概率PCA多元统计方法在过程监控中的应用研究

《概率PCA多元统计方法在过程监控中的应用研究》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、江南大学硕士学位论文概率PCA多元统计方法在过程监控中的应用研究姓名：杨沛武申请学位级别：硕士专业：检测技术与自动化装置指导教师：刘飞20080601AbstractAfterlongtimerunning，therearesomechangesinanyproducingsystem，whichwillunavoidablyinfluencethequalityofproductsandevenresultingreataccidents．Therefore，thetraditionalmethodsent

2、irelybasedonmanpowerareoutdatedandcannotsatisfythecomplicateddesireofqualitycontr01．Havingnouseofcomplexmechanismmodel，multivariablestatisticalprocessmonitoringmethodcanmonitorprocessthroughextractingimportantinformationfromrawdatausingstatisticalmethodandt

3、hentransformingthemintoseveralsignificativeindices．Themethodnotonlytakessufficientuseoftheexistinginformationandiswellrealizable，butalsogreatlyreducestheprocedureofprocessmonitoringsystem，whichhasbeendevelopedmorethanthirtyyears，inwhichlotsofresearchresults

4、havebeenacquiredandappliedwidely．Probabilisticprincipalcomponentanalysis(PPCA)firstlyassumethedistributionoflatentvariablesanderrorvector,secondlyevaluatethegenerativemodelbytheexpectationandmaximization(EM)algorithm，SOitcalldetectfaulteffectivelyandperform

5、son—linefaultidentification。whichmakeitattractivebothinindustryandacademia．HoweverPPCAisalinearway,thepreconditionsofitsapplicationarethatprocessarenormallydistributedandnoauto-correlationamongthem．Butmostofindustrialprocessarecomplicatedandalwaysviolatethe

6、preconditions，SOthePPCA．basedmethodbehavesunsatisfactorily．AimingatthedisadvantagesofthePPCA—basedmethod，themaincontributionsareasfollows：1．Thedissertationproposesallimprovedmonitoringwaywhichmonitorsthenormofwhitenmeasurementvariablesandperformon—linefault

7、identificationbymonitoringeverywhitenvariable，SOtheloadofmonitoringisreduced．AtlastPCAandPPCAarecomparedinmonitoringprocessofchemicalseparation．2．Aimingatthestrongdynamiccharacteristicoftheindustrialprocess，usingEMalgorithm，thedynamicPPCAmodeliSbuilttocopew

8、iththedatamatrixextendbytimeseries．Accordingtothetechnique，staticPPCAcanbeextendedtomonitordynamicmultivariateprocess，andauto—correlationamongprocessvariablesiseffectivelyeliminated．3．Onthemonitoringof