Feature Extraction And Recognition Of Rail Damage Acoustic Emission Signal Based On Optimized VMD

By the end of 2019,the total mileage of China’s high-speed railway has reached35000 kilometers and ranks first in the world.With the rapid development of high-speed railway technology,it must be considered to ensure the safe operation of trains.The operating environment of high-speed train is complex,and rails and wheels work under high load for a long time.In the case of stress concentration,even small damages are easy to spread and grow.It will lead to irreversible consequences.Therefore,the health of the rail is one of the most important requirements for the safe operation of the high-speed train.At present,a variety of railway damage detection methods have been developed.Acoustic emission(AE)technology has shown superior performance in railway damage detection due to its sensitivity and dynamics.Since the AE technology passively receives the signal generated by the break of the rail,it means that a large amount of wheel-rail noise will submerge the damage signal and make the damage signal event unavailable.This is one of the difficulties of rail damage detection based on AE Technology.Therefore,based on AE technology,this paper carries out the following research work on rail damage detection:First of all,based on the time domain features,non-linear permutation entropy feature and Mel spectrum feature,the damage signal,environmental noise signal and the mixed signal of the two are analyzed,and the features of rail damage detection which are different from wheel-rail noise are found through comparison.At the same time,because the damaged signal is submerged by noise,it is difficult to extract features.The variational mode decomposition(Variational Mode Decomposition,VMD)algorithm is introduced to decompose the AE signal,and the damaged signal is separated from the original signal.Then,when VMD algorithm is used to decompose signals,different decomposition parameters need to be adjusted for signals with different components.However,for AE signals to be decomposed,the signal components are complex and updated in real time,and manual debugging of parameters is very time-consuming and inefficient.Therefore,in this paper,for the decomposition problem caused by wrong parameters,the decomposition results that can completely decompose the useful information of signals without mode mixing and over-decomposition are used as the optimization objective of particle swarm optimization algorithm to search for the best combination of parameters and realize the adaptive decomposition of different components of AE signals.The validity of the proposed algorithm is verified by the experimental signal decomposition results and the comparison with the othertwo optimization algorithms.Finally,according to a large amount of AE signal data and no label for the damage signal,an unsupervised training model based on auto-encoder and K-means clustering algorithm is designed to identify the damage.In order to keep the time sequence information of AE signal,the auto-encoder structure is built with double layer LSTM network.In this paper,the optimized VMD algorithm is used to decompose 1000 groups of acoustic emission signals,and the decomposition results are sent to the designed model for training through feature engineering.Finally,the K-means algorithm is used to identify rail damage.The comparison with the single-feature recognition results verifies that the structure designed in this paper can improve the accuracy of injury recognition,and the recognition results are robust to noise.In this paper,the optimized variational mode decomposition algorithm is combined with the depth neural network to improve the reliability of the detection results,which has a certain significance for the actual rail damage detection work.