Acoustic Emission Behavior Characteristics Of Rock Micro-fracture Evolution Based On Deep Learning

Various large rock projects(such as dams,underground chambers,tunnels,mining,etc.)are being built in full swing in all parts of the country,but the instability of surrounding rock in the process of construction leads to the increasing problems of geological disasters,which has become one of the important factors restricting the safety,construction progress and economic cost of rock engineering.Acoustic emission technology is widely used in various rock engineering and geological disaster monitoring and early warning.Aiming at the problems of AE in studying the evolution mechanism of rock micro-fracture,the micro fracture evolution mechanism of red sandstone deformation and failure under the action of hydraulic coupling is studied by using the methods of indoor test,theoretical analysis,and numerical simulation.The main achievements are as follows:(1)In view of the characteristics of rock acoustic emission signals such as low signal-to-noise ratio,strong randomness and non-stationarity,a method of denoising rock acoustic emission(AE)signals based on ensemble empirical mode decomposition(EEMD)and single channel blind source separation(SCBSS)is proposed.In order to eliminate the influence of the speed to the AE positioning accuracy during the rock internal propagation process,an unknown wave speed AE positioning algorithm based on time difference PSO is proposed.In order to eliminate the influence of the velocity of elastic wave on the accuracy of AE location during the rock internal propagation process,an unknown velocity AE positioning algorithm based on time difference PSO is proposed.A set of "moment tensor parameter inversion and source rupture mechanism analysis of AE source" is developed based on the MATLAB computing platform,which realizes the automatic extraction,filtering,unknown wave velocity AE positioning and inversion of source parameters and analysis of source rupture mechanism of AE monitoring waveform data.(2)It was found that the decrease rate due to the AE signal faster during the propagation process in water,causing several AE receiving sensors to becomes more difficult to collect the same AE event.Therefore,the micro-fracture evolution mechanism in the process of the deformation and failure of red sandstone specimen in saturated state is not studied by the moment tensor inversion theory.In view of the problems existing in the analysis of the physical and mechanical mechanism of AE source in the deformation and failure of red sandstone specimens under saturated state by the moment tensor inversion theory,a two-dimensional depth residual convolutional neural network for AE source mechanism identification is proposed.By converting 1D AE signal into 2D digital image,the advanced and abstract AE source features in 2D digital image are extracted by using the two-dimensional depth residual convolutional neural network.And it has been successfully applied to the intelligent identification task of AE source mechanism during the microfracture evolution of red sandstone under saturated state.(3)The ModeⅠ and Ⅱ fracture experiments were applied for an investigation of the mechanical parameters and AE characteristics in the process of micro-fracture.The relationships between AE behavior,load stress,fracture toughness,damage in nonfracture region and water content with different fracture modes were systematically studied.The mathematical formula between damage amount in non-fracture damaged area under different fracture modes and water content was established.From the point of view of the main failure modes in the fracture process of ModeⅠ and Ⅱ of CCNBD specimens with different water contents,the variation characteristics of AE signals under different fracture modes were studied.Based on the GMTS criterion,the relationships among critical fracture radius,initial crack initiation angle and water content of CCNBD specimens were deduced.Through SEM imaging results and AE source spatial distribution information,the feasibility of the NBPM proposed in this paper was verified.The micro-fracture propagation mechanism and the fracture process zone during the fracture process of ModeⅠ were investigated by using the heterogeneous cementation model,and the micro-fracture evolution mechanism during the fracture process of ModeⅠ was revealed.(4)Based on the theory of statistical mechanics and damage mechanics,the creep acoustic emission model of red sandstone is established.The creep model reveals the internal relationship between the AE behavior of decelerating creep and steady creep stages and the AE behavior of accelerating creep stage.The singular value decomposition method was used to analyze the cumulative AE events and realized the quantitative identification of the accelerating creep stage.The results show that the AE signals in the deceleration creep and constant creep stage are all abrupt acoustic emission signals,while the AE signals in the accelerated creep stage are mixed signals composed of abrupt and continuous AE signals.Therefore,the quantitative recognition of accelerated creep stage is realized from the angle of acoustic emission waveform characteristics.The method is relatively small affected by the noise of the external environment.The relationship between the characteristics of AE signal waveform and the crack length in rock is discussed.The characteristics of AE signal wave in different creep stages are explained from the perspective of elastic wave motion mechanics.(5)By means of numerical simulation,AE,SEM imaging,fractal theory and ResNet50 model,the microfracture evolution mechanism of red sandstone under different water contents was revealed.The results show that with the increase of water content,the failure mode of rock specimen changes from tensile fracture to shear fracture,and the number of macroscopic cracks on the surface decreases gradually.Tensile cracks are more likely to aggregate and form macroscopic cracks,while shear cracks are relatively dispersed.By comparing and analyzing the numerical simulation results,SEM imaging results and the recognition results of the ResNet50 model,it is confirmed that the ResNet50 model can monitor and interpret the generation mechanism of AE source in the process of rock micro-fracture under different water contents.