Reliable detection of faults in measurement systems

International audienceThe goal of this paper is to discuss two special aspects of fault detection in measurement systems. First, the main goal of fault detection in measurement systems is to detect the system degradation when it leads to an unacceptable growth of the output errors. All other faults are of no importance; moreover, their detection can be considered by the user as a false alarm. Second, the traditional on‐line fault detection methods are formulated as that of the quickest detection of abrupt changes in a stochastic process. A sensor fault also should be detected quickly but sometimes (especially for safety‐critical applications) the user imposes the constraint on the maximum delay for the detection. Specifically, the user fixes the probability of missed detection within a given time‐to‐alarm.A new performance criterion for this reliable detection of changes in stochastic systems is proposed and two suboptimal change detection algorithms are considered in the paper. The application of these algorithms as Receiver Autonomous Integrity Monitoring (RAIM) schemes to the problem of the Global Positioning Systems (GPS) integrity monitoring is discussed — these self‐contained schemes are of great interest today for many safety‐critical applications