PFEKF Filter-Based Fault Detection and Isolation in Non-Linear Systems

In this article, we propose a robust fault detection method for non-Gaussian non-linear systems, isolating various fault types and accurately estimating faults in robotic systems. Leveraging recent advancements in particle filter algorithms, we adopt the combined particle filter with an extended Kalman filter (PFEKF) for its robustness in mitigating estimation errors in strongly non-linear systems. The PFEKF method, integrating the Monte Carlo method and extended Kalman filter equations, is employed to estimate conditional likelihood. We propose a novel fault detection method that utilizes the PFEKF-based CUSUM rule, and we compare its performance to that of the intelligent particle filter (IPF). We introduce a recursive algorithm to calculate an adaptive threshold that minimizes false alarms and undetected fault rates while enhancing detection speed. A fault isolation method for highly non-linear systems uses the enhanced CUSUM rule, with likelihoods calculated by the PFEKF filter. Finally, we accurately estimate faults in robotic systems using only the PFEKF filter, comparing it with the adaptive exogenous Kalman filter (AXKF) approach.