Research On Analysis Method Of Axle Acoustic Emission Characteristic With Streaming Big Data

In daily life,in order to ensure the normal operation and work of rail vehicles,the fault detection of rail vehicles is particularly important,and various detection methods have been developed and optimized.Nowadays,the mainstream fault monitoring methods include both damage and non-damage detection.Acoustic emission detection is one of the main applications of non-destructive monitoring.First,the acoustic emission signal is endogenous.When the stress accumulates to a certain extent,it will produce acoustic emission phenomenon.Acoustic Emission(AE)is widely used because it does not use external stress to cause damage to the detected parts,which ensures the safety of the detected parts.It is also the advantage of AE for axle fault detection.Therefore,the use of AE signal for rail vehicle fault detection has become an important issue.By collecting AE signal data of railway vehicle axle,this paper mainly studies the change of axle from small cracks to axle splitting b.Firstly,the AE data collected are denoised to eliminate the noise data types such as knocking,and then the AE data are processed to identify the cracks and the damage stage of axle.In addition,an online and offline analysis method is proposed in this paper.In the first place,the online time-frequency domain characteristics of AE data are analyzed,and then,the off-line information entropy characteristic parameters of AE data are classified.Due to the huge data scale of AE signal,in order to analyze data quickly and accurately,the concept of data flow and distribution is introduced in this paper.It not only guarantees real-time and fast data processing,but also can accurately analyze the data without losing data items.Acoustic emission data can be counted through an improved sliding window to reduce the amount of parameter calculation,speed up data real-time processing,and identify and classify the faults of axle acoustic emission data.LMD(Local Mean Decomposition)algorithm and BP neural network are implemented in the framework of distributed Spark.The waveform analysis of AE data in time domain and frequency domain is also used to directly show the change of characteristic parameters in each damage state.In order to achieve real-time and fast fault analysis.In this paper,the whole experiment process is deployed in the Spark distributed framework.Spark Streaming data flow framework is used to calculate the parameters of time-frequency domain and information entropy of AE data.Based on the improved LMD method,the problem of introducing endpoint effect into the sliding window is dealt with,which further improves the robustness of the algorithm.By the way,it speeds up the efficiency of data processing and shortens the time of data analysis.