Control system design for a gas turbine engine usi

ControlsystemdesignforagasturbineengineusingevolutionarycomputingformultidisciplinaryoptimizationValceresV.R.eSilvaI;WaelKhatibII;PeterJ.FlemingIIIUniversidadeFederaldeSãoJoãodelRei-PraçaFreiOrlando170-36307352-SãoJoã[e-mail]ieldMappinStreet,[e-mail]aryoptimization(MDO)isconcernedwithcomplexsystemsexhibitingchallengesintermsoforganizationandscale.Thus,itiswellsuitedtobeappliedtocomplexmultivariablecontroldesign.Collaborativeoptimizationisoneapproachfordealingwithcomplexmultidisciplinaryoptimizationproblems.ThreeMDOarchitectures,includingcollaborativeoptimization,areappliedtocontrolsystemdesignforagasturbineengine,inordertoimprovethedesignsearchprocessbyexploringpossiblesolutionswithparallel,butindependentsearchstrands.Theoptimizationiscarriedoutthroughamultiobjectivegeneticalgorithmframework.Keywords:Geneticalgorithms,gasturbines,optimization,PIcontrollers.1INTRODUCTIONThereisasignificantbodyofresearchdevotedtothestudyofdesignandoptimizationofanumberofinteractingorcoupledsystems.Mostofthisresearchtendstoberelatedtoaerostructuraldesignandiscalledmultidisciplinaryoptimization.Thedesignofanairplanerequiresthebringingtogetherofresourcesrepresentingstructures,metallurgy,aerodynamics,performance,controlandotherdisciplinesinordertoproduceanoptimaldesign.ThemainchallengesfacedinMDOdesignproblemsarecomputationalcostandorganizationalcomplexity(Sobieszczanski-SobieskiandHaftka,1996).Thecomplexityofdesignoptimizationdependsonthecomplexityofthepertinentdisciplines,thesizeoftheproblem,andthenatureoftheobjectivesandconstraints.Comparingwithanaggregationofmanydisciplines,theproblemgrowsverymuchincomplexity,ifthereismorethanonedisciplinecontrollingthesamedesignvariablesforaparticularobjective.Thisismainlyduetotheeffectofcouplingbetweenthevariables.Organizationalcomplexityisduetothefactthatthevariousdisciplinestraditionallyreflectdifferentanalysismethods,schoolsofthought,softwareandhardwareplatforms,standards,etc.TheorganizationalchallengeinMDOisforanefficientexchangeofdata,systemsintegrationandotheraspectsofcommunication.Evolutionarycomputingreferstocomputer-basedproblemsolvingsystemsthatuseevolutionaryalgorithms(EAs).EAsgenerallyusecomputationalmodelsthatexploitmechanismsbasedontheneo-Darwiniantheoryofevolution.ThemaintechniquesusedinEAsinclude:geneticalgorithms(GAs),evolutionaryprogramming(EP),evolutionstrategies(ES)andgeneticprogramming(GP).EAsareamenabletoparallelizationandcanhelpreducethecomputationalcost.Thesealgorithmsarestochasticinnatureandcanusuallystartanoptimizationprocesswithoutmuchaprioriknowledge.NoderivativeinformationisrequiredasinthetraditionalgradientbasedmethodsandthishelpsEAsdealwithdifficultsearchspacescharacterizedbymultimodaldisjointfeasibleareas.Mostdesignproblemsaremultipleobjectivesinnature,includingMDOproblems.Theseobjectivesareoftenconflictingorcompeting.TheconceptofParetooptimalityisapowerfulmethodfordealingwithmultipleobjectives.Usingthisapproach,thedesignerisnolongersearchingforasingleoptimum,ratheracompromisesatisfyingthevariousobjectives.andconstraints.Thecollectionofcompromisesolutionsisreferredtoasthenon-dominatedset.Withinthisset,attemptedimprovementinoneobjectivewillresultindegradationinoneormoreoftheothers.EAsareamenabletomultiobjectiveoptimization(MO).ThisisbecauseanEAworksonapopulationofsolutionsinsteadofthetraditionalsinglepointsearch.Thesearchwiththispopulationcanhelpachieveafasterandmorecomprehensivemappingofthetrade-offhypersurface.AnoverviewoftheapplicationofEAstoMDOcanbefoundinKhatibandFleming(1997).2THEGASTURBINEENGINEGasturbineengines(GTE)arehighlynonlinearplantsthathavemultipleinputsandoutputs.Theoperatingconditionsspanextremesoftemperature,pressureandloadconditions.TheengineperformancerequirementscoverawideflightenvelopethatincludesacontinuumofsetpointsofaltitudeandspeedintermsoftheMachnumber.Theserequirementsaddtothecomplexityofdesigningsuitablecontrollersthatcanachievehighperformancelevelswhilemaintainingstabilityandsafeoperationwithminimumoverallcost.HandPI(proportionalandintegral)controllershavebeendesignedfortheGTEusingsimplifiedmodelsobtainedthroughtheresponsesurfacevariablecomplexitymodellingtechnique(Silvaetal.,2007).Itwasalsoobtainedinprovementsonthisengine'sperformancebyreducingfuelconsumption,increasingthrustindashmissionsandminimizingturbinebladetemperature(Silvaetal.,2005).Inthiswork,multidisciplinarycollaborativeoptimizationstructuresareusedtosplitthePIcontrollerdesignprobleminthreeandthus,optimizationiscarriedout.Theenginemodelhasthreeinputs:fuelflow(wFE),exhaustnozzlearea(A8)andinletguidevaneangle(IGV).Sensorsprovidedfromengineoutputsinclude:highpressurespoolspeed(NH),lowpressurespoolspeed(NL),enginepressureratio(EPR)andfanpressureratio.Thesemeasurementscanbeusedtoprovidevariouspairingsofinput-outputforclosed-loopcontrol.Importantenginevariablessuchasthrust(XGN)andsurgemargin(LPSM)cannotbemeasureddirectly.Suchvariablesarecontrolledimplicitlythroughotherrel