Research On Some Related Problems About Model-based Diagnosis

During1970’s, in order to avdid the disaster which produced by the failures, researchersused professional system to diagnosis the system of aeronautics and astronautics andmanufacturing. Model based diagnosis is an intelligent diagnostic reasoning technologywhich overcomes the serious deficiency of the professional system. MBD proves theresearching of artificial intelligence a lot. In the beginning, researchers paid more attention onstatic systems, MBD dignoses the system through the behaviour and structure. According tothe difference between the model’s prediction and the real observation and using logicreasoning, the failure components sets can be figured out. Some researchers paid moreattention on dynamic systems since the researching on static systems has been more thorough.Discrete event dynamic systems attract a lot of researchers. DES can be viewed as a brigebetween static and dynamic systems. It is a foundation of MBD from hybrid system tocontinous dynamic systems.Moreover, many continous dynamic systems can be transformedto DES by quantify. The diagnosis of DES is a kind of dynamic dianosis. We can get thetransition of the states through the events.In this paper, we firstly introduce the basic thoughts of MBD, the processing of diagnosis,the diangosis of DES and the background of researching. Then we discribe the system modelof des. We main discribe and compare the des model based on Petri and automato. Whether asystem model can be clearly diagnosis is confirmed by the diagnosability of DES, that is tosay, any occurred fault can be found in a limited amount of time. A discrete event system canbe called non-diagnostic, iff there are two excutable paths whose observations are the same,and one of the two is faulty. Then we give the methods of testing diagnosability and codeoptimazation.Through the function optimazation based on model, we can reduce theefficiency of the function. In many cases, the dynamic system is unknown. So we need thetest to get the models of the system. Model-Based Diagnosis is a method that derives thefaulty component(s) from the system model and observations. However, in most dynamicsystems, the system models are unknown, therefore, we have to obtain the part modelidentification of the system resulted from the experimental data. Many complex systems arecapable of performing different actions. Each of these actions requires activating differentsubsets of the components. Some actions can be performed in more than one way. Given sucha model, MBD algorithms can diagnose the faulty components utilizing reasoning methods, by comparing the observed system behavior and the expected behavior by the model.Assuming that it is possible to control the activity of the systems’ components, we canactivate a subset of the components, try to perform an action and then analyze the results. Wecan repeat this iteratively for various actions to learn a partial model. Naturally, such anapproach leads to massive failure injection into the studied system. It would take a lot of timeand other resources to get enough experimental data. To solve this problem, by adding threetimes of filtering we propose an algorithm called DePrune-DFBnB, which can effectivelyreduce the time of model identification. Furthermore, we give a novel algorithm, calledReWDBnB, which can filter the input queue on the counter and the set of result to achievepruning. This algorithm can significantly reduce the times of experiment when there are manyactive sets in the component set. Finally, we propose a method called ReSE-tree based onenumerated set tree. The ReSE-tree method can effectively solve the minimal activecomponent support set. The results of the experiments show that our algorithms outperformthe existing ones.