Research On Uncertain Reasoning Of Intelligent Fault Diagnosis Based On Ontology

With the improvement of machinery equipment on automation, it is very important to intelligent fault diagnosis based on the check-up of the running status. But for large and complex equipment, there are many problems about the complexity of fault mechanism, inaccurate knowledge representation, lack fault sample and so on, which brings to uncertain factor for fault diagnosis. Therefore, it is usually used to realize the process of intelligent diagnosis by uncertain representation and reasoning in intelligent fault diagnosis system. Ontology, as a model tool, can describe knowledge model on semantic and knowledge level, provides a formal description of the concept, and lays a strong foundation for knowledge share. However, Ontology could neither represent overlaps or intersection degree among concepts, nor support the reasoning of concepts or individual partial information. Bayesian Network theory is proved to one of the most effective way in uncertain knowledge reasoning. Accordingly this thesis adopts probability extend of the Web Ontology Language so that it can support the reasoning of incomplete and inaccurate knowledge based on Bayesian network.Combining with the NSFC project, the thesis takes example for rotor-vibrated diagnosis by the application of Ontology theory in the field of knowledge engineering and Bayesian network in uncertain reasoning, and makes in-depth research and analysis from the following contents:(1) The thesis analyses the ubiquitous uncertain problems and some solutions in the fault diagnosis of rotary machine, and builds a semantic represention model of fault diagnosis by which encodes the probability informations used to descripe uncertain degree to fault ontology by augmenting ontology semantic.(2) Such a translation would be based on a set of rules and processes that would transtalate the ontology with probabilistic information to Bayesian network, which can support Bayesian network reasoning. The rules are the translation from concepts to nodes, the translation from relations to arcs, the translation of the property values and the structure of Condition Probability Tables.(3) The thesis takes Bayesian network as the bottom reasoning and lays down the uncertain seasoning mechanism. It is verified the feasibility and the reasoning accuracy of the ontology_based Bayesain network method for resolving uncertain knowledge on search_based approximate reasoning algorithm in the fault diagnosis of rotary machine. The results show that the combination of Ontology and Bayesian Network theory can give full play to strong points on knowledge description and Bayesian network reasoning, which can promote knowledge sharing and get the knowledge according to partial probabilistic information to improve the accuracy of reasoning.