Nonlinear Statistical Process Monitoring and Fault Detection Using Kernel ICA.pdfVIP

维普资讯 JournalofDonghuaUniversity (Eng．Ed．)Vo1．24．No．5 (2007) Nonlinear StatisticalProcess Monitoring and FaultDetection Using KernelICA ZHANGXi(张 曦)，YANWei．wu(阎威武)，ZHAOXu(赵 旭)，SHAOHui_he(邵惠鹤) DepartmentofAutomation，ShanghaiJiaotongUniversity．Shanghai200240，China Abstract：A novelnonlinearprocessmonitoring and fault independence up to second order statistics information detectionmethod basedon kernelindependentcomponent (mean and variance) while constraining the direction analysis(ICA)isproposed．ThekernelICA method isa vectorstobeorthogonal一 ． tw~phasealgorithm ：whitenedkernelprincipalcomponent In recent years， independent component analysis (KPCA) plusICA．KPCA spheresdata and makes the (ICA)hasarousedwideresearch interestsandbecomea data structurebecomeaslinearlyseparableaspossibleby popular toolfor blind source separation and feature virtueofan implicitnonlinear mapping determined by extraction．Itisalso being used asa powerfultoo lfor kerne1．ICA seekstheprojectiondirectionsintheKPCA processmonitoring[一 ． ICA hasno orthogonalconstraint whitenedspace，makingthedistributionoftheprojected and involves higher—order statistics， i．e．， not only daat asnon-gaussian aspossible．The application to the decorrelates the data (second order statistics) butalso fluid caatlyticcracking unit (FCCU) simulated process reduces higher order statistical dependencies． Hence， indicates that the proposed process monitoring method in