Drum-Boiler Control Performance

Master Thesis
Drum-boiler control performance

optimization using an observer-based
state-feedback controller within
MATLAB/Simulink environment
Ahmed Elguindy
December 11, 2013
Tutor: 1st Examiner:

Dipl.-Ing. Simon Rünzi Prof. Dr.-Ing. Kai Michels
2nd Examiner:
Prof. Dr.-Ing. Bernd Orlik
Universität Bremen
Acknowledgment
It gives me great pleasure in expressing my sincere gratitude to everyone who
have supported and contributed into making this thesis possible.
I would like first to acknowledge my direct supervisor Dipl.-Ing. Simon Rünzi
for his enthusiasm, inspiration and huge efforts to explain things clearly and
simply. His in-depth knowledge regarding the CHP plant in Munich, related to
his PhD research, was quite helpful and beneficial for my work. Furthermore I
would like to thank my examiner Prof. Dr.-Ing Kai Michels for offering me the
project which have evolved over the course of time into an interesting thesis
topic. I wish also to address their constructive criticism following initial review
of the thesis.
My appreciation for SWM Services GmbH, specially Mr. Julian Niedermeier
for his willingness to perform experiments on the plant, its priceless valuable
information contributed significantly to improve my understanding of the real
process.
I wish to acknowledge the scholarship support provided by the Katholischer
Akademischer Ausländer-Dienst (KAAD). In particular I am very grateful to
Dr. Christina Pfestroff as I do believe that my master studies in Germany
wouldn’t have been possible without her guidance when applying for the schol-
arship. I thank as well Prof. Dr.-Ing Rainer Laur, Mr. Hans Landsberg, Mr.
Raphael Nabholz and Mrs. Claudia Dillmann for their continuous follow-up
and assistance.
Lastly and most importantly, I dedicate this thesis to my parents who raised,
supported, taught and loved me throughout my entire life.
Abstract
This thesis presents the development of an observer-based state-feedback con-
troller designed using LQ and pole placement methods to optimize pressure
and water level control performance of a drum-boiler unit that belongs to a
450 MW CHP plant in Germany. The Åström-Bell nonlinear model is initially
built within MATLAB/Simulink environment, later enlarged to include the
process PID-controllers and control valves regulating mass flow rates before
being validated against data measurements with very rich excitation. The con-
cluded simulation results adopting the newly proposed control strategy shows
that the suggested multivariable control technique outperforms the existing
PID-controller in many aspects improving the control performance significantly
and yielding much tighter reference value tracking during load changes.
Keywords: drum-boiler level control; optimal control; multivariable feedback

control; power plants simulation
Contents
Contents
1. Introduction 6
2. Process modelling 8
2.1. Combined cycle process overview . . . . . . . . . . . . . . . . . 8
2.1.1. Gas turbine . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.1.2. Heat recovery steam generator . . . . . . . . . . . . . . 9
2.1.3. Steam turbine . . . . . . . . . . . . . . . . . . . . . . . 10
2.1.4. Surface condenser . . . . . . . . . . . . . . . . . . . . . 10
2.2. Steam generation process description . . . . . . . . . . . . . . . 11
2.2.1. Drum-boiler mass and energy balance . . . . . . . . . . 11
2.2.2. Drum-boiler nonlinear state equations . . . . . . . . . . 14
2.2.3. Mass flow control valve . . . . . . . . . . . . . . . . . . 15
2.2.4. Process PID-controller . . . . . . . . . . . . . . . . . . . 16
2.3. MATALB/Simulink model . . . . . . . . . . . . . . . . . . . . . 18
2.3.1. Drum-boiler model . . . . . . . . . . . . . . . . . . . . . 19
2.3.2. Control valve and actuator model . . . . . . . . . . . . . 21
2.3.3. Process PID-controller model . . . . . . . . . . . . . . . 22
3. Process analysis and validation 23

3.1. Theoretical overview . . . . . . . . . . . . . . . . . . . . . . . . 23
3.1.1. Concept of stability . . . . . . . . . . . . . . . . . . . . 23
3.1.2. Linearization . . . . . . . . . . . . . . . . . . . . . . . . 24
3.1.3. Poles and zeros . . . . . . . . . . . . . . . . . . . . . . . 24
3.2. Stability analysis . . . . . . . . . . . . . . . . . . . . . . . . . . 26
3.2.1. Linear state-space model . . . . . . . . . . . . . . . . . 26
3.2.2. I/O pole-zero plot . . . . . . . . . . . . . . . . . . . . . 27
3.3. Open loop step response . . . . . . . . . . . . . . . . . . . . . . 29
3.3.1. Change of gas turbine electrical output power . . . . . . 29
3.3.2. Change of butterfly valve position . . . . . . . . . . . . 30
3.3.3. Change of feedwater control valve position . . . . . . . . 30
3.4. Validation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
3.4.1. Assumptions . . . . . . . . . . . . . . . . . . . . . . . . 32
4
Contents
3.4.2. Comparison with measurement data . . . . . . . . . . . 33

3.5. Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . 36
4. Process optimization 38
4.1. Concept of state-feedback control . . . . . . . . . . . . . . . . . 38
4.1.1. Controllability and observability . . . . . . . . . . . . . 38
4.1.2. Observer-based control . . . . . . . . . . . . . . . . . . . 39
4.1.3. PI-based state-feedback control . . . . . . . . . . . . . . 40
4.2. Controller design methods . . . . . . . . . . . . . . . . . . . . . 41
4.2.1. Pole placement method . . . . . . . . . . . . . . . . . . 41
4.2.2. Linear-Quadratic method . . . . . . . . . . . . . . . . . 42
4.3. Observer-based state-feedback controller design . . . . . . . . . 44
4.3.1. Riccati controller . . . . . . . . . . . . . . . . . . . . . . 44
4.3.2. Luenberger observer . . . . . . . . . . . . . . . . . . . . 45
4.4. Simulation results . . . . . . . . . . . . . . . . . . . . . . . . . 45
5. Conclusion and future work 52
A. Appendix 53
A.1. Nomenclature . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
A.2. MATLAB Control System Toolbox . . . . . . . . . . . . . . . . 54
A.2.1. Linear analysis functions . . . . . . . . . . . . . . . . . . 54
A.2.2. Controller design functions . . . . . . . . . . . . . . . . 54
A.3. MATLAB script . . . . . . . . . . . . . . . . . . . . . . . . . . 55
A.3.1. Drum-boiler model . . . . . . . . . . . . . . . . . . . . . 55
A.3.2. Controller design . . . . . . . . . . . . . . . . . . . . . . 57
A.4. Heat engines . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58
A.4.1. Brayton cycle . . . . . . . . . . . . . . . . . . . . . . . . 58
A.4.2. Rankine cycle . . . . . . . . . . . . . . . . . . . . . . . . 59
A.5. Non-minimum phase systems . . . . . . . . . . . . . . . . . . . 60
A.6. Integral anti-windup control . . . . . . . . . . . . . . . . . . . . 61
A.7. Drum-boiler state equations coefficients . . . . . . . . . . . . . 63
A.8. Operator interface . . . . . . . . . . . . . . . . . . . . . . . . . 64
B. List of Figures 66
C. List of Tables 68
D. Bibliography 69
5
1. Introduction
1. Introduction
Energy market deregulation and integration of renewable energy resources
into the electrical grid have led to dramatic changes in the power industry which
escalated rapidly new challenges that have to be met by conventional power
plants. Such evolution caused a noticeable process modification regarding how
power plants operate, as they should become more flexible to fulfill their load
requirements which are more frequent nowadays. The process controllers have
to be designed in a way which can simultaneously fulfill the load demand as
soon as possible while at the same time bearing in mind safety and life span of
the plant crucial elements.
One common challenge is control of steam drum-boiler units handling supply
of the steam turbine continuously with steam at high pressure and temperature.
The controller should maintain drum pressure and water level within acceptable
ranges for all operating conditions. If the level exceeds upper limits, water
would be carried over to the superheater or the turbine leading to outage
in either of the turbine or the boiler. Surpassing lower limits would cause
overheating of the water wall tube resulting in serious tube rupture and severe
damage.
Drum level control in particular is quite tough due to the process physical
phenomena known as shrink/swell of steam bubbles under the water level which
causes the system to react with an initial inverse response known as a non-
minimum phase behaviour.
Classical control design methods using 2-element or 3-element PID-controllers
can behave fairly well to compensate such effect. However as the process is
quite complicated, dealing with several input variables to regulate each process
variable separately might end up with bad parameter tuning and poor level
performance observed during load changes, eventually leading the boiler unit
to trip or even worse cause emergency shutdown of the power plant. It is stated
that about 30% of the emergency shutdowns in French pressurized water reac-
tors (PWR) plants were caused by poor level control of a steam drum-boiler
unit [21].
An ongoing research project is taking place at the moment in collabora-
tion with Stadtwerke München GmbH - Munich City Utilities (SWM) in re-
gards with the process PID-controllers of the low pressure drum-boiler unit,
6
1. Introduction
located within the combined cycle plant GuD 2, short for Gas-und-Dampf-
Kombikraftwerk at Heizkraftwerk Süd (HKW) - combined heat and power
(CHP) facility. The main objective is drum level and pressure closed loop
performance optimization which have been reported to behave very poorly un-
der huge load changes taking place frequently following energy deregulation in
Germany.
The thesis is presented as follows, initially the complete process is briefly
introduced before being simplified to highlight the significant elements domi-
nating the steam generation process which are mainly focused on during mod-
elling procedure. Derivation of the differential equations is carried out for each
established featured element to develop a mathematical model capable of cap-
turing most of the system nonlinearities and later on suitable for model-based
control.
The model parameterized and implemented within MATLAB/Simulink en-
vironment will be subjected to a detailed analysis by examining stability, simu-
lating the model open loop response and validating the closed loop against data
measurements from the plant. The investigation concluded results will offer a
good insight into the system inner dynamics and shall inspect the model abil-
ity to catch the plant dynamical behaviour for a wide spectrum of operating
conditions.
In the end, the proposed control strategy is addressed. First, state-feedback
control concept and the numerous methods which applies it shall be briefly
described to illustrate their applicability and major difference between them.
The most convenient and suitable approach shall be employed to compute the
state-feedback and observer gain matrices. Finally, simulation results of the
process utilizing the newly designed observer-based state-feedback controller is
presented for various sequences to ensure stability of the optimized closed loop.
7
2. Process modelling
2. Process modelling
HKW Süd plant is classified as a combined cycle cogeneration plant, it can
handle concurrent production of electrical power and useful heat utilizing a
class of sustainable integrated technologies progressively being used.
Cogeneration plants reduce thermal and mechanical losses, harmful carbon
dioxide (CO2 ) emissions and more importantly increases the overall plant effi-
ciency to approximately 81% in comparison to stand alone plants which don’t
exceed 45%. The German government is planning to double its share of CHP
plants from approximately 12% to 25% by 2020, as part of the Integrated
Energy and Climate Protection Program (IECPP) [8].
GuD 2 at HKW Süd manages electrical power generation by combining both
Brayton and Rankine thermodynamic theoretical cycles (A.4) [12] [20] using
gas and steam turbines. Exhaust gas emitted from the gas turbine can be
reused as the heat source for steam production required to operate the steam
turbine, therefore more useful energy can be extracted, supplying additional
electricity to the grid.
Further energy can by even withdrawn from the low pressure steam leaving
the turbine when condensed using a heat exchanger where the low temperature
steam released can be utilized for district heating or water desalination.
In this chapter, the overall combined cycle process is being narrowed down
to draw the focus on one particular key element within the plant. The process
is further simplified in order to spotlight primarily our aim interest which is
the steam production using the low pressure drum-boiler unit along side with
its process PID-controllers.
2.1. Combined cycle process overview

GuD 2 at HKW süd plant combined cycle principle is shown in figure (2.1),
it consists of the following main elements briefly described
2 General Electric gas turbine units producing a total of 278 MW

1 Heat recovery steam generator equipped with supplementary firing
1 Alstom steam turbine unit producing additional 139 MW
1 heat-exchange surface condenser supported with an auxiliary unit
8
Low pressure water Waste heat

to atmosphere
Electricl power
Cooling water
Condenser
Steam turbine
Superheated steam
Feedwater pump
Fresh air
Electricl power
Gas turbine
Exhaust heat
Heat Recovery Steam Generator

Gas
Figure 2.1.: Combined cycle working principle
2.1.1. Gas turbine

The combined cycle starts at the gas turbine unit whose process is based on
the Brayton open cycle (A.4.1). Continuous fresh air is compressed then mixed
with the supplied natural gas before being burned inside the combustion cham-
ber at around 1124◦ C. The hot compressed air expands within the turbine
driving its blades which eventually turns the generator shaft producing elec-
trical power and the exhaust low pressure gas leaving the turbine at 535◦ C is
used as the heating source for the HRSG. Gas turbines typically have capacities
between 500 kW and 250 MW.
2.1.2. Heat recovery steam generator

HRSG acts as a heat exchanger between exhaust heat supplied from the gas
turbine and the liquid/vapour mixture circulating into finned tubes through
3 heat exchangers highlighted in figure (2.2) where additional firing can take
place if necessary. Production of high pressure steam is carried out using high
and low pressure drum-boiler units according to the following process.
1. Economizer stage Water fed by the pump supplied to the drum inlet
9
is preheated in order to reduce energy consumption.

2. Evaporator stage Due to the gravity water flows down through a
downcomer-riser closed loop producing saturated steam which flows along
the riser tubes before being collected and fed back into the drum.
3. Superheater stage The saturated steam flows through the water level
till it exits upon reaching the drum outlet. Then it is reheated one more
time producing superheated steam supplied the turbine
Figure 2.2.: Heat Recovery Steam Generator (HRSG) [14]
2.1.3. Steam turbine

The theory of operation is based on the Rankine cycle (A.4.2) where high pres-
sure and temperature superheated steam enters the turbine converting thermal
energy into rotational mechanical energy capable of moving its blades and gen-
erator shaft producing additional electricity. The steam losing most of its tem-
perature during the conversion process is collected and fed into the condenser.
Steam turbines typically have capacities between 50 kW and 250 MW.
2.1.4. Surface condenser

Also known as water-cooled shell and tube heat exchanger, it installed at the
turbine outlet handling the last phase of the combined cycle by condensing
the exhaust steam to achieve maximum attainable efficiency. Water is used to
carry off waste heat from the steam due to its availability, high specific thermal
capacity and heat transfer properties.
10
2.2. Steam generation process description

The differential equations describing dynamics associated with the drum-
boiler, feedwater and steam regulating valves with their actuators, in addition
to the existing process PID-controllers shall be explained and addressed in
details throughout the following sections.
The simplified process relevant to our analysis concerning steam generation
using the drum-boiler unit is illustrated1 in figure (2.3). Supplied inflow from
the feedwater pump is regulated using one control valve. As for the steam
flow rate leaving the drum, it can be regulated using five valves2 connected in
parallel with distinctive construction and functionalities.
1. Water tank control valve always kept opened at a predefined position
2. Bypass valve A butterfly valve handling supply of heat to city districts
3. Security valve for safety matters when the drum pressure exceeds limits
4. Steam turbine control valve feeds the steam turbine
5. Condenser control valve bypasses the steam turbine feeding directly
the condenser
1 3
Outflow - qs
Feedwater Tank
2
4 5
Inflow - qf
Downcomer-riser loop
Exhaust heat - Q
Figure 2.3.: Schematic diagram of the low pressure steam generation process
2.2.1. Drum-boiler mass and energy balance

Figure (2.4) illustrates the detailed process of steam generation within the
drum. Its complex geometry, number of riser and downcomer tubes and spe-
1
Process PID-controllers are excluded
2
In steady state only one valve is operational while the others are closed
11
cially the two phase flow modelling attempt is usually quite complicated requir-
ing typically usage of partial differential equations. In literature there exists
a lot of research papers that were devoted into developing relatively simple
physical models [2] [7] [13] [14].
In particular the well developed Åström - Bell model3 is being considered.
The majority of the system attitude can be captured through a 4th order non-
linear model by means of defining mass flow and energy balance with the help
a physical mechanism introduced under the following elementary assumptions.
Steam demand
(to downstream) Sat. steam
Drum Internal
Upper void Separation
(saturated steam) Device Mixture
from riser
Steam rises
Steam-water
Condensation
Sat.
steam
Steam-water
mixture
Feedwater Sat. water
Upper collecting header
Water boils
and flows
upward
Downcomer
Riser
Heat from
hot medium
Lower distribution header
Figure 2.4.: Schematic diagram of the downcomer-riser circulation loop [13]
Most of the system parts will be under thermal equilibrium due to their
direct contact with saturated liquid/vapour mixture. The energy stored in the
mixture is either absorbed or released quickly following drum pressure changes,
meaning that various metal parts of the system would adapt their temperatures
in the same manner.
This agrees with experimental observation which have proven that the differ-
ence between both temperatures is very small, thus a detailed representation
3
Part of an ongoing research project which started back in the early seventies
12
of the temperature distribution within the metal isn’t necessary.

Equations (2.1) presents the mass and global energy balance for the drum
in terms of feedwater qf , steam qs and heat Q̇ flow rates respectively. It
describes the drum pressure dynamical behavior quite well by simply computing
properties of liquid/vapour mixture using steam tables. Condensation of the
steam within the drum causes the coupling between the drum pressure P and
water total volume Vwt .
d
(ρs Vst + ρw Vwt ) = qf − qs
dt (2.1)
d
(ρs hs Vst + ρw hw Vwt − P Vwt + mt cp tsat ) = Q̇ + qf hf w − qs hs
dt
Distribution of steam along the riser tubes was carried out using a lumped
model which represents the energy and mass balance caused by the naturally
circulated downcomer-riser closed loop as seen in equation (2.2). The steam
mass fraction αr assumed to vary linearly from the inlet to the outlet of the
riser is characterized in response to changes in the downcomer qdc , riser qr and
heat Q̇ flow rates respectively.
d
[ρs ᾱv Vr + ρw (1 − ᾱv )Vr ] = qdc − qr
dt
d (2.2)
[ρs hs ᾱv Vr + ρw hw (1 − ᾱv )Vr − P Vr + mr Cp tsat ]
dt
= Q̇ + qdc hw − qr (hw + αr hc )
The empirical equation (2.3) resulted from various attempts to fit with the
experimental data. It defines mass balance of the steam bubbles under the
water level in terms of condensation flow qcd and steam flow through the liquid
surface qsd driven by density difference of the mixture and momentum of the
flow qr entering through the riser tubes. It can capture most of the process
dynamics by proper parameterizations of residence time of steam inside the
◦ and empirical
drum Td , the bubbles steam volume at hypothetical situation4 Vsd
coefficient β correspondingly.
d
(ρs Vsd ) = αr qr − qcd − qsd
dt (2.3)
ρs ◦
qsd = (Vsd − Vsd ) + αr qdc + αr β(qdc − qr )
Td
4
Theoretical state that assumes no condensation of steam inside the drum
13
2.2.2. Drum-boiler nonlinear state equations

To derive a state model, chosen state variables should have a good physical
interpretation. Drum pressure P is obviously chosen as it describes the total
energy of the system. The accumulation of water related to total water volume
Vwt in the system is selected since it represents the storage of mass.
Steam quality αr in the riser tubes and steam bubbles volume under the
liquid level Vsd are chosen as well to describe distribution of steam under the
water, thus estimating the level. The resulting nonlinear state-space model
would be a 4th order system whose states are x = [ P , Vwt , αr , Vsd ].
The model actuating variables are u = [ qf , qs ], the feedwater and steam flow
rates are manipulated to control primarily the drum water level and pressure
respectively, whereas the heat flow rate Q̇ is rather considered as a model input
disturbance z due to the fact that its amount is associated with the gas turbine
exhaust heat which in return corresponds to its electrical output power as we
elaborated concisely the combined cycle working principle in section (2.1). On
the contrary heat flow rate becomes a control variable in thermal plants as it
can be regulated directly by adjusting the boiler firing rate.
Arrangement of the mass and energy balance differential equations was car-
ried out in order to derive the algebraic state equations. The liquid/vapour
mixture properties time derivative in terms of the drum pressure are calculated
using the coefficients enm provided in appendix (A.7).
dP e12 Q̇ + qf (e12 hf w − e22 ) − qs (e22 − e12 hs )

= (2.4)
dt e12 e21 − e11 e22

dVwt 1 dP
= qf − qs − e11 (2.5)
dt e12 dt

dαr 1 dP
= Q̇ − αr hc qdc − e31 (2.6)
dt e33 dt

dVsd 1 ρs ◦ h f w − hw dP dαr
= (Vsd − Vsd ) − qf − e41 − e43 (2.7)
dt e44 Td hc dt dt
Equations (2.4), (2.5) rearrange the drum mass and energy balance, equation
(2.6) combines the mass and energy balance of the downcomer-riser closed loop
in a single equation and equation (2.7) considers only the mass balance of steam
bubbles under water level. The interesting feature of this model is that the
states can be grouped in the form ((P, Vwt ), αr , Vsd ), where each term can be
computed separately in a nested manner treating the system as 2nd , 3rd or 4th
order according to modelling requirements.
14
2.2.3. Mass flow control valve

The process concerned with regulation of feedwater flow rate supplied from
the pump can be simplified and highlighted as seen in figure (2.5). The flow
is computed with the aid of the nonlinear equation (2.8) essentially used by
mechanical engineers to size their valves and meet mass flow requirements.
The pressure drop ΔP across the valve would be the difference between the
feedwater pump and drum pressures, xf is its percentage opening ranging from
0% to 100% and finally Kv is the valve sizing coefficient.
p1 p2
H100
H0
t1
Q W
Figure 2.5.: Flow through control valve for liquid service [22]
The dynamics related to regulation of steam flow rate are quite complicated
where additional considerations have to be taken care of when compared to
feedwater mainly due to the difference in properties between both. One good
approximation to describe the flow rate through a control valve meeting prac-
tical needs can be achieved using equation (2.9) where P is the drum pressure,
the head loss coefficient m and the compressibility factor Z are taken into
account to distinguish between saturated and superheated steam.
√
Kvf · ρw · ΔP
qf = xf · (2.8)
3600
Kvs · Z · m
qs = x s · (2.9)
3600
Clearly the valve position value would vary according to the type of valve
being used. The inherent flow characteristic depicted in figure (2.6) highlight
the comparison between the commonly used control valve demonstrating that
mass flow rate for the same opening position and pressure drop across it is
obviously altered according to the category it belongs to.
Examining halfway opened linear, butterfly and relief valves correspondingly,
15
undoubtedly the butterfly valve would supply approximately one-third of the

total amount provided using the linear valve while the relief valve employed for
safety precautions would grant roughly twice the flow afforded by the linear
valve.
The actuators used to operate the control valve handle the positioning im-
posed by the controller using electrical motors with 3 basic states which are
opening, closing or holding the same opening percentage. The rate of open-
ing/closing is correlated to the motor maximum speed.
Figure 2.6.: Inherent flow characteristics of typical control valves [24]
2.2.4. Process PID-controller

There exists two major classifications in regards with implementation of
PID-controller algorithm [3] commonly known in industry as series (2.10) and
parallel (2.11). It is mainly introduced to identify the controller realization
and not to describe it, since the algorithms are identical to one another where
the overall transfer behaviour from the controller input to the output is always
the same, regardless of how the derivative action is being handled by different
manufactures.

u(s) 1 1 + s · Td
= Kp · 1 + · (2.10)
e(s) s · Ti 1 + s · Tf

u(s) 1 s · Td
= Kp · 1 + + (2.11)
e(s) s · Ti 1 + s · Tf
16
The controller parameter Kp corresponds to the proportional gain element,

Ti and Td represent the time constants assigned to the integrator and deriva-
tive elements respectively and finally the time constant Tf relates to the filter
frequency applied on the derivative term which is mandatory from practical
and theoretical aspects.
Practically the measurement sensors produce noise at high frequency further
getting amplified due to the derivative action leading to very large unusable
controller output. The additional low pass filter pole is placed in a manner
which attenuates high frequency noise.
From the theoretical point of view, the PID-controller transfer function with-
out the additional pole cannot be realized since the nominator would be higher
than the dominator.
The series algorithm -still being adopted in digital controllers- was first intro-
duced within the early analog controllers, which were realized using electrical
circuits or pneumatic elements. Its corresponding transfer function can be rep-
resented easily in the frequency domain where the poles and zeros correspond
to the inverse of the corner frequencies.
In the parallel form referred to as non-interacting, the unity feedforward
signal and derivative action predict the error at the moment assigned by its time
constant Td . The integrator intends to eliminate the error between reference
and process output completely where the resulting action from both is modified
afterwards using the proportional gain Kp .
It is worth mentioning that such minor difference in implementation would
have a major impact when attempting to tune the controller parameters using
analytical methods such as Cohen-Coon or Lambda since they can only be
applied on the parallel algorithm.
One notable problem using PID-controller is integrator windup (A.6) neces-
sitating usage of an anti-windup mechanism to prevent the integral element
from growing up further as soon as the controller output hits the saturation
limits entering the nonlinear region. This would occur when the control signal
exceeds the predefined physical boundaries related to the control valve opening
range and allowable amount of mass flow rate which can be supplied.
The drum pressure and water level PID-controllers adopts the parallel algo-
rithm and their set values are always kept constant regardless of the supplied
amount of heat flow rate. Each output is controlled with its separate con-
trol loop without considering any sort of coupling or interaction between both
outputs.
17
2.3. MATALB/Simulink model
Water level and pressure control

Drum level control can be realized using 3 different industry-standard strate-
gies with typical application for each noted as single, two and three-element
control respectively [1]. The numbered term corresponds to the set of mea-
surements being utilized to control the system. GuD 2 at HKW Süd currently
implements the 2-element structure which employees a cascaded control archi-
tecture using level and feedwater flow rate as process variables.
Such strategy is useful as it addresses disturbance imposed on the level and
improves set point response performance when compared to 1-element control.
When directly controlling the level it isn’t enough for the controller by itself
to directly open or close the valve since it have to decide as well whether it
should be feeding more or less feedwater into the drum.
By considering the feedwater flow rate as well, the outer loop compares
the current level with the specified reference and the computed error signal
generates using the PID-controller a new set value for feedwater flow rate. The
inner loop examines the current flow with the amount established by the outer
loop in order to adjust accordingly the control valve percentage opening using
a PI-controller.
During normal operation the pressure is regulated by modifying the linear or
butterfly valves’ position using identical control loop structures consisting of a
simple feedback loop which compares the reference value with the drum actual
pressure. The error is subjected to unity negative gain that ensures an inverse
response to the valve position where its value is altered using a PI-controller
equipped with a dead zone.
The control valve should open if the drum pressure increases to relief the
pressure inside. The same holds if it drops, where the required action is steam
valve closure, thus increasing the pressure within the drum and restoring it
back to the defined set value. The dead zone guarantees a region of zero
output causing the PI-controller to hold its previous state as it’s only allowed
to react when the error signal exceeds certain limits.

The complete physical model is realized within MATLAB/Simulink environ-
ment carrying out direct computation of the differential and algebraic state
equations describing the process elaborated in section (2.2). The parameters
were either extracted from the construction data and control schemes of GuD2
at HKW Süd plant or estimated following their unavailability.
18
2.3.1. Drum-boiler model

A simple MATLAB code was written to implement directly the established
algebraic state equations. The Simulink model uses the user-defined MATLAB
function block to manage the developed script5 provided in appendix (A.3.1).
It requires input of the states, heat and mass flow rates current values in order
to calculate the state variables. They are integrated before being fed back
again as shown in figure (2.7).
States
1
1 State variables (dx/dt)
s
Heat flow rate (z)
2 Model inputs Pressure (bar) 1
Mass flow rates (u) Pressure (bar)
Level (mm) 2
States (x) Level (mm)
Drum-Boiler m-file
Figure 2.7.: Simulink model of the Drum-boiler unit
The liquid/vapour mixture properties are calculated using functions of water

properties and derivatives based on the International Association on Properties
of Water and Steam (IAPWS). These functions were realized using two im-
plemented MATLAB functions which are XSteam [10] and IAPWS-IF97 [19].
The partial derivatives of water and steam densities with pressure were ap-
proximated linearly within the drum pressure operating range as they weren’t
implemented in both of the above mentioned functions.
The riser and downcomer tubes volumes Vr /Vdc is computed by knowledge
of their cylindrical pipe length and cross sectional area. The downcomer-riser
closed loop consists of two tubes that belong to the downcomer supplying water
to the evaporator, the saturated steam is fed back into the drum through six
collectors related to the riser. The drum oval shape was approximated and
considered as a cylinder whose volume Vd would be calculated similarly. The
drum area Ad is associated with the water surface area assumed to be constant
even though it would vary according to the level. The downcomer area Adc is
the pipe cross sectional area already obtained while computing the volume.
Parameters related to empirical equations or lumped models such as the
friction coefficient in downcomer-rise loop K, empirical coefficient β and the
5
Three different subsystems were constructed separately block-by-block in Simulink during
early development of the model
19
hypothetical volume Vsd◦ were quite hard to obtain, therefore were either kept
constant or scaled down relative to the Åström-Bell model6 .

The residence time of steam within the drum Td was identified7 following
system analysis which shall be illustrated in details throughout the next chapter
in section (3.4). It have a huge influence on the overall behaviour which is
reasonably expected since it can be interpreted as the time constant of water
and steam separation process. Simulation results shows that a residence time
higher than 6 sec leads to closed loop instability when utilizing the existing
PID-controllers.
Drum-boiler construction data

Variable Drum Downcomer Riser
Volume 20.204 m3 0.9 m3 20 m3
Mass8 1363 kg 580 kg 1300 kg
Area 14.7 m 2 0.0637 m 2 -
Drum-boiler model parameters
Residence time in drum 4s
Hypothetical volume 2 m3
Friction coefficient 25
Empirical coefficient 0.3
Table 2.1.: Drum-boiler model parameters
The amount of mass flow rates qf and qs at a given pressure P are first
specified in order to compute initial values. This allows computation of the
necessary heat flow rate Q̇ that preservers energy balance. A primary simula-
tion can run once these values are assigned as the model drives by itself the
variables αr and Vsd to steady state by solving equations (2.12). Finally, the
total volume Vwt is the amount required to keep the water at the relative zero
level.
2ρw Adc (ρw − ρs )g ᾱv Vr
Q̇ = qs hs − qf hf w = αr hc
K (2.12)
ρw ρs ρ w − ρs
ᾱv = 1− ln 1 + αr
ρw − ρs (ρw − ρs )αr ρs
6 ◦
K and β were kept the same, whereas Vsd is chosen as a rule of thumb
7
Changed within the range [ 2 sec - 6 sec ] until the model closed loop behaviour matched
the plant real measurements
8
Total mass including the evaporator mt = 98888 kg
20
2.3.2. Control valve and actuator model

The actuator dealing with the positioning is modeled as a 1st order lag element
whose integrator saturation corresponds to the valve position ranging between
0% and 100%. The discontinuous rate limiter block afforded by Simulink library
is used to model the motor rate of opening and closing. Finally, the amount of
feedwater flow rate varies according to the pressure drop across the valve and
its opening percentage as illustrated in equation (2.8).
The pressure ratio across the valve was estimated not to exceed 0.7, therefore
the head loss m according to steam service tables would be 0.96 [22]. The di-
mensionless compressibility factor Z is treated as a function of the superheated
steam pressure and density. Therefore, the amount of steam flowing through a
control valve can be rewritten as described in equation (2.13).

Z = 14.2 ρs P
√ (2.13)
Kvs xs ρs P
qs = 13.6
3600
The feedwater and steam turbine control valves sizing coefficients were ob-
tained directly from their corresponding data sheets. Unfortunately data about
the butterfly valve was missing, its sizing had to be estimated using table charts
from [24] and its inherent flow characteristic curve was simplified as linear. Such
assumption is still very plausible as the valve in the real process never opens
beyond 30%.
1 Pdrop (bar)
Pressure drop qf (kg/s) 1

1
2 xf (%) qf (Kg/s)
s
Postion
Set Point Actuator Electrical Motor
Feedwater Control Valve
Figure 2.8.: Simulink model of the control valve combined with its actuator
Control valves and actuators

Variable Feedwater Steam turbine Butterfly
3
Sizing coefficient 20.368 mhr 364 kg
hr 1363 kg
hr
Rate of opening ± 3.333 %s ± 0.166 %s ± 0.555 %s
Table 2.2.: Control valve and actuator parameters
21
2.3.3. Process PID-controller model

Simulink continuous PID block offers functionalities which meets exactly our
needs, thus a separate realization wasn’t required regarding implementation
of the process controllers. It can simulate the non-interacting PID algorithm
according to equation (2.14)9 , provide output saturation when required, reset
integrators and more importantly equipping an anti-windup mechanism.

u(s) Ki N ·s
= Kp · 1 + + Td · (2.14)
e(s) s s+N
The anti-windup can be handled using the back-calculation method or a log-

ical clamping circuit. Back-calculation feedback loop when employed attempts
to discharge the PID-Controller internal integrator when the controller hits
specified saturation limits by proper tuning of the highlighted coefficient Kb as
shown in figure (2.9) [3].
1
1 Ki Kp 1
s
Error Controller
Integrator gain Integrator Proportional gain Saturation output
Kb
Back-calculation
Td N coefficient
Derivative gain Filter coefficient
1
s
Filter
Figure 2.9.: Simulink model of a parallel PID-controller equipped with an anti-

windup mechanism
PID-Controller parameters
Controller Kp Ti Td Tf
Level 0.05 300 100 50
Feedwater valve 2.3 25 - -
Pressure10 1.8 12 - -
Table 2.3.: PID-controller parameters

9
Ki = T1i and N = T1f
10
Pressure control is carried on during model validation using only the bypass butterfly valve
22
3. Process analysis and validation
3. Process analysis and validation

In this chapter a detailed analysis of the drum-boiler unit shall be carried out
keeping in mind future plans and design considerations. First, essential the-
oretical aspects required to effectively analyze the system are briefly covered.
Then stability of the process is examined analytically by linearizing the non-
linear model at various operating conditions in order to predict its open loop
behavior. Later on, the expected behaviour shall be addressed in details by
conducting several simulation scenarios. Finally, the closed loop response is
validated against real measured data from GuD 2 at HKW Süd.
3.1. Theoretical overview

3.1.1. Concept of stability
Stability of linear systems can be roughly summarized as follows, a system
output will be limited and restricted for any applied bounded input referred
to as Bounded-Input Bounded-Output (BIBO) stability. Examining stability
of linear systems is fairly simple and straightforward as they can be described
either as transfer function or in state-space form, thus stability can be de-
termined by direct computation and graphical visualisations of its eigenvalues
within the complex plane. Furthermore, stability of the controlled closed loop
system can be predicated by merely inspecting the system in open loop while
applying well established methods such as the Nyquist criterion.
On the other hand, stability analysis for nonlinear systems is relatively com-
plicated and requires a high level of mathematical understanding since further
matters have to be considered. The analysis should address stability of equi-
librium points, known as position of rest xR , instead of the overall system.
Steady state takes place for a constant input u0 if and only if the state vari-
ables remains constant as defined in equation (3.1).
ẋ = f (xR , u0 ) = 0 (3.1)
Obviously nonlinear systems positions of rest - referred to from now on as

operating points - have a finite number associated with the solution of equa-
23
tion (3.1), hence requires a more generalized definition offered by Lyapunov

[18] discriminating between different stability forms for each operating points
classified as stable, asymptotically stable and unstable.
3.1.2. Linearization
For simplicity the intended stability analysis shall be performed by lineariza-
tion of the nonlinear model at typical operating points of the drum whose state
algebraic equations can be summarized into the generalized description shown
in equation (3.2).
ẋ = f (x, u)
(3.2)
y = g(x, u)
The resulting linearized model can be described in state-space form (3.3)
where A, B, G, C and D are the system, input, disturbance, output and
feedforward matrices respectively. This will come in handy when attempting
to optimize the controller since algorithm execution of the proposed strategy
requires these matrices. In addition they would reduce the nonlinear state
equations complexity offering a rather simplified overview of the states and
inputs dominating the process outputs. The open loop response of both lin-
ear and nonlinear models should be compared to inspect if both still match,
therefore answering the crucial question concerned with the linearized model
reliability during design of the optimal controller.
ẋ = Ax + Bu + Gz
(3.3)
y = Cx + Du
The matrices are computed with the help of Taylor series approximation
neglecting quadratic and higher order terms (3.4). The method intuitive basis
is that a smooth curve differs very little from its tangent line as long as the
variable doesn’t wander from the point of tangency.

∂fi ∂fi
aij = , bij =
∂xj xR ,u0 ∂uj xR ,u0
(3.4)
∂gi ∂gi
cij = , d =
∂xj ∂uj
ij
xR ,u0 xR ,u0
3.1.3. Poles and zeros

System zeros affects only shape of the output which can lead to minimum or
non-minimum phase behaviour according to their position within the complex
24
plane (A.5) [9] [23]. Alternatively poles determine stability as they are directly
associated with the system eigenvalues. That’s why inspection of the system
poles and zeros is quite efficient while analysing and predicting the system
response.
If the eigenvalues are located on the left-hand side (LHS) of the complex
plane the states will converge to zero stabilizing over the course of time. How-
ever if located on the right-hand side (RHS) the states will keep growing due
to the exponential product as depicted throughout equation (3.5) where ci
are constants coefficients related to the solution of the homogenous differential
equation describing dynamics of the system.

n
y(t) = ci eλi t (3.5)
i=1
We still need to define the relationship between the zeros, poles and eigen-
values, in addition understand how it differs when comparing Multiple-input
Multiple-Output (MIMO) systems to Single-Input Single-Output (SISO).
SISO systems
Commonly input-output (I/O) behavior is presented using transfer functions
(3.6) where zeros zi and poles pi are simply the roots of the numerator N (s)
and dominator D(s) respectively. The transfer function dominator is exactly
equivalent to the characteristic polynomial evaluated by solving equation (3.7),
that why all poles corresponds to the system eigenvalues λ.
N (s) (s − z1 )(s − z2 )...(s − zn )

G(s) = = (3.6)
D(s) (s − p1 )(s − p2 )...(s − pn )
.
det(λI − A) = 0 (3.7)
MIMO systems
Zeros in multivariable systems do play an additional role besides affecting
system shape and performance since it might gravely influence the ability to
fully control the system [5] [6]. They are redefined with the help of Rosenbrock
matrix which benefits from the state-space description distinguishing between
transfer and decoupling zeros. The complete set consisting of both known as
invariant zeros1 is examined by computing the rank of the matrix (3.8). Not
1
Only under the assumption that feedforward matrix D = 0
25
3.2. Stability analysis
all of the system eigenvalues necessary appear as poles due to existence of

decoupling zeros compensating poles in I/O transfer functions Gij (s).
If this occurs in a system, it would be impossible to fully control the system
since some eigenvalues are no longer influenced by the controller. It gets even
worse if the uncontrollable eigenvalue is located on the RHS of the complex
plane because there no way to stabilize the plant with its current setup using
any control technique.
sI − A −B
P (s) = (3.8)
C 0

The drum-boiler unit stability can be easily comprehended from the physical
point of view with basic understanding of the drum mass and energy balance
equations discussed during modelling in section (2.2.1). For example, if the
outflow leaving the drum is less than the amount which is supplied by the
inflow then water will start filling the drum and vice-versa. Alternatively while
assuming constant mass flow rate, if additional firing takes place by the HRSG
providing more heat leading the drum temperature to rise and causing the
pressure to build up in return, thus reaching dangerous limits which will cause
explosion of the drum ultimately.
Even though this shortened explanation could be enough, additional ana-
lytically driven investigations needs to conducted by performing a stability
analysis to extend our understanding of the expected behaviour and establish
the foundation necessary for the controller design.
3.2.1. Linear state-space model

The model is linearized at 3 operating points shown in table (3.2.1) us-
ing MATALB Control System ToolboxTM linear time-invariant (LTI) functions
dedicated for continuous systems time-domain analysis (A.2). They cover the
drum operating range whose upper limit is specified by the maximum amount
of saturated steam allowed to flow through the pipes upon exiting the drum.
The input B and disturbance G matrices show that the dominant inputs
which affects the drum pressure P dynamics are heat Q̇ and steam qs flow
rates as expected. The water total volume Vwt is obviously affected mainly by
the mass flow rates. Steam mass fraction αr depends heavily on condensation
enthalpy hc , downcomer qdc and heat Q̇ flow rates according to state equation
(2.6). As for the steam bubbles volume Vsd in the water level, it can be seen
that it’s associated with all states and input variables as it follows an empirical
26
equation derived through continuous observation of the process to capture the

drum complicated dynamics.
⎡ ⎤
−7.148e-5 1.051e-14 0 0
⎢−5.094e-11 −9.723e-21 0 0 ⎥
A=⎢ ⎣ 1.021e-9
⎥
⎦
−2.684e-21 −0.1827 0
1.752e-6 7.424e-18 −297.3 −0.3333
⎡ ⎤ ⎡ ⎤
−21.09 −216.6 103
⎢ 0.001085 −0.001255⎥ ⎢7.341e-5⎥
B=⎢ ⎣ 5.386e-6
⎥ G = ⎢ ⎥
5.534e-5 ⎦ ⎣7.904e-5⎦
−0.01486 0.2023 0.07526

1e-5 0 0 0
C=
−4.86e-4 68.027 2035 68.027
States Inputs
Operation 3 3
P (bar) 2
Vwt (m ) αr (%) Vsd (m ) Q̇(MW) qf ( kg
s ) qs ( kg
s )
Low 5.5 21.501 0.0098 1.378 13.8473 6 6
Medium 5.5 20.391 0.0138 1.067 20.771 9 9
High 5.5 19.736 0.0178 0.756 27.6947 12 12
Table 3.1.: Drum-boiler operating points for low, medium and high load
3.2.2. I/O pole-zero plot

The input-output pole-zero map illustrated in figure (3.1) concerning the
transfer behaviour from inputs to the water level shows that all four eigenvalues
appear as poles. The first three are always located at − T1d and the origin3 ,
associated with the drum pressure, water volume and dynamics of the steam
bubbles under water level respectively. The last pole which depends on the
operating point is situated at − hec33
qdc
, it relates to the steam dynamics flowing
through the riser tubes. It keeps shifting to the left along the negative real axis
towards infinity as long as the load increases.
This was quite expected from our basic understanding regarding steam gen-
eration working principle using the drum-boiler. Higher loads require more
electrical power generated by the gas turbine which in return provides addi-
tional heat to the riser tubes, thus accelerating conversion process of feedwater
into steam within the naturally circulated downcomer-riser loop. If an enor-
mous amount of heat is supplied the pole keeps approaching negative infinity,
2
The indicated pressure through the thesis is the absolute pressure
3
Assuming constant residence time of steam within the drum
27
when inspected in the complex plane, since the conversion shall take place
instantaneously.
Since no compensation of eigenvalues have occurred, all invariant zeros are
classified as transfer zeros. In addition one can stay assured that the system is
completely controllable because any eigenvalue can be influenced by affecting its
corresponding pole. The transfer zeros which are located on the right hand side
(RHS) of the complex plane have been anticipated earlier from experimental
observation and physical understanding. They are directly correlated with the
shrink and swell physical phenomena leading the system to react in a non-
minimum phase behaviour.
In particular zeros related to the transfer behavior from steam flow rate to
water level are very close to the origin when compared with zeros linked to
feedwater and heat flow rates transfer functions respectively as seen in figure
(3.1). Therefore we should be definitely expecting a significant difference in
regards with amplitude of the water level initial inverse response when stimu-
lated by the input variables. This shall verified in the next section concerned
mainly with the open loop response to a step input.
Heat flow rate Q (MW)
1
3 2 1 1 23 1 2 3
0
1
0.28 0.21 0.14 0.07 0 0.07
Flow rate qf (kg/s)

1
3 2 1 3 2 1
0
1
0.28 0.21 0.14 0.07 0 0.07
Flow rate qs (kg/s)

1
3 2 1 3 21
0
1
0.28 0.21 0.14 0.07 0 0.07
Real axis
Figure 3.1.: Pole-zero plot of the linearized models at low (1), medium (2) and
high (3) load
28
3.3. Open loop step response

The open loop response shall be studied by simulating the model4 consisting
of the drum-boiler unit and control valves5 when subjected to a step change of
the gas turbine electrical output power and control valves position respectively.
One input at a time is stimulated using a step function while the other inputs
remain intact. The mass flow rates would vary according to pressure dynamics.
3.3.1. Change of gas turbine electrical output power

Figure (3.2) shows the response to a step input of gas turbine output electrical
power equivalent to a decrease of 20 MW. The amount of heat flow rate Q̇
supplied to the drum, required as an input to run the simulation, is assumed
to vary instantaneously following the change of the gas turbine power. Butterfly
and feedwater control valves positions were kept constant.
The pressure P starts decreasing following the declination of heat flow rate
associated with gas turbine output power. It affects the amount of steam flow
rate qs leaving the drum as the valve position haven’t changed. On the other
hand, the pressure drop across the feedwater valve starts building up since the
feedwater pump pressure is kept constant, hence causing more feedwater qf to
flow into the drum.
The water total volume Vwt initially decreases due to evaporation caused by
sudden pressure drop before incrementing eventually following the increase of
feedwater. The steam mass fraction αr in the riser tubes immediately steps
down once the heat supplied is smaller than its initial state then keeps sliding
down gradually as the amount of water being vaporized by the evaporator
within the downcomer-riser loop was reduced.
The level response l depends on a combination of complicated dynamics
related to distribution of water and steam. The step-like change of steam mass
fraction αr leads to the initial undershoot as the quantity of steam bubbles
fed back to the drum rapidly drops. The swelling effect is then noticed once
the pressure starts to decrease resulting in steam bubbles expansion causing
the level to rise. Finally following this transient effect, water keeps filling in
the drum due mass imbalance where feedwater supplied to drum inlet is much
higher than the steam leaving from the outlet.
4
All simulations were conducted in Simulink using a fixed step size of 1 s
5
In [2] the open loop response considers only the drum-boiler unit
29
3.3.2. Change of butterfly valve position

Figure (3.3) shows the step response of the system due to opening of the
butterfly valve equivalent to 10% while heat flow rate and position of feedwater
control valve were kept constant.
The steam flow rate qs rapidly increases according to the valve rate of open-
ing after the unexpected rapid change in valve positions. The opening of the
valve relieves the pressure inside the drum and hence it starts decreasing. Once
the valve reaches its designated opening percentage, the pressure P starts dom-
inating the behaviour of the steam flow rate thus reducing the amount of steam
leaving the drum because both are related to each other. Feedwater flow rate
qf increments due to the increased pressure drop across the feedwater valve.
The water total volume Vwt decreases for two reasons; one is the evaporation
caused by the pressure drop and the other being the relatively high difference
in mass flow rates.
The steam mass fraction αr steps up once the pressure have decreased then
starts sliding gradually until it approaches its original state following the tran-
sient effect occurring to downcomer qdc and riser qr flow rates. Finally the
level l initial inverse response is caused by the bubbles swelling and volume
expansion then it falls constantly due to mass imbalance.
3.3.3. Change of feedwater control valve position

Figure (3.4) shows the step response of the system due to closing of the
feedwater control valve equivalent to 10% while the heat flow rate and position
of butterfly control valve were kept constant.
The feedwater qf drops in step fashion since the control valve reaches its
designated position very quickly with its fast rate of opening/closing. The
decrease of cold feedwater fed into the drum increases its temperature which
in return affects the pressure allowing more steam qs to leave the drum. The
mass balance inflow and outflow was disturbed within the drum, therefore the
water total volume Vwt declines at high rate.
The steam mass fraction αr behaviour is similar to the open loop response
of the steam control valve initially dropping following pressure rise then sliding
gradually upwards towards its initial state. The sudden drop of feedwater flow
rate resulted as expected in the level l initial inverse response corresponding to
the predicated system non-minimum phase behaviour.
30
Pressure (bar) Flow rate qs (Kg/s) Steam quality (%)

9 0.014
5.5
0.013
8.5
5 0.012
8
0.011
4.5 7.5 0.01

0 100 200 0 100 200 0 100 200
Level (mm) Flow rate qf (Kg/s) Volume Vwt (m3)

50 9.2 20.8
20.7
9
0 20.6
8.8
20.5
50 8.6 20.4

0 100 200 0 100 200 0 100 200
Figure 3.2.: Open loop response for a step change equivalent to decrease of
20 MW of the gas turbine electrical output power
Pressure (bar) Flow rate qs (Kg/s) Steam quality (%)

12 0.0145
5.5
11
0.014
5 10
0.0135
9
4.5 8 0.013
0 100 200 0 100 200 0 100 200
Level (mm) Flow rate qf (Kg/s) Volume Vwt (m3)

9.2 20.6
100
9 20.4
50
8.8 20.2
0
8.6 20
0 100 200 0 100 200 0 100 200
Figure 3.3.: Open loop response for a step change equivalent to 10% opening
of butterfly valve position
31
3.4. Validation
r esuu es(ba) e7 Flow(e) ts(qu(bKg/u7 Sts) m(q ) lity(b%7

4.P4 8.84 0.0135
4.P 8.8
4.44 8.94 0.0135
4.4 8.9
4.54 8.P4 0.0133

0 100 200 0 100 200 0 100 200
Lsvsl(bmm7 Flow(e) ts(qf(bKg/u7 Vol ms(Vwt(bm37

6
10 20.P
0 8 20.5
10 9 20.2
20 P 20
30 16.8
4
0 100 200 0 100 200 0 100 200
Figure 3.4.: Open loop response for a step change equivalent to 10% closing of
feedwater control valve position
3.4. Validation
The system closed loop response will be validated and examined against data6
from the real plant for different scenarios to experiment its ability to capture
the real process dynamics at various operating conditions. The complete model
with the PID-controllers is shown in figure (3.5).
3.4.1. Assumptions
The heat flow rate required as an input of the model cannot be measured in
reality, yet can be predicted from the gas turbine electrical output power which
1 MW
changes as ramp function with a slope of 12 s
. The transfer function relating
st
both is assumed to be 1 order lag element whose time constant was identified
τ = 280 s assuming that the supplied heat behaviour is directly associated with
the evaporator temperature.
The feedwater valve position is always kept half-way opened in the plant
without considering the amount of feedwater which flows through it. Therefore
the pressure drop across the valve should increase or decrease accordingly to
preserve such condition which is achieved using he feedwater pump controller.
6
The measurements of the plant are being filtered and sampled at a rate of 1 Hz
32
3.4. Validation
Implementation of the controller was neglected for simplicity and it is assumed

1 bar
that the pump output would vary ramp-wise with a slope of 60 s
followed by
PT1 element with a time constant τ = 15 s chosen as a rule of thumb.
The tests conducted in the plant were using only the butterfly valve to reg-
ulate the steam flow rate therefore the pressure control loop will consist only
of the corresponding PI-controller. Additionally steam flow rate measurement
from the plant doesn’t take into account the amount supplied back to the
feedwater tank which was assumed to be approximately around 1 kgs .
The corresponding valve is usually kept open at a predefined position and
its contribution to control the drum pressure can be neglected. However as
the drum-boiler model requires the total flow rate which leaves the drum as an
input variable, the estimated amount flowing into the feedwater tank is simply
related to the drum pressure dynamics with a low pass transfer behaviour which
is similarly chosen as another rule of thumb.
The pressure loop PI-controller proportional gain Kp was adjusted from 1.8
to 4. The model closed loop performance improved and matched much better
the measurements when compared to its initial value when observing the sim-
ulations results. This is due to several factors discussed when concluding the
chapter in section (3.5).
3.4.2. Comparison with measurement data

The pressure controller senses the pressure decrease within the drum as less
heat is being supplied as shown in figure (3.6), thus it tries to close the butterfly
valve to restore pressure back to its set point.
Once the valve starts closing, the water level l drops due to the shirking
effect of steam bubbles. It experiences an undershoot followed by an overshoot
since the cascade controller is simultaneously trying to restore the level back
to its set point and to reestablish energy balance for the drum as well as mass
balance for inflow and outflow.
Figure (3.7) illustrates the comparison considering gas turbine power in-
crease. The controllers react on the pressure rise within the drum caused by
the additional heat supplied, therefore opening the corresponding valve to re-
lief drum pressure allowing more steam to leave from the drum outlet in the
process.
The water level l increases due to steam bubbles swelling, yet a smaller
overshoot is observed since the change of electrical power is less when compared
to the previous scenario, thus in return permit the controllers to settle and drive
the process back to steady-state faster.
33
3.4. Validation
Steam to 0.18182
feedwater
tank 5s+1
P_drum (bar)
qs (kg/s) qs (Kg/s)
1
5.5 PI(s) xs (%)
s P (Bar)
Pressure
Dead Zone Actuator xs Electrical Motor Butterfly valve
Reference value qs PI control
-0.04 to 0.04 10% by 18sec
Kp = 4 Ti = 1/15
1e+006
Q (W)
280s+1
Feedwater Heat flow rate (Watt)
Low pass Low pass
pump Saturation
Gain watt to MW
1
15s+1
Level (mm)
Pdrop (bar)
qf (kg/s) qf (Kg/s)
1
0 PID(s) PI(s) xf (%)
s
Level
Actuator xf Electrical Motor Feedwater valve
Reference value Level PID Controller qf PI Controller
10 % by 3 sec
Kp = 0.05 Ti = 1/300 Kp = 2 Ti = 1/25 Drum-boiler model
Td =50 Tv = 100
Figure 3.5.: Simulink validation model
34
3.4. Validation
Pressure (bar) Valve position xs (%) Flow rate qs (Kg/s)
20 8
5.6
7
5.5
5.4 15 6
5.3 5
5.2 10 4
0 2000 4000 0 2000 4000 0 2000 4000
Level (mm) Valve position xf (%) Power Q (MW)

100
80 100
50
0 60 90
50
40 80
100
0 2000 4000 0 2000 4000 0 2000 4000
Figure 3.6.: Comparison between model (dashed) and plant data (solid line)
for a decrease of the gas turbine electrical output power equivalent
to 20 MW
r esuu es(r (ba) e1 i ) pvs(t nux%

xn7(8u(bw 1 Kpng (e) %
s(/ u(bL mfu1
q.4
4.P 2q q
4.6 P.4
26
4.4 P
23 6.4
4.3
6
0 2400 5400 0 2400 5400 0 2400 5400
l svsp(p(bo o 1 i ) pvs(t nux%

xn7(8F(bw 1 r ng se(Q(bMW1
P0
40 90
60
V4
40
0 q4
30
V4 50 q0
40 V0
0 2400 5400 0 2400 5400 0 2400 5400
Figure 3.7.: Comparison between model (dashed) and plant data (solid line) for
an increase of the gas turbine electrical output power equivalent
to 10 MW
35
3.5. Concluding remarks

In the end of the process comprehensive study, we shall wrap up the analysis
and summarize results as follows.
The resulting linearized model of the drum-boiler unit predicted the open
loop response of the dynamic realistic nonlinear system quite well and can be
relied on safely in regards with intended future plans.
The I/O pole-zero plot assures full controllability of the system as all of its
eigenvalues are appearing as poles. It clarified as well how the shrink and swell
phenomena is associated with the transfer zeros located on the RHS of the
complex plane leading to an initial inverse response which should be handled
by the PID process controllers.
The identified main problem with the existing level control can be outlined
as follows
Assuming drop of exhaust heat provided to the drum leading to decrease
of pressure and water level. The feedwater control valve supplies more
water through the inlet yet unintentionally contributing into additional
drop of the level
From the physical point of view this takes place since the cold water fed
into the drum decrease its temperature and as result its pressure as well
The pressure controller tries to close the steam valve even more to track
the set value leading eventually to further water drop due to steam bub-
bles shirking
Clearly the level control isn’t considering the initial inverse response iden-
tified while examining the system open loop behaviour
The comparison results show that the model can capture the drum dynamics
to a great extent. However, a relatively small deviation from real measurements
and slightly faster response specially regarding pressure and level is still notice-
able. The error arises due to the uncertainty of some model parameters such
as Td and K, in addition to the suggested assumptions necessary in order to
simplify the complete process simulation.
Nevertheless, the model current attitude can be regarded as satisfactory,
bearing in mind that several control loops were omitted for simplification.
Sooner or later, a realization within the real plant would ultimately require
an observer gain, whose design shall be discussed in section (4.3.2), to correct
the states based upon the difference between real measurements and model
outputs.
Figure (3.8) shows the closed loop behaviour for different values of the pres-
sure controller proportional gain. As mentioned earlier in the initial assump-
tions (3.4.1), the gain doesn’t correspond to the current real value and had to
36
be increased so that the simulation results match the plant measurements. The
drum pressure drops more when the gain isn’t changed and as a consequence
it causes higher overshoot of the level due to steam bubbles swelling.
The reasons behind such modification which as seen in the previous simu-
lation results improved the model overall performance could be summarized
throughout the following
The treatment of the bypass butterfly valve as a linear valve, therefore

in reality the valve opens more to allow flow of additional steam, yet in
the model it doesn’t open with the same percentage due to the different
nature of both valves, which was already been clarified in figure (2.6)
Neglecting storage of low pressure steam within the pipes and superheater
in the current model, as a result, introduction of an additional state
variable7 might be necessary
Pressure P (bar) Flow rate qs (kg/s)

8
5.6 7
6
5.4
5
5.2 4
0 1000 2000 3000 4000 0 1000 2000 3000 4000
Level l (mm) Flow rate qf (kg/s)

100 14
50
10
0
6
50
100 2
0 1000 2000 3000 4000 0 1000 2000 3000 4000
Figure 3.8.: Comparison between model and plant data (solid line) for a de-
crease of the gas turbine electrical output power equivalent to
20 MW when Kp = 1.8 (dotted dashed) and Kp = 5 (dashed)
7
System identification shows that a 5th order system matches better the plant measurements
when compared to the current 4th order model
37
4. Process optimization
4. Process optimization
The concluded results brought to our attention during system modelling
and analysis suggests that an optimization of the process is achievable using
a multivariable control technique. The strategy would account for synergy
between feedwater and steam flow rates instead of just decoupling the MIMO
system into several coupled SISO systems regulated by their own noninteracting
control loops.
A state-feedback controller is suggested in order to consider the internal
variables of the system instead of the process outputs, therefore accounting for
additional aspects which were discarded using the classical control methodol-
ogy. The inner dynamics of the drum-boiler unit correspond to the developed
nonlinear model state variables which were defined in section (2.2.2).
The control concept shall be addressed presenting the available control meth-
ods and algorithms applying the approach while highlighting advantages and
disadvantages for each.
4.1. Concept of state-feedback control

For a continuous linear system described in state-space form (3.3) whose
states are available for feedback, it can be subjected to a multidimensional
proportional gain element F compared with the reference value to compute
the actuating variables u according to the control law (4.1) where the resulting
system matrix would be A − BF .
The designed state-feedback matrix would place the poles of the closed loop
system in a desired position within the complex plane, thus directly influencing
its rise and settling time, damping and transient oscillations.
u = −F x (4.1)
4.1.1. Controllability and observability

The concept of controllability was briefly hinted to while examining the sys-
tem poles and zeros in section (3.1.3). The term was introduced to investigate
whether the actuating variable are perfectly able to drive the system from any
38
initial state to the desired state. Alternatively observability was proposed to

examine ability of estimating the system states from a set of available measure-
ments. The terms are dual to each other, thus any criterion or control design
method can be applied for both by adjusting A AT and B C T .
State-space description offers the opportunity to investigate both properties
analytically by simply inspecting the system, input and output matrices respec-
tively using Kalman, Hautus or Gilbert criterions [18]. Kalman criteria only
investigate controllability/observability of the entire system without mention-
ing any particular eigenvalue. Fortunately Hautus and Gilbert criterions can
identify the non-controllable/observable eigenvalue, thus allowing the control
designer to adjust the plant structure accordingly if necessary.
4.1.2. Observer-based control

The states have to be estimated with the help of a plant model due to the
fact that they are most likely hard to be measured in the real process. The
difference between the real and estimated output is subjected to the so-called
Luenberger observer gain matrix [4] [16] [17] before being fed back to the model,
thus correcting the states and matching the reality as much as possible.
The newly established objective similarly to state-feedback is computation
of a feedback matrix L which modifies the system matrix into A − LC. The
observer gain is designed in such a manner that ensures convergence of the
estimation error to zero which allows usage of the corrected states for feedback.
B C
Luenberger
Observer L
B C
Figure 4.1.: State estimation using Luenberger observer
39
4.1.3. PI-based state-feedback control

One major disadvantage with the basic state-feedback structure is lack of
reference value tracking even without existence of external disturbance due to
the absence of an integrator element. An enlargement of the basic structure
to integrate the error is mandatory to ensure that the output can follow the
set value, thus guaranteeing steady-state accuracy. Low control speed draw-
back would arise similar to the classical I-controller which necessitates further
enlargement by incorporating a supplementary P-controller which improves
control performance. The complete enlarged PI-based state-feedback control
structure is shown in figure (4.2).
- -I B C
-F
State Controller
Figure 4.2.: PI-based state-feedback control structure
The previous control law have to be slightly modified considering error signals
as additional states h. The newly computed state-feedback matrix K would
consist of 3 parameters F , I and P affecting the states, integrated error signals
and reference tracker correspondingly.
The tunable parameters are assigned upon computation of the feedback ma-
trix K. By default I is uniquely defined as it corresponds to the left-hand
side of the matrix. However F and P can be freely selected due to the un-
derdeterministic nature of equation (4.2). Ignoring P element would lead to a
simple I-controller, while setting F to zero isn’t relevant when attempting to
design a state controller, besides, this would normally introduce an unsolvable
equation1 (4.3).

x
u = [P C − F , −I] = −Kx (4.2)
h
P = −KC −1 (4.3)
1
A solution might exist under very special certain conditions
40
4.2. Controller design methods

4.2.1. Pole placement method
Several algorithms do exist to apply the pole placement method which directly
assigns poles of the closed loop at the desired positions chosen by the designer.
The method however is quite hard and time consuming when applied in practice
with real systems due to the huge amount of freedom offered.
Additionally it doesn’t take into consideration the limitations imposed from
the real actuating variables and no clear guidelines do exist regarding how and
where the poles should be placed. Finally predicting the system dynamical
behaviour by just positioning the poles is a complicated task and the design
process requires usually tedious work even for low order systems before satis-
factory results can be achieved.
Ackermann’s formula
The algorithm is regarded as a standard design procedure which computes
the unique state-feedback f row vector or observer gain l column vector using
the formula (4.4) [18] where n is the system order, p are the coefficients of
characteristic polynomial calculated while defining positions of the eigenvalues
for the closed loop and finally t is the last column/row obtained from the
computed controllability/observability matrix inverse.
Unfortunately such method cannot be applied for MIMO systems as the al-
gorithm requires inversion of controllability/observability matrix which is only
attainable with square matrices with full rank.
f = t1 (p0 + p1 A + ... + pn−1 An−1 + An )

(4.4)
l = (p0 + p1 A + ... + pn−1 An−1 + An )t1
Kautsky, Nichols, and Van Dooren algorithm

The major difference which should be pointed out when designing a state con-
troller for a MIMO system is that the state and observer feedback gain matrices
are no longer unique, therefore offering an extra degree of freedom for design-
ers. The algorithm [11] tries to find a solution which improves robustness of
the resulting state controller by computation the matrix F and estimating in
an iterative manner how closely are the eigenvalues of the closed loop system
matrix A − BF from the desired position.
41
4.2.2. Linear-Quadratic method

Optimal control theory handles issues related to computation of a control law
for a given system bearing in mind that certain optimality criterion has to be
achieved, mainly focusing on how to operate a dynamical process at minimum
cost. The calculus of variations maximum principle formulated an abstract
framework which describes the optimal control problem trying to minimize the
cost function (4.5) subjected to a dynamical linear system with zero initial
conditions where ψ and L are the terminal cost and lagrangian2 respectively.
tf
J = ψ(x(tf ), tf ) + L(x(t), u(t), t)dt
t0 (4.5)
ẋ = f (x(t), u(t), t)
LQ method is a significant result of the theory that manages problems associ-

ated with quadratic performance criteria for state-space systems. The method
algorithm computes a Linear-Quadratic Regulator/Estimator (LQR/LQE) re-
ferred to as Riccati controller when designing a state-feedback matrix or Kalman
filter when attempting to estimate states of a real system.
Riccati controller
The Riccati controller allows trade off between regulation performance and
control effort compared with the pole placement method. It’s regarded as
a robust controller since it attains infinite marginal gain and offers a phase
margin δ ≥ 60◦ [15] which is aligned with practical guidelines for control system
design. The resulting optimal feedback gain should drive the closed-loop system
without external input from any initial state to the zero state minimizing the
cost function described by equation (4.6).
∞
J= (xT (t)Qx(t) + uT (t)Ru(t))dt (4.6)
0
Q and R matrices are positive definite matrices assigned as weighting factors

for the course of states and input variables. Faster convergence of a particular
state towards zero should increase its equivalent coefficient inside the matrix
Q. If a slower response of the actuating variables is preferred to lower the
energy consumption and minimize control effort then coefficients of R matrix
have to be chosen larger.
2
Function that summarizes dynamics of the system
42
The control law (4.1) tries to minimize the quality function (4.6) whose
optimal feedback matrix F requires the symmetrical positive definite matrix
P resulting from solution of the Matrix Algebraic Riccati Equation (MARE)
(4.7) and hence the reason why a LQ regulator is named after Riccati.
F = R−1 B T P
(4.7)
AT P + P A − P BR−1 B T P + Q = 0
Choosing the values of Q and R matrices in principle is similar to tuning

of PID-controller parameters where the weighting matrices are varied until a
satisfactory response is reached. Tuning a Riccati controller first carries inves-
tigations concerning the range of values for each system state and actuating
variable. Once these limits are established, an initial guess can take place by
constructing diagonal matrices with the normalized values as seen in (4.8).
⎡ ⎤ ⎡ ⎤
q11 · · · 0 r11 · · · 0
⎢ .. ⎥ ⎢ .. ⎥
Q = ⎣ ... . . . . ⎦ R = ⎣ ... . . . . ⎦
0 · · · qnn 0 · · · rnn (4.8)
1 1
qii = max , rii = max , i = 1, 2, ..., n
xi ui
Kalman filter
The Kalman filter optimizes the estimation of the system states using a series
of process measurements. It takes into account the input w and measurement
v noises assumed to be random unbiased white noise. It constructs an optimal
state estimator that minimizes the cost function (4.9) where E() calculates the
expected value based on the assumed random noise.
J = E(eT W e) (4.9)
The observer gain L computed by solving the modified MARE (4.10) at-
tempts to minimize the difference between estimated and real states considering
noise influence on the process. The weighting matrices Q and R aren’t con-
sidered as punishing factors anymore as they define intensities of the expected
process noise. Choice of their values usually starts with identity matrices as
initial guess then adjusted repeatedly until a decent estimation is achieved.
L = P C T R−1
(4.10)
AP + P T A − P C T R−1 CP + Q = 0
43
4.3. Observer-based state-feedback controller design
4.3. Observer-based state-feedback controller design

4.3.1. Riccati controller
The state-feedback matrix K is computed using LQ method due to its advan-
tages over the pole placement method. The existing control valves PI-control
loops would remain the same, leading to smooth implementation of the pro-
posed structure when performed at the distributed control system (DCS) of
the real plant as seen in the block diagram shown in figure (4.10).
This idea was suggested since the state controller would be only trying to
generate the mass flow rates signals required to stabilize the process. These
signals can be treated as the set values for their corresponding control valves
handled by their own actual controllers, therefore further modifications in the
plant control schemes would be skipped.
Setting the drum-boiler model states limits were straight forward and helped
performing the initial simulation. The drum pressure limit would be 9 bar as
specified by construction data, the error between reference and actual measure-
ment shouldn’t exceed 0.3 bar. The water level maximum allowable deviation
from set point isn’t allowed to surpass ±150 mm. The remaining states and
actuating variables were kept at unity.
The main challenge was system stabilization considering the limitations and
constraints imposed by the control valves position range and fixed rate of open-
ing/closing. Several simulation took place varying mainly r11 , r22 , q56 and q66
associated with mass flow rates, pressure and level error signals respectively
until good results were obtained with the weighting matrices provided along-
side with the MATLAB script to calculate the state-feedback matrix K in
appendix (A.3.2).
The tunable parameters from equation (4.2) were assigned as follows, the
integrator gain Q as mentioned is uniquely defined by the last two columns.
The proportional gain P was set to zero3 and as a result the state-feedback
matrix F doesn’t require any modifications since it’s described by the first four
columns when solving equation (4.2).

7.1296e-5 21.786 −319.78 0.29002
F =
−1.5083e-4 −3.8327 36.181 −1.2566e-2

1.0356e-2 −1.4135e-3
I=
0.18248 2.6739e-5
3
P should be a 2-by-2 identity matrix in the real process, since the proportional gain of
the PID-controller reacts on the sum of all actions and not the error signal
44
4.4. Simulation results
4.3.2. Luenberger observer

The drum-boiler unit nonlinear model is already estimating drum-boiler unit,
however only lacking the observer gain. It shall be designed to eliminate the
error between measured data and model outputs noticed during validation in
section (3.4). The observer gain isn’t calculated using Kalman filter because
efficient practice of the method requires continuous testing of the estimator
alongside the real process. Further, the optimal choice of the filter weight-
ing matrices depend on reliable prediction of the process noise which is only
guaranteed by regular observation. Pole placement method is applied despite
its disadvantages mentioned in section (4.2.1) with respect to the following
practical consideration.
The observer response must be faster than the closed loop employing the
state-feedback controller because the estimation error have to decay to zero
causing the state variables to converge before the states can be used for control.
As a rule of thumb, the observer slowest pole A − LC should be faster than the
state-feedback controller poles A − BF . This would guide us in a certain way
to assign its position assuring suitable and decent estimation. The observer
poles are provided within the MATLAB script to calculate the matrix L in
appendix (A.3.2).
T
992.85 1.0584e-2 2.6068e-4 0.47076
L=
2.2837e-16 2.1728e-4 3.2667e-6 6.0672e-3

The simulated optimized system performance will be shown in the follow-
ing figures. First we shall examine the estimated states when employing the
observer to check if the new poles associated with A − LC introduce noise
into the system. Then the model behaviour with state correction is validated
against new measurement data with very rich excitation covering the drum-
boiler operating range. Therefore the observer stability can be investigated,
providing a good indication of the proposed control strategy applicability since
the states are crucial for feedback.
Later on, comparison between both controllers is addressed to check if the
newly proposed state-feedback controller did handle efficiently the main prob-
lems identified with the current controller causing its poor performance which
was discussed when concluding the process analysis in section (3.5). Further,
the states and flow rates at different load conditions are inspected to ensure
stability of state-feedback matrix K for the drum operating range.
45
States estimation
The observer pole placement was successful as shown in figure (4.3) since the
estimated states noise is almost negligible therefore they can be efficiently fed
back to the state-feedback controller matrix.
Observer performance subjected to perturbations in gas turbine electrical

output power
Significant improvement of the model closed loop response when combined
with the observer was achieved. Both outputs are almost matching perfectly
when compared to the simulation conducted during primary validation in sec-
tion (3.4). Figure (4.4) shows the comparison between the observer and real
process when electrical power of the gas turbine was switched in between
120 MW and 80 MW for approximately one hour.
Even for different initial conditions as seen in figure (4.5), the observer gain
was still being able to adjust the states and accordingly the process controllers
to track the real output. Finally, the noticeable error which occurred due to
model uncertainty and assumptions discussed in section (3.4.1) was eliminated
when adopting the observer gain as illustrated in figure (4.6).
State controller performance subjected perturbations in gas turbine

electrical output power
The drum pressure and water level were vastly enhanced when analyzing both
behaviours depicted in figure (4.7). The level maximum peak overshoot/un-
dershoot didn’t exceeds ±100 mm during transients and the pressure never
surpasses the safety limits which might lead to operation of the security safety
valve. Obviously the steam flow rate performance is the same using both con-
trollers but the feedwater flow rate behaviour was modified in a way which
boosted the overall closed loop performance.
This is no surprise and should have been expected following process analysis
which diagnosed the drum level cascade controller and highlighted its particular
weakness. The optimal state controller was smart enough paying attention to
the initial inverse response and shrink/swell physical phenomena by considering
the inner dynamics of the system instead of the output. It clearly solved one
of the main problems reported by the plant engineers.
46
State controller performance at different loading conditions

One should note that we designed the optimal state controller feedback matrix
using a linearized model which is normally valid for one particular operating
point. Therefore we have to investigate if it is still able to stabilize the system at
various loading conditions and whether the closed loop response is still tolerable
in regards with the requirements we assigned while weighting the matrices Q
and R.
Figure (4.8) illustrates the performance of the drum pressure and level for the
same loading conditions utilized to linearize the model, which were illustrated
in table (3.2.1). Obviously the level drop was much higher at low load as
the feedback gain matrix wasn’t computed in order to optimize this particular
operation, nevertheless we still have a decent better response when compared
with the existing process PID-controllers. Figure (4.9) shows how nicely the
mass flows set points considers the limitations imposed by the control valves
opening and closing rates allowing feedwater and steam flow rates to track
them smoothly.
Water volume Vwt (kg/s)
22
20
0 1000 2000 3000 4000 5000 6000 7000

Steam quality (%)
0.015
0.01
0 1000 2000 3000 4000 5000 6000 7000
Steam volume under water level Vsd (m3)
1.7
0.95
0.2
0 1000 2000 3000 4000 5000 6000 7000
Figure 4.3.: Estimated states using the observer for perturbations in gas tur-
bine electrical output power
47
Pressue P (bar) Flow rate qs (Kg/s)
5.8 10
5.6 8
5.4
6
5.2
4
5
0 2000 4000 6000 0 2000 4000 6000
Level l (mm) Flow rate qf (Kg/s)
200
100 15
0 10
100 5
200
0
0 2000 4000 6000 0 2000 4000 6000
Figure 4.4.: Comparison between state observer (dashed) and plant (solid line)
for perturbations in gas turbine electrical output power
Pressure (bar) Flow rate qs (Kg/s)
5.6
8
5.5
5.4 7
5.3
6
0 1000 2000 3000 0 1000 2000 3000
Level (mm) Flow rate qf (Kg/s)
60
12
40
20 10
0 8
20
6
40
0 1000 2000 3000 0 1000 2000 3000
to 10 MW
48
Pressure P (bar) Flow rate qs (Kg/s)
8
5.6
5.4 6
5.2 4
0 1500 3000 4500 0 1500 3000 4500
100 15
50
10
0
50
5
100
0 1500 3000 4500 0 1500 3000 4500
to 20 MW
Pressure (bar) Flow rate qs (kg/s)

5.8
10
5.6
8
5.4
6
5.2
4
5
0 2000 4000 6000 0 2000 4000 6000
Level (mm) Flow rate qf (kg/s)
15
100
0 10
100 5
200
0
0 2000 4000 6000 0 2000 4000 6000
Figure 4.7.: Comparison between PI-based state-feedback controller (dashed)

and plant (solid line) for perturbations in gas turbine electrical
output power
49
Pressure (bar) Flow rate qs (kg/s)

10
5.5
5.45
5
5.4
5.35 0
0 500 1000 1500 0 500 1000 1500
12
0 10
8
50
6
4
100
2
0 500 1000 1500 0 500 1000 1500
Figure 4.8.: Model closed loop response using the PI-based state-feedback con-
troller for high (dashed), medium (solid line) and low load (dotted
dashed)
Flow rates qf,qs (kg/s) Water Volume Vwt (m3)

10
20.8
9
20.6
8
20.4
7 20.2
0 500 1000 1500 0 500 1000 1500
Steam quality (%) Steam bubbles volume Vsd (m3)
0.016
1.3
0.014 1.2
1.1
0.012
1
0.01 0.9
0 500 1000 1500 0 500 1000 1500
Figure 4.9.: States and input variables behaviour using the PI-based state-
feedback controller at medium load
50
Set value tracking Flow rates
set value
Existing PI Mass flow

-I Plant
controllers control valves
Drum-boiler
model
Observer-based
State-feedback controller
-F
Figure 4.10.: Block diagram of the proposed multivariable feedback control strategy
51
5. Conclusion and future work
5. Conclusion and future work

A mathematical nonlinear model which describes the dynamical process of
steam generation using steam drum-boiler units including its control valves
and process PID-controllers was fitted into to a real drum-boiler unit which
corresponds to 450 MW CHP in Munich, in order to analyse the pressure and
water level control performance which was reported to behave very poorly
during transients corresponding to huge load changes.
The model was implemented within MATLAB/Simulink environment and
examined intensively throughout various scenarios with very rich excitation
from the plant covering a wide operating range to ensure its validity and reli-
ability. Further it pointed out very clearly the main drawbacks of the existing
control strategy employed to stabilize the process.
Stability analysis was conducted by linearization at typical operating points
of the drum-boiler unit. It predicts the plant open loop response quite well
and clarifies the reason behind its non-minimum behaviour which is associated
with the steam bubbles shrink/swell physical phenomena.
A multivariable feedback control strategy is proposed in order to optimize the
process using a PI-based state-feedback controller designed using LQ method
ensuring steady-state accuracy and set value tracking. Additionally an ob-
server gain which guarantees correct estimation of the state variables required
for feedback is realized using pole placement method. Simulation results shows
that the state-feedback controller outperforms the PID-control in terms of con-
trol behaviour and performance.
Unfortunately the complete control structure which combines both the state-
feedback and observer together cannot be examined at the moment within the
simulation environment because the observer gains requires new measurements
from the plant while being handled by the proposed control strategy.
In the near future, the nonlinear model shall be realized within the real
plant Distributed Control System (DCS) of GuD 2 at HKW Süd to act as
an observer of the process, thus offering in return a great opportunity to test
and examine the model more closely before being combined with the suggested
state-feedback controller.
52
A. Appendix
A. Appendix
A.1. Nomenclature
Symbol Unit Description

A m2 Area
Cp J/kg K Metal specific heat capacity
h J/kg Specific enthalpy
Kv kg/hr Valve sizing coefficient
m kg Mass
P Pascal Pressure
Q̇ W Heat flow rate
q kg/s Mass flow rate
Tsat ◦C Saturation temperature
V m3 Volume
V◦ m3 Volume in hypothetical situation
Td s Residence time of steam in drum
ρ kg/m3 Density
Table A.1.: Physical units
Symbol Description Symbol Description

K Friction coefficient c Condensation
m Head loss d Drum
x Valve opening percentage dc Downcomer
Z Compressibility factor fw Feedwater
αr Steam-mass fraction r Riser
α¯v Steam-volume fraction t Total
β Empirical coefficient s Steam
ζ Normalized length w Water
Table A.2.: Dimensionless units Table A.3.: Subscripts
53
A.2. MATLAB Control System Toolbox
A.2. MATLAB Control System Toolbox

Control System ToolboxTM offers various functionality to design, analysis and
tuning of linear controllers. In this section, the tools adopted in the thesis will
be briefly featured.
A.2.1. Linear analysis functions

The embedded functions can be utilized using the graphical user interface ”LTI
viewer” or MATLAB command window.
linio defines the linearization input/output points

operspec specifies operating point requirements for states, inputs
and outputs
findop computes steady-state operating point meeting predefined
specifications
linearize performs linear approximation of a non-linear model
pzplot computes poles and zeros of a dynamic linear system and
plot them in the complex plane
tzero computes invariant zeros of a linear MIMO system
A.2.2. Controller design functions

The functions are only accessible using MATLAB command window.
ss creates state-space model given the system matrices

eig computes eigenvalues for a system
ctrb computes the controllability matrix for state-space model
obsv computes the observability matrix for state-space model
place places the desired closed-loop poles at a desired position in
the complex plane
lqr computes an optimal state-feedback controller given the state-
space model and weighting matrices
lqi computes an I-based optimal state-feedback controller
54
A.3. MATLAB script
A.3. MATLAB script

A.3.1. Drum-boiler model
The following function implements the state algebraic
equations (2.2.2) to cal-
dP dVwt dαr dVsd
culate the states derivatives ẋ = dt , dt , dt , dt and the level l.
function y = DrumBoiler SWM (Q,qf,qs,P,Vwt,Alpha,Vsd)

%% Model inputs
% Q = Amount of heat flow rate added to the system (Watt)
% qf = Feedwater flow rate (Kg/s)
% qs = Steam flow rate (Kg/s)
% P = Pressure (Pascal)
% Vwt = Water total volume (m3)
% Alpha = Steam quality (%)
% Vsd = Steam bubbles volume under water level (m3)
%% Drum−boiler parameters and construction data

Vd = 20.204; %Drum volume (m3)
Vr = 20; %Drum riser volume (m3)
Vdc = 0.9; %Drum downcomer volume (m3)
Vt = Vd + Vr + Vdc; %Total drum volume (m3)
Ad = 14.7; %Drum area (m2)
Adc = 0.0637; %Downcomer area (m2)
mr = 1300; %Riser mass (Kg)
md = 1363; %Drum mass (Kg)
mt = mr+md+98888; %Total metal mass (kg)
K = 25; %Friction coefficient in downcomer
Td = 3; %Residence time of steam in drum(s)
Beta = 0.3 ; %Empirical coefficient
Vsd0 = 2; %Steam bubbles volume in the hypothetical situation (m3)
Cp = 550; %Metal specific heat capacity (Pascal.m3/Kg.K)
%% Liquid/Vapour mixture properties

P = P*1e−5; %Pascal to Bar
%% Temperature
Tfw = 104; %Feedwater (C)
T Sat = XSteam('Tsat p',P); %Saturation (C)
dT Sat dP = IAPWS IF97('dTsatdpsat p',P*0.1) * 1e−6; %(K/Pa)
%% Density
rhoV = XSteam ('rhoV p',P); %Steam (Kg/m3)
rhoL = XSteam ('rhoL p',P); %Water (Kg/m3)
%Partial derivative with pressure
drhoL dP = (2*P*0.0148 − 3.7836) * 1e−5; %Water (Kg/J)
drhoV dP = (2*P*0.0010 + 0.4450) * 1e−5; %Steam (Kg/J)
%% Specific Enthalpy
hfW = XSteam('hL T',Tfw) *1e3; %Feedwater (J/Kg)
55
A.3. MATLAB script
hL = XSteam ('hL p',P) * 1e3; %Water (J/Kg)

hV = XSteam ('hV p',P) * 1e3; %Steam (J/Kg)
hC = hV−hL; %Condensation (J/Kg)
%Partial derivative with pressure
dhL dP = IAPWS IF97('dhLdp p',P*0.1) * 1e−3; %Water (J/Kg.Pa)
dhV dP = IAPWS IF97('dhVdp p',P*0.1) * 1e−3; %Steam (J/Kg.Pa)
%% Coefficients Values
Eta = (Alpha*(rhoL−rhoV))/rhoV;
AlphaV = (rhoL / (rhoL−rhoV)) * (1 − ((rhoV/((rhoL−rhoV)*Alpha)) ...
* log(1+(((rhoL−rhoV)*Alpha)/rhoV))));
dAlphaV dP = (1/((rhoL−rhoV)ˆ2))*(rhoL*drhoV dP − ...
rhoV*drhoL dP)*(1 + (rhoL/(rhoV *(1+Eta))) − ...
(((rhoV+rhoL)*log(1+Eta))/(rhoV*Eta)));
dAlphaV dAlpha = (rhoL/(rhoV*Eta))*(((1/Eta)*log(1+Eta)) − ...
(1/(1+Eta)));
qdc = sqrt((2*rhoL*Adc*(rhoL−rhoV)*9.81*AlphaV*Vr)/K);
Vwd = Vwt − Vdc − (1−AlphaV)*Vr;
%% State equations coefficients

e11 = rhoL − rhoV;
e12 = Vwt*drhoL dP + (Vt−Vwt)*drhoV dP;
e21 = rhoL*hL − rhoV*hV;
e22 = Vwt*(hL*drhoL dP + rhoL* dhL dP) + (Vt−Vwt)*(hV*drhoV dP + ...
rhoV* dhV dP) − Vt + mt*Cp* dT Sat dP;
e32 = (rhoL* dhL dP − Alpha*hC*drhoL dP)*(1−AlphaV)*Vr + ...
((1−Alpha)*hC*drhoV dP + rhoV* dhV dP)*AlphaV*Vr + (rhoV + ...
(rhoL−rhoV)*Alpha)*hC*Vr*dAlphaV dP − Vr + mr*Cp* dT Sat dP;
e33 = ((1−Alpha)*rhoV + Alpha*rhoL)*hC*Vr*dAlphaV dAlpha;
e42 = Vsd*drhoV dP + (1/hC)*(rhoV*Vsd* dhV dP + rhoL*Vwd* dhL dP − ...
Vsd − Vwd + md*Cp* dT Sat dP) + ...
Alpha*Vr*(1+Beta)*(AlphaV*drhoV dP + (1−AlphaV)*drhoL dP + ...
(rhoV−rhoL)*dAlphaV dP);
e43 = Alpha*(1+Beta)*(rhoV−rhoL)*Vr*dAlphaV dAlpha;
e44 = rhoV;
%% State variables
dP dt = (1/(e11*e22 − e12*e21))*(e11*Q + qf*(hfW*e11 − e21) + ...
qs*(e21 − hV*e11));
dVwt dt = (1/(e11*e22 − e12*e21))*(qf*(e22 − hfW*e12) + ...
qs*(hV*e12 − e22) − e12*Q);
dAlpha dt = (1/e33)*(Q − Alpha*hC*qdc − e32* dP dt);
dVsd dt = (1/e44)*(((rhoV/Td)*(Vsd0−Vsd)) + ((hfW − hL)*qf/hC) − ...
e42* dP dt − e43*dAlpha dt);
Level = (Vwd+Vsd)/Ad −0.875;
%% Model outputs
y = [ dP dt dVwt dt dAlpha dt dVsd dt Level ];
end
56
A.3. MATLAB script
A.3.2. Controller design

The following m.file was used to obtain a linearized model in state-state form at
a common operating point to compute the state-feedback and observer matrices
gains using LQ and pole placement methods.
% Specify the model name
model = 'Drum Model';
% Create the linearization I/O as specified in Linearize Model

ios(5) = linio('Drum Model/Drum model',2,'out');
ios(4) = linio('Drum Model/Drum model',1,'out');
ios(3) = linio('Drum Model/qs (Kg//s)',1,'in');
ios(2) = linio('Drum Model/qf (Kg//s)',1,'in');
ios(1) = linio('Drum Model/Q (MW)',1,'in');
% Create the operating point specification object

opspec = operspec(model);
% Set the constraints in the model

opspec.Inputs(2).u = 9; % qf (Kg/s)
opspec.Inputs(2).Known = true;
opspec.Outputs(1).y = 5.5; % P (bar)
opspec.Outputs(1).Known = true;
opspec.Outputs(2).y = 0; % l (mm)
opspec.Outputs(2).Known = true;
% Perform the operating point search

% Linearize the model
op = findop(model,opspec,opt);
sys = linearize(model,op,ios);
% Create state−space model excluding Q(MW)

Drum ss = ss(sys.A,sys.B(:,2:3),sys.C,sys.D(:,2:3));
% Assign weighting matrices Q and R

% Assign pole position for the observer
Q = [ 1.23e−12 0 0 0 0 0;
0 1 0 0 0 0;
0 0 1 0 0 0;
0 0 0 0.25 0 0;
0 0 0 0 5 0;
0 0 0 0 0 1e−4 ];
R = [ 50 0; 0 150 ];
P = [ −0.03 −0.03 −0.2 −0.9];
% Compute optimal state−feedback controller K and observer gain L

K = lqi(Drum ss,Q,R);
L = place(Drum ss.A',Drum ss.C',P)';
57
A.4. Heat engines
A.4. Heat engines

Heat engine plays an essential role in electrical power generation as they con-
vert the thermal energy into mechanical energy required to drive the generator
shaft producing electricity.
The heat engine can be a closed or open loop system, which involves typically
four thermodynamic basic processes shown in figure (A.1). It converts the state
of the working fluid into another before returning it to its original state. The
processes are compression, heat addition, expansion and heat rejection, each
can be carried out under one or more of the following conditions:
Isothermal At constant temperature
Isobaric At constant pressure
Isometric / Isochoric / Iso-volumetric At constant volume
Adiabatic At constant entropy, no heat is added or removed from the
system and no work done.
Isentropic At constant entropy, reversible adiabatic conditions
Heat addition
Compression Closed system Expansion
Heat rejection
Fluid Exhaust
heat Open system
Figure A.1.: Heat engine typical closed/open loop heat cycle
A.4.1. Brayton cycle

The cycle shown in figure (A.2) is mathematical model used describe the
thermodynamics and heat cycle for the operation of the gas turbine.
58
A.4. Heat engines
Process 1-2 Fresh air is being supplied for an open cycle, as for a closed
one it’s drawn back from the turbine to a compressor increasing its pres-
sure in an adiabatic compression process.
Process 2-3 The compressed air is mixed with fuel or natural gas before
being burned inside the combustion chamber at constant pressure.
Process 3-4 The heated pressurized air is supplied to the heat engine,
where it’s allowed to expand through the turbine driving its blades, in
an adiabatic expansion process.
Process 4-1 Finally, heat rejection to the surrounding atmosphere takes
place at constant pressure.
P T 3
q in
t.
ns
co
2 3 q in
=
p
s 4
s=
=
co
ns
co
t.
ns
t.
1 4 2
q out ns
t.
1 = co q out
p
P-v Diagram v T-s Diagram s
Figure A.2.: T-S and P-V diagram of a typical ideal Brayton cycle
A.4.2. Rankine cycle

The cycle shown in figure (A.3) is a mathematical model used to describe
closed cycle heat conversion into mechanical energy using two phase working
fluids that drive a steam turbine blades producing electricity.
Process 1-B High pressure water is pre-heated at constant pressure at
the economizer stage, until it reaches its boiling point converting it to
water-vapour mixture.
Process B-2 A second heating phase takes place using evaporator and
superheater, to convert the converting the mixture into superheated steam.
Process 2-3 The vapour at high pressure and temperature enters the
turbine, where vapour energy is converted into mechanical work driving
the turbine blades.
59
A.5. Non-minimum phase systems
Process 3-4 Vapour leaving the turbine at low pressure and temperature
is condensed, converting it into wet saturated water.
Process 4-1 Saturated water is pumped back, feeding the boiler at high
pressure, where the cycle is repeated.
Figure A.3.: T-S diagram of a typical ideal Rankine cycle
A.5. Non-minimum phase systems

A plant whose poles and zeros are real numbers and located within the left-
hand side (LHS) of the complex plane is known as a minimum phase system.
This is due to the fact that the phase shift have a minimum range restricted
within 0◦ to −90◦ degrees for a given amplitude response when being examined
in frequency domain. If a stable plant have one or more zeros in the right hand
side (RHS) of the complex plane, then phase shift range is always greater
than −90◦ . Such systems are known as non-minimum phase, where an inverse
response always exists, leading to an initial overshoot or undershoot delaying
the output behaviour.
Assume a Single-Input, Single-Output (SISO) plant whose closed loop trans-
fer function is assumed to be internally stable and given by equation (A.1), only
one zero lies in RHS of the complex plane for the sake of simplicity. A step in-
put is applied to the plant closed loop and the output Y (s) is given by equation
(A.2). Due to the assumption that the plant is internally stable, the open-loop
zero z0 lies within of the region of convergence (ROC) of Y (s), which yields
the unilateral laplace transform to (A.3).
60
A.6. Integral anti-windup control
Since the output signal has an initial value y(0) = 0, final value y(∞) = 1
and its area under the curve evaluated by the integral in equation (A.3) is equal
to 0, then this implies that the output signal y(t) must take negative values
over time.
Y (s) (s − z0 )
Gcl (s) = = (A.1)
R(s) s + p0
(s − z0 )
Y (s) = Gcl (s)R(s) = (A.2)
s(2s + p0 − z0 )
∞
(z0 − z0 )
y(t)e−z0 t dt = Y (s)|s=z0 = Y (z0 ) = =0 (A.3)
0 s(2s + z0 − z0 )
Figure (A.4) illustrates the step and bode responses of a simple 2nd order
minimum and non-minimum phase plant. The zero of the first plant lies in the
LHS leading to the normal expected step response, however, as it’s is shifted to
the RHS for the second plant, there exists an undershoot in the initial response
and delay in overall response, caused by the phase shift difference as seen in
the bode plot.
Step response Bode Diagram

1.2
1
Magnitude (dB)
0.8 20
0.6 40
0.4 180
135
Phase (deg)
0.2 90
45
0 0
45
0.2 90
2 1 0 1 2
0 5 10 15 20 10 10 10 10 10
Figure A.4.: Step response and bode plot for a minimum (solid line) and a
non-minimum (dashed) system

Most PID-controllers in practical applications are equipped with a nonlinear
saturation element which saturates the controller output once it attains a cer-
tain values, imposed by the physical limitations of the actuators. The nonlin-
61
earity might lead to integral windup.

Such occurring phenomena take place when the PID-controller integral ele-
ment builds up and accumulates the error signal even if the controller output
is saturated, it might degrade the controller performance or even lead to closed
loop instability if neglected.
Consider a simple linear motor positioner modeled as an integrator, whose
input and output are velocity and position respectively. Clearly, the motor ve-
locity will be physically limited according to its type and manufacturer, there-
fore the equipped PI-controller output should be limited when used to control
the closed loop. The controller parameters weren’t tuned since this is merely an
explanatory example focused on effects of windup phenomena and saturation
limit is set to be ±0.3 ms associated with the motor allowable maximum speed.
Position (m) Velocity (m/s)

0.4
4
0.2
3
0
2
0.2
1
0 0.4
0 10 20 30 40 50 0 10 20 30 40 50
Figure A.5.: Motor velocity and position behaviours with (dashed) and without
(solid line) anti-windup
Figure (A.6) illustrates the closed-loop behaviour of the motor positioner

while equipping an anti-windup mechanism using back-calculation method and
compares it when no anti-windup is utilized. The position set point changed
to 3 m, due to this large switch, the controller tries to track the reference value
as fast as possible, however it was limited by the motor velocity upper limit.
Without anti-windup, the integrator element output keeps growing and the
motor position would require more time to reach the steady-state.
On the other hand, while equipping an anti-windup, the back-calculation gain
starts discharging the PI-controller integrators and prevents it from building up
once the controller output is saturated, therefore vastly improving the output
settling time while eliminating undesired overshoots.
62
A.7. Drum-boiler state equations coefficients
A.7. Drum-boiler state equations coefficients
∂ρw ∂ρs
e11 = Vwt + Vst
∂p ∂p
e12 = ρw − ρs

∂ρw ∂hw ∂ρs ∂hs ∂tsat
e21 = Vwt hw + ρw + Vst hs + ρs − Vt + mt Cp
∂p ∂p ∂p ∂p ∂p
e22 = ρw hw − ρs hs

∂hw ∂ρw ∂hs ∂ρs
e31 = ρw − α r hc (1 − ᾱv )Vr + ρs + (1 − αr )hc ᾱv Vr +
∂p ∂p ∂p ∂p
∂ ᾱv ∂tsat
(ρs + (ρw − ρs )αr ))hc Vr − Vr + mr Cp
∂p ∂p
∂ ᾱv
e33 = ((1 − αr )ρs + αr ρw )hc Vr
∂p

∂ρs ∂ρs ∂ρw ∂ ᾱv
e41 = Vsd + αr (1 + β)Vr ᾱv + (1 − ᾱv ) + (ρs − ρw ) +
∂p ∂p ∂p ∂p

1 ∂hs ∂hw ∂tsat
ρs Vsd + ρw Vwd − Vsd − Vwd + md Cp
hc ∂p ∂p ∂p
∂ ᾱv
e43 = αr (1 + β)(ρw + ρs )V r
∂p

2ρw Adc (ρw − ρs )g ᾱv Vr
qdc =
K
h w − hf w 1 ∂hw ∂hs ∂tsat dP
qct = qf + ρw Vwt + ρs Vst − Vt + mt Cp
hc hc ∂p ∂p ∂p dt

∂ρs ∂ρw ∂ ᾱv dP ∂ α¯v dαr
qr = qdc − Vr ᾱv + (1 − ᾱv ) + (ρw − ρs ) + (ρw − ρs )Vr
∂p ∂p ∂p dt ∂αr dt

ρw ρs ρ w − ρs
ᾱv = 1− ln 1 + αr
ρ w − ρs (ρw − ρs )αr ρs
(ρw − ρs )
ζ = αr
ρs

∂ ᾱv ρw ln(1 + ζ) 1
= −
∂αr ρs ζ ζ 1+ζ

∂ ᾱv 1 ∂ρs ∂ρw ρw ρs + ρw
= ρ w − ρ s 1 + − ln(1 + ζ)
∂p (ρw − ρs )2 ∂p ∂p ρs (1 + ζ) ζρs
63
A.8. Operator interface
Figure A.6.: Screenshot of the drum-boiler unit in the real process DCS
64
Figure A.7.: Screenshot of the low pressure steam distribution network in the
real process DCS
65
B. List of Figures
B. List of Figures
2.1. Combined cycle working principle . . . . . . . . . . . . . . . . . 9

2.2. Heat Recovery Steam Generator (HRSG) [14] . . . . . . . . . . 10
2.3. Schematic diagram of the low pressure steam generation process 11
2.4. Schematic diagram of the downcomer-riser circulation loop [13] 12
2.5. Flow through control valve for liquid service [22] . . . . . . . . 15
2.6. Inherent flow characteristics of typical control valves [24] . . . . 16
2.7. Simulink model of the Drum-boiler unit . . . . . . . . . . . . . 19
2.8. Simulink model of the control valve combined with its actuator 21
2.9. Simulink model of a parallel PID-controller equipped with an
anti-windup mechanism . . . . . . . . . . . . . . . . . . . . . . 22
3.1. Pole-zero plot of the linearized models at low (1), medium (2)
and high (3) load . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2. Open loop response for a step change equivalent to decrease of
20 MW of the gas turbine electrical output power . . . . . . . . 31
3.3. Open loop response for a step change equivalent to 10% opening
of butterfly valve position . . . . . . . . . . . . . . . . . . . . . 31
3.4. Open loop response for a step change equivalent to 10% closing
of feedwater control valve position . . . . . . . . . . . . . . . . 32
3.5. Simulink validation model . . . . . . . . . . . . . . . . . . . . . 34
3.6. Comparison between model (dashed) and plant data (solid line)
for a decrease of the gas turbine electrical output power equiva-
lent to 20 MW . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.7. Comparison between model (dashed) and plant data (solid line)
for an increase of the gas turbine electrical output power equiv-
alent to 10 MW . . . . . . . . . . . . . . . . . . . . . . . . . . . 35
3.8. Comparison between model and plant data (solid line) for a de-
crease of the gas turbine electrical output power equivalent to
20 MW when Kp = 1.8 (dotted dashed) and Kp = 5 (dashed) . 37
4.1. State estimation using Luenberger observer . . . . . . . . . . . 39

4.2. PI-based state-feedback control structure . . . . . . . . . . . . 40
66
B. List of Figures
4.3. Estimated states using the observer for perturbations in gas tur-
bine electrical output power . . . . . . . . . . . . . . . . . . . . 47
4.4. Comparison between state observer (dashed) and plant (solid
line) for perturbations in gas turbine electrical output power . 48
line) for a decrease of the gas turbine electrical output power
equivalent to 10 MW . . . . . . . . . . . . . . . . . . . . . . . . 48
line) for a decrease of the gas turbine electrical output power
equivalent to 20 MW . . . . . . . . . . . . . . . . . . . . . . . . 49
4.7. Comparison between PI-based state-feedback controller (dashed)
and plant (solid line) for perturbations in gas turbine electrical
output power . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.8. Model closed loop response using the PI-based state-feedback
controller for high (dashed), medium (solid line) and low load
(dotted dashed) . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.9. States and input variables behaviour using the PI-based state-
feedback controller at medium load . . . . . . . . . . . . . . . . 50
4.10. Block diagram of the proposed multivariable feedback control
strategy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
A.1. Heat engine typical closed/open loop heat cycle . . . . . . . . . 58

A.2. T-S and P-V diagram of a typical ideal Brayton cycle . . . . . 59
A.3. T-S diagram of a typical ideal Rankine cycle . . . . . . . . . . 60
A.4. Step response and bode plot for a minimum (solid line) and a
non-minimum (dashed) system . . . . . . . . . . . . . . . . . . 61
A.5. Motor velocity and position behaviours with (dashed) and with-
out (solid line) anti-windup . . . . . . . . . . . . . . . . . . . . 62
A.6. Screenshot of the drum-boiler unit in the real process DCS . . 64
A.7. Screenshot of the low pressure steam distribution network in the
real process DCS . . . . . . . . . . . . . . . . . . . . . . . . . . 65
67
C. List of Tables
C. List of Tables
2.1. Drum-boiler model parameters . . . . . . . . . . . . . . . . . . 20

2.2. Control valve and actuator parameters . . . . . . . . . . . . . . 21
2.3. PID-controller parameters . . . . . . . . . . . . . . . . . . . . . 22
3.1. Drum-boiler operating points for low, medium and high load . 27
A.1. Physical units . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53

A.2. Dimensionless units . . . . . . . . . . . . . . . . . . . . . . . . . 53
A.3. Subscripts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
68
D. Bibliography
D. Bibliography
[1] Asea Brown Boveri: Drum Level Control Systems in the Process Industries,
1997. ABB Download Center.
[2] Åström, Karl Johan und Rodney D. Bell: Drum Boiler Dynamics. Auto-
matica, 36:363–378, März 2000.
[3] Åström, K.J. und T. Hägglund: PID Controllers - Theory Design and
Tuning. International Society of America, 1995, ISBN 9781556175169.
[4] Ellis, G.: Observers in Control Systems - A Practical Guide. Academic

Press, 2002, ISBN 9780122374722.
[5] Emami, A. und P. Van Dooren: Computation of zeros of linear multiva-

riable systems. Automatical, 18:412–430, 1982.
[6] Falb, P. L. und W. A. Wolovich: On the decoupling of multivariable sys-

tems. Proc. JACC, Philadelphia, Pennsylvanial, 41:791–796, 1967.
[7] Flynn, M.E. und M.J. O Malley: A drum Boiler Model for Long Term
Power System Dynamic Simulation. IEEE Transaction Power System,
14(1):209–217, 1999.
[8] G., Westner und Madlener R.: Development of Cogeneration in Germany:

A Dynamic Portfolio Analysis Based on the New Regulatory Framework.
FCN Working Paper, 2009.
[9] Goodwin, G. C., S. F. Graebe und M. E. Salgado: Control System Design.

Prentice Hall, Upper Saddle River, New Jersey, international Auflage,
2001.
[10] Holmgren, Magnus: X Steam, Thermodynamic properties of water and

steam, 2006. MATLAB Central File Exchange.
[11] Kautsky, J., N.K. Nichols und P. Van Dooren: Robust Pole Assignment
in Linear State Feedback. International Journal of Control, 41:1129–1155,
1985.
69
D. Bibliography
[12] Kehlhofer, R., B. Rukes, F. Hannemann und F.X. Stirnimann: Combined-

Cycle Gas and Steam Turbine Power Plants. PennWell, 2009,
ISBN 9781593701680.
[13] Kim, H. und S. Choi: A model on water level dynamics in natural circula-
tion drum-type boilers. International Communications in Heat and Mass
Transfer, 32:786 – 796, 2005.
[14] Kim, T.S., D.K. Lee und S.T. Ro: Analysis of thermal stress evolution
in the steam drum during start-up of a heat recovery steam generator.
Applied Thermal Engineering, 20(11):977 – 992, 2000, ISSN 1359-4311.
[15] Levine, William S.: The control handbook. The electrical enginee-
ring handbook series. CRC Press New York, Boca Raton (Fl.), 1996,
ISBN 0-8493-8570-9.
[16] Luenberger, D. G.: Observing the State of a Linear System. IEEE Tran-
sactions on Military Electronics, 8:74–80, 1964.
[17] Luenberger, D. G.: Observers for multivariable systems. IEEE Transacti-
ons on Automatic Control, 11:190–197, 1966.
[18] Michels, K.: Regelungstechnik (Vorlesungsmanuskript). Institut für Auto-
matisierungstechnik, Universität Bremen, 2013.
[19] Mikofski, Mark: IAPWS IF97 functional form with no slip, 2012. MAT-
LAB Central File Exchange.
[20] Moran, M.J. und H.N. Shapiro: Fundamentals of engineering thermodyna-
mics. John Wiley and Sons Inc., New York, NY, 2009.
[21] Parry, A., Petetrot J. F. und M. J Vivier: Recent progress in sg level control
in french pwr plants. British Nuclear Energy Society, Seiten 81–88, 1995.
[22] Samson AG: Application Notes for Valve Sizing, 2012. Samson Product
Documentation.
[23] Skogestad, Sigurd und Ian Postlethwaite: Multivariable Feedback Control:
Analysis and Design. John Wiley & Sons, 2005, ISBN 0470011688.
[24] Spirax-sarco: The Steam and Condensate Loop Book, 2011.
[25] Triantafyllou, Michael S. und Franz S. Hover: Maneuvering and Control of
Marine Vehicles (Lecture Notes). Massachusetts Institute of Technology,
2003. MIT OpenCourseWare.
70

Drum-Boiler Control Performance

Uploaded by

Copyright:

Available Formats

Drum-Boiler Control Performance

Uploaded by

Document Information

Original Description:

Original Title

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

Drum-Boiler Control Performance

Uploaded by

Copyright:

Available Formats

Master Thesis

Drum-boiler control performance

Tutor: 1st Examiner:

Keywords: drum-boiler level control; optimal control; multivariable feedback

3. Process analysis and validation 23

3.4.2. Comparison with measurement data . . . . . . . . . . . 33

5. Conclusion and future work 52

2.1. Combined cycle process overview

2 General Electric gas turbine units producing a total of 278 MW

Low pressure water Waste heat

Heat Recovery Steam Generator

Figure 2.1.: Combined cycle working principle

2.1.1. Gas turbine

2.1.2. Heat recovery steam generator

is preheated in order to reduce energy consumption.

Figure 2.2.: Heat Recovery Steam Generator (HRSG) [14]

2.1.3. Steam turbine

2.1.4. Surface condenser

2.2. Steam generation process description

2.2.1. Drum-boiler mass and energy balance

Feedwater Sat. water

Upper collecting header

Lower distribution header

Figure 2.4.: Schematic diagram of the downcomer-riser circulation loop [13]

of the temperature distribution within the metal isn’t necessary.

2.2.2. Drum-boiler nonlinear state equations

dP e12 Q̇ + qf (e12 hf w − e22 ) − qs (e22 − e12 hs )

2.2.3. Mass ﬂow control valve

undoubtedly the butterﬂy valve would supply approximately one-third of the

Figure 2.6.: Inherent ﬂow characteristics of typical control valves [24]

2.2.4. Process PID-controller

The controller parameter Kp corresponds to the proportional gain element,

Water level and pressure control

2.3. MATALB/Simulink model

2.3.1. Drum-boiler model

Figure 2.7.: Simulink model of the Drum-boiler unit

The liquid/vapour mixture properties are calculated using functions of water

constant or scaled down relative to the Åström-Bell model6 .

Drum-boiler construction data

Table 2.1.: Drum-boiler model parameters

2.3.2. Control valve and actuator model

Pressure drop qf (kg/s) 1

Control valves and actuators

Table 2.2.: Control valve and actuator parameters

2.3.3. Process PID-controller model

The anti-windup can be handled using the back-calculation method or a log-

Derivative gain Filter coefficient

Figure 2.9.: Simulink model of a parallel PID-controller equipped with an anti-

Table 2.3.: PID-controller parameters

3. Process analysis and validation

3.1. Theoretical overview

Obviously nonlinear systems positions of rest - referred to from now on as

tion (3.1), hence requires a more generalized deﬁnition oﬀered by Lyapunov

3.1.3. Poles and zeros

N (s) (s − z1 )(s − z2 )...(s − zn )

all of the system eigenvalues necessary appear as poles due to existence of

3.2. Stability analysis

3.2.1. Linear state-space model