Research On Acquisition And Recognition Of Microseismic Signal And Source Location Based On Optical Fiber Sensing

Microseismic monitoring is an effective means for monitoring and early warning of mine dynamic disasters.Traditional electronic microseismic monitoring systems,which need supplying power to sensors,are vulnerable to electromagnetic interference in harsh mine environment and the transmission distance is limited.It is urgent to develop a microseismic monitoring system which can adapt to the harsh underground environment and can transmit microseismic signals over a long distance.The microseismic signals of mines include the signals caused by rock burst and the signals produced by blasting.Most of them are processed and analyzed as effective signals.That will lead to misleading and misjudgement for researchers and managers.Conventional signal recognition methods need to extract the characteristic parameters of microseismic signals manually.Some effective features will be lost without training and learning from the signals themselves.Therefore,it is necessary to study the technology to distinguish the microseismic signal from rock mass rupture and the blasting signal which lays a foundation for the study of parameters such as inversion of stratum structure,source location and mechanism and so on.Microseismic source location is the essential factor in microseismic monitoring technology,and the location precision will greatly influence the application.Accurate location of microseismic source in coal mine is of great significance for monitoring and early warning of dynamic disasters.Microseismic source information is generally extracted through inversion of the data acquired by the underground sensors.The installation of the sensors is limited around underground roadway.The source location precision will be greatly decreased in case of unreasonable arrangement of sensors along the roadway and approximately in a plane.Optical fiber sensing technology has natural advantages such as no power supply,anti-electromagnetic interference,and long transmission distance and so on.It is very competent for mine microseismic monitoring under harsh conditions.In this paper,S-Transform with good two-dimensional time-frequency characteristics is introduced into the original microseismic digital signal conversion.The microseismic signal is displayed in the form of images,which contain both time-domain and frequency-domain features.Then the convolutional neural network(CNN)algorithm designed for image recognition is utilized for mine microseismic signals recognition.So as to solve the problem of lower recognition accuracy rate that result from the loss of features caused by manual extraction the feature parameters.The results show that Tikhonov regularization method based on L-curve criterion can improve the accuracy rate of ill-posed matrix solution.Consequently,in this paper we study microseismic monitoring system based on optical fiber sensing,microseismic signal recognition based on depth learning,Tikhonov regularization and high precision source location technology.By the means of theoretical analysis,formula derivation,numerical simulation,experimental testing and field monitoring as the study methods/tools.The self-developed programs realize microseismic signal acquisition,recognition and source location.The main research contents and research results are as follows:1.A criterion for identifying effective microseismic signals based on energy eigenvalue and sliding time-window width is established.The principle of Fiber Bragg Grating(FBG)sensor with cantilever structure is analysed.The influence of vibration on FBG reflectance spectrum is detected by Distributed Feedback Laser(DFB),and the aquisition of vibration signals is realized.According to the great difference of energy before and after the first arrival point and end point of seismic wave,the sliding time-window energy ratio method is used to pick the first arrival and end points.The quantitative relationship among sliding time-window width(TWL),sampling frequency and main frequency of signal is proposed for the first time.Based on the energy eigenvalue and sliding window width,the criterion of effective microseismic signals is designed.The coupling parameters between sensor and rock stratum are analyzed to realize the high efficiency coupling.The software and hardware system of optical fiber microseismic monitoring has been developed and tested outdoors.The performance of FBG sensor and traditional piezoelectric vibration sensor has been compared and analyzed in shaking table and field test.The system is installed in underground coal mine for microseismic monitoring.Signal deviation is used initiatively to establish effective signal discrimination criterion to achieve effective microseismic events screening and storage.2.The conversion technology of monitoring data of different systems based on Discrete Fourier Transform(DFT)is deduced,and the comparable and universal data of multiple systems are realized.The feasibility of the algorithm is verified by signal simulation.3.A deep learning algorithm based on CNN and S-Transform is proposed and trained which is suitable for the characteristics of microseismic signals.In this paper,CNN and S-Transform are combined innovatively to train,test and predict the original time-domain signal and the time-frequency signal by S-Transform.The training results demonstrate that the accuracy rate of the prediction is approximately 50%when the time-domain signals are utilized as training samples,therefore,it cannot be popularized.After S-Transform,data augmentation,batch standardization function and over-fitting function are added to the original time-domain signal,the prediction results are still unfavorable when the time-frequency images are with three color channels.Nevertheless,when the image with three color channels is changed to gray-scale image with single color channel,the prediction accuracy rate is increased to 84.6%.When the size of the training samples is reduced to 45×35 pixel(px),the number of neuron connections is reduced and the prediction accuracy rate is further increased to 93.75%accordingly.It is feasible to classify blasting signal and microseismic signal based on S-Transform gray-scale image and the CNN model architecture and training parameters are designed by synthesizing LeNet-5 and AlxNet.4.The arrival time difference formula based on signal cross-correlation is deduced,and the velocity of first arrival wave is corrected.The microseismic location characteristics in mine are analyzed according to the characteristics of the mine 's microseismic wave field.The influences of sensor distribution,first arrival time picking and the velocity model on microseismic source location are analyzed.On account of the complex environment in the underground mine,the microseismic waves generated by fracture of the deep strata are usually disturbed by clutter.At the same time,the first arrival time of signals cannot be accurately picked up due to the influence of system noise.Conventional manual picking or automatic arrival time picking based on long and short windows will cause picking errors to a certain extent.The triangular geometry relation is utilized to establish the distinction method of the first arrival wave type.An innovative time difference algorithm based on signal cross-correlation is proposed,and the corresponding velocity is given.When the original vibration signal is contaminated by noise,the calculation precision of signal delay time can be greatly improved by cross-correlation method compared with manual picking and sliding energy ratio method.5.The regularization processing method for inversion of ill-posed problems of microseismic source location.As for the ill-posed problems caused by the inversion for source location with sensor-received information,high precision location algorithm of micro-seismic source based on monitoring points arrangement optimization is presented in this paper.Firstly,the ill-posed problems are judged by calculating the coefficient matrix condition numbers.The ill-posed matrix is then pretreated by centralization and row balance jointly.The regularization parameters of pretreated matrix A and b are calculated by the L-curve method.The source coordinates regularization solution is obtained by Tikhonov regularization algorithm.The results show that the matrix magnitude is decreased effectively with centralization method,and ill-posed condition numbers are reduced with row balance pretreatment.The minimum source coordinate error of Tikhonov regularization solution after pretreatment is 3.09 m,and is greatly decreased compared to the error by Gaussian elimination solution before pretreatment.Comprehensive analysis of the above results it can be seen that optical fiber sensing microseismic system is suitable for the monitoring requirements of underground harsh environment.Underground FBG sensors are no power supply and the acquisition signals are not disturbed by electromagnetic interference signals.And the distance between underground monitoring points and substation can reach up to 10 km.The deep learning algorithm based on CNN and S-Transform does not require to extract the characteristic parameters manually,which realizes the recognition of microseismic signal of rock mass fracture and blasting signal.The Tikhonov regularization method can be utilized to source location after pretreatment for the recognized microseismic signal of rock mass fracture.The high precision of ill-posed problems caused by the arrangement of monitoring points can be achieved by means of the above optimization processing.