Online Fault Prognosis of Labeled Petri Nets

This paper addresses the problem of online fault prognosis of discrete event systems modeled by bounded and unbounded-labeled Petri nets (LPNs). Fault prognosis, also known as fault prediction, has as its main goal to detect that an unobservable fault event will inevitably occur in the future given the current event observation. Using the concepts of basis markings and coverability trees, we show that the prognosis can be performed online with a copy of the system LPN, referred to as the fault-driven Petri net. The contributions of this paper are the following: we present a general formulation and a solution to the problem of fault prognosis for both bounded and unbounded LPNs whose unique assumption is that the LPN has no cycles of unobservable transitions. The proposed method can perform an online fault prognosis even after fault occurrences, making it suitable for the prognosis of repetitive fault occurrences.