Sensors Incipient Fault Detection and Isolation Using Kalman Filter and Kullback–Leibler Divergence

Abstract

This paper presents a real-time statistical technique for sensors incipient fault detection and isolation (FDI). The proposed approach comprises fault detection index and fault signature formulation using Kalman filter under relaxed assumption on the monitored system stability. A fault decision statistics is generated by combining the Kullback-Leibler divergence of considered hypotheses with an exponential weighted moving average. Furthermore, fault detection performance has been characterized using missed detection and false alarm probabilities. Fault-to-noise ratio (FNR) is acting as a comparative criterion between the fault and noise level for statistical characterization. Numerical results of single and multiple sensors incipient FDI for pressurized water reactors (PWRs) pressurizer illustrate the effectiveness of the proposed method.