Detection and repair faults of sensors in sampled control system

In order to improve reliability and safety of computer control system, it is very important to detect and to diagnose faults in sensors. Generally, pulse-type faults in sensors result in outliers in a sampled time series of computer control system. Based on the fact that the median estimator is the most robust one among all estimators, a new algorithm for detection and magnitude identification of abrupt outliers is established in this paper. The algorithm is composed of four parts: outlier-tolerant filtering, detection of pulse-type faults, magnitude identification and outliers repairing. This new algorithm can be used directly not only to deal with pulse-type faults in sensors but also to eliminate isolated outliers as well as patches in a sampled time series. Results from some simulated examples indicate that this new algorithm is practical and efficient.