Research On Aluminum Alloy Materials And Bearing Defects Of High-speed Train Based On Acoustic Emission Detection

Acoustic Emission can be used to capture the dynamic process of crack generation and propagation in real time compared with other conventional non-destructive testing methods,and has more localized wave form;the signal frequency range is wider and the information contained is richer,which means it can detect the structural failure early.Also,using multisensor positioning analysis can locate the orientation of the crack(sound source).This provides acoustic emission technology a unique advantage during the high-speed train "state monitoring".However,the weakness and sensitivity of acoustic emission make it difficult to extract and analyze them under the influence of noise.There are some common problems of acoustic emission detection involved in aluminum alloy materials and bearing defects of high-speed train,mainly include waveform extraction,signal de-noising,characterization method and fault classification.Among them,the crack of large structural parts such as car body etc,also involves acoustic emission positioning problem.The envelope analysis technique,which is represented by empirical mode decomposition(EMD)the signal from the Fourier analysis and wavelet analysis,,is the first real method which does not depend on the "basis function" to decompose the signals.The EMD method has a significant advantage for the decomposition of all kinds of signals,especially unsteady signals decomposition because of the adaptive decomposition of the data itself.Based on the analysis of the performance of EMD、local mean decomposition(LMD)and CEEMDAN decomposition algorithm,an improved Modified Complementary Ensemble Empirical Mode Decomposition with Adaptive Noise(MCEEMDAN)method is proposed.In the EEMD decomposition,it is combined with the frequency of the signal itself and the coupled linear sine curve adaptive structure to add noise,while limiting the number of screening on the IMF(intrinsic mode function)can reduce the modal aliasing and residual noise interference.Aiming at the long-term acoustic emission monitoring of the actual operation environment of high-speed trains,try to extract the characterize and statistics the acoustic emission signals at a small computational cost.We propose an EMD noise reduction method using the 3σ criterion segmentation threshold to avoid the small amplitude of the acoustic emission signal masked from the global threshold.Meanwhile,according to the intrusion mode function,the recognition and extraction of the AE signal are compared with the autocorrelation and KL divergence.At the same time,a statistical and characterization method of acoustic emission signals based on the kurtosis of the IMF component envelope is proposed.The results of the method have been analyzed under the influence of different intensity noise and the effectiveness is proved.Based on the decomposition of MCEEMDAN,which under the influence of noise,the difference of the permutation entropy in noise signals and the AE signals in the IMF modal components have been studied.A method of judging whether or not the acoustic emission signal is included in the high frequency IMF component distribution entropy distribution is proposed.The robustness of the method is verified by comparison.Considering the real moniting scenario for the vehicle body material,it contains multiple types of acoustic emission signals,a strong SVM classifier construction method which can identify a variety of acoustic emission sources is proposed.The measured signal verifies the performance of the classifier.During the position of the acoustic emission source,the IMF component with the typical acoustic emission mode is used to determine the time difference of the sound emission to reduce the influence of the multi-source signal and the interference.The range of wide frequency has better adaptability.In the IMF cross-correlation latency estimation,a second order singular value difference spectrum method is proposed to obtain the noise reduction signal,and the correlation peak is more definite and the accuracy of the time difference estimation is improved.Time reversal imaging is carried out by using the time difference and velocity of the IMF.There is no dependence on the positioning sample and the wave transfer model,.Therefore,it has some advantages in dealing with multi-source,multi-mode signal and positioning accuracy.High-speed train wheels on the bearing failure directly affect operating safety,compared to the body crack detection,has more attention to early fault detection and early warning.Due to complexity classification state of bearing fault signal.It is difficult to set the parameters and cope with the multi-classification problem by artificial neural network and SVM.Therefore,the group sparse representation-based classification(GSRC)is used to realize the multi-fault state Identification.In the atom obtain,MCEEMDAN combined with variational mode decomposition(VMD)is designed to adaptively obtain the initial atom of the dictionary and is optimized by K-SVD,which preserves the intrinsic characteristics of the training sample.The least absolute shrinkage and selection operator(LASSO)constraint optimization algorithm with One-order method is used to improve the convergence and the calculation speed in the sparse matching of atoms.At the same time,an interval translation sparse coding(ITSC)is proposed to relax the requirements of the training samples and the samples to be classified in the data interception.This shows the advantages of compact and sparseness in dictionary.Considering from dictionary structure,a kind of Complex failure redundant dictionary with index is designed.Due to the small volume advantage of index dictionary,the classification of the fault class can be narrowed by pre-matching based on the multi-scale arrangement entropy of the training sample signal,which accelerates the convergence speed of classification.The defect of the bearing is carried out on the test bench,which further validates the advantages of the method.In the frequency domain analysis,the effectiveness of this method on bearing fault extraction is calculated and verified by the Hilbert time spectrum.Finally,the Phenomenon of "frequency modulation" during the faulted bearing acoustic emission signal in the power spectrum domain is observed in the experiment ",using VMD method to decompose its power spectrum signal to extract the hidden low-frequency modulation waveform,can quickly and easily determine whether the bearing defects.