基于過程能力指數的貝葉斯分析及其應用.pdf

1、湖南大學碩士學位論文基于過程能力指數的貝葉斯分析及其應用姓名：孫雄志申請學位級別：碩士專業(yè)：統(tǒng)計學指導教師：許鵬20071110AbstractStatisticalprocessqualitycontrolisausefulandeffectivemeansinproductqualitycontrolofmechanicalprocessprocesscapabilityindices(PCIS)aresinglenumbermea

2、sureforthecapabilityofaprocessinmeetingspecificationlimitsTheseindiceshavebeenwidelyusedinmanufacturingindustrialTheindividualizeddemandfortheproductsincreaseconstantlyandthesmallbatchmanufacturinghasalreadybecomethemain

3、modetllatenterprisesmanagegraduallyItisagreatchallengeundoubtedlyforPCIS，ToprobethePCISmethodthatissuitableforthesmallbatchmanufacturingbecomeaheavyfocusofcontrolledfieldaboutpresentqualityalreadyThispapermainlyfocuseson

4、themostwidelyusedPCISCp，Cm，C嘲l(fā)byBayesianmethodFirstlyweanalyzedthethreetypesofprocesscapabilityindicesCv，c恥，％inthetraditionalfrequentiststatisticaltheoryofthepointestimatorandconfidenceintervalsandthepercentagenonconform

5、ing(NC)associatedwithPCISThemostcommonlyrecommendedestimatorofPCISisbiased，wederiveditsunbiasedfactorSecondlywestudiedofdatanormalassumptionsanduseJohnsoncurvefittingofnonnoFfflaldatatransformationThUSwealsoCanmakereason

6、ableestimatorsundertheconditionthattheprocesscapabilityindicesarenon—normaldistributionItprovedthatthemethodismoresimpleandtheresultsmoreaccurateandreliableFurthermorewetheresearchedtheproblemofBayesianestimatorandthelow

7、erBayesianconfidencelimitontheprocesscapabilityindexCpWederivedconditionalexpectationandthehighestposteriorestimatorofPCISCpbyusingnoninformativeandconjugatepriordistributionsrespectivelyItshowedthroughusingBayesianmetho

8、dsthantraditionalfrequentiststatisticalmethodsofcalculationofprocesscapabilityindexCallreflectedtheactualproductionssituationmoretrulyFinallywestudiedtheproblemofthelowerBayesianconfidencelimitontheprocesscapabilityindex

9、C。TheJeffreys’non—informativepriordistributiondependingonFisher’SinformationisexploredTheposteriordistributionofC。isconductedThelowerBayesianconfidencelimitdesignedisbasedontheposteriordistributionofC。combinedwiththeappl