Model for Unanticipated Fault Detection by OCPCA

Abstract

Detection and diagnosis of unanticipated fault has inevitably become a critical issue for PHM (Prognostics and Health Management), especially in the fields of robot, spacecraft and industrial system. It is difficult to overcome this problem since there is lack of history information, prior knowledge and dealing strategy for unanticipated fault. In this paper, a general processing model for unanticipated fault detection and diagnosis is constructed, then, a detection method, named OCPCA (One-class Principal Component Analysis), is proposed. Every OCPCA detector is trained by data from single pattern, and the testing task is to determine whether the testing data is from the very pattern. If the unanticipated fault data is rejected by all OCPCA detectors, then the detection task is accomplished. TEP (Tennessee-Eastman Process), a widely used simulated system based on an actual industrial process, is used to verify the detection of unanticipated fault. The results demonstrate the validity of the proposed model and method.