Abstract: Kernel principal component analysis(KPCA) is one of multivariate statistical control methods for solving nonlinear chemical process fault diagnosis. In this paper, it improves KPCA from two aspects. First, in order to improve the accuracy of KPCA for fault detection, a new method combined with wavelet was developed. Second, feature vector sclection(FVS) scheme was adopted to reduce the computational complexity of KPCA whereas preserve the geometrical structure of the data Tennessec Eastman process(TEP) simulations were carried out to show the given approach's effectiveness in process monitoring performance.

Key words: Fault diagnosis, Kernel principal component analysis, Feature extraction, Wavelet transform