Nonlinear Filters Based Fault Diagnosis in Nonlinear Stochastic Systems

Abstract

Three methods for fault diagnosis in nonlinear stochastic systems are studied in this paper, which are based on extended Kalman filter (EKF), unscented Kalman filter (UKF), and particle filter (PF). Multiple-model based fault detector and fault classification method with the likelihood ratio test are applied to fault diagnosis of nonlinear systems. In various noise environments, the comparative studies have been carried out on the fault detection and diagnosis for nonlinear stochastic systems with EKF, UKF and PF, respectively. Simulations demonstrate the effectiveness of the fault diagnosis method based on the PF in nonlinear and non-Gauassian stochastic systems.