You are on page 1of 28

Closed-loop PI/PID Controller Tuning

for Stable and Unstable Processes


Shamsuzzoha, Moonyong Lee*, Hiroya Seki**


King Fahd University of Petroleum and Minerals,
Saudi Arabia
*Yeungnam University, Kyongsan, Korea
**Tokyo Institute of Technology, Yokohama, Japan


2012 American Control Conference (ACC), 27
th
June 2012
Motivation
Desborough and Miller (2001): More than 97% of controllers are PI/PID
Vast majority of the PID controllers do not use D-action.
PI controller: Only two adjustable parameters
But still not easy to tune
Many industrial controllers poorly tuned
Ziegler-Nichols closed-loop method (1942) is popular, but
Requires sustained oscillations
Tunings relatively poor


Big need for a fast and improved closed-loop tuning procedure

2
Outline
1. Existing approaches to PID tuning
2. PID tuning rule for proposed study
3. Closed-loop setpoint experiment
4. Correlation between setpoint response and proposed PID-settings
5. Final choice of the controller settings (detuning)
6. Analysis and Simulation
7. Conclusion
3
1. Common approach:
PID-tuning based on open-loop model

Step 1: Open-loop experiment:
Most tuning approaches are based on open-loop plant model
gain (k),
time constant ()
time delay ()
Problem: Loose control during identification experiment

Step 2: Tuning
Many approaches
IMC-PID (Rivera et al., 1986): good for setpoint change
SIMC-PI (Skogestad, 2003): Improved for integrating disturbances
IMC-PID (Shamsuzzoha and Lee, 2007&2009) for disturbance
rejection for different type of processes
Shamsuzzoha and Skogestad (2010): The setpoint overshoot
method (Closed-loop tuning method)



-
( )
1
s
ke
g s
s
u
t
=
+
4
Alternative approach:
PI-tuning based on closed-loop data
Ziegler-Nichols (1942) closed-loop method
Step 1. Closed-loop experiment
Use P-controller with sustained oscillations. Record:
1. Ultimate controller gain (K
u
)
2. Period of oscillations (P
u
)

Step 2. Simple PI rules: K
c
=0.45K
u
and
I
=0.83Pu.

Advantages ZN:
Closed-loop experiment
Very little information required
Simple tuning rules

Disadvantages:
System brought to limit of instability
Many trials are required to obtain K
u
Relay test (strm) can avoid this problem but requires the feature of switching to
on/off-control
Settings not very good: Aggressive for lag-dominant processes (Tyreus and Luyben) and
quite slow for delay-dominant process (Skogestad).
Only for processes with phase lag > -180
o
(does not work on second-order)
5
Want to develop improved and
simpler alternative to Z-N:
Closed-loop setpoint
response with P-controller
Use P-gain about 50% of
Z-N
Identify key parameters
from setpoint response:
Simplest to observe is first
peak!


p
y A
p
t
y

A
s
y A
t
0 t =
u
u
y A
s
y
y
This work.
Improved closed-loop PID-tuning method
Idea: Derive correlation between key parameters and proposed
PID-settings for corresponding process
6
2. PID tuning rule based on IMC
y
s
d
c g
y u
+
-
First-order process with
time delay:
-
( )
1
s
ke
g s
s
u
t
=
+
PID controller:
PID controller based on IMC approach:
c
=
Fast and robust setting:
( )
1
1
c D
I
c s K s
s
t
t
| |
= + +
|
\ .
( )
2
2
c
c
K
k
t u
t u
+
=
+
2
D
tu
t
t u
=
+
I c
=min , 4( +)
2
c
u
t

| |
+
`
|
\ .
)
2
3
c
K
k
t u
u
+
=
min , 8
2
I
u
t t u

| |
= +
`
|
\ .
)
2
D
tu
t
t u
=
+
7
2
I c
u
t t = +
Skogestad (2003) recommended modifying the integral time



p
y A
p
t
y

A
s
y A
t
0 t =
u
u
y A
s
y
Procedure:
Switch to P-only mode and make
setpoint change
Adjust controller gain to get
overshoot about 0.30 (30%)

Record key parameters:
1. Controller gain K
c0
2. Overshoot = (y
p
-y

)/y

3. Time to reach peak (overshoot), t
p
4. Steady state change, b = y

/y
s
.

Estimate of y

without waiting to settle:


y

= 0.45(y
p
+ y
u
)

Advantages compared to ZN:
* Not at limit to instability
* Works on a simple second-order process.

3. Closed-loop setpoint experiment
Closed-loop step setpoint response with P-only control.
8
Various overshoots (10%-60%)
10 1
s
e
g
s

=
+
0 5 10 15 20 25 30
0
0.25
0.5
0.75
1
1.25
time
O
U
T
P
U
T

y


Setpoint
t/u=1 (K
c0
=0.855)
t/u=0.4 (K
c0
=0.404)
t/u=100 (K
c0
=79.9)
t/u=2 (K
c0
=1.636)
t/u=5 (K
c0
=4.012)
t/u=10 (K
c0
=8.0)
t/u=0.2 (K
c0
= 0.309)
t/u=0 (K
c0
= 0.3)
Closed-loop setpoint experiment
Overshoot of 0.3 (30%) with different s
30%
=0
=100
=2
Small : K
c0
small and b small
0 5 10 15 20 25
0
0.25
0.5
0.75
1
1.25
1.5
time
O
U
T
P
U
T


y


overshoot=0.10 (K
c0
=5.64)
overshoot=0.20 (K
c0
=6.87)
overshoot=0.30 (K
c0
=8.0)
overshoot=0.40 (K
c0
=9.1)
overshoot=0.50 (K
c0
=10.17)
overshoot=0.60 (K
c0
=11.26)
setpoint
9
Estimate of y

using undershoot y
u


p
y A
p
t
y

A
s
y A
t
0 t =
u
u
y A
s
y
y
Line: y

= 0.8947(y
p
+ y
u
)/2
0
0,2
0,4
0,6
0,8
1
1,2
0 0,2 0,4 0,6 0,8 1 1,2
(y
p
+ y
u
)/2

overshoot=0.6
overshoot=0.5
overshoot=0.4
overshoot=0.3
overshoot=0.2
Data: 15 first-order with delay processes using 5 overshoots each (0.2, 0.3, 0.4, 0.5, 0.6). y
s
=1
Conclusion:
y

0.45(y
p
+y
u
)
10
4. Correlation between Setpoint
response and proposed PID-settings
Goal: Find correlation between proposed PID-settings and key
parameters from 90 setpoint experiments.

Consider 15 first-order plus delay processes:
/ = 0.1, 0.2, 0.4, 0.8, 1, 1.5, 2, 2.5, 3, 5, 7.5, 10, 20,
50, 100

For each of the 15 processes:
Obtain proposed PID-settings (K
c
,
I
)
Generate setpoint responses with 6 different overshoots (0.10, 0.20,
0.30, 0.40, 0.50, 0.60) and record key parameters(K
c0
, overshoot, t
p
, b)
-
( )
1
s
e
g s
s
u
t
=
+
11
Fixed overshoot:
Slope K
c
/K
c0
= A approx. constant,
independent of the value of /
c
c0
K
=A
K
Correlation Setpoint response and proposed PID-settings
Controller gain (K
c
)
K
c0
K
c
Agrees with ZN (approx. 100% overshoot):
Original: K
c
/K
cu
= 0.45
Tyreus-Luyben: K
c
/K
cu
= 0.33

90 cases: Plot K
c
as a function of K
c0
0 20 40 60 80 100 120
0
10
20
30
40
50
60
70
kK
c0
k
K
c


0.10 overshoot
kK
c
=1.1621kK
c0
0.20 overshoot
kK
c
=0.9701kK
c0
0.30 overshoot
kK
c
=0.841kK
c0
0.40 overshoot
kK
c
=0.7453kK
c0
0.50 overshoot
kK
c
=0.6701kK
c0
0.60 overshoot
kK
c
=0.6083kK
c0
12
2
A= 1.55(overshoot) - 2.159(overshoot) + 1.35
(

Overshoots between 0.1 and 0.6
(should not be extended outside this range).
Conclusion: K
c
= K
c0
A
A = slope
overshoot
0.1 0.2 0.3 0.4 0.5 0.6
0.7
0.8
0.9
1
1.1
overshoot (fractional)
A



y = 1.55*(overshoot)
2
- 2.159*(overshoot) + 1.35
13
( )
I1
b
=1.5A
1-b

Proposed PID-rules
Case 1 (large delay):
I1
= +/2
Case 2 (small delay):
I2
= 8
Case 1 (large delay):

c c0 c c0 c0
kK =kK K K kK A =
c0
b
kK =
(1-b)
Correlation Setpoint response and proposed PID-settings
Integral time (
I
)
(from steady-state offset)
Conclusion so far:
Still missing: Correlation for
I
1.5
c
kK t u =
14
c
2+
K =
3k
1.5 0.5
c
kK t u u =
I
1.5
c
kK t u =
(substitute
I
= +/2 into
the proposed rule for K
c
)
0.1 0.3 0.5 0.6
0
0.1
0.2
0.3
0.4
0.5
Overshoot
u
/
t
p
0.43 (t
I1
)
0.305 (t
I2
)
t/u=0.1
t/u=8
t/u=100
t/u=1
( )
I I1 I p 2 p
b
=min( , ) min 0.645A , 2.44 t
1-b
t
| |
= |
|
\ .
Correlation between and t
p


p
y A
p
t
y

A
s
y A
t
0 t =
u
u
y A
s
y
/t
p
overshoot
Use:
/t
p
= 0.43 for
I1
(large delay)
/t
p
= 0.305 for
I2
(small delay)
Conclusion:
t
p

15
Derivative action (
D
)
The derivative action can increase stability and improve the
closed-loop performance.
Case I: For approximately integrating process (>> )

Case II: The processes with a relatively large delay ( )

Summary: The derivative action for both the cases i.e.,
D1
and
D2
are approximately same
( )
0.14 1
1-
D p
b
t if A
b
t = >
1
0.305
0.15
2 2 2
p
D p
t
t
tu u
t
t
~ = = =
2 2
2
0.43
0.1433
2 3 3 3
p
D p
t
t
u u u
t
u u u
~ = = = =
+
2
D
tu
t
t u
=
+
16
Selection of Controller Gain (K
c0
)
An overshoot of around 0.3 is recommended,

Achieving the P-controller gain (K
c0
) via trial and error
can be sometime time consuming.

Lets assume for the first closed-loop test P-controller gain
of K
c01
is applied and resulting overshoot OS
1
is achieved
that is between 0.1 to 0.60 but not around 0.30.
17
Selection of Controller Gain (K
c0
)
18
First closed-loop test with P-controller gain of
K
c01
is applied and resulting overshoot is OS1
K
c01
=K
c0


Goal achieved!

Overshoot not around 0.30
but between 0.1 to 0.60
Overshoot around 0.30
( ) ( ) ( ) ( )
2 2
1 1 01 0
1.55 OS 2.159 OS 1.35 1.55 OS 2.159 OS 1.35
c c
K K
( (
+ = +
( (

As it is mentioned earlier the proposed
method is good agreement with the IMC-
PID for the overshoot around 0.3.
( ) ( )
( )
2
0 1 1 01
1.19 1.45 OS 2.02 OS 1.27
c c
K K = +
K
c01
=K
c0

Goal achieved!

Choice of detuning factor F:
F=1. Good tradeoff between fast and robust (
c
=)
F>1: Smoother control with more robustness
F<1 to speed up the closed-loop response.
0 c c
K =K A F
( )
p p I
=min 0.645A ,
b
t t
1-b
2.44
| |
|
|
\ .
F
2
A= 1.55( ) - 2.15 overshoot oversh 9( ) + 1.3 oot 5
(

From P-control setpoint experiment record key parameters:
1. Controller gain K
c0
2. Overshoot = (y
p
-y

)/y

3. Time to reach peak (overshoot), t
p
4. Steady state change, b = y

/y
s

Proposed PID settings (including detuning factor F)
5. Summary setpoint overshoot method
( )
D p
b
= 0.14t if A 1
1-b
>
19
6. Analysis: Simulation PID-control
5 1
s
e
g
s

=
+
First-order + delay process
t=0: Setpoint change
t=40: Load disturbance
in training set
better response
0 20 40 60 80
0
0.25
0.5
0.75
1
1.25
time
O
U
T
P
U
T


y


Shamsuzzoha and Skogestad method with F=1(overshoot=0.10)
Shamsuzzoha and Skogestad method with F=1(overshoot=0.298)
Shamsuzzoha and Skogestad method with F=1 (overshoot=0.599)
Proposed method with F=1(overshoot=0.10)
Proposed method with F=1(overshoot=0.298)
Proposed method with F=1(overshoot=0.599)
20
s
g e s

=
Integrating process
Analysis: Simulation PID-control
in training set
0 20 40 60 80 100
0
0.5
1
1.5
2
2.5
3
time
O
U
T
P
U
T


y


Shamsuzzoha and Skogestad method with F=1 (overshoot=0.108)
Shamsuzzoha and Skogestad method with F=1 (overshoot=0.302)
Shamsuzzoha and Skogestad method with F=1 (overshoot=0.60)
Proposed method with F=1 (overshoot=0.108)
Proposed method with F=1 (overshoot=0.302)
Proposed method with F=1 (overshoot=0.60)
21
( )( )
1
1 0.2 1
g
s s
=
+ +
Second-order process
Analysis: Simulation PID-control
Not in training set
0 2 4 6 8 10
0
0.25
0.5
0.75
1
1.25
1.5
time
O
U
T
P
U
T


y


Shamsuzzoha and Skogestad method with F=1 (overshoot=0.127)
Shamsuzzoha and Skogestad method with F=1 (overshoot=0.322)
Shamsuzzoha and Skogestad method with F=1 (overshoot=0.508)
Proposed method with F=1(overshoot=0.127)
Proposed method with F=1(overshoot=0.322)
Proposed method with F=1(overshoot=0.508)
22
( )( )( )( )
1
1 0.2 1 0.04 1 0.008 1
g
s s s s
=
+ + + +
High-order process
Analysis: Simulation PID-control
Not in training set
0 5 10 15 20
0
0.25
0.5
0.75
1
1.25
time
O
U
T
P
U
T


y


Shamsuzzoha and Skogestad method with F=1 (overshoot=0.104)
Shamsuzzoha and Skogestad method with F=1 (overshoot=0.292)
Shamsuzzoha and Skogestad method with F=1(overshoot=0.598)
Proposed method with F=1(overshoot=0.104)
Proposed method with F=1(overshoot=0.292)
Proposed method with F=1(overshoot=0.598)
23
( )
2
1
1
g
s s
=
+
Third-order integrating
process
Analysis: Simulation PID-control
Not in training set
0 40 80 120 160 200
0
1
2
3
4
5
time
O
U
T
P
U
T


y


Shamsuzzoha and Skogestad method with F=1 (overshoot=0.106)
Shamsuzzoha and Skogestad method with F=1 (overshoot=0.307)
Shamsuzzoha and Skogestad method with F=1 (overshoot=0.610)
Proposed method with F=1 (overshoot=0.106)
Proposed method with F=1 (overshoot=0.307)
Proposed method with F=1 (overshoot=0.610)
24
5 1
s
e
g
s

First-order unstable process


Analysis: Simulation PID-control
Not in training set

0 20 40 60 80
0
0.5
1
1.5
2
time
O
U
T
P
U
T


y


Shamsuzzoha and Skogestad method with F=1 (overshoot=0.10)
Shamsuzzoha and Skogestad method with F=1 (overshoot=0.30)
Shamsuzzoha and Skogestad method with F=1 (overshoot=0.607)
Proposed method with F=1(overshoot=0.10)
Proposed method with F=1(overshoot=0.30)
Proposed method with F=1(overshoot=0.607)
25
6. Conclusion
From P-control setpoint experiment obtain:
1. Controller gain K
c0
2. Overshoot = (y
p
-y

)/y

3. Time to reach peak (overshoot), t
p
4. Steady state change, b = y

/y
s,

Estimate: y

= 0.45(y
p
+ y
u
)
PID-tunings for Revised Setpoint Overshoot Method:


p
y A
p
t
y

A
s
y A
t
0 t =
u
u
y A
s
y
y
c c0
K =K A , F
2
A= 1.55(overshoot) - 2.159(overshoot) + 1.35 (

( )
I p p
b
=min 0.645A t , 2.44t
1-b
| |
|
|
\ .
F
F=1: Good trade-off between performance and robustness
F>1: Smoother
F<1: Speed up
Probably the fastest PID-tuning approach in the world
( )
D p
b
= 0.14t if A 1
1-b
>
26
References
strm, K. J., Hgglund, T. (1984). Automatic tuning of simple regulators with specifications on phase and
amplitude margins, Automatica, (20), 645651.
Desborough, L. D., Miller, R. M. (2002). Increasing customer value of industrial control performance
monitoringHoneywells experience. Chemical Process ControlVI (Tuscon, Arizona, Jan. 2001), AIChE
Symposium Series No. 326. Volume 98, USA.
Kano, M., Ogawa, M. (2009). The state of art in advanced process control in Japan, IFAC symposium
ADCHEM 2009, Istanbul, Turkey.
Rivera, D. E., Morari, M., Skogestad, S. (1986). Internal model control. 4. PID controller design, Ind. Eng.
Chem. Res., 25 (1) 252265.
Seborg, D. E., Edgar, T. F., Mellichamp, D. A., (2004). Process Dynamics and Control, 2nd ed., John Wiley
& Sons, New York, U.S.A.
Shamsuzzoha, M., Skogestad. S. (2010). Report on the setpoint overshoot method (extended version)
http://www.nt.ntnu.no/users/skoge/.
Skogestad, S., (2003). Simple analytic rules for model reduction and PID controller tuning, Journal of
Process Control, 13, 291309.
Tyreus, B.D., Luyben, W.L. (1992). Tuning PI controllers for integrator/dead time processes, Ind. Eng. Chem.
Res. 26282631.
Yuwana, M., Seborg, D. E., (1982). A new method for on-line controller tuning, AIChE Journal 28 (3) 434-
440.
Ziegler, J. G., Nichols, N. B. (1942). Optimum settings for automatic controllers. Trans. ASME, 64, 759-768.
Shamsuzzoha, M., Skogestad, S., (2010). The setpoint overshoot method: A simple and fast closed-loop
approch for PI tuning, Journal of Process Control 20 (2010) 12201234.

27
Thank You
28

You might also like