Design-Expert-8.0使用指南-Multifactor-Response-Surface

DX8-04E-MultifactorRSM-P3.docxxRev.5/17/10Design-Expert8User’sGuideMultifactorRSMTutorial–Part31MultifactorRSMTutorial(Part3–AdvancedTopics)AddingPropagationofError(POE)totheOptimizationDetailsaboutthevariationinyourinputfactorscanbefedintoDesign-Expert®software.Thenyoucangeneratepropagationoferror(POE)plotsshowinghowthaterroristransmittedtotheresponse.Lookforconditionsthatminimizetransmittedvariation,thuscreatingaprocessthat’srobusttofactorsettings.ThistutorialshowshowtogeneratePOEfromanexperimentdesignedbyresponsesurfacemethods(RSM).IfyoustillhavetheRSMdataactiveinDesign-ExpertsoftwarefromPart2inthisseriesoftutorials,continueon.Ifyouexitedtheprogram,re-startitanduseFile,OpenDesigntoopenyourdatafile(tut-RSM-opt.dxp).Otherwise,gobackandsetitupasinstructedinMultifactorRSMTutorialsParts1and2.StartbyclickingtheDesignnodeontheleftsideofthescreentogetbacktothedesignlayout.ThenselectView,ColumnInfoSheet.EnterthefollowinginformationintotheStd.Dev.column:time:0.5,temperature:1.0,catalyst:0.05,asshownbelow.ColumnInfoSheetwithfactorstandarddeviationsfilledinNoticethesoftwarealreadyenteredthestandarddeviationfortheanalyzedresponse,Conversion(4.1…).Becauseyouhaven’tchangedanyotherdata,Design-Expertremembersyourpreviousanalysischoices–youcansimplyclickthroughtheanalysisbuttons.Foryourinformation,right-clickingthebuttonstotheleftofresponsenamesallowsyoutospecifyadifferentstandarddeviation.Optiontoenteradifferentstandarddeviationforresponse2MultifactorRSMTutorial–Part3Design-Expert8User’sGuideUndertheAnalysisbranchclicktheConversionnode.ThenjumppasttheintermediatebuttonsforanalysisandclicktheModelGraphsbutton.SelectView,PropagationofError.(Thisoptionwaspreviouslygrayedout–unavailable–becausethestandarddeviationsforthefactorshadnotyetbeenentered.)Ifthegraph’sbackgroundissolidyellow,switchtoGraduatedintheContourgraphshadingoptionthat’sfoundbyright-clickingoverthegraphandchoosingSurfaceGraphsinGraphPreferences.ContourgraphforPOENowselectView,3DSurfaceorsimplyclick3DSurfaceonthefloatingGraphsTool.3DSurfaceviewofthePOEGraphDX8-04E-MultifactorRSM-P3.docxxRev.5/17/10Design-Expert8User’sGuideMultifactorRSMTutorial–Part33ThelowerthePOEthebetter,becauselessoftheerrorincontrolfactorswillbetransmittedtotheselectedresponse,whichresultsinamorerobustprocess.However,POEwillonlyworkwhentheresponsesurfaceisnon-linear,suchasfortheConversion.Whenthesurfaceislinear,suchasthatforActivity,erroristransmittedequallythroughouttheregion.Therefore,Design-Expertsoftware‘graysout’theoptionasshowninthescreenshotbelow.POEoptionnotenabledforlinearmodelssuchasthatforActivityresponseForanin-depthexplanationofPOE,attendStat-Ease’sworkshopRobustDesignandToleranceAnalysis.Call,orcheckourwebsiteforaschedule.Nowthatyou’vegeneratedPOEforConversion,let’sgobackandaddittotheoptimizationcriteria.UndertheOptimizationbranchclicktheNumericalnode.ForthePOE(Conversion)settheGoaltominimizewithaLowerLimitof4andanUpperLimitof5asshownbelow.SetGoalandLimitsforPOE(Conversion)NowclicktheSolutionsbuttonatopthescreentogeneratenewsolutionswithadditionalcriteria.OnthefloatingSolutionsToolclickRampsorselectViewandchoosetheRampFunctionGraphoption.(Note:Duetorandomstartingpointsforthesearches,youmayseeslightdifferencesonyourscreenversustheshotbelow.)Clickthealternativesolutions(2,3,…).Watchthereddots.Whatchangesdoyousee?4MultifactorRSMTutorial–Part3Design-Expert8User’sGuideRampsviewforoptimizationwithPOE(Yourresultsmaydiffer)GobacktotheSolutionsToolandpressthePop-OutView.PopoutviewshowsreportalongsideoframpsExitoutofthePop-OutViewbypressingtheXbuttonatitsupperrightcorner.PresstheGraphsbuttonatopyourscreen.Tomaketheviewsimilartowhatwehadbefore,ontheFactorsToolpaletteright-clicktheC:catalystandchangeittotheX2axis.DX8-04E-MultifactorRSM-P3.docxxRev.5/17/10Design-Expert8User’sGuideMultifactorRSMTutorial–Part35OptimalSolutionwithaddedPOEcriteriaTheaboveoptimalsolutionrepresentstheformulationthatbestmaximizesconversionandachievesatargetvalueof63foractivity,whileatthesametimefindsthespotwiththeminimumerrortransmittedtotheresponses.So,thisshouldrepresentprocessconditionsthatarerobusttoslightvariationsinfactorsettings.DesignEvaluationDesign-ExpertofferspowerfultoolstoevaluateRSMdesigns.Designevaluationoughttobeaccomplishedpriortocollectingresponsedata,butitcanbedoneafterthefact.Forexample,youmayfinditnecessarytochangesomefactorlevelstoreflectsignificantdeviationsfromtheplannedsetpoint.Oryoumaymissrunsentirely–atleastforsomeresponses.Thenitwouldbewellworthwhiletore-evaluateyourdesigntoseethedamage.Forare-capofwhat’sbeendonesofar,gototheDesignbranchandclicktheSummarynode.DesignsummaryThesummaryreportsthattheexperimenterplannedacentralcompositedesign(CCD)intwoblocks,whichwasgearedtofitaquadraticmodel.ClicktheEvaluationnodeandnoticeDesign-Expertassumesyouwantdetailsonthisdesigned-fororderofmodel.6MultifactorRSMTutorial–Part3Design-Expert8User’sGuideDesignevaluation–modelchoiceClicktheResultsbuttonforaninitialreportshowingannotationsonbydefault.DesignevaluationresultsDX8-04E-MultifactorRSM-P3