Research On Extraction And Recognition Method Of AE Signals Based On Coal And Rock Fracture

Using acoustic emission(AE)to monitor mine dynamic disasters is a feasible plan to prevent and control mine disasters.The actual acoustic emission signals are doped with various noises,which affect the accuracy of monitoring results.The data collected by acoustic emission signals are large and difficult to monitor.For this reason,this paper studies the denoising,feature extraction and recognition of AE signals.The time-frequency localization advantage of wavelet threshold analysis is used to denoise non-stationary nonlinear AE signals.In view of the problem that the wavelet hard threshold function causes signal oscillation and the wavelet soft threshold function has a constant deviation problem,the wavelet threshold function is improved.However,the method is limited by the selection deviation of the wavelet basis function,and the signal denoising effect is not ideal for the signal with low signal to noise ratio.Thus,another denoising method,the particle filter(Particle Filter,PF),is introduced.This method is not controlled by the condition of linear system and Gauss noise,and has strong signal approximation ability.The fruit fly algorithm(FOA)can be used to solve the resampling problem of SIS,and use the neighborhood dynamic adjustment factor(D)to enhance the optimization ability of FOA and accelerate its convergence,alleviate the particle dilution in SIS and improve the denoising precision.The deep belief network(DBN)is used to extract and recognize the denoised signals.The depth confidence network trained by AE signal sample data is used as an evaluation function of FOA algorithm,so that the basic parameters of the model can be improved.It solves the shortcomings of traditional feature extraction and recognition methods which rely on expert experience and signal processing technology.It solves the shortcomings of traditional feature extraction and recognition methods which rely on expert experience and signal processing technology.Taking Baliancheng coal mine as the research background and carrying out the simulation experiment,the denoising results show that the improved wavelet threshold method and the D-FOA-PF method can reduce the noise amplitude in AE.Compared with the non optimized DBN,AE parameter-SVM,AE parameter-BP neural network method,its feature recognition results show that the separation precision of the FOA-DBN model is improved,the accuracy of feature recognition is high,and the foundation for monitoring the hazard degree of the mine dynamic disaster is laid.