You are on page 1of 9

D.E.

Rivera, ChE 461/598: Process Dynamics and Control; Introduction to Internal Model Control with Application to PID Controller Tuning
Session Focus
We will emphasize the use of the Internal Model Control design procedure as a means for tuning PID controllers:
d r + e

Introduction to Internal Model Control with Application to PID Tuning


Daniel E. Rivera, Ph.D.
Department of Chemical, Bio and Materials Engineering and Control Systems Engineering Laboratory Computer-Integrated Manufacturing Systems Research Center

+ y

Arizona State University, Tempe AZ 85287-6006


daniel.rivera@asu.edu

Integral Derivative Proportional Kc t de u(t) = u + Kc e(t) + e(t )dt + KcD I 0 dt e(t) = r(t) y (t)

Honeywell PID Eqn Forms

PID Controller Forms


Ideal Form

Equation A

PID -y

c(s) = Kc 1 + 1s + Ds I
Equation B

Interactive "Electronic" Form

Ds+1 c(s) = Kc 1 + 1 s I Ds + 1

PI

Ideal Form with Filter

1 c(s) = Kc 1 + 1 + Ds I s F s + 1
Low-Order Difference Equation Form

-y
Equation C

PD

uk =K'c ek - 'I ek-1 + 'D ek-2 + 'F uk-1

-y

Copyright 1999 by D.E. Rivera, All Rights Reserved

D.E. Rivera, ChE 461/598: Process Dynamics and Control; Introduction to Internal Model Control with Application to PID Controller Tuning
Internal Model Control Structure d
r + e

IMC-Closed Loop Transfer Functions


d r +e=r-y

+ y

P ~ P

+ y + +

C = Q(I-PQ)-1 Q = C(I+PC)-1
d r +

In the absence of plant/model mismatch (p = p ): y = = u = = e = = p q r + (1 p q ) d r +d q rq d rp 1 d p 1 (1 p q ) r (1 p q ) d r d = (1 + p c)1

P ~ P

+ y + +

=p c(1 + p c)1

Requirements for Physical Realizability of the IMC Controller q(s)


Internal Stability: Bounded inputs to the control system must lead result in bounded outputs elsewhere in the control system. "Properness": The controller must not differentiate step changes (e.g., avoid subjecting step inputs subject to pure derivative action). Causality: The controller must not require prediction, i.e., it must rely on past and current measurements of the process.

Internal Model Control Design Procedure


Step 1 (Nominal Performance). Obtain an optimal Q must specify input type (e.g., step, ramp) and objective function closed-form solutions available. Step 2 (Robust Stability and Performance). Augment Q with a filter F(s) Qfinal(s) = Q(s) F(s) F(s) contains adjustable parameters which can be tuned for Robust Stability and Performance

Copyright 1999 by D.E. Rivera, All Rights Reserved

D.E. Rivera, ChE 461/598: Process Dynamics and Control; Introduction to Internal Model Control with Application to PID Controller Tuning
Performance and Robustness Measures
Integral Square Error (ISE): J = ISE =
0 (y

Internal Model Control Design Procedure (Cont)

r)2dt

Step 1: Factor out all deadtime and Right-Half Plane Zeros in the model P(s) = P+(s) P-(s) IAE-Optimal (for step inputs) P+(s) =e-s -is + 1
i n

Integral Absolute Error (IAE): J2 = IAE =


0 |y

Re(i)>0

r|dt

Maximum Peak of the Complementary Sensitivity Function: M = sup

ISE-Optimal (for step inputs) n -is + 1 P+(s) =e-s is + 1 i

Re(i)>0

which leads to a controller that is stable and causal Q(s) = P-1 - (s)

Internal Model Control Design Procedure (Cont) Step 2: Augment the optimal controller with a filter F(s) which insures that the final control system is proper: Qfinal (s) = Q(s) F(s) Example: F(s) = 1 ( s + 1) nf

Some IMC-PID Tuning Rules


Selected rules for plants without integrator, > 0, no offset for step setpoint/disturbance changes, is an adjustable parameter.
Plant K(- s + 1) s + 1 : : K(- s + 1) 2s2 + 2 s + 1 : : K(- s + 1) 2s2 + 2 s + 1 KK c + : : 2 + : : 2 2 + I : : 2 : : 2 D F Controller

0 : : 2 : : 2

0 : : 0 : : 2 +

PI : : PID : : PID w/ filter

To note: Closed-loop transfer function P C 1+P C -1= P+(s) F(s) Filter time constants reflect closed-loop speed-of-response Filter time constants can be optimized for robust performance (Zafiriou and Morari, 1986)

1 c(s) = Kc 1 + 1s + Ds I F s + 1

Copyright 1999 by D.E. Rivera, All Rights Reserved

D.E. Rivera, ChE 461/598: Process Dynamics and Control; Introduction to Internal Model Control with Application to PID Controller Tuning
Example 1: PI Control
A PI tuning rule arises from applying IMC to the rst-order model: p = K s + 1 >0 (1)
We can now solve for the classical feedback controller equivalent c(s) to obtain q 1 c= = (1 + ) (3) 1 pq K s which leads to the tuning rule for a PI controller (4) Kc = K (5) I = The corresponding nominal closed-loop transfer functions for this control system are = 1 s + 1 p 1 = s + 1 k (s + 1) = s s + 1 (6)

Example 1: PI Control (Continued)

under the condition that d and r are step input changes. Step 1: Factor and invert p ; since p + = 1, we obtain: q = s + 1 K 1 (s + 1) (2)

Step 2: Augment with a rst-order lter f= The nal form for q is

s + 1 q= K (s + 1)

Example 1b: PI Control


Consider now the rst-order model with Right Half Plane (RHP) zero: p (s) = K (s + 1) ( s + 1) , > 0

Example 1c: PI with Filter Control


Consider now the rst-order model with Left Half-Plane (LHP) zero: p (s) = K (s + 1) ( s + 1) >0 >0

again under the assumption that the inputs to r and d are steps. Step 1: Use the IAE-optimal factorization for step inputs: p + = (s + 1) Step 2: Use a rst-order lter f= 1 (s + 1) q= ( s + 1) K (s + 1) K p = ( s + 1) ( s + 1) q = K

again under the assumption that the inputs to r and d are steps. Step 1: No nonminimum phase behavior in p ; since p + = 1, we obtain: p = K (s + 1) ( s + 1) q = ( s + 1) K (s + 1)

Step 2: Use a rst-order lter (q is now strictly proper). f= 1 (s + 1) q= ( s + 1) K (s + 1)(s + 1)

Solving for the classical feedback controller leads to another tuning rule for a PI controller: 1 ) c(s) = Kc(1 + I s Kc = I = K ( + )

Copyright 1999 by D.E. Rivera, All Rights Reserved

D.E. Rivera, ChE 461/598: Process Dynamics and Control; Introduction to Internal Model Control with Application to PID Controller Tuning
Example 1c: PI with Filter Control (cont.)
Solving for the classical feedback controller c = for an PI with lter controller:

Example 2: PID Control


Consider now the second-order model with RHP zero: p (s) = K (s + 1) (1s + 1)(2s + 1) , 1, 2 > 0 (7)

q 1p q

leads to a tuning rule

again under the assumption that the inputs to r and d are steps. Step 1: Use the IAE-optimal factorization for step inputs: p + = (s + 1) p = K (1s + 1)(2s + 1) (8)

c(s) = Kc 1 +

1 1 I s (F s + 1)

Kc = K I = F =

In IMC design, the presence of a Left-Half Plane zero in the model leads a low-pass lter element in the classical feedback controller!

(1s + 1)(2s + 1) (9) K Step 2: Use a rst-order lter (even though this means that q will still be improper). q = f= 1 (s + 1) q= (1s + 1)(2s + 1) K (s + 1) (10)

Example 2: PID Control (Continued)


q Solving for the classical feedback controller c = 1 p q leads to a tuning rule for an ideal PID controller: 1 + D s) (11) c(s) = Kc(1 + I s

Example 3: PID with Filter Control


Consider a second-order model with RHP zero p (s) = K (s + 1) (1s + 1)(2s + 1) , 1, 2 > 0 (16)

Kc =

1 + 2 K ( + ) I = 1 + 2 1 2 D = 1 + 2

(12) (13) (14)

and subject to step inputs to the closed-loop system. Applying the IMC design procedure gives: Step 1: Use the ISE-optimal factorization p + = s + 1 s + 1 p = K (s + 1) (1s + 1)(2s + 1) (17)

Step 2: A rst-order lter leads to q which is semiproper:


This PID tuning rule can be rearranged to conform to the interactive form 1 c(s) = Kc(1 + )(1 + D s) (15) I s

q=

(1s + 1)(2s + 1) K (s + 1)(s + 1)

f=

1 s + 1

(18)

Copyright 1999 by D.E. Rivera, All Rights Reserved

D.E. Rivera, ChE 461/598: Process Dynamics and Control; Introduction to Internal Model Control with Application to PID Controller Tuning
Example 3: PID with Filter Control (Continued)
Solving for c(s) as before results in a ltered ideal PID controller c = Kc(1 + with the associated tuning rule Kc = ( 1 + 2 ) K (2 + ) I = 1 + 2 1 2 D = 1 + 2 F = 2 + (19) (20) (21)
q=

Example 4: Deadtime Compensation (PI + Smith Predictor)


Consider the rst-order with delay plant p (s) = Kes s + 1 (23)

1 1 + D s) I s (F s + 1)

and step setpoint/output disturbance changes to the closed-loop system. Step 1: The optimal factorization (IAE, ISE, or otherwise) is p + = es, resulting in: s + 1 1 q = p = K Step 2: A rst-order lter makes q semiproper; s + 1 K (s + 1) = es (s + 1) (24)

(22)

The corresponding feedback controller is c(s) = s + 1 K (s + 1 es) (25)

Note the insight given by IMC design procedure regarding on-line adjustment (by changing the value for the IMC lter parameter ).

which can be expressed as a PI controller using the Smith Predictor structure (see Figure 17.4, page 605 in Ogunnaike and Ray).

Plants With Integrator


For plants with integrator, the practical problem will most likely demand A no oset for ramp output disturbances (d = s 2 ). This requires a Type-2 lter meeting the requirement d d (p +f ) = ( )|s=0 = 0 ds ds

Plants with Integrator (Cont.)


d One such lter transfer function which meets the condition ds ( )|s=0 = 0 is (2 p +(0))s + 1 f (s) = (26) (s + 1)2 Specic forms for p +(0) for various simple factorizations of nonminimum phase elements are shown below:

Plant
K (s+1) s K (s+1) s K (s+1) s K (s+1) s K (s+1) s( s+1)

= pq =p +f
s+1 s+1 (s+1) (s+1)(s+1) (s+1)[( +2)s+1] (s+1)2 (s+1) [2( +)s+1] s+1) (s+1)2 (s+1) [2( +)s+1] s+1) (s+1)2

Controller c(s) No Oset Conditions P P with lter PI PI with lter PID with lter Steps only Steps Only Steps and Ramps Steps and Ramps Steps and Ramps

d s (e )|s=0 = ds d (s + 1)|s=0 = ds d s + 1 ( )|s=0 = 2 ds s + 1

(27) (28) (29)

Equation (26) for simple plants with integrator leads to IMC-PID rules, as summarized in the previous table.

Note the progression in controller sophistication as closed-loop performance requirements increase!

Copyright 1999 by D.E. Rivera, All Rights Reserved

D.E. Rivera, ChE 461/598: Process Dynamics and Control; Introduction to Internal Model Control with Application to PID Controller Tuning
IMC-PID Tuning Rules for First-Order With Deadtime Processes
Ideal Form Parameters
Controller KKc I D F
Recommended ( > 0.2 always )

IMC-PID Tuning - FODT Plant


We can obtain a PID tuning rule for plants with deadtime by using a rst-order Pad e approximation in lieu of the time delay. p=
K ( 2 Kes s + 1) s + 1 (2 s + 1)( s + 1)

(26)

PI

2 + 2 2 + 2 + 2 + 2( + )

+ 2 + 2 + 2

> 1.7

The Pad e-approximated plant is a second-order plant with RHP zero; using the analysis from Example 2: PID Control leads to a PID tuning rule: Kc = 2 + K (2 + ) I = + 2 D = 2 + (27) (28) (29)

PID

2 + 2 +

> 0.8

PID w / filter

2( + )

> 0.25

p(s) = K e- s , is an adjustable parameter s+1 Rivera, Morari, Skogestad, "Internal Model Control. 4. PID Controller Design," Ind. Eng. Chem. Proc. Des. Dev., Vol 25, No. 1, 1986, pgs. 252-265.

As shown in Rivera et al. the ISE objective function for this control system can be plotted as a function of , independent of .

IMC-PID Tuning - Effect of Filter Parameter on Closed-loop Response


2 1.8 1.6 1.4 1.2 1 0.8 0.6 0.4 0.2 0 0

IMC-PID Tuning Performance Chart


J = ISE for a unit step input Jopt = ISE for "optimal" controller (perfect Smith Predictor) M = Maximum peak, sensitivity function

Controlled Variable Response Solid: / = 0.8 Dashed: / = 0.4 Dotted: / = 2.5

10

12

Copyright 1999 by D.E. Rivera, All Rights Reserved

D.E. Rivera, ChE 461/598: Process Dynamics and Control; Introduction to Internal Model Control with Application to PID Controller Tuning
IMC-PID Tuning - Comparison With Other Rules

"Original" IMC-PI tuning

= ISE for a unit step input

Jopt = ISE for "optimal" controller (perfect Smith Predictor)

O.L. Z-N: Open-Loop Ziegler-Nichols; C.L. Z-N: Closed-loop Ziegler-Nichols, C-C: Cohen-Coon

"Improved" IMC-PI Tuning


The Improved PI rule arises by incorporating the delay in the time constant of the internal model p p= K Kes s + 1 ( + 2 )s + 1 (30)

Improved IMC-PI Tuning - Worst-Case Analysis (over all Theta/Tau)

resulting, as shown in Example 1: PI control in the tuning rule: 2 + (31) 2K I = + (32) 2 For the Improved PI rules, closed-loop performance varies as a function of / . Kc =

Copyright 1999 by D.E. Rivera, All Rights Reserved

D.E. Rivera, ChE 461/598: Process Dynamics and Control; Introduction to Internal Model Control with Application to PID Controller Tuning
"Improved" IMC-PI Tuning Comparison (lambda/theta = 1.7)
Solid: J/Jopt Dashed: M

IMC-PID with Filter


The tuning rule for a PID with lter controller can be obtained from the Pad e-approximated plant and the analysis of Example 3: PID with Filter Control, Kc = 2 + 2K ( + ) I = + 2 D = 2 + F = 2( + ) (37) (38) (39) (40)

The IMC-PID with lter tuning parameters lead to higher ISE than the IMC-PID for the same value of /; however, the PID with lter settings display much smoother closed-loop responses. In industrial practice, the smoothness of the response may well be worth the loss of performance in terms of ISE.

C.L. Z-N: Closed-loop Ziegler-Nichols, C-C: Cohen-Coon

IMC-PID with Filter Performance

IMC-PID with filter (solid) versus IMC-PID (dashed) response (lambda/theta = 0.45 vs lambda/theta = 0.8)

Copyright 1999 by D.E. Rivera, All Rights Reserved

You might also like