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

Because of the ability to handle uncertain information,DS evidence theory has become one of the most widely applied information fusion methods.However,there are two difficult problems in the application of DS theory.One is how to survive the basic probability assignment function(BPA function),the second one is that it will appear paradox results when dealing with conflict evidence.This paper mainly studies and improves the DS method in these two aspects.Based on the improved DS method,a multi classifier decision fusion model is proposedFirst of all,because of orbital vibration and environmental factors,the information obtained by the sensor may have a great conflict.The original improvement method mainly focused on the degree of convergence,but they did not consider conflict resolution.In this paper,an improved DS method based on discounted operator is proposed,and the optimization model of discount coefficient is established to achieve weight optimization.Conflict resolution is carried out while the evidence source discount is corrected.In the next integration,the risk of fusion is reduced,and the degree of focus and the degree of conflict resolution can be flexibly adjusted according to actual needs.The effectiveness of the improved method is illustrated through a classical example.After generating BPA,the improved DS method is compared with other fusion methods,which shows that the DS improvement method based on discounted operator can better deal with evidence conflict,and the recognition rate is higher than other fusion methods.Secondly,previous studies have designed BPA functions from their application background,so there is no generally accepted method for modeling BPA generation.In this paper,A BPA generation method suitable for bogie single fault data is proposed,which used generalized triangular fuzzy function to construct BPA of each fault category.Due to the large number of yawdamper fault categories leads to computational complexity problems such as explosion,this paper improves the original method by setting threshold and reducing the number of focal elements.Then,the improved DS method is used to fuse the generated BPA to get the final decision results.The method is applied to UCI data set to verify the effectiveness of the method.The method is applied to the fault data of high speed train bogie and compared with other BPA generation methods.It shows that the BPA generation method in this paper is simple and effective,and it can obtain higher classification accuracy.Finally,based on the improved DS method,a multi classifier decision fusion model(Mul-DS)is proposed.In order to better evaluate the difference of classifier system,this paper uses the evidence distance as a new index,and it combines with Q statistics to form a comprehensive index to select the optimal multi classifier system.The reliability of the quantum classifier is checked with the confusion matrix,and the credibility is weighted into the BPA function.The DS method based on the improved discount operator in this paper is used as the decision fusion method.The method is applied to UCI standard dataset to verify the effectiveness of the method.In the fault experiment of high speed train bogie,the Mul-DS method is compared with a single classifier,which shows that the Mul-DS method can improve the classification accuracy steadily.