Real-time Statistical Detection and Identification of Sensor Incipient Fault using Kalman Filter

This article describes a real-time statistical approach of fault detection for linear time-invariant (LTI) system subject to incipient fault profile. The proposed methodology combines Kalman filter for fault detection index generation and Generalized Likelihood Ratio (GLR) test that cancels out the monitored fault effect dynamics. Maximum Likelihood Estimation (MLE) has been used for signature correlation, statistical distance measure between defined hypotheses and estimation of an unknown parameter to carry out fault identification. To illustrate the effectiveness of proposed method numerical results for temperature sensor incipient fault detection and identification have been presented for a pressurizer of the Pressurized Water Reactor (PWR).