Research On Fault Detection Of Hub Bearing Of High Speed Train Based On Multi-source Data Fusion

The hub bearing of high-speed train is the core component to drive the operation of high-speed railway.However,with the increase of high-speed railway running mileage and the increasingly complex driving environment,it is prone to damage,which lays a hidden danger for the safety of high-speed train operation.Therefore,the fault diagnosis of high-speed train hub bearing is of great significance.Due to the characteristics of multi-dimensional,massive and multi-modal monitoring data of high-speed train hub bearings,it is difficult for traditional fault diagnosis methods to accurately identify fault categories.The diagnostic method combined with multi-sensor fusion can effectively improve the ability of model feature extraction,thereby improving the model diagnostic accuracy and diagnostic efficiency,At present,there are still some shortcomings in multi-sensor fusion,such as low fusion degree and large redundancy of fusion results.Therefore,this paper starts from the defects of multi-sensor fusion methods and conducts in-depth research on multi-level fusion fault diagnosis.The main research contents are as follows:(1)Aiming at the problem that the traditional data-level fusion method has the problem of large fusion signal redundancy and information loss in the fusion process,a variational Bayesian inference fusion model based on information factor measurement is proposed.Firstly,the Gaussian scale transformation formula is used to modify the traditional variational Bayesian model to improve the feature retention rate of the fusion model.Secondly,the model is further updated by combining the total variation regularization penalty factor to enhance the robustness of the model.Finally,the entropy information factor is constructed to update the model input and reduce the information redundancy in the fusion model.The experimental results on the bearing data set and simulation signal data set of Case Western Reserve University show that the proposed method can obtain good fusion results.(2)Aiming at the problems of insufficient feature utilization and redundant model design in traditional feature-level fusion methods,a parallel multi-channel convolutional network model is proposed.Firstly,the inception-v2 module is used to extract the features of the network input data.Then,the underlying features are transmitted step by step in combination with the Densenet network to improve the feature utilization rate of the network model.Finally,a parallel network structure is established to enhance the learning ability of the network for high-level features.The network model integrates features by complementary fusion,and has the ability to prevent overfitting.The experimental results on the bearing data set of Case Western Reserve University and the bearing data set of Jiangnan University show that the proposed method has high accuracy and verifies the effectiveness of the method.(3)Aiming at the problem of low diagnostic accuracy of single-level fusion model,a multi-level fusion model based on data-level fusion and feature-level fusion is proposed.Firstly,the model proposed in research content(1)is used to fuse and integrate the initial number input data.Secondly,the fused data is input into the model proposed in research content(2)for feature complementary fusion,and finally the decision result is output.The model combines the advantages of different fusion level methods and effectively improves the robustness and fusion accuracy of the model.Experiments are carried out on the simulated high-speed train hub bearing data set.The experimental results show that the multi-level fusion method can more accurately characterize the running state of different damage types and damage degrees of the bearing,which is better than the unfused state,data-level fusion and feature-level fusion,and achieves the expected fault detection effect.