Determination method of rock compressive strength based on deep learning springback and audio

Abstract

The invention relates to a deep learning springback and audio-based rock compressive strength determination method. The method comprises the following steps: S1, collecting rock springback, springback instrument striking audio recording and rock uniaxial compressive strength test results through a large sample test; s2, carrying out data processing on the collected rock springback data and the springback striking audio set; s3, establishing a neural network calculation model based on the collected rock springback value, the characteristic value of the springback striking audio spectrum and the corresponding data of the uniaxial compressive strength of the rock; and S4, by collecting a rock rebound value and a characteristic value of a rebound strike audio spectrum, based on the established neural network calculation model, substituting the rock rebound value and the characteristic value of the rebound strike audio spectrum obtained by testing into the prediction model, and calculating the rock uniaxial compressive strength of the rock. The method performs deep learning training to obtain information and summarizes rules to establish the rock strength prediction model, and is suitable for various types of rocks.

Description

Technical Field

[0001] The present invention relates to the fields of geotechnical engineering, rock acoustics, and artificial intelligence, and particularly relates to a method for measuring the compressive strength of rocks based on deep learning rebound and audio. Background Art

[0002] The compressive strength of rocks is an important index for analyzing the mechanical properties of rocks (rock masses), and is also the basic basis for the design and construction plan formulation of geotechnical engineering. At present, the most commonly used method for testing the compressive strength of rocks is the uniaxial compressive strength test of rocks. However, this method has high requirements for conditions such as the sample selection of specimens, specimen size, surface smoothness, and loading speed, and the testing process will damage the specimens and cannot be re-inspected, belonging to a destructive test. In addition, this test requires a large number of core drillings, it is difficult to form soft rocks, and the press equipment is bulky and cannot carry out tests at the engineering site. It is necessary to transport the samples from the field to the laboratory for sample preparation and testing, resulting in poor economy and timeliness. In contrast, the rebound hammer has been widely used in fields such as railways, expressways, water conservancy and hydropower projects for the preliminary survey of rock mass strength due to its small size, light weight, and convenient operation.

[0003] Although the rebound method has obvious advantages, the accuracy of predicting the compressive strength of rocks still has great limitations. First, due to the wide variety of rock types, the differences in the weathering degree, structure and structure of different rocks are significant, and the operation methods of different testers are not unified, resulting in large deviations in rebound values, and there is currently no unified standard specification. Secondly, although the correlation between the rebound value and the uniaxial compressive strength is good, most of the existing studies are based on linear equation fitting, which is only applicable to specific types of rocks and cannot cover all rock types. Finally, a single rebound value is easily affected by factors such as rock surface cracks, roughness, and moisture content, and the prediction accuracy is limited. Research shows that if multi-parameter prediction methods such as sound waves and audio spectrograms are combined, the prediction accuracy of rock strength will be significantly improved.

[0004] In the prior art, a patent for invention with the application number CN201811088260.3 discloses a method for determining the surface strength of rocks based on deep learning of spectrograms. This method obtains an audio spectrogram by hitting the rock surface with a geological hammer and trains a neural network model in combination with the data of a rebound hammer to determine the surface strength of the rock. However, this method can only predict the surface strength of the rock and cannot evaluate the overall strength of the rock. At the same time, due to the uncertainty of the hitting force, angle and position of the geological hammer, it is difficult to ensure the accuracy of the audio spectrogram features. In addition, a patent for invention with the application number CN202110726582.1 discloses a method for predicting the uniaxial compressive strength of rocks by using the rebound value measured by an L-shaped Schmidt hammer and through a random forest model. However, this method only relies on a single rebound value parameter, with a low fitting correlation and insufficient prediction accuracy. It can be seen that there are still problems in the prior art such as low accuracy, limited applicable range, and insufficient model reliability in the determination of rock compressive strength. There is an urgent need for a method for determining rock compressive strength that combines multi-parameter analysis and has high prediction accuracy to meet the actual needs of engineering applications. Summary of the Invention

[0005] Aiming at the problems of low accuracy, limited applicable range, and insufficient model reliability in the method for determining rock compressive strength in the prior art, the present invention provides a method for determining rock compressive strength based on deep learning of rebound and audio. By combining the multi-parameter features of rock rebound values and impact audio, and using a neural network model for deep learning training, the uniaxial compressive strength of the rock is accurately predicted, overcoming the prediction errors caused by single parameters or unstable factors in the prior art. It is applicable to the non-destructive determination of various types of rocks and significantly improves the accuracy and applicability of the test.

[0006] To achieve the above object, the technical solution of the present invention is as follows:

[0007] A method for determining rock compressive strength based on deep learning of rebound and audio, comprising the following steps:

[0008] S1, collecting rock rebound, rebound hammer impact audio recordings, and the test results of rock uniaxial compressive strength through large-sample tests;

[0009] S2, processing the collected rock rebound data and rebound impact audio sets;

[0010] S3, establishing a neural network calculation model based on the collected rock rebound values, characteristic values of rebound impact audio spectra, and the data of the corresponding rock uniaxial compressive strength;

[0011] S4. By collecting the rebound value of the rock and the characteristic values of the rebound impact audio spectrum, substituting the obtained rock rebound value and the characteristic values of the rebound impact audio spectrum into the prediction model based on the established neural network calculation model, calculate the uniaxial compressive strength of the rock.

[0012] Preferably, the step S1 includes:

[0013] S11. When conducting the test, select three major types of representative typical rocks, including igneous rocks, sedimentary rocks, and metamorphic rocks. The rock samples mainly consist of core samples, with a total of 600 groups of core samples. Among them, 400 groups are used to train the neural network calculation model, and 200 groups are used to verify the accuracy of the network calculation model;

[0014] S12. Number the core samples, grind the upper and lower end faces of the core cylindrical specimens flat, divide them into 3 circles, namely the center, inner circle, and around the axis, according to the bottom center plane. Arrange 4 measuring points in each circle, and a total of 12 measuring points are arranged on each specimen for testing;

[0015] S13. Select core samples in the natural state for simple fixation to ensure that they will not move during the test. Conduct rebound tests at the measuring point positions, and the test angle of the rebound instrument is perpendicular to the rock surface;

[0016] S14. While making impacts, use a recording pen with a bit rate greater than 1500 kbps to record the impact audio files and store the audio files;

[0017] S15. Conduct uniaxial compressive strength tests on the core samples.

[0018] Preferably, in the step S13, select a suitable rebound instrument for different rocks. If the rebound value is lower than 40, use an L-type rebound instrument with an impact energy of 0.735 Nm; if the rebound value is higher than 40, use an N-type rebound instrument with an impact energy of 2.207 Nm, and record the test results.

[0019] Preferably, in the step S15, select a hydraulic testing machine with a load of 2000 kN and a minimum resolution of 0.01 kN for the rock uniaxial compressive strength testing machine.

[0020] Preferably, the step S2 includes:

[0021] S21. Conduct value analysis on the rock rebound data. Exclude 2 critical values from the 12 rebound readings in the 3 circles of the measuring area, and the arithmetic mean of the remaining 10 rebound readings is the rebound value of this measuring area, forming a rock rebound value data set;

[0022] S22. Digitally process the rebound audio.

[0023] Preferably, the step S22 includes:

[0024] S221, use short-time Fourier transform on the recorded rebound impact audio to produce the spectrogram of the audio file;

[0025] S222, analyze the spectrogram obtained in step S221. To better reflect the attenuation rate of the sound signal, intercept the time length starting from the time point when the impact begins. Define 0 - 0.5s as the large amplitude segment of the sound wave signal, and 0 - 0.4s as the attenuation segment of the sound wave signal;

[0026] S223, calculate the integral area of the segments intercepted at both ends in step S222. The area between the large amplitude segment curve and the horizontal axis is a1, and the area between the attenuation segment curve and the horizontal axis is a2. Define the ratio of a1 to a2 as the amplitude attenuation coefficient i;

[0027] S224, the frequency of the spectrogram obtained in step S221 is mainly concentrated in the range of 0 - 20000Hz. Define 0 - 5000Hz as the low-frequency region, and 5000 - 20000Hz as the high-frequency region;

[0028] S225, calculate the integral area of the segments intercepted at both ends in step S224. Use the integral method to find the area between the two curves and the horizontal axis. The area of the low-frequency region is b1, and the area of the high-frequency region is b2. Define the ratio of b1 to b2 as the high-low frequency ratio j;

[0029] S226, record the amplitude attenuation coefficient i in step S223 and the high-low frequency ratio j in step S225. So far, all the spectral features in a rebound impact audio have been extracted and digitized, forming a rebound audio feature value dataset. The audio dataset corresponds one-to-one with the rock rebound value and the rock uniaxial compressive strength.

[0030] Preferably, the neural network calculation model in step S3 includes an input layer, a hidden layer, and an output layer. Each layer has several nodes, and the connection state of the nodes between layers is reflected by weights. Step S3 includes:

[0031] S31, accept data input. The number of nodes in the input layer is 3, corresponding to the rock rebound value, the characteristic value of the rebound impact audio spectrogram, and the rock uniaxial compressive strength;

[0032] S32, process the input data;

[0033] S33, output the calculation result.

[0034] Preferably, the number of nodes in the hidden layer is set to 3 by default, and each node contains a perceptron, which includes input items, weights, biases, activation functions and outputs; in the forward propagation process, the input data is calculated by the perceptron node and then processed by the activation function to obtain the output result; in the backward propagation process, the output result is compared with the expected result, and the weights of each node on the network are continuously adjusted through multiple iterations to optimize the model performance.

[0035] Preferably, the rock samples selected in step S11 also include boulders and natural rock outcrops, and a relatively flat and fresh outcrop surface is selected as the measurement area. The measurement area is at least 5 cm away from the edge of the rock mass. 16 measurement points are selected in one measurement area, and the distance between each measurement point is greater than 2 cm.

[0036] Preferably, in step S21, for the block stone or natural rock outcrop, 3 maximum values ​​and 3 minimum values ​​are removed from the 16 rebound values ​​in the measurement area, and the arithmetic mean of the remaining 10 rebound readings is the rebound value of the measurement area.

[0037] Compared with the prior art, the advantages of the present invention are:

[0038] (1) The traditional uniaxial compressive strength test of rock requires drilling and coring of rocks, which is complicated and costly. In addition, the samples need to be transported long distances from the field to the laboratory for sample preparation and testing, which seriously affects the economy and timeliness. The present invention uses a rock compressive strength determination method based on deep learning of rebound and audio. It does not require the preparation of complete samples and can be directly tested quickly and easily at the field geological survey site, avoiding destructive testing of specimens, significantly improving the detection efficiency, and greatly reducing the detection cost.

[0039] (2) When the rebound method is used alone to estimate the compressive strength of rock, the internal differences or defects of the object may cause large test errors, and the prediction accuracy is limited. The present invention combines the two parameters of rebound value and impact audio spectrum characteristics, and quickly determines the compressive strength of rock through multi-parameter collaborative analysis, which significantly improves the accuracy and reliability of the prediction.

[0040] (3) The linear equations in the prior art cannot be applied to the strength prediction of all types of rocks. Different functional relationships need to be set separately for different types of rocks, and the differences in rock structure and properties need to be considered. The present invention inputs rock rebound value, audio spectrum characteristic value and compressive strength test data into the BP neural network, and automatically summarizes the law of rock strength prediction through deep learning training model, which can meet the measurement needs of various types of rocks, especially suitable for actual field geological survey work scenes, and has stronger applicability and versatility.

[0041] (4) The present invention adopts a non-destructive testing method, which will not damage the rock specimens, and solves the problem that the traditional testing method cannot be rechecked. At the same time, by combining the rebound value and the audio characteristic value for multi-scenario determination, it does not rely on laboratory conditions, greatly expanding the application scenarios for determining the compressive strength of rocks and providing a more efficient and accurate basis for engineering design and construction. Description of the Drawings

[0042] Figure 1 It is the flowchart for determining the compressive strength of rocks of the present invention;

[0043] Figure 2 It is the schematic diagram for extracting the rebound audio characteristic value of the present invention;

[0044] Figure 3 It is the flowchart for training the BP neural network of the present invention;

[0045] Figure 4 It is the neural network model diagram of the present invention;

[0046] Figure 5 It is the rebound audio waveform diagram of the present invention;

[0047] Figure 6 It is the waveform diagram after the amplitude positive normalization processing of the present invention;

[0048] Figure 7 It is the spectrogram of the present invention;

[0049] Figure 8 It is the comparison diagram between the original result and the predicted result of the neural network model. Detailed Embodiments

[0050] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art belong to the present invention.

[0051] In addition, the described features, structures or characteristics can be combined in any suitable manner in one or more embodiments. In the following description, many specific details are provided to give a full understanding of the embodiments of the present application. However, those skilled in the art will realize that the technical solutions of the present application can be practiced without one or more of the specific details, or other methods, components, devices, steps, etc. can be adopted. In other cases, well-known methods, devices, implementations or operations are not shown or described in detail to avoid obscuring various aspects of the present application.

[0052] Such as Figure 1As shown in the figure, this embodiment discloses a method for measuring the compressive strength of rocks based on deep learning rebound and audio, which is characterized by including the following steps:

[0053] S1. Collect rock rebound, rebound hammer impact audio recordings, and the results of rock uniaxial compressive strength tests through large-sample tests;

[0054] S2. Process the collected rock rebound data and rebound impact audio sets;

[0055] S3. Based on the collected rock rebound values, characteristic values of rebound impact audio spectra, and the data of the corresponding rock uniaxial compressive strength, establish a neural network calculation model;

[0056] S4. By collecting the rock rebound values and characteristic values of rebound impact audio spectra, based on the established neural network calculation model, substitute the obtained rock rebound values and characteristic values of rebound impact audio spectra into the prediction model to calculate the rock uniaxial compressive strength of the rock.

[0057] Among them, step S1 specifically includes:

[0058] S11. When conducting the test, select three major types of typical rocks with representativeness, including igneous rocks, sedimentary rocks, and metamorphic rocks. The rock samples mainly include core samples. The representative rocks of igneous rocks mainly include granite, diorite, diabase, andesite, basalt, etc. The representative rocks of sedimentary rocks mainly include sandstone, mudstone, shale, limestone, dolomite, tuff, etc. The representative rocks of metamorphic rocks mainly include slate, schist, gneiss, etc. The total number of core samples is 600 groups, of which 400 groups are used to train the neural network calculation model, and 200 groups are used to verify the accuracy of the network calculation model. It should be noted that the total number of core samples is not a fixed limit, about 600 groups are sufficient, and the total number of samples can be appropriately adjusted according to actual needs. In some embodiments, block stones and natural rock outcrops can also be selected. Select a relatively flat and fresh outcrop surface as the measuring area. If the surface is covered with moss or sediment, it needs to be polished with sandpaper or a grinding wheel. Avoid cracks during the test. The measuring area should be at least 5 cm away from the edge of the rock mass. Take 16 measuring points in one measuring area, and the distance between each measuring point is greater than 2 cm.

[0059] S12. Number the core samples, grind the upper and lower end faces of the core cylindrical specimens flat, and divide them into 3 circles: the center, the inner circle, and the perimeter around the axis according to the bottom center plane. Arrange 4 measuring points in each circle, and a total of 12 measuring points are arranged on each specimen for testing.

[0060] S13. Select core samples in their natural state for simple fixation to ensure they do not move during testing. Conduct rebound tests at the measuring points, with the rebound instrument testing angle perpendicular to the rock surface. Select suitable rebound instruments for different rocks. For soft rocks such as mudstone, siltstone, and weathered and altered rocks, if the rebound value is less than 40, use an L-type rebound instrument with an impact energy of 0.735 Nm; for hard rocks such as granite, andesite, limestone, basalt, and shale, if the rebound value is higher than 40, use an N-type rebound instrument with an impact energy of 2.207 Nm, and record the test results.

[0061] S14. While performing the rebound impact, use a recording pen with a bit rate greater than 1500 kbps to record the rebound impact audio file and store the audio file. When the rebound values are relatively stable during the rebound test, use a recording pen to record the audio file while performing the rebound impact, and keep the recording environment as quiet as possible to avoid noise.

[0062] S15. Conduct a uniaxial compressive strength test on the core samples. Specifically, the test procedures can refer to Section 2.7 of the "Standard for Test Methods of Engineering Rock Masses" GB / T 50266 - 2013 for the test. Select a hydraulic testing machine with a load of 2000 kN and a minimum resolution of 0.01 kN for the rock uniaxial compressive strength testing machine.

[0063] Among them, step S2 specifically includes:

[0064] S21. Analyze the values of the rock rebound data. For core samples, among the 12 rebound readings of the 3 circles in the measuring area, eliminate 2 critical values, and the arithmetic mean of the remaining 10 rebound readings is the rebound value of this measuring area, forming a rock rebound value data set; for block stones or natural rock outcrops, eliminate 3 maximum values and 3 minimum values from the 16 rebound values in the measuring area, and the arithmetic mean of the remaining 10 rebound readings is the rebound value of this measuring area, and conduct statistics on the rebound values to form a rock rebound value data set.

[0065] S22. Digitally process the rebound audio.

[0066] As Figure 2 shown, among them, step S22 specifically includes:

[0067] S221. Use the short-time Fourier transform to make a spectrogram of the recorded rebound impact audio file.

[0068] S222. Analyze the spectrogram obtained in step S221. In order to better reflect the attenuation speed of the sound signal, intercept the time length starting from the time point when the impact starts. Define 0 - 0.5 s as the large amplitude segment of the sound wave signal, and define 0 - 0.4 s as the attenuation segment of the sound wave signal.

[0069] S223. Calculate the integral area of the segments intercepted at both ends in step S222. The area between the curve of the large-amplitude segment and the horizontal axis is a1, and the area between the curve of the attenuation segment and the horizontal axis is a2. Define the ratio of a1 to a2 as the amplitude attenuation coefficient i.

[0070] S224. The frequencies of the spectrogram obtained in step S221 are mainly concentrated in the range of 0 - 20000 Hz. Consider 0 - 5000 Hz as the low-frequency region and 5000 - 20000 Hz as the high-frequency region.

[0071] S225. Calculate the integral area of the segments intercepted at both ends in step S224. Use the integral method to find the areas between the two curves and the horizontal axis. The area of the low-frequency region is b1, and the area of the high-frequency region is b2. Define the ratio of b1 to b2 as the high-low frequency ratio j.

[0072] S226. Record the amplitude attenuation coefficient i in step S223 and the high-low frequency ratio j in step S225. So far, all the spectrogram features in a rebound impact audio have been extracted and digitized, forming a rebound audio feature value dataset. The audio dataset corresponds one-to-one with the rock rebound value and the rock compressive strength.

[0073] Among them, the neural network calculation model in step S3 includes an input layer, a hidden layer, and an output layer. Each layer has several nodes, and the connection state between the nodes of different layers is reflected by weights. Step S3 specifically includes:

[0074] S31. Accept data input. The number of nodes in the input layer is 3, corresponding to the rock rebound value, the feature value of the rebound impact audio spectrogram, and the uniaxial compressive strength of the rock.

[0075] S32. Process the input data.

[0076] As Figure 3-4 shown, the number of nodes in the input layer is equal to the dimension of the input, the number of nodes in the output layer is equal to the dimension of the output, and the number of nodes in the hidden layer can be set according to the actual situation. The default setting is 3. Each node contains a perceptron (i.e., a single neuron). The perceptron includes an input term, weights, a bias, an activation function, and an output. In the forward propagation process, the input data is processed through the perceptron node and then through the activation function to obtain the output result. In the backward propagation process, the output result is compared with the expected result, and the weights of each node on the network are continuously adjusted through multiple iterations to optimize the model performance.

[0077] S33. Output the calculation result.

[0078] In this embodiment, the learning of the neural network mainly aims to solve a set of W and b to minimize the error function of the neural network. The calculation process of the neural network includes two stages: forward propagation and backward propagation. In the forward propagation stage, the input data enters the network through the input layer and obtains the output result after being calculated by the hidden layer and the output layer. There are connection relationships between the neurons in each layer, and the weights and biases of the connections can be learned and adjusted through training. In the forward propagation process, after the input data is calculated by the perceptron nodes, the output result is obtained through the processing of the activation function. In the backward propagation stage, the output result is compared with the expected result to calculate the error, and then the error is propagated backward layer by layer from the output layer to the hidden layer, and the weights and biases of each neuron in the network are updated to reduce the error. This process will be iterated continuously until the stopping criterion is met. In the backward propagation process, according to the gradient descent method, gradient search technology is used to update the weights and biases.

[0079] The following verifies the measurement method of this embodiment through an example:

[0080] First, obtain the test data through testing: For the selected various types of rocks mentioned above, carry out rebound tests to obtain the rebound values and the corresponding rebound impact audio spectrograms; carry out uniaxial compression tests on the rocks to obtain the uniaxial compressive strength of the rocks. For soft rocks with a rebound value below 40 (such as mudstone, shale, weathered and altered rocks, etc.), use a rebound hammer with an L-type impact energy of 0.735 Nm, and for hard rocks with a rebound value above 40 (such as granite, andesite, limestone, etc.), use an N-type rebound hammer with an impact energy of 2.207 Nm; select a recording pen with a bit rate greater than 1500 kbps as the recording device to record and sample the audio when each group of rocks is impacted by the rebound hammer; select a hydraulic testing machine with a load of 2000 kN and a minimum resolution of 0.01 kN for the rock uniaxial compressive strength testing machine. The final effective test data obtained from the experiment is shown in Table 1.

[0081] Table 1 Test data of rock rebound value, rebound audio, and uniaxial compressive strength

[0082]

[0083] Second, process the audio data to obtain audio feature values: Perform positive value processing on the rebound audio waveform diagram obtained from the experiment to fully present the change process of the sound wave amplitude over time, and obtain the acoustic feature values - amplitude attenuation coefficient i and high-low frequency ratio j from the audio time-domain signal. Taking the 7-64.wav audio as an example, through analysis, the rebound audio waveform, the waveform after amplitude positive value processing, and the spectrum are respectively as Figure 5-7 shown.

[0084] Third, process all the audio files to obtain their amplitude attenuation coefficients and high-low frequency ratios as shown in Table 2 below.

[0085] Table 2 Amplitude attenuation coefficients and high-low frequency ratios of each audio

[0086]

[0087] Finally, using the rebound values, amplitude attenuation coefficients, high-low frequency ratios, and compressive strength values (Tables 1 and 2) obtained in the above tests as training samples, the neural network was used for training to obtain a prediction model for rock compressive strength. The comparison between the original results and the prediction results of the model is as Figure 8 shown.

[0088] It can be seen from the training results that most of the prediction results in the samples are relatively close to the original test results, and the prediction effect is good. There are certain differences between some prediction results and the original results. This difference can be further reduced by further increasing the number of test samples and adjusting the model training parameters to meet the accuracy requirements of the project.

[0089] Taking the test results of rock samples numbered 10 - 75 as an example to verify the accuracy of the model. The test obtained the rebound value of the rock as 75, the amplitude attenuation coefficient as 0.80, the high-low frequency ratio as 0.78, and the compressive strength as 90.90 MPa. Inputting the rebound value, amplitude attenuation coefficient, and high-low frequency ratio into the model, the predicted strength value can be obtained as 86.22 MPa, with a difference of 4.68 MPa from the test result of 90.90 MPa, an error of 5.1%, and an accuracy rate of 94.9%. The model prediction result is relatively accurate.

[0090] In summary, the present invention discloses a method for measuring the compressive strength of rocks based on deep learning rebound and audio. By collecting rock rebound values, impact audio, and uniaxial compressive strength test data of rocks, characteristic values such as amplitude attenuation coefficients and high-low frequency ratios of the audio are extracted, and deep learning training is carried out using a neural network model to construct a compressive strength prediction model applicable to various rock types. This method introduces multi-parameter collaborative analysis and intelligent prediction technology in the analysis of rock mechanical properties, and can quickly and accurately predict the uniaxial compressive strength of rocks. By implementing the technical solution of the present invention, the prediction accuracy and reliability of rock compressive strength can be significantly improved. Compared with the traditional uniaxial compressive strength test of rocks, this method does not require damaging the rock sample, does not require cumbersome coring and sample preparation operations, can directly and quickly detect the compressive strength of rocks in the field, significantly improves the detection efficiency and reduces the detection cost. In addition, the present invention combines the rebound value and audio spectrogram characteristics, overcomes the limitations of single-parameter prediction methods, has a wider application range, significantly reduces the error of prediction results, and is suitable for various rock types and actual engineering requirements.

[0091] The present invention is of great significance in the fields of geotechnical engineering, rock acoustics, and artificial intelligence. On the one hand, the present invention solves the limitations of traditional methods in terms of economy, timeliness, and applicability, providing an efficient and reliable technical means for geotechnical engineering design, construction plan formulation, and on-site rapid assessment. On the other hand, the multi-parameter collaborative analysis method based on deep learning provides a new perspective and technical route for the study of rock mechanical properties, promoting the intelligent and modern development of rock mechanics detection technology, and having broad industry application prospects and popularization value.

[0092] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it; under the idea of the present invention, the technical features in the above embodiments or different embodiments can also be combined, and the steps can be implemented in any order, and there are many other variations in different aspects of the present invention as described above. For the sake of brevity, they are not provided in detail; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for determining the compressive strength of rock based on deep learning rebound and audio, characterized in that: The following steps are involved: S1, collect rock rebound, rebound hammer impact audio recording and rock uniaxial compressive strength test results through large sample tests; S2, data processing of the collected rock rebound data and rebound impact audio set; S3, establishing a neural network calculation model based on the collected rock rebound value, characteristic value of the rebound impact audio spectrum and corresponding rock uniaxial compressive strength data; S4, by collecting the characteristic values ​​of rock rebound value and rebound impact audio spectrum, based on the established neural network calculation model, the characteristic values ​​of rock rebound value and rebound impact audio spectrum obtained from the test are substituted into the prediction model to calculate the rock uniaxial compressive strength of the rock.

2. The measuring method according to claim 1, characterized in that The step S1 comprises: S11, when testing the target, select three representative types of typical rocks, including igneous rocks, sedimentary rocks and metamorphic rocks. The rock samples are mainly core samples. The total number of core samples is 600 groups, of which 400 groups are used to train the neural network calculation model and 200 groups are used to verify the accuracy of the network calculation model; S12, the core samples are numbered, the upper and lower end surfaces of the core cylindrical specimen are ground flat, and three layers are divided according to the center plane of the bottom circle, namely the center, the inner circle, and the surrounding of the axis. Four measuring points are arranged in each layer, and a total of 12 measuring points are arranged for each sample for testing; S13, select a core sample in a natural state and simply fix it to ensure that it will not move during the test, and perform a rebound test at the measuring point. The test angle of the rebound hammer is perpendicular to the rock surface; S14, recording an audio file of the tapping using a recording pen with a bit rate greater than 1500 kbps, and storing the audio file; S15, uniaxial compressive strength test is performed on the core samples.

3. The measuring method according to claim 2, characterized in that In step S13, a suitable rebound hammer is selected for different rocks. If the rebound value is lower than 40, an L-type rebound hammer with an impact energy of 0.735 Nm is used. If the rebound value is higher than 40, an N-type rebound hammer with an impact capacity of 2.207 Nm is used. The test results are recorded.

4. The measuring method according to claim 3, characterized in that In step S15, the rock uniaxial compressive strength testing machine is a hydraulic testing machine with a load of 2000 kN and a minimum resolution of 0.01 kN.

5. The measuring method according to claim 4, characterized in that The step S2 comprises: S21, performing value analysis on the rock rebound data, removing two critical values ​​from the 12 rebound readings of the three layers in the survey area, and taking the arithmetic mean of the remaining 10 rebound readings as the rebound value of the survey area, thereby forming a rock rebound value data set; S22, digitizes the rebound audio.

6. The measuring method according to claim 5, characterized in that The step S22 comprises: S221, using short-time Fourier transform to generate a spectrogram of the audio file of the recorded rebound audio; S222, analyzing the sound spectrogram obtained in step S221, in order to better reflect the attenuation speed of the sound signal, the time length is intercepted with the time point of the start of the impact as the starting point, 0 to 0.5s is defined as the sound wave signal large amplitude section, and 0 to 0.4s is defined as the sound wave signal attenuation section; S223, performing integral area calculation on the segments intercepted at both ends in step S222, the area of ​​the large amplitude segment curve and the horizontal axis is a1, the area of ​​the attenuation segment curve and the horizontal axis is a2, and the ratio of a1 to a2 is defined as the amplitude attenuation coefficient i; S224, the frequencies of the spectrogram acquired in step S221 are mainly concentrated in the range of 0 to 20,000 Hz, with 0 to 5,000 Hz being the low frequency region and 5,000 to 20,000 Hz being the high frequency region; S225, performing integral area calculation on the segments intercepted at both ends in step S224, and using the integration method to find the area between the two segments of the curve and the horizontal axis, the area of ​​the low-frequency region is b1, the area of ​​the high-frequency region is b2, and the ratio of b1 to b2 is defined as the high-low frequency ratio j; S226, recording the amplitude attenuation coefficient i in step S223 and the high-low frequency ratio j in step S225. At this point, all spectral features in a rebound audio have been extracted and digitized to form a rebound audio feature value data set. The audio data set corresponds one-to-one to the rock rebound value and the rock compressive strength value.

7. The measuring method according to claim 6, characterized in that The neural network calculation model in step S3 includes an input layer, a hidden layer and an output layer. Each layer has a number of nodes. The link state of the nodes between layers is reflected by weights. Step S3 includes: S31, accepts data input, the number of nodes in the input layer is 3, corresponding to the rock rebound value, the characteristic value of the rebound impact audio spectrum and the uniaxial compressive strength of the rock; S32, processing the input data; S33, output the calculation result.

8. The measuring method according to claim 7, characterized in that The number of nodes in the hidden layer is set to 3 by default, and each node contains a perceptron, which includes input items, weights, biases, activation functions and outputs. In the forward propagation process, the input data is calculated by the perceptron node and then processed by the activation function to obtain the output result. In the backward propagation process, the output result is compared with the expected result, and the weights of each node on the network are continuously adjusted through multiple iterations to optimize the model performance.

9. The measuring method according to claim 8, characterized in that The rock samples selected in step S11 also include boulders and natural rock outcrops. A relatively flat and fresh outcrop surface is selected as the measurement area. The measurement area is at least 5 cm away from the edge of the rock mass. 16 measurement points are selected in one measurement area, and the distance between each measurement point is greater than 2 cm.

10. The measuring method according to claim 9, characterized in that In step S21, for the block stone or natural rock outcrop, 3 maximum values ​​and 3 minimum values ​​are removed from the 16 rebound values ​​in the measurement area, and the arithmetic mean of the remaining 10 rebound readings is the rebound value of the measurement area.

Images