Decentralized Fault Prognosis For Labeled Petri Net Systems

To deal with the fault prognosis problem in systems with complex logical structures,e.g.,large scale cyber physical systems,this paper proposes the use of discrete event system as the model of the CPS's logic.Then,for the fault prognosis problem in DES,this paper transfers it into a standard Petri net language form of description with the help of labeled Petri nets.Since most large scale CPSs have distributed structures,the control and fault prognosis theory under the centralized setting,e.g.,the property of prognosability,can not be applied directly.Therefore,this paper presents the notion of decentralized fault prognosis and extends the existing results in the field of automaton into Petri net language.Specifically,this paper investigates the problem of decentralized fault prognosis of discrete event systems modeled by unbounded labeled Petri nets.We assume that the system is monitored by a set of local agents(prognosers)with local observations so that they can predict the occurrence of fault in the system as a team.It is known in the literature that the notion of coprognosability provides the necessary and sufficient condition for the existence of a set of decentralized prognosers so that any fault can be predicted before its occurrence without false alarm.In this paper,we mainly investigate the verification of coprognosability for systems modeled by labeled Petri nets.We show that coprognosability is decidable even when the Petri net is unbounded.Also,a concept of coprognosability with guaranteed performance bound is proposed,instead of knowing merely whether a fault will occur or not,now we are informed with how far a fault can be predicted in advance and how soon a fault will actually happen since an issued alarm,which makes the property more powerful during a prognoser design.Our result provides a critical theoretical background for the decentralized fault prognosis problem in DES modeled by labeled Petri nets.