Diagnosis of Temporal Faults in SPN with Trajectory-Observers and Control Charts

This paper deals with the problem of detection and isolation of temporal faults in timed stochastic discrete event systems. Partially observed timed Petri nets are used to model the considered systems. Temporal faults are considered that correspond to significant variations of the probability density function used to compute the firing durations. Moving average control charts and trajectory-observers are combined in order to detect the variations of the firing durations mean value. The advantages of the proposed analysis are to detect variations in time series when parameters vary slowly and to isolate the faults thanks to a set of elementary trajectories consistent with the successive measurements that are of minimal-size.