Sie sind auf Seite 1von 33

Internal Model Control: A Comprehensive View

Daniel E. Rivera Departm e of Chemical,Bio and Materials nt Engineering Collegeof Engineering and App lied S ciences Arizona State University, Tempe, Arizona 8 5 2 8 7 6006 October 2 7 , 1 9 9 9

Copyright c 1 9 9 9by DanielE. Rive ra

The assistance of Amanda J. Wruble and Kyoung-Shik Jun in putting together this document is greatly appreciated

Co ntents
1 Internal Model Co ntrol Structure - (IMC) 4 4 4 7 7 . . 1 .1 C lo se d oop tra n sfe r n ction s, -l fu IMC structure. . . . . . . . . . . . . . . . . 1 .2 Inte rn a l Stability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 .3 R e g a rd in g ple mnta tio n . . . . . . . . . . . . . . . . . . . . . . . . . . . . Im e 1 .4 A sym ptoticclosed -l p behavior (S yste mT ype ) . . . . . . . . . . . . . . . . oo 1.5 Requiremnts for Physical Realizabili on q, the IMC Controller e ty . . . . 2 Internal Model Co ntrol Design Procedure 2 .1 S ta te m nt o f th e IMC D e s ig nP roc e d u re . . . . . . . . . . . . . . . . . . . . e 2 .2 Why fa c to rp? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Application of IMC Design to PID controller tuning 7 8 8 10 11 11 12 12 13 14 . . 15 15 20

3 .1 E x a m p le1 : PI Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 .2 E x a m p le1 b : PI Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 .3 E x a m p le1 c : PI with lter control . . . . . . . . . . . . . . . . . . . . . . . . 3 .4 E x a m p le2 : PID Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 .5 E x a m p le3 : PID with Filter C ontrol . . . . . . . . . . . . . . . . . . . . . . . 3 .6 E x a m p le4 : D e a d tim e compensation (PI controller + Sm ith Predictor) . 15 3 .7 PID c ontro l fo r p lants w ith inte g ra to r. . . . . . . . . . . . . . . . . . . . . . 4 PID Tuning Rules for 1st-order References with Deadtime Pla nts

IMC: A Comprehensive View

List of Figures
1 2 3 4 5 6 7 Classicaland InternalModel Control Feedba Structures.. . . . . . . . . . ck Evolution of the Internal Model Control Structure. 6 5 . .

. . . . . . . . . . . .

J/J opt and M for theIMC-PID controller,an d com p arison ith otherm eth w ods. 1 8 IMC-PID controlle d variab leresponsesfor a ste psetpoint cha n ge , various for setting s ; s o lid : = 0.8 ; d otted: = 2.5 ; d a s h e d = 0.4. . . . . . . . . of : 21 J/J opt and M for theoriginalIMC-PI controller. . . . . . . . . . . . . . . 21 Wo rs t-ca s e opt and M for the im pr ved IMC-PI controller,and comparJ/J o is o n w ith o th e rm e th s. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . od 22 J/J opt and M for the IMC-PID w ith lter controller,and controlled variable responsecom parison with IMC-PID ru le . . . . . . . . . . . . . . . . . . . . . 23

List of Tables
1 2 3 PID tu n in g ru le sfo r p lants w ith inte g ra to r . . . . . . . . . . . . . . . . . . . 16 IM C-BasedTuning for IdealPID Controllers Using Sim pleModels . . . . . 1 7 . IM C -b asedtun ing rules for PI, PID, and PID with lter controllers for a rs t-o rd e w ith d e a d tim e y s te m. . . . . . . . . . . . . . . . . . . . . . . . . r s 19

IMC: A Comprehensive View

Internal Model Co ntrol Structure

- (IM C)

Internal Model Control (IMC) fo rm s the b asis for the system aticcontrol syste m d e sig nmethodology that is the prim aryfocus of this te xt. Th e rst issueone needsto understan d ard in g reg IMC is the IMC structure (to be d istin gu ishe d from th e IMC d e sig n proce d u re ). Figure 1B is the Internal Model Control or Q-param etrization structure.The IMC structure and the classical feedba structure (Figure 1A) are e q u va lent ck i rep re setatio ns; ig u re 2 d e m o n stra tethe evolution of the IMC structure. n F s We will show that the design of q(s) is m ore straig htforward and intuitive than the desig nof c(s). Having designe dq(s), its equi alent classicalfeedba controller c(s) ca n be readily v ck obtainedvia algebraic transfor- ations, m and vice- e rs a v c = q = q 1 pq c 1 + pc (1 ) (2)

1.1

Closed-l oop transfer

functions,

IMC structure

A statem e t of the sensitivi and com plem ntary sensitivi in te rm sof the n ty e ty internal model p and controller q(s) corres onds to: p y = pq r+ 1 + q(p p) 1 d pq 1 + q(p p) (3 )

= (s)r(s) + (s)d(s) In th e ab sen ce plant/model m ism at (p = p), the sefu n ctio ns of ch sim p lifyto (s) = pq (s) = 1 (s) = 1 pq p1 = q

(4 )

(5 )

w hich lead to the follow ing expressions the input/outputrelationships for between y, u, e and r, d, and n: y = pqr + (1 pq)d pqn u = qr qd qn e = (1 pq)r (1 pq)d (1 pq)n (6 ) (7 ) (8)

1.2

Internal

Stabili ty

1 . A ssu m e pe rfe ctmod el (p = p). Th e IMC system(Figu re1 B ) is internally stab le a (IS)

if and only if both p and q a re stable. 2 . A s s u m ethat p is stableand p = p. Then the classicalfeedba system(Figure ck 1A) with controller according Equation(1) is IS if and only if q is stable. to

IMC: A Comprehensive View

d r+ e

(A)

d r +

p p
~

+ +

~
d

(B)
F ig u re1 : C la ssical(A) and InternalModel Control (B) Feedba Structures. ck

IMC: A Comprehensive View

d(s) r(s) + c(s) p(s) y(s)

d(s) y(s) c(s) p(s)

r(s) +

p(s) +

p(s) -

d(s) r(s) + p(s) p(s) -

c(s)

p(s)

y(s)

Figure2: Evolution of the Internal Model Control Feedba Structure. ck

IMC: A Comprehensive View The IMC structure s o ersth e follow in g bene ts w ith thu respect to classical fe e d b ak : c n o n e edto solve fo r roots of th e ch aracte ristic polyn o m ia 1 + pc; o ne sim p ly l ex a m in e s e pole s of q; th on e ca n se ar h fo r q in steadof c w ithoutany loss of generali c ty.

1.3

Regarding

Impleme ntation

For linear, stable plants in th e ab sen ceof constrai nts on u, it m akes n o d i ere nce to im ple-m ent the controller either throughc o r q. Howeve r, in th e p re se n ce f ac tu a to r o co n strai nts, o n e c a n u se th e IMC structureto avoid saturationproblem sw ithout the need for special anti-w indupm easures.

1.4

Asymptotic

closed-l oop behavior (System Ty pe)

We n e e d to insure that the feedba control system leads to no o set for setpoint or ck distur-bancech a n g e swe thus needto d e n eso -ca lle d ; Type 1 a n d Type 2 inputs: Ty pe 1 (Step Inputs): is obtained if No o setto asym ptotically step setpoint/disturbancechanges lim pq = (0 ) =
s0

Ty pe 2 (Ramp that

Inputs):

For n o o se t to ram p inputs, it is re qu ired

s0

lim pq = (0) = 1 d (pq) = =0 ds


s=0

d
s0

lim

ds

1.5

Requirem nts for Physical Realizabili ty on q, the IM Cone C troller

In orde r for q, the IMC controller, to result in physically rea liza b lem a n ipu late d varia b lerespon se s, must satisfythe follow ing criteria: it 1 . Stability . T h e c ontro lle r mu st g e n e ra te u n d e d spo n se sto bo u n d e d p u ts ; bo re in th e re fo re po le s o f q must lie in the open Left-Half all Plane. 2 . Properness. We e s ta b lis h e in th e se c o n d b p re pse ss io n d la that di ere ntiation of step inputsby a fe e d bak controller le ad sto im p u lsecha n ge s u, w hi h are not c in c physically

IMC: A Comprehensive View realizable. ord erto avoid pu re di ere In ntiation of sign als, must req uirethat we q(s) be pro per , w hi h m ean sth at th e quantity c
|s|

lim q(s)

must be nite. We say q(s) is strictly pro per if


|s|

lim |q(s)| = 0

A strictlyproper tran sfer nctionhas a denom inator fu ordergreater than the num erator order.q(s) is semi-pro per , that is,
|s|

lim |q(s)| > 0

if th e d enom in ator orderis equalto the num erator order. A systemthat is not strictly proper or sem ipro is calledimpro per . per 3 . Causality . q(s) must be ca u sal, w hi h m ea n s that the controller must not c re qu irep re d ictio n ,i.e ., it must rely on current and previous plant m e a su re m e nts. A sim p le e xa m p le of n on ca usal a transfer functionis the inverse of a tim e delay transfer function q(s) = u(s) = K e+s c e(s) (9 )

Th e inverse tran sform (9 ) relieson future inputsto g e n e ra te current output;it of a is clearlynot realizable: u(t) = Kc e(t + ) (1 0 )

Internal Model Co ntrol Design Pr ocedure

The IMC d esig n proced u re is a two-step approa that, although sub-optim alin a ch g e n e ra (n o rm )se n s e ,p rovid e s a re a so n a b le d e o betwe en pe rfo rm a n ce n d ro b u stne ss. l tra a T h e m a in be n e t o f th e IMC approa is the abili y to directly specify the ch t com plem ntary sensitivi and sensitivi functions and , w hich as n oted previousl e ty ty y, directlyspecify the natureof the closed-l re spo n s e . oop

2.1

Statem nt of the IM Design Pr ocedure e C

The IMC d esig nproced ureco nsistsof two m a in step s. The rst step will in surethat q is stableand ca u sa l; secon d p will req uireq to be prope r. the ste Step1: Factor the model p into two parts:

p = p+ p

(1 1 )

IMC: A Comprehensive View p+ contains all Nonmini mum Phase Eleme nts in the p lant model, that is all RightHalf-Plane(RHP) ze ro sand tim e delays. The p , m e a h ile , is nw factor Mini mum Phase and invertible;an IMC controller de n e d s a
1 q = p

is stable and causal. The factorization p+ from p is depende upon the objecti ve function ch o s e n . of nt For example, p+ = es (i s + 1 ) Re(i ) > 0 (1 2 )
i

is Integral-Absolute-Error (IAE)-optim alfor step setpoint and output disturbance a n g e s M eanw hile, the factorization ch . p+ = es
i

(i s + 1) (i s + 1)

Re(i ) > 0

(1 3 )

is Integral-Square-Error (ISE)-optim al for step setpoint/output disturbance ch a n g e s As noted in Morari a n d Z a rio u [2 ] using ram p, exponential, or other . inputs would im ply di ere factorizations. nt Step 2: Augm e q w ith a lter f (s) such that the nal IMC controller q = qf (s) is nt now , in addition to stable and causal, proper. W ith the inclusion of the lter transfer function, the nal form for the closed-loop transfer fun ction s characterizin g e systemis th = pqf = 1 pqf (1 4 ) (1 5 )

The inclusion of the lter transfer function in S te p 2 m e a n s that we no longer obtain optimal control, as im plied in Step 1 . We w ish to d e n e lter form s that allow for n o o se tto Type 1 a n d Type 2 in p u ts;fo r n o o1 tto step inputs (Type 1 ), se we mu st re q u ireth a t (0 ) = 1 , w hi h re quires t q(0) = p (0 ) a nd fo rces c tha f (0 ) = 1 A com m on lter choice that conform s this to requirem nt is e f (s) = 1 (s + 1)n (1 6 )

(1 7 )

The lter ord er n is se le cte dla rge e n o u g hto m ake q proper, w hile is an adjustable parameter w hich determ ines speed-of-res the ponse. Incre asing in c re a se th e c lo se d oo p s -l tim e c o n stat a n d slow s th e spe e d o f re spo n s e ;d ecre asin g does th e opposite. can be n be ad justed -lineto compensa te for on plant/model m ism at in the de sig no f th e contro l ch sy ste m th e h ig he rth e va lu e o f , the higherthe robustness control system . ; the

IMC: A Comprehensive View

10

For n o o se tto T ype -2 (ra m p ) p u ts ,in a d d itio nto th e re q u ire m e (1 6 ),th e c lo s e d -l p in nt oo s y ste mmust satisfythe following d (pq) |s=0 = d ds ds =0
s=0

(1 8 )

By substituting expression q obtain edfrom th e two-ste pIMC d e sig np roce d u re , the for we ca n w rite (1 8 ) speci ca llyas d (p+ f )|s=0 = 0 (1 9 ) ds O n e such lte r tran sferfu nctionw hi h m ee tsth e cond ition(18)is c s (2 p (0 )) +1 + (s + 1)2

f (s) =

(2 0 )

Specic form sfor p (0 ) for va riou ssim p lefacto riza tio ns f n o n m in i m p h ase le m e o mu e nts + a re show n be low : d s (e )|s=0 = (2 1 ) ds d (s + 1)|s=0 = (2 2 ) ds d s + 1 )|s=0 = 2 (2 3 ) ( ds s + 1 Eq uation(20 )will enableus to obtainPID rulesfor plants w ith integrator, will be as show n la ter in this document.

2.2

Why factor

p?

Recallthat for classical ed b ak fe c y = r + d = (1 + pc)1 pc = (1 + pc)1 Usingthe IMC structure, no plant/model m istm at (p = p), we have for ch = pq = 1 pq (2 4 ) (2 5 ) (2 6 )

Perfect control (m eaningy = r for all tim e) is ach ieved w hen = 1 a n d = 0 , w h ch implies that i q = p1 (2 7 ) Howeve r, in ord er for u = q(r d), the m anipulatedvariable response, to be physically realizable, must be stable, proper, and causal. Nom ini q mum phase behavior

(deadtim eand RHP ze ro s ) will c a u se q = p1 to be n oncausa land un stab le , re specti ely; if p is strictly proper, then q will be im p ro v per a swell. H e nceth e ne e dfo r factorization.

IMC: A Comprehensive View

11

O ne can betterunderstand discussion exam ining sim pleexam ple. onsider this by a C the plant model s p(s) = K (s + 1)e (2 8 ) 2 s2 + 2 s +1 w h e re > 0 , w h ch im p lie s the p res n ceof a R ight-H alf Plane zero. N onm ini i se mum phaseelemets for this plant a re (e (s + 1 ). The perfect IMC controller for this syste mco rrre pon ds to s 2 s2 + 2 s + 1 +s e q = p1 = K (s + 1) While y = r u sin g th is contro lle r, th e m a n ipu la te dva ria b le re spo n se is phy sica lly u n re aliza b le for two re a so n s. First, q is u n sta b le a s a result of a Right-Half Plane pole arising from (s + 1 ). S ec o n d l q is non causal ca u seo f th e p re se n ce f th e y, be o tim e lead term e+s . Applying an appropriate factorization this model as descri to bed earlier results in sta ble,causal control action;a correctlychosenlter orderwill in surepropernessan d a physically realizableresponse. O ne must ke e p in m ind that the nonm ini mum p h a se e le m e ts es (s+ n 1 ) will always form part of the closed-l oop re spo n s e !

Application

of IM Design to PID controller tuning C

The IMC control designprocedure,w henappliedto low-ordermodels, will often result in PID and PID-like controllers. D eveloping these is the focus of this section:

3.1

Example

1: PI Co ntrol

A PI tuning rule arises from applying IMC to the rst-order model: p = K s + 1 >0 (2 9 )

underthe conditionthat d and r are step inp ut ch a n g e s . Step 1: Factor and invert p; sin cep+ = 1, we obtain: q = Step 2: Augme with a rst-order nt lter 1 f = (s + 1) s + 1 K

The nal form for q is q=

s + 1 K (s + 1)

(3 0 )

IMC: A Comprehensive View We can now solve for the classical feedba controller equi alent c(s) to obtain ck v c= q 1 pq = K K (1 + 1 s )

12

(3 1 )

w hich leadsto the tuningrule for a PI controller Kc = (3 2 ) (3 3 )

I =

The corres pondingnominal closed-l transferfunctions this control systemare oop for = 1 s + 1 p1 s + 1 k(s + 1) = s = s + 1 (3 4 )

3.2

Example

1b: PI Co ntrol
p(s) = K (s + 1) ( s + 1) , > 0 (3 5 )

Consider now the rst-order model w ith Right Half Plane (RHP) z e ro :

againunder the assum ption that the inputsto r and d a re s te p s . Step 1: Use the IAE-optim alfactorization step for inputs: p+ = (s + 1) p = 1) Step 2: U se a rst-order lter K ( s + q =

( s + 1) K (3 6 )

f =

1 (s + 1)

q=

( s + 1) K (s + 1)

(3 7 )

Solvingfor the classical feedba controller leadsto anothertuningrule for a PI ck controller: c(s) = Kc (1 + Kc = K ( + ) 1 ) I s I = (3 8 )

3.3

Example

1c: PI w ith lter control

Considernow the rst-order model with Left Half-Plane(LHP) z e ro : p(s) = K (s + 1) ( s + 1) > 0 > 0 (3 9 )

againunder the assum ption that the inputsto r and d a re s te p s .

IMC: A Comprehensive View Step 1: No nonm ini mum phasebehavior in p; sincep+ = 1 , we obtain: p = K (s + 1) ( s + 1) ( s + 1) q = K (s + 1)

13

(4 0 )

Step 2: U se a rst-order lter (q is now strictly proper). f = 1 (s + 1) q= ( s + 1) K (s + 1 )(s + 1) (4 1 )

Solvingfor the classicalfe e d b ak controller c = q lea dsto a tuning rule for an PI with c 1pq lter controller: c(s) = Kc 1 + 1 1 (4 2 )

I s (F s + 1)

Kc = K I = F = It is interestingto note that in IMC de sig n ,th e p re se n ce a Left-HalfPlane zero in of th e model leadsa low -p a ss lter elem e in the classical nt feedba controller! ck

3.4

Example

2: PID Co ntrol
, 1 , 2 > 0

C on sid er now th e se co nd -o rd er e l w ith RHP mod z e ro : p(s) = K (s + 1 ) (1 s + 1 )(2 s + 1)

againunder the assum ption that the inputsto r and d a re s te p s . Step 1: Use the IAE-optim alfactorization step inputs: for p+ = (s + 1) p = K ( s + 1 )( s + 1 )
1 2

(4 3 )

q =

(1 s

+ 1 )(2 s + 1)

(4 4 )

Step 2: U se a rst-ord er lter (even thou ghth is m ean sthat q will still be im pro r). pe f = q= 1 (s + 1) (4 5 ) (46 )

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

IMC: A Comprehensive View Solvingfor the classicalfe e d b ak controller c = q lea d sto a tuning rule for an c 1pq ideal PID controller: c(s) = Kc (1 + 1 + D s) I s

14

(4 7 )

Kc

1 + 2 K ( + )

(4 8 ) (4 9 ) (5 0 )

I = 1 + 2 1 2 D = 1 + 2

3.5

Example

3: PID w ith Filter Co ntrol


K (s + 1 ) (1 s + 1 )(2 s + 1) , 1 , 2 > 0 (5 1 )

Considera secon d -orde r e l w ith RHP ze ro mod p(s) =

> 0 , a s be fo re ,a n d subje c t to step inputs to the closed-l system .Applying the oop IMC d esig np roced uregive s : Step 1: Use the ISE-optim al factorization p+ = s + 1 s + 1 K (s + 1) p ( s + 1 )( s + 1) 1 2 = (5 2 )

Step 2: A rst-order lter leadsto q w hich is sem ip ro e r: p q= (1 s + 1 )(2 s + 1 ) K (s + 1 )(s + 1) f = 1 s + 1 (5 3 )

Solvingfor c(s) a s be fo rere su ltsin a ltered ideal PID controller c = Kc (1 + ( w ith the associatedtuning rule 1 + D s) I s 1 F s + 1)

Kc

(1 + 2 ) K (2 + )

(5 4 )

I = 1 + 2 1 2 D = 1 + 2 F = 2 +

(5 5 ) (5 6 )

(5 7 )

Note the insight given by IMC desig n pr ocedurere gard ing -lin eadju stm nt (by on e ch angin gthe value for the IMC lter param eter ).

IMC: A Comprehensive View

15

3.6

Example 4: Deadtim e Predictor)

com pensation

(PI controller

+ S ith m

Considerthe rst-order ith delay plant w p(s) = Kes s + 1 and step setpoint/output disturbance a n g e s th e closed -l system . ch to oop Step 1: The optim alfactorization (IAE, ISE, or otherw ise) p+ = es , re sultin gin : is s q = p1 = +1 K Step 2: A rst-order lter m akes q sem ip ro e r; p q= s + 1 K (s + 1) = es (s + 1) (5 8 )

The corres ondingfeedba controller p ck s + 1 is K (s + 1 es ) c(s) = w h ch ca n be e x p re sse d s a PI controller u sing the Sm ith Pred ictorstructu re(see i a Figure 1 7 .4 , a g e6 0 5in Ogunnai and Ray). p ke

(5 9 )

3.7

PID control for pla nts with integrator

For plants with inte g ra to r,we n e e dto ke e p in m ind that the practicalproblemwill m ostlikely dem andno o setfor Type-2 inputs,for exam ple, ram poutputdisturbances (d = A ). s2 The application a Type-2 lter m eeting requirement of the d d ()|s=0 = 0 (6 0 )

a s de sc ri d in S e c tio n2 .1 is n e c e s sa ry orde rto m eetthis requirement. be in Various c a s e so f PI, PID, and PID with lter controller tuning rules arising from plants w ith integrator are de scri d in re fe re n c e [1 ] a n d [2 ] ,and sum m arized T a b le be s in 1 ; n o te th e p ro g re ssio nin controller so ph isticatio n as clo se d -l oop perfo rm an ce req u ire mnts incre ase ! e

PID Tuning Rules for 1st-order

with Deadtime

Pla nts

A sum m aryof the PI, PID, and PID with lter tuning rules for rst-orderplants w ith deadtim e found in Table 3. The PID tuning rule for plants with d e ad tim e is a rise sfrom

IMC: A Comprehensive View

16

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 Controller c(s) No O set


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

P P with lter PI PI with lter PID with lter

Ste p sonly Ste psOnly S tepsand Ramps S tepsand Ramps S tepsand Ramps

T a b le 1 : PID tuningrules for pla w ith integrator nts usinga rst-order Pade appr oxim ationin lieu of the tim e de lay. p = Ke s + 1 K ( s + 1) 2 (2 s + 1 )( s + 1)
s

(6 1 )

(6 2 )

The Pade-appr xim atedplant (6 2 )is a sec o n d -o rde rlant w ith RHP ze ro ;u s in gth e o p a n a ly s isfrom the Example 2: PID Co ntrol sub se ctio n ad sto a PID tu nin grule: le Kc I D 2 + K (2 + ) = + 2 = 2 + = (6 3 ) (6 4 ) (6 5 )

As show n in Rivera et al. [1 ] th e ra tio of the ISE objecti e functionfor the PID v control system J = ISE = (y r)2 dt (6 6 )
0

versusthe optim alISE for a rst-order ith deadtim e w plant Jopt = 2 (6 7 ) can be p lo tte das a fun ctio nof inde pende of , a sn o te din Fig u re3 . F ig ure3 also nt s how s M , w hch represe ts the m axi u m peak of th e n om in alcom plemntary sensitivi i n m e ty function M = sup

(6 8 )

This m e a s u re a n be re la te dto ro b u s tn e ssf th e clo s e doo p s y s te m a s d e s c ri d in [1 ]. c o -l , be

Note that at

0.8 the IMC-PID controller resultsin an ISE value that is o n ly 1 0 % greater

IMC: A Comprehensive View

17

Table 2: IM C -BasedTuning for IdealPID Controllers Using Sim ple Mod e ls c(s) = Kc (1 + ( 1 + D s) I s 1 F s + 1)

Model
K s+1 K 2 s2 +2 s+1 K (s+1) s+1

Input vM
1 s 1 s 1 s 1 s 1 s 1 s 1 s 1 s 1 s2 1 s2 1 s2 1 s2 1 s2 1 s2 1 s2 1 s2

= pq = pqf
1 s +1 1 s +1 s+1 s +1 (s+1) (s+1)( s +1) 1 s +1 (s+1) s +1 (s+1) (s+1)( s +1) 1 s +1 2s +1 (s +1) 2 2s +1 (s +1) 2 (s+1)(2 +)s+1 (s +1) 2 (s+1)(2( +)s+1) (s+1)( s +1) 2 2s +1 (s +1) 2 (s+1)(( +2 )s+1) (s +1) 2 (s+1)(2( +)s+1) (s+1)( s +1) 2 2s +1 (s +1) 2

Kc K
2 + 2+ 2 + 2 2+ 2 2 2+ 2 2+ (+)2 2(+) 2 2 +4 +2 2 +2 + (+)2 2(+)+ 2 2 +4 +2 2+ 2

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

D 2

F 2+

2 2 2

>0 >0 <0 >0 >0 <0


K s

K (s+1) s+1 K (s+1) s+1 K (s+1) 2 s2 +2 s+1 K (s+1) 2 s2 +2 s+1 K (s+1) 2 s2 +2 s+1

2+

2 2 2 +4 +2

2 2+

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

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

>0 >0 <0 >0 >0 <0

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

2 2 2 +4 +2

IMC: A Comprehensive View

18

F ig u re3 : J/J opt and M for the IMC-PID controller (top),and com parison ith otherm ethw ods (bottom): open-l op Ziegler-N i o chols (O-L Z -N ), closed-l Ziegler-Ni oop chols (C-L ZN ), and Cohen-C oon (C-C).

IMC: A Comprehensive View

19

Controller always) Original PI Improved PI PID PID with lter

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

I
+2

D
(2 +) 2 +

F
2(+)

R ecom m end ed( > 0.2

> 1.7 > 1.7 > 0.8 > 0.2 5

+ 2
+2

Tab le 3: IM C -based ingrulesfor PI, PID, and PID w ith lter controllersfor a rsttun orderw ith deadtime syste m than optim al,while m ai ntaining a low value for M . Th e controlled variab lerespon seof the IMC-PID controller for varioussettings is show n in Fig u re 4 . of

The originalPI tuningrule is foundby appr oxim atingthe rst-order delay plant w ith just the rst-order term ,w ithoutd elay : lag p= Kes K s + 1 s + 1 (6 9 )

F ig u re 5 show s a m ar ked deterioration achievable ISE perform ance, in relati e to the v PID tuningrule. At its best setting( 1.3 5 ) the IMC-PI controller resultsin an ISE value that is ove r 5 0 % g re a ter a n o p tim a l,w ith a high value for the com plemntary th e sensitivity function,M 1.4 . The Impr oved PI ru le arise sby incorporating the delay in the time consta of the internal model p nt p= Kes K s + 1 ( + 2)s + 1 (7 0 )

resultin ga s show n in Example 1: PI control in th e tu ningrule: , Kc I = 2 + 2K = + 2 (7 1 ) (7 2 )

The im pr oved PI ru le s, a s th e n a m e im p lie s ,re su lt in s upe rio r pe rfo rm a n c e r th e ove

s ta n d a rdIMC-PI ru le s ;howeve r, th e pe rform a nce b ta in e dfro m th ese rule s va ries a s a o functionof / . A worst-case perform a nce and robustness analysiswith respect to / for a w id e ra n g eof / is prese nted in Fig ure 6 (top). Evaluating the im pr oved PI tuning rule for a specic choice of / = 1.7 show s that the corres on din g p perform ance superior to that of the C ohen-C on and closed-l Ziegler-N i is o oop chols rules over m ostof the / ra n g e , s n o te din Fig u re 6 (bottom). a

IMC: A Comprehensive View

20

Tuning rulesfor a PID w ith lter controller (show n in T a b le 3 ) c a n be o b ta in e da s we ll u sin g(6 2 )and the analysis Example 3: PID with Filter Co ntro l, leadingto the of result Kc I D F 2 + 2K ( + ) = + 2 = 2 + = 2 ( + ) = (7 3 ) (7 4 ) (7 5 ) (7 6 )

Fig ure 7 show s the IS E perform ance tain ed from th ePID w ith lter tuning rule. ob Com par-ing Figure 7 with Figure 3, one noticesthat the IMC-PID w ith lter tuning leads to higherISE than the IMC-PID for the sam e valu e of / ; however, th e PID w ith lter se ttin g sd is- p lay much smoo th er clo se d -l oop respon se s,a s e v id e n ce d Figu re 7 in (bottom ). In in du strial pra ctice,th e smoothn essof the response may well be worth the lo sso f pe rfo rm a n ce te rm sof ISE. in

References
[1 ] R ive ra , D .E ., M. M orari,and S. Skogestad, Internal Model Control 4 . PID Controller Design,Ind. Eng. Chem. Process Des. Dev. 25, 2 5 2 ,1 9 8 6 . [2 ] M o ra ri,M. and E. Zariou.Robust Process Co ntro l, Prentice-Hall,Engle wood Cli s, NJ, 1 9 8 9 .

IMC: A Comprehensive View

21

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

10

12

F ig u re 4 : IMC-PID controlled variable responsesfor a step setpoint ch ang e,for various settings ; s o lid : = 0.8 ; dotted : = 2.5 ; d as h e d = 0.4 . of :

F ig u re 5 : J/J opt and M for theoriginalIMC-PI controller.

IMC: A Comprehensive View

22

Figure 6 : Wo rs t-c a s eJ/J opt and M for the im pr ved IM C-PI controller (top), and o com par-ison (for / = 1.7 ) w ith other m eth ods: closed-l Z iegler-Ni oop chols (Z -N ),and C ohen-C on (C -C ) (bottom ).Solid: J/Jopt ; D a sh e d: . o M

IMC: A Comprehensive View

23

F ig u re7 : J/J opt and M for the IMC-PID with lter controller(top),and controlled variablere spo n s ecom parison ith the IMC-PID rule (bottom ). For bottom g ure,solid: w IMC-PID with lter ( = 0.4 5 );dotted :IMC-PID ( = 0.8 );

Das könnte Ihnen auch gefallen