dt
This ODE describes the behavior of some variable, x, with respect to time.
A Taylor series is a series expansion of a function about a point. If x = a , an expansion of a real function is given by:
∞ n
f (a)
n
f(x) = ∑ (x − a )
n!
n=0
2 3 n
f (a) f (a) f (a)
′ 2 3 n
f(x) = f(a) + f (a)(x − a) + (x − a ) + (x − a ) −…+ (x − a ) +…
2! 3! n!
When x = 0, the function is also known as Maclaurin series. Taylor’s theorem states that any function satisfying certain conditions can be
expressed as a Taylor series.
For simplicity’s sake, only the first two terms (the zero- and first-order) terms of this series are used in Taylor approximations for linearizing
ODEs. Additionally, this truncation (ie. "chopping" off the n = 2 and higher terms from the polynomial shown above) assures that the
Taylor Series is a linear polynomial. If more terms are used, the polynomial would have (x − a) and higher order terms and become a
2
nonlinear equation. The variable ‘a ’ in the Taylor series is the point chosen to linearize the function around. Because it is desired that most
processes run at steady state, this point will be the steady state point. So, our differential equation can be approximated as:
dx ′
= f(x) ≈ f(a) + f (a)(x − a) = f(a) + 6a(x − a)
dt
Since a is our steady state point, f(a) should always be equal to zero, and this simplifies our expression further down to:
dx ′
= f(x) ≈ f (a)(x − a) = 6a(x − a)
dt
The graph shown above shows the approximation of f(x) at (x,f(x)). As mentioned previously, linearization is only an approximation for any
given function near a continuous point. When working with a system of ODEs, the Jacobian is written as a matrix. It is the matrix of
constants needed to describe a system's linearity. The Jacobian may be thought of as how much a system is distorted to take on a linear
identity. A Jacobian matrix will always be a square(#rows = #columns) and it shows how each equation varies with each variable. The
Jacobian matrix is defined as:
∂y ∂y
1 1
⎡ ⋯ ⎤
∂ x1 ∂ xn
⎢ ⎥
J (x1 , … , xn ) = ⎢ ⎥
⎢ ⋮ ⋱ ⋮ ⎥
⎢ ⎥
∂ yn ∂ yn
⎣ ⋯ ⎦
∂ x1 ∂ xn
And is used as such for this non-linear set of equations

dx 2
= 2x + cos(5y) − 12 arctan(z − 3)
dt
dy 3
= 7y + sin(5x)
dt
dz
= 1.3 log(x + y)
dt
The linearized system is

′
x J1 J2 J3 x c1
⎡ ⎤ ⎡ ⎤⎡ ⎤ ⎡ ⎤
′
⎢y ⎥ = ⎢ J4 J5 J6 ⎥ ⎢ y ⎥ + ⎢ c2 ⎥
⎣ ′ ⎦ ⎣ ⎦⎣ ⎦ ⎣ ⎦
z J7 J8 J9 z c3
The square matrix is the Jacobian.
 EXAMPLE 10.2.1
Lets say you have the following set of equations and you want to find its Jacobian matrix around the point \(A=3,\,B=2).
dA 2
= 3A − A − AB
dt
dB 2
= 6B − AB − 2B
dt
We find the Jacobian by taking the derivative of each equation with respect to each variable.
2
d (3A − A − AB)
= 3 − 2A − B
dA
2
d (3A − A − AB)
= −A
dB
2
d (6B − AB − 2B )
= −B
dA
2
d (6B − AB − 2B )
= 6 − A − 4B
dB
These are the equations in the matrix. The values of the variables from whatever point we are linearizing are then put into these
equations and calculated out to get the Jacobian.
3 − 2A − B −A
J ac = [ ] A = 4, B = 2
−B 6 − A − 4B
−7 −4
J ac = [ ]
−2 −6
2. Change the approximation by linearizing around a steady state point in order to describe how the process deviates from steady
state.
The following substitution can be made:
dx d(x − a)
=
dt dt
d(x − a)
≈ 6a(x − a)
dt
This substitution is allowed because ‘a’ is a constant, and the derivative of a constant is zero.
Substituting (x-a) for x signifies that our differential equation now shows how our function, x, deviates away from the steady state value, a,
with respect to time. This deviation, (x-a), is commonly expressed as x′. It should also be noted that the quantity ‘6a’ is a constant, and thus
will be further recognized as ‘A’.
Our final linearized equation becomes:
′
dx
′
≈ Ax
dt
The once nonlinear ODE,

dx
2
= f(x) = 3x
dt
has now been simplified into a linear differential equation.

The procedure of linearization typically occurs around the steady state point or points of a specified process. Engineers anticipate a certain
change in output for the particular steady state point, and may proceed to linearize around it to complete their approximation. Please follow
the graphic below for further detail. Note that a steady state point occurs when dx/dt = 0
\rrorP lotColorN ew1.jpg
As can be seen, moving farther away from the steady state point results in significantly larger deviation, and thus error (actual-linearization).
Occasionally, for very unique operating conditions, plant management may decide to momentarily run a process outside of strict steady state
conditions (perhaps a unique start-up procedure, shut-down recovery, etc.). To reinforce the concept of linearization around an unsteady
state point (arbitrarily chosen in this example), please consider the following visual representation. Note that a=50 and C=7500 in the first
plot.
\ewImage2.jpg
The second plot is a magnification of a small section of the first plot. As one can readily notice, both the linear approximation and the actual
graph overlap almost exactly over this small range. This illustrates how this particular linearization could be used to approximate this region
of the function and can describe its behavior quite accurately. This approximating technique using the linearization of ODE can be
performed around different points (different values of a) using the same method in order to model the behavior of the rest of the dx/dt vs x
function. Note: frequent recalculation of the integration constant, C, permits for increased accuracy in approximation.
10.2.5: EXAMPLE OF A SIMPLE LINEARIZATION PROCESS IN USE

Provided a circumstance in which developing a rigorous analytical solution is unfeasible due to time, access to computing resources, or
mathematical ability, the linearization process offers a convenient and swift alternative.
An appropriate introduction to linearization application is featured in section 5.3 of the Bequette reading and has been described below.
The following model describes the behavior of tank height as a function of time (t), faced with a steady- state incoming liquid flowrate.
dh/dt represents the change in height of the fluid contained within the tank as a function of time, h indicates the height (or level) or the fluid
contained within the tank, F represents the magnitude of incoming flow. The following model describes the tank level/height fluctuation as
function of both the current height (h) and flow (F). Fluid enters the tank as flow per area of the inlet(F/A), and leaves as a function of a
corrective proportional term, β, also per area of the outlet, multiplied by the square root of the current height.
Instead of fluid leaving through this term, we may also consider it a form of feedback control, which subsequently reduces the amount of
inlet fluid instead ( depends on the application, the final result is identical). Therefore:
dh F β −
= f(h, F ) = − ∗ √h
dt A A
Important note about the relevance of the provided variables:

Fs indicates the initial flow rate for the described system and is important in calculating the intercept and subsequently the calculative
corrective term for the linearization (if the linearization is not performed at steady-state, as in this example). hs describes the tank height for
the specified flow rate, Fs. Upon plugging in the value of hs into the linearized formula, you will notice that f(h,Fs) = 0, the point on the plot
where both the linearization and the characteristic equation coincide (i.e. are tangent).
h remains a variable for the derivation so that students may notice that it is a dependant variable (dependent upon the flowrate). Bequette
must (and does) provide both the values for Fs, hs so that a linearization may proceed (otherwise we will simply have an equation describing
the slope with no characteristic point to pass through).
System Parameters (Fs is an example flow rate at steady state, this information is used to determine the intercept of the linearized formula):
2.5 3
2
1 ft 1ft
A = 1f t ; hs = 5ft; β = – ; Fs =
√5 min min
Rewriting the system with the input variables:

dh 1 −
= f (h, Fs ) = 1 + – ∗ √h
dt √5
To linearize around a certain point, simply evaluate the derivative of the desired function and add in a corrective constant, C, represented by
the value of the function at the initial (specified) condition.
∂f
\(h, Fs ) ≈ f(h, Fs ) + | ,Fs (h − hs )
h
∂h
\[\(h,F_s) \approx 0+ \frac{-1}

|_h, _{F_s}(h-h_s) \nonumber \]
Selecting, h=5 to linearize around, we present an adapted version of the text example (linearization is in red):
\[\(h,1) \approx 0+ \frac{-1}
|_h, _{F_s}(h-5) \nonumber \]
\[\(5,1) \approx 0+ \frac{-1}
|_h, _{F_s}(5-5)= 0 \nonumber \]
f(h,1)approx. = − 0.1 * (h − 5)
\isual3.jpg
The physical significance of this example:

The model describes a situation where a tank maintains the steady state set point, a level (height) of 5 feet, by regulating the input (and/or
output) flow rate(s). A height above the preferred level of 5 ft. results in a negative height change per time; conversely, a height below the
preferred level of 5 ft. results in a positive height change per time, so as to always approach the steady state level.
Also, please note that the linearized system responds relatively slower than the characteristic equation system (nonlinear)when tank level
falls below 5 ft. and faster than the characteristic equation system when the tank level is above 5 ft.
Please read on below to learn a step-by-step technique on how to complete this method for most equations.
10.2.6: LINEARIZATION USING MATHEMATICA

Most ODEs are not as simple as the example worked out above. For those that can be solved analytically, the 'by-hand' method from the
previous section can be used, although the math becomes tricky and tedious when one is dealing with systems of ODEs. For those that
cannot be solved analytically and can only be approximated numerically, it is sometimes impossible to do the numerical analysis by hand.
For these ODEs, powerful computational tools such as Mathematica can be used to perform linearization.
The following Mathematica file contains a worked out example (CSTR w/ heat exchange). This example contains multiple ODEs and
multiple types of variables. When dealing with ODEs, there are three main type of variables: state, input, and output. State variables
describes the system at any given time (ie. Ca, Cb, T, etc.). Input variables are simply inputs into the system. Output variables are those that
are not directly related to the inputs, but are dependent on the state variables (ie. k, the reaction constant). You will discover in the CSTR
example that it is necessary to linearize state variables and output variables separately.
Before proceeding to the Mathematica file we will first present some of the Mathematica commands that will be necessary to linearize the
CSTR example.
Note that you need to press shift+enter (shift+return on Mac) after a command to evaluate the expression.
Example:
In[1]= ((0.04587*54.32497+4.59058433734947^7)/48700043)^131.94564
Out[1]= 0.000411241
Another thing to note is the use of assignment and equality operators.
= is called immediate assignment operator and := is called delayed assignment operator. When immediate assignment operator is used,
the right hand side of the operator is evaluated every time an assignment is made. However, when delayed assignment operator is used, the
right hand side is evaluated only when the value of the expression on the left hand side is requested.
lhs=rhs rhs is intended to be the "final value" of lhs (e.g., f[x_]=1-x^2) lhs:=rhs rhs gives a "command" or "program" to be executed
whenever you ask for the value of lhs (e.g., f[x_]:=Expand[1-x^2])
p := x^3 - 6 x^2 + 11 x - 6
This input, which produces no output, assigns the polynomial expression to the variable p. The assignment is done because this polynomial
may be used again later, so the assignment saves retyping it. The equality operator ==, which is used here to form the equation we are
asking Mathematica to solve:
In[2]:= Solve[p == 0]
Out[2]=
When solving nonlinear ODEs using Mathematica, it is necessary to form a matrix and there are several commands that can be used to
create matrices.
D[X,Y]
This command takes the partial derivative of an expression (X) with respect to a variable (Y) defined by the user. The first term in the
square brackets is the experession you wish to differentiate, the second is the variable you wish to differentiate with respect to.
AppendRows[column1,column2]
This command creates a matrix by combining two columns. In order to use this command you must first load the Matrix Manipulation
package using the command <<LinearAlgebra`MatrixManipulation`
M=
This command creates a matrix, in this case a 3x3 matrix.
MatrixForm[X]
This command displays a specified matrix in matrix form.
These commands will allow us to proceed with the linearization of the CSTR problem shown below.
The first three commands input the equations that govern the behavior of the concentration, Ca, the temperature, T, and the rate of reaction,
k. Once the equations have been entered into Mathematica a matrix, M, with one column and two rows containing the equations for Ca and
T is created. Next two new matrices are created by taking the derivative of matrix M is with respect to both Ca and T. These matrices, called
column1 and column2 respectively, together form the Jacobian of the system and is then displayed in matrix form. The Jacobian matrix is
the matrix of all first-order partial derivatives of a vector-valued function. Its importance lies in the fact that it represents the best linear
approximation to a differentiable function near a given point. For example, given a set y=f(x) of n equations in variables x1, x2,...,xn, the
Jacobian matrix is defined as following:
\acobianmatrix. gif
\athematicapg2.J P G
Now that we have the Jacobian we need to create a matrix containing the deviation from steady state for each of the variables. This matrix,
SS, contains the actual concentration and temperature, C and T, minus the steady state concentration and temperature, Cas and Ts. The
matrix is then displayed in matrix form. We now have both the Jacobian and the deviation matrix for the state variables. The next four
commands create the Jacobian and deviation matrix for the output variable, k. The first command creates the Jacobian matrix by taking the
derivative of the k equation with respect to Ca and T. The Jacobian is then shown in matrix form. Finally the deviation matrix for k is
created in the same manner as above and then displayed in matrix form. Note that because k is defined above, this expression is substituted
in for k in the deviation matrix. The following Mathematica file contains the code shown above with extra comments explaining why each
step is performed Media:cstr_example.nb. It may be useful to downolad this file and run the program in Mathematica yourself to get a feel
for the syntax. Downloading the file will also allow you to make any changes and edits to customize this example to another example of
interest.
Note: This file needs to be saved to your computer, and then opened using Mathematica to properly run.
To see another example lets linearize the first 4 differential equations given in the introduction section.
Non-linear system of equations:
dA 2 3
= 3A + 2B + C − 7D
dt
dB 2
= A+ C + 2D
dt
dC 2 2
= A + 4B −C
dt
dD
= 2C − D
dt
After linearization (around the steady state point {-0.47,-0.35,0.11,0.23}:

′
A −2.83 2 1 −1.10 A 0.50
⎛ ⎞ ⎛ ⎞⎛ ⎞ ⎛ ⎞
′
⎜ B ⎟ ⎜ 1 0 0.23 2 ⎟⎜B ⎟ ⎜ 0.01 ⎟
⎜ ⎟ = ⎜ ⎟⎜ ⎟+⎜ ⎟
⎜ C′ ⎟ ⎜ 1 −2.78 −0.23 0 ⎟⎜C ⎟ ⎜ 0.47 ⎟
⎝ ′ ⎠ ⎝ ⎠⎝ ⎠ ⎝ ⎠
D 0 0 2 −1 D 0
The 4 differential equations above are added into a Mathematica code as “eqns” and “s1” is the fixed points of the differentials. The steady
state values found for “a, b, c, and d” are called "s1doubleBrackets(7)” After the steady state values are found, the Jacobian matrix can be
found at those values.
To find “k1, k2, k3, and k4” the constants of the Linearization matrix equation, “m1” must be defined, which is the 2nd matrix on the right-
hand side of the Linearization matrix equation.
To determine the k values (in matrix form), execute the dot product of "m1" and the “Jac” matrix, which is done by the "." operator.
Therefore it should look like "Jac.m1"
To obtain the k values, determing the "Jac.m1" at the steady state values, which is done by the "/." operator. Therefore it should look like
“Jac.m1/.s1doubleBrackets(7)”
 EXAMPLE 10.2.1
Let's say you have a system governed by the following system of equations:
dXa
2
= 3Xa + 2Xb + Fo
dt
dXb
2
= 6Xa + 9X + Xa Xb + Fo
b
dt
2 2
Y = Xa + 3 ∗ Fo
In this case, X and X are state variables (they describe the state of the system at any given time), F is an input variable (it enters the
a b o
system and is independent of all of the state variables), and Y is the output of the system and is dependent on both the state and input
variables. Please linearize this system using Mathematica.
 EXAMPLE 10.2.2
Use the Mathematica file presented in the article to linearize the CSTR example presented in the ODE & Excel CSTR model with heat
exchangewiki around the steady state (T = 368.4K and Ca = 6.3mol/L). Use the constants presented in their example. Use the
linearization to approximate Ca at 369K.
Solution
The linear approximation gives concentration of a (Ca) of 7.13 mol/L. The Mathematica file used to solve this problem can be
downloaded here. Media:LinearizationExample.nb If you are having trouble running the Mathematica file try right clicking the link and
using the Save Target As option.
Exercise 10.2.1
In an ODE linearization, what point in the process is the linearization generally centered around?
a. Starting point
b. Ending point
c. Steady-state point
d. Half-way point
Answer
Answer: c
Exercise 10.2.2
What does the D[] function in Mathematica do?
a. Find the determinant of a matrix
b. Finds the partial derivative a function or functions
c. Find the dot product of two matrices
d. Integrate a matrix or function
Answer
Answer: b
10.2.7: REFERENCES
Bequette, B. Wayne. Process Dynamics: Modeling, Analysis, and Simulation. Prentice- Hall PTR, Upper Saddle River, NJ 07458 (c)
2006.
Kravaris, Costas. Chemical Process Control: A Time Domain Approach, Appendix A. Department of Chemical Engineering, University
of Michigan, Ann Arbor, MI.
Bhatti, M. Asghar. Practical Optimization Methods With Mathematica Applications. Springer Telos.
10.2: Linearizing ODEs is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Navin Raj Bora, Dallas Burkholder, Nina
Mohan, Sarah Tschirhart, & Sarah Tschirhart via source content that was edited to conform to the style and standards of the LibreTexts platform; a detailed
10.3: EIGENVALUES AND EIGENVECTORS
10.3.1: WHAT ARE EIGENVECTORS AND EIGENVALUES?
Eigenvectors ( ) and Eigenvalues (λ ) are mathematical tools used in a wide-range of applications. They are used to solve differential
equations, harmonics problems, population models, etc. In Chemical Engineering they are mostly used to solve differential equations and to
analyze the stability of a system.
 DEFINITION: EIGENVECTOR AND EIGENVALUES

An Eigenvector is a vector that maintains its direction after undergoing a linear transformation.
An Eigenvalue is the scalar value that the eigenvector was multiplied by during the linear transformation.
Eigenvectors and Eigenvalues are best explained using an example. Take a look at the picture below.
In the left picture, two vectors were drawn on the Mona Lisa. The picture then under went a linear transformation and is shown on the right.
The red vector maintained its direction; therefore, it’s an eigenvector for that linear transformation. The blue vector did not maintain its
director during the transformation; thus, it is not an eigenvector. The eigenvalue for the red vector in this example is 1 because the arrow
was not lengthened or shortened during the transformation. If the red vector, on the right, were twice the size than the original vector then
the eigenvalue would be 2. If the red vector were pointing directly down and remained the size in the picture, the eigenvalue would be -1.
Now that you have an idea of what an eigenvector and eigenvalue are we can start talking about the mathematics behind them.
Fundamental Equation
The following equation must hold true for Eigenvectors and Eigenvalues given a square matrix A :
A⋅v = λ⋅v (10.3.1)
where:
A is a square matrix
v is the Eigenvector
λ is the Eigenvalue
Let's go through a simple example so you understand the fundamental equation better.
 EXAMPLE 10.3.1
Is v an eigenvector with the corresponding λ = 0 for the matrix A?

1
v = [ ]
−2
6 3
A = [ ]
−2 −1
Solution
A⋅v = λ⋅v
6 3 1 1
[ ] ⋅[ ] = 0[ ]
−2 −1 −2 −2
0 0
[ ] = [ ]
0 0
Therefore, it is true that v and λ = 0 are an eigenvector and eigenvalue respectively, for A . (See section on Matrix operations, i.e.
matrix multiplication)
10.3.2: CALCULATING EIGENVALUES AND EIGENVECTORS

Calculation of the eigenvalues and the corresponding eigenvectors is completed using several principles of linear algebra. This can be done
by hand, or for more complex situations a multitude of software packages (i.e. Mathematica) can be used. The following discussion will
work for any nxn matrix; however for the sake of simplicity, smaller and more manageable matrices are used. Note also that throughout this
article, boldface type is used to distinguish matrices from other variables.
10.3.2.1: LINEAR ALGEBRA REVIEW

For those who are unfamiliar with linear algebra, this section is designed to give the necessary knowledge used to compute the eigenvalues
and eigenvectors. For a more extensive discussion on linear algebra, please consult the references.
10.3.2.1.1: BASIC MATRIX OPERATIONS
An m x n matrix A is a rectangular array of mn numbers (or elements) arranged in horizontal rows (m) and vertical columns (n):
a11 a1j a1n
⎡ ⎤
A = ⎢ ai1 aij ain ⎥
⎣ ⎦
am1 amj amn
To represent a matrix with the element aij in the ith row and jth column, we use the abbreviation A = [aij]. Two m x n matrices A = [aij] and
B = [bij] are said to be equal if corresponding elements are equal.
10.3.2.1.2: ADDITION AND SUBTRACTION
We can add A and B by adding corresponding elements:
A + B = [aij ] + [bij ] = [aij + bij
This will give the element in row i and column j of C = A + B to have

cij = aij + bij .
More detailed addition and subtraction of matrices can be found in the example below.
1 2 6 8 3 5 1+ 8 2+ 3 6+ 5 9 5 11
⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
⎢4 5 10 ⎥ + ⎢ 5 4 4⎥ = ⎢ 4+ 5 5+ 4 10 + 4 ⎥ = ⎢ 9 9 14 ⎥
⎣ ⎦ ⎣ ⎦ ⎣ ⎦ ⎣ ⎦
5 3 11 3 0 6 5+ 3 3+ 0 11 + 6 8 3 17
10.3.2.1.3: MULTIPLICATION
Multiplication of matrices are NOT done in the same manner as addition and subtraction. Let's look at the following matrix multiplication:
A× B = C
A is an m × n matrix, B is an n × p matrix, and C is an m × p matrix. Therefore the resulting matrix, C , has the same number of rows as
the first matrix and the same number of columns as the second matrix. Also the number of columns in the first is the same as the number of
rows in the second matrix.
The value of an element in C (row i, column j) is determined by the general formula:
n
ci,j = ∑ ai,k bk,j
k=1
Thus,
1 2 6 3 0 1× 3+ 2× 0+ 6× 5 1× 0+ 2× 1+ 6× 1
⎡ ⎤⎡ ⎤ ⎡ ⎤
⎢4 5 10 ⎥ ⎢ 0 1 ⎥ = ⎢ 4 × 3 + 5 × 0 + 10 × 5 4 × 0 + 5 × 1 + 10 × 1 ⎥
⎣ ⎦⎣ ⎦ ⎣ ⎦
5 3 11 5 1 5 × 3 + 3 × 0 + 11 × 5 5 × 0 + 3 × 1 + 11 × 1
33 8
⎡ ⎤
= ⎢ 62 15 ⎥
⎣ ⎦
70 14
It can also be seen that multiplication of matrices is not commutative (A B ≠B A). Multiplication of a matrix by a scalar is done by
multiplying each element by the scalar.
cA = Ac =[caij]
1 2 6 2 4 12
⎡ ⎤ ⎡ ⎤
2⎢4 5 10 ⎥ = ⎢ 8 10 20 ⎥
⎣ ⎦ ⎣ ⎦
5 3 11 10 6 22
10.3.2.1.4: IDENTITY MATRIX

The identity matrix is a special matrix whose elements are all zeroes except along the primary diagonal, which are occupied by ones. The
identity matrix can be any size as long as the number of rows equals the number of columns.
1 0 0 0
⎡ ⎤
⎢0 1 0 0⎥
I = ⎢ ⎥
⎢0 0 1 0⎥
⎣ ⎦
0 0 0 1
10.3.2.1.5: DETERMINANT
The determinant is a property of any square matrix that describes the degree of coupling between equations. For a 2x2 matrix the
determinant is:
∣a b ∣
det(A) = ∣ ∣ = ad − bc
∣c d∣
Note that the vertical lines around the matrix elements denotes the determinant. For a 3x3 matrix the determinant is:
∣a b c ∣
∣ ∣ ∣ e f ∣ ∣d f ∣ ∣d e ∣
det(A) = d e f = a∣ ∣− b∣ ∣+ c∣ ∣ = a(ei − fh) − b(di − fg) + c(dh − eg)
∣ ∣ ∣h ∣ ∣g ∣ ∣g
i i h∣
∣g h i ∣
Larger matrices are computed in the same way where the element of the top row is multiplied by the determinant of matrix remaining once
that element’s row and column are removed. Terms where the top elements in odd columns are added and terms where the top elements in
even rows are subtracted (assuming the top element is positive). For matrices larger than 3x3 however; it is probably quickest to use math
software to do these calculations since they quickly become more complex with increasing size.
10.3.2.2: SOLVING FOR EIGENVALUES AND EIGENVECTORS
The eigenvalues (λ) and eigenvectors (v), are related to the square matrix A by the following equation. (Note: In order for the eigenvalues
to be computed, the matrix must have the same number of rows as columns.)
(A − λI) ⋅ v = 0
This equation is just a rearrangement of the Equation 10.3.1 . To solve this equation, the eigenvalues are calculated first by setting det(A-λI)
to zero and then solving for λ. The determinant is set to zero in order to ensure non-trivial solutions for v, by a fundamental theorem of
linear algebra.
4 1 4
⎡ ⎤
A = ⎢1 7 1⎥
⎣ ⎦
4 1 4
4 1 4 −λ 0 0
⎡ ⎤ ⎡ ⎤
A − λI = ⎢ 1 7 1⎥+⎢ 0 −λ 0 ⎥
⎣ ⎦ ⎣ ⎦
4 1 4 0 0 −λ
∣ 4− λ 1 4 ∣
∣ ∣
det(A − λI) = 1 7− λ 1 = 0
∣ ∣
∣ 4 1 4− λ ∣
2 3
−54λ + 15λ −λ = 0
−λ(λ − 6)(λ − 9) = 0
λ = 0, 6, 9
For each of these eigenvalues, an eigenvector is calculated which will satisfy the equation (A-λI)v=0 for that eigenvalue. To do this, an
eigenvalue is substituted into A-λI, and then the system of equations is used to calculate the eigenvector. For λ = 6
4− 6 1 4 x −2 1 4 x
⎡ ⎤⎡ ⎤ ⎡ ⎤⎡ ⎤
(A − 6I)v = ⎢ 1 7− 6 1 ⎥⎢y ⎥ = ⎢ 1 1 1 ⎥⎢y ⎥ = 0
⎣ ⎦⎣ ⎦ ⎣ ⎦⎣ ⎦
4 1 4− 6 z 4 1 −2 z
Using multiplication we get a system of equations that can be solved.
Equating (1) and (3) ...
Plugging this into (2)...
There is one degree of freedom in the system of equations, so we have to choose a value for one variable. By convention we choose x = 1
then
A degree of freedom always occurs because in these systems not all equations turn out to be independent, meaning two different equations
can be simplified to the same equation. In this case a small number was chosen (x = 1) to keep the solution simple. However, it is okay to
pick any number for x, meaning that each eigenvalue potentially has an infinite number of possible eigenvectors that are scaled based on the
initial value of x chosen. Said another way, the eigenvector only points in a direction, but the magnitude of this pointer does not matter. For
this example, getting an eigenvector that is is identical to getting an eigenvector that is or an eigenvector that is
scaled by some constant, in this case 2.
Finishing the calcualtions, the same method is repeated for λ = 0 and λ = 9 to get their corresponding eigenvectors.
For λ = 0,
For λ = 9,
In order to check your answers you can plug your eigenvalues and eigenvectors back into the governing equation . For this
example, λ = 6 and was double checked.
Therefore, λ = 6 and are both an eigenvalue-eigenvector pair for the matrix .
10.3.3: CALCULATING EIGENVALUES AND EIGENVECTORS USING NUMERICAL SOFTWARE

10.3.3.1: EIGENVALUES IN MATHEMATICA
For larger matrices (4x4 and larger), solving for the eigenvalues and eigenvectors becomes very lengthy. Therefore software programs like
Mathematica are used. The example from the last section will be used to demonstrate how to use Mathematica. First we can generate the
matrix A. This is done using the following syntax:
A = {{4, 1, 4}, {1, 7, 1}, {4, 1, 4}}
It can be seen that the matrix is treated as a list of rows. Elements in the same row are contained in a single set of brackets and separated by
commas. The set of rows are also contained in a set of brackets and are separated by commas. A screenshot of this is seen below. (Note: The
"MatrixForm[]" command is used to display the matrix in its standard form. Also in Mathematica you must hit Shift + Enter to get an
output.)
Next we find the determinant of matrix A-λI, by first subtracting the matrix λI from A (Note: This new matrix, A-λI, has been called A2).
The command to find the determinant of a matrix A is:

Det[A]
For our example the result is seen below. By setting this equation to 0 and solving for λ, the eigenvalues are found. The Solve[] function is
used to do this. Notice in the syntax that the use of two equal signs (==) is used to show equivalence whereas a single equal sign is used for
defining a variable.
Solve[{set of equations},{variables being solved}]
Alternatively the eigenvalues of a matrix A can be solved with the Mathematica Eigenvalue[] function:
Eigenvalues[A]
Note that the same results are obtained for both methods.
To find the eigenvectors of a matrix A, the Eigenvector[] function can be used with the syntax below.
Eigenvectors[A]
The eigenvectors are given in order of descending eigenvalues.
One more function that is useful for finding eigenvalues and eigenvectors is Eigensystem[]. This function is called with the following
syntax.
Eigensystem[A]
In this function, the first set of numbers are the eigenvalues, followed by the sets of eigenvectors in the same order as their corresponding
eigenvalues.
The Mathematica file used to solve the example can be found at this link.Media:Eigen Solve Example.nb
10.3.3.2: MICROSOFT EXCEL

Microsoft Excel is capable of solving for Eigenvalues of symmetric matrices using its Goal Seek function. A symmetric matrix is a square
matrix that is equal to its transpose and always has real, not complex, numbers for Eigenvalues. In many cases, complex Eigenvalues cannot
be found using Excel. Goal Seek can be used because finding the Eigenvalue of a symmetric matrix is analogous to finding the root of a
polynomial equation. The following procedure describes how to calculate the Eigenvalue of a symmetric matrix in the Mathematica tutorial
using MS Excel.
(1) Input the values displayed below for matrix A then click menu INSERT-NAME-DEFINE “matrix_A” to name the matrix.
(2) Similarly, define identity matrix I by entering the values displayed below then naming it “matrix_I.”
(3) Enter an initial guess for the Eigenvalue then name it “lambda.”
(4) In an empty cell, type the formula =matrix_A-lambda*matrix_I. Highlight three cells to the right and down, press F2, then press
CRTL+SHIFT+ENTER. Name this matrix “matrix_A_lambda_I.”
(5) In another cell, enter the formula =MDETERM(matrix_A_lambda_I). This is the determinant formula for matrix_A_lambda_I.
(6) Click menu Tools-Goal Seek… and set the cell containing the determinant formula to zero by changing the cell containing lambda.
(7) To obtain all three Eigenvalues for matrix A, re-enter different initial guesses. Excel calculates the Eigenvalue nearest to the value of the
initial guess. The Eigenvalues for matrix A were determined to be 0, 6, and 9. For instance, initial guesses of 1, 5, and 13 will lead to
Eigenvalues of 0, 6, and 9, respectively.
The MS Excel spreadsheet used to solve this problem, seen above, can be downloaded from this link: Media:ExcelSolveEigenvalue.xls.
10.3.4: CHEMICAL ENGINEERING APPLICATIONS

The eigenvalue and eigenvector method of mathematical analysis is useful in many fields because it can be used to solve homogeneous
linear systems of differential equations with constant coefficients. Furthermore, in chemical engineering many models are formed on the
basis of systems of differential equations that are either linear or can be linearized and solved using the eigenvalue eigenvector method. In
general, most ODEs can be linearized and therefore solved by this method. Linearizing ODEs For example, a PID control device can be
modeled with ODEs that may be linearized where the eigenvalue eigenvector method can then be implemented. If we have a system that
can be modeled with linear differential equations involving temperature, pressure, and concentration as they change with time, then the
system can be solved using eigenvalues and eigenvectors:
dP
= 4P − 4T + C
dt
dT
= 4P − T + 3C
dt
dC
= P + 5T − C
dt
Note: This is not a real model and simply serves to introduce the eigenvalue and eigenvector method.
A is just the matrix that represents the coefficients in the above linear differential equations. However, when setting up the matrix, A, the
order of coefficients matters and must remain consistent. Namely, in the following representative matrix, the first column corresponds to the
coefficients of P, the second column to the coefficients of T, and the third column corresponds to the coefficients of C. The same goes for
the rows. The first row corresponds to , the second row corresponds to , and the third row corresponds to :
4 −4 1
⎡ ⎤
A = ⎢4 −1 3 ⎥
⎣ ⎦
1 5 −1
It is noteworthy that matrix A is only filled with constants for a linear system of differential equations. This turns out to be the case because
each matrix component is the partial differential of a variable (in this case P, T, or C). It is this partial differential that yields a constant for
linear systems. Therefore, matrix A is really the Jacobian matrix for a linear differential system.
Now, we can rewrite the system of ODE's above in matrix form.
where
P (t)
⎡ ⎤
x(t) = ⎢ T (t) ⎥
⎣ ⎦
C(t)
We guess trial solutions of the form

λt
x = ve
since when we substitute this solution into the matrix equation, we obtain
λt λt
λv e = Av e
After cancelling the nonzero scalar factor eλt, we obtain the desired eigenvalue problem.
Av = λv
Thus, we have shown that

λt
x = ve
will be a nontrivial solution for the matrix equation as long as v is a nonzero vector and λ is a constant associated with v that satisfies the
eigenvalue problem.
In order to solve for the eigenvalues and eigenvectors, we rearrange the Equation 10.3.1 to obtain the following:
4− λ −4 1 x
⎛ ⎡ ⎤ ⎡ ⎤
⎜Λ λI)v = 0 ⎢ 4 1 λ 3 ⎥⋅⎢ y ⎥ = 0
⎝ ⎣ ⎦ ⎣ ⎦
1 5 −1 − λ z
For nontrivial solutions for v, the determinant of the eigenvalue matrix must equal zero, det(A − λI) = 0 . This allows us to solve for the
eigenvalues, λ. You should get, after simplification, a third order polynomial, and therefore three eigenvalues. (see section on Solving for
Eigenvalues and Eigenvectors for more details) Using the calculated eignvalues, one can determine the stability of the system when
disturbed (see following section).
Once you have calculated the three eigenvalues, you are ready to find the corresponding eigenvectors. Plug the eigenvalues back into the
equation and solve for the corresponding eigenvectors. There should be three eigenvectors, since there were three
eigenvalues. (see section on Calculating Eigenvalues and Eigenvectors for more details)
The solution will look like the following:
P (t) x1 x2 x3
⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
λ1 t λ2 t λ3 t
⎢ T (t) ⎥ = c1 ⎢ y1 ⎥ e + c2 ⎢ y2 ⎥ e + c3 ⎢ y3 ⎥ e
⎣ ⎦ ⎣ ⎦ ⎣ ⎦ ⎣ ⎦
C(t) z1 z2 z3
where
x1, x2, x3, y1, y2, y3, z1, z2, z3 are all constants from the three eigenvectors. The general solution is a linear combination of these three
solution vectors because the original system of ODE's is homogeneous and linear. It is homogeneous because the derivative expressions
have no cross terms, such as PC or TC, and no dependence on t. It is linear because the derivative operator is linear. To solve for c1, c2, c3
there must be some given initial conditions (see Worked out Example 1).
This Wiki does not deal with solving ODEs. It only deals with solving for the eigenvalues and eigenvectors. In Mathematica the Dsolve[]
function can be used to bypass the calculations of eigenvalues and eigenvectors to give the solutions for the differentials directly. See Using
eigenvalues and eigenvectors to find stability and solve ODEs for solving ODEs using the eigenvalues and eigenvectors method as well as
with Mathematica.
This section was only meant to introduce the topic of eigenvalues and eigenvectors and does not deal with the mathematical details
presented later in the article.
10.3.5: USING EIGENVALUES TO DETERMINE EFFECTS OF DISTURBING A SYSTEM

Eigenvalues can help determine trends and solutions with a system of differential equations. Once the eigenvalues for a system are
determined, the eigenvalues can be used to describe the system’s ability to return to steady-state if disturbed.
The simplest way to predict the behavior of a system if disturbed is to examine the signs of its eigenvalues. Negative eigenvalues will drive
the system back to its steady-state value, while positive eigenvalues will drive it away. What happens if there are two eigenvalues present
with opposite signs? How will the system respond to a disturbance in that case? In many situations, there will be one eigenvalue which has a
much higher absolute value than the other corresponding eigenvalues for that system of differential equations. This is known as the
“dominant eigenvalue”, and it will have the greatest effect on the system when it is disturbed. However, in the case that the eigenvalues are
equal and opposite sign there is no dominant eigenvalue. In this case the constants from the initial conditions are used to determine the
stability.
Another possible case within a system is when the eigenvalue is 0. When this occurs, the system will remain at the position to which it is
disturbed, and will not be driven towards or away from its steady-state value. It is also possible for a system to have two identical
eigenvalues. In this case the two identical eigenvalues produce only one eigenvector. Because of this, a situation can arise in which the
eigenvalues don’t give the complete story of the system, and another method must be used to analyze it, such as the Routh Stability
Analysis Method.
Eigenvalues can also be complex or pure imaginary numbers. If the system is disturbed and the eigenvalues are non-real number, oscillation
will occur around the steady state value. If the eigenvalue is imaginary with no real part present, then the system will oscillate with constant
amplitude around the steady-state value. If it is complex with a positive real part, then the system will oscillate with increasing amplitude
around the function, driving the system further and further away from its steady-state value. Lastly, if the eigenvalue is a complex number
with a negative real part, then the system will oscillate with decreasing amplitude until it eventually reaches its steady state value again.
Below is a table of eigenvalues and their effects on a differential system when disturbed. It should be noted that the eigenvalues developed
for a system should be reviewed as a system rather than as individual values. That is to say, the effects listed in the table below do not fully
represent how the system will respond. If you were to pretend that eigenvalues were nails on a Plinko board, knowing the location and angle
of one of those nails would not allow you to predict or know how the Plinko disk would fall down the wall, because you wouldn't know the
location or angle of the other nails. If you have information about all of the nails on the Plinko board, you could develop a prediction based
on that information. More information on using eigenvalues for stability analysis can be seen here, Using eigenvalues and eigenvectors to
find stability and solve ODEs_Wiki.
The above picture is of a plinko board with only one nail position known. Without knowing the position of the other nails, the Plinko disk's
fall down the wall is unpredictable.
Knowing the placement of all of the nails on this Plinko board allows the player to know general patterns the disk might follow.
10.3.5.1: REPEATED EIGENVALUES

A final case of interest is repeated eigenvalues. While a system of N differential equations must also have N eigenvalues, these values may
not always be distinct. For example, the system of equations:
dCA −−
= fA,in ρCA = fout ρCA √V1 − V1 k1 CA CB
dt
dCB −−
= fB,in ρCBin − fout ρCB √V1 − V1 k1 CA CB
dt
dCC −−
= − fout ρCc √V1 + V1 k1 CA CB
dt
dV1 −−
= fA,in + fB,in − fout √V1
dt
dV2 −− −−
= fout √V1 − fcustomer √V2
dt
dCC2 −− −−
= fout , ρCC √V1 − fcustomer , ρCC2 √V2
dt
May yield the eigenvalues: {-82, -75, -75, -75, -0.66, -0.66}, in which the roots ‘-75’ and ‘-0.66’ appear multiple times. Repeat eigenvalues
bear further scrutiny in any analysis because they might represent an edge case, where the system is operating at some extreme. In
mathematical terms, this means that linearly independent eigenvectors cannot be generated to complete the matrix basis without further
analysis. In “real-world” engineering terms, this means that a system at an edge case could distort or fail unexpectedly.
However, for the general solution:
Y (t) = k1 exp(λt)V1 + k2 exp(λt) (t V1 + V2 )
If λ < 0 , as t approaches infinity, the solution approaches 0, indicating a stable sink, whereas if λ > 0, the solution approaches infinity in
the limit, indicating an unstable source. Thus the rules above can be roughly applied to repeat eigenvalues, that the system is still likely
stable if they are real and less than zero and likely unstable if they are real and positive. Nonetheless, one should be aware that unusual
behavior is possible. This course will not concern itself with resultant behavior of repeat eigenvalues, but for further information, see:
http://math.rwinters.com/S21b/supplements/newbasis.pdf
http://www.sosmath.com/diffeq/system/linear/eigenvalue/repeated/repeated.html
 EXAMPLE 10.3.1
Your immediate supervisor, senior engineer Captain Johnny Goonewadd, has brought you in on a project dealing with a new silcone-
based sealant that is on the ground level of research. Your job is to characterize the thermal expansion of the sealant with time given a
constant power supply. Luckily, you were given a series of differential equations that relate temperature and volume in terms of one
another with respect to time (Note: T and V are both dimensionless numbers with respect to their corresponding values at t=0). Solve
the system of differentials and determine the equations for both Temperature and Volume in terms of time.
Solution
You are given the initial condition at time t=0, T=1 and V=1
dT
= 4T − 3V
dt
dV
= 3T + 4V
dt
By defining a matrix for both the coefficients and dependant variables we are able to rewrite the above series of differentials in matrix
form
4 −3
A = [ ]
3 4
T
X = [ ]
V
dT
dV
A∗X = [ ]
dV
dt
Lambda is inserted into the A matrix to determine the Eigenvalues
For each eigenvalue, we must find the eigenvector. Let us start with λ1 = 4 − 3i
Now we find the eigenvector for the eigenvalue λ2 = 4 + 3i
The general solution is in the form
A mathematical proof, Euler's formula, exists for transforming complex exponentials into functions of sin(t) and cos(t)
Thus
Simplifying
Since we already don't know the value of c1, let us make this equation simpler by making the following substitution
Thus, we get have our solution in terms of real numbers
Or, rewriting the solution in scalar form
Now that we have our solutions, we can use our initial conditions to find the constants c3 and c4
First initial condition: t=0, T=1
First initial condition: t=0, V=1
We have now arrived at our solution
See Using eigenvalues and eigenvectors to find stability and solve ODEs_Wiki for solving ODEs using the eigenvalues and
eigenvectors.
 EXAMPLE 10.3.2
Process Engineer, Dilbert Pickel, has started his first day for the Helman's Pickel Brine Factory. His first assignment is with a pre-
startup team formulated to start up a new plant designed to make grousley sour pickle brine. Financial constraints have demanded that
the process begin to produce good product as soon as possible. However, you are forced to reflux the process until you reach the set
level of sourness. You have equations that relate all of the process variable in terms of one another with respect to time. Therefore, it is
Dill Pickles job to characterize all of the process variables in terms of time (dimensionless Sourness, Acidity, and Water content; S, A,
& W respectively). Below is the set of differentials that will be used to solve the equation.
\[\
dS
= S + A + 10W
dt
dA
= S + 5A + 2W (10.3.2)
dt
dW
= 4S + 3A + 8W
dt
]
Thus the coefficient matrix
1 1 10
⎡ ⎤
A = ⎢1 5 2 ⎥
⎣ ⎦
4 3 8
Using mathematica it is easy to input the coefficients of the system of equations into a matrix and determine both the eigenvalues and
eigenvectors.
The eigenvectors can then be used to determine the final solution to the system of differentials. Some data points will be necessary in
order to determine the constants.
S 0.88 2 −2.74
⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤
(5+ √89 k 4t
⎢ A ⎥ = C1 ⎢ 0.38 ⎥ e + C2 ⎢ −4 ⎥ e + C3 ⎢ 0.10 ⎥
⎣ ⎦ ⎣ ⎦ ⎣ ⎦ ⎣ ⎦
W 1 1 1
See Using eigenvalues and eigenvectors to find stability and solve ODEs_Wiki for solving ODEs using the eigenvalues and
eigenvectors.
 EXAMPLE 10.3.3
It is possible to find the Eigenvalues of more complex systems than the ones shown above. Doing so, however, requires the use of
advanced math manipulation software tools such as Mathematica. Using Mathematica, it is possible to solve the system of ODEs
shown below.
dX 10XY F
= 8X +
dt X+Z
dY 3XY
= 4F − Y − Z −
dt X+Y
dZ
= 9X − 2Z + F
dt
Obviously, this system of ODEs has 4 variables and only 3 equations.
Obviously, this is a more complex set of ODEs than the ones shown above. And even though they will create a more complex set of
Eigenvalues, they are solved for in the same way when using Mathematica.
Using the code shown below:
The equations can be entered into Mathematica. The equations are shown again in the output
Then, using the next bit of code:
The it is possible to find where the equations are equal to 0 (i.e. the fixed points). The results of this is also shown in the image above.
It's notable that 3 solutions are found. This makes sense as the system is 3 ODEs.
The Jacobian can then be found by simply using the code shown below.
The ersults of finding the Jacobian are shown in the equation above.
Finally, to find one of the Eigenvalues, one can simply use the code shown below.
This gives the Eigenvalue when the first fixed point (the first solution found for "s") is applied. The other two solutions could be found
by simply changing the fixed blade that is referred to when finding t1. The other Eigenvalues are not shown because of their large size.
Exercise 10.3.1: Calculating Eigenvalues and Eigenvectors using Numerical Software

What are the eigenvalues for the matrix A?
4 2
A = [ ]
3 −1
a. λ1 = −2 and λ = 5
1
b. λ1 = 2 and λ = −5
1
c. λ1 = 2 and λ = 5
1
d. λ1 = −2 and λ = −5
1
Answer
a.
Exercise 10.3.2: Using Eigenvalues to Determine Effects of Disturbing a System

When a differential system with a real negative eigenvalue is disturbed, the system is...
a. Driven away from the steady state value
b. Unchanged and remains at the disturbed value
c. Driven back to the steady state value
d. Unpredictable and the effects can not be determined
Answer
c. A real negative eigenvalue is indicative of a stable system that will return to the steady state value after it is disturbed.
10.3.6: REFERENCES
Kravaris, Costas: Chemical Process Control: A Time-Domain Approach. Ann Arbor: The University of Michigan, pp 1-23, A.1-A.7.
Bhatti, M. Asghar: Practical Optimization Methods with Mathematica Applications. Springer, pp 75-85, 677-691.
Strang, Prof. Gilbert: “Eigenvalues and Eigenvectors.” Math 18.06. Lord Foundation of Massachusetts. Fall 1999.
Edwards, C. Henry and David E. Penney: Differential Equations: Computing and Modeling. Upper Saddle River: Pearson Education,
Inc, pp 299-365.
Teknomo, Kardi. Finding Eigen Value of Symmetric matrix Using Microsoft Excel. http:\\people.revoledu.com\kardi\
tutorial\Excel\EigenValue.html

Authors: (October 19, 2006) Tommy DiRaimondo, Rob Carr, Marc Palmer, Matt Pickvet
Stewards: (October 22, 2007) Shoko Asei, Brian Byers, Alexander Eng, Nicholas James, Jeffrey Leto
10.3: Eigenvalues and Eigenvectors is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source content
10.4: USING EIGENVALUES AND EIGENVECTORS TO FIND STABILITY AND
SOLVE ODES
Eigenvalues and eigenvectors are very useful in the modeling of chemical processes. When designing the controls for a process it is
necessary to create a program to operate these controls. Differential equations are used in these programs to operate the controls based on
variables in the system. These equations can either be solved by hand or by using a computer program. The solutions for these differential
equations will determine the stability of the system. After finding this stability, you can show whether the system will be stable and damped,
unstable and undamped (so that there is constant fluctuation in the system), or as an unstable system in which the amplitude of the
fluctuation is always increasing. For the first case, a stable and damped system, if there is a change, the system will adjust itself properly to
return to steady state. For the other two cases, the system will not be able to return to steady state. For the undamped situation, the constant
fluctuation will be hard on the system and can lead to equipment failure. The final situation, with the ever increasing amplitude of the
fluctuations will lead to a catastrophic failure.
There are a couple ways to develop the differential equation used to determine stability. First, you can create a differential equation to guide
the system where the variables are the readings from the sensors in the system. A second method would be using actual data found from
running the system. You could fit a differential equation to this data and use that equation for stability determination.
In this section on Eigenvalue Stability, we will first show how to use eigenvalues to solve a system of linear ODEs. Next, we will use the
eigenvalues to show us the stability of the system. After that, another method of determining stability, the Routh stability test, will be
introduced. For the Routh stability test, calculating the eigenvalues is unnecessary which is a benefit since sometimes that is difficult.
Finally, the advantages and disadvantages of using eigenvalues to evaluate a system's stability will be discussed.
10.4.2: SOLVING ODES

Eigenvalues and eigenvectors can be used as a method for solving linear systems of ordinary differential equations (ODEs). The method is
rather straight-forward and not too tedious for smaller systems. See The Eigenvector Eigenvalue Method for solving systems by hand and
Linearizing ODEs for a linear algebra/Jacobian matrix review. When trying to solve large systems of ODEs however, it is usually best to use
some sort of mathematical computer program. Mathematica is a program that can be used to solve systems of ordinary differential equations
when doing them by hand is simply too tedious. Once one overcomes the syntax of Mathematica, solving enormous systems of ordinary
linear differential equations becomes a piece of cake!
10.4.2.1: USING EIGENVALUES TO SOLVE A SYSTEM

A linear system will be solve by hand and using Eigenvalues[ ] expression in Mathematica simultaneously. Note that, in the Mathematica
inputs below, "In[]:=" is not literally typed into the program, only what is after it. The syntax needed to be typed is the line following
"In[]=" . The term is used here to more accurately demonstrate coding in Mathematica. To find a general solution of the linear system of
ordinary differential equation:
dx
= 4x + 8y
dt
dy
= 10x + 2y
dt
We first put the system in matrix form:

dx
dt 4 8 x
A = [ ] = [ ][ ]
dy
10 2 y
dt
Where we can see that

4 8
A = [ ]
10 2
In mathematica, we can use the following code to represent A:
In[1]:= MatrixForm [{{4,8},{10,2}}]
Out[1]:=
The eigenvalues λ1 and λ2, are found using the characteristic equation of the matrix A, det(A- λI)=0.
Therefore, λ1 = 12 and λ2 = − 6
We can use Mathematica to find the eigenvalues using the following code:
In[2]:= Eigenvalues[{{4,8},{10,2}}]
Out[2]:={12,-6}
Now, for each eigenvalue (λ1=12 and λ2=-6), an eigenvector associated with it can be found using , where is an
eigenvector such that
i) For λ1=12
This will lead to the equations (1) &(2):

(1)
(2)
The Mathematica input is:
In[3]:= eqn1= -8x+8y==0
In[4]:= eqn2= 10x-10y==0
In[5]:= Solve[{eqn1,eqn2},{x,y}]
Out[5]:=
Equations (1) & (2) lead to the solution
ii) For λ2=-6,
This will lead to the equations (3) & (4):

(3)
(4)
The Mathematica input is:
In[6]:= eqn3= 10x+8y==0
In[7]:= eqn4= 10x+8y==0
In[8]:= Solve[{eqn3,eqn4},{x,y}]
Out[8]:=
Equations (3) & (4) lead to the solution .
Recall that the direction of a vector such as is the same as the vector or any other scalar multiple. Therefore, to get the eigenvector,
we are free to choose for either the value x or y.
i) For λ1 = 12
We have arrived at y = x. As mentioned earlier, we have a degree of freedom to choose for either x or y. Let’s assume that x=1. Then, y=1
and the eigenvector associated with the eigenvalue λ1 is

ii) For λ2 = − 6
We have arrived at . Let’s assume that x = 4. Then, y = -5 and the eigenvector associated with the eigenvalue λ2 is .
These two eigenvalues and associated eigenvectors yield the solution:
x(t) 1 4
12t −6t
[ ] = c1 [ ]e + c2 [ ]e
y(t) 1 −5
Hence a general solution of the linear system in scalar form is:

12t −6t
lx(t) = c1 e + c2 4e
12t −6t
y(t) = c1 e − c2 5e
10.4.2.2: SOLVING A SYSTEM USING DSOLVE

Using the same linear system of ordinary differential equations:
dx
= 4x + 8y
dt
dy
= 10x + 2y
dt
We input the differential equations to Mathematica with the following command:

In:= ODEs={x'[t]==4x[t]+8y[t],y'[t]==10x[t]+2y[t]}
where x'[t] = and y'[t] =

After entering the equations, we use the DSolve function:
In:= DSolve[ODEs, {x[t], y[t]}, t]
Mathematica will output the solution:
Out:=
This set of equations, although looks more complicated than the first one, is actually the same.
10.4.3: STABILITY
Eigenvalues can be used to determine whether a fixed point (also known as an equilibrium point) is stable or unstable. A stable fixed point
is such that a system can be initially disturbed around its fixed point yet eventually return to its original location and remain there. A fixed
point is unstable if it is not stable. To illustrate this concept, imagine a round ball in between two hills. If left alone, the ball will not move,
and thus its position is considered a fixed point. If we were to disturb the ball by pushing it a little bit up the hill, the ball will roll back to its
original position in between the two hills. This is a stable fixed point. Now image that the ball is at the peak of one of the hills. If left
undisturbed, the ball will still remain at the peak, so this is also considered a fixed point. However, a disturbance in any direction will cause
the ball to roll away from the top of the hill. The top of the hill is considered an unstable fixed point.
The eigenvalues of a system linearized around a fixed point can determine the stability behavior of a system around the fixed point. The
particular stability behavior depends upon the existence of real and imaginary components of the eigenvalues, along with the signs of the
real components and the distinctness of their values. We will examine each of the possible cases below.
10.4.3.1: IMAGINARY (OR COMPLEX) EIGENVALUES
−
−−
When eigenvalues are of the form a + bi , where a and b are real scalars and i is the imaginary number √−1, there are three important
cases. These three cases are when the real part is positive, negative, and zero. In all cases, when the complex part of an eigenvalue is non-
zero, the system will be oscillatory.
10.4.3.1.1: POSITIVE REAL PART
When the real part is positive, the system is unstable and behaves as an unstable oscillator. This can be visualized as a vector tracing a spiral
away from the fixed point. The plot of response with time of this situation would look sinusoidal with ever-increasing amplitude, as shown
below.
This situation is usually undesirable when attempting to control a process or unit. If there is a change in the process, arising from the process
itself or from an external disturbance, the system itself will not go back to steady state.
10.4.3.1.2: ZERO REAL PART

When the real part is zero, the system behaves as an undamped oscillator. This can be visualized in two dimensions as a vector tracing a
circle around a point. The plot of response with time would look sinusoidal. The figures below should help in understanding.
Undamped oscillation is common in many control schemes arising out of competing controllers and other factors. Even so, this is usually
undesirable and is considered an unstable process since the system will not go back to steady state following a disturbance.
10.4.3.1.3: NEGATIVE REAL PART

When the real part is negative, then the system is stable and behaves as a damped oscillator. This can be visualized as a vector tracing a
spiral toward the fixed point. The plot of response with time of this situation would look sinusoidal with ever-decreasing amplitude, as
shown below.
This situation is what is generally desired when attempting to control a process or unit. This system is stable since steady state will be
reached even after a disturbance to the system. The oscillation will quickly bring the system back to the setpoint, but will over shoot, so if
overshooting is a large concern, increased damping would be needed.
While discussing complex eigenvalues with negative real parts, it is important to point out that having all negative real parts of eigenvalues
is a necessary and sufficient condition of a stable system.
10.4.3.1.4: COMPLEX PART OF EIGENVALUES
As previously noted, the stability of oscillating systems (i.e. systems with complex eigenvalues) can be determined entirely by examination
of the real part. Although the sign of the complex part of the eigenvalue may cause a phase shift of the oscillation, the stability is unaffected.
10.4.3.2: REAL EIGENVALUES

We've seen how to analyze eigenvalues that are complex in form, now we will look at eigenvalues with only real parts.
10.4.3.2.1: ZERO EIGENVALUES
If an eigenvalue has no imaginary part and is equal to zero, the system will be unstable, since, as mentioned earlier, a system will not be
stable if its eigenvalues have any non-negative real parts. This is just a trivial case of the complex eigenvalue that has a zero part.
10.4.3.2.2: POSITIVE EIGENVALUES
When all eigenvalues are real, positive, and distinct, the system is unstable. On a gradient field, a spot on the field with multiple vectors
circularly surrounding and pointing out of the same spot (a node) signifies all positive eigenvalues. This is called a source node.
Graphically, real and positive eigenvalues will show a typical exponential plot when graphed against time.
10.4.3.2.3: NEGATIVE EIGENVALUES

When all eigenvalues are real, negative, and distinct, the system is unstable. Graphically on a gradient field, there will be a node with
vectors pointing toward the fixed point. This is called a sink node.
Graphically, real and negative eigenvalues will output an inverse exponential plot.
10.4.3.2.4: POSITIVE AND NEGATIVE EIGENVALUES
If the set of eigenvalues for the system has both positive and negative eigenvalues, the fixed point is an unstable saddle point. A saddle point
is a point where a series of minimum and maximum points converge at one area in a gradient field, without hitting the point. It is called a
saddle point because in 3 dimensional surface plot the function looks like a saddle.
10.4.3.3: REPEATED EIGENVALUES

If the set of eigenvalues for the system has repeated real eigenvalues, then the stability of the critical point depends on whether the
eigenvectors associated with the eigenvalues are linearly independent, or orthogonal. This is the case of degeneracy, where more than one
eigenvector is associated with an eigenvalue. In general, the determination of the system's behavior requires further analysis. For the case of
a fixed point having only two eigenvalues, however, we can provide the following two possible cases. If the two repeated eigenvalues are
positive, then the fixed point is an unstable source. If the two repeated eigenvalues are negative, then the fixed point is a stable sink.
10.4.3.4: SUMMARY OF EIGENVALUE GRAPHS

Below is a table summarizing the visual representations of stability that the eigenvalues represent.
Note that the graphs from Peter Woolf's lecture from Fall'08 titled Dynamic Systems Analysis II: Evaluation Stability, Eigenvalues were
used in this table.
10.4.4: ANOTHER METHOD OF DETERMINING STABILITY

The process of finding eigenvalues for a system of linear equations can become rather tedious at times and to remedy this, a British
mathematician named Edward Routh came up with a handy little short-cut.
First, recall that an unstable eigenvalue will have a positive or zero real part and that a stable eigenvalue will have a negative real part.
The first test is to take an n-th degree polynomial of interest:
n n−1
P (λ) = a0 λ + a1 λ + ⋯ + an−1 λ + an
and look to see if any of the coefficients are negative or zero. If so, there is at least one value with a positive or zero real part which refers to
an unstable node.
The way to test exactly how many roots will have positive or zero real parts is by performing the complete Routh array. Referring to the
previous polynomial, it works as follows:
Row
1
2
3
4
n-1
n
n+1
An array of n+1 rows and the coefficients placed as above. After the first two rows, the values are obtained as below:
a1 a2 − a0 a3 a1 a4 − a0 a5 a1 a6 − a0 a7 b1 a3 − a1 b2 b1 a5 − a1 b3
b1 = , b2 = , b3 = , ⋯ c1 = , c2 = , c3
a1 a1 a1 b1 b1
b1 a7 − a1 b4
= ,⋯
b1
Routh’s theorem says:
 ROUTH’S THEOREM
1. For all of the roots of the polynomial to be stable, all the values in the first column of the Routh array must be positive.
2. If any of the values in the first column are negative, then the number of roots with a positive real part equals the number of sign
changes in the first column.
So considering the following example,

4 3
f(x) = 9x + 14x + 7x + 10
Preliminary test: All of the coefficients are positive, however, there is a zero coefficient for x2 so there should be at least one point with a
negative or zero real part.
Routh array:
Row
1 ;9 0 10
2 ; 14 7
3
4
5
Since Row 3 has a negative value, there is a sign change from Row 2 to Row 3 and again from Row 3 to Row 4. Thus, there are 2 roots with
positive or zero real part.
10.4.5: STABILITY SUMMARY
The following image can work as a quick reference to remind yourself of what vector field will result depending on the eigenvalue
calculated.
The table below gives a complete overview of the stability corresponding to each type of eigenvalue.
Eigenvalue Type Stability Oscillatory Behavior Notation
All Real and + Unstable None Unstable Node
All Real and - Stable None Stable Node
Mixed + & - Real Unstable None Unstable saddle point
+a + bi Unstable Undamped Unstable spiral
-a + bi Stable Damped Stable spriral
0 + bi Unstable Undamped Circle
Repeated values Depends on orthogonality of eigenvectors
10.4.6: ADVANTAGES AND DISADVANTAGES OF EIGENVALUE STABILITY

There are several advantages of using eigenvalues to establish the stability of a process compared to trying to simulate the system and
observe the results. However, there are situations where eigenvalue stability can break down for some models.
1. High accuracy for linear systems.
2. General method that can be applied to a variety of processes.
3. Can be used even if all variables are not defined, such as control parameters.
1. Only applicable for linear models.
2. Linear approximations of nonlinear models break down away from the fixed point of approximation.
 EXAMPLE 10.4.1
For the system described by:

dx
= y
dt
dy
= 2x + y
dt
Find the fixed points and determine their stability.

Solution
Step 1. Determining the fixed points
At the fixed points, nothing is changing with respect to time. Therefore, set the derivatives to zero to find the fixed points.
0 = y
0 = 2x + y
Solving these two equations simultaneously, we see that we have one fixed point at {0,0}
Step 2. Determine the eigenvalue of this fixed point
First, let us rewrite the system of differentials in matrix form.
dx
dt 0 1 x
[ ] = [ ][ ]
dy
2 1 y
dt
Next, find the eigenvalues by setting det(A − λI) = 0
Using the quadratic formula, we find that and

Step 3. Determine the stability based on the sign of the eigenvalue
The eigenvalues we found were both real numbers. One has a positive value, and one has a negative value. Therefore, the point {0, 0}
is an unstable saddle node.
The stability can be observed in the image below. The fixed point is seen at (0,0). All solutions that do not start at (0,0) will travel away
from this unstable saddle point. The solution was found by using the two-dimensional system in PPlane 2005.10 PPlane.
 EXAMPLE 10.4.2
Determine the Routh array and the number of positive or zero roots of the following equation.
5 4 3 2
f(x) = 6x + 12x + 5x + 3x + 17x
Solution
Routh Array:
Row
1 ;6 5 17
2 ; 12 3 0
3
4
5
6
Since you go from a positive value in row three, to a negative value in row four, and back to a positive value in row five, you will have
a positive or zero real part for two of your roots.
 EXAMPLE 10.4.1
Use Mathematica to find the eigenvalues of the system defined by:

dx1
= 8x1 + 15x2 − 3x3 + 7x4 + 2x5
dt
dx2
= −22x1 − 21x2 + 3x3 − 12x4 + 11x5
dt
dx3
= 10x1 + 6x2 + 24x3 + 3x4 − 6x5
dt
dx4
= 0x1 − 2x2 + 21x3 + 0x4 + 4x5
dt
dx5
= 4x1 + 9x2 + x3 − 22x4 − 7x5
dt
And comment on the stability of this system

Solution
The matrix that corresponds with this system is the square matrix:
8 15 −3 7 2
⎡ ⎤
⎢ −22 −21 3 −12 11 ⎥

⎢ ⎥
⎢ 10 6 24 3 −6 ⎥
⎢ ⎥
⎢ ⎥
⎢ 0 −2 21 0 4 ⎥
⎣ ⎦
4 9 1 −22 −7
Using the Eigenvalues[ ] function in Mathematica the input is:

In[1]:= Eigenvalues[
]
In[2]:= N[%] This step produces numerical results
The output contains the 5 eigenvalues:
out[2]:= {27.0612, -10.7653 + 10.0084 , -10.7653 - 10.0084 , -0.765272 + 7.71127 , -0.765272 - 7.71127 }
So the five eigenvalues are:
27.0612
-10.7653 + 10.0084
-10.7653 - 10.0084
-0.765272 + 7.71127
-0.765272 - 7.71127
Looking at these eigenvalues it is clear that the system as a whole is unstable. This is because one of the eigenvalues has a positive real
part.

A system is stable if and only if all of the system's eigenvalues:
1. a. are real
2. b. are complex
3. c. have negative real parts
4. d. have negative imaginary parts
Answer: c
What would the following set of eigenvalues predict for the system's behavior?
−4, −2 + 3i, −2 − 3i, 3
a. An unstable oscillation
b. A damped oscillation
c. An undamped oscillation
d. A source
e. A saddle point
Answer: d
10.4.9: REFERENCES
Kravaris, Costas. Chemical Process Control: A Time Domain Approach.
10.4.10: CONTRIBUTORS
Daniel Katzman, Jessica Moreno, Jason Noelanders, and Mark Winston-Galant
10.4: Using Eigenvalues and Eigenvectors to Find Stability and Solve ODEs is shared under a CC BY 3.0 license and was authored, remixed, and/or
10.5: PHASE PLANE ANALYSIS - ATTRACTORS, SPIRALS, AND LIMIT CYCLES
10.5.1: INTRODUCTION TO ATTRACTORS, SPIRALS AND LIMIT CYCLES
We often use differential equations to model a dynamic system such as a valve opening or tank filling. Without a driving force, dynamic
systems would stop moving. At the same time dissipative forces such as internal friction and thermodynamic losses are taking away from
the driving force. Together the opposing forces cancel any interruptions or initial conditions and cause the system to settle into typical
behavior. Attractors are the location that the dynamic system is drawn to in its typical behavior. Attractors can be fixed points, limit cycles,
spirals or other geometrical sets.
Limit cycles are much like sources or sinks, except they are closed trajectories rather than points. Once a trajectory is caught in a limit
cycle, it will continue to follow that cycle. By definition, at least one trajectory spirals into the limit cycle as time approaches either positive
or negative infinity. Like a sink, attractive (stable) limit cycles have the neighboring trajectories approaching the limit cycle as time
approaches positive infinity. Like a source, non-attractive (unstable) limit cycles have the neighboring trajectories approaching the limit
cycle as time approaches negative infinity. Below is an illustration of a limit cycle [1].
Spirals are a similar concept. The attractor is a spiral if it has complex eigenvalues. If the real portion of the complex eigenvalue is positive
(i.e. 3 + 2i), the attractor is unstable and the system will move away from steady-state operation given a disturbance. If the real portion of
the eigenvalue is negative (i.e. -2 + 5i), the attractor is stable and will return to steady-state operation given a disturbance.
Given the following set of linear equations we will walk through an example that produces a spiral:
dx
= 2x + 5y
dt
dy
= −5x + 2y
dt
The Jacobian matrix would be the coefficients:

∣ 2 5∣
A = ∣ ∣
∣ −5 2∣
Next we found the eigenvalues:

∣ 2 5∣ ∣1 0∣ ∣ (2 − λ) 5 ∣
(A − λI) = ∣ ∣− λ∣ ∣ = ∣ ∣
∣ −5 2∣ ∣0 1∣ ∣ −5 (2 − λ) ∣
where I is the identity matrix

∣1 0∣
I = ∣ ∣
∣0 1∣
2
det(A − λI) = (2 − λ) + 25 = 0
Eigenvalues:
λ = 2 ± 5i
The system is unstable because the real portion of the complex eigenvalues is positive.
To find the first eigenvector we continue by plugging in 2 − 5i :
∣ (2 − λ) 5 ∣ ∣ 2 − (2 − 5i) 5 ∣ ∣ 5i 5 ∣
∣ ∣ = ∣ ∣ = ∣ ∣
∣ −5 (2 − λ) ∣ ∣ −5 2 − (2 − 5i) ∣ ∣ −5 5i ∣
∣ 5i 5 ∣
(A − λI)v = ∣ ∣v = 0
∣ −5 5i ∣
let
∣x∣
v = ∣ ∣
∣y ∣
∣ 5i 5 ∣∣x 0∣
∣ ∣∣ | = | ∣
∣ −5 5i ∣ ∣ y 0∣
We now have a system of equations which we can solve for x, y:

5ix + 5y = 0
−5x + 5iy = 0
Dividing both equations by 5:

ix + y = 0
−x + iy = 0
Solution
∣ −1 ∣
v1 = ∣ ∣
∣ i ∣
Following the same procedure using the second eigenvalue of 2 + 5i, we find the second eigenvector to be:
∣ i ∣
v2 = ∣ ∣
∣ −1 ∣
Now plugging both eigenvalues and eigenvectors into the characteristic equation:
2t
x(t) = e (C1 cos 5t + C2 sin 5t)
2t
y(t) = e (C3 cos 5t + C4 sin 5t)
For more on this procedure, see: Eigenvalues and Eigenvectors

The phase-plane plot is shown below:
10.5.2: INTRODUCTION TO PPLANE

Phase-plane analysis is an important tool in studying the behavior of nonlinear systems since there is often no analytical solution for a
nonlinear system model.
PPlane is a JAVA applet for phase plane analysis of two-dimensional systems. It starts in your web browser and you can directly input your
equations and parameter values. PPLANE plots vector fields for systems of differential equations. At each point, (x,y), of a grid, PPLANE
draws an arrow indicating the direction and magnitude of the vector (x',y'). This vector equals dy/dt / dx/dt = dy/dx, and is independent of t;
therefore, it must be tangent to any solution curve through (x,y).
It allows the user to plot solution curves in the phase plane by simple clicking on them. It also enables the user to plot these solutions in a
variety of plots. There are a number of advanced features, including finding equilibrium points, eigenvalues and nullclines, that you will
find useful later.
10.5.2.1: HOW TO USE PPLANE
In the PPlane equation window you can enter a system of differential equations of the form dx/dt = f(x, y) and dy/dt = g(x, y) , define
parameters and resize the display window. Under the Gallery pull down from the menu, you can switch to a linear system.
Note, if your differential equations contain constant parameters, you can enter them in the "Parameter Expressions" boxes below the
differential equations as seen in the figure below (A, B, and C are used as example parameters). This is a convenient feature to use when
considering the effect of changed parameters on the steady state of a system because it eliminates the redundancy of re-entering the
parameter values multiple times within the differential equations.
In the PPlane Phase Plane window below you will see the vector fields for the system. By clicking on the field you will plot solution curves
in the phase plane. If you are interested in a plot of your solution vs. time or a 3-D view, click on graph:
If you choose the x-t and y-t option, you have to pick a specific solution curve. The result will look like this:
Use the crop function to zoom in on a point of interest

Choose Find Equillibrium Point under the Solution pull down menu. Then when you click on an orbit in the phaseplane, the Pplane
Messages window will display the eigenvalues and possible equillibrium points.
Additional Things you can Change in PPlane
Changing the Slope Field
By clicking on the "Options" tab, then by selecting "Direction Field Settings," you can change the number of rows and columns plotted, the
way the field is made up, as well as the computational settings of PPlane.
Erasing Made Orbits
On the "Edit" tab, there are options that say "Delete Orbit" or "Delete All Orbits". These options act as their names imply.
Changing the Direction of Graphing
By clicking on the "Options" tab, then by selecting the "Solution Direction" option, you can then change the way that PPlane graphs a line
when you click on the field. You can change the graphing to plot forwards (for values of t>0), backwards (for values of t<0) or in both
directions of t.
10.5.2.2: MORE USES FOR PPLANE

Of the many uses that PPLANE has to offer, some of the most helpful functions involve:
Finding eigenvalues/eigenvectors for an equilibrium point.
After graphing a series of differential equations, find a particular equilibrium point within the graphed data (“Find an Equilibrium Point,”
under the Solution tab)
Now, by selecting a point on the field that has ben graphed by pplane, pplane will find the closest equilibrium point on the graph, and
highlight this point on the graph in red.
The PPLANE Messages box in the upper left hand corner of the screen should pop up with some new information. This information
provides eigenvalues and the corresponding eigenvectors to the selected equilibrium value:
Stability of a Equilibrium Point

Similarly to before, the Messages Box will provide the stability features (ie: is it a nodal sink?) of the found equilibrium point, immediately
after using the “Find an Equilibrium Point.”
10.5.2.3: OTHER CONCEPTS OF PHASE PLANE ANALYSIS
10.5.2.4: SEPARATRIX
A separatrix is any line in the phase-plane that is not crossed by any trajectory. The unstable equilibrium point, or saddle point, below
illustrates the idea of a separatrix, as neither the x or y axis is crossed by a trajectory. If you picture a topographic map, the seperatrix would
be a mountain ridge; if you fall a little of the edge, you will never come back. Plotting your phase plane in Pplane would be useful to
identify impossible set points, for example.
10.5.2.4.1: NULLCLINES
A nullcline is a curve where x'=0 or y'=0, thus indicating where the phase plane is completely horizontal or completely vertical. The point at
which two nullclines intersect is an equilibrium point. Nullclines can also be quite useful for visualization of a phase plane diagram as they
split the phase plane into regions of similar flow. To display nullclines on the Phase Plane window, select Nullclines under the Solutions
drop down menu. The screenshot below is an example.
Notice that the red nullcline shows where the flow is completely vertical (x'=0) and the yellow nullcline shows where the flow is completely
horizontal (y'=0).
10.5.2.4.2: LIMIT CYCLE
Below you will find a solution curve for a limit cycle. The limit cycle contains the response in a set range, which is something you may
want to take advantage for certain engineering applications. On the other hand it is always rotating and may not be stable enough for your
purposes.
10.5.2.5: TAKING SCREEN SHOTS TO COPY PPLANE PHASE PORTRAITS
With the introduction of Windows Vista, the Snipping Tool was introduced. This tool allows much greater flexibility with taking screen
shots and editing them. This article will talk about the Snipping Tool as well as the Windows Print Screen key which can be used to take
photos of your computer screen. When pressing the key, your computer copies the image of your screen and onto your computer’s
clipboard. The image can then be pasted into multiple programs. There are many instances throughout the CHE 466 course in which taking
a screen shot of your work will come in handy. Examples include copying phase portraits created in Pplane, graphs created in Mathematica,
or your Mathematica code.
To enable the Snipping Tool on your Vista computer go to the Windows button in the bottom left of your screen and click Accessories ->
Snipping Tool.
Figure 1. How to enable the Snipping Tool
A window will appear asking if you would like to add the Snipping Tool to your Quicklaunch. This provides a simple and quick way to take
screenshots.
To take a picture of your graph, just press the Snipping Tool button in the Quicklaunch area and a window like this will appear:
Figure 2. The Snipping Tool Window
Automatically, the Snipping Tool will default to a crosshair from which you can click and drag to make a selection of the section of the
screen you would like represented by a red rectangle.
WARNING: In the Options section you should uncheck "Show selection ink after snips are captured" in order to eliminate the red edge
around your photos.
Figure 3. Snipping Tool Option Menu (Uncheck the selection ink)
The Snipping Tool will open up a new window with your selection and copy the image to your clipboard. Feel free to edit your image or
save it where it is convenient.
Figure 4. Snipping Tool Editing Window
If not using Windows Vista you can still use Print Screen:
Follow these simple steps to copy and paste your phase portrait into a Microsoft Word document:
1. Pull up the window containing your phase portrait so that it is displayed on the screen.
2. Find the Print Screen or PrtSc button in the upper-right hand portion of your keyboard. (The key may appear slightly different depending
on your Windows keyboard manufacturer).
3. Open Microsoft Word to the document of your choice (i.e. CHE 466 Homework 7).
4. Paste the image into the Word document. Figure 1 below indicates how your phase portrait will look in Word.
5. To crop or resize the image as you like, you may use the Picture toolbar (seen in Figure 2) by selecting View -> Toolbars -> Picture.
If you prefer to take a screen shot of just your phase portrait rather than the entire computer screen, follow these simple steps:
1. Pull up the window containing your phase portrait so that it is displayed on the screen.
2. Press Alt-Print Screen to capture a photo of the window you selected.
3. Open Microsoft Word to the document of your choice (i.e. CHE 466 Homework 7).
4. Paste the image into the Word document. Figure 3 below indicates how your phase portrait image will look.
Figure 5. Initial screen shot
Figure 6. Microsoft Word document containing the screen shot
Figure 7. Screen shot of phase portrait window
 EXAMPLE 10.5.1: LINEAR SYSTEM OF EQUATIONS
Use PPLANE to calculate the following information of the system given below: location and type of equilbrium point, Jacobian matrix,
eigenvectors, and eigenvalues.
′
x = y
′
y = 2x + y
Solution
Using PPLANE's "Find an Equlibrium Point" feature and clicking on the phase plane, the following equilibrium point will be indicated:
The location, and type of the equilibrium point is given in the "PPLANE Messages" window, along with the Jacobian matrix,
eignvectors and eigenvalues.
For our second example problem we would like you to try a non-linear system of equations.
 EXAMPLE $\PAGEINDEX{2$: NONLINEAR SYSTEM OF EQUATIONS
Solve for the set of equations on PPlane. Consider the trends of change in rate of the differential equations and subsequently solve the
equations on Mathematica to compare the trends. The following two differential equations are going to be used to walk through the
solutions on PPlane and Mathematica:
dx
$ = x − (5xy)
dt
dy
= −2x + 2y$
dt
Solution
The above equations were put in to the PPlane window and solved. The following two windows show the solution for the set of
differential equations:
Looking at the phase-plane plot, at low values of x and y, t increases slowly. However at higher values of y, the increase in t is rapid.
When x is high and y is low, however, t increases slowly. Mathematica will help us visualize the relative rates of change better.
The following is the code used in Mathematica to solve and plot the set of differential equations:
ODEs = {x'[t] == (x[t] - 5*x[t]*y[t]), y'[t] == (-2*x[t]) + (2*y[t]), x[0] == 9, y[0] == 370} numericalSol = NDSolve[ODEs, {x[t], y[t]},
{t, 1, 100}] Plot[y[t] /. numericalSol, {t, 1, 100}, PlotRange -> All]
The graphs obtained on Mathematica for x versus t and y versus t are shown below. Please note the difference between the axes scales.
Also, for a closer look, here are the plots of the slopes of graphs above.
Comparing the increase in t with increase in x, we see a similar trend in the PPlane diagram. Mathematica shows a constantly
increasing x' with t. At higher values of x, the value for t increases in the PPlane diagram. However, as seen clearly in the PPlane
diagram and the graphs of the slope of x with respect to t and the slope of y with respect to t, the slope of x does not compare to the
very large slope shown for y' versus t. Therefore the results using Mathematica and PPlane are consistant.
This modeling system could be used to view trends of variables in a CSTR or any other system which can be modeled using differential
equations.
10.5.3: MULTIPLE CHOICE QUESTIONS
10.5.3.1: QUESTION 1
Open PPLANE and enter the following equations into the PPLANE Equation Window:
x '= sin(x)
y' = cos(y)
What does the resulting phase plane look like? (Note: Click on image to enlarge)
A.
B.
C.
D.
10.5.3.2: QUESTION 2
If you have a disturbance in your system and the system is driven right back to equilibrium, that fixed point's eigenvalue is most likely a:
A. complex number with negative real number component
B. 0
C. negative real number
D. positive real number
10.5.4: ANSWERS TO THE MULTIPLE CHOICE QUESTIONS

Question 1: C
Question 2: A
Authors: Erin Knight, Dipti Sawalka, Matt Russell, Spencer Yendell
Stewards: Eric Black, Megan Boekeloo, Daniel Carter, Stacy Young
10.5: Phase Plane Analysis - Attractors, Spirals, and Limit cycles is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter
request.
10.6: ROOT LOCUS PLOTS - EFFECT OF TUNING
Root locus plots show the roots of the systems characteristic equation, (i.e. the Laplacian), as a function of the control variables such as K c.
By examining these graphs it is possible to determine the stability of different values of the control variable. A typical transfer function is of
the form G(s) = Y (s)/U (s).
Poles: U(s) = 0
Zeros: Y(s) = 0
In other words, after factorization the poles are the roots of the denominators and the zeros are the roots of the numerator. Stability only
depends on the value of the poles. The system is stable for all values of the control variables that result in the value of the real part of every
pole value being less than zero.The lines of a Root locus plot display the poles for values of the control variable(s) from zero to infinity on
complex coordinate system. These plots will always have a line of symmetry at i = 0 .
10.6.1.1: CLOSED-LOOP VS. OPEN-LOOP

A closed-loop system uses feedback control where the output has an effect on the input. With a closed-loop, oscillations are usually
introduced, and therefore can become unstable. Unlike the open-loop systems, the closed-loop incorporates valves and controllers.
In an open-loop system, the output is not compared with and has no effect on the input. In an open-loop system, oscillations are not
introduced and therefore cannot become unstable. An open-loop system, however, can be inaccurate because it does not take into account
the control dynamics. Open-loop systems will include feed forward control schemes or timed control schemes. The two diagrams below
depict this difference when trying to control the temperature in your apartment. (Click on images for increased resolution).
In this diagram a thermostat presents feedback to the heater to turn it on or off.
Since the heater only turns on at 10 pm, it assumes that the amount of time you open the door will be the same each day. It is considered an
open system since the temperature in the room is independent of the heater controller.
Note that all the examples presented in this web page discuss closed-loop systems because they include all systems with feedback.
10.6.1.2: COMPLEX COORDINATE SYSTEMS
Root locus plots are a plot of the roots of a characteristic equation on a complex coordinate system. A complex coordinate system allows the
plotting of a complex number with both real and imaginary parts. The real component is plotted on the x-axis and the imaginary component
is plotted on the y-axis. When creating root locus plots imaginary roots must be solved for. These imaginary roots come in complex
conjugate pairs (this can be seen below in the section on "Plotting Poles on a Complex Coordinate System to make Root Locus Plot").
For example a plot of the following complex numbers is shown below. These complex numbers can be broken into the real and imaginary
components to make it easier to plot.
10.6.1.3: DEVELOPING A CHARACTERISTIC EQUATION
Although the focus of this article is to discuss root locus plots, it is necessary to mention briefly how to determine the characteristic equation
for a system in order to obtain the root locus plot. In general, most chemical engineering processes can be described by a system of ordinary
differential equations. Follow the following steps to determine the characteristic equation for the system (which will allow you to develop a
root locus plot).
1. If the ODEs are not linear, linearize them. (See this link for more information on linearizing ODEs
<controls.engin.umich.edu/wiki/index.php/LinearizingODEs>)
2. After linearizing the ODEs, use matrix algebra to find the eigenvalues of your system. Be careful here not to insert numerical values for
your control parameters, (e.g. leave Kc as Kc, not Kc=1). (For more information on finding eigenvalues
<controls.engin.umich.edu/wiki/index.php/EigenvaluesEigenvectors>)
3. The polynomial equation obtained for the eigenvalue should contain lambda and the control parameters. This equation is the
characteristic equation. Obtain solutions to this equation by setting values for the control parameters and solving for the eigenvalues.
The roots obtained will be used to create the root locus diagram.
This 3-step process is valid to obtain a characteristic equation for any closed loop control system. A more traditional method to develop
characteristic equations is by applying Laplace transforms.
10.6.2: LAPLACE TRANSFORMS

Laplace transforms are a method to change linear ordinary differential equations into transfer function. All transfer functions used in root
locus plots are independent of time because the $L[f(t)] ≣ F(s)$. Formally, the equation below shows that the time function is integrated,
leaving only the variable s .
∞
−st
F(s) = ∫ f(t)e dt
0
s is a complex number, therefore allowing us to construct complex coordinate system graphs. The exact solution to most disturbances and
controllers can be found in any controls book. Please always refer to your Ordinary differential equations mathematics book for more
information on Laplace transforms.
 EXAMPLE 10.6.1
The stability of the series chemical reactors is to be determined. The reactors are well mixed and isothermal, and the reaction is first-
order in component A. The outlet concentration of the second reactor is controlled with a PI feedback algorithm that manipulates the
flow of the reactant, which is very much smaller than the flow of the solvent. The sensor and final element are assumed fast, and
process data is as follows.
Process
V= 5m3
Fs=5m3/min >> FA
k = 1 min-1
vs= 50% open
CA0 = 20 mole/m3
CA0(s)/v(s) = Kv= 0.40 (mole/ m3) (%open)
PI Controller Kc = ?? Tf = 1 min
Formulation The transfer function for the process and controller are
Ky
Gy (s) =
(τs + 1)(τs + 1)
1
Gc (s) = Kc (1 + )
Tf s
3
F mole/m
Ky = Kv ( ) = 0.1
F +Vk %
V
τ = = 0.5min
F +VR
The individual transfer functions can be combined to give the closed-loop transfer function for a set point change, which includes the
characteristic equation. (where CV= Control variable & SP = set-point signal)
1 0.1
Kp (1 + ) 2
CV (s) Gp (s)Gv (s)Gc (s) s
(0.5s+1)
= =
1 0.1
S P (s) 1 + Gr (s)Gv (s)Gc (s)Gs (s) 1 + Kp (1 + )
s 2
(0.5s+1)
Characteristic equation
1 0.1
0 = 1 + Kp (1 + )
2
s (0.5s + 1)
10.6.3: ROOT LOCUS DIAGRAMS

Root Locus plots are a method of evaluating the behavior of a control system. The creation of a root locus plot begins by determining the
poles of the control system for a given set of control parameters. These poles are then plotted on a complex coordinate system as seen in the
previous section and analyzed to determine the behavior of the system.
10.6.3.1: DETERMINING THE POLES OF A CONTROL SYSTEM

The “poles” of a system are the roots of the demoninator of the transfer function. In other words the poles are the values of "s" when the
transfer function go to infinity (when the demoninator equals zero). For a system of differential equations, poles are the eigenvalues of the
equation system. Consider the following solution to a system of differential equations:
3 2
f(s) = 48s + 44s + 12s + 1 (10.6.1)
This equation is a third order polynomial, therefore it will have three poles (be aware that some of these poles may be imaginary numbers).
These poles can be obtained by factoring the expression or using a computer program such as Maple. The three poles, or roots, of this
equation are s = -0.167, -0.25, -0.5.
For a system of differential equations, finding the eigenvalues can be time consuming and the use of Matlab, Maple, or Mathematica is
more efficient. One method would be to use the Mathematica eigenvalues[] function to solve the system for you. This function is explained
in Eigenvalues and Eigenvectors section of the wiki text.
10.6.3.2: PLOTTING POLES ON A COMPLEX COORDINATE SYSTEM TO MAKE ROOT LOCUS PLOT
A root locus plot is created by plotting the resulting poles on a complex coordinate system. For system of P-only control, the governing
differential equations will depend on the proportional gain, Kc. Consider the following solution to a system of differential equations:
3 2
f(s) = 48s + 44s + 12s + 1 + 6K c (10.6.2)
Notice that Kc is a term in this equation. Therefore, there exists a set of poles for each value of Kc. If Kc = 0, then equation #2 reduces to
equation #1, and poles are as listed above. Table 1 lists the three poles of the system for given values of Kc. These poles were calculated
using a computer algebra system because it is impractical to try and evaluate these functions with analytical methods.
Table 1- Poles of Characteristic Equation
To create a root locus plot, each pole is broken down into its real (x-axis) and imaginary (y-axis) component (Table 2):
Table 2- Real and Imaginary Components of each Pole
Figure 1- Root Locus Diagram of the Characteristic Equation

By convention, red arrows are drawn on the plot in the direction of increasing Kc values. They help to illustrate how the roots of the system
vary by changing the Kc value. Also by convention, the points in which Kc = 0 are represented with ‘x’ marks instead of dots.
10.6.3.3: INTERPRETING A ROOT LOCUS DIAGRAM
The primary use of a Root Locus Diagram is to evaluate how differing values of Kc affect the stability and behavior of a control system.
The stability of the control system depends on the sign of the real component of the pole. If the real components of all poles are negative,
then the system is said to be stable for that value of Kc. If the real component of the pole is positive, the system is unstable for that value of
Kc, meaning the output signal will diverge from the set point.
The behavior of the control system depends on the presence of an imaginary component of the pole. If any of the three poles contains an
imaginary number component, then that value of Kc will cause the output signal to oscillate. If all of the poles are real (contain no
imaginary components), the output signal will not oscillate at that Kc value.
Please refer back to the Root Locus diagram in Figure 1. The system becomes unstable (i.e. the real component of the poles becomes
positive) for Kc > 1.67 and Kc < 0. Within the range of stability, no oscillations (i.e. no imaginary component) are observed when 0 < Kc <
0.004.
10.6.4: ROOT LOCUS DIAGRAMS FOR PID CONTROL

Root Locus diagrams are much more difficult to create for PID control. The characteristic equation will contain unknown variables Kc, Ti,
and Td. Therefore, each point on the Root Locus Diagram will represent a set of tuning parameters. In order to show the progression as each
tuning parameter changes, the resulting diagram will be a three-dimensional surface plot. Due to the complexity of this diagram, we will not
create one, as it is beyond the scope of this text. However, stability analysis can still be applied to the characteristic equation. Take for
example Equation 10.6.3 :
3 2
48s + 44s + 12s + 1 + 6K c
f(s) = (10.6.3)
T i + 3T d
In this theoretical instance, equation 3 is the characteristic equation governing the same system mentioned above, only now with PID
control. Notice the additional presence of the Ti and Td terms. Under P- only control, this system was stable for Kc values between 0 and
1.67. Will the same be true if this system were tuned with PID control?
Let’s suppose we want to test the conditions Kc = 1.0, Ti = 0.3 and Td = 0.1. These values were randomly chosen. Using a computer
software package, the three roots were determined to be: r1 = -0.827, r2 = -0.044 + 0.417i, r3 = -0.044 – 0.417i. Therefore, since the real
components are all negative, the system is still stable. Since complex roots are present, the response is expected to oscillate around the set
point.
10.6.5: CREATING ROOT LOCUS PLOTS WITH MATHEMATICA

Mathematica allows you to develop root locus plots for polynomials since the math involved in solving for the solutions can become very
tedious. Before you are able to obtain the root locus plot, you need to solve for the roots. We will use the same equation as was used in the
“Determining the Poles of a Control System” section.
3 2
f(s) = 48s + 44s + 12s + 1 + 6Kc
The Solve[ ] function can be used to determine the roots, both real and imaginary, for each corresponding Kc value. The syntax in
Mathematica is shown below for the specified equation when Kc = 0. It is important to note that in order to obtain an output you must press
“Shift Enter.”
The following is just to show how Mathematica formats the output when Kc ≠ 0 and when there are imaginary roots. In this case,
K c = −0.167 .
This can be done for all Kc values to obtain the corresponding real and imaginary roots. After all of the roots have been calculated, a table
can be made to format the real roots (x axis) verses imaginary roots (y axis). When the syntax, A={{x,y}{a,b}…}, is used, you are inputting
all of the x and y values and naming those values A. When you call A in the TableForm[ ] function, the table will be made according to all
of the x and y values input into A. The syntax and corresponding outputs are shown below.
In order to produce the table shown above, you only need TableForm[A] and nothing else. The other descriptions (ie. TableAlignments and
TableHeadings) are just used for formatting purposes.
Once the real (x) and imaginary (y) roots have been determined and put into table format for your equation, the ListPlot[ ] function can be
used to develop your root locus diagram by calling A.
As you can see, the ListPlot[ ] function returned the same plot as shown in the “Complex Coordinate System to make Root Locus Plot”
section (this plot was created in Excel). The arrows are not shown in this Mathematica plot; however the Excel plot shows the direction of
increasing Kc values to show how the roots of the system vary by changing the Kc values.
The Mathematica Help Browser is a very useful tool for understanding the syntax. For example, when using the ListPlot[ ] function you can
search how to label the axes and how to make the points larger (more visible) on the plot. The following examples are used to illustrate the
use of root locus plots. Click the following link if you would like to access the actual Mathematica file: Media:RLPFinalaa.nb
10.6.6: SECOND PLOT METHOD USING ARRAYS

The goal here is to be able to understand exactly what is happening with these Mathematica inputs in order to get a better grasp on the
outputs. Basically, once you have found the Eigenvalues of the system, which examples can be found for in other areas of the wiki, we then
want to solve explicitly for them. After we solve for the Eigenvalue in terms of the variables, we can create an array by varying this value
and solving for the real and imaginary terms of the Eigenvalues.
10.6.6.1: EIGENVALUE OUTPUTS IN MATHEMATICA

The Eigenvalues can either be returned as a function of 'l', which is the designation of the Eigenvalue or simply as a complex function in
terms of variables. When returned as a function of 'l' there will be a polynomial input that has to be simplified to find 'l' as a function of the
variable. In this case: Kc.
This example is taken directly from lecture 17 and its supplementary Mathematica file. What is happening here is that the Jacobian was
inputted, and there were multiple eigenvalue arrays that were found. The last array has three complex polynomial functions that equal to 0.
The last line is simplifying them, solving for 'l'. This is putting the simplified Eigenvalues into the array 'ss'.
Another output of Eigenvalues is simply in the form of complex numbers themselves. Rather than worry about the simplification steps,
these Eigenvalues can be directly inputted as a complex number into an array, again termed 'ss':
This is a variation of the sample in lecture 17.
10.6.6.2: PLOT ARRAY CREATION

Now that we have seen the two ways of Eigenvalue array creation, we can use these to create another array that will be the points for the
plot. Here we will be substituting in values of the variable Kc to find different values of the Eigenvalues, which will give us information on
stability changes as Kc changes.
In the example above, 'ss' is in the form :
Each solution in this array is in the form l -> a+bi. For example, ss refers to solution2, which is the object in the second position in ss.
Re[l/.ss] applies solution2 to the variable l and then returns only the real part. Im[l/.ss] applies solution2 to the variable l and then returns
only the imaginary part. The effect of the Append function above is that the ordered pair {a,b} (from a solution of the form a+bi) is added to
the array called 'a'.
However, not all Eigenvalue arrays will have the same number of Eigenvalues. For example, as mentioned in the above paragraph, 'ss' is in
the form :
There could be 2, or 3, or more separate 'solution' values in the array. Depending on the number of the 'solution' values, there will be a
different format for your 'append' function in Mathematica. For example, in the image below, there are only two eigenvalues in our arrat that
we want to solve for. Thus, there should only be two rows of the addition to the array for the this method:
These different methods both output arrays of numbers that act essentially as 'x' and 'y' values, or in this case Real and Imaginary, for the
plot. The array 'a' is plotted as such:
ListPlot[a, PlotStyle -> PointSize[0.02]]
This is inputted directly to Mathematica, which creates the Root Locus Plot.
10.6.7: DIFFERENTIAL EQUATION EXAMPLE OF ROOT LOCUS PLOTS IN MATHEMATICA

An example problem pulled from the Fall 2008 second exam:
"You have just been put in charge of designing a large-scale bioreactor for the production of a blockbuster protein based drug. After some
research, you have come up with the following model to describe the reactor system:"
dX 10XS
= −2X +
dt S +2
dS 3XS
= 4F in − S −
dt S +2
dP
= 9X − 2P
dt
where X is the cell concentration, S is the nutrient concentration, and P is the protein product. F in is the flow rate of nutrient into the
system.
In this system there are two fixed points found by setting all the differential equations equal to zero and solving in Mathematica.
The two fixed points depend on the value of Fin and with that varying values of Fin will change eventual eigenvalues. Since this problem
will eventually want root locus plots for both fixed points it is easier to save both points as parameters to be applied to general equations
later. Overall it will help save the need for redundant code. The entire mathematica file being referenced in this example is available here:
Media:Exam2Prob1.nb
From this point a Jacobian matrix should be created to linearize the system around the fixed points (as applied to the Jacobian). Once
created the stored fixed point values for X, S, and P can be applied. Once applied, the eigenvalues for each fixed point can be solved for.
These will still contain a variable 'Fin' since Fin has not been defined yet.
From here the Table[] function can be used to tabulate eigenvalues for various Fin values for visual sake of inspection. This is not necessary
since the eigenvalue results have already been solved for in terms of Fin. To create the root locus plots, values of Fin need to be defined and
applied to the eigenvalue equations with the results stored. That step is next. In this table step it is for the user's sake to see with increasing
Fin values what the eigenvalues look like. The tabulated eigenvalues are stored as a matrix and displayed as such with the function
'MatrixForm[
These tables make it easy to see with increasing Fin (goes from 0 to 2 down the table) what the changes are in the stability of each fixed
point. The first table is the first fixed point, the second table the second FP.
Now, to make the root locus plots, a really long string of values needs to be created in order for the real and imaginary roots to be plotted.
This is a long string of code that is basically brute force. The overall action occurring here is setting Fin to a value, evaluating that Fin
through the eigenvalues, and saving the real portion separate from the imaginary portion by different columns. This is repeated for as many
Fin values as you desire. It is extremely repetitive, but works. After evaluating as many Fin values as you desired you can display the two
columns through the ListPlot[] function. Example snippets of the code look like such:
And the ListPlot[] function can look like this:
And would make graphs like these (Fixed Point 1 on top and FP2 on bottom in this example):
Looking back at the tables of eigenvalues the plots can be interpreted. For FP1 the values start real negative and increase to real positive.
There are no imaginary values. For FP2, increasing Fin values go from real positive to negative with imaginary values occurring in the
transition from low to high Fin values.
10.6.8: ALTERNATIVE MATHEMATICA METHOD

Some Mathematica 6 programs have an add-on or supplementary application known as ANALOG INSYDES which contains special
options to create Root Locus Plots with ease. The Function is called RootLocusPlot. Enter the function that is to be analyzed, into
Mathematica with the following format:
RootLocusPlot[tfunc,{k,k_0 ,k_1}]
tfunc is the transfer function in the frequency variable s and one real parameter k. k_0 and k_1 are the range for the real parameter k that is
to be varied int he Root Locus Plot. Follow an example below.
Example: Equation to make a Root Locus Plot with:
2 3
H(s) = \draca + 2s + s210 + 3as + 4s +s
To make a Root Locus Plot follow the example code below:

H4[s_, a_] := (a + 2*s + s^2)/(10 + 3*a*s + 4*s^2 + s^3)
RootLocusPlot[H4[s, a], {a, 3, 5}]
Image: Wolfram.com
Other Mathematica forms for using RootLocusPlot[]:
RootLocusPlot[func] This form displays the a pole/zero diagram of a function i.e. func without parameters k and k_1.
RootLocusPlot[rootloc] This form displays a root locus calculated with function RootLocusByQZ[].
Refer to Mathematica link for further discuss about RootLocusByQZ
Many options for the plot can be varied including the number of points, plot style, and plot color. All of these options and more can be
viewed online with a Mathematica Tutorial
10.6.9: CREATING ROOT LOCUS PLOTS WITH MATLAB

10.6.9.1: ROOT LOCUS GENERATION IN MATLAB
Three matlab files have been given to obtain the root locus plot and poles of the root locus plot at specified Kc values for a specific transfer
function with relative ease. These three files are polesfunction.m, transferfunction.m, and Locusplotpoles.m .[1]. The transferfunction file is
where the specific transfer function should be input. As written the only line needing variation is line 4 (Gs). The polesfunction file finds the
poles of the given transfer function using built in matlab utilities and needs no variation, even after changing the transfer function. The
Locusplotpoles file is what is called in matlab to generate the root locus plot as well as the value of the poles and integrates the two previous
files. To use this file you type in the values of the specific Kc you require the pole values for as well as a single value of Kc for which you
would like to see the root locus plot. An example of the inputs to and outputs of this file are given below.
>> Locusplotpoles([0 50 100 200],0)
ans =
1.0e+002 *
Columns 1 through 3
0 -0.0020 -0.0020 + 0.0000i 0.5000 -0.0045 -0.0008 + 0.0022i 1.0000 -0.0051 -0.0004 + 0.0027i 2.0000 -0.0060 -0.0000 + 0.0034i
Columns 4 through 6
-0.0020 - 0.0000i 0 0 -0.0008 - 0.0022i -0.0020 -0.0020 + 0.0000i -0.0004 - 0.0027i -0.0020 + 0.0000i -0.0020 - 0.0000i -0.0000 - 0.0034i
-0.0020 -0.0020 + 0.0000i
Column 7
0 -0.0020 - 0.0000i -0.0020 -0.0020 - 0.0000i
The values inside the brackets are the specific Kc values for which poles are desired. The last value is the Kc which the root locus plot will
be plotted for. Notice that the value of Kc for the root locus plot to be generated in this example is zero. This is so that the entire range of Kc
can be examined using the interactive graph produced by matlab. If other specific Kc plots would like to be observed this value can be
changed to generate these as well. The column lines give the values of Kc input and then value of the poles for each Kc. Note that the actual
values are the values matlab gives multiplied by 100.
Now that a basic understanding of root locus plots has been obtained here are a few websites which go further in depth into the underlying
principles and generation of these plots. www.facstaff.bucknell.edu/mastascu/eControlHTML/RootLocus/RLocus1A.html
This site gives animated examples which progress the value of the control variable through a root locus plot to better explain the
fundamental function of these plots. It would be beneficial to go through a few of these examples to observe patterns of fluctuation of the
control variable. The parameters for each animation are given to the left of the frame.
http://www.engin.umich.edu/group/ctm/rlocus/rlocus.html
This page gives examples of how you generate a root locus plot of a given transfer function using matlab. While you go through the content
on this page take special notice of the fact that none of the given transfer functions have an integrated Kc variable for which to generate a
locus plot. To see how these plots can are generated when a value of Kc is specified see the code of Locusplotpoles.m. For clarification on
the application of root locus plots see the given examples.
10.6.10: CREATING ROOT LOCUS PLOTS WITH EXCEL AND PPLANE

Though this tactic may remain slightly more time consuming and work involved, there is a method in computing the Root Locus plot using
PPLANE and Excel. This particular model really aids in better understanding of how to read and acknowledge a Root Locus Plot, and is
also useful if alternative options for making a Root Locus Plot are not currently available.
This method involves using PPLANE in order to find the eigenvalues for each equilibrium value, and then plotting these points with Excel
software. In order to do this, suppose that you have the following differential equations that represent a reactor:
dX 2+ X+ Y
= −X +
dt Y +3
dY
= Y × F in + 7 − (2X)
dt
Graphing the following differential equations provides something similar to the following chart (a few lines were drawn, just to better show
the graphs flow):
Using the ability to “Find an Equilibrium Point” in PPLANE (if you need help with PPLANE, click PhasePlaneAnalysis), one can select a
particular equilibrium point on the field, and be able to provide the following results:
This window “pops up” in the higher left corner of the screen, when the equilibrium point is found. This window provides the eigenvalues
for this equilibrium point.
Depending what you are varying (in this case, Fin), all you would have to do is enter various values of this Fin into PPLANE, and plot the
new set of differential formulas. After this is done, just find the same equilibrium point, and record the new given eigenvalues in excel.
These values should be entered with an x-component (in one column) and a y-component (in another column) into excel, with the following
rules:
Assuming the example 3+3i, 3-3i were given as eigenvalues.
For any given eigenvalue, the x-value of a particular eigenvalue is the real part of that number (ie: the “3” part).
For any given point, the y-value of a particular eigenvalue is the imaginary part of that number (ie: the “3i” part). If there is an imaginary
component, ignore the “i,” though the number should be recorded in the y-component. If there is no imaginary number, the y component
is zero.
For both cases, a positive is a positive, and a negative is a negative. Remember: imaginary numbers have 2 parts, a positive and negative
“i” value.
These points are simply recorded according to these rules, then the points are plotted against each other, with x components on the x-axis,
and the y in the y-axis. Here is an example of an excel diagram, taken from the proposed diagram:
And, the excel data is then plotted, x against y, and is shown below:
As you can see, this particular model did not have any imaginary data. The root locus plot has been formed, and for all accounts of Fin, the
eigenvalues are along the x-axis for the root locus plot.
10.6.11: PRACTICAL APPLICATION

In the past, it was necessary for engineers to master the techniques required to efficiently construct root locus diagrams. In today's
engineering world, this is not the case for one of two reasons. In many instances, root locus diagrams are not used industrially because they
require models of the system which are generally not available. If a model is available to develop a root locus diagram, there are computer
applications that can develop the diagrams much faster than a person. Thus, energy and effort should be placed on understanding and
interpreting a root locus diagram and understanding the general rules of stability for a root of the characteristic equation.
 EXAMPLE 10.6.1
Given the following table of roots to the characteristic equation for a system:
1. Construct a root locus plot (Making the plot by hand initially may give some a better understanding of what is going on as you will
see the changes in the plot as you add each set of roots).
2. Determine the ranges of Kc that would lead to the following system responses: no oscillations, dampened oscillations, stable
oscillations & unstable oscillations. Also, sketch what the output response would look like for a given step change in input for each
range of Kc.
3. How would the roots to the characteristic equation be different if this were PI control?
Solution
1) Root locus plot:
First, we break the roots into their imaginary and real components:
Then, we plot each point on a complex coordinate system (x-axis = real componebt, y-axis = imaginary component).
Also, it is useful to add lines tracing the path of increasing Kc. These lines are added below. Also, the points where Kc= zero are
generally labeled with an x. This has also been done in the figure below.
2) Ranges of Kc with explanations:

We know that the range where there will be no oscillatory response is when Kc provides all real roots. This occurs when Kc is between
0 and 0.66. (Sketch will look like figure a in the "interpreting a root locus diagram" section above).
Also, Kc yields dampened oscillations when the two of the roots of the characteristic equation are complex with negative real
components. This occurs when Kc is greater than 0.66 but less than 2.33. (Sketch will look like figure b in the "interpreting a root locus
diagram" section above).
Unstable oscillations are achieved when two of the roots yield complex numbers with positive real components. This occurs when Kc is
greater than 2.33. Graphically, we see that when the points move past the dashed line, we will have unstable oscillations. This is
represented in the figure below. (Sketch will look like figure c in the "interpreting a root locus diagram" section above).
Based on the data provided, our best assumption is that at a Kc just below 2.33, we will be able to achieve stable oscillations. When Kc
= 2.33 the real component of the complex root is 0.02 (slightly greater than zero). Ideally, we will achieve stable oscillations when the
real component of the complex number is zero. (Oscillations should be consistent in magnitude).
Lastly, it is a good idea to label some values of Kc on the root locus plot. This will help orient the reader when they are trying to
determine which point correspond to different values of Kc.
3) If this were PI control, each root to the characteristic equation would be accompanied by not only a Kc value but also a value for
integral control parameter, KI.
 EXAMPLE 10.6.2
You've been recently hired at MichChem Controls. In your first week at the company, a virus spreads through the computer network,
effectively disabling most of the computational programs. Fortunately, the virus was halted before it was able to disable Mathematica.
Shortly after the chaos from the virus subsides, your boss needs to know how to use Mathematica to create a root locus plot for several
new Proportional only level control systems that are being tested. Being the highly skilled engineer that you are, he doesn't want you to
waste time developing the root locus plots, rather, he wants you to write out example code, with a brief description that someone else
could use to develop the root locus plot.
The general characteristic equation for the new level control systems is:
3 2
f(s) = Ax + Bx − Cx + D + EKc
Please, provide your boss with an example of the exact script that would need to be entered into Mathematica in order to develop the
root locus plot. You should also brief descriptions explaining what the code does and what steps would need to be repeated to develop
the plot.
Solution
First, the user should develop a table of roots for the characteristic equation. This can be done using the solve function in Mathematica.
The code would appear like this for Kc = 0.
(note: the In[#] appears for each new line of code. This will not be typed in by the user.)
In[1]:= Solve[Ax^3 +Bx^2-Cx+D+E*(0)==0,x]
This should be repeated for Kc = 0.1, 0.15, 0.225, 0.5, 0.75, 1.5, etc. (until the roots of the equation begin to yield two complex
solutions whose real components are positive). After a sufficient number of roots have been obtained for the characteristic equation at
different values of Kc, these will need to be tabulated. To develop a table of data in Mathematica input the following code:
In[2]:= A =
This tabulates the real components and the imaginary components at each value of Kc. Once all of the values are tabulated, the root
locus plot can be developed using the ListPlot function.
In[3]:= ListPlot[A, PlotStyle -> PointSize[0.02]
This will plot all of the tabulated data in In[2] which is the root locus plot.
Exercise 10.6.1
A root locus plot is used to:
a. Identify the solutions to the characteristic transfer function.
b. Show the response time of the feedback system to a disturbance.
c. Graphically illustrate the effect of Kc on the dynamic characteristics of the feedback system.
d. Choose the optimal Kc for the feedback system.
Answer
c
Exercise 10.6.2
When a root is complex with positive real parts, the oscillations are:
a. Decreasing
b. Increasing
c. Unstable
d. A & C
e. B & C
Answer
e
10.6.12: REFERENCES
Luyben, Michael L.; Luyben, William L. "essential of Process Control". McGraw-Hill:New York, 1997.
Marlin, Thomas E. "Process Control: Designing Processes and Control Systems for Dynamic Performace 2nd Edition". McCGraw-Hill:
New York, 2000.
Ogunnaike, Babatunde A.; Ray, W. Harmon. Process Dynamics, Modeling, and Control. New York Oxford: Oxford UP, 1994. 488-507.
Riggs, James B.; Karim, M. Nazmul. Chemical and Bio-Process Control. 3rd ed. Ferret. 279-295.
Luyben, Michael L.; Luyben, William L.; Tyreus, Bjorn D. "Plantwide Process Control". Mcgraw-Hill:New York, 1999.
Seborg, Dale E.; Edgar, Thomas F.; Mellichamp, Duncan A. "Process Dynamics and Control". John Wiley and Sons, Inc. 2004.
http://www.engin.umich.edu/group/ctm/rlocus/rlocus.html as viewed 10/23/2007
www.facstaff.bucknell.edu/mastascu/eControlHTML/RootLocus/RLocus1A.html as viewed 10/23/2007

Authors: Halley Crast, Andrew Laskowski, Maurice Telesford, Emily Yatch
Stewards: Victoria Cardine, Anthony Campell, David Hines, Stephen Kerns
10.6: Root Locus Plots - Effect of Tuning is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source
10.7: ROUTH STABILITY - RANGES OF PARAMETER VALUES THAT ARE
STABLE
The stability of a process control system is extremely important to the overall control process. System stability serves as a key safety issue
in most engineering processes. If a control system becomes unstable, it can lead to unsafe conditions. For example, instability in reaction
processes or reactors can lead to runaway reactions, resulting in negative economic and environmental consequences.
The absolute stability of a control process can be defined by its response to an external disturbance to the system. The system may be
considered stable if it exists at a consistent state or setpoint and returns to this state immediately after a system disturbance. In order to
determine the stability of a system, one often must determine the eigenvalues of the matrix representing the system’s governing set of
differential equations. Unfortunately, sometimes the characteristic equation of the matrix (the polynomial representing its eigenvalues) can
be difficult to solve; it may be too large or contain unknown variables. In this situation, a method developed by British mathematician
Edward Routh can yield the desired stability information without explicitly solving the equation.
Recall that in order to determine system stability one need only know the signs of the real components of the eigenvalues. Because of this, a
method that can reveal the signs without actual computation of the eigenvalues will often be adequate to determine system stability.
To quickly review, negative real eigenvalue components cause a system to return to a steady state point (where all partial derivatives equal
zero) when it experiences disturbances. Positive real components cause the system to move away from a stable point, and a zero real
component indicates the system will not adjust after a disturbance. Imaginary components simply indicate oscillation with a general trend in
accordance with the real part. Using the method of Routh stability, one can determine the number of each type of root and thus see whether
or not a system is stable. When unknown variables exist in the equation, Routh stability can reveal the boundaries on these variables that
keep the roots negative.
10.7.2: THE ROUTH ARRAY

The Routh array is a shortcut to determine the stability of the system. The number of positive (unstable) roots can be determined without
factoring out any complex polynomial.
10.7.2.1: GENERATING THE ARRAY

The system in question must have a characteristic equation of a polynomial nature. as shown below:
n n−1
P (S ) = an S + an−1 S + ⋯ + a1 S + a0
In order to examine the roots, set P(S)=0, which will allow you to tell how many roots are in the left-hand plane, right hand plane, and on
the j-omega axis. If the system involves trigonometric functions it needs to be fit to a polynomial via a Taylor series expansion. One
necessary condition for stability is that a_n 0 \qquad " src="/@api/deki/files/18713/image-273.png">. (If , all coefficients may be
multiplied by -1 before checking). The other condition is that all values in column 1 of the Routh array must be positive for the system to be
stable.
This flow diagram shows the generation of a Routh array for an idealized case with m,n representing the location in the matrix.
The coefficients of the polynomial are placed into an array as seen below. The number of rows is one more than the order of the equation.
The number of sign changes in the first column indicate the number of positive roots for the equation.
Row
an an− 2 an− 4 ...
1
Row
an− 1 an− 3 an− 5 ...
2
Row
b1 b2 b3 . . .
3
Row
c1 c2 c3 . . .
4
⋮ ⋮ ⋮ ⋮
Row
p1 p2
5
Row
q1
6
Row
v1
7
In the array, the variables b1,b2,c1,c2,etc. are determined by calculating a determinant using elements from the previous two rows as shown
below:
The general expression for any element x after the first two rows with index (m,n) is as follows:
For
Note, that if the Routh array starts with a zero, it may still be solved (assuming that all the other values in the row are not zero), by replacing
the zero with a constant, and letting that constant equal a very small positive number. Subsequent rows within that column that have this
constant will be calculated based on the constant choosen.
Once the array is complete, apply the following theorems to determine stability:
1. If all of the values in the fist column of the Routh array are >0, then P(S) has all negative real roots and the system is stable.
2. If some of the values in the first column of the Routh array are <0, then the number of times the sign changes down the first column will
= the number of positive real roots in the P (S ) equation.
3. If there is 1 pair of roots on the imaginary axis, both the same distance from the origin (meaning equidistant), then check to see if all the
other roots are in the left hand plane. If so, then the imaginary roots location may be found using AS2 + B = 0, where A and B are the
elements in the Routh array for the 2nd to last row.
To clarify even further, an example with real numbers is analyzed.
 EXAMPLE ARRAY
The following polynomial was generated from a sample system.

3 2
P (S ) = 5S − 10S + 7S + 20
The preceding polynomial must be investigated in order to determine the stability of the system. This is done by generating a Routh
array in the manner described above. The array as a result of this polynomial is,
Row
5 7
1
Row
−10 20
2
Row
17
3
Row
20
4
In the array shown above, the value found in the third row is calculated as follows. The array can now be
analyzed. When looking down the first column, it can be seen that 5 is positive in magnitude, then the sign changes in the -10 entry, and
the sign changes a second time to positive 17. This counts as two changes in sign, which corresponds to two positive roots, making the
system unstable.
10.7.3: FINDING STABLE CONTROL PARAMETER VALUES

Often, for a unit operation, a PID parameter such as controller gain (Kc), the integral time constant (Ti), or the derivative time constant (Td)
creates an additional variable in the characteristic equation. This can be carried through the computations of the Routh array to indicate
which values of the variable will provide stability to the system through by preventing positive roots from occuring in the equation. For
example, if a controller output is governed by the function:
3 2
10s + 5s + 8s + (Td + 2)
The stable values of Td can be found via a Routh array:

Row 1 10 8
Row 2 5 (Td + 2)
Row 3 4 − 2Td
Row 4 (Td + 2)
We reveal −2 < T < 2

d in order to keep the first column elements positive, so this is the stable range of values for this parameter.
If multiple parameters were in the equation, they would simply be solved for as a group, yielding constraints along the lines of "Ti + Kc > 2"
etc, so any value chosen for one parameter would give a different stable range for the other.
10.7.4: SPECIAL CASES

There are a few special cases that one should be aware of when using the Routh Test. These variances can arise during stability analysis of
different control systems. When a special case is encountered, the traditonal Routh stability solution methods are altered as presented below.
10.7.4.1: ONE OF THE COEFFICIENTS IN THE CHARACTERISTIC EQUATION EQUALS ZERO

If the power of the 0*Sn is , replace the zero with a quantity, ε, which would be positive and will approach zero. Then continue with
your analysis as normal. Essentially this gives the limit of the roots as that coefficient approaches zero. (If the power = 0, see Case 3). For
example,
Equation:
Working Equation:
Row
2 − 24
1
Row
ϵ 32
2
Row
f rac−24ϵ − 64ϵ 0
3
Row
q q uad32
4
Since ε is positive we know that in the first column row 2 will be positive, row 4 will be positive, and row 3 will be negative. This means
we will have a sign change from 2 to 3 and again from 3 to 4. Because of this, we know that two roots will have positive real components. If
you actually factor out the equation you see that , showing that we do have 2 positive roots. Both of these roots are
equal to 2, so there is technically only one root, but in any case we know the system is unstable and must be redesigned.
10.7.4.2: ONE OF THE ROOTS IS ZERO
This case should be obvious simply from looking at the polynomial. The constant term will be missing, meaning the variable can be
factored from every term. If you added an ε to the end as in case 1, the last row would be ε and falsely indicate another sign change. Carry
out Routh analysis with the last zero in place.
Equation:
Row
1 − 2
1
Row
2
Row
3
Row
0
4
As you can see in column one we have row 1 positive, row 2 and 3 negative, and row 4 zero. This is interpreted as one sign change, giving
us one positive real root. Looking at this equation in factored form,
we can see that indeed we have only one positive root equal which equals 2. The zero in the last row indicates an additional unstable root of
zero. Alternatively, you may find it easier to just factor out the variable and find the signs of the remaining eigenvalues. Just remember there
is an extra root of zero.
10.7.4.3: ROW FULL OF ZEROS

When this happens you know you have either a pair of imaginary roots, or symmetric real roots. The row of zeros must be replaced. The
following example illustrates this procedure.
Equation:
Row
1 10 9
1
Row
2
Row
9 9
3
Row
0 0
4
Row 4 contains all zeros. To determine its replacement values, we first write an auxiliary polynomial A determined by the entries in Row 3
above.
Notice that the order decreases by 1 as we go down the table, but decreases by 2 as we go across.
We then take the derivative of this auxiliary polynomial.
The coefficients obtained after taking the derivative give us the values used to replace the zeros. From there, we can proceed the table
calcuations normally. The new table is
Row
1 10 9
1
Row
2
Row
9 9
3
Row
18 0
4
Row
9
5
In fact, the purely imaginary or symmetric real roots of the original polynomial are the same as the roots of the auxiliary polynomial. Thus,
we can find these roots.
Because we have two sign changes, we know the other two roots of the original polynomial are positive.
In fact, after factoring this polynomial, we obtain
Therefore, the roots are , where in this case, the root 3 has multiplicity 2.
10.7.5: LIMITATIONS
Routh arrays are useful for classifying a system as stable or unstable based on the signs of its eigenvalues, and do not require complex
computation. However, simply determing the stability is not usually sufficient for the design of process control systems. It is important to
develop the extent of stability as well as how close the system is to instability. Further stability analysis not accounted for in the Routh
analysis technique include finding the degree of stability, the steady state performance of the control system, and the transient response of
the system to disturbances.
More involved techniques, such as those discussed in Eigenvalues and Eigenvectors, must be used to further characterize the system
stability (with the exception of system polynomials resulting in pure imaginary roots). Another limitation of the Routh method occurs when
the polynomial in question becomes so large that Routh stability is too computationally time consuming (a personal judgment). For this
situation another method, such as a root locus plotmust be used.
Note that for defining stability, we will always start out with a polynomial. This polynomial arises from finding the eigenvalues of the
linearized model. Thus we will never encounter other functions, say exponenential functions or sin or cos functions in general for stability
analysis in control theory.
10.7.6: ADVANTAGES OVER ROOT LOCUS PLOTS

Routh stability evaluates the signs of the real parts of the roots of a polynomial without solving for the roots themselves. The system is
stable if all real parts are negative. Therefore unlike root locus plots, the actual eigenvalues do not need to be calculated for a Routh stability
analysis. Furthermore, sometimes the system has too many unknowns to easily construct and interpret a root locus plot (e.g. with two PID
controllers there are the variables Kc1, Kc2, τi1, τi2, τd1, and τd2).
 EXAMPLE 10.7.1
Assume the following polynomial describes the eigenvalues of a linearized model of your process. For this polynomial, complete a
Routh array and determine the system's stability?
Answer
Since P(X) is a fourth-order polynomial, the Routh array contains five rows.
Row 1
Row 2
Row 3
Row 4
Row 5
Rows 1 and 2 correspond to the coefficients of the polynomial terms:

Row 1
Row 2
Rows 3, 4, and 5 contain the determinants using elements from the previous two rows.
Row 3
Row 4
Row 5
Therefore,
Row 3
Row 4
Row 5
The complete Routh array:

Row 1
Row 2
Row 3
Row 4
Row 5
Since all the values in the first column are positive, the equation P(x) has all negative roots and the system is stable.
 EXAMPLE 10.7.1
Consider a system with the following characteristic equation:

3 2
20s + 59s + 46s + (4 + Kc ) = 0
Using a P-only controller, find the range of controller gain that will yield a stable system.
Answer
Since the equation is a third-order polynomial, the Routh array has four rows. Routh Array:
Row
20 46
1
Row
59 (4 + Kc )
2
Row 20
46 − (4 + Kc ) 0
3 59
Row
(4 + Kc ) 0
4
For the system to be stable, all the elements of the first column of the Routh array have to be positive.
The first column will contain only positive elements if:
6 - \frac{20}{59}(4+K_c)0" src="/@api/deki/files/18855/image-349.png">
4+K_c0 \qquad" src="/@api/deki/files/18859/image-351.png">

_c-4 \qquad" src="/@api/deki/files/18861/image-352.png">
Acceptable stable range
 EXAMPLE 10.7.3
Consider a system with the following characteristic equation:

5 4 3 2
s − 3s +s +s +4 = 0
Determine the stability of this system.

Answer
One of the coefficients in the characteristic equation equals 0. We replace the zero with a quantity which would be positive (approach
zero from the right-hand side) and continue with the analysis as normal.
Working equation:
5 4 3 2
s − 3s +s +s + ϵs + 4 = 0
Row
1 1 ϵ
1
Row
2
Row 4 4
ϵ+
3 3 3
Row 9ϵ
4 + 4
4 4
Row 4 64
ϵ+ − 0
5 3 27ϵ + 48
Row
4
6
Since ε is positive, in the first column, there are two sign changes, from row 1 to row 2 and from row 2 to row 3. Thus, we know that
the roots will have two positive real components. If you actually factor out the equation you will see that,
, showing that we do have two positive roots, both equal to 2.
Additional complication exists because at row 5, as ε goes to zero, the term also goes to zero, which means that for row 5, we are
getting a row full of zeros. This means that we have a pair of imaginary roots, and this situation can be solved using the equation,
.
In this case, the working equation is,
The imaginary roots are,
 EXAMPLE 10.7.4
You are an engineer at an icecream factory. There is a storage vat with a cooling system that has a PI controller scheme. It has the
following characteristic equation:
Your job is to determine the constraints on the values Kc and Ti such that the system is stable.
Answer
The goal is to make the matrix such that the first column has no sign changes. Since the first two entries in the first column are numbers
and positive, therefore all other values in this column must be positive.
Working equation:
Row
10 − Kc Ti + 6
1
Row
3 3
2
Row
10 − Kc Ti + 6
3
Row 3(Ti + 6)
3− 0
4 (10 − Kc )
Row
Ti + 6
5
Since is in the first column, it gives the constraint . Likewise, must be positive, giving the constraint _i-6\,"
src="/@api/deki/files/18907/image-375.png">. Which brings the final entry of . By dividing by three and
rearrangement of the terms, it is seen that . The terms are then rearranged to arrive at the inequality: .
10.7.7: REFERENCES
Bequette, W.B. Process Control Modeling Design and Simulation., New Jersey: Prentice Hall, pp 170-178.
Foecke, H.A. and Weinstein, A. "Complex roots via real roots and square roots using Routh's stability criterion." arxiv.org, January 5,
2007.
Ogunnaike, Babatunde A.; Ray, W. Harmon. Process Dynamics, Modeling, and Control. New York Oxford: Oxford University Press,
1994.
Contributors
Authors: John D'Arcy, Matt Hagen, Adam Holewinski, and Alwin Ng
Stewards: Jeff Falta, Taylor Lebeis, Shawn Mayfield, Marc Stewart, Tom Welch
10.7: Routh Stability - Ranges of Parameter Values that are Stable is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter
request.
CHAPTER OVERVIEW
11: CONTROL ARCHITECTURES

11.1: Feedback Control
11.2: Feed Forward Control
11.3: Cascade Control
11.4: Ratio Control
11.5: Common control loops and model for liquid pressure and liquid level
11.6: Common Control Loops and Model for Temperature Control
11.7: Common control architectures and model for reactors
11: Control Architectures is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source content that was
1
11.1: FEEDBACK CONTROL
There are many different control mechanisms that can be used, both in everyday life and in chemical engineering applications. Two broad
control schemes, both of which encompass each other are feedback control and feed-forward control. Feedback control is a control
mechanism that uses information from measurements to manipulate a variable to achieve the desired result. Feed-forward control, also
called anticipative control, is a control mechanism that predicts the effects of measured disturbances and takes corrective action to achieve
the desired result. The focus of this article is to explain application, advantages, and disadvantages of feedback control.
Feedback control is employed in a wide variety of situations in everyday life, from simple home thermostats that maintain a specified
temperature, to complex devices that maintain the position of communication satellites. Feedback control also occurs in natural situations,
such as the regulation of blood-sugar levels in the body. Feedback control was even used more than 2,000 years ago by the Greeks, who
manufactured such systems as the float valve which regulated water level. Today, this same idea is used to control water levels in boilers
and reservoirs.
11.1.2: FEEDBACK CONTROL

In feedback control, the variable being controlled is measured and compared with a target value. This difference between the actual and
desired value is called the error. Feedback control manipulates an input to the system to minimize this error. Figure 11.1.1 shows an
overview of a basic feedback control loop. The error in the system would be the Output - Desired Output. Feedback control reacts to the
system and works to minimize this error. The desired output is generally entered into the system through a user interface. The output of the
system is measured (by a flow meter, thermometer or similar instrument) and the difference is calculated. This difference is used to control
the system inputs to reduce the error in the system.
Figure 11.1.1: Feedback control loop (Adapted from Lee, Newell, and Cameron 1998: 6)
To understand the principle of feedback control, consider Figure 11.1.2 In order to bake cookies, one has to preheat an electric oven to
350°F. After setting the desired temperature, a sensor takes a reading inside the oven. If the oven is below the set temperature, a signal is
sent to the heater to power on until the oven heats to the desired temperature. In this example, the variable to be controlled (oven
temperature) is measured and determines how the input variable (heat into oven) should be manipulated to reach the desired value.
Figure 11.1.2: Feedback control in an electric oven (Adapted from Lee, Newell, and Cameron 1998: 6)
Feedback control can also be demonstrated with human behavior. For example, if a person goes outside in Michigan winter, he or she will
experience a temperature drop in the skin. The brain (controller) receives this signal and generates a motor action to put on a jacket. This
minimizes the discrepancy between the skin temperature and the physiological set point in the person. The example is illustrated below:
There are two types of feedback control: negative and positive. Negative feedback is the most useful control type since it typically helps a
system converge toward an equilibrium state. On the other hand, positive feedback can lead a system away from an equilibrium state thus
rendering it unstable, even potentially producing unexpected results. Unless stated explicitly, the term feedback control most often refers to
negative feedback.
11.1.2.1: NEGATIVE FEEDBACK

By definition, negative feedback is when a change (increase/decrease) in some variable results in an opposite change (decrease/increase) in
a second variable. This is demonstrated in Figure 11.1.3 where a loop represents a variation toward a plus that triggers a correction toward
the minus, and vice versa. Negative feedback leads to a tight control situation whereby the corrective action taken by the controller forces
the controlled variable toward the set point, thus leading the system to oscillate around equilibrium.
Figure 11.1.3: Negative Feedback: Maintenance of equilibrium and convergence

Figure 11.1.4 shows the negative feedback mechanism between Duck Population and Duck Death Rate. For a given flock, we say that if the
death rate increases, the duck flock will decrease. On the contrary, if the duck flock increases, the death rate of the flock will decrease.
Figure 11.1.4: Duck population negative feedback loop
11.1.2.2: POSITIVE FEEDBACK

As opposed to negative feedback, positive feedback is when a change (increase/decrease) in some variable results in a subsequently similar
change (increase/decrease) in a second variable. In some cases, positive feedback leads to an undesirable behavior whereby the system
diverges away from equilibrium. This can cause the system to either run away toward infinity, risking an expansion or even an explosion, or
run away toward zero, which leads to a total blocking of activities (Figure 11.1.5 ).
Figure 11.1.5: Positive Feedback: Exponential Growth and divergent behavior, no intermediate situation
Figure 11.1.6 shows the feedback mechanism responsible for the growth of a duck flock via births. In this example, we consider two system
variables: Duck Births and Ducks Population. For a given flock, we state that if the birth rate increases, the duck flock will increase.
Similarly, if the duck flock increases, the birth rate of the flock will increase.
Figure 11.1.6: Duck population positive feedback loop
11.1.3: APPLICATIONS
Control mechanisms are used to achieve a desired output or operating condition. More specifically, feedback control is a useful mechanism
in numerous applications across many engineering fields. In chemical engineering, feedback control is commonly used to manipulate and
stabilize the conditions of a CSTR. Figure 11.1.7 shows how feedback control can be effectively used to stabilize the concentrations of
reactants in a CSTR by adjusting the flow rates.
11.1.3.1: CSTR WITH FEEDBACK CONTROL
Several types of feedback control can be used to manipulate the conditions in a CSTR: positive feedback, negative feedback, or a
combination of both. Figure 11.1.7 illustrates each of these possible situations. As depicted below, each CSTR is equipped with two
electrodes that measure the voltage of the solution contained inside the reactor. A computer adjusts the flow rates of the pump(s) in response
to any changes in the voltage.
Figure 11.1.7: CSTR with feedback control: equipment and control configuration (Dolnik et al. 1997)
(a) All of the reagents are pumped into the reactor by the same pump. The flow rate through the pump is adjusted constantly by a
negative feedback mechanism; when level of the iodide solution is low, the computer detects the insufficiency and increases the flow
rate of all the reactants.
(b) The iodide solution is pumped into the reactor by one pump, while the other two reactants are pumped in at a constant flow rate. The
flow rate of the iodine solution is controlled by a negative feedback mechanism; when the computer detects an insufficient amount of
iodine in the CSTR, it signals the pump. The flow rate of iodine into the CSTR is therefore increased.
(c) Two pumps are used to feed the reactor: one pump for the iodine solution and one for the ClO2 and H+ solutions. The flow of the
iodine solution is controlled by a negative feedback mechanism; if the computer detects an iodine deficiency, it signals Pump 2 to
increase the flow rate of the iodide solution. The flow rate of the ClO2 and H+ solutions is controlled by a positive feedback mechanism;
if the computer detects an iodide deficiency, it will signal Pump 1 to decrease the flow rates of ClO2 and H+,thereby increasing the
concentraion of iodide.
It is easy to see that by combining feedback controls, such as in Figure 11.1.1c , output concentrations and operating conditions can be
maintained at the desired state much more effectively than by only using one type of feedback control.
11.1.4: ADVANTAGES AND DISADVANTAGES

The unique architecture of the feedback control provides for many advantages and disadvantages. It is important to look at the exact
application the control will be used for before determining which type of control will be the best choice (see Cascade Control Systemsand
Feed Forward Control).
Advantages: The advantages of feedback control lie in the fact that the feedback control obtains data at the process output. Because of this,
the control takes into account unforeseen disturbances such as frictional and pressure losses. Feedback control architecture ensures the
desired performance by altering the inputs immediately once deviations are observed regardless of what caused the disturbance. An
additional advantage of feedback control is that by analyzing the output of a system, unstable processes may be stabilized. Feedback
controls do not require detailed knowledge of the system and, in particular, do not require a mathematical model of the process. Feedback
controls can be easily duplicated from one system to another. A feedback control system consists of five basic components: (1) input, (2)
process being controlled, (3) output, (4) sensing elements, and (5) controller and actuating devices. A final advantage of feedback control
stems from the ability to track the process output and, thus, track the system’s overall performance.
Disadvantages: Time lag in a system causes the main disadvantage of feedback control. With feedback control, a process deviation
occurring near the beginning of the process will not be recognized until the process output. The feedback control will then have to adjust the
process inputs in order to correct this deviation. This results in the possibility of substantial deviation throughout the entire process. The
system could possibly miss process output disturbance and the error could continue without adjustment. Generally, feedback controllers
only take input from one sensor. This may be inefficient if there is a more direct way to control a system using multiple sensors. Operator
intervention is generally required when a feedback controller proves unable to maintain stable closed-loop control. Because the control
responds to the perturbation after its occurrence, perfect control of the system is theoretically impossible. Finally, feedback control does not
take predictive control action towards the effects of known disturbances.
11.1.5: CLOSED LOOP CONTROL VERSUS OPEN LOOP CONTROL

Although there are various types of controllers, most of them can be grouped into one of the two broad categories: closed loop and open
loop controllers. The subsections below summarize the differentiation.
Closed Loop System
In a closed loop control system, the input variable is adjusted by the controller in order to minimize the error between the measured output
variable and its set point. This control design is synonymous to feedback control, in which the deviations between the measured variable
and a set point are fed back to the controller to generate appropriate control actions.The controller C takes the difference e between ther
reference r and the output to change the inputs u to the system. This is shown in figure below. The output of the system y is fed back to the
sensor, and the measured outputs go to the reference value
Open Loop System

On the other hand, any control system that does not use feedback information to adjust the process is classified as open loop control. In open
loop control, the controller takes in one or several measured variables to generate control actions based on existing equations or models.
Consider a CSTR reactor that needs to maintain a set reaction temperature by means of steam flow (Figure 11.1.8 ): A temperature sensor
measures the product temperature, and this information is sent to a computer for processing. But instead of outputting a valve setting by
using the error in temperature, the computer (controller) simply plugs the information into a predetermined equation to reach output valve
setting. In other words, the valve setting is simply a function of product temperature.
Figure 11.1.8: Example of CSTR open loop controller. (A) System diagram. (B) Open control loop.
Note that the open loop controller only uses the current state of the measured variable (product temperature) and a model to generate its
control output (valve setting), as opposed to monitoring errors that have already taken place. As the result, the quality of the control system
depends entirely upon the accuracy of the implemented model, which is challenging to develop. For this reason, feedback, or closed loop,
controllers are generally recognized as the more reliable control system.
Short Summary on Closed and Open Loop Controllers
Feedback Controller = Closed Loop Controller
Non-Feedback Controller = Open Loop Controller
 EXAMPLE 11.1.1
Your company is designing a plant to produce biodiesel. Several of the unit processes require heated water that is under tight
temperature control. The desired temperature of the water is 65°C ± 1°C. To accomplish this, steam will be used in a shell and tube heat
exchanger to heat the incoming water.
A member on the project team suggests the following feed forward control system to regulate the temperature of the water stream.
Temperature and flow rate data for the incoming steam and water streams will be used along with energy balance calculations to
determine the flow rate of steam required to keep the exiting water stream at the appropriate temperature.
You on the other hand suggest a feedback control system. Place the appropriate sensors, actuators, and controllers to implement a
feedback control system that will allow for the control of the exiting water temperature. What are some of the advantages and
disadvantages of using a feedback control system over a feed forward control system?
Solution
In the feedback control system, the temperature of the variable to be controlled is measured directly. If the exiting water temperature is
too high, the controller will decrease the amount of steam entering the shell side of the heat exchanger. Likewise, if the exiting water
temperature is too low, the controller will increase the amount of steam entering the shell side of the heat exchanger. This is an example
of negative feedback at work to keep the exiting temperature at the setpoint.
Advantages
Effective regardless of the origin of the disturbance.
Relatively simple to implement. A mathematical model of the exiting water temperature is not needed.
Disadvantages
Corrective action taken only after there is a deviation in the output from the set point.
Can cause instability if incorrectly tuned
 EXAMPLE 11.1.1
Of the following two scenarios, which would be the best to use a feedback control?
Solution
A feed-forward control or feed-forward/feedback system would most appropriate for the first scenario. In this situation, it is
critical that the pH of the waste stream not drop below the setpoint of pH 6. It is also desirable to minimize the use of lime, since
this reduces the cost of the waste treatment. Therefore very tight process controls are desired, for which a feed-forward or feed
forward/feedback system would be most appropriate. A feed-forward system can adjust to changes in inputs before they cause
deviations in the output stream. Additionally, since there are important fluctuations two process inputs (pH variations in both the
lime slurry and the acidic waste), feed-forward controls would be favored over feedback control.
A feedback control would be most useful for the second scenario. There is only one monitored variable (concentration of
reduced wine) and only one manipulated variable (steam). Since minor fluctuations about the wine concentration set point are
not critical, the performance of a typical feedback control system would be acceptable. It would be hard to implement a feed-
forward control system since the concentration of the wine feed to the evaporator may fluctuate and is not monitored.
 EXAMPLE 11.1.3
Using a shell and tube heat exchanger, a fuel stream is pre-heated using a steam stream from the utilities part of the plant. The plant is
located in Ann Arbor, Michigan. Originally, a feedback controller is implemented to control the amount of steam coming into the shell
and tube exchanger based on the temperature of the flow out of the exchanger. This is depicted below.
Upon implementation in the process, it was found that the feed temperature was not the only factor in the variance of the outlet stream.
Because the plant is operated in Michigan, the steam temperature varies between seasons, colder in the winter and warmer in the
summer. Propose a control scheme to account for both of these variances.
Solution
Here a cascade control scheme would be optimal to control the outlet temperature of the stream to a desired temperature, not just a
feedback controller. The scheme is depicted below.
The feedback control previously used to control the system is still implemented, however it is now reporting to the flow controller
before the steam control valve. Also, there is a temperature controller now incorporated in the control scheme. This is a feed forward
controller that also controls the flow controller for the steam control valve. Thus the primary control loops, the feed forward and
feedback control loops, comprise the secondary control loop in the cascade control. This incorporates both the temperature of the
incoming feed stream, which is controlled by feedback control via the temperature sensor after the exchanger. This is because the
temperature of the outlet stream would vary with the temperature of the inlet stream if the steam stream were not accounted for. But
accounting for the temperature of the steam stream, via a feed forward controller that is based on a temperature sensor before the
control valve, will give a better approximation and more control over the temperature of the outlet stream. This is also a very good
example of combining several control architectures and how they function in a real-world application, and why a feedback controller in
not necessarily the most accurate control.
 EXAMPLE 11.1.4
A blending system is used to mix a process stream (Stream 1, pure A) with liquid catalyst (Stream 2, pure B). The mixture then flows
into a reactor. The flowrate of stream 1 (f1) can vary but it is possible to manipulate the flowrate of stream 2 (f2). The outflow rate from
the reactor (f3) also can be manipulated via a pump. The level in the measures by a level transmitter that sends its signal to a level
controller. All three flow rates can be measured, but measuring the composition of the streams is impossible. In order to control the
level of the tank, and engineer has two options:
i. manipulate f2 while holding f3 constant
ii. manipulate f3 while holding f2 constant
Could method (i) utilize feedforward or feedback control of the level? If so, explain how it can be done.
Could method (ii) utilize feedforward or feedback control of the level? If so, explain how it can be done.
Solution
For configuration (i)
Feedback: If you measure h, when it goes up, you can reduce f2, and when it goes down, you can increase f2
Feedforward: If you measure f1, when it goes up, you can reduce f2, and and when it goes down, you can increase f2. You are
indirectly controlling h due to the mass balance.
For configuration (ii)
Feedback: If you measure h, when it goes up, you can increase f3, and when it goes down, you can decrease f3.
Feedforward: If you measure f1, when it goes up, you can reduce f3, and when it goes down, you can reduce f3.
This problem was meant to explore different strategies for both feedback and feedforward control configurations. Additionally, it
demonstrates how controlling the manipulated variables (either f2 for configuration (i) or f3 for configuration (ii)) affects the desired
outcome (maintaining consistent tank height).
11.1.6: REFERENCES
Lee, P., Newell, R., Cameron, I.(1998).Process Control and Management, London:Blackie Academic & Professional. ISBN 0-7514-
0457-8
Marlin, T.(2000). Process Control: Designing Processes and Control Systems for Dynamic Performance, Boston:Mcgraw Hill. ISBN 0-
07-039362-1
Palazoglu, A., Romagnoli, J.(2006).Introduction to Process Control, Boca Raton:Taylor & Francis. ISBN 0-8493-3696-9
Perry,R.H.(1997). Perry's Chemical Engineer's Handbook, 7th Edition, McGraw-Hill. ISBN 978-0-07-049841-9
National Lime Association Article. Using Lime for Acid Neutralization. September 2000.
MIT Open Courseware. (2006). "Lesson 8: Cascade and Feedforward Control Schemes," [1]

Authors: Eddy Kim, Bing Liu, Tyler Roehm, Samah Tout
Stewards: Gillian Berberich, Katie Feldt, Christopher Mark, Jason Wong
Cara Canady, David Carpenter, Che Martinez, Jeremy Minty, Bradley Novak
11.1: Feedback Control is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source content that was
11.2: FEED FORWARD CONTROL
Feed-forward control is a useful tool in the field of chemical engineering when there is a known set of deviations occurring upstream of
the system. This would allow engineers to account for that particular deviation within the controller and reduce the effects of the deviation
on the system. An example would be a car's cruise-control system. If it has feedback control, when there is a slope and therefore a speed
reduction, the feedback controller would compensate by applying additional throttle to the engine. If it uses a feed-forward controller
instead, the controller would calculate the slope beforehand and throttle the engine before any speed reduction occurs. In this sense, the
controller predicts the incoming deviation and compensates for it.
The following block diagram shows a feed-forward controller implemented on a idealized process with a setpoint R and load U :
where:
^
GP represents the process operator,
^
GM represents the measurement operator,
^
G C represents the controller operator, and
F F is the feed-forward controller.
−1
The perfect feed-forward controller is the inverse of the process operator, G

^
P
For example:
−1 δ
^ ^
GP = KP (τ + 1) = GF F
δt
In General:
−1
^ ^ ^
GP Y = (U + GF F U ) + GC (R − Y )
(in the language of operator or state space).
11.2.2: FEED-FORWARD CONTROL

The objective of feed-forward control is to measure disturbances and compensate for them before the controlled variable deviates from the
setpoint. Feed-forward control basically involves a control equation which has certain corrective terms which account for predicted
disturbances entering the system. The equation is only effective for gains in a steady state process. Dynamic compensation should be used
in the control equation if there are any dynamic deviations with the process response to the control action. This dynamic compensation
ability will be discussed further in the next section.
One form of a feed-forward control would be a Derivative (D) control which calculates the change in error and compensates
proportionately. But a D-control can’t perform by itself and usually requires working in conjunction with a Proportional (P) or Proportional
Integral (PI) control. More information can be found in the PID Intro article.
Simulated below is a typical shell and tube heat exchanger which heats up liquid water using steam.
Adding a feed-forward control into the system manipulates the amount of steam required to compensate for the varying amounts of liquid
feed coming in.
Overall heat balance:

Heat into the system is equal to the heat leaving the system.
Heat gained by liquid:
qout = ml Cp (T2 − T1 ) (11.2.1)
where:
-- mass flow rate of the liquid
-- the heat capacity of the liquid
-- the input temperature of the liquid
-- the desired temperature setpoint for the liquid
Heat lost by steam:
qin = ms λ (11.2.2)
where:
-- heat of vaporization
-- mass flow rate of the steam
Therefore:
Cp
ms = ml (T2 − T1 ) (11.2.3)
λ
To compensate for a deviation from the desired inlet temperature of the liquid water, the amount of steam must be adjusted. Equation (3)
acts as the control equation for the feed-forward controller. This controller actuates the inlet valve of the steam and acquires temperature and
flow rate data from the inlet streams. Based on the desired temperature setpoint, T2, and the actual input temperature, T1, we can use this
equation to calculate the mass flow rate of the steam required to compensate for the temperature deviation.
11.2.2.1: ACCOUNTING FOR SYSTEM NON-IDEALITIES
Often, one of the most difficult tasks associated with created a functional feed forward controller is determining the necessary equations that
govern the system. Even more complexities arise when the system is not, and cannot be treated as, ideal. This is the case with many real and
practical systems. There is simply too much heat lost or too many unforeseen effects to safely assume ideal conditions. The pure math of the
example above does not account for these effects.
The equations will output a value to a control valve (often in voltage). That voltage will be derived from some previously determined
relationship between voltage, valve %open, and steam flow rate. A very simple way to begin managing the issue of non-ideality is by
including a "non-ideality constant". This can be an additive or a multiplicative constant that adjusts the voltage output determined by the
equations.
Voltage output is some function of the calculated required steam flow
V = f (ms )
Voltage output is adjusted by some constant

V = cN ∗ f (ms )
or
V = cN + f (ms )
This non-ideality constant c often must be determined by trial. One way to accomplish this is to use manual control to determine the
N
output voltage needed at various inlet conditions. Using the data from the manual trials, and the resulting voltage that your unadjusted feed
forward controller would output, it is possible to determine by what factor your feed forward voltage needs to be adjusted. For example, if
by manual control you determine that for inlet conditions X the required voltage is 300 mV and your feed forward controller is only
outputting 270 mV for conditions X, you need some factor to adjust.
It may also happen that your "non-ideality constant" will not turn out to be constant when you begin to look at the data. In this situation,
consider using a linear relationship between the non-ideality factor and some inlet condition or implementing CASE or IF statements
covering discrete ranges of inlet conditions and giving different constants.
11.2.3: DYNAMIC COMPENSATION

Dynamic compensation is a method to account for factors such as lead and lag times when using feed-forward control. For instance, in the
example above, when the feed forward controller monitors a temperature decrease in the liquid feed, it will increase the steam flow rate -
the manipulated variable - to compensate and thus maintain the temperature of the exiting liquid flow - the controlled variable. However, the
steam may enter the heat exchanger faster than the liquid feed and this will cause a transient increase in the controlled variable from the
setpoint. In an ideal case the steam and liquid feed would enter the heat exchanger at the same time and no deviation from the set exiting
temperature would be observed. Therefore, dynamic compensation involves predicting non-ideal behavior and accounting for it. While
perfect feed-forward control is nearly impossible, dynamic compensation is one step closer.
11.2.4: OPEN LOOP SYSTEM

Feed-forward control is a open loop system. In an open loop system, the controller uses current, or live, information of the system to
generate appropriate actions by using predetermined models. The sensor providing the reference command to the closed loop actuator is not
an error signal generated from a feedback sensor but a command based on measurements. This is the defining characteristic of an open loop
system, in which the controller does not manipulate the system by trying to minimize errors in the controlled variable.
Because a feed-forward controller listens to the system and calculates adjustments without directly knowing how well the controlled
variable is behaving (if it does, it is getting feedback), it is open loop in nature. In other words, the controller operates on "faith", that its
mathematical models are able to accurately generate responses (valve setting, motor speed, etc.) that lead to expected results. Therefore, it is
critical to implement good models in feed-forward systems. This is often the most difficult part of a feed-forward design, as mentioned
previously.
However, not all open loop systems are feed-forward. Open loop simply means the system is not getting feedback information, in which the
controlled variable tells the controller how well it is doing compared to its set point (i.e. error). Feed-forward control, on the other hand,
takes an extra step by using non-feedback information to produce predictive actions.
This simple open loop system is an example of a open loop system that is not feed-forward. The controller simply takes the current state of
the controlled variable to generate a valve setting. It is not getting a feedback, because the product temperature is not compared to a set
pont. It is not feed-forward either, because there is no mechanism that produces actions that may yield expected results in the future.
11.2.5: FEED-FORWARD APPLICATIONS

Feed-forward control is used in many chemical engineering applications. These include heat exchangers, CSTRs, distillation columns and
many other applications. A typical furnace, shown below, is heating up an input fluid using fuel gas.
One possible disturbance is the flow rate of the incoming fluid. For example: If the fluid input rate was increased by 10%, then the required
heat duty should also be increased by approximately 10%. This feed-forward control strategy immediately changes the fuel gas flow rate
and therefore the heat duty. The performance of feed-forward controls is limited by model uncertainty, and in practice feed-forward control
is combined with feedback control in order to ensure better control of the system. Feed-forward control will compensate for fluid input
disturbances while feedback control will compensate for other disturbances—such as the fuel gas flow rate or the temperature of the furnace
—and model uncertainty. This system can also be represented by the following block diagram.
11.2.5.1: PROS & CONS OF FEED-FORWARD CONTROL

Different applications require different types of control strategies. Sometimes feed-forward solutions are required for proper system control;
sometimes only feedback solutions are necessary. Feedback systems don’t always maintain the setpoint as well because of the lag that
comes with waiting for the disturbance to propagate through the system. As a result, many control systems use a combination of feed-
forward and feedback strategies, such as PID controllers. PID controllers use the Proportional-Integral control for feedback and the
Derivative control for feed-forward control. This forms a system with multiple loops, otherwise known as a cascading system. A critical
advantage of running both forward and backward controls is that the system is still somewhat able to adjust a variable if one mechanism
fails since the two loops use different sensors. As a result, PID controllers are great for controlling processes, however, they require a
number of equations to determine feed-forward and feedback correction.
Feed-forward systems work by checking the conditions of an incoming stream and adjusting it before the system is adversely affected. If the
controller is told the traits of an acceptable incoming stream, then it can compare that standard to whatever is coming down the pipe. The
feed-forward controller can look at this error and send a corrective signal to the automatic valve responsible for that pipe (or any other
control device). In order to have this kind of predictive ability, the controller must have explicitly defined equations that account for the
effects of a disturbance on the system. In addition, these equations must also then prescribe action to counter-act the disturbance. This can
become even more difficult when there are several incoming stream traits that are being observed. When there are multiple inputs, the feed-
forward system will require non-linear equations, leading to the development of neural networks.
Neural networks are based on “neurons”, which are representations of non-linear equations. This concept is based on the brain's use of
neurons to process and transmit information. The neuron is actually comprised of a set of sigmoidal equations relating inputs to outputs.
Sigmoid functions are non-linear equations that take inputs and apply constants, or weights, to transform the value to make an output.
Below is a picture that illustrates the function that these neurons serve.
The picture above shows a multiple input-output system. Most feed-forward systems have to deal with more than one input.
Neurons can be “connected” in ways that allow the inputs to be transformed any number of times. Neurons that are connected indicate that
one sigmoidal function's output becomes the input of another one. Although the concept of neurons is easily understandable, the difficulty
lies in the potential complexity of real systems. For example: the number and type of inputs go to which neurons, the initial weights be, the
number of neurons needed, etc. As you can see, there are a lot of design questions that make developing the neural network difficult.
Coupled with an expected set of upstream conditions, the feed-forward system can continually adjust the method it uses to control an output
variable. The system is capable of doing this by measuring sensor inputs coming into the controller, using neurons to transform the data, and
comparing the resulting output(s) to a set of known or desired outputs. If the actual output is different from the desired output, the weights
are altered in some fashion and the process repeats until convergence is achieved. This is how a controller is able to "learn". Learning is just
discovering the weighting factors that allow the transformed ouputs to match the desired outputs.
The following table shows a list of feed-forward pros and cons:
Feed-Forward Pros Feed-Forward Cons
Does not require disturbance to propagate through system for detection

Able to predict incoming disturbance before they hit the system Explicit non linear equation required for adequate disturbance prediction. Much
Neutral network utilized to familiarize control system with a changing harder to implement as a result
environment Cannot compensate for any disturbances that occur WITHIN the system
Provides better set point management than feed backward control Requires infinitely accurate models
Can be used in conjunction with teed-back control to provide multiple input Requires infinitely accurate measurements
single input MISO control
A CSTR with a given volume with heat-exchange capability has a coolant water system to maintain a specific system temperature (368K).
To maintain this temperature, the flow of coolant water oscillates. Let a temperature disturbance of 100K be introduced over a period of 10
minutes. For this simulation, the Excel spreadsheet from PID-Tuning Optimization was used to create feed-forward and feed-backward data.
In the following plot, observe the difference between using feed-forward control versus feed-backward control in an example.
As shown, the feed-forward strategy works more effectively than the feed-backward strategy at nullifying the disturbance (over the time
frame). The temperatures reached in feedback mode vary more than the ones seen in feed-forward mode. However, the controller response
is solely defined by the parameters used in the Excel spreadsheet. If one assumes that the PID parameters used by the PID-optimization
group were optimized, then we can say feed-forward would be the best option here.
 FEED-FORWARD DESIGN PROCEDURE
This section provides an outline of the different steps for designing a feed-forward control strategy. The steps pay close attention to
designing a feed-forward controller where there are multiple disturbances. Feed-forward design can be broken down into eight steps
(Corripio,1990).
Step 1. State the control objective. This step includes defining which variable needs to be controlled and what the set point is. The
setpoint should be adjustable by the operator.
Step 2. List the possible measured disturbances. This step includes identifying which disturbances are easily measured and how fast
each disturbance should be expected to vary.
Step 3. State which variable is going to be manipulated by the feed-forward controller.
Step 4. The feed-forward controller consists of two parts: steady-state and dynamic compensators. Develop the steady-state
compensator first. The compensator should be an equation where the manipulated variable, identified in step 3, can be calculated
from the measured disturbances, identified in step 2, and the control objective (set point), identified in step 1.
Step 5. Reevaluate the list of disturbances. The effect of a disturbance on the controlled variable can be calculated from the
equation. Three criteria will be used to determine which disturbance the feed-forward controller will correct: the effect the
disturbance has on the controlled variable, the frequency and magnitude of variation, and the capital cost and maintenance of the
sensor.
Step 6. Introduce the feedback compensation. This depends on the physical significance assigned to the feedback signal.
Step 7. Decide whether dynamic compensation, lead/lag, and/or dead time is required, and decide how to introduce it to the design.
Step 8. Draw the instrumentation diagram from the feed forward control strategy. The details of the diagram depend largely on the
control system being used.
 EXAMPLE 11.2.1
For the first example, a water stream enters an evaporator. A discrete amount of the water stream is to be evaporated. Below is the
design diagram for the evaporator. The incoming stream is being used as the input to the feed forward control box.
Determine what additional input will be needed to regulate the amount of steam produced and draw this on the design diagram.
Solution
By regulating the heating coil, we can regulate how much water will boil off in accordance with any change of the feed stream.
 EXAMPLE 11.2.2
For the second example, a particular process liquid from a previous reaction will be heated before proceeding to the next phase of
reactions. A shell and tube heat exchanger has been installed for this purpose with steam providing the heat duty. A feedback controller
is used to control the output temperature of the process liquid stream, but the flowrate and the temperature of the input liquid stream
vary too much for the feedback controller to be effective.
What sort of controller would be appropriate in addition to the feedback controller to account for the deviations of the input liquid
stream? How would this controller be implemented? For simplicity it is assumed that all the steam condenses.
Solution
The solution is to implement a feed-forward controller that gathers the input temperature and flowrate data from TC1 and FC2 and
adjusts the steam valve V1 in order to account for any deviation that may occur in the input liquid stream. This sort of feedback feed-
forward combination is a typical setup for the majority of control systems. The feed-forward controller accounts for any upstream
disturbances while the feedback controller accounts for the disturbances within the system.
 EXAMPLE 11.2.3
Using the P&ID shown below, write out the feed-forward controller expressions for v3. List any assumptions used in your controller
expression.
Solution
To maintain LC1 by measuring FC1 and FC2 and preemptively adjust v3, we must use the controller expression below.
v3 = kvalves (F C1 + F C2)
The controller expression above assumes that the valve is linear. Also, it assumes that FC1, FC2, and the model are infinitely accurate.
11.2.6: REFERENCES
Svrcek, William Y., Mahoney, Donald P. & Young, Brett R. (2006). A Real-Time Approach to Process Control. John Wiley & Sons, Ltd.
Astrom, Karl J. & Hagglund, Tore (1998). Advanced PID Control. Automatic Tuning of PID Controllers. The Instrumentation Systems
and Automation Society.
Smith, Carlos A. & Corripio Armando B. (1997). Principles and Practice of Automatic Process Control. John Wiley & Sons, Inc.
Authors: Anwar Stephens, Christopher Garcia, Winardi Kusumaatmaja, Meng Yang Ng
11.2: Feed Forward Control is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source content that
was edited to conform to the style and standards of the LibreTexts platform; a detailed edit history is available upon request.
11.3: CASCADE CONTROL
In the previous chapters, only single input, single output (SISO) systems are discussed. SISO involves a single loop control that uses only
one measured signal (input). This signal is then compared to a set point of the control variable (output) before being sent to an actuator (i.e.
pump or valve) that adjusts accordingly to meet the set point. Cascade controls, in contrast, make use of multiple control loops that involve
multiple signals for one manipulated variable. Utilizing cascade controls can allow a system to be more responsive to disturbances.
Before venturing further into the topic of cascade controls, the terms 'manipulated variables', 'measured variables' and 'control variables'
should be clarified. The definitions of these terms commonly found in literature are often interchangeable; but, they typically refer to either
the input or output signal. For the purpose of this article, 'control variables' will refer to inputs like flow rates, pressure readings, and
temperature readings. 'Manipulated variables' and 'measured variables' will refer to the output signals which are sent to the actuator.
The simplest cascade control scheme involves two control loops that use two measurement signals to control one primary variable. In such a
control system, the output of the primary controller determines the set point for the secondary controller. The output of the secondary
controller is used to adjust the control variable. Generally, the secondary controller changes quickly while the primary controller changes
slowly. Once cascade control is implemented, disturbances from rapid changes of the secondary controller will not affect the primary
controller.
11.3.2: CASCADE CONTROL

To illustrate how cascade control works and why it is used, a typical control system will be analyzed. This control system is one that is used
to adjust the amount of steam used to heat up a fluid stream in a heat exchanger. Then an alternative cascade control system for the same
process will be developed and compared to the typical single loop control. The figure below shows the performance of cascade control vs.
single-loop control in CST heater
Cascade control gives a much better performance because the disturbance in the flow is quickly corrected
11.3.2.1: EXAMPLE OF CASCADE CONTROL
Figure 1. Single loop control for a heat exchanger
In the above process, the fluid is to be heated up to a certain temperature by the steam. This process is controlled by a temperature controller
(TC1) which measures the temperature of the exiting fluid and then adjusts the valve (V1) to correct the amount of steam needed by the heat
exchanger to maintain the specified temperature. Figure 2 shows the flow of information to and from the temperature controller.
Figure 2. Flow of information when single loop feedback control is used for a heat exchanger
Initially, this process seems sufficient. However, the above control system works on the assumption that a constant flow of steam is
available and that the steam to the heat exchanger is solely dependent on opening the valve to varying degrees. If the flow rate of the steam
supply changes (i.e. pipeline leakage, clogging, drop in boiler power), the controller will not be aware of it. The controller opens the valve
to the same degree expecting to get a certain flow rate of steam but will in fact be getting less than expected. The single loop control system
will be unable to effectively maintain the fluid at the required temperature.
Implementing cascade control will allow us to correct for fluctuations in the flow rate of the steam going into the heat exchanger as an inner
part of a grander scheme to control the temperature of the process fluid coming out of the heat exchanger. A basic cascade control uses two
control loops; in the case presented below (see Figure 3), one loop (the outer loop, or master loop, or primary loop) consists of TC1 reading
the fluid out temperature, comparing it to TC1set (which will not change in this example) and changing FC1set accordingly. The other loop
(the inner loop, or slave loop, or secondary loop) consists of FC1 reading the steam flow, comparing it to FC1set (which is controlled by the
outer loop as explained above), and changing the valve opening as necessary.
Figure 3. Cascade control for a heat exchanger
The main reason to use cascade control in this system is that the temperature has to be maintained at a specific value. The valve position
does not directly affect the temperature (consider an upset in the stream input; the flow rate will be lower at the same valve setting). Thus,
the steam flow rate is the variable that is required to maintain the process temperature.
The inner loop is chosen to be the inner loop because it is prone to higher frequency variation. The rationale behind this example is that the
steam in flow can fluctuate, and if this happens, the flow measured by FC1 will change faster than the temperature measured by TC1, since
it will take a finite amount of time for heat transfer to occur through the heat exchanger. Since the steam flow measured by FC1 changes at
higher frequency, we chose this to be the inner loop. This way, FC1 can control the fluctuations in flow by opening and closing the valve,
and TC1 can control the fluctuations in temperature by increasing or decreasing FC1set .
Thus, the cascade control uses two inputs to control the valve and allows the system to adjust to both variable fluid flow and steam flow
rates. The flow of information is shown in figure 4.
Figure 4. Flow of information when cascade control is used for a heat exchanger
In order to accomplish this, relationships between the primary and secondary loops (see definitions below) must be defined. Generally, the
primary loop is a function of the secondary loop. A possible example of such relations is:
11.3.2.2: PRIMARY AND SECONDARY LOOPS

In Figure 3, there are two separate loops. Loop 1 is known as the primary loop, outer loop, or the master, whereas loop 2 is known as the
secondary loop, inner loop, or the slave. To identify the primary and secondary loops, one must identify the control variable and the
manipulated variable. In this case, the control variable is the temperature and the reference variable is the steam flow rate. Hence, the
primary loop (loop 1) involves the control variable and the secondary loop (loop 2) involves the reference variable. The information flow
for a two loop cascade control system will typically be as shown in Figure 5. Please note that the user sets the set point for loop 1 while the
primary controller sets the set point for loop 2.
Figure 5: Information flow of a two loop cascade control

In addition to this common architecture, cascade control can have multiple secondary loops; however, there is still one primary loop and a
main controlled variable. Unfortunately, with multiple inner loops, tuning the PID becomes even more challenging, making this type of
cascade less common. The secondary loops can be either independent of each other, or dependent on each other, in which case each
secondary loop affects the set point of the other secondary loop. When tuning such controller, the inner most loop should be tuned first. The
loop that manipulates the set point of the inner-most loop should be tuned next and so for. The figure below shows an example of using two
secondary loops, independent of each other, in a fuel combustion plant. In this combustion furnace, the master controller controls the
temperature in the furnace by changing the set point for the flow of fuels A and B. The secondary loops correspond to the change in the set
point for the flow, by opening or closing the valves for each fuel.
Cascade control is generally useful when

A system error affects the primary control variable only after a long period of time as it propagates through dead time and lag time.
A system has long dead times and long lag times.
Multiple measurements with only one control variable are required for better response to a disturbance of a system.
Variance occurs in multiple streams
11.3.3: GENERAL CASCADE CONTROL SCHEMATIC

The reactor below needs to be cooled during continuous-feed operation of an exothermic reaction. The reactor has been equipped with a
cooling water jacket with the water flow rate being controlled by cold water valve. This valve is controlled by two separate temperature
controllers. An “inner-loop” or “slave” (highlighted in orange) temperature transmitter communicates to the slave controller the
measurement of the temperature of the jacket. The “outer-loop” or “master” (green) temperature controller uses a master temperature
transmitter to measure the temperature of the product within the reactor. The output from the slave controller is fed into the master
controller and used to adjust the cold water valve accordingly.
The cascade control loop used to control the reactor temperature can be generalized with the schematic below. We will use this main
diagram to go through the formal derivation of the equations describing the behavior of the system when there are changes in the loads U1
and U2 but with no change in the set point, R1(t). The general equations derived below can be used to model any type of process (first,
second, third order differential equations, ect.) and use any type of control mechanism (proportional-only, PI, PD, or PID control). See wiki
pages titled “first-order differential equations” and “P, I, D, PI, PD, PID Control” for more details.
Step 1: Write down all the equations for each stage of the control loop
Master Loop
Slave Loop
Gp1 and Gp2 are the process operators and are usually of the form:
Where n is a natural number.

Gc1 and Gc2 are the control operators and depend on the type of controller used. For PID controllers, they would be:
Gm1 and Gm2 are the measurement operators and usually are just equal to 1. Note that there are no equations for the "intersections" A and
B shown on the diagram.
Step 2: Simplify the equations for the slave loop
Solve for Y2(t)
Step 3: Simplify the equations for the master loop
Solve for Y1(t)
(Note: here the G's are written as operators rather than Laplace transforms, and as such they shouldn't be divided. Thus, expressions in the
denominator should be interpreted as inverse operators in the numerator.)
11.3.4: CONDITIONS FOR CASCADE CONTROL

In order to have a smooth flow of information throughout the control system, a hierarchy of information must be maintained. In a double
loop cascade system, the action of the secondary loop on the process should be faster than that of the primary loop. This ensures that the
changes made by the primary output will be reflected quickly in the process and observed when the primary control variable is next
measured. This hierarchy of information can be preserved by applying the following conditions when setting up the cascade controls.
1. There must be a clear relationship between the measured variables of the primary and secondary loops.
2. The secondary loop must have influence over the primary loop.
3. Response period of the primary loop has to be at least 4 times larger than the response period of the secondary loop.
4. The major disturbance to the system should act in the primary loop.
5. The primary loop should be able to have a large gain, Kc.
Cascade control is best when the inner loop is controlling something that happens at fairly high frequency. Cascade control is designed to
allow the master controller to respond to slow changes in the system, while the slave controller controls disturbances that happen quickly. If
set up in reverse order, there will be a large propagation of error. Hence, it is important to maintain the hierarchy of information. In
summary, the master controller responds to SLOW changes in the system, while the slave controller responds to the high frequency, or
FAST changes in the system. This also requires that the inner control scheme be tuned TIGHTLY so error is not allowed to build.
Commonly, the inner loops controls a flow controller, which will reduce the effect of changes such as fluctuations in steam pressure.
11.3.5: CASCADE CONTROL DESIGN CONSIDERATIONS

Open Loop and Closed Loop
Although cascade control generally incorporates information from several secondary loops, the system overall does not automatically
become a closed loop or an open loop process. For a cascade control to be open loop, all control loops in the system should be open loop in
nature. On the other hand, if any of the control loops in the system is feedback-based, the system overall is considered as closed loop. This
is due to the fact that the system is getting some sort of feedback, no matter how little the “fedback” variable influences the system.
Cascade with Feed-Forward and Feedback Controls
Cascade controllers have a distinct advantage over other kinds of controller due to its ability to combine both feedback and feed-forward
controls. While feed-forward loops have the potential to adjust the controlled variables to the ideal states, the feedback loops in a mixed
system check deviations to make sure the system is on track. The figure below is an example of a mixed cascade control:
Figure 6. Cascade control with both feed-forward and feedback controls
An endothermic reaction takes place in a jacketed CSTR, in which the reactor is heated by steam. A reactant stream feeds into the CSTR,
which serves as the wild stream for the ratio controller to predict a required steam flow rate (feed-forward). On the product stream side, a
temperature controller manipulates the ratio setting of the ratio controller in order to minimize the product temperature errors (feedback).
The temperature controller is the primary loop, whereas the ratio controller is the secondary loop.
11.3.6: ADVANTAGES AND DISADVANTAGES OF CASCADE CONTROL

The following table shows a list of cascade control pros and cons:
11.3.7: STARTING UP A CASCADE SYSTEM

A cascade system needs to be set up properly in order to function. The inner loop should be tuned before the outer loop. The following are
the suggested steps for starting up a cascade system (both controllers start in automatic mode):
1. Place the primary controller in manual mode. This will break the cascade and isolate the secondary controller so that it can be tuned.
2. Tune the secondary controller as if it were the only control loop present.
3. Return the secondary controller to the remote set point and/or place the primary controller in automatic mode. This will isolate the
primary controller so that it can be tuned.
4. Tune the primary control loop by manipulating the set point to the secondary controller. If the system begins to oscillate when the
primary controller is placed in automatic, reduce the primary controller gain.
Cascade control systems can be tuned using conventional methods. Ziegler-Nichols tuning method can be used to tune the secondary
controller. Then the parameters (depending on the system, it can be P, PI, or PID) need to be fine tuned by introducing disturbances to the
system and changing the parameters accordingly. The secondary controller must be tuned tightly (no oscillations when disturbance is
added), otherwise the primary controller will be responding to oscillations due to the secondary controller as well as from disturbances to
the system. This will cause the parameters for the primary controller to be inaccurate, which will cause a high controller effort. This is
undesirable because it will wear down the controller. The primary controller often utilizes internal model control (IMC), which allows for
improved performance by incorporating the process model into the controller setup. Another way to tune the primary controller is by using
trial and error. For example, if the system has a PID primary controller, the integral and derivative gains should be set at small values. Then
a small proportional gain should be introduced. The proportional gain should be tuned first then the integral and derivative gains can be
added. For more information about the trial and error method refer to PID Tuning Classical section.
11.3.7.1: STARTUP EXAMPLE

A simple example of starting up a cascade system is shown using the heat exchange system seen in Figure 3. To begin startup the
temperature controller (primary) is set to manual mode. The flow controller (secondary) is then tuned by adjusting the set point of the flow
controller. This is shown in Figure 6 below.
Figure 6: Automatic mode for secondary controller (manual mode for primary controller)
Now the temperature controller is set to automatic mode and the flow controller is set to manual mode. In this system the temperature
controller outputs a set point to the flow controller just like it would in regular cascade mode. The temperature controller is then tuned by
adjusting the output to the valve. This is shown in Figure 7 below.
Figure 7: Manual mode for secondary controller (automatic mode for primary controller)
11.3.7.2: DEVELOPING THE STRUCTURE OF A CASCADE ALGORITHM

Below is a step-by-step method through which the basic structure of a Cascade Algorithm can be developed. The examples in each step
refer to Figure 3.
1 – Determine the goal of your algorithm. That is, which ultimate output you would like to ultimately end up changing. Also determine
what “tool” (or aspect of the system) will most directly and physically allow you to accomplish this desired change.
This “tool” can be any physical property of the system: temperature, pressure, surface area, flow regime, flow rate, etc. of any part of any
physical component of the system. The “tool” doesn’t need to be something read by a sensor, though it can be (as it is in this example, the
temperature of the heated fluid leaving the heat exchanger is read by sensor TC1).
For instance, the end goal of the process of Figure 3 is to regulate the temperature of the fluid leaving a heat exchanger. In this case, the only
way to physically accomplish this is to change the flow rate of the steam through the heat exchanger (this flow rate just so happens to be
read by the sensor FC1).
2 – Determine how the “tool” physically affects the ultimate output.

For instance, an increased flow rate of steam (read by FC1) will physically increase the temperature of the fluid leaving the heat
exchanger, TC1.
3 – Determine how the ultimate output needs to affect the “tool” to achieve the desired ultimate output change.
For instance, we presume that we would like to, in general, resist any change in TC1 (i.e. we don’t want an increase in TC1 to further
increase the value of TC1, that would be bad!). For this reason, we want an increase in TC1 to decrease FC1, which will then decrease TC1
(negative feedback).
4 – Determine which valve or sensor’s output will most directly affect the “tool”. Also, determine how the sensor and “tool” are related.
For instance, the setting of valve V1 will most directly affect the flow rate of steam entering the heat exchanger (it is the "closest"
controllable component of the process to the "tool", the steam flow rate). An increase in V1 will increase flow rate.
5 – Determine all valves/sensors in between the ultimate output and valve/sensor that will directly affect the “tool”.
For instance, between the ultimate output (TC1) and the valve (V1) that will directly affect the “tool”, there is only the FC1 sensor.
6 – Make an incidence graph. Make sure that you use the set points of the valves/sensors that directly affect the “tool”, because you
cannot control what these sensors read (you can only control their set points). Confirm that the relationship between the “tool” and the
ultimate output is consistent with the incidence graph!
7 – Use this incidence graph to construct the algorithm. The incidence graph is particularly helpful in determining what “sign” changes in
one component should result in a the “sign” in another component. If we use incidence diagrams, and the equations for the cascade
controller, we can determine a relationship between the components of the system.
For instance, if TC1 were to increase, we would want the steam flow rate to decrease, which ultimately results in a decreased temperature.
An increase (a “sign” of +) in TC1 should result in a decrease in FC1set (a “sign of -). This can be seen from equation 2, where if TC1set
decreases, the difference between TC1set and TC1 will be smaller. This causes FC1set to decrease relative to its previous set point. A
decrease (a “sign” of -) in FC1set should result in a decrease in V1 (a “sign of -). This is shown by equation 1, where if FC1set decreases
while FC1 is the same, V1 will decrease relative to its previous position. A decrease (a “sign” of -) in V1 should result in a decreased flow
rate (a “sign of -). A decrease (a “sign” of -) in steam flow rate should result in a decreased TC1 (a “sign of -), completing the negative-
feedback mechanism initially desired.
(1)
(2)
11.3.8: FAILURE
A cascade system is not to be confused with a fail-safe system. The sole purpose of implementing a cascade control system is to make the
system more responsive, not more robust.
In a cascade system, if the master controller fails, the entire system will fail too. Just like for any other design, one should anticipate failure
and build in redundancy. An example of redundancy could be having multiple controllers to be the master controller and then using
selectors to choose which reading to use.
Alternatively, if the cascade system fails and has no built-in redundancy, there are a couple of ways to keep the cascade system running
while the controller is being repaired. For example, the controller can be operated manually by an employee, or an average of previous
readings can be used to temporarily send a constant signal to the system.
 EXAMPLE 11.3.1
The diagram below shows the top portion of a distillation column used to separate the components of the inlet liquid mixture. Reflux is
used for the column to improve the separation efficiency, as the desired product stream must have a mass fraction of component C that
is greater than 0.8. At the top of the distillation column the temperature is 200oC. Components A, B, and C, which are gases at that
temperature, accumulate and flow out of the top of the distillation column through the total condenser. The total condenser cools the
gas stream to 150oC, causing A, B, and C to liquefy. The liquid mixture is collected in a distillate receiver. Finally, the liquid stream
exits the receiver and is either collected as product or sent back to the column. A control valve is in place on the product stream to
ensure that the necessary purity is attained.
On the diagram, draw the necessary controllers and valves required to implement a cascade control for the exiting stream. Also,
describe the cascade control that you have implemented.
Note that this is only one possible solution to the example question. Alternatives to this solution would be to only use a composition
controller with a flow controller or a temperature controller but the below solution was chosen to give an example of a multiple cascade
control.
Solution
The system can most effectively be controlled with a cascade scheme. A composition controller can be installed to control the
composition of the product stream. However, the composition controllers are typically slow to react since it takes time to analyze the
samples to obtain compositions. In order to correct for the time lag of the composition controller, a temperature controller, located at
the top of the distillation column, measures the temperature of the product stream giving a rough idea of the product composition.
Additionally, a flow controller is needed on the reflux stream.
The input from the composition controller measures the composition of the product stream and sends a signal to the temperature
controller at the top of the column. The temperature controller compares the set point to the measured temperature of the stream and
outputs a signal to the flow controller on the recycle stream. The flow controller compares this set point to the measured flow rate and
adjust the control valve on the recycle stream accordingly. This cascade control allows the system to react to disturbances quickly and
to more effectively control the reflux process.
 EXAMPLE 11.3.2
You just graduated from the University of Michigan and escaped the cold winter of Ann Arbor for a sweet full-time position in Orange
County. Your company manufactures a specialty drug to cure an epidemic in the penguin population in Antarctica.
The endothermic reaction A + B --> C is run in a CSTR. A and B are toxic in even dilute concentrations. Therefore, a conversion above
99.95% has to be maintained in order to ensure the safety of the process output.
Steam from a boiler is utilized to maintain the optimum reaction temperature of 100C. A PID cascade control scheme is used to
maintain this temperature.
This morning, your plant manager came to you with a problem. He explained that production has been delayed because the conversion
dropped below the acceptable level. He noticed that the reactor temperature was well below 100C, even though the steam valve (V2) is
open completely.
As can be seen in the diagram below, a cascade control is already present in the system, but it isn't set up properly. Can you fix the
problem and save the penguins?
Hint: The boiler also supplies steam to other equipment. You can increase its steam output by adding more fuel to the boiler (assume
that water supply is unlimited).
Solution
Before: TC is cascaded with FC1, which controls V2. (TC-->FC1-->V2) Even when V2 is open all the way, there is not enough steam
to maintain the reaction at optimum temperature.
In this situation the controller surpasses the saturation limit of the actuator (V2). In other words, the controller is demanding more
steam from the boiler than it can supply. A likely cause of this is a disturbance in the steam supply from the boiler because of increased
steam demand from other processes. A possible solution is to increase the steam output from the boiler by adding more fuel (controlled
by FC2 and V4). A side effect of this control situation would be that the I- part of a PID controller would also windup, preventing the
system from responding.
An additional control loop can be added to enable the boiler output to communicate with the fuel input. In addition to controlling V2,
FC1 also controls FC2, which in turn will control V4. This additional controller is shown in the figure below.
Congratulations, you just saved the penguins!
 EXAMPLE 11.3.3
You have a holding tank for a warm process fluid that is cooled by a water-filled jacket. The process stream enters at 75-85 C and
should be cooled to 60 C. The temperature of the cooling water also varies, since it comes from another process, but its average
temperature is about 29 C. You have thermocouples inside the holding tank and on the cooling water jacket exit stream. Unfortunately,
the tank is not well mixed, so the thermocouple reading inside the tank lags behind the actual temperature of the outlet stream. When
the cooling jacket is first turned on, the tank contents are initially at 70 C. Examine the dynamics of a feedback control stream versus a
cascade control scheme for this system.
Solution
One setup for this system is given in the following Excel file: Media:Cascade.xls
Under the feedback control scheme, the flow rate of coolant is adjusted based on the temperature inside the holding tank. The system
overshoots setpoints more significantly, because of the lag time before disturbances are recognized.
In cascade control, the temperature sensor in the process fluid outputs a setpoint for the jacket outlet (based on a knowledge of the
process fluid setpoint and the average temperature of the cooling fluid). The signal from the temperature controller on the jacket outlet
controls the flow rate of cooling fluid. Since the temperature in the cooling fluid has a shorter time lag, the process remains closer to its
setpoint.
 EXAMPLE 11.3.4
We have a P&ID as shown below. Write out the equations necessary to connect LC1 to FC1 to adjust v1. Which equation is the slave
and which one is the master?

Solution
The first equation is the slave and the second equation is the master. We have to use the FC1set to determine what V1 should be set as.
And these are both based on the level in the tank.
(1)
(2)
 EXAMPLE 11.3.5
Write the general equation that describes the behavior of the cascade control system shown below. Include changes in U1 and U2.
Hint: Start by writing the inner loop and outer loop equations separately.
Solution
For inner cycle:
ε1 = Z1 + GM2Y2
Z2 = Z1GC2 − Y2GC2GM2
X2 = U2 + Z2 = U2 + Z1GC2 − Y2GC2GM2
Y2 = GP2X2 = U2GP2 + Z1GP1GC2 − Y2GC2GM2GP2
For outer cycle:
Z1 = R1GC1 − Y1GM1GC1
Y1 = U1GP1 + Y2GP1
Combining Z1 from outer cycle with Y2 from inner cycle:
Combining Y2 above with Y1 from outer cycle:
Rearranging and solving explicitly for Y1:
11.3.9: PRACTICE QUIZ

Answer the five questions below with the most appropriate solution. There is one correct answer for each question. Correct answers are
awarded 4 points, whereas a loss of 2 points occurs for incorrect answers. Good luck!
1. Which of the following is never true in a process controlled by a cascade control scheme?
a) Changes in the secondary control variable affect the primary measured variable.
b) Disturbances in the secondary loops propagate at least four times faster than disturbances in the primary loop.
c) One control variable is used to influence the primary and secondary measured variables.
d) Major disturbances in the process are detected in the primary loop.
2. Which of the following in not an advantage of cascade control?
a) The system reacts to disturbances more quickly.
b) More efficient control of the primary variable.
c) It is a cheap control scheme to implement.
d) The effects of dead time and phase lag time are reduced in the system.
3. True or False: During start-up of a cascade control system, after the primary and secondary controllers are turned and placed in automatic
mode, the primary controller gain should be increased to reduce system oscillations.
4. The primary loop, or outer loop, involves which of the following variables?
a) Control variable
b) Temperature variable
c) Reference variable
d) I love penguins (variable)
5. True or False: The primary loop is generally a function of the secondary loop.
11.3.9.1: ANSWERS
Based on your selections, calculate your score on this quiz.
1. C- Cascade control schemes use multiple control variables for one manipulated variable.
2. C- Increased amounts of equipment and instrumentation may drive up the cost of the process.
3. False- The controller gain (Kc) should be reduced to reduce system oscillations
4. A- The best answer is (a) since the control variable may or may not be temperature. The reference variable is involved in the secondary
loop. Penguins are really cute.
5. True- The output of the primary controller is the set point for the secondary controller. The output of the secondary controller adjusts the
control variable.
11.3.9.2: SCORING
If you got…
20 points: CONGRATULATIONS! You are a cascade genius!
13-19 points: Make a list of confusing topics and bring your questions to a cascade control sage.
0-12 points: You may want to re-read this article and try again…
-10 points: Choose a new major. You may be better off in LS&A.
11.3.10: REFERENCES
Astrom, Karl J. and Hagglund, Torr. "Advanced PID Control" The Instrumentation, Systems and Automation Society.
Bequette, B. Wayne Process Dynamics: Modeling, Analysis, and Simulation Prentice- Hall PTR, Upper Saddle River, NJ 07458 (c)
2006.
Mahoney, D.P., Svrcek, W.Y., and Young, B.R. (2006) A Real-Time Approach to Process Control, 2nd Edition. John Wiley & Sons, Ltd.
ISBN 0470025336
Industrial Controller: basic theory. Retrieved from http://www.sapiensman.com/control/index.htm. Last accessed on 12/12/2008

Stewards: Irene Brockman, Robert Chockley, Scott Dombrowski, Chloe Sweet
11.3: Cascade Control is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source content that was
11.4: RATIO CONTROL
Ratio control architecture is used to maintain the relationship between two variables to control a third variable. Perhaps a more direct
description in the context of this class is this: Ratio control architecture is used to maintain the flow rate of one (dependent controlled feed)
stream in a process at a defined or specified proportion relative to that of another (independent wild feed stream) (3) in order to control the
composition of a resultant mixture.
As hinted in the definition above, ratio control architectures are most commonly used to combine two feed streams to produce a mixed flow
of a desired composition or physical property downstream (3). Ratio controllers can also control more than two streams. Theoretically, an
infinite number of streams can be controlled by the ratio controller, as long as there is one controlled feed stream. In this way, the ratio
control architecture is feedfoward in nature. In this context, the ratio control architecture involves the use of an independent wild feed
stream and a dependent stream called the controlled feed.
Ratio Control is the most elementary form of feed forward control. These control systems are almost exclusively applied to flow rate
controls. There are many common usages of ratio controls in the context of chemical engineering. They are frequently used to control the
flows on chemical reactors. In these cases, they keep the ratio of reactants entering a reaction vessel in correct proportions in order to keep
the reaction conditions ideal. They are also frequently used for large scale dilutions.
11.4.2: RATIO CONTROL BASED UPON ERROR OF A VARIABLE RATIO

The first type of ratio control architecture uses the error of a variable ratio (R ) from a set ratio (R ) to manipulate y (the controlled
actual set
stream). R actualis a ratio of the two variables wild stream and controlled stream. The controller adjusts the flow rate of stream y (controlled
stream) in a manner appropriate for the error (R actual −Rset ).
The error in this system would be represented using the following equations:
y
= Ractual
yw
or
yw
= Ractual
y
with
Error = Ractual − Rset
This error would be input to your general equation for your P, PI, or PID controller as shown below.
1 d(Error)
Vy = Offset + Kc (Error) + ∫ (Error)dt + τD
τI dt
NOTE: Vy is used instead of y, because y is not directly adjustable. The only way to adjust y is to adjust the valve (V) that affects y.
4.2.1 Diagram of Ratio Dependant System. Image taken from Svreck et al.
11.4.3: RATIO CONTROL BASED UPON ERROR OF THE CONTROLLED STREAM

The second type of ratio control architecture uses the error of a setpoint (yset, setpoint of the control variable) from y (controlled variable)
to control, once again, y. The controller adjusts the flow rate of stream y in a manner appropriate for the error (y-yset).
The error in this system would be represented using the following equation:
yw Rset = yset
or
yw
= yset
Rset
with
Error = y − yset
This error would be input to your general equation for your P, PI, or PID controller as shown below.
1 d(Error)
Vy = Offset + Kc (Error) + ∫ (Error)dt + τD
τI dt
4.3.1 Diagram of Flowrate Dependant System. Image taken from Svreck et al.
11.4.4: COMPARING THE TWO TYPES OF RATIO CONTROL

The main difference between the two aforementioned types of ratio control architecture is that they respond to changes in the monitored
variable (Ractual or y - the variables through which the error is determined) differently.
The first method mentioned that defines error as (R_{actual}-R_{set}\) (monitored variable of R ) responds sluggishly when y is
actual
relatively large and too quickly when y is relatively small. This is best explained by examining the equations below.
yw ∂ (R) 1
R = ⟶ = (11.4.1)
y ∂ (yw ) y
y ∂ (R) y
R = ⟶ = − (11.4.2)
yw 2
∂ (yw ) yw
Unlike the first method, the second method mentioned that defines error as y − y (monitored variable of y) does not respond differently
set
depending upon the relative amounts of y (or anything else, for that matter). This is best explained by examining the equations below.
∂ (y)
y = yw Rytt ⟶ = Rytt (11.4.3)
∂ (yw )
yw ∂ (y) 1
y = ⟶ = (11.4.4)
Rset ∂ (yw ) Ryet
11.4.5: DIFFICULTIES WITH RATIO CONTROLLERS

11.4.5.1: STEADY STATE ISSUES
A common difficulty encountered with ratio controllers occurs when the system is not at steady state. Unsteady state conditions can cause a
delay in the adjustment of the manipulated flow rate and thus the desired state of the system cannot be met.
Blend Stations
To overcome the difficulty with steady state, a Blend station can be used, which takes into account both the set ratio and the wild stream. As
can be seen by the diagram below, the Blend station takes in the wild stream flow rate, the set ratio, and the initial set point for the wild
stream (r1) as inputs. Using a weighting factor, called gain, the Blend station determines a new set point for the manipulated stream (r2).
The relationship that determines this new set point can be seen below.
∗
r2 = set ratio [γ r1 + (1 − γ)yw ]
where γ is the gain factor.
Image taken from Astrom and Hagglund article on "Advanced PID Control."
11.4.5.2: ACCURACY ISSUES
Another problem (which is an issue with all "feed-forward" controllers) is that the variable under control (mix ratio) is not directly
measured. This requires a highly accurate characterization of the controlled stream's valves so that the desired flow rate is actually matched.
One way to address this problem is to use two levels of PID control (Cascade Control). The first level monitors the controlled streams flow
rate and adjusts it to the desired set point with a valve. The outer level of control monitors the wild stream’s flow rate which adjusts the set
point of the controlled stream by multiplying by the desired ratio. For example, if stream A’s flow rate is measured as 7 gpm, and the desired
ratio of A:B is 2, then the outer level of control will adjust B’s set point to 3.5. Then the inner level of control will monitor B’s flow rate
until it achieves a flow rate of 3.5.
11.4.6: RATIO CONTROL SCHEMES

The following subsections introduce three ratio controller schematics. Each schematic illustrates different ways that the controlled feed
stream can be manipulated to account for the varying wild feed stream.
11.4.6.1: RATIO RELAY CONTROLLER

The wild feed flow rate is received by the ratio relay and then multiplied by the desired mix ratio. The ratio relay outputs the calculated
controlled feed flow rate which is compared to the actually flow rate of the controlled feed stream. The flow controller then adjusts the
controlled feed flow rate so that it matches the set point (3).
Image adapted from Houtz, Allen and Cooper, Doug "The Ratio Control Architecture"
The mix ratio (Fc/Fw) is not easy to access, so it requires a high level of authorization to change. This higher level of security may be an
advantage so that only permitted people can change the mix ratio and decrease the chance that an accidental error occurs. A disadvantage is
that if the mix ratio does need to be changed quickly, operation may be shut down while waiting for the appropriate person to change it.
Another disadvantage is that linear flow signals are required. The output signals from the flow transmitters, Fw and Fc, must change linearly
with a change in flow rate. Turbine flow meters provide signals that change linearly with flow rates. However, some flow meters like orifice
meters require additional computations to achieve a linear relationship between the flow rate and signal.
11.4.6.2: FLOW FRACTION CONTROLLER
Flow fraction and ratio relay controllers are very similar except that the flow fraction controller has the advantage of being a single-input
single-output controller. A flow fraction controller receives the wild feed and controlled feed flow rates directly. The desired ratio of the
controlled feed to wild feed is a preconfigured option in modern computer control systems (3).
11.4.6.3: RATIO RELAY WITH REMOTE INPUT

The ratio relay with remote input model of a ratio controller is similar to the cascade control model in that it is part of a larger control
strategy. The purpose of this model is to account for any unexpected or unmeasured changes that have occurred in the wild and controller
feed streams. The diagram below illustrates how the composition of the mixed flow is used to change the relay ratio (3).
The main advantage of having remote input is that the mix ratio is constantly being updated. A disadvantage of using an additional analyzer
is that the analysis of the mixed stream may take a long time and decrease the control performance. The lag time depends on the type of
sensor being used. For example, a pH sensor will most likely returned fast, reliable feedback while a spectrometer would require more time
to analyze the sample.

There are some pros and cons to using ratio control where there is a ratio being maintained between two flow rates.
1. Allows user to link two streams to produce and maintain a defined ratio between the streams
2. Simple to use
3. Does not require a complex model
1. Often one of the flow rates is not measured directly and the controller assumes that the flows have the correct ratio through the control
of only the measured flow rate
2. Requires a ratio relationship between variables that needs to be maintained
3. Not as useful for variables other than flow rates
11.4.8: SELECT ELEMENTS IN RATIO CONTROL
A select element enables further control sophistication by adding decision-making logic to the ratio control system. By doing so, a select
variable can be controlled to a maximum or minimum constraint. The figure below depicts the basic action of both a low select and high
select controller.
11.4.8.1: SINGLE SELECT OVERRIDE CONTROL

Often times, these select elements are used as an override element in ratio control. Take, for example, ratio control system (with remote
input) of a metered-air combustion process in the figure below.
In this design, the fuel set point, SPw, comes in as firing demand from a different part of the plant, so fuel flow rate cannot be adjusted
freely. As the fuel set point (and therefore fuel mass flow rate, PVw) fluctuates, a ratio relay is employed to compute the combustion air set
point, SPc. If the flow command outputs (COw and COc) are able to respond quickly, then the system architecture should maintain the
desired air/fuel ratio despite the demand set point varying rapidly and often. However, sometimes the final control element (such as the
valve on the fuel feed stream and blower on the combustion air feed stream in the diagram) can have a slow response time. Ideally, the flow
rates would fluctuate in unison to maintain a desired ratio, but the presence of a slow final control element may not allow the feed streams to
be matched at that desired ratio for a significant period of time. Valves often have quick response times, however, blowers like the one
controlling the combustion air feed stream can have slow response times. A solution to this problem is to add a “low select” override to the
control system, as shown in the figure below.
The second ratio controller will receive the actual measured combustion air mass flow rate and compute a matching fuel flow rate based on
the ideal design air/fuel mixture. This value is then transmitted to the low select controller, which also receives fuel flow rate set point based
from the firing demand. The low select controller then has the power to pass the lower of the two input signals forward. So, if the fuel flow
rate firing demand exceeds combustion air availability required to burn the fuel, then the low select controller can override the firing
demand and pass along the signal of the fuel flow rate calculated based on the actual air flow (from the second ratio relay).
This low select override strategy ensures that the proper air/fuel ratio will be maintained when the firing demand rapidly increases, but does
not have an effect when the firing demand is rapidly decreasing. While the low select override can help eliminate pollution when firing
demand rates increase quickly, a rapid decrease in firing demand can cause incomplete combustion (as well as increased temperature) and
loss of profit.
11.4.8.2: CROSS-LIMITING OVERRIDE CONTROL

Cross-limiting override control utilizes the benefit of multiple select controls. Most commonly, a control system using the cross-limiting
override strategy will implement both a high select and low select control. As a result, the system can account for quickly increasing firing
demand as well as quickly decreasing firing demand. The figure below depicts a such a scenario using the inlet combustion air and fuel for a
boiler as the example.
It is assumed that same firing demand enters the high select and low select control and the flow transmitters have been calibrated so that the
ideal air/fuel ratio is achieved when both signal match. As a result, the set point of air will always be the greater value of the firing demand
and current fuel flow signals. Likewise, the set point of fuel will always be the lower value of the firing demand and current air flow signals.
So, as the firing demand increases, the set point of air will increase. At the same time, the low select will keep the set point of fuel to the
signal set by the present flow or air. Therefore, the set point of fuel will not match the increasing firing demand, but will follow the
increasing air flow rate as it is responding upwards. Similarly, if the firing demand decreases, the low select control will listen to the firing
demand and the high select controller will not. As a result, the firing demand will directly cause the set point of fuel to decrease while the
set point of air will follow the decreasing fuel rate as it responds downwards. The cross-limiting override strategy allows for greater balance
in a ratio control system.

In the following process shown below, a concentrated solution of product is diluted continuously to be sold as a final 10% solution.
Flowrates coming from the unit feeding the pure product to this mixing tank are not constant (the wild stream), and therefore a ratio
controller is used to properly dilute the solution.
In this example, the ratio controller would be set to a value of FC1/FC2 = RC1 = 9. The ratio contoller in this case would then work by the
following logic:
For an additional challenge, determine an error equation that can is related to this controller.

One of the most common uses of a ratio controller is used for correcting the air-fuel ratio in automobiles. A specified amount of air is
always taken up by the engine and the amount of fuel injected is controlled by a computer through the use of a ratio controller. The
stoichiometric ratio for air to gasoline by mass is 14.7:1. There is an oxygen sensor in the exhaust manifold as well as sensors for all of the
exhaust components. This sensor is connected to the valve for the fuel injection. If there is too much oxygen measured in the exhaust, then
the mixture is said to be lean. Is the ratio of air to gas in the original mixture greater than or less than 14.7:1 and what should the valve
controlling the fuel injection do?
Note: This is an indirect form of a ratio controller as it measuring a physical property of downstream in order to determine if the ratio is too
high or low. Further information may be found here about Air-fuel Ratio and Oxygen Sensors.
Answer: Since there is leftover oxygen in the exhaust, there must have been too much in the original mixture. The ratio of air to gas in the
original mixture is greater than 14.7:1 and thus the computer should open the valve more for the fuel injection.

Many processes in industry require multiple process streams entering at different locations in the system. A senior engineer gives you the
project to design the control scheme for a new process to be implemented at the company. The process involves 2 reactions and 3 different
feed streams.
Reactions: 2A + B --> D
C+3D ⟶ E
Assume the reactions go to 100% completion (you only need stoichiometric amounts).
Complete the P&ID with the control scheme and write the ratios (with respect to B) that the other molar flow rates will be set to. Also
explain how the system will take action when it detects excess C or D in the product stream.
Solution:
A: 2 B: 1 C: 1 D: 1/3 E: 1/3
If the AC detects greater or less amounts of C expected, it will adjust the ratio setting in RY connected to C's flow controller, reducing or
decreasing the ratio of C to B, respectively.
 EXERCISE 11.4.1
Why is a Blend Station used?

a. to create the set point for the ratio controller
b. to reduce lag time in adjusting flow rates
c. to measure flow rates
d. to multiply one flow rate by the desired ratio
Answer
TBA
 EXERCISE 11.4.2
How is the set ratio used in the first type of ratio controller described above?
a. Outputs how much each stream should be flowing.
b. It is compared against two flow rates to adjust one if needed.
c. Tells the system to shut down.
d. It adjusts for lag time when the system is not at steady-state.
Answer
Add texts here. Do not delete this text first.
11.4.12: REFERENCES
Astrom, Karl J. and Hagglund, Torr. "Advanced PID Control" The Instrumentation, Systems and Automation Society.
Mahoney, Donald P., Svrcek, William Y. and Young, Brent R. "A Real-Time Approach to Process Control", Second Edition. John Wiley
& Sons, Ltd.
Houtz, Allen and Cooper, Doug "The Ratio Control Architecture" http://www.controlguru.com
(http://www.controlguru.com/2007/120207.html)
11.4: Ratio Control is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Stephanie Fraley, Michael Oom, Benjamin Terrien,
John Zalewski, & John Zalewski via source content that was edited to conform to the style and standards of the LibreTexts platform; a detailed edit history
is available upon request.
11.5: COMMON CONTROL LOOPS AND MODEL FOR LIQUID PRESSURE AND
LIQUID LEVEL
11.5: Common control loops and model for liquid pressure and liquid level is shared under a CC BY 3.0 license and was authored, remixed, and/or curated
by Peter Woolf et al. via source content that was edited to conform to the style and standards of the LibreTexts platform; a detailed edit history is available
upon request.
11.6: COMMON CONTROL LOOPS AND MODEL FOR TEMPERATURE
CONTROL
Temperature, pressure, flow, and level are the four most common process variables. Temperature is important because it provides a critical
condition for processes such as combustion, chemical reactions, fermentation, drying, distillation, concentration, extrusion, crystallization,
and air conditioning. Poor temperature control can cause major safety, quality, and productivity problems. Although highly desirable, it is
often difficult to control the temperature because its measurement must be within a specified range of accuracy and have a specified degree
of speed of response, sensitivity, and dependability. Additionally, temperature measurements must also be representative of true operating
conditions in order to achieve successful automated control. The instrument selected, installation design, and location of the measuring
points determine these specifications.
This article will serve as a guide to select the best location of measuring points to achieve the best automatic control. It will consider
temperature control for three common process types: a CSTR, distillation column, and heat exchanger.
11.6.1.1: TEMPERATURE CONTROL LOOPS
Before temperature control loops for specific processes are explained, we must discuss the general considerations common for all
temperature control loops.
Temperature control loops can either be endothermic (requiring heat energy) or exothermic (generating heat energy). Both types are similar
in that they both result in a response representing a process with a dominant capacitance plus a dead time. For both types of processes, one
of the following devices is used to measure temperature:
thermocouple
filled thermal well system
Resistance temperature detector (RTD)
The measurement device, or thermal well, should be selected so that is minimizes additional lag to the overall process lag. Minimizing
temperature measurement lag in the temperature control loop is important in both slow and fast loops. Some general rules of thumb for
reducing temperature measurement lag are:
1. Use a small-diameter bulb or thermal well to minimize the thermal resistance and thermal capacity of the measuring element.
2. Use a thermal well made from a material that minimizes thermal resistance and thermal capacity of the measuring element.
3. Use a small pipe or orfice near the measuring device to increase velocity of the passing flow. Increasing flow will increase the rate of
heat transfer between the process fluid and the measuring device.
4. Place the measuring element in the liquid phase when measuring temperature in a two-phase system since thermal resistance is smaller
in the liquid phase than in the vapor.
5. Use a transmitter with derivative action to cancel out some of the lag in the measuring element. Compensate for this added derivative
gain in the transmitter by reducing the derivative gain in the controller.
11.6.2: CSTR TEMPERATURE CONTROL

11.6.2.1: ENDOTHERMIC REACTOR TEMPERATURE CONTROL LOOPS
Endothermic CSTR reactors are generally easier to control than exothermic CSTR reactors because when the temperature reaches a critical
minimum, the reaction does not proceed until there is adequate heat. In this sense, endothermic CSTR reactors are self-regulating. A good
way to think of an endothermic CSTR’s controls is that of a heat exchanger being used to heat a reaction solution. This heat exchanger’s
response is controlled by the dead time, so typically a PI or PID controller is used.
There are two main types of temperature control methods for endothermic CSTRs: control via steam flow rate and control via steam
pressure. For the steam flow rate control case, the temperature control sends a signal to the flow control, which then controls the steam inlet
valve. For the steam pressure control case, the temperature control sends a signal to the pressure control, which controls the steam inlet
valve. Examples of an endothermic CSTR temperature control loop can be seen below in Figure 1 a-b.
Image Modified From: Riggs, James B., Karim, Nazmul M.. Chemical and Bio-Process Control. Third Edition Chapter 18. Ferret
Publishing.
11.6.2.2: EXOTHERMIC REACTOR TEMPERATURE CONTROL LOOPS
In an exothermic reaction, energy is released in the form of heat. In some cases, a cooling system is require to bring the temperature back to
a set point and also to ensure that the temperature does not rapidly increase, or runaway. Some steps can be taken to prevent runaway
temperatures, such as reducing the feed rates or concentrations of reactants. Also, the ratio of the heat transfer area to the reactor volume can
be increased to help increase the controllability of the CSTR. Typically, the temperature is controlled using a PID controller, which is
described in the previous section PID Control.
Exothermic CSTRs are very difficult to control because they are very unstable and the temperature can easily runaway. The relationship
between heat generation and temperature is non-linear, while the relationship between heat removal and temperature is linear. This
relationship is illustrated in Figure 2. The stability of the temperature control loops depend on the rate at which heat can be removed from
the system. For example, an exothermic CSTR that generates heat at a slow steady rate is more stable than a reactor that rapidly produces
heat. In this example, the rate at which the heat can be removed from the system depends on the rate at which the temperature can be
changed in the cooling jacket surrounding the CSTR.
Figure 2. Heat generation or removal as a function of temperature in an exothermic CSTR

The intersections of the two curves, labeled A, B and C, represent steady state operating points. A and C are stable operation points while B
is unstable. A and C are stable because as the temperature increases from this point, the rate of heat generation is less than the rate of heat
removal. This means that as the temperature increases, heat is removed faster than it is generated so the temperature will be brought back
down to that operating point. The same will happen if the temperature decreases from that point. As the temperature decreases, the heat
generation rate is greater than the heat removal rate so the temperature will be brought back up to the operating point.
Point B is unstable because when the temperature increases from that point, the heat generation is greater than the heat removal. This means
that as the temperature increases, heat is continuously added to the system and the temperature will rise until it reaches one of the stable
operating points, in this case C. If the temperature decreases from point B, the heat generation rate is less than the heat removal rate so the
temperature will continue to decrease until it reaches the lower, stable operating point, A.
This graph would be similar for an endothermic CSTR but there would be only one stable steady state operating point.
Very rapid exothermic reactions are the most difficult to control and they are sometimes carried out in a semi-batch reactor so the addition
of reactants can be carefully controlled and runaway temperatures can be avoided.
Figure 3.Temperature control loops for an exothermic CSTR
Figure 3 illustrates a common control loop for an exothermic CSTR with a cooling jacket. A temperature transmitter (TT) sends a signal to
the temperature controller (TC), which controls the hot and cold water valves on the jacket. The cold and hot water are pumped into the
jacket which controls the reactor at a set temperature.
11.6.3: TEMPERATURE CONTROL IN DISTILLATION

In a distillation column, temperature control is used as a means to control composition because temperature sensors are cheaper, more
reliable, provide continuous measurements, and respond quicker than composition analyzers. Through equilibrium relationships,
temperature measurements can be used to infer composition of the product. Inferential temperature control is only effective when the
relative volatility of the components is high (greater than 2.0). Temperature controllers are used as feedback composition controllers to
adjust column operation to meet production requirements. They must be able to satisfy the constraints defined by production requirements at
all times, even in the face of disturbances.
11.6.3.1: INFERENTIAL TEMPERATURE CONTROL
To control composition via temperature, a correlation must be made between temperature in the tray and composition of the key
components. Determining a correlation is challenging because the temperature-composition relationship is affected by process nonlinearity
and disturbances in feed composition, flowrates, and occurrence of entrainment or fouling. For multi-component separations, temperature
does not determine a unique composition, so either an online composition analyzer or periodic lab tests must be utilized to verify
composition at the temperature setpoint. Additionally, the column pressure greatly affects the measurement of tray temperature. For most
systems, the following linear equation can be used to correct for variations in pressure:
Tpc = Tmeas − Kpr (P − P0 ) (11.6.1)
Where,
=corrected temperature
=measured tray temperature
=pressure correction factor
=column pressure
=reference pressure
The pressure correction factor, , can be estimated using a steady-state column simulator for two different operating pressures and the
equation:
Ti (P1 ) − Ti (P2 )
Kpr = (11.6.2)
P1 − P2
where T is temperature of tray i predicted by the column simulator.

i
Successful temperature control in the column depends on the dynamic response of measuring the tray temperature with respect to the
manipulated energy source used to actuate temperature. The energy source is either the reboiler or the reflux. To have tight process control
means that the equivalent dead time in the loop is small compared to the shortest time constant of a disturbance with significant amplitude.
According to Svreck, et al, the following observations from experimental tests are cited:
Temperature control is made less stable by measurement lag or response times.
The speed of response and control stability of tray temperature, when controlled by reboil heat, is the same for all tray locations.
The speed of response and control stability of the tray temperature, when controlled by reflux, decrease in direct relation with the
number of trays below the reflux tray.
When pressure is controlled at the temperature control tray, the speed of response of the temperature control instrument can vary
considerably with tray location, and is normally slower.
To achieve the best composition control, you need to determine the tray(s) whose temperature(s) show the strongest correlation with product
composition. The following procedure using a steady-state column model can be used:
1. Run the column at the base conditions (x BC

and y BC
at steady-state and record the temperature, , of each tray.
2. Increase the impurity level in the bottoms product ( and , so that Δx is about 25-50% of the impurity level for the base
case. Record the temperature, , of each tray.
3. Increase the impurity level in the overhead product and , so that Δy is about 25-50% of the impurity level for the base
case. Record the temperature, , of each tray.
4. The best tray for inferential temperature control of the stripping section will be the least sensitive to variations in the composition of the
overhead product. This tray is the one that maximizes:
net Δx BC Δy BC
ΔTi = ( Ti − Ti ) − (Ti − Ti ) (11.6.3)
5. The best tray for inferential temperature control of the rectifying section will be the least sensitive to variations in the composition of the
bottoms product. This tray is the one that maximizes:
net Δy BC Δx BC
ΔT = (T −T ) − (T −T ) (11.6.4)
i i i i i
6. Repeat this procedure for a range of feed compositions. Then, select the tray that maximizes the equations most often for the
range of feed compositions.
11.6.3.2: SINGLE COMPOSITION CONTROL

Once you have determined the best tray to use for inferential temperature control, you must choose a manipulated variable (MV) that will
have a significant gain on the tray temperature, and thus on the composition. Usually in distillation, the composition of only one product
stream is controlled while the composition of the other product stream is allowed to drift. This is called single composition control. The
chemical industry primarilly uses sing composition control.
When you are interested in controlling the composition of the overhead product, the reflux is used to control the purity, while the reboiler
duty is held constant. The bottom composition is not directly controlled and will vary as the feed composition varies. The reflux, L, is the
manipulated variable that will result in the tightest control, rather than either the distillate flowrate, D, or the reflux ratio, L/D. This is
because L is the fastest-acting MV for the overhead and is the least sensitive to changes in the feed composition This scenario is depicted
below in Figure 4a.
Figure 4a. Control diagram for inferential temperature-single composition control of overhead product stream by L.
Figure 4b. Control diagram for inferential temperature-single composition control of bottom product stream by V.
When you are interested in controlling the composition of the bottom product, the boilup rate, V, is used to control the purity, while the
reflux rate is held constant. The overhead composition is not directly controlled and will vary as the feed composition varies. V is the
manipulated variable that will result in the tightest control, rather than either the bottoms product flowrate, B, or the boilup ratio, V/B. This
is because V is the fastest-acting MV for the bottoms and is the least sensitive to changes in the feed composition This scenario is depicted
above in 4b.
11.6.3.3: DUAL COMPOSITION CONTROL

Although single composition control is most commonly used, there are some industries that need to simultaneously control the composition
of both the overhead and bottom product streams. This is called dual composition control and is harder to implement, tune, and maintain
than single composition control. Dual composition control increases product recovery and reduces utility costs and is used for refinery
columns that generally have high associated energy usage. Industries that have refineries, such as the oil industry, use dual composition
control because they are interested in using both the overhead and bottom streams as product.
In dual composition control, there are many possible control configurations. The control configuration is described by the control objective
and by the MV. There are four possible control objectives: bottom product x, overhead product y, reboiler level, and accumulator level. The
chosen control objectives are then paired with a MV. There are a variety of choices for the MV. These include: L, D, L/D, V, B, V/B, B/L,
and D/V. Therefore, in dual composotion control, there are many possible configurations to consider, although most are not practical.
The most commonly used configuration for dual composition control is (L, V) which uses L to control y (overhead composition) and V to
control x (bottom composition), because it provides good dynamic response, is the least sensitive to changes in feed composition, and is the
easiest to implement. However, it is highly susceptible to coupling. In this configuration, depicted below, the setpoint for the reflux flow
controller is set by the overhead composition controller, and the setpoint for the flow controller on the reboiler duty is set by the bottom
composition controller. D is then used to control the accumulator level, and B is used to control the reboiler level.
Figure 5. Control diagram for inferential temperature-dual composition configuration (L,V), the control of overhead product stream by L
and bottom product stream by V.
Since there are so many possibilities for configurations, there is no clear best choice for a configuration of MVs in dual composition control
in distillation columns. It is impossible to know theoretically which configuration is optimum for a particular process, but there are some
rules of thumb to follow to increase the possibility of choosing a good configuration. These rules are summarized in the chart below.
Figure 6. Chart to determine a good configuration for your process when using dual composition inferential temperature control
For the gray area where 5<L/D<8, follow the rules of thumb for the closest L/D, and use experimental data to determine the best
configuration for your particular process.
11.6.3.4: CONTROLLER TUNING AND CONSTRAINTS

When temperature is used for inferential composition control, a PID controller is usually used because of the significant sensor lag.
Common constraints on the extent of temperature control in a distillation column include:
Capacity of reboiler power
Capacity of overhead condenser
Flooding or entrainment
Maximum (undesired) column temperature
In general, temperature control can be used to control composition in a distillation column. The temperature measuring device should be
chosen to minimize the lag time. A steady-state model can be used to correct for pressure changes in the column and to determine the best
tray at which to measure temperature. The choice of the MV depends on if you are operating under single or dual composition control and
which product composition you want to control.
11.6.4: HEAT EXCHANGER CONTROL

In heat exchanger control, the temperature of the process exit stream is the controlled variable (CV) and can be adjusted by one of four
possible manipulated variables: cool side entry stream, cool side exit stream, hot side entry stream, or hot side exit stream. The selection of
where control will be implemented is a combined result of three factors:
1. Whether we want to heat or cool the process stream
2. The response time of the controller
3. The capital cost
To illustrate the selection process, first it is shown that the hot stream preferred for temperature control, then two case studies are discussed,
one for heating up the process stream and the other for cooling the process stream.
11.6.4.1: CONTROLLING THE COOL SIDE STREAM

It is not advisable to place temperature control on the cool side stream. This is explained below.
The particular MV in each stream is the flow rate. It is known that the temperature of the hot stream will change once the flow rate of the
cool stream is adjusted.
Figure 7. Image modified from Riggs & Karim, shows coolant temperature control loop for a Liquid/Liquid Heat Exchanger.
It is easier to think of the temperature change as the magnitude of the process gain (| |). This just is the magnitude of the
temperature change over a specified interval. This way the discussion can be general for both heating or cooling the process stream.
Figure 8. Taken from Riggs & Karim, demonstrates the limit that increasing the flow rate of the coolant stream has on the exit temperature
of the process stream for three heat exchanger areas
As the rate of the cool stream is increased from zero, the magnitude of the process gain continuously decreases. Above a certain rate, the
gain will not be a measurable change in the process stream temperature. Because of this, making the coolant stream the MV will make the
process uncontrollable at a critical point and is not a good idea.
11.6.4.2: CONTROLLING THE HOT SIDE STREAM

As indicated above it is never recommended to place the temperature control on the cool side stream. This reduces controls consideration to
only the hot stream. There are two cases where the control can be on the hot stream; when heating up and when cooling down the process
stream.
Heating Up The Process Stream
CASE STUDY 1: Steam-Heat Heat Exchanger
In this case the process stream is the cooler of the two entering streams, the previous section states that the MV would have to on the stream
used to heat up the process stream.
The temperature control loop can either be the direct steam flow or the steam pressure. Both controls the amount of steam flowing into the
heat exchanger by adjusting a valve, however, the preferred of the two is the steam pressure as the MV in the temperature control loop. This
is because a change in the steam supply can quickly cause the internal pressure of the heat exchanger to change hence affecting the
temperature exchange. The steam pressure control responds and corrects this faster than the steam flow control. This becomes especially
useful for heavy duty requirements when the change in response to change in flow rate of the process stream. The two figures below show
two preferred configurations;
Figure 9. Image modified from Riggs & Karim, shows one arrangement for temperature control for heating up a process stream
Figure 10. Image modified from Riggs & Karim, shows an alternative arrangement for temperature control for heating up a process stream
Notice that Figure 10 shows the control valve on the condensate stream. This is an acceptable control placement, however, in most practical
application the first of the configurations in Figure 9 is preferred. In addition there are some ramifications to making the condensate flow
the MV. The temperature loop in this case would not be as responsive as in the previous case. This is because the level responds slower than
the pressure to the changes in the respective valves. The capitol cost involved in placing the valve in the condensate stream is much lower
that that in the steam stream because the steam stream is generally larger and requires a larger, more expensive valve. In addition a steam
trap is required down stream of the condensate valve, for that configuration.
Cooling Down The Process Stream
CASE STUDY 2: Liquid/Liquid Heat Exchanger
In this case the process stream is the hotter of the two entering stream so the controls will have to be on this stream. As in case one we are
setting up the temperature control loop for varying flows in the process stream because in a plant the flow rate of this stream is dependent
on a process before the heat exchanger. Hence when creating control loop for cooling down the process stream the coolant flow is kept
constant (as stipulated in the section "Coolant Stream Controls") and maintain the CV at the set point. The best choice for the MV in this
temperature control loop is having a feed bypass stream, as illustrated in Figure 11 below. The advantages of having this bypass stream are
that the process dynamics are considerably faster with low levels of dead time and this condition is more linear, as opposed to varying the
coolant flow rate. With this configuration the coolant flow rate can be maintained very high which reduced the tendency for fouling on the
process fluid side of the heat-transfer surface.
Figure 11. Image modified from Riggs & Karim, shows the common arrangement for temperature control for cooling down a process stream
 EXAMPLE 11.6.1
a. Is an exothermic CSTR easier or harder to control than an endothermic CSTR? What are some reasons for this, and what are a few
steps that can be taken to increase control?
b. In Graph 1, why are points A and C stable operating points, while point B is an unstable operating point?
Solution:
(a) An exothermic CSTR is harder to control. The relationship between heat generation and temperature is non-linear for an exothermic
system (Graph 1). If the rate of heat generation exceeds the maximum rate of heat removal then the temperature will runaway.
An endothermic CSTR is much easier to control because these reactors are self-regulating and there is no concern for runaway
temperatures.
The control of an exothermic CSTR can be increased by reducing the feed flow of the reactants, reducing the concentration of reactants
and increasing the ratio of the heat transfer surface area to the reactor volume.
(b)
Figure E1-1. Relationship between heat generation or heat removal and temperature for an exothermic system.
Points A and C are stable because if you increase the temperature from this point, the heat generation rate is lower than the heat
removal rate. If you increase the temperature from point B, the heat generation rate is higher than that heat removal rate, shich is not
desired. If the rate of heat generation is greater than the heat removal then the temperature is rising and could possible runaway. This is
an unstable state for the reactor.
 EXAMPLE 11.6.2
Refer to Heat Exchanger Modeling Worked Example 1 problem statement.

After you successfully created your ODE model, you are assigned a partner who will work along with you to put temperature controls
on the heat exchanger. Your partner, a new college graduate, seems to remember that in his process controls class, the best way to
control a temperature loop for a heat exchanger is to make the flow rate of the coolant stream the manipulated variable. Since you are
such a process guru, you know that this is not right. However, the hot shot college graduate just feels that you are over the hill and he
knows his stuff.
Vary the coolant flow rate in the Euler's Method heat exchanger to demonstrate to your new partner why it is not a good idea to place
the controls on the coolant side flow of a stream. Produce a process gain plot similar to that in the article to show this. Also, find the
magnitude of the process gain for three different points in the plot: low, medium and high coolant flow rate, to show process gain goes
to zero as the coolant flow rate increases.
Solution
Note that in this case we are dealing with a liquid/liquid heat exchanger and the hot stream is the process stream. The table below
summarizes the results.
Table E2-1. Process stream exit temperature as a function of coolant flow rate.
Figure E2-1. A continuous picture of how the magnitude of process gain decreases to zero as the coolant flow rate is increased.
From the graph or table above one can show that the magnitude of process gain decreases as the coolant flow rate increases. This is
done as follows;
Magnitude of process gain for;
Low flow rates, choose from 0 to 0.1 m3 / s.
= |330 - 274.50| = 55.5K
Medium flow rates, choose (same interval size) from 0.4 to 0.5 m3 / s.
= |262.01 - 261.20| = 0.81K
High flow rates, choose (again same interval size) from 0.9 to 1 m3 / s.
= |259.798 - 259.626| = 0.172 K
Clearly, as the flow rate increases, the magnitude of process gain decreases towards zero. This puts a limit on the amount of control of
the heat exchanger if the coolant flow rate is chosen as the MV in the temperature control loop.
What two types of endothermic CSTR temperature control methods are there?
a. steam flow rate and reactor level
b. reactor composition and reactor level
c. steam flow rate and steam pressure
d. reactant flow rate and reactor level

What are temperature controllers commonly used for in distillation columns?
a. To control composition
b. To control reflux ratio
c. To control pressure
d. None of the above
11.6.7: REFERENCES
Svrcek, W., Mahoney, D., Young, B.. A Real-Time Approach to Process Control. Second Edition. John Wiley and Sons.
Riggs, James B., Karim, Nazmul M.. Chemical and Bio-Process Control. Third Edition Chapter 18. Ferret Publishing.

Authors: Danesh Deonarain, Carly Ehrenberger, Georgina Mang, Teresa Misiti
11.6: Common Control Loops and Model for Temperature Control is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter
request.
11.7: COMMON CONTROL ARCHITECTURES AND MODEL FOR REACTORS
Reactors are the central focus of many chemical plants. Many parameters must be controlled in a reactor for proper operation. Temperature
is of great importance because it affects reaction rates and equilibrium relationships. A major challenge for temperature control is handling
the nonlinear nature of temperature inside most reactors. Therefore, it is important to design an effective control architecture in order to
ensure optimal operation of the reactor.
This article discusses the common control architectures and topologies in CSTRs. The control architectures are designed based on whether
the reactor is endothermic or exothermic. The same concepts introduced in the wiki can be applied to other reactors as well. However, only
CSTRs will be discussed for simplicity.
11.7.1: COMMON TOPOLOGIES

Here we will introduce a few of the most common control topologies which will be examined in applications to endothermic and
exothermic CSTRs below. Please refer to the specific wiki pages for each type for detail.
11.7.1.1: FEEDBACK AND FEED-FORWARD

Feedback and feed-forward refer to the direction in which the sensor information is transfered to an actuator valve. Feedback control
dictates that sensor information is "fed back" to a previous part of the process. For example, the reading from a level sensor of a filling tank
can be "fed back" to the valve controlling the input to the tank. Feed-forward control means the sensor information is used to control
something downstream from where the reading was taken. For example, the measured flow rate of water going into an evaporator can be
used to control the heating coil inside of the evaporator. For detail about these two control topologies, please see the respective pages:
Feedback Control Feed-Forward Control.
11.7.1.2: RATIO CONTROL

Ratio Control is used when the ratio between two measured process variables has an optimal value. In the context of two input streams with
optimal flow ratio going into a reactor, one stream is designated as the control stream, and one stream is designated as the wild stream. The
wild stream fluctuates, and a valve on the control stream is opened or closed to maintain the ratio between the two stream flows. For detail
about ratio control, see Ratio Control.
11.7.1.3: CASCADE CONTROL
Cascade control simply means that instead of one control loop found in simple control topologies where a sensor's measurement directly
controls an actuator valve, multiple loops are used so that sensors' measurements can control set points for other controllers. For example,
the temperature sensor measurement of process fluid exiting a reactor can be used to modify the set point of the flow controller of steam
feeding the heating jacket, which then sets the steam valve. This multiple-loop system eliminates some problems caused by variable-
pressure feeds, for instance. For detail about cascade control, see Cascade Control.
11.7.2: DISTURBANCES TO CSTRS

There a few very common disturbances that CSTR may be subjected to. When designing a control architecture for a CSTR, you must
invertigate the possibility of all of these disturbances, determine the magnitude of each possible disturbance, and address how each will be
handled.
Changes in feed properties
flow rate
composition
temperature
Changes in enthalpy of heat exchange medium
Change in heat transfer properties (ex: fouling)
11.7.3: DISTURBANCES TO PFRS

Plug Flow Reactors (PFRs) behave differently than CSTRs and will have different properties to consider when designing a control
architecture for them. The biggest difference is that the temperature, flows, and compositions will all vary along the reactor. There is danger
of exceeding a design limitation in temperature or flow in certain parts of the PFR, so control of the reactor is important along its length. To
design a control architecture for a PFR, the following disturbances and changes to the system must be addressed:
Temperature control in multiple places along reactor (hot spots can easily occur in PFRs)
Flow control in multiple places along the reactor
Inlet/outlet pressures
Feed property disturbances:
flow rate
composition
temperature
Changes in enthalpy of heat exchange medium
Change in heat transfer properties (ex: fouling)
11.7.4: ENDOTHERMIC REACTORS

DISCLAIMER: All reactors shown posess two controllers. We realize that in the real-world a reactor would require more than two
temperature controllers but we are concerned primarily with the common placement of controllers outside of the reactor in this
wiki. Thank you and continue reading.
Endothermic reactors tend to be easier to control than exothermic ones because they are much less prone to runaway. There are two
commonly used methods to control an endothermic CSTR. These two methods are differentiated by the variables they manipulate. In the
first method, the steam pressure in the reactor jacket is in the manipulated variable, whereas in the second method, the steam flowrate is the
manipulated variable. In endothermic reactors, as well as exothermic ones as we will later see, feedback is the most common type of control
system used. This is done to ensure that changes are being based on what is actually happening in the reactor and not what is predicted to
happen. This is important in the endothermic case to ensure that the proper amount of heat is being added to the reactor to obtain a desired
conversion. Feedback is more useful for control since the amount of heat needed can change quickly based on the amount or concentration
of reactants being added. Generally the temperature control of the reactor is independent of the reactant feed, so the system needs a way to
adjust to changes within reactant feed, and this is accomplished by using feedback control in the streams that control the temperature (i.e.
the steam feed stream).
11.7.4.1: CONTROLLED BY STEAM PRESSURE

Heat duty can be more effectively removed with the use of steam pressure as the manipulated variable because changes in the heat duty of
the reactor will cause an immediate change in the steam pressure. By using the steam pressure, it also linearizes the temperature control
system which is not the case when the steam flowrate is the manipulated variable. Using the steam pressure as the manipulated variable
does not provide a direct measurement of the heat load, the amount of heat needed by the reactor. If a direct measurement is required, a
steam flow indicator can be installed which would allow for the direct measurement of the applied heat load. An added benefit, which those
of you who have worked with the temperature controller in ChE 460 might have noticed, is that often times the pressure of steam being
provided by the utilities plant is variable. By using the steam pressure as the manipulated variable the control system will automatically
adjust to this type of disturbance from the source. If a large change in steam pressure were to occur, the pressure controller would measure
this change, and the control scheme would intake the change in steam pressure and translate this into an appropriate change in the valve
setting.
Figure: Image Modified From: Riggs, James B., Karim, Nazmul M.. Chemical and Bio-Process Control. Third Edition Chapter 18. Ferret
Publishing.
As can be seen in the figure above, when steam pressure is used as the manipulated variable the control system is run in a feedback mode.
One of two things can control the amount of steam fed into the jacket. The first is the temperature of the product, and the second is the
steam pressure in the jacket. As mentioned above, a change in the heat duty required by the reactor will quickly change the steam pressure
which is why it is commonly used over the steam flowrate.
Notice that there are two controllers that are responsible for adjusting the steam valve. The pressure controller is sensitive to changes in the
heat duty required by the reactor and is used to adjust the steam according to the needs of the temperature controller on the product stream.
This is a prime example of cascade control. The temperature of the product stream would output a setpoint to the pressure controller for the
amount of steam needed to attain the desired temperature setpoint. The pressure controller would then communicate to the valve what needs
to be done in order to achieve this temperature setpoint based on the steam pressure.
11.7.4.2: CONTROLLED BY STEAM FLOWRATE

The second method for controlling a endothermic CSTR is by manipulating the steam flow rate. Using flow rate as the manipulated variable
makes the control system prone to changes in heat load and changes in the enthalpy of supplied steam. These changes require direct action
of the temperature control system. One advantage of using flowrate as the manipulated variable is that heat load is directly measured, and
thus, conversion is directly measured.
As can be seen in the figure above the system works in a feedback mode when the steam flowrate is used as the manipulated variable. The
temperature of the product stream is the primary factor in adjusting the amount of steam fed into the system, and therefore this system is
less responsive to changes in the amount of heat duty required for the process.
Note here that there are again two controllers used to adjust the steam valve. This setup is similar to the previous case, but now the flow
controller is the "slave" controller to the temperature controller. The cascade control scheme is again at work!
11.7.5: EXOTHERMIC REACTORS

Exothermic reactors are harder to control because safety is dependent on heat removal. There are two common control architectures for an
exothermic CSTR based on the temperature of the coolant. The first method uses the outlet temperature of the coolant as the manipulated
variable, while the second method uses the inlet temperature of the coolant. Just like the endothermic reactors, feedback control is very
commonly used when controlling the temperature of exothermic reactors. The reasoning for feedback control in the exothermic case is
slightly different but similar to that of endothermic reactors. In endothermic reactors we wanted to ensure that the proper amount of heat
was being supplied to our reactor, and if the amount needed changed; we needed a way to adjust our temperature accordingly. In the
exothermic case we need to ensure that we are removing the proper amount of heat, not only to ensure an optimal reaction temperature, but
also to prevent a runaway reaction. Therefore, we need to know what is actually going on within the reactor instead of forecasting possible
temperature changes within it. Cascade control is also used. In this case the temperature of the product stream outputs a setpoint to the
temperature controller whether it be located on the outlet or inlet coolant feed. The "slave" controller can then adjust the valve position
based on the temperature of the coolant water.
11.7.5.1: CONTROLLED BY OUTLET COOLANT TEMPERATURE
The figure above shows the control architecture for CSTR reactor temperature controlled by outlet coolant temperature in a feedback
mechanism. There is a temperature control on both the product stream and outlet coolant temperature. Both of the temperature controls are
used to control the valve on the inlet coolant stream. The advantage of this setup is that it responds faster to fouling on heat-transfer surfaces
than the setup with the temperature controller on the inlet. This is because in order for the inlet temperature to adjust to fouling, the fouling
must first affect the temperature of the product stream. Therefore the fouling has less direct effect on the controller, but in this case fouling
will immediately affect the temperature of the exiting cooling water. However, this setup has the disadvantage of responding slower to
changes in the inlet coolant temperature.
As can be seen in the above diagram you will notice that again cascade control is utilized in this particular system. The temperature sensor
on the product stream provides information on whether the stream needs to be cooled more or less and outputs a setpoint to the temperature
controller on the recycle stream. This controller can then take into account the temperature of the recycled coolant water and make an
adjustment to the amount of fresh coolant water that is added to the system.
11.7.5.2: CONTROLLED BY INLET COOLANT TEMPERATURE
As seen in the figure above, the control configuration for CSTR reactor temperature is controlled by inlet coolant temperature in a
feedforward mechansim. There are temperature controls on both the product stream and inlet coolant temperature; both of the controls are
used to control the valve on the inlet coolant stream. One advantage of this setup is that it responds faster to changes in inlet coolant
temperature. A disadvantage is that it responds slower to fouling on heat-transfer surfaces.
11.7.5.3: MORE ON EXOTHERMIC REACTORS

How do you decide which exothermic reactor control architecture to use? It depends on whether a faster response to fouling on heat-transfer
surfaces or a faster response to changes in inlet coolant temperature is more important.
In addition to stabilizing the temperature within an exothermic reactor, it is often a goal to maximize production rate. This can be achieved
by placing a valve on the feed stream. As the temperature control adjusts the cold water valve, the cold water valve position is then
communicated to the feed valve and adjusts it.
Cascade control is again utilized. In this case the secondary or "slave" controller is located in a different location, which is the only
difference between the two scenarios.
 EXAMPLE 11.7.1
As part of a community outreach program, your company is sponsoring a haunted house. The haunted house is being constructed on
site in an old, dilapidated warehouse located on the plant property. You have been put in charge of designing a scene in which a witch is
creating a witch’s brew. It has been proposed that you use a reaction between “Witch’s Brew Deluxe”, a commercial compound used in
the haunted house industry, dry ice, and water.
Dry Ice + “Witch’s Brew Deluxe” + Water --> Bubbles + Mist
The manufacturer of “Witch’s Brew Deluxe” supplies optimal conditions for real life cauldron imitation at 30ºC water temperature.
Temperatures in excess of this result in a surplus of cauldron bubbles and mist which impair the visual experience. Temperatures below
this do not allow for large enough reaction rates and do not produce any steam or bubbles.
The reaction is highly endothermic, so a supply of steam has been diverted from the main plant to supply heat to the reactor. However,
due to the great distance this steam is traveling, the supply steam pressure is highly variable. An uncovered, jacketed CSTR is available
for use as a cauldron. A mixture of water and “Witch’s Brew Deluxe” is constantly fed into the reactor, and solid dry ice is manually
feed by the operator.
What control scheme is most suitable for temperature control and why? Specifically, what should you use as the manipulated variable?
Solution
Because the steam supply pressure will vary greatly, a steam pressure control architecture is preferred. By monitoring the jacket
pressure, changes in steam supply pressure will immediately be observed and corrected for.
 EXAMPLE 11.7.2
You have just recently taken a job offer from a haunted engineering firm run by a group of witches located in Salem, Massachusetts. As
the new guy on the job they would like your input on which variable they should treat as the manipulated variable for the reactor
temperature controller. Their reactor is an exothermic CSTR. The following reaction is what takes place within this CSTR.
3 frog's legs + 2 cow hooves --> 4 moles of magic elixir
The magic elixir is generally used as a wart remover but also shows promise as an antacid.
The head witch at the plant tells you that significant fouling occurs on the heat transfer surfaces because the city only allows them
access to dirty water with high mineral content because they are witches and generally considered substandard citizens. She (the head
witch) also points out that they have recently developed a spell to maintain the inlet temperature and pressure of their coolant water.
Which variable should be the manipulated one and why?'
Solution
The correct manipulated variable for the reactor temperature controller would be the outlet coolant temperature. This allows for a
quicker response to fouled heat-transfer surfaces. Since the head witch told you that inlet temperature and pressure are constant, the
configuration's weakness to changes in these two parameters can be overlooked.
Exercise 11.7.2A
Which type of REACTION is generally much easier to control?
a. Endothermic
b. Distillation of alcohol
c. Cold Fusion
d. Exothermic
Answer
endothermic
Exercise 11.7.2B
By using the __________ as the manipulated variable, it linearizes the temperature control system which is not the case when the steam
flowrate is the manipulated variable.
a. feed flow rate
b. steam pressure
c. product composition
d. rate of mixing
11.7.6: REFERENCES
Riggs, James B. and Karim, M. Nazmul. Chemical and Bio-Process Control. Chapter 18: Control Case Studies. Third Edition. Ferret
Publishing.
Svrcek, William Y., Mahoney, Donald P., and Young, Brent R. A Real-Time Approach to Process Control. Chapter 7: Common Control
Loops. Second Edition. John Wiley and Sons, Ltd.
11.7: Common control architectures and model for reactors is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Brian
McQuillan, Crystal Miranda, Brandon Quigley, John Zhang, & John Zhang via source content that was edited to conform to the style and standards of the
LibreTexts platform; a detailed edit history is available upon request.
CHAPTER OVERVIEW
12: MULTIPLE INPUT, MULTIPLE OUTPUT (MIMO) CONTROL

12.3: MIMO using Model Predictive Control
12.5: Understanding MIMO Control Through Two Tanks Interaction
12: Multiple Input, Multiple Output (MIMO) Control is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al.
1
12.1: DETERMINING IF A SYSTEM CAN BE DECOUPLED
Authors: Wen Chung Lim, James Bennett, Jamila Grant, Ajay Bhasin
A system of inputs and outputs can be described as one of four types: SISO (single input, single output), SIMO (single input, multiple
output), MISO (multiple input, single output), or MIMO (multiple input, multiple output).
Multiple input, multiple output (MIMO) systems describe processes with more than one input and more than one output which require
multiple control loops. Examples of MIMO systems include heat exchangers, chemical reactors, and distillation columns. These systems
can be complicated through loop interactions that result in variables with unexpected effects. Decoupling the variables of that system will
improve the control of that process.
An example of a MIMO system is a jacketed CSTR in which the formation of the product is dependent upon the reactor temperature and
feed flow rate. The process is controlled by two loops, a composition control loop and a temperature control loop. Changes to the feed rate
are used to control the product composition and changes to the reactor temperature are made by increasing or decreasing the temperature of
the jacket. However, changes made to the feed would change the reaction mass, and hence the temperature, and changes made to
temperature would change the reaction rate, and hence influence the composition. This is an example of loop interactions. Loop interactions
need to be avoided because changes in one loop might cause destabilizing changes in another loop. To avoid loop interactions, MIMO
systems can be decoupled into separate loops known as single input, single output (SISO) systems. Decoupling may be done using several
different techniques, including restructuring the pairing of variables, minimizing interactions by detuning conflicting control loops, opening
loops and putting them in manual control, and using linear combinations of manipulated and/or controlled variables. If the system can’t be
decoupled, then other methods such as neural networks or model predictive control should be used to characterize the system.
There are two ways to see if a system can be decoupled. One way is with mathematical models and the other way is a more intuitive
educated guessing method. Mathematical methods for simplifying MIMO control schemes include the relative gain array (RGA) method,
the Niederlinski index (NI) and singular value decomposition (SVD). This article will discuss the determination of whether a MIMO control
scheme can be decoupled to SISO using the SVD method. It will also discuss a more intuitive way of decoupling a system using a variation
of the RGA method.
12.1.1.1: DEFINITIONS OF INPUT AND OUTPUT SYSTEM TYPES

1. SISO- Single Input, Single Output
These systems use data/input from one sensor to control one output. These are the simplest to design since they correspond one sensor to
one actuator. For example, temperature (TC) is used to control the valve state of v1 through a PID controller.
2. SIMO- Single Input, Multiple Output
These systems use data/input from one sensor to control multiple outputs. For example, temperature (TC) is used to control the valve state
of v1 and v2 through PID controllers.
3. MISO- Multiple Input, Single Output
These systems use data/input from multiple sensors to control one ouput. For example, a cascade controller can be considered MISO.
Temperature (TC) is used in a PID controller (#1) to determine a flow rate set point i.e. FCset. With the FCset and FC controller, they are
used to control the valve state of v1 through a PID controller (#2).
4. MIMO- Multiple Input, Multiple Output
These systems use data/input from multiple sensors to control multiple outputs. These are usually the hardest to design since multiple sensor
data is integrated to coordinate multiple actuators. For example, flow rate (FC) and temperature (TC) are used to control multiple valves
(v1, v2, and v3). Often, MIMO systems are not PID controllers but rather designed for a specific situation.
12.1.2: SINGULAR VALUE DECOMPOSITION

Singular value decomposition is a useful mathematical way to determine if a system will be prone to interactions due to sensitivity problems
that result from small errors in process gains. It is a matrix technique that determines if a system is able to be decoupled.
12.1.2.1: TWO INPUT TWO OUTPUT SYSTEM
The SVD method starts with the steady state gain matrix, which is obtained from the relative gain array (RGA) method for a given 2 input-2
output control loop system:
,
Using [G], we obtain the eigenvalues for the system. This can be done through two methods; either the eigenvalues can be obtained through
numerical software as described in the Eigenvalues and Eigenvectors section or the eigenvalues can be calculated by hand. The hand
calculations are shown here. [G] gives the following values for the system, which have been derived from theoretical work and are used to
find the eigenvalues:
b = g112 + g122 --------------(1)
c = g11g21 + g12g22---------(2)
d = g212 + g222--------------(3)
b, c, and d are parameters created for this method that do not have a direct physical meaning. Using the formulas below with the values of b,
c, and d gives the eigenvalues for the system:
λ1 = s12 = ------(4)
λ2 = s22 = ------(5)
s1 and s2 are the positive square roots of the respective eigenvalues. The condition number CN is defined as the ratio of the larger of the two
values to the smaller value:
CN = if , or CN = if
The greater the CN value, the harder it is for the system in question to be decoupled. As a rule of thumb, a system with a CN number of
more than 50 is impossible to decouple. Such a system has manipulated variables which all have significant and similar relative impact on
the controlled variables they affect and hence the control equations for the system cannot be restructured to obtain a simpler control system.
In the worst case scenario for a two input two output system, using the exact same control variable twice would give a CN number of
infinity, because both control variables would have the same effect on the controlled variables. An ideal system would have a CN number of
one, where each control variable controls a single distinct output variable.
12.1.2.2: MIMO SYSTEMS WITH TWO OR MORE INPUTS AND OUTPUTS

To use this method, it is necessary to know how to create a transpose matrix. A transpose matrix is created by writing the columns of the
original matrix as the rows of the transpose matrix. Mathematically, the transpose matrix is denoted by [A]T, where [A] is the original
matrix. An example is shown here:

To use singular value decomposition on systems with two or more inputs and outputs, the m x m steady state gain matrix [G] of the form
g1,1 g1,2 ⋯ g1,m
⎡ ⎤
⎢ g2,1 g2,2 ⋯ g2,m ⎥

⎢ ⎥
$[G] = ⎢ ⎥ $
⎢ ⎥
⎢ ⋮ ⎥
⎣ ⎦
gn,1 gn,2 ⋯ gn,m
is broken down into the form:

⊤
$G = U ΣV $
where:
U is an m x m matrix , the column vectors of which are the unit eigenvectors of the m x m matrix GGT,GT being the transpose matrix of
G.
V is an m x m matrix, the column vectors of which are the unit eigenvectors of the m x m matrix GTG.
Σ is an m x m diagonal matrix containing singular values, where the values are arranged in descending order from the top left most
corner to the bottom right corner.
for a 2 input 2 output system if , or for a 2 input 2 output system if .

The CN number can then be determined by the ratio of the largest value in Σ to the smallest value. Notice in the previous two input two
output example, there was no need to calculate the eigenvalue matrix Σ for a two by two matrix, since the eigenvalues can be calculated
directly with the parameters b,c, and d.
For more information on how to derive the three matrices U, Σ and V by hand, refer to the Eigenvalues and Eigenvectors section.
Mathematica can also be used to find U, Σ and V for a given m x m Matrix G. This is accomplished by typing in the command
Map[MatrixForm, {u, w, v}=SingularValueDecomposition[G]],
giving U, Σ and V in that order.
12.1.2.3: INTUITIVE DECOUPLING USING THE RGA

With this technique, the first step is to construct the RGA by placing all of the controllers in manual, changing the value of the output
manually, and then recording the final value of the controller after making the change. This will give the gain matrix, which is then
multiplied by its inverse to give the relative gain matrix. The goal here is to pair measurements and control valves so that the relative gain
elements are close to 1, and all other combinations are close to zero. A values between 0 and 1 tells that the net effect of the other loops on
this pairing is to change the measurement signal in the same direction as the control valve signal (cooperation). A value greater than one
means that the measurement signal will go in the opposite direction of the control valve signal (conflict). Finally a relative gain element less
than zero means that the net effect is conflict, but here the gain will change sign, not the measurement signal. Therefore, if a loop is stable
when the other loops are in manual, it will probably not be stable when the other loops are in automatic, and vice versa. To intuitively
decouple this system then would require a guess and check of different pairings, with the goal being to get the pairings to a relative gain
element close to one. If this is not possible and no pairings give a value close to one, then the system will need to be decoupled using
decoupling control.
12.1.2.4: DECOUPLING A SYSTEM USING DECOUPLING CONTROL

The goal of decoupling control is to eliminate complicated loop interactions so that a change in one process variable will not cause
corresponding changes in other process variables. To do this a non-interacting or decoupling control scheme is used. In this scheme, a
compensation network called a decoupler is used right before the process. This decoupler is the inverse of the gain array and allows for all
measurements to be passed through it in order to give full decoupling of all of the loops. This is shown pictorially below.
For a simple jacketed reactor example, if changing the flow rate of reactant to obtain a desired composition also increases the reactor
temperature by five degrees, the decoupler would decrease the temperature of the jacket to decrease the temperature inside the reactor by
five degrees, offsetting the original change. This jacket temperature decrease also changes the composition inside the reactor, so a properly
designed controller accounts for both effects in the control scheme to obtain the desired composition without changing the reactor
temperature.

Hot water at 110°C is pumped into a mixer at 500 kg/h. The water is used to dilute a concentrated caustic substance at 50°C flowing in at
200 kg/h. Below is a diagram of the process. Check, by the singular value decomposition method, to see if decoupling of the system is
feasible. Assume that the heat capacities of both streams are the same with no heat of mixing.
boldS olution :
From the question you know:

T1 = 383.15 K
T2 = 323.15 K
m1 = 500 kg/h
m2 = 200 kg/h
y1 = outgoing mass flow rate (kg/h)
y2 = temperature of outflowing stream (K)
From mass balances you know:
So:
y1 = 700 kg/h
y2 = 366.01 K
The steady state gain matrix (G) needs to be calculated as explained in the RGA method:
By plugging in all known values, the steady state gain matrix becomes:
From here you can solve for the values b, c, d λ1, λ2, s1 and s2 by using the equations (1) through (5) in the wiki:
b = g112 + g122 = 12 + 12 = 2
c = g11g21 + g12g22 = (1)(.0245) + (1)( − .0612) = -0.0367
d = g212 + g222 = .02452 + − .06122 = 0.004346
λ1 = = =2
λ2 = = = 0.003673
s1 = = = 1.414
s2 = = = 0.0606
= 23.33
CN is less than 50, therefore decoupling for this system is feasible.

A jacketed reactor with a single feed stream (Stream 1) is monitored by 3 proportional (P) controllers as indicated in the figure.
These controllers all act as inputs to each of 3 valves according to the relationships:
Where:
C1 = Kc1e1(t) + b1, C1 is the output from the level controller
C2 = Kc2e2(t) + b2, C2 is the output from the temperature controller
C3 = Kc3e3(t) + b3, C3 is the output from the composition sensor
Is it feasible to decouple this system?
boldS olution :
The gain array is calculated according to the RGA method and is:
Input the following commands in Mathematica for SVD analysis:

G=
G = G*1.0 Map[MatrixForm, {u,w,v}=SingularValueDecomposition[G]]
The output of the SVD decomposition is the following:
, ,
The condition number is calculated from the highest and lowest values of the ∑ matrix:
CN < 50, therefore the system may be decoupled.

As a more complicated problem, consider how this problem would change if the controllers used a cascade control style. For a P controller,
the output may be similar to:
C2 = Kc2e2(t) + b2 = Kc2 * (LCactual − f(C1)) + b
What if we used a PID controller instead?

The CN, or condition number
a. Is the ratio of the smaller number (s2) to the larger number (s1)
b. Determines the feasibility of decoupling a system.
c. Is the unit eigenvectors of the m x m matrix GTG
d. Is always less than 50.
Answer: b

For MIMO systems
a. Control loops are isolated

b. Each controlled variable is only manipulated by one variable
c. Decoupling the system makes it more complicated
d. Manipulated variables may affect several controlled variables
Answer: d

www.youtube.com/v/j-heeeVmEAM
Unnarated Powerpoint Presentation
12.1.8: REFERENCES
Mahoney, D.P., Svrcek, W.Y., and Young, B.R. (2006) A Real-Time Approach to Process Control, 2nd Edition. John Wiley & Sons, Ltd.
ISBN 0470025336
Gwak, K.W., Masada, G.Y. (2005) "Structural Analysis and Optimization of Nonlinear Control Systems Using Singular Value
Decomposition". American Society of Mechanical Engineers Journal of Dynamic Systems, Measurement and Control, Vol. 127, No. 1,
pp. 105-113, March 2005.
McMillan, Gregory K. (1983) Tuning and Control Loop Performance. Instrument Society of America. ISBN 0-87664-694-1.
Tham, M.T. (1999). "Multivariable Control: An Introduction to Decoupling Control". Department of Chemical and Process Engineering,
University of Newcastle upon Tyne. Available online at: lorien.ncl.ac.uk/ming/mloop/mloop.pdf.
Lee, Jay H., Choi, Jin Hoon, and Lee, Kwang Soon. (1997). "3.2 Interaction and I/O Pairing". Chemical Engineering Research
Information Center. Available online at: www.cheric.org/education/practical/mpc16473.pdf.
12.1: Determining if a system can be decoupled is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via
12.2: MIMO CONTROL USING RGA
Single variable Input or Single variable Output (SISO) control schemes are just one type of control scheme that engineers in industry use to
control their process. They may also use MIMO, which is a Multi-Input-Multi-Output control scheme. In MIMO, one or more manipulated
variables can affect the interactions of controlled variables in a specific loop or all other control loops. A MIMO control scheme is
important in systems that have multiple dependencies and multiple interactions between different variables- for example, in a distillation
column, where a manipulated variable such as the reflux ratio could directly or indirectly affect the feed flow rate, the product composition,
and the reboiler energy. Thus, understanding the dependence of different manipulated and controlled variables in a MIMO control scheme
could be extremely helpful in designing and implementing a control scheme for a process.
One method for designing and analyzing a MIMO control scheme for a process in steady state is with a Relative Gain Array (RGA). RGA
is useful for MIMO systems that can be decoupled (see the article about determining if a system can be decoupled). For systems that cannot
be decoupled, model predictive control or neural networks are better choices of analysis tool than RGA. A good MIMO control scheme for
a system that can be decoupled is one that can control a process variable without greatly affecting the other process variables. It must also be
stable with respect to dynamic situations, load changes, and random disturbances. The RGA provides a quantitative approach to the analysis
of the interactions between the controls and the output, and thus provides a method of pairing manipulated and controlled variables to
generate a control scheme.
12.2.2: WHAT IS RGA?

Relative Gain Array is an analytical tool used to determine the optimal input-output variable pairings for a multi-input-multi-output
(MIMO) system. In other words, the RGA is a normalized form of the gain matrix that describes the impact of each control variable on the
output, relative to each control variable's impact on other variables. The process interaction of open-loop and closed-loop control systems
are measured for all possible input-output variable pairings. A ratio of this open-loop 'gain' to this closed-loop 'gain' is determined and the
results are displayed in a matrix.
λ11 λ12 ⋯ λ1n
⎡ ⎤
⎢ λ21 λ22 ⋯ λ2n ⎥

RGA = Λ = ⎢ ⎥
⎢ ⎥
⎢ ⋮ ⎥
⎣ ⎦
λn1 λn2 ⋯ λnn
The array will be a matrix with one column for each input variable and one row for each output variable in the MIMO system. This format
allows a process engineer to easily compare the relative gains associated with each input-output variable pair, and ultimately to match the
input and output variables that have the biggest effect on each other while also minimizing undesired side effects.
12.2.2.1: UNDERSTANDING THE RESULTS OF THE RGA
The closer the values in the RGA are to 1 the more decoupled the system is
The maximum value in each row of the RGA determines which variables should be coupled or linked
Also each row and each column should sum to 1
Example
The table below includes the RGA results. The values highlighted in red are the maximum values in the row. These values indicated that the
valve for that row should be used to control the variable that is listed in the corresponding column.
12.2.3: CALCULATING RGA
There are two main ways to calculate RGA:
1. Experimentally determine the effect of input variables on the output variables, then compile the results into an RGA matrix.
2. Use a steady-state gain matrix to calculate the RGA matrix.
Method (1) should be used when it is possible to carry out the experiments as detailed in the Calculating RGA with Experiments section.
This method will generally yield the most accurate RGA matrix of the system because it is based on actual data taken from the operating
control system. If performing these extensive experiments is not possible, method (2) can be used. If a process model is available, method
(2) can be used with no experimental data. If there is no process model available, some experimental data must be taken (though less
extensively than in method (1)) and used in conjunction with method (2).
12.2.3.1: METHOD 1: CALCULATING RGA WITH EXPERIMENTS

This method of calculating the RGA can be used when it is possible to run experiments on each of the input-output pairings. Below is a
step-by-step explanation of how to experimentally compile the RGA for a simple MIMO system.
The simplest MIMO system is one that has two inputs and two outputs. Remember that by definition, a change in one of the inputs in a
MIMO system will change both of the outputs. This system can be expressed mathematically as written below.
y1 = a11 m1 + a12 m2
y2 = a21 m1 + a22 m2
Where yi is the output for loop i, the m variables are the inputs for each loop, and the a variables are the transfer functions. It also helps to
see this system as a control diagram, as shown here.
Clearly, both y1 and y2 are affected by both inputs (m1 and m2). Because of this, there are two choices for control. y1 could be controlled by
m1, or y1 could be controlled by m2. y2 would then be controlled by the input that is not controlling y1. For a MIMO system with n input
variables, there are n! possible control configurations. The question is: Which configuration will give you the best control? The answer can
be determined by finding the relative gain array (RGA). There are a few ways to determine the RGA, one of which is by doing two
experiments, repeated for every possible configuration. The RGA can be assembled from the data collected. These experiments are
explained below using the above drawing and the configuration where m1 controls y1.
Experiment 1
For the first experiment, the objective is to observe the effect of m1 on y1. To do this, all of the loops must be open. In other words, the
feedback loop is removed and the system is run manually with no control. This configuration is shown below.
Now, since the system is under manual control, it is possible to introduce a step change Δm1 while keeping m2 constant. Because m2 is held
constant, the change in y1, Δy1, can be attributed entirely to m1. Although a change in m1 will also affect y2, it is the relationship between m1
and y1 that must be observed at this point. Then, the gain of y1 with respect to m1 with all loops open is defined as g11. The calculation of g11
is shown below.
Δy1 ( all - loops -open )
g11 =
Δm1 ( all − loops -open )
The final objective of experiment one is obtaining the value gij where the controlled variable i is controlled by manipulated variable j.
Experiment 2
The goal of experiment two is to determine the effect of m2 on y1. To do this, loop 1 must remain open, but all other loops remain closed.
This configuration is shown below.
Now, the same step change that was introduced to m1 in experiment one must be introduced again. The difference is, now loop 2 is in
perfect control which means that when m1 changes, it affects y2, so the control loop on loop two will adjust m2 in order to compensate and
keep y2 constant. When m2 changes it, in turn, affects y1. It is this change that is being observed. The amount y1 changes is Δy1 and the
amount m1 was stepped up is Δm1. Now:
Δy1 ( only − loop − one − open )
∗
g =
11
Δm1 ( only − loop − one − open )
The objective of experiment two is to obtain this value of g where controlled variable i is being controlled by manipulated j.
∗
ij
Compiling the Array

Once the experiments are run for every possible configuration, the results can be used to compile the relative gain array (RGA). To do this,
we first have to find the relative gain. The relative gain is the ratio of to and is represented by .
δyi
( )
gij δmj
all-loops-open
λij = = (12.2.1)
∗ δyi
g
ij ( )
δmj
only-loop-i-open
The value for λ must be computed for every possible combination of i and j. After every value is found, the RGA matrix can be formed:
λ11 λ12 ⋯ λ1n
⎡ ⎤
⎢ λ21 λ22 ⋯ λ2n ⎥

RGA = Λ = ⎢ ⎥ (12.2.2)
⎢ ⎥
⎢ ⋮ ⎥
⎣ ⎦
λn1 λn2 ⋯ λnn
12.2.3.2: METHOD 2: CALCULATING RGA WITH STEADY-STATE GAIN MATRIX

Sometimes it is not convenient or possible to perform the experiments described above on every control pairing possibility in the system. If
that is the case, a steady-state gain matrix can be used to determine the RGA. If a process model is available, the steady-state gain matrix
can be calculated from the model equations. If a process model is not available, the steady-state gain matrix can be calculated from
experimental data (fewer experiments are required than when the RGA is calculated directly from experimental data as above). Once the
steady-state gain matrix is calculated, it can be used to compute the RGA. Beware that a RGA has to have the same number of controlled
variables and manipulated variables (same number of inputs and outputs) to be in an m x m matrix form.
Process Model Available
If a process model is available, the steady-state gain matrix relates the manipulated variables to the controlled variables according to the
following equation:
y = Gm
where y is a vector of the controlled variables:

y1
⎡ ⎤
⎢ y2 ⎥
y = ⎢ ⎥
⎢ ⎥
⎢ ⋮ ⎥
⎣ ⎦
yn
m is a vector of the manipulated variables:

m1
⎡ ⎤
⎢ m2 ⎥
m = ⎢ ⎥
⎢ ⎥
⎢ ⋮ ⎥
⎣ ⎦
mn
and G is the steady-state gain matrix:

g11 g12 ⋯ g1n
⎡ ⎤
⎢ g21 g22 ⋯ g2n ⎥

G = ⎢ ⎥
⎢ ⎥
⎢ ⋮ ⎥
⎣ ⎦
gn1 gn2 ⋯ gnn
The values of gij are calculated by taking partial derivatives of the equations governing the controlled variables. Specifically:
∂ yi
gij =
∂ mj
During differentiation, hold constant all manipulated variables other than mj. To fill in matrix G with numerical values, simply substitute the
appropriate steady state values for the process into the expressions derived from differentiation.
Now that the steady-state gain matrix has been found, the RGA can be calculated with just a little bit more math. However, now that you
have the steady state gain matrix, this might be a good time to determine if your system can even be decoupled! See the article on singular
value decomposition for methods to check this. Assuming SVD gives a condition number of less than about 50, we can proceed to
calculation of the RGA. First, define a matrix R to be the transpose of the inverse of matrix G:
−1 T
R = (G )
The calculation of the inverse of a matrix is quite complicated for anything larger than a 2x2 matrix. Fortunately Mathematica will do this
with the following command for a square matrix M:
Inverse[M]
The transpose of a matrix is when the rows become columns and the columns become rows. For a square matrix, this just means reflecting
across the diagonal running from the top left to the bottom right corner of the matrix. The transpose can be found in Mathematica with the
following command:
Transpose[M]
The RGA can now be obtained one element at a time according to this equation:
λij = gij rij
Note that this is not your usual matrix multiplication! Here you multiply corresponding elements of the G and R matrices to get the
corresponding element of the RGA. This is the type of multiplication Mathematica does with the standard multiplication operator.
See Example 1 below for a worked-out RGA calculation using this approach.
Process Model Not Available
In case there is no process model available and it is not feasible to determine the RGA by carrying out both experiments necessary for full
experimental determination, it still may be possible to develop a steady-state gain matrix experimentally which can then be used to derive
the RGA. In this case, just carry out experiment 1 as described above to determine the elements of the steady-state gain matrix:
Δyi
gij =
Δmj
with all loops open.

Each element of the steady-state gain matrix can be determined this way, and then the RGA can be calculated from the steady-state gain
matrix as shown in the previous section.
12.2.4: INTERPRETING THE RGA

There are some important properties and guidelines in understanding and analyzing the RGA, and what the different values of the RGA
mean:
1) All elements of the RGA across any row, or down any column will sum up to one:
n n
∑ λij = ∑ λij = 1
i=1 j=1
This makes calculating the RGA easier because:

in 2X2 case, only 1 element must be calculated to determine all elements
in 3X3 case, only 4 elements must be calculated to determine all elements
and so on.
2) The λij calculated from steady-state matrix is dimensionless and unaffected by scaling.
3) Each of the rows in the RGA represent one of the outputs. Each of the columns represent a manipulated variable.
If λij = 0: The manipulated variable (mj) will have no effect on the output or the controlled variable (yi).
If λij = 1: The manipulated variable mj affects the output yi without any interaction from the other control loops in the system. From the
definition of λij, this implies that the gain loop with all loops open is equal to the gain loop with all other loops closed, ie: g11 = g*11.
If λij < 0: The system will be unstable whenever mj is paired with yi, and the opposite response in the actual system may occur if other
loops are opened in the system.
If 0 < λij < 1: This implies that other control loops (mj- yi) are interacting with the manipulated and controlled variable control loop.
Three different relationships based on λ=0.5 imply different interpretations of pairing and the RGA:
If λij = 0.5: The control pairing effect is equal to the retaliatory effect of other loops.
If λij < 0.5: The other control loops are influencing the control pair, and the influence from the other control loops are greater than the
influence from the control pair.
If λij > 0.5: This means that the control pair has a greater influence on the system than the other control loops.
If λij >1: The open-loop gain of the control pair is greater than the gain with all other loops closed, ie: g11 > g*11. The positive value of
RGA indicates that the control pair is dominant in the system, but the other loops are still affecting the control pair in the opposite
direction. The higher the value of λij, the more correctional effects the other control loops have on the pair.
λij Possible Pairing
λij = 0 Avoid pairing mj with yi
λij = 1 Pair mj with yi
λij < 0 Avoid pairing mj with yi
λij = or < 0.5 Avoid pairing mj with yi
λij >1 Pair mj with yi
12.2.5: ANALYSIS WITH RGA
The Niederlinski Index (NI) is a calculation used to analyze the stability of the control loop pairings using the result of the RGA, evaluated
at Steady State:
|G|
NI =
n
∏i=1 gii
A negative NI value indicates instability in the control loop. For a 2 2 matrix, a positive NI value indicates stability in the pairings, but
this is not necessarily true for larger matrices! For matrices larger than 2 2, a conclusion can only be drawn from a negative NI, which
indicates instability. NI should not be used for systems that have time delays (dead time), as the NI stability predictions assume immediate
feedback to the controller. However, the index can be adapted for systems with dead time, although that application is beyond the scope of
this article.
Here's an example NI calculation: given the steady state gain matrix

−0.002 0.001
G = [ ]
0.002 0.003
the NI can be calculated from the following expression:

(−0.002 × 0.003) − (0.001 × 0.002)
NI = = 1.333
−0.002 × 0.003
Since this is a 2 2 matrix, the positive value of the NI indicates stability in the control loop pairings.
12.2.6: OPTIMIZING A MIMO CONTROL SCHEME: PAIRING RULES

The goal of the RGA and NI analysis is to quantitatively determine the optimal variable pairing for a given process. Some basic rules to
remember when attempting to obtain an optimal pairing of control loops in a system are:
Rule 1: The positive RGA elements that are closest to 1.0 should have the corresponding manipulated and controlled variables paired.
When the CN number is large, implying a less decoupled system, one should look for the max RGA elements.
Rule 2: If the NI value is negative, the loop pairing for that control system configuration is unacceptable.
 EXAMPLE 12.2.1
A blending unit is used to dilute and cool down the product stream of a reactor. Three streams are combined in the mixer: the hot,
concentrated stream from the reactor, a room temperature stream containing none of the product A, and a second room temperature
stream containing some A produced elsewhere in the process. It is desired to control the flow rate, temperature, and concentration of A
in the product stream by manipulating the flow rates of the three input streams. A schematic of the process follows, filled in with the
steady state values of flow rate, temperature, and concentration. The manipulated variables are indicated with the letter m and the
controlled variables are indicated with the letter y:
The equations used to model the system are:
y1 = m1 + m2 + m3
T1 m1 + T2 m2 + T3 m3 T1 m1 + T2 m2 + T3 m3
y2 = =
y1 m1 + m2 + m3
c1 m1 + c2 m2 + c3 m3 c1 m1 + c2 m2 + c3 m3
y3 = =
y1 m1 + m2 + m3
Calculate the RGA using the steady-state gain matrix method and use it to determine the best control scheme for the system.
Solution
First you must take the partial derivatives of the control variable equations in order to determine the elements of the steady-state gain
matrix:
∂ y1
g11 = = 1
∂ m1
∂ y2 (T1 − T2 ) m2 + (T1 − T3 ) m3
g21 = =
2
∂ m1 (m1 + m2 + m3 )
∂ y3 (c1 − c2 ) m2 + (c1 − c3 ) m3
g31 = =
2
∂ m1 (m1 + m2 + m3 )
∂ y1
g12 = = 1
∂ m2
∂ y2 (T2 − T1 ) m1 + (T2 − T3 ) m3
g22 = =
2
∂ m2 (m1 + m2 + m3 )
∂ y3 (c2 − c1 ) m1 + (c2 − c3 ) m3
g32 = =
2
∂ m2 (m1 + m2 + m3 )
∂ y1
g13 = = 1
∂ m3
∂ y2 (T3 − T1 ) m1 + (T3 − T2 ) m2
g23 = =
2
∂ m3 (m1 + m2 + m3 )
∂ y3 (c3 − c1 ) m1 + (c3 − c2 ) m2
g33 = =
2
∂ m3 (m1 + m2 + m3 )
Now you can substitute in the given steady state values of the variables to find the numerical values of gij, obtaining:
In matrix form:
1 1 1
⎡ ⎤
11 2 2
G = ⎢
⎢ 13
−
13
−
13
⎥
⎥
⎣ 47 8 18 ⎦
−
845 845 845
Now calculate matrix R:
2 4 7
⎡ 13 13 13 ⎤
T
−1 5 3
R = (G ) = ⎢ 1 − ⎥
⎢ 2 2 ⎥
⎣ 0 65 65 ⎦
−
2 2
You can now calculate the RGA by multiplying matrices G and R piecewise, according to the formula:
λij = gij rij
So the RGA is:

2 4 7
⎡ 13 13 13 ⎤
11 5 3
RGA = ⎢ − ⎥
⎢ 13 13 13 ⎥
⎣ 0 4 9 ⎦
13 13
Note that all rows and columns sum to 1, as they should.

Now it's time to determine the optimal control scheme for the system. The closer an element in the matrix is to 1, the better the pairing
is between the manipulated and controlled variables represented by that element. With this in mind, we can immediately see that m1, the
flow rate of stream 1, should be used to control y2, the temperature of the output stream, because the value of λ21 is the closest to 1 in
the matrix ( ). Next in line is λ33, with a value of . This indicates that m3, the flow rate of stream 3, should control y3, the
concentration of A in the product stream. So far so good, but what should control y1, the flow rate of the product stream? The best
option would be m3, but this is already controlling the concentration of A in the product, so the next best option is m2, the flow rate of
stream 2. This is not optimal because the relative gain is less than 0.5, but it is the best option available given the constraints of the
system.
 EXAMPLE 12.2.2
In a distillation column, reflux rate (R) and steam rate (S)are used to control the compositions in the distillate (xD)and the bottom
product (xB). The control loops are opened in order to conduct experiment one as described above, and the following set of data is
collected:
Trial R (kg/min) S (kg/min) xD xB
1 75 20 0.96 0.05
2 85 20 0.94 0.06
3 95 20 0.92 0.08
4 85 15 0.95 0.06
5 85 25 0.97 0.03
Calculate the RGA for this system, and determine which is the best pairing between the manipulated (R and S) and controlled variables
(xD and xB).
Solution
From the experiment data, it is observed that the reflux rates (R) are constant for trials 2, 4 and 5, while the steam rates (S) are constant
for trials 1, 2 and 3.
First, calculate the steady-state gain matrix for this system:
with S constant
with S constant
with R constant
with R constant
Organize the gain value into the matrix form and compute R :
−0.002 0.001
G = [ ]
0.002 0.003
−1
T −375 250
R = (G ) = [
125 250
Multiply each value in the R matrix by its corresponding gain value to obtain λij:
λ11 = g11 R11
λij is then organized into the RGA matrix form:

3 1
4 4
RGA = [ ]
1 3
4 4
The RGA matrix indicates that the composition of the distillate (xD) would be best controlled by the reflux flow (R); while xB would be
best controlled by the steam rate (S)from the reboiler.
 EXAMPLE 12.2.3
An example problem based from Fall 2008 second exam:

"Imagine that you are developing a control scheme for a new low temperature reactor system. The physics governing the system is
complicated, so you decide to start out with an experiment on the system to see if you can develop a simple set if SISO controllers. To
evaluate this possibility, you begin with your five valves at their steady state operating conditions (v1=0.5, v2=0.55, v3=0.5, v4=0.6,
v5=0.4) and measure the five key variables of this process T1, T2, T3, P1 and P2. Next you slightly adjust each valve to see the steady
state effects on each output variable"
As an approximation, you fit this model to the following system:
T1 −0.2 0.06 0.03 1.0 0.02 V1
⎡ ⎤ ⎡ ⎤⎡ ⎤
⎢ T2 ⎥ ⎢ 0.01 0.1 −0.143 0.012 1.0 ⎥ ⎢ v2 ⎥

⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢T ⎥ = ⎢ 0.5 1.0 −0.23 0.04 0.02 ⎥ ⎢ v ⎥
⎢ 3 ⎥ ⎢ ⎥ ⎢ 3 ⎥
⎢ ⎥ ⎢ ⎥ ⎢ ⎥
⎢P ⎥ ⎢ −0.2 −0.1 1.0 0.123 −0.023 ⎥ ⎢ v ⎥
1 4
⎣ ⎦ ⎣ ⎦⎣ ⎦
P2 1.0 −0.01 0.1 0.01556 −0.013 v5
"Perform an RGA analysis to find the best pairing between valves (v1, v2, v3, v4, and v5) and outputs (T1,T2, T3, P1, and P2)."
In Mathematica, you can input the matrix of coefficients by the following:
Note: I assign the matrix of coefficients to the variable, G.

It can be viewed in a matrix form in Mathematica by using function, "MatrixForm[]":
With the gain matrix, G, you can use the a singular value decomposition (SVD) of the gain matrix to determine how linearly dependent
the model is. The linearity is based on the condition number (CN). Mathematica has a built-in function to do singular value
decomposition. The following shows how to do singular value decomposition for this example.
The Mathematica syntax is : N[Map[MatrixForm, {u, w, v} = SingularValueDecomposition[G]]]. The "SingularValueDecomposition[]"
pops out three arrays assigned to variable u, w, and v . The "Map[MatrixForm, ..." portion pops out the three arrays into a matrix view
and "N[.." portion forces numerical values in the matrices.
To the condition number(CN), you divide the biggest value by the smallest value in matrix, w. The general rule thumb is that if the
condition number(CN) is greater than 50, then the system/model is too difficult to decouple.
This illustrates that this system can be decoupled because the CN number is below 50. To determine the best pairing, RGA analysis is
used in which it is a normalized form of the gain matrix, "G" that determines the impact of each control variable on its respective
output e.g. how much impact does v1 have on T1, etc. The Mathematica syntax is N[G*Transpose[Inverse[G]]]. The RGA
normalization is essential since at times, not in this case however, the values of the gain matrix may represent that a valve has great
influence on multiple variables (ie. a column may contain the highest value from multiple rows). In order to distinguish unique
separations between these influences, the RGA is used to normalize the gain matrix as seen below. This allows us to determine the
different magnitudes of influence the valve has on an output in order to find the predominant effect on an output. The RGA illustrates
these marginal differences in impact by the valves on the variable outputs, and lets us make the best possible pairings.
The best pairing is discovered by taking the maximum value of RGA Matrix for each row.
The bold values correspond to the valve that is the best pairing for the individual output. By looking at the bold values, V1->P2 V2-
>T3, V3-> P1, V4-> T1, V5-> T2.
Exercise 12.2.1
For a MIMO control system containing 4 different manipulated and 4 different controlled variables, how many control configurations
are possible for the system?
a. 24
b. 4
c. 64
d. 256
Answer
a
Explanation: Mathematically, for an nxn manipulated by controlled variable system, the number of control configurations possible
will be n! or n factorial. For a 4x4 variable system, 4x3x2x1 = 24. Physically, the number of control configurations translates to the
number of possible input-output pairings.
Exercise 12.2.2
In a case of a 2x2 matrix system, if one RGA element in first row calculated to be . According to the RGA properties, what is the
other element in that same row?
a.
b.
c.
d. Cannot be determined with given information.
Answer
b
Explanation: As explained above, all elements of the RGA across any row, or down any column should sum up to unity (one). For a
2x2 matrix, if one element in a row is , the other element must be 1- =
Exercise 12.2.3
RGA is useful for process engineers because it allows us to determine:
a. Which matrix size to use.
b. What eigenvalues and eigenvectors are paired.
c. What input and output pairs to use.
d. When feed forward control is preferred over feedback control.
Answer
c
Explanation: Using RGA, a process engineer will assess the relative gain of pairing each input with each output to minimize
undesired interactions and maximize desired interactions between controlled and manipulated variables in a system. See
"Interpreting the RGA" section if you are confused on how this analysis is achieved.
12.2.7: REFERENCES
Berber, Ridvan. Methods of Model Based Process Control, Kluwer Academic Publishers, 1994.
Lee, P.L., Newell, R.B., Cameron, I.T. Process Control and Management, Blackie Academic and Professional.
Romagnoli, Jose A., Palazoglu, Ahmet. Introduction to Process Control, Taylor and Francis Group.
Shinskey, F.G. Process Control Systems: Application, Design, and Tuning, Fourth Edition, McGraw-Hill, Inc., 1996.
Svrcek, W.Y., Mahoney, D.P., and Young, B.R. A Real-Time Approach to Process Control, John Wiley and Sons, Ltd.
Edinburgh University, Exercise: Relative Gain Arrays, 2006
Edinburgh University, Multivariable Systems, 2007
12.2: MIMO control using RGA is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Ben van Kuiken, Curt Longcore,
Angela Yeung, Jeffrey Carey, & Jeffrey Carey via source content that was edited to conform to the style and standards of the LibreTexts platform; a detailed
12.3: MIMO USING MODEL PREDICTIVE CONTROL
This section describes how to control a system with multiple inputs and outputs using Model Predictive Control (MPC). MPC is a linear
algebra method for predicting the result of a sequence of control variable manipulations. Once the results of specific manipulations are
predicted, the controller can then proceed with the sequence that produces the desired result. One can compare this controller method to
"look ahead" in chess or other board games. In look ahead, you foresee what an action might yield some time in the future using specific
knowledge of the process (or game in the case of chess), and are thereby able to optimize your actions to select for the best long term
outcome. MPC methods can prevent an occurrence with conventional PID controllers in which actions taken achieve short term goals, but
end up very costly in the end. This phenomenon can be described as "winning the battle but losing the war."
The open ended nature of MPC allows the process control engineer use MPC to control any system for which models can be generated.
12.3.2: MODEL PREDICTIVE CONTROL

MPC is a widely used means to deal with large multivariable constrained control issues in industry. The main aim of MPC is to minimize a
performance criterion in the future that would possibly be subject to constraints on the manipulated inputs and outputs, where the future
behavior is computed according to a model of the plant. The model predictive controller uses the models and current plant measurements to
calculate future moves in the independent variables that will result in operation that honors all independent and dependent variable
constraints. The MPC then sends this set of independent variable moves to the corresponding regulatory controller set-points to be
implemented in the process.
MPC uses the mathematical expressions of a process model to predict system behavior. These predictions are used to optimize the process
over a defined time period. An MPC controller can operate according to the following algorithm.
1. Development of a process model by the control engineers.
2. At time t, previous process inputs and outputs are used, along with the process model, to predict future process outputs u(f) over a
"prediction horizon."
3. The control signals that produce the most desired behavior are selected.
4. The control signal is implemented over a pre-defined time interval.
5. Time advances to the next interval, and the procedure is repeated from step 2.
This is one of the many algorithm possibilities, which can be applied to systems with any number of inputs or outputs. The process model
can be variable as well. Examples include physical models, input-output models, and state models which are all derived from the specific
system being controlled.
When comparing predicted behavior to desired behavior, there are multiple techniques. A common procedure is to generate a second
mathematical model that describes your desired behavior. When process behavior is predicted in step 2 of the MPC algorithm, the control
signals which produce the predicted behavior that minimize deviations from your desired behavior over the [t, t + h] interval, are selected.
The generation of models and optimization process is repeated continuously as the algorithm is repeated.
As seen in the figure above, depending on the algorithm, MPC may generate possible outcomes given possible controller action. These
generations are either based on past process outputs, or the process model. After many possible outcomes are generated, the controller can
pick one based on the optimization goals. This generation and optimization process is repeated at every time step.
The flow diagram below depicts the flow of information used by the controller.
The figure above shows the basic structure of a Model Predictive Controller. The model takes data from past inputs and outputs, and
combines it with the predicted future inputs, and gives a predicted output for the time step. This predicted output is combined with the
reference trajectory, giving the predicted future errors of the system. These errors are fed into an optimizer, which enforces the constraints of
the system (for instance, ensuring that a flow rate calculation for the model is not greater than the maximum flow) on the predicted outputs
and minimizes the operating cost function. This gives the predicted future inputs, which are fed back into the main model, restarting the
cycle.
12.3.2.1: MOTIVATION
The motivation for the development and implementation of MPC controllers is compelling. MPC is very simple for sampled systems in
which the control signal is constant over the interval [t,t+h]. The value of h can then be taken as the sampling interval, and the prediction
horizon can become a small number of sampling intervals. This can reduce the computational power needed to implement a model
predictive controller.
A highly desired feature of MPC is that constraints can be implemented in the controller. These constraints include heaters and valves that
have a finite operating range, actuators with finite states (on/off or low/high), and cost or energy limits for the process. MPC can incorporate
these constraints and eliminate the possibility of variables exceeding their limits. This helps the process operate efficiently, prevents damage
to equipment, and prevents the system from running away by continuously increasing a variable's setting when the limit for the equipment
has already been reached. Another benefit of MPC controllers incorporating system constraints is the ability of the model to dynamically
react to system changes. For example, if a valve is stuck open, it can be added as a constraint, and the model will compensate accordingly.
This allows the controller to continue effectively controlling the system after an equipment malfunction. In a simpler control, such as PID
control, this would not be possible.
MPC is a highly specific method for controlling a process. Each controller is specific to the system it was designed for and the model
equations, constraints, and set points will change for different systems. This means that a controller developed for a tank reactor will not be
able to control an evaporation unit as the process model will be very different. This weakness in MPC brings out its biggest strength. The
specificity and customizable nature of the controller will empower you with the freedom to design for exactly what is desired.
MPC is a flexible control technique that uses discrete time segments and is the most commonly applied advanced control technique in the
chemical process industry. MPC helps to simplify or completely eliminate controller design and instead works as a system modeling
controller. With MPC the designer does not have to worry about optimizing control parameters such as with PID control. The ability of
MPC controllers to handle constraints in an optimal fashion is also a contributor to its success.
12.3.2.2: MODEL PREDICTIVE CONTROL EXAMPLE
To demonstrate the concepts of MPC, a general example for the development of a sampled process and a first-order system will be shown.
General Model
Take Equation 12.3.1 as our process model:
y(t) + a1 y(t − h) + … + an y(t − nh) = b1 u(t − h) + b2 u(t − 2h) + … + bn (t − nh) (12.3.1)
where u is controller input, y is process output, and h is the time interval. This is a general equation that relates previous process output
y(t − h) and previous controller input u(t − h) . In some situations this equation may be created using fitted experimental data, but is most
often a derivation using knowledge of your specific system and fundamentals of chemical engineering. At time t, total previous behavior yp
is shown as Equation 12.3.2 .
yp = f(y(t), y(t − h), … , u(t − h), u(t − 2h), …) (12.3.2)
Future process output y can be predicted using current and future control signals, u(t) and u(t + h) respectively:
f
yf = f(u(t), u(t − h), … , u(t + N h)) (12.3.3)
Both y and y could possibly be created by fits of experimental data, but are more likely to be derived from specific equations related to
p f
your system. Deviations from the desired behavior y , either specified by another mathematical model or reference trajectory, produce an
d
error function e(t) = y(t) − y (t) for increments of control actions Δu(t) = u(t) − u(t − h) . The loss function J to be minimized is
d
shown as Equation 12.3.4 .

t∣N
2 2
J (u(t), u(t − h), … , u(t + N h)) = ∑ e(t + kh) + ρ(Δu(t + (k − 1)h)) (12.3.4)
l=1
The control inputs that minimize Equation 12.3.4 are then applied to the system by the controller over the time interval, and the process is
repeated. The control input function F in Equation 12.3.5 is determined implicitly by the optimization.
u(t) = F (y(t), y(t − h), … y(t − nh), u(t − h), y(t − 2h), … , u(t − nh)) (12.3.5)
This general model is meant to be a guideline, and the equations listed representative, for the thought process required to create a model
predictive controller.
First-Order System Example
Take the process model to be Equation 12.3.6 :
Δy(t + h) = −aΔy(t) + bΔu(t) (12.3.6)
Let us define Δy(t) = y(t) − y(t − h) and Δu(t) = u(t) − u(t − h) . Let us also define our desired system behavior y as a function d
which starts at y(t) and exponentially approaches a set point y with time constant T . Our desired behavior y then becomes Equation
sp d
12.3.7 :
−h
yd (t + h) = y(t) + (1 − e T ) (ysp − y(t)) (12.3.7)
Assuming that our controller can take as much action as needed to produce the desired behavior, the desired behavior can be realized in the
next sampling period. This is done by setting y(t + h) equal to y (t + h) , and can be seen in Equation 12.3.8 :
d
−h
y(t + h) = y(t) + Δy(t + h) = y(t) − αΔy(t) + bΔu(t) = y(t) + (1 − e T ) (ysp − y(t)) (12.3.8)
Solving Equation 12.3.8 for u(t) gives Equation 12.3.9 :

−h
a 1− e T
Δu(t) = Δy(t) + (ysp − y(t)) (12.3.9)

b b
−h
Upon examination of this result, you can see that we have produced a PI controller with gains k = and k = . It should be noted
a 1−e T
i
b b
that the proportional gain k will only depend upon the developed process model, and the integral gain ki depends on both the process model
and the desired response rate T.
This process can be modified to include multiple inputs and outputs via the process model and desired behavior. We may also add system
constraints to our MPC example by adding conditional statements limiting the maximum controller output discussed in the logical programs
wiki. As discussed before, there are as many variations on the MPC process as you can think of. See the worked examples 1 and 2 for more
MPC instances.
 QUESTION
Is the above expression a PI control system or more of a PD control system? The first term is k*(y(present0)-y(past)) which seems
more like a derivative term. The second term is more like k*(y(set)-y(present)) which seems more like a P term.
12.3.3: DIFFERENCES FROM OTHER CONTROLLERS TYPES

Model predictive control uses a mathematical model to simulate a process. This model then fits the inputs to predict the system behavior. In
this way, MPC is a type of feed forward control. It uses system inputs as a basis of control. MPC is more complex than most other feed
forward control types because of the way these predictions are used to optimize a process over a defined amount of time. Most feed forward
control types do not take into account the process outputs much past a residence time. The MPC algorithm will compare predicted outputs
to desired outputs and select signals that will minimize this difference over the time selected. This control type can see ahead into multiple
time steps in the future in order to optimize the process. Normal PID type controllers use mathematical expressions based on error from a
set point. The governing equation for MPC controllers are based on set points, system properties, and desired outcomes and optimization.
MPC is very specific to the process it is modeling. Unlike ratio or cascade control set ups, where it is simple to implement and change set
points in various situations, MPC will model one specific process and optimize it. As previously mentioned this can either be an advantage
or disadvantage. MPC is great for selecting one type of operation on one system and perfecting it to the desired conditions. This also has a
downside in that the model equation will work for one and only one situation.
12.3.4: LIMITATIONS OF MPC

12.3.4.1: ADVANTAGES OF MPC
1. MPC can be used to handle multivariable control programs.
2. MPC can consider actuator limitations.
3. MPC can increase profits by allowing for operation close to the system constraints.
4. MPC can perform online computations quickly.
5. MPC can be used for non-minimal phase and unstable processes.
6. MPC is easy to tune.
7. MPC is able to handle structural changes.
12.3.4.2: DISADVANTAGES OF MPC

1. Several MPC models are limited to only stable, open-loop processes.
2. MPC often requires a large number of model coefficients to describe a response.
3. Some MPC models are formulated for output disturbances, and they may not handle input disturbances well.
4. Some forms of MPC use a constant output disturbance assumption. This corrects for the fact that the output predicted by the model is
not exactly equal to the actual measured output. This method assumes the correction term is constant in the future, which may not yield a
good performance if there is a real disturbance at the plant input.
5. If the prediction horizon is not formulated correctly, control performance will be poor even if the model is correct.
6. Some systems have a wide range of operating conditions that change frequently. Some examples of this include exothermic reactors,
batch processes, and any systems where different consumers have different product specifications. An MPC linear model will not be able
to handle the dynamic behavior of these processes. A nonlinear model must be used for better control performance.
12.3.5: INDUSTRIAL MPC APPLICATIONS

There are many industrial applications that incorporate model predictive control in order to effectively control a multivariable system. In
order to effectively do this, one needs to set up a working model by testing many different parameters in a plant. This is usually done by
starting up a plant, varying many different parameters, and having the MPC program analyze the data.
In order to test the plant, one may vary parameters such as:
1. feed flow rate/composition
2. steam pressure
3. heat duty
4. recycle ratio
5. reactor temperatures
This is only a small sample of parameters that can be changed. Once these parameters have been changed, the data is analyzed, and
downstream effects of these parameters are characterized as a function of these variables. This relationship can be a combination of many
different relationships (linear, nonlinear, logarithmic, exponential, power, etc.).
By testing the plant thoroughly and coming up with a robust model, the engineers ensure that an MPC controller will be able to much more
effectively run the plant. Barring any major process changes, this model should be accurate for normal use.
Some industrial MPC applications are:
1. Model Predictive Heuristic Control by Richard et al. 1976 (Adersa)
2. Dynamic Matrix Control (DMC) by Cutler and Ramaker 1979 (Shell Oil)
3. Quadratic-Program Dynamic Matrix (QDMC) Control by Cutler et al. 1983 (Shell Oil)
4. IDCOM-M by Setpoint, Inc (part of ASPEN Technology)
5. Generalized Predictive Control (GPC)
12.3.6: IMPLEMENTING MPC USING EXCEL

In MPC, values of the control variables will be optimized for a given time interval in order to best tell the system how it should act. The
control variables will be optimized by optimizing some characteristic. Usually, this characteristic is simply the least squared error between
an actual state and a "set" or desired state. This can easily be done using the Solver tool in Excel. See MPC_in_Excel for an example of
implementing MPC to control the valve, v1, in the following system (keep reading to learn about everything in the Excel file).
To describe this system in words, there is a variable feed, Fin, with time that pours into a storage tank. The flow out of the tank, Fout, is
controlled by the valve, v1, multiplied by a constant of proportionality, k1 = 50. The tank volume is to be maintained at 150.
There are three scenarios contained in the example Excel file. In the first scenario, the valve is continuous and in the second, it is binary
(can be open or closed only). In the third scenario, the valve is still binary; however, Solver cannot recognize this setup even though this
setup may seem intuitive (this is discussed further below). As it turns out, it doesn't really matter if a binary or continuous valve is used.
Both types of valves yield very similar results; however, the continuous valve yields slightly tighter control over the tank volume (see
screenshot below).
As you can see from this screen shot, the sum of the least squares between the actual volume and the set volume (Row 22) is lowest for the
1st scenario, when the valve is continuous. This measure of deviation from the set value is only slightly lower than in the 2nd scenario,
when the valve is binary. In the 3rd scenario, Solver didn't work properly, resulting in all of the v1 variables (Column P) to be set to 0.
At the start of the simulation, the volume of the tank is 0, so v1 stays closed until time = 5.4 when the volume is very close to the set
volume. If the deviations that occurred during the time v1 was closed (which is the same time length for all 3 scenarios) is removed from the
sum of the least squares (Row 23) the 1st scenario suddenly looks much better at controlling the volume than the 2nd scenario.
There are a couple things to note while using Solver in Excel to perform MPC:
1) This note only applies if your system has a binary control variable. If you intend to use a conditional statement (such as an IF() function)
that depends on the optimization of the control variable, you must leave open the possibility that this control variable can be any real
number, whether it's binary or not. This may seem unnecessary because Solver will output only optimized discreet numbers. To put this in
the context of the example Excel file, you might think you could have simply done the following to control Fout:
= IF(v1 = 1, k1*1, 0)
(The variables "v1" and "k1" are not recognized by Excel. Instead of typing the variables into Excel, cells containing the variable values
must be referenced.)
This statement says that if the valve is open, allow a flow of k1 (or 50), otherwise, the flow is 0. This would not allow Solver to work
properly. If Solver chose a value of v1 = 0.1, measured the target cell, and then it chose v1 = 0.2, it wouldn't see any difference because in
both cases the flow would be 0. Solver needs to be able to see a change in the system, whether it's practical or not, to determine the actual
gradient of your system. This gradient cannot be determined if an IF() statement turns the continuous output to discreet output.
The proper way to handle this is to program a constraint into solver that allows only binary outputs of v1 values (select bin from the pull
down menu when adding the constraint). F can then be programmed as:
out
= v1 * k1
Using this method, Solver can calculate the gradient of the system to define how it would react if v1 changed to a non-binary value. By
using this method, Solver will output only binary values for the valve state as is desired.
2) Solver can only handle manipulating 200 variables at a time, so if your model requires looking at more than 200 time steps, you will have
to optimize your system in sections.
 EXAMPLE 12.3.1
You have a tank outside your home used to supply water for various household functions (i.e. flushing the toilet). The optimal water
pressure from the tank is achieved when the height of water in the tank is at 15, and the tank can hold water up to a height of 30,
because when it rains, it pours. When it does rain, the level of water in this tank exceeds 15 and a model predictive controller and valve
is in place to release water so the height will drop back down to this value (don't worry, this water is not wasted, but used as drinking
water on your llama farm). At time t=25, there is a downpour that increases the flow into the tank by 150 and lasts for 100 time steps.
Will this tank overflow, and how long will it take for the tank to return to the desired set point? Your controller output is 25 maximum.
Other key pieces of info:
H(0)= 15
A= 100
B= 1
T= 15
Hset= 15
h= 1
Solution
To obtain the required information, we will produce a model predictive controller for the water tank system, and simulate how the
system will behave during the downpour. The first step is to generate a mathematical model that describes the relevant parameters.
Equation 12.3.10 below relates the fluid height in the tank H to the flow in, and controller signal u.
dH −−−−
A = Fin − Bu√H(t) (12.3.10)
dt
where A and B are physical constants of the system. Next we will define the way we wish the height of fluid in the tank to act in
Equation 12.3.11 .
−h/T
H(t + h) = H(t) + (1 − e ) (Hset − H(t)) (12.3.11)
Where h is the time interval, T is an exponential time constant, and Hset is the desired liquid set point of 15. It can be seen by examining
equation 11 that as time proceeds, the exponential term will reduce to zero, resulting in H(t + h) = Hset. The next step is to figure out
how to produce a controller input that will give the desired behavior. This may be done by calculating the error between process model
and desired behavior, and optimizing for u. Fortunately for us, this is a relatively simple set of equations that can be solved for u
explicitly. By expanding Equation 12.3.10 using the fundamental theorem of calculus to produce Equation 12.3.12 , and plugging
Equation 12.3.11 in for H(t + h), equation 13 results solved for u.
H(t + h) − H(t) −−−−
A = Fin − Bu√H(t) (12.3.12)
h
−h −h
(−AHset + AH(t) + Ae T ∗ Hset − Ae T H(t) + Fin h)
u = −−−
− (12.3.13)
(B√Hset h)
These equations can be implemented in a spreadsheet to model the behavior of the height of liquid in the tank, as well as the actions
taken by the controller to restore the set point. Keep in mind the 0 to 25 output restriction on your controller.
This spreadsheet has been made with the relevant equations and can be used to solve this problem and model tank behavior. This
spreadsheet can be found here:MPC example 1
So yes, the tank will overflow, and will return to its set point after 200 time points.
 EXAMPLE 12.3.2
What happens if the actuator on the valve breaks and must be replaced with an older one you happened to have in your basement? This
new (although old) actuator cannot operate the valve fast enough, and B is decreased from 1 to 0.8. How does the graph from example
1 change? What does this tell you about the physical meaning of parameter B?
Solution
This B can easily be changed in the spreadsheet used in example 1. With this different actuator, and hence different B, we can see that
the tank acquires more volume, and it takes longer for the tank to return to its optimal height of 15. This is because the valve on the
tank cannot be opened as fast. Parameter B represents the physical properties/limitations of the valve, and pipe properties.
www.youtube.com/v/XNccO4neVfc
www.youtube.com/v/2HkpKgXlyfE
 EXERCISE 12.3.1
Which of the following is a highly desired feature for model predictive control?
a. It is ideal in determining short-term behavior.
b. It is the simplest of all feed forward controller types.
c. It accounts for system constraints.
d. Controllers of this type are universal and can be used without modification for any process.
Answer
C
 EXERCISE 12.3.2
Will a model predictive controller developed for a tank reactor be able to control an evaporation unit?
a. Yes, as they are both standard pieces of chemical engineering equipment.
b. Maybe, it depends if the reaction taking place in the reactor is exothermic or endothermic.
c. No, the governing equations will be very different.
d. Yes, because evaporation units and tank reactors can both use steam to supply heat.
Answer
C
 EXERCISE 12.3.3
Which of the following statements about MPC is false?

a. MPC uses a defined time interval.
b. MPC utilizes feed forward control.
c. MPC requires a certain number of inputs and outputs.
d. MPC can generate more than one possible outcome.
Answer
C
12.3.8: REFERENCES
Astrom, Karl J. and Hagglund, Tore. Advanced PID Control, Department of Automatic Control, Lund Institute of Technology, Lund
University.
Bequette, B. Wayne. Process Control: Modeling, Design and Simulation. Prentice Hall PTR. pp. 487-511, December 26, 2002.
Garcia C.E, Prett D.M. and Morari M. Model Predictive Control: Theory and Practice - a Survey. Automatica, Vol.25, No.3, pp. 335-
348, 1989.
Orukpe, P.E. Basics of Model Predictive Control. Imperial College, London. April 14, 2005.
12.3: MIMO using Model Predictive Control is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Bradley Anderson, Aaron
Bennick, Michael Salciccioli, & Michael Salciccioli via source content that was edited to conform to the style and standards of the LibreTexts platform; a
12.4: NEURAL NETWORKS FOR AUTOMATIC MODEL CONSTRUCTION
Authors: Group E - Ardemis Boghossian, James Brown, Sara Zak
Multiple Input Multiple Output (MIMOs) are systems that require multiple inputs and generate multiple outputs. MIMOs are controlled by
controllers that combine multiple input readings in an algorithm to generate multiple output signals. MIMOs can be used with a variety of
algorithms. The most versatile algorithm used to date is the neural network. Neural networks, which were initially designed to imitate
human neurons, work to store, analyze, and identify patterns in input readings to generate output signals. In chemical engineering, neural
networks are used to predict the ouputs of systems such as distillation columns and CSTRs. This article will discuss how neural networks
work, the advantages and disadvantages of neural networks, and some common applications of the networks.
12.4.2: MIMOS
As mentioned, Multiple Inputs Multiple Outputs (MIMOs) are systems that require multiple inputs and generate multiple outputs, as shown
schematically in Figure 1.
Figure 1. Schematic diagram of MIMOs

As shown in the figure, MIMOs are controlled by controllers that combine multiple input readings in an algorithm to generate multiple
output signals. Typically, MIMOs do not require that the the number of inputs and outputs be the same. Instead, any number of input
readings could be used to generate any number of output signals.
12.4.3: NEURAL NETWORKS

Various types of controllers can be used to control MIMOs. One of the most accurate and versatile controllers used for MIMOs is the neural
network. Neural networks are controllers that crudely imitate the human neuron. Initially, these networks were designed to model neural
brain activity. However, as people began to recognize the advantages of neural networks, these networks were applied to controller
algorithm design. Like a human neuron, neural networks work to store, analyze, and identify patterns in data by performing learning tasks.
The ability of these networks to "learn" parallels the human neuron's ability to learn, making these automatic controllers the closest analog
to a human controller.
12.4.3.1: NEURONS
Like neurons in the body, network neurons receive inputs, store this data, and transmit outputs to either another neuron or directly to the
MIMO. In order to transmit this data, the neuron must relate the multiple inputs to the multiple outputs. A simple mathematical
representation of this relationship is shown below.
n
y = f (w1 a1 + w2 a2 + w3 a3 + … + wn an ) = f (∑ wi ai )
i=1
where
wi = weight
ai = input
y = output
f = sigmoid function (any nonlinear function)
According to this relationship, the multiple input parameters are each multiplied by the corresponding weight factor, wi. These weight
factors "weigh" the significance of each input, scaling each input proportionally to the effect it will have on the output. These weighted
inputs are then added and the sum is input into the sigmoid function to generate an output. This output can then be sent to multiple neurons
that, in turn, each generate its own output.
The sigmoid function in this relationship is a nonlinear, empirical function that relates the input readings to the output signals. This
empirical function can take on many forms depending on the data set. The equation that is best able to predict the outputs for the given
system will be used (polynomial, sine, logarithmic, etc.). For example, one form this function may take is the hyperbolic sine function,
where
f(x) = sinh(αx)
where
x = sum of weighted inputs =
α = empirical parameter
In this sigmoid function, α is an empirical parameter that adjusts the function outputs. The effect of α on this sigmoid function is shown in
Figure 2.
Figure 2. Plot of hyperbolic sine sigmoid function with varying αx

As shown in Figure 2, increasing α increases the output for this particular sigmoid function. Like most empirical functions, the hyperbolic
sine sigmoid function can only be used within a specified range of values (x values between the vertical asymptotes). In this case, the range
would depend on the value of α.
12.4.3.2: COMBINING NEURONS INTO NEURAL NETWORKS
Once neurons have been programmed to correlate input and output data, they can be connected in a feedforward series to produce a neural
network, or neural net (NN). A schematic diagram of a neural network is shown in Figure 3.
Figure 3. Schematic diagram of a neural network.

Figure 3 shows MIMO parameters, such as temperature, pressure, and flow readings, are first processed in the first layer of neurons. The
outputs of the first layer of neurons then serve as the inputs to the second layer. The outputs of the second layer then become the inputs to
the third layer, and so on, until the final output of the network is used to directly affect MIMO controls such as valves. The layers of neurons
between the initial and final layers are known as hidden layers.
The way neurons within hidden layers correlate the inputs and outputs is analogous to the way individual neurons correlate these variables.
As shown in the flowchart inset of the diagram, a neuron within the network receives multiple inputs. Each of these input parameters are
then weighted, having the most effective parameters weighted more heavily. These weighted input values are added and input into the
particular sigmoid function the neuron is programmed to follow. The output of this function is then sent to other neurons as input. These
input values are then reweighed for that particular neuron. This process is continued until the final output of the neuron network is used to
adjust the desired controls. Although the diagram shows only one hidden layer, these neuron networks can consist of multiple layers.
Although almost all continuous functions can be approximated by a single hidden layer, incorporating multiple hidden layers decreases the
numbers of weights used. Since more layers will result in more parameters, each of these individual parameters will be weighted less. With
more parameters and less weights, the system becomes more sensitive to parameter change (a greater "rippling" effect within the network).
The "rippling" effect of a neuron network makes the system difficult to model analytically. A small change in a single input variable would
result in multiple changes throughout the the entire network. Although modelling these complex networks is beyond the scope of the class,
only a basic, qualitative understanding of how neural networks function is necessary to analyze neural network controllers and their effects
on input and output parameters.
12.4.3.3: LEARNING PROCESS

The ability of neural networks to learn distinguishes them from most automatic controllers. Like humans, neural networks learn by example,
and thus need to be trained. Neural networks are usually configured to specific applications and have the ability to process large amounts of
data. Complex trends and patterns can be detected by neural networks that would otherwise be imperceptible to humans or other computing
programs. Within neural networks, there are learning procedures that allow the device to recognize a certain pattern and carry out a specific
task. These learning procedures consist of an algorithm that enables the network to determine the weighting parameters in order to match
the given data (inputs and outputs) with a function. In this iterative procedure, the initial input values are used to generate initial output
values. Based on these input and output values, the weights within the network are adjusted to match the data. These adjusted weights are
then used to correlate the next pair of input and output values. Again, these values are used to adjust weights. This process continues until
the network obtains a good fit for the data. A flow chart summarizing this iterative process is shown in Figure 4.
Figure 4. Flowchart for training a network.

The neural network learns by tracing the path of an object over a specific area and sequence. For this application, inputs for the neural
network included the position, velocity, and direction. Once the the neural network has completed its training, it will use what it learned to
predict trajectories. The neural network can predict where the object will move given the position, velocity, and direction of the object.
The trajectories shown in these videos demonstrate an important characteristic of neural networks. In these videos, the neural network can
only predict outputs for objects that behave similar to the output it encountered during training. For instance, if a person was to start running
forward, then stop, spin in a circle, and start running backwards, the neural network output would be unpredictable.
Once the neural network becomes capable of predicting system outputs, it is run like any other controller. In real-time, this often entails the
chemical process and a computer system of some sort. Often, such as in LabVIEW, the chemical engineer is presented with a user-friendly
data acquisition program that allows the user to set desired temperatures, flowrates, etc. and displays the system's input and outputs.
Although the neural network continually sends signals to the system controllers (such as valves), the network algorithm is embedded in the
acquisition program. Therefore, once the system is up and running, the user does not directly see the algorithm working.

Given the interesting, human-like behavior of neural networks, one would expect all process applications to be controlled by neural
networks. However, the advantages and disadvantages of neural networks limit their use in applications. The following lists summarize
these advantages and disadvantages.
Advantages
Neural networks are very general and can capture a variety of patterns very accurately
The static, non-linear function used by neural networks provide a method to fit the parameters of a particular function to a given set of
data.
A wide variety of functions can be used to fit a given set of data
Neural networks do not require excessive statistical training
There is no need to assume an underlying input data distribution when programming a neural network
Neural networks can detect all possible, complex nonlinear relationships between input and outputs
Neural networks are the closest thing to having an actual human operate a system (i.e. they can "learn")
Disadvantages
Neural networks are difficult to design. One must determine the optimal number of nodes, hidden layers, sigmoid function, etc.
Neural networks are diffcult to model analytically because a small change in a single input will affect the entire network
The operation of neural networks is limited to the training process. If the network is trained poorly, then it will operate poorly and the
outputs cannot be guaranteed.
There is a great computational burden associated with neural networks
Neural networks require a large sample size in order to empirically fit data
Neural networks have a “black box” nature. Therefore, errors within the complex network are difficult to target.
Outside of their data training range, neural networks are unpredictable. This occurs because neural networks may "overfit" data. For
instance, during training, a neural network may fit a 10th order polynomial to only 5 data points. When using this model to make
predictions of values outside this 5-point range, the neural network behaves unpredictably.
12.4.5: APPLICATIONS OF NEURAL NETWORKS

Each neural network will generate a different algorithm based on the inputs and outputs to the system. Because neural networks fit a specific
function to the given data, they can be used in a variety of applications. Within chemical engineering, neural networks are frequently used to
predict how changing one input (such as pressure, temperature, etc.) on a distillation column will influence the compositions and flow rates
of the streams exiting the column. The network training is performed on various inputs to the column, and thus can predict how changing
one input will affect the product streams. Within a CSTR, neural networks can be used to determine the effect of one input parameter such
as temperature or pressure on the products.
In addition to its applications to chemical equipment, neural networks can also be applied to model material responsess as a function of
various loads under various conditions. These models can then be used in product development to create a device for a particular
application, or to improve an existing device. For example, by modeling the corrosion of steel under different temperature and pH
conditions, implanted biomedical devices can be manufactured or improved.
Neural networks are also often used in biology and biological applications to predict the outcome of a certain event. For example, neural
networks can be used to predict the growth of cells and bacteria in cell culture labs, given a set of varying conditions, such as temperature
and pH. In addition, neural network models have been used to predict the mortality rate in intensive care units in hospitals. Data was
collected from different patients, and a neural network model was created to predict the mortality of future patients given a set of specified
conditions. Neural networks have also been used to diagnose breast cancer in patients by predicting the effects of a tumor provided specified
input conditions of the patient.
Neural networks are also used in applications beyond the chemical aspect of controllers. For instance, neural networks are used to predict
travel time based on different travel conditions. Signs on the highways that give estimated travel times are examples of neural networks.
They predict the amount of time required to reach a certain destination given the varying traffic volume, road conditions, and weather
conditions.
Neural networks are not limited to the applications listed above. They can be used to model most predictable events, and the complexity of
the network will increase depending on the situation. For more information on the uses listed above, refer to the journal articles listed in the
references section of this wiki.
 EXAMPLE 12.4.1
Hypothetical Industries has expanded to now include a biology lab! As an employee, you are working on developing a new antibiotic,
and thus have been assigned to predict the growth of bacteria. Your boss wants to know how the growth of the bacteria is affected by
different conditions, such as temperatures, pH, and provided nutrients. You don't really feel like manipulating all these conditions, and
then sitting and watching bacteria grow on a petri dish, so you decide to come up with a way to predict the how the bacteria will grow.
Using the information presented in the wiki, determine what the inputs and outputs to this neural network are.
Solution
The inputs to the neural network are each of the parameters that may affect the growth of the bacteria. In this situation, the inputs are
the temperature, pH, and nutrients provided (such as sugars, amino acids, and antibiotics). The outputs from this system include the
growth of the bacteria.
 EXAMPLE 12.4.2
As seen in the example above, a neural network can be used to predict bacterial growth. Given the information presented in the wiki,
explain the advantages and disadvantages of using a neural network to model bacterial growth.
Solution
Advantages
Because there are so many inputs (temperature, pH, etc.), a neural network fits a function to this data that is able to predict how
future conditions would affect the bacterial growth.
Neural networks provide a mechanical method of modeling the system of bacterial growth that is extremely similar to have an
acutal human predict the growth of bacteria.
Someone monitoring the system does not need much statistical training to use the neural network.
Disadvantages
You must first run a large number of samples with varying conditions in order to obtain a good fit for the neural network. Based on
how well you would like your network to predict bacterial growth, the number of samples that must be run in order to create the
network changes. The network will only function as well as it is trained, so if a high degree of accuracy is desired, then more data
inputs will be required. This becomes time comsuming and expensive.
Because of the black box nature of the neural networks, it is difficult to determine how the individual parameters such as
temperature or pH will affect the bacterial growth.
Neural networks cannot be used to predict the growth patterns of the bacteria outside of the given data ranges.
 EXERCISE 12.4.1
Why would someone want to increase the number of hidden layers when combining neurons into a neural network?
a. To decrease the amount of programming needed
b. To decrease the number of weights required
c. To increase the cost associated with the system
d. To increase the aesthetic structure of the system
 EXERCISE 12.4.2
What does the neural network output if its inputs are outside its training range?
a. Zero
b. Input^2
c. sqrt(input)
d. The outputs outside of range are unpredictable
12.4.6: REFERENCES
Accurate freeway travel time prediction with state-space neural networks under missing data. Van-zuylen. Transportation research part c.
2005/10/12. 13(5-6) pp347-369.
Astrom, K.J., Hagglund, T. Advanced PID Control. ISA- The Istrumentation Systems and Automation Society.
Campos, Lucio P.A., Silva, Aristorfanes C. Barros, Allan Kardec. Diagnosis of breast cancer in digital mammograms using independent
component analysis and neural networks. Lecture Notes in Computer Science, vol 3773, pub 2005, p460-469.
Chan, CH, Chow, PY. Application of artificial neural networks to establish a predictive mortality risk model in children admitted to
pediatric intensive care unit. Singapore Med J. Volume 47, Issue 11, Date 2006/10/31, pages 928-934.
Yu, C. Davidson, VJ, Yang, SX. A neural network approach to predict durvival/death and growth/no-growth interfaces for Escherichia
coli O157:H7. Food Microbiology [Food Microbiol.]. Vol. 23, no. 6, pp. 552-560. Sep 2006
12.4: Neural Networks for automatic model construction is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et
12.5: UNDERSTANDING MIMO CONTROL THROUGH TWO TANKS
INTERACTION
We have been familiar with the models of single surge tank manipulated by first order process and two tanks in series manipulated by
second order process, both of which are typical examples of Single Input Single Output (SISO) control. However, in the real chemical
processes, there are always interations between the reactors. The following page will discuss the two tanks model by taking into
consideration the interaction between the two tanks. To manipulate this model, we need to use Multiple Input Multiple Output (MIMO)
control, which will add more complexity in understanding the overall process.
12.5.1: TWO TANKS INTERACTION MODEL

The following figure, Figure 1, demonstrates how two tanks model works to interact each other. From Figure 1, we can see that the input of
Tank1, x1, not only affects the level in Tank1, but also affects that in Tank2. The same goes for the input of Tank2, x2. We use the
resistance, R3, to account for the interaction between the two tanks.
Figure 1: Two tanks interaction model

In the above figure,
x1 = Input of Tank1
x2 = Input of Tank2
A1 = Cross section area of Tank1
A2 = Cross section area of Tank2
h1 = Level of Tank1
h2 = Level of Tank2
R1 = Resistance of Tank1
R2 = Resistance of Tank2
R3 = Interaction Resistance
12.5.1.1: MATHEMATICAL EQUATIONS FOR THE PROCESS

Now, we will derive the mathematical equations to describe the process, now we assume the direction of the flow goes from Tank1 to
Tank2:
To begin with, we write down the governing equation for each of the tanks, taking into consideration the interaction term:
dh1 h1 (h1 − h2 ) 1 1 1
A1 = x1 − dt − = x1 + h1 (− − ) + h2 ( )
dt R1 R3 R1 R3 R3
dh2 h2 (h1 − h2 ) 1 1 1
A2 = x2 − dt + = x2 + h2 (− − ) + h1 ( )
dt R2 R3 R2 R3 R3
Under steady state, the time derivatives, i.e. the left hand side of the above equations, go to zero:
1 1 1
0 = x1 (0) + h1 (0) (− − ) + h2 (0) ( )
R1 R3 R3
1 1 1 1
0 = x2 (0) + h1 (0) ( − ) + h2 (0) (− − )
R3 R3 R2 R3
Now, we define the deviation variables as follows:
X1 = x1 − x1 (0)
Y1 = h1 − h1 (0)
X2 = x2 − x2 (0)
Y2 = h2 − h2 (0)
Set y1 = h1 and y
2 = h2 , we can obtain the following equations:
dy1 x1 1 1 1
A1 = + y1 (− − ) + y2 ( )
dt A1 A1 R1 A1 R3 A1 R3
dy2 x2 1 1 1
A2 = + y1 ( ) + y2 (− − )
dt A1 A2 R3 A2 R2 A2 R3
It will be more general if we write down the above equations into a form of Matrix:
1 1 1 1
⎡ 0 ⎤ ⎡ − − ⎤
ẏ 1
⎢ A1 ⎥ x1
⎢ A1 R1 A1 R2 A1 R3 ⎥ y1
[ ] = ⎢ ⎥[ ]+⎢ ⎥[ ]
ẏ 2 ⎢ 1 ⎥ x2 ⎢ 1 1 1 ⎥ y2
⎣ 0 ⎦ ⎣ − − ⎦
A2 A2 R3 A2 R2 A2 R3
We can represent the above equation using the following equation:

→ →
⃗ ⃗
y ⃗ = B x⃗ + A y ⃗
→
⃗ ⃗
y ⃗ − Ay ⃗ = B x⃗
→ →−1
−
−1 ∂
⃗ ⃗ ⃗
x⃗ = (B )( − A ) y ⃗ = Gp y ⃗
∂t
Invert the above equation:
Now, set
We get the following equation:

^ ^
y1 Gp1,1 Gp1,2 x1
[ ] = [ ][ ] (12.5.1)
y2 ^ ^ x2
Gp2,1 Gp2,2
12.5.1.2: CONTROL DIAGRAM
Figure 2: Contol Diagram for Two tanks interaction model.
12.5.1.3: DECOUPLE THE PROCESS
12.5: Understanding MIMO Control Through Two Tanks Interaction is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by
Omar Ismail, Yuan Chen, Erin Baragar, & Erin Baragar via source content that was edited to conform to the style and standards of the LibreTexts platform;
a detailed edit history is available upon request.
CHAPTER OVERVIEW
13: STATISTICS AND PROBABILITY BACKGROUND

13.1: Basic statistics- mean, median, average, standard deviation, z-scores, and p-value
13.2: SPC- Basic Control Charts- Theory and Construction, Sample Size, X-Bar, R charts, S charts
13.4: Bayes Rule, Conditional Probability and Independence
13.5: Bayesian Network Theory
13.6: Learning and analyzing Bayesian networks with Genie
13.7: Occasionally Dishonest Casino? Markov Chains and Hidden Markov Models
13.8: Continuous Distributions- normal and exponential
13.9: Discrete Distributions - Hypergeometric, Binomial, and Poisson
13.11: Comparisons of two means
13.12: Factor analysis and ANOVA
13.13: Correlation and Mutual Information
13.14: Random sampling from a stationary Gaussian process
13: Statistics and Probability Background is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source
1
13.1: BASIC STATISTICS- MEAN, MEDIAN, AVERAGE, STANDARD DEVIATION,
Z-SCORES, AND P-VALUE
Statistics is a field of mathematics that pertains to data analysis. Statistical methods and equations can be applied to a data set in order to
analyze and interpret results, explain variations in the data, or predict future data. A few examples of statistical information we can calculate
are:
Average value (mean)
Most frequently occurring value (mode)
On average, how much each measurement deviates from the mean (standard deviation of the mean)
Span of values over which your data set occurs (range), and
Midpoint between the lowest and highest value of the set (median)
Statistics is important in the field of engineering by it provides tools to analyze collected data. For example, a chemical engineer may wish
to analyze temperature measurements from a mixing tank. Statistical methods can be used to determine how reliable and reproducible the
temperature measurements are, how much the temperature varies within the data set, what future temperatures of the tank may be, and how
confident the engineer can be in the temperature measurements made. This article will cover the basic statistical functions of mean, median,
mode, standard deviation of the mean, weighted averages and standard deviations, correlation coefficients, z-scores, and p-values.
13.1.1: WHAT IS A STATISTIC?

In the mind of a statistician, the world consists of populations and samples. An example of a population is all 7th graders in the United
States. A related example of a sample would be a group of 7th graders in the United States. In this particular example, a federal health care
administrator would like to know the average weight of 7th graders and how that compares to other countries. Unfortunately, it is too
expensive to measure the weight of every 7th grader in the United States. Instead statistical methodologies can be used to estimate the
average weight of 7th graders in the United States by measure the weights of a sample (or multiple samples) of 7th graders.
Parameters are to populations as statistics are to samples.
A parameter is a property of a population. As illustrated in the example above, most of the time it is infeasible to directly measure a
population parameter. Instead a sample must be taken and statistic for the sample is calculated. This statistic can be used to estimate the
population parameter. (A branch of statistics know as Inferential Statistics involves using samples to infer information about a populations.)
In the example about the population parameter is the average weight of all 7th graders in the United States and the sample statistic is the
average weight of a group of 7th graders.
A large number of statistical inference techniques require samples to be a single random sample and independently gathers. In short, this
allows statistics to be treated as random variables. A in-depth discussion of these consequences is beyond the scope of this text. It is also
important to note that statistics can be flawed due to large variance, bias, inconsistency and other errors that may arise during sampling.
Whenever performing over reviewing statistical analysis, a skeptical eye is always valuable.
Statistics take on many forms. Examples of statistics can be seen below.
13.1.2: BASIC STATISTICS

When performing statistical analysis on a set of data, the mean, median, mode, and standard deviation are all helpful values to calculate.
The mean, median and mode are all estimates of where the "middle" of a set of data is. These values are useful when creating groups or bins
to organize larger sets of data. The standard deviation is the average distance between the actual data and the mean.
13.1.2.1: MEAN AND WEIGHTED AVERAGE

The mean (also know as average), is obtained by dividing the sum of observed values by the number of observations, n. Although data
points fall above, below, or on the mean, it can be considered a good estimate for predicting subsequent data points. The formula for the
mean is given below as Equation 13.1.1 . The excel syntax for the mean is AVERAGE(starting cell: ending cell).
i=n
∑ Xi
i=1
X̄ = (13.1.1)
n
However, equation (1) can only be used when the error associated with each measurement is the same or unknown. Otherwise, the weighted
average, which incorporates the standard deviation, should be calculated using equation (2) below.
∑ wi xi
Xwav = (13.1.2)
∑ wi
where
1
wi =
2
σ
i
and x is the data value.

i
13.1.2.2: MEDIAN
The median is the middle value of a set of data containing an odd number of values, or the average of the two middle values of a set of data
with an even number of values. The median is especially helpful when separating data into two equal sized bins. The excel syntax to find
the median is MEDIAN(starting cell: ending cell).
13.1.2.3: MODE
The mode of a set of data is the value which occurs most frequently. The excel syntax for the mode is MODE(starting cell: ending cell).
13.1.2.4: CONSIDERATIONS
Now that we've discussed some different ways in which you can describe a data set, you might be wondering when to use each way. Well, if
all the data points are relatively close together, the average gives you a good idea as to what the points are closest to. If on the other hand,
almost all the points fall close to one, or a group of close values, but occasionally a value that differs greatly can be seen, then the mode
might be more accurate for describing this system, whereas the mean would incorporate the occasional outlying data. The median is useful
if you are interested in the range of values your system could be operating in. Half the values should be above and half the values should be
below, so you have an idea of where the middle operating point is.
13.1.2.5: STANDARD DEVIATION AND WEIGHTED STANDARD DEVIATION
The standard deviation gives an idea of how close the entire set of data is to the average value. Data sets with a small standard deviation
have tightly grouped, precise data. Data sets with large standard deviations have data spread out over a wide range of values. The formula
for standard deviation is given below as Equation 13.1.3 . The excel syntax for the standard deviation is STDEV(starting cell: ending cell).
−−−−−−−−−−−−−−−− −
 i=n
 1 2
¯
σ =  ∑ ( Xi − X ) (13.1.3)
⎷ n− 1
i=1
Side Note: Bias Estimate of Population Variance

The standard deviation (the square root of variance) of a sample can be used to estimate a population's true variance. Equation 13.1.3 above
is an unbiased estimate of population variance. Equation 13.1.4 is another common method for calculating sample standard deviation,
although it is an bias estimate. Although the estimate is biased, it is advantageous in certain situations because the estimate has a lower
variance. (This relates to the bias-variance trade-off for estimators.)
−−−−−−−−−−−− −

i=n
1 2
σn =  ∑ ( Xi − X̄ ) (13.1.4)
⎷n
i=1
When calculated standard deviation values associated with weighted averages, Equation 13.1.5 below should be used.
1
σwav = −−−− (13.1.5)
√∑ wi
13.1.2.6: THE SAMPLING DISTRIBUTION AND STANDARD DEVIATION OF THE MEAN

Population parameters follow all types of distributions, some are normal, others are skewed like the F-distribution and some don't even have
defined moments (mean, variance, etc.) like the Chaucy distribution. However, many statistical methodologies, like a z-test (discussed later
in this article), are based off of the normal distribution. How does this work? Most sample data are not normally distributed.
This highlights a common misunderstanding of those new to statistical inference. The distribution of the population parameter of interest
and the sampling distribution are not the same. Sampling distribution?!? What is that?
Imagine an engineering is estimating the mean weight of widgets produced in a large batch. The engineer measures the weight of N widgets
and calculates the mean. So far, one sample has been taken. The engineer then takes another sample, and another and another continues until
a very larger number of samples and thus a larger number of mean sample weights (assume the batch of widgets being sampled from is near
infinite for simplicity) have been gathered. The engineer has generated a sample distribution.
As the name suggested, a sample distribution is simply a distribution of a particular statistic (calculated for a sample with a set size) for a
particular population. In this example, the statistic is mean widget weight and the sample size is N. If the engineer were to plot a histogram
of the mean widget weights, he/she would see a bell-shaped distribution. This is because the Central Limit Theorem guarantees that as the
sample size approaches infinity, the sampling distributions of statistics calculated from said samples approach the normal distribution.
Conveniently, there is a relationship between sample standard deviation (σ) and the standard deviation of the sampling distribution ( -
also know as the standard deviation of the mean or standard error deviation). This relationship is shown in Equation 13.1.6 below:
σX
σ¯ = −
− (13.1.6)
X
√N
An important feature of the standard deviation of the mean, is the factor in the denominator. As sample size increases, the standard
deviation of the mean decrease while the standard deviation, σ does not change appreciably.
Microsoft Excel has built in functions to analyze a set of data for all of these values. Please see the screen shot below of how a set of data
could be analyzed using Excel to retrieve these values.
13.1.2.7: EXAMPLE BY HAND

You obtain the following data points and want to analyze them using basic statistical methods. {1,2,2,3,5}
Calculate the average: Count the number of data points to obtain n = 5
1+ 2+ 2+ 3+ 5
mean = = 2.6
5
Obtain the mode: Either using the excel syntax of the previous tutorial, or by looking at the data set, one can notice that there are two 2's,
and no multiples of other data points, meaning the 2 is the mode.
Obtain the median: Knowing the n=5, the halfway point should be the third (middle) number in a list of the data points listed in ascending
or descending order. Seeing as how the numbers are already listed in ascending order, the third number is 2, so the median is 2.
Calculate the standard deviation: Using Equation 13.1.3 ,
−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−− −
1 2 2 2 2 2
σ = √ (1 − 2.6) + (2 − 2.6) + (2 − 2.6) + (3 − 2.6) + (5 − 2.6) = 1.52
5− 1
13.1.2.8: EXAMPLE BY HAND (WEIGHTED)

Three University of Michigan students measured the attendance in the same Process Controls class several times. Their three answers were
(all in units people):
Student 1: A = 100 ± 3
Student 2: A = 105 ± 4
Student 3: A = 102 ± 2
What is the best estimate for the attendance A?
, , ,
Therefore,
A = 101.92 ± 0.65 students
13.1.2.9: GAUSSIAN DISTRIBUTION

Gaussian distribution, also known as normal distribution, is represented by the following probability density function:
2
(x −μ)
1 −
2
P DFμ,σ (x) = −−e
2σ
σ√2π
where μ is the mean and σ is the standard deviation of a very large data set. The Gaussian distribution is a bell-shaped curve, symmetric
about the mean value. An example of a Gaussian distribution is shown below.
In this specific example, μ = 10 and σ = 2.

Probability density functions represent the spread of data set. Integrating the function from some value x to x + a where a is some real value
gives the probability that a value falls within that range. The total integral of the probability density function is 1, since every value will fall
within the total range. The shaded area in the image below gives the probability that a value will fall between 8 and 10, and is represented
by the expression:
Gaussian distribution is important for statistical quality control, six sigma, and quality engineering in general. For more information see
What is 6 sigma?.
13.1.2.10: ERROR FUNCTION
A normal or Gaussian distribution can also be estimated with a error function as shown in the equation below.
2
10 (x −μ)
1 −
2
P (8 ≤ x ≤ 10) = ∫ −−e
2σ
dx = erf(t)
8 σ√2π
Here, erf(t) is called "error function" because of its role in the theory of normal random variable. The graph below shows the probability of
a data point falling within t*σ of the mean.
For example if you wanted to know the probability of a point falling within 2 standard deviations of the mean you can easily look at this
table and find that it is 95.4%. This table is very useful to quickly look up what probability a value will fall into x standard deviations of the
mean.
13.1.2.11: CORRELATION COEFFICIENT (R VALUE)

The linear correlation coefficient is a test that can be used to see if there is a linear relationship between two variables. For example, it is
useful if a linear equation is compared to experimental points. The following equation is used:
∑ (Xi − Xmean ) (Yi − Ymean )
r = −−−−−−−−−−−−−− −−−−−−−−−−−−−
2 2
√∑ (Xi − Xmean ) √∑ (Yi − Ymean )
The range of r is from -1 to 1. If the r value is close to -1 then the relationship is considered anti-correlated, or has a negative slope. If the
value is close to 1 then the relationship is considered correlated, or to have a positive slope. As the r value deviates from either of these
values and approaches zero, the points are considered to become less correlated and eventually are uncorrelated.
There are also probability tables that can be used to show the significant of linearity based on the number of measurements. If the
probability is less than 5% the correlation is considered significant.
13.1.2.12: LINEAR REGRESSION
The correlation coefficient is used to determined whether or not there is a correlation within your data set. Once a correlation has been
established, the actual relationship can be determined by carrying out a linear regression. The first step in performing a linear regression is
calculating the slope and intercept:
\[\mathit{Slope} = \frac{n\sum_i X_iY_i -\sum_i X_i \sum_j Y_j }
\nonumber \]
\[\mathrm{Intercept} = \frac{(\sum_i X_i^2)\sum_i(Y_i)-\sum_i X_i\sum_i X_iY_i }
\nonumber \]
Once the slope and intercept are calculated, the uncertainty within the linear regression needs to be applied. To calculate the uncertainty, the
standard error for the regression line needs to be calculated.
−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−− −

 1
2
S =  ((∑ Y ) − intercept ∑ Yi − slope(∑ Yi Xi ))
i
⎷ n− 2
i i
The standard error can then be used to find the specific error associated with the slope and intercept:
−−−−−−−−−−−−−−−−−
n
Sslope = S
√ 2 2
n ∑i X − (∑i Xi )
i
−−−−−−−−−−−−−−−−−− −
 2
 ∑ (X )
i
Sintercept = S 
2
⎷ n (∑ X 2 ) − (∑ X Y )
i i i i
Once the error associated with the slope and intercept are determined a confidence interval needs to be applied to the error. A confidence
interval indicates the likelihood of any given data point, in the set of data points, falling inside the boundaries of the uncertainty.
∗
β = slope ± Δslope ≃ slope ± t Sslope
∗
α = intercept ± Δintercept ≃ intercept ± t Sintercept
Now that the slope, intercept, and their respective uncertainties have been calculated, the equation for the linear regression can be
determined.
Y = βX + α
13.1.2.13: Z-SCORES
A z-score (also known as z-value, standard score, or normal score) is a measure of the divergence of an individual experimental result from
the most probable result, the mean. Z is expressed in terms of the number of standard deviations from the mean value.
X−μ
z = (13.1.7)
σ
X = ExperimentalValue
μ = Mean
σ = StandardDeviation
Z-scores assuming the sampling distribution of the test statistic (mean in most cases) is normal and transform the sampling distribution into
a standard normal distribution. As explained above in the section on sampling distributions, the standard deviation of a sampling
distribution depends on the number of samples. Equation (6) is to be used to compare results to one another, whereas equation (7) is to be
used when performing inference about the population.
Whenever using z-scores it is important to remember a few things:
Z-scores normalize the sampling distribution for meaningful comparison.
Z-scores require a large amount of data.
Z-scores require independent, random data.
X−μ
zobs = (13.1.8)
σ
√n
n = SampleNumber
13.1.2.14: P-VALUE
A p-value is a statistical value that details how much evidence there is to reject the most common explanation for the data set. It can be
considered to be the probability of obtaining a result at least as extreme as the one observed, given that the null hypothesis is true. In
chemical engineering, the p-value is often used to analyze marginal conditions of a system, in which case the p-value is the probability that
the null hypothesis is true.
The null hypothesis is considered to be the most plausible scenario that can explain a set of data. The most common null hypothesis is that
the data is completely random, that there is no relationship between two system results. The null hypothesis is always assumed to be true
unless proven otherwise. An alternative hypothesis predicts the opposite of the null hypothesis and is said to be true if the null hypothesis is
proven to be false.
The following is an example of these two hypotheses:
4 students who sat at the same table during in an exam all got perfect scores.
Null Hypothesis: The lack of a score deviation happened by chance.
Alternative Hypothesis: There is some other reason that they all received the same score.
If it is found that the null hypothesis is true then the Honor Council will not need to be involved. However, if the alternative hypothesis is
found to be true then more studies will need to be done in order to prove this hypothesis and learn more about the situation.
As mentioned previously, the p-value can be used to analyze marginal conditions. In this case, the null hypothesis is that there is no
relationship between the variables controlling the data set. For example:
Runny feed has no impact on product quality
Points on a control chart are all drawn from the same distribution
Two shipments of feed are statistically the same
The p-value proves or disproves the null hypothesis based on its significance. A p-value is said to be significant if it is less than the level of
significance, which is commonly 5%, 1% or .1%, depending on how accurate the data must be or stringent the standards are. For example, a
health care company may have a lower level of significance because they have strict standards. If the p-value is considered significant (is
less than the specified level of significance), the null hypothesis is false and more tests must be done to prove the alternative hypothesis.
Upon finding the p-value and subsequently coming to a conclusion to reject the Null Hypothesis or fail to reject the Null Hypothesis, there is
also a possibility that the wrong decision can be made. If the decision is to reject the Null Hypothesis and in fact the Null Hypothesis is true,
a type 1 error has occurred. The probability of a type one error is the same as the level of significance, so if the level of significance is 5%,
"the probability of a type 1 error" is .05 or 5%. If the decision is to fail to reject the Null Hypothesis and in fact the Alternative Hypothesis is
true, a type 2 error has just occurred. With respect to the type 2 error, if the Alternative Hypothesis is really true, another probability that is
important to researchers is that of actually being able to detect this and reject the Null Hypothesis. This probability is known as the power
(of the test) and it is defined as 1 - "probability of making a type 2 error."
If an error occurs in the previously mentioned example testing whether there is a relationship between the variables controlling the data set,
either a type 1 or type 2 error could lead to a great deal of wasted product, or even a wildly out-of-control process. Therefore, when
designing the parameters for hypothesis testing, researchers must heavily weigh their options for level of significance and power of the test.
The sensitivity of the process, product, and standards for the product can all be sensitive to the smallest error.
 IMPORTANT NOTE ABOUT SIGNIFICANT P-VALUES
If a P-value is greater than the applied level of significance, and the null hypothesis should not just be blindly accepted. Other tests
should be performed in order to determine the true relationship between the variables which are being tested. More information on this
and other misunderstandings related to P-values can be found at P-values: Frequent misunderstandings.
Calculation
There are two ways to calculate a p-value. The first method is used when the z-score has been calculated. The second method is used with
the Fisher’s exact method and is used when analyzing marginal conditions.
First Method: Z-Score
The method for finding the P-Value is actually rather simple. First calculate the z-score and then look up its corresponding p-value using the
standard normal table.
This table can be found here: Media:Group_G_Z-Table.xls
This value represents the likelihood that the results are not occurring because of random errors but rather an actual difference in data sets.
To read the standard normal table, first find the row corresponding to the leading significant digit of the z-value in the column on the
lefthand side of the table. After locating the appropriate row move to the column which matches the next significant digit.
Example:
If your z-score = 1.13
Follow the rows down to 1.1 and then across the columns to 0.03. The P-value is the highlighted box with a value of 0.87076.
Values in the table represent area under the standard normal distribution curve to the left of the z-score.
Using the previous example:
Z-score = 1.13, P-value = 0.87076 is graphically represented below.
Second Method: Fisher's Exact

In the case of analyzing marginal conditions, the P-value can be found by summing the Fisher's exact values for the current marginal
configuration and each more extreme case using the same marginals. For information about how to calculate Fisher's exact click the
following link:Discrete_Distributions:_hypergeometric,_binomial,_and_poisson#Fisher.27s_exact
13.1.2.15: CHI-SQUARED TEST

A Chi-Squared test gives an estimate on the agreement between a set of observed data and a random set of data that you expected the
measurements to fit. Since the observed values are continuous, the data must be broken down into bins that each contain some observed
data. Bins can be chosen to have some sort of natural separation in the data. If none of these divisions exist, then the intervals can be chosen
to be equally sized or some other criteria.
The calculated chi squared value can then be correlated to a probability using excel or published charts. Similar to the Fisher's exact, if this
probability is greater than 0.05, the null hypothesis is true and the observed data is not significantly different than the random.
Calculating Chi Squared
The Chi squared calculation involves summing the distances between the observed and random data. Since this distance depends on the
magnitude of the values, it is normalized by dividing by the random value
N 2
(observed − random)
2
χ = ∑
random
k=1
or if the error on the observed value (sigma) is known or can be calculated:

N 2
2
observed − theoretical
χ = ∑( )
sigma
k=1
Detailed Steps to Calculate Chi Squared by Hand

Calculating Chi squared is very simple when defined in depth, and in step-by-step form can be readily utilized for the estimate on the
agreement between a set of observed data and a random set of data that you expected the measurements to fit. Given the data:
Step 1: Find
2
(yi − A − Bxi )
2
χo = ∑
2
σ
i yi
When:
2
Δ = S Sxx − (Sx )
The Excel function CHITEST(actual_range, expected_range) also calculates the value. The two inputs represent the range of data the
actual and expected data, respectively.
Step 2: Find the Degrees of Freedom
When: df = Degrees of Freedom

n = number of observations
k = the number of constraints
Step 3: Find
~2
χo
= the established value of obtained in an experiment with df degrees of freedom

Step 4: Find using Excel or published charts.
The Excel function CHIDIST(x,df) provides the p-value, where x is the value of the chi-squared statistic and df is the degrees of freedom.
Note: Excel gives only the p-value and not the value of the chi-square statistic.
= the probability of getting a value of that is as large as the established
Step 5: Compare the probability to the significance level (i.e. 5% or 0.05), if this probability is greater than 0.05, the null hypothesis is true
and the observed data is not significantly different than the random. A probability smaller than 0.05 is an indicator of independence and a
significant difference from the random.
Chi Squared Test versus Fisher's Exact
For small sample sizes, the Chi Squared Test will not always produce an accurate probability. However, for a random null, the Fisher's
exact, like its name, will always give an exact result.
Chi Squared will not be correct when:
1. fewer than 20 samples are being used
2. if an expected number is 5 or below and there are between 20 and 40 samples
For large contingency tables and expected distributions that are not random, the p-value from Fisher's Exact can be a difficult to
compute, and Chi Squared Test will be easier to carry out.
13.1.2.16: BINNING IN CHI SQUARED AND FISHER’S EXACT TESTS
When performing various statistical analyzes you will find that Chi-squared and Fisher’s exact tests may require binning, whereas ANOVA
does not. Although there is no optimal choice for the number of bins (k), there are several formulas which can be used to calculate this
number based on the sample size (N). One such example is listed below:
k = 1 + log2 N
Another method involves grouping the data into intervals of equal probability or equal width. The first approach in which the data is
grouped into intervals of equal probability is generally more acceptable since it handles peaked data much better. As a stipulation, each bin
should contain at least 5 or more data points, so certain adjacent bins sometimes need to be joined together for this condition to be satisfied.
Identifying the number the bins to use is important, but it is even more important to be able to note which situations call for binning. Some
Chi-squared and Fisher's exact situations are listed below:
Analysis of a continuous variable:
This situation will require binning. The idea is to divide the range of values of the variable into smaller intervals called bins.
Analysis of a discrete variable:
Binning is unnecessary in this situation. For instance, a coin toss will result in two possible outcomes: heads or tails. In tossing ten coins,
you can simply count the number of times you received each possible outcome. This approach is similar to choosing two bins, each
containing one possible result.
Examples of when to bin, and when not to bin:
You have twenty measurements of the temperature inside a reactor: as temperature is a continuous variable, you should bin in this case.
One approach might be to determine the mean (X) and the standard deviation (σ) and group the temperature data into four bins: T < X –
σ, X – σ < T < X, X < T < X + σ, T > X + σ
You have twenty data points of the heater setting of the reactor (high, medium, low): since the heater setting is discrete, you should not
bin in this case.
 EXAMPLE 13.1.1
Say we have a reactor with a mean pressure reading of 100 and standard deviation of 7 psig. Calculate the probability of measuring a
pressure between 90 and 105 psig.
Solution
To do this we will make use of the z-scores.
b− μ a−μ
Pr(a ≤ z ≤ b) = F (b) − F (a) = F ( )−F ( )
σ σ
where a is the lower bound and b is the upper bound

Substitution of z-transformation equation (3)
Look up z-score values in a standard normal table. Media:Group_G_Z-Table.xls
So:
= 0.76155 - 0.07636
= 0.68479.
The probability of measuring a pressure between 90 and 105 psig is 0.68479.
A graphical representation of this is shown below. The shaded area is the probability
13.1.2.17: ALTERNATE SOLUTION

We can also solve this problem using the probability distribution function (PDF). This can be done easily in Mathematica as shown
below. More information about the PDF is and how it is used can be found in the Continuous Distribution article
As you can see the the outcome is approximately the same value found using the z-scores.
 EXAMPLE 13.1.2
You are a quality engineer for the pharmaceutical company “Headache-b-gone.” You are in charge of the mass production of their
children’s headache medication. The average weight of acetaminophen in this medication is supposed to be 80 mg, however when you
run the required tests you find that the average weight of 50 random samples is 79.95 mg with a standard deviation of .18.
a. Identify the null and alternative hypothesis.
b. Under what conditions is the null hypothesis accepted?
c. Determine if these differences in average weight are significant.
Solution
a)
Null hypothesis: This is the claimed average weight where Ho=80 mg
Alternative hypothesis: This is anything other than the claimed average weight (in this case Ha<80)
b) The null hypothesis is accepted when the p-value is greater than .05.
c) We first need to find Zobs using the equation below:
X−μ
zobs = σ
√n
Where n is the number of samples taken.

79.95 − 80
zobs = = −1.96
.18
√50
Using the z-score table provided in earlier sections we get a p-value of .025. Since this value is less than the value of significance (.05)
we reject the null hypothesis and determine that the product does not reach our standards.
 EXAMPLE 13.1.3
15 students in a controls class are surveyed to see if homework impacts exam grades. The following distribution is observed.
Determine the p-value and if the null hypothesis (Homework does not impact Exams) is significant by a 5% significance level using the
P-fisher method.
Solution
To find the p-value using the p-fisher method, we must first find the p-fisher for the original distribution. Then, we must find the p-
fisher for each more extreme case. The p-fisher for the original distribution is as follows.
9!6!8!7!
pfisher = = 0.195804
15!6!3!2!4!
To find the more extreme case, we will gradually decrease the smallest number to zero. Thus, our next distribution would look like the
following.
The p-fisher for this distribution will be as follows.

9!6!8!7!
pfisher = = 0.0335664
15!7!2!1!5!
The final extreme case will look like this.
The p-fisher for this distribution will be as follows.

9!6!8!7!
pfisher = = 0.0013986
15!8!1!0!6!
Since we have a 0 now in the distribution, there are no more extreme cases possible. To find the p-value we will sum the p-fisher values
from the 3 different distributions.
pvalue = 0.195804 + 0.0335664 + 0.0013986 = 0.230769
Because p-value=0.230769 we cannot reject the null hypothesis on a 5% significance level.
13.1.3: APPLICATION: WHAT DO P-VALUES TELL US?

13.1.3.1: POPULATION EXAMPLE
Out of a random sample of 400 students living in the dormitory (group A), 134 students caught a cold during the academic school year. Out
of a random sample of 1000 students living off campus (group B), 178 students caught a cold during this same time period.
Population table
Graup A Group B
Sick a = 134 b = 178 a + b = 312
Not Sick c = 266 d = 822 c + d = 1088
a + c = 400 b + d = 1000 a + b + c + d = 1400
Fisher's Exact:
(a + b)!(c + d)!(a + c)!(b + d)!
pf =
(a + b + c + d)!a!b!c!d!
Solve:
(312)!(1088)!(400)!(1000)!
pf =
(1400)!134!178!266!822!
pf = 2.28292 * 10 − 10
Comparison and interpretation of p-value at the 95% confidence level
This value is very close to zero which is much less than 0.05. Therefore, the number of students getting sick in the dormitory is significantly
higher than the number of students getting sick off campus. There is more than a 95% chance that this significant difference is not random.
Statistically, it is shown that this dormitory is more condusive for the spreading of viruses. With the knowledge gained from this analysis,
making changes to the dormitory may be justified. Perhaps installing sanitary dispensers at common locations throughout the dormitory
would lower this higher prevalence of illness among dormitory students. Further research may determine more specific areas of viral
spreading by marking off several smaller populations of students living in different areas of the dormitory. This model of significance
testing is very useful and is often applied to a multitude of data to determine if discrepancies are due to chance or actual differences between
compared samples of data. As you can see, purely mathematical analyses such as these often lead to physical action being taken, which is
necessary in the field of Medicine, Engineering, and other scientific and non-scientific venues.
 EXERCISE 13.1.1
You are given the following set of data: {1,2,3,5,5,6,7,7,7,9,12} What is the mean, median and mode for this set of data? And then the z
value of a data point of 7?
a. 5.82, 6, 7, 0.373
b. 6, 7, 5.82, 6.82
c. 7, 6, 5, 0.373
d. 7, 6, 5.82, 3.16
Answer
a
 EXERCISE 13.1.2
What is n and the standard deviation for the above set of data {1,2,3,5,5,6,7,7,7,9,12}? And then consulting the table from above, what
is the p-value for the data "12"?
a. 12, 3.16, 5.82
b. 7, 3.16, 0.83
c. 11, 3.16, 0.97
d. 11, 5.82, 0
Answer
c
13.1.4: REFERENCES
Woolf P., Keating A., Burge C., and Michael Y.. "Statistics and Probability Primer for Computational Biologists". Massachusetts
Institute of Technology, BE 490/ Bio7.91, Spring 2004
Smith W. and Gonic L. "Cartoon Guide to Statistics". Harper Perennial, 1993.
Taylor, J. "An Introduction to Error Analysis". Sausalito, CA: University Science Books, 1982.
http://www.fourmilab.ch/rpkp/experiments/analysis/zCalc.html
13.1: Basic statistics- mean, median, average, standard deviation, z-scores, and p-value is shared under a CC BY 3.0 license and was authored, remixed,
and/or curated by Andrew MacMillan, David Preston, Jessica Wolfe, Sandy Yu, & Sandy Yu via source content that was edited to conform to the style and
standards of the LibreTexts platform; a detailed edit history is available upon request.
13.2: SPC- BASIC CONTROL CHARTS- THEORY AND CONSTRUCTION,
SAMPLE SIZE, X-BAR, R CHARTS, S CHARTS
Authors: Chris Bauman, Jennifer De Heck, Evan Leonard, Merrick Miranda
Stewards: Eric Black, Stacy Young, Dan Carter, Megan Boekeloo
Control charts are one of the most commonly used methods of Statisical Process Control (SPC), which monitors the stability of a process.
The main features of a control chart include the data points, a centerline (mean value), and upper and lower limits (bounds to indicate where
a process output is considered "out of control").They visually display the fluctuations of a particular process variable, such as temperature,
in a way that lets the engineer easily determine whether these variations fall within the specified process limits. Control charts are also
known as Shewhart charts after Walter Shewhart, who developed them in the early 1900’s.
13.2.2: CONTROL CHART BACKGROUND

A process may either be classified as in control or out of control. The boundaries for these classifications are set by calculating the mean,
standard deviation, and range of a set of process data collected when the process is under stable operation. Then, subsequent data can be
compared to this already calculated mean, standard deviation and range to determine whether the new data fall within acceptable bounds.
For good and safe control, subsequent data collected should fall within three standard deviations of the mean. Control charts build on this
basic idea of statistical analysis by plotting the mean or range of subsequent data against time. For example, if an engineer knows the mean
(grand average) value, standard deviation, and range of a process, this information can be displayed as a bell curve, or population density
function (PDF). The image below shows the control chart for a data set with the PDF overlay.
Figure I. Control chart showing PDF for a data set

The centerline is the mean value of the data set and the green, blue and red lines represent one, two, and three standard deviations from the
mean value. In generalized terms, if data points fall within three standard deviations of the mean (within the red lines), the process is
considered to be in control. These rules are discussed in greater detail later in this section.
Control Charts are commonly used in six sigma control today, as a means of overall process improvement. For more on six-sigma control,
see six sigma.
13.2.3: CONTROL CHART FUNCTIONS

The main purpose of using a control chart is to monitor, control, and improve process performance over time by studying variation and its
source. There are several functions of a control chart:
1. It centers attention on detecting and monitoring process variation over time.
2. It provides a tool for ongoing control of a process.
3. It differentiates special from common causes of variation in order to be a guide for local or management action.
4. It helps improve a process to perform consistently and predictably to achieve higher quality, lower cost, and higher effective capacity.
5. It serves as a common language for discussing process performance.
13.2.4: SAMPLE SIZE AND SUBGROUPING

There are a few key conditions that must be met when constructing control charts:
The initial predictions for the process must be made while the process is assumed to be stable. Because future process quality will be
compared to these predictions, they must be based off of a data set that is taken while the operation is running properly.
Multiple subsets of data must be collected, where a subset is simply a set of n measurements taken over a specific time range. The
number of subsets is represented as k. A subset average, subset standard deviation, and subset range will be computed for each subset.
From these subsets, a grand average, an average standard deviation, and an average range are calculated. The grand average is the
average of all subset averages. The average standard deviation is simply the average of subset standard deviations. The average range is
simply the average of subset ranges.
The upper and lower control limits for the process can then be determined from this data.
Future data taken to determine process stability can be of any size. This is because any point taken should fall within the statistical
predictions. It is assumed that the first occurrence of a point not falling within the predicted limits shows that the system must be
unstable since it has changed from the predictive model.
The subsets are defined, based on the data and the process. For example, if you were using a pH sensor, the sensor would most likely
output pages of data daily. If you know that your sensor has the tendency to drift every day, you might select a 30 minute subset of data.
If it drifts monthly you might set your subset to be 24 hours or 12 hours.
Finally, the population size, N is assumed to be infinite. Alternatively, if the population is finite but the sample size is less than 5% of the
population size, we can still approximate the population to be near infinite. That is, n/N <= 0.05 where n is the sample size and N is the
population size. [5]
13.2.5: X-BAR, R-CHARTS, AND S-CHARTS

There are three types of control charts used determine if data is out of control, x-bar charts, r-charts and s-charts. An x-bar chart is often
paired with either an r-chart or an s-chart to give a complete picture of the same set of data.
Pairing X-Bar with R-Charts
X-Bar (average) charts and R (range) -charts are often paired together. The X-Bar chart displays the centerline, which is calculated using the
grand average, and the upper and lower control limits, which are calculated using the average range. Future experimental subsets are plotted
compared to these values. This demonstrates the centering of the subset values. The R-chart plots the average range and the limits of the
range. Again, the future experimental subsets are plotted relative to these values. The R-chart displays the dispersion of the subsets. X-
Bar/R-Chart plot a subgroup average. Note that they should only be used when subgroups really make sense. For example, in a Gage R&R
study, when operators are testing in duplicates or more, subgrouping really represents the same group.
Pairing X-Bar with S-Charts
Alternatively, X-Bar charts can be paired with S-charts (standard deviation). This is typically done when the size of the subsets are large.
For larger subsets, the range is a poor statistic to estimate the distributions of the subsets, and instead, standard deviation is used. In this
case, the X-Bar chart will display control limits that are calculated using the average standard deviation. The S-Charts are similar to the R-
charts; however, instead of the range, they track the standard deviation of multiple subsets.
Smoothing Data with a Moving Average
If it is desired to have smooth data, the moving average method is one option. This method involves taking the average of a number of
points, and using that average for the middle data point. From this point on, the data is treated the same as any normal group of k subsets.
Though this method will produce a smoother curve, it has a lag in detecting points, which may be problematic if the points are out of the
acceptable range. This time lag would keep the control system from reacting to the problem until after the average is found. For this reason,
moving average charts are appropriate mainly for slower processes that can handle the lag.
For example, let us calculate a value for a set of data which takes samples every second. We will use an average of 10 points to find this,
however, in practice there is no set number of data points that should be used. For the point t = 50, we must wait until data has been
collected through t = 54. The points are then averaged for t = 45-54 and used as the function value. For the next point, t = 51, the average of
the points for t = 46-55 are used, and so on. If this is still confusing, please see moving average for a more detailed explanation.
Reading Control Charts
Control charts can determine whether a process is behaving in an "unusual" way.
Note: The upper and lower control limits are calculated using the grand average and either the average range and average sigma. Example
calculations are shown in the Creating Control Charts Section.
The quality of the individual points of a subset is determined unstable if any of the following occurs:
Rule 1: Any point falls beyond 3σ from the centerline(this is represented by the upper and lower control limts).
Rule 2: Two out of three consecutive points fall beyond 2σ on the same side of the centerline.
Rule 3: Four out of five consecutive points fall beyond 1σ on the same side of the centerline.
Rule 4: Nine or more consecutive points fall on the same side of the centerline.
Figure III. Quality control rules.

The quality of a subset is determined unstable according to the following rules:
1. Any subset value is more than three standard deviations from the centerline.
2. Two consecutive subset values are more than two standard deviations from the centerline and are on the same side of the centerline.
3. Three consecutive subset values are more than one standard deviation from the centerline and are on the same side of the centerline.
Creating Control Charts
To establish upper and lower control limits on control charts, there are a number of methods. We will discuss the method for the number of
components in a subset, n, less than 15. For methods involving n > 15 and other techniques, see Process Control and Optimization, Liptak,
2.34. Here, the table of constants for computing limits, and the limit equations are presented below.
Please note that Table A below does NOT contain data for a sample problem. Any time you make a control chart, you refer to this table. The
values in the table are used in the equations for the upper control limit (UCL), lower control limit (LCL), etc. This will be explained in the
examples below. If you are interested in how these constants were derived, there is a more detailed explanation in Control Chart Constants.
Table A: Table of Constants
To determine the value for n, the number of subgroups
In order to determine the upper (UCL) and lower (LCL) limits for the x-bar charts, you need to know how many subgroups (n) there are in
your data. Once you know the value of n, you can obtain the correct constants (A2, A3, etc.) to complete your control chart. This can be
confusing when you first attend to create a x-bar control chart. The value of n is the number of subgroups within each data point. For
example, if you are taking temperature measurements every min and there are three temperature readings per minute, then the value of n
would be 3. And if this same experiment was taking four temperature readings per minute, then the value of n would be 4. Here are some
examples with different tables of data to help you further in determining n:
Example 1:
n= 4 since there are four readings of kg.
Example 2:
n= 4 since there are four readings of pH.
Example 3:
n= 3 since there are three readings of temperature.
After creating multiple control charts, determining the value of n will become quite easy.
Calculating UCL and LCL
For the X-Bar chart the following equations can be used to establish limits, where is the grand average, is the average range, and
is the average standard deviation.
Calculating Grand Average, Average Range and Average Standard Deviation
To calculate the grand average, first find the average of the n readings at each time point. The grand average is the average of the averages
at each time point.
To calculate the grand range, first determine the range of the n readings at each time point. The grand range is the average of the ranges at
each time point.
To calculate the average standard deviation, first determine the standard deviation of the n readings at each time point. The average standard
deviation is the average of the standard deviations at each time point.
Note: You will need to calculate either the grand range or the average standard deviation, not both.
For X-bar charts, the UCL and LCL may be determined as follows:
Alternatively, can be used as well to calculate UCL and LCL:
The centerline is simply .

For R-charts, the UCL and LCL may be determined as follows:
The centerline is the value .

For S-charts, the UCL and LCL may be determined as follows:
The centerline is .
The following flow chart demonstrates the general method for constructing an X-bar chart, R-chart, or S-chart:
Calculating Region Boundaries

To determine if your system is out of control, you will need to section your data into regions A, B, and C, below and above the grand
average. These regions are shown in Figure III. To calculate the boundaries between these regions, you must first calculate the UCL and
LCL. The boundaries are evenly spaced between the UCL and LCL. One way to calculate the boundaries is shown below.
Boundary between A and B above XGA = XGA + (UCL − XGA) * 2 / 3
Boundary Between B and C above XGA = XGA + (UCL − XGA) * 1 / 3
Boundary Between A and B below XGA = LCL + (XGA − LCL) * 2 / 3
Boundary Between B and C below XGA = LCL + (XGA − LCL) * 2 / 3
13.2.6: EXAMPLE 1
Assume that in the manufacture of 1 kg Mischmetal ingots, the product weight varies with the batch. Below are a number of subsets taken at
normal operating conditions (subsets 1-7), with the weight values given in kg. Construct the X-Bar, R-charts, and S-charts for the
experimental data (subsets 8-11). Measurements are taken sequentially in increasing subset number.
Solution:
First, the average, range, and standard deviation are calculated for each subset.
Next, the grand average XGA, average range RA, and average standard deviation SA are computed for the subsets taken under normal
operating conditions, and thus the centerlines are known. Here n=4.
X-Bar limits are computed (using ).
X-Bar limits are computed (using ).
Note: Since n=4 (a relatively small subset size), both and can be used to accurately calculate the UCL and LCL.
R-chart limits are computed.
S-chart limits are computed.
The individual points in subsets 8-11 are plotted below to demonstrate how they vary with in comparison with the control limits.
Figure E-1: Chart of individual points in subsets 8-11.
The subgroup averages are shown in the following X-Bar chart:
Figure E-2: X-Bar chart for subsets 8-11.

The R-chart is shown below:
Figure E-3: R-chart for subsets 8-11.

The S-chart is shown below:
Figure E-4: S-chart for subsets 8-11.
The experimental data is shown to be in control, as it obeys all of the rules given above.
13.2.7: EXAMPLE 2
It’s your first day on the job as a chemical engineer in a plant, and one of your responsibilities is to monitor the pH of a particular process.
You are asked by your boss to monitor the stability of the system. She gives you some baseline data for the process, and you collect data for
the process during your first day. Construct X-bar and R-Charts to report your results.
Table 1: Baseline data
To be consistent with the baseline data, each hour you take four pH readings. The data you collect is displayed below.
Table 2: Experimental data
Solution
For this situation, there are k=24 subsets because there are 24 data sets. For each subset, n=4 because there are four pH measurements taken
each hour. The first thing you do is calculate the mean and range of each subset. The means are calculated using the AVERAGE() Excel
function and the ranges are calculated using MAX() – MIN(). Once these values are calculated, the Grand Average XGA and average range
RA are calculated. These values are simply the means of each subset’s mean and range. This data is displayed below.
Table 3: Data used to calculate and grand average and Range.
Now that you know XGA = 7.01 and RA = 0.12, you can calculate the upper control limit, UCL, and lower control limit, LCL, for the X-bar
control chart.
From Table A, A2 = 0.729 when n=4. Using equations UCL and LCL for X-bar charts listed above:
Then the UCL = 7.0982, LCL = 6.9251 and XGA = 7.01 are plotted in Excel along with the average values of each subset from the
experimental data to produce the X-bar control chart.
Table 4: Average subset values and ranges plotted on the X-bar and R-chart
Figure E-5: X-bar control chart

Then, to construct the Range charts, the upper and lower control limits were found. For n=4, D3 = 0 and D4 = 2.282 so then:
Figure E-6).
Figure E-6: Range control chart

From both of these charts, the process is in control because all rules for stabilty are met.
Rule 1: No point falls beyond the UCl and LCL.
Rule 2: Two out of three consecutive points do not fall beyond 2σ on the same side of the centerline.
Rule 3: Four out of five consecutive points do not fall beyond 1σ on the same side of the centerline.
Rule 4: Nine or more consecutive points do not fall on the same side of the centerline.
It's important that both of these charts be used for a given set of data because it is possible that a point could be beyond the control band in
the Range chart while nothing is out of control on the X-bar chart.
Another issue worth noting is that if the control charts for this pH data did show some points beyond the LCL or UCL, this does not
necessarily mean that the process itself is out of control. It probably just means that the pH sensor needs to be recalibrated.
13.2.8: EXAMPLE 3
A simple out-of-control example with a sample constructed control chart.
You have been analyzing the odd operation of a temperature sensor in one of the plant's CSTR reactors. This particular CSTR's temperature
sensor consists of three small thermocouples spaced around the reactor: T1, T2, and T3. The CSTR is jacketed and cooled with industrial
water. The reaction taking place in the reactor is moderately exothermic. You know the thermocouples are working fine; you just tested
them, but a technician suggests the CSTR has been operating out of control for the last 10 days. There have been daily samples taken and
there is a control chart created from the CSTR's grand average and standard deviation from the year's operation.
You are assigned to see if the CSTR is operating out of control. The grand average is 307.47 units of temperature and the grand standard
deviation is 4.67 units of temperature. The data is provided for construction of the control chart in Table 1 and the data from the last 10
troublesome days is shown in Table 2. You decide to plot the troublesome data onto the control chart to see if it violates any stability rules.
Table 3-1. Data for Construction of Control Chart

The way I found A_3 or in this case, R_3, I used the control charts constants table which is found on this wiki page. I decided to use the x-
bar using the standard deviations) but you can also use use the range). I found that the value for n(number of subgroups) is three since the
CSTR's temperature sensor consists of three small thermocouples (T1,T2,T3). Therefore by looking at the constant chart, I get A_3( or
R_3 in this case) to be 1.954. Here's the table below:
Also, you will notice if you used the range instead of the standard deviation to determine the UCL,LCL, etc. that the values will be roughly
the same. Here's the table in comparing the values of UCL and LCL using either A_2 (range) or A_3(stdev):
Note: These values were using the same grand average (307.47), the grand standard deviation (4.67) and the grand range (8.80)
Table 3-2. Sample Data from Past 10 Troublesome Days
Solution
When the sample data was graphed onto the control chart, the image below was seen.
Figure 3-1. 10-Day Data Graphed Onto Control Chart

We can see from the control chart that the CSTR system is clearly out of control. Each thermocouple was tested to see which stability rules
it violates.
The first thermocouple (T1) violates every stability rule.
Rule 1 - Several points from the T1 data fall above the upper control line.
Rule 2 - There are many instances where at least two out of three consecutive points fall above the zone AB threshold.
Rule 3 - There are eight consecutive points falling above the BC threshold.
Rule 4 - Nine consecutive points fall above the mean value.
Judging on this thermocouple's performance, we can say that the system is out of control, but we will analyze the other thermocouples'
performance for good measure.
The second thermocouple (T2) violates stability rule 1, 2, and 3.
Rule 1 - One point falls below the lower control line.
Rule 2 - Two consecutive points (samples 9 and 10) fall beyond the AB threshold.
Rule 3 - Of the last five samples from T2, four are beyond the BC threshold.
The third thermocouple (T3) does not violate any stability rules and the results it displays are within control.
This system is out of control because the data from the thermocouples falls beyond the threshold rules for the unit's control chart. This
could be explained with many potential situations. One is explained below.
If the CSTR's agitator is knocked loose, the agitation could become erratic. The erratic agitation could create eddy currents and hot spots in
the CSTR.
The entire system is out of control because you know that the thermocouples are operating fine and more than one thermocouple violates the
stability rules.

When is it useful to use a moving average data set?
A. Never
B. When instant response to process is desired
C. When you have a process that changes slowly
D. Always, as it is a far superior method

What does n signify?
A. The number of subsets in collected data
B. The number of data points in a subset
C. The upper control limit for an X-bar chart
D. The centerline for an R-chart

Of the four stability rules for reading control charts, how many must occur in order for a subset to be determined unstable?
A. 1
B. 2
C. 3
D. 4
13.2.12: MULTIPLE CHOICE ANSWERS

Question 1: C
Question 2: B
Question 3: A

SPC Control Presentation
Slides without narration SPC control
13.2.14: REFERENCES
Wheeler, Donald J., and David S. Chambers. Understanding Statistical Process Control. 2nd ed. Knoxville: SPC P. 37-88. [1]
Box, George E., William G. Hunter, and J S. Hunter. Statistics for Experimenters: An Introduction to Design, Data Analysis, and Model
Building. New York: John Wiley & Sons. 43-45. [2]
Liptak, Bela G. "Process Control and Optimization." Instrument Engineers' Handbook 4: 405-413. [3]
Woolf, Peter, Amy Keating, Christopher Burge, and Michael Yaffe. Statistics and Probability Primer for Computational Biologists.
Massachusetts Institute of Technology. 2004. [4]
Anderson, David R., Sweeney, Dennis J. and Williams, Thomas A. Statistics for Business and Economics, 10e. Thompson South-
Western. 2008. [5]
13.2: SPC- Basic Control Charts- Theory and Construction, Sample Size, X-Bar, R charts, S charts is shared under a CC BY 3.0 license and was authored,
13.3: SIX SIGMA- WHAT IS IT AND WHAT DOES IT MEAN?
Every generation of business strives for a new level of quality. The quality program that is currently in vogue and being widely used and
recognized by industry is the Six Sigma program. Six Sigma is a relatively new program, and was only started in 1986. It was first put into
implementation at Motorola, but is now in use by most large corporations. Some of these other large companies include GE, Honeywell, and
Bank of America. The Six Sigma program is in place to eliminate any abnormalities, failures, or defects that occur within a given process.
These are problems that already exist. DFSS (Design for six sigma) starts earlier, to develop or redesign the process itself, so fewer wrinkles
show up in the first place, thus systematically preventing downstream errors. Six Sigma is also used in developing new processes. The Six
Sigma program strives to achieve six standard deviations between the mean and the closest specification limit on a short term study. Studies
run to obtain this goal are short term capability studies, which include common cause or random variation, such as operator control, and
long term studies, which include random and special types of variation. Both of these studies are evaluated on a Z-scale. The short term data
variability which makes up long term variability tends to cause the mean to shift. This shift of the mean is by 1.5 standard deviations. Most
companies are looking on a long term scale, because they would rather have a good/safe product in the long run/for a long time rather than
for a short amount of time. Using this idea, the goal for the Six Sigma program is to have fewer than 3.4 failures per one million
opportunities when the data is evaluated to include the shifted mean from process variability (6 standard deviations - 1.5 standard deviations
= 4.5 standard deviations). The 4.5 vs 6 standard deviations is the same goal, but the 4.5 represents data variation in the long run, which is
used in most processes. We will be using the shifted mean scenario for the rest of this article when referring to opportunity goals. This
leaves very slight room for error on a process and leads to a very high level of quality in the products.
The often-used six sigma symbol.

To reiterate, the term "Six Sigma" comes from the standard deviation and the Gaussian distribution. 3.4 failures per one million
opportunities represents 4.5 standard deviations (sigma) away from the median value, either up or down, under a bell curve. This will be
discussed further below along with what the Six Sigma program represents, what mathematically this means, and finally what a Gaussian
distribution is. After the basics have been covered we will move onto Statistical Process Control. Statistical Process Control is different
ways to analyze the data that you have obtained. Finally, we will relate Six Sigma to process control and explain how you can tell if
something is in Six Sigma control or not.
13.3.2: THE SIX SIGMA PROGRAM

Six Sigma is defined as a measure of quality that organizations strive to achieve. It is a disciplined, data-driven approach used to eliminate
defects in any process – from manufacturing to transactional and from product to service. Six Sigma does not refer to a process operating
within 6 or 4.5 standard deviations of the desired state. To achieve Six Sigma, a process must not produce more than 3.4 defects per million
opportunities, where a defect is defined as anything outside of the customer specification. One million opportunities, and only 3.4 defects?
Sounds like an efficient process. Here is an example that will help demonstrate. Let’s say a surgeon at the top of his field has a 5.51 sigma
ranking. Sounds pretty good, right? That means that if he operates on 1,000,000 patients per year he loses 30 of them. When a process is
operating at Six Sigma, it is seen that there are essentially zero defects within a process.
In chemical engineering, many different processes require strict, robust control. For example, a reactor may need to be maintained between
a specified temperature range to keep the plant safe. However, there is also an optimum temperature at which the reaction will occur.
Engineers will want to keep a tight control around this temperature to make the system more efficient, while also closely watching the
variable and keeping track of how close it gets to the control limits. Crossing the control limits is not the only problem. In order for a system
to be in statistical control as stated by the Six Sigma program, the data points must not frequently approach the same control limit. The Six
Sigma program helps engineers monitor processes to keep them within their desirable ranges. Many industries employ the Six Sigma system
as well as its hierarchy of positions to champion data-driven problem solving.
Originally invented by Motorola, the Six Sigma system contains belt rankings that represent a person’s achievements utilizing Six Sigma
methodologies and associated roles and responsibilities. These rankings are modeled after the martial arts system, as detailed below.
13.3.2.1: YELLOW BELTS (YB)
Yellow belt is the first level in the Six Sigma program. To get yellow belt status one must go through a training course in Six Sigma and
pass an online certification. Yellow belt team members do not lead projects on their own, but they are usually a core team member with
special knowledge on a project. Yellow belts are responsible for:
The development of process maps to support Six Sigma projects.
Running smaller process improvement projects using the Plan, Do, Check, Act (PDCA) method
Being a subject matter expert on the Six Sigma project
Yellow Belt Projects
usually limited to identifying small projects the could benefit from improvement
13.3.2.2: GREEN BELTS (GB)
Green belts are new team members that achieve certification after completing training in a Six Sigma course. To complete green belt
training, new members must complete a relevant project that utilizes the DMAIC model outlined below. To maintain green belt
certification, green belts generally complete DMAIC projects once every twelve months. Green belts are also responsible for:
Recommending Six Sigma projects
Participating on Six Sigma project teams
Leading Six Sigma teams in local improvement projects
Green Belt Projects
Green belts complete six sigma projects while still holding their current job duties. Therefore, a six sigma project undertaken by a green belt
could include:
Statistical analysis to improve reliability on a reactor within the green belt's current responsibility
Root cause analysis and elimination of quality control problems that chronically affect the green belt's department
13.3.2.3: BLACK BELTS (BB)
Black belts are intermediate team members that employ in-depth technical expertise to achieve Six Sigma objectives. Furthermore, black
belts can become temporary, full-time change agents for a moderate period of time (two-three years) to produce change using Six Sigma
methods. Black belts are also responsible for:
Demonstrating mastery of black belt knowledge
Demonstrating proficiency at achieving results through the application of the Six Sigma approach
Consultation for functional area Internal Process Improvement
Coaching, Mentoring, and Recommending green belts
Black Belt Projects
Black belts complete six sigma projects outside of their previous job duties. Therefore, a six sigma project undertaken by a black belt could
include:
Plant-level process controls improvement at a facility that greatly affects the finished product at a facility
Assessment of total warranty/quality control issues at a facility through large scale statistical data analysis and improvement
13.3.2.4: MASTER BLACK BELTS (MBB)
The major duty of master black belts include acting as project leaders, as well as:
Training and mentoring Black Belts and Green Belts
Helping to decide large-scale Six Sigma projects
Maintaining the quality of Six Sigma measurements
Developing Six Sigma training
Master Black Belt Projects
Most master black belts would only be involved with directing black and/or green belts in their projects. Their effect on process would
generally be limited to pointing out and suggesting large projects for appointed black belts to undertake.
To complete Six Sigma projects, belt candidates utilize the DMAIC model. DMAIC (Define, Measure, Analyze, Improve, Control) is an
improvement system for existing processes falling below specification and looking for incremental improvement. DMADV (Define,
Measure, Analyze, Design, Verify) is an improvement system used to develop new processes at Six Sigma quality level. Below, the steps for
DMAIC and DMADV are outlined.
Table 1 - Outline of the steps used in DMAIC (improve an existing process)
Table 2 - Outline of the steps used in DMADV (develop a new process)
All charts and descriptions referenced from Pyzdek, Thomas, “DMAIC and DMADV Explained”
http://www.pyzdek.com/DMAICDMADV.htm , 2003
13.3.3: STATISTICS AND SIX SIGMA

You are probably already familiar with the concepts of average, standard deviation, and Gaussian distribution. However, they are very
important concepts in Six Sigma, so they are reviewed in this section. This section also discusses analysis methods used in Six Sigma.
13.3.3.1: AVERAGE
The equation for calculating an average is shown below.
N
1 x1 + x2 + ⋯ + xN
x̄ = ∑ xi =
N N
i=1
where
is the average
x̄
xiis the measurement for trial i

N is the number of measurements
This equation relates to Six Sigma because it is the value that you aim for when you are creating your product. After millions of products
made, you will have a very good idea of what your exact average product specification is. The average is combined with the specification
limits, which are the limits that determine if your product is in or out of spec. The wider the specification limits are, the more room for
deviation from the average there is for your product. A product specification would be written like this:
10 ± 2 mm
Where the first number (10) represents the average and the second number (2) represents the amount of error allowable from the average
without violating 4.5 standard deviations (on a long term scale). Thus, your product can range from 8 to 12 mm for this example.
The average is a good representation of the a set of data. However, the main problem it experiences is the how it is strongly influenced by
outlier values. For example, say a town had a population of 50 people with an average income of $30,000 per person. Now say that a person
moves into the town with an income of $1,000,000. This would move the average up to approximately $50,000 per person. However, this is
not a good representation of the average income of the town. Hence, outlier values must be taken into account when analyzing data. In
contrast to the mean, sometimes the median can be a good representation of a set of data. The median is defined as the middle value of a set
of data are arranged in order. The median is immune to outlier values as it basically one value and is not calculated it any way so it cannot
be influenced by a wide range of numbers. Both the mean and median can be taken into account when analyzing a set of data points.
13.3.3.2: STANDARD DEVIATION

The equation for standard deviation is shown below.
−−−−−−−−−−−− −

N
 1
2
σ =  ∑ (xi − x̄)
⎷N
i=1
where σ = standard deviation, and the other variables are as defined for the average.
For each measurement, the difference between the measured value and the average is calculated. This difference is called the residual. The
sum of the squared residuals is calculated and divided by the number of samples minus 1. Finally, the square root is taken.
The standard deviation is the basis of Six Sigma. The number of standard deviations that can fit within the boundaries set by your process
represent Six Sigma. If you can fit 4.5 standard deviations within your process specifications then you have obtained a Six Sigma process
for a long term scale. However, the number of errors that you can have for your process as you move out each standard deviation continues
to decrease. The table below shows the percentage of data that falls within the standard deviations and the amount of defects per sigma, in
terms of "Defects Per Million Opportunities" or DPMO. The percentage of errors that you are allowed is one minus the percentage
encompassed by the percent of the total.
Data from: www.isixsigma.com/library/content/c020813a.asp
# of Standard Deviations % of Total DPMO
1 68.27 690,000
2 95.45 308,537
3 99.73 66,807
4 99.9937 6,210
5 99.99994 233
6 99.9999998 3.4
The image below shows an example data set (see #Gaussian Distribution below) with lines marking 1 to 6 standard deviations from the
mean. In this example, the mean is approximately 10 and the standard deviation is 1.16.
13.3.3.3: GAUSSIAN DISTRIBUTION

The normal, or Gaussian, distribution is a family of continuous probability distributions. The normal distribution function was first
introduced in 1733, and since has become the most widely used family of distribution functions in statistical analysis. These distribution
functions are defined by two parameters: a location (most commonly the "mean",μ), and a scale (most commonly the "variance", σ2). The
use of normal distributions in statistics can be seen as advantageous due to its ability to maximize information entropy around a given mean
and variance. Information entropy is the measure of uncertainty associated with a random variable.
For the purposes of this class and Chemical Engineering applications, you can model a given set of data using the probability density
function which is shown below.
2
1 (x − μ)
φ = −− exp(− )
σ√2π 2σ 2
where φ)istheprobabilitydensity, \(μ is the average and the other variables are as defined in the above two sections.
The above PDF equation gives graphs, depending on mean and variance, similar to those below.
Above are four examples of different distributions given different values for mean and standard deviation. An important case is the standard
normal distribution shown as the red line. The standard normal distribution is the normal distribution with a mean of 0 and a variance of 1.
It is also important to note some special properties of probability density functions:
symmetry about the mean, μ
the mode and mean are both equal to the mean
the inflection points always occur one standard deviation away from the mean, at μ − σ and μ + σ
Suppose we have a process where we make a product of a certain concentration and we have good control over the process. After analyzing
a set of data from a time period we see that we have a standard deviation of only 0.01 and our product concentration is required to be within
0.05. In order to say our product is essentially defect-free, 4.5 standard deviations away from the average must be less than our required
product tolerance (± 0.05). In this case 4.5 standard deviations is equal to 0.045 and our product tolerance is 0.05. This is more easily seen
graphically, as seen in the figure below.
As you can see 4.5 standard deviations on either side of the averages falls just slightly inside of the required tolerances for our product. This
means that 99.997% of our product will fall inside of our product tolerances.
13.3.3.4: ANALYSIS METHODS

Two ways to analyze data for similarities (percent confidence) would be running a regression analysis or an analysis of variance (ANOVA).
The difference between these two methods of analysis is that the ANOVA is for comparing independent groups such as 4 different reactor
compositions that are supposed to make the same product (in reality this doesn’t happen). A linear regression would be used for data that is
dependent on each other, such as looking at the difference between slurry product permeabilities when a one-reactant composition in the
process is changed. The percent confidence is the same as the student’s t-test, where you would want a t value (error) that is less than 0.05 to
have statistically similar data.
One statistical program that is widely used in Six Sigma projects is MINITAB. MINITAB is similar to Microsoft Excel but has much greater
statistical analysis capabilities. This program allows you to run linear regressions and ANOVAs by the click of an option. It also graphs not
only in 2-D but 3-D as well and runs statistical analysis methods that are more in depth than those offered in Excel. MINITAB graphical and
analytical tools can provide insight and information throughout the DMAIC process. MINITAB can be used to:
Identify - visualize sources of variation and relationships among variables
Verify - having statistical confidence behind conclusions
13.3.3.5: KEY TOOL BAR DESCRIPTIONS ON MINITAB

File - typical of most programs - save, open, print, etc.
Data - can subset data easily (stratification), transpose data, stach and unstack, convert text to numberic and back, code data
Calc - calculator function - one time calc (doesn't update as you update your data)
Stat - t-tests, regression, ANOVA, DOE, control charts, quality tools (Pareto and Gage R&R)
Graph - plots, histogram, matrix plot
Editor - find/replace (if you need to find items)
Tools - options and links to some external tools
Window - session window shows detains on what you have open
MINITAB is a powerful tool to help us understand the language of measurement and the different uses of measurement system. Some
unique features of MINITAB include: Measurement System Analysis (MSA) and the Gage Reproducibility and Repeatability Study (Gage
R&R). These two complicated Six Sigma analysis can be accomplished within steps using MINITAB.
13.3.4: STATISTICAL PROCESS CONTROL

Statistical process control (SPC) is an effective method of determining whether a system has deviated from its normal statistical behavior
by the use of control charts. Walter Shewhart first pioneered the use of statistical techniques when studying variations in product quality and
consistency at Bell Laboratories in the early 1920’s. SPC utilizes statistical tools in the form of control charts (such as Shewhart charts) to
observe the performance of the production process in order to predict significant deviations that may result. Dr. Shewhart created concepts
for statistical control and the basis for control charts through a series of carefully designed experiments. While he primarily drew from pure
mathematical statistical theories, he understood that data from physical processes seldom produced a normal distribution curve (a Gaussian
distribution, also known as a bell curve). Shewhart determined that every process displays variation: some display controlled variation that
is natural to the process, while others display uncontrolled variation that is not always present in the process causal system.
SPC, most often used for manufacturing processes, is used in conjunction with experimental designs, process capability analyses, and
process improvement plans to monitor product quality and maintain process targets. Six Sigma programs often depend on statistical process
controls to provide their data with supportive information. If a process falls outside its preferred operating range, the statistical control can
help you determine where and why the process failed.
Benefits of the SPC Method:
Provides surveillance and feedback for keeping processes in control
Signals when a problem with the process has occurred
Detects assignable causes of variation
Accomplishes process characterization
Reduces need for inspection
Monitors process quality
Provides mechanism to make process changes and track effects of those changes
Once a process is stable (assignable causes of variation have been eliminated), provides process capability analysis with comparison to
the product tolerance
Capabilities of the SPC Method:
All forms of SPC control charts
Variable and attribute charts
Average (X), Range (R), standard deviation (s), Shewhart, CuSum, combined Shewhart-CuSum, exponentially weighted moving average
(EWMA)
Selection of measures for SPC
Process and machine capability analysis (Cp and Cpk)
Process characterization
Variation reduction
Experimental design
Quality problem solving
Cause and effect diagrams
SPC is used to monitor the consistency of processes used to manufacture a product as designed. It aims to get and keep processes under
control. No matter how good or bad the design, SPC can ensure that the product is being manufactured as designed and intended. Thus, SPC
will not improve a poorly designed product's reliability, but can be used to maintain the consistency of how the product is made and,
therefore, of the manufactured product itself and its as-designed reliability.
13.3.4.1: METHODS AND CONTROL CHARTS

Control charts are the basis of statistical process controls methodologies, and are used to distinguish between random/uncontrollable
variations and controlled/correctable variations in a system. Control chart detects and monitors process variation over time. It also plays a
role as a tool for ongoing control of a process. There are many types of SPC control charts, though centering charts are used most often. The
remainder of this article will focus on the different types of centering and dispersion charts. Figure 1 displays single point centering charts
and Figure 2 displays subgroup sampling charts. Each figure has a brief description of the chart type within each of the two chart families
(centering and dispersion). In the centering chart, the centerline is given by the average of the data samples, where as in the dispersion chart
the centerline is the frequency distribution from that average.
Figure 1: Control chart types and descriptions for the single point sampling method(click to enlarge)[1].
Figure 2: Control chart types and descriptions for the subgroup sampling method(click to enlarge)[1].
A very useful type of centering chart is the Xbar chart, also known as the Shewhart chart mentioned earlier in the article. Figure 3 gives an
example of an Xbar chart (on top), as well as an example of a dispersion chart (on bottom). The charts are used in combination with a
system of rule based checks to determine whether the system is in statistical control. These rules can be seen in Figure 4 for both the single
point centering and subgroup sampling methods. Using these rules, one can inspect a given chart over a period of time and determine
whether the system is in control. If the system is not in control they must then check the controllable aspects of the system to see where the
problem is occurring.
Figure 3: The centering Xbar chart (Shewhart Chart) on top is used to determine whether the centering of the process is stable. The
dispersion (R) chart on bottom is used to determine whether the spread of the process is stable (click to enlarge)[1].
Figure 4: Table showing the criteria that indicate if a system is not in statistical control for different types of centering charts (click to
enlarge)[1].
Note: We only need to see if any one of the criteria objectives are achieved to claim that the system is statistically out of control.
As can be seen from the figures above, the primary tool of SPC is the control chart. Several different descriptive statistics are used in control
charts along with several different types of control charts that test for different causes of failure. Control charts are also used with product
measurements to analyze process capability and for continuous process improvement efforts.
 EXAMPLE 13.3.1
KLMM Industries has recent data on their widget making process. The data is shown below. The allowable range of widget width is
19.5-22.5 cm.
a. Please from this data calculate the mean and write the mean with its allowable range?
b. Next calculate the standard deviation for this data.
c. Does this process fall within 4.5 standard deviations?
d. 6 standard deviations?
Solution
a
N
1 x1 + x2 + ⋯ + xN
x̄ = ∑ xi =
N N
i=1
This is also calculated using the average function in excel, from this the average is calculated to be 20.9945
b
−−−−−−−−−−−− −
 N
 1
2
σ =  ∑ (xi − x̄)
⎷N
i=1
This is also calculated using the STDEV function in excel, from this the standard deviation is calculated to be 0.3206
3) Yes
4) No
 EXAMPLE 13.3.2
A chemical engineer at Cool Processes Inc. is in charge of monitoring the flowrate of the cooling water for a reaction. He's new at the
company and has asked you for your input. He says that the process is currently in statistical control and does not need to be changed.
Below is the centering control chart he has made of the process. Do you agree or disagree with his opinion? Explain.
Solution
Even though the process appears to be within a reasonable range of oscillation, and none of the data points are exceeding the process
limitations, the system is not in statistical control. Centering (X) charts show that a process is not in statistical control if "three or more
consecutive points fall on the same side of the centerline and all are located closer to the control limit than to the centerline."(Liptak
2006). The data points circled in the plot below violate that rule.
 EXAMPLE 13.3.3
This example will also serve as a small tutorial for using Minitab.
ABC Production is coming out with a new product. One of the chemical engineers has hypothesized that the product thickness is
dependent on the process temperature and the concentration of “compound Z.” There are currently controls regulating the temperature
between 77 and 80°C, and the concentration of Z between 0.0066 and .0073 g/L. Product specifications require that the thickness limits
are 1.94 to 2.10 cm. You have been assigned the task of determining if these are good parameters to control. Use the data from random
samples and use Minitab with linear regression to make your conclusion.
Solution
Conclusion: Temperature is much more influential on the thickness than the concentration, so it may be more worthwhile to refine the
temperature control than to focus on the concentration controllers. Another option would be to investigate other system parameters (i.e.,
time in reactor, pressure, etc.) to see if they hold more influence on the thickness.
Using Minitab
1) Open Minitab and input data
2) Select the “Regression” option from the “Stat” menu and the “Regression” tab
3)The regression window will open; add the “thickness” column for the response, and the Temperature and Concentration columns to
the predictors.
4)Click the “Options” button and select “PRESS and predicted R-squared,” then click “OK” on both dialog boxes.
5)The results will appear in the session window:
Interpreting the data

Some of the most important values in the returned data are the p values. These values test the null hypothesis that there is no effect by
the inputs (concentration and temperature here) on the output (thickness). The lower the value of p, the more correlation there is
between each variable and the result. In this case, the regression has a p value of 0 (under analysis of variance), so the model is
significant and at least one of the coefficients is not zero. The p values in the predictor section show a zero value for Temperature (high
correlation with thickness) but a 0.3 value for concentration (low correlation with thickness. This indicates that the concentration
controller is not significant.
Next we will in spect the R2 values. The R-Sq value of 97.7% indicates that the predictors (concentration and temperature) explain that
percent of the variance in thickness. The adjusted value (97.3%) accounts for the number of predictors (here, 2) in the model. Since
both are high, the model that is given at the beginning of the readout fits the data well.
The predicted R value is also important because it measures the amount of overfit of the model. If this value were not close to the R-Sq
value, then the model would only be applicable for this set of data points but would probably not be accurate for any other combination
of data points. Since the “R-Sq(pred)” value is close to the R-Sq value, this model is good.
13.3.5: REFERENCES
1. Instrument Engineer's Handbook:Process Control and Optimization Vol.IV (2005) edited by Liptak, Bela G. New York: Taylor and
Francis Group
2. Perlmutter, Barry (2005) A Six Sigma Approach to Evaluating Vacuum Filtration Technologies BHS-Filtration INC.
3. Chambers & Wheeler Understanding Statistical Process Control Tennessee: SPC Press
13.3: Six Sigma- What is it and what does it mean? is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via
13.4: BAYES RULE, CONDITIONAL PROBABILITY AND INDEPENDENCE
Probability is the likely percentage of times an event is expected to occur if the experiment is repeated for a large number of trials. The
probability of rare event is close to zero percent and that of common event is close to 100%. Contrary to popular belief, it is not intended to
accurately describe a single event, although people may often use it as such. For example, we all know that the probability of seeing the
head side of a coin, if you were to randomly flip it, is 50%. However, many people misinterpret this as 1 in 2 times, 2 in 4 times, 5 in 10
times, etc. of seeing the head side of the coin appear. So if you were to flip a coin 4 times and it came up heads every single time, is this
50% probability incorrect? No! It's just that your sample size is small. If you were to flip a coin 10,000 times, you would begin to see a
more even distribution of heads and tails. (And if you don't, you should probably get a different coin.) Now, as engineers, even if we know
the probability of the system, we don't have time to perform 10,000 trials to verify it. However, you'd be surprised at the small number of
trials that are required to get an accurate representation of the system. The following sections describe the relationship between events and
their probabilities. These relationships will then be used to describe another probability theory known as Bayes’ Theorem.
13.4.2: TYPES OF PROBABILITY

13.4.2.1: COMBINATION
Combinatorics is the study of all the possible orderings of a finite number of objects into distinct groups. If we use combinatorics to study
the possible combinations made from ordering the letters A, B, and C we can begin by counting out all the orderings
{(ABC), (ACB), (BCA), (BAC), (CBA), (CAB)}
giving us a total of 6 possible combinations of 3 distinct objects. As you can imagine the counting method is simple when the number of
objects is small yet, when the number of objects being analyzed increases the method of counting by hand becomes increasingly tedious.
The way to do this mathematically is using factorials. If you have n distinct objects then you can order them into n! groups. Breaking the
factorial down for our first example we can say, first there are 3 objects to choose from, then 2, then 1 no matter which object we choose
first. Multiplying the numbers together we get 3*2*1=3!. Now consider finding all the possible orderings using all the letters of the
alphabet. Knowing there are 26 letters in the English alphabet, the number of possible outcomes is simply 26!, a number so large that
counting would be difficult.
Now what if there are n objects and m that are equivalent and you wish to know the number of possible outcomes. By example imagine
finding the number of distinct combinations from rearranging the letters of PEPPER. There are 6 letters, 2 Es and 3 Ps but only 1 R.
Starting with 6! we need to divide by the repeat possible outcomes
6! 6 × 5 × 4 × 3! 6× 5× 4
= = = 6 × 5 × 2 = 60 possible arrangements
3! × 2! 3! × 2! 2
where on the bottom, the 3! is for the repeated Ps and the 2! is for the repeated Es.
 NOTE
You can cancel same integer factorials just like integers.
The next topic of importance is choosing objects from a finite set. For example, if four hockey teams are made from 60 different players,
how many teams are possible? This is found using the following relation:
60 × 59 × 58 × 57
= 487, 635 possible teams
4× 3× 2× 1
Generally, this type of problem can be solved using this relation:

n!
(n, r) = (13.4.1)
(n − r)!r!
called n choose r where n is the number of objects and r is the number of groups
Using Equation 13.4.1 , for the example above the math would be:
60!
=
56! × 4!
60 × 59 × 58 × 57 × 56!
=
56! × 4!
60 × 59 × 58 × 57
= 487, 635 possible teams
4!
13.4.2.2: JOINT PROBABILITY

Joint probability is the statistical measure where the likelihood of two events occurring together and at the same point in time are
calculated. Because joint probability is the probability of two events occurring at the same time, it can only be applied to situations where
more than one observation can be made at the same time. When looking at only two random variables, A and B , this is called bivariate
distribution, however this can be applied to numerous events or random variables being measured at one time (multivariate distribution).
The probability of two events, A and B , both occurring is expressed as:
P (A, B)
Joint probability can also be expressed as:

P (A ∩ B)
This is read as the probability of the intersection of A and B .

If A , B , and C are independent random variables, then
P (A, B, C) = P (A)P (B)P (C)
 EXAMPLE 13.4.1
Two cards are selected randomly from a standard deck of cards (no jokers). Between each draw the card chosen is replaced back in the
deck. What is the probability of choosing a four then a five? Let P (A) denote the probability that the first card is a four and P (B)
denote the probability that the second card is a five.
Solution
If there are 52 cards in a standard card deck with 4 suits, then P (A) = 4/52 and P (B) = 4/52 . Knowing that the events are
independent, each probability is multiplied together to find the overall probability for the set of events. Therefore:
P (A, B) = P (A) × P (B)
4 4
= ( )( )
52 52
= 1/169
The probability of choosing a four then a five from the deck with replacement is 1 out of 169.
13.4.2.3: CONDITIONAL PROBABILITY

Conditional probability is the probability of one event occurring, given that another event occurs. The following expression describes the
conditional probability of event A given that event B has occurred:
P (A ∣ B)
If the events A and B are dependent events, then the following expression can be used to describe the conditional probability of the events:
P (A, B)
P (A ∣ B) =
P (B)
P (A, B)
P (B ∣ A) =
P (A)
This can be rearranged to give their joint probability relationship:

P (A, B) = P (B ∣ A) × P (A)
= P (A ∣ B) × P (B)
This states that the probability of events A and B occurring is equal to the probability of B occurring given that A has occurred multiplied
by the probability that A has occurred.
A graphical representation of conditional probability is shown below:
Conditional probability is often derived from tree diagrams or contingency tables. Suppose you manufacture 100 piston shafts. Event A:
feature A is not defective Event B: feature B is not defective
P (A Not Def|B is Def) = 6/(80 + 6) = 0.0698
P (A Not Def|B Not Def) = 5/(9 + 5) = 0.3571
 EXAMPLE 13.4.2
Two fair, or unbiased, dice are tossed. Some example outcomes are (1,6) and (3,2). The probability of each possible outcome of the dice
is 1/36. When the first die is rolled it results in a value of 2. Once the second die is rolled, what it the probability that the sum of the
dice is 7?
Solution
Since it is known that the first value is 2, the possible combination of the two die are as follows:
(2,1) (2,2) (2,3) (2,4) (2,5) (2,6)
This results in six outcomes with equal probabilities since the second die is fair. Therefore, the conditional probability of the outcomes
above is 1/6. The conditional probability of the remaining 30 combinations is 0 since the first die is not a 2 in these cases. Finally, since
only one of these six outcomes can sum up to 7, (2,5), the probability is 1/6 for rolling a sum of 7 given the value of the first die is a 2.
 EXAMPLE 13.4.3
The probability that a rare species of hamster will give birth to one male and one female is 1/3. The probability that the hamster will
give birth to a male is 1/2. What is the probability that the hamster will give birth to a female knowing that the hamster has given birth
to a male? Let A denote the probability of giving birth to a male and B denote the probability of giving birth to a female.
Solution
P (A is the probability of giving birth to a male
P (B ∣ A is the probability of giving birth to a female given that birth of a male has already occurred
P (A, A is the probability of giving birth to one male and one female
These events are dependent so the following equation must be used:
P (A, B) = P (B ∣ A) ∗ P (A)
Rearranging this equation to find P (B ∣ A would give:

P (A, B)
P (B ∣ A) =
P (A)
Plugging in the known values would give:

P (B ∣ A) = (1/3)/(1/2)
P (B ∣ A) = 2/3
Therefore, the probability of giving birth to a female, given that birth of a male already occurred is 2/3.
13.4.3: LAW OF ITERATIVE EXPECTATION

An important application of conditional probability is called the "Law of Iterative Expectation".Given simply, it is: E[X]=E[E[X|Y]]. If the
random variable distribution of X is unknown, but we are given the distribution of the conditional variable of X, then by finding the
expection of the conditional variable twice, we can return to the expectation of the original random variable X. By looking at the
expectation of the random variable X, we can deduce the distribution of the random variable X.
The Law of Iterative Expectation is quite useful in mathematics and often used to prove important relationships. Note the example below:
 EXAMPLE 13.4.1
Use the Law of Iterative Expection to find Var[X] given only X|Y.
Solution
E[Var(X|Y)] = E[X^2]-E[(E[X|Y])^2]
Var(E[X|Y)])=E[(E[X|Y])^2]-(E[X])^2
E[Var(X|Y)]+ Var(E[X|Y]) = E[X^2]-E[(E[X|Y])^2] + E[(E[X|Y])^2]-(E[X])^2
Var[X]= E[(E[X|Y])^2] + E[(E[X|Y])^2] (by definition)
thus Var[X] = E[Var(X|Y)]+ Var(E[X|Y])
13.4.3.1: MARGINAL PROBABILITY

Marginal probability is the unconditional probability of one event; in other words, the probability of an event, regardless of whether
another event occurs or not. Finding the marginal probability of an event involves summing all possible configurations of the other event to
obtain a weighted average probability. The marginal probability of an event A is expressed as:
P (A) = ∑ P (A, B) = ∑ P (A ∣ B) ∗ P (B)
B B
The marginal probability (of A) is obtained by summing all the joint probabilities. Marginal probability can be used whether the events are
dependent or independent. If the events are independent then the marginal probability is simplified to simply the probability. The following
example will clarify this computation.
 EXAMPLE 13.4.4
The joint probabilities for rain and wind are given in the following table
Marginal probability of no rain = sum of Joint probabilities = .1 + .2 + .05+.01 = .36

Similarly, marginal probability of light rain = .05+.1+.15+.04 = .34
Similarly, marginal probability of heavy rain = .3.
13.4.3.2: MARGINALIZING OUT A FACTOR
In a system with two or more factors affecting the probability of the output of another factor, one of these initial factors can be marginalized
out to simplify calculations if that factor is unknown.
For instance, consider a system where A and B both affect the output of C. If the condition of B is unknown but its probability is known, it
can be marginalized out to put the system in terms of how only A affects C, using the equation below:
P (C ∣ A) = ∑ P (C ∣ A, Bi ) P (Bi )
Example Problem 2
The table below shows the probablitiy of having a large, small or no sand storm, if there is high, medium or no wind, and depending on if
there is rain. The next table show the probability of rain.
From this it is possible to calculate the probability of a large, small or no sand storm defendant just on the wind speed:
Similarly as to above:
P(Sandstorm Size|Wind Speed)= P(Sandstorm Size|Wind Speed, Rain)*P(Rain)+P(Sandstorm Size|Wind Speed, No Rain)*P(No Rain)
13.4.4: RELATIONSHIPS BETWEEN EVENTS

Knowing whether two events are independent or dependent can help to determine which type of probability (joint, conditional, or marginal)
can be calculated. In some cases, the probability of an event must be calculated without knowing whether or not related events have
occurred. In these cases, marginal probability must be used to evaluate the probability of an event.
13.4.4.1: INDEPENDENCE
If the two events are considered independent, each can occur individually and the outcome of one event does not affect the outcome of the
other event in any way.
Let's say that A and B are independent events. We'll examine what this means for each type of probability.
Independence in Conditional Probability
Independent events technically do not have a conditional probability, because in this case, A is not dependent on B and vice versa.
Therefore, the probability of A given that B has already occurred is equal to the probability of A (and the probability of B given A is equal
to the probability of B). This can be expressed as:
Independence in Joint Probability

Independent events can have a joint probability, even though one event does not rely on the other. For example, joint probability is used to
describe an event such as tossing a coin. The outcome when the coin is tossed the first time is not related to the outcome when the same coin
tossed a second time. To calculate the joint probability of a set of events we take the product of the individual probabilities of each event in
the set. This can be expressed as:
Independent events can also occur in a case where there are three events. For example, if two dice are being rolled with a sum of 6 (Event
A). Let event B represent that the first die is a 1 and let event C represent that the second rolled die is a 5. To prove if these events are
independent, the following relation is considered:
If any of these relations are false than the event is not independent. When considering events of more than three, the same relation would
follow but with an additional relation to event D.
13.4.4.2: DEPENDENCE
If the two events are considered dependent, then the outcome of the second event depends on the probability of the first event. The
probabilities of the individual events must be analyzed with conditional probability.
Let's now say that A and B are dependent events. We'll examine what this means for each type of probability.
Dependence in Conditional Probability
Conditional probability only applies to dependent events. In other words, A must depend on B in order to determine the probability of A
occurring given that B has already occurred. Therefore, for dependent events A and B, one can just apply the equations as seen in the
conditional probability section.
P (A, B)
P (A ∣ B) =
P (B)
and
P (A, B)
P (B ∣ A) =
P (A)
13.4.4.3: DEPENDENCE IN JOINT PROBABILITY

Joint probability can also be calculated for dependent events. For example, dependent joint probability can be used to describe cards being
drawn from a deck without replacing the cards after each consecutive drawing. In this case, the probability of A and B happening is more
complex, since the probability of the B happening depends on the probability of A happening. It can be expressed as:
P (A, B) = P (A)P (B ∣ A)
Note that this equation is found by rearranging the conditional probability equation.
13.4.5: BAYES’ THEOREM

Most probability problems are not presented with the probability of an event "A," it is most often helpful to condition on an event A"." At
other times, if we are given a desired outcome of an event, and we have several paths to reach that desired outcome, Baye’s Theorem will
demonstrate the different probabilities of the pathes reaching the desired outcome. Knowing each probability to reach the desired outcome
allows us to pick the best path to follow. Thus, Baye’s Theorem is most useful in a scenario of which when given a desired outcome, we can
condition on the outcome to give us the separate probabilities of each condition that lead to the desired outcome.
The following is Bayes' Theorem:
P (A ∣ Bj ) P (Bj )
P (Bj ∣ A) = (13.4.2)
∑j P (A ∣ Bj ) P (Bj )
where P (A ∣ B j) is the probability of A conditioned on B and ∑

j j
P (A ∣ Bj ) P (Bj ) is the law of total probability.
13.4.5.1: DERIVATION OF BAYES’ THEOREM

The derivation of Bayes’ theorem is done using the third law of probability theory and the law of total probability.
Suppose there exists a series of events: B , B ,..., B and they are mutually exclusive; that is, B
1 2 n 1 ∩ B2 ∩ … ∩ Bn = 0 .
This means that only one event, B , can occur. Taking an event "A" from the same sample space as the series of B , we have:
j i
A = ∪j ABj
Using the fact that the events AB are mutually exclusive and using the third law of probability theory:
i
P (A) = ∑ P (ABj )
Conditioning on the above probability, the result below is also called "the law of total probability"
P (A) = ∑ P (A|Bj )P (Bj )
Using the definition of conditional probabilities:

P (A ∣ Bj ) P (Bj )
P (Bj ∣ A) =
P (A)
Putting the above two equations together, we have the Bayes' Theorem (Equation 13.4.2 ):
P (A ∣ Bj ) P (Bj )
P (Bj ∣ A) =
∑j P (A ∣ Bj ) P (Bj )
13.4.5.2: REAL WORLD/CHEMICAL APPLICATIONS

Bayes' rule can be used to predict the probability of a cause given the observed effects. For example, in the equation assume B represents an
underlying model or hypothesis and A represents observable consequences or data. So,
P (model ∣ data) ∗ P (data)
P (data ∣ model) =
P (model)
Where
: probability of obtaining observed data given certain model
: probability that certain model gave rise to observed data
: probability of occurence of the model prior to taking the data into account
Another application: Bayes' rule estimation is used to identify species in single molecule Fluorenscence microscopy. More information
can be found on [1]
13.4.5.3: UNDERLYING PRINCIPLES AND SIGNIFICANCE OF BAYES’ RULE

As stated previously Bayes’ Rule allows for changing the probability of an event based on new information about existing knowledge or
expertise. For example, in the previous subsection the probability of the occurrence of the model, P(model), would be obtained through
many trials conducted previously and thus be considered existing knowledge. The probability of the data, P(data), would then be considered
new information. Bayes’ Rule essentially uses this new information to upgrade the existing knowledge and then determine the probability of
the new information based on the upgraded existing knowledge. Traditional, or Frequentist, statistics would differ from Bayesian statistics
by comparing P(data) to P(model) and determine, with 95% confidence, if P(data) is statistically significant to P(model).
Bayesian theory is being used by many companies and institutions to better classify errors and calculate uncertainty. It has proven to
perform better than averaging techniques and is used in safety systems as well as computing end states.
 EXAMPLE 13.4.1
A true-false question is posed to a team of two ChE students on a quiz show. Both the students will, independently, choose their answer.
Team member A knows the correct answer, while team member B has the wrong answer. Use Marginal Probability to find the best
strategy for the team?
Strategy 1: Choose one of them and let that person answer the question without knowing what the other student thought was the
correct answer.
Strategy 2: Have both students consider the question and then give the common answer they agree on, or if they disagree, flip a coin
to determine which answer to give.
13.4.5.4: SOLUTION
Strategy 1
We will break down the different possibilities for this situation using the following variables:
C is the event that they submit the correct answer
A is the event that student A is chosen to answer
B is the event that student B is chosen to answer
P (C) = P (C, A) + P (C, B) = P (C ∣ A) ∗ P (A) + P (C ∣ B) ∗ P (B)
because there is an equal chance for either student to be chosen

because student A is correct
because student B is incorrect
P (C) = 1 ∗ (1/2) + 0 ∗ (1/2) = 1/2
P (C) = 1/2
Strategy 2
Since we know the students will disagree on their answer, they must flip a coin in order to decide. Therefore the probability is simply
1/2.
Both strategies provide a 1/2 chance in getting the answer correct, therefore they are equally accurate.
 EXAMPLE 13.4.2
A biologist is studying human cells that are virally infected by a deadly disease, which is present among 0.01% of the population.
When preparing a sample, he mishandles a vial with infected cells and it ends up breaking, cutting through the protective gloves and
cutting his hand. Even though he immediately washed his hands, he is worried that he has contracted the disease. Luckily for him, his
laboratory has been developing a test for the disease and has performed trials with infected and non-infected patients alike. So, the test
is administered and to the biologist’s surprise, the test turns up negative. He becomes relieved, but then remembers that the test does not
always work, so he decides to use Bayes’ Rule to determine the probability that he has the disease. He goes into the laboratory database
to get all of the data he needs to determine this.
Solution
He wants to determine the probability that he tested negative but is actually positive for the disease, or P ( DISpos ∣ TESTneg ) .
From Bayes' Rule,
P (T ES T neg|DIS pos) ∗ P (DIS pos)
P (DIS pos|T ES T neg) =
P (T ES T neg)
To find the overall probability of a negative test, you need to use marginal probability and sum over all events, that is both having the
disease and not having the disease:
P (T ES T neg) = P (T ES T neg|DIS pos) ∗ P (DIS pos) + P (T ES T neg|DIS neg) ∗ P (DIS neg)
Assuming that all people without the virus do not test positive, and that 99% of infected patients test postive, what is the probability
that the biologist is actually positive with the disease?
P (T ES T neg) = .01 ∗ .0001 + 1 ∗ .9999 = 0.999901
(0.01 ∗ 0.0001)
P (DIS pos|T ES T neg) = = 1.0001E − 06
.999901
The chance that the biologist has the disease is roughly one in a million.
 EXAMPLE 13.4.3: INDEPENDENCE IN JOINT PROBABILITY
If there are 52 cards in a standard card deck with 4 suits, then the probably of picking a card of a particular numerical (5 of spades, 5 of
clubs, 5 of diamonds, 5 of hearts) value is P(A) = 1/13. The probability associated with any subsequent draw of a card is only
dependent upon the remaining cards in the deck. If the same card is taken from the deck four times in a row and not returned (resulting
in no fives in the deck), what is the probability of picking a four, returning it to the deck and then picking a 6.
Solution
The probability of each independent event is given below.
P (4) = 4/48
P (6) = 4/48
To find the probability of both events happening, one must multiply the probability of each independent event.
P = P (4) ∗ P (6) = 1/144.
 EXAMPLE 13.4.4
A Gambler has two coins in his pocket, a fair coin and a two-headed coin. He picks one at random from his pocket, flips it and gets
heads.
a. What is the probability that he flipped the fair coin?
b. If he flips the same coin a second time and again gets heads, what is the probability that he flipped the fair coin?
Solution
Call F the event he picked the fair coin and B the event he picked the biased coin. There are two coins in his pocket so the probability
of picking the either coin is 0.50. Let H_1 be the event "the first flip is heads" and H_2 be the event "the second flip is heads".
For (a)
P (F ) P (H1 ∣ F ) P (F )
P (F ∣ H1 ) = P (H1 ∣ F ) ∗ =
P (H1 ) P (H1 ∣ F ) P (F ) + P (H1 ∣ B) P (B)
Where and , so
For (b)
This question is solved in the same way as (a)
 EXAMPLE 13.4.5
Consider the situation where you're going on a bus tour of the local refinery and the person you like has asked you to find them because
they're going to save a seat for you. There are three buses that are taking the ChemE students on the tour and you need to immediately
decide which is the best one to get on. The first bus has room for 25 students, the second has space for 45 and the third has space for 55.
To decide which is the best bus, find the expected value (or expected number) of passengers for the bus your friend is on, given the
following expression:
E(X) = ∑ xi p (xi )
Solution
The expected value of X is the sum of the products of x_i multiplied by p(x_i)
E(passengers) = (25 ∗ (1/3) − 45 ∗ (1/3) + 55 ∗ (1/3i) = (25 + 45 + 55)/3 = 135/3 = 4.5
Therefore, the expected number of passengers for the bus your friend is on equals the sum of the passengers on each bus divided by the
total number of buses, assuming equal probability of your friend being on any of the 3 buses. This shows that you should get on the
second bus in order to find your friend.
Alternative method Now consider that you don't want to give each bus an equal chance that your friend is on it. The objective in this
alternative method is to find the probability that your friend is on each bus.
P ( Bus 1) = 25/135 = 0.185185
P ( Bus2 ) = 45/135 = 0.333333
P ( Bus3 ) = 55/135 = 0.407407
Now using the new probability we can find a more accurate expected number of passengers on the bus your friend is on.
′
E ( passengers ) = 25 ∗ (0.18518) + 45 ∗ (0.333333) + 55 ∗ (0.407407) = 42.037
This shows a similar result, you should get on the second bus, but the probabilities associated to each bus are adjusted to account for the
number of passengers on each one.
 EXERCISE 13.4.1
What is conditional probability?

a. A way of calculating probability under varying conditions
b. The probability of an event, given that another event has occurred
c. Probability that can not be determined
d. Probability that is always very small
 EXERCISE 13.4.1
In Bayes' rule, what is P(A)?

a. The probability of A
b. The probability of A, given B
c. The probability of B, given A
d. The marginal probability of A
Answer
A
 EXERCISE 13.4.3
Which type of probability does this statement infer to: What is the probability that a temperature sensor fails given that a flow sensor
has failed?
a. Joint Probability
b. Conditional Probability
c. Marginal Probability
Answer
B
13.4.6: REFERENCES
Ross, Sheldon, A First Course in Probability: Seventh Edition, 2006, Pearson Education, Inc.
Woolf, Keating, Burge, Yaffe,Statistics and probability primer for computational biologists, spring 2004, MIT.
[2]University of Oslo
[3], Wikipedia
13.4: Bayes Rule, Conditional Probability and Independence is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Brandon
Kountz, Ashwini Miryala, Kyle Scarlett, Zachary Zell, & Zachary Zell via source content that was edited to conform to the style and standards of the
13.5: BAYESIAN NETWORK THEORY
Bayesian network theory can be thought of as a fusion of incidence diagrams and Bayes’ theorem. A Bayesian network, or belief
network, shows conditional probability and causality relationships between variables. The probability of an event occurring given that
another event has already occurred is called a conditional probability. The probabilistic model is described qualitatively by a directed
acyclic graph, or DAG. The vertices of the graph, which represent variables, are called nodes. The nodes are represented as circles
containing the variable name. The connections between the nodes are called arcs, or edges. The edges are drawn as arrows between the
nodes, and represent dependence between the variables. Therefore, for any pairs of nodes indicate that one node is the parent of the other so
there are no independence assumptions. Independence assumptions are implied in Bayesian networks by the absence of a link. Here is a
sample DAG:
The node where the arc originates is called the parent, while the node where the arc ends is called the child. In this case, A is a parent of C,
and C is a child of A. Nodes that can be reached from other nodes are called descendents. Nodes that lead a path to a specific node are
called ancestors. For example, C and E are descendents of A, and A and C are ancestors of E. There are no loops in Bayesian networks,
since no child can be its own ancestor or descendent. Bayesian networks will generally also include a set of probability tables, stating the
probabilities for the true/false values of the variables. The main point of Bayesian Networks is to allow for probabilistic inference to be
performed. This means that the probability of each value of a node in the Bayesian network can be computed when the values of the other
variables are known. Also, because independence among the variables is easy to recognize since conditional relationships are clearly
defined by a graph edge, not all joint probabilities in the Bayesian system need to be calculated in order to make a decision.
13.5.2: JOINT PROBABILITY DISTRIBUTIONS

Joint probability is defined as the probability that a series of events will happen concurrently. The joint probability of several variables can
be calculated from the product of individual probabilities of the nodes:
n
P (X1 , … , Xn ) = ∏ P (Xi ∣ parents (Xi ))
i=1
Using the sample graph from the introduction, the joint probability distribution is:
P (A, B, C, D, E) = P (A)P (B)P (C ∣ A, B)P (D ∣ B)P (E ∣ C)
If a node does not have a parent, like node A, its probability distribution is described as unconditional. Otherwise, the local probability
distribution of the node is conditional on other nodes.
13.5.3: EQUIVALENCE CLASSES

Each Bayesian network belongs to a group of Bayesian networks known as an equivalence class. In a given equivalence class, all of the
Bayesian networks can be described by the same joint probability statement. As an example, the following set of Bayesian networks
comprises an equivalence class:
Network 1
Network 2
Network 3
The causality implied by each of these networks is different, but the same joint probability statement describes them all. The following
equations demonstrate how each network can be created from the same original joint probability statement:
Network 1
P (A, B, C) = P (A)P (B ∣ A)P (C ∣ B)
Network 2
P (A, B, C) = P (A)P (B ∣ A)P (C ∣ B)
P (A ∣ B)P (B)
= P (A) P (C ∣ B)
P (A)
= P (A ∣ B)P (B)P (C ∣ B)
Network 3
Starting now from the statement for Network 2
P (A, B, C) = P (A ∣ B)P (B)P (C ∣ B)
P (B ∣ C)P (C)
= P (A ∣ B)P (B)
P (B)
= P (A ∣ B)P (B ∣ C)P (C)
All substitutions are based on Bayes rule.

The existence of equivalence classes demonstrates that causality cannot be determined from random observations. A controlled study –
which holds some variables constant while varying others to determine each one’s effect – is necessary to determine the exact causal
relationship, or Bayesian network, of a set of variables.
13.5.4: BAYES' THEOREM

Bayes' Theorem, developed by the Rev. Thomas Bayes, an 18th century mathematician and theologian, it is expressed as:
P (H ∣ c) ⋅ P (E ∣ H, c)
P (H ∣ E, c) =
P (E ∣ c)
where we can update our belief in hypothesis H given the additional evidence E and the background information c . The left-hand term,
P (H|E, c) is known as the "posterior probability," or the probability of H after considering the effect of E given c. The term P (H|c) is
called the "prior probability" of H given c alone. The term P (E|H, c) is called the "likelihood" and gives the probability of the evidence
assuming the hypothesis H and the background information c is true. Finally, the last term P (E|c) is called the "expectedness", or how
expected the evidence is given only c . It is independent of H and can be regarded as a marginalizing or scaling factor.
It can be rewritten as
P (E ∣ c) = ∑ P (E ∣ Hi , c) ⋅ P (Hi ∣ c)
where i denotes a specific hypothesis Hi, and the summation is taken over a set of hypotheses which are mutually exclusive and exhaustive
(their prior probabilities sum to 1).
It is important to note that all of these probabilities are conditional. They specify the degree of belief in some proposition or propositions
based on the assumption that some other propositions are true. As such, the theory has no meaning without prior determination of the
probability of these previous propositions.
13.5.5: BAYES' FACTOR

In cases when you are unsure about the causal relationships between the variables and the outcome when building a model, you can use
Bayes Factor to verify which model describes your data better and hence determine the extent to which a parameter affects the outcome of
the probability. After using Bayes' Theorem to build two models with different variable dependence relationships and evaluating the
probability of the models based on the data, one can calculate Bayes' Factor using the general equation below:
p(data∣model1)p(model1)
p(model1 ∣ data) p(data) p(data ∣ model1)

BF = = =
p(model2 ∣ data) p(data∣model 2)p(model 2) p(data ∣ model 2)
p(data)
The basic intuition is that prior and posterior information are combined in a ratio that provides evidence in favor of one model verses the
other. The two models in the Bayes' Factor equation represent two different states of the variables which influence the data. For example, if
the data being studied are temperature measurements taken from multiple sensors, Model 1 could be the probability that all sensors are
functioning normally, and Model 2 the probability that all sensors have failed. Bayes' Factors are very flexible, allowing multiple
hypotheses to be compared simultaneously.
BF values near 1, indicate that the two models are nearly identical and BF values far from 1 indicate that the probability of one model
occuring is greater than the other. Specifically, if BF is > 1, model 1 describes your data better than model 2. IF BF is < 1, model 2 describes
the data better than model 1. In our example, if a Bayes' factor of 5 would indicate that given the temperature data, the probability of the
sensors functioning normally is five times greater than the probability that the senors failed. A table showing the scale of evidence using
Bayes Factor can be found below:
Although Bayes' Factors are rather intuitive and easy to understand, as a practical matter they are often quite difficult to calculate. There are
alternatives to Bayes Factor for model assessment such as the Bayesian Information Criterion (BIC).
The formula for the BIC is:
−2 ⋅ ln p(x ∣ k) ≈ BIC = −2 ⋅ ln L + k ln(n).
x = the observed data; n = the number of data points in x, the number of observations, or equivalently, the sample size;
k = the number of free parameters to be estimated. If the estimated model is a linear regression, k is the number of regressors, including
the constant; p(x|k) = the likelihood of the observed data given the number of parameters;
L = the maximized value of the likelihood function for the estimated model.
This statistic can also be used for non-nested models. For further information on Bayesian Information Criterion, please refer to:
13.5.6: ADVANTAGES AND LIMITATIONS OF BAYESIAN NETWORKS

The advantages of Bayesian Networks are as follows:
Bayesian Networks visually represent all the relationships between the variables in the system with connecting arcs.
It is easy to recognize the dependence and independence between various nodes.
Bayesian networks can handle situations where the data set is incomplete since the model accounts for dependencies between all
variables.
Bayesian networks can maps scenarios where it is not feasible/practical to measure all variables due to system constraints (costs, not
enough sensors, etc.)
Help to model noisy systems.
Can be used for any system model - from all known parameters to no known parameters.
The limitations of Bayesian Networks are as follows:
All branches must be calculated in order to calculate the probability of any one branch.
The quality of the results of the network depends on the quality of the prior beliefs or model. A variable is only a part of a Bayesian
network if you believe that the system depends on it.
Calculation of the network is NP-hard (nondeterministic polynomial-time hard), so it is very difficult and possibly costly.
Calculations and probabilities using Baye's rule and marginalization can become complex and are often characterized by subtle wording,
and care must be taken to calculate them properly.
13.5.7: INFERENCE
Inference is defined as the process of deriving logical conclusions based on premises known or assumed to be true. One strength of
Bayesian networks is the ability for inference, which in this case involves the probabilities of unobserved variables in the system. When
observed variables are known to be in one state, probabilities of other variables will have different values than the generic case. Let us take
a simple example system, a television. The probability of a television being on while people are home is much higher than the probability of
that television being on when no one is home. If the current state of the television is known, the probability of people being home can be
calculated based on this information. This is difficult to do by hand, but software programs that use Bayesian networks incorporate
inference. One such software program, Genie, is introduced in Learning and analyzing Bayesian networks with Genie.
13.5.8: MARGINALIZATION
Marginalization of a parameter in a system may be necessary in a few instances:
If the data for one parameter (P1) depends on another, and data for the independent parameter is not provided.
If a probability table is given in which P1 is dependent upon two other system parameters, but you are only interested in the effect of one
of the parameters on P1.
Imagine a system in which a certain reactant (R) is mixed in a CSTR with a catalyst (C) and results in a certain product yield (Y). Three
reactant concentrations are being tested (A, B, and C) with two different catalysts (1 and 2) to determine which combination will give the
best product yield. The conditional probability statement looks as such:
P (R, C, Y ) = P (R)P (C)P (Y ∣ R, C)
The probability table is set up such that the probability of certain product yield is dependent upon the reactant concentration and the catalyst
type. You want to predict the probability of a certain product yield given only data you have for catalyst type. The concentration of reactant
must be marginalized out of P(Y|R,C) to determine the probability of the product yield without knowing the reactant concentration. Thus,
you need to determine P(Y|C). The marginalization equation is shown below:
P (Y ∣ C) = ∑ P (Y ∣ Ri , C) P (Ri )
where the summation is taken over reactant concentrations A, B, and C.

The following table describes the probability that a sample is tested with reactant concentration A, B, or C:
This next table describes the probability of observing a yield - High (H), Medium (M), or Low (L) - given the reactant concentration and
catalyst type:
The final two tables show the calculation for the marginalized probabilities of yield given a catalyst type using the marginalization equation:
13.5.9: DYNAMIC BAYESIAN NETWORKS
The static Bayesian network only works with variable results from a single slice of time. As a result, a static Bayesian network does not
work for analyzing an evolving system that changes over time. Below is an example of a static Bayesian network for an oil wildcatter:
www.norsys.com/netlibrary/index.htm
An oil wildcatter must decide either to drill or not. However, he needs to determine if the hole is dry, wet or soaking. The wildcatter could
take seismic soundings, which help determine the geological structure at the site. The soundings will disclose whether the terrain below has
no structure, which is bad, or open structure that's okay, or closed structure, which is really good. As you can see this example does not
depend on time.
Dynamic Bayesian Network (DBN) is an extension of Bayesian Network. It is used to describe how variables influence each other over
time based on the model derived from past data. A DBN can be thought as a Markov chain model with many states or a discrete time
approximation of a differential equation with time steps.
An example of a DBN, which is shown below, is a frictionless ball bouncing between two barriers. At each time step the position and
velocity changes.
www.norsys.com/netlibrary/index.htm
An important distinction must be made between DBNs and Markov chains. A DBN shows how variables affect each other over time,
whereas a Markov chain shows how the state of the entire system evolves over time. Thus, a DBN will illustrate the probabilities of one
variable changing another, and how each of the individual variables will change over time. A Markov chain looks at the state of a system,
which incorporates the state of each individual variable making up the system, and shows the probabilities of the system changing states
over time. A Markov chain therefore incorporates all of the variables present in the system when looking at how said system evolves over
time. Markov chains can be derived from DBNs, but each network represents different values and probabilities.
There are several advantages to creating a DBN. Once the network has been established between the time steps, a model can be developed
based on this data. This model can then be used to predict future responses by the system. The ability to predict future responses can also be
used to explore different alternatives for the system and determine which alternative gives the desired results. DBN's also provide a suitable
environment for model predictive controllers and can be useful in creating the controller. Another advantage of DBN's is that they can be
used to create a general network that does not depend on time. Once the DBN has been established for the different time steps, the network
can be collapsed to remove the time component and show the general relationships between the variables.
A DBN is made up with interconnected time slices of static Bayesian networks. The nodes at certain time can affect the nodes at a future
time slice, but the nodes in the future can not affect the nodes in the previous time slice. The causal links across the time slices are referred
to as temporal links, the benefit of this is that it gives DBN an unambiguous direction of causality.
For the convenience of computation, the variables in DBN are assumed to have a finite number of states that the variable can have. Based
on this, conditional probability tables can be constructed to express the probabilities of each child node derived from conditions of its parent
nodes.
Node C from the sample DAG above would have a conditional probability table specifying the conditional distribution P(C|A,B). Since A
and B have no parents, so it only require probability distributions P(A) and P(B). Assuming all the variables are binary, means variable A
can only take on A1 and A2, variable B can only take on B1 and B2, and variable C can only take on C1 and C2. Below is an example of a
conditional probability table of node C.
The conditional probabilities between observation nodes are defined using a sensor node. This sensor node gives conditional probability
distribution of the sensor reading given the actual state of system. It embodies the accuracy of the system.
The nature of DBN usually results in a large and complex network. Thus to calculate a DBN, the outcome old time slice is summarized into
probabilities that is used for the later slice. This provides a moving time frame and forms a DBN. When creating a DBN, temporal
relationships between slices must be taken into account. Below is an implementation chart for DBN.
The graph below is a representation of a DBN. It represents the variables at two different time steps, t-1 and t. t-1, shown on the left, is the
initial distribution of the variables. The next time step, t, is dependent on time step t-1. It is important to note that some of these variables
could be hidden.
Where Ao, Bo, Co are initial states and Ai, Bi, Ci are future states where i=1,2,3,…,n.
The probability distribution for this DBN at time t is…
N
i i
P (Zt ∣ Zt−1 ) = ∏ P ( Zt ∣ π ( Zt ))
i=1
If the process continues for a larger number of time steps, the graph will take the shape below.
Its joint probability distribution will be...

T N
i i
P (Z1:T ) = ∏ ∏ P ( Z ∣ π ( Z ))
t t
t=1 i=1
DBN’s are useful in industry because they can model processes where information is incomplete, or there is uncertainty. Limitations of
DBN’s are that they do not always accurately predict outcomes and they can have long computational times.
The above illustrations are all examples of "unrolled" networks. An unrolled dynamic Bayesian network shows how each variable at one
time step affects the variables at the next time step. A helpful way to think of unrolled networks is as visual representations of numerical
solutions to differential equations. If you know the states of the variables at one point in time, and you know how the variables change with
time, then you can predict what the state of the variables will be at any point in time, similar to using Euler's method to solve a differential
equation. A dynamic Bayesian network can also be represented as a "rolled" network. A rolled network, unlike an unrolled network, shows
each variables' effect on each other variable in one chart. For example, if you had an unrolled network of the form:
then you could represent that same network in a rolled form as:
If you examine each network, you will see that each one provides the exact same information as to how the variables all affect each other.
13.5.10: APPLICATIONS
Bayesian networks are used when the probability that one event will occur depends on the probability that a previous event occurred. This is
very important in industry because in many processes, variables have conditional relationships, meaning they are not independent of each
other. Bayesian networks are used to model processes in a wide variety of applications. Some of these include…
1. Gene regulatory networks
2. Protein structure
3. Diagnosis of illness
4. Document classification
5. Image processing
6. Data fusion
7. Decision support systems
8. Gathering data for deep space exploration
9. Artificial Intelligence
10. Prediction of weather
11. On a more familiar basis, Bayesian networks are used by the friendly Microsoft office assistant to elicit better search results.\
12. Another use of Bayesian networks arises in the credit industry where an individual may be assigned a credit score based on age, salary,
credit history, etc. This is fed to a Bayesian network which allows credit card companies to decide whether the person's credit score
merits a favorable application.
13.5.11: SUMMARY: A GENERAL SOLUTION ALGORITHM FOR THE PERPLEXED

Given a Bayesian network problem and no idea where to start, just relax and try following the steps outlined below.
Step 1: What does my network look like? Which nodes are parents (and are they conditional or unconditional) and which are children? How
would I model this as an incidence diagram and what conditional probability statement defines it?
Step 2: Given my network connectivity, how do I tabulate the probabilities for each state of my node(s) of interest? For a single column of
probabilities (parent node), does the column sum to 1? For an array of probabilities (child node) with multiple possible states defined by the
given combination of parent node states, do the rows sum to 1?
Step 3: Given a set of observed data (usually states of a child node of interest), and probability tables (aka truth tables), what problem am I
solving?
Probability of observing the particular configuration of data, order unimportant
Solution: Apply multinomial distribution
Probability of observing the particular configuration of data in that particular order
Solution: Compute the probability of each individual observation, then take the product of these
Probability of observing the data in a child node defined by 2 (or n) parents given only 1 (or n-1) of the parent nodes
Solution: Apply marginalization to eliminate other parent node
Probability of a parent node being a particular state given data in the form of observed states of the child node
Solution: Apply Bayes' Theorem
Solve for Bayes' Factor to remove incalculable denominator terms generated by applying Bayes’ Theorem, and to compare the parent node
state of interest to a base case, yielding a more meaningful data point
Step 4: Have I solved the problem? Or is there another level of complexity? Is the problem a combination of the problem variations listed in
step 3?
If problem is solved, call it a day and go take a baklava break
If problem is not solved, return to step 3
Graphically:
 EXAMPLE 13.5.1
A multipurpose alarm in a plant can be tripped in 2 ways. The alarm goes off if the reactor temperature is too high or the pressure in a
storage tank is too high. The reactor temperature may be too high because of a low cooling water flow (1% probability), or because of
an unknown side reaction (5% probability). The storage tank pressure might be too high because of a blockage in the outlet piping (2%
probability). If the cooling water flow is low and there is a side reaction, then there is a 99% probability that a high temperature will
occur. If the cooling water flow is normal and there is no side reaction, there is only a 3% probability a high temperature will occur. If
there is a pipe blockage, a high pressure will always occur. If there is no pipe blockage, a high pressure will occur only 2% of the time.
Create a DAG for the situation above, and set up the probability tables needed to model this system. All the values required to fill in
these tables are not given, so fill in what is possible and then indicate what further values need to be found.
Solution
The following probability tables describe the system, where CFL = Cold water flow is low, SR = Side reaction present, PB = Pipe is
blocked, HT = High temperature, HP = High pressure, A = Alarm. T stands for true, or the event did occur. F stands for false, or the
event did not occur. A blank space in a table indicates an area where further information is needed.
An advantage of using DAGs becomes apparent. For example, you can see that there is only a 3% chance that there is a high
temperature situation given that both cold water flow is not low and that there is no side reaction.However, as soon as the cold water
becomes low, you have at least a 94% chance of a high temperature alarm, regardless of whether or not a side reaction occurs.
Conversely, the presence of a side reaction here only creates a 90% chance of alarm trigger. From the above probability calculations,
one can estimate relative dominance of cause-and-effect triggers. For example you could now reasonably conjecture that the cold water
being low is a more serious event than a side reaction.
 EXAMPLE 13.5.2
The DAG given below depicts a different model in which the alarm will ring when activated by high temperature and/or coolant water
pipe leakage in the reactor.
The table below shows the truth table and probabilities with regards to the different situations that might occur in this model.
The joint probability function is:

P (A, HT , CP ) = P (A|HT , CP )P (HT |CP )P (CP )
A great feature of using the Bayesian network is that the probability of any situation can be calculated. In this example, write the
statement that will describe the probability that the temperature is high in the reactor given that the alarm sounded.
Solution
P(A = T , CP = T ) ∑HT C[T ,P ] P (A − T , HT , CP − T )

P(CP = T ∣ Δ = T ) = [=
P(A = 1) ∑ P (A = T , HT , CP )
HT ,CP C[T ,P ]
 EXAMPLE 13.5.3
Certain medications and traumas can both cause blood clots. A blood clot can lead to a stroke, heart attack, or it could simply dissolve
on its own and have no health implications. Create a DAG that represents this situation.
Solution
b. The following probability information is given where M = medication, T = trauma, BC = blood clot, HA = heart attack, N = nothing,
and S = stroke. T stands for true, or this event did occur. F stands for false, or this event did not occur.
What is the probability that a person will develop a blood clot as a result of both medication and trauma, and then have no medical
implications?
Answer
P(N, BC, M, T) = P(N ∣ BC)P(BC ∣ M, T)P(M)P(T) = (0.25)(0.95)(0.2)(0.05) = 0.2375%
 EXAMPLE 13.5.4
Suppose you were given the following data.

Catalyst p(Catalyst)
A 0.40
B 0.60
Temperature Catalyst P(Yield = H) P(Yield = M) P(Yield = L)

H A 0.51 0.08 0.41
H B 0.30 0.20 0.50
M A 0.71 0.09 0.20
M B 0.92 0.05 0.03
L A 0.21 0.40 0.39
L B 0.12 0.57 0.31
How would you use this data to find p(yield|temp) for 9 observations with the following descriptions?
# Times Observed Temperature Yield
4x H H
2x M L
3x L H
An DAG of this system is below:
Solution
Marginalization! The state of the catalyst can be marginalized out using the following equation:
p(yield | temp) = ∑ p(yield | temp,cati)p(cati)
i = A,B
The two tables above can be merged to form a new table with marginalization:
Temperature P(Yield = H) P(Yield = M) P(Yield = L)
H 0.51*0.4 + 0.3*0.6 = 0.384 0.08*0.4 + 0.2*0.6 = 0.152 0.41*0.4 + 0.5*0.6 = 0.464
M 0.71*0.4 + 0.92*0.6 = 0.836 0.09*0.4 + 0.05*0.6 = 0.066 0.20*0.4 + 0.03*0.6 = 0.098
L 0.21*0.4 + 0.12*0.6 = 0.156 0.40*0.4 + 0.57*0.6 = 0.502 0.39*0.4 + 0.31*0.6 = 0.342
9! 4 2 3
p(yield ∣ temp) = ∗ ( 0.384 ∗ 0.098 ∗ 0.156 ) = 0.0009989
4!2!3!
 EXAMPLE 13.5.5
A very useful use of Bayesian networks is determining if a sensor is more likely to be working or broken based on current readings
using the Bayesian Factor discussed earlier. Suppose there is a large vat in your process with large turbulent flow that makes it difficult
to accurately measure the level within the vat. To help you use two different level sensors positioned around the tank that read whether
the level is high, normal, or low. When you first set up the sensor system you obtained the following probabilities describing the noise
of a sensor operating normally.
Tank Level (L) p(S=High) p(S=Normal) p(S=Low)
Above Operating Level Range 0.80 0.15 0.05
Within Operating Level Range 0.15 0.75 0.10
Below Operating Level Range 0.10 0.20 0.70
When the sensor fails there is an equal chance of the sensor reporting high, normal, or low regardless of the actual state of the tank. The
conditional probability table for a fail sensor then looks like:
Tank Level (L) p(S=High) p(S=Normal) p(S=Low)
Above Operating Level Range 0.33 0.33 0.33
Within Operating Level Range 0.33 0.33 0.33
Below Operating Level Range 0.33 0.33 0.33
From previous data you have determined that when the process is acting normally, as you believe it is now, the tank will be operating
above the level range 10% of the time, within the level range 85% of the time, and below the level range 5% of the time. Looking at the
last 10 observations (shown below) you suspect that sensor 1 may be broken. Use Bayesian factors to determine the probability of
sensor 1 being broken compared to both sensors working.
Sensor 1 Sensor 2
High Normal
Normal Normal
Normal Normal
High High
Low Normal
Low Normal
Low Low
High Normal
High High
Normal Normal
From the definition of the Bayesian Factor we get.
For this set we will use the probability that we get the data given based on the model.
If we consider model 1 both sensors working and model 2 sensor 2 being broken we can find the BF for this rather easily.
p(data | model 1) = p(s1 data | model 1)*p(s2 data | model 1)
For both sensors working properly:
The probability of the sensor giving each reading has to be calculated first, which can be found by summing the probability the tank
will be at each level and multiplying by probability of getting a specific reading at that level for each level.
p(s1 = high | model 1) = [(.10)*(.80) + (.85)*(.15) + (.05)*(.10) = 0.2125
p(s1 = normal | model 1) = [(.10)*(.15) + (.85)*(.75) + (.05)*(.20) = 0.6625
p(s1 = low | model 1) = [(.10)*(.05) + (.85)*(.10) + (.05)*(.70) = 0.125
Probability of getting sensor 1's readings (assuming working normally)
p(s1data | model1) = (.2125)4 * (.6625)3 * (.125)3 = 5.450 * 10 − 6
The probability of getting each reading for sensor 2 will be the same since it is also working normally
p(s2data | model1) = (.2125)2 * (.6625)7 * (.125)1 = 3.162 * 10 − 4
p(data | model1) = (5.450 * 10 − 6) * (3.162 * 10 − 4) = 1.723 * 10 − 9
For sensor 1 being broken:
The probability of getting each reading now for sensor one will be 0.33.
p(s1data | model2) = (0.33)4 * (0.33)3 * (0.33)3 = 1.532 * 10 − 5
The probability of getting the readings for sensor 2 will be the same as model 1, since both models assume sensor 2 is acing normally.
p(data | model2) = (1.532 * 10 − 5) * (3.162 * 10 − 4) = 4.844 * 10 − 9
A BF factor between 1/3 and 1 means there is weak evidence that model 2 is correct.
13.5.12: TRUE OR FALSE?

1. Is the other name for Bayesian Network the "Believer" Network?
2. The nodes in a Bayesian network can represent any kind of variable (latent variable, measured parameter, hypothesis..) and are not
restricted to random variables.
3. Bayesian network theory is used in part of the modeling process for artificial intelligence.
Answers:
1. F
2. T
3. T
13.5.13: REFERENCES
1. Aksoy, Selim. "Parametric Models Part IV: Bayesian Belief Networks." Spring 2007.
<www.cs.bilkent.edu.tr/~saksoy/courses/cs551/slides/cs551_parametric4.pdf>
2. Ben-Gal, Irad. “BAYESIAN NETWORKS.” Department of Industrial Engineering. Tel-Aviv University.
<http://www.eng.tau.ac.il/~bengal/BN.pdf>http://www.dcs.qmw.ac.uk/~norman/BBNs/BBNs.htm
3. Charniak, Eugene (1991). "Bayesian Networks without Tears", AI Magazine, p. 8.
4. Friedman, Nir, Linial, Michal, Nachman, Iftach, and Pe’er, Dana. “Using Bayesian Networks to Analyze Expression Data.” JOURNAL
OF COMPUTATIONAL BIOLOGY, Vol. 7, # 3/4, 2000, Mary Ann Liebert, Inc. pp. 601–620
<www.sysbio.harvard.edu/csb/ramanathan_lab/iftach/papers/FLNP1Full.pdf>
5. Guo, Haipeng. "Dynamic Bayesian Networks." August 2002.<www.kddresearch.org/Groups/Probabilistic-Reasoning/258,1,Slide 1>
6. Neil, Martin, Fenton, Norman, and Tailor, Manesh. “Using Bayesian Networks to Model Expected and Unexpected Operational Losses.”
Risk Analysis, Vol. 25, No. 4, 2005 <http://www.dcs.qmul.ac.uk/~norman/papers/oprisk.pdf>
7. Niedermayer, Daryle. “An Introduction to Bayesian Networks and their Contemporary Applications.” December 1, 1998. <
http://www.niedermayer.ca/papers/bayesian/bayes.html>
8. Seeley, Rich. "Bayesian networks made easy". Application Development Trends. December 4, 2007 <www.adtmag.com/article.aspx?
id=10271&page=>.
9. http://en.Wikipedia.org/wiki/Bayesian_network

Authors: Sarah Hebert, Valerie Lee, Matthew Morabito, Jamie Polan
13.5: Bayesian Network Theory is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source content
13.6: LEARNING AND ANALYZING BAYESIAN NETWORKS WITH GENIE
Genie (Graphical network interface) is a software tool developed at the University of Pittsburgh for Microsoft Windows and available free
of charge at Genie. It is useful for decision analysis and for graphically representing the union of probability and networked occurrences.
Particularly, Genie can be used for the analysis of Bayesian networks, or directed acylic graphs (i.e. occurrences in a web of happenings are
conditionally independent of each other). Bayesian networks or Dynamic Bayesian Networks (DBNs) are relevant to engineering controls
because modelling a process using a DBN allows for the inclusion of noisy data and uncertainty measures; they can be effectively used to
predict the probabilities of related outcomes in a system. In Bayesian networks, the addition of more nodes and inferences greatly increases
the complexity of the calculations involved and Genie allows for the analysis of these complicated systems. Additionally, the graphical
interface facilitates visual understanding of the network (Charniak, 1991).
This link gives an example of a complex Bayesian network depicted in the graphical interface of Genie. As can be seen, Genie arranges the
network of nodes and inferences in a topology that is easily visualized and is useful for both simple and extremely complex systems.
13.6.2: USING GENIE TO CONSTRUCT AND ANALYZE DYNAMIC BAYESIAN NETWORKS

This section will include instructions for downloading and installing the Genie software as well as describe the procedure for using Genie to
analyze simple Bayesian networks.
13.6.2.1: DOWNLOADING AND INSTALLING GENIE

The Genie software download is available the Genie website, depicted below:
To download Genie, click on the “Downloads” link on the left menu of the page and highlighted in red in the figure above.
A registration page will appear which requires some basic user information. After this has been entered, click on the Register link.
A Downloads page with several download options will appear. Click on the GeNIe 2.0 link highlighted in red on the figure below. This will
initiate the download.
NOTE: The Genie Software download is only available for Windows.
Install the software by following the steps indicated in the Genie installation program.
The installation of the Genie software is now complete. Please note the help section of the software features many tutorials describing how
to use a wide array of functions. However, this article will be focused on the analysis of Dynamic Bayesian Networks.
All informational materials used to create this tutorial come from GeNIe
13.6.2.2: USING GENIE TO ANALYZE DYNAMIC BAYESIAN NETWORKS

If the program was installed in the default location given in the installation wizard a shortcut to GeNIe can be found in the start menu listed
under All Programs. Click on the GeNIe 2.0 Icon as seen below:
The main interface will appear as shown in the figure below:
Note the important tool bars highlighted on the figure above: the menu bar (yellow), the standard toolbar (red), and the format toolbar
(blue).
Consider this simple illustration of how to build a Dynamic Bayesian Network using Genie:
Polly popular is the most popular girl on the UM campus who only dates other UM students. It is common knowledge that Polly only
accepts 30% of the invitations she receives to go on a date. Of the invitations that she accepts, 40% of them come from guys with high
GPAs, 40% of them come from guys with medium GPAs and 20% of them come from guys with low GPAs. Of the invitations that she
rejects, 10% of them come from guys with high GPAs, 30% of them come from guys with medium GPAs, and 60% of them come from guys
with low GPAs.
NOTE: In this example, the variable of GPA is discretized into categories of high, medium, and low. This is necessary for analysis using
Genie because otherwise there would exist infinite states for each node in the DBN (Charniak, 1991).
Creating a Bayesian network allows for determinations to be made about probabilities of Polly accepting or rejecting certain date
invitations. Shown below is how to use Genie to find the probability of Polly accepting a date invitation from a guy if she knows the guy
inviting her has a high GPA.
First, a node is created for the variable called acceptance of invitation. Select the “chance” node from the standard toolbar as is shown in the
figure below highlighted in red.
Left-click on a clear part of the graph area of the screen. An oval will appear with “Node1” inside as seen in the figure below.
The edit mode for the node should come up automatically; if not,simply double-click on the node to pull up the edit screen as depicted in the
figure below.
Enter an identifier for the node (must be a valid variable name in a programming language) and enter a name for the node (can be a string of
any length). For this example, the identifier is entered as “Acceptance” and the name is entered as “Acceptance of Invitation”.
To define the probabilities of the node, click on the definition tab highlighted in blue as shown below. The names of the states can be edited
by double clicking on the name fields (highlighted in red in the figure below) and the probabilities can be edited by clicking on the
probability field (highlighted in yellow in the figure below).
In this example, “State0” is changed to “Accept” and “State1” is changed to “Reject”. The default probabilities are entered as 0.3 for Accept
and 0.7 for Reject.
To create a second node for the variable GPA, simply click the change node again and place a second node under the “Acceptance” node.
Define the name and identifier of this node as done with the “Acceptance” node. Then define the probabilities for the “GPA” node by first
adding another outcome by clicking the “add outcome” button highlighted in blue in the figure below. Change state0, state1, and state2 to
HighGPA, MediumGPA, and LowGPA respectively. To finish defining this node, fill in the probabilities listed in the problem statement
(shown in the figure below) and press ok.
After the creation and definition of both nodes, connect these two nodes with an influence arc to represent that GPA affects how Polly
accepts or rejects the invitations she receives. To do this, click on the Arc tool (found on the standard toolbar), , and click on the
“Acceptance” node and drag the arrow to the new “GPA” node.
The Bayesian network describing this problem is now fully defined within the Genie program. To determine the probability of Polly
accepting a date invitation from a guy if she knows the guy has a high GPA, first right click on the GPA node, scroll down to "Set
Evidence", and select "HighGPA". This is shown in the figure below.
Figure below.
The results can also be accessed by double clicking the Acceptance node and selecting the value tab, shown in the figure below.
As can be seen in the figure above, Polly will accept the date invitation 63.16% of the time when she knows the guy inviting her has a high
GPA. Other probabilities related to this example can be determined similarly.
The logic and procedure involved in this simple problem can be applied to complex systems with many interconnected nodes. Please see the
Worked Out Examples sections for more examples of how to use Genie to analyze Bayesian networks.
All informational materials used to create this tutorial come from GeNIe
 EXAMPLE 13.6.1
For the reactor shown below, the probability that the effluent stream will contain the acceptable mole fraction of product is 0.5. For the
same reactor, if the effluent stream contains the acceptable mole fraction, the probability that the pressure of the reactor is high is 0.7. If
the effluent stream does not contain the acceptable fraction, the probability that the pressure is low is 0.85. If the pressure of the reactor
is high, the probability that the temperature of the feed is high is 0.9 and if the pressure of the reactor is low, the probability that
temperature of the feed is low is 0.75. Given that the temperature of the feed is low, what is the probability that the effluent stream
contains the acceptable mole fraction of product?
Solution
The variables to be included in the Bayesian network are the acceptable mole fraction of product in the effluent stream, the pressure of
the reactor, and the temperature of the reactor. The variables are connected in an acyclic network as shown in the image below. After
the nodes were created and the incidence arcs were defined, the probability was calculated by updating the network and moving the
pointer over the checkmark on the node. As can be seen in the figure, the probability of the effluent stream containing the acceptable
mole fraction of product given that the feed temperature is low is 67%.
A GeNIe file containing the full solution of the problem is located here. In this file, the values entered into Genie for the probabilities
and stages can be accessed by double-clicking on the nodes.
 EXAMPLE 13.6.2
The following example is complicated and contains many layers of information. It is intended to illustrate how many elements can be
put together in a Bayesian Network and analyzed using Genie such that the problem is not impossibly complicated. Assume that
detailed statistical experiments were conducted to find all the stated probabilities.
Consider a CSTR with a cooling jacket and an exothermic reaction with A --> B. The feed stream is pure A and the cooling stream in
the jacket has a sufficient flow rate high such that the ambient temperature is constant.
The pumps transporting the reactor feed stream are old, and sometimes do not pump at their full capacity. 98% of the time the pumps
work normally, but at other times the feed flow rate is slightly lower than normal. The preheating process for the feed stream is
inefficient, and sometimes does not heat the reactants to 80°C. There is a 95% chance the feed stream is slightly cooler than desired.
Finally, the pumps for the cooling water sometimes lose power, causing the ambient temperature for the reactor to climb higher than
usual. There is a 1% chance of this power loss occurring.
The concentration of A in the reactor depends on the feed flow rate. If this flow rate is normal, then there is a 98% chance CA is
normal, and a 1% chance each that it is slightly higher or lower than normal. If the flow rate is low, then there is a 40% chance that CA
will be higher than normal and a 60% chance that CA will be normal.
The reactor temperature depends on the feed temperature and the ambient temperature. If both T0 and Ta are normal, there is a 98%
chance T is normal, and a 1% chance each that it is slightly higher or lower than normal. If T0 is normal and Ta is high, there is a 90%
chance T is normal and a 10% chance it is high. If Ta is normal and T0 is low, there is a 80% chance T is normal and a 20% T is low.
Finally, if T0 is low and Ta is high, there is a 90% chance T is normal, a 8% chance T is low, and a 2% chance T is high.
The conversion depends on CA and T. If one of these variables is high and the other low, there is an equal chance that X will be low,
normal, or high. If both CA and T are normal, there is a 98% chance X is normal, and a 1% chance each that it is slightly higher or
lower than normal. If both CA and T are low, there is equal chance that X will be low or normal. If both CA and T are high, there is
equal chance that X will be normal or high. If CA is normal and T is low, there is a 75% chance X is normal and a 25% chance X is
low. If CA is normal and T is high, there is a 75% chance X is normal and a 25% chance X is high. Finally, if T is normal and CA is
low or high, there is a 85% chance X is normal and a 15% chance that X is low or high, respectively.
Create a model in Genie for this system that can be used to determine probabilities of related events occuring.
Solution
A GeNIe file containing the Bayesian Network of this problem is located here.
This model can be used to answer questions such as (1) if a composition sensor for B in the exit stream tells us that the conversion is
slightly low, what is the probability that the feed temperature is normal and (2) if a temperature sensor in the reactor tells us that the
reactor temperature is high what is the probability that the ambient temperature is normal?
13.6.3: MINITUBA
MiniTuba is a program that allows the creation of Bayesain Networks thought time, with only the final data sets. This was created by
Zuoshuang Xiang, Rebecca M. Minter, Xiaoming Bi, Peter Woolf, and Yongqun He, to analysis medical data, but can be used to create
Bayesian Networks for any propose. In oder to use this program go to www.minituba.org/ and go to the Sandbox demo Tab. From here one
can either start a new project or merly modify an old project. This wiki will talk though how to create a new project.
First Log in: to do this click the Log In link at the bottom of the page and enter [email protected] as the user name and demo as the password this
should bring up this screen:
and click My Projects and then Creat New Project.

The user needs to know how many variables (EX: Temperature, Pump speed, Yield, flowrate in ChemE), data sets (EX: Reactors, Tanks
etc.) and the average number of time steps that will be analyzed.
In this example we will have a CSTR whose yield may be affected by the reactor Volume, the motor speed, the temperature or the
concentration. This means that there are 5 variable. Lets say that we have 5 reactors running this experiment with differing either high or
low Moter speed, temperature, and yield and either high medium or low volume and concentration. In this example Yield is only affected by
the motor speed and Temperature with the following probability
Enter the requested data in miniTuba and then click open project from the list. Then click "load/updata data"
this example will use the following data:
Media:CSTRExample.xls
To insert the data, simply copy and paste in the data to open box, then click LOAD DATA. In Order to run the analysis click "Start DBN
botton"
Here you can slect which data series and variables you would like to analize. Now miniTuba gives several different options for the data. The
data can be analyises and a child a parent or both. A Child means that the data isnt effected by the other datathat , Parent means the data
doesnt affect the other data, and both means that the data can be affected and effect other variables. In our example, the motor speed is the
only thing that we know is not affected by teh other variables, so we will select it as a parent.
Discretization Policy tells MiniTuba how to bin the. In our case, as things are exactly high, med or low represented by 0,1,2 it is easy to
desice how to descretize the data, but it is ok if the data isnt allready bin (ie in 0s 1s and 2s). Quantile spits the data into even chucks (ie if
you select quantile 2, it will find the mean and then everything above that is in one bin and everything below that mean is in a different bin).
Interval bins things that are in equal sized intervals around each other.
Select Natural fit for slipe fitting (this will allow you to have some data points missing ex: no volume reading in 1 data set) Lets select 2
intervals for moter speed, temperature and yeild adn then 3 intervals for volume and concentration.
MiniTuba also give you the option to force it to come up with some relationships weather MiniTuba thinks they should or not. MiniTuba
also allows you to choose the method used to solve this, we will select simulated anneling Number of Instances allows you to choose the
number of computers used to solve the problem, we will just leave it at 1 In the Demo version, the max calculation time is 1 min.
When everything is done, click Run Bayesian Analysis. By clicking Check Progress teh following screen will show up. Click Updata every
so often to check weather the solution has been computed.
This will output the following screen when it is done
As you can see, the yield depends on the motor speed and the temperature alone, just like we crated the module to show. Top Ten networks,
shows the 10 most likely networks. Click Score Distribution of Top Ten Networks shows the relative probability of each network, and the
probability that the number 1 network is the correct network out of the top 10. In our example the best network has a probablitity of 0.77 as
seen below.
13.6.4: REFERENCES
Charniak, Eugene (1991). "Bayesian Networks without Tears", AI Magazine, 50-62.
GeNIe Helpfile at GeNIe & SMILE
Murphy, Kevin. (2002). "Dynamic Bayesian Networks."
Xiang Z., Minter R., Bi X., Woolf P., and He Y.(2007). "miniTUBA: medical inference by network integration of temporal data using
Bayesian analysis," Bioinformatics, v. 23, 2423-2432, 2007.

Authors: Kevin Dahlberg, Paul Jantzen, Genevieve Lampinen, Albert Sawalha, and David Toronto
Stewards: Ross Bredeweg, Jessica Morga, Ryan Sekol, Ryan Wong
13.6: Learning and analyzing Bayesian networks with Genie is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf
13.7: OCCASIONALLY DISHONEST CASINO? MARKOV CHAINS AND HIDDEN
MARKOV MODELS
Basic probability can be used to predict simple events, such as the likelihood a tossed coin will land on heads rather than tails. Instinctively,
we could say that there is a 50% chance that the coin will land on heads.
But let’s say that we’ve just watched the same coin land on heads five times in a row, would we still say that there is a 50% chance of the
sixth coin toss will results in heads? No – intuitively, we sense that there is a very low probability (much less than 50%) that the coin will
land on heads six times in a row.
What causes us to rethink the probability of the coin landing on heads? The same coin is being tossed, but now we are taking into account
the results of previous tosses. Basic probability can be used to predict an isolated event. But more advanced probability concepts must be
used to assess the likelihood of a certain sequence of events occurring. This section provides an introduction to concepts such as Bayes rule,
Markov chains, and hidden Markov models, which are more useful for predicting real world scenarios (which are rarely singular, isolated
events).
13.7.2: BAYES' RULE

Bayes’ rule is a foundational concept for finding the probability of related series of events. It relates the probability of event A conditional
to event B and the probability of event B conditional to event A, which are often not the same probabilities.
P (B|A)P (A)
P (A|B) =
P (B)
Another interpretation of Bayes’ rule: it’s a method for updating or revising beliefs (i.e. probability of landing on heads) in light of new
evidence (knowing that the coin landed on heads the last five tosses).
13.7.3: MARKOV CHAINS

A Markov chain is a particular way of modeling the probability of a series of events. A Markov chain is a sequence of random values whose
probabilities at a time interval depend only upon the value of the number at the previous time. A Markov chain, named after Andrey
Markov, is a stochastic process with the Markov property. This means that, given the present state, future states are independent of the past
states. In other words, the description of the present state fully captures all the information that could influence the future evolution of the
process, so that future states of the system will be reached through a probabilistic (determined from current data) process instead of a
deterministic one (from past data).
At each point in time, the system could either change its state from the current state to a different state, or remain in the same state. This
happens in accordance with the specified probability distribution. These changes of state are called transitions, and the probabilities
associated with various state changes are called transition probabilities.
For example, let’s say we want to know the probability of an elevator moving more than five floors up in a 30-floor building (including the
lobby). This is an example of a Markov chain because the probability does not depend on whether the elevator was on the 6th floor at 7:00
am or on the 27th floor at 10:30 pm. Instead the probability is dependent on what floor the elevator is on right before it starts to move. More
specifically, the fact that the elevator is on the top floor at t results in a zero probability that the elevator will be five floors higher at t .
1 2
The figure below helps to visualize the concept of a Markov Model.
x1, x2, x3 = states

a12, a21, a23, a32 = transition probabilities
A transition probability is simply the probability of going from one state to another state. Please note that there may be other paths or
transition probabilities for state x1, say to state x3 or back to itself. The sum of the transition probabilities must always equal to 1.
a1,i = a11 + a12 + a13 + … a1n
∑ = 1
13.7.4: TRANSITION PROBABILITY
The transition probabilities of each state for one system can be compiled into a transition matrix. The transition matrix for a system can
represent a one-step transition or a multiple-step transition. A n-step transition matrix is simply a one-step transition matrix raised to the n-th
power.
For state i to make a transition into state j, the transition probability Pij must be greater than 0. so:
Pi,j ≥ 0, i, j ≥ 0
∑ Pij = 1
j=0
with i = 0, 1, ...
Let P be the matrix of a one-step transition matrix,
P=
Pi,1 would denote the probability of going from state i to 1
 EXAMPLE 13.7.1: WILL IT RAIN TOMORROW?
Suppose that whether or not it rains today depends on previous weather conditions through the last two days. Specifically, suppose that
if it has rained for the past two days, then it will rain tomorrow with probability 0.7; if it rained today but not yesterday, then it will rain
tomorrow with probability 0.5; if it rained yesterday but not today, then it will rain tomorrow with probability 0.4; if it has not rained in
the past two days, then it will rain tomorrow with probability 0.2.
Solution
The transition matrix for the above system would be:
state 0: if it rained both today and yesterday
state 1: if it rained today but not yesterday
state 2: if it rained yesterday but not today
state 3: if it did not rain either yesterday or today
\mathbf{P}=\left| \begin{array}{cccc}
0.7 & 0 & 0.3 & 0 \\
0.5 & 0 & 0.5 & 0 \\
0 & 0.4 & 0 & 0.6 \\
0 & 0.2 & 0 & 0.8 \right|
\end{array} \nonumber
13.7.5: APPLICATIONS OF MARKOV CHAINS

Since this concept is quite theoretical, examples of applications of this theory are necessary to explain the power this theory has. Although
these following applications are not chemical control related, they have merit in explaining the diversity of operations in which Markov
chains can be used. Hopefully, by seeing these examples, the reader will be able to use Markov chains in new, creative approaches to solve
chemical controls problems.
Queuing Theory
Markov chains can be applied to model various elements in something called queuing theory. Queuing theory is simply the mathematical
study of waiting lines (or queues).
This theory allows for the analysis of elements arriving at the back of the line, waiting in the line, and being processed (or served) at the
front of the line. This method uses the variables including the average waiting time in the queue or the system, the expected number waiting
or receiving service and the probability of encountering the system in certain states, such as empty, full, having an available server or having
to wait a certain time to be served. These variables are all unique to the system at hand. A link that explains this theory in more detail can be
found below:
13.7.6: QUEUEING THEORY
Although Queuing Theory is generally used in customer service, logistics and operations applications, it can be tailored to chemical
processing. For instance, it is essential to model how long a particular product will be in an assembly line before it makes it to the desired
end process. Thus, this theory is robust for all kinds of uses.
 EXAMPLE 13.7.1: QUEUING PROBLEM
On a sunny Tuesday morning in the Computer Engineering Building, a group of Chemical Engineering students are finishing their
engineering laboratory reports and have stayed up all night writing ambivalent fluid mechanics questions, they need some prestigious
coffee for refreshment instead of Red Bull. Coffee shop called "Dreamcaster's Cafè" was opened by a University of Michigan Chemical
Engineering graduate. Business is going very well for "Dreamcaster's Cafè" and there are many people, m people, waiting in line in
front of the Chemical Engineering students. The service time for the first person in line (not necessarily a Chemical Engineering
student) is an exponential distribution with an average waiting time of . Since the Chemical Engineering students have to return to
the Computer Engineering Building to finish their report, they have a finite patience waiting in line and their patience is an exponential
distribution with an average of . The group of Chemical Engineering students does not care how long the service time is. Find the
probability that the ith Chemical Engineering student to wait long enough to get coffee.
Solution
Add text here.
Google’s Page Ranking System

The PageRank of a webpage as used by Google is defined by a Markov chain. PageRank is the probability to be at page i in the following
Markov chain on all (known) webpages. If N is the number of known webpages, and a page i has k links then it has transition probability
i
(1 − q) q
+
ki N
for all pages that are linked to and q/N for all pages that are not linked. The constant q is around 0.15.
Children’s Games
Markov chains are used in the children’s games “Chutes and Ladders,” “Candy Land” and “Hi Ho! Cherry-O!” These games utilize the
Markov concept that each player starts in a given position on the game board at the beginning of each turn. From there, the player has fixed
odds of moving to other positions.
To go into further detail, any version of Snakes and Ladders can be represented exactly as a Markov chain, since from any square the odds
of moving to any other square are fixed and independent of any previous game history. The Milton Bradley version of Chutes and Ladders
has 100 squares, with 19 chutes and ladders. A player will need an average of 45.6 spins to move from the starting point, which is off the
board, to square 100.
One last game of interest is Monopoly. This can be modeled using Markov chains. Check out the following link to see a simulation of the
probability of landing on a particular property in the Monopoly game:
Monopoly Simulation
13.7.7: HIDDEN MARKOV MODELS

Building further on the concept of a Markov chain is the hidden Markov model (HMM). A hidden Markov model is a statistical model
where the system being modeled is assumed to be a Markov process with unknown parameters or casual events. HMMs are primarily
helpful in determining the hidden parameters from the observable parameters.
Again, the figure below may help visualize the Hidden Markov Model concept.
x1, x2, x3 = hidden/unknown states

a12, a21, a23, a32 = transition probabilities
y1, y2, y3 = observable states
e1, e2, e3 = emission probabilities
The key point of the figure above is that only the blue circles are seen, but you suspect or know that these observable states are directly
related or dependent on some hidden state. These hidden states (the red circles) are what actually dictates the outcome of the observable
states. The challenge is to figure out the hidden states, the emission probabilities and transition probabilities.
For example, say you have a data acquisition program that continuously records the outlet temperature of your product stream which you
are heating up through a heat exchanger. The heat exchanger provides a constant supply of heat. You observe that the temperature has some
step disturbances and is not always constant. The temperature is the observable parameter, but the cause of its deviations/states is actually
the stream's flowrate. When the flowrate fluctuates, the outlet temperature also fluctuates since the heat exchanger is providing the same
amount of heat to the stream. So, the stream flowrate is the hidden parameter (since you do not have access to this measured variable) and
the probability that the flowrates switches from say 100 to 150 to 300 L/h are the transition probabilities. The emission probability is the
probability that the observed outlet temperature is directly caused by the flowrate. This emission probability is not necessarily 1 since
temperature variations could also be due to noise, etc.
Another common scenario used to teach the concept of a hidden Markov model is the “Occasionally Dishonest Casino”. If a casino uses a
fair die, each number has a 1/6 probability of being landed on. This type of probability that represents a known parameter is known as
emission probability in HMMs. Sometimes a casino can tip the odds by using a loaded die, where one number is favored over the other five
numbers and therefore the side that is favored has a probability that is higher than 1/6.
But how does one know that the casino is being dishonest by using a loaded die? Pretend that the die is loaded in such a way that it favors
the number 6. It would be difficult to differentiate between a fair die and loaded die after watching only one roll. You may be able to get a
better idea after watching a few tosses and see how many times the die landed on 6. Let’s say that you saw the following sequence of die
tosses:
465136
It is still difficult to say with certainty whether the die is loaded or not. The above sequence is a feasible for both a fair die and a loaded die.
In this particular case, the above numbers that represent what was rolled are the observable parameters. The hidden parameter is the type of
die used just because we do not know which type produced the above sequence of numbers.
Instead of relying on a sneaking suspicion that the casino is being dishonest, one can use a hidden Markov model to prove that a loaded die
is being used occasionally. Please note that if a loaded die was used all the time it would be more blatantly obvious. In order to get away
with this “slight” unfairness, the casino will switch the fair die with a loaded one every so often. Let’s say that if a fair die is being used
there is a 5% chance that it will be switched to a loaded die and a 95% chance that it will remain fair. These are known as transition
probabilities because they represent the likelihood of a causal event changing to another causal event.
Below is a picture representing the transition probabilities (probability of staying or changing a type of die) as well as the emission
probabilities (probability of landing on a number).
Using the basic diagram above that show the emission and transition properties, conditioned statements in Excel can be used to model the
probability that a loaded die is being used. Please refer to example two for the hidden Markov model Excel sheet.
HHMs can be modeled using Matlab. Follow the link to the site which contains information in installing the HHM toolbox in Matlab:
HHMs in Matlab
 EXAMPLE 13.7.1: "WHAT SHOULD I WEAR?"
The weather affects important issues such as, choice of clothing, shoes, washing your hair, and so many more. Considering the bizarre
weather in Ann Arbor, you wish to be able to calculate the probability that it will rain given the data of the last three consecutive days.
Applying your chemical engineering skills and empowered by the knowledge from this Wiki article, you sit down and decide to be the
next Weather-Person.
You know the following information since you have been an avid fan of the Weather Channel since age 5.
You then record the weather for the past 3 days.
Solution
Since the last day was a blizzard you use a matrix expressing a 100% chance in the past, and multiply this by the probabilities of each
weather type.
8 15 05
⎡ ⎤
[0 1 1 ] ⎢ .6 25 .15 ⎥ = [ .6 .25 .15 ]

⎣ ⎦
3 .6 .1
This means that there is a 60% chance of sun, a 25% chance of rain and a 15% chance of a blizzard. This is because a Markov process
works based on the probability of the immediately previous event. Thus, you decide that is a low enough risk and happily decide to
wear your best new outfit AND wash your hair.
 EXAMPLE 13.7.2: "WHAT SHOULD I WEAR FOR THE WEEKEND?"
Turns out it does rain the next day and you get your favorite clothes soaked. However, you are leaving for a weekend with your family
and must decide what to wear for the next two days. What is the prediction for the next two days?
Solution
Predicting for the first day is like the previous example substituting rain in for a blizzard.
.8 .15 .05
⎡ ⎤
[0 0 1 ] ⎢ .6 .25 .15 ⎥ = [ .3 .6 .1 ]
⎣ ⎦
.3 .6 .1
In order to predict for the second day we must use our previous data to predict how the weather will act. Since we predict two days
ahead we multiply the probabilities twice, effectively squaring them.
2
8 15 05
⎡ ⎤
[0 0 1 ]⎢ 6 25 .15 ⎥ = [ .63 .255 .115 ]
⎣ ⎦
3 6 .1
Alternatively, we can take the prediction from the previous day and multiply that by the probability of weather.
L. 3 .6 1⌋ .0 .15 .05
.6 .25 .15 = .63 .255 .115
Both ways give you the same answer. Therefore, you decide to pack for both sunny and rainy weather.
 EXAMPLE 13.7.3: OSEH EXAMPLE

Now, back to REAL chemical engineering applications, the probability theories described above can be used to check if a piece of
equipment is working. For example, you are an OSEH inspector and you have to analyze a set of data recording the pollutant levels of a
plant. If they use the scrubber they have, the probability of pollutant levels being high is low. However, if the scrubber is not used, the
probability of pollutant levels being high increases. You have gotten some inside information that the operators use the scrubbers only
70% off the time. Your task is to create a model based on the data recordings and using probability theories to predict WHEN they
switch their scrubbers ON and OFF, so that you can catch them red-handed in the act!
There are 6 possible pollutant levels: none, low, med-low, med, med-high, and high. If the level = high, then this is above the allowable
limit. Every other level is considered allowable. If the scrubber is ON, there is equal chance that any of the states will occur. If the
scrubber is OFF, there is a 50% chance that a high pollutant level will be recorded.
Given the following data:
Solution
Formulas used are listed below:
(1)
(2)
(3)
(4)
Please refer to the attached Excel File for the predictive model. OSEHExample
From our results in the Excel file, we graphed the predicted vs actual times when the scrubber was turned off. The value of "1" was
chosen arbitrarily to represent when the scrubber was OFF and "2" when it was "ON".

What is the key property of a Markov chain?
A) Probability of each symbol depends on FUTURE symbols.
B) Probability of each symbol depends on ALL preceding symbols.
C) Probability of each symbol depends on NONE of the preceding symbols.
D) Probability of each symbol depends only on the value of the PREVIOUS symbol.

What kind of parameters does emission probability represent?
A) unknown parameters
B) Markov parameters
C) known parameters
D) Bayes' parameters
13.7.10: REFERENCES
Woolf P., Keating A., Burge C., and Michael Y. "Statistics and Probability Primer for Computational Biologists". Massachusetts Institute
of Technology, BE 490/ Bio7.91, Spring 2004
Smith W. and Gonic L. "Cartoon Guide to Statistics". Harper Perennial, 1993.
Kemeny J., Snell J. "Finite Markov Chains". Springer, 1976. ISBN 0387901922
en.Wikipedia.org/wiki/Snakes_and_Ladders
Page, Lawrence; Brin, Sergey; Motwani, Rajeev and Winograd, Terry (1999). "The PageRank citation ranking: Bringing order to the
Web"
Ross, Sheldon M. "Introduction to Probability Models". Academic Press, 2007.
Authors: Nicole Blan, Jessica Nunn, Pamela Roxas, Cynthia Sequerah
Stewards: Kyle Goszyk, So Hyun Ahn, Sam Seo, Mike Peters
13.7: Occasionally Dishonest Casino? Markov Chains and Hidden Markov Models is shared under a CC BY 3.0 license and was authored, remixed, and/or
13.8: CONTINUOUS DISTRIBUTIONS- NORMAL AND EXPONENTIAL
Authors: Navin Raj Bora, Dallas Burkholder, Nina Mohan, Sarah Tschirhart
Random variables whose spaces are not composed of a countable number of points but are intervals or a union of intervals are said to be of
the continuous type. Continuous distributions are probability models used to describe variables that do not occur in discrete intervals, or
when a sample size is too large to treat each individual event in a discrete manner (please see Discrete Distributions for more details on
discrete distributions). The main difference between continuous and discrete distributions is that continuous distributions deal with a sample
size so large that its random variable values are treated on a continuum (from negative infinity to positive infinity), while discrete
distributions deal with smaller sample populations and thus cannot be treated as if they are on a continuum. This leads to a difference in the
methods used to analyze these two types of distributions: continuous and discrete distributions is continuous distributions are analyzed using
calculus, while discrete distributions are analyzed using arithmetic. There are many different types of continuous distributions including
some such as Beta, Cauchy, Log, Pareto, and Weibull. In this wiki, though, we will only cover the two most relevant types of continuous
distributions for chemical engineers: Normal (Gaussian) distributions and Exponential distributions.
In chemical engineering, analysis of continuous distributions is used in a number of applications. For example in error analysis, given a set
of data or distribution function, it is possible to estimate the probability that a measurement (temperature, pressure, flow rate) will fall
within a desired range, and hence determine how reliable an instrument or piece of equipment is. Also, one can calibrate an instrument (eg.
temperature sensor) from the manufacturer on a regular basis and use a distribution function to see of the variance in the instruments'
measurements increases or decreases over time.
13.8.2: NORMAL DISTRIBUTIONS

13.8.2.1: WHAT IS A GAUSSIAN (NORMAL) DISTRIBUTION CURVE?
A Gaussian distribution can be used to model the error in a system where the error is caused by relatively small and unrelated events.
This distribution is a curve which is symmetric about the mean (i.e. a Bell shaped curve) and has a range measured by standard deviations
above and below the mean of the data set (please see Basic Statistics for a further discussion on these statistical parameters). To better
explain, consider that a certain percentage of all data points will fall within one standard deviation of the mean. Likewise, more data points
will fall within two standard deviations of the mean, and so on. However, under this model it would require an infinite range to capture ALL
the data points thus presenting a minor difficulty in this appraoch.
The figure below is a possible Gaussian distribution, where the mean (μ) is 10 and standard deviation (σ) is 2. F(x) is the number of times a
certain value of x occurs in the population. The mean is simply the numerical average of all the samples in the population, and the standard
deviation is the measure of how far from the mean the samples tend to deviate. The following sections explain how and why a normal
distribution curve is used in control and what it signifies about sets of data.
Figure 1: Generic Normal Distribution curve
13.8.2.2: THE PROBABILITY DENSITY FUNCTION (PDF) FOR A NORMAL DISTRIBUTION

As was mentioned in the Introduction section, distribution curves can be used to determine the probability, P(x), of a certain event
occurring. When this is done, the distribution curve is known as a Probability Density Function (PDF). In the figure shown above, the x-
axis represents the range of possible events (eg. the range of ages in a sample population or the magnitude of noise generated by a
temperature sensor). The y-axis represents the number of times a certain x value occurs in a population. The PDF can be described
mathematically as follows:
In some cases, it might not be necessary to know the probability of just one event occurring. Rather, you may want to know the probability
of a range of events (eg. what is the probability that the noise generated by my temperature sensor will fall in the range of 5-10 Hz?). When
this happens, you must integrate the above PDF over the desired range, in the following manner:
where k1 and k2 are the limits of your desired range. This integral results in the following expression:
The Erf function can be found in most scientific calculators and can also be calculated using tables of Erf[] values. For example, its use here
parallels what we learned in ChE 342, Heat and Mass Transfer. Determine the value inside the brackets of the erf function through simple
arithmetic, then take this value and find the corresponding Erf number from a table. Finally use this value in the calculation to determine the
probability of a certain point, x, falling within a range bound from k1 to k2.
Sample Calculation
Given a data set with an average of 20 and a standard deviation of 2, what is the probability that a randomly selected data point will fall
between 20 and 23?
Solution 1
To solve simply substitue the values into the equation above. This yeilds the following equation:
These Erf values must be looked up in a table and substituted into the equation. Doing this yeilds
Thus there is a 43.3255% chance of randomly selecting a number from the data set with a value between 20 and 23.
Graphically speaking, the PDF is just the area under the normal distribution curve between k1 and k2. So, for the distribution shown above,
the PDF for 8<x<12 would correspond the area of the orange shaded region in the figure below:
Figure 2: PDF for 8<x<12

Solution 2
Alternatively, rather than using the error function, Mathematica's built-in probability density function can be used to solve this problem.
This probability density function can be applied to the normal distribution using the syntax shown below.
NIntegrate[PDF[NormalDistribution[μ,σ],x],{x,x1,x2}]
In this syntax, μ represents the mean of the distribution, σ represents the standard deviation of the distribution, and x1 and x2 represent the
limits of the range.
This function can be applied to this problem as shown below.
This shows that the probability of a randomly selected data point falling between 20 and 23 is 0.4332, or 43.32%. As expected, this value
calculated using the built-in probability density function in Mathematica matches the value calculated from using the error function.
Mathematica provides a faster solution to this problem.
An important point to note about the PDF is that when it is integrated from negative infinity to positive infinity, the integral will always be
equal to one, regardless of the mean and standard deviation of the data set. This implies that there is a 100% chance that your random
variable x will fall between negative infinity and positive infinity. Likewise, the integral between negative infinity and the mean is 0.5, or
there is a 50% chance of finding a value in this region due to the symmetric nature of the distribution.
13.8.2.3: THE CUMULATIVE DENSITY FUNCTION (CDF) FOR A NORMAL DISTRIBUTION

To transition our arguement to a more holistic perspective regarding the probability density function for a normal distribution, we present
the cumulative density function, which represents the integral (area under the curve) of the PDF from negative infinity to some given value
on the y- axis of the graph incrementally accoring to the x axis, which remains the same as before. Because of this type of definition, we
may circumvent the rigorous error function analysis presented above by simply subtracting one CDF points from another. For example, if
engineers desire to determine the probability of a certain value of x falling within the range defined by k1 to k2 and posses a chart
feauturing data of the relevant CDF, they may simply find CDF(k2)- CDF(k1) to find the relevant probability.
The Cumulative Density Function (CDF) is simply the probability of the random variable, x, falling at or below a given value. For example,
this type of function would be used if you wanted to know the probability of your temperature sensor noise being less than or equal to 5 Hz.
The CDF for a normal distribution is described using the following expression:
where k is the maximum allowable value for x.

The main difference between the PDF and CDF is that the PDF gives the probability of your variable x falling within a definite range,
where the CDF gives the probability of your variable x falling at or below a certain limit, k. The following figure is the CDF for a normal
distribution. You'll notice that as x approaches infinity, the CDF approaches 1. This implies that the probability of x falling between
negative and positive infinity is equal to 1.
Figure 3: CDF plot for a Normal Distribution
13.8.2.4: STANDARD NORMAL DISTRIBUTION

A special type of probability distribution curve is called the Standard Normal Distribution, which has a mean (μ) equal to 0 and a
standard deviation (σ) equal to 1.
This simplified model of distribution typically assists engineers, statisticians, business strategists, economists, and other interested
professionals to model process conditions, and to associate the attention and time needed to adress particular issues (i.e. higher probability
for a failed condition necessitate additional attention, etc.). Also, our grades in many of the courses here at the U of M, both in and outside
of the college of engineering, are based either strictly or loosely off of this type of distribution. For example, the B range typically falls
within +/- of one standard deviation.
The following figure is a Standard Normal Distribution curve:
Figure 4: Standard Normal Distribution Curve
The benefit of the standard normal distribution is it can be used in place of the Erf[] function, if you do not have access to a scientific
calculator or Erf[] tables. To use the Standard Normal Distribution curve, the following procedure must be followed:
1.Perform a z-transform. This is a transformation which essentially normalizes any normal distribution into a standard normal distribution.
It is done using the following relationship:
Mathematically speaking, the z transform normalizes the data by changing all the raw data points into data points that dictate how many
standard deviations they fall away from the mean. So, regardless of the magnitude of the raw data points, the standardization allows
multiple sets of data to be compared to each other.
2.Use a standard normal table to find the p-value. A standard normal table has values for z and corresponding values for F(x), where
F(x) is known as the p-value and is just the cumulative probability of getting a particular x value (like a CDF). A standard normal table may
look like the following table (it should be noted that if you are dealing with a Standard Normal Table that contains only positive z values,
the following property can be used to convert your negative z values into positive ones: F(-z)=1-F(z)):
Table 1: Standard Normal Table
A more detailed standard normal table can be found here (Note: This table is the same table used in the 'Basic Statistics' wiki).
3. What if I want the probability of z falling between a range of x=a and x=b?. First, find your two z values that correspond to a and b.
So these would be and , respectively. The probability of x falling in between a and b is just: F(zb) – F(za), where
F(zb) and F(za) are found from the standard normal tables.
Sample Calculation
Lets take the same scenario as used above, where you have a data set with an average of 20 and standard deviation of 3 and calculate the
probability of a randomly selected data point being between 20 and 23.
Solution
To do this simply subtract the two Z scores:
These Z scores correspond to probabilities of 0.933 and 0.5 respectively. Their difference, 0.433, is the probability of a randomly selected
value falling between the two. Notice that this is almost identical to the answer obtained using the Erf method. Note that obtaining this
answer also required much less effort. This is the main advantage of using Z scores.
13.8.3: PROPERTIES OF A NORMAL DISTRIBUTION

There are several properties for normal distributions that become useful in transformations.
1 If X is a normal with mean μ and σ2 often noted then the transform of a data set to the form of aX + b follows a
.
2 A normal distribution can be used to approximate a binomial distribution (n trials with probability p of success) with parameters μ = np
and σ2 = np(1 − p). This is derived using the limiting results of the central limit theorem.
3 A normal distribution can be used to approximate a Poisson distribution (parameter λ) with parameters μ = σ2 = λ. This is derived using
the central limit theorem.
4 The a sum of random variables following the same normal distribution ˜N(nμ,nσ2). This result shows how the sample mean .
5 The square of a standard normal variable is a chi-squared variable with one degree of freedom.
6 Combining 4 and 5 yields the sum of n squared standard normal variables is a chi-squared variable with n degrees of freedom. This has
application for chi-square testing as seen other sections of this text.
13.8.4: EXPONENTIAL DISTRIBUTION

The exponential distribution can be thought of as a continuous version of the geometric distribution without any memory. It is often used to
model the time for a process to occur at a constant average rate. Events that occur with a known probability for a given x value build the
theory developed previously (i.e. F(x) vs. x, F(x) in this case is different than before), where x may indicate distance between mutations on a
DNA strand., for one example application, rate of instrument failure, or time required for a radioactive particle to decay (the decay RATE)
for the below applications.
However, do remember that the assumption of a constant rate rarely holds as valid in actuality. The rate of incoming phone calls differs
according to the time of day. But if we focus on a time interval during which the rate is roughly constant, such as from 2 to 4 p.m. during
work days, the exponential distribution can be used as a good approximate model for the time until the next phone call arrives.
One can implement the exponential distribution function into Mathematica using the command: ExponentialDistribution[lambda].
Or, for a more grass-roots understanding of the function reference the following website, detailing the number of sharks seen in area one
square mile in different one hour time periods. Use the "Fish" button to run the applet. To change the parameter lambda, type in the value
and hit the "Clear" button.
www.math.csusb.edu/faculty/stanton/m262/poisson_distribution/Poisson_old.html
13.8.4.1: THE PROBABILITY DENSITY FUNCTION (PDF)
A few notes are worth mentioning when differentiating the PDF from the two-parameter Exponential Distribution function. As λ is
decreased in value, the distribution is stretched out to the right, and as λ is increased, the distribution is pushed toward the origin. This
distribution has no shape parameter as it has only one shape, i.e. the exponential, and the only parameter it has is the failure rate, λ. The
distribution starts at T = 0 at the level of f(T = 0) = λ and decreases thereafter exponentially and monotonically as T increases, and is
convex. The PDF for an exponential distribution is given in the form below, where λ is the rate parameter and x is random variable:
Figure 5: PDF (λ = 0.25) for Exponential Distribution
Here, x could represent time while the rate parameter could be the rate in which decay occurs. The rate parameter must be constant and
greater than 0. The PDF decreases continuously in this diagram because of its definition as a decay example. Exponential decay typically
models radioactive particles which lose mass per unit of time. Thus F(x) represents the mass of the particle with x equalling elapsed time
since the start of the decay. As time passes, the mass falls due to radioactive decay (particle emits radiation as a form of energy release, any
energy that results subtracts from mass, E = MC2).
13.8.4.2: THE CUMULATIVE DISTRIBUTION FUNCTION (CDF)

The CDF for an exponential distribution is expressed using the following:
Figure 6: CDF (λ = 1) for Exponential Distribution

Following the example given above, this graph describes the probability of the particle decaying in a certain amount of time (x).
13.8.5: PROPERTIES OF THE EXPONENTIAL DISTRIBUTION

Among the distribution functions, the exponential distribution funtion has two unique properties, they are the memoryless property and a
constant hazard rate.
--The Memoryless Property--
If a random variable, X, survived for "t" units of time, then the probability of X surviving an additional "s" units of time is the same as the
probability of X suriving "s" units of time. The random variable has "forgotten" that it has survived for "t" units of time, thus this property is
called the "memoryless" property.
A random variable, X, is memoryless if
P { X > s+t | X > t} = P { X > s } for all s,t (1)
From using (1) above and the definition of conditional probability:
\frac{P( X s, X > t)}{P (X > t)} = P(x > s)" src="/@api/deki/files/19503/image-794.png">(2)
and so:
P{X > s+t} = P{X > s}P{X > t} (3)
to prove the above property, we suppose that there is a function such that F(x)=P{X > x}.
by equation 3 above, we have
F(x+t) = F(s)F(t) (4)
By elementary calculus, the only continuous solution for this kind of functiona equation is
F(x) = e − λx (5)
It is thus proven the only distribution that can solve equation (4), or the memoryless property, is an exponential function.
--The Constant Hazard Rate Property--
Suppose the random variable, X, survived for "t" units of time, and r(t) represents the conditional probability that the t-year-old variable will
fail. Mathematically the hazard rate, or the failure rate, is defined as:
(6)
For the exponential distribution, the hazard rate is constant since
f(t) = e − λx,
and
r(t) = λ
the above proved that r(t) is constant for exponential distribution.
13.8.5.1: STANDARD EXPONENTIAL DISTRIBUTION
When an exponential distribution has λ = 1, this is called the standard exponential distribution. The equation and figure for this function is
described below:
Figure 7: Standard Exponential Distribution Curve

This type of distribution is a way of standardizing your graph. This parallels our previous example of standard normal distribution, however,
since time is now the x variable, it may not be negative (as an assumption of our hypothetical scenario). Thus, decay occurs only for
positive time (as the radioactive particle emits mass).

You are the lead process engineer for a leading pharmaceutical company, and your newest responsibility is to characterize one of your
bioreactors. Part of this characterization is to determine the temperature variance within the reactor during the last step in an isothermal
reaction. After taking data during this reaction for two weeks straight, you determine that the average temperature of the reaction is 30 C,
with a standard deviation of your set of data being 3 C. Also, the data fits an normal distribution (ie. it is not skewed either side of the
average) The higher-ups, ie. the ones who give you the raises, decide that this reactor needs to operate within 6 degrees of its average
operating temperature at least 60% of the time. Given the data that you have taken, does the reactor meet these standards, or will you be
pulling some 80 hour weeks in the near future trying to fix the process?
Solution
The information given in this problem is as follows:
Average: 30
Standard Deviation: 3
We want to find the probability of the temperature falling between 24 and 36 C
Using the PDF for a Normal Distribution, we can say that:
This expression can easily be solved using a table of Erf[] values as well as the property of the Erf function that states Erf(-x)=-Erf(x).
Another simple approach would be to use Mathematica with the syntax shown below.
Solving for P(24<T<36), we get the probability of the temperature falling within two standard deviations of the average to be about 95%
Since this value is much greater than the required 60%, you can be rest assured that your reactor is running well.

You are an engineer at a plant producing a volatile fuel. In your first week on the job both the primary control and redundancy control fail in
the same day, necessitating the significant inconvenience of shutting down the reactor in order to prevent a much more inconvenient
explosion. Your boss wants you to decide whether to simply replace both sensors or add an additional sensor to ensure that such problems
are avoided in the future. While adding another sensor would certainly help solve the problem there is an extra cost associated with
purchasing and installing the sensor. A very helpful fellow engineer with a penchant for statistics helps you out by informing you that these
dual instrument failures occur at the rate of one per 3.8 * 108 hours. What is the probability that this type of failure will happen again in the
next 50 years (the estimated life of the plant)? What is your recommendation to your boss?
Solution
Because you are given the probability of the event occuring and are interested in the amount of time separating two events using an
exponential distribution is appropriate. To solve this you use the CDF expression:
In this case λ is given in hours and x in years, so you must first convert one of the units. Also the rate give is in events per hour. λ must be
entered into the equation as the inverse (probability per time unit). Entering the appropriate numbers yeilds the following equation:
There is a 0.115% chance of both sensors failing in the same day sometime in the next 50 years. Based in this data you can conclude that
you are probably safe simply replacing the two sensors and not adding a third.

You work at a chemical plant that uses the famous reaction A + B -> C to produce C. As part of your process you operate a mixing tank to
dilute reactant A before sending it to the reactor. Reactant B is a byproduct from another process in the plant and comes in at a variable rate.
This causes fluctuations in the level of the mixing tank for reactant A as the flow in and out of the tank has to be adjusted. In the past this
has caused overflows in mixing tank A that your boss has tasked you with avoiding. You'ved decided the easiest way to avoid overflows is
to simply buy a larger mixing tank. Currently the mixing tank is 50 gallons. After some experimentation you find the normal operating
conditions of the tank are 45 gallons of reactant A with a 3 gallon standard deviation. However, to minimize costs the tank needs to be as
small as possible. What is smallest tank you can buy that avoids overflows 99.99% of the time?
Solution:
Known values:
μ = 45gal.
σ = 3gal.
Using a z-table you find that the smallest z-value that gives an F(x) of 0.9999 is 3.71. You then rearrange the z equation given below to
solve for x and plug in all known values.
x=z*σ+μ
x = 3.71 * 3 + 45 = 56.1
In order to prevent overflows 99.99% of the time the smallest tank that can be used is 56.1 gal.

Please briefly summarize the primary difference between continuous and discrete distributions, and their respective methods of analysis:
a. Continuous distributions and discrete distributions are always both analyzed through calculus methods only. Continuous distributions are
distinct from discrete distributions in that the sample size is too small to treat each individual event in a discrete manner.
b. Continuous distributions are analyzed by applying calculus methods whereas discrete distributions are commonly addressed via
arithmetic methods. Continuous distributions are distinct from discrete distributions in that the sample size is too large to treat each
individual event in a discrete manner.
c. By performing a standard z- transform, the interested student may begin the process of characterizing the probability of z falling between
the specific range from x = ‘a’ and x = ‘b’.
d. Error functions allow us to determine the likelihood of a data point falling within certain bounds without having to complete the rigorous
of the probability distribution function. Continuous distributions may be addressed in this manner since their sample size is too confusing to
be addressed in a discrete manner.

Please explain the primary philosophical difference between the parameters of the probability distribution function (PDF) and the
cumulative distribution function (CDF).
a. Who uses the word philosophical in a chemical engineering problem? The author of this question is most likely confused and trying too
hard, so I will protest by not answering.
b. The PDF and CDF are identical in application, and as such, must also have identical parameters.
c. The PDF and CDF both commonly measure the probability of sensor noise being at or below a specific value. They are different in the
fact that the CDF contains a specific range of use, ranging from k1 to k2, whereas the PDF does not, as it rises from negative infinity to a
value, k.
d. The major difference the PDF and CDF is that the PDF provides the probability of the variable x falling within a defined range.
Alternatively, CDF gives the probability of the variable x falling at or below a certain limit, k.

Therapeutic Insulin Production
www.youtube.com/v/b3aDqdOzP5I
A Brief Problem on Continuous Exponential Distribution Functions
A copy of the slides can be found here:Slides without narration
Normal Distribution Functions in Excel
If YouTube doesn't work, click here Normal Distribution Functions in Excel
13.8.12: REFERENCES
Gernie, L. and W. Smith "Cartoon Guide to Statistics" Harper Perennial, c. 1993
13.8: Continuous Distributions- normal and exponential is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et
13.9: DISCRETE DISTRIBUTIONS - HYPERGEOMETRIC, BINOMIAL, AND
POISSON
13.9.1: WHAT ARE DISCRETE DISTRIBUTIONS?
In order to explain the meaning and utility of dicrete distributions, it is necessary to first introduce the topic of Random Variables. The
term random in Random Variable refers to the idea that each individual outcome has an equal chance of occuring. Therefore, each outcome
is randomly determined. For example, a fair coin has an equal probability of flipping up heads as it does flipping up tails. Hence, the
random variable is determined by the outcome of flipping a coin. However, it is frequently the case in chemical engineering that when a
process is performed we are only interested in some function of the outcome rather than the outcome itself. For example, in a chemical
reactor we might want to know the probability that two molecules collide and react to form a product at a certain temperature, however it is
more useful to know the sum of the these collisions that form product rather than the individual occurences. This is analgous to rolling two
dice and wanting to know the sum of the dice rather than the individual numbers rolled. These discrete values defined on some sample space
compose what are called Random Variables.
13.9.1.1: RANDOM VARIABLE EXAMPLE

Suppose that our experiment consists of tossing 4 fair coins. If we define a variable X and let it denote the number of heads flipped, then X
is a random variable with possible discrete values of 0,1,2,3,4 with respective probabilities:
We can do the same for the probability of rolling sums of two dice. If we let Y denote the sum, then it is a random variable that takes on the
values of 2, 3, 4, 5, 6, 7, 8, 9, 10, & 12. Rather than writing out the probability statements we can represent Y graphically:
The graph plots the probability of the Y for each possible random value in the sample space (y-axis) versus the random value in the sample
space (x-axis). From the graph one can infer that the sum with the greatest probability is Y = 7.
These are just two ways one can describe a random variable. What this leads into is representing these random variables as functions of
probabilities. These are called the discrete distributions or probability mass functions. Furthermore, independent random events with known
probabilities can be lumped into a discrete Random Variable. The Random Variable is defined by certain criteria, such as flipping up heads a
certain number of times using a fair coin. The probability of a certain random variable equaling a discrete value can then be described by a
discrete distribution. Therefore, a discrete distribution is useful in determining the probability of an outcome value without having to
perform the actual trials. For example, if we wanted to know the probability of rolling a six 100 times out of 1000 rolls a distribution can be
used rather than actually rolling the dice 1000 times.
Note: Here is a good way to think of the difference between discrete and continuous values or distributions. If there are two continuous
values in a set, then there exists an infinite number of other continuous values between them. However, discrete values can only take on
specific values rather than infinite divisions between them. For example, a valve that can only be completely open or completely closed is
analogous to a discrete distribution, while a valve that can change the degree of openness is analogous to a continuous distribution.
The three discrete distributions we discuss in this article are the binomial distribution, hypergeometric distribution, and poisson distribution.
13.9.2: BINOMIAL DISTRIBUTION

This type of discrete distribution is used only when both of the following conditions are met:
1. The test has only two possible outcomes
2. The sample must be random
If both of the above conditions are met, then one is able to use this distribution function to predict the probability of a desired result.
Common applications of this distribution range from scientific and engineering applications to military and medical ones. For example, a
binomial distribution may be used in determining whether a new drug being tested has or has not contributed ("yes" or "no" labels) to
alleviating symptoms of a disease.
Since only two outcomes are possible, they can be denoted as MS and MF for the number of successes and number of failures respectively.
The term p is the frequency with which the desired number, MS or MF will occur. The probability of a collection of the two outcomes is
determined by the following equation.
MS MF
P (MS , MF ) = k p (1 − p) (13.9.1)
k in the above equation is simply a proportionality constant. For the binomial distribution it can be defined as the number of different
combinations possible
MS + MF (MS + MF )!
k = C ( ) = (13.9.2)
MS MS !MF !
! is the factorial operator. For example, 4! = 4 * 3 * 2 * 1 and x! = x * (x − 1) * (x − 2) * … * 2 * 1. In the above equation, the term (MS +
MF)! represents the number of ways one could arrange the total number of MS and MF terms and the denominator, NS!MF!, represents the
number of ways one could arrange results containing MS successes and MF failures. Therefore, the total probability of a collection of the
two outcomes can be described by combining the two above equations to produce the binomial distribution function.
(MS + MF )! MS MF
P (MS , MF ) = p (1 − p) (13.9.3)
MS !MF !
This can be simplified as the equation

n! k n
Γ{X − k} − p (1 − p) (13.9.4)
(n − k)!k!
for k = k = −0, 1, 2, 3 … .
In the above equation, n represents the total number of possibilities, or MS + MF, and k represents the number of desired outcomes, or MS.
These equations are valid for all non-negative integers of MS , MF , n , and k and also for p values between 0 and 1.
Below is a sample binomial distribution for 30 random samples with a frequency of occurrence being 0.5 for either result. This example is
synonymous to flipping a coin 30 times with k being equal to the number of flips resulting in heads.
An important note about these distributions is that the area under the curve, or the probability of each integer added together, always sums
up to 1.
One is able to determine the mean and standard deviation, which is described in the Basic Statistics article, as shown below.
An example of a common question that may arise when dealing with binomial distributions is “What is the probability of a certain result
occurring 33 – 66% of the time?” In the above case, it would be synonymous to the likelihood of heads turning up between 10 and 20 of the
trials. This can be determined by summing the individual probabilities of each integer from 10 to 20.
qquad P(66% X > 33%)=\sum_{possible values}P(N_S)=P(10)+P(11)+...+P(19)+P(20)" src="/@api/deki/files/19527/image-813.png">
The probability of heads resulting in 33 – 66% of the trials when a coin is flipped 30 times is 95.72%.
The probability for binomial distribution can also be calculated using Mathematica. This will simplify the calculations, save time, and
reduce the errors associated with calculating the probability using other methods such as a calculator. The syntax needed to calculate the
probability in Mathematica is shown below in the screen shot. An example is also given for various probabilities based on 10 coin tosses.
where k is the number of events, n is the number of independent samples, and p is the known probability. The example above also generates
a set of values for the probabilities of getting exactly 0 to 10 heads out of 10 total tosses. This will be useful because it simplifies calculating
probabilities such as getting 0-6 heads, 0-7 heads, etc. out of 10 tosses because the probabilities just need to be summed according to the
exact probabilities generated by the table.
In addition to calling the binomial function generated above, Mathematica also has a built in Binomial Distribution function that is
displayed below:
PDF[BinomialDistribution[n,p],k] where n,p, and k still represent the same variables as before
This built in function can be applied to the same coin toss example.
As expected, the probabilities calculated using the built in binomial function matches the probabilities derived from before. Both methods
can be used to ensure accuracy of results.
13.9.3: POISSON DISTRIBUTION

Note: The variables used in this section are defined the same way as seen above in the "Binomial Distribution" section.
A Poisson distribution has several applications, and is essentially a derived limiting case of the binomial distribution. It is most applicably
relevant to a situation in which the total number of successes is known, but the number of trials is not. An example of such a situation would
be if you know the mean expected number of cancer cells present per sample and wanted to determine the probability of finding 1.5 times
that amount of cells in any given sample, you would use a Poisson distribution.
In order to derive the Poisson distribution, you must first start off with the binomial distribution function.
n! k n
P {X − k} = p (1 − p) (13.9.5)
(n − k)!k!
for k = 1, 2, 3 … .
Due to the fact that the Poisson distribution does not require an explicit statement of the total number of trials, you must eliminate n out of
the binomial distribution function. This is done by first introducing a new variable (μ ), which is defined as the expected number of
successes during the given interval and can be mathematically described as:
μ = np (13.9.6)
We can then solve this equation for p, substitute into Equation 13.9.6 , and obtain Equation 13.9.7 , which is a modified version of the
binomial distribution function.
n! μ k μ n−k
P {X = k} = ( ) (1 − ) (13.9.7)
(n − k)!k! n n
If we keep μ finite and allow the sample size to approach infinity we obtain Equation 13.9.8 , which is equal to the Poisson distribution
function. (Notice to solve the Poisson distribution, you do not need to know the total number of trials)
k −μ
μ e
P {X = k} = (13.9.8)
k!
In the above equation:

e is the base of the natural logarithm (e = 2.71828...)
k is the number of occurrences of an event
k! is the factorial of k
μ is a positive real number, equal to the expected number of occurrences that occur during the given interval.
Using the Poisson distribution a probability versus number of successes plot can be made for several different numbers of successes.
This graph represents several key characteristics of a Poisson process.
1. The number of successes is independent of the time since the last event.
2. Probabilities, or outcomes, are discrete.
3. The probability of two or more successes in a short interval is extremely small (approximately zero).
These rules hint that the Poisson processes only model rare events, yet this is not true. What these characteristics say is that if you are
limited to a small interval window, you will observe less than two events. It is important to know that the Poisson distribution is actually an
approximation of the binomial distribution. As n increases and p decreases, the Poisson distribution becomes equal to the binomial
distribution.
13.9.4: HYPERGEOMETRIC DISTRIBUTION

Note: The definitions of the variables in this section are different than the previous sections.
A hypergeometric distribution function is used only if the following three conditions can be met:
1. Only two outcomes are possible
2. The sample must be random
3. Selections are not replaced
Hypergeometric distributions are used to describe samples where the selections from a binary set of items are not replaced. This distribution
applies in situations with a discrete number of elements in a group of N items where there are K items that are different. As a simple
example imagine we are removing 10 balls from a jar of mixed red and green balls without replacing it. What is the chance of selecting 2
green balls? 5 red balls? In general terms, if there are K green balls and N total balls , the chance of selecting k green balls by selecting n
balls is given by
(# of ways to select k green balls )(# of ways to select n − k red balls )
P (x = k) = (13.9.9)
( total number of ways to select)
As another example, say there are two reactants sitting in a CSTR totaling to N molecules. The third reactant can either react with A or B to
make one of two products. Say now that there are K molecules of reactant A, and N-K molecules of reactant B. If we let n denote the
number of molecules consumed, then the probability that k molecules of reactant of A were consumed can be described by the
hypergeometric distribution. Note: This assumes that there is no reverse or side reactions with the products.
In mathematical terms this becomes
(N − K )!K !n!(N − n)!
P {X = k} = label10
k!(K − k)!(N − K + k − n)!(n − k)!N !
where,
N = total number of items
K = total number of items with desired trait
n = number of items in the sample
k = number of items with desired trait in the sample
This can be written in shorthand as
K N −K
( )( )
k n− k
P {X = k} = (13.9.10)
N
( )
n
where
A A!
( ) = (13.9.11)
B (A − B)!B!
The formula can be simplified as follows: There are possible samples (without replacement). There are ways to obtain k green
balls and there are ways to fill out the rest of the sample with red balls.
If the probabilities P are plotted versus k, then a distribution plot similar to the other types of distributions is seen.
 EXAMPLE 13.9.1
Suppose that you have a bag filled with 50 marbles, 15 of which are green. What is the probability of choosing exactly 3 green marbles
if a total of 10 marbles are selected?
Solution
\[P\{X=3\}=\frac{\left(
15
(13.9.12)
3
\right)\left(
50 − 15
(13.9.13)
10 − 3
\right)}{\left(
50
(13.9.14)
10
\right)}]
(50 − 15)!15!10!(50 − 10)!
P {X = 3} =
3!(15 − 3)!(50 − 15 + 3 − 10)!(10 − 3)!50!
P {X = 3} = 0.2979
13.9.4.1: FISHER'S EXACT
A special case of the hypergeometric distribution is the Fisher's exact method. Fisher's exact is the probability of sampling a specific
configuration of a 2 by 2 table with constrained marginals. Marginals in this case refer to the sums for each row and each column.
Therefore, every Fisher's exact will have 4 marginals, those 4 being the sum of the first column, the sum of the second column, the sum of
the first row, and the sum of the second row. Since these values are constant, that also means that the sum of all the elements in the table
will always equal the same thing. This is clarified with the image below:
In the image above, constant marginals would mean that E, F, G, and H would be held constant. Since those values would be constant, I
would also be constant as I can be thought of as the sum of E and F or G and H are of which are constants.
In theory this test can be used for any 2 by 2 table, but most commonly, it is used when necessary sample size conditions for using the z-test
or chi-square test are violated. The table below shows such a situation:
From this pfisher can be calculated:
(a + b)!(c + d)!(a + c)!(b + d)!
pfisher =
(a + b + c + d)!a!b!c!d!
where the numerator is the number of ways the marginals can be arranged, the first term in the denominator is the number of ways the total
can be arranged, and the remaining terms in the denominator are the number of ways each observation can be arranged.
As stated before, this calculated value is only the probability of creating the specific 2x2 from which the pfisher value was calculated.
Another way to calculate pfisher is to use Mathematica. The syntax as well as an example with numbers can be found in the screen shot
below of the Mathematica notebook. For further clarification, the screen shot below also shows the calculated value of pfisher with numbers.
This is useful to know because it reduces the chances of making a simple algebra error.
Another useful tool to calculate pfisher is using available online tools. The following link provides a helpful online calculator to quickly
calculate pfisher. [1]
Using Mathematica or a similar calculating tool will greatly simplify the process of solving for pfisher.
After the value of pfisher is found, the p-value is the summation of the Fisher exact value(s) for the more extreme case(s), if applicable. The
p-value determines whether the null hypothesis is true or false. An example of this can be found in the worked out hypergeometric
distribution example below.
Finding the p-value
As elaborated further here: [2], the p-value allows one to either reject the null hypothesis or not reject the null hypothesis. Just because the
null hypothesis isn't rejected doesn't mean that it is advocated, it just means that there isn't currently enough evidence to reject the null
hypothesis.
In order to fine the p-value, it is necessary to find the probabilities of not only a specific table of results, but of results considered even more
"extreme" and then sum those calculated probabilities. An example of this is shown below in example 2.
What is considered "more extreme" depends on the situation. In the most general sense, more extreme means further from the expected or
random result.
Once the more extreme tables have been created and the probability for each value obtained, they can be added together to find the p-value
corresponding to the least extreme table.
It is important to note that if the probabilities for every possible table were summed, the result would invariably be one. This should be
expected as the probability of a table being included in the entire population of tables is 1.
 EXAMPLE 13.9.1
The example table below correlates the the amount of time that one spent studying for an exam with how well one did on an exam.
There are 5 possible configurations for the table above, which is listed in the Mathematica code below. All the pfisher for each
configuration is shown in the Mathematica code below.
 EXAMPLE 13.9.1
What are the odds of choosing the samples described in the table below with these marginals in this configuration or a more extreme
one?
Solution
First calculate the probability for the initial configuration.
6!19!6!19!
pfisher = = 0.000643704
25!5!1!1!18!
Then create a new table that shows the most extreme case that also maintains the same marginals that were in the original table.
Then calculate the probability for the most extreme case.

6!19!6!19!
pfisher = = 0.00000564653
25!6!0!0!19!
Finally add the two probabilities together.
pfisher = 0.000643704 + 0.00000564653 = 0.00064
13.9.5: MAXIMUM ENTROPY FUNCTION

The maximum entropy principle uses all the testable information about multiple given probability distributions to describe the true
probability distribution by maximizing the information entropy. To derive a distribution with the maximum entropy principle, you obtain
some testable information I about some quantity x. This information is expressed as m constraints on the expectation values of the functions
fk. This makes the Bayesian probability distribution to satisfy:
n
Fk = ∑ Pr(xi ∣ I)fk (xi )
i=1
with k = 1, ⋯ , m .
We also know that all of these probabilities must sum to 1, so the following constraint is introduced:
n
∑ Pr(xi ∣ I) = 1
i=1
Then the probability distribution with maximum information entropy that satisfies all these constraints is:
1
Pr(xi ∣ I) = exp[λ1 f1 (xi ) + ⋯ + λm fm (xi )]
Z (λ1 , ⋯ , λm )
The normalization constant is determined by the classical partition function:

n
Z (λ1 , ⋯ , λm ) = ∑ exp[λ1 f1 (xi ) + ⋯ + λm fm (xi )]
i=1
The λk parameters are Lagrange multipliers with values determined by:

∂
Fk = log Z (λ1 , ⋯ , λm )
∂ λk
All of the well-known distributions in statistics are maximum entropy distributions given appropriate moment constraints. For example, if
we assume that the above was constrained by the second moment, and then one would derive the Gaussian distribution with a mean of 0 and
a variance of the second moment.
The use of the maximum entropy distribution function will soon become more predominant. In 2007 it had been shown that Bayes’ Rule
and the Principle of Maximum Entropy were compatible. It was also shown that maximum entropy reproduces every aspect of orthodox
Bayesian inference methods. This enables engineers to tackle problems that could not be simply addressed by the principle of maximum
entropy or Bayesian methods, individually, in the past.
13.9.6: SUMMARY
The three discrete distributions that are discussed in this article include the Binomial, Hypergeometric, and Poisson distributions. These
distributions are useful in finding the chances that a certain random variable will produce a desired outcome.
13.9.6.1: BINOMIAL DISTRIBUTION FUNCTION

The Binomial distribution function is used when there are only two possible outcomes, a success or a faliure. A success occurs with the
probability p and a failure with the probability 1-p. Suppose now that in n independent trials the binomial random variable X represents the
number of successes. The following equation applies to binomial random variables:
13.9.6.2: POISSON DISTRIBUTION FUNCTION

The Poisson distribution can be used in a large variety of situations. It can also be used to approximate the Binomial Distribution when n is
large and p is small yielding a moderate np.
The following example is a situation in which the Poisson Distribution applies: Suppose a CSTR is full of molecules and the probability of
any one molecule reacting to form product is small (say due to a low temperature) while the number of molecules is large, then the
probability distribution function would fit the Poisson Distribution well (Note: all the molecules are the same here, unlike in the
hypergeometric example to follow). Other examples include, the number of wrong telephone numbers that are dialed each day or the
number of people in a community that live to 100 years of age. The following equation is the Poisson Distribution function:
13.9.6.3: SUMMARY OF KEY DISTRIBUTIONS
13.9.6.4: SUMMARY OF DISTRIBUTION APPROXIMATIONS

There are many useful approximations for discrete random variables. An important approximation is that which yields a normal distribution
because it allows for confidence intervals and probabilities to be continuous. Additionally, since the normal distribution is so well
understood it is used for convenience when speaking to those who have yet to study probability theory. Below is a table of useful
approximations. The majority of the theory behind these approximations is in the application of the Central Limit Theorem and Large
Number Law but these are outside the scope of this text.
The above chart can be read as "A random variable under the following distribution can is approximately a <blank> random variable when
the following requirements are met." For example, the final line of the table can be read: "A random variable following a Poisson
distribution with parameter λ is approximately a continuous Normal random variable where the mean is λ and the variance is λ assuming
that there are at least 100 samples that have been taken."
 EXAMPLE 13.9.1
In order for a vaccination of Polio to be efficient, the shot must contain at least 67% of the appropriate molecule, VPOLIO. To ensure
efficacy, a large pharmaceutical company manufactured a batch of vaccines with each syringe containing 75% VPOLIO. Your doctor
draws a syringe from this batch which should contain 75% VPOLIO. What is the probability that your shot, will successfully prevent
you from acquiring Polio? Assume the syringe contains 100 molecules and that all molecules are able to be delivered from the syringe
to your blood stream.
Solution
This can be done by first setting up the binomial distribution function. In order to do this, it is best to set up an Excel spreadsheet with
values of k from 0 to 100, including each integer. The frequency of pulling a VPOLIO molecule is 0.75. Randomly drawing any
molecule, the probability that this molecule will never be VPOLIO, or in other words, the probability of your shot containing 0
VPOLIO molecules is
100!
0 100−0
P = (0.75) (1 − 0.75)
(100 − 0!)!0!
Note that 0!=1
Your spreadsheet should contain probabilities for all off the possible values of VPOLIO in the shot. A link to our spreadsheet can be
seen at the end of this article. A graph of our distribution is shown below.
The next step is to sum the probabilities from 67 to 100 of the molecules being VPOLIO. This is calculation is shown in the sample
spreadsheet. The total probability of at least 67 of the molecules being VPOLIO is 0.9724. Thus, there is a 97.24% chance that you will
be protected from Polio.
 EXAMPLE 13.9.1: GAUSSIAN APPROXIMATION OF A BINOMIAL DISTRIBUTION
Calculation of the binomial function with n greater than 20 can be tedious, whereas calculation of the Gauss function is always simple.
To illustrate this, consider the following example.
Suppose we want to know the probability of getting 23 heads in 36 tosses of a coin. This probability is given by the following binomial
distribution:
36! 36
P = (0.5) = 3.36%
23!13!
To use a Gaussian Approximation, we must first calculate the mean and standard deviation.
Now we can approximate the probability by the Gauss function below.

2
(x −X)
1 −
2
P = e 2σ = 3.32%
−−
σ√2π
This approximation is very close and requires much less calculation due to the lack of factorials. The usefulness of the Gaussian
approximation is even more apparent when n is very large and the factorials involve very intensive calculation.
 EXAMPLE 13.9.1: WORKED OUT HYPERGEOMETRIC DISTRIBUTION
A teacher has 12 students in her class. On her most recent exam, 7 students failed while the other 5 students passed. Curious as to why
so many students failed the exam, she took a survey, asking the students whether or not they studied the night before. Of the students
who failed, 4 students did study and 3 did not. Of the students who passed, 1 student did study and 4 did not. After seeing the results of
this survey, the teacher concludes that those who study will almost always fail, and she proceeds to enforce a "no-studying" policy. Was
this the correct conclusion and action?
Solution
This is a perfect situation to apply the Fisher's exact test to determine if there is any association between studying and performance on
the exam. First, create a 2 by 2 table describing this particular outcome for the class, and then calculate the probability of seeing this
exact configuration. This is shown below.
(4 + 3)!(1 + 4)!(4 + 1)!(3 + 4)!
pfisher ,1 = = 0.221
(4 + 3 + 1 + 4)!4!3!1!4!
Next, create 2 by 2 tables describing any configurations with the exact same marginals that are more extreme than the one shown
above, and then calculate the probability of seeing each configuration. Fortunately, for this example, there is only one configuration that
is more extreme, which is shown below.
(5 + 2)!(0 + 5)!(5 + 0)!(2 + 5)!

pfisher ,2 = = 0.0265
(5 + 2 + 0 + 5)!5!2!0!5!
Finally, test the significance by calculating the p-value for the problem statement. This is done by adding up all the previously
calculated probabilities.
p = pfisher ,1 + pfisher ,2 = 0.221 + 0.0265 = 0.248
Thus, the p-value is greater than 0.05, which is the standard accepted level of confidence. Therefore, the null hypothesis cannot be
rejected, which means there is no significant association between studying and performance on the exam. Unfortunately, the teacher
was wrong to enforce a "no-studying" policy.
 EXAMPLE 13.9.1: POISSON
The hormone PREGO is only found in the female human body during the onset of pregnancy. There is only 1 hormone molecule per
10,000 found in the urine of a pregnant woman. If we are given a sample of 5,000 hormone molecules what is the probability of finding
exactly 1 PREGO? If we need at least 10 PREGO molecules to be 95% positive that a woman is pregnant, how many total hormone
molecules must we collect? If the concentration of hormone molecules in the urine is 100,000 molecules/mL of urine,what is the
minimum amount of urine (in mL) necessary to insure an accurate test(95% positive for pregnancy)?
Solution
This satisfies the characteristics of a Poisson process because
1. PREGO hormone molecules are randomly distributed in the urine
2. Hormone molecule numbers are discrete
3. If the interval size is made smaller (i.e. our sample size is reduced), the probability of finding a PREGO hormone molecule goes to
zero
Therefore we will assume that the number of PREGO hormone molecules is distributed according to a Poisson distribution.
To answer the first question, we begin by finding the expected number of PREGO hormone molecules:
μ = np
1
μ = (5, 000) ( ) = 0.5
10, 000
Next we use the Poisson distribution to calculate the probability of finding exactly one PREGO hormone molecule:
k −μ
μ e
P {X = k} = = 0.303
k!
where k = 1 and μ = 0.5 .

The second problem is more difficult because we are calculating n. We begin as before by calculating the expected number of PREGO
hormone molecules:
n
μ = np =
10, 000
Now we apply the Poisson Distribution Function equation:

u
−(
⎛ k,00! ⎞
∞ −5 10 k
e )
k μ k μ ′
μ e μ e ⎜ n ⎟
F {X ≥ 10} − 0.95 − ∑ − 1.0 − ∑ − 1.0 − ∑ ⎜( ) ⎟
k! k! ⎜ 10.000 k! ⎟
k=1 k=0 k=1
⎝ ⎠
It is easiest to solve this in Mathematica:
The function FindRoot[] was used in Mathematica because the Solve[] function has problems solving polynomials of degree order
greater than 4 as well as exponential functions. However, FindRoot[] requires an initial guess for the variable trying to be solved, in
this case n was estimated to be around 100,000. As you can see from the Mathematica screen shot the total number of hormone
molecules necessary to be 95% sure of pregnancy (or 95% chance of having atleast 10 PREGO molecules) was 169,622 molecules.
For the last step we use the concentration of total hormone molecules found in the urine and calculate the volume of urine required to
contain 169,622 total hormone molecules as this will yield a 95% chance of an accurate test:
 EXAMPLE 13.9.1: GAUSSIAN APPROXIMATION TO A POISSON DISTRIBUTION
To illustrate the Gaussian approximation to the Poisson distribution, consider a distribution where the mean (μ ) is 64 and the number of
observations in a definite interval (N ) is 72. Determine the probability of these 72 observations?
Solution
Using the Poisson Distribution
N
μ
−μ
P (N ) = e ⋅
N!
7
−64
64 2
P (72) = e ⋅ = 2.9%
72!
This can be difficult to solve when the parameters N and μ are large. An easier approximation can be done with the Gaussian function:
2
(x −X)
1 −
2
P (72) = G64,8 = −−e
2σ = 3.0%
σ√2π
where, X = μ and σ = √−
μ.
 EXERCISE 13.9.1
All of the following are characteristics of the Poisson distribution EXCEPT:

a. The number of successes is independent of the interval
b. There are imaginary numbers
c. Probabilities, or outcomes, are discrete
d. Two or more success in a short interval is extremely small
Answer
TBA
 EXERCISE 13.9.1
If there are K blue balls and N total balls , the chance of selecting k blue balls by selecting n balls in shorthand notation is given by:
a)
b)
c)
d)
Answer
TBA
13.9.7: DISCRETE DISTRIBUTION PRESENTATION: CLOWN TIME

www.youtube.com/v/6WOieU664Ew
13.9.8: REFERENCES
Ross, Sheldon: A First Course in Probability. Upper Saddle River: Prentice Hall, Chapter 4.
Uts, J. and R. Hekerd. Mind on Statistics. Chapter 15 - More About Inference for Categorical Variables. Belmont, CA: Brooks/Cole -
Thomson Learning, Inc. 2004.
Weisstein, Eric W.: MathWorld - Discrete Distributions. Date Accessed: 20 November 2006. MathWorld
Woolf, Peter and Amy Keating, Christopher Burge, Michael Yaffe: Statistics and Probability Primer for Computational Biologists.
Boston: Massachusetts Institute of Technology, pp 3.1 - 3.21.
Wikipedia-Principle of maximum entropy. Date Accessed: 10 December 2008. [3]
Giffin, A. and Caticha, A., 2007,[4].
13.9: Discrete Distributions - Hypergeometric, Binomial, and Poisson is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by
Tommy DiRaimondo, Rob Carr, Marc Palmer, Matt Pickvet, & Matt Pickvet via source content that was edited to conform to the style and standards of the
13.10: MULTINOMIAL DISTRIBUTIONS
Statistical events generate two types of outcomes: continuous or discrete. Continuous outcomes can take an infinite number of values; for
instance, a person's height can take any real number, so there are infinite possibilities. Typical events generating continuous outcomes may
follow a normal, exponential, or geometric distribution. Discrete outcomes can only on take prescribed values; for instance, a dice roll can
only generate an integer between 1 to 6. Discrete outcomes are typically distributed either binomially or multinomially. It is with
multinomial distribution that this article is concerned.
13.10.2: MULTINOMIAL DISTRIBUTIONS: MATHEMATICAL REPRESENTATION

Multinomial distributions specifically deal with events that have multiple discrete outcomes. The Binomial distribution is a specific subset
of multinomial distributions in which there are only two possible outcomes to an event.
Multinomial distributions are not limited to events only having discrete outcomes. It is possible to categorize outcomes with continuous
distributions to different degrees (high, medium, low). For instance, the water level - a continuous entity - in a storage tank can be made
discrete by categorizing them into either "desirable" or "not desirable." Multinomial distributions, therefore, have expansive applications in
process control.
13.10.2.1: PROBABILITY DENSITY FUNCTION

One way of describing the probability of an outcome occurring in a trial is the probability density function. The probability density
function (PDF) mathematically represents the probability of having a specified outcome. The probability density function is a useful way to
find the probability of simultaneous occurrence of specific results (i.e. having n = 1, n = 1, and n = 1 with 3 trials as opposed to other
1 2 3
outcomes, such as n = 3, n = 0, n = 0 or n = 0, n = 1, n = 2).

1 2 3 1 2 3
N!
ni n1 n2 nk
P (n1 , n2 , … , nk ) = (n1 !n2 ! … nk !) ∏ pi = (n1 !n2 ! … nlℓ !) ∗ ( p1 p2 …p )
k
i=1
where
N is the number of trials
k is the number of possible outcomes
n is the number of occurrences of outcome i
i
p is the probability of observing outcome i

i
We know that the the sum of the probabilities of all possible outcomes that can occur in a trial must be unity (since one outcome must
occur). The probability density function yields this result for both continuous and discrete outcomes. However, it is important to note that to
get this result for continuous outcomes, one must take the integral of the probability density function over all possible outcomes. To get this
unity result for discrete outcomes, one must sum the probabilities of each outcome (similar to taking Riemann sums).
13.10.2.2: CUMULATIVE DISTRIBUTION FUNCTION

While the probability density function calculates the probability of a single outcome, the cumulative distribution function (CDF) is a useful
way to find the probability that an outcome lies within a given range of values.
where
ni is the number of occurrences of outcome i
pi is the probability of seeing outcome i
ci is the maximum number of occurrences of outcome i
13.10.2.3: VISUALIZING PROBABILITY DENSITY FUNCTION WITH MATHEMATICA
Before using the functions for multinomial probability distributions, a special package must be loaded using the following command
(depending on the version of Mathematica):
<< Needs["MultivariateStatistics`"] (Mathematica 6.0)
<< Statistics`MultiDiscreteDistributions` (Mathematica 5.2)
Note that the ` (tilde) is the key next to 1 and not the single quote character '.
Table of pertinent Mathematica commands:
Command Description
PDF[multinomial,x] probability density function at x, where x is scalar-, vector-, or matrix-valued depending on distribution
CDF[multinomial,x] cumulative distribution function at x
Mean[multinomial] mean
Variance[multinomial] variance
StandardDeviation[multinomial] standard deviation
To plot the multinomial distribution probability density function (PDF) in Mathematica, follow three simple steps:
Defining the Multinomial Distribution
multinomial = MultinomialDistribution[n,{p1,p2,...pk}] where k is the number of possible outcomes, n is the number of outcomes, and p1
to pk are the probabilities of that outcome occurring. n and p1 to pk are usually given as numbers but can be given as symbols as long as
they are defined before the command.
Defining the PDF of the Multinomial Distribution
pdf=PDF[multinomial, {x1,x2,...,xk}]; The x here simply refers to the variable so this command can be typed as is, and leave the x as a
variable not a number.
Plotting the PDF
Plot3D[pdf, {x1, 0, 6}, {x2, 0, 5}, AxesLabel -> {x1, x2, probability}] the 0,6 and 0,5 are the ranges of x1 and x2 on the plot respectively,
and the AxesLabel argument is to help us see which is which on the plot created. This command can also be typed as is, by leaving all the
x's as variables.
Note that this will only work if you have 2 variables. If there are more variables, constraints can be set so that it can be plotted. For 3
variables, set the third variable x3 as n-x1-x2. See the attached Mathematica notebook for more information.
Media:Multinomial.nb
13.10.2.4: OTHER CHARACTERISTICS

Statistics have historically been useful in descriptive and inferential analysis of data. Likewise, multinomial distribution is also applicable to
the aforementioned areas: descriptive statistics, inferential statistics, and six-sigma. Several key variables are used in these applications:
The expected value below describes the mean of the data. Discrete random variables can take on a range of values; the mean of the data
describes the location of the data within this range.
E (Xi ) = npi
The data's standard deviation describes the variance of the data. The standard deviation of the data describes the spread of the data with
respect to the center value (the mean of the data).
var(Xi ) = npi (1 − pi )
Using the above parameters, it is possible to find the probability of data lying within n standard deviations of the mean. By setting n equal to
6, six sigma quality control can be implemented on the event and outcomes in question.
13.10.2.5: DERIVATION OF BINOMIAL DISTRIBUTION
As was stated above, the binomial distribution is simply a special case of the multinomial distribution. Using the multinomial distribution,
the probability of obtaining two events n1 and n2 with respective probabilities p and p from N total is given by:
1 2
N! n1 n2
P (n1 , n2 ) = (p p )
1 2
n1 !n2 !
If we label the event of interest, say n1 in this case, as "k," then, since only two outcomes are possible, n2 must equal N-k. Finally, If we
label the probability of obtaining k as simply "p," then the probability of obtaining event n2 (p2) must be 1-p, because again only two
outcomes are possible. With these subsitutions, the above equation simplifies to
N! k N−k
P (k, N , p) = p (1 − p)
k!(N − k)!
Which is the familiar binomial distribution, where k is the number of events of interest, N is the total number of events, and p is the
probability of attaining the event of interest.
13.10.3: APPLICATIONS OF MULTINOMIAL DISTRIBUTIONS
As mentioned before, multinomial distributions are a generalized version of binomial distributions. In chemical engineering applications,
multinomial distributions are relevant to situations where there are more than two possible outcomes (temperature = {high, med, low}).
Multinomial systems are a useful analysis tool when a “success-failure” description is insufficient to understand the system. A continuous
form of the multinomial distribution is the Dirichlet distribution.
Using Bayes' Rule is one of the major applications of multinomial distributions. For example, Bayes' Rule can be used to predict the
pressure of a system given the temperature and statistical data for the system. Bayes' Rule can be used to determine the probability of an
event or outcome as mentioned above. Additional details on Bayes' Rule can be found at Bayes' Rule, conditional probability, independence.
13.10.3.1: BAYES' RULE EXAMPLE

In case you’ve forgotten how to use Bayes’ Rule, here is an example that shows how to use it to solve a problem with a multinomial
outcome which results from combining dual possibilities (such as desired flow vs. undesired flow and configuration A vs. configuration B).
You are given a flow apparatus with two possible specific valve configurations, A and B. For specific valve configuration A, desired flow
rates are achieved 98.5% of the time. And for specific valve configuration B, the desired flow rates are achieved 89.3% of the time. An
operator might use the configuration from A 79% of the time when desired flow rates are achieved. Can you determine the probability that a
randomly chosen configuration is the same as the one in specific valve configuration B and produces undesirable flow rates?
Worked Out Solution
We'll be using the following symbols:
FU: {undesirable flow rates}
FD: {desirable flow rates}
CB: {configuration from Apparatus B}
Thus we want to find : , which is the probability of undesirable flow rates and valve configuration B.
If we use Apparatus A 79% of the time then the probability of using Apparatus B would be 21% or : , which is the
probability of using configuration B and getting desired flow rates.
Additionally, we are given : , which is the probability of a desired flow rate, given that configuration B is used.
Then , which is the probability that an undesirable flow rate is obtained, given
that configuration B is used.
The information from above is then used below: , which is the
probability of getting an undesirable flow rate and configuration B which is found by taking the probability of getting an undesirable flow
rate given that configuration B is used and multiplying that by the probability of a desirable flow rate and configuration B.
 EXAMPLE 13.10.1
Consider the scenario in which you toss a fair die 12 times. What is the probability that each face value (1-6) will occur exactly twice?
Solution
The probability can be determined using a multinomial distribution in which 6 outcomes are possible. The individual probabilities are
all equal given that it is a fair die, p = 1/6. The total number of trials N is 12, and the individual number of occurrences in each category
n is 2.
2 2 2 2 2 2
12! 1 1 1 1 1 1
P (2, 2, 2, 2, 2, 2) − ∗( ) ( ) ( ) ( ) ( ) ( ) − 0.003488
2!2!2!2!2!2! 6 6 6 6 6 6
Therefore, the probability of rolling exactly 2 of each face value on a fair die is about 0.35%.
 EXAMPLE 13.10.2
A bowl has 2 maize marbles, 3 blue marbles and 5 white marbles. A marble is randomly selected and then placed back in the bowl. You
do this 5 times. What is the probability of choosing 1 maize marble, 1 blue marble and 3 white marbles?
Solution
N is the number of trials = 5
k is the number of possible outcomes = 3
ni is the number of occurrences of outcome i
The three possible outcomes are choosing a maize marble, a blue marble or a white marble.
We must determine ni and pi to solve the multinomial distribution.
The number of occurrences of the outcome are the number of times we wish to see each outcome. These are given in the problem
statement.
nmaize = 1
nblue = 1
nwhite = 3
The probability of seeing each outcome is easy to find. For example, there are two maize marbles in the bowl of 10, so the probability
of choosing a maize marble is .
pmaize =
pblue =
pwhite =
We can now solve the multinomial distribution as shown below. The probability of choosing 1 maize marble, 1 blue marble and 3 white
marbles is 0.15.
1 1 3
5! 2 3 5
P (1, 1, 3) = ∗( ) ( ) ( ) = 0.15
1!1!3! 10 10 10
 EXAMPLE 13.10.3
Two valves are used to control the flow of liquid out of a storage tank (Tank 1) in to another storage tank (Tank 2) and controlled by
Valves 1 and 2 as seen below. The valves exist in two states: open and closed.
The following table describes the four different valve configurations and the frequency of the desired flow for each valve configuration
based on experimental data.
Configuration Valve 1 Valve 2 Probability of that configuration having the best flow
1 Open Closed 0.305
2 Open Open 0.476
3 Closed Open 0.218
4 Closed Closed 0.001
To reduce costs, two different apparatuses with a similar configuration, but with different pipe dimensions are being considered to
replace the original apparatus. Below are data tables for each potential replacement apparatus with different valve configurations and
the number of times out of 100 measurements that the valve configuration gave the desired flow.
Apparatus 1
# of Best
Configuration Valve 1 Valve 2
Flow
1 Open Closed 42
2 Open Open 37
3 Closed Open 21
4 Closed Closed 0
Apparatus 2
# of Best
Configuration Valve 1 Valve 2
Flow
1 Open Closed 20
2 Open Open 22
3 Closed Open 56
4 Closed Closed 2
Based on the information above, which apparatus is more like the desired model? By how much?
Solution
With the 4 different valve configurations, multinomial distribution can be utilized to calculate the probability of a measurement. The
goal here is to find the model with the highest probability density function, because that one is more like the desired model.
The probability of a measurement can be calculated as follows given that there are 4 different possible valve configurations (see section
on pdf for more information on where this equation came from):
N! n1 n2 n3 n4
P (n1 , n2 , n3 , n4 ) = p p p p
1 2 3 4
n1 !n2 !n3 !n4 !
where
ni = total number of measurements with the best flow rate from each valve configuration
i = configuration
pi = probability of that configuration having the best flow
N = Total # of observations
The calculation for the probability of each apparatus can be done similarly:
Apparatus 1:
Apparatus 2:
Based on the calculations above for both apparatuses, Apparatus 1 is more like the desired model. Apparatus 1 has a higher probability
density function, based on the relative likelihood of each configuration flow.
This figure shows Mathematica code that can be used in order to solve the probability of a multinomial distribution. The n values are
the number of occurrences of each outcome and the p values are the probabilities of eachout come. The function is set for a multinomial
distribution with five different outcomes. However, it can be used for multinomial distributions with fewer outcomes by setting the
unused n values to 0 and the unused p values to any number other than 0. This figure also shows the probabilities calculated from
Apparatus 1 and Apparatus 2.
 EXAMPLE 13.10.4
A runaway reaction occurs when the heat generation from an exothermic reaction exceeds the heat loss. Elevated temperature increases
reaction rate, further increasing heat generation and pressure buildup inside the reactor. Together, the uncontrolled escalation of
temperature and pressure inside a reactor may cause an explosion.
The precursors to a runaway reaction - high temperature and pressure - can be detected by the installation of reliable temperature and
pressure sensors inside the reactor. Runaway reactions can be prevented by lowering the temperature and/or pressure inside the reactor
before they reach dangerous levels. This task can be accomplished by sending a cold inert stream into the reactor or venting the reactor.
Les Lloyd is a process engineer at the Miles Reactor Company that has been assigned to work on a new reaction process. Using
historical data from all the similar reactions that have been run before, Les has estimated the probabilities of each outcome occurring
during the new process. The potential outcomes of the process include all permutations of the possible reaction temperatures (low and
high) and pressures (low and high). He has combined this information into the table below:
Outcome Temperature Pressure Probability
1 high high 0.013
2 high low 0.267
3 low high 0.031
4 low low 0.689
Worried about risk of runaway reactions, the Miles Reactor Company is implementing a new program to assess the safety of their
reaction processes. The program consists of running each reaction process 100 times over the next year and recording the reactor
conditions during the process every time. In order for the process to be considered safe, the process outcomes must be within the
following limits:
Outcome Temperature Pressure Frequency
1 high high n1 = 0
2 high low
3 low high
4 low low n4 = 100 − n1 − n2 − n3
Help Les predict whether or not the new process is safe by answering the following question: What is the probability that the new
process will meet the specifications of the new safety program?
Solution
The probability of the safety guidelines being met is given by the following CDF expression:
where
ni is the number of occurrence of outcome i
ci is the maximum number of occurrence of outcome i
This CDF expression can be evaluated using the following commands in Mathematica:
<< Needs["MultivariateStatistics`"]
<< multinomial = MultinomialDistribution[100, {0.013, 0.267, 0.031, 0.689}]
<< CDF[multinomial, {0, 20, 2, 78}]
The result is:
P (n− = 0, n2 ≤ 20, n3 ≤ 2, n4 = 1000 − n1 − n2 − n3 ) = 0.00132705
Based on this probability calculation, it appears unlikely that this new process will pass the new safety guidelines.

www.youtube.com/v/aAlQpREhy5c
For slides of this presentation by Group Si:Media:Multinomial Distribution.ppt
13.10.5: REFERENCES
Woolf, Peter, Amy Keating, Christopher Burge, and Michael Yaffe. (2004). Statistics and Probability Primer for Computational
Biologists. Massachusetts Institute of Technology.
"Multinomial Distribution." Wikipedia. en.Wikipedia.org/wiki/Multinomial_distribution.
Perry,R.H.(1997). Perry's Chemical Engineer's Handbook, 7th Edition, McGraw-Hill. ISBN 978-0-07-049841-9
Multinomial Calculator: Online Statistical Calculator. Accessed 27 Nov. 2007. http://stattrek.com/Tables/multinomial.aspx#calculator
Authors: Hillary Kast, Andrew Kim, Adhi Paisoseputra, Stephanie Van Kirk
Stewards: Gillian Berberich, Katie Feldt, Christopher Mark, Jason Wong
13.10: Multinomial Distributions is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source content
13.11: COMPARISONS OF TWO MEANS
Engineers often must compare data sets to determine whether the results of their analyses are statistically equivalent. A sensor outputs a
series of values and the engineer must both determine whether the sensor is precise and whether the values are accurate according to a
standard. To make this evaluation, statistical methods are used. One method compares the probability distributions and another uses the
Students t-test on two data sets. Microsoft Excel also has functions that perform the t-test that output a fractional probability to evaluate the
null hypothesis (Basic Statistics).
13.11.1: DISTRIBUTIONS
13.11.1.1: GENERAL DISTRIBUTIONS
Distributions are governed by the probability density function:
2
1 1 x−μ
f(x) = −− exp(− ( ) )
σ√2π 2 σ
where
σ is the standard deviation of a data set
σ is the mean of the data set
is the input value
This equation gives a typical bell curve. Changing σ will alter the shape of the curve, as shown in the graph below.
Changing μ will shift the curve along the x-axis, as shown below.
Changing both variables will have a result similar to the graph below.
For a more in-depth look at distributions, go to (Distributions).
13.11.1.2: OVERLAPPING DISTRIBUTIONS

The overlap between two distribution curves helps determine the probability of the two data sets being from the same distribution. The
probability value increases as overlapping increases. The overlap is shown as the purple region in the graph below.
13.11.2: COMPARISON OF TWO MEANS
13.11.2.1: PROBABILITY
The similarity of two data sets can be determined by finding the probability of overlap. This is illustrated by the following equation:
∞
ps (k ∣ θs )
P ( overlap ) = ∫ min {
−∞ po (k ∣ θ0 )
The functions contained within the integral are the probability distributions of each respective data set. The equation sums the lesser
probability distribution from each data set. After using this equation, the solution will be a value with a range between 0 and 1 indicating the
magnitude of overlapping probability. A value of 0 demonstrates that the two data sets do not overlap at all. A value of 1 demonstrates that
the two data sets completely overlap. Values between 0 and 1 give the extent of overlap between the two data sets. This probability is not
the same as the confidence interval that can be computed with t-tests.
13.11.2.2: STUDENT'S T-TEST
The Student’s t-test is extremely useful for comparing two means. There are various versions of the student t-test, depending on the context
of the problem. Generally, the test quantifies the signal to noise ratio - where the signal is the difference of means and the noise is a function
of the error around the means. If the signal is large and the noise is small, one can be more confident that the difference between the means
is "real" or significant. To prove a significant difference, we need to disprove the null hypothesis. The "null hypothesis" (Ho) is that there is
no difference between the two means. If we are able to disprove the "null hypothesis" then we will be able to say that the two groups are
statistically different within a known level of confidence.
The extremities at either ends of a probability distribution are referred to as the "tails" of the distribution. The assessment of the significance
of a calculated t-value will depend upon whether or not both tails of the distribution need to be considered. This will depend on the form of
the null hypothesis. If your null hypothesis is an equality, then the case where one mean is larger and smaller must be considered; i.e. only
one tail of the distribution should be accounted for. Conversely, if the null hypothesis is an inequality, then you are only concerned with the
domain of values for a mean either less than or greater than the other mean; i.e. both tails of the distribution should be accounted for.
More Info: the student t-distribution
The t-distribution is the resulting probability distribution with a small sample population. This distribution is the basis for the t-test, in order
to find the statistical significance between two means of data. This distribution is in the form of a generalised hyperbolic function (Which
goes into details that would only clutter here. For more information, the Wikipedia site holds a lot of information on the subject:
en.Wikipedia.org/wiki/Generalised_hyperbolic_distribution).
The t-distribution is commonly used when the standard deviation is unknown, or cannot be known (ie: a very small population set). When
the data sets are large, or a standard deviation is assumed, the t-distribution is not very useful for a statistical analysis, and other methods of
analysis should be used. An example of the t-distribution can be seen below:
Comparing Two Unknown True Means when Sample Standard Deviations are approximately Equal
The first accepted assumption is that when two sample means are being compared, the standard deviations are approximately equal. This
method requires the average, standard deviation, and the number of measurements taken for each data set. The deviations are then pooled
into one standard deviation. The equation for this t-test is as follows:
−−−− −− −
Signal x̄1 − x̄2 n1 n2
t = = √
Noise Spooled n1 + n2
where:
−−−−−−−−−−−−−−−−−−−−
2 2
s (n1 − 1) + s (n2 − 1)
1 2
Spooled = √
n1 + n2 − 2
where:
¯¯
x̄1 is the average of the first data set
¯¯
x̄2 is the average of the second data set
n1 is the number of measurements in the first data set
n2 is the number of measurements in the second data set
s1 is the standard deviation of the first data set
\(s_2) is the standard deviation of the second data set
t is a result of the t-test; it relates to values from the Student's t-distribution
Also note that the variance is defined as the square of the standard deviation.
Using t-distribution tables (sample shown below), the confidence level for the two means can then be determined. This confidence level
determines whether the two means are significantly different. The confidence level can be found with the degrees of freedom for the
measurements and the t-value computed above. The degree of freedom is equal to two less than the total number of measurements from the
two data sets, as shown below:
DOF = n1 + n2 − 2
The following table is an image of a t value table, which can also be found (here):
For example, if you had two data sets totaling 10 measurements and you calculated a t-value of 2.305 the confidence level would be 95%.
This means that there is a 95% chance that the two data sets are statistically different and only a 5% chance that the two data sets are
statistically similar. Also, degrees of freedom between the values listed on the table can be found by interpolating between those two values.
Note that there are some drawbacks when evaluating two means to see if they are significant or not. This problem mainly stems from the
standard deviation. If say a set of values has a certain mean x, but the standard deviation was high due to the fact that some numbers in may
have been greatly out of the range of the mean. This standard deviation may imply, from the student's t-test, that the mean x is significantly
different from the mean of another set of data, when in actuality it may not seem that different. Hence, this must be taken into account when
comparing two means using the student's t-test.
Comparing Two Unknown True Means (μ1 = ? and μ2 = ?) with Known True Unequal Standard Deviations (
sigma1 ≠ σ2 (13.11.1)
)
The z-test is used when the difference between a sample mean and the population mean is large enough to be statistically significant. The t-
test and z-test are essentially the same but in the z-test the actual population means ( ) and standard deviations( ) are known. Since the
estimate for difference in standard deviation used here is biased, two sample z-tests are rarely used.
The two sample z-statistic is described by:
Signal x̄1 − x̄2
z = =
−−−−−−−
Noise σ
2
σ
2
1 2
√ +
n1 n2
where:
¯¯
x̄1 is the average of the first data set
¯¯
x̄2 is the average of the second data set
n1 is the number of measurements in the first data set
n2 is the number of measurements in the second data set
σ1 is the known standard deviation of the first population
σ2 is the known standard deviation of the second population
A different table is used to look up the probability of significance, please refer to Z-score table. If p < 0.05 (using a 95% confidence
interval), we can declare a significant difference exists. The p-value is the probability that the observed difference between the means is
caused by sampling variation, or the probability that these two samples came from the same population.
Comparing Two Unknown True Means (μ1 = ? and μ2 = ?) with Unknown True Standard Deviations (σ1 = ? and σ2 = ?)
This is known as the two sample t-statistic, which is used in statistical inference for comparing the means of two independent, normally
distributed populations with unknown true standard deviations. The two sample t-statistic is described by:
Signal x̄1 − x̄2
t = = −−−−−−
Noise s
2
s
2
1 2
√ +
n1 n2
Where:
overlinex1
is the average of the first data set
overlinex2
is the average of the second data set

is the number of measurements in the first data set
is the number of measurements in the second data set
is the standard deviation of the first data set
is the standard deviation of the second data set
13.11.3: COMPARING THE MEAN OF DIFFERENCES FOR PAIRED DATA

This is used in statistical analysis for the case where a single mean of a population occurs when two quantitative variables are collected in
pairs, and the information we desire from these pairs is the difference between the two variables.
Two examples of paired data:
1. Each unit is measured twice. The two measurements of the same data are taken under varied conditions.
2. Similar units are paired prior to the experiment or run. During the experiment or run each unit is placed into different conditions.
The mean of differences t-statistic is described by:
¯
d
t = sd
√n
where:
¯
¯¯
d is the the mean of the differences for a sample of the two measurements
is the standard deviation of the sampled differences
is the number of measurements in the sample
Summary of Two Sample Mean Tests
13.11.3.1: EXCEL METHOD
Instead of using the t-distribution tables to interpolate values, it is often easier to use built-in tools in Excel. The following three functions
can be used for most of the common situations encountered when comparing two means:
The TDIST function is useful when you have calculated a t-value and you want to know the probability that the t-value is significant.
The TINV function is useful when you know the probability of significance you are interested in, and you desire the t-value (essentially
the reverse of the TDIST function). This is helpful if you are designing an experiment and would like to determine the number of
experimental runs needed to test for the difference of two means.
The TTEST function is useful if you have two sets of data and you would like to know the probability that the mean of the two data sets
are significantly different.
TDIST Function
The TDIST function has the syntax "=TDIST(x,deg_freedom,tails)"
Where:
x is the t-value of the statistic
deg_freedom is the number of degrees of freedom of the t-statistic. For comparing sample means and , with sample sizes and
respectively, has degrees of freedom.
tails is the number of tails to be summed for probability. If null hypothesis is an equality, 2 tails will be used. If the null hypothesis is an
inequality, 1 tail will be used.
The output of the function is the fractional probability of the Students t-distribution. For example, if the function returned a value of 0.05,
this would correspond to a 95% or equivalently (1 - 0.05) confidence level for rejecting the null hypothesis.
13.11.3.2: TINV FUNCTION
The TINV function has the syntax "=TINV(probalility, deg_freedom)"
where:
probability is the fractional probability of the Students t-distribution. This is identical to the output of the "TDIST" function.
deg_freedom is the number of degrees of freedom of the t-statistic. For comparing sample means and , with sample sizes and
respectively, has degrees of freedom.
The output of the function is the t-value of the Student's t-distribution.
TTEST Function
The TTEST function has the syntax "=TTEST(array 1, array 2, tails, type)"
where:
array 1 is the first data set
array 2 is the second data set
tails is the number of tails to be summed for probability (1 or 2). If null hypothesis is an equality 2 tails will be used, if the null
hypothesis is an inequality 1 tail will be used.
type is the type of t-test to be performed the values that correspond to each type of test are listed below.
If type equals | This test is performed 1 | Paired 2 | Two-sample equal variance (homoscedastic) 3 | Two-sample unequal variance
(heteroscedastic)
For our purposes we will only be concerned with type = 3. This corresponds to unequal variance (independent data sets). The other two
types are useful and may prove interesting for the curious, but are beyond our scope.
Alternatively if you are not fond of Excel, a website located here will do the TTEST calculation for you.
The output of the function is the fractional probability of the Student's t-distribution. For example, if the function returned a value of 0.05,
this would correspond to a 95% or equivalently (1 - 0.05) confidence level for rejecting the null hypothesis.
This function is very useful when the data-sets of the two means to be compared are known.

You are a product quality engineer at a company that manufactures powdered laundry detergent in 100 ounce boxes. You are in charge of
determining whether the product meets the specifications the company promises the customer. In the past, 100 samples of the process
established normal process conditions. The data is given below. Note that weighing samples using an imprecise scale does not affect your
results. Also assume that the standard deviations are close enough to pool them.
You randomly sample 25 of the products and you get the following data:
The data can also be found in the first tab of this Excel file: Data File
Is the process running significantly different from normal with 95% confidence? In this example, do not use built-in Excel functions.
13.11.4.1: SOLUTION
Using the data, the normal process produces an average product mass of 102.37 and a variance of 3.17.
The sample has an average mass of 100.00 with a variance of 4.00.

These were calculated using the AVERAGE function and VAR function in Excel. See the second tab of this Excel file for further
explanation: Example 1
The null hypothesis is that the means are identical and the difference between the two means is due purely to chance.
Using the t-test:
−−−− −− −
x̄o − x̄s no ns
t = −−−−−−−−−−−−√
2 2
so (no −1)+ ss (ns −1)
no + ns
√
no + ns −2
−−− −− −−−
102 − 100 25 ∗ 100
t = − − −−− −−− − −−−√ = 5.00
3(100−1)+4(25−1) 25 + 100
√
100+25−2
Looking at the table that is reprinted below for reference:
The number of degrees of freedom is:
The t-value corresponding to 95% confidence level at 123 degrees of freedom is between 1.960 and 1.980. Since the calculated t-value,
5.00, is much greater than 1.980, the null hypothesis is rejected at the 95% confidence level.
We conclude that the two means are significantly different. Thus, the process is not running normally and it is time to troubleshoot to find
the problems occurring in the system.
Same problem as "Worked Out Example 1." Instead, use the TTEST function in Excel.
13.11.5.1: SOLUTION
The solution can also be seen in the third tab of this Excel file: Example 2
Using the TTEST function, with tails = 2 and test type = 3, the function gives a T-test value of .
Since is less than 0.05 (from our 95% confidence level) we can once again conclude that we can reject the null hypothesis.
 EXERCISE 13.11.1
If there are two sets of data, one with 15 measurements and another with 47 measurements, how many degrees of freedom would you
enter in the Excel functions?
a. 20
b. 47
c. 15
d. 60
Answer
The number of degrees of freedom is calculated as follows:
where n1 is the number of measurements in the first data set and n2 is the number of measurements in the second data set.
Therefore,
the answer is d.
What does the statement "Accurate within 95% confidence" mean?

a. The average of the data set is statistically equal to the true value within 95%
b. The value is the average plus or minus 95%
c. The average is statistically equal to the true value within 5%
Answer
c
13.11.6: REFERENCES
"Comparison of Two Means." State.Yale.Edu. Yale. 19 Nov. 2006 <http://www.stat.yale.edu/Courses/1997-98/101/meancomp.htm>.
Excel Help File. Microsoft 2006.
Harris, Daniel C. Exploring Chemical Analysis. 3rd ed. New York: W. H. Freeman and Company, 2005. 77-151.
Woolf, Peter, et al. Statistics and Probability Primer for Computational Biologists. Massachusetts Institute of Technology. BE 490/Bio
7.91. Spring 2004. 52-68.
"Z-test." Wikipedia. en.Wikipedia.org/wiki/Z-test
13.11: Comparisons of two means is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Daniel Katzman, Jessica Moreno,
Jason Noelanders, Mark Winston-Galant, & Mark Winston-Galant via source content that was edited to conform to the style and standards of the LibreTexts
13.12: FACTOR ANALYSIS AND ANOVA
First invented in the early 1900s by psychologist Charles Spearman, factor analysis is the process by which a complicated system of many
variables is simplified by completely defining it with a smaller number of "factors." If these factors can be studied and determined, they can
be used to predict the value of the variables in a system. A simple example would be using a person's intelligence (a factor) to predict their
verbal, quantitative, writing, and analytical scores on the GRE (variables).
Analysis of variance (ANOVA) is the method used to compare continuous measurements to determine if the measurements are sampled
from the same or different distributions. It is an analytical tool used to determine the significance of factors on measurements by looking at
the relationship between a quantitative "response variable" and a proposed explanatory "factor." This method is similar to the process of
comparing the statistical difference between two samples, in that it invokes the concept of hypothesis testing. Instead of comparing two
samples, however, a variable is correlated with one or more explanatory factors, typically using the F-statistic. From this F-statistic, the P-
value can be calculated to see if the difference is significant. For example, if the P-value is low (P-value<0.05 or P-value<0.01 - this
depends on desired level of significance), then there is a low probability that the two groups are the same. The method is highly versatile in
that it can be used to analyze complicated systems, with numerous variables and factors. In this article, we will discuss the computation
involved in Single-Factor, Two-Factor: Without Replicates, and Two-Factor: With Replicates ANOVA. Below, is a brief overview of the
different types of ANOVA and some examples of when they can be applied.
13.12.1: OVERVIEW AND EXAMPLES OF ANOVA TYPES

ANOVA Types
Single-Factor ANOVA (One-Way):
One-way ANOVA is used to test for variance among two or more independent groups of data, in the instance that the variance depends on a
single factor. It is most often employed when there are at least three groups of data, otherwise a t-test would be a sufficient statistical
analysis.
Two-Factor ANOVA (Two-Way):
Two-way ANOVA is used in the instance that the variance depends on two factors. There are two cases in which two-way ANOVA can be
employed:
Data without replicates: used when collecting a single data point for a specified condition
Data with replicates: used when collecting multiple data points for a specified condition (the number of replicates must be specified and
must be the same among data groups)
When to Use Each ANOVA Type
Example: There are three identical reactors (R1, R2, R3) that generate the same product.
One-way ANOVA: You want to analyze the variance of the product yield as a function of the reactor number.
Two-way ANOVA without replicates: You want to analyze the variance of the product yield as a function of the reactor number and the
catalyst concentration.
Two-way ANOVA with replicates: For each catalyst concentration, triplicate data were taken. You want to analyze the variance of the
product yield as a function of the reactor number and the catalyst concentration.
ANOVA is a Linear Model
Though ANOVA will tell you if factors are significantly different, it will do so according to a linear model. ANOVA will always assumes a
linear model, it is important to consider strong nonlinear interactions that ANOVA may not incorporate when determining significance.
ANOVA works by assuming each observation as overall mean + mean effect + noise. If there are non-linear relationship between these (for
example, if the difference between column 1 and column 2 on the same row is that column2 = column1^2), then there is the chance that
ANOVA will not catch it.
13.12.2: KEY TERMS

Before further explanation, please review the terms below, which are used throughout this Wiki.
13.12.3: COMPARISON OF SAMPLE MEANS USING THE F-TEST
The F-Test is the ratio of the sample variances. The F-statistic and the corresponding F-Test are used in single-factor ANOVA for purposes
of hypothesis testing.
Null hypothesis (Ho): all sample means arising from different factors are equal
Alternative hypothesis (Ha): the sample means are not all equal
Several assumptions are necessary to use the F-test:
The samples are independent and random
The distribution of the response variable is a normal curve within each population
The different populations may have different means
All populations have the same standard deviation
13.12.3.1: INTRODUCTION TO THE F-STATISTIC

The F-statistic is the ratio of two variance estimates: the variance between groups divided by the variance within groups. The larger the F-
statistic, the more likely it is that the difference between samples is due to the factor being tested, and not just the natural variation within a
group. A standardized table can be used to find Fcritical for any system. Fcritical will depend on alpha, which is a measure of the confidence
level. Typically, a value of alpha = 0.05 is used, which corresponds to 95% confidence. If Fobserved > Fcritical, we conclude with 95%
confidence that the null hypothesis is false. For an explanation of how to read an F-Table, see Interpreting the F-statistic (below). In a
similar manner, F tables can also be used to determine the p-value for a given data set. The p-value for a given data set is the probability that
you could obtain this data set if the null hypothesis were true: that is, if the results were strictly due to chance. When Ho is true, the F-
statistic has an F distribution.
13.12.3.2: F-DISTRIBUTIONS
The F-distribution is important to ANOVA, because it is used to find the p-value for an ANOVA F-test. The F-distribution arises from the
ratio of two Chi squared distributions. Thus, this family has a numerator and denominator degrees of freedom. (For information on the Chi
squared test, click here.) Every function of this family has a a skewed distribution and minimum value of zero.
Figure 1 - F distribution with alpha and Fcritical indicated
13.12.4: SINGLE-FACTOR ANALYSIS OF VARIANCE

In the case of single-factor analysis, also called single classification or one-way, a factor is varied while observing the result on the set of
dependent variables. These dependent variables belong to a specific related set of values and hence, the results are expected to be related.
This section will describe some of the computational details for the F-statistic in one-way ANOVA. Although these equations provide
insight into the concept of analysis of variance and how the F-test is constructed, it is not necessary to learn formulas or to do this analysis
by hand. In practice, computers are always used to do one-way ANOVA
13.12.4.1: SETTING UP AN ANALYSIS OF VARIANCE TABLE

The fundamental concept in one-way analysis of variance is that the variation among data points in all samples can be divided into two
categories: variation between group means and variation between data points in a group. The theory for analysis of variance stems from a
simple equation, stating that the total variance is equal to the sum of the variance between groups and the variation within groups –
Total variation = variation between groups + variation within groups
An analysis of variance table is used to organize data points, indicating the value of a response variable, into groups according to the factor
used in each case. For example, Table 1 is an ANOVA table for comparing the amount of weight lost over a three month period by dieters
on various weight-loss programs.
Table 1 - Amount of weight lost by dieters on various programs over a 3 month period
Program 1 Program 2 Program 3
7 9 15
9 11 12
5 7 18
7
A reasonable question is, can the type of program (a factor) be used to predict the amount of weight a dieter would lose on that program (a
response variable)? Or, in other words, is any one program superior to the others?
13.12.4.2: MEASURING VARIATION BETWEEN GROUPS
The variation between group means is measured with a weighted sum of squared differences between the sample means and the overall
mean of all the data. Each squared difference is multiplied by the appropriate group sample size, ni, in this sum. This quantity is called sum
of squares between groups or SS Groups.
2 2 2 ′2
SSGroups = n1 (x1 − x) + n2 (x2 − x) + … + nk (xk − x) = ∑ nj ( x̄jj − x̄
с
yr ups
The numerator of the F-statistic for comparing means is called the mean square between groups or MS Groups, and it is calculated as -
SSGroups
MSGroups =
k −1
13.12.4.3: MEASURING VARIATION WITHIN GROUPS

To measure the variation among data points within the groups, find the sum of squared deviations between data values and the sample mean
in each group, and then add these quantities. This is called the sum of squared errors, SSE, or sum of squares within groups.
2 2 2 2
S S E = (n1 − 1) s + (n2 − 1) s + … + (nk − 1) s = ∑ (nj − 1) s
1 2 k j
allgroups
where
2
(xij − x̄j )
2
sj = ∑ =
nj − 1
groupj
and is the variance within each group.
The denominator of the F-statistic is called the mean square error, MSE, or mean squares within groups. It is calculated as
2 2 2
SSE (n1 − 1) s + (n2 − 1) s + … + (nk − 1) s
1 2 k
MSE = =
N −k n1 + n2 + … + nk − k
MSE is simply a weighted average of the sample variances for the k groups. Therefore, if all ni are equal, MSE is simply the average of the
k sample variances. The square root of MSE (sp), called the pooled standard deviation, estimates the population standard deviation of the
response variable (keep in mind that all of the samples being compared are assumed to have the same standard deviation σ).
13.12.4.4: MEASURING THE TOTAL VARIATION
The total variation in all samples combined is measured by computing the sum of squared deviations between data values and the mean of
all data points. This quantity is referred to as the total sum of squares or SS Total. The total sum of squares may also be referred to as
SSTO. A formula for the sum of squared differences from the overall mean is
2
SSTotal = ∑ (xij − x̄)
values
where xij represents the jth observation within the ith group, and is the mean of all observed data values. Finally, the relationship between
SS Total, SS Groups, and SS Error is
SS Total = SS Groups + SS Error

Overall, the relationship between the total variation, the variation between groups, and the variation within a group is illustrated by Figure 2.
Figure 2: Visual illustration of variance in ANOVA, adapted from www.uwsp.edu/psych/stat/12/anova-1w.htm

A general table for performing the one-way ANOVA calculations required to compute the F-statistic is given below
Table 2 - One-Way ANOVA Table
Source Degrees of Freedom Sum of Squares Mean Sum of Squares F-Statistic
Between groups k-1
Within groups (error) N-k
Total N-1
13.12.4.5: INTERPRETING THE F-STATISTIC

Once the F-statistic has been found, it can be compared with a critical F value from a table, such as this one: F Table. This F table is
calculated for a value of alpha = 0.05, indicating a 95% confidence level. This means that if Fobserved is larger than Fcritical from the table,
then we can reject the null hypothesis and say with 95% confidence that the variance between groups is not due to random chance, but
rather due to the influence of a tested factor. Tables are also available for other values of alpha and can be used to find a more exact
probability that the difference between groups is (or is not) caused by random chance.
13.12.4.6: FINDING THE CRITICAL F VALUE

In this F Table, the first row of the F table is the number of degrees of between groups (number of groups - 1), and the first column is the
number of degrees of freedom within groups (total number of samples - number of groups).
For the diet example in Table 1, the degree of freedom between groups is (3-1) = 2 and and the degree of freedom within groups is (13-3) =
10. Thus, the critical F value is 4.10.
13.12.4.7: COMPUTING THE 95% CONFIDENCE INTERVAL FOR THE POPULATION MEANS
It is useful to know the confidence interval at which the means of the different groups are reported. The general formula for calculating a
confidence interval is . Because it is assumed that all populations have the same standard
deviation can be used to estimate the standard deviation within each group. Although the population standard deviation is
assumed to be the same, the standard error and the multiplier may be different for each group, due to differences in group size and degrees
of freedom. The standard error of a sample mean is inversely proportional to the square root of the number of data points within the sample.
It is calculated as . The multiplier is determined using a t-distribution where the degrees of freedom are calculated as df =
N-k. Therefore,Insertformulahere the confidence interval for a population mean is . More details on confidence intervals
can be found in Comparison of two means
An example for using factor analysis is the following:
You have two assembly lines. Suppose you sample 10 parts from the two assembly lines. Ho: s12 = s2x2 Ha: variances are not equal Are the
two lines producing similar outputs? Assume a=0.05 F.025,9,9 = 4.03 F1-.025,9,9 = ?
Are variances different?
How would we test if the means are different?
13.12.5: TWO-FACTOR ANALYSIS OF VARIANCE

A two factor or two-way analysis of variance is used to examine how two qualitative categorical variables (male/female) affect the mean of
a quantitative response variable. For example, a psychologist might want to study how the type and volume of background music affect
worker productivity. Alternatively, an economist maybe be interested in determining the affect of gender and race on mean income. In both
of these examples, there is interest in the effect of each separate explanatory factor, as well as the combined effect of both factors.
13.12.5.1: ASSUMPTIONS
In order to use the two-way ANOVA, the following assumptions are required:
Samples must be independent.
Population variances must be equal.
Groups must have same sample size. The populations from which the samples were obtained must be normally distributed (or at least
approximately so).
The null hypothesis is assumed to be true.
The null hypothesis is as follows:
The population means for the first factor have to be equal. This is similar to the one-way ANOVA for the row factor.
The population means for the second factor must also be equal. This is similar to the one-way ANOVA for the column factor.
There isn’t an interaction between the two factors. This is similar to performing an independence test using contingency tables.
More simply, the null hypothesis implies that the populations are all similar and any differences in the populations are caused by chance, not
by the influence of a factor. After carrying out two-way ANOVA it will be possible to analyze the validity of this assumption.
13.12.5.2: TERMS USED IN TWO-WAY ANOVA

The interaction between two factors is the most unique part of a two-way analysis of variance problem. When two factors interact, the
effect on the response variable depends on the value of the other factor. For example, the statement being overweight caused greater
increases in blood pressure for men than women describes an interaction. In other words, the effect of weight (factor) on blood pressure
(response) depends on gender (factor).
The term main effect is used to describe the overall effect of a single explanatory variable. In the music example, the main effect of the
factor "music volume" is the effect on productivity averaged over all types of music. Clearly, the main effect may not always be useful if the
interaction is unknown.
In a two-way analysis of variance, three F-statistics are constructed. One is used to test the statistical significance of the interaction, while
the other two are used to test the significance of the two separate main effects. The p-value for each F-statistic is also reported--a p-value of
<.05 is usually used to indicate significance. When an F-factor is found to have statistical significance, it is considered a main effect. The p-
value is also used as an indicator to determine if the two factors have a significant interaction when considered simultaneously. If one factor
depends strongly on the other, the F-statistic for the interaction term will have a low p-value. An example output of two-way analysis of
variance of restaurant tip data is given in Table 4.
Table 4 - Two-Way Analysis of Variance of Restaurant Tipping Data
Source DF Adj SS Adj MS F-Statistic P-Value
Message 1 14.7 14.7 .13 .715
Sex 1 2602.0 2602.0 23.69 0.00
Interaction 1 438.7 438.7 3.99 .049
Error 85 9335.5 109.8
Total 88 12407.9
In this case, the factors being studied are sex (male or female) and message on the receipt ( :-) or none). The p-values in the last column are
the most important information contained in this table. A lower p-value indicates a higher level of significance. Message has a significance
value of .715. This is much greater than .05, the 95% confidence interval, indicating that this factor has no significance (no strong
correlation between presence of message and amount of tip). The reason this occurs is that there is a relationship between the message and
the sex of the waiter. The interaction term, which was significant with a value of p= 0.049, showed that drawing a happy face increased the
tip for women but decreased it for men. The main effect of waiter sex (with a p-value of approximately 0) shows that there is a statistical
difference in average tips for men and women.
13.12.5.3: TWO-WAY ANOVA CALCULATIONS
Like in one-way ANOVA analysis the main tool used is the square sums of each group. Two-way ANOVA can be split between two
different types: with repetition and without repetition. With repetition means that every case is repeated a set number of times. For the above
example that would mean that the :-) was given to females 10 times and males 10 times, and no message was given to females 10 times and
males 10 times
Using the SS values as a start the F-statistics for two-way ANOVA with repetition are calculated using the chart below where a is the
number of levels of main effect A, b is the number of levels of main effect B, and n is the number of repetitions.
Source SS DF Adj MS F-Statistic
Main Effect A From data given a-1 SS/df MS(A)/MS(W)
Main Effect B From data given b-1 SS/df MS(B)/MS(W)
Interaction Effect From data given (a-1)(b-1) SS/df MS(A*B)/MS(W)
Within From data given ab(n-1) SS/df
Total sum of others abn-1
Without repetition means there is one reading for every case. For example is you were investigating whether or not difference in yield are
more significant based on the day the readings were taken or the reactor that the readings were taken from you would have one reading for
Reactor 1 on Monday, one reading for Reactor 2 on Monday etc... The results for two-way ANOVA without repetition is slightly different in
that there is no interaction effect measured and the within row is replaced with a similar (but not equal) error row. The calculations needed
are shown in the table below.
Source SS DF MS F-Statistic
Main Effect A From data given a-1 SS/df MS(A)/MS(E)
Main Effect B From data given b-1 SS/df MS(B)/MS(E)
Error From data given (a-1)(b-1) SS/df
Total sum of others ab-1
These calculations are almost never done by hand. In this class you will usually use Excel or Mathematica to create these tables. Sections
describing how to use these programs are found later in this chapter.
13.12.6: OTHER METHODS OF COMPARISON

Unfortunately, the conditions for using the ANOVA F-test do not hold in all situations. In this section, several other methods which do not
rely on equal population standard deviations or normal distribution. It is important to realize that no method of factor analysis is appropriate
if the data given is not representative of the group being studied.
13.12.6.1: HYPOTHESES ABOUT MEDIANS

In general, it is best construct hypotheses about a population median, rather than the mean. Using the median accounts for the sample being
skewed based on extreme outliers. Median hypotheses should also be used for dealing with ordinal variables (variables which are described
only as being higher or lower than one other and do not have a precise value). When several populations are compared, the hypotheses are
stated as -
H0: Population medians are equal
Ha: Population medians are not all equal
13.12.6.2: KRUSKAL-WALLIS TEST FOR COMPARING MEDIANS

The Kruskal-Wallis Test provides a method of comparing medians by comparing the relative rankings of data in the observed samples.
This test is therefore referred to as a rank test or non-parametric test because the test does not make any assumptions about the distribution
of data.
To conduct this test, the values in the total data set are first ranked from lowest to highest, with 1 being lowest and N being highest. The
ranks of the values within each group are averaged, and the test statistic measures the variation among the average ranks for each group. A
p-value can be determined by finding the probability that the variation among the set of rank averages for the groups would be as large or
larger as it is if the null hypothesis is true. More information on the Kruskal-Wallis test can be found [here].
13.12.6.3: MOOD'S MEDIAN TEST FOR COMPARING MEDIANS
Another nonparametric test used to compare population medians is Mood's Median Test. Also called the Sign Scores Test, this test
involves multiple steps.
1. Calculate the median (M) using all data points from every group in the study
2. Create a contingency table as follows
A B C Total
Number of values greater than M
Number of values less than or equal to M
3. Calculate the expected value for each data set using the following formula:
( rowtotal )( columntotal )
expected =
grandtotal
4. Calculate the chi-square value using the following formula

2
( actual - expected )
χ =
expected
A chi-square statistic for two-way tables is used to test the null hypothesis that the population medians are all the same. The test is
equivalent to testing whether or not the two variables are related.
13.12.7: ANOVA AND FACTOR ANALYSIS IN PROCESS CONTROL

ANOVA and factor analysis are typically used in process control for troubleshooting purposes. When a problem arises in a process control
system, these techniques can be used to help solve it. A factor can be defined as a single variable or simple process that has an effect on the
system. For example, a factor can be temperature of an inlet stream, flow rate of coolant, or the position of a specific valve. Each factor can
be analyzed individually to determine the effect that changing the input has on the process control system as a whole. The input variable can
have a large, small, or no effect on what is being analyzed. The amount that the input variable affects the system is called the “factor
loading”, and is a numerical measure of how much a specific variable influences the system or the output variable. In general, the larger the
factor loading is for a variable the more of an affect that it has on the output variable.
A simple equation for this would be:
Output = f1 * input1 + f2 * input2 + ... + fn * inputn
where fn is the factor loading for the nth input.
Factor analysis is used in this case study to determine the fouling in an alcohol plant reboiler. This article provides some additional insight
as to how factor analysis is used in an industrial situation.
13.12.8: USING MATHEMATICA TO CONDUCT ANOVA
Mathematica can be used for one-way and two-way factor anaylses. Before this can be done, the ANOVA package must be loaded into
Mathematica using the following command:
Needs["ANOVA`"]
Once this command is executed, the 'ANOVA' command can be utilized.
13.12.8.1: ONE-WAY FACTOR ANALYSIS
The basic form of the 'ANOVA' command to perform a one-way factor analysis is as follows:
ANOVA[data]
An example set of data with five elements would look like:

ANOVA[
]
An output table that includes the degrees of freedom, sum of the squares, mean sum of the squares, F-statistic, and the P-value for the
model, error, and total will be displayed when this line is executed. A list of cell means for each model will be displayed beneath the table.
13.12.8.2: TWO-WAY FACTOR ANALYSIS
The basic form of the 'ANOVA' command to perform a two-way factor analysis is as follows:
ANOVA[data, model, vars]
An example set of data with seven elements would look like:

ANOVA[
,{x,y},{x,y}]
An output table will appear similar to the one that is displayed in the one-way analysis except that there will be a row of statistics for each
variable (i.e. x,y).
13.12.9: ANOVA IN MICROSOFT EXCEL 2007

In order to access the ANOVA data analysis tool, install the package:
1. Click on the Microsoft Office button (big circle with office logo)
2. Click 'Excel Options'
3. Click 'Add-ins' on the left side
4. In the manage drop-down box at the bottom of the window, select 'Excel Add-ins'
5. Click 'Go...'
6. On the Add-Ins window, check the Analysis ToolPak box and click 'OK'
To use this package:
1. Click on the 'Data' tab and select 'Data Analysis'
2. Choose the desired ANOVA type- 'Anova: Single Factor', 'Anova: Two Factor with Replication', or 'Anova: Two Factor without
Replication'(see note below for when to use replication)
3. Select the desired data points including data labels at top of the corresponding columns. Make sure the box is checked for 'Labels in first
row' in the ANOVA parameter window.
4. Specify alpha in the ANOVA parameter window. Alpha represents the level of significance.
5. Output the results into a new worksheet.
NOTE: Anova: Two Factor with Replication is used in the cases where there are multiple readings for a single factor. For instance, the input
below, there are 2 factors, control architecture and unit. This input shows how there are 3 readings corresponding to each control
architecture (FB, MPC, and cascade). In this sense, the control architecture is replicated 3 times, each time providing different data relating
to each unit. So, in this case, you would want to use the Anova Two Factor with Replication option.
Anova: Two Factor without Replication is used in cases where there is only one reading pertaining to a particular factor. For example, in the
case below, each sample (row) is independent of the other samples since they are based on the day they were taken. Since multiple readings
were not taken within the same day, the "without Replication" option should be chosen.
Excel outputs:
Summary:
1. Count- number of data points in a set
2. Sum- sum of the data points in a set
3. Average- mean of the data points in a set
4. Variance- standard deviation of the data points in a set
ANOVA:
1. Sum of squares (SS)
2. The degree of freedom (df)
3. The mean squares (MS)
4. F-statistic (F)
5. P-value
6. Fcritical
See the figure below for an example of the inputs and outputs using Anova: Single Factor. Note the location of the Data Analysis tab. The
data was obtained from the dieting programs described in Table 1. Since the F-statistic is greater than Fcritical, the null hypothesis can be
rejected at a 95% confidence level (since alpha was set at 0.05). Thus, weight loss was not random and in fact depends on diet type chosen.
 EXAMPLE 13.12.1
Determine the fouling rate of the reboiler at the following parameters:

T = 410K
Cc = 16.7g / L
RT = 145min
Which process variable has the greatest effect (per unit) on the fouling rate of the reboiler?
Note that the tables below are made up data. The output data for a single input was gathered assuming that the other input variables
provide a negligible output. Although the factors that affect the fouling of the reboiler are similar to the ones found the the article linked
in the "ANOVA and Factor Analysis in Process Controls" section, the data is not.
Temperature of Reboiler (K) 400 450 500
Fouling Rate (mg/min) 0.8 0.86 0.95
Catalyst Concentration (g/L) 10 20 30

Residence Time (min) 60 120 180

Solution
1) Determine the "factor loading" for each variable.
This can be done using any linearization tool. In this case, the factor loading is just the slope of the line for each set of data. Using
Microsoft Excel, the equations for each set of data are the following:
Temperature of Reboiler
y = 0.0015 * x + 0.195
Factor loading: 0.0015
Catalyst Concentration
y = 0.0805 * x − 0.2833
Residence Time
y = 0.0238 * x − 0.5067
2) Determine the fouling rate for the given process conditions and which process variable affects the fouling rate the most (per unit).
Note that the units of the factor loading value are always the units of the output divided by the units of the input.
Plug in the factor loading values into the following equation:
Output = f1 * input1 + f2 * input2 + ... + fn * inputn
You will end up with:

FoulingRate = 0.0015 * T + 0.0805 * Cc + 0.0238 * RT
Now plug in the process variables:

FoulingRate = 0.0015 * 410 + 0.0805 * 16.7 + 0.0238 * 145
FoulingRate = 5.41mg / min
The process variable that affects the fouling rate the most (per unit) is the catalyst concentration because it has the largest factor loading
value.
 EXAMPLE 13.12.2
The exit flow rate leaving a tank is being tested for 3 cases. The first case is under the normal operating conditions, while the second
(A) and the third (B) cases are for new conditions that are being tested. The flow value of 7 (gallons /hour) is desired with a maximum
of 10. A total of 24 runs are tested with 8 runs for each case. The tests are run to determine whether any of the new conditions will
result in a more accurate flow rate. First, we determine if the new conditions A and B affect the flow rate. The results are as follows:
The recorded values for the 3 cases are tabulated. Following this the values for each case are squared and the sums for all of these are
taken. For the 3 cases, the sums are squared and then their means are found.
These values are used to help determine the table above (the equations give an idea as to how they are calculated). In the same way
with the help of ANOVA, these values can be determine faster. This can be done using the mathematica explained above.
Conclusion:
Fcritical equals 3.4668, from an F-table. Since the calculated F value is greater than Fcritical, we know that there is a statistically
significant difference between 2 of the conditions. Thus, the null hypothesis can be rejected. However we do not know between which
2 conditions there is a difference. A post-hoc analysis will help us determine this. However we are able to confirmed that there is a
difference.
 EXAMPLE 13.12.3
As the new engineer on site, one of your assigned tasks is to install a new control architecture for three different units. You test three
units in triplicate, each with 3 different control architecture: feedback (FB), model predictive control (MPC) and cascade control. In
each case you measure the yield and organize the data as follows:
Do the units differ significantly? Do the control architectures differ significantly?

Solution
This problem can be solved using ANOVA Two factor with replication analysis.
ANOVA analysis works better for?

a. Non-linear models
b. Linear models
c. Exponential models
d. All of the above
Answer
B
Two-Way ANOVA analysis is used to compare?
a. Any two sets of data
b. Two One-Way ANOVA models to each other
c. Two factors on their effect of the output
d. D. B and C
Answer
C
13.12.10: REFERENCES
Ogunnaike, Babatunde and W. Harmon Ray. Process Dynamics, Modeling, and Control. Oxford University Press. New York, NY: 1994.
Uts, J. and R. Hekerd. Mind on Statistics. Chapter 16 - Analysis of Variance. Belmont, CA: Brooks/Cole - Thomson Learning, Inc. 2004.
Charles Spearman. Retrieved November 1, 2007, from www.indiana.edu/~intell/spearman.shtml
Plonsky, M. "One Way ANOVA." Retrieved November 13, 2007, from www.uwsp.edu/psych/stat/12/anova-1w.htm
Ender, Phil. "Statistical Tables F Distribution." Retrieved November 13, 2007, from www.gseis.ucla.edu/courses/help/dist3.html
Devore, Jay L. Probability and Statistics for Engineering and the Sciences. Chapter 10 - The Analysis of Variance. Belment, CA:
Brooks/Cole - Thomson Learning, Inc. 2004.
"Mood's Median Test (Sign Scores Test)" Retrieved November 29, 2008, from
www.micquality.com/six_sigma_glossary/mood_median_test.htm
13.12: Factor analysis and ANOVA is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Alexander Voice, Andrew Wilkins,
Rohan Parambi, Ibrahim Oraiqat, & Ibrahim Oraiqat via source content that was edited to conform to the style and standards of the LibreTexts platform; a
13.13: CORRELATION AND MUTUAL INFORMATION
The application of control networks to engineering processes requires an understanding of the relationships between system variables. One
would expect, for example, to find a relationship between the steam flow rate to a heat exchanger and the outlet stream temperature. How,
then, are engineers to quantify the degree of relation between a pair of variables? Often times, investigations into the correlation and mutual
information of data sets can provide the answer. Correlation analysis provides a quantitative means of measuring the strength of a linear
relationship between two vectors of data. Mutual information is essentially the measure of how much “knowledge” one can gain of a certain
variable by knowing the value of another variable. By utilizing these techniques, engineers can increase the selectivity and accuracy of their
control systems.
13.13.1: CORRELATION
13.13.1.1: POPULATION CORRELATION COEFFICIENT
The correlation coefficient is an important measure of the relationship between two random variables. Once calculated, it describes the
validity of a linear fit. For two random variables, X and Y, the correlation coefficient, ρxy, is calculated as follows:
cov(X, Y )
ρxy =
σx σy
That is, the covariance of the two variables divided by the product of their standard deviations. Covariance serves to measure how much the
two random variables vary together. This will be a positive number if both variables consistently lie above the expected value and will be
negative if one tends to lie above the anticipated value and the other tends to lie below. For a further description of covariance, as well as
how to calculate this value, see Covariance on Wikipedia.
The correlation coefficient will take on values from 1 to -1. Values of 1 and -1 indicate perfect increasing and decreasing linear fits,
respectively. As the population correlation coefficient approaches zero from either side, the strength of the linear fit diminishes.
Approximate ranges representing the relative strength of the correlation are shown below. See the Correlation article on Wikipedia for more
detailed information about the correlation theory establishing these ranges. The ranges also apply for negative values between 0 and -1.
Correlation
Small: 0.10 to 0.29
Medium: 0.30 to 0.49
Large: 0.50 to 1.00
13.13.1.2: SAMPLE CORRELATION COEFFICIENT
A modified form of the expression for the correlation coefficient describes the linearity of a data sample. For n measurements of variables
X and Y , the sample correlation coefficient is calculated as follows:
∑ (xi − x̄) (yi − ȳ )

ρxy =
(n − 1)sx sy
where and are the sample averages, and sx and sy are the samples’ standard deviations. This correlation coefficient will still take on
values from 1 to -1, corresponding to the strength of linearity in the same way as the population correlation coefficient. This expression is
often the more useful for process controls applications; an engineer can measure two variables, calculate their means and standard
deviations, and ultimately determine the validity of a linear relationship between the two. This method is extremely useful when applied to a
sample of experimental data that can be modeled by a normal distribution function. It is simpler and easier to use compared to the
population correlation coefficient because it uses only well defined statistical inputs, mean and standard deviation, rather than the complex
covariance function.
13.13.1.3: CORRELATION COEFFICIENT ASSUMPTIONS: LINEARITY, NORMAL DISTRIBUTION

The correlation coefficient indicates the strength of a linear relationship between two variables with random distribution; this value alone
may not be sufficient to evaluate a system where these assumptions are not valid.
The image below shows scatterplots of Anscombe's quartet, a set of four different pairs of variables created by Francis Anscombe. The four
y variables have the same mean (7.5), standard deviation (4.12), correlation coefficient (0.81) and regression line (y = 3 + 0.5x).
[Image from en.Wikipedia.org/wiki/Correlation Correlation]
However, as can be seen on the plots, the distribution of the variables is very different.
The scatterplot of y1 (top left) seems to exhibit random distribution. It corresponds to what one would expect when considering two
variables linearly correlated and following normal distribution.
The scatterplot of y2 (top right) does not exhibit random distribution; a plot of its residuals would not show random distribution.
However, an obvious non-linear relationship between the two variables can be observed.
The scatterplot of y3 (bottom left) seems to exhibit a near perfect linear relationship with the exception of one outlier. This outlier exerts
enough influence to lower the correlation coefficient from 1 to 0.81. A Q test should be performed to determine if a data point should be
retained or rejected.
The scatterplot of y4 (bottom right) shows another example when one outlier is enough to produce a high correlation coefficient, even
though the relationship between the two variables is not linear.
These examples indicate that the correlation coefficient alone should not be used to indicate the validity of a linear fit.
13.13.1.4: ENGINEERING APPLICATIONS

Measuring the correlation of system variables is important when selecting manipulated and measured control pairs. Say, for instance, that an
engineer takes several measurements of the temperature and pressure within a reactor and notes that the data pairs have a correlation
coefficient of approximately one. The engineer now understands that by applying a control system to maintain the temperature, he will also
be controlling the reactor pressure by way of the linear relationship between the two variables.
Here we begin to see the role correlations play in defining MIMO versus SISO systems. Generally, engineers seek control systems with
independent pairs of manipulated (u) and measured variables (x). If two inputs strongly influence a single output as measured by the
correlation value, then the system will be difficult to decouple into a SISO controller. Correlation values closer to 0 for all but one
manipulated variable means the system will be easier to decouple because here a single system input controls a single system output and
there is limited effect on other process variables.
13.13.1.5: CORRELATION IN MATHEMATICA

Correlation coefficients can be quickly calculated with the aid of Mathematica. To illustrate by example, suppose that an engineer measures
a series of temperatures inside a bacterial chemostat and the corresponding protein concentration in the product. She measures temperatures
of 298, 309, 320, 333, and 345 K corresponding to protein concentrations of 0.4, 0.6, 0.7, 0.85, and 0.9 weight percent protein (these values
are for illustrative purposes and do not reflect the performance of an actual chemostat). The engineer enters these measurements into
Mathematica as a pair of vectors as follows:
where T is a vector containing the temperature measurements and C is a vector containing the concentration measurements. To determine
the correlation coefficient, simply use the syntax below.
In this case, the correlation coefficient of nearly 1 indicates a strong degree of linearity between the chemostat temperature and the protein
concentration, informing the engineer that she can essentially control this concentration through temperature control. The sample correlation
function in Mathematica uses an approximate solution with the mean and standard deviation of the data set, but the value is very close to the
true correlation function value. This can be seen in the figure below as the dashed line represents the sample correlation function, which
deviates slightly from the true correlation function. As the sample becomes larger (i.e. higher n) the approximation gets closer to the true
value. This represents a limitation of the sample correlation function and demonstrates how Mathematica can be used to determine multiple
correlations between sets of data.
13.13.2: MUTUAL INFORMATION

Mutual information (also referred to as transinformation) is a quantitative measurement of how much one random variable (Y) tells us about
another random variable (X). In this case, information is thought of as a reduction in the uncertainty of a variable. Thus, the more mutual
information between X and Y, the less uncertainty there is in X knowing Y or Y knowing X. For our purposes, within any given process,
several parameters must be selected in order to properly run the process. The relationship between variables is integral to correctly
determine working values for the system. For example, adjusting the temperature in a reactor often causes the pressure to change as well.
Mutual information is most commonly measured in logarithms of base 2 (bits) but is also found in base e (nats) and base 10 (bans).
13.13.2.1: EXPLANATION OF MUTUAL INFORMATION
The mathematical representation for mutual information of the random variables A and B are as follows:
p(a, b)
I(A; B) = ∑ ∑ p(a, b) ∗ log( )
p(a)p(b)
b∈B a∈A
where,
p(a, b) is the joint probability distribution function of A and B .
p(a) is the marginal probability distribution function of A .
p(b) is the marginal probability distribution function of B .
For a review and discussion of probability distribution functions, please refer to Probability Distribution (Wikipedia).
When referring to continuous situations, a double integral replaces the double summation utilized above.
p(a, b)
I(A; B, c) = ∫ ∫ p(a, b) ∗ log( )dadb
B A p(a)p(b)
where,
p(a, b) is the joint probability distribution function of A and B .
p(a) is the marginal probability distribution function of A .
p(b) is the marginal probability distribution function of B .
For a review and discussion of probability density functions, please refer to Probability Density Function (Wikipedia).
By performing the functions above, one measures the distance between the joint distribution/density functions of A and B. Since mutual
information is always measuring a distance, the value obtained from this calculation will always be nonnegative and symmetric. Therefore,
In both cases, the information shared between both A and B is being measured. If A and B are independent of each other, then the difference
between the two would be zero. In terms of chemical engineering processes, two independent variables would share no mutual information
which means that one variable provides no information about the other variable. If A and B were identical, then all the information derived
from obtaining variable A would supply the knowledge needed to get variable B.
By knowing the mutual information between two variables A and B, modeling of a multivariable system can be accomplished. If two or
more variables provide the same information or have similar effects on one outcome, then this can be taken into consideration while
constructing a model. For example, if the level of a tank was observed to be too high, pressure and temperature could be plausible variables.
If manipulating both of these variables causes a change in the fluid level in the tank, then there exists some mutual information between the
pressure and temperature. It would then be important to establish the relationship between temperature and pressure to appropriately control
the level of fluid in the tank.
13.13.2.2: VISUAL REPRESENTATION OF MUTUAL INFORMATION
Figure 1. This figure demonstrates how y provides information for x. The blue is p(x) and the red is p(x|y) for a specific y. A. y shares
mutual information with x such that given a y1 the values of x are now narrowed down to the five values. B. y shares mutual information
with x but it shares less information than A and therefore is not as helpful as A, as it does not eliminate as many potential x values. C. y
does not share any mutual information. Given the y3 value, no x values are eliminated as potential values. Since y3 does not provide any
information there is no information between the two variables. D. y shares mutual information with x. Because of the bimodal distribution,
one more piece of information is required to describe which distribution lies within. Therefore, D gives as much information as part B. B
gives 11 potential values while D gives 10 potential values. However, one additional bit of information is required to describe which of the
two curves the x values lie on, thus there are 11 needed inputs.
13.13.2.3: RELATING MUTUAL INFORMATION TO OTHER QUANTITIES/CONCEPTS

Mutual information can be related to other concepts to gain a comprehensive understanding of the information obtained from any given
parameter. Mutual information discussions are most commonly coupled with a discussion on Shannon entropy (See Shannon Entropy for
more information on entropy).
Relative Entropy
In terms of probability theory, the relative entropy of a system measures the distance between two distinct probability functions. In chemical
engineering applications, mutual information can be used to determine the entropy inherent in a system. Mutual information in this aspect
can now be defined simply as:
I(A; B) = H(A) − H(A ∣ B) = H(B) − H(BA) = I(B; A) = H(A) + H(B) − H(A, B)
where,
H(A) = marginal entropy of A.
H(B) = marginal entropy of B.
H(A | B) = conditional entropy of A.
H(B | A) = conditional entropy of B.
H(A,B) = joint entropy of A and B.
As described above, the marginal entropy still represents the uncertainty around a given variable A and B. Conditional entropy describes the
uncertainty in the specified variable that remains after the other variable is known. In terms of mutual information, the conditional entropies
of A and B tell us that a certain number of bits needs to be transferred from A in order to determine B and vice versa. Once again, these bits
of information are determined by the logarithmic base chosen for analysis. One can determine the uncertainty in several different process
parameters with respect to any other process variable.
13.13.2.4: MUTUAL INFORMATION AND BOOLEAN MODELS

Recall that via Boolean Models, a complex series of sensors and valves can be modeled into a simple control system. A Boolean variable,
however, will never return a value because it can only return True or False. This knowledge can be combined to determine a relationship
between two different variables (from Boolean model) and then their dependency can be quantified (mutual information). Utilizing the same
logic that goes into constructing truth tables, data can be input into tables to determine the mutual information value.
13.13.3: CORRELATION EXAMPLE

Suppose an engineer wants to test the applicability of the ideal gas law to a closed tank containing a fixed amount of his company’s gaseous
product. Given that the vessel is of constant volume, and given that no change in moles will occur in this fixed system, one would
hypothesize that the ideal gas law would predict a linear relationship between the temperature and pressure within the tank.
The engineer makes the following measurements and notes his findings in Excel.
Before proceeding with his analysis, the engineer decides to calculate the correlation coefficient for these values in Mathematica.
Noting the correlation coefficient’s approximate value of 1, the engineer proceeds by creating a linear fit of the data in Excel.
Indeed, the data are nearly linear.
13.13.4: SUMMARY
If an engineer were trying to determine if two random variables are linearly related, calculating correlation coefficients would be the first
step to determine this. If the mean and standard deviation for both variables are known values (i.e. calculated from thousands of
measurements), then the population correlation coefficient should be used. To determine the correlation coefficient for a smaller set of data,
the sample correlation coefficient equations should be used. If the two variables were shown to have a high degree of correlation, the next
step would be to plot the data and graphically check the linear relationship of this data. A high correlation coefficient does not necessarily
mean the two variables are linearly related as shown by Anscombe’s quartet, so this graphical check is necessary.
Mutual information helps reduce the range of the probability density function (reduction in the uncertainty) for a random variable X if the
variable Y is known. The value of I(X;Y) is relative, and the larger its value, the more information that is known of X. It is generally
beneficial to try to maximize the value of I(X;Y), thus minimizing uncertainty. The concept of mutual information is quite complex and is
the basis of information theory.
13.13.5: REFERENCES
Larsen, Richard J. and Marx, Morris L. (2006) An Introduction to Mathematical Statistics and Its Applications, New Jersey: Prentice
Hall. ISBN 0-13-186793-8
Wikipedia Correlation.
Wikipedia Mutual Information.
Scholarpedia Mutual Information.
Kojadinovic, Ivan (no date) "On the use of mutual information in data analysis: an overview"
13.13: Correlation and Mutual Information is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Daniel Wheeler, Daniel
McClure, Jeremy Goldfarb, Lilian Ngobi, & Lilian Ngobi via source content that was edited to conform to the style and standards of the LibreTexts
13.14: RANDOM SAMPLING FROM A STATIONARY GAUSSIAN PROCESS
Gaussian distributions are powerful tools for data analysis of randomly generated samples. Gaussian distributions are typically associated
with a “Bell curve”, a graphical representation of a probability density function that is characterized by a peak at the data mean and a width
determined by the standard deviation of the data set. The distribution of the curve around the mean is symmetric, with certain percentages of
the data falling within one standard deviation of the mean. The probability that a data point exists within a certain range of values is equal to
the area under the probability density curve between those two values. The total area under the curve is equal to one because all data points
can be found to exists somewhere in that data set. For more detailed information on Gaussian distributions, consult the section on
Continuous Distributions.
The random number sampler is a powerful tool and is very useful in process modeling. One example of a use for a random number samplers
is to generate weather data from a model to simulate changing climate throughout the year, in order to properly maintain the temperature of
a storage vessel. Another way it can be used is to create a control chart to monitor how well a process is controlled over time based on some
information about normal system operation. The goal of random number generation is to be able to use large amounts of data based on
limited experimentation to test a control scheme.
The objective of this article is to explain how random sampling to form a Gaussian data set is useful in process control. First is discussion of
how to use a Gaussian distribution to analyze probability and compare the statistical similarities/differences between different data sets.
Then is an explanation of how to use the Central Limit Theorem and show how to create a Gaussian distribution from non-Gaussian data
using average values. Last are examples of how to use this theory in process controls, including how to generate and use control charts.
13.14.1: RANDOM NUMBER SAMPLER

The purpose of the random number sampler is to take randomly generated points and turn them into a Gaussian distribution (i.e. a PDF). It
is important to use points distributed throughout the range of 0 to 1. This task is accomplished by assuming a random number (R ) is equal
to the ‘y’ value of the cumulative distribution function (CDF). The CDF is the cumulative area under the curve of the probability density
function (PDF). By taking an integral from −∞ to x in the PDF, this ‘y’ value of the CDF can be found. Since the value of R is already
known, we can find the x values of the CDF and thus, the PDF. With the x values of the PDF known, we can find P (x) and construct the
Gaussian distribution. The relationship between R and x is shown below:
x 2
−(X−μ)
2 1 2
2(σ )
R = ∫ −−−−−−e dX
2
−∞ √2π(σ)
1 x−μ −∞ − μ
R = [Erf( – ) − Erf( – )]
2 σ√2 σ√2
1 x−μ
R = [Erf( – ) + 1]
2 σ√2
–
x = σ√2 InverseErf(2R − 1) + μ
The process of choosing a random number and implementing it into the PDF can be done using Mathematica. Excel is not recommended
because it cannot compute the Erf() of a negative number, so we cannot use it for data points below the mean in a CDF.
The first step for random sampling a stationary Gaussian process is to input the mean (µ ) and the standard deviation (σ) into the equation
below. Then, you can determine the random points either with Random[ ] function in Mathematica or via user input to develop a list of
random numbers between 0 and 1. Using a random number input as R , Mathematica can be used to determine the corresponding data point
x using the syntax below:
Solve[R == (1/2)*(Erf[(x- µ)/( σ*Sqrt[2])]+1), x]
The syntax can be repeated to determine as many random numbers and their corresponding x values as is necessary for your problem. A
CDF plot can be created by plotting all of the random R values versus their corresponding x values.
We will demonstrate how the procedure works in the following example. First, we chose 10 random numbers and found corresponding x
values, given in Table 1 with µ = 10 and σ = 5. See Figure 1 for the plot of the CDF.
Table 1: CDF Random Numbers and Corresponding x Values
Figure 1: Continuous Distribution Function
In order to develop the PDF you would input the mean and standard deviation into the following PDF equation, and find P (x) for each
value of x.
2
−(X−μ)
1
2σ 2
P (μ, σ) = −− −−e
√2πσ 2
The syntax used in Mathematica is:

Solve[P(x) == (1/Sqrt[2*Pi*σ^2])*e^(-(x-µ)^2/(2*σ^2)), P(x)]
The syntax outputs values for P (x) which are then plotted against the x values calculated from the CDF. The following table shows the
random numbers generated in Mathematica, the corresponding x values, and the P (x) values for the PDF. The plot of the PDF is also
shown below.
Table 2: For Each R, x Values and Corresponding P (x) Values
Figure 2: Probability Density Function
As you can see the plot is in fact Gaussian. The distribution will become more Gaussian as more random numbers are used. The take home
message from the random number generator is that a data set with numbers ranging between 0 and 1, obtained through random number
generation, can be converted into a Gaussian distribution.
13.14.2: PROBABILITY PRIMERS

One major advantage of the random number sampler is the ability to generate different data sets without having to actually gather data.
These data sets can be used to help you better understand the workings of many of the different statistical comparisons. For instance, if the
sample size increases, does that lead to more or less significance in the difference of two means? Most of these analysis tools are the topics
of other wikis, so we will revisit a few only briefly with the intent on asking more in-depth questions in example 1.
13.14.2.1: PROBABILITY
Recall that the area under any Gaussian function is related to the probability. Suppose we use our sampler to generate a random set of 100
data points with a mean of 10 and a standard deviation of 5. The probability of creating an additional data point between ‘a’ and ‘b’ is the
integral of the Gaussian distribution function from ‘a’ to ‘b’ as follows:
2
b −(x −μ)
1
2
P (a ≤ x ≤ b) = ∫ −−−−e
2σ
dx
a √2πσ 2
Further information on probabilities can be found in the section on Continuous Distributions.
13.14.2.2: ERROR IN THE MEAN

Assume we have the same data set as described above. Let’s say we wish to add a new data point and compute the probability that the new
mean of the data set lies between ‘c’ and ‘d.’ This calculation is very similar to that described above with one difference. The standard
deviation, which describes the variance in individual data points, is replaced with the standard error in the mean, which describes the
variance in the mean as data sample size changes. The standard error in the mean is calculated as follows:
σ
σμ =
−
√n
Once this value is obtained, we can solve for the probability as follows:
2
−(x −μ)
d
1 2
2σ
μ
P (a ≤ μ ≤ b) = ∫ −−−−e dx
c 2
√2πσμ
Further information on standard error can be found in section on the Comparison of Two Means.
13.14.2.3: COMPARISON OF TWO DATA SETS

Now suppose we have two distinct data sets with different means and standard deviations and we wish to determine if one data set is
statistically different from the second. To do this, we will compute a p-value. The p-value is computed as follows:
2
−(x −μ)
∞ 1 2σ
2
μ1
P (μ1 → μ2 , μ1 < μ2 ) = ∫ e dx
m 2
√2πσμ1
2
−(x −μ)
m 1 2σ
2
μ1
P (μ1 → μ2 , μ1 > μ2 ) = ∫ e dx
−∞ 2
√2πσμ1
Further information on p-values and mean comparisons can be found in these wiki articles: Comparison of Two Means and P-Values
13.14.3: CENTRAL LIMIT THEOREM

Recall that one property of the random number sampler was that as the set of randomly generated numbers increased, errors became more
random and the distribution became more Gaussian. The central limit theorem states that the sampling of a given distribution mean will
approach a normal or Gaussian distribution as the sample size increases. This theorem is proven by the following example illustrated in the
“Statistics for Experimenters: An Introduction to Design, Data Analysis, and Model Building.”
13.14.3.1: ROLLING OF DICE

The probability that a certain face will land upright when rolling a six-sided die is the same for each face. This is shown below in Figure 3
(a). The mean value of a roll can be calculated to be 3.5 by summing up the value of each face and dividing by 6. As the sample size is
increased, this mean value will be shown to have the highest density of occurance. When the sample size is increased to two dice you can
notice in Figure 3 (b) that the density distribution of the average score begins to take the shape of a curve. The density distribution of the
average score for sample size increases to three, five, and ten dice are shown in Figure 3 (c, d, and e respectively). For each increase it can
be noted that the density of the extreme values decreases and the overall density distribution appears more like a Gaussian distribution, as
predicted by the central limit theorem.
Figure 3: Distribution from Increasing Sample Size (Box, "Statistics for Experimenters")
13.14.3.2: RANDOM NUMBER GENERATION

The above dice example is analogous to random number generation. Consider a set of random numbers between 0 and 10. If these numbers
were truly chosen randomly, then you would expect an even distribution of numbers between 0 and 10. The average of these numbers would
be about 5, with a large standard deviation. However, let’s suppose your data set now consists of an average of two random numbers. In this
case, the data will be less evenly distributed, as averaging tends to lessen the contribution of terms further away from 5. The shape of the
distribution will begin to appear Gaussian.
As you increase the number of random numbers averaged together (‘n’) to create each data point, the distribution becomes more Gaussian in
nature. Please refer to figure 4. When n=1 (one random number is used to compute each data point), the density distribution is relatively
even. As ‘n’ increases, the curves become more Gaussian in nature until we have a very smooth Gaussian distribution when n=100 (100
random numbers were averaged to obtain one data point).
Figure 4: Distribution of data points generated by averaging random numbers
 EXAMPLE 13.14.1
In a healthy adult female, studies have shown that the mean fraction of deoxygenated blood leaving the lungs is 0.05 with a standard
deviation 0.01. To obtain this data, 100 healthy females had been sampled using our random number sampler.
a. Assuming this distribution is Gaussian, 95% of all healthy females will have deoxygenated blood fractions in what range (assume
this range is centered over the mean of the data)?
b. What is the expected error in the mean to a 95% level of confidence?
c. A second study has been conducted on pregnant adult females. It was found that the mean fraction of deoxygenated blood leaving
the lungs was 0.06 with a standard deviation of 0.02. Is the deoxygenated blood concentration in this population of women
statistically different?
d. Now assume that only 10 adult females had been sampled instead of 100. Analyze how the answer to part (c) may change.
Solution
a) To solve this problem, recall the equation of a Gaussian distribution. Substitute in the value of the mean (0.05) and standard
deviation (0.01). Next, recall that the area under the curve corresponds to probability, so we can set integrate this function to determine
the range at which the probability equals 0.95. The bounds on this integral are 0.05 + k and 0.05 – k, since we are assuming this data
range is centered about the mean. This integral can be solved in integral form or in error function form, depending on the commands
you choose to use in inputting the function into a computer algebra system solver. Maple, Excel, and Mathematics can be used to solve
the expression simply by entering the last line of math text shown in the solution below.
2
−(x −μ)
1 2
P (μ, σ) = e 2σ
√2πσ 2
2
−(x −0.05)
1 2
P (μ, σ) = e 2(0.01)
2
√2π(0.01)
2
−(x −0.05)
005+k 1 2
2(001)
0.95 = ∫ e dx
005−k 2
√2π(0.01)
1 0.05+k−0.05 0.05−k−0.05
0.95 = [Erf( ) − Erf( )]
2 0.01√2 0.01√2
k = 0.02
Therefore, 95% of healthy adult females have deoxygenated blood levels between 0.03 and 0.07.
b) This problem is solved much in the same way we solved part (a). However, in this case we are looking for the 95% confidence
interval of the mean, and not the entire population. Therefore, the standard deviation must be converted to the standard error in the
mean. Then, all the above calculations are repeated.
σ 0.01
σμ = = = 0.001
− −−−
√n √100
2
−(x −0.05)
1
3
2(0.001)
P (μ, σμ ) = −−−−−− −−−e
√2π(0.001)2
2
0.05+k −(x −0.09)
1
2
2(0.001)
0.95 = ∫ −−−−−− −−−e dx
0.05−k √2π(0.001)2
1 0.05 + k − 0.05 0.05 − k − 0.05

0.95 = [Erf( ) − Erf( )]
– –
2 0.001√2 0.001√2
k = 0.002
Therefore, the 95% confidence interval of the mean is 0.05 ± 0.002.

c) In order to compare the statistical significance of two different data sets, the concept of p-values must be used. Since we are
interested comparing the means of these two data sets, standard deviation will be replaced by standard error in the mean. To find the
probability (or p-value) that pregnancy results in higher levels of deoxygenated blood, we need to calculate the area under the Gaussian
curve for healthy females that is 0.06 or more. Remember, we are evaluating the Gaussian function describing healthy females, so
standard deviation data for the pregnant females is not needed.
σ 0.01
σμ = = = 0.001
− −−−
√n √100
2
−(x −μ)
1 2
2σ
P (μ, σμ ) = −−−−e
k
2
√2πσμ
2
−(x −005)
1 2
2(0.001)
P (μ, σμ ) = −−−−−− −−−e
√2π(0.001)2
2
∞ −(x −0.05)
1
1
2(0.001)
P (x ≥ 0.06) = ∫ −−−−−− −−−e dx
0.06 √2π(0.001)2
1 ∞ − 0.05 0.06 − 0.05

P (x ≥ 0.06) = [Erf( – ) − Erf( – )]
2 0.001√2 0.001√2
1 0.06 − 0.05
P (x ≥ 0.06) = [1 − Erf( – )]
2 0.001√2
P (x ≥ 0.06) = 0
The p-value is equal to zero. By convention, p-values less than 0.05 are considered to be statistically significant. Therefore, we
conclude that pregnancy statistically affects the level of deoxygenated blood in an adult women’s body.
d) The size of the sample only has an affect on the standard error in the mean. To solve this problem, recalculate the standard error and
repeat the calculations above.
σ 0.01
σμ = − = −− = 0.003
√n √10
2
−(x −μ)
1 2
2σ
P (μ, σμ ) = e k
−−−−
2
√2πσμ
3
−(x −0.09)
1
2
2(0000)
P (μ, σμ ) = −−−−−− −−−e
√2π(0.003)2
2
∞ −(x −0.00)
1
2
2(0.003)
P (x ≥ 0.06) = ∫ −−−−−− −−−e dx
0.06 √2π(0.003)2
1 ∞ − 0.05 0.06 − 0.05

P (x ≥ 0.06) = [Erf( ) − Erf( )]
– –
2 0.003√2 0.003√2
1 0.06 − 0.05
P (x ≥ 0.06) = [1 − Erf( – )]
2 0.003√2
P (x ≥ 0.06) = 0.0004
In this case, the p-value is still less than 0.05, so we still arrive at the same conclusion. We also conclude that as sample size decreases,
differences in sample means become less significant because the p-value has slightly increased.
 EXAMPLE 13.14.2: COMPARISON OF TWO DATA SETS
This example is intended to demonstrate how increased sample size affects the comparison of two data sets.
Using a random number generator, four data sets (A,B,C,D) were generated. Each data set contains 100 total data points. For data sets
A and B, two random numbers were averaged to attain each data point, while in C and D, five random numbers were averaged for each
point. This difference resulted in smaller standard deviations for data sets C and D. A summary of the four data sets created is shown in
the following chart. M is the number of random samples averaged to obtain each data point. N is the total number of data points. The
mean and standard deviation for each of the data sets is also given.
a. Compare the means of data sets A and B by calculating the p-value

b. Compare the means of data sets C and D by calculating the p-value
c. Compare the p-values obtained for parts a) and b) above. Explain why the values make sense.
d. What general effect does increasing sample size have on the comparison of two data sets?
Solution
a) The p-value comparing data sets A and B is computed as follows:
σ 2.3
σμ = − = − −− = 0.23
√n √100
2
−(x −μ)
1 2
2σ
P (μ, σμ ) = −−−−e
k
2
√2πσμ
2
−(x −5)
1 2
2(0.23)
P (μ, σμ ) = −−−−−− −−e
2
√2π(0.23)
3
−(x −5)
∞
1 3
2(0.23)
P (x ≥ 5.5) = ∫ −−−−−− −−e dx
2
55 √2π(0.23)
1 ∞− 5 5.5 − 5
P (x ≥ 5.5) = [Erf( – ) − Erf( – )]
2 0.23√2 0.23√2
1 5.5 − 5
P (x ≥ 5.5) = [1 − Erf( – )]
2 0.23√2
P (x ≥ 5.5) = 0.015
From this p-value, we can see that it is very unlikely that these two data sets are statistically the same. There is only a 1.5% chance of
randomly getting a data set with a mean as high as 5.5! It is far more likely that these two data sets are actually statistically different.
b) The p-value comparing data sets C and D is computed as follows:
σ 1.1
σμ = − = − −− = 0.11
√n √100
2
−(x −μ)
1 2
2σ
P (μ, σμ ) = −−−−e
F
2
√2πσμ
3
−(x −5)
1 2
2(0.11)
P (μ, σμ ) = −−−−−− −−e
√2π(0.11)2
2
∞ −(x −5)
1
3
2(0.11)
P (x ≥ 5.5) = ∫ −−−−−− −−e dx
5.5 √2π(0.11)2
1 ∞− 5 5.5 − 5
P (x ≥ 5.5) = [Erf( ) − Erf( )]
– –
2 0.11√2 0.11√2
1 5.5 − 5
P (x ≥ 5.5) = [1 − Erf( – )]
2 0.11√2
P (x ≥ 5.5) = 0
From this p-value, we can see that these two data sets are statistically different. There is an almost 0 percent chance of randomly getting
a data set with a mean as high as 5.5!
c) Comparing the two p-values, we can see that the value for b) is smaller than a), indicating that we are more confident about a
statistical difference between sets C and D than between A and B. Going back to the description of the problem, we can find a reason
for this result. Because of the larger sampling of random numbers in data sets C and D (five numbers averaged for each data point,
compared to just two), these data sets have smaller standard deviations. These smaller standard deviations mean that we have a higher
level of confidence that the sample mean is the true mean. Because each data set is more likely to show the true mean, there is also an
increased likelihood that one data set is statistically different from the other.
d) In a Gaussian distribution, as sample size increases, standard error decreases. This indicates that the sample mean is closer to the true
mean, and two data sets are more likely to be statistically different from each other than if there are smaller sample sizes and higher
standard deviations in the data sets.
13.14.4: CONTROL CHARTS

This section will provide a brief overview on control charts in order to complete example 3, in which our random number sampler is applied
to a control chart situation. For a more detailed explanation on control charts, and the methodology behind them, please see this wiki page:
Control Charts
13.14.4.1: BACKGROUND
Controls charts are tools used to determine whether a particular part of a process is predictable or not. A process is predictable when it is in
a state of statistical control and unpredictable when it is not in a state of statistical control. A state of statistical control simply means that we
can accurately predict what the output of a process will be in the future based on past measurements; it does not mean that the product is on
target or within any limits of consumer acceptability. The random number sampler is a useful tool that can be used to analyze control charts.
As previously discussed, a data set is more Gaussian in nature as the amount of data points is increased. Therefore, one will expect that a
large set of data generated with our random number sampler will always be in statistical control, but smaller sets may contain an element of
unpredictability.
To create a control chart, we begin by looking at historical data measurements on the measurement (variable) of importance, e.g. acetic acid
concentration. Once a reasonable amount of data has been gathered, it is used to calculate appropriate limits for the variable. If the historical
data and future measurements fall within the range of the limits, it is safe to predict that future measurements will also continue to be within
the range of the limits. Conversely, if the historical data does not fall within the range of the limits, it can be safely predicted that future
measurements will not fall within the range of the limits.
13.14.4.2: CONSTRUCTING CONTROL CHARTS

The best way to illustrate how to create a control chart is to go through an example. This is done below. The initial assumption in
developing a control chart is that the process is stable over period of time where a set of measurements can be made on a particular variable
of the process. In the example below, this variable is acetic acid concentration.
 EXAMPLE 13.14.1
Chemical engineers often combine acetic anhydride and water to produce acetic acid in a moderately exothermic reaction. The reaction
in this particular process is supposed to produce a product stream of acetic acid at 5.5 wt%. The composition of the product stream is
measured four times every hour for ten hours. Each hour’s measurements are viewed as a subgroup of data. Table 1 shows the data
obtained from the 40 measurements. Table 1 also listed the average concentration and the range of concentrations for each subgroup of
measurements.
Table 3. Sample Data - Acetic Acid Concentration
To create a control chart for this process, we must first compute the average concentration for each subset. Additionally, we must
determine the range of data for each subset. These steps have already been completed and their values are listed in Table 3 in the last
two columns on the right. Next the grand average concentration and average range should be computed by averaging the average
concentrations of each subset and by taking the average of each subset's ranges. For the data in table 3, the grand average concentration
is 5.51 wt% and the average range is 0.74 wt%. At this point, we are ready to determine our upper and lower control limits for our
acetic acid concentration. To determine these limits we will need to use a table of constants that have been mathematically derived for
Gaussian distributions. These numbers can then be inserted into equations that find the upper and lower average control limits (UCLx
and LCLx) and the upper and lower range control limits (UCLR andLCLR). The following equations provide the control limits for the
average concentrations and for their ranges.
Table 5. Constants for Average and Range Charts Based on the Average Range
For this example, our subgroup size, n, is 4, A2 is 0.729, D3 is 0 and D4 is 2.282. When plugged into the above equations, they yield
control limits of:
Graphically the control chart is represented by creating a plot of the average concentration values versus their subgroup number and
creating a plot of each range value versus the subgroup number. Horizontal lines should be included on these plots which indicate the
control limits. Control charts for this example are plotted in figures 3 and 4.
Figure 6. Control Chart for Average Acetic Acid Concentration
Figure 7. Control Chart for Ranges of Acetic Acid Concentration
13.14.4.3: INTERPRETING CONTROL CHARTS

If any of the following rules hold true for the centering control chart, the process is not in statistical control.
1. One or more points fall outside the control limits.
2. Seven or more consecutive points fall on the same side of the centerline.
3. Ten of 11 consecutive points fall on the same side of the centerline.
4. Three or more consecutive points fall on the same side of the centerline and all are located closer to the control limit than the centerline.
Observe that in both the control chart for the average concentrations, the historical data never exceeds or goes below the control limits.
Also, none of the other rules above hold true for this chart. This suggests that the process is stable; however, to confirm this thought, future
observations must be made. If the average acetic acid concentration and its range continues to stay within the control limits, the process is
said to be stable.
In the event that the historical data did not stay within the control limits, the plots could have looked like figures 5 & 6.
Figure 8. Control Chart for Average Acetic Acid Concentration (Unstable)
Figure 9. Control Chart for Ranges of Acetic Acid Concentration (Unstable)

Observe that in each of these control charts, the historical data does not stay within the control limits for the process. Based on this
observation, it can be predicted that future data will also not stay within the control limits for the process and the process is not stable. In
conclusion, control charts allow you to use samples of data to determine the control limits for a process and evaluate if the process is stable
or not.
 EXAMPLE 13.14.1: RANDOM SAMPLING PROBLEM
In an industrial bioengineering process, it is desired to obtain a 75% proline to 25% hydroxyproline ratio in a specialized enzyme. The
relative compositions can be analyzed using a fluorescent dye, where a fluorescence of 10 corresponds to the correct ratio of these two
amino acids. To monitor the stability of this process, five reading are taken every 30 minutes for the duration of the 10 hour production
process. The results were obtained using our random number sampler with a specified mean (10) and standard deviation (in this case
1.6). The results are in the table below
a. Calculate the average and range for each data subgroup.
b. Create an average and range chart for the entire time of production.
c. Is this process in control?
Solution
The average and range can be calculated using the =AVERAGE() and =MAX()-MIN() functions of Microsoft excel.
b) The grand average is 10.2 and the average range is 3.8. Using Table 2, A2 = 0.577, D3 = 0, and D4 = 2.114. Therefore:
The charts are as follows:
c) The first rule is not violated as none of the points fall outside the upper and lower control limits. Seven or more points do not fall on
one side of the centerline (the maximum was six), so rule two is not violated. Rule three was not violated, as 10 of 11 points did not fall
on the same side of the centerline (in ouir case, eight was the mx). Finally, rule four was not violated as none of the points were closer
to the control limits then the centerline. Therefore, this process is under statistic control since none of the rules were violated.
Exercise 13.14.1
Our random number sampler was designed based on which key principle:
a. Data points obtained by taking the average of several random numbers are more likely to fit a Gaussian distribution.
b. The area under a Gaussian distribution is always between zero and one and equal to the 'y' value of the CDF.
c. The distribution of data points in a Gaussian distribution is caused by random uncertainties.
d. A process under Gaussian statistical control will not have a data sample exceed the upper or lower control limit
Answer
A
Exercise 13.14.2
Which of the following would decrease the p-value if comparing two different data sets:
a. Smaller sample size
b. Smaller count of averaged random numbers
c. Smaller standard deviation
d. Smaller difference in mean values
Answer
C
13.14.5: REFERENCES
Box, George E., William G. Hunter, and J S. Hunter. Statistics for Experimenters: An Introduction to Design, Data Analysis, and Model
Building. New York: John Wiley & Sons. 43-45.
Liptak, Bela G. "Process Control and Optimization." Instrument Engineers' Handbook 4: 405-413.
Wheeler, Donald J., and David S. Chambers. Understanding Statistical Process Control. 2nd ed. Knoxville: SPC P. 37-88.
Woolf, Peter, Amy Keating, Christopher Burge, and Michael Yaffe. Statistics and Probability Primer for Computational Biologists.
Massachusetts Institute of Technology. 2004.
13.14: Random sampling from a stationary Gaussian process is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Halley
Crast, Andrew Laskowski, Maurice Telesford, Emily Yatch, & Emily Yatch via source content that was edited to conform to the style and standards of the
CHAPTER OVERVIEW
14: DESIGN OF EXPERIMENTS

14.1: Design of Experiments via Taguchi Methods - Orthogonal Arrays
14.2: Design of experiments via factorial designs
14.3: Design of Experiments via Random Design
14.4: Summary- Summary on Control Architectures’ philosophies, advantages, and disadvantages.
14: Design of Experiments is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via source content that was
1
14.1: DESIGN OF EXPERIMENTS VIA TAGUCHI METHODS - ORTHOGONAL
ARRAYS
The Taguchi method involves reducing the variation in a process through robust design of experiments. The overall objective of the
method is to produce high quality product at low cost to the manufacturer. The Taguchi method was developed by Dr. Genichi Taguchi of
Japan who maintained that variation. Taguchi developed a method for designing experiments to investigate how different parameters affect
the mean and variance of a process performance characteristic that defines how well the process is functioning. The experimental design
proposed by Taguchi involves using orthogonal arrays to organize the parameters affecting the process and the levels at which they should
be varies. Instead of having to test all possible combinations like the factorial design, the Taguchi method tests pairs of combinations. This
allows for the collection of the necessary data to determine which factors most affect product quality with a minimum amount of
experimentation, thus saving time and resources. The Taguchi method is best used when there is an intermediate number of variables (3 to
50), few interactions between variables, and when only a few variables contribute significantly.
The Taguchi arrays can be derived or looked up. Small arrays can be drawn out manually; large arrays can be derived from deterministic
algorithms. Generally, arrays can be found online. The arrays are selected by the number of parameters (variables) and the number of levels
(states). This is further explained later in this article. Analysis of variance on the collected data from the Taguchi design of experiments can
be used to select new parameter values to optimize the performance characteristic. The data from the arrays can be analyzed by plotting the
data and performing a visual analysis, ANOVA, bin yield and Fisher's exact test, or Chi-squared test to test significance.
In this article, the specific steps involved in the application of the Taguchi method will be described and examples of using the Taguchi
method to design experiments will be given.
14.1.1: SUMMARY OF TAGUCHI METHOD

14.1.1.1: PHILOSOPHY OF THE TAGUCHI METHOD
1. Quality should be designed into a product, not inspected into it. Quality is designed into a process through system design, parameter
design, and tolerance design. Parameter design, which will be the focus of this article, is performed by determining what process
parameters most affect the product and then designing them to give a specified target quality of product. Quality "inspected into" a
product means that the product is produced at random quality levels and those too far from the mean are simply thrown out.
2. Quality is best achieved by minimizing the deviation from a target. The product should be designed so that it is immune to
uncontrollable environmental factors. In other words, the signal (product quality) to noise (uncontrollable factors) ratio should be
high.
3. The cost of quality should be measured as a function of deviation from the standard and the losses should be measured system
wide. This is the concept of the loss function, or the overall loss incurred upon the customer and society from a product of poor quality.
Because the producer is also a member of society and because customer dissatisfaction will discourage future patronage, this cost to
customer and society will come back to the producer.
14.1.1.2: TAGUCHI METHOD DESIGN OF EXPERIMENTS

The general steps involved in the Taguchi Method are as follows:
1. Define the process objective, or more specifically, a target value for a performance measure of the process. This may be a flow rate,
temperature, etc. The target of a process may also be a minimum or maximum; for example, the goal may be to maximize the output
flow rate. The deviation in the performance characteristic from the target value is used to define the loss function for the process.
2. Determine the design parameters affecting the process. Parameters are variables within the process that affect the performance measure
such as temperatures, pressures, etc. that can be easily controlled. The number of levels that the parameters should be varied at must be
specified. For example, a temperature might be varied to a low and high value of 40 C and 80 C. Increasing the number of levels to vary
a parameter at increases the number of experiments to be conducted.
3. Create orthogonal arrays for the parameter design indicating the number of and conditions for each experiment. The selection of
orthogonal arrays is based on the number of parameters and the levels of variation for each parameter, and will be expounded below.
4. Conduct the experiments indicated in the completed array to collect data on the effect on the performance measure.
5. Complete data analysis to determine the effect of the different parameters on the performance measure.
See below for a pictorial depiction of these and additional possible steps, depending on the complexity of the analysis.
A detailed description of the execution of these steps will be discussed next.
14.1.2: TAGUCHI LOSS FUNCTION

The goal of the Taguchi method is to reduce costs to the manufacturer and to society from variability in manufacturing processes. Taguchi
defines the difference between the target value of the performance characteristic of a process, τ, and the measured value, y, as a loss function
as shown below.
2
l(y) = kc (y − τ )
The constant, kc, in the loss function can be determined by considering the specification limits or the acceptable interval, delta.
C
kc =
Δ2
The difficulty in determining kc is that τ and C are sometimes difficult to define.

If the goal is for the performance characteristic value to be minimized, the loss function is defined as follows:
2
l(y) = kc y
where τ = 0 .
If the goal is for the performance characteristic value to maximized, the loss function is defined as follows:
kc
l(y) =
2
y
The loss functions described here are the loss to a customer from one product. By computing these loss functions, the overall loss to society
can also be calculated.
14.1.3: DETERMINING PARAMETER DESIGN ORTHOGONAL ARRAY

The effect of many different parameters on the performance characteristic in a condensed set of experiments can be examined by using the
orthogonal array experimental design proposed by Taguchi. Once the parameters affecting a process that can be controlled have been
determined, the levels at which these parameters should be varied must be determined. Determining what levels of a variable to test requires
an in-depth understanding of the process, including the minimum, maximum, and current value of the parameter. If the difference between
the minimum and maximum value of a parameter is large, the values being tested can be further apart or more values can be tested. If the
range of a parameter is small, then less values can be tested or the values tested can be closer together. For example, if the temperature of a
reactor jacket can be varied between 20 and 80 degrees C and it is known that the current operating jacket temperature is 50 degrees C, three
levels might be chosen at 20, 50, and 80 degrees C. Also, the cost of conducting experiments must be considered when determining the
number of levels of a parameter to include in the experimental design. In the previous example of jacket temperature, it would be cost
prohibitive to do 60 levels at 1 degree intervals. Typically, the number of levels for all parameters in the experimental design is chosen to be
the same to aid in the selection of the proper orthogonal array.
Knowing the number of parameters and the number of levels, the proper orthogonal array can be selected. Using the array selector table
shown below, the name of the appropriate array can be found by looking at the column and row corresponding to the number of parameters
and number of levels. Once the name has been determined (the subscript represents the number of experiments that must be completed), the
predefined array can be looked up. Links are provided to many of the predefined arrays given in the array selector table. These arrays were
created using an algorithm Taguchi developed, and allows for each variable and setting to be tested equally. For example, if we have three
parameters (voltage, temperature, pressure) and two levels (high, low), it can be seen the proper array is L4. Clicking on the link L4 to view
the L4 array, it can be seen four different experiments are given in the array. The levels designated as 1, 2, 3 etc. should be replaced in the
array with the actual level values to be varied and P1, P2, P3 should be replaced with the actual parameters (i.e. voltage, temperature, etc.)
Array Selector
Links to Orthogonal Arrays

The following links are connected to images of the orthogonal array named in the link title:
L4 Array
L8 Array
L9 Array
L12 Array
L16 Array
L'16 Array
L18 Array
L25 Array
L27 Array
L32 Array
L'32 Array
L36 Array
L50 Array
14.1.3.1: IMPORTANT NOTES REGARDING SELECTION + USE OF ORTHOGONAL ARRAYS

Note 1
The array selector assumes that each parameter has the same number of levels. Sometimes this is not the case. Generally, the highest value
will be taken or the difference will be split.
The following examples offer insight on choosing and properly using an orthogonal array. Examples 1 and 2 focus on array choice, while
Example 3 will demonstrate how to use an orthogonal array in one of these situations.
Example 1:
# parameter: A, B, C, D = 4 # levels: 3, 3, 3, 2 = ~3 array: L9
Example 2:
# parameter: A, B, C, D, E, F = 6 # levels: 4, 5, 3, 2, 2, 2 = ~3 array: modified L16
Example 3:
A reactor's behavior is dependent upon impeller model, mixer speed, the control algorithm employed, and the cooling water valve type. The
possible values for each are as follows:
Impeller model: A, B, or C
Mixer speed: 300, 350, or 400 RPM
Control algorithm: PID, PI, or P
Valve type: butterfly or globe
There are 4 parameters, and each one has 3 levels with the exception of valve type. The highest number of levels is 3, so we will use a value
of 3 when choosing our orthogonal array.
Using the array selector above, we find that the appropriate orthogonal array is L9:
When we replace P1, P2, P3, and P4 with our parameters and begin filling in the parameter values, we find that the L9 array includes 3
levels for valve type, while our system only has 2. The appropriate strategy is to fill in the entries for P4=3 with 1 or 2 in a random,
balanced way. For example:
Here, the third value was chosen twice as butterfly and once as global.
Note 2
If the array selected based on the number of parameters and levels includes more parameters than are used in the experimental design,
ignore the additional parameter columns. For example, if a process has 8 parameters with 2 levels each, the L12 array should be selected
according to the array selector. As can be seen below, the L12 Array has columns for 11 parameters (P1-P11). The right 3 columns should
be ignored.
14.1.4: ANALYZING EXPERIMENTAL DATA

Once the experimental design has been determined and the trials have been carried out, the measured performance characteristic from each
trial can be used to analyze the relative effect of the different parameters. To demonstrate the data analysis procedure, the following L9 array
will be used, but the principles can be transferred to any type of array.
In this array, it can be seen that any number of repeated observations (trials) may be used. Ti,j represents the different trials with i =
experiment number and j = trial number. It should be noted that the Taguchi method allows for the use of a noise matrix including external
factors affecting the process outcome rather than repeated trials, but this is outside of the scope of this article.
To determine the effect each variable has on the output, the signal-to-noise ratio, or the SN number, needs to be calculated for each
experiment conducted. The calculation of the SN for the first experiment in the array above is shown below for the case of a specific target
value of the performance characteristic. In the equations below, yi is the mean value and si is the variance. yi is the value of the performance
characteristic for a given experiment.
2
ȳ
i
S Ni = 10 log
2
si
where
Ni
1
ȳ i = ∑ yi,u
Ni
u=1
Ni
1
2
s = ∑ (yi,u − ȳ i )
i
Ni − 1
u=1
and
iis the experimental number,
u is the trial number, and
N is the numb er of trials for experiment i
i
For the case of minimizing the performance characteristic, the following definition of the SN ratio should be calculated:
Ni 2
yu
S Ni = −10 log(∑ )
Ni
u=1
For the case of maximizing the performance characteristic, the following definition of the SN ratio should be calculated:
Ni
1 1
S Ni = −10 log[ ∑ ]
2
Ni yu
u=1
After calculating the SN ratio for each experiment, the average SN value is calculated for each factor and level. This is done as shown
below for Parameter 3 (P3) in the array:
Once these SN ratio values are calculated for each factor and level, they are tabulated as shown below and the range R (R = high SN - low
SN)of the SN for each parameter is calculated and entered into the table. The larger the R value for a parameter, the larger the effect the
variable has on the process. This is because the same change in signal causes a larger effect on the output variable being measured.
Please refer to the Worked Out Example for a numeric example of how the data analysis procedure described here is applied.

An advantage of the Taguchi method is that it emphasizes a mean performance characteristic value close to the target value rather than a
value within certain specification limits, thus improving the product quality. Additionally, Taguchi's method for experimental design is
straightforward and easy to apply to many engineering situations, making it a powerful yet simple tool. It can be used to quickly narrow
down the scope of a research project or to identify problems in a manufacturing process from data already in existence. Also, the Taguchi
method allows for the analysis of many different parameters without a prohibitively high amount of experimentation. For example, a
process with 8 variables, each with 3 states, would require 6561 (38) experiments to test all variables. However using Taguchi's orthogonal
arrays, only 18 experiments are necessary, or less than .3% of the original number of experiments. In this way, it allows for the identification
of key parameters that have the most effect on the performance characteristic value so that further experimentation on these parameters can
be performed and the parameters that have little effect can be ignored.
The main disadvantage of the Taguchi method is that the results obtained are only relative and do not exactly indicate what parameter has
the highest effect on the performance characteristic value. Also, since orthogonal arrays do not test all variable combinations, this method
should not be used with all relationships between all variables are needed. The Taguchi method has been criticized in the literature for
difficulty in accounting for interactions between parameters. Another limitation is that the Taguchi methods are offline, and therefore
inappropriate for a dynamically changing process such as a simulation study. Furthermore, since Taguchi methods deal with designing
quality in rather than correcting for poor quality, they are applied most effectively at early stages of process development. After design
variables are specified, use of experimental design may be less cost effective.
14.1.6: OTHER METHODS OF EXPERIMENTAL DESIGN

Two other methods for determining experimental design are factorial design and random design. For scenarios with a small number of
parameters and levels (1-3) and where each variable contributes significantly, factorial design can work well to determine the specific
interactions between variables. However, factorial design gets increasingly complex with an increase in the number of variables. For large
systems with many variables (50+) where there are few interactions between variables, random design can be used. Random design assigns
each variable a state based on a uniform sample (ex: 3 states = 0.33 probability) for the selected number of experiments. When used
properly (in a large system), random design usually produces an experimental design that is desired. However, random design works poorly
for systems with a small number of variables.
To obtain a even better understanding of these three different methods, it's good to get a visual of these three methods. It will illustrate the
degree of efficiency for each experimental design depending on the number of variables and the number of states for each variable. The
following will have the three experimental designs for the same scenario.
Scenario. You have a CSTR that has four(4) variables and each variable has three or two states. You are to design an experiment to
systematically test the effect of each of the variables in the current CSTR.
Experimental Design #1: Factorial Design By looking at the # variables and # states, there should be a total of 54 experiments because
(3impellers)(3speeds)(3controllers)(2valves)=54. Here's a list of these 54 experiments:
Experimental Design #2: Taguchi Method Since you know the # of states and variables, you can refer to the table above in this wiki and
obtain the correct Taguchi array. It turns out to be a L9 array.
With the actual variables and states, the L9 array should look like the following:
Experimental Design #3: Random Design
Since we do not know the number of signal recoveries we want and we don't know the probabilities of each state to happen, it will be
difficult to construct a random design table. It will mostly be used for extreme large experiments. Refer to the link below to help you obtain
a better grasp on the random design concept.
http://groups.csail.mit.edu/drl/journal_club/papers/CS2-Candes-Romberg-05.pdf
 EXAMPLE 14.1.1
A microprocessor company is having difficulty with its current yields. Silicon processors are made on a large die, cut into pieces, and
each one is tested to match specifications. The company has requested that you run experiments to increase processor yield. The factors
that affect processor yields are temperature, pressure, doping amount, and deposition rate.
Determine the Taguchi experimental design orthogonal array. The operating conditions for each parameter and level are listed below:
A: Temperature
A1 = 100ºC
A2 = 150ºC (current)
A3 = 200ºC
B: Pressure
B1 = 2 psi
B2 = 5 psi (current)
B3 = 8 psi
C: Doping Amount
C1 = 4%
C2 = 6% (current)
C3 = 8%
D: Deposition Rate
D1 = 0.1 mg/s
D2 = 0.2 mg/s (current)
D3 = 0.3 mg/s
Conducting three trials for each experiment, the data below was collected. Compute the SN ratio for each experiment for the target
value case, create a response chart, and determine the parameters that have the highest and lowest effect on the processor yield.
Solution
a
The L9 orthogonal array should be used. The filled in orthogonal array should look like this:
This setup allows the testing of all four variables without having to run 81 [=34=(3 Temperatures)(3 Pressures)(3 Doping Amounts)(3
Deposition rates)] separate trials.b) Question:
b) Shown below is the calculation and tabulation of the SN ratio.
Shown below is the response table. This table was created by calculating an average SN value for each factor. A sample calculation is
shown for Factor B (pressure):
The effect of this factor is then calculated by determining the range:
It can be seen that deposition rate has the largest effect on the processor yield and that temperature has the smallest effect on the
processor yield.
NOTE: The data for the following example is not real, and details pertaining to microprocessor fabrication may not be completely
accurate.
a) Solution:
 EXAMPLE 14.1.2: SESAME SEED SUFFERING
You have just produced one thousand 55 gallon drums of sesame oil for sale to your distributors. However, just before you are to ship
oil, one of your employees remembers that one of the oil barrels was temporarily used to store insecticide and is almost surely
contaminated. Unfortunately, all of the barrels look the same.
One barrel of sesame oil sells for $1000, while each assay for insecticide in food oil costs $1200 and takes 3 days. Tests for insecticide
are extremely expensive. What do you do?
Solution
Extreme multiplexing. This is similar to using a Taguchi method but optimized for very sparse systems and specific cases. For example,
instead of 1000 barrels, let us consider 8 barrels for now, one of which is contaminated. We could test each one, but that would be
highly expensive. Another solution is to mix samples from each barrel and test the mixtures.
Mix barrels 1,2,3,4 ---> Sample A
Mix barrels 1,2,5,6 ---> Sample B
Mix barrels 1,3,5,7 ---> Sample C
We claim that from testing only these three mixtures, we can determine which of the 8 barrels was contaminated. Let us consider some
possible results of these tests. We will use the following label scheme: +/-,+/-,+/- in order of A, B, C. Thus, +,-,+ indicates A and C
showed contamination but not B.
Possible Result 1: -,-,- The only barrel not mixed in was #8, so it is contaminated.
Possible Result 2: +,-,- Barrel #4 appears in A, but not in B and C. Since only A returned positive, barrel #4 was contaiminated.
Possible Result 3: -,+,- Barrel #6 appears in B, but not in A and C. Since only B returned positive, barrel #6 was contaminated.
We can see that we have 23 = 8 possible results, each of which corresponds to a particular barrel being contaminated. We leave the rest
of the cases for the reader to figure out.
Solution with 1,000 barrels: Mix samples from each barrel and test mixtures. Each mixture will consist of samples from a unique
combination of 500 barrels. Experiments required = log2(1000)=~10.
Solution with 1,000,000 barrels: Experiments required = log2(1000000)=~20.
Thus, by using extreme multiplexing, we can greatly reduce the # of experiments needed, since the # of experiments scales with log2(#
of barrels) instead of # of barrels.
 EXERCISE 14.1.1
You are producing decaffeinated coffee using supercritical carbon dioxide as a solvent. To optimize the effectiveness of caffeine
removal, you would like to test 2 different possible pressures of CO2, 3 possible temperatures, 3 ratios of CO2 to coffee beans, 3
residence times of supercritical CO2 with beans, and 2 different procedures for preroasting the beans prior to caffeine extraction. What
is the most appropriate method to design your optimization experiments?
a. Factorial design
b. Taguchi methods
c. Random design
d. None of the above. Coffee should never be decaffeinated
Answer
To perform a full factorial experiment, you’d need to account for: (2 pressures)(3 temperatures)(3 ratios)(3 times)(2 roasts) = 108
experiments! Thus, factorial design is not a practical choice: a good rule of thumb is 1-3 variables with few states for a manageable
factorial analysis. However, selecting 3 for the number of levels and consulting the array selector, we see that an L18 array will
suffice for a Taguchi analysis. 18 is a much more feasible number of experiments than 108. Finally, random design is only practical
when >50 variables are involved and few will have significant impact on the system, which is not the case here. Thus, although the
author is of the opinion that choice ‘D’ has significant merit, the correct answer is ‘B’.
 EXERCISE 14.1.2
Which of the following is inconsistent with the Taguchi philosophy of quality control?
a. Variation is the opposite of quality
b. Interactions between parameters affecting product quality are unimportant and should never be considered
c. Customer dissatisfaction is the most important measure for process variation, and should be minimized
d. A high signal to noise ratio (SN) should be present in a process to make it robust against uncontrollable factors that would cause
variation
Answer
Choice ‘A’ is another way of expressing point 3 of the Taguchi philosophy above. Similarly, choice ‘C’ is another way of expressing
the cost function, and choice ‘D’ is consistent with philosophy 2. Although it is true that the Taguchi methods do not account well
for interactions between variables, choice ‘B’ is incorrect for two reasons. 1) Pair-wise interactions between variables are accounted
for in Taguchi orthogonal arrays. 2) Negligibility of interactions between parameters is not so much a central tenet of Taguchi
philosophy as a necessary simplifying assumption. An analog is decoupling of MIMO systems. Since applying this assumption
yields good results, it is consistent with Taguchi thinking to say that makes it valid.
14.1.7: REFERENCES
Vuchkov, I.N. and Boyadjieva, L.N. Quality Improvement with Design of Experiments: A Response Surface Approach. Kluwer
Academic Publishers. Dordrecht, 2001.
http://www.ee.iitb.ac.in/~apte/CV_PRA_TAGUCHI_INTRO.htm
http://www.york.ac.uk/depts/maths/tables/orthogonal.htm
www.freequality.org/sites/www_freequality_org/documents/tools/Tagarray_files/tamatrix.htm
Roy, R. K., A Primer on the Taguchi Method, Van Nostrand Reinhold, New York 1990.
Ravella Sreenivas Rao, C. Ganesh Kumar, R. Shetty Prakasham, Phil J. Hobbs (2008) The Taguchi methodology as a statistical tool for
biotechnological applications: A critical appraisal Biotechnology Journal 3:510–523.
14.1: Design of Experiments via Taguchi Methods - Orthogonal Arrays is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by
Stephanie Fraley, John Zalewski, Mike Oom, Ben Terrien, & Ben Terrien via source content that was edited to conform to the style and standards of the
14.2: DESIGN OF EXPERIMENTS VIA FACTORIAL DESIGNS
Factorial design is an important method to determine the effects of multiple variables on a response. Traditionally, experiments are
designed to determine the effect of ONE variable upon ONE response. R.A. Fisher showed that there are advantages by combining the study
of multiple variables in the same factorial experiment. Factorial design can reduce the number of experiments one has to perform by
studying multiple factors simultaneously. Additionally, it can be used to find both main effects (from each independent factor) and
interaction effects (when both factors must be used to explain the outcome). However, factorial design can only give relative values, and to
achieve actual numerical values the math becomes difficult, as regressions (which require minimizing a sum of values) need to be
performed. Regardless, factorial design is a useful method to design experiments in both laboratory and industrial settings.
Factorial design tests all possible conditions. Because factorial design can lead to a large number of trials, which can become expensive and
time-consuming, factorial design is best used for a small number of variables with few states (1 to 3). Factorial design works well when
interactions between variables are strong and important and where every variable contributes significantly.
14.2.1: WHAT IS FACTORIAL DESIGN?

14.2.1.1: FACTORIAL DESIGN EXAMPLE
The easiest way to understand how factorial design works is to read an example. Suppose that you, a scientist working for the FDA, would
like to study and measure the probability of patients suffering from seizures after taking a new pharmaceutical drug called CureAll. CureAll
is a novel drug on the market and can cure nearly any ailment of the body. You along with your co-workers at the FDA have decided to test
two dosage levels: 5 mg and 10 mg. You are also interested in determining whether the drug side-effects differ between younger adults (age
20) and older adults (age 40). Based on the given information, you see that there are two factors: dosage and age. Factors are the main
categories to explore when determining the cause of seizures in patients. Under each of these factors, there are different levels: 5 and 10 mg
for the dosage; 20 and 40 years for age. A level is basically one of the subdivisions that make up a factor. From this information, we can see
that we have a 2 x 2 factorial design, which means that we will have 2 * 2 = 4 groups. A group is set of conditions that will make up that
particular experiment.
14.2.1.2: NULL OUTCOME

A null outcome situation is when the outcome of your experiment is the same regardless of how the levels within your experiment were
combined. From the example above, a null outcome would exist if you received the same percentage of seizures occurring in patients with
varying dose and age. The graphs below illustrate no change in the percentage of seizures for all factors, so you can conclude that the
chance of suffering from a seizure is not affected by the dosage of the drug or the age of the patient.
14.2.1.3: MAIN EFFECTS
A main effects situation is when there exists a consistent trend among the different levels of a factor. From the example above, suppose you
find that as dosage increases, the percentage of people who suffer from seizures increases as well. You also notice that age does not play a
role; both 20 and 40 year olds suffer the same percentage of seizures for a given amount of CureAll. From this information, you can
conclude that the chance of a patient suffering a seizure is minimized at lower dosages of the drug (5 mg). The second graph illustrates that
with increased drug dosage there is an increased percentage of seizures, while the first graph illustrates that with increased age there is no
change in the percentage of seizures. Both of these graphs only contain one main effect, since only dose has an effect the percentage of
seizures. Whereas, graphs three and four have two main effects, since dose and age both have an effect on the percentage of seizures.
14.2.1.4: INTERACTION EFFECTS
The interaction effects situation is the last outcome that can be detected using factorial design. From the example above, suppose you find
that 20 year olds will suffer from seizures 10% of the time when given a 5 mg CureAll pill, while 20 year olds will suffer 25% of the time
when given a 10 mg CureAll pill. When 40 year olds, however, are given a 5 mg pill or a 10 mg pill, 15% suffer from seizures at both of
these dosages. This correlation can be seen in the graphs below. There is an increasing chance of suffering from a seizure at higher doses for
20 year olds, but no difference in suffering from seizures for 40 year olds. Thus, there must be an interaction effect between the dosage of
CureAll, and the age of the patient taking the drug. When you have an interaction effect it is impossible to describe your results accurately
without mentioning both factors. You can always spot an interaction in the graphs because when there are lines that are not parallel an
interaction is present. If you observe the main effect graphs above, you will notice that all of the lines within a graph are parallel. In
contrast, for interaction effect graphs, you will see that the lines are not parallel.
14.2.2: MATHEMATICAL ANALYSIS APPROACH

In the previous section, we looked at a qualitative approach to determining the effects of different factors using factorial design. Now we are
going to shift gears and look at factorial design in a quantitative approach in order to determine how much influence the factors in an
experiment have on the outcome.
14.2.2.1: HOW TO DEAL WITH A 2N FACTORIAL DESIGN

Suppose you have two variables A and B and each have two levels a1, a2 and b1, b2. You would measure combination effects of A and B
(a1b1, a1b2, a2b1, a2b2). Since we have two factors, each of which has two levels, we say that we have a 2 x 2 or a 22 factorial design.
Typically, when performing factorial design, there will be two levels, and n different factors. Thus, the general form of factorial design is 2n.
In order to find the main effect of A , we use the following equation:
A = (a2 b1 − a1 b1 ) + (a2 b2 − a1 b2 )
Similarly, the main effect of B is given by:
B = (b2 a1 − b1 a1 ) + (b2 a2 − b1 a2 )
By the traditional experimentation, each experiment would have to be isolated separately to fully find the effect on B. This would have
resulted in 8 different experiments being performed. Note that only four experiments were required in factorial designs to solve for the eight
values in A and B. This shows how factorial design is a timesaver.
By taking the coefficients in A and B, the table below was created.
AB is found by multiplying the coefficients of axbx to get the new coefficient effect.
The additional complication is the fact that more than one trial/replication is required for accuracy, so this requires adding up each sub-
effect (e.g adding up the three trials of a1b1). By adding up the coefficient effects with the sub-effects (multiply coefficient with sub-effect),
a total factorial effect can be found. This value will determine if the factor has a significant effect on the outcome. For larger numbers, the
factor can be considered extremely important and for smaller numbers, the factor can be considered less important. The sign of the number
also has a direct correlation to the effect being positive or negative.
To get a mean factorial effect, the totals needs to be divided by 2 times the number of replicates, where a replicate is a repeated experiment.
total factorial effect
mean factorial effect =
2r
By adding a third variable (C ), the process of obtaining the coefficients becomes significantly complicated. The main factorial effect for A :
A = (a2 b1 c1 − a1 b1 c1 ) + (a2 b2 c1 − a1 b2 c1 ) + (a2 b1 c2 − a1 b1 c2 ) + (a2 b2 c2 − a1 b2 c2 )
The table of coefficients is listed below
It is clear that in order to find the total factorial effects, you would have to find the main effects of the variable and then the coefficients.
Yates Algorithm can be used to simplify the process.
14.2.2.2: YATES ALGORITHM
Frank Yates created an algorithm to easily find the total factorial effects in a 2n factorial that is easily programmable in Excel. While this
algorithm is fairly straightforward, it is also quite tedious and is limited to 2n factorial designs. Thus, modern technology has allowed for
this analysis to be done using statistical software programs through regression.
Steps:
1. In the first column, list all the individual experimental combinations
According to the yates order, such as follows for a 23 factorial design
---
+--
-+-
++-
--+
+-+
-++
+++
2. In the second column, list all the totals for each combination
3. The 1st four entries in the third column (Stage 1) are obtained by adding pairs together from the "totals list" (previous column). The next
four numbers are obtained by subtracting the top number from the bottom number of each pair.
4. The fourth column (Stage 2) is obtained in the same fashion, but this time adding and subtracting pairs from Stage 1.
5. The fifth column (Stage 3) is obtained in the same fashion, but this time adding and subtracting pairs from Stage 2.
6. Continue with Stages until reaching n, or the number of factors. This final column is the Effect Total. A positive value means a positive
correlation, and a negative values means a negative correlation. These values are all relative, however, so there is no way to compare
different effect totals from different experiments.
Ignoring the first row, look in the last stage and find the variable that has the largest relative number, then that row indicates the MAIN
TOTAL EFFECT. The Main Total Effect can be related to input variables by moving along the row and looking at the first column. If the
row in the first column is a2b1c1 then the main total effect is A. The row for a1b2c1 would be for B. The row for a2b1c2 would be for AC.
This main total effect value for each variable or variable combination will be some value that signifies the relationship between the output
and the variable. For instance, if your value is positive, then there is a positive relationship between the variable and the output (i.e. as you
increase the variable, the output increases as well). A negative value would signify a negative relationship. Notice, however, that the values
are all relative to one another. So while the largest main total effect value in one set of experiments may have a value of 128, another
experiment may have its largest main total effect value be 43. There is no way to determine if a value of 128 in one experiment has more
control over its output than a value of 43 does, but for the purposes of comparing variables within an experiment, the main total effect does
allow you to see the relative control the variables have on the output.
14.2.2.3: FACTORIAL DESIGN EXAMPLE REVISITED

Recall the example given in the previous section What is Factorial Design? In the example, there were two factors and two levels, which
gave a 22 factorial design. The Yates Algorithm can be used in order to quantitatively determine which factor affects the percentage of
seizures the most. For the use of the Yates algorithm, we will call age factor A with a1 = 20 years, and a2 = 40 years. Likewise, we will call
dosage factor B, with b1 = 5 mg, and b2 = 10 mg. The data for the three outcomes is taken from the figures given in the example, assuming
that the data given resulted from multiple trials.
Null Outcome
The following Yates algorithm table using the data for the null outcome was constructed. As seen in the table, the values of the main total
factorial effect are 0 for A, B, and AB. This proves that neither dosage or age have any effect on percentage of seizures.
Main Effect
The following Yates algorithm table using the data from the first two graphs of the main effects section was constructed. Besides the first
row in the table, the row with the largest main total factorial effect is the B row, while the main total effect for A is 0. This means that
dosage (factor B) affects the percentage of seizures, while age (factor A) has no effect, which is also what was seen graphically.
The following Yates algorithm table using the data from second two graphs of the main effects section was constructed. Besides the first
row in the table, the main total effect value was 10 for factor A and 20 for factor B. This means that both age and dosage affect percentage
seizures. However, since the value for B is larger, dosage has a larger effect on percentage of seizures than age. This is what was seen
graphically, since the graph with dosage on the horizontal axis has a slope with larger magnitude than the graph with age on the horizontal
axis.
Interaction Effect
The following Yates algorithm table was constructed using the data from the interaction effects section. Since the main total factorial effect
for AB is non-zero, there are interaction effects. This means that it is impossible to correlate the results with either one factor or another;
both factors must be taken into account.
14.2.3: CHEMICAL ENGINEERING APPLICATIONS

It should be quite clear that factorial design can be easily integrated into a chemical engineering application. Many chemical engineers face
problems at their jobs when dealing with how to determine the effects of various factors on their outputs. For example, suppose that you
have a reactor and want to study the effect of temperature, concentration and pressure on multiple outputs. In order to minimize the number
of experiments that you would have to perform, you can utilize factorial design. This will allow you to determine the effects of temperature
and pressure while saving money on performing unnecessary experiments.
A 2007 study on converting wheat straw to fuel utilized factorial design to study the effect of four factors on the composition and
susceptibility to enzyme hydrolysis of the final product. (Perez, et.al.). The table below shows the full factorial design for the study. The
four factors that were studied all had only two levels and dealt with pretreatment parameters. They were: Water temperature, residence time,
solid fraction and overpressure in the reactor. It is not necessary to understand what each of these are to understand the experimental design.
As seen in the table below, there were sixteen trials, or 2^4 experiments.
Source: Perez, et. al.
14.2.4: MINITAB DOE EXAMPLE

Minitab 15 Statistical Software is a powerful statistics program capable of performing regressions, ANOVA, control charts, DOE, and much
more. Minitab is especially useful for creating and analyzing the results for DOE studies. It is possible to create factorial, response surface,
mixture, and taguchi method DOEs in Minitab. The general method for creating factorial DOEs is discussed below.
14.2.4.1: CREATING FACTORIAL DOE

Minitab provides a simple and user-friendly method to design a table of experiments. Additionally, analysis of multiple responses (results
obtained from experimentation) to determine which parameters significantly affect the responses is easy to do with Minitab. Minitab 15
Statistical Software can be used via Virtual CAEN Labs by going to Start>All Programs>Math and Numerical Methods>Minitab
Solutions>Minitab 15 Statistical Software.
The first step is creating the DOE by specifying the number of levels (typically 2) and number of responses. To do this, go to
Stat>DOE>Factorial>Create Factorial Design as shown in the image below.
The next image is the "Create Factorial Design" options menu.
For a 2 level design, click the "2-level factorial (default generators)" radio button. Then specify the number of factors between 2 and 15.
Other designs such as Plackett-Burman or a General full factorial design can be chosen. For information about these designs, please refer to
the "Help" menu.
After the number of factors is chosen, click on the "Designs..." option to see the following menu.
In this menu, a 1/2 fraction or full factorial design can be chosen. Although the full factorial provides better resolution and is a more
complete analysis, the 1/2 fraction requires half the number of runs as the full factorial design. In lack of time or to get a general idea of the
relationships, the 1/2 fraction design is a good choice. Additionally, the number of center points per block, number of replicates for corner
points, and number of blocks can be chosen in this menu. Consult the "Help" menu for details about these options. Click "Ok" once the type
of design has been chosen.
Once the design has been chosen, the "Factors...", "Options..." and "Results..." buttons become active in the "Create Factorial Designs"
option menu. Click on "Factors..." button to see the following menu.
The above image is for a 4 factor design. Factors A - D can be renamed to represent the actual factors of the system. The factors can be
numeric or text. Additionally, a low and high value are initially listed as -1 and 1, where -1 is the low and 1 is the high value. The low and
high levels for each factor can be changed to their actual values in this menu. Click "OK" once this is completed.
The necessary steps for creating the DOE are complete, but other options for "Results..." and "Options..." can be specified. The menus for
"Results..." and "Options..." are shown below.
In the main "Create Factorial Design" menu, click "OK" once all specifications are complete. The following table is obtained for a 2-level, 4
factor, full factorial design. None of the levels were specified as they appear as -1 and 1 for low and high levels, respectively.
The above table contains all the conditions required for a full factorial DOE. Minitab displays the standard order and randomized run order
in columns C1 and C2, respectively. Columns A-D are the factors. The first run (as specified by the random run order) should be performed
at the low levels of A and C and the high levels of B and D. A total of 16 runs are required to complete the DOE.
14.2.4.2: MODIFYING DOE TABLE

Once a table of trials for the DOE has been created, additional modifications can be made as needed. Some typical modifications include
modifying the name of each factors, specifying the high and low level of each factor, and adding replicates to the design. To being
modifications of a current design, go to Stat>DOE>Modify Design... as seen in the figure below.
The following menu is displayed for modifying the design.
In the "Modify Design" menu, users can modify factors, replicate design, randomize design, renumber design, fold design, and add axial
points. Additionally, any changes made can be put into a new worksheet. To change the factors, click the "Modify factors" radio button and
then "Specify" to see the following options menu.
The default factors are named "A", "B", "C", and "D" and have respective high and low levels of 1 and -1. The name of the factors can be
changed by simply clicking in the box and typing a new name. Additionally, the low and high levels for each factor can be modified in this
menu. Since the high and low levels for each factor may not be known when the design is first created, it is convenient to be able to define
them later. Click "OK" after modifications are complete.
Another typical modification is adding replicates to a design. Replicates are repeats of each trial that help determine the reproducibility of
the design, thus increasing the number of trials and accuracy of the DOE. To add replicates, click the "Replicate design" radio button in the
"Modify Design" menu. The following menu will be displayed.
The only option in this menu is the number of replicates to add. The number ranges between 1 and 10. To have a total of 3 trials of each, the
user should add 2 replicates in this menu. If 4 replicates are added, there will be a total of 5 trials of each. Typically, if the same
experimentation will occur for 3 lab periods, 2 replicates will be added.
Additional modifications to the design include randomizing and renumbering the design. These are very straightforward modifications
which affect the ordering of the trials. For information about the "Fold design" and "Add axial points", consult the "Help" menu.
14.2.4.3: ANALYZING DOE RESULTS
After the complete DOE study has been performed, Minitab can be used to analyze the effect of experimental results (referred to as
responses) on the factors specified in the design. The first step in analyzing the results is entering the responses into the DOE table. This is
done much like adding data into an Excel data sheet. In the columns to the right of the last factor, enter each response as seen in the figure
below.
The above figure contains three response columns. The names of each response can be changed by clicking on the column name and
entering the desired name. In the figure, the area selected in black is where the responses will be inputted. For instance, if the purity, yield,
and residual amount of catalyst was measured in the DOE study, the values of these for each trial would be entered in the columns.
Once the responses are entered, statistical analysis on the data can be performed. Go to Stat>DOE>Factorial>Analyze Factorial Design... as
seen in the following image.
The menu that appears for analyzing factorial design is shown below.
In the "Analyze Factorial Design" menu, the responses are shown on the left of the screen. The first step is to choose the responses to be
analyzed. All of the responses can be chosen at once or individually. To choose them, click (or click and drag to select many) and then click
"Select" to add them into the "Responses:" section as seen below.
The next step is selecting which terms will be analyzed for the responses. To do this, click on "Terms..." and the following menu will
appear.
The types of interactions between factors are chosen in this menu. For a first order model which excludes all factor-to-factor interactions,
"1" should be chosen from the drop-down menu for "Include terms in the model up through order:". To include higher order terms and
account for factor interactions, choose 2, 3, or 4 from the drop-down menu. Unless significant factor-to-factor interactions are expected, it is
recommended to use a first order model which is a linear approximation.
Once the terms have been chosen, the next step is determining which graphs should be created. The types of graphs can be selected by
clicking on "Graphs..." in the main "Analyze Factorial Design" menu.
In the Graphs menu shown above, the three effects plots for "Normal", "Half Normal", and "Pareto" were selected. These plots are different
ways to present the statistical results of the analysis. Examples of these plots can be found in the Minitab Example for Centrifugal Contactor
Analysis. The alpha value, which determines the limit of statistical significance, can be chosen in this menu also. Typically, the alpha value
is 0.05. The last type of plots that can be chosen is residual plots. A common one to select is "Residuals versus fits" which shows how the
variance between the predicted values from the model and the actual values.
The final option that must be specified is results. Click "Results..." from the "Analyze Factorial Design" menu to see the following screen.
In this menu, select all of the "Available Terms" and click the ">>" button to move them to the "Selected Terms". This will ensure that all
the terms will be included in the analysis. Another feature that can be selected from this menu is to display the "Coefficients and ANOVA
table" for the DOE study.
Other options can be selected from the "Analyze Factorial Design" menu such as "Covariates...", "Prediction...", "Storage...", and
"Weights...". Consult the "Help" menu for descriptions of the other options. Once all desired changes have been made, click "OK" to
perform the analysis. All of the plots will pop-up on the screen and a text file of the results will be generated in the session file.
14.2.4.4: MINITAB EXAMPLE FOR CENTRIFUGAL CONTACTOR ANALYSIS

Centrifugal Contactors, also known as Podbielniak (POD) centrifugal contactors, are used to purify a contaminated stream by counter-
current, liquid-liquid extraction. Two immiscible fluids with different specific gravities are contacted counter-currently and the solute from
the dirty stream is extracted by the clean stream. A common use for PODs methanol removal from biodiesel by contacting the stream with
water. The amount of methanol remaining in the biodiesel (wt% MeOH) after the purification and the number of theoretical stages (No.
Theor. Stages) obtained depend on the operating conditions of the POD. The four main operating parameters of the POD are rotational
speed (RPM), ratio of biodiesel to water (Ratio), total flow rate of biodiesel and water (Flow Rate), and pressure (Pressure). A DOE study
has been performed to determine the effect of the four operating conditions on the responses of wt% MeOH in biodiesel and number of
theoretical stages achieved. (NOTE: The actual data for this example was made-up)
A 4-factor, 2-level DOE study was created using Minitab. Because experiments from the POD are time consuming, a half fraction design of
8 trial was used. The figure below contains the table of trials for the DOE.
After all the trials were performed, the wt% methanol remaining in the biodiesel and number of theoretical stages achieved were calculated.
The figure below contains the DOE table of trials including the two responses.
Analysis was performed on the DOE study to determine the effects of each factor on the responses. Only first order terms were included in
the analysis to create a linear model. Pareto charts for both wt% MeOH in biodiesel and number of theoretical stages are shown below.
The Pareto charts show which factors have statistically significant effects on the responses. As seen in the above plots, RPM has significant
effects for both responses and pressure has a statistically significant effect on wt% methanol in biodiesel. Neither flow rate or ratio have
statistically significant effects on either response. The Pareto charts are bar charts which allow users to easily see which factors have
significant effects.
Half Normal Plots for wt% methanol in biodiesel and number of theoretical stages are shown below.
Like Pareto plots, Half Normal plots show which factors have significant effects on the responses. The factors that have significant effects
are shown in red and the ones without significant effects are shown in black. The further a factor is from the blue line, the more significant
effect it has on the corresponding response. For wt% methanol in biodiesel, RPM is further from the blue line than pressure, which indicates
that RPM has a more significant effect on wt% methanol in biodiesel than pressure does.
The final plot created is the Normal Effect Plot. The Normal Plot is similar to the Half Normal plot in design. However, the Normal Plot
displays whether the effect of the factor is positive or negative on the response. The Normal Plots for the responses are shown below.
As seen above, RPM is shown with a positive effect for number of theoretical stages, but a negative effect for wt% methanol in biodiesel. A
positive effect means that as RPM increases, the number of theoretical stages increases. Whereas a negative effect indicates that as RPM
increases, the wt% methanol in biodiesel decreases. Fortunately for operation with the POD, these are desired results. When choosing
operating conditions for the POD, RPM should be maximized to minimize the residual methanol in biodiesel and maximize the number of
theoretical stages achieved.
In addition to the above effects plots, Minitab calculates the coefficients and constants for response equations. The response equations can
be used as models for predicting responses at different operating conditions (factors). The coefficients and constants for wt% methanol in
biodiesel and number of theoretical stages are shown below.
Since this is a first order, linear model, the coefficients can be combined with the operating parameters to determine equations. The
equations from this model are shown below.
These equations can be used as a predictive model to determine wt% methanol in biodiesel and number of theoretical stages achieved at
different operating conditions without actually performing the experiments. However, the limits of the model should be tested before the
model is used to predict responses at many different operating conditions.
 EXAMPLE 14.2.1
You have been employed by SuperGym, a local personal training gym, who want an engineer's perspective on how to offer the best
plans to their clients. SuperGym currently categorizes her clients into 4 body types to help plan for the best possible program.
Type 1 - Very healthy
Type 2 - Needs tone
Type 3 - Needs strength
Type 4 - Needs tone and strength
In addition, SuperGym offers 4 different workout plans, A through D, none of which are directly catered to any of the different types.
Create an experimental factorial design that could be used to test the effects of the different workout plans on the different types of
people at the gym.
Solution
In order to solve this problem, we need to determine how many different experiments would need to be performed. In order to solve
this, we can see that we have two different factors, body type and workout plan. For each factor, there exist four different levels. Thus,
we have a 42 factorial design, which gives us 16 different experimental groups. Creating a table of all of the different groups, we arrive
at the following factorial design:
Solution
A1 B1 C1 D1
A2 B2 C2 D2
A3 B3 C3 D3
A4 B4 C4 D4
Where A-D is the workout plan and 1-4 is the types
 EXAMPLE 14.2.2
Suppose that you are looking to study the effects of hours slept (A), hours spent with significant other (B), and hours spent studying (C)
on a students exam scores. You are given the following table that relates the combination of these factors and the students scores over
the course of a semester. Use the Yates method in order to determine the effect each variable on the students performance in the course.
Given Information
1 17 24 19 21 22 28 25 24
2 18.5 21 20 19 26 22 27 19
3 16.5 22.5 22 25 24 26 21 20
Total 52 67.5 61 65 72 76 73 63
Solution
Using the approach introduced earlier in this article, we arrive at the following Yates solution.
Stage Main Total
Combination Total 1 2 3 Factorial Effect
a1b1c1 52 119.5 245.5 529.9 Doesn't matter
a2b1c1 67.5 126 284 13.5 A
a1b2c1 61 148 19.5 -5.5 B
a2b2c1 65 136 -6 -25.5 AB
a1b1c2 72 15.5 6.5 38.5 C
a2b1c2 76 4 -12 -25.5 AC
a1b2c2 73 4 -11.5 -18.5 BC
a2b2c2 63 -10 -14 -2.5 ABC
From this table, we can see that there is positive correlation for factors A and C, meaning that more sleep and more studying leads to a
better test grade in the class. Factor B, however, has a negative effect, which means that spending time with your significant other leads
to a worse test score. The lesson here, therefore, is to spend more time sleeping and studying, and less time with your boyfriend or
girlfriend.
 EXAMPLE 14.2.3
Your mom is growing a garden for the state fair and has done some experiments to find the ideal growing condition for her vegetables.
She asks you for help interpreting the results and shows you the following data:
Make plots to determine the main or interaction effects of each factor.
Solution
Here is the plot you should have gotten for the given data.
From this one can see that there is an interaction effect since the lines cross. One cannot discuss the results without speaking about both
the type of fertilizer and the amount of water used. Using fertilizer A and 500 mL of water resulted in the largest plant, while fertilizer
A and 350 mL gave the smallest plant. Fertilizer B and 350 mL gave the second largest plant, and fertilizer B and 500 mL gave the
second smallest plant. There is clearly an interaction due to the amount of water used and the fertilizer present. Perhaps each fertilizer is
most effective with a certain amount of water. In any case, your mom has to consider both the fertilizer type and amount of water
provided to the plants when determining the proper growing conditions.
 EXERCISE 14.2.1
Which of the following is not an advantage of the use of factorial design over one factor design?
a. More time efficient
b. Provides how each factor effects the response
c. Does not require explicit testing
d. Does not require regression
Answer
TBA
 EXERCISE 14.2.2
In a 22 factorial design experiment, a total main effect value of -5 is obtained. This means that
a. there is a relative positive correlation between the two factors
b. there is no correlation between the two factors
c. there is a relative negative correlation between the two factors
d. there is either a positive or negative relative correlation between the two factors
Answer
TBA
14.2.5: REFERENCES
Box, George E.P., et. al. "Statistics for Engineers: An Introduction to Design, Data Analysis, and Model Building." New York: John
Wiley & Sons.
Trochim, William M.K. 2006. "Factorial Designs." Research Methods Knowledge Base.
<http://www.socialresearchmethods.net/kb/expfact.htm>
Perez, Jose A., et. al. "Effect of process variables on liquid hot water pretreatment of wheat straw for bioconversion to fuel-ethanol in a
batch reactor." Journal of Chemical Technology & Biotechnology. Volume 82, Issue 10, Pages 929-938. Published Online Sep 3, 2007.
14.2: Design of experiments via factorial designs is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Jocelyn Anleitner,
Stephanie Combs, Diane Feldkamp, Heeral Sheth, Jason Bourgeois, Michael Kravchenko, Nicholas Parsons, Andrew Wang, & Andrew Wang via source
14.3: DESIGN OF EXPERIMENTS VIA RANDOM DESIGN
Random design is an approach to designing experiments. As the name implies, random experimental design involves randomly assigning
experimental conditions. However, numbers should not be picked without any thought. This type of experimental design is surprisingly
powerful and often results in a high probability to create a near optimal design.
The simplified steps for random design include the following:
1. Choose a number of experiments to run (NOTE: This may be tricky to pick a number because it is dependent upon the amount of signal
recovery you want.)
2. Assign to each variable a state based on a uniform sample. For instance, if there are 5 states, each state has a probability of 20%.
Random designs typically work well for large systems with many variables, 50 or more. There should be few interactions between variables
and very few variables that contribute significantly. Random design does not work very well with relatively smaller systems. Generally
speaking, Taguchi and random designs often perform better than factorial designs depending on size and assumptions. When choosing the
design for an experiment, it is important to determine an efficient design that helps optimize the process and determines factors that
influence variability.
There is more than one type of random design, randomized block design and completely randomized design. Randomized block design
involves blocking, which is arranging experimental units into groups so they have a common similarity. The blocking factor is usually not a
primary source of variability. An example of a blocking factor may include eye color of a patient, so if this variability source is controlled,
greater precision is achieved. Completely randomized design is where the groups are chosen at random.
In various technological fields, it is important to design experiments where a limited number of experiments is required. Random design is
practical for many design applications. Extensive mathematical theory has been used to explore random experimental design. Examples of
random design include areas of data compression and medical imaging. The research conducted to support the practical application of
random design can be found at <http://groups.csail.mit.edu/drl/journal_club/papers/CS2-Candes-Romberg-05.pdf>.
Other research has been conducted recently on random design, and more information can be found at:
http://ieeexplore.ieee.org/xpls/abs_all.jsp?arnumber=1614066
14.3.2: COMPLETELY RANDOMIZED DESIGN (CRD)

14.3.2.1: DESCRIPTION OF DESIGN
Completely randomized design (CRD) is the simplest type of design to use. The most important requirement for use of this design is
homogeneity of experimental units.
14.3.2.2: PROCEDURE FOR RANDOMIZATION

1. Assign treatments to experimental units completely at random.
2. Verify that every experimental unit has the same probability of receiving any treatment.
3. Perform randomization by using a random number table, computer, program, etc.
14.3.2.3: EXAMPLE OF CRD

If you have 4 treatments (I, II, III, IV) and 5 replicates, how many experimental units do you have?
{I} {IV} {III} {II} {II} {III} {III} {II} {I} {III} {I} {IV} {III} {IV} {I} {IV} {II} {I} {II} {IV} =20 randomized experimental units
14.3.3: RANDOMIZED BLOCK DESIGN (RBD)

14.3.3.1: DESCRIPTION OF DESIGN
Randomized block design (RBD) takes advantage of grouping similar experimental units into blocks or replicates. One requirement of RBd
is that the blocks of experimental units be as uniform as possible. The reason for grouping experimental units is so that the observed
differences between treatments will be largely due to “true” differences between treatments and not random occurrences or chance.
14.3.3.2: PROCEDURE FOR RANDOMIZATION

1. Randomize each replicate separately.
2. Verify that each treatment has the same probability of being assigned to a given experimental unit within a replicate.
3. Have each treatment appear at least once per replicate.
14.3.3.3: ADVANTAGES OF RBD
1. Generally more precise than the CRD.
2. Some treatments may be replicated more times than others.
3. Missing plots are easily estimated.
4. Whole treatments or entire replicates may be deleted from the analysis.
14.3: Design of Experiments via Random Design is shared under a CC BY 3.0 license and was authored, remixed, and/or curated by Peter Woolf et al. via
14.4: SUMMARY- SUMMARY ON CONTROL ARCHITECTURES’ PHILOSOPHIES,
ADVANTAGES, AND DISADVANTAGES.
14.4.1: SUMMARY ON CONTROL ARCHITECTURES
A summary of the philosophies, advantages, and disadvantages of the different control architectures is shown below in Table 1.
Table 1. Control architectures’ philosophy, advantages, and disadvantages.
Philosophy Advantages Disadvantages
simple to design connects for error after they have already occurred
Feedback Control adjusts errors as they occur
no process model needed usually only takes input from one sensor
corrects potential errors before they occur

Feed infinitely accurate models are needed
Control connects errors they occur Ideally can produce a perfect control where
Forward infinitely accurate measurements are needed
there is never any offset
The flow you control will never truly measured and

Controls connects two flow to maintain a constant Generate defined ratio
Ratio ratio model is a assumed to be accurate
ratio Simple Model
does not take into consideration of varying pressure
Responds quickly to high frequency changes can oe very complex model

Controls set points of sensors by controlling other
Cascade Incorporate multiple sensor readings tuning model is difficult due to multiple set points and
sensors to integrate multiple information
together (incidence graphs) parameters
pick the right combination of controls to

even more complex model than cascade alone
Control integrating combinations of control accomplish main control in zo objective
Mixed tuning model is even more complex than cascade alone,
architectures incorporate any number of sensors in a
Zeigler Nichols tuning won't work by itself
logical manner
Liquid level and liquid pressure control are fundamental aspects of many control processes because of the reliance on the stability and
convenience of liquid usage. The following article will introduce the main concepts behind liquid pressure control, liquid level control,
examining the loop characteristics, response, tuning and limitations. Two models are included to illustrate the application of a liquid
pressure control loop and a level control loop in idealized systems.
14.4.3: PRESSURE CONTROL BASICS

Like its counterparts of temperature, level and flow, pressure is one of the most common process variables. Pressure is a key process
variable as it provides a critical condition for processes such as any chemical reaction, extrusion, boiling, distillation, air conditioning and
vacuuming. Poor pressure control may likely lead to problems associated with quality, safety and productivity. For example, the presence of
high-pressure conditions in a sealed vessel may result in an explosion and thus it is imperative that pressure be kept under good control
within its safety limits.
Liquid pressure control is one of the easier control loops in the sense that it is shares many of the same characteristics of a common flow
loop. The aim of the liquid pressure loop is to control the pressure at the needed set point pressure by controlling the flow with changing
process needs. Because dealing with a liquid, the flow present is that of an incompressible fluid and as a result the pressure changes very
quickly leading to a fast-responding process with small dead time and capacitance. This process behaves as a fixed restriction where the
change in pressure is a function of the flow through the process. Illustrated below is a diagram with liquid pressure control on the pipe
leading to the process.
There are a few additional considerations for the aforementioned liquid pressure loop. Firstly the controller can be proportional plus integral
(PI) or integral only (I-only) if Kc < 2 otherwise a proportional only (P-only) controller is to be used. This controller is tuned similarly to the
flow controller. Another consideration is that process gain is not constant, a square-root extractor or the highest loop gain should be used in
tuning the controller. This highest loop gain is employed to prevent the process loop from ever becoming unstable. Finally, this liquid
pressure loop is noisy like its flow loop counterpart and as a result it is recommended that derivative action in the controller not be used.
If you wish to see the various issues associated with pressure controllers please refer to previous wiki articles.
14.4.4: LEVEL CONTROL BASICS

Controlling the level of liquid in a tank is one of the most basic requirements for a chemical process to run effectively. Most chemical
engineering processes require some sort of liquid to be held for usage, and the most convenient way of ensuring the proper amount of liquid
is available is through level control. A specific example of a process that requires level control is seen around the world. Water towers must
have a designated amount of water within them to ensure sufficient water pressure to surrounding neighborhoods and hence liquid level
control (LLC) is used.
Fortunately Liquid level control is one of the easier control loops available. The systems are fairly simple, a sensor reads the level of liquid
within a tank, which then is conveyed to the controller. Subsequently, the controller changes a valve to increase or decrease the flowrate into
or out of the tank, depending on the required action and controlled valve placement. Below is a figure with LLC on the pipe leaving the
tank.
The ease of LLC also lies in its large capacitance and nominal dead time. Usually hold-up times vary from 5-15 minutes. Problems may
arise when signal noise becomes a factor, as it usually does with level controllers. Refer to previous wiki articles to the issues associated
with level controllers(), however it is important to note that a measurement of weight instead of directly measuring level can be a effective
way of eliminating noise.
14.4.4.1: P-ONLY CONTROLLERS
Oftentimes using a P-only controller is the best way to ensure proper level control, in most cases only a small percent error will result, and it
reduces problems associated with noise. P-only controllers should only be used when gain is very small and the tank has a large capacity.
This will be the basis for the rest of the text.
A P-only controller works off the following principle of control:
Output = Gain × Error + Bias
The output directly effects a valve to control the flowrate into or out of the tank. Refer to the previous wiki articles to learn more about P-
only control.
14.4.4.2: LEVEL MEASUREMENT NOISE

When controlling liquid levels in drums or vessels, an important aspect to consider is that there may be noise in the level measurement due
to disturbances such surface turbulence, boiling of the liquid, or agitation. Due to the existence of this noise, using a derivative action
controller may not be appropriate. However, if this type of controller action is used, there are various methods that may be employed to
minimize the noise.
Method 1: Use a displacer in a stilling well
Advantage: Filters high frequency noise due to turbulence in the tank
Disadvantage: Bobbing level due to the low-frequency movement of tank liquid into the well resulting from the formation of a U-
tube between tank and well
Method 2: Use an ultrasonic level measurement with electronic filtering of the signal
Advantage: Works well when the level response period is much lower than the noise frequency
Method 3: Use a tank weighing method
In the this method, a loading device is placed beneath the tank supports to measure the mass of the tank. Output weight values are averaged
by a transmitter which are then sent to a converter which outputs the corresponding level to the controller.
Advantage: Effectively eliminates noise in the level measurement due to turbulence
14.4.5: MODELS
The following are Excel models of Pressure and Level control in simple systems. Proportional only control is used primarily for these
models. Controller noise has been eliminated for sake of simplicity.
14.4.5.1: LIQUID PRESSURE CONTROL MODEL
The model at hand illustrates both the feed backwards system of the liquid pressure in a pipline of a process and its process gain
relationships. This liquid pressure control loop is regulated by the position of the valve on the pipeline leading to the process. A P-only
control is used here to change the valve position resulting in the need response in attaining the set point value. The following are any
assumptions made and the equations used in modeling this process. An assumption is made that the process behaves like a fixed restriction
such as an orifice plate whose Dp is a function of flow through the process. Another assumption that should be made is that the valve
responds linearly to the flow-rate through it. In determining the process gain the following equations where used.
P = Δp + P0
where P is the downstream pressure at zero flow

0
2
F
Δp =
2
R
where R is the process flow resistance. As a result we obtain:

2
F
P = + P0
R2
The gain of the process is then determined by the following expression:

dP 2F
Kp = =
2
dF R
A model for the feed backwards system for liquid pressure control introduced in this article is very complicated and might possibly be out
of the scope of our discussion. A possible alternative model would be very similar to the liquid level model presented below with the head
in the tank determining the liquid pressure in the pipeline leaving the reactor.
14.4.5.2: LIQUID LEVEL CONTROL MODEL

The following model represents a feed backward system of a liquid holding tank. The control loop is regulated by the exiting valve position
on the pipe leaving the tank. A maximal flowrate out is assumed and the height of water in the tank does not have any effect on the flowrate
out of the tank. This assumption can be made in tanks which are sufficiently wider than the pipe exiting operating with little liquid level
change, as often is the case. Another assumption made is the linear response of the valve to flowrate out.
The pipe entering the tank is set by the user at a fixed flowrate. The P-only control is used to change the exiting valve position in order to
cause the desired response to obtain the setpoint value. The P control follows the equation listed above and the bias is designated by the
point at which the valve is open to allow a steady state to occur. Gain in Level control systems is usually low, however this can be changed
by the user as well. The setpoint for the tank is determined by the user, as are the entering liquid flowrates and the maximal exiting flowrate.
The user can input a set point change in order to view the corresponding response in the controller.
 EXAMPLE 14.4.1
Using the Liquid Level control Excel Model, what do you find when you change the level set point at a given point in time?
What happens if you change the gain to an large number?
Solution
When you change the level set point, you should notice a corresponding increase or decrease in the level as the P-control kicks in and
beings to compensate for the error. The rate at which this set point is obtained is determined by the controller gain. At high controller
gains the system responds extremely rapidly in an unreal fashion, this is why controller gain is usually small in large capacity tanks
with flow control.
The above figure changes the setpoint to 100 @ t=100
The above figure uses a very high controller gain. Note the unreal response.
 EXAMPLE 14.4.2
What is the primary difference between the model for the process gain of the pressure flow loop introduced earlier in the article and a
typical head flow response curve?
Solution
The response of pressure to flow is shaped similarly to the response of the head flow device and as a result the rules that govern the
liquid pressure process gain model are in essence the same for the head flow device response. The only difference between the two
response curves is that whereas the pressure varies from Po to 100 per cent for the process gain curve for the pressure flow loop, the
pressure varies from 0 to 100 percent for the head flow device.
Below is the response for the pressure loop followed by that of head flow device.
Figure 14.4.1: esponse for the pressure loop followed by that of head flow device.
14.4.5.3: MULTIPLE CHOICE QUESTION 1
When should P-only controllers be used in Liquid Level control?
a. Whenever, because it's really easy.
b. Never because it is too simple.
c. Sometimes, when the gain is small and the tank capacity is large.
d. None of the above
14.4.5.4: MULTIPLE CHOICE QUESTION 2

Various methods may be used to minimize the noise in a level measurement when using which type of controller action:
a. None, you can never minimize noise in a level measurement
b. Integral
c. Derivative
d. PI-control
14.4: Summary- Summary on Control Architectures’ philosophies, advantages, and disadvantages. is shared under a CC BY 3.0 license and was authored,
remixed, and/or curated by Jennifersue Bowker, George Cater, Kibwe John Date, & Kibwe John Date via source content that was edited to conform to the
style and standards of the LibreTexts platform; a detailed edit history is available upon request.
INDEX
A Bourdon tubes E
acid 3.3: Pressure Sensors eigenvalues
3.7: pH and Viscosity Sensors brown noise 10.3: Eigenvalues and Eigenvectors
activity coefficient 2.5: Noise Modeling - White, Pink, and Brown Noise, eigenvectors
Pops and Crackles
3.7: pH and Viscosity Sensors 10.3: Eigenvalues and Eigenvectors
butterfly valves
Analysis of variance (ANOVA) 3.9: Valves - Types, Kinds, and Selection
endothermic reactors
13.12: Factor analysis and ANOVA 11.7: Common control architectures and model for
reactors
attractors C
exothermic reactors
10.5: Phase Plane Analysis - Attractors, Spirals, and completely randomized design (CRD)
Limit cycles 11.7: Common control architectures and model for
14.3: Design of Experiments via Random Design reactors
B conditional probability
13.4: Bayes Rule, Conditional Probability and F
ball valves Independence factorial design
control chart 14.2: Design of experiments via factorial designs
base 13.2: SPC- Basic Control Charts- Theory and
3.7: pH and Viscosity Sensors Construction, Sample Size, X-Bar, R charts, S charts
Fisher's exact
13.9: Discrete Distributions - Hypergeometric, Binomial,
Bayes factor correlation and Poisson
13.5: Bayesian Network Theory 13.13: Correlation and Mutual Information fixed points
Bayes rule correlation coefficient 10.1: Finding fixed points in ODEs and Boolean models
13.4: Bayes Rule, Conditional Probability and 13.13: Correlation and Mutual Information
Independence
flow
crackling (noise) 3.5: Flow Sensors
Bayes' Theorem
2.5: Noise Modeling - White, Pink, and Brown Noise, flow sensors
13.4: Bayes Rule, Conditional Probability and Pops and Crackles
Independence 3.5: Flow Sensors
CSTR
Bayesian Information Criterion (BIC) 9.4: PID tuning via Frequency Responses with Bode
frequency based noise
13.5: Bayesian Network Theory Plots 2.5: Noise Modeling - White, Pink, and Brown Noise,
Pops and Crackles
Bayesian network theory cumulative distribution function (CDF)
13.5: Bayesian Network Theory 13.10: Multinomial Distributions G
Bayesian networks Gaussian noise
13.6: Learning and analyzing Bayesian networks with
D
2.5: Noise Modeling - White, Pink, and Brown Noise,
Genie dead time Pops and Crackles
belief network 9.2: P, I, D, PI, PD, and PID control
13.5: Bayesian Network Theory degrees of freedom analysis H
Bergman minimal model 4.3: Piping and Instrumentation Diagrams - Location of heat exchanger
Controls and Standard Control Structures 6.6: ODE and Excel model of a Heat Exchanger
binomial distribution determinant hidden Markov models
10.3: Eigenvalues and Eigenvectors 13.7: Occasionally Dishonest Casino? Markov Chains
13.9: Discrete Distributions - Hypergeometric, Binomial,
and Poisson Dirac delta function and Hidden Markov Models
blend stations 7.1: Dirac delta (impulse) function humidity sensors

11.4: Ratio Control directed acyclic graph 3.8: Miscellaneous Sensors
blue noise 13.5: Bayesian Network Theory hypergeometric distribution

2.5: Noise Modeling - White, Pink, and Brown Noise, distillation column 13.9: Discrete Distributions - Hypergeometric, Binomial,
Pops and Crackles and Poisson
6.5: ODE and Excel model of a Simple Distillation
Bode plot Column
I
9.4: PID tuning via Frequency Responses with Bode doppler meters
Plots incidence graphs
3.5: Flow Sensors
boolean 2.3: Incidence Graphs -Interpretations, Consistency, and
drift noise Inconsistency
5.1: Boolean Models - Truth Tables and State Transition
Diagrams 2.5: Noise Modeling - White, Pink, and Brown Noise, independence
Pops and Crackles
Boolean modeling 13.4: Bayes Rule, Conditional Probability and
Dynamic Bayesian Network (DBN) Independence
13.5: Bayesian Network Theory inherent flow characteristic
Boolean network
5.1: Boolean Models - Truth Tables and State Transition
Diagrams
1 https://eng.libretexts.org/@go/page/28370
J nullcline Proton Exchange Membrane (PEM) fuel cell
joint probability 10.5: Phase Plane Analysis - Attractors, Spirals, and 5.1: Boolean Models - Truth Tables and State Transition
Limit cycles Diagrams
13.4: Bayes Rule, Conditional Probability and
Independence purple noise
O
2.5: Noise Modeling - White, Pink, and Brown Noise,
L ODE modeling Pops and Crackles
level sensors 10.1: Finding fixed points in ODEs and Boolean models
Q
3.4: Level Sensors orifice meter
3.5: Flow Sensors
queuing theory
limit cycle 13.7: Occasionally Dishonest Casino? Markov Chains
10.5: Phase Plane Analysis - Attractors, Spirals, and orthogonal arrays and Hidden Markov Models
Limit cycles 14.1: Design of Experiments via Taguchi Methods -
linear optimization Orthogonal Arrays R
8.2: Linear Optimization overspecification random design
linearizing nonlinear ordinary differential 4.4: Piping and Instrumentation Diagrams - Standard 14.3: Design of Experiments via Random Design
Pitfalls
equations randomized block design (RBD)
10.2: Linearizing ODEs P 14.3: Design of Experiments via Random Design
logical equivalence PageRank ratio control
5.1: Boolean Models - Truth Tables and State Transition 13.7: Occasionally Dishonest Casino? Markov Chains 11.4: Ratio Control
Diagrams and Hidden Markov Models
ratio control architecture
logical inversion pH 11.4: Ratio Control
5.1: Boolean Models - Truth Tables and State Transition 3.7: pH and Viscosity Sensors
Diagrams reboiler
photometric analysis 6.5: ODE and Excel model of a Simple Distillation
M 3.6: Composition Sensors Column
MIMO PID relative gain array (RGA)

12.5: Understanding MIMO Control Through Two Tanks 9.6: PID Downsides and Solutions 12.2: MIMO control using RGA
Interaction
PID controller resistance temperature detector (RTD)
MIMO (multiple input, multiple output) 9.2: P, I, D, PI, PD, and PID control 3.2: Temperature Sensors
pink noise root locus plots
2.5: Noise Modeling - White, Pink, and Brown Noise, 10.6: Root Locus Plots - Effect of Tuning
MISO (multiple input, single output) Pops and Crackles
rotameter
Piping and Instrumentation Diagrams
3.5: Flow Sensors
Model Predictive Control (P&ID)
12.3: MIMO using Model Predictive Control Routh array
4.2: Piping and Instrumentation Diagram Standard
Notation 10.7: Routh Stability - Ranges of Parameter Values that
Monod equation are Stable
4.3: Piping and Instrumentation Diagrams - Location of
6.3: Bacterial Chemostat Controls and Standard Control Structures
4.4: Piping and Instrumentation Diagrams - Standard
Routh test
monotonicity Pitfalls 10.7: Routh Stability - Ranges of Parameter Values that
2.3: Incidence Graphs -Interpretations, Consistency, and are Stable
Inconsistency
plinko
Routh’s theorem
Moser equation 10.4: Using Eigenvalues and Eigenvectors to Find
plug flow reactors Stability and Solve ODEs
multinomial distributions 11.7: Common control architectures and model for S
13.10: Multinomial Distributions reactors
Poisson distribution separatrix

Multiple Input Multiple Output
10.5: Phase Plane Analysis - Attractors, Spirals, and
12.5: Understanding MIMO Control Through Two Tanks 13.9: Discrete Distributions - Hypergeometric, Binomial, Limit cycles
Interaction and Poisson
pops (noise) Shewhart chart
N 2.5: Noise Modeling - White, Pink, and Brown Noise,
negative feedback Pops and Crackles shot noise

11.1: Feedback Control positive feedback 2.5: Noise Modeling - White, Pink, and Brown Noise,
Pops and Crackles
neural network 11.1: Feedback Control
probability density function SIMO (single input, multiple output)

Niederlinski index
process gain Single Input Single Output
12.5: Understanding MIMO Control Through Two Tanks
noise Interaction
2.5: Noise Modeling - White, Pink, and Brown Noise, proportional control Single variable Input or Single variable
Pops and Crackles 9.6: PID Downsides and Solutions Output (SISO)
SISO T U
12.5: Understanding MIMO Control Through Two Tanks Taguchi loss function ultimate gain
Interaction
14.1: Design of Experiments via Taguchi Methods - 9.3: PID Tuning via Classical Methods
SISO (single input, single output) Orthogonal Arrays
12.1: Determining if a system can be decoupled Taguchi method V
Six Sigma program 14.1: Design of Experiments via Taguchi Methods - valve coefficient
13.3: Six Sigma- What is it and what does it mean? Orthogonal Arrays
snap (noise) Taylor series
valve rangeability
2.5: Noise Modeling - White, Pink, and Brown Noise, 7.4: Taylor Series
Pops and Crackles temperature control loops
viscosity
sonic nozzle 11.6: Common Control Loops and Model for
Temperature Control 3.7: pH and Viscosity Sensors
3.5: Flow Sensors
spirals Tessier equation
W
10.5: Phase Plane Analysis - Attractors, Spirals, and 6.3: Bacterial Chemostat
Limit cycles
white noise
thermometer 2.5: Noise Modeling - White, Pink, and Brown Noise,
stationary Gaussian process 3.2: Temperature Sensors Pops and Crackles
13.14: Random sampling from a stationary Gaussian Transmitter Symbols (P&IDs)
process
4.2: Piping and Instrumentation Diagram Standard
Y
step function Notation Yates algorithm
7.1: Dirac delta (impulse) function truth tables 14.2: Design of experiments via factorial designs
surge tank 5.1: Boolean Models - Truth Tables and State Transition
6.1: Surge Tank Model Diagrams Z
zero buoyancy force level
3.4: Level Sensors

Full Eigenvectors

Uploaded by

Copyright:

Available Formats

Full Eigenvectors

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Full Eigenvectors

Uploaded by

Copyright:

Available Formats

CHEMICAL PROCESS

Peter Woolf et al.

Peter Woolf et al.

This text was compiled on 07/04/2022

3: SENSORS AND ACTUATORS

4: PIPING AND INSTRUMENTATION DIAGRAMS

7: MATHEMATICS FOR CONTROL SYSTEMS

9: PROPORTIONAL-INTEGRAL-DERIVATIVE (PID) CONTROL

10: DYNAMICAL SYSTEMS ANALYSIS

11: CONTROL ARCHITECTURES

13: STATISTICS AND PROBABILITY BACKGROUND

14: DESIGN OF EXPERIMENTS

1.1.2: PROCESS CONTROL BACKGROUND

1.1.3: THE OBJECTIVES OF CONTROL

1.1.4: DEFINITIONS AND TERMINOLOGY

1.1.5: DESIGN METHODOLOGY FOR PROCESS CONTROL

1.4.1: BHOPAL, INDIA DISASTER

1.4.2: THREE MILE ISLAND DISASTER

1.4.3: TEXAS CITY DISASTER

1.5.2: EXAMPLES OF PROCESS CONTROL FOR THE COMMON MAN

2.1.2: PREREQUISITE INFORMATION REGARDING A PROCESS

2.1.3: STEP-BY-STEP METHOD FOR DESCRIBING CONTROLS AND THEIR PURPOSE

2.1.4: ALTERNATIVE METHOD OF VERBAL MODELING

2.1.5: THE BARKEL METHOD OF VERBAL MODELING

2.1.6: COMMON ERRORS

2.1.7: WORKED OUT EXAMPLE 1

2.1.9: SAGE'S CORNER

2.2.3: THE PROCESS

2.2.5: EXAMPLE 1: BLENDER

Figure 2: Blender with controls

Monotonic Increase : If x≥y, then f(x)≥f(y)

Monotonic Decrease : If x≥y, then f(x)≤f(y)

2.3.2: INCIDENCE GRAPHS

This incidence graph indicates that:

This incidence graph indicates that:

This incidence graph indicates that:

2.3.3: CONSISTENT, INCONSISTENT, AND PARTIALLY CONSISTENT

2.3.3.1: CONSISTENT GRAPHS

Paths(s) Sign series

All pathways leading from node 1 to node 3 -

Paths(s) Sign series

2.3.3.2: INCONSISTENT GRAPHS

Paths(s) Sign series

Paths(s) Sign series

All pathways leading from node 1 to node 2 -

2.3.3.3: PARTIALLY CONSISTENT GRAPHS

Paths(s) Sign series

All pathways leading from node 1 to node 4 -

2.3.4: WORKED OUT EXAMPLE 1

2.3.5: WORKED OUT EXAMPLE 2

2.3.6: WORKED OUT EXAMPLE 3

2.3.8: CONTRIBUTORS AND ATTRIBUTIONS

2.4.1.1: LOGICAL METHODS

3) Click solve to find the solution.

2.4.1.3: NONLINEAR REGRESSION

2.4.1.3.1: ANOVA ANALYSIS

2.4.1.4: RANDOM NUMBER SAMPLING

Microsoft Excel 2007:

2.4.1.5: WORKED OUT EXAMPLE 1

3. Click the Min radio button