State estimation and fault prediction with partially observed Petri nets

This article concerns the prevention of fault in discrete event systems (DES). For this purpose, DES are modeled with partially observed Petri nets (POPNs) that include the definition of sensors used to measure the events and markings. Observation sequences result from this modeling. The firing sequences and initial markings consistent with a given observation sequence are systematically obtained. Future states and events are predicted and degrees of confidence are computed for these predictions. State estimation and fault prediction result from this computation. Finally detectability and predictability are defined for POPNs and discussed with respect to the sensor configuration in order to quantify the quality of estimation and prediction.