Research Of Fault Diagnosis Algorithm Based On Multivariate Statistics Analysis

Past three decades, with the rapid development of mass production and complexity in process industries, reliability and security are being greatly needed to avoid large economical loss brought by accidents and even breakdowns of industrial productions. Therefore, it is very important for industrial integrated automation to research and develop process supervisory systems which integrate the function of control, supervision and diagnosis, which has important theoretical and practical value. Multivariate statistical method is an important research branch of fault diagnosis area, which is more practical because it does not rely on mathematical models, as well as easy access to a large number of process data.In this paper, important aspects of process fault detection using multivariate statistical theoty are presented and studied systematically, which based on Tennessee Eastman(TE) process data for research background:1 Convention principal component analysis (PCA) is introduced and fault diagnosis is made using PCA, which shows a good performance in fault detection and a general effect in fault diagnosis.2 For non-linear characteristics of TE, a method of non-linear—kernel principal component analysis (KPCA) is introduced. A new fault identification method which is built on the basis of differential contribution plots and the derivative of kernel function is presented for KPCA's difficult identification, and possesses higher accuracy than the methed of PCA in identifying faulty variables.3 This article also discuss to a new fault diagnosis method—correspondence analysis (CA), which is an expansion of PCA. Significat performance improvements are showed in monitoring and diagnosis using CA over PCA and useful attempts is made in fault diagnosis based on multivariate statistical analysis.