Research On Rock Acoustic Emission Signal Recognition Based On Improved VMD Denoising And Multi Feature Fusion

By processing and analyzing the acoustic emission signals generated in the process of rock fracture,the structural changes in the rock during mining can be obtained,which is of great practical significance for the early warning of rock fracture instability.However,the working environment of the mine site is complex,and all kinds of noise signals and acoustic emission signals generated by rock fracture are mixed with each other,which seriously affects the accuracy of acoustic emission monitoring results.This paper focuses on the relationship between rock fracture process and multiple characteristic quantities of rock acoustic emission signals,from the aspects of time domain,frequency domain and time-frequency domain,the multi domain features of rock acoustic emission signals in four stages are extracted and fused.Finally,the classification model of extreme learning machine(MVO-ELM)optimized by multi universe optimization algorithm is established,in order to realize the automatic identification of rock acoustic emission signal,so as to achieve the purpose of monitoring rock fracture state.All the research results of this paper are as follows:(1)Aiming at the problem that the IMF component number k needs to be set in advance when the VMD algorithm decomposes the signal,an improved VMD algorithm is proposed.The simulation results show that the improved VMD algorithm can obtain the optimal K value;meanwhile,the permutation entropy of each IMF component after improved VMD decomposition is calculated,and then the optimal IMF component is selected and the signal is reconstructed to complete the signal denoising.(2)When the rock is in different fracture stages,calculate the time-domain characteristic parameters of acoustic emission signal,including average amplitude,margin index,waveform index and pulse index,it is found that the eigenvalue of acoustic emission signal in the first stage is the largest,the eigenvalue of acoustic emission signal in the fourth stage is the smallest,and there are obvious differences between acoustic emission signals in the first stage and the fourth stage.The spectrum distribution of acoustic emission signal is obtained based on Fourier transform,and then the spectrum centroid of acoustic emission signal in each stage is calculated,the result is that the spectrum centroid of the first stage AE signal is the smallest,and the distribution range is 135 ～ 140;in the second stage and the third stage,the spectral centroids of a small number of acoustic emission signals are similar,but it does not affect the overall characteristic difference of the signals.Improve VMD＿MFE algorithm extracts the difference characteristics of AE signals in time-frequency domain.It is found that when the embedding dimension m = 3,the fuzzy function gradient n = 2,the scale factor s = 10 and the similarity tolerance r = 0.15SD(SD is the standard deviation of the original data),the MFE values of the reconstructed four types of AE signals are obviously different.(3)An classification model based on MVO-ELM is constructed.The multi-domain fusion feature vector of rock acoustic emission signal is used as the input of the model,and the signal recognition accuracy of different feature vectors and different classification methods are compared.The results show that the recognition accuracy is the highest when using multi-domain fusion features,which can reach 95%;When sigmoid function is used as the activation function of ELM,the recognition accuracy of acoustic emission signal is the highest;The recognition accuracy of MVO-ELM model is significantly higher than ELM,Bayes and SVM.