Kernel local fisher discriminant analysis for fault diagnosis in chemical process

Abstract

Though Fisher discriminant analysis (FDA) is an outstanding method for fault diagnosis, it is difficult to extract the discriminant information in complex industrial environment. One of the reasons is that FDA can not remain the geometric structure information of the sample space truly due to non-Gaussian and nonlinear structures characteristics of data in industrial process. In this paper, kernel local fisher discriminant analysis (KLFDA) is proposed to solve the problem. The proposed approach is applied to Tennessee Eastman process (TEP). The results demonstrate that KLFDA shows better fault diagnosis performance than conventional FDA.