Sensor Fault Detection of Linear Time-Varying Systems Under Stochastic Noise With Unknown Probability Distributions

This article focuses on the sensor fault detection problem in stochastic linear time-varying systems (SLTVSs). Unlike previous studies that required complete knowledge of noise distributions, this study considers the case where the probability distributions of stochastic noise are unknown. On the basis of the differences in the distributions of faulty and fault-free measurements, the design of residuals and thresholds is transformed into a two-category classification problem modeled by a minimum error minimax probability machine (MEMPM). By solving this formulation, a novel fault detection scheme is developed that achieves an optimal tradeoff between the upper bound of the false alarm rate (FAR) and that of the missed detection rate (MDR). The effectiveness and feasibility of the proposed fault detection method are conclusively validated through simulations and experiments conducted on a rotary steerable drilling tool system.