Active Diagnosis of Incipient Actuator Faults for Stochastic Systems

In this article, the active diagnosis problem of incipient actuator faults is investigated for a class of discrete-time stochastic systems. The input is exclusively designed for diagnosis purposes to optimally discriminate among multiple mode hypotheses that correspond to the multiple incipient actuator fault scenarios. Due to infeasibility of expressing the misdiagnosis probability in the closed form, its new upper bound is derived as an alternative. It is proven that this upper bound is a tighter bound on the misdiagnosis probability for incipient actuator faults, compared with the widely used Bhattacharyya upper bound. For computation complexity reduction and diagnosis performance improvement purposes, a novel input design criterion is constructed for active fault diagnosis (AFD) based on the derived upper bound, under which the input design problem can be solved to global optimality. By injecting the optimal input, to fully utilize the diagnostically relevant information in the output data so as to draw a reliable diagnosis conclusion, the resulting measurement output sequence is exploited to make the AFD decision on the basis of the minimum misdiagnosis probability decision rule. Finally, experimental results on a four-tank system illustrate the effectiveness and superiority of the proposed AFD approach.