Research On The Method Of Discrete Event System Model-based Incremental Diagnosis

Today, the function of man-made systems in industrial and large network has becomemore and more powerful, and the behaviors of system also become more complex. When thesystem is faulty, and the fault that is not timely treated will result in catastrophicconsequences. Obviously, it is particularly important to timelydiagnose and exclude failures,so artificial intelligence community has made many methods of diagnosis system's fault.Model-based diagnosis is a new intelligent reasoning technology, and largely promotesdiagnostic techniques improvement. Before diagnosing system, model-based diagnosticmethod firstly create a logical method for the system according to the structure and behaviorof the system. Moreover, the diagnoser diagnoses the system that whether is failure. If a faultoccurs,diagnoserwillexplainthefault.Model-based diagnosis methods are generally divided into static and dynamic diagnosismethod.Discreteeventsystemmodel-baseddiagnosisisadynamicdiagnosismethod,andthefocus study of the artificial intelligence community. However, when modeling large-scalediscrete event systems, the general global diagnostic methods exist the bottleneck problem,and the efficiency of calculating diagnostic is lower. In order to improving diagnosticefficiency, reducing the difficulty of modeling and reducing diagnostic space are to beresolved.Nowadays, the incremental diagnosis method is one way to solve these problems. Notonly because the process of incremental diagnosis is only limited to automaton model of anobservation windows, but also previous diagnostic result can be reused, which improve theefficiency of diagnosis. The incremental diagnosis is one of the most effective diagnosticmethods.Inthediagnosisprocess,basedonnon-exhaustiveengine(NEDE)incrementaldiagnosticmethods may produce many multi-candidate diagnostic results. However, non-exhaustivediagnosis engine will not return all of these candidates, only return the highest prioritycandidate diagnosis, and continually extend diagnosis on each local diagnostic result thatalready be elected. If the incremental diagnosis can not be extended, combining observationwindowandimmediatelyreturningtotheinitialstaterestartdiagnosis.In order to overcome the defect of NEDE, we propose method of based on backtrackingincremental diagnosis is extended on the NEDE method. The method use automata chainmodel to modeling for the system, narrowing the space of diagnosis. When producingmulti-candidate diagnosis, this method will choose only one diagnostic result among thesecandidates, and reserve the terminate state of the result as backtracking point and already ordered candidate diagnoses. Incremental diagnosis based on backtracking has established agood backtrack mechanism, which make not hesitate to return to the initial state re-diagnosis,but return to the location of backtracking point according to the current location and re-selectthe priorityof the diagnosis, when the selected diagnostic result conflict with the observationwindow.