Laboratory comprehensive evaluation method for surrounding rock control effect of coal mine spraying supporting material

Abstract

The present application provides a kind of coal mine spraying supporting material surrounding rock control effect laboratory comprehensive evaluation method.The method comprises: obtaining the acoustic emission data and image data synchronously collected when the rock sample is subjected to mechanical test;Wherein, the surface of the rock sample comprises a sprayed surface and an untreated control surface, and the sprayed surface and the control surface are isolated from each other on the surface of the rock sample;Based on the acoustic emission data, calculate the acoustic emission evaluation index reflecting the internal fracture process;And based on the image data, calculate the kinetic evaluation index reflecting the behavior of the block ejection;The acoustic emission evaluation index and the kinetic evaluation index are input into the trained comprehensive evaluation model, and the comprehensive evaluation grade of the spraying supporting material surrounding rock control effect is output.The present application can effectively avoid the interference of rock sample dispersion, and can scientifically reveal the real supporting effect of the spraying material in a relatively simple and repeatable way.

Description

Technical Field

[0001] This invention relates to the field of coal mine roadway support technology, and in particular to a laboratory comprehensive evaluation method for the surrounding rock control effect of coal mine sprayed support materials. Background Technology

[0002] In underground engineering projects such as coal mine roadways and tunnels, the stability of the surrounding rock directly affects the safety of workers and production efficiency. Traditional support methods, such as anchor bolts, anchor cables, and metal supports, while widely used, often have limitations in dealing with complex geological conditions such as high ground pressure and weak, fractured rock, often resulting in untimely support. In recent years, spray-applied support technology has shown significant advantages as a new type of active support method. This technology involves spraying polymer materials onto the surface of the surrounding rock to quickly form a continuous, thin support shell that combines strength and toughness. This shell not only seals surface cracks and prevents rock degradation but also significantly enhances the overall load-bearing capacity through synergistic deformation with the surrounding rock.

[0003] However, the current evaluation system for the support effect of this material has significant shortcomings. Existing laboratory methods mainly focus on testing the physical and mechanical properties of the material itself (such as preparing standard test blocks). This method cannot truly reflect the mechanical response of the composite structure of the support layer and surrounding rock under actual stress. More importantly, natural rock, as the carrier of the surrounding rock, inevitably contains inherent defects such as fissures and joints due to its geological formation. This results in significant dispersion in key mechanical parameters such as uniaxial compressive strength and deformation modulus even for standard samples processed from the same rock layer and batch. This inherent heterogeneity problem poses a huge challenge to the scientific and accurate evaluation of the support effect of sprayed materials. Currently, to avoid this problem, researchers usually adopt the method of preparing a large number of samples for statistical averaging, but this undoubtedly increases the complexity, time, and economic cost of the experiment. Moreover, in terms of testing standards, directly borrowing the testing standards for homogeneous materials in the field of concrete or shotcrete completely ignores the core characteristic of the high heterogeneity of natural rock, leading to a serious disconnect between the evaluation results and engineering practice.

[0004] Therefore, there is an urgent need in this field for a laboratory testing method that can effectively avoid the interference of discrete rock samples and scientifically reveal the true support effect of sprayed materials in a relatively simple and repeatable manner. Summary of the Invention

[0005] This invention provides a comprehensive laboratory evaluation method for the surrounding rock control effect of sprayed support materials in coal mines, addressing the shortcomings of existing technologies where the use of homogeneous material testing standards due to the discrete interference of rock samples leads to a significant disconnect between evaluation results and engineering realities. This invention effectively avoids the discrete interference of rock samples and can scientifically reveal the true support effect of sprayed materials in a relatively simple and repeatable manner. The technical solution proposed by this invention is as follows: In a first aspect, the present invention provides a laboratory comprehensive evaluation method for the surrounding rock control effect of coal mine sprayed support materials, comprising: Acoustic emission data and image data are acquired simultaneously during mechanical testing of a rock sample; wherein the surface of the rock sample includes a coated surface and an untreated control surface; Based on the acoustic emission data, an acoustic emission evaluation index reflecting the internal fracture process is calculated; and based on the image data, a dynamic evaluation index reflecting the fragment ejection behavior is calculated. The acoustic emission evaluation index and the dynamic evaluation index are input into the trained comprehensive evaluation model, which outputs a comprehensive evaluation level of the rock control effect of the sprayed support material.

[0006] Optionally, the mechanical tests include uniaxial compression tests, split Hopkinson bar tests, and biaxial compression tests.

[0007] Optionally, the acoustic emission evaluation index includes a breakage suppression index based on the cumulative count of acoustic emission ringing; The breakage suppression index based on the cumulative count of acoustic emission ringing is determined in the following way: Based on the acoustic emission data, the cumulative count of acoustic emission ringing on the coated surface and the untreated control surface within a set depth range is extracted respectively; The fracture suppression index is determined based on the cumulative count of acoustic emission ringing on the coated surface and the cumulative count of acoustic emission ringing on the control surface.

[0008] Optionally, the dynamic evaluation index includes: average ejection velocity of fragments, momentum control efficiency index based on average ejection momentum of fragments, kinetic energy decay safety factor based on average kinetic energy of fragments, fragmentation degree coefficient based on fragment mass distribution, and performance dispersion improvement coefficient. The performance dispersion improvement coefficient is used to evaluate the effect of the sprayed material on improving the consistency of the mechanical properties of the rock sample; the performance dispersion improvement coefficient is determined in the following manner: The average kinetic energy data of fragments from multiple rock samples on the sprayed surface and the untreated control surface were obtained; Based on the average kinetic energy data of fragments from the multiple rock samples on the sprayed surface, the coefficient of variation and the mean kinetic energy of the sprayed surface are calculated; and based on the average kinetic energy data of fragments from the multiple rock samples on the untreated control surface, the coefficient of variation and the mean kinetic energy of the control surface are calculated. Based on the coefficient of variation of the sprayed surface, the coefficient of variation of the control surface, the mean kinetic energy of the sprayed surface, and the mean kinetic energy of the control surface, the performance dispersion improvement coefficient is determined.

[0009] Optionally, the comprehensive evaluation model is trained in the following manner: Obtain a training dataset, which contains multiple sets of sample data. Each set of sample data includes acoustic emission evaluation index and dynamic evaluation index calculated from the mechanical response data of sprayed and untreated samples, as well as the corresponding benchmark evaluation level label. The training dataset is augmented using a variational autoencoder model to generate augmented samples for model training, resulting in an augmented dataset. Using the acoustic emission evaluation index and dynamic evaluation index in the enhanced dataset as input features, and the corresponding benchmark evaluation level label as the training objective, the deep neural network model is trained by minimizing the composite loss function, and the trained deep neural network model is used as the comprehensive evaluation model. The composite loss function is determined based on the classification error loss and the physical consistency constraint loss. The classification error loss is determined based on the predicted probability distribution of the model output and the true benchmark rating label. The physical consistency constraint loss is determined based on the predicted score relationship of sample pairs in the training batch. The sample pairs are selected based on the physical performance relationship of the input features of the sample pairs, and the sample with better physical performance should have a predicted score no lower than that of the sample with worse physical performance.

[0010] Optionally, the physical consistency constraint loss is determined in the following manner: For all selected sample pairs, the hinge loss is calculated based on the prediction score relationship of each sample pair, and the physical consistency constraint loss is determined based on the hinge loss of all sample pairs. Specifically, for a sample pair consisting of a sample with better physical performance and a sample with worse physical performance, based on the input features, the prediction score of the sample with better physical performance and the prediction score of the sample with worse physical performance are calculated respectively. If the prediction score of the sample with better physical performance is less than or equal to the prediction score of the sample with worse physical performance, the hinge loss is calculated based on the prediction scores of the sample with better physical performance and the sample with worse physical performance. Otherwise, the hinge loss of the sample pair is zero.

[0011] Secondly, the present invention also provides a laboratory comprehensive evaluation device for the surrounding rock control effect of coal mine sprayed support materials, comprising the following modules: The data acquisition module is used to acquire acoustic emission data and image data simultaneously collected during mechanical testing of rock samples; wherein, the surface of the rock sample includes a coated surface and an untreated control surface; The index calculation module is used to calculate an acoustic emission evaluation index reflecting the internal fracture process based on the acoustic emission data; and to calculate a dynamic evaluation index reflecting the fragment ejection behavior based on the image data. The effect evaluation module is used to input the acoustic emission evaluation index and the dynamic evaluation index into the trained comprehensive evaluation model and output the comprehensive evaluation level of the control effect of the sprayed support material on the surrounding rock.

[0012] Thirdly, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and running on the processor, wherein the processor executes the computer program to implement a laboratory comprehensive evaluation method for the control effect of coal mine sprayed support materials on surrounding rock as described in the first aspect above.

[0013] Fourthly, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a laboratory comprehensive evaluation method for the control effect of coal mine sprayed support materials on surrounding rock as described in the first aspect above.

[0014] Fifthly, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements a laboratory comprehensive evaluation method for the control effect of coal mine sprayed support materials on surrounding rock as described in the first aspect above.

[0015] Based on the above technical solution, the beneficial effects of the present invention compared with the prior art are as follows: This invention provides a laboratory comprehensive evaluation method for the surrounding rock control effect of sprayed support materials in coal mines. During sample preparation, a sprayed support layer is applied to two adjacent sides of the same standard rock sample to form a sprayed surface, while the remaining two sides are left untouched as control surfaces. This design directly utilizes the self-comparison of the same sample, fundamentally eliminating the interference of mechanical parameter dispersion caused by the heterogeneity of different rock samples due to primary fissures, joints, etc., in a relatively simple and repeatable manner. This scientifically reveals the true support effect of the sprayed material, enabling the evaluation results to accurately reflect the true support effectiveness of the sprayed material. This method ensures that the experimental conditions closely resemble the local interaction scenario between the support layer and the surrounding rock in actual engineering, effectively avoiding the applicability deviation of homogeneous material standards. Simultaneously, multi-source data fusion technology is employed, namely, the collaborative monitoring of acoustic emission data and image data. During mechanical loading tests, acoustic emission data and image data are simultaneously acquired. Acoustic emission signals are used to capture the dynamic processes of internal fissure initiation, propagation, and frictional sliding in real time to reflect the microscopic damage evolution law, while image data is used to record the ejection trajectory of surface fragments and the macroscopic fracture morphology, thereby quantifying the overall stability changes. Abandoning the homogeneous assumption, this approach replaces theoretical derivations based on homogeneous materials by directly monitoring the actual fracture process and fragment dynamics, ensuring the evaluation results closely align with engineering realities. Finally, acoustic emission evaluation indicators and dynamic evaluation indicators extracted from imagery are input into a trained comprehensive evaluation model. Through multi-parameter fusion analysis, the model outputs a comprehensive support effectiveness level. Trained on a large-scale experimental dataset, this model not only boasts high accuracy but also overcomes the limitations of single indicators, comprehensively revealing the synergistic mechanism between support materials and rock mass.

[0016] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained through the structures particularly pointed out in the description and the drawings.

[0017] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the laboratory comprehensive evaluation method for the surrounding rock control effect of coal mine sprayed support materials provided by the present invention.

[0020] Figure 2 This is a technical roadmap for the laboratory comprehensive evaluation method of the surrounding rock control effect of coal mine sprayed support materials provided by the present invention.

[0021] Figure 3 This is a schematic diagram of the rock sample provided by the present invention.

[0022] Figure 4 This is a schematic diagram of the acoustic emission probe arrangement during uniaxial compression testing provided by the present invention.

[0023] Figure 5 This is a schematic diagram of the high-speed camera arrangement during single-axis compression testing provided by the present invention.

[0024] Figure 6 This is a schematic diagram of the split Hopkinson's experiment system provided by the present invention.

[0025] Figure 7 This is a schematic diagram of the high-speed camera arrangement in the split Hopkinson experiment provided by the present invention.

[0026] Figure 8 This is a schematic diagram of the high-speed camera arrangement during bidirectional compression testing provided by the present invention.

[0027] Figure 9 This is a flowchart of the comprehensive evaluation process provided by the present invention.

[0028] Figure 10 This is a schematic diagram of the laboratory comprehensive evaluation device for the surrounding rock control effect of coal mine sprayed support materials provided by the present invention.

[0029] Figure 11 This is a schematic diagram of the structure of the electronic device provided by the present invention. Detailed Implementation

[0030] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0031] The following describes in detail the laboratory comprehensive evaluation method for the surrounding rock control effect of coal mine sprayed support materials provided by the present invention, with reference to specific embodiments.

[0032] Reference Figure 1 As shown, the laboratory comprehensive evaluation method for the surrounding rock control effect of the sprayed support material in coal mine includes the following: S110. Acquire acoustic emission data and image data simultaneously collected during mechanical testing of the rock sample; wherein the surface of the rock sample includes a coated surface and an untreated control surface.

[0033] Reference Figure 2 As shown, the test preparation begins first. First, rock samples are processed by taking rock blocks from the site and shaping them into samples that meet the experimental requirements. Specifically, the rock blocks can be processed into corresponding rock samples according to different mechanical test requirements. For example, for impact tests, samples can be processed into specimens with a height × base square side length = 50mm × 50mm; for static compression tests, samples can be processed into specimens with a height × base square side length = 100mm × 50mm.

[0034] Due to their geological formation, natural rocks inevitably contain inherent structural defects such as fissures and cleavage. When these rocks are processed into standard specimens, the randomness of the number, size, and distribution of these internal defects directly leads to significant dispersion in the mechanical properties of the specimens. This inherent heterogeneity causes large fluctuations in test results, severely interfering with the objective evaluation of the actual support effect of the sprayed material, and may even mask the performance differences between different materials. To eliminate the interference of individual differences in rock specimens on experimental results and ensure the reliability and comparability of the tests, this method introduces a specimen homogenization screening step before the formal test. This involves non-destructive testing and screening of the processed specimens to ensure good internal consistency and comparability of their physical and mechanical properties. Specifically, non-destructive testing techniques such as ultrasonic testing or CT three-dimensional crack scanning can be used to assess the internal integrity of the specimens. Based on the assessment results, rock specimens with excessively developed defects are eliminated to ensure good consistency in the physical and mechanical properties of the specimens used. The final number of test specimens is determined according to the experimental design, but should not be less than 150.

[0035] Next, test specimens are prepared. A protective layer is applied to a portion of the surface of each selected specimen to form a sprayed surface; the remaining surface areas of the specimen are left untreated as an untreated control. For example, such as... Figure 3 As shown, for a cuboid specimen, a support material can be uniformly sprayed onto its two adjacent sides A and D to form a sprayed surface; the two adjacent sides B and C remain unchanged as untreated control surfaces. During spraying, it is essential to ensure that the support material adheres evenly to the surface of the specimen, and the amount of material sprayed onto each surface should be consistent. The prepared test specimen is then cured in a constant temperature and humidity environment to obtain the test sample. Specifically, the specimen can be placed in a curing chamber (90% humidity, 25℃) for curing for 3 days. Curing the specimen is to fully utilize the mechanical properties of the sprayed material.

[0036] Mechanical testing and data acquisition were conducted on the test samples. Mechanical property testing was performed according to standard methods for rock mechanics testing. During testing, two types of data were simultaneously acquired on both the coated surface and the untreated control surface of the sample: acoustic emission data and image data. An acoustic emission acquisition system was used to capture the elastic waves released by the generation and propagation of internal cracks during the stress process, thus obtaining acoustic emission data. Testing was conducted in test groups; specifically, three test samples could be selected as one test group, and the experiment was divided into 10 test groups.

[0037] S120. Based on the acoustic emission data, calculate the acoustic emission evaluation index reflecting the internal fracture process; and based on the image data, calculate the dynamic evaluation index reflecting the fragment ejection behavior.

[0038] Evaluation indices are calculated based on the collected data. Based on the acoustic emission data, characteristic parameters related to the internal fracture process are extracted, and acoustic emission evaluation indices are calculated. Based on the image data, motion analysis techniques are used to extract the flight trajectory and morphological information of the fragments, and dynamic evaluation indices reflecting the ejection behavior of the fragments are calculated.

[0039] S130. Input the acoustic emission evaluation index and the dynamic evaluation index into the trained comprehensive evaluation model, and output the comprehensive evaluation level of the control effect of the sprayed support material on the surrounding rock.

[0040] The neural network model employs a multilayer perceptron architecture and is obtained through supervised training using historical experimental data. The training data comes from a large number of experimental samples with known support effectiveness. Each sample contains a complete dataset of acoustic emission and dynamic evaluation indicators, along with corresponding expert evaluation level labels. During model training, the backpropagation algorithm is used to optimize the network parameters until the model can accurately fit the complex nonlinear relationship between the input indicators and the support effectiveness level.

[0041] In practical applications, the obtained acoustic emission evaluation indicators and dynamic evaluation indicators are used as input feature vectors and fed into a trained neural network model. The model first performs automatic feature extraction and fusion on the input features, and mines the intrinsic correlation between different indicators through multi-layer nonlinear transformation; then, it generates a corresponding support effect level prediction at the output layer. This level evaluation comprehensively considers multiple factors such as the effect of sprayed materials on surrounding rock strength enhancement, deformation control, and failure mode improvement, and finally outputs a comprehensive evaluation level of the surrounding rock control effect of sprayed support materials.

[0042] The specific evaluation process includes: The aforementioned acoustic emission and dynamic evaluation indices are combined into a feature vector. The Z-Score method is used to standardize the feature vector to eliminate the influence of dimensions. The standardized feature vector is then input into the comprehensive evaluation model, which outputs a probability distribution vector. Based on this probability distribution vector p, the level of rock control effect of the sprayed support material is determined. If max(p) = p I And p I >θ, classified as Level I (excellent control effect); If max(p) = p II It was classified as Level II (good control effect); If max(p) = p III It is classified as Level III (moderate control effect). If max(p) = p IV It is classified as Level IV (poor control effect).

[0043] Where, p I p II p III p IV These correspond to the predicted probabilities of Level I, Level II, Level III, and Level IV, respectively, with θ being the confidence threshold, such as 0.8.

[0044] This method breaks through the traditional experimental paradigm by adopting a differentiated design of double-sided support and double-sided control on the same sample. During sample preparation, a sprayed support layer is applied to two adjacent sides of the same standard rock sample to form the sprayed surface, while the remaining two sides are left untouched as control surfaces. This design directly utilizes the self-comparison of the same sample, fundamentally eliminating the interference of mechanical parameter dispersion caused by the heterogeneity of different rock samples due to primary fractures, joints, etc., in a relatively simple and repeatable way. This scientifically reveals the true support effect of the sprayed material, enabling the evaluation results to accurately reflect the true support effectiveness of the sprayed material. This method makes the experimental conditions closely resemble the local interaction scenario between the support layer and the surrounding rock in actual engineering, effectively avoiding the applicability bias of homogeneous material standards. Simultaneously, multi-source data fusion technology is employed, namely, the collaborative monitoring of acoustic emission data and image data. During the mechanical loading test, acoustic emission data and image data are simultaneously acquired. The acoustic emission signal is used to capture the dynamic processes of internal fracture initiation, propagation, and frictional sliding in real time to reflect the microscopic damage evolution law, while the image data is used to record the ejection trajectory of surface fragments and the macroscopic fracture morphology, thereby quantifying the overall stability changes. Abandoning the homogeneous assumption, this approach replaces theoretical derivations based on homogeneous materials by directly monitoring the actual fracture process and fragment dynamics, ensuring the evaluation results closely align with engineering realities. Finally, acoustic emission evaluation indicators and dynamic evaluation indicators extracted from imagery are input into a trained comprehensive evaluation model. Through multi-parameter fusion analysis, the model outputs a comprehensive support effectiveness level. Trained on a large-scale experimental dataset, this model not only boasts high accuracy but also overcomes the limitations of single indicators, comprehensively revealing the synergistic mechanism between support materials and rock mass.

[0045] In an alternative embodiment, refer to Figure 2 As shown, the mechanical tests include uniaxial compression tests, split Hopkinson bar experiments, and biaxial compression tests.

[0046] The uniaxial compression test process is as follows: Three specimens are selected as one test group, and the experiment is divided into 10 test groups. Uniaxial compression strength tests are performed on each group according to the standard methods for rock mechanics testing. Acoustic emission monitoring is also conducted during the test. At least two acoustic emission probes are used on each surface of the specimen, and the total number of probes should not be less than eight. The acoustic emission probes are arranged as follows: Figure 4 As shown. A partition was placed at the interface between the sprayed surface and the control surface. After the experiment, rock fragments were collected from both surfaces after damage. During the test, a high-speed camera was used to monitor the size and ejection velocity of the fragments when the rock sample was damaged. One high-speed camera should be placed on each side of the sample to monitor the damage to the sprayed surface and the control surface at different angles. The high-speed camera arrangement is as follows. Figure 5 As shown.

[0047] Reference Figure 6As shown, the split-type Hopkinson test system mainly consists of a launching system, an impact rod (bullet), an incident rod, a transmission rod, a data acquisition system, and a data processing system. The test sample is clamped between the incident rod and the transmission rod. The launching system releases high-pressure gas to drive the bullet to impact the incident rod, generating an incident stress pulse (ε). i The pulse propagates along the incident rod, and when it reaches the interface between the incident rod and the test sample, a portion of the pulse is reflected back to the incident rod (reflected pulse ε). r Another portion of the pulse propagates through the sample to the transmission rod (transmission pulse ε). t The three pulse values ​​can be measured using high-precision strain gauges attached to the rod.

[0048] Three test specimens were selected as one test group, and the experiment was divided into 10 test groups in total. Hopkinson bar impact tests were conducted according to the above experimental steps and principles. A partition was placed at the interface between the sprayed surface and the control surface. After the experiment, rock fragments were collected from both the sprayed surface and the control surface after damage. Simultaneously, a high-speed camera was used to monitor the size and ejection velocity of the fragments during the test. One high-speed camera was placed on each side of the specimen to monitor the damage to the sprayed surface and the control surface at different angles. The high-speed camera arrangement is as follows: Figure 7 As shown.

[0049] The biaxial compression test was performed using a true triaxial testing machine. A continuous axial load was applied to the test specimen, and lateral loads were applied to both sides of the specimen for fixed constraints. The front and rear sides were respectively coated and untreated (the front side was the coated surface, and the rear side was the control surface). The high-speed camera was arranged as follows: Figure 8 As shown. During the experiment, two high-speed cameras were used to monitor the size and ejection velocity of the fragments when they were damaged on the front and rear sides. The purpose of this experiment was to simulate the mechanical environment of the surrounding rock in a real underground coal mine roadway, and to more realistically evaluate the support performance of the sprayed material in a real underground mechanical environment. Three test samples were selected as one test group, and the experiment was divided into 10 test groups in total. Acoustic emission monitoring was also used during the test. Acoustic emission probes were placed on the front and rear sides, with at least 2 acoustic emission probes on each surface and a total of no less than 4 probes. Three test samples were selected as one test group, and the experiment was divided into 10 test groups in total. The principal stresses are in the axial loading direction. The stress on the laterally constrained surface (surface B or surface D).

[0050] This invention achieves comprehensive testing coverage from static to dynamic, and from simple stress to complex stress states, through three testing methods: uniaxial compression, Hopkinson bar compression, and biaxial compression. It can simulate different mechanical environments in underground coal mines. Both uniaxial and biaxial compression tests simultaneously acquire acoustic emission signals and high-speed image data, obtaining a complete record of the evolution process from microscopic damage to macroscopic failure, providing sufficient data support for comprehensive evaluation. The biaxial compression test simulates the actual stress state of the surrounding rock in underground roadways, allowing laboratory test results to better reflect the support effect of the sprayed material in actual engineering. By setting up a self-comparison area on the same sample (i.e., the sprayed surface and the untreated control surface), the influence of the discreteness of the rock sample is effectively avoided. Compared with the traditional method of statistical averaging of a large number of samples, this method significantly improves testing efficiency while ensuring the reliability of the results. Each testing method uses standardized experimental equipment and standardized testing procedures, ensuring the repeatability and comparability of the test results, which is conducive to the objective comparison of the performance of different sprayed materials.

[0051] In an optional embodiment, the acoustic emission evaluation index includes a breakage suppression index based on the cumulative count of acoustic emission ringing; The breakage suppression index based on the cumulative count of acoustic emission ringing is determined in the following way: Based on the acoustic emission data, the cumulative count of acoustic emission ringing on the coated surface and the untreated control surface within a set depth range is extracted respectively; the fracture suppression index is determined according to the cumulative count of acoustic emission ringing on the coated surface and the cumulative count of acoustic emission ringing on the control surface.

[0052] The dynamic evaluation indicators include: average ejection velocity of fragments, momentum control efficiency index based on average ejection momentum of fragments, kinetic energy decay safety factor based on average kinetic energy of fragments, fragmentation degree coefficient based on fragment mass distribution, and performance dispersion improvement coefficient.

[0053] (1) The evaluation indicators of the uniaxial compression test are calculated in the following manner: ① The calculation process of the uniaxial acoustic emission fracture suppression index is as follows: A spatial distribution map of acoustic emission events is plotted using acoustic emission data monitored during the test. An acoustic emission event refers to a transient elastic wave released by an independent acoustic emission source. The cumulative acoustic emission ringing count is calculated based on the signal waveform of the acoustic emission event. An acoustic emission event generates a signal waveform; the number of oscillations exceeding a threshold is the ringing count for that event. The cumulative ringing count is obtained by summing the ringing counts of all acoustic emission events. The cumulative acoustic emission ringing count reflects the degree of internal fracture damage during rock loading and can be included in the evaluation index. The cumulative acoustic emission ringing counts within a 2cm radius from the surface to the interior of the sprayed surface and the control surface are calculated separately. The cumulative acoustic emission ringing counts for both the sprayed and control surfaces are taken as the average of 10 test groups. The uniaxial acoustic emission fracture suppression index is calculated based on the cumulative acoustic emission ringing counts. The evaluation level is divided into four levels based on the uniaxial acoustic emission breakage suppression index, as shown in Table 1.

[0054] (1) In the formula, , The cumulative counts of acoustic emission ringing are for the sprayed surface and the control surface, respectively.

[0055] Table 1 ② The average ejection velocity of uniaxial fragments is calculated by tracking the ejection process of fragments from the sprayed and control surfaces through image analysis, and calculating the velocity of the fragments as they leave the main sample. ,Depend on Calculate the total velocity of all fragments ejected from the sprayed surface and the control surface respectively. ,Depend on Calculate the average ejection velocity of the fragments respectively , where n is the total number of fragments from different treated surfaces (i.e., the coated surface or the control surface). This index directly reflects the severity of rock fragment ejection. The lower the ejection velocity, the better the buffering effect of the sprayed material on the impact energy, and the more stable the rock damage. The average ejection velocity of the fragments is taken as the average value of the surfaces under the same conditions in each test group. Based on the average ejection velocity of fragments along a single axis. The rating is divided into five levels, as shown in Table 2.

[0056] (2) In the formula, , , These represent the average ejection velocity of fragments on the sprayed surface, the average ejection velocity of fragments on the control surface, and the average ejection velocity of fragments on a single axis, respectively.

[0057] Table 2 ③ The calculation process of the momentum control effectiveness index is as follows: The experimental process is recorded by two high-speed cameras, and the ejection process of the fragments on the sprayed surface and the control surface is analyzed by image analysis software. The motion velocity of the fragments when they leave the main body of the sample is tracked. The longest and shortest side lengths of the fragments during the ejection and rotation process are captured by image analysis, and the volume of the fragments is calculated. ,Depend on Calculate the mass m of each fragment. For the density of the fragments, according to Calculate the momentum of each fragment. ,Depend on Calculate the total momentum of all fragments on both the painted surface and the control surface. ,Depend on Calculate the average momentum of the fragments on the painted surface and the control surface respectively. Momentum reflects the ability to transmit the ejected motion of rock fragments during rock fracture. The greater the momentum, the better the transmission of ejected rock fragment motion, and the more severe the fracture. The average ejection momentum of the fragments is taken as the average value of the rock surface under the same conditions in each test group. The first momentum control effectiveness index is used. The evaluation levels are divided into four levels, as shown in Table 3.

[0058] (3) Table 3 ④ The calculation process of the kinetic energy decay safety factor based on the average kinetic energy of the fragments is as follows: The experimental process is recorded using two high-speed cameras. Image analysis software is used to analyze the fragment ejection process on the sprayed surface and the control surface respectively, tracking the velocity of the fragments as they leave the main sample. Based on image analysis, the longest and shortest side lengths of the fragments during the ejection rotation are captured, and the fragment volume is calculated. Calculate the mass of each fragment, based on Calculate the kinetic energy of each fragment. ,Depend on Calculate the total kinetic energy of all fragments on both the sprayed and control surfaces. , Let be the kinetic energy of the i-th fragment. Calculate the average kinetic energy of the fragments on the sprayed surface and the control surface, respectively. Kinetic energy reflects how much impact energy is converted into ejected fragment energy during rock fracture. Lower kinetic energy indicates that the coating material absorbs more impact energy and dissipates it through crack penetration and fracturing, rather than converting it into fragment kinetic energy, resulting in more stable rock fracture. The average fragment kinetic energy is taken as the average value of surfaces under the same conditions in each test group. This is based on the first kinetic energy decay safety factor. The evaluation levels are divided into four levels, as shown in Table 4.

[0059] (4) Table 4 ⑤ The calculation process for the fragmentation degree coefficient D based on the fragment mass distribution is as follows: After the experiment, all fragments from the sprayed surface and the control surface are collected, weighed, and the fragmentation degree coefficients of the sprayed surface and the control surface are calculated respectively. , where m z m is the total mass of the painted surface and the control surface. s The total mass of the ejected fragments is represented by D. A larger D value indicates a greater degree of overall rock fragmentation. This suggests that the sprayed material failed to effectively inhibit the propagation and penetration of cracks. The fragmentation coefficient is the average value of the surfaces under the same conditions in each test group. The evaluation of the fragmentation coefficient (D) needs to consider both the degree of fragmentation and the integrity of the rock mass. The percentage of the maximum fragment mass, M, is introduced. max The percentage (the ratio of the largest mass fragment to the total mass of fragments) is used as an auxiliary indicator to calculate the uniaxial breakage coefficient. : (5) (6) Table 5 (2) The evaluation indices of the split Hopkinson bar test are calculated in the following manner: ① The process for determining the average ejection velocity of the impact fragments is as follows: The experimental process is recorded using two high-speed cameras, and image analysis software is used to analyze the ejection process of the fragments on the sprayed surface and the control surface, respectively, and to track the velocity of the fragments when they leave the main body of the sample. ,Depend on Calculate the total velocity of all fragments ejected from the sprayed surface and the control surface respectively. ,Depend on Calculate their average speed respectively , where n is the total number of fragments from different treated surfaces. This indicator directly reflects the severity of rock fragment ejection. A lower ejection velocity indicates a better buffering effect of the sprayed material on impact energy, resulting in more stable rock breakage. The average ejection velocity of the fragments is taken as the average value of surfaces under the same conditions in each test group. Based on the average ejection velocity of the impact fragments... The evaluation levels are divided into five levels, as shown in Table 6: (7) Table 6 ② The process for determining the second momentum control effectiveness index is as follows: The experimental process is recorded using two high-speed cameras. Image analysis software is used to analyze the ejection process of the fragments from the sprayed surface and the control surface, tracking the velocity of the fragments as they leave the main sample. Based on image analysis, the longest and shortest side lengths of the fragments during the ejection and rotation process are captured, and the fragment volume is calculated. Calculate the mass of each fragment. ,according to Calculate the momentum of each fragment. ,Depend on Calculate the total momentum of all fragments on both the painted and control surfaces separately. Calculate the average ejection momentum of impact fragments on the coated surface and the control surface respectively. Momentum reflects the ability to transmit the ejection motion of rock fragments during rock fracture. The greater the momentum, the better the transmission of ejection motion, and the more severe the fracture. The average ejection momentum of the fragments is taken as the average value of the rock surface under the same conditions for each test group. A second momentum control effectiveness index is determined based on the average ejection momentum of the fragments, and then... The evaluation levels are divided into four levels, as shown in Table 7.

[0060] (8) Table 7 ③ The process for determining the kinetic energy decay safety factor is as follows: The experimental process is recorded using two high-speed cameras. Image analysis software is used to analyze the ejection process of the fragments from the sprayed surface and the control surface, tracking the velocity of the fragments as they leave the main sample. Based on image analysis, the longest and shortest side lengths of the fragments during the ejection and rotation process are captured, and the fragment volume is calculated. Calculate the mass of each fragment, based on Calculate the kinetic energy of each fragment. ,Depend on Calculate the total kinetic energy of all fragments on both the sprayed and control surfaces. ,Depend on Calculate the average kinetic energy of the fragments on the sprayed surface and the control surface respectively. Kinetic energy reflects how much impact energy is converted into ejected fragment energy during rock fracture. Lower kinetic energy indicates that the coating material absorbs more impact energy and dissipates it through crack penetration and fracturing, rather than converting it into fragment kinetic energy, resulting in more stable rock fracture. The average fragment kinetic energy is taken as the average value of surfaces under the same conditions in each test group. The second kinetic energy decay safety factor is determined based on the average fragment kinetic energy. The evaluation level is divided into four levels based on the kinetic energy decay safety factor, as shown in Table 8.

[0061] (9) Table 8 ④ The process for determining the impact fracture coefficient is as follows: After the experiment, collect all fragments from the sprayed surface and the control surface, weigh them, and calculate the fracture coefficients of the sprayed surface and the control surface respectively. , where m z m is the total mass of the painted surface and the control surface. s The total mass of the ejected fragments is represented by D'. A larger D' value indicates a greater degree of overall rock fragmentation. This suggests that the sprayed material failed to effectively inhibit the propagation and penetration of cracks. The fragmentation coefficient is the average value of the surfaces under the same conditions in each test group. The evaluation of the fragmentation coefficient (D') needs to consider both the degree of fragmentation and the integrity of the rock mass. The percentage of the maximum fragment mass, M, is introduced. max The percentage (the ratio of the largest mass fragment to the total mass of the fragments) is used as an auxiliary indicator to calculate the impact fragmentation coefficient. : (10) (11) Table 9 (3) The evaluation indicators for the bidirectional compression test are calculated in the following manner: ① The process for determining the acoustic emission fracture suppression index is as follows: A spatial distribution map of acoustic emission events is plotted using acoustic emission data monitored during the test. The cumulative count of acoustic emission ringing can reflect the degree of internal fracture damage during rock loading and can be included in the evaluation index. The total number of acoustic emission events within a 2cm radius from the surface to the interior of the sprayed surface is counted; the total number of acoustic emission events within a 2cm radius from the surface to the interior of the control surface is also counted. The cumulative count of acoustic emission ringing for both the sprayed and control surfaces is taken as the average of 10 test groups. The acoustic emission fracture suppression index under bidirectional compression testing is determined based on the cumulative count of acoustic emission ringing. Based on the acoustic emission breakage suppression index under bidirectional compression test, the evaluation level is divided into four levels, as shown in Table 10.

[0062] (12) Table 10 ② The experimental process was recorded using two high-speed cameras, and image analysis software was used to analyze the fragment ejection process on the sprayed surface and the control surface, respectively, and to track the velocity of the fragments as they left the main sample. ,Depend on Calculate the total velocity of all fragments ejected from the sprayed surface and the control surface respectively. ,Depend on Calculate their average speed respectively , where n is the total number of fragments from different treated surfaces. This indicator directly reflects the severity of rock fragment ejection. A lower ejection velocity indicates a better buffering effect of the coating material on impact energy, resulting in more stable rock breakage. The average ejection velocity of the fragments is taken as the average value of surfaces under the same conditions in each test group. Based on the average ejection velocity of fragments under bidirectional compression testing... The evaluation levels are divided into five levels, as shown in Table 11.

[0063] (13) Table 11 ③ The third momentum control effectiveness index was determined as follows: The experimental process was recorded using two high-speed cameras. Image analysis software was used to analyze the ejection process of the fragments from the sprayed surface and the control surface, tracking the velocity of the fragments as they left the main sample. Based on the image analysis, the longest and shortest side lengths of the fragments during the ejection rotation were captured, and the fragment volume was calculated. Calculate the mass of each fragment, based on Calculate the momentum of each fragment. ,Depend on Calculate the total momentum of all fragments on both the painted surface and the control surface. ,Depend on Calculate the average momentum of the fragments on the sprayed surface and the control surface respectively. Momentum reflects the ability to transmit the ejection motion of rock fragments during rock fracture. The greater the momentum, the better the transmission of ejection motion, and the more severe the fracture. The average ejection momentum of the fragments is taken as the average value of the rock surface under the same conditions for each test group. The third momentum control effectiveness index is determined based on the average ejection momentum of the fragments. The evaluation level is divided into four levels based on the third momentum control effectiveness index, as shown in Table 12.

[0064] (14) Table 12 ④ The process for determining the third kinetic energy attenuation safety factor is as follows: The experimental process is recorded using two high-speed cameras. Image analysis software is used to analyze the ejection process of the fragments from the sprayed surface and the control surface, tracking the velocity of the fragments as they leave the main sample. Based on image analysis, the longest and shortest side lengths of the fragments during the ejection rotation are captured, and the fragment volume is calculated. Calculate the mass of each fragment, based on Calculate the kinetic energy of each fragment. ,Depend on Calculate the total kinetic energy of all fragments on both the sprayed and control surfaces. ,Depend on Calculate the average kinetic energy of the fragments on the sprayed surface and the control surface respectively. Kinetic energy reflects how much impact energy is converted into ejected fragment energy during rock fracture. Lower kinetic energy indicates that the coating material absorbs more impact energy and dissipates it through crack penetration and fracturing, rather than converting it into fragment kinetic energy, resulting in more stable rock fracture. The average fragment kinetic energy is taken as the average value of surfaces under the same conditions in each test group. The third kinetic energy decay safety factor is determined based on the average fragment kinetic energy. The evaluation level is divided into four levels based on the third kinetic energy decay safety factor, as shown in Table 13.

[0065] (15) Table 13 ⑤ The process for determining the degree of breakage coefficient is as follows: After the experiment, all fragments from the sprayed surface and the control surface are collected, weighed, and the degree of breakage coefficients of the sprayed surface and the control surface are calculated respectively. , where m z m is the total mass of the painted surface and the control surface. s The total mass of the ejected fragments is represented by D1. A larger D1 value indicates a greater degree of overall rock fragmentation. This suggests that the sprayed material failed to effectively inhibit the propagation and penetration of cracks. The fragmentation coefficient is the average value of the surfaces under the same conditions in each test group. The evaluation of the fragmentation coefficient (D1) needs to consider both the degree of fragmentation and the integrity of the rock mass. The percentage of the largest fragment mass (Mmax%) (the ratio of the largest mass fragment to the total mass of the fragments) is introduced as an auxiliary indicator to determine the fragmentation coefficient under bidirectional compression testing. The evaluation level is divided into four levels based on the fracture degree coefficient under bidirectional compression test, as shown in Table 14.

[0066] (16) (17) Table 14 The performance dispersion improvement coefficient is used to evaluate the effect of the sprayed material on improving the consistency of the mechanical properties of the rock sample; the performance dispersion improvement coefficient is determined in the following manner: The average kinetic energy data of fragments from multiple rock samples on the sprayed surface and the untreated control surface were obtained; Based on the average kinetic energy data of fragments from the multiple rock samples on the sprayed surface, the coefficient of variation and the mean kinetic energy of the sprayed surface are calculated; and based on the average kinetic energy data of fragments from the multiple rock samples on the untreated control surface, the coefficient of variation and the mean kinetic energy of the control surface are calculated; based on the coefficient of variation of the sprayed surface, the coefficient of variation of the control surface, the mean kinetic energy of the sprayed surface, and the mean kinetic energy of the control surface, the performance dispersion improvement coefficient is determined.

[0067] The process for determining the performance dispersion improvement coefficient is as follows: To evaluate the effect of sprayed materials on improving the consistency of mechanical properties of rock samples, a performance dispersion improvement coefficient is introduced. This index assesses the ability of sprayed materials to control performance dispersion by comparing the coefficients of variation between different samples on the sprayed surface and between different samples on the control surface. This index evaluates whether sprayed materials can reduce the fluctuation of mechanical properties of rock samples and improve the reliability and consistency of support effects. This index uses the average kinetic energy (Ea) of impact fragments, which most directly reflects the severity of sample failure, as the most direct indicator. K As the basis for discreteness analysis, the calculation method is as follows: First, calculate the standard deviation between different samples on the sprayed surface and between different samples on the control surface. and mean μ.

[0068] (18) In the formula: Let be the average kinetic energy of the fragments on the sprayed or control surface of the i-th sample.

[0069] Then, the coefficients of variation between different samples on the sprayed surface and between different samples on the control surface are calculated using formula (19). : (19) Finally, the performance dispersion improvement coefficient S is calculated using formula (20). Cv : (20) The evaluation level is divided into five levels based on the performance dispersion improvement coefficient, as shown in Table 15.

[0070] Table 15 This invention achieves a multi-dimensional comprehensive evaluation of the rock mass control effect of sprayed support materials through the joint analysis of acoustic emission indicators and multiple kinetic indicators, overcoming the limitations of single-indicator evaluation and demonstrating a comprehensive effect. Secondly, the self-comparison design based on the same sample effectively eliminates the interference of individual differences in rock samples on the evaluation results, significantly improving the accuracy and reliability of the evaluation results, achieving a high level of accuracy. Regarding engineering applicability, the introduction of the performance dispersion improvement coefficient enables the evaluation results to truly reflect the improvement effect of the sprayed material on rock mass stability in actual engineering projects, greatly improving the correlation between laboratory evaluation and engineering practice. By establishing a mathematical calculation model, the complex support effect is transformed into a quantifiable evaluation index, achieving an objective and standardized evaluation of the support effect and demonstrating a quantitative evaluation effect. Compared with traditional methods involving a large number of samples, this invention significantly reduces the number of samples and testing workload while ensuring evaluation accuracy, improving evaluation efficiency and demonstrating a significant efficiency improvement effect.

[0071] In an optional embodiment, this method proposes a deep learning evaluation system based on physical constraints and data augmentation (Physics-Constrained Deep Neural Network with VAE Augmentation, PC-DNA). This method aims to address the problem of limited sample size in rock mechanics experiments, which makes it difficult to meet the training requirements of traditional deep learning, while ensuring that the evaluation results conform to the fundamental principles of rock mechanics through a physical constraint mechanism. (Refer to...) Figure 9 As shown, the comprehensive evaluation model described in S130 above is trained in the following manner: S1301. Obtain the training dataset, which contains multiple sets of sample data. Each set of sample data includes acoustic emission evaluation index and dynamic evaluation index calculated from the mechanical response data of the sprayed sample and the untreated sample, as well as the corresponding benchmark evaluation level label. Obtain a training dataset containing multiple sets of sample data. Each set of sample data includes input features and corresponding benchmark evaluation level labels. The input features are acoustic emission evaluation indicators and dynamic evaluation indicators calculated based on the mechanical response data of sprayed and untreated samples. A structured feature vector x∈R is constructed based on each evaluation indicator. 15 The feature vector contains 15 dimensions of physical evaluation index data: uniaxial acoustic emission fragmentation suppression index η. AEcount Average ejection velocity of single-axis fragments First Momentum Control Efficiency Index Safety factor for first kinetic energy decay Single-axis crushing degree coefficient S D Average ejection velocity of impact fragments Second momentum control efficiency index Safety factor for second kinetic energy decay Impact fracture degree coefficient S D' Acoustic emission fracture suppression index η under bidirectional compression test AEcount' Average ejection velocity S of fragments under bidirectional compression test 1v Third Momentum Control Efficiency Index Third kinetic energy decay safety factor Fracture degree coefficient under bidirectional compression test Performance dispersion improvement coefficient S Cv It also includes the corresponding benchmark evaluation level labels: Level I, Level II, Level III, and Level IV.

[0072] The Z-Score standardization method is used to process the feature vectors, eliminating the influence of data differences on the neural network weight updates, resulting in standardized feature vectors. For the observed values ​​x of the j-th acoustic emission evaluation index and the dynamic evaluation index...j The standardized value x j ∗ (Right now Figure 9 The standardized score calculation formula is as follows: (twenty one) Where, μ j Let σ be the mean of the acoustic emission evaluation index and the dynamic evaluation index across all samples. j Let be the standard deviation of the acoustic emission evaluation index and the dynamic evaluation index. After standardization, all input features are mapped to a distribution space with a mean of 0 and a variance of 1.

[0073] Based on the individual acoustic emission evaluation indicators and dynamic evaluation indicators, the initial comprehensive score S is calculated using a weighted average formula. total The baseline evaluation level label is obtained by directly assigning a label based on the score's range. The calculation formula is as follows: (twenty two) Wherein, U1, U2, and U3 are the comprehensive scores of the uniaxial compression test, the split Hopkinson bar test, and the bidirectional compression test, respectively, and w1, w2, and w3 are the weights corresponding to the uniaxial compression test, the split Hopkinson bar test, and the bidirectional compression test, respectively. In this embodiment, the weights are set as follows: w1=0.2, w2=0.35, and w3=0.45.

[0074] The scores for each dimension are calculated as a weighted sum of its internal individual evaluation indicators. The overall score U1 for the uniaxial compression test is calculated using the following formula: (twenty three) The overall score U2 of the split Hopkinson bar test is calculated using the following formula: (twenty four) The overall score U3 of the bidirectional compression test is calculated using the following formula: (25) S1302. The training dataset is augmented using a variational autoencoder model to generate augmented samples for model training, resulting in an augmented dataset.

[0075] To address the problem of insufficient physical sample quantity in rock mechanics experiments, this invention constructs a Variational Autoencoder (VAE) model for probabilistic modeling and expansion. The VAE model comprises an encoder network and a decoder network. The encoder network receives the standardized feature vectors and maps them to Gaussian distribution parameters in a low-dimensional latent space, namely the mean vector μ and the log-variance vector logσ.2 σ represents the standard deviation. A reparameterization technique is used to sample in the latent space, introducing standard normal distribution noise ε ~ N(0, I) to generate a latent vector z = μ + σ⊙ε, simulating the uncertainty and randomness of the internal microstructure of natural rock materials. I represents the identity matrix. The decoder network maps the sampled latent vector z back to the original feature space, generating virtual experimental data. By interpolating and adding small perturbations in the latent space, a large number of virtual samples conforming to the physical distribution of the original data are generated, which, together with the real samples, constitute the augmented training dataset D. train .

[0076] S1303. Using the acoustic emission evaluation index and dynamic evaluation index in the enhanced dataset as input features, and the corresponding benchmark evaluation level label as the training target, the deep neural network model is trained by minimizing the composite loss function, and the trained deep neural network model is used as the comprehensive evaluation model. The composite loss function is determined based on the classification error loss and the physical consistency constraint loss. The classification error loss is determined based on the predicted probability distribution of the model output and the true benchmark rating label. The physical consistency constraint loss is determined based on the predicted score relationship of sample pairs in the training batch. The sample pairs are selected based on the physical performance relationship of the input features of the sample pairs, and the sample with better physical performance should have a predicted score no lower than that of the sample with worse physical performance.

[0077] Specifically, the comprehensive evaluation model in this invention is obtained by training a deep neural network model. The network architecture of the deep neural network model specifically adopts a Deep Fully Connected Network (DFCN). First, a deep fully connected neural network is constructed to establish a nonlinear mapping model from multidimensional physical indicators to support effect evaluation levels. The deep fully connected neural network includes an input layer, hidden layers, and an output layer. The input layer receives a 15-dimensional standardized feature vector. The hidden layer consists of four fully connected layers, with the number of neurons arranged in an inverted pyramid structure (64→32→16→8) to compress the feature dimension layer by layer and extract higher-order abstract features. The output layer contains four neurons, corresponding to the four evaluation levels: Level I (Excellent), Level II (Good), Level III (Medium), and Level IV (Poor). The hidden layers uniformly use the LeakyReLU activation function (α=0.01) to enhance the model's nonlinear expression ability in the negative value range. Dropout layers (dropout rate 0.3) and batch normalization layers are introduced between the fully connected layers to prevent overfitting and improve generalization ability. The output layer uses the Softmax function, and the output sample is the probability distribution vector p=[p I ,p II ,p III ,pIV ].

[0078] A composite loss function incorporating physical consistency constraints is defined to guide the parameter updates of a deep fully connected neural network, ensuring that the evaluation results conform to the fundamental laws of rock mechanics. The composite loss function L... total Classification error loss L class And physical consistency constraint loss L phy Weighted composition: (26) Where λ is the weighting coefficient.

[0079] L class This is used to measure the difference between the predicted probability distribution of the output of a deep fully connected neural network and the true benchmark rating level labels. Specifically, it is obtained by calculating the cross-entropy loss between the predicted probability distribution and the true labels. Assume the current training batch size is N, and there are C = 4 rating levels (Level I, Level II, Level III, Level IV). class Calculated using formula (25): (27) In the formula, y i,c Let y be the baseline rating label for the i-th sample. For example, if sample i is at level I, then y i =[1,0,0,0]. p i,c This is the predicted probability of the i-th sample calculated by the Softmax function from the output layer of the deep fully connected neural network, which is the probability that the i-th sample is predicted by the model to be of the c-th level.

[0080] Physical consistency constraint loss L phy The model is constructed based on the principle of monotonicity constraints. During training, sample pairs (i,j) are randomly selected. If sample i has better acoustic emission and dynamic evaluation indicators (such as higher energy absorption efficiency index and lower fragmentation score) than sample j, then the model is forced to give a higher prediction score (or level I probability) for sample i than for sample j. If the model output violates this rule, a penalty loss is incurred.

[0081] Using augmented dataset D train The AdamW optimizer is used for iterative training. When the validation set loss no longer decreases and the physical consistency constraint loss approaches zero, training is stopped and the optimal model parameters are saved. The trained deep neural network model is then used as the final comprehensive evaluation model.

[0082] This invention utilizes probabilistic modeling with variational autoencoders and reparameterized sampling techniques to generate a large number of virtual samples conforming to physical distribution laws based on limited experimental data, effectively improving the applicability and generalization ability of deep learning models in the field of rock mechanics. By introducing physical consistency constraint loss, the basic principles of rock mechanics and expert experience are integrated into the model training process, ensuring that the evaluation results possess both high data-driven accuracy and strict physical rationality. Through the inverted pyramid structure of a deep fully connected neural network, high-order abstract features of 15 evaluation indicators are effectively extracted, establishing a nonlinear mapping relationship from multi-dimensional indicators to comprehensive performance levels. Combined with regularization techniques such as Dropout and batch normalization, the model's anti-overfitting ability and stability are enhanced, ensuring the consistency and comparability of evaluation results for support materials with different formulations and batches.

[0083] In an optional embodiment, the physical consistency constraint loss is determined as follows: For all selected sample pairs, the hinge loss is calculated based on the prediction score relationship of each sample pair, and the physical consistency constraint loss is determined based on the hinge loss of all sample pairs. Specifically, for a sample pair consisting of a sample with better physical performance and a sample with worse physical performance, based on the input features, the prediction score of the sample with better physical performance and the prediction score of the sample with worse physical performance are calculated respectively. If the prediction score of the sample with better physical performance is less than or equal to the prediction score of the sample with worse physical performance, the hinge loss is calculated based on the prediction scores of the sample with better physical performance and the sample with worse physical performance. Otherwise, the hinge loss of the sample pair is zero.

[0084] Specifically, since the output of a deep fully connected neural network is a probability vector, in order to perform numerical comparisons, the probability vector is first converted into a continuous scalar score. The weight vectors for each level are defined as w = [1.0, 0.7, 0.4, 0.1] (corresponding to levels I to IV). For the i-th sample, its predicted score The calculation is as follows: (28) In the formula: represents the weight of the c-th evaluation level. A higher predicted score indicates that the model considers the material to provide better support. c is the evaluation level index, representing the four evaluation levels (Level I, Level II, Level III, Level IV). i is the sample index, representing the i-th sample in the current training batch.

[0085] In the current training batch, M pairs of samples (i,j) are randomly selected. For each pair of samples, based on its feature vector x, the relative merits of their physical performance are determined. The determination method is as follows: Calculate the weighted Euclidean distance between the feature vectors of two samples or directly compare their initial composite scores S. total If the physical properties of sample i are significantly better than those of sample j (denoted as x), then... i x j In theory, the model's prediction score must satisfy the following conditions: > .

[0086] Hinge loss and its variants are used to penalize predictions that violate the above physical laws.

[0087] (29) In the formula, L phy The loss represents the physical consistency constraint, and `max` indicates taking the maximum value. Ω represents the loss satisfying the physical dominance relation x. i >x j The set of sample pairs. M is the number of sample pairs in the set Ω. Let be the predicted score for the i-th sample. Let ζ be the predicted score for the j-th sample. ζ is the safety margin, usually set to 0.1. This means that the score for a good sample must not only be higher than that of a bad sample, but also at least 0.1 points higher; otherwise, a small loss will occur. If sample i has better physical performance, but the model gives sample j a higher predicted score (i.e., ...), then... > ), then S safe (j) -S safe (i) A positive value indicates an error. `max(0,…)` is the ReLU function. If the prediction order is correct ( > If the expression in parentheses is negative, the loss is zero; a loss only occurs when the prediction order is incorrect or the discrimination is insufficient.

[0088] This invention, through the aforementioned physical consistency constraint loss, transforms the fundamental mechanical principle that samples with superior physical performance should receive higher evaluation scores into a calculable loss function term, ensuring that the AI ​​model strictly adheres to domain knowledge. Physical consistency constraints effectively prevent the model from making absurd predictions that violate common sense and physical laws, ensuring reasonable evaluation results even in data-sparse regions. As a powerful regularization method, physical constraints guide the model to learn essential characteristics that conform to physical laws, rather than simply fitting surface correlations in the training data, thus improving the prediction stability for unknown materials. In situations with limited data, physical constraints provide guiding signals for model optimization, effectively preventing overfitting, accelerating model convergence, and enhancing practical value in small-sample scenarios.

[0089] The following describes the laboratory comprehensive evaluation device for the surrounding rock control effect of coal mine sprayed support material provided by the present invention. The laboratory comprehensive evaluation device for the surrounding rock control effect of coal mine sprayed support material described below can be referred to in correspondence with the laboratory comprehensive evaluation method for the surrounding rock control effect of coal mine sprayed support material described above.

[0090] The laboratory comprehensive evaluation device for the surrounding rock control effect of coal mine sprayed support materials provided by this invention is based on... Figure 10 As shown, it includes: The data acquisition module 210 is used to acquire acoustic emission data and image data synchronously collected during mechanical testing of the rock sample; wherein, the surface of the rock sample includes a sprayed surface and an untreated control surface; The index calculation module 220 is used to calculate an acoustic emission evaluation index reflecting the internal fracture process based on the acoustic emission data; and to calculate a dynamic evaluation index reflecting the fragment ejection behavior based on the image data. The effect evaluation module 230 is used to input the acoustic emission evaluation index and the dynamic evaluation index into the trained comprehensive evaluation model and output the comprehensive evaluation level of the control effect of the sprayed support material on the surrounding rock.

[0091] Figure 11 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 11 As shown, the electronic device may include a processor 310, a communications interface 320, a memory 330, and a communication bus 340. The processor 310, communications interface 320, and memory 330 communicate with each other via the communication bus 340. The processor 310 can call logical instructions from the memory 330 to execute a laboratory comprehensive evaluation method for the control effect of coal mine sprayed support materials on surrounding rock.

[0092] Furthermore, the logical instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0093] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the laboratory comprehensive evaluation method for the control effect of coal mine sprayed support materials on surrounding rock provided by the above methods.

[0094] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a laboratory comprehensive evaluation method for the control effect of coal mine sprayed support materials on surrounding rock provided by the methods described above.

[0095] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0096] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A laboratory comprehensive evaluation method for the surrounding rock control effect of sprayed support materials in coal mines, characterized in that, include: Acoustic emission data and image data are acquired simultaneously during mechanical testing of a rock sample; wherein the surface of the rock sample includes a coated surface and an untreated control surface; Based on the acoustic emission data, an acoustic emission evaluation index reflecting the internal fracture process is calculated; and based on the image data, a dynamic evaluation index reflecting the fragment ejection behavior is calculated. The acoustic emission evaluation index and the dynamic evaluation index are input into the trained comprehensive evaluation model, which outputs a comprehensive evaluation level of the rock control effect of the sprayed support material.

2. The laboratory comprehensive evaluation method for the surrounding rock control effect of coal mine sprayed support materials according to claim 1, characterized in that, The mechanical tests include uniaxial compression tests, split Hopkinson bar experiments, and biaxial compression tests.

3. The laboratory comprehensive evaluation method for the surrounding rock control effect of coal mine sprayed support materials according to claim 1, characterized in that, The acoustic emission evaluation index includes a breakage suppression index based on the cumulative count of acoustic emission ringing; The breakage suppression index based on the cumulative count of acoustic emission ringing is determined in the following way: Based on the acoustic emission data, the cumulative count of acoustic emission ringing on the coated surface and the untreated control surface within a set depth range is extracted respectively; The fracture suppression index is determined based on the cumulative count of acoustic emission ringing on the coated surface and the cumulative count of acoustic emission ringing on the control surface.

4. The laboratory comprehensive evaluation method for the surrounding rock control effect of coal mine sprayed support materials according to claim 1, characterized in that, The dynamic evaluation indicators include: average ejection velocity of fragments, momentum control efficiency index based on average ejection momentum of fragments, kinetic energy decay safety factor based on average kinetic energy of fragments, fragmentation degree coefficient based on fragment mass distribution, and performance dispersion improvement coefficient. The performance dispersion improvement coefficient is used to evaluate the effect of the sprayed material on improving the consistency of the mechanical properties of the rock sample; the performance dispersion improvement coefficient is determined in the following manner: The average kinetic energy data of fragments from multiple rock samples on the sprayed surface and the untreated control surface were obtained; Based on the average kinetic energy data of fragments from the multiple rock samples on the sprayed surface, the coefficient of variation and the mean kinetic energy of the sprayed surface are calculated; and based on the average kinetic energy data of fragments from the multiple rock samples on the untreated control surface, the coefficient of variation and the mean kinetic energy of the control surface are calculated. Based on the coefficient of variation of the sprayed surface, the coefficient of variation of the control surface, the mean kinetic energy of the sprayed surface, and the mean kinetic energy of the control surface, the performance dispersion improvement coefficient is determined.

5. The laboratory comprehensive evaluation method for the surrounding rock control effect of coal mine sprayed support materials according to claim 1, characterized in that, The comprehensive evaluation model is trained in the following manner: Obtain a training dataset, which contains multiple sets of sample data. Each set of sample data includes acoustic emission evaluation index and dynamic evaluation index calculated from the mechanical response data of sprayed and untreated samples, as well as the corresponding benchmark evaluation level label. The training dataset is augmented using a variational autoencoder model to generate augmented samples for model training, resulting in an augmented dataset. Using the acoustic emission evaluation index and dynamic evaluation index in the enhanced dataset as input features, and the corresponding benchmark evaluation level label as the training objective, the deep neural network model is trained by minimizing the composite loss function, and the trained deep neural network model is used as the comprehensive evaluation model. The composite loss function is determined based on the classification error loss and the physical consistency constraint loss. The classification error loss is determined based on the predicted probability distribution of the model output and the true benchmark rating label. The physical consistency constraint loss is determined based on the predicted score relationship of sample pairs in the training batch. The sample pairs are selected based on the physical performance relationship of the input features of the sample pairs, and the sample with better physical performance should have a predicted score no lower than that of the sample with worse physical performance.

6. The laboratory comprehensive evaluation method for the surrounding rock control effect of coal mine sprayed support materials according to claim 5, characterized in that, The physical consistency constraint loss is determined in the following manner: For all selected sample pairs, the hinge loss is calculated based on the prediction score relationship of each sample pair, and the physical consistency constraint loss is determined based on the hinge loss of all sample pairs. Specifically, for a sample pair consisting of a sample with better physical performance and a sample with worse physical performance, based on the input features, the prediction score of the sample with better physical performance and the prediction score of the sample with worse physical performance are calculated respectively. If the prediction score of the sample with better physical performance is less than or equal to the prediction score of the sample with worse physical performance, the hinge loss is calculated based on the prediction scores of the sample with better physical performance and the sample with worse physical performance. Otherwise, the hinge loss of the sample pair is zero.

7. A laboratory comprehensive evaluation device for the surrounding rock control effect of coal mine sprayed support materials, characterized in that, include: The data acquisition module is used to acquire acoustic emission data and image data simultaneously collected during mechanical testing of rock samples; wherein, the surface of the rock sample includes a coated surface and an untreated control surface; The index calculation module is used to calculate an acoustic emission evaluation index reflecting the internal fracture process based on the acoustic emission data; and to calculate a dynamic evaluation index reflecting the fragment ejection behavior based on the image data. The effect evaluation module is used to input the acoustic emission evaluation index and the dynamic evaluation index into the trained comprehensive evaluation model and output the comprehensive evaluation level of the control effect of the sprayed support material on the surrounding rock.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements a laboratory comprehensive evaluation method for the surrounding rock control effect of coal mine sprayed support materials as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements a laboratory comprehensive evaluation method for the surrounding rock control effect of coal mine sprayed support materials as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements a laboratory comprehensive evaluation method for the surrounding rock control effect of coal mine sprayed support materials as described in any one of claims 1 to 6.

Images