Research On Acoustic Emission Signal Fault Diagnosis Of Axle Fatigue Crack Based On LMD And SVM

Due to the rapid development of China's economy,high-speed railway is developing rapidly,and rail transit vehicles have become an important part of transportation for Chinese cities.Train axle as an important part of the rail transit vehicle,it's failure occurs mainly in the form of fatigue crack.If the fatigue crack of the train axle is not found in time,it may lead to the further diffusion of the crack and even the phenomenon of the broken axis,so that the train derailment and overturning will occur.Therefore,it is of great significance to recognize the train axle fatigue crack in time for the safe and stable operation of the train.The fault diagnosis process of train axle includes three stages:obtaining state signal,extracting state signal characteristic and state pattern recognition.In this paper,the train axle state signal is obtained by simulating the train axle fatigue crack experiment and using AE detection technology of nondestructive testing.Because the sample data of axle fatigue crack is large,this paper extracts small sample data of axle fatigue crack,and adds the background noise and percussion signal of axle as recognition interference.On the basis of this,the feature extraction of signal and the pattern recognition of the state are studied in the following stages:Due to the AE signal of the train axle fatigue crack is non-stationary and the conventional signal analysis method has limitations when dealing with non-stationary signals,this paper introduces a local mean decomposition(LMD)of time-frequency analysis.Then propose a feature extraction method which is combined with LMD and AR model for AE signal of axle.The method first decomposes the train shaft AE signal with LMD and obtains the sum of several PF components whose instantaneous frequencies have physical meaning,the PF components have stationarity After processing.Then combines with the autoregressive model(AR model)that has sensitive characteristics to the change of state information,a suitable AR model is established for PF1 component,and the first 6 order autoregressive coefficients and residual are extracted as feature vectors input support vector machine(SVM)to identify the state.In order to further improve the effect of classification,this paper proposes grid search;genetic algorithm(GA)and particle swarm algorithm(PSO)to find the optimal parameters of SVM parameters in the sense of CV.After parameter optimization,therecognition rate of SVM for train axle acoustic emission signal has been significantly improved.Through analysis,the LMD-AR model and SVM method can be used for fatigue crack diagnosis of train axle.