An efficient ICA-DW-SVDD fault detection and diagnosis method for non-Gaussian processes

Abstract

Independent Component Analysis (ICA) has been extensively used for detecting faults in industrial processes. While applying ICA to process monitoring, the inability of identifying the important components affect the fault diagnosis ability. For further improving the competence of ICA, this paper proposes an approach integrating ICA, Durbin Watson (DW) criterion and Support Vector Data Description (SVDD) to monitor non-Gaussian process for detecting faults. In the proposed approach, namely ICA–DW–SVDD, ICA is a non-Gaussian information extractor from original variables, DW identifies dominating ICs, and SVDD plays the role of fault detector. This paper also discusses the retracing method to detect original variables causing disturbance in the process. One simulation case and the Tennessee Eastman Process are used to demonstrate the effectiveness of our proposed approach.