Experimental Study On Acoustic Emission Location Of Rock Failure Based On Data Driven

Acoustic emission(AE)is a kind of transient elastic wave produced by the rapid release of energy from a local source in a material.In the process of mining,a certain number of transducers are arranged to receive the acoustic emission signal of rock mass,and then the source point of acoustic emission can be located through real-time monitoring and calculation.However,the accuracy of this detection method is affected by the physical properties of the rock mass itself,the load pressure and action mode of the surrounding environment and the geographical environment,so the positioning results often have a large deviation from the actual acoustic emission source coordinates.At the same time,from the point of location method,the traditional method of using mathematical calculation to locate the acoustic emission source can not consider the complexity of the actual situation of the project,and a single calculation model can not be applied to the changeable engineering environment.To sum up,the accuracy of acoustic emission source location is generally poor at present.In order to solve this problem,this paper uses the laboratory to carry out the lead breaking test to explore the method of acoustic emission source location under the data-driven mode.In order to reduce the demand of model data,this paper studies the differences of acoustic emission signal waveform,spectrum characteristics and energy characteristics caused by different acoustic emission signal propagation modes.According to the difference as a criterion,we use python programming to write the criterion code,preprocess and judge the received acoustic data,determine its plane,reduce the target dimension of the machine learning model from three-dimensional to two-dimensional positioning,reduce the data demand and improve the positioning accuracy.The main research work and achievements of this paper are as follows:1)The acoustic emission signals of sandstone and granite are collected by PCI-2acoustic emission tester.The differences of acoustic emission signals of sandstone and granite in time domain and frequency domain are explored through the study of waveform and spectrum after short-time Fourier transform,The results show that the spectrum differences of sandstone and granite are concentrated in the opposite xoz plane,and the sandstone signal is in the form of wave signal,while the granite signal is in the form of sudden signal.2)The acoustic emission signals of sandstone and granite are decomposed by wavelet transform and Fourier transform,and the frequency domain information of each decomposition level is derived.This paper analyzes the differences of different AE signals at different levels,discusses the method of distinguishing different surfaces by using AE signal data,and analyzes the influence of different signal transmission modes on their properties.The results show that: with the change of signal source from front to side and then to the opposite,the amplitude values of granite and sandstone gradually decrease,and the amplitude of granite decreases more than that of sandstone.3)By using the method of energy spectrum analysis,the difference of energy spectrum coefficient distribution of acoustic emission signal with different transducer arrangement points is studied,and the relationship between transducer distance change and signal energy spectrum coefficient distribution is analyzed.Based on this,the discrimination method of acoustic emission signal difference between two sides of transducer arrangement point is discussed.The results show that with the increase of the distance between the transducer and the side,the energy level distribution of granite will change from D4 and D5 scattered distribution to D5 concentrated distribution;The energy and distribution of sandstone will change from D4 to D4 and D5.4)The vgg-16 model is optimized to adapt to the regression goal of machine learning in this paper,and a new regression full connection and model combination layer is added to carry out the regression goal of machine learning in this paper,The relationship between model regression error and data samples is discussed,and the visual image of regression prediction is realized.The results show that the accuracy of regression positioning is better in the front and side,but worse in the opposite.The quality of data set has the greatest impact on the accuracy.