DOE基礎(chǔ)及其在Molding工序參數(shù)優(yōu)化中的應(yīng)用FundamentalsofExperimentDesignProcessModelProcessControllableinputfactorsUncontrollableinputfactorsInputOutputTheConceptofDOEAdesignedexperimentisatestorseriesoftestsinwhichpurposefulchangesaremadetotheinputvariablesofaprocesssothatwemayobserveandidentifycorrespondingchangesintheoutputresponse.ScreeningExperimentstocharacterizethemainfactorsEverycontrollablefactorsoftheprocessScreeningDesignMainFactorsApplicationofDOEToProcessImprovementQualityLevelDOEImportantPracticalConsiderationsCheckperformanceofgauges/measurementdevicesfirst.

Keeptheexperimentassimpleaspossible.

Checkthatallplannedrunsarefeasible.

Watchoutforprocessdriftsandshiftsduringtherun.Avoidunplannedchanges(e.g.,swapoperatorsathalfway)

ImportantPracticalConsiderationsPreservealltherawdata--donotkeeponlysummaryaverages!

Recordeverythingthathappens.Resetequipmenttoitsoriginalstateaftertheexperiment.CheckTheAssumptionsArethemeasurementsystemscapableforallofyourresponses?Isyourprocessstable?Areyourresponseslikelytobeapproximatedwellbysimplemodels?Aretheresidualswellbehaved?ResidualsResidualsareestimatesofexperimentalerrorobtainedbysubtractingtheobservedresponsesfromthepredictedresponses.Howtoselectandscaletheprocessvariables?IncludeallimportantfactorsBebold,butnotfoolish,inchoosingthelowandhighfactorlevels.

CheckthefactorsettingsforimpracticalorimpossiblecombinationsIncludeallrelevantresponses.Avoidusingonlyresponsesthatcombinetwoormoremeasurementsoftheprocess.DesignSelectionGuidelineNumberofFactorsComparativeObjectiveScreeningObjectiveResponseSurfaceObjective11-factorcompletelyrandomizeddesign

2~4RandomizedblockdesignFullorfractionalfactorialCentralcompositeorBox-Behnken5ormoreRandomizedblockdesignFractionalfactorialorPlackett-BurmanScreenfirsttoreducenumberoffactorsCompletelyRandomizedDesignwithOnePrimaryFactork=numberoffactors(=1forthesedesigns)L=numberoflevelsn=numberofreplicationsandthetotalsamplesize(numberofruns)isN=kxLxn.Compareasinglefactorthathasadifferentlevelsusingthecompletelyrandomizeddisigns.Examplewith3levelsand3replicationsTherunorderoftrialsarecompletelyrandomized,thustheeffectiveofexperimentordercanbeeliminated.Asinthisexample,thereareall9!Waystorunthisexperiment.Thelevelsofthefactoriscalled“1”,“2”and“3”.X1

Response3

1

2

2

1

3

1

2

3

ResultAnalysisAnalysisofVarianceH0:H1:OtherwiseTableofANOVASourceofVariationSumofSquaresDegreesofFreedomMeanSquareF0BetweenfactorlevelsSSFactora-1MSFactorF0=Error(withinfactorlevels)SSEa(n-1)MSE

TotalSSTan-1

IfF0>Fa-1,a(n-1)(0.05)thenwecanconcludethefactor-levelmeansaredifferent.I.e.werejecttheH0RandomizedBlockDesignWhenwecannotcontrolnuisancefactors,animportanttechniqueknownasblockingcanbeusedtoreduceoreliminatethecontributiontoexperimentalerrorcontributedbynuisancefactors.ARBDTableThemodelforaRBDYi,j

=+Ti

+Bj

+randomerrorwhereYi,j

isanyobservationforwhichX1=iandX2=jX1istheprimaryfactorX2istheblockingfactorμisthegenerallocationparameter(i.e.,themean)Tiistheeffectforbeingintreatmenti(offactorX1)Bjistheeffectforbeinginblockj(offactorX2)FullFactorialDesignIftherearekfactors,eachat2levels,afullfactorialdesignhas2kruns.Two-levelfullfactorialdesigns(3factors)BlockingofFullFactorialDesignFractionalFactorialDesignsuseonlyafractionoftherunsspecifiedbythefullfactorialdesign.ResponseSurfaceDesignsCentralCompositeDesignCCC:CentralCompositeCircumscribedCCF:CentralCompositeFaceCenteredCCI:CentralCompositeInscribedBox-BehnkenDesignsBlockingaResponseSurfaceDesignAddingCenterPointsWeaddcenterpointrunsinterspersedamongtheexperimentalsettingrunsfortwopurposes:ToprovideameasureofprocessstabilityandinherentvariabilityTocheckforcurvatureIngeneral,3~5centerpointsareaddedinaFactorialDesign.AnalysisofDOEDataOneExample“Lookat”theDOEData殘差的直方圖(殘差正態(tài)性檢驗(yàn))ProcessModelR-sq:Thecoefficientofdeterminationforamodelrepresentstheproportionofvariationintheresponsedatathatisexplainedbythepredictorsinthemodel.Itiscalculatedastheratioofthesumsofsquaresforregressionoverthetotalsums