Experimental Data Analysis Of Acoustic Emission Signal Of Axle Fatigue Crack Based On CEEMD And GRNN-DBN-LSTM Network

Under the background that nearly 30000 km high-speed railway network has been built,the national railway industry continues to develop at a high speed,and a large number of high-speed railway vehicles have been put into operation.The faster the train runs,the greater the risk of safe operation.Therefore,it is very necessary to realize the real-time fault classification of the important components of the train,and then predict its development situation.As the main load-bearing part of railway vehicle,the state of axle directly affects the running state of bogie.Generally,the fatigue crack of the axle can only be detected when it develops to a certain extent.At this time,the available maintenance time is not much,which will have a serious impact on the safe operation of the train.In order to realize the accurate and quick classification and waveform prediction of the fatigue crack of the train axle,this paper puts forward the experimental data identification and prediction model of the acoustic emission signal of the fatigue crack of the axle based on the CEEMD-GRNN-DBN-LSTM.The difficulty of the research is the accurate recognition and waveform prediction of the large volume sample of the acoustic emission signal of the axle.After comparing with three common data preprocessing methods,the complementary set empirical mode decomposition(CEEMD)is used to preprocess the three acoustic emission signal data sets to reflect the feature information,and improved deep belief network GRNN-DBN is introduced to identify.And compared with other networks.After that,the long and short-term memory network(LSTM)is constructed to predict partial data of the acoustic emission signal of the fatigue crack of the axle,and the results are compared with those of other networks.At last,the performance of the model is verified by using strange data sets.Through the identification and waveform prediction of the acoustic emission signal of the fatigue crack of the train axle,it is concluded that in the case of preprocessing data by CEEMD,the GRNN-DBN network established in this paper has higher recognition accuracy and shorter operation time compared with the DBN network of other structures and types;the LSTM network established in this paper has higher recognition accuracy and shorter operation time compared with the LSTM network of other structures and three kinds of artificial neural networks.