Sensor Fault Detection and Isolation Based on Variable Moving Window KPCA

Abstract

Kernel principal component analysis (KPCA) has been widely applied for fault detection. The time-varying property of industrial processes requires the adaptive ability of the KPCA model. This paper introduces online KPCA methods for fault detection. The moving window KPCA (MWKPCA) and variable moving window KPCA (VMWKPCA) methods update the KPCA model according to the process status. To locate the faulty sensor, a fault isolation algorithm should be carried out once a fault is detected. Thus, a partial VMWKPCA is proposed to achieve the fault isolation task. To demonstrate the performance of the proposed method, it is applied for fault diagnosis of air quality monitoring networks. The simulation results show that the proposed method effectively identifies the source of the fault.