反应釜温度智能控制系统设计 2.docx
《反应釜温度智能控制系统设计 2.docx》由会员分享,可在线阅读,更多相关《反应釜温度智能控制系统设计 2.docx(17页珍藏版)》请在冰豆网上搜索。
反应釜温度智能控制系统设计2
Gain-schedulingfuzzytemperaturecontrollerforone-wayinputsystem
Abstract:
Inmanychemicalandsemiconductormanufacturingprocesses,temperatureisanimportantcontrolparameterforobtainingthedesiredproductquality.Generally,thetemperaturecontrolsystemhasnon-lineartime-varying,slowresponse,time-delayandone-waycontrolinputcharacteristics.Itisdifficulttoestimateaccuratelythedynamicmodelanddesignageneral-purposetemperaturecontrollertoachievegoodcontrolperformance.Hereamodel-freeintelligentgain-schedulingfuzzycontrolstrategyisproposedtodesignatemperaturecontrollerforanironclosedchamberwithheaterinputonly.Theconceptofgainschedulingisemployedtoadjustthemappingrangesoffuzzymembershipfunctionsduringthecontrolprocessforimprovingthecontrolperformance.Theexperimentalresultsshowthatthesteady-stateerrorsofthestepinputresponsesarealwayslessthan0.28Cwithoutovershootbyusingthiscontrolscheme.Itissuitableforindustrialtemperaturecontrolsystems
Keyword:
fuzzycontrol,gainscheduling,temperaturecontrolandone-wayinput.
1Introduction
Temperatureisanimportantcontrolparameterinchemical,materialandsemiconductormanufacturingprocesses.Forexample,materialannealing,thinfilmdepositionandTVglassmeltingfurnaceallneedappropriatetemperaturecontrolsystems.Someofthetemperaturecontrolsystemshaveheatingandcoolingcontrolphasesandothersonlyhaveaheatinginputcontrolphase.Theirdynamicbehaviourshavesignificantdifferences.Thetemperaturecontrolsystemwithheaterinputonlyismoredifficulttomonitorthantwophasecontrolsystemstoobtaingoodcontrolperformance.Howtodesignageneral-purposetemperaturecontrollerwithgoodresponsespeed,smallersteady-stateerrorandwithoutovershootforindustrialimplementationisstillachallengeinthecontrolresearchfield.Currently,on–off
controlandPIDcontrolschemesareemployedincommercialproducts.APIDcontrollerwasproposedin1936.Ithasbeenwidelyusedinindustrialautomaticcontrolsystems.However,howtoadjustthecontrolgainsisthekeyfactorofimplementingaPIDcontroller.Iftheaccuratedynamicmodelofacontrolsystemisavailable,theZieglerandNicholsrule(ZieglerandNichols,1942)andIMCcontrolstrategy(ChienandFruehauf,1990;Riveraetal.,1986)canbeusedtocalculatetheappropriategains.However,theheatingplanthastime-delayandtemperaturedependencenon-linearbehaviours.ItishardtoestablishanaccuratedynamicmodelforaPIDcontrollerdesign.Generally,itneedsatrial-and-errorprocesstoobtainagoodcontrolresponse.Whenthesystemhasexternaldisturbanceorset-pointchange,itstransientresponsemaydeteriorate.Itneedsanonlineoperatortoreadjustitorswitchittomanualcontrol.Thisisnotaconvenientapplicationandtheproductionparametersmaynotmaintaininagoodlevelinmanufacturingprocesses.Hencemodel-freeintelligentcontrolschemeshavegainedtheattentionofresearchers.
Aself-tuningPIDcontrolstrategywasproposedtocontrolwaterbathtemperature(Yusofetal.,1994).Afrequencyloop-shapingtechniqueisemployedtotunethePIDgainsofthetemperaturecontrollerofachemicalvapourdeposition(CVD)diffusionfurnace(GrassiandTsakalis,2000).Theappropriategainsofthisapproacharesearchedbasedonthesystemoutputresponseofanon–offopen-looprelaycontrol.Inaddition,fuzzycontrolhasbeensuccessfulemployedinmanyindustrialprocesses,asithasmodel-freeintelligentcharacteristics.Recently,fuzzycontroltheoryhasbeenusedtoimprovetheadaptivityandrobustnessofaPIDcontroller.AhybridfuzzyandPIcontrolforTVglassmeltingfurnacetemperaturecontrolhasbeenproposed(MoonandLee,2000,2003).ThefuzzylogicschemealsoisappliedtoadjustPIDcontrollergains(ChenandLi,2003;Heetal.,1993;Visioli,2001).ThePIDgainsoftheseapproachesarenon-linearfunctionsoftrackingcontrolperformance.Theycanbeadjustedautomaticallybasedontheoutputerror.Itcanachievebetterrobustness,quickresponseandsmallerovershootthanthatofatraditionalPIDcontroller.However,itisdifficulttodesignageneral-purposefuzzycontrolrulestableandthecorrespondingfuzzyparameters.Hence,aself-organizingfuzzycontroller(LuandChen,1994)andadaptivefuzzycontrolscheme(Haissig,1999)wereemployedtodesignthetemperaturecontrollertoobtainthestableadaptivebehaviour,butthesecontrolstrategiesstillcannotachievebothaquicktransientresponseandanaccuratesteady-stateresponse.
Usually,temperaturecontrolsystemshavenon-lineartime-varyingandtime-delaycharacteristics.Itisdifficulttoestimateanappropriatedynamicmodelformodel-basedcontrollerdesign.Especially,thetemperaturecontrolproblemwithone-wayheaterinputonlyhastime-delayandasymmetriccontrolbehaviour.Itisdifficulttoachieveprecisetemperaturecontrolaccuracywithagoodtransientresponsebasedontraditionalcontrolalgorithms.Herethemodel-freegain-schedulingfuzzycontrolschemeisemployedtodesigntheheaterinputsingle-phasetemperaturecontrollerwithfuzzygainparameters,auto-switchingtoachieveagoodtransientresponseandsmallsteady-stateerror.Ageneral-purposeintelligenttemperaturecontrolleristhedesignedgoalofthispaper.Thetemperaturedynamicresponseperformanceofthefuzzygain-schedulingcontrollerwillbecomparedwiththatofanauto-tuningPIDcontroller.Theexperimentaltestrigisahollowmetalchamberwithheatercontrolinputonly.
2Systemstructure
ThePC-basedcontrolsystemstructureofatemperaturemonitoringsystemisshowninFigure1.ThePCsendsthecontrolvoltageintoanSCRdriverthroughtheD/Acard.A12-bitlowcostASIO-113AD/DA/8255interfacecardisselectedforthiscontrolsystem.TheMAXTHERMOSCRdriverhasa250innerresistancetotransformthe1–5Vcontrolvoltageinto4–20mAcontrolinputcurrentfortheheatingrod.ThisSCRcanmonitorthepoweroutputofasingle-phase110-Vand20-Aalternativecurrentpowersource.Itregulatesthecurrentinputoftheheatingrodforraisingthetemperatureofthehollowmetalchamber.Themediuminsidethehollowcylinderchamberisair.Thetemperatureofthehollowmetalcylinderismeasuredusingaresistancetemperaturesensor(RTD)andfedbackintothePCthroughtheA/Dcard.ThetemperaturecontrolalgorithmisimplementedwithCþþprogram.Ahollowmetalchamberexperimentaltestrigisbuiltforevaluatingthecontrolperformance.Thedimensionofthehollowironcylinderis250mminheightand51mmindiameterwitha10-mmdiameterhollowholeforinstallingtheheatingrodandtheRTDPT100temperaturesensor.TheaccuracyoftheselectedPT100RTSsensorisabout0.28Cwithinthe3008Ctemperaturemeasuringrange.ThesensitivityofthisRTDsensoris0.0015per8C.Thesamplingfrequencyofthefollowingexperimentsissetas40Hz.
3Auto-tuningPIDcontrol
ThekeyfactorswhichinfluencethecontrolsystemperformanceofaPIDcontrollerishowtofindtheoptimalproportionalgain,integraltimeconstantandderivativetimeconstant.Forpracticalimplementation,thesegainsadjustmentsareachievedbyexpertiseoranexperiencedengineer,andatrial-and-errormodification.Itistime-consumingworkandthedynamicresponsebehaviourcannotbeguaranteed.Hence,arelayfeedbackevaluationmethodwasproposedtofindthegainparametersofaPIDcontroller(AstromandHagglund,1984).Firstly,anon–offswitchingcontrolisemployedforthefirsttwocycles.Whenthesystemtemperatureislessthanthesettingcommand,thecontrolpowerisfullyopenedtodrivethetemperatureup.
Figure1PC-basedtemperaturecontrolsystemstructure
Whenthetemperaturereachesthesettingvalue,thecontrolinputisswitchedoffimmediately.Thenthecriticalgain,Ku,andcriticalperiod,Pu,canbefoundfromtheseinput–outputresponsecurvesasFigure2,where
Hence,thegainparameterscanbecalculatedbyusingtheexperiencedformulaofZiegler–Nichols.
Figure2Theestimatingofcriticalperiodandcriticalgainfromon–offcontrol
4Gain-schedulingfuzzylogiccontroller
Sincethistemperaturecontrolsystemhasobvioustime-delayandone-wayinputnon-linearbehaviour,itisdifficulttoestablishanappropriatedynamicmodelfortheprecisemodel-basedcontrollerdesign.Inaddition,theovershoottransientresponseofthetemperaturecontrolsystem,withtime-delayandsingle-phaseheatinginputdynamicfeatures,isdifficulttoavoidandeliminatequickly.Hence,howtodesignageneral-purposetemperaturecontrollerwithsmallovershootandquickresponsewillbeachallengeresearchtopic.Herethemodel-freegain-schedulingfuzzycontrolstrategyisproposedtosolvethisproblem.ThecontrolblockdiagramisshowninFigure3.
Usuallythemotivationofafuzzyapproachisthattheknowledgeisinsufficientandthedynamicmodelhasuncertainty.Fuzzysettheorywasemployedtosimulatethelogicreasoningofhumanbeings.Themajorcomponentsofafuzzycontrollerareasetoflinguisticfuzzycontrolrulesandaninferenceenginetointerprettheserules.
Figure3Fuzzycontrolblockdiagramofthetemperaturecontrolsystem
Thesefuzzyrulesofferatransformationbetweenthelinguisticcontrolknowledgeofanexpertandtheautomaticcontrolstrategiesofanactivator.Everyfuzzycontrolruleiscomposedofanantecedentandaconsequent;ageneralformoftherulescanbeexpressedas
R:
IFXisAandYsiA,THENUisCI
inputvariableseandce.TheyareNB,NM,NS,ZO,PS,PMandPB.Themembershipfunctionsofthesefuzzyvariablesa