Research On Fault Diagnosis Methods Of High-speed Train Bogie Based On Improved-CEEMDAN And 1-D CNN

Recent years have witnessed the development of high-speed railway,and the corresponding demand for trains,along with it,has jumped considerably.Thus,the security and comfortableness of the trains are of great importance under the high-speed operation.As one of the most crucial parts of HSTs,the bogie carries the weight and load of the train and guarantees the normal operation of the train on the rail.However,during the long-term operation period,the track irregularity and wheel rail wear may contribute to the bogie’s performance degradation and even components failure,which brings great threat to the safe operation of trains.Therefore,the efficient and effective health assessment and performance monitoring approach for the train bogies has been highlighted.In order to improve the accuracy and robustness of bogies fault diagnosis for HSTs,a bogie fault diagnosis model based on ICEEMDAN(Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise)and 1-D CNN(One-Dimensional Convolutional Neural Network)is proposed in this paper.The main contents are as follows:First,the fundamental principles and development processes of the series of empirical mode decomposition algorithms are briefly introduced.In order to select the algorithm which is suitable for the subsequent bogie vibration signal decomposition,the performance of different decomposition methods is presented through the artificial simulation signal experiments.This part provides a theoretical basis for fault feature analysis of the signal.Secondly,the structure of the bogie,the dataset and the channel selection model are described at length.The function of three key parts of bogies and the corresponding fault forms are introduced.Moreover,the train operation simulation experiments and the acquisition method of fault data on the basis of SIMPACK are illustrated.Then,a channel selection model combining the results of multiple methods is designed.Through the analysis of the original signal of each channel,the better channel combination is selected,so as to reduce the coupling and redundancy between different channels.In order to deal with the complexity of bogie vibration signal and make full use of the information contained in the original signal,a fault diagnosis model based on 1-D CNN is proposed.Using one-dimensional convolution kernel with remarkable feature extraction ability to extract the fault features,the accuracy of 96.62% and 72% are achieved in the HST bogie complete disassembly dataset and single component fault dataset respectively,which verifies that 1-D CNN has good applicability in bogie fault diagnosis.Finally,considering that the vibration signal of bogie contains strong random noise,a fault diagnosis model based on ICEEMDAN and 1-D CNN is proposed.Firstly,the original signals of 10 channels are decomposed into multiple IMF components by ICEEMDAN algorithm,and then the effective IMF components are input into 1-D CNN to extract more abstract fault features.Moreover,the performance of the model in fault classification and fault location tasks is evaluated by different evaluation indexes.The comparative experimental results show that the proposed model outperforms single neural network models in both the accuracy of fault diagnosis and the robustness of the model.