Kernel Fisher Discriminant Analysis Using Feature Vector Selection for Fault Diagnosis

Abstract

Kernel-based Fisher discriminant analysis (KFDA) has been widely applied in pattern recognition and classification such as face recognition. It is proved which is a powerful method for nonlinear discriminant. In this paper, it is used for fault diagnosis. It has two aspects in this work. First, the wavelet de-noising preprocessing with KFDA scheme is proposed. Second, a geometry-based feature vector selection (FVS) scheme is adopted to reduce the computational complexity of KFDA whereas preserve the geometrical structure of the data. Tennessee Eastman process (TEP) simulation are carried out to show the given approachpsilas effectiveness in process monitoring performance.