A method, system, device and medium for calibrating microscopic parameters of parallel bonding model

Abstract

The present invention discloses a parallel bonding model meso-parameter calibration method, system, equipment and medium, which relates to the field of rock mechanics simulation technology. It includes: obtaining the meso-mechanical parameter range of a rock specimen; and performing random interpolation processing on the meso-mechanical parameter range to obtain multiple random meso-mechanical parameters; establishing a parallel bonding model, performing a uniaxial compression test on the parallel bonding model, and obtaining multiple simulated macro-mechanical parameters corresponding to the random meso-mechanical parameters; training a BP neural network through input layer training samples and output layer training samples to obtain a trained BP neural network; inputting actual macro-mechanical parameters into the trained BP neural network to obtain actual meso-mechanical parameters of the rock specimen to be tested after calibration. The present invention simplifies the process of meso-parameter calibration, reduces the time required for meso-parameter calibration, and overcomes the shortcomings of the traditional trial and error method in quantitatively evaluating the matching degree between simulated cracks and laboratory test cracks.

Description

Technical Field

[0001] The invention relates to the technical field of rock mechanics simulation, and in particular to a parallel bonding model microscopic parameter calibration method, system, equipment and medium. Background Art

[0002] Particle Flow Code (PFC) is widely used in geotechnical engineering. Among them, the parallel bond model (PBM) is the most widely used. Calibration of its mesoscopic parameters is the premise for studying the macro-mesomechanical properties of rocks and revealing the mesoscopic mechanism of damage and destruction of rock materials.

[0003] In the prior art, the calibration of microscopic parameters of parallel bonding models is still based on the "trial and error method". Its implementation process includes: first, the initial values ​​of the microscopic parameters are set based on experience or preliminary analysis. Then, the parallel bonding model is run and the difference between its output and the experimental data or benchmark data is observed. According to the difference, the parameter values ​​are adjusted one by one, and after each adjustment, the parallel bonding model is re-run and the performance improvement is evaluated. By continuously iterating this "run-evaluate-adjust" process, the optimal parameter combination is gradually approached. When the model output is highly consistent with the experimental data, the calibration is completed.

[0004] The defects of the above-mentioned prior art are: the prediction of the fractal dimension of the crack is not considered when the trial and error method is used for parameter calibration; the degree of matching between the simulated crack characteristics of the rock after destruction and the indoor test results is mainly based on human qualitative verification, and no quantitative verification is given mathematically. Therefore, there is a certain degree of blindness and uncertainty in the process of parameter iteration, which makes the micro-parameter calibration process cumbersome and time-consuming. Summary of the invention

[0005] Based on this, it is necessary to provide a parallel bonding model micro-parameter calibration method, system, equipment and medium to address the above technical problems.

[0006] The embodiment of the present invention provides a method for calibrating microscopic parameters of a parallel bonding model, comprising:

[0007] Obtaining the range of mesomechanical parameters of the rock specimen; and performing random interpolation processing on the range of mesomechanical parameters to obtain multiple random mesomechanical parameters;

[0008] A parallel bonding model is established based on multiple random mesomechanical parameters, and a uniaxial compression test is performed on the parallel bonding model to obtain multiple simulated macromechanical parameters corresponding to the random mesomechanical parameters.

[0009] The simulated macroscopic mechanical parameters are used as input layer training samples, and the random microscopic mechanical parameters are used as output layer training samples; the BP neural network is trained through the input layer training samples and the output layer training samples to obtain a trained BP neural network;

[0010] The actual macroscopic mechanical parameters of the rock specimen to be tested are input into the trained BP neural network to obtain the actual microscopic mechanical parameters of the rock specimen to be tested after calibration.

[0011] Optionally, the obtaining of a plurality of random micromechanical parameters specifically includes:

[0012] Obtain the value range of random micromechanical parameters of the particle flow model corresponding to the rock specimen, including bond elastic modulus, friction coefficient, bond cohesion, internal friction angle, bond tensile strength and stiffness ratio;

[0013] Based on the value range of random micromechanical parameters, a random function RAND is used to generate multiple random micromechanical parameters, and the formula is:

[0014] X(i,n)=X(i,min)+RAND(X(i,max)-X(i,min))

[0015] Where: X(i, n) is the value of the random micromechanical parameter, i is the micromechanical parameter, n is the sample number, X(i, min) is the minimum value of the sample, and X(i, max) is the maximum value of the sample.

[0016] Optionally, the obtaining of a plurality of simulated macroscopic mechanical parameters corresponding to the random microscopic mechanical parameters specifically includes:

[0017] The uniaxial compression test was carried out on the parallel bonding model to obtain the stress-strain curve and the crack distribution diagram after the specimen was destroyed;

[0018] The elastic modulus, Poisson's ratio and peak stress were calculated from the stress-strain curve;

[0019] The crack distribution diagram of the specimen after destruction is binarized, and then the crack fractal dimension is calculated;

[0020] The simulated macroscopic mechanical parameters were obtained by integrating the elastic modulus, Poisson's ratio, peak stress and crack fractal dimension.

[0021] Optionally, before training the BP neural network using the input layer training samples and the output layer training samples, the data is pre-processed by normalization, and the normalization formula is:

[0022]

[0023] Among them, x newis the normalized value of each physical quantity, x is the original data, x min is the minimum value of each physical quantity, x max The maximum value of the sample data.

[0024] The embodiment of the present invention further provides a parallel bonding model micro-parameter calibration system, comprising:

[0025] The data collection module is used to obtain the range of micromechanical parameters of the rock specimen; and to perform random interpolation processing on the range of micromechanical parameters to obtain multiple random micromechanical parameters;

[0026] A sample determination module is used to establish a parallel bonding model based on multiple random mesoscopic mechanical parameters, perform a uniaxial compression test on the parallel bonding model, and obtain multiple simulated macroscopic mechanical parameters corresponding to the random mesoscopic mechanical parameters;

[0027] The model training module is used to use the simulated macroscopic mechanical parameters as input layer training samples and the random microscopic mechanical parameters as output layer training samples; the BP neural network is trained through the input layer training samples and the output layer training samples to obtain a trained BP neural network;

[0028] The calibration module is used to input the actual macroscopic mechanical parameters of the rock specimen to be tested into the trained BP neural network to obtain the actual microscopic mechanical parameters of the rock specimen to be tested after calibration.

[0029] An embodiment of the present invention further provides a computer device, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the steps of the parallel bonding model micro-parameter calibration method are implemented.

[0030] An embodiment of the present invention further provides a storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the parallel bonding model micro-parameter calibration method are implemented.

[0031] Compared with the prior art, the above-mentioned parallel bonding model micro-parameter calibration method, system, device and medium provided by the embodiment of the present invention have the following beneficial effects:

[0032] The present invention uses simulated macroscopic mechanical parameters as input layer training samples and random microscopic mechanical parameters as output layer training samples; trains the BP neural network through the input layer training samples and the output layer training samples to obtain a trained BP neural network; performs a uniaxial compression test on the parallel bonding model, and uses the obtained crack fractal dimension as part of the simulated macroscopic mechanical parameters to ensure the consistency of the simulated crack morphology with the laboratory test results, thereby overcoming the shortcomings of the traditional trial and error method in quantitatively evaluating the matching degree between simulated cracks and laboratory test cracks. The trained BP neural network is used for prediction, which effectively solves the complexity and time-consuming problems of the traditional trial and error method in the parallel bonding model microscopic parameter calibration process, makes up for the dependence on manual qualitative judgment in the traditional calibration method, avoids the blindness and uncertainty in the parameter iteration process, thereby simplifying the microscopic parameter calibration process and reducing the time required for microscopic parameter calibration.

[0033] In addition, the existing BP neural network model was optimized and trained to improve the efficiency and accuracy of micro-parameter calibration of the parallel bonding model. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 A schematic flow chart of a method for calibrating microscopic parameters of a parallel bonding model provided in an embodiment;

[0035] Figure 2 A schematic diagram of a model structure of a method for calibrating microscopic parameters of a parallel bonding model provided in an embodiment;

[0036] Figure 3 A schematic diagram of a neural network structure of a method for calibrating microscopic parameters of a parallel bonding model provided in one embodiment;

[0037] Figure 4 A schematic diagram of crack binarization of a method for calibrating microscopic parameters of a parallel bonding model provided in an embodiment;

[0038] Figure 5 An activation function diagram of a parallel bonding model micro-parameter calibration method provided in one embodiment;

[0039] Figure 6 A schematic diagram of a post-destruction crack fractal dimension of a parallel bonding model micro-parameter calibration method provided in an embodiment; Figure 6 (a) is a schematic diagram of the crack after failure. Figure 6 (b) is the calculation curve of crack fractal dimension;

[0040] Figure 7 A mean square error curve diagram of a microscopic parameter calibration method of a parallel bonding model provided in an embodiment;

[0041] Figure 8 A neural network performance diagram of a parallel bonding model micro-parameter calibration method provided in one embodiment;

[0042] Fig. 9 A test set performance diagram of a parallel bonding model micro-parameter calibration method provided in an embodiment;

[0043] Fig.10 A stress-strain curve comparison diagram of a parallel bonding model micro-parameter calibration method provided in an embodiment;

[0044] Fig.11 A mechanical parameter comparison diagram of a parallel bonding model microscopic parameter calibration method provided in an embodiment;

[0045] Fig.12 A crack cloud fractal dimension diagram of a parallel bonding model micro-parameter calibration method provided in an embodiment;

[0046] Fig.13 A diagram showing simulation results of a parallel bonding model micro-parameter calibration method provided in an embodiment without considering the fractal dimension of cracks; Fig.13 (a) is the stress-strain curve. Fig.13 (b) in the figure is the crack morphology diagram. DETAILED DESCRIPTION

[0047] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0048] In one embodiment, a method for calibrating microscopic parameters of a parallel bonding model is provided, such as Figure 1 As shown, the method includes:

[0049] The range of mesomechanical parameters of the rock specimen is obtained; and the range of mesomechanical parameters is randomly interpolated to obtain multiple random mesomechanical parameters.

[0050] A parallel bonding model is established based on multiple random mesomechanical parameters. A uniaxial compressive test is carried out on the parallel bonding model to obtain multiple simulated macromechanical parameters corresponding to the random mesomechanical parameters.

[0051] The simulated macroscopic mechanical parameters are used as input layer training samples, and the random microscopic mechanical parameters are used as output layer training samples; the BP neural network is trained by the input layer training samples and the output layer training samples to obtain a trained BP neural network.

[0052] The actual macroscopic mechanical parameters of the rock specimen to be tested are input into the trained BP neural network to obtain the actual microscopic mechanical parameters of the rock specimen to be tested after calibration.

[0053] Exemplarily, the technical solution includes three parts: determining the BP neural network structure, establishing training samples and determining the BP neural network parameters.

[0054] 1. Determine the BP neural network structure

[0055] Use Deep Learning Toolbox in MATLAB to create, learn, and train neural networks.

[0056] Schematic diagram of parallel bonding model Figure 2 As shown in the figure, the mechanical and deformation behavior characteristics of the parallel bonding model under external loads mainly depend on six microscopic mechanical parameters: bonding elastic modulus, friction coefficient, bonding cohesion, internal friction angle, bonding tensile strength, and stiffness ratio. Therefore, these six microscopic mechanical parameters are used as output layer training samples.

[0057] Elastic modulus, Poisson's ratio, and peak strength are often used to evaluate the macroscopic mechanical characteristics of rocks. Cracks after rock destruction have fractal characteristics, and fractal dimension can better digitize irregular cracks. Therefore, elastic modulus, Poisson's ratio, peak strength, and crack fractal dimension are used as macroscopic mechanical parameters, and macroscopic mechanical parameters are used as input layer training samples.

[0058] The structure of BP neural network is shown as follows Figure 3 As shown in Figure 1, the main principle of the hidden layer is to adjust the weight and bias of each hidden node according to the prediction error through the error back propagation algorithm. Specifically, when there is an error between the prediction result of the output layer and the true value, the error will be propagated layer by layer in the opposite direction of the weight to update the weight and bias of the hidden layer nodes.

[0059] 2. Establishment of training samples

[0060] (1) Determine the random micromechanical parameters

[0061] ① Through literature search, the value ranges of six random micromechanical parameters of the particle flow model corresponding to the rock specimens, including bond elastic modulus, friction coefficient, bond cohesion, internal friction angle, bond tensile strength, and stiffness ratio, were obtained.

[0062] ② On this basis, based on the value range of random mesomechanical parameters, a random function RAND is used to generate multiple random mesomechanical parameters, and the formula is:

[0063] X(i,n)=X(i,min)+RAND(X(i,max)-X(i,min))

[0064] Where X(i, n) is the value of the random mesomechanical parameter; i is the mesomechanical parameter; n is the sample number; X(i, min) is the minimum value of the sample; X(i, max) is the maximum value of the sample.

[0065] (2) Determine the simulation macroscopic mechanical parameters

[0066] A. Obtain the random mesomechanical parameters of the rock specimen, construct a parallel bonding model based on the random mesomechanical parameters, and conduct a uniaxial compression test on the parallel bonding model in accordance with the Engineering Rock Test Methods to obtain the stress-strain curve and the crack distribution diagram of the specimen after failure.

[0067] B. Calculate the elastic modulus, Poisson's ratio and peak stress through the stress-strain curve, such as Figure 4 As shown in the figure, the crack distribution map after the specimen is damaged is binarized, and then the crack fractal dimension is calculated to better analyze the crack shape and contour. On this basis, it is recommended to use imageJ software to convert the crack distribution map to 8bit and binarize it, and then use Fractal BoxCount to calculate the crack fractal dimension.

[0068] C. The simulated macroscopic mechanical parameters are obtained by combining elastic modulus, Poisson's ratio, peak stress and crack fractal dimension.

[0069] (3) Data normalization

[0070] The dimensions of each physical quantity are different. In order to reduce the training error, the data is preprocessed by normalization before training the BP neural network through the input layer training samples and the output layer training samples. The normalization formula is:

[0071]

[0072] In the formula, x new is the normalized value of each physical quantity, x is the original data, x min is the minimum value of each physical quantity, x max The maximum value of the sample data.

[0073] (4) Training process

[0074] The input layer training samples and the output layer training samples are input into the BP neural network, and the estimated micromechanical parameters are obtained according to the input layer training samples.

[0075] The estimated mesomechanical parameters are compared with the random mesomechanical parameters used as the output layer training samples to obtain the loss function. Based on the gradient of the loss function, the gradient of each layer's weights and biases is calculated through network back propagation, so that the neural network can learn how to adjust the parameters to reduce prediction errors.

[0076] Repeat the iteration until the error between the estimated micromechanical parameters output by the output layer and the training samples of the output layer is less than the set threshold, and a trained BP neural network is obtained.

[0077] 3. BP neural network parameter selection

[0078] When calibrating the mesoscopic parameters, the actual macroscopic mechanical parameters of the rock specimen to be tested are obtained, and the actual macroscopic mechanical parameters are input into the trained BP neural network to obtain the actual mesoscopic mechanical parameters of the rock specimen to be tested after calibration. The parameter settings of the trained BP neural network mainly include the neuron parameters and the hidden layer structure parameter settings.

[0079] (1) Neuron parameter setting

[0080] ① Training algorithm selection

[0081] The Levenberg-Marquardt algorithm is recommended. This algorithm combines the advantages of the Gauss-Newton method and the gradient descent method, and has both high convergence speed and global convergence properties.

[0082] ②Activation function selection

[0083] It is recommended to use the ReLu activation function for nonlinear transformation, such as Figure 5 As shown. This function is a piecewise linear function, when x>0, its derivative is 1; when x≤0, its derivative is 0.

[0084] ③Weight initialization

[0085] It is recommended to use the equal-proportional coefficient initialization method and initialize the weights according to the following formula.

[0086] IW sr =rand()*1 / (R+1)

[0087] LW ks =rand()

[0088] Where: IW sr is the initial weight between the input and hidden layer; LW ks is the initial weight between the hidden layer and the output layer; rand() is a random number in [0,1), and R is the number of elements in the input vector.

[0089] ④Learning rate setting

[0090] It is recommended to use the optimal initial learning rate determination method proposed by Leslie N. Smith. Initially, set a very small initial learning rate and record the mean square error. Repeat the above steps in multiples of 10. According to the mean square error curve, the optimal learning rate is obtained.

[0091] (2) Hidden layer structure parameter setting

[0092] At present, the number of hidden layer units is often determined by the following formula:

[0093]

[0094] n1=log2n

[0095] Where: n1 is the number of hidden layer units; n is the number of input units; m is the number of output units; k is the number of samples; i is a constant between [0, n]; a is a constant between [1, 10].

[0096] According to the calculated number of hidden layer units, the range of the number of hidden layer units can be known by rounding the minimum value in the opposite direction and the maximum value in the positive direction. The number of hidden layer units in this range was trained respectively, and the mean square error of the training set, the mean square error of the validation set and the mean square error of the test set were compared to determine the optimal number of hidden layer units.

[0097] 4. Conclusion

[0098] The present invention uses simulated macroscopic mechanical parameters as input layer training samples and random microscopic mechanical parameters as output layer training samples; trains the BP neural network through the input layer training samples and the output layer training samples to obtain a trained BP neural network; performs a uniaxial compression test on the parallel bonding model, and uses the obtained crack fractal dimension as part of the simulated macroscopic mechanical parameters to ensure the consistency of the simulated crack morphology with the laboratory test results, thereby overcoming the shortcomings of the traditional trial and error method in quantitatively evaluating the matching degree between simulated cracks and laboratory test cracks. The trained BP neural network is used for prediction, which effectively solves the complexity and time-consuming problems of the traditional trial and error method in the parallel bonding model microscopic parameter calibration process, makes up for the dependence on manual qualitative judgment in the traditional calibration method, avoids the blindness and uncertainty in the parameter iteration process, thereby simplifying the microscopic parameter calibration process and reducing the time required for microscopic parameter calibration.

[0099] In addition, the existing BP neural network model was optimized and trained to improve the efficiency and accuracy of micro-parameter calibration of the parallel bonding model.

[0100] 5. Embodiment

[0101] 5.1 Indoor test

[0102] Through uniaxial compression test on three standard granite specimens with a diameter of 50mm and a height of 100mm, the loading rate was set to 0.02mm / s. The test results show that the average peak strength of granite is 83.16MPa, the average elastic modulus is 71.02GPa, and the average Poisson's ratio is 0.225. A specimen is selected as the research object, and the crack fractal dimension of the specimen after failure is calculated to be 1.3924, such as Figure 6 shown. Figure 6 (a) is a schematic diagram of the crack after failure. Figure 6 (b) in the figure is the calculation curve of crack fractal dimension.

[0103] 5.2 Training Sample Creation

[0104] The particle parameters of the present invention are set as follows: the minimum particle size is 0.5 mm, the maximum to minimum particle size ratio is 1.5, and the density is 2750 kg / m 3 ; The particle arrangement characteristics are kept consistent by fixing the random seed number, and its value is 10001.

[0105] The ranges of values ​​of six main microscopic parameters, such as elastic modulus, cohesion, and friction coefficient, of the granite parallel bonding model in the field of geotechnical engineering in the past 10 years are statistically analyzed (see Table 1). Among them, the range of elastic modulus is 30-60GPa, the range of friction coefficient is 0.4-0.7, the range of cohesion is 50-100MPa, the range of internal friction angle is 30-60°, the range of tensile strength is 50-100MPa, and the range of normal to tangential stiffness ratio is 1-3.

[0106] On this basis, the random function RAND in Excel software was used to generate 200 sets of samples for numerical simulation. -1 Each sample was subjected to a uniaxial compression test, and the rock elastic modulus, Poisson's ratio and peak strength were obtained based on the stress-strain curve. The process is as follows: on the stress-strain curve, the ratio of stress to strain is usually measured in the straight line segment (i.e., the elastic stage) to obtain the elastic modulus, the ratio of transverse strain to longitudinal strain to obtain the Poisson's ratio, and the peak strength is the maximum stress value reached by the rock on the stress-strain curve.

[0107] The crack distribution map was converted to 8-bit and binarized using imageJ software, and the fractal dimension was calculated using Fractal Box Count.

[0108] Table 1. Parallel bonding model parameter calibration training samples and value ranges

[0109]

[0110] According to Table 1, the random micro-parameter samples generated by Excel and their corresponding elastic modulus, Poisson's ratio, peak strength and crack fractal dimension are normalized and used as training samples.

[0111] 5.3 BP neural network parameter selection

[0112] (1) Neuron parameter setting

[0113] The present invention adopts Levenberg-Marquardt algorithm for training, selects ReLu activation function as activation function, and adopts equal-proportional coefficient type initialization method for weight initialization.

[0114] According to the optimal initial learning rate determination method proposed by Leslie N. Smith, the present invention calculates the learning rate to be 1×10 -5 ~1×10 -1 The mean square error of the training samples (such as Figure 7 As shown in the figure, we can see that the learning rate corresponding to the minimum mean square error of the training sample is 1×10 -3 ~1×10 -2 Considering the learning rate, the initial learning rate of the present invention is 0.01.

[0115] (2) Hidden layer structure parameter setting

[0116] The training sample parameters are substituted into formula (5) to formula (7) to determine the number of hidden layer units, and the calculated numbers of hidden layer units are 13, 5.16, and 2.5, respectively. By rounding the minimum value in the opposite direction and the maximum value in the positive direction, it can be seen that the number of hidden layer units is between 3 and 13. For the number of hidden layer units of 3, 4, 5...13, respectively, Table 2 shows the mean square error of the training samples, the mean square error of the validation set, and the mean square error of the test set. When the number of hidden layer units is 12, the mean square error of the training set is 0.0020 and the mean square error of the validation set is 0.0024, both of which are minimum values. The mean square error of the test set is 0.0051, which is close to the average mean square error of the test set of 0.0058. Therefore, the number of hidden layer units of the present invention is determined to be 12.

[0117] Table 2 Relationship between the number of hidden layer units and mean square error

[0118]

[0119]

[0120] 5.4 Neural Network Performance Evaluation

[0121] Based on the setting of neuron parameters and hidden layer structure parameters, the neural network is trained using the train function of matlab software. The number of training sets is 140, the number of validation sets is 30, and the number of test sets is 30. Figure 8 As shown in the figure, as the number of iterations increases, the mean square error of the training set decreases from 0.8128 to 0.0020; when the number of iterations is less than 5, the mean square errors of the test set and the validation set both decrease as the number of iterations increases; when the number of iterations is greater than 5, as the number of iterations increases, the mean square error of the test set generally stabilizes at 0.0081, but the mean square error of the validation set increases from 0.0027 to 0.0061.

[0122] The test set does not participate in the learning and training process, such as Fig. 9 As shown, the fitting results of the neural network output data and the expected output data are given when the number of iterations is 5. From the figure, it can be seen that the slope of the fitting line is 0.9666, which is approximately 1, and the error between the test set output data and the expected output data is about 3.34%.

[0123] 5.5 Evaluation of Neural Network Prediction Structure

[0124] The four parameters of the specimen, namely, peak strength, elastic modulus, Poisson’s ratio and post-fractal dimension of the crack, were normalized and input into the neural network. After the output layer results were denormalized, the actual mesomechanical parameters after calibration were obtained, as shown in Table 3.

[0125] Table 3 Neural network prediction denormalization results

[0126]

[0127]

[0128] PFC simulation was carried out according to the predetermined particle parameters and the microscopic parameters in Table 3. Usually, the stress-strain curve obtained by simulation can be translated, that is, the strain value in the compaction stage is deducted, and then compared with the stress-strain curve obtained by indoor test. Fig.10 As shown in the figure, the stress-strain curve after translation is basically consistent with the stress-strain curve of the indoor test, indicating that the microscopic parameters of the above parallel bonding model are reasonable to a certain extent.

[0129] Depend on Fig.11As shown in Table 4, the peak strength obtained from the indoor test is 80.80MPa, and the peak strength obtained from the simulation is 82.45MPa, with a relative error of 2.04%; the peak strain obtained from the indoor test after deducting the compaction stage is 1.23%, and the peak strain obtained from the simulation is 1.22%, with a relative error of 0.80%; the elastic modulus obtained from the indoor test is 66.50GPa, and the elastic modulus obtained from the simulation is 63.08GPa, with a relative error of 0.51%; the Poisson's ratio obtained from the indoor test is 0.221, and the Poisson's ratio obtained from the simulation is 0.22, with a relative error of 0.00%.

[0130] Table 4 Comparison of mechanical parameters of predicted results

[0131]

[0132] like Fig.12 As shown in the figure, after the specimen is destroyed, there are mainly three through-tension cracks along the loading direction, and the crack fractal dimension is 1.3920. The simulation shows that there are multiple vertical cracks, which is similar to the indoor test results, and the crack fractal dimension is 1.3928.

[0133] 5.6 Comparison of parameter calibration results without considering the crack fractal dimension

[0134] Depend on Fig.13 From Table 5, we can see that Fig.13 (a) is the stress-strain curve. Fig.13 (b) in the figure is the crack morphology diagram.

[0135] Table 5 Macroscopic mechanical parameters obtained by simulation without considering the fractal dimension of crack

[0136]

[0137]

[0138] When the fractal dimension of cracks after rock failure is not considered, the peak strength obtained by simulation is 83.50MPa, and the relative error with the indoor test results is 3.30%; the peak strain obtained by test after deducting the compaction stage is 1.23%, and the peak strain obtained by simulation is 1.21%, with a relative error of 1.60%; the elastic modulus obtained by simulation is 71.56GPa, and the relative error with the indoor test results is 0.76%; the Poisson's ratio obtained by simulation is 0.22, which is the same as the indoor test results.

[0139] Compared with the simulation results of the fractal dimension of the crack after considering rock failure, the relative errors of the peak strength and peak strain elastic modulus increased by 1.26%, 0.80% and 0.25% respectively. More importantly, the simulated cracks are mainly distributed in the upper and middle parts of the specimens, and no obvious through cracks are formed, which is significantly different from the indoor test results. At the same time, the fractal dimension of the crack obtained by simulation is 1.4118, which is 0.0198 lower than the indoor test result. The difference in fractal dimension accounts for about 9.90% of the variation range of the crack fractal dimension.

[0140] Based on the same inventive concept, the present invention also provides a parallel bonding model micro-parameter calibration system, comprising:

[0141] The data collection module is used to obtain the range of micromechanical parameters of the rock specimen; and to perform random interpolation processing on the range of micromechanical parameters to obtain multiple random micromechanical parameters;

[0142] A sample determination module is used to establish a parallel bonding model based on multiple random mesoscopic mechanical parameters, perform a uniaxial compression test on the parallel bonding model, and obtain multiple simulated macroscopic mechanical parameters corresponding to the random mesoscopic mechanical parameters;

[0143] The model training module is used to use the simulated macroscopic mechanical parameters as input layer training samples and the random microscopic mechanical parameters as output layer training samples; the BP neural network is trained through the input layer training samples and the output layer training samples to obtain a trained BP neural network;

[0144] The calibration module is used to input the actual macroscopic mechanical parameters of the rock specimen to be tested into the trained BP neural network to obtain the actual microscopic mechanical parameters of the rock specimen to be tested after calibration.

[0145] Based on the same inventive concept, the present invention also provides a computer device, including a memory and a processor, wherein a computer program is stored in the memory, and when the processor executes the computer program, the steps of the parallel bonding model micro-parameter calibration method are implemented.

[0146] Based on the same inventive concept, the present invention also provides a computer-readable storage medium containing instructions, and a computer program is stored on the storage medium. For example, a memory containing instructions, the above instructions can be executed by a processor of a computer device to complete the above method. For example, a non-temporary computer-readable storage medium can be a ROM, a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, and an optical data storage device. When the computer program is executed by the processor, the steps in the embodiment of the parallel bonding model micro-parameter calibration method can be implemented. The specific implementation method can be found in the method embodiment, which will not be repeated here.

[0147] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A method for calibrating microscopic parameters of a parallel bonding model, characterized in that: include: Obtain the range of micromechanical parameters of rock specimens; And the range of mesomechanical parameters is randomly interpolated to obtain multiple random mesomechanical parameters; A parallel bonding model is established based on multiple random mesomechanical parameters, and a uniaxial compression test is performed on the parallel bonding model to obtain multiple simulated macromechanical parameters corresponding to the random mesomechanical parameters. Among them, a plurality of simulated macroscopic mechanical parameters corresponding to the random microscopic mechanical parameters are obtained, which specifically include: A uniaxial compression test was carried out on the parallel bonding model to obtain the stress-strain curve and the crack distribution diagram after the specimen was destroyed; the elastic modulus, Poisson's ratio and peak stress were calculated through the stress-strain curve; the crack distribution diagram after the specimen was destroyed was binarized and the crack fractal dimension was calculated; the elastic modulus, Poisson's ratio, peak stress and crack fractal dimension were combined to obtain multiple simulated macroscopic mechanical parameters corresponding to the random mesomechanical parameters; The simulated macroscopic mechanical parameters are used as input layer training samples, and the random microscopic mechanical parameters are used as output layer training samples; the BP neural network is trained through the input layer training samples and the output layer training samples to obtain a trained BP neural network; The actual macroscopic mechanical parameters of the rock specimen to be tested are input into the trained BP neural network to obtain the actual microscopic mechanical parameters of the rock specimen to be tested after calibration.

2. A method for calibrating microscopic parameters of a parallel bonding model as claimed in claim 1, characterized in that: The multiple random micromechanical parameters are obtained, which specifically include: Obtain the value range of random micromechanical parameters of the particle flow model corresponding to the rock specimen, including bond elastic modulus, friction coefficient, bond cohesion, internal friction angle, bond tensile strength and stiffness ratio; Based on the value range of random micromechanical parameters, a random function RAND is used to generate multiple random micromechanical parameters, and the formula is: ； in: X ( i , n ) is the value of the random micromechanical parameter, i is the micromechanical parameter, n is the sample number, X ( i , min ) is the minimum value of the sample, X ( i , max ) is the maximum value of the sample.

3. A method for calibrating microscopic parameters of a parallel bonding model as claimed in claim 1, characterized in that: Before training the BP neural network through the input layer training samples and the output layer training samples, the data is pre-processed by normalization, and the normalization formula is: ； in, x new is the normalized value of each physical quantity, x is the original data, x min is the minimum value of each physical quantity, x max The maximum value of the sample data.

4. A parallel bonding model mesoscopic parameter calibration system based on a parallel bonding model mesoscopic parameter calibration method according to any one of claims 1 to 3, characterized in that: include: Data collection module, used to obtain the range of micromechanical parameters of rock specimens; And the range of mesomechanical parameters is randomly interpolated to obtain multiple random mesomechanical parameters; A sample determination module is used to establish a parallel bonding model based on multiple random mesoscopic mechanical parameters, perform a uniaxial compression test on the parallel bonding model, and obtain multiple simulated macroscopic mechanical parameters corresponding to the random mesoscopic mechanical parameters; A model training module is used to use simulated macroscopic mechanical parameters as input layer training samples and random microscopic mechanical parameters as output layer training samples; The BP neural network is trained through input layer training samples and output layer training samples to obtain a trained BP neural network; The calibration module is used to input the actual macroscopic mechanical parameters of the rock specimen to be tested into the trained BP neural network to obtain the actual microscopic mechanical parameters of the rock specimen to be tested after calibration.

5. A computer device comprising a memory and a processor, wherein a computer program is stored in the memory, wherein: When the processor executes the computer program, the steps of the parallel bonding model micro-parameter calibration method described in any one of claims 1-3 are implemented.

6. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the parallel bonding model micro-parameter calibration method described in any one of claims 1 to 3 are implemented.

Images