Experimental Data Analysis Of Acoustic Emission Signal Of Train Axle Based On SDAE

In recent years,China's high-speed rail industry has achieved rapid development.On June 26,2017,China's self-developed Fuxing EMU CR400AF officially launched in Beijing and Shanghai high-speed rail,at a fast speed of 350km/h.This is also the highest commercial running speed in the world.The faster speed it runs,the greater risk we will take.The train axle,which was assumed to be the core part of the wheel set,plays an important part in the train body structure.In the process of running,it is inevitable that cracks appear in the axle.If the axle can not be monitored in time,leading to cracking,the consequences will be disastrous once the ax.Therefore,it is necessary to monitor the train axle in real time.Concerning the traditional troubleshooting,a large number of problems such as complex model,difficult extraction of the fault features.This paper will carry out on-line detection of axle fatigue cracks,using the acoustic emission technology and the popular deep learning algorithm.In addition,in the case of nonlinear and non-stationary transmitted signal,the author put forward a method for classification and recognition of axle crack acoustic emission signals based on SDAE.In-depth study of the feasibility of deep learning network in the diagnosis of axle faults,the author recognize the commonly used vibration signal time domain signal feature data as the input of deep neural network.With powerful autonomous learning function of SDAE,unsupervised training is carried out on the input training,setting axle crack,tapping and noise signals as targets and the signal can be extracted by itself.As a result,the author can predictably classify the research data.After the comparison experiments,the impact of each network parameter on network performance will be analyzed.For example,the number of network layers,the number of iterations,Batchsize,the proportion of training samples and so on.Finally,the author had the clues that the method can effectively classify and recognize the axle acoustic emission signals by many calculations and experiments.