Professional Documents
Culture Documents
MohammadShamsuzzoha
SigurdSkogestad
DepartmentofChemicalEngineering
NorwegianUniversityofScienceandTechnology(NTNU)
Trondheim
Motivation
DesboroughandMiller(2001):Morethan97%ofcontrollersarePID
VastmajorityofthePIDcontrollersdonotuseDaction.
PIcontroller:Onlytwoadjustableparameters
butstillnoteasytotune
Manyindustrialcontrollerspoorlytuned
ZieglerNicholsclosedloopmethod(1942)ispopular,but
Requiressustainedoscillations
Tuningsrelativelypoor
Bigneedforafastandimprovedclosedlooptuningprocedure
Outline
1.
2.
3.
4.
5.
6.
7.
ExistingapproachestoPItuning
SIMCPItuningrules
Closedloopsetpointexperiment
CorrelationbetweensetpointresponseandSIMCsettings
Finalchoiceofthecontrollersettings(detuning)
AnalysisandSimulation
Conclusion
1.Commonapproach:
PItuningbasedonopenloopmodel
Step1:Openloopexperiment:
Mosttuningapproachesarebasedonopenloopplantmodel
gain(k),
timeconstant()
timedelay()
ke-s
g(s)=
s+1
Problem:Loosecontrolduringidentificationexperiment
Step2:Tuning
Manyapproaches
IMCPID(Riveraetal.,1986):goodforsetpointchange
SIMCPI(Skogestad,2003):Improvedforintegratingdisturbances
IMCPID(ShamsuzzohaandLee,2007&2009)fordisturbancerejection
Alternativeapproach:
PItuningbasedonclosedloopdataonly
ZieglerNichols(1942)closedloopmethod
Step1.Closedloopexperiment
UsePcontrollerwithsustainedoscillations.Record:
1. Ultimatecontrollergain(K u)
2. Periodofoscillations(Pu)
Step2.SimplePIrules:Kc=0.45KuandI=0.83Pu.
AdvantagesZN:
Closedloopexperiment
Verylittleinformationrequired
Simpletuningrules
Disadvantages:
Systembroughttolimitofinstability
Relaytest(strm)canavoidthisproblembutrequiresthefeatureofswitchingtoon/off
control
Settingsnotverygood:Aggressiveforlagdominantprocesses(TyreusandLuyben)and
quiteslowfordelaydominantprocess(Skogestad).
Onlyforprocesseswithphaselag>180 o(doesnotworkonsecondorder)
Thiswork.
ImprovedclosedloopPItuningmethod
Want to develop improved and
simpleralternativetoZN:
Closedloop setpoint response
withPcontroller
Identifykeyparametersfrom
setpointresponse:
UsePgainabout50%ofZN
2.SIMCPItuningrules
Firstorderprocesswithtimedelay:
ke-s
g(s)=
s+1
PIcontroller:
1
c s =K c 1+
Is
SIMCPIcontrollerbasedondirectsynthesis:
Kc =
k +
c
I =min , 4( c +)
Fastandrobustsetting:
c =
7
3.Closedloopsetpointexperiment
Procedure:
SwitchtoPonlymodeandmake
setpointchange
Adjustcontrollergaintoget
overshootabout0.30(30%)
Recordkeyparameters:
1.ControllergainKc0
2.Overshoot=(ypy)/y
3.Timetoreachpeak(overshoot),tp
4.Steadystatechange,b=y/ys.
Estimateofywithoutwaitingtosettle:
y=0.45(yp+yu)
AdvantagescomparedtoZN:
*Notatlimittoinstability
*Worksonasimplesecondorderprocess.
ClosedloopstepsetpointresponsewithPonlycontrol.
Closedloopsetpointexperiment
Overshootof0.3(30%)withdifferents
Variousovershoots(10%60%)
1.5
e s
g
10s 1
30%
1.25
1.25
=100
Setpoint
/=100 (Kc0=79.9)
0.75
OUTPUT y
OUTPUT y
/=10 (Kc0=8.0)
overshoot=0.10 (Kc0=5.64)
overshoot=0.20 (Kc0=6.87)
0.5
/=5 (Kc0=4.012)
=2
0.75
/=2 (Kc0=1.636)
/=1 (Kc0=0.855)
0.5
overshoot=0.30 (Kc0=8.0)
overshoot=0.40 (Kc0=9.1)
/=0.4 (Kc0=0.404)
overshoot=0.50 (Kc0=10.17)
0.25
0.25
overshoot=0.60 (Kc0=11.26)
=0
setpoint
0
0
10
15
time
20
25
0
0
10
15
time
20
25
Small:Kc0smallandbsmall
30
Estimateofyusingundershootyu
Conclusion:
y0.45(yp+yu)
10
Data:15firstorderwithdelayprocessesusing5overshootseach(0.2,0.3,0.4,0.5,0.6).y s=1
4.CorrelationbetweenSetpointresponse
andSIMCsettings
Goal:FindcorrelationbetweenSIMCPIsettingsandkey
parametersfrom90setpointexperiments.
e- s
g (s)
Consider15firstorderplusdelayprocesses:
s 1
/=0.1,0.2,0.4,0.8,1,1.5,2,2.5,3,5,7.5,10,20,50,100
Foreachofthe15processes:
ObtainSIMCPIsettings(Kc,I)
Generatesetpointresponseswith6differentovershoots(0.10,
0.20,0.30,0.40,0.50,0.60)andrecordkeyparameters(Kc0,
overshoot,tp,b)
11
CorrelationSetpointresponseandSIMCPIsettings
Controllergain(Kc)
90cases:PlotKcasafunctionofKc0
Kc
10%:
30%:
A=0.87 A=0.63
60%:
A=0.45
Kc
=A
K c0
Kc0
12
Fixedovershoot:
SlopeKc/Kc0=Aapprox.constant,
independentofthevalueof/
AgreeswithZN(approx.100%overshoot):
Original:
Kc/Kcu=0.45
TyreusLuyben: Kc/Kcu=0.33
Conclusion:Kc=Kc0A
0.9
2
A=1.152(overshoot) -1.607(overshoot)+1.0
0.8
A=slope
A
0.7
0.6
0.5
0.4
0.1
overshoot
0.2
0.3
0.4
overshoot (fractional)
0.5
0.6
(shouldnotbeextendedoutsidethisrange).
CorrelationSetpointresponseandSIMCPIsettings
Integraltime(I)
SIMCrules
Case1(largedelay):I1=
Case2(smalldelay):I2=8
Kc =
I =min , 4( c +)
Case1(largedelay):
=2kKc(substitute=IintotheSIMCruleforKc)
kK c =kK c0 K c K c0 kK c0 A
kK c0 =
b
(1-b)
(fromsteadystateoffset)
Conclusionsofar: I1 =2A
14
1-b
Stillmissing:Correlationfor
k +
c
c =
Kc =
0.5
k
Correlationbetweenandtp
/=0.1
0.5
/tp
/=1
0.43 (I1)
0.4
/tp
/=8
0.3
0.305 (I2)
/=100
0.2
Use:
/tp=0.43forI1(largedelay)
/tp=0.305forI2(smalldelay)
0.1
overshoot
0
0.1
0.3
0.5
tp
0.6
Overshoot
Conclusion:
15
b
I =min( I1 , I2 ) min 0.86A
t p , 2.44t p
1-b
5.Summarysetpointovershootmethod
FromPcontrolsetpointexperimentrecordkeyparameters:
1.ControllergainKc0
2.Overshoot=(ypy)/y
3.Timetoreachpeak(overshoot),tp
4.Steadystatechange,b=y/ys
ProposedPIsettings(includingdetuningfactorF)
K c = K c0 A F
A= 1.152(overshoot) 2 - 1.607(overshoot) + 1.0
b
I =min 0.86A
t p , 2.44t p F
1-b
16
6.Analysis:SimulationPIcontrol
1.25
Firstorder+delayprocess
s
e
g
5s 1
OUTPUT y
0.75
intrainingset
similarresponseasSIMC
0.5
Proposed method with F=1 (overshoot=0.10)
Proposed method with F=1 (overshoot=0.298)
Proposed method with F=1 (overshoot=0.599)
SIMC (c==1)
0.25
0
0
20
t=0:Setpointchange
17
40
time
60
80
t=40:Loaddisturbance
Analysis:SimulationPIcontrol
2
Puretimedelayprocess
OUTPUT y
1.5
g e s
intrainingset
0.5
0
0
12
18
time
18
24
30
Analysis:SimulationPIcontrol
3
Integratingprocess
2.5
g e s s
OUTPUT y
1.5
intrainingset
1
Proposed method with F=1 (overshoot=0.108)
Proposed method with F=1 (overshoot=0.302)
Proposed method with F=1 (overshoot=0.60)
SIMC (c==1)
0.5
0
0
20
40
60
time
19
80
100
Analysis:SimulationPIcontrol
1.5
Secondorderprocess
1.25
OUTPUT y
g
1
Notintrainingset
0.75
0.5
Proposed method with F=1 (overshoot=0.127)
Proposed method with F=1 (overshoot=0.322)
Proposed method with F=1 (overshoot=0.508)
SIMC (c=effective=0.1)
0.25
0
0
10
time
20
1
s 1 0.2s 1
ResponsesforPIcontrolofsecondorderprocessg=1/(s+1)(0.2s+1).
Analysis:SimulationPIcontrol
1.25
Highorderprocess
g
OUTPUT y
0.75
Notintrainingset
0.5
0.25
0
0
21
1
s 1 0.2s 1 0.04s 1 0.008s 1
10
time
15
20
ResponsesforPIcontrolofhighorderprocessg=1/(s+1)(0.2s+1)(0.04s+1)(0.008s+1).
Analysis:SimulationPIcontrol
Proposed method with F=1 (overshoot=0.106)
Proposed method with F=1 (overshoot=0.307)
Proposed method with F=1 (overshoot=0.610)
SIMC (c=effective=1.5)
5
Thirdorderintegrating
process
1
g
2
s s 1
OUTPUT y
4
3
2
Notintrainingset
1
0
0
40
80
120
time
22
160
200
Analysis:SimulationPIcontrol
2
Firstorderunstableprocess
e s
g
5s 1
OUTPUT y
1.5
Notintrainingset
NoSIMCsettingsavailable
0.5
0
0
23
20
40
time
60
80
Analysis:SimulationPIcontrol
EffectofdetuningfactorF
1.25
Secondorderprocess
g
OUTPUT y
0.75
0.5
F=1.0
F=2.0
F=3.0
F=0.8
0.25
0
0
6
time
24
(overshoot=0.322)
(overshoot=0.322)
(overshoot=0.322)
(overshoot=0.322)
8
10
1
s 1 0.2s 1
Kc
Ms
0.8
11.29
0.77
1.96
1.0
9.031
0.958
1.74
2.0
4.52
1.92
1.36
3.0
3.01
2.87
1.24
6.Conclusion
ProbablythefastestPItuningapproachintheworld
FromPcontrolsetpointexperimentobtain:
1.ControllergainKc0
2.Overshoot=(ypy)/y
3.Timetoreachpeak(overshoot),tp
4.Steadystatechange,b=y/ys,
Estimate:y=0.45(yp+yu)
PItuningsforSetpointOvershootMethod:
(Shamsssetpointmethod)
K c = K c0 A F ,
25
b
I =min 0.86A
t p , 2.44t p F
1-b
REFERENCES
26
strm,K.J.,Hgglund,T.(1984).Automatictuningofsimpleregulatorswithspecificationson
phaseandamplitudemargins,Automatica,(20),645651.
Desborough,L.D.,Miller,R.M.(2002).Increasingcustomervalueofindustrialcontrolperformance
monitoringHoneywellsexperience.ChemicalProcessControlVI(Tuscon,Arizona,Jan.2001),
AIChESymposiumSeriesNo.326.Volume98,USA.
Kano,M.,Ogawa,M.(2009).ThestateofartinadvancedprocesscontrolinJapan,IFACsymposium
ADCHEM2009,Istanbul,Turkey.
Rivera,D.E.,Morari,M.,Skogestad,S.(1986).Internalmodelcontrol.4.PIDcontrollerdesign,Ind.
Eng.Chem.Res.,25(1)252265.
Seborg,D.E.,Edgar,T.F.,Mellichamp,D.A.,(2004).ProcessDynamicsandControl,2nded.,John
Wiley&Sons,NewYork,U.S.A.
Shamsuzzoha,M.,Skogestad.S.(2010).Reportonthesetpointovershootmethod(extendedversion)
http://www.nt.ntnu.no/users/skoge/.
Skogestad,S.,(2003).SimpleanalyticrulesformodelreductionandPIDcontrollertuning,Journalof
ProcessControl,13,291309.
Tyreus,B.D.,Luyben,W.L.(1992).TuningPIcontrollersforintegrator/deadtimeprocesses,Ind.
Eng.Chem.Res.26282631.
Yuwana,M.,Seborg,D.E.,(1982).Anewmethodforonlinecontrollertuning,AIChEJournal28
(3)434440.
Ziegler,J.G.,Nichols,N.B.(1942).Optimumsettingsforautomaticcontrollers.Trans.ASME,64,
759768.
SIMCPItuningrules
Ondimensionlessform,
theSIMC(c=)
K c ' =kK c =0.5
K I' =
kK c
1
=max 0.5,
I
16
ScaledproportionalandintegralgainforSIMCtuningrule.
K I =Kc
Note:
27
isknownastheintegralgain.
Integralterm(KI?)ismostimportantfordelaydominantprocesses(/<1).
ProportionaltermKc?ismostsignificantforprocesseswithasmallertimedelay.
Abstract
28
ThePIcontrolleriswidelyusedintheprocessindustriesduetosimplicityand
robustness,Ithaswiderangesofapplicabilityintheregulatorycontrollayer.
TheproposedmethodissimilartotheZieglerNichols(1942)tuningmethod.
Itisfastertouseanddoesnotrequirethesystemtoapproachinstabilitywith
sustainedoscillations.
Theproposedtuningmethod,originallyderivedforfirstorderwithdelay
processesandtestedonawiderangeofotherprocessesandtheresultsare
comparablewiththeSIMCtuningsusingtheopenloopmodel.
Basedonsimulationsforarangeoffirstorderwithdelayprocesses,simple
correlationshavebeenderivedtogivePIcontrollersettingssimilartothoseof
theSIMCtuningrules.
ThedetuningfactorFthatallowstheusertoadjustthefinalclosedloop
responsetimeandrobustness.
TheproposedmethodisthesimplestandeasiestapproachforPIcontroller
tuningavailableandshouldbewellsuitedforuseinprocessindustries.