Acoustic Emission Signal Identification Of Axle Fatigue Cracks Based On VMD Permutation Entropy-CNN

With the vigorous development of the national railroad industry,traveling by high-speed trains has become the choice of more and more people,and at the same time,the safety of highspeed trains should be given attention to.The axle is an important aspect of the train’s travel section,and as long as it works,it will be affected by the load,so cracks in the train’s axle are inevitable.In order to ensure the safety of train operation,it is extremely important to identify the signals of fatigue cracks in the axle in time.In order to correctly identify the fatigue cracks of train axles,this paper proposes a recognition method based on VMD alignment entropy and CNN.Firstly,three types of signals,namely,axle cracks,noise and knocks,are collected and saved by acoustic emission technology,and the three types of signals are decomposed separately by using VMD decomposition method,and several IMF components are obtained for each type of signal after decomposition,and then the alignment entropy value is obtained for each IMF.The three types of feature data are input to the one-dimensional convolutional neural network for classification and recognition.The experimental results show that the proposed method can be accurately and efficiently identify the axle fatigue crack signals.Through four sets of comparison experiments,the proposed 1D convolutional neural network can identify the crack signals more efficiently than other neural networks;through a set of validation experiments,the proposed classification method also applies to other axle crack acoustic emission signals.In order to meet the needs of simplifying the diagnosis steps and facilitating humancomputer interaction,this paper also developed a set of axle fatigue crack acoustic emission signal recognition system together with the members of this group using Python programming language on the basis of the above mentioned classification method,which mainly realizes the functions of user login,data acquisition and signal classification.The system can collect signals in real-time,and the collected data can be saved to the database for future use.For different usage scenarios,the system also provides a variety of classification methods for users to choose when using.Finally,the system is verified by the lead break experiment to prove that the system can complete the function of classification and identification.