Process monitoring and fault diagnosing of TE process based on the integration method of PCA and cluster analysis

Abstract

Many studies applied the multivariate statistical methods on process monitoring and fault diagnosis of the industrial process. However, a single traditional method cannot obtain an ideal result on detecting some special faults including many kinds of variables. In this paper, an integration of principal component analysis (PCA) and k-nearest neighbor (kNN), a kind of supervised learning methods in Cluster Analysis, is proposed. It combines the advantages of principal component analysis and k-nearest neighbor to solve this problem. PCA-kNN employs PCA to monitor the appearance of fault in the process and applies kNN to diagnose the type of faults. In order to illustrate the efficiencies of the method, the data of the Tennessee Eastman (TE) process is utilized for this study. The simulation result demonstrates the availability of the proposed method.