Failure Detection, Isolation, and Recovery of Multifunctional Self-Validating Sensor

A novel strategy based on a relevance vector machine (RVM) coupled with principal component analysis (PCA) is proposed for failure detection, isolation, and recovery (FDIR) of a multifunctional self-validating sensor. The working principle and the online updating algorithm of the RVM predictor are emphasized to identify and recover faults. The proposed predictor can effectively isolate multiple simultaneous faults of multifunctional sensors and accomplish failure recovery with high accuracy and good timeliness. Further, it also possesses a good ability of tracking fault-free signals with sudden changes. Failure detection is carried out by using PCA-based squared prediction error statistics. The PCA-RVM method can distinguish the normal signals with sudden changes from faulty signals. The performance of the strategy is compared with other different predictors, and it is evaluated in a real multifunctional self-validating sensor experimental system. Results demonstrate that the proposed methodology provides a better solution to the FDIR of multifunctional self-validating sensors.