Fault Diagnosis Of High-Speed Train Bogies Based On D-S Theory

Information fusion has been widely used in the field of pattern recognition.DS evidence theory is one of the important ways to deal with the problem of information fusion,but DS evidence theory will produce contradictory results when merging conflict information.In order to avoid the appearance of paradox results when merging conflict information,therefore researchers at home and abroad have proposed a number of ways to improve the theory of evidence.Firstly,this paper takes the feature set of the bogie fault data of high-speed train as the research basis.There are different feature rankings because of using different single feature selection method.It indicates that conflicts exist between results obtained from different single feature selection method.Therefore a multi-criteria feature selection method MCF-fgoalattain is proposed.In this paper,we propose a multi-criteria feature selection method MCF-fgoalattain by combining the four different single criterion feature selection methods based on improved DS evidence theory by the multi-objective optimization.The validity of MCF-fgoalattain is verified by applying MCF-fgoalattain method on the bearing inner degradation fault data and UCI standard data set and calculating the robustness of MCF-fgoalattain.Then the MCF-fgoalattain method is applied to the feature set of the bogie fault data of the high-speed train.The obtained results show that the MCF-fgoalattain method has improved the fault recognition accuarcy stably.The comparison between MCF-fgoalattain and four single criterion feature selection methods shows that the MCF-fgoalattain method take advantages over the single criterion feature selection method.The comparison between MCF-fgoalattain and other multi-criteria feature feature selection methods shows that the improved DS theory based on multi-objective optimization can better deal with the conflict between single criterion feature selection methods.Secondly,after the feature selection of the fault data feature set of the high-speed train with MCF-fgoalattain method,this paper proposed a fusion method of SVM and DS theory to realize fault identification.The hard output of linear kernel function SVM,RBF kernel function SVM,polynomial kernel function SVM classifier is transformed into probability output by mapping function.And the confusion matrix obtained by the above three classifiers are used to calculate the recognition credibility of the classifier for different categories of targets.Then probability assignment function is constructed according to probability output and credibility,combining the probabilistic assignment function with DS theory fusion rule to form the decision fusion method.After verifying the validity of the decision fusion method on the bearing failure data and the UCI standard data set,we apply it to the sample to be classified,which is selected by MCF-fgoalattain method of the high-speed train bogie fault data.The Comparison of classification results of decision fusion method and SVM classifier with different kernel functions indicates that decision fusion method stably improves the fault recognition accuracy.