Application of Kernel Independent Component Analysis for Multivariate Statistical Process Monitoring.pdfVIP

JournalofDonghuaUniversity (Eng．Ed．)Vo／．26，No．5 (2009) p n 卸№ lication of Kernel Independent Component Analysis for tivariateStatisticalProcessMonitoring WANGLi(T 丽) ，SHIHong—bo(侍洪波) ResearchInstituteofAutomation，EastChinoUniversityofScienceandTechnology，Shanghai200237．China Abstract：In this research，a new faultdetection method data(secondorderstatistics)butalsoreducinghigherorder based on kernelindependentcomponentanalysis (kernel statisticaldepe ndences，[Csrevealmoreusefulinformation ICA) is developed． Kernel ICA is an improvement of from observed data than PCsE ． However， the ICA independentcomponentanalysis (ICA)，and isdifferent approachmodelsthe data ashaving been generated by from kernelprincipalcomponentanalysis(KPCA)proposed independentsourcesthrough alinearmapping，thereality fornonlinearprocessmonitornig．The basic idae of onl- is not the casel4_． Hyvarine first presented nonlinear approach is tousethe kernelICA to extractindependent independent component analysis (NICA ) approach in componentseffi cientlyandtocombinetheselectedessential l999 ．W hereas．theNICA approach usually hastodeal indepe ndent compo nents with process monitoring withacomplicatenonlinearoptimizationproblem ，andthe tcehniques．I (theslim ofthesquarde independentscores) computationalcostisslower．There are some researches and squared predictionerror (SPE)chartsareadopted as aboutkernelICA ，which have been employed to face statisticalquantities．Tbeproposed monitoring method is recognitionL ， and for nonlinear statisticaJ proc ess applide t0TennesseeEastman process，and thesimulation monitoring[ ．However，thekernelICA theyused isa resultsclaerlyshowtheadvantagesofkernelICA monitoring two—phase algorithm： whitened KPCA plus ICA． The incomparisontoICA monitoring．