21st European Symposium on Computer Aided Process Engineering – ESCAPE 21

E.N. Pistikopoulos, M.C. Georgiadis and A.C. Kokossis (Editors)

© 2011 Elsevier B.V. All rights reserved.

Optimization of IMC-PID Tuning Parameters for

Adaptive Control: Part 1
Chih-Wei Chua, *B. Erik Ydstie a, Nikolaos V. Sahinidis a
̙

a
Carnegie Mellon University, 5000 Forbes Ave., Pittsburgh, PA 15213, USA,
[email protected], *[email protected], ̙[email protected]

Abstract
This paper describes Part 1 of a two-part strategy for robust certainty equivalence
adaptive PID control. In Part 1 we develop the strategy for simple PID controller tuning
which maximizes the bandwidth subject to gain and phase margin constraints. An
implementation of the non-adaptive strategy in a real time environment using model
estimation based on non-convex optimization is described. The test shows the potential
of the tuning method. In the next part, which due to space limitations could not be
included here, we describe adaptive implementation.

Keywords: IMC-PID controller, robustness, optimization, adaptive control.

1. Introduction
Surveys on PID control [1] show that the majority of PIDs are left on factory settings.
This observation shows that the PID has inherent robustness properties when it is
applied to typical chemical processes. However, one might suspect that significant gains
could be achieved if the controllers were optimized since the accumulated effect of
millions of poorly tuned PIDs may be large.

Many methods have been proposed for on and off line PID tuning. Most of these are not
suitable for adaptive control since they do not tune performance subject to robust
performance. For example, classical methods for PID tuning taught in undergraduate
classes on process control (e.g. [2]) do not include any tuning knobs. In this respect the
Internal Model Control (IMC) tuning procedure by Rivera et al. [3] is better suited since
includes the filter parameter ߬௖ which tune closed loop performance [4,5] to achieve
robust stability.

In this paper we develop a tuning procedure for IMC which minimizes the filter-
parameter to maximize bandwidth subject to pre-specified gain and phase margin [6-9].
An analytical solution is developed for the first order dead time process. In the next
section we show that the approach meshes with certainty equivalence adaptive control.

2. Robustness of Certainty Equivalence Adaptive Control

A compelling paradigm for adaptive control was developed in the 1970s under the
banner of Certainty Equivalence Adaptive Control. In this approach the parameters of a
transfer function model ‫ܩ‬௣ ሺ‫ݏ‬ሻ is estimated in real time by matching the model to
process data. The resulting model is used to update the controller as shown in Figure 1A.
The figure does not highlight that it is critical update the controller tuning to achieve
robust performance. This property is better illustrated in the equivalent diagram Figure
1B where the adaptive system is viewed as two composite systems. The first system
Optimization of PID 759

shows the controller in feedback with the model. The second system shows how the
model adapts to the plant. Robust performance is achieved when the nominal feedback
system on the left does not generate high frequency inputs.

Figure1. Figure 1A on the left shows the classical representation of the certainty equivalence
approach to adaptive control. Figure 1B on the right shows the structure used in stability analysis.

The controller tuning only needs to be robust with respect to unstructured (additive)
uncertainty. Closed loop stability is ensured if ȁ ሺ•ሻୡ୪ ȟሺ‫ݏ‬ሻȁ ൏ ͳ where ȟሺ•) is the model
uncertainty. It follows that the PID controller should be tuned so that it has pre-specified
gain and phase margins to compensate for given unstructured uncertainty. In this sense
adaptive control achieves better performance and robustness than robust control theory
alone since we do not need to tune for parametric uncertainty.

3. PID Control with Pre-specified Gain and Phase Margins

The IMC design achieves optimal performance robustness by minimizing ߬௖ subject to
gain margin and phase margin constraints. Thus we want to solve the problem
݉݅݊ ߬௖
‫ݏ‬Ǥ ‫ݐ‬Ǥ‫ܣ‬௠ ൒ ‫כܣ‬௠
‫כ‬
ߔ௠ ൒ ߔ௠
߬௖ ൐ Ͳ
where ‫כܣ‬௠ is the desired phase margin (typically ͳȀ͵Ɏሻƒ† Ȱ୫ ‫כ‬
is the desired gain
margin (typically 1.7). Denoting the process and controller transfer functions by ‫ܩ‬௣ ሺ‫ݏ‬ሻ
and ‫ܩ‬௖ ሺ‫ݏ‬ሻ we get
ଵ
‫ܣ‬௠ ൌ (1)
หீ೎ ሺ௝ఠ೛ ሻீ೛ ሺ௝ఠ೛ ሻห

ܽ‫݃ݎ‬ൣ‫ܩ‬௖ ሺ݆߱௣ ሻ‫ܩ‬௣ ሺ݆߱௣ ሻ൧ ൌ െߨ (2)

ߔ௠ ൌ ܽ‫݃ݎ‬ൣ‫ܩ‬௖ ሺ݆߱௚ ሻ‫ܩ‬௣ ሺ݆߱௚ ሻ൧ ൅ ߨ (3)
ห‫ܩ‬௖ ሺ݆߱௚ ሻ‫ܩ‬௣ ሺ݆߱௚ ሻห ൌ ͳ (4)
where ߱௣ , ߱௚ are the phase and gain crossover frequency. Below we simplify this
problem.

4. Tuning algorithm
A first-order-plus-time-delay (FOPTD) plant model models process control systems in
this paper. The first order Pade approximation gives
భ
௄೛ ௄೛ ଵିమഇೞ
‫ܩ‬௣ ሺ‫ݏ‬ሻ ൌ ݁ ିఏ௦ ؆ (5)
ఛ௦ାଵ ఛ௦ାଵ ଵାభ
మഇೞ
760 Chih-Wei Chu et al.

The IMC-PID formula for FOPTD is given as [9]:

ഇ
ቀଵା మ ௦ቁሺఛ௦ାଵሻ
‫ܩ‬௖ ሺ‫ݏ‬ሻ ൌ ഇ (6)
௄೛ ቀఛ೎ ା ቁ௦
మ
ഓ
ଵ ଶቀഇቁାଵ
‫ܭ‬௖ ൌ (7)
௄೛ ଶቀഓ೎ ቁାଵ
ഇ

ఏ
ܶ௜ ൌ ൅ ߬ (8)
ଶ
ఛ
ܶௗ ൌ ഓ (9)
ଶቀ ቁାଵ
ഇ

The open-loop and closed-loop transfer function are given by

ഇ
ቀଵା మ ௦ቁ
‫ܩ‬௢௟ ሺ‫ݏ‬ሻ ൌ ‫ܩ‬௣ ሺ‫ݏ‬ሻ‫ܩ‬௖ ሺ‫ݏ‬ሻ ൌ ഇ ݁ ିఏ௦ (10)
ቀఛ೎ ା మ ቁ௦

భ
ீ೚೗ ሺ௦ሻ ଵି ఏ௦
మ
‫ܩ‬௖௟ ሺ‫ݏ‬ሻ ൌ ؆ (11)
ଵାீ೚೗ ሺ௦ሻ ఛ೎ ௦ାଵ

Substituting Eq. (10) into (1) – (4) results in

ഇ
ቀఛ೎ ା మ ቁఠ೛
‫ܣ‬௠ ൌ (12)
ഘ ഇ మ
ඨ൬ ೛ ൰ ାଵ
మ

ఠ೛ ఏ గ
ܽ‫ ݊ܽݐܿݎ‬ቀ ቁ െ ߱௣ ߠ ൌ െ (13)
ଶ ଶ
ఠ೒ ఏ గ
ߔ௠ ൌ ܽ‫ ݊ܽݐܿݎ‬ቀ ቁ െ ߱௚ ߠ ൅ (14)
ଶ ଶ
ଵ
߱௚ ൌ (15)
ඥఛ೎ మ ାఛ೎ ఏ

Thus the optimization problem becomes

݉݅݊ ߬௖
ഇ
ቀఛ೎ ା మ ቁఠ೛
‫ݏ‬Ǥ ‫ݐ‬Ǥ ൒ ‫כܣ‬௠
ഘ ഇ మ
ඨ൬ ೛ ൰ ାଵ
మ

ఠ೛ ఏ గ
ܽ‫ ݊ܽݐܿݎ‬ቀ ቁ െ ߱௣ ߠ ൌ െ
ଶ ଶ
ఠ೒ ఏ ‫כ‬ గ
ܽ‫ ݊ܽݐܿݎ‬ቀ ቁ െ ߱௚ ߠ ൅ ൒ ߔ௠
ଶ ଶ
ଵ
߱௚ ൌ ߬௖ , ߱௣ , ߱௚ ൐ Ͳ
ඥఛ೎ మ ାఛ೎ ఏ

This problem has 3 variables (߬௖ , ߱௣ , ߱௚ ) and 3 parameters (ߠ, ‫כܣ‬௠ , ߔ௠

‫כ‬
). However, the
problem needs the time delay information, and an explicit analytical relation between
the tuning parameter, ߬௖ , gain margin, ‫ܣ‬௠ , and phase margin, ߔ௠ can be found from Eq.
(12) – (15). Solving Eq. (12) gives a constant and for convenience denoted as ߙ:
߱௣ ߠ ൌ ߙ ൌ ʹǤͶͷͺ (16)
Optimization of PID 761

By substituting Eq. (12), (15) and (16) into (14) and using Eq. (12) and (16) we express
ߔ௠ and ߬௖ as functions of ‫ܣ‬௠ so that
గ ଶ ଵ
ߔ௠ ൌ െ ൅ ܽ‫݊ܽݐܿݎ‬ (17)
ଶ ర ర
ට஺మ೘ ቀଵା మ ቁିଵ ට஺మ೘ ቀଵା మ ቁିଵ
ഀ ഀ

ఏ ସ
߬௖ ൌ ቆ‫ܣ‬௠ ටͳ ൅ െ ͳቇ (18)
ଶ ఈమ

The plot in Fig. 2 shows that for a given process that the gain and phase margins are
coupled so that only one of the two constraints will be active.
85 1.2

80 1
2.5ș
75 2ș 0.8
1.5ș
70 0.6
P h a se m a rg in ( Φ m )

65 T =ș 0.4
c

O u tp u t
60 0.2

55 0
0.5ș
50 -0.2 A *=2
m
45 -0.4 A * = 2.5
m
40 -0.6 A *=3
m
T = 0.3ș
c
35 -0.8
1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 0 1 2 3 4 5 6 7 8 9 10
Gain margin( A ) Time (sec)
m

Figure 2. ‫ܣ‬௠ vs. ߔ௠ respect to ߬௖ Figure 3. Closed-loop step responses.

The bandwidth ߱஻ௐ is is defined as the frequency at which

ଵ
‫ܴܣ‬௖௟ ሺ݆߱஻ௐ ሻ ൌ ȁ‫ܩ‬௖௟ ሺ݆߱஻ௐ ሻȁ ൌ (19)
ξଶ

Substituting Eq. (11) and (18) into (19) gives

ଶ
߱஻ௐ ൌ (ʹͲ)
ర ర
ఏඨ஺೘ మ ቀଵା మ ቁିଶ஺೘ ටଵା మ ିଵ
ഀ ഀ

The relation between ߱஻ௐ and ‫ܣ‬௠ provides an estimate for closed-loop performance.
Now we can propose a tuning method based on gain margin specification. According to
Eq. (12), ‫ܣ‬௠ is proportional to ߬௖ . So for given ‫כܣ‬௠ , the minimal ߬௖ can be located when
‫ܣ‬௠ equals to its minimal value, ‫כܣ‬௠ . Then the PID controller parameters, corresponding
phase margin and bandwidth can be calculated from Eq. (7) – (9), (17) and (20). Fig. 3
shows the simulation result for the closed-loop step responses of the controller designed
by different gain margin specifications. As ‫כܣ‬௠ getting larger, the performance of the
controller gets more conservative. Substituting into (20) gives
ξଶାଵ
‫כܣ‬௠ ൐ ర
ൌ ͳǤͺ͹ (21)
ටଵା మ
ഀ

5. Real-time experiments
The experimental set up comprises of a countercurrent shell and tube heat exchanger.
Hot water flows through the shell side and the cold water flows through the tube side.
Temperatures and flow rates are recorded at a sampling time of 0.1 seconds. The
FOPTD model
762 Chih-Wei Chu et al.

ି଴Ǥଷ଺ହ
‫ܩ‬௖௖ ሺ‫ݏ‬ሻ ൌ ݁ ିଷ௦ (22)
ଵǤ଺ଷ௦ାଵ

was identified using global optimization as shown [10]. Fig. 3 shows the response of the
system output for a set-point change followed by a load disturbance on hot water flow
rate change. The controller gives quick set-point response and well disturbance rejection.
63.6

63.4

63.2
Temp (F)

63

62.8

62.6

62.4

62.2
0 50 100 150 200 250 300 350 400
Time (sec)

10
Cold water
9
Hot water
Flow rate (gpm)

4
0 50 100 150 200 250 300 350 400
Time (sec)

Figure 3. Real-time experimental result. The precision is limited by 8bit AtoD conversion

6. Conclusions
An optimization problem for the IMC-PID controller suitable for adaptive control is
developed. The analytical solution for the optimization of bandwidth from gain and
phase margin constraints is derived. We show that that gain and phase margins are
coupled. The real time experiment result gives satisfied satisfactory set-point response
and disturbance rejection. The proposed approach is ideally suited for application to
adaptive control since the tuning criteria (gain margin and phase margin) are based on
closed rather than open-loop performance.

References
[1] K.J. Astrom, T. Hagglund(2nd ed.), PID Controllers: Theory, Design, and Tuning, Instrument
Society of America, Research Triangle Park, NC, 1995.
[2] J.G. Ziegler, N.B. Nichols, Optimum settings for automatic controllers, Trans. A.S.M.E., 64
(1942) 759–768.
[3] D.E. Rivera, M. Morari, S. Skogestad, Internal model control. 4. PID controller design, Ind.
Eng. Chem. Res., 25 (1) (1986) 252–265.
[4] S. Skogestad, Simple Analytic Rules for Model Reduction and PID Controller Tuning, J.
Process Control, 13(2003) 291-309.
[5] I.L. Chien, P.S. Fruehauf, Consider IMC tuning to improve controller performance, Chemical
Engineering Progress (1990) 33–41.
[6] W.K. Ho, C.C. Hang, L.S. Cao, Tuning of PID controllers based on gain and phase margin
specifications, Automatica, 31 (3) (1995) 497-502.
[7] Q.G. Wang, H.W. Fung, Y. Zhang, PID tuning with exact gain and phase margins, ISA
Transactions, 38 (1999) 243-249.
[8] W. K. Ho, T. H. Lee, H. P. Han, and Y. Hong, Self-Tuning IMC-PID Control with Interval
Gain and Phase Margins Assignment, IEEE Transactions on Control Systems Technology, 9
(3) (2001).
[9] D.E. Seborg, T.F. Edgar, and D.A. Mellichamp(2nd ed.), Process Dynamics and Control,
Wiley, New York, 2003.
[10] G.H. Staus, L.T. Biegler, and B.E. Ydstie, Global optimization for identification, Proceeding
of the 36th Conference on Decision and Control, (1997) 3010-3015.