Robust Diagnosability Analysis Of Discrete Event Systems Against Intermittent Loss Of Observations

With the rapid development of electronic information technology and industrial automation control,automatic control systems are widely applied in various fields such as household appliances,computer networks,power grids,intelligent manufacturing,aerospace,military industry,etc.Many of them are event-driven,random,and dynamic systems called discrete event systems.Due to a large number of internal components and states,discrete event sys-tems are often complex and precise.Fault events are unexpected behaviors that may have destructive effects on a control system.Therefore,fault-freeness is one of the necessary guar-antees for healthy and stable operations of systems.However,fault events are inevitable in practice.To minimize the damage and loss to systems and people caused by fault events,the research and application of fault diagnosis technology are indispensable.As a hot field of fault diagnosis research,diagnosability analysis focuses on a system's abil-ity to determine whether a fault event has occurred.However,traditional diagnosis mod-els relying on sensors may fail due to the occurrences of an electronic component failure,communication failure,or atmospheric electromagnetic interference,etc.As a result,robust diagnosability has attracted more and more attention since it improves reliability of the di-agnosis technology.This work investigates the robust diagnosability problem of discrete event systems in the presence of the intermittent loss of observations.The main contributions are summarized as follows:1.As a vital attribute of discrete event systems,diagnosability represents the ability to know whether a fault has occurred in the system within a finite delay after its occurrence.Observ-ing and analyzing system language is a commonly used technique in fault diagnosis.For systems modeled by automata or Petri nets,the notion of diagnosability using their respec-tive languages is recalled.As a run-time diagnosis tool,a diagnoser can determine whether a fault has occurred and help analyze the system's diagnosability.For the system modeled by Petri nets,a reachability graph is used in this work to construct the reachability diagnoser.With the help of diagnosers and the notion of indeterminate cycles,a necessary and sufficient condition for the diagnosability is proposed and proved.2.As a result of sensor malfunctions or communication failures between sensors and the di-agnoser,or any other case,a system will experience intermittent loss of observations.When the system loses observations,the diagnoser may stall or give wrong diagnosis conclusions.This problem calls for a robust diagnoser.Through comprehensive analysis and modeling of intermittent observation loss,an operation called language dilation is proposed.With the language dilation operation,a robust diagnoser is developed,which eliminates the influence of intermittent observation loss.With robust diagnosers,the notion of indeterminate cycles is updated.A necessary and sufficient condition for robust diagnosability is proposed and proved.3.When generating the Petri net reachability graph necessary for constructing a reachability diagnoser,one may face the state explosion problem.As a Petri net gets more complicated,the complexity of its reachability graph also increases exponentially.In this case,a large amount of computing power and storage space are consumed or even wasted.To solve this problem,a compact representation of a system's reachable markings is proposed,namely a basis reachability graph.With the basis reachability graph,a lightweight diagnoser named a basis reachability diagnoser is developed,together with a robust basis reachability diagnoser obtained by dilating the basis reachability graph.Compared with reachability graphs,the advantages of basis reachability graphs in construction complexity are confirmed and the efficiency in robust diagnosability analysis is improved.