Method and system for representing progressive rock failure based on stress and acoustic wave changes

Abstract

The application discloses a method and system for representing rock progressive failure based on stress and acoustic wave change, and relates to the technical field of rock mass mechanics, and comprises the following steps: by applying an incremental constant clamping pressure to the same type of rock test block, obtaining the response data thereof, and performing a yield test to obtain the peak strength of the rock, taking the strain and acoustic wave response as the indication response data, determining the initial training weight and dynamically correcting by analyzing the correlation of the indication response and the damage response in each damage stage. A multilayer perceptron model is established, the preprocessed indication response data is taken as the input, the damage response data is taken as the label for training, the damage response data prediction value is obtained by inputting the indication response data of the rock to be analyzed, and then the internal damage coefficient and the external indication coefficient are calculated, so that the progressive damage degree of the rock is comprehensively analyzed, and the prediction deviation caused by the static weight in the traditional method is reduced.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of rock mechanics, in particular to a method and system for jointly characterizing rock progressive failure based on stress and acoustic wave changes. BACKGROUND

[0002] During the process of being subjected to external stress loading, the internal structure of rock will gradually be damaged until reaching the final yield instability state. This process is usually referred to as rock progressive failure. The essence of rock progressive failure is the continuous accumulation of internal damage of rock under external force, as well as the formation and expansion of macroscopic cracks, which widely exists in the fields of geotechnical engineering, mining, tunnel construction, and earthquake prediction, and has important significance for engineering safety. In order to study the mechanical behavior of rock in the process of progressive failure, experimental research usually obtains the mechanical response of rock through loading test. However, due to the heterogeneity and complex mechanical properties of rock, it is difficult to accurately reveal the damage characteristics of rock in each stage from the initial loading to the failure instability in the whole life cycle by simply relying on mechanical parameters such as stress-strain curve.

[0003] In recent years, acoustic wave monitoring technology has been widely used in the monitoring of rock failure process as a non-destructive testing method. The acoustic wave propagation process can reflect the changes of the internal microstructure of rock, for example, the decrease of acoustic velocity is often related to the expansion of cracks, and the attenuation of acoustic amplitude usually corresponds to the damage degree of the material. Therefore, the time series changes of acoustic wave parameters can be an important supplement to mechanical parameters, helping researchers to track the microstructure changes of rock in real time.

[0004] In the prior art, the method for jointly characterizing rock progressive failure based on stress and acoustic wave changes disclosed in CN112986020A includes a TH100 series acoustic wave parameter tester, a PC, and a microcomputer-controlled electro-hydraulic servo rock triaxial shear rheological testing machine. First, the sample is placed on the microcomputer-controlled electro-hydraulic servo rock triaxial shear rheological testing machine. During the test, the acoustic wave pressure head is placed at both ends of the rock test block, and the number, time interval, and cycle of the excited acoustic wave signal are set. During the test, the time information and longitudinal wave information during the triaxial compression test are collected simultaneously. The method processes the acoustic wave data on the PC to obtain the wave velocity-stress-time change curve, further processes the acoustic wave data, and uses the "wave velocity-stress characteristic value" σvpc, σvpi, and σvpd to describe the progressive failure process of rock. However, in this method, the relationship between acoustic wave and stress characteristic value is relatively fixed, and the changes in response parameters at different stages of the experimental process and the influence of signal errors on the analysis are not considered. Therefore, the accuracy and effectiveness of the failure characterization are reduced.

[0005] The above information disclosed in the background section is only for the purpose of enhancing the understanding of the background of the present disclosure and, as such, it can include information that does not constitute prior art that is already known to those of ordinary skill in the art. SUMMARY

[0006] The purpose of the present application is to provide a method and system for jointly characterizing the progressive failure of rock based on stress and acoustic wave changes, in order to solve the problems raised in the background.

[0007] To achieve the above-mentioned purpose, the present application provides the following technical solutions:

[0008] The method for jointly characterizing the progressive failure of rock based on stress and acoustic wave changes comprises the following specific steps:

[0009] A plurality of rock test blocks of the same type as the rock to be analyzed are subjected to an increasing constant clamping pressure, and the strain response, acoustic wave response and surface crack morphology response data are obtained. The rock peak strength of the rock test block under the influence of different constant clamping pressures is obtained through a yield test. The strain response and acoustic wave response are used as indicator response data, and the surface crack morphology response and rock peak strength are used as damage response data.

[0010] Based on the strain response and acoustic wave response data, each rock test block is divided into a corresponding damage stage, and the correlation between the indicator response data and the damage response data in each damage stage is analyzed to determine the initial training weight of the indicator response data in each damage stage.

[0011] A multilayer perceptron model is established, the preprocessed indicator response data is used as input, the damage response data is used as label, and the corresponding initial training weight is corrected through the real-time response credibility of the indicator response data to obtain a dynamic weight. The multilayer perceptron model is trained according to the dynamic weight of each group of response data.

[0012] The indicator response data of the rock to be analyzed within the detection period is obtained and input into the multilayer perceptron model. The damage response data prediction value at different time points within the detection period is obtained, and the internal damage coefficient and the external indicator coefficient are calculated according to the damage response data prediction value.

[0013] The internal damage coefficient and the external indicator coefficient are comprehensively analyzed to jointly characterize the progressive failure degree of the rock to be analyzed.

[0014] Further, the strain response data is specifically a strain value, and the specific acquisition method is as follows: a plurality of monitoring points arranged with strain detection units are arranged on the surface of the rock test block, the strain values at each monitoring point on the surface of the rock test block are collected by the strain detection units, the strain value after stabilization is taken as the final strain value, and the acoustic wave response data and the surface crack morphology response data are obtained after the strain value stabilizes.

[0015] The acoustic wave response data is specifically longitudinal wave speed data of the rock test block, and the surface crack morphology response data includes crack length and crack width data.

[0016] A specific method for obtaining the surface crack morphology response data is: acquiring a rock test block surface image through a CCD camera system, performing crack edge detection and length quantification based on the rock test block surface image, and obtaining the surface crack morphology response data.

[0017] Further, the rock progressive failure process is divided into a plurality of failure stages according to the strain response and the acoustic wave response, wherein the entire rock progressive failure process is divided into an elastic deformation stage, a yield fracture stage and a macroscopic failure stage through the longitudinal wave speed and the strain value, wherein the longitudinal wave speed variation factor is calculated by detecting the change in the corresponding wave speed under different constant clamping pressures. The formula for calculating the longitudinal wave speed variation factor is:

[0018] ;

[0019] In the formula, is the longitudinal wave speed variation factor of the i-th rock test block, is the longitudinal wave speed of the i-th rock test block, is the longitudinal wave speed of the i+1-th rock test block, is the constant clamping pressure applied to the i-th rock test block, is the constant clamping pressure applied to the i+1-th rock test block, wherein i is the index of the rock test block, and each rock test block is sequentially indexed in ascending order of constant clamping force, , is the total number of rock test blocks;

[0020] The strain concentration factor is calculated by the strain value of the strain detection unit, and the strain concentration factor represents the strain concentration degree and damage condition of the test piece. The formula for calculating the strain concentration factor is:

[0021] ;

[0022] In the formula, is the strain concentration factor of the i-th rock test block, is the difference between the average strain value of the selected monitoring point in the i-th rock test block and the average strain value of all monitoring points, is the maximum strain value of the i-th rock test block, wherein the formula for calculating is as follows:

[0023] ;

[0024] In the formula, is the strain value of the gth selected monitoring point on the ith rock test block, is the strain value of the hth monitoring point on the ith rock test block, n is the total number of selected monitoring points in the rock test block, N is the total number of all monitoring points on the surface of the rock test block, wherein g is the index of the selected monitoring point on the rock test block, and h is the index of the monitoring point on the rock test block.

[0025] Further, the logic for selecting the monitoring points is as follows: based on the strain values of the surface monitoring points, arranging in descending order to form a strain sequence, and selecting the monitoring points in a preset proportion according to the strain sequence;

[0026] The logic for dividing the entire test process into several damage stages is as follows: setting the condition interval of each damage stage, and dividing the rock test block into the corresponding damage stage according to the condition interval into which the longitudinal wave velocity variation factor , the strain concentration factor change rate and the longitudinal wave velocity of the rock test block fall;

[0027] The formula for calculating the strain concentration factor change rate is as follows:

[0028] ;

[0029] In the formula, is the strain concentration factor of the ith rock test block, is the strain concentration factor of the ith+1 rock test block;

[0030] The condition interval of the elastic deformation stage is as follows:

[0031] ;

[0032] In the formula, is the acoustic wave change threshold value of the elastic deformation stage, is the strain concentration change threshold value of the elastic deformation stage, and are the minimum and maximum reference values of the longitudinal wave velocity in the elastic deformation stage, respectively;

[0033] The condition interval of the yield and fracture stage is as follows:

[0034] ;

[0035] In the formula, is the maximum reference value of the longitudinal wave velocity in the yield and fracture stage;

[0036] The condition interval of the macroscopic damage stage is as follows:

[0037] ;

[0038] Determine the condition interval that each rock sample meets and divide the rock sample into the corresponding damage stage.

[0039] Further, the correlation between the indicator response data and the damage response data is characterized by the Spearman rank correlation coefficient, and the logic is that:

[0040] 1) In each damage stage, the indicator response data and the damage response data of all rock samples in the damage stage are counted to form the longitudinal wave velocity sequence, the strain sequence, the rock peak strength sequence, the crack length sequence and the crack width sequence, and the elements in each sequence are sorted in ascending order to determine the ranking of each rock sample in each sequence in the damage stage.

[0041] 2) Based on the ranking of each element value in the sequence, the Spearman rank correlation coefficient is calculated, and the specific formula is:

[0042] ;

[0043] In the formula, is the rank correlation coefficient of the qth indicator response data to the Yth damage response data in the kth damage stage, is the ranking difference of the u th rock sample in the k th damage stage in the q th indicator response data sequence and the Y th damage response data sequence, is the number of rock samples in the k th damage stage, where u is the index of the rock sample in the damage stage, is the index of the indicator response data, and k is the index of the damage stage, is the index of the damage response data.

[0044] The logic for setting the initial training weight of different indicator response data is: according to the rank correlation coefficient, the initial training weight of each indicator response data is determined, and the specific formula for calculating the initial training weight is:

[0045] ;

[0046] In the formula, is the initial training weight of the qth indicator response data to the Yth damage response data in the kth damage stage, is the total number of indicator response data.

[0047] Further, the multi-layer perceptron model is divided into a plurality of sub-models based on the damage stages, each of the sub-models is established based on a LightGBM model, a ReLU is used as an activation function, and a mean absolute error is used as a loss function, wherein the response data is used as an input of each of the sub-models, the corresponding surface crack morphology data and the rock peak strength are used as labels, an attention mechanism is constructed through corresponding weights, and the sub-models are trained;

[0048] The logic for correcting the initial training weight according to the response credibility of the response data is as follows: the initial training weight is corrected based on the signal error of the acoustic wave response and the strain response, specifically including correction based on the average error of the signal in a single training round, wherein the acoustic wave response is corrected through the signal-to-noise ratio of the acoustic wave signal, and the formula for the correction is as follows:

[0049] ;

[0050] wherein, is a dynamic weight of the acoustic wave response data to the Yth damage response data in the kth damage stage, is a signal-to-noise ratio of the zth training sample, is an initial training weight of the acoustic wave response data to the Yth damage response data in the kth damage stage, and are a minimum signal-to-noise ratio and a maximum signal-to-noise ratio of the calibration respectively, wherein z is an index of the training sample in a single training round, , is a batch size of a single training round;

[0051] wherein the formula for correcting the initial training weight of the strain response is as follows:

[0052] ;

[0053] wherein, is a dynamic weight of the strain response data to the Yth damage response data in the kth damage stage, is an initial training weight of the strain response data to the Yth damage response data in the kth damage stage, is an actual clamping pressure of the zth training sample, is a set clamping pressure value of the zth training sample.

[0054] Further, the formula for calculating the internal damage coefficient is as follows:

[0055] ;

[0056] wherein, is the internal damage coefficient at time t, is the wave impedance damage at time t, which is calculated by the P-wave velocity at time t combined with the rock test density, is the rock peak strength prediction value at time t, and are proportional constants, is the intercept constant, and t is the time variable within the detection period;

[0057] The formula on which the calculation of the external indication coefficient is based is:

[0058] ;

[0059] In the formula, is the external indication coefficient at time t, is the surface crack area, which is calculated according to the crack length and width prediction values, is a proportional constant, is an intercept constant;

[0060] The logic on which the joint representation of the progressive damage degree of the rock to be analyzed is based is as follows: a damage risk coefficient is generated based on the comprehensive changes of the external indication coefficient and the internal damage coefficient, the progressive damage degree of the rock to be analyzed is represented based on the damage risk coefficient, and the formula on which the calculation of the damage risk coefficient is based is:

[0061] ;

[0062] In the formula, is the damage risk coefficient of the rock to be analyzed, and are the variances of the internal damage coefficient and the external indication coefficient of the rock to be analyzed within the detection period, and are the means of the internal damage coefficient and the external indication coefficient of the rock to be analyzed within the detection period, wherein and are weight coefficients, wherein , and and are both greater than 0;

[0063] If , it is determined that the rock to be analyzed has a collapse risk;

[0064] If , it is determined that the rock to be analyzed currently does not have a collapse risk, and observation should be continued, is a preset damage threshold.

[0065] The application further provides a rock progressive failure system based on stress and acoustic wave change joint representation, which is used for executing the rock progressive failure method based on stress and acoustic wave change joint representation.

[0066] A data acquisition module is configured to apply an incremental constant clamping pressure to a plurality of rock test blocks of the same type as the rock to be analyzed, and obtain strain response, acoustic wave response and surface crack morphology response data of the rock test blocks, obtain rock peak strength of the rock test blocks under the influence of different constant clamping pressures through a yield test, take the strain response and the acoustic wave response as the indication response data, and take the surface crack morphology response and the rock peak strength as the damage response data.

[0067] An initial weight determination module is configured to divide each rock test block into a corresponding damage stage based on the strain response and the acoustic wave response data, analyze the correlation between the indication response data and the damage response data in each damage stage, and determine the initial training weight of the indication response data in each damage stage.

[0068] A dynamic weight analysis module is configured to establish a multilayer perceptron model, take the preprocessed indication response data as input, take the damage response data as a label, correct the corresponding initial training weight through real-time response credibility correction of the indication response data, obtain a dynamic weight, and train the multilayer perceptron model according to the dynamic weight of each group of response data.

[0069] A damage condition representation module is configured to obtain the indication response data of the rock to be analyzed in a detection time period, input the multilayer perceptron model, obtain a damage response data prediction value at different time points in the detection time period, and calculate and generate an internal damage coefficient and an external indication coefficient according to the damage response data prediction value.

[0070] A risk warning module is configured to comprehensively analyze the internal damage coefficient and the external indication coefficient, and jointly represent the progressive damage degree of the rock to be analyzed.

[0071] Compared with the prior art, the application has the following beneficial effects:

[0072] The application divides the entire test process into a plurality of damage stages, and sets different initial training weights according to the correlation between the indication response data in different stages and the surface crack morphology and the rock peak strength. By dynamically updating these weights, the relationship between the acoustic wave and the mechanical response in different states can be better reflected. This dynamic weight adjustment mechanism improves the sensitivity of the model to the progressive damage characteristics, so that the model can adapt to the complex rock failure process and reduce the prediction deviation that may be caused by the static weight in the traditional method.

[0073] The multilayer perceptron model uses the pretreated response data as input to predict the crack morphology and rock peak strength, which can extract deep features from a large amount of data and form a multi-dimensional understanding of rock failure. Through machine learning, the model can not only provide prediction values of crack morphology and peak strength, but also promote the calculation of internal damage coefficient and external indicator coefficient, comprehensively reflect the damage characteristics inside and outside the rock, and realize the comprehensive characterization of the progressive damage degree of the rock. BRIEF DESCRIPTION OF DRAWINGS

[0074] Figure 1 It is a whole method flowchart of the present application;

[0075] Figure 2 It is a dynamic weight fitting curve chart of receiving clamping pressure-strain response;

[0076] Figure 3 It is a dynamic weight adjustment statistical histogram;

[0077] Figure 4 It is a dynamic weight training model accuracy comparison chart;

[0078] Figure 5 It is a whole system structure schematic diagram of the present application. DETAILED DESCRIPTION

[0079] In order to make the purpose, technical scheme and advantages of the present application clearer and more apparent, the present application will be further described in detail below in combination with specific embodiments.

[0080] It should be noted that, unless otherwise defined, the technical terms or scientific terms used in the present application should be understood as the usual meaning understood by those skilled in the art to which the present application belongs. The "first", "second" and similar words used in the present application do not represent any order, quantity or importance, but are only used to distinguish different components. "Include" or "contain" and similar words mean that the elements or objects before the word cover the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connected" or "connected" and similar words are not limited to physical or mechanical connection, but can include electrical connection, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to represent relative positional relationship, when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0081] EMBODIMENT:

[0082] Please refer to Figures 1-4 The present application provides a technical scheme:

[0083] A method for representing progressive failure of rock based on stress and acoustic wave changes, the specific steps comprising:

[0084] Step 1: Apply an increasing constant clamping pressure to several rock test blocks of the same type as the rock to be analyzed, obtain their strain response, acoustic response and surface crack morphology response data, obtain the rock peak strength of the rock test block under the influence of different constant clamping pressures through the yield test, take the strain response and acoustic response as the indication response data, and take the surface crack morphology response and rock peak strength as the damage response data.

[0085] A plurality of monitoring points arranged with strain detection units are arranged on the surface of the rock test block, the strain values at each monitoring point on the surface of the rock test block are collected by the strain detection units, the values after the strain values are stable are taken as the final strain values, and the acoustic response data and surface crack morphology response data are obtained after the strain values are stable;

[0086] A plurality of monitoring points are uniformly or selectively arranged on the surface of the rock test block, each monitoring point is equipped with a strain detection unit, the number of monitoring points is determined according to the size of the rock test block and experimental requirements, and the monitoring range needs to cover the entire test block area, each monitoring point is equipped with a strain detection unit such as a strain gauge or other strain sensing device for monitoring local strain changes on the rock surface, and the strain detection unit transmits the strain value to the data acquisition system through an electrical signal or other means, during the loading process, the strain value usually changes with time and gradually tends to be stable, until the strain value tends to be stable and is recorded, a judgment criterion is set to determine that the strain value reaches a stable state, within a certain time window, the change amplitude of the strain value is lower than a threshold value, which is specifically expressed as:

[0087] ;

[0088] Wherein, and are the strain values at two consecutive time points, is the change amplitude threshold value;

[0089] The acoustic response data is specifically the longitudinal wave velocity data through the rock test block, and the surface crack morphology response data includes crack length and crack width data;

[0090] After the strain value is stable, the rock test block surface is irradiated with sound waves by a sound wave detection device, and the response data of the rock surface to the sound waves is recorded, the longitudinal wave velocity information of the acoustic response data can reflect the change of the physical properties inside the rock, and the acoustic response measurement process includes: sound wave emission, selection of sound wave frequency and emission direction, emission of sound waves to the rock surface through a sensor; sound wave receiving, arranging a sound wave receiver on the rock surface to record the changes in the sound wave transmission process, and data acquisition, converting the received sound wave signal into analyzable velocity data.

[0091] The specific method for obtaining the surface crack morphology response data is: acquiring the surface image of the rock test block through the CCD camera system, detecting the crack edge and quantifying the length based on the surface image of the rock test block, and obtaining the surface crack shape response data.

[0092] After the strain value stabilizes, the crack morphology of the rock surface is captured through image processing technology or other crack detection equipment, the surface image of the rock test block is collected using a high-precision camera, and the cracks in the image are identified and analyzed using computer vision algorithms to extract the geometric characteristics of the cracks, including width and length.

[0093] The specific method for obtaining the rock peak strength of the rock test block under the influence of different constant clamping pressures through the yield test is to gradually apply axial load, continuously monitor the strain response of the test block, record the stress-strain curve, identify the stress value when the stress reaches the peak on the stress-strain curve, and record it as the rock peak strength under the constant clamping pressure.

[0094] Step 2: Based on the strain response and acoustic response data, each rock test block is divided into the corresponding damage stage, and the correlation between the indicator response data and the damage response data in each damage stage is analyzed to determine the initial training weight of the indicator response data in each damage stage.

[0095] According to the strain response and acoustic response, the progressive damage process of the rock is divided into several damage stages, wherein the entire progressive damage process of the rock is divided into an elastic deformation stage, a yield fracture stage and a macroscopic damage stage by means of longitudinal wave speed and strain value, wherein under the application of different constant clamping pressures, the change of the corresponding wave speed is detected to calculate the longitudinal wave speed change factor, and the formula for calculating the longitudinal wave speed change factor is:

[0096] ;

[0097] In the formula, is the longitudinal wave speed change factor of the i-th rock test block, is the longitudinal wave speed of the i-th rock test block, is the longitudinal wave speed of the i+1-th rock test block, is the constant clamping pressure applied to the i-th rock test block, is the constant clamping pressure applied to the i+1-th rock test block, wherein i is the index of the rock test block, and each rock test block is sequentially indexed in ascending order of constant clamping force, , is the total number of rock test blocks;

[0098] It should be noted that the corresponding acoustic velocity characteristics of each stage can help researchers understand the mechanical behavior of the rock during the loading process. The elastic deformation stage is usually characterized by relatively stable acoustic velocity, the yield and fracture stage is characterized by rapid change of acoustic velocity, and the macroscopic failure stage is characterized by significant decrease of acoustic velocity, indicating that the strength of the rock has been destroyed.

[0099] The longitudinal wave velocity change factor of the i-th rock test block The longitudinal wave velocity change factor of the i-th rock test block The greater the value, the faster the longitudinal wave velocity changes, reflecting the characteristics of different damage stages.

[0100] The numerator of the formula represents the change in longitudinal wave velocity of the i-th test block and the i-th test block, reflecting the change in the internal state of the rock after applying different constant clamping pressures. The denominator represents the change in clamping pressure, and by dividing the acoustic velocity change by the pressure change, the influence of different clamping pressure changes on acoustic velocity change can be determined, thereby determining the amplitude of acoustic velocity change under different clamping pressures.

[0101] The strain concentration factor is calculated by the strain value of the strain detection unit, and the strain concentration factor is used to characterize the strain concentration degree and damage condition of the test piece. The formula for calculating the strain concentration factor is as follows:

[0102]

[0103] In the formula, is the strain concentration factor of the i-th rock test block, is the difference between the average strain value of the selected monitoring point in the i-th rock test block and the average strain value of all monitoring points, is the maximum strain value of the i-th rock test block.

[0104] It should be noted that The strain concentration factor is used to characterize the strain concentration degree of different regions of the rock test block, and thus reflects the internal damage of the rock. Under the action of external force, the rock usually produces stress and strain concentration phenomenon, especially at the crack tip or weak area. The strain concentration factor is a quantitative index that can describe the strain concentration degree and damage condition of the local area of the rock, The size of the strain concentration factor represents the strain concentration degree of the local area of the test block. The greater the value, the closer the local strain value to the maximum value, indicating that the stress concentration in this area is more serious and the damage risk is higher.

[0105] ​​By analyzing the local strain deviation at selected monitoring points and comparing it with the global strain and maximum strain value, the degree of strain concentration in local areas on the surface of the rock specimen is quantified.

[0106] Maximum strain value These are extreme values ​​in the strain distribution of the rock specimen, typically reflecting the area of ​​most severe damage. Local strain deviations are normalized. With maximum strain value In comparison, the strain concentration factor has a clear physical meaning.

[0107] Where calculation The formula used is:

[0108] ;

[0109] In the formula, Let g be the strain value at the g-th selected monitoring point on the i-th rock specimen. Let g be the strain value at the h-th monitoring point on the i-th rock specimen, n be the total number of selected monitoring points in the rock specimen, N be the total number of all monitoring points on the surface of the rock specimen, where g is the index of the selected monitoring point on the rock specimen, and h is the index of the monitoring point on the rock specimen.

[0110] It should be noted that the strain distribution on a rock surface after being subjected to stress is usually non-uniform. This can be addressed by comparing the average strain values ​​at locally selected monitoring points. Average value of overall surface strain This can determine whether strain is concentrated in a local area. If the difference between the two is large, it means that the strain in the local area is significantly higher than the overall strain, that is, there is a stress concentration phenomenon.

[0111] The specific logic for selecting monitoring points is as follows: based on the strain values ​​of each surface monitoring point, they are arranged in descending order to form a strain sequence; and according to the strain sequence, a preset proportion of monitoring point units are selected in sequence.

[0112] The logic behind dividing the entire experimental process into several failure stages is as follows: setting the condition range for each failure stage, based on the longitudinal wave velocity variation factor of each rock specimen. , rate of change of strain concentration factor Based on the condition range in which the longitudinal wave sound velocity falls, the rock specimen is classified into the corresponding failure stage;

[0113] The formula used to calculate the rate of change of strain concentration factor is:

[0114] ;

[0115] In the formula, a strain concentration factor of the i-th rock test block, a strain concentration factor of the i+1-th rock test block;

[0116] wherein the condition interval of the elastic deformation stage is:

[0117] ;

[0118] wherein, a sound wave change threshold value of the elastic deformation stage, a strain concentration change threshold value of the elastic deformation stage, and are minimum and maximum reference values of the longitudinal wave sound velocity of the elastic deformation stage, respectively;

[0119] wherein the condition interval of the yield fracture stage is:

[0120] ;

[0121] wherein, a maximum reference value of the longitudinal wave sound velocity of the yield fracture stage;

[0122] wherein the condition interval of the macroscopic damage stage is:

[0123] ;

[0124] judging the condition interval met by each rock test block, and dividing the rock test block into the corresponding damage stage.

[0125] In the elastic deformation stage, the internal structure of the rock is basically complete, the state of the fissure and the pore does not change obviously, the longitudinal wave sound velocity is mainly determined by the elastic modulus and the density of the rock matrix, and these physical properties do not change in the elastic deformation stage, so the sound velocity is in a slow rising state, and since the internal and surface have not yet cracked and damaged, the strain distribution is not uniform, so the strain concentration factor changes slowly.

[0126] In the yield fracture stage, although the local area has crack propagation, the overall material may have stress redistribution phenomenon, at this time, the stress reduction of some areas will lead to the increase of the elastic modulus, so that the sound velocity rises, and in this stage, the internal crack of the test piece begins to expand, the crack expands obviously, and the strain is further concentrated.

[0127] When the load reaches the ultimate strength of the rock, the local crack propagates through to form a macroscopic crack, causing large-scale damage inside the rock, high development of cracks inside the rock, great reflection and scattering effect of the crack surface, and complete interruption of the wave propagation path. In the macroscopic damage stage, the sound wave propagation is significantly attenuated by the cracks and pores, resulting in a large decrease in sound velocity. Moreover, due to the crack penetration, the specimen is broken, and the strain concentration condition gradually becomes blurred.

[0128] The correlation between the indicator response data and the damage response data is characterized by the Spearman rank correlation coefficient, and the logic is as follows:

[0129] 1) In each damage stage, the indicator response data and the damage response data of all rock specimens in the damage stage are counted to form a longitudinal wave velocity sequence, a strain sequence, a rock peak strength sequence, a crack length sequence, and a crack width sequence, and the elements in each sequence are sorted in ascending order to determine the ranking of each rock specimen in each sequence in the damage stage.

[0130] 2) Based on the ranking of each element value in the sequence, the Spearman rank correlation coefficient is calculated, and the specific formula is as follows:

[0131] ;

[0132] In the formula, is the rank correlation coefficient of the qth indicator response data to the Yth damage response data in the kth damage stage, is the ranking difference of the qth indicator response data sequence and the Yth damage response data sequence for the u th rock specimen in the kth damage stage, is the number of rock specimens in the kth damage stage, where u is the index of the rock specimen in the damage stage, is the index of the indicator response data, and k is the index of the damage stage, is the index of the damage response data;

[0133] It should be noted that the formula evaluates the correlation by ranking the data and then calculating the difference between the rankings. First, the values of the indicator response data and the Yth damage response data are ranked, which means that all the values are arranged in ascending order, and each value is assigned a rank. The difference between the ranking of each indicator response data and the ranking of the Yth damage response data is calculated. The Spearman rank correlation coefficient is a non-parametric statistical indicator, which means that it does not require the data to meet the assumption of normal distribution. Therefore, it is suitable for processing non-normal distribution or ordinal data. In the study of rock and crack morphology, the data is often scattered and does not follow a normal distribution. Therefore, the Spearman correlation coefficient is more appropriate.

[0134] Using the sum of squares of rank differences The correlation can be calculated by using the sum of squares of rank differences, which can effectively reduce the influence of outliers on the results, and by calculating the Spearman rank correlation coefficient, the correlation between different indicator response data and fracture morphology or rock peak strength can be effectively evaluated, and a complex nonlinear relationship can be revealed.

[0135] The logic for setting the initial training weight of different indicator response data is as follows: the initial training weight of each indicator response data is determined according to the rank correlation coefficient, and the formula for calculating the initial training weight is as follows:

[0136] ;

[0137] In the formula, is the initial training weight of the qth indicator response data to the Yth damage response data in the kth damage stage, is the total number of types of indicator response data.

[0138] Step 3: Establish a multilayer perceptron model, use the preprocessed indicator response data as input, use the damage response data as label, and modify the corresponding initial training weight through the real-time response credibility of the indicator response data to obtain a dynamic weight, and train the multilayer perceptron model according to the dynamic weight of each group of response data.

[0139] The multilayer perceptron model is divided into several submodels based on the damage stage, each submodel is established based on the LightGBM model, the activation function is ReLU, and the loss function adopts the mean absolute error, wherein the indicator response data and the corresponding applied clamping pressure are used as the input of each submodel, and the corresponding surface crack morphology data and rock peak strength are used as the label, and the attention mechanism is constructed through the corresponding weight, and each submodel is trained.

[0140] The core idea of the attention mechanism is to dynamically adjust the weight according to the importance of the input features, so that the model can pay more attention to the features that have a significant impact on the target prediction when making predictions. In deep learning, the attention mechanism is usually realized by calculating a series of weight coefficients, which determine the contribution of each input feature to the output.

[0141] The specific training process includes: taking the preprocessed indicator response data as input features, taking the P-wave velocity and the applied clamping pressure dynamic weight as attention weights to weight the input features, inputting the weighted features into the corresponding submodel, and outputting the label data through the output layer for training. The above process is repeated for each submodel, that is, the attention weight is calculated and applied for each indicator response data, so as to focus on the most important features for predicting crack morphology and rock peak strength, and improve the accuracy and reliability of crack morphology and rock strength prediction.

[0142] Meanwhile, the attention weight is adjusted only during the training process to improve the accuracy and reliability of the fracture morphology and rock strength prediction and to improve the model training convergence speed.

[0143] The logic for modifying the initial training weight according to the response reliability of the response data is as follows: the initial training weight is modified based on the signal error of the acoustic response and the strain response, specifically including modification based on the average error of the signals in a single training round, wherein the acoustic response is modified by the signal-to-noise ratio of the acoustic signal, and the formula according to which the modification is specifically made is:

[0144] ;

[0145] wherein, is the dynamic weight of the acoustic response data to the Yth damage response data in the kth damage stage, is the signal-to-noise ratio of the zth training sample, is the initial training weight of the acoustic response data to the Yth damage response data in the kth damage stage, and are the minimum and maximum signal-to-noise ratios of the calibration, respectively, wherein z is the index of the training sample in a single training round, , is the batch size of a single training round;

[0146] It should be noted that, is used to represent the dynamic weight of the acoustic response data in a single training round, wherein The greater the value of, the higher the reliability of the acoustic response data in the training round, and the higher the corresponding weight, so as to improve the prediction accuracy of the model;

[0147] signal-to-noise ratio is an important indicator of signal quality, reflecting the ratio of signal strength to background noise. High signal-to-noise ratio indicates good signal quality, and low signal-to-noise ratio indicates that the signal is disturbed by a lot of noise. By introducing the difference and the standardization expression, the reliability of the signal can be quantified and reflected in the weight, and the signal-to-noise ratio is normalized using to ensure consistency in the comparison of different signals. This standardization process enables the signal error under different conditions to be comparable, and in the formula, the average value of is used to calculate, indicating the total influence of all signals in a single training, and by averaging multiple signals, the influence of individual signal noise on weight correction can be reduced, enhancing overall stability.

[0148] By adjusting the signal-to-noise ratio (SNR), the model's adaptability to different signal qualities can be improved, ensuring that the model can learn effective information better during training and improving prediction accuracy. The properties and responses of rock materials may change at different stages of damage; by dynamically adjusting the weights, these changes can be adapted to in a timely manner, improving the model's learning efficiency. The introduction of SNR reduces over-reliance on low-quality signals, avoiding learning from noise during training and thus reducing the risk of overfitting.

[0149] The specific formula used to correct the initial training weights of the strain response is as follows:

[0150] ;

[0151] In the formula, This refers to the dynamic weights of the strain response data and the Y-th damage response data after correction during the k-th failure stage. The initial training weights for the strain response data and the Y-th damage response data within the k-th damage stage. Let be the actual clamping pressure of the z-th training sample, specifically representing the clamping pressure received by the clamping pressure sensor. Set the clamping pressure value for the z-th training sample; the clamping pressure sensor is specifically set on the clamped surface of the rock sample to collect the actual clamping pressure.

[0152] It should be noted that the corrected dynamic weights Based on initial weights The adjustment is based on the received clamping pressure response data. With set value The difference reflects the deviation between the actual applied pressure and the target pressure. (Calculation) This reflects the absolute difference between each response data point and the set pressure. This difference is used to evaluate the stability and reliability of the strain response data. The formula uses... Standardization makes the magnitude of the deviation more comparable relative to the set pressure.

[0153] The applied clamping pressure is a crucial parameter that directly affects the strain response data of rock materials. By adjusting the weights, the model can better reflect the changes in the actual applied pressure, which helps improve the accuracy of the model's prediction of strain response data. At different failure stages, the effect of clamping pressure may vary due to changes in the material state. Adjusting the dynamic weights allows the model to adapt to these changes in a timely manner, improving the model's flexibility and adaptability. By correcting the deviation between the applied clamping pressure and the set value, it can be ensured that the model pays attention to the influence of the set pressure during the learning process, reducing the uncertainty caused by errors.

[0154] The sub-model is composed of an input layer, a convolution layer, a pooling layer, a full connection layer, and an output layer; for the full connection layer, the number of full connection layer neurons is set to 32, the initial neural network learning rate is set to 0.001, and the training round number is 100. The mean absolute error is used as the loss function, and when the model damage function value is less than the target damage precision, the model training is completed.

[0155] Step 4: Obtain the indication response data of the rock to be analyzed within the detection time period, input the multilayer perceptron model, obtain the damage response data prediction value at different times within the detection time period, and calculate the internal damage coefficient and external indication coefficient according to the damage response data prediction value.

[0156] The formula according to which the internal damage coefficient is calculated is:

[0157] ;

[0158] In the formula, is the internal damage coefficient at time t, is the wave impedance damage at time t, which is calculated by the P-wave velocity at time t combined with the rock test density, is the rock peak strength prediction value at time t, and are proportional constants, is the intercept constant, and t is the time variable within the detection time period;

[0159] It should be noted that the internal damage coefficient represents the damage degree of the rock material at the internal microstructure level, such as pores and cracks. Under the action of external force, the internal microstructure of the rock will change, including crack initiation and propagation. This damage will affect the overall mechanical properties of the material. The greater the internal damage coefficient, the more serious the internal damage of the rock.

[0160] The wave impedance is the product of the P-wave velocity and the rock density, and is usually used to represent the change in the mechanical properties of the rock. When the rock is damaged, the wave impedance will decrease due to the generation of internal cracks and pores. The change in wave impedance is an important internal feature of the degree of rock damage, reflecting the physical degradation of the material at the micro level. Therefore, the internal damage coefficient is calculated by the change in wave impedance.

[0161] The peak intensity refers to the maximum stress that the rock can withstand during the loading process, and is an important indicator of the rock's resistance to damage. As internal damage accumulates, the peak intensity of the rock decreases. This decrease reflects the decline in the rock's carrying capacity due to the expansion of micro-cracks and damage, and therefore, the peak intensity prediction value, combined with the wave impedance, can provide a more comprehensive description of the internal damage of the rock: the wave impedance is used to describe the changes in the microstructure, and the peak intensity reflects the impact of these changes on the macroscopic mechanical properties. The weighted average form The model allows the contribution ratio of wave impedance and peak intensity to be adjusted according to actual needs, and the square root is used to limit the growth trend of the damage coefficient while enhancing the sensitivity to small changes.

[0162] The formula used to calculate the external indicator coefficient is as follows:

[0163] ;

[0164] In the formula, is the external indicator coefficient at time t, is the surface crack area at time t, which is calculated based on the crack length and width prediction values, is a proportional constant, is an intercept constant;

[0165] It is worth noting that the external indicator coefficient represents the degree of surface cracks or visible damage of the rock material. It reflects the expansion of surface macro-cracks of the rock after being subjected to load or environmental influence, as well as the impact of surface damage on the overall performance of the rock. The external indicator coefficient is mainly related to the geometric characteristics of the rock surface cracks, and the crack area directly reflects the degree of damage to the rock surface. The larger the crack, the more serious the external damage to the rock. Cracks are the most significant representation of rock instability and destruction, and the degree of external damage can be quantified through the crack area.

[0166] Before the appearance of external cracks, the peak intensity of the rock will begin to decrease, and the expansion of external cracks is often synchronized with the decline in the carrying capacity of the rock. Therefore, the peak intensity can provide a link between internal and external damage.

[0167] The square root operation makes the change of the damage coefficient more smooth, and the intercept adjusts the baseline level of the coefficient, ensuring that the damage coefficient still has a reasonable baseline value when there is no damage.

[0168] Step 5: Comprehensive analysis of the internal damage coefficient and the external indicator coefficient to jointly represent the degree of progressive damage of the rock being analyzed.

[0169] The logic for jointly representing the progressive damage degree of the rock to be analyzed is as follows: a damage risk coefficient is generated based on the comprehensive change rates of the external indication coefficient and the internal damage coefficient, the progressive damage degree of the rock to be analyzed is represented based on the damage risk coefficient, and the formula for calculating the damage risk coefficient is as follows:

[0170] ;

[0171] In the formula, is the damage risk coefficient of the rock to be analyzed, and are the variances of the internal damage coefficient and the external indication coefficient of the rock to be analyzed within the detection time period, and are the mean values of the internal damage coefficient and the external indication coefficient of the rock to be analyzed within the detection time period, wherein and are weight coefficients, wherein , and and are both greater than 0;

[0172] It should be noted that the calculation of the damage risk coefficient comprehensively represents the change rates of the internal damage coefficient and the external indication coefficient , aiming to evaluate the progressive damage degree of the rock, wherein the greater the value of , the greater the damage fluctuation amplitude of the rock to be analyzed within the detection time period, the more unstable the mechanical characteristics of the rock to be analyzed, and the greater the collapse risk.

[0173] The variances ( and ) are used to reflect the change degree of damage and external influence, and the greater the variance, the more significant the fluctuation of the index, which may mean that the damage risk of the rock is also increasing, therefore, the variance effectively captures the progressive damage process of the rock as a dynamic index.

[0174] Standardization by dividing the variances by the mean values and can eliminate the influence of different units and magnitudes, so that different indexes can be compared on the same scale. This standardization enables the damage risk coefficient to more accurately reflect the relative risk level.

[0175] The external indication coefficient of the rock is usually more intuitive and directly affects the stability of the rock than the internal damage coefficient. External damage is often a direct factor causing collapse risk, therefore, the value of is set to be greater than or equal to is meaningful, in contrast, internal damage usually needs to be indirectly evaluated by the result of external manifestation, although the importance of internal damage cannot be ignored, but its influence is relatively small in the actual collapse risk assessment, so the influence of internal damage is set less than or equal to .

[0176] If , it is judged that the rock to be analyzed has a collapse risk;

[0177] If , it is judged that the rock to be analyzed does not have a collapse risk at present, and observation should be continued, is a preset damage threshold, which can be set according to increasing experience.

[0178] Please refer to Figure 5 , the application also provides a system for jointly representing progressive damage of rock based on stress and acoustic wave change, which is used to execute the above-mentioned method for jointly representing progressive damage of rock based on stress and acoustic wave change, and comprises:

[0179] A data acquisition module is configured to apply an increasing constant clamping pressure to a plurality of rock test blocks of the same type as the rock to be analyzed, obtain strain response, acoustic wave response and surface crack morphology response data of the rock test blocks, obtain rock peak strength of the rock test blocks under the influence of different constant clamping pressures through a yield test, take the strain response and the acoustic wave response as indication response data, and take the surface crack morphology response and the rock peak strength as damage response data.

[0180] An initial weight determination module is configured to divide each rock test block into a corresponding damage stage based on the strain response and acoustic wave response data, analyze the correlation between the indication response data and the damage response data in each damage stage, and determine initial training weights of the indication response data in each damage stage.

[0181] A dynamic weight analysis module is configured to establish a multilayer perceptron model, take the preprocessed indication response data as input, take the damage response data as label, correct the corresponding initial training weights through real-time response credibility correction of the indication response data, obtain dynamic weights, and train the multilayer perceptron model according to the dynamic weights of each group of response data.

[0182] A damage condition representation module is configured to obtain indication response data of the rock to be analyzed within a detection time period, input the multilayer perceptron model, obtain damage response data prediction values at different time points within the detection time period, and calculate internal damage coefficients and external indication coefficients according to the damage response data prediction values.

[0183] The risk warning module is used for comprehensively analyzing the internal damage coefficient and the external indication coefficient to jointly represent the progressive damage degree of the rock to be analyzed.

[0184] The above formulas are dimensionless values calculated, the formulas are obtained by collecting a large amount of data to simulate a formula of the most recent real situation, and preset parameters in the formulas are set by a person skilled in the art according to actual conditions.

[0185] The above embodiments can be realized wholly or partially by software, hardware, firmware or any other combination. When realized by software, the above embodiments can be realized wholly or partially in the form of a computer program product. Those skilled in the art can realize that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized by hardware or software methods depends on the specific application and design constraints of the technical solutions.

[0186] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, which can be located in one place or distributed on multiple network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiments.

[0187] The above is merely specific embodiments of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application.

Claims

1. A method for characterizing progressive rock failure based on a combination of stress and acoustic wave variations, characterized in that, The specific steps include: A constant clamping pressure of increasing increments was applied to several rock specimens of the same type as the rock to be analyzed, and their strain response, acoustic response and surface crack morphology response data were obtained. The rock peak strength of the rock specimens under different constant clamping pressures was obtained through yield test. The strain response and acoustic response were used as indicator response data, and the surface crack morphology response and rock peak strength were used as damage response data. Based on strain response and acoustic response data, each rock specimen was divided into the corresponding failure stage. The correlation between indicator response data and damage response data in each failure stage was analyzed to determine the initial training weights of indicator response data in each failure stage. A multilayer perceptron model is established, with preprocessed indication response data as input and damage response data as labels. The initial training weights are corrected by the real-time response confidence of the indication response data to obtain dynamic weights. The multilayer perceptron model is trained based on the dynamic weights of each group of response data. The indicator response data of the rock to be analyzed during the detection period is obtained and input into the multilayer perceptron model to obtain the damage response data prediction values ​​at different times during the detection period. Based on the damage response data prediction values, the internal damage coefficient and external indicator coefficient are calculated and generated. By comprehensively analyzing the internal damage coefficient and the external indicator coefficient, the degree of progressive failure of the rock under analysis is jointly characterized. The logic behind using Spearman's rank correlation coefficient to characterize the correlation between indication response data and damage response data is as follows: 1) At each failure stage, statistical analysis is performed on the indicator response data and damage response data of all rock specimens located within that failure stage to form a longitudinal wave velocity sequence, strain sequence, rock peak strength sequence, crack length and crack width sequence. The elements in each sequence are sorted in ascending order of their values ​​to determine the ranking of each rock specimen in each sequence at that failure stage. 2) Calculate the Spearman rank correlation coefficient based on the ranking of each element value within the sequence. The specific formula used is as follows: In the formula, Let be the rank correlation coefficient between the q-th indicator response data and the Y-th damage response data within the k-th damage stage. For the u-th rock specimen within the k-th failure stage, this represents the ranking difference between the q-th indicator response data sequence and the Y-th damage response data sequence. Let u be the number of rock samples located in the k-th failure stage, where u is the index of the rock sample in the failure stage. The index indicates the type of response data, and k is the index for the destruction phase. An index for the types of damage response data; The logic behind setting the initial training weights for different indicator response data is as follows: The initial training weights for each indicator response data are determined based on the rank correlation coefficient, and the specific formula for calculating the initial training weights is as follows: In the formula, Let be the initial training weights for the q-th indicator response data and the Y-th damage response data during the k-th damage stage. This indicates the total number of types of response data.

2. The method for characterizing progressive rock failure based on a combination of stress and acoustic wave changes according to claim 1, characterized in that: The strain response data is specifically the strain value. The specific acquisition method is as follows: several monitoring points with strain detection units are set on the surface of the rock sample block. The strain value at each monitoring point on the surface of the rock sample block is collected through the strain detection unit. The value after the strain value stabilizes is taken as the final strain value. After the strain value stabilizes, the acoustic response data and surface crack morphology response data are acquired. The acoustic response data specifically refers to the longitudinal wave velocity data passing through the rock sample, and the surface crack morphology response data includes crack length and crack width data. The specific method for obtaining surface crack morphology response data is as follows: acquire surface images of rock specimens using a CCD camera system, and perform crack edge detection and length quantization based on the surface images of the rock specimens to obtain surface crack morphology response data.

3. The method for characterizing progressive rock failure based on a combination of stress and acoustic wave changes according to claim 2, characterized in that: Based on strain response and acoustic response, the progressive failure process of rock is divided into several failure stages. Specifically, the entire progressive failure process is divided into elastic deformation stage, yielding and fracturing stage, and macroscopic failure stage by using longitudinal wave velocity and strain value. Under different constant clamping pressures, the changes in corresponding wave velocity are detected to calculate the longitudinal wave velocity variation factor. The specific formula used to calculate the longitudinal wave velocity variation factor is as follows: In the formula, Let be the longitudinal wave velocity variation factor for the i-th rock sample. Let be the longitudinal wave velocity of the i-th rock sample. Let be the longitudinal wave velocity of the (i+1)th rock sample. To provide a constant clamping pressure applied to the i-th rock specimen, Let be the constant clamping pressure applied to the (i+1)th rock specimen, where i is the index of the rock specimen, and the rock specimens are indexed sequentially in ascending order of constant clamping force. , The total number of rock test blocks; The strain concentration factor is calculated using the strain values ​​from the strain detection unit. The strain concentration factor characterizes the degree of strain concentration and damage in the specimen. The specific formula used to calculate the strain concentration factor is as follows: In the formula, Let be the strain concentration factor of the i-th rock specimen. Let be the difference between the mean strain value at the selected monitoring point in the i-th rock specimen and the mean strain value at all monitoring points. Let be the maximum strain value of the i-th rock specimen, where is calculated. The formula used is: In the formula, Let g be the strain value at the g-th selected monitoring point on the i-th rock specimen. Let g be the strain value at the h-th monitoring point on the i-th rock specimen, n be the total number of selected monitoring points in the rock specimen, N be the total number of all monitoring points on the surface of the rock specimen, where g is the index of the selected monitoring point on the rock specimen, and h is the index of the monitoring point on the rock specimen.

4. The method for characterizing progressive rock failure based on a combination of stress and acoustic wave changes according to claim 3, characterized in that: The specific logic for selecting monitoring points is as follows: based on the strain values ​​of each surface monitoring point, they are arranged in descending order to form a strain sequence; and according to the strain sequence, a preset proportion of monitoring points are selected in sequence. The logic behind dividing the entire experimental process into several failure stages is as follows: setting the condition range for each failure stage, based on the longitudinal wave velocity variation factor of each rock specimen. , rate of change of strain concentration factor Based on the condition range in which the longitudinal wave sound velocity falls, the rock specimen is classified into the corresponding failure stage; The formula used to calculate the rate of change of strain concentration factor is: In the formula, Let be the strain concentration factor of the i-th rock specimen. Let be the strain concentration factor of the (i+1)th rock specimen; The condition interval for the elastic deformation stage is as follows: In the formula, The threshold for acoustic wave change during the elastic deformation stage. This represents the threshold for strain concentration during the elastic deformation stage. and These are the minimum and maximum reference values ​​for the longitudinal wave sound velocity during the elastic deformation stage, respectively. The condition interval for the yield fracture stage is as follows: In the formula, This is the maximum reference value for the longitudinal wave velocity during the yielding and fracturing stage; The condition range for the macroscopic destruction phase is as follows: Determine the condition range that each rock specimen meets, and classify the rock specimen into the corresponding failure stage.

5. The method for characterizing progressive rock failure based on a combination of stress and acoustic wave changes according to claim 4, characterized in that: The multilayer perceptron model is divided into several sub-models based on the damage stage. Each sub-model is built based on the LightGBM model, with ReLU as the activation function and mean absolute error as the loss function. The indicator response data is used as the input of each sub-model, and the corresponding surface crack morphology data and rock peak intensity are used as labels. An attention mechanism is constructed through corresponding weights to train each sub-model. The logic behind correcting the initial training weights based on the reliability of the indicated response data is as follows: The initial training weights are corrected based on the signal errors of the acoustic response and strain response. Specifically, this includes correction based on the average signal error in a single training round, where the acoustic response is corrected using the signal-to-noise ratio of the acoustic signal. The specific formula used for this correction is: In the formula, This refers to the dynamic weights of the acoustic response data and the Y-th damage response data after correction during the k-th damage stage. Let be the signal-to-noise ratio of the z-th training sample. The initial training weights for the acoustic response data and the Y-th damage response data within the k-th damage stage are: and These are the calibrated minimum and maximum signal-to-noise ratios, respectively, where z is the index of the training sample in a single training round. , This represents the batch size for a single training epoch. The specific formula used to correct the initial training weights of the strain response is as follows: In the formula, This refers to the dynamic weights of the strain response data and the Y-th damage response data after correction during the k-th failure stage. The initial training weights for the strain response data and the Y-th damage response data within the k-th damage stage. The actual clamping pressure of the z-th training sample. Set the clamping pressure value for the z-th training sample.

6. The method for characterizing progressive rock failure based on a combination of stress and acoustic wave changes according to claim 5, characterized in that: The specific formula used to calculate the internal damage coefficient is as follows: In the formula, Let be the internal damage coefficient at time t. The wave impedance damage at time t is calculated specifically by combining the longitudinal wave velocity at time t with the rock test density. The predicted peak rock strength at time t. and These are proportionality constants. The intercept constant is t, and t is the time variable within the detection period. The specific formula used to calculate the external indicator coefficient is as follows: In the formula, Let be the external indicator coefficient at time t. The surface crack area at time t is calculated based on the predicted crack length and width. It is a proportionality constant. The intercept constant; The logic behind the joint characterization of the progressive failure degree of the rock under analysis is as follows: a failure risk coefficient is generated based on the combined rate of change of the external indicator coefficient and the internal damage coefficient; the progressive failure degree of the rock under analysis is then characterized based on this failure risk coefficient; and the specific formula used to calculate the failure risk coefficient is as follows: In the formula, To determine the failure risk factor of the rock to be analyzed, and These represent the variances of the internal damage coefficient and the external indicator coefficient of the rock being analyzed during the testing period. and These represent the average internal damage coefficient and external indicator coefficient of the rock to be analyzed during the detection period, respectively. and Here are the weighting coefficients, where ,and and All are greater than 0; If The analysis determined that the rock to be analyzed was at risk of collapse. like The rock being analyzed is currently not at risk of collapse and should be continuously monitored. This is the preset damage threshold.

7. A system for characterizing progressive rock failure based on a combination of stress and acoustic wave variations, characterized in that: The system for characterizing progressive rock failure based on a combination of stress and acoustic wave changes is used to execute the method for characterizing progressive rock failure based on a combination of stress and acoustic wave changes as described in any one of claims 1-6, comprising: The data acquisition module is used to apply increasing constant clamping pressure to several rock specimens of the same type as the rock to be analyzed, and to obtain their strain response, acoustic response and surface crack morphology response data. The peak rock strength of the rock specimens under different constant clamping pressures is obtained through yield test. The strain response and acoustic response are used as indicator response data, and the surface crack morphology response and peak rock strength are used as damage response data. The initial weight determination module is used to classify each rock specimen into the corresponding failure stage based on strain response and acoustic response data, analyze the correlation between indicator response data and damage response data in each failure stage, and determine the initial training weight of indicator response data in each failure stage. The dynamic weight analysis module is used to build a multilayer perceptron model. It takes preprocessed indication response data as input, damage response data as labels, and corrects the corresponding initial training weights by the real-time response confidence of the indication response data to obtain dynamic weights. The multilayer perceptron model is trained according to the dynamic weights of each group of response data. The damage characterization module is used to acquire the indication response data of the rock to be analyzed during the detection period, input it into the multilayer perceptron model, obtain the predicted values ​​of the damage response data at different times during the detection period, and calculate and generate the internal damage coefficient and the external indication coefficient based on the predicted values ​​of the damage response data. The risk warning module is used to comprehensively analyze the internal damage coefficient and the external indicator coefficient to jointly characterize the degree of progressive damage of the rock under analysis.

Images