Research On AE Signal Of Axle Fatigue Crack Based On EEMD Multi Information Entropy And DBN

Since the reform and opening up,under the guidance of national policies,rapid progress has been achieved in various fields.Rail transit construction has made tremendous progress in terms of speed and comfort.At the same time,high-speed operation safety issues have become more and more serious with the development of rail transit.Attract attention;the axle,which is an important component in the running department,bears the important responsibility for the smooth operation of the vehicle.Therefore,it is of great significance to monitor the state of the train axles and find faults in time to ensure the safety of the train at high speed,the personal safety of the passengers and the property of the country.In view of the above problems and research significance.In this paper,the method of classification and identification of axle fatigue crack based on EEMD and multi information entropy combined with DBN and the method of axle life prediction based on EEMD and multi information entropy combined with DBN-NN are proposed.In condition monitoring and fault diagnosis,acoustic emission technology is better than other traditional methods.Therefore,this paper uses acoustic emission detection technology to collect three types of acoustic emission signals of axle fatigue crack,noise and knocking.In the work of feature extraction: firstly,the three kinds of data signals are processed by grouping and EEMD decomposition,and the decomposition components are optimized by kurtosis criterion.Then,the multi information entropy value of the optimal component is obtained,and it is used as the feature data input network to reduce the data input.In the establishment of the model: in the classification and recognition network,through the construction of DBN structure model and the selection of parameters such as structure layer number,node number,activation function,learning rate,number of iterations,batch size and distribution proportion of test set and training set,the optimized training classification and recognition network model is obtained.In the process of life prediction: the DBN-NN structure model is used in the prediction.Firstly,the data is trained unsupervised by DBN training model,and then the establishment of life prediction model and network parameter adjustment are finally determined by transferring to NN structure.According to the above work content and the proposed classification identification and life prediction methods,the comparative analysis of relevant schemes and the verification of the proposed method by replacing experimental data are carried out.