Abstract: The fault diagnosis model based on principal component analysis has limitation in nonlinear time varying process.Based on the characteristics of the batch process,we introduced the theory of kernel transformation into the data extraction of nonlinear space,and proposed an improved fault diagnosis model based on multiple kernel principal component analysis.This method shows good performance for the nonlinear problem of process data and the full extraction of nonlinear information,where the nonlinear principal element can be rapidly extracted in the high dimensional feature space.The method was tested by comparison.The results show that the method has good accuracy and real-time performance in the process of slow time varying batch process.

Key words: Batch process,Fault detection,Multiway kernel principal component analysis,Piecewise modeling