Demystifying PID

Controller Tuning for

Performance and Robust
Stability
Jeff Arbogast
Standards
Certification
American Air Liquide
Education & Training
Publishing
Newark, DE
Conferences & Exhibits
Presenter

Jeff Arbogast, Ph.D.

Research Scientist, Process Control & Logistics R&D
American Air Liquide
Newark, DE

• Ph.D. in Chemical Engineering -

University of Connecticut The world leader in gases
– Advisor: Prof. Doug Cooper for industry, health, and the
– Dissertation focused on Intuitive environment.
Analysis of Robust Stability
• Air Liquide project manager for:
Delaware Research & Technology Center
– Alarm Management
– Human-Machine Interface Design
– Fault Detection & Analysis
• Called upon within Air Liquide for
expertise in Control Loop Tuning

2
Outline

• Model-based PID Controller Tuning

– First Order plus Dead Time (FOPDT) Linear Model
– Internal Model Control (IMC) Tuning Correlations
• Introduction to Robustness
– What do Robust Stability and Robust Performance mean?
– Understanding Robust Stability with Plots and Metrics
– Application for two Industrial Pressure Control Loops
• Application for Industrial Blowdown Drum Level Control
(an Integrating Process)
• Novel Contributions (Arbogast, 2008)
– Quantitative Robust Stability Factor Metric (RSF2D and RSF3D)
– Plant-Model Mismatch in Process Time Constant (τP)
Significantly Impacts Robust Stability (RSF3D < RSF2D)
3
Modeling Dynamic Process Behavior

• When a First Order Plus Dead Time (FOPDT) linear model is fit to dynamic
process data

dPV (t )
τP + PV (t ) = K P ⋅ OUT (t − θ P )
dt

• The important parameters that result are:

– Steady State Process Gain, KP (How Far)
– Overall Process Time Constant, τP (How Fast)
– Apparent Dead Time, ӨP (With How Much Delay)

4
 Example 1 - The Process is Nonlinear
... Consider the Level of Operation

Level of Operation: Level of Operation: Level of Operation:

OUT = 35% OUT OUT = 18.25% OUT OUT = 18.25% OUT
PV = 205 psig PV = 205 psig PV = 205 psig
direction: ↑ & ↓ direction: ↓ direction: ↑

Model Parameters: Model Parameters: Model Parameters:

KP = 25.9 psig / %OUT KP = 39.8 psig / %OUT KP = 30.6 psig / %OUT
τP = 0.13 min τP = 0.45 min τP = 0.28 min
θP = 0.058 min θP = 0.033 min θP = 0.16 min
Example 1 - Why is this nonlinearity
important?

• Control Loop Tuning Unchanged …but Performance Changes

• Performance on the Step Down (left)
– Slight overshoot of SP (~ 20%)
– No second peak beyond SP (no decay ratio)
• Performance on the Step Up (right)
– Significant overshoot of SP (~ 50%)
– High second peak beyond SP (decay ratio of ~ 75%)
PID Controller Form

• The Ideal Form is one form of the Proportional-Integral-

Derivative (PID) Controller

⎡ 1 dPV (t ) ⎤
OUT (t ) = bias + K C ⎢e(t ) + ∫ e(t )dt − τ D ⎥
⎣ τ I dt ⎦
where e(t ) = SP (t ) − PV (t )

• PID Controller Tuning Parameters

– Controller Gain, KC
– Reset Time, τI
– Reset Rate, τD
• Important - Know the form of your PID controller!

7
IMC / SIMC Tuning

• Internal Model Control (IMC) Tuning introduced by

Rivera, et al. (1986)
• Closed-Loop Time Constant, τC
– Single Adjustable Parameter in IMC / SIMC Tuning Correlations
– Specifies Performance and Robust Stability
• IMC-PI Tuning Correlation:
1 τP
KC = ; τI =τP
K P (θ P + τ C )

• Skogestad IMC (SIMC) PI Tuning (2003):

(especially useful for time constant dominant processes)

1 τP
KC = ; τ I = min ⎡⎣τ P , 4 (τ C + θ P ) ⎤⎦
K P (θ P + τ C )
8
Introduction to Robustness

• What is Controller Robustness?

– Controller Robustness measures the Ability to Tolerate
Variations in Process Behavior (e.g., Nonlinearity)
– Robust Stability and Robust Performance are of interest

• Variation in terms of Plant-Model Mismatch:

a = KP,Plant / KP,Model
b = τP,Plant / τP,Model
c = θP,Plant / θP,Model

9
Introduction to the Robust Stability
Factor (RSF) Metric
• Robustness Plot introduced by 16
Gerry and Hansen (1987)

Plant-Model Mismatch in Gain

• Generally applies to a

a = KP,Plant / KP,Model
12
Moderately tuned PI controller
using the IMC tuning
correlation 8
Square Lies Entirely
Within Stable Region

• Dot (a = 1, c = 1) indicates
perfect plant-model match 4
UNSTABLE
RSF2D REGION
3.75
• Robust Stability Factor, RSF2D STABLE REGION
0
– Quantitative Metric 0 4 8 12 16
– Defines square within stable Plant-Model Mismatch in Dead Time
region c = θP,Plant / θP,Model
– Value ≥ 1 (if stable at design
point, otherwise undefined)
10
What About Plant-Model Mismatch
in τP (b ≠ 1)?

• What have the Robustness Plots shown?

– Impact of Plant-Model Mismatch in KP and θP
– Assume no Plant-Model Mismatch in τP (b = 1)
• What does RSF2D indicate?
– Amount of Plant-Model Mismatch in both KP and θP
– Provides No Information about Plant-Model Mismatch in τP
• What does RSF3D indicate?
– Amount of Plant-Model Mismatch in All 3 Parameters (KP, τP, θP)
– By definition, RSF3D ≤ RSF2D
– If RSF3D is significantly different from RSF2D, Plant-Model
Mismatch in τP significantly impacts Robust Stability

11
 Effect of Plant-Model Mismatch in τP
(at b = 1 and b = 1/RSF3D)

(for b = τP,Plant / τP,Model = 1) (for b = τP,Plant / τP,Model = 1/RSF3D = 1/1.77)

(a) UNSTABLE UNSTABLE

REGION (b)
REGION
Plant-Model Mismatch in Gain

Plant-Model Mismatch in Gain

4 4
a = KP,Plant / KP,Model

a = KP,Plant / KP,Model
Both RSF2D and RSF3D Only RSF3D
RSF2D = 2.17 Squares Lie Entirely RSF2D = 2.17 Square Lies Entirely
2 Within Stable Region 2 Within Stable Region
RSF3D RSF3D
= 1.77 = 1.77

STABLE REGION STABLE REGION

0 0
0 2 4 0 2 4
Plant-Model Mismatch in Dead Time Plant-Model Mismatch in Dead Time
c = θP,Plant / θP,Model c = θP,Plant / θP,Model
τC = 2 θP,Model
θP,Model = 0.25 τP,Model

12
 Example 2 – Confirming the Validity
of a Model Fit to Closed-Loop Data

Is the model fit good? More confirmation of model fit:

Qualitatively: it looks good Simulated closed-loop
Quantitatively: R2 of 0.936 is good performance with existing
tuning matches actual
Model Parameters: performance
KP = 23.3 psig / %OUT
τP = 4.03 min
θP = 0.126 min

τP / θP = 32 … a time constant dominant process!

Example 2 – Using τC to specify
Performance and Robust Stability

Settling Time Settling Time Settling Time

2.5 min 2.9 min 5.5 min
Overshoot Overshoot Overshoot
10% 0% 0%

RSF2D = 1.71 RSF2D = 2.47 RSF2D = 4.09

RSF3D = 1.43 RSF3D = 1.85 RSF3D = 2.60

τC = 0.11 min τC = 0.36 min τC = 1.2 min

(selected tuning)
 Example 2 – Effect of Tuning on
Performance
Original Final SIMC IMC
KC 0.01 0.1 0.2 0.355 0.355
τI (sec) 3 30 60 118 242

SP
Tracking

Disturbance
Rejection

RSF2D 1.55 3.21 3.00 2.44 2.49

RSF3D 1.55 2.50 2.16 1.83 1.85

Modeling Integrating
(Non-Self Regulating) Behavior
• When a First Order Plus Dead Time Integrating (FOPDT Integrating) linear
model is fit to dynamic process data

dPV (t )
= K P* ⋅ OUT (t − θ P )
dt

• The important parameters that result are:

– Integrator Gain, KP* (How Far, How Fast)
– Apparent Dead Time, ӨP (With How Much Delay)

• IMC-PI Tuning Correlation:

2τ C + θ P
KC = ; τ I = 2τ C + θ P
K P* (τ C + θ P )

16
Example 3 – FOPDT Integrating
Model Fit to Open-Loop Data

• Model Parameters
– KP* = -0.326 %level / %OUT-min
– θP = 0.215 min
• Model Fit can be computed manually (Arbogast, 2007)
– Does not require Process Variable (level) ever to be at steady-state
– Requires two periods of constant Controller Output values
Example 3 – Blowdown Drum
Disturbance Rejection Performance

• Every 1.5 – 2 hours, a

disturbance caused
the pressure and level
to spike.
• In responding, the
loop overreacted.
• Poor tuning can turn
an unimportant loop
into an important one
• RSF2D improved from
about 1.0 to 3.4
Example 3 – Blowdown Drum
Set Point Tracking Performance

• SP Tracking is not the

objective for this
control loop
• However, the SP
response is of great
use
– Closed-Loop Data
for Model Fit
– Descriptive Statistics
Summary

• FOPDT & FOPDT Integrating Models provide useful information

• Consider Nonlinearity – Tune over Range of Operation
• IMC and SIMC Tuning use a single adjustable parameter, τC, to
specify performance and robust stability
• SIMC Tuning is useful for time constant dominant processes
• Robust Stability indicated using various tools
– Graphically – Robustness Plot
– Quantitatively – Robust Stability Factor (RSF) Metric
• Robust Stability and Performance are related
• Significance of Plant-Model Mismatch in τP
– RSF3D can be significantly smaller than RSF2D
– Depends on ratios: τP-to-θP and τC-to-θP
20
Thank You!
Any Questions?

Thank You! Any Questions?

Jeff Arbogast
[email protected]
(302) 286-5417

Delaware Research & Technology Center

The world leader in gases

for industry, health, and the
environment.

21
Back-up Slides
SIMC vs. IMC Tuning

SIMC IMC

SP
Tracking

Disturbance
Rejection
Introduction to the Robustness Plot

• Relates Plant-Model Mismatch to Stability

• Increase τC to Increase Robust Stability (Increase Size of
Stable Region)
(for b = τP,Plant / τP,Model = 1) (for b = τP,Plant / τP,Model = 1)

(a) (for τC = 0.77 min) UNSTABLE (b) (for τC = 2.87 min) UNSTABLE
8 REGION 8 REGION
Plant-Model Mismatch in Gain

Plant-Model Mismatch in Gain

a = KP,Plant / KP,Model

a = KP,Plant / KP,Model
6 6

4 4

2 2

STABLE REGION
STABLE REGION
0 0
0 2 4 6 8 0 2 4 6 8
Plant-Model Mismatch in Dead Time Plant-Model Mismatch in Dead Time
c = θP,Plant / θP,Model c = θP,Plant / θP,Model

24
Robustness Plot - Background

• History
– Introduced by Gerry and Hansen (1987)
– Procedures for Producing Robustness Plot
– Simulations / Experimentation - Shinskey (1990, 1994)
– Bode Stability Criterion (not detailed) – Fruehauf, et al. (1994)
– Bode Stability Criterion (detailed) – Arbogast, et. al. (2007, 2008)
• For analysis, assume the following:
– Actual Plant is represented by a FOPDT model
– FOPDT Model is used to Tune the Controller
• Robustness Plots
– Fix the Controller Tuning (the Model)
– Vary the Actual Plant

25
 Relationship between Robust Stability
and Robust Performance

τC = 8 θP,Model 60

θP,Model = 0.2 τP,Model 16

PV
55

Plant-Model Mismatch in Gain

(a = 1, b = 1, c = 1)

a = KP,Plant / KP,Model
12
50
60

55
PV

55 8

50

CO
50
4
UNSTABLE
RSF2D REGION
3.75 45
55
STABLE REGION
0
0 4 8 12 16 0 25 50 75 100
50 Time, t (min)
CO

Plant-Model Mismatch in Dead Time

c = θP,Plant / θP,Model
(a = 3.75, b = 1, c = 3.75)
45

0 25 50 75 100
Time, t (min)

26
 Relationship between Robustness
and Nyquist Plots

0.6
τC = 8 θP,Model

Imaginary Part of GOL,Plant(jω)

16 0.4
θP,Model = 0.2 τP,Model

Plant-Model Mismatch in Gain

0.2

a = KP,Plant / KP,Model
12 0.0

-0.2

(a = 1, b = 1, c = 1) 8 -0.4

-1.5 -1.0 -0.5 0.0 0.5

0.05 Real Part of GOL,Plant(jω)
Imaginary Part of GOL,Plant(jω)

0.00
4
RSF2D
UNSTABLE (a = 3.75, b = 1, c = 3.75)
REGION
3.75
-0.05 STABLE REGION
0
-0.10 0 4 8 12 16
Plant-Model Mismatch in Dead Time
-0.15 c = θP,Plant / θP,Model
-1.0 -0.5 0.0
Real Part of GOL,Plant(jω)

27
 Other Robust Stability Metrics

• Safety Factor (SF) region

– Introduced by Gerry (1988)
For b = τP,Plant / τP,Model = 1
– In robustness plot bounded by
8 (for any KP,Model, τP,Model, and θP,Model)
dotted line (SF = 2)
Plant-Model Mismatch in Gain
a = KP,Plant / KP,Model

⎧ c SF
⎪⎪ SF ≤ a ≤ c ,1 ≤ c ≤ SF
IMC-PI Tuning ∀a, c : ⎨
⎪ 1 1
where SF ≥ 1.
≤ a ≤ cSF , ≤ c <1
4 τC = 2.4θP ⎪⎩ cSF SF

RSF2D = 2.31
UNSTABLE • RSF region
REGION

STABLE REGION 1
0 ∀a, c : ≤ a, c ≤ RSF2D , b = 1
RSF2D
0 4 8
Plant-Model Mismatch in Dead Time
c = θP,Plant / θP,Model 1
∀a , b , c : ≤ a, b, c ≤ RSF3D .
RSF3D