The Study On A Fuzzy Reasoning Method Orienting Fault Diagnosis

The fault diagnosis is a technology of fault judgement and diagnosis through analyzing the logic relationship among fault information. How to construct effective diagnosis model, and how to eliminate the effect of the uncertain problems, are key problems to be solved in fault diagnosis research. Among them, the analysis of the uncertain problems is a difficulty. In view of different backgrounds, the researchers at home and abroad introduce many methods of analysis and processing from different angles.Aimed at solving the series of uncertain problems existing in the fault diagnosis system, especially the distribution problem of impact factor and the gathering time uncertainty of fault information, based on the Fuzzy sets theory and Petri nets, the paper presents an extended fuzzy Petri nets (EFPN) reasoning method. This method converts each uncertain problem into quantitative numerical description, and constructs EFPN model for diagnosis reasoning. The EFPN model defines two parts in the structure according to the different roles. One part named "analysis", is used to analyze the accuracy and time attribute of the fault information, and calculate the credibility of the information. The other part named "diagnosis", is used to calculate the credibility of diagnostic conclusion based on the results of "analysis" part. In order to quantitative calculate the uncertainties, the model sets up a series of functions. Two points are the most important. Aimed at the distribution problem of impact factor, the paper adopts analytic hierarchy process(AHP) method to distribute the impact factor of information in the rule, to ensure that the distrubution results achieve the satisfied consistency and reliability, and then sets the weight function on the input arcs of diagnosis transitions. Aimed at the time uncertainty of information, on the analysis transtition, the model defines fuzzy timestamp and membership function to analyze and calculate the influence of information time factor for information credibility, and make the optimal distribution scheme. Each function in the model complements each other. On that basis, the model gives different startup strategies for different transitons to limit the evolution of the status marking strictly. The fault diagnosis example of substation shows that, this method can do modeling for fault reasoning of substation, conduct qualitative analysis and quantitative calculation, and can complete the fuzzy reasoning smoothly when the informations are incomplete. The reasoning reasults of this model can be assisted to judge the fault of protection device, and the model can avoid the problem of combinatorial explosion by setting the life cycle of transiton. This method has good semantic description ability, strong fault-tolerance and adaptability. It can be used in more complex equipment fault diagnosis process.