Research On Failure Diagnosis In Model-based Diagnosis

The system, whose internal component is independent, operates as an wholeindividual. A simple physical circuit is a system. An aerospace equipment is also asystem. Regardless of the size of a system, the internal component of the system oftenfails. Especially in large-scale industrial systems, a slight malfunction may causeincalculable consequences, which causing irreparable damage and disaster. In order toavoid unnecessary losses and disasters, a new branch of Artificial Intelligence raisesup---fault diagnosis.In the earlier years, among the existing methods of fault diagnosis, the mosttypical one is based on expert system. The method needs strong professionalknowledge, thus it has a lower portability. The method is too restrictive. Then manyscholars proposed a lot of fault diagnosis methods. By far the most widely used one ismodel-based diagnosis. Model-based diagnosis raised up in the1980s. It is a new typeof intelligent diagnostic reasoning technology. Model-based diagnosis is an importantfield of Artificial Intelligence. The principle of model-based diagnosis is based on thedifferences between the structure and behavior of the system model, and the actualbehavior and expected behavior, to deduce the final diagnostic results. Finally, we canget the failure components. In model-based diagnosis, computing minimal hitting setsis an important procedure to express the final diagnosis results.Initial methods of model-based diagnostic only confined in static systems.However, the most of the actual systems in our life are dynamic. Therefore, alongwith the development of fault diagnosis technology, many experts and scholarsproposed some methods of model-based diagnosis which adapt to dynamic systems.Especially in recent years, discrete event system, which quantified the dynamicsystem as a static system. Then treat the dynamic system as a static one to diagnose. Discrete event systems attract more and more attention of scholars.In the reality, there are many discrete event systems, such as Queueing Systems,Communication Systems, Manufacturing Systems and Database Systems.Model-based diagnosis for these kinds of discrete event systems, we usually use thefinite state automata to model them. And then we establish the diagnosers to diagnosethe system.In large-scale discrete event systems, the complexity of the fault diagnosis isvery large, no matter from space complexity or from the time complexity. Therefore,component-based discrete event systems with decentralized information have receiveda lot of attention. However, when diagnose a failure system, the method whichseparates the system into components is very few.In this paper, based on the difference between the actual behavior and expectedbehavior of the model-based diagnosis, as well as simplification of systems under thediscrete event systems, model separation, the merger of the system model, thediagnosability of the systems, we research the following studies:1) for static systems,we propose a new method to compute minimal hitting sets, based on the matrix. Westorage the information of the cluster of conflict collections into the matrix, andintroduce the frequency of the elements as heuristic information, to compute theminimal hitting sets.2) for discrete event systems, according to the system ofdecentralized information, we insert communication events into the system.3) dividedthe system into a number of components, and diagnose the components.4) fornon-diagnostic components, we refine the component firstly. And then we merge therefined components. Finally we can get the final diagnosability of the system.