A rock shear failure whole process prediction method based on acoustic emission monitoring

Abstract

The present application relates to a kind of rock shear failure whole process prediction method based on acoustic emission monitoring, first, rock sample is collected, and is processed into multiple rock test pieces, acoustic emission mold is made, second, acoustic emission test platform is installed, and multiple rock shear failure whole process acoustic emission test is carried out using it, the data of multiple acoustic emission tests are recorded, and the calculation parameters of rock shear failure whole process prediction model are selected and calculated based on experimental data, finally, rock shear failure whole process prediction model is constructed based on XGBoost algorithm, and the calculation parameters of rock shear failure whole process prediction method are input into model, and the rock shear failure prediction result is output.The method of the present application can monitor rock mass stability in real time, continuously, classify and predict different stages in the process of rock direct shear, which helps to improve the safety and reliability of rock mass engineering such as tunnel engineering.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of mine exploitation, and particularly discloses a rock shear failure whole-process prediction method based on acoustic emission monitoring. BACKGROUND

[0002] In the process of tunnel construction in rock mass engineering such as underground tunnel engineering and slope engineering, adverse geological conditions such as fault fracture zones are often encountered. In order to ensure that the tunnel can be excavated safely and smoothly, protect the normal development of the transportation industry, and protect the safety of workers' lives and property, it is crucial to accurately predict the stability of the tunnel fault. The traditional rock stability monitoring methods mainly include field observation, displacement monitoring, stress monitoring, etc. The field observation method largely depends on manual experience. Construction personnel observe some characteristics of the rock mass surface, such as the appearance of cracks and the deformation of rocks, to judge the stability of the rock mass. However, this method has obvious limitations because the structure and changes inside the rock mass cannot be directly observed, making it difficult to accurately predict potential damage inside the rock mass. Displacement monitoring and stress monitoring usually require the installation of acoustic emission probes on the surface or inside the rock. Displacement monitoring mainly evaluates the stability of the rock by measuring the displacement changes of the rock in different directions, and stress monitoring detects the stress distribution inside the rock. However, these methods can only obtain local information and cannot fully understand the overall condition of the rock mass. Moreover, for the initiation and expansion of small cracks inside the rock mass, these monitoring methods are not ideal. In addition, traditional monitoring methods are usually intermittent and cannot monitor the stability of the rock mass in real time and continuously. In actual engineering, the stability of the rock mass is a dynamic process, and intermittent monitoring may miss early signs of rock shear failure, making it difficult to capture key information in a timely manner. Under the complex geological conditions of tunnel faults, the limitations of traditional monitoring methods are more prominent, and they cannot meet the requirements of engineering practice for the accuracy and timeliness of rock stability prediction.

[0003] Therefore, the application applies acoustic emission technology to rock mass engineering, deeply processes and analyzes acoustic emission parameter data, and uses the XGBoost algorithm model in machine learning to classify and predict the collected acoustic emission parameter data and the rock direct shear test stage, ultimately proposes a rock shear failure whole-process prediction method based on acoustic emission monitoring, which meets the requirements of engineering practice for the accuracy and timeliness of rock stability prediction. SUMMARY

[0004] The application proposes a rock shear failure whole-process prediction method based on acoustic emission monitoring to solve the problem that existing rock mass monitoring methods cannot monitor the stability of the rock mass in real time and continuously, miss early signs of rock shear failure, and are difficult to capture key information in a timely manner.

[0005] The application provides a rock shear failure whole process prediction method based on acoustic emission monitoring, characterized by comprising the following steps:

[0006] S1. Collect rock samples and process them into multiple rock test pieces, manufacture an acoustic emission mold according to the size of the rock test pieces, and open square holes on the acoustic emission mold;

[0007] S2. Install an acoustic emission test platform, which comprises a mechanical loading device, an acoustic emission monitoring system and the acoustic emission mold manufactured in step S1;

[0008] S3. Perform multiple acoustic emission tests of the rock shear failure whole process by using the acoustic emission test platform installed in step S2, place one rock test piece processed in step S1 into the acoustic emission mold, arrange acoustic emission probes in the acoustic emission mold, set parameters of the acoustic emission monitoring system, and record data of multiple acoustic emission tests;

[0009] S4. Select three acoustic emission parameters with low correlation as calculation parameters of a rock shear failure whole process prediction model, and calculate the calculation parameters of the rock shear failure whole process prediction model based on the acoustic emission test data obtained in step S3;

[0010] S5. Determine input features and prediction target variables of the XGBoost algorithm, select data from the acoustic emission test data obtained in step S3, calculate the calculation parameters of the rock shear failure whole process prediction model selected in step S4, construct a data set for training the XGBoost algorithm, and train the XGBoost algorithm;

[0011] S6. Perform initialization setting on the XGBoost algorithm trained in step S5, determine basic operation modes and operation parameters of the algorithm, and construct the rock shear failure whole process prediction model;

[0012] S7. Input the calculation parameters of the rock shear failure whole process prediction method obtained in step S4 into the rock shear failure whole process prediction model constructed in step S6, and output a rock shear failure prediction result.

[0013] According to the rock shear failure whole process prediction method based on acoustic emission monitoring, the rock test piece in step S1 is a cubic test piece, the surface of the rock test piece is smooth, and the error of the end non-parallelism and non-perpendicularity is ±0.02 mm.

[0014] The acoustic emission mold further comprises a wedge-shaped plug for fixing the acoustic emission probes.

[0015] According to the rock shear failure whole process prediction method based on acoustic emission monitoring, the mechanical loading device is arranged outside the acoustic emission mold in step S2, and is used for applying axial pressure and shear stress to the rock sample.

[0016] The acoustic emission monitoring system is arranged outside the acoustic emission mold, and comprises an acoustic emission probe, a preamplifier, an acoustic emission acquisition card and acoustic emission processing software running on an electronic device, and is used for collecting acoustic emission transient waveforms of the rock sample in real time.

[0017] The acoustic emission mold is arranged between the rock sample and the mechanical loading device, and is used for bearing the rock sample and fixing the acoustic emission probe.

[0018] The acoustic emission probe has a mounting interval of 10-20 mm.

[0019] According to the rock shear failure whole process prediction method based on acoustic emission monitoring, the step S3 comprises:

[0020] Step 301. The rock sample processed in step S1 is placed in the acoustic emission mold, the acoustic emission probe is contacted with the rock sample through the square hole of the acoustic emission mold, and the wedge-shaped plug is inserted into the square hole to fix the acoustic emission probe.

[0021] Step 302. The acoustic emission monitoring system parameters are set, and the rock sample is preloaded mechanically before formal mechanical loading of the mechanical loading device. After the mechanical preloading is completed, the vertical force data, the vertical displacement data, the horizontal force data and the horizontal displacement data are zeroed.

[0022] Step 303. The normal force applied by the mechanical loading device is loaded to an initial normal stress value, and then the mechanical loading device is used to perform mechanical loading at a shear rate of 0.2 mm / min. The mechanical loading process is kept synchronous with the data monitoring of the acoustic emission monitoring system during the mechanical loading process. The initial normal stress value is set according to the particle size and normal stress condition of the rock sample.

[0023] Step 304. When the mechanical loading of step 303 reaches the peak value, that is, the acoustic emission test rock sample failure stage is ended, the mechanical loading of the mechanical loading device and the acoustic emission test data monitoring of the acoustic emission monitoring system are stopped at the same time. The acoustic emission test data obtained by monitoring is collected and saved.

[0024] Step 305. The acoustic emission probe and the damaged rock sample are taken out, and the residual in the acoustic emission mold is cleaned.

[0025] According to the rock shear failure whole process prediction method based on acoustic emission monitoring, the acoustic emission probes are uniformly arranged at positions 5 cm away from upper and lower surfaces of the rock sample in step S301.

[0026] The contact part between the rock sample and the acoustic emission probe is uniformly smeared with vaseline with a thickness of 0.1-0.2 mm.

[0027] According to the rock shear failure whole process prediction method based on acoustic emission monitoring, the acoustic emission monitoring system parameters in step S302 include a channel number, a threshold value, an amplification gain, and a sampling rate, the channel number is set to 8, the threshold value is set to 45 dB, the amplification gain is set to 40 dB, and the sampling rate is set to 1 MSPS.

[0028] The preloading force of the mechanical preloading is 10%-20% of the maximum expected bearing force of the rock sample.

[0029] According to the rock shear failure whole process prediction method based on acoustic emission monitoring, the calculation parameters of the rock shear failure whole process prediction model in step S4 include a logarithm of acoustic emission cumulative energy, a b value, and a main frequency A.

[0030] The acoustic emission cumulative energy is shown in formula (1).

[0031] (1)

[0032] In the formula, E (T) represents acoustic emission cumulative energy from 0 time to T time, E (t) represents energy carried by an acoustic emission signal at t time.

[0033] The logarithm of the acoustic emission cumulative energy is .

[0034] The b value is shown in formula (2).

[0035] (2)

[0036] In the formula, B represents the b value, m represents a total number of magnitude grades, M j j represents a magnitude median of the jth grade, and N j j represents a number of acoustic emission events of the jth grade. As shown in formula (3).

[0037] (3)

[0038] In the formula, a and b are constants, M represents a magnitude, and N represents a total number of earthquakes with a magnitude greater than M.​

[0039] (4)

[0040] in the formula, A dB represents the amplitude of acoustic emission;

[0041] The main frequency A is in KHz, which is extracted from the acoustic emission signal in the acoustic emission test data collected in step S3 by using fast Fourier transform.

[0042] According to the rock shear failure whole process prediction method based on acoustic emission monitoring according to some embodiments of the present application, the step S5 comprises:

[0043] Step 501. Select data from the acoustic emission test data collected in step S3, calculate the logarithm of the acoustic emission cumulative energy, the b value and the main frequency A, and select the rock shear failure stage as the prediction target variable;

[0044] Step 502. Divide the rock shear failure stage into the compaction stage, the elastic stage, the inelastic stage and the post-peak failure stage, encode the compaction stage as 1, the elastic stage as 2, the inelastic stage as 3, and the post-peak failure as 4;

[0045] Step 503. Encode the logarithm of the acoustic emission cumulative energy, the b value and the main frequency A of each rock sample according to the corresponding rock shear failure stage to form labeled data and construct a data set for training the XGBoost algorithm;

[0046] Step 504. Divide the stress stage of the rock sample into the elastic stage, the yield stage, the strengthening stage and the failure stage, and take the elastic stage, the yield stage, the strengthening stage and the failure stage as the input features of the XGBoost algorithm;

[0047] Step 505. Take the compaction stage, the elastic stage, the inelastic stage or the post-peak failure stage of the rock sample as the target variable of the XGBoost algorithm;

[0048] Step 506. Train the XGBoost algorithm.

[0049] According to the rock shear failure whole process prediction method based on acoustic emission monitoring according to some embodiments of the present application, the step S6 comprises:

[0050] Step 601. Determine the basic operation mode, set the algorithm to run data in tree structure gbtree, construct an algorithm similar to tree structure to analyze data step by step, and set general parameters;

[0051] Step 602. Setting the XGBoost algorithm boost parameters, wherein,

[0052] The maximum number of iterations n_estimator is set to 100,

[0053] The iteration step learning_rate is set to 0.7,

[0054] The step size of each iteration is set to 0.7,

[0055] The maximum depth of the tree max_depth is set to 1.5,

[0056] The proportion of random sampling of each tree subsample is set to 1,

[0057] The proportion of random sampling column number of each tree colsample_bytree is set to 1,

[0058] The proportion of column sampling of each tree per node split colsample_bylevel is set to 1,

[0059] The weight L1 regularization term reg_alpha is set to 0,

[0060] The weight L2 regularization term reg_lambda is set to 0,

[0061] The minimum leaf node sample weight is set to 1,

[0062] The min_child_weight is set to 1;

[0063] Step 603. Setting the XGBoost algorithm learning target parameters, wherein,

[0064] The objective is used to specify the logical relationship between the input features and the rock shear failure stage, the input features including the logarithm of the acoustic emission cumulative energy, the b value and the main frequency A three input features, the XGBoost algorithm selects binary: logistic binary logistic regression, and the output is the probability of the rock specimen being in the unbroken, the compaction stage, the elastic stage, the inelastic stage and the post-peak damage stage,

[0065] The eval_metric is the quantitative evaluation of the effective data, the XGBoost algorithm takes "auc" which is the area under the ROC curve, and is set to evaluate the performance of the model by calculating the area under the receiver operating characteristic curve ROC curve,

[0066] The seed is a random number seed, which is set to 1.

[0067] According to the rock shear failure whole process prediction method based on acoustic emission monitoring provided by the embodiment 1 of the present application, the step S601 of setting general parameters comprises: setting the silent mode as 0, and outputting algorithm running information; and setting the multi-thread control parameter nthread as the maximum thread number of a computer.

[0068] The technical solution provided by the present application deeply processes and analyzes acoustic emission parameter data, studies the relationship between acoustic emission parameter data and different stages of rock direct shear, and adopts a comprehensive analysis method of three acoustic emission parameters with low correlation. The three parameters reflect the changes in the rock from different angles. Through comprehensive consideration of the energy parameter, the b value and the main frequency, the limitations of single parameter analysis can be avoided, and the change information in the rock can be captured more comprehensively and accurately. This comprehensive analysis method can improve the accuracy and reliability of rock shear failure prediction, and make the prediction results more consistent with the actual situation. Whether in the laboratory rock direct shear test or in the actual engineering application scenario, it can provide more powerful support for the evaluation and prediction of rock stability. In addition, the technical solution provided by the present application establishes a rock direct shear stage classification prediction model based on acoustic emission data using the XGBoost algorithm in machine learning. In the model establishment process, a large amount of rock direct shear test data is analyzed and processed, appropriate feature parameters are selected, and the rock shear failure stage is reasonably divided, so that the related information of the internal structure change of the rock can be collected in real time and stably. By continuously optimizing the parameter setting of the algorithm, the accuracy of the model reaches 77%. This high accuracy indicates that the model has good performance and can effectively classify and predict different stages in the rock direct shear process, meeting the requirements of engineering practice for the accuracy and timeliness of rock stability prediction. BRIEF DESCRIPTION OF DRAWINGS

[0069] Figure 1 A flowchart of the rock shear failure whole process prediction method based on acoustic emission monitoring provided by the embodiment 1 of the present application is provided.

[0070] Figure 2 A schematic diagram of the acoustic emission mold and wedge-shaped plug model provided by the embodiment 3 of the present application is provided.

[0071] Figure 3 A flowchart of the XGBoost algorithm provided by the embodiment 3 of the present application is provided.

[0072] Figure 4 A comparison diagram of the number of real values and correct prediction values after prediction provided by the embodiment 3 of the present application is provided.

[0073] Figure 5 A prediction effect diagram of the CSY-3 test piece provided by the embodiment 3 of the present application is provided.

[0074] Figure 6 This is a prediction effect diagram of the ZSY-4 specimen provided in Embodiment 3 of the present invention;

[0075] Figure 7 The image shows the predicted effect of the XSY-3 specimen provided in Embodiment 3 of the present invention. Detailed Implementation

[0076] The embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and examples. The following examples are for illustrative purposes only and should not be construed as limiting the scope of the invention.

[0077] Example 1: This example provides a method for predicting the entire process of rock shear failure based on acoustic emission monitoring, such as... Figure 1 As shown, it includes the following steps:

[0078] S1. Collect rock samples and process them into multiple rock specimens. Make an acoustic emission mold according to the size of the rock specimens and open a square hole in the acoustic emission mold.

[0079] Preferably, in this embodiment, based on the engineering requirements, yellow sandstone rocks from Chuxiong, Yunnan Province, are selected as samples. The rocks in this area are primarily yellow sandstone, but also include samples with varying degrees of weathering to better simulate the diversity of rocks in actual engineering projects. The rock specimens are 100*100*100 cubic specimens with good integrity, smooth surfaces, and no obvious defects. The non-parallelism and non-perpendicularity errors at the ends are ±0.02mm. The size and position of the square holes are determined according to the size of the acoustic emission probes and the monitoring requirements. The spacing between the acoustic emission probes is 10-20mm. The acoustic emission mold also includes wedge-shaped plugs for fixing the acoustic emission probes. The size and shape of the wedge-shaped plugs match the square holes and the acoustic emission probes to ensure stable fixation of the probes during the experiment, achieving the goal of acoustic emission monitoring during the experiment.

[0080] S2. Install the acoustic emission test platform, which includes: a mechanical loading device, an acoustic emission monitoring system, and the acoustic emission mold prepared in step S1. Preferably, in this embodiment, the mechanical loading device is a rock direct shear rheological testing machine manufactured by Jilin Guanteng Automation Technology Co., Ltd., with a GTC550 fully digital controller and rheological testing software. The AE monitoring device adopts the PCI-2 multi-channel AE monitoring system manufactured by PAC, Inc. In this embodiment, the mechanical loading device provides axial pressure and shear stress to the rock, the acoustic emission monitoring system acquires the transient waveform of rock acoustic emission in real time, and the improved acoustic emission mold is used to fix the acoustic emission probe to realize the acoustic emission monitoring of the rock specimen during the test.

[0081] S3. Use the acoustic emission test platform installed in step S2 to perform multiple acoustic emission tests of the whole process of rock shear failure. Place the rock specimen processed in step S1 into the acoustic emission mold, arrange the acoustic emission probe in the acoustic emission mold, set the parameters of the acoustic emission monitoring system, and record the data of multiple acoustic emission tests.

[0082] S4. Select three acoustic emission parameters with low correlation as calculation parameters of the rock shear failure whole process prediction model.

[0083] S5. Determine the input features and prediction target variables of the XGBoost algorithm, select data from the acoustic emission test data obtained in step S3, calculate the calculation parameters of the rock shear failure whole process prediction model selected in step S4, construct a data set for training the XGBoost algorithm, and train the XGBoost algorithm.

[0084] S6. Initialize the XGBoost algorithm trained in step S5, determine the basic operation mode and operation parameters of the algorithm, and construct a rock shear failure whole process prediction model.

[0085] S7. Input the acoustic emission parameters selected in step S4 into the rock shear failure whole process prediction model constructed in step S6, and output the rock shear failure prediction result.

[0086] The rock shear failure whole process prediction method based on acoustic emission monitoring provided in the embodiment fully utilizes the phenomenon that micro-cracks are initiated, expanded and fractured in the internal stress deformation process of rock, and instantaneous strain energy is released in the form of elastic stress wave. By installing acoustic emission probes on the surface or inside of the rock, these acoustic emission signals are monitored in real time, and relevant information about the internal structure changes of the rock is obtained. The acoustic emission technology is used to study the relationship between acoustic emission parameter data and different stages of rock direct shear, and the XGBoost algorithm model in machine learning is used to classify and predict the collected acoustic emission parameter data and rock direct shear test stages. Finally, a rock shear failure whole process prediction method based on acoustic emission monitoring is proposed, which can collect relevant information about the internal structure changes of rock in real time and stably, and meet the requirements of engineering practice for the accuracy and timeliness of rock stability prediction.

[0087] In addition, the embodiment adopts a comprehensive analysis method of three low-correlation acoustic emission parameters, which reflect the changes in the rock interior from different angles. Through comprehensive consideration of the three parameters of energy, b value and main frequency, the limitations of single parameter analysis can be avoided, and the changes in the rock interior can be captured more comprehensively and accurately. This comprehensive analysis method can improve the accuracy and reliability of the prediction of rock shear failure, and make the prediction results more in line with the actual situation. Whether in the laboratory rock direct shear test or in the actual engineering application scenario, it can provide more powerful support for the evaluation and prediction of rock stability.

[0088] In embodiment 2, a rock shear failure whole-process prediction method based on acoustic emission monitoring is provided, comprising the following steps:

[0089] S1. Collect rock samples and process them into multiple rock test pieces. According to the size of the rock test pieces, an acoustic emission mold is made, and a square hole is opened in the acoustic emission mold.

[0090] S2. Install an acoustic emission test platform, which comprises a mechanical loading device, an acoustic emission monitoring system and the acoustic emission mold made in step S1. Preferably, in the embodiment, the mechanical loading device is arranged outside the acoustic emission mold and is used to apply axial pressure and shear stress to the rock test piece. The acoustic emission monitoring system is arranged outside the acoustic emission mold and comprises an acoustic emission probe, a preamplifier, an acoustic emission acquisition card and acoustic emission processing software running on an electronic device, which is used to collect the acoustic emission transient waveform of the rock test piece in real time. The acoustic emission mold is arranged between the rock test piece and the mechanical loading device, and is used to carry the rock test piece and fix the acoustic emission probe. The acoustic emission probe installation spacing is 10-20 mm.

[0091] Step 301. Put the rock test piece processed in step S1 into the acoustic emission mold, make the acoustic emission probe contact with the rock test piece through the square hole of the acoustic emission mold, and insert the wedge-shaped plug into the square hole to fix the acoustic emission probe. Preferably, in the embodiment, the acoustic emission probes are uniformly arranged at a position 5 cm away from the upper and lower surfaces of the rock test piece, and the contact parts of the rock test piece and the acoustic emission probes are uniformly smeared with vaseline with a thickness of 0.1-0.2 mm.

[0092] Step 302. Set the acoustic emission monitoring system parameters, and perform mechanical preloading on the rock sample before the formal mechanical loading of the mechanical loading device. After the mechanical preloading is completed, the vertical force data, vertical displacement data, lateral force data and lateral displacement data are zeroed. Preferably, in the present embodiment, the acoustic emission monitoring system parameters include: channel number, threshold value, amplification gain and sampling rate. The channel number is set to 8, the threshold value is set to 45dB, the amplification gain is set to 40dB, and the sampling rate is set to 1MSPS. The preloading force of the mechanical preloading is 10%-20% of the maximum expected bearing force of the rock sample.

[0093] Step 303. Load the normal force applied by the mechanical loading device to the initial normal stress value, and then use the mechanical loading device to perform mechanical loading at a shear rate of 0.2mm / min. The mechanical loading process is kept in synchronization with the data monitoring of the acoustic emission monitoring system during the mechanical loading process. The initial normal stress value is set according to the particle size and normal stress conditions of the rock sample.

[0094] Step 304. When the mechanical loading of step 303 reaches the peak value, i.e. the end of the rock sample failure stage of the acoustic emission test, simultaneously stop the mechanical loading of the mechanical loading device and the acoustic emission test data monitoring of the acoustic emission monitoring system. The monitored acoustic emission test data is collected and saved.

[0095] Step 305. Remove the acoustic emission probe and the damaged rock sample, and clean the residue in the acoustic emission mold to prepare for the next test.

[0096] S4. Select three acoustic emission parameters with low correlation as calculation parameters of the rock shear failure whole process prediction model.

[0097] Preferably, in the present embodiment, the calculation parameters of the rock shear failure whole process prediction model in step S4 include: the logarithm of acoustic emission cumulative energy, b value and main frequency A;

[0098] wherein the acoustic emission cumulative energy is as shown in formula (1):

[0099] (1)

[0100] In the formula, represents the acoustic emission cumulative energy from time 0 to time T, represents the energy carried by the acoustic emission signal at time t, and the logarithm of the acoustic emission cumulative energy is .

[0101] The b value is as shown in formula (2):

[0102] (2)

[0103] wherein B represents the b value, m represents the total number of magnitude bins, M j represents the median magnitude of the jth bin, N j represents the number of acoustic emission events of the jth magnitude bin, as formula (3):

[0104] (3)

[0105] wherein a and b are constants, M represents the magnitude, and N represents the total number of earthquakes with a magnitude greater than M,

[0106] (4)

[0107] wherein A dB represents the acoustic emission amplitude;

[0108] The main frequency A is in KHz, and is extracted from the acoustic emission signal in the acoustic emission test data collected in step S3 by using fast Fourier transform. In this embodiment, the acoustic emission signal in the acoustic emission test data is subjected to fast Fourier transform to convert the acoustic emission time-domain waveform into a frequency-domain waveform, and the frequency corresponding to the maximum amplitude point in the frequency-domain waveform is the main frequency of the acoustic emission.

[0109] Step 501. Select data from the acoustic emission test data collected in step S3, calculate the logarithm of the acoustic emission cumulative energy, the b value and the main frequency A, and select the rock shear failure stage as the prediction target variable. Preferably, in this embodiment, whether the rock specimen is broken or in a certain stage of breaking is taken as the target variable.

[0110] Step 502. Divide the rock shear failure stage into a compaction stage, an elastic stage, a non-elastic stage and a post-peak failure stage, encode the compaction stage as 1, the elastic stage as 2, the non-elastic stage as 3, and the post-peak failure as 4.

[0111] Step 503. Encode the logarithm of the acoustic emission cumulative energy, the b value and the main frequency A of each rock specimen according to the corresponding rock shear failure stage to form labeled data and construct a data set for training the XGBoost algorithm.

[0112] Step 504. Divide the stress stage of the rock specimen into an elastic stage, a yield stage, a strengthening stage and a failure stage, and take the elastic stage, the yield stage, the strengthening stage and the failure stage as the input features of the XGBoost algorithm.

[0113] Step 505. The compaction phase, the elastic phase, the inelastic phase, or the post-peak failure phase of the rock specimen is taken as a target variable of the XGBoost algorithm.

[0114] Step 506. The XGBoost algorithm is trained.

[0115] Step 601. A basic running mode is determined, the algorithm is set to run data in a tree structure gbtree, an algorithm similar to a tree structure is constructed to analyze data step by step, and general parameters are set. Preferably, the general parameters in the embodiment include: the silent mode is opened, the silent is 0, and the algorithm running information is output; the multi-thread control parameter nthread is set to the maximum thread number of the computer.

[0116] Step 602. The XGBoost algorithm boost parameter is set, wherein,

[0117] The maximum iteration number n_estimator is set to 100,

[0118] The iteration step length learning_rate is set to 0.7,

[0119] The step length of each iteration is set to 0.7,

[0120] The maximum depth of the tree max_depth is set to 1.5,

[0121] The proportion of random sampling of each tree subsample is set to 1,

[0122] The proportion of column sampling of each tree colsample_bytree is set to 1,

[0123] The proportion of column sampling of each tree colsample_bylevel is set to 1,

[0124] The weight L1 regularization term reg_alpha is set to 0,

[0125] The weight L2 regularization term reg_lambda is set to 0,

[0126] The minimum leaf node sample weight is set to 1,

[0127] The min_child_weight is set to 1.

[0128] Step 603. The XGBoost algorithm learning target parameter is set, wherein,

[0129] objective is used to specify the learning input features and the logical relationship between the rock shear failure stage, the input features including the logarithm of the acoustic emission cumulative energy, the b value and the main frequency A three input features, the XGBoost algorithm selects binary: logistic binary logistic regression, the output is the probability of the rock sample in the compression stage, the elastic stage, the inelastic stage and the post-peak failure stage,

[0130] eval_metric is the quantitative evaluation of valid data, the XGBoost algorithm takes "auc" which is the area under the ROC curve, and is set to evaluate the performance of the model by calculating the area under the receiver operating characteristic curve ROC curve,

[0131] seed is a random number seed, set to 1.

[0132] S7. The rock shear failure whole process prediction method obtained in step S4 is input into the rock shear failure whole process prediction model constructed in step S6, and the rock shear failure prediction result is output.

[0133] In example 3, a rock shear failure whole process prediction method based on acoustic emission monitoring is provided, and an example of establishing a rock direct shear process classification prediction model based on machine learning of different particle size yellow sandstone is taken as an example to illustrate the rock shear failure whole process prediction method based on acoustic emission monitoring in detail.

[0134] The design background of the rock shear failure whole process prediction method based on acoustic emission monitoring in this embodiment is as follows: In the excavation and construction process of numerous rock engineering projects, mine engineering, underground tunnel engineering and slope engineering projects often face complex geological conditions. Among them, fault is a common adverse geological structure, and the rock mass has poor mechanical properties, and there are a large number of cracks and weak planes in the rock, which makes the rock mass prone to shear failure when subjected to external stress. This shear failure can cause deformation, collapse and other disasters of the rock mass, which not only seriously affects the progress of the project, but also threatens the life safety of the construction personnel and the stability of the surrounding environment.

[0135] Yellow sandstone is a kind of rock that is relatively common in geological structure and widely exists in many tunnel engineering strata. These physical and mechanical properties of yellow sandstone are representative, and it is widely distributed in the strata involved in tunnel engineering. Through detailed indoor test research on yellow sandstone, the mechanical behavior and acoustic emission characteristics under different conditions are analyzed, the mechanism of rock shear failure is deeply understood, and then an effective rock stability evaluation and prediction method is established, which provides strong technical support for tunnel fault stability prediction.

[0136] The mechanical properties of yellow sandstone are as follows: the density of yellow sandstone is usually between 2.2-2.6 g / cm 3 (about 2.4 g / cm 3 on average in some areas), and the porosity is generally about 10%-25% (about 12% for fine particles and about 20% for coarse particles). In terms of mechanical properties, the uniaxial compressive strength varies with particle size, cementation degree and other factors, and is 20-50 MPa for coarse particles, 30-60 MPa for medium particles, and 40-80 MPa for fine particles, and the compressive strength increases with the increase of confining pressure.

[0137] The embodiment provides a rock shear failure whole-process prediction method based on acoustic emission monitoring, comprising the following steps:

[0138] Step S1. Collect rock samples and process them into multiple rock test pieces, manufacture an acoustic emission mold according to the size of the rock test pieces, and open a square hole on the acoustic emission mold.

[0139] Preferably, in the embodiment, yellow sandstone rock taken from Chuxiong in Yunnan is selected as a sample according to the actual engineering situation. The rock in this area is mainly yellow sandstone, and also contains samples with different degrees of weathering, so as to better simulate the diversity of rock in actual engineering. During the collection process, professional rock collection tools are used to avoid unnecessary damage to the rock. The collected rock samples are transported to the laboratory, and the yellow sandstone test pieces are processed into cubic rock test pieces with a size of 100*100*100. There are 14 coarse-grained yellow sandstone test pieces, 5 medium-grained yellow sandstone test pieces, and 5 fine-grained yellow sandstone test pieces, which are numbered as CSY1-14, ZSY1-5 and XSY1-5 respectively. CSY represents coarse sandstone, ZSY represents medium sandstone, and XSY represents fine sandstone. The rock test pieces are in good integrity, the surface is smooth, there is no obvious defect, and the end parallelism and perpendicularity are controlled within ±0.02 mm.

[0140] Preferably, in the embodiment, a direct shear acoustic emission mold as shown in Figure 2 is developed. A square hole is opened on the acoustic emission mold, and the size and position of the hole are determined according to the size of the acoustic emission probe and the monitoring requirements. The distance between the acoustic emission probes is reasonably selected, and the distance between the acoustic emission probes is selected to be 10-20 mm according to the size of the rock test piece and the physical properties of the rock test piece, so as to ensure the strength and accuracy of the acoustic emission signal and avoid mutual interference between the acoustic emission probes. A wedge-shaped plug for fixing the acoustic emission probe is provided, and the size and shape of the wedge-shaped plug are matched with the hole and the probe, so as to ensure that the acoustic emission probe can be stably fixed during the test process and achieve the goal of acoustic emission monitoring during the test process.

[0141] S2. Install the acoustic emission test platform, which comprises a mechanical loading device, an acoustic emission monitoring system and the acoustic emission mold manufactured in step S1.

[0142] Preferably, the present embodiment includes debugging installation and straight shear acoustic emission mold, etc. The instruments used in this test include: the mechanical loading device used is the rock straight shear rheological testing machine produced by Jilin Guanteng Automation Technology Co., Ltd., GTC550 full-digital controller and rheological test software. The AE monitoring device uses the PCI-2 type multi-channel AE monitoring system produced by PAC, a company of physical acoustics in the United States. In this embodiment, the mechanical loading device is used to provide axial pressure and shear stress for the rock, the acoustic emission monitoring system is used to collect rock acoustic emission transient waveform in real time, the improved acoustic emission mold is used to fix the acoustic emission probe, and the acoustic emission monitoring of the rock specimen during the test is realized.

[0143] S3. The test scheme is implemented, the acoustic emission test platform installed in step S2 is used to perform acoustic emission test of rock shear failure in the whole process for multiple times, the rock specimen processed in step S1 is placed into the acoustic emission mold, the acoustic emission probe is arranged in the acoustic emission mold, the parameters of the acoustic emission monitoring system are set, and the data of multiple acoustic emission tests are recorded.

[0144] Preferably, in this embodiment, the processed rock specimen is placed into the installed acoustic emission mold, 8 acoustic emission probes are arranged according to the pre-designed acoustic emission probe arrangement scheme, and the 8 acoustic emission probes are in contact with the specimen through the mold holes and are uniformly arranged at a position 5 cm away from the upper and lower surfaces of the specimen. Vaseline is uniformly applied to the contact part between the rock specimen and the acoustic emission probe, and the amount of application is appropriate to ensure good coupling effect, and the general application thickness is about 0.1-0.2 mm. The wedge-shaped plug is inserted into the hole to fix the acoustic emission probe. The parameter setting scheme of the sliding window length and the sliding distance in the main frequency data processing is shown in Table 1.

[0145] Table 1 Parameter setting scheme of sliding window length and sliding distance in main frequency data processing

[0146]

[0147] Preferably, in this embodiment, the parameters of the acoustic emission monitoring system are set, the channel number is set to 8, the threshold value is set to 45 dB, the amplification gain is set to 40 dB, and the sampling rate is set to 1 MSPS. The rock sample is preloaded before loading, and the preloading force is determined according to the size and properties of the sample and is 10%-20% of the maximum force expected to be borne by the sample. For example, for a cubic specimen with a side length of 100 mm, the maximum force expected to be borne is 100 kN, and the preloading force is 10-20 kN. After preloading, the vertical force and displacement data and the horizontal force and displacement data are zeroed. Then, the normal force is applied according to the test requirements, and the normal force setting value is determined according to the actual engineering situation and the test requirements. In this embodiment, four situations of 6 MPa, 2 MPa, 4 MPa and 8 MPa are set. The loading is performed at a shear rate of 0.2 mm / min.

[0148] Preferably, in this embodiment, the normal force is applied to a set value according to the test design. In this embodiment, different initial normal stress values are set for samples of different particle sizes and different normal stress conditions, including: 6 MPa, 2 MPa, 4 Mpa and 8 MPa, etc. The direct shear testing machine is then loaded at a shear rate of 0.2 mm / min, and the loading process is synchronized with the acoustic emission monitoring to ensure that the mechanical response and acoustic emission signals of the rock can be obtained simultaneously. When the loading is stopped at the end of the post-peak failure stage, the mechanical loading system and the acoustic emission monitoring system are stopped simultaneously, and the test data is saved. Then, the acoustic emission probe is carefully removed, the damaged specimen is taken out, and the rock residue in the mold is carefully cleaned to prepare for the next test.

[0149] S4. Select three acoustic emission parameters with low correlation as the calculation parameters of the rock shear failure whole process prediction model.

[0150] Preferably, in this embodiment, after obtaining the acoustic emission test data, according to the correlation of the acoustic emission parameters, three acoustic emission parameters with low correlation, i.e. energy parameter, b value and main frequency, are calculated and used as the parameters used in the rock shear failure whole process prediction method.

[0151] Preferably, in this embodiment, for the acoustic emission energy Es, the energy utilization rate and the acoustic emission cumulative energy are solved according to the acoustic emission energy Es, and the logarithm of the acoustic emission energy contribution rate and the logarithm of the acoustic emission cumulative energy (i.e. 1g acoustic emission cumulative energy) are taken as two parameters. The parameter setting scheme of the sliding window length and the sliding distance in the main frequency data processing is shown in Table 2.

[0152] Table 2 Parameter setting of sliding window length and sliding distance in main frequency data processing

[0153]

[0154] S5. Determine the input features and prediction target variables of the XGBoost algorithm, select data from the acoustic emission test data obtained in step S3, calculate the calculation parameters of the rock shear failure whole process prediction model selected in step S4, construct a data set for training the XGBoost algorithm, and train the XGBoost algorithm.

[0155] Preferably, the present embodiment selects data from the acoustic emission data during the rock direct shear process to establish the data set required for the training algorithm, and selects the 1g acoustic emission cumulative energy, b value and dominant frequency as the three characteristics in the acoustic emission parameters, and takes the rock shear failure stage as the classification prediction object. According to the mechanical properties and failure modes during the rock direct shear process, the rock shear failure stage is divided into the compaction stage, elastic stage, inelastic stage and post-peak failure stage. The four stages are coded as 1, 2, 3 and 4, respectively.

[0156] Preferably, the present embodiment establishes a detailed correspondence relationship for the test data, that is, for each rock sample, the three characteristic values of the 1g acoustic emission cumulative energy, b value and dominant frequency are explicitly coded for the corresponding rock shear failure stage, thereby forming labeled data. At the same time, the four stress stages (such as the elastic stage, yield stage, strengthening stage, failure stage, etc.) are taken as input characteristics, and whether the sandstone is broken or in a certain stage of breakage is taken as the target variable.

[0157] S6. The XGBoost algorithm trained in step S5 is initialized and set, the basic operation mode and operation parameters of the algorithm are determined, and the rock shear failure whole-process prediction model is constructed. Preferably, the general settings of the XGBoost algorithm used in the present embodiment are as follows. The algorithm is divided into two parts of model training and test verification, the model is trained by iteration, new classifiers are added to fit the training error in the current iteration, the fitting effect of the model on the sample is optimized, and the classification prediction performance of the model is verified in the test set verification stage. The XGBoost algorithm process is shown in Figure 3 The basic operation mode is determined first, the algorithm is set to run data in the tree structure gbtree, and an algorithm similar to a tree structure is constructed to analyze the data step by step. At the same time, general parameters are set, such as whether to start the silent mode silent, which is set to 0, that is, to output the algorithm running information, so as to facilitate understanding of the process; the multi-thread control parameter nthread is set to the maximum number of threads of the computer, so as to fully utilize the computer resources and improve the running speed.

[0158] Preferably, the boost parameters of the XGBoost algorithm used in the present embodiment model are as follows:

[0159] (1) The maximum number of iterations n_estimator is set to 100, and 100 iterations are performed to optimize the model.

[0160] (2) The iteration step learning_rate is set to 0.7, and the step size of each iteration is set to 0.7, the model prediction accuracy is high, and the model running speed is fast.

[0161] (3) The maximum depth of the tree max_depth is set to 1.5 to control the complexity of the model and prevent overfitting. A tree that is too deep may over-learn the noise in the training data, affecting the performance on new data.

[0162] (4) The proportion of random sampling of each tree subsample, the proportion of column sampling of each tree colsample_bytree, and the proportion of column sampling of each tree at each node split colsample_bylevel are all set to 1 to ensure that the model fully learns the data features and avoid overfitting or underfitting.

[0163] (5) The weight L1 regularization term reg_alpha is set to 0 to speed up the algorithm running speed; the weight L2 regularization term reg_lambda is set to 0 to prevent overfitting.

[0164] (6) The minimum leaf node sample weight and min_child_weight are set to 1 to avoid overfitting.

[0165] Preferably, the XGBoost algorithm learning target parameters used in the embodiment are as follows:

[0166] (1) objective: used to specify the learning goal and learning target, the model in this paper selects binary: logistic, binary classification logistic regression, and the output is probability. When predicting the rock shear failure stage, the model will learn the logical relationship between the input features (lg acoustic emission cumulative energy, b value and main frequency) and the rock shear failure stage (coded as 1-4), and output the probability of the rock being in a certain stage.

[0167] (2) eval_metric: quantitative evaluation of effective data, the model in this paper takes "auc" which is the area under the ROC curve. Set to evaluate the performance of the model by calculating the area under the receiver operating characteristic curve ROC curve.

[0168] (3) seed: random number seed, set to 1 as the random number seed. When the model is trained, by setting a fixed random number seed, the data is randomly shuffled, some random parameters are initialized, etc.

[0169] S7. The calculated lg acoustic emission cumulative energy, b value and main frequency and other three acoustic emission parameters are input into the XGBoost algorithm model set up, and the rock shear failure stage can be predicted, and the prediction result is output.

[0170] Preferably, the embodiment also includes step S8. ACC evaluation index model evaluation and application.

[0171] Step S8 in this embodiment includes:

[0172] S801. Set up indexes and two performance curves to evaluate the effect of the model.

[0173] Accuracy (ACC): defined as the ratio of the number of correctly classified samples to the total number of samples, reflecting the overall data classification accuracy, but the ACC index is more affected by the majority class classification data set. For class-imbalance data sets, the ACC evaluation index is not comprehensive enough. The ACC evaluation index is shown in equation (5):

[0174] (5)

[0175] In the formula, ACC is the accuracy; TP, TN, FP, and FN are the numbers of true, true negative, false positive, and false negative samples, respectively.

[0176] According to the prediction results of the three characteristics of lg acoustic emission cumulative energy, b value and main frequency A, it is judged whether the sample is correctly classified or not. For example, if the ACC evaluation index model predicts that a rock sample is in the damage stage 3 (inelastic stage) and the actual value is also 3, the sample is correctly classified and the number of correctly classified samples is counted.

[0177] (2) Balanced Accuracy, BACC: defined as the average accuracy, since the ACC evaluation index is sometimes not comprehensive enough, BACC is calculated. The classification accuracy of each rock shear failure stage is calculated and averaged. That is, the classification accuracy of the compaction stage, elastic stage, inelastic stage and post-peak damage stage is calculated, and the average BACC is obtained, as shown in equation (6):

[0178] (6)

[0179] In the formula, BACC is the average accuracy; TP, TN, FP, and FN are the numbers of true, true negative, false positive, and false negative samples, respectively.

[0180] (3) F1-score: defined as the arithmetic mean of precision and recall, which combines the results of precision and recall together, and can comprehensively and objectively reflect the model performance. According to the prediction results of the three characteristics of lg acoustic emission cumulative energy, b value and main frequency, the precision and recall are calculated, and then the F1-score is calculated, as shown in equation (7):

[0181] (7)

[0182] In the formula, F1-score is the arithmetic mean of precision and recall; Precision is the precision, and Recall is the recall.

[0183] (4) Receiver Operating Characteristic (ROC) curve: it is one of the indicators used to reflect the performance of the classifier, and is the relationship curve between the True Positive Rate (TPR) and the False Positive Rate (FPR). The size of the Area Under Curve (AUC) is a quantitative value for evaluating the performance of the classifier. The FPR is shown in equation (8):

[0184] (8)

[0185] In the formula, FPR is the false positive rate of classification, TN is the number of true negative samples, FP is the number of false positive samples, and FN is the number of false negative samples.

[0186] TPR is shown in equation (9):

[0187] (9)

[0188] In the formula, TPR is the true positive rate of classification, TP is the number of true positive samples, FP is the number of false positive samples, and FN is the number of false negative samples.

[0189] Precision-Recall (PR) curve: it can also reflect the performance of the classifier, and is the relationship curve between the precision and the recall. The AUC can be calculated to evaluate the performance of the model.

[0190] Step S802. The effect of the rock shear failure whole process prediction model is evaluated using the set indicators and performance curves, and the performance of the rock shear failure whole process prediction model is analyzed. After evaluation, the rock direct shear different stage classification prediction model based on acoustic emission data established in this embodiment has an accuracy of 77%. The number of true values and predicted correct values in each stage is shown in Figure 4 . The rock shear failure whole process prediction model is applied to the rock direct shear test of the prepared rock sample. The prediction effect of the CSY-3 test piece is shown in Figure 5 , the prediction effect of the ZSY-4 test piece is shown in Figure 6 , and the prediction effect of the XSY-3 test piece is shown in Figure 7 . As can be seen from the figures, the rock shear failure whole process prediction model of this embodiment has good effect on single-class rock direct shear stage classification prediction.

[0191] The embodiments of the present application are presented by way of example and description, and are not intended to be exhaustive or to limit the application to the form disclosed. Many modifications and variations will be apparent to those skilled in the art. Embodiments are chosen and described in order to best explain the principles of the application and its practical application, and to thereby enable others skilled in the art to best utilize the application in various embodiments and with various modifications as are suited to the particular use contemplated.

Claims

1. A rock shear failure whole process prediction method based on acoustic emission monitoring, characterized in that, The method comprises the following steps: S1. Collecting rock samples and processing them into multiple rock test pieces, making an acoustic emission mold according to the size of the rock test pieces, and opening a square hole on the acoustic emission mold; S2. Installing an acoustic emission test platform, which comprises a mechanical loading device, an acoustic emission monitoring system, and the acoustic emission mold made in step S1; S3. Using the acoustic emission test platform installed in step S2 to perform multiple acoustic emission tests of the whole process of rock shear failure, placing a rock test piece processed in step S1 into the acoustic emission mold, arranging acoustic emission probes in the acoustic emission mold, setting parameters of the acoustic emission monitoring system, and recording data of multiple acoustic emission tests; S4. Selecting three acoustic emission parameters with low correlation as calculation parameters of a rock shear failure whole process prediction model; S5. Determining input features and prediction target variables of an XGBoost algorithm, selecting data from the acoustic emission test data obtained in step S3, calculating the calculation parameters of the rock shear failure whole process prediction model selected in step S4, constructing a data set for training the XGBoost algorithm, and training the XGBoost algorithm; S6. Initializing the XGBoost algorithm trained in step S5, determining the basic operation mode and operation parameters of the algorithm, and constructing the rock shear failure whole process prediction model; S7. Inputting the calculation parameters of the rock shear failure whole process prediction method obtained in step S4 into the rock shear failure whole process prediction model constructed in step S6, and outputting rock shear failure prediction results; The calculation parameters of the rock shear failure whole process prediction model in step S4 comprise the logarithm of acoustic emission cumulative energy, the b value, and the main frequency A; The acoustic emission cumulative energy is shown in formula (1): （1） In the formula, represents the cumulative energy of acoustic emission from 0 time to T time, represents the energy carried by the acoustic emission signal at the t time. The logarithm of the cumulative energy of the acoustic emission is ; The b value is shown in formula (2): （2） where B represents the b value, m represents the total number of magnitude bins, M j represents the median magnitude of the jth bin, N j represents the number of acoustic emission events for the jth bin of magnitude, and as formula (3): （3） In the formula, a and b are constants, M represents the magnitude shown in formula (4), and N represents the total number of earthquakes with a magnitude greater than M, （4） In the formula, A dB represents the acoustic emission amplitude; The main frequency A is in KHz and is extracted from the acoustic emission signal in the acoustic emission test data collected in step S3 by using fast Fourier transform; Step S5 comprises: Step 501. Selecting data from the acoustic emission test data collected in step S3, calculating the logarithm of acoustic emission cumulative energy, the b value, and the main frequency A, and selecting the rock shear failure stage as the prediction target variable; Step 502. Dividing the rock shear failure stage into a compaction stage, an elastic stage, a non-elastic stage, and a post-peak failure stage, encoding the compaction stage as 1, the elastic stage as 2, the non-elastic stage as 3, and the post-peak failure as 4; Step 503. Encoding the logarithm of acoustic emission cumulative energy, the b value, and the main frequency A of each rock test piece according to the corresponding rock shear failure stage to form labeled data and construct a data set for training the XGBoost algorithm; Step 504. Dividing the stress stage of the rock sample into an elastic stage, a yield stage, a strengthening stage and a failure stage, and taking the elastic stage, the yield stage, the strengthening stage and the failure stage as input features of the XGBoost algorithm; Step 505. Taking the compaction stage, the elastic stage, the inelastic stage or the post-peak failure stage of the rock sample as a target variable of the XGBoost algorithm; Step 506. Training the XGBoost algorithm.

2. The rock shear failure whole process prediction method based on acoustic emission monitoring according to claim 1, characterized in that, In the step S1, the rock sample is a cubic sample, the surface of the rock sample is smooth, and the end parallelism and perpendicularity error is ±0.02mm; The acoustic emission mold further comprises a wedge-shaped plug for fixing the acoustic emission probe.

3. The rock shear failure whole process prediction method based on acoustic emission monitoring according to claim 2, characterized in that, In the step S2, the mechanical loading device is arranged outside the acoustic emission mold and is used to apply axial pressure and shear stress to the rock sample; The acoustic emission monitoring system is arranged outside the acoustic emission mold and comprises an acoustic emission probe, a preamplifier, an acoustic emission acquisition card and acoustic emission processing software running on an electronic device, and is used to collect acoustic emission transient waveforms of the rock sample in real time; The acoustic emission mold is arranged between the rock sample and the mechanical loading device and is used to carry the rock sample and fix the acoustic emission probe; The acoustic emission probe installation spacing is 10-20mm.

4. The rock shear failure whole process prediction method based on acoustic emission monitoring according to claim 3, characterized in that, The step S3 comprises: Step 301. Placing the rock sample processed in the step S1 into the acoustic emission mold, contacting the acoustic emission probe with the rock sample through the square hole of the acoustic emission mold, and inserting the wedge-shaped plug into the square hole to fix the acoustic emission probe; Step 302. Setting the acoustic emission monitoring system parameters, and preloading the rock sample mechanically before formal mechanical loading by the mechanical loading device, and after the mechanical preloading is completed, zeroing the vertical force data, vertical displacement data, lateral force data and lateral displacement data; Step 303. Loading the normal force applied by the mechanical loading device to an initial normal stress value, and then using the mechanical loading device to perform mechanical loading at a shear rate of 0.2mm / min, keeping the mechanical loading process and the data monitoring of the acoustic emission monitoring system synchronous during the mechanical loading process, and the initial normal stress value is set according to the particle size and normal stress condition of the rock sample; Step 304. When the mechanical loading of step 303 reaches the peak value, i.e. the failure stage of the acoustic emission test rock sample ends, simultaneously stopping the mechanical loading of the mechanical loading device and the acoustic emission test data monitoring of the acoustic emission monitoring system, collecting and saving the monitored acoustic emission test data; Step 305. Taking out the acoustic emission probe and the broken rock sample, and cleaning the residues in the acoustic emission mold.

5. The rock shear failure whole process prediction method based on acoustic emission monitoring according to claim 4, characterized in that, In the step 301, the acoustic emission probes are uniformly arranged at positions 5cm away from the upper and lower surfaces of the rock sample; The contact parts of the rock sample and the acoustic emission probe are uniformly smeared with vaseline with a thickness of 0.1-0.2mm.

6. The rock shear failure whole process prediction method based on acoustic emission monitoring according to claim 5, characterized in that, The parameters of the acoustic emission monitoring system in the step 302 include: channel number, threshold value, amplification gain and sampling rate, the channel number is set to 8, the threshold value is set to 45dB, the amplification gain is set to 40dB, and the sampling rate is set to 1MSPS; The preloading force of the mechanical preloading is 10%-20% of the maximum expected bearing force of the rock sample.

7. The rock shear failure whole process prediction method based on acoustic emission monitoring according to claim 6, characterized in that, The step S6 comprises: Step 601. Determine the basic operation mode, set the algorithm to run data in tree structure gbtree, construct an algorithm similar to tree structure to analyze data step by step, and set general parameters; Step 602. Set the XGBoost algorithm boost parameter, wherein, The maximum iteration number n_estimator is set to 100, The iteration step length learning_rate is set to 0.7, The step length of each iteration is set to 0.7, The maximum depth of the tree max_depth is set to 1.5, The proportion of random sampling of each tree subsample is set to 1, The proportion of random sampling column number of each tree colsample_bytree is set to 1, The proportion of column sampling of each tree per node split colsample_bylevel is set to 1, The weight L1 regularization term reg_alpha is set to 0, The weight L2 regularization term reg_lambda is set to 0, The minimum leaf node sample weight is set to 1, min_child_weight is set to 1; Step 603. Set the XGBoost algorithm learning target parameter, wherein, objective is used to specify the logical relationship between the input features and the rock shear failure stage, the input features include the logarithm of the acoustic emission cumulative energy, b value and main frequency A three input features, the XGBoost algorithm selects binary:logistic binary logistic regression, and the output is the probability of the rock sample being in the unbroken, compaction stage, elastic stage, non-elastic stage and post-peak damage stage, eval_metric is the quantitative evaluation of effective data, the XGBoost algorithm takes "auc" which is the area under the ROC curve, and is set to evaluate the performance of the model by calculating the area under the receiver operating characteristic curve ROC curve, seed is a random number seed, which is set to 1.

8. The rock shear failure whole process prediction method based on acoustic emission monitoring according to claim 7, characterized in that, The general parameters set in the step 601 include: the silent mode is opened, the silent is 0, the algorithm running information is output, and the multi-thread control parameter nthread is set to the maximum thread number of the computer.

Images