Research On Automatic Recognition Technology Of Waveform In The Process Of Rock Microfracture By Continuous Active Source Scanning

A large number of microseismic events are usually generated in the process of underground rock microfracture.Microseismic monitoring technology can effectively record and analyze microseismic signals to obtain a variety of information about rock fracture.This technology has already played an active role in fracturing logging,mining,and deep-buried tunnels.However,due to the inaccuracy of the wave velocity of the microseismic signal propagating inside the rock,there are always inherent errors in the location of the microseismic event,and the complexity of the microseismic signal itself makes it more difficult to separate the wave field and pick up the first arrival of the effective wave.It also brings many challenges to the use of microseismic signals for source location.Studies have shown that the elastic wave excited by the active source has good scanning characteristics for the internal structure of the rock.Therefore,the continuously excited vibration wave can be used to scan the rock fracture deformation process to study the rock micro-fracture mechanism and the improved method of seismic source location.However,the frequency components of the microseismic signals generated by rock microfractures are complex and easy to mix with the signals of the active seismic source.Therefore,how to identify and classify the waveform signals is the first problem to be solved,which is related to the reliability of subsequent research.In view of the above problems,this paper mainly completed the research work of waveform signal acquisition,feature analysis and classification recognition of continuous active source scanning of rock micro-fracture process:First,in order to collect experimental data of continuous active source scanning rock microfracture waveforms,this paper selects large-scale sandstone as the main body of the experiment,and uses static hydraulic press to slowly split the rock.In the stage,scanning is performed by manually tapping the rock continuously,and the IMS microseismic monitoring system is used for waveform data acquisition to record the three types of waveform signals generated during the experiment in real time and completely.The three types of signals are the active source scanning signal under the intact rock,the mixed signal generated by the active source and the rock micro-fracture during the rock rupture process,and the active source scanning signal under the broken rock.Then,this paper uses time-frequency analysis,ensemble empirical mode decomposition,and polarization attribute methods to analyze the different characteristics of the three types of experimental waveform signals,including dominant frequency,different IMF components,and particle motion trajectories.The results show that the active source scanning signal under the intact rock and the active source scanning signal under the fractured rock are similar in the above-mentioned feature comparison,while the mixed signals produced by the active source and the rock micro-fracture during the rock fracture process are relatively different.Finally,this paper uses the deep learning method to complete the recognition and classification of the three types of waveform data.According to the different types of input neural network data,this paper will study from two perspectives of waveform classification based on pictures and waveform classification based on time series data.The improved Alex Net is used in the image-based waveform classification.After adjusting the parameters,the accuracy of the model applied to the new waveform data classification reaches 91.7%,which is better than the recognition result of artificial neural network;The CNN-GRU network model is used for waveform classification based on time series data.After the input sequence length is selected,the classification accuracy of new waveform data reaches 94.5%,which is higher than that of GRU.