Fault Detection in Non-Gaussian Processes Based on Mutual Information Weighted Independent Component Analysis

Abstract

A novel method which integrates mutual information (MI) with weighted independent component analysis (MI-WICA) is proposed to highlight useful information for non-Gaussian process monitoring. Since the traditional independent component analysis (ICA) may not function well for non-Gaussian process monitoring, the MI-WICA uses MI technology to evaluate the importance of each independent component (IC) within a moving window, and then set different weighting values on the selected ICs to highlight the fault information for fault detection. The proposed method is applied to a simple multivariate process and the Tennessee Eastman benchmark process, and process simulation results demonstrate that the method is superior to those of the regular principal component analysis, ICA methods.