An improved KPCA algorithm of chemical process fault diagnosis based on RVM

KPCA-SVM algorithm is a combination of kernel principal component analysis (KPCA) and support vector machine (SVM). It could increase the diagnosis time and decrease the diagnosis efficiency, because more relevant vectors are needed when it is used to monitor the on-line complex chemical process. According to this problem, another combined algorithm which is composed of kernel principal component analysis and relevance vector machine (RVM) is proposed in this paper. Firstly, KPCA-RVM algorithm uses KPCA to structure T² statistics and SPE statistics in the feature space to detect fault, and then it takes the non-linear principal component score vector of samples as the input of relevance vector machine to identify the fault modes. KPCA-RVM algorithm is applied to Tennessee Eastman (TE) chemical process and many kinds of fault mode simulation results show that this algorithm not only can obtain higher fault diagnosis accuracy than KPCA-SVM, but also can raise the speed of fault diagnosis obviously owing to the less necessary relevant vectors.