Simulation-Supportd Decision Making仿真支持决策

Simulation-SupportedDecisionMakingGeneAllenPresidentDecisionInciteInc.Simulation–AToolforDecisionMaking•QuicklyIdentifyandUnderstandHowaProductFunctions:–Whatarethemajorvariablesdrivingfunctionality?–Whatarethecombinationsofvariablesthatleadtoproblemsincomplexsystems?•AbilityExistsToday–DuetoadvancesincomputecapabilityCorrelationMapsGenerationofCorrelationMapsCorrelationMap–a2-DviewofaResultsDatageneratedfromMonteCarloAnalysis–IncorporatesVariabilityandUncertainty–UpdatedLatinHypercubesampling–IndependentoftheNumberofVariables–Resultswith100runs–DoesNotViolatePhysics–Noassumptionsofcontinuity–“Notelegant,onlygivestherightanswers.”CorrelationMapstoUnderstandCause&EffectInputVariablesOutputVariables•Ranksinputvariablesandoutputresponsesbycorrelationlevel•FollowsMIT-developedDesignStructureMatrixmodelformat•FiltersVariablesBasedonCorrelationLevelUpperright–positivecorrelationLowerleft–negativecorrelationACorrelationMapMetaModelofDesignAlternativesCorrelationMap:-IncludesAllResults-HighlightsKeyVariablesStochasticSimulationTemplate100MCSrunsGenerationofCorrelationMapsMonteCarloAnalysisSolution:Establishtolerancesfortheinputanddesignvariables.Measurethesystem’sresponseinstatisticalterms.SourcesofVariability•MaterialProperties•Loads•Boundaryandinitialconditions•Geometryimperfections•Assemblyimperfections•Solver•Computer(round-off,truncation,etc.)•Engineer(choiceofelementtype,algorithm,meshband-width,etc.)x1x2x3y1y2TheFundamentalProblem…VariabilityStructuralMaterialScatterMATERIALCHARACTERISTICCVMetallicRupture8-15%Buckling14%CarbonFiberRupture10-17%Screw,Rivet,WeldingRupture8%BondingAdhesivestrength12-16%Metal/metal8-13%HoneycombTension16%Shear,compression10%Facewrinkling8%InsertsAxialloading12%Thermalprotection(AQ60)In-planetension12-24%In-planecompression15-20%Source:Klein,M.,Schueller,G.I.,et.al.,ProbabilisticApproachtoStructuralFactorsofSafetyinAerospace,ProceedingsoftheCNESSpacecraftStructuresandMechanicalTestingConference,Paris,June1994,CepaduesEdition,Toulouse,1994.TheDeceptionofPreciseGeometryGeometryimperfectionsshouldbedescribedasstochasticfields.MonteCarloResultsshowRealityUnderstandingthephysicsofaphenomenonisequivalenttotheunderstandingofthetopologyandstructureoftheseclouds.Singlecomputerrun=AnalysisCollectionofcomputerruns=SimulationUnderstandingMCSResults•Simulationgeneratesalargeamountofdata.•Atypicalsimulationrunrequiresaround100solverexecutions.•Eachcombinationofhundredstothousandsofvariablesproducesapointcloud.Ineachcloud:•POSITIONprovidesinformationonPERFORMANCE•SCATTERrepresentsQUALITY•SHAPErepresentsROBUSTNESSKEY:•REDUCEtheMulti-DimensionalCloudtoEASILYUNDERSTOODINFORMATION•CondenseintoaCORRELATIONMAP•VariablesaresortedbythestrengthoftheirrelationshipMonteCarloSimulationResults12ofthe782Dviewsthatresultedfromasimulationwith6outputsfromascanof7inputswithuniformdistributions.Numberof2DViewsofResults=Sumofallintegersfrom1to(NumberofVariables-1)•Displayscondensedinformationfromhundredsofanalysisruns.•CorrelationMap=StructuredInformation=Knowledge•ACorrelationMaphelpsanengineer:–Understandhowasystemworks.•Howinformationflowswithinthesystem.•howvariablesandcomponentscorrelate.–Makedecisionsonhowadesignmaybeimproved.•Identifydominantdesignvariables.•Useasinputforstochasticdesignimprovement.–Findtheweakpointsinasystem.–Findredundanciesinadesign.–Identifyrulesthatgoverntheperformance(“ifAandBthenC”).ThereareNOalgorithmstolearn.Theengineerconcentratesonengineering,notonnumericalanalysis.CorrelationMaps:UnderstandingCauseandEffectDesignImprovementProcess1234TargetPerformanceIteration•AutomotiveandAerospacecompanieshavecontinuedtoexpanduseofprocesssince1997•BMW,Audi,Toyota,Mecedes,NissanandJaguarhaveexpandedComputerClustersforStochasticCarCrashSimulationtaking10’sofpoundsfromcarmodeldesigns.•Aerospacecompaniesapplyingtoimproveaerospacedesigns.Aleniareducedweightofnewcommercialairlinertailby6%.APPLICATIONSCourtesy,AleniaAeronauticaCourtesyofBMWAGProcessforDecisionSupport•Modelamulti-disciplinarydesign-analysisprocess•Randomizetheprocessmodel•RunMonteCarlosimulationofthemodel•ProcessResults–CorrelationMapsshowingCauseandEffect–Outlieridentificationshowinganomalies–DirectionforDesignImprovement•Identifywhatinfluencesfunctionality•AddressUncertaintyandVariation•Providescredibilityinmodeling&simulation•Resultscloudsrepresentwhatispossible•Easytouse•Nomethodsoralgorithmstolearn•Reducesriskthroughbetterengineering•Takesallinputsintoaccountviceusinginitialassumptions•ChangingthegeneralengineeringprocessCorrelationMaps-FilterComplexitywhileModelingReality