Research On Model Based Diagnosis In Discrete Event Systems

Model-based diagnosis is a diagnostic method which only depends on system’s structureand actural observations, and it is an important and active field in artrificial intelligence. Inearlier days, static model-based diangosis paradigm describe the beahviours of system andobservation of components by first-order logic based state transition systems. Basd on theresaoning of first-order logic, the faulty components can be derivated.As the model based diagnosis method apply in real-world applications, the staticdiagnosis paradigm which need applied offline becoming obsolete. Therefore, themodel-based diagnosis in dynamic system proposed. The dynamic diagnosis paradigm can beapplied when the target system running, it can capture the faulty result instantly (withing thetemporal limitions). At the same time, the dynamic method can monitior and review the wholeprocess of the system, the reason of the faults can be gotten also, the diagnostic process aredeeper.Model-based diagnosis in discrete event systems is a dynamic diagnostic method. Thediscrete event systems like the static system when the states last, and some continuoussystems can be modeled by discrete event systems. Model-based diagnosis in discrete eventsystem is paid attention by researchers and engineers as an easy and effeciency method inmodeling for dynamic system.The research on model-based diagnosis in discrete event system can be seperated intothree stages: model buliding, determining the diagnosability and diagnosis. The whole processof model-based diagnosis conclude these stages. First, the model of actual system is built, inorder to described by logic or math fomulations. Second, the diagnosabilty of the model isdetermined, which assure the diagnostic result is correct and exclusive. At the end, diagnosisis carried online, get the result of whether a fault exists, and the reason of fault is given.In the modeling stage, the main works focus on discrete event systems with differentproperty. The models are built, the framwork is constructed. The special properties in discreteevent systems such as fuzzy, stochatic and temporal. The researches on model such as abstract,hierarchical, approximate and equivalent are carried, as the basement of diagnoses.Based on the model in modeling stage, the diagnosability is determined. The research ondiagnosability can be divided as diagnosability in different models and the accordinglyeffeciency promoting. In different model, the diagnosability is determined. Base on thesedifferent models and the methods of diagnosability judging, the effeciency will be promoted.After a system is determined to be diagnosable, the diagnostic system will run, the online process of diagnosis will start. The research on diagnosis will be divided as promotingeffeciency of diagnosis, diagnostic method on different models, and diagnosatic forecast.The following parts will be dicussed: the effecienty promoting of judging diagnosability,the property of diagnosability, incremental method of diagnosis, and incomplete model.A method of judging diagnosability is proposed, which build the diagnoser incrementaly.The construction of diagnoser is divided into two parts: compile offline faulty labels and theincremental fake online judgement. At the offline part, the faulty labels are generated fromfinal states to inicial states, the backward pre-diagnoser is generated. At the online part,traverse pre-diagnoser forward, the faulty label is judged at the branch, then decide whetherprune the searching branch or traverse forward.When the acquired precision of fault in the system is not precise, the acquirment ofdiagnosability decrece also. Diagnosability does not need to distinguish arbitrary two faults,but distinguish whether the faults belong to one set. In this case, the diangosability is calledfaulty set diagnosability. A method which judge the faulty set diagnosability is proposed, andthe according property is proposed, which assure the continuous diagnosability in the processof faults dividing.When the system is diagnosable, the diagnostic system runs. Different from the offlineprocess of modeling and diagnosability judging, the diagnostic method in discrete eventsystems is online. By the offline model and online sequence of observations, the diagnosaticsystem deduces whether a fault exist in system. Once the fault is discovered, the trajectoryfrom the initial state to current state which include this fault is deduced. Usually, thediagnoses start at the initial state.Incremental diagnosis method promote the tridational method. The main idea is choosingin the diagnostic results and save the best one, once a new observations come, the newdiagnostic results will be deduced on the existing diagnostic result and the new observations.The method avoids diagnosis every time from scratch when new observations arrival, theeffecienty and the reused precetage of information are higher.A conflict-based incremental diagnostic method is proposed. At the stage of choosingcandidate diagnoses, the most possible trajectory is chosen by probability as current diagnoses.The other candidate trajectories is saved as a backtrick point, these trajectories are ordered byprobability. Once the current diagnoses is not compatible with the new observations, thebacktrick will happen. At the backtrick point, every candidate diagnoses is reordered by theposterior probability of current observations, and chose to backtrick one more step ordiagnosis at this poion. The method avoids the problem when a candidate is wrong, the wholediagnoses will be carried at the beginning.Incremental diagnosis is a new method to promote effecienty of diagnosis bydecomposing model and reducing diagnostic process. As the size and complexity of modelincreasing, the modeling process from actual system to model may have some incompletecorrespondence, which called incomplete model. Incomplete model is a model which have unknown behaviors. At the stage of offlinemodeling, diagnostic system consider all the behaviors that the system may have. But atactual running, some behaviors which can not compatible with arbitrary trajectory are found.The traditional diagnostic system can not produce this case, so a series of methods whichfocus on incomplete model are proposed.Three cases of incomplete model is proposed: temporal incomplete, event incompleteand casual incomplete. And the methods of diagnosability judging and diagnosis is proposed.The time axis is proposed to help judging diagnosability, the process of judgingdiagnosability is divided into offlin and online. The offline process is before the origin of timeaxis, the discovery method of online incomplete events is proposed. All the incomplete eventsfound online will be added to model by their time points on time axis. In the diagnoser whichbuilds offlin, the incomplete events are added incrementally, and the the diagnoser isconstruted incrementaly, the diagnosability is judged online.A diagnosis method which used in incomplete model is proposed. By adding temporalrelationship into model, and relaxing some temporal constraint, the temporal incompletemodel is diagnosis. The method relaxes the constraint between observation and model, thetemporal incomplete diagnostic result is gotten. At the same time, the incomplete eventswhich are found online will be added into model. The dianogstic result is gotten online.Casual incomplete exists in distributed system. Some relationship among sub-systems isnot enough to describe the local diagnostic results, but the actual faulty reason relates to othersub-systems. The casual graph is built to diagnosis distributed systems together, and thejointed diagnsis is limited in the case that one sub-system gets one fault frequently.