Operation Control And Fault Detection Of Discrete Event Systems Using Labeled Petri Nets

Discrete event systems are a class of dynamic systems with event-driven state changes.Generally speaking,flexible manufacturing systems,network communication systems,transportation systems,etc.all conform to the characteristics of discrete-event systems.From the reliability point of view,it is necessary to maintain the non-blocking operation of discrete event systems,improve the efficiency,and realize the detection of faults.Petri nets are effective tools for modeling and analyzing discrete event systems,since they do not only reflect the evolution of the system state intuitively,but also have rigorous mathematical explanations.In the actual discrete event system,the sensors cannot be arranged considering the cost and environmental factors.Therefore,the occurrence of some events is not observable,which introduces a corresponding class of net system,namely,labeled Petri nets.This dissertation focuses on supervisory control,sequence planning and fault detection problems of discrete event systems using labeled Petri net modeling.The main contributions of this dissertation are summarized as follows:1.The problem of multi-property supervisory controller design for λ-free labeled Petri nets is studied.First,the λ-free label Petri net for the corresponding discrete event system is constructed,which does not satisfy boundedness and liveness initially.Then,the solvability of the problem is confirmed,and a necessary condition for the existence of a feasible monitor based on transition priority,i.e.,a necessary condition that the problem is solvable,is proposed and proved.The necessary condition is based on the notion of liveness and transition-invariants.Then,a cyclic behavior graph is constructed that maximally shows the states of the system that satisfy the liveness within a given bound.Finally,an online supervisory controller design method that can realize both boundedness and liveness of the system is proposed,which is a logic-type control of the net system using transition priority.The real-time control based on the dynamic transition priority obtained by the monitor construction algorithm can ensure that the operation of the system is maintained in a range that meets both boundedness and activity.Experiments have proved the effectiveness of the method,which has the advantage of realizing real-time control of the system and maintaining the structure of the original net without increasing the structural complexity of the system.2.The problem of forbidden state supervisory control for labeled Petri nets is addressed.Considering insufficient sensor deployments in an actual system,the unobservable transitions are added to the Petri net model.First,the labeled Petri net model for the corresponding discrete event system is constructed.Next,the basis reachability graph of the Petri net system is calculated,which is a compact representation of the traditional reachability graph.Based on the pre-defined generalized mutual exclusion constraints and the notion of deadlock,an integer linear programming is established to find out the forbidden states,including deadlocks,token-overflow markings and weakly illegal markings.Then,the forbidden markings are recorded on the basis reachability graph.Since the deadlocks and token-overflow markings can be reached by firing an unobservable sequence at a weakly illegal marking,the control objective of the monitor is to avoid the system from entering the weakly illegal states.Finally,a supervisory control algorithm is proposed,which can realize real-time control by restricting the transition priority of controllable transitions in the current state.Experiments demonstrate that the method can effectively mitigate state-space explosion and can achieve the maximally permissive behavior of the system.3.The sequence planning problem for labeled Petri net systems with time and resource constraints is touched upon.For systems with resource constraints,it is a necessary task to improve resource utilization and reduce operation time.First,the corresponding labeled Petri net model is established based on the time constraints and resource limitations of the discrete event system.Due to the consumption characteristics of resources,there exist deadlocks in the reachable states of this net system as soon as the system runs out of resources.Then,based on the basis reachability graph and integer linear programming,we find the target marking of the sequence planning problem,i.e.,the deadlock due to resource exhaustion,and this step can effectively accelerate the determination of the target marking compared with the traditional traversal method.Finally,a class of heuristic search algorithm based on the extended basis reachability graph is proposed,and experiments confirm that the algorithm can effectively reduce the search space and accelerate the search speed.4.The design problem of fault diagnoser for labeled Petri nets is investigated.First,the faults of discrete event systems are modeled as unobservable fault transitions in the corresponding labeled Petri nets.Then,the notion of minimal marking is proposed for an observable transition to characterize all minimal unobservable paths and corresponding markings associated with the observable transition.Then,an algorithm for constructing a fault diagnoser based on the net structure is proposed,which can construct a diagnoser based on the minimal markings and their corresponding paths.The diagnoser visualizes the unobservable firing paths of the original net in the form of observable transitions.The design of the fault diagnoser is aimed at inferring whether a fault transition has fired at the current state,and thus inferring the results of the fault detection,which can be categorized into three categories: normal,abnormal,and possibly abnormal.Finally,the current state of the system can be estimated and the fault detection results can be obtained based on the real-time observations.The experiments prove that this method is more practical than the diagnostic based on reachability analysis in the application of unbounded Petri nets,and the structure of the diagnostic is fixed,so there is no infinite expansion based on the event observation.