Intelligent Diagnosis Based On Bayesian Network Technology Research And System Development

Railway transportation industry of china has obtained great development in recent years. Railway ministry's "speed raising plan" and the rise of subway require trains and corresponding equipments to be more advanced. Automatic door system (ADS) is an important constituent part of a train, and whether it is on working order or not is very important to the run of a train.It is difficult to obtain adequate knowledge for fault diagnosis, while some knowledge resource isn't being fully utilized. Aimed at this problem, this paper attempts to make use of design information, especially reliability analysis results, for fault diagnosis. This paper researched intelligent fault diagnosis (IFD) technique from aspects of system architecture, knowledge acquisition and organization, diagnosis reasoning and system implementation. The main contents of this paper are as following:Firstly, after the analysis of structure and working characteristic of ADS, the architecture of intelligent diagnosis system is constructed, and the functional model is introduced in detail.Secondly, after the analysis of source of diagnostic knowledge and characteristics of fault, this paper proposed to do diagnosis using reliability analysis results. The methods to obtain diagnostic knowledge from experts, FMEA tables and FTA analysis results are expatiated. Based on the hiberarchy characteristic of products' structure, this paper put forward to organize diagnostic knowledge using product structure tree (PST) and expatiated on its organization model and storage model.Thirdly, based on the causality among 'failure cause', 'failure mode' and 'failure effect', this paper brought forward the method of constructing 'CME (Cause-Mode-Effect)' diagnosis model for a diagnosis object level by level. The construction method, diagnosis reasoning ability and diagnosis reasoning process of 'CME' diagnosis model are expounded. After that this paper proposed to combine 'Rule based diagnosis' with 'Bayesian Networks (BN) based diagnosis'.Fourthly, this paper introduced the implementation of PST based knowledge organization and management and BN based diagnosis.Finally, all works in this paper are briefly summaried and a prospect of next work is listed.