Microcosmic parameter calibration method, device and equipment of brittle material and medium

Abstract

The invention relates to a microscopic parameter calibration method, device and equipment of a brittle material and a medium, and belongs to the technical field of discrete element numerical simulation, the microscopic parameter calibration method of the brittle material comprises the following steps: establishing a bonded particle model of the brittle material, and determining a plurality of microscopic parameters based on the bonded particle model; determining the value range of the microscopic parameters and the simulation time step length; a Plackett-Burman method is adopted to carry out dimension reduction on the microscopic parameters, and saliency parameters are obtained; obtaining a plurality of groups of stress-strain data based on the value range, the simulation time step length and the significance parameters; training the initial neural network model by adopting the multiple groups of stress-strain data to obtain a completely trained neural network model; the stress-strain curve to be calibrated is input into the neural network model, the value of the significance parameter is obtained, the calibration of the microscopic parameter is completed, and the efficiency and accuracy of the microscopic parameter calibration are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of computer software engineering, and in particular to a method, device, equipment and medium for calibrating microscopic parameters of brittle materials. Background Art

[0002] The discrete element method (DEM) has considerable advantages in simulating the damage and cracking processes of brittle materials. The bonded particle model based on the DEM has been widely used to simulate the fracture behavior of brittle or quasi-brittle materials such as ceramics, iron ore, corn kernels, and wheat straw. In DEM analysis, accurate model parameters are a necessary prerequisite for ensuring the accuracy of DEM simulation.

[0003] Uniaxial compression experiments are widely used to calibrate micro-parameters. The response obtained by simulating with calibrated micro-parameters should be consistent with the response obtained from the uniaxial compression experiment. Initially, the trial-and-error method was often used to calibrate the micro-parameters of the bonded particle model. This experience-based method is difficult to capture the mapping relationship between the micro-parameters and the macro-properties of the material. In order to overcome the shortcomings of the trial-and-error method, various parameter calibration methods have been proposed, including the response surface method, deep neural network method and convolutional neural network method. When dealing with calibration problems, the response surface method, deep neural network method and convolutional neural network method first establish a mapping relationship between the macro-properties and the micro-parameters of the bonded particle model, and then predict the micro-parameters of the bonded particle model based on the established mapping relationship.

[0004] However, response surface method, deep neural network method and convolutional neural network method have problems of poor calibration accuracy, narrow application scope and low efficiency when dealing with calibration problems. Summary of the Invention

[0005] In view of this, it is necessary to provide a method, device, equipment and medium for calibrating the microscopic parameters of brittle materials to solve the technical problems of poor calibration accuracy and low efficiency when calibrating microscopic parameters.

[0006] In order to solve the above problems, the present invention provides a method for calibrating microscopic parameters of brittle materials, comprising:

[0007] establishing a cohesive particle model of the brittle material, and determining a plurality of microscopic parameters based on the cohesive particle model;

[0008] Determining the value range of the microscopic parameters and the simulation time step;

[0009] Plackett-Burman method is used to reduce the dimension of the microscopic parameters to obtain significant parameters;

[0010] Obtaining multiple sets of stress and strain data based on the value range, the simulation time step, and the significance parameter;

[0011] Using the multiple sets of stress-strain data to train the initial neural network model to obtain a fully trained neural network model;

[0012] The stress-strain curve to be calibrated is input into the neural network model to obtain the value of the significance parameter.

[0013] In one possible implementation, the microscopic parameters include contact parameters and parallel bond parameters. The contact parameters include the particle-particle collision recovery coefficient, the particle-particle static friction coefficient, and the particle-particle rolling friction coefficient. The parallel bond parameters include the normal stiffness per unit area of the parallel bond, the tangential stiffness per unit area of the parallel bond, the critical normal stress of the parallel bond, and the critical shear stress of the parallel bond.

[0014] In a possible implementation, determining the value range of the microscopic parameter includes:

[0015] Step 1: Setting an initial simulation time step, wherein the initial simulation time step is a preset multiple of the Rayleigh time step;

[0016] Step 2: presetting an initial value range of the micro-parameters, and adjusting the micro-parameters based on the initial value range to obtain multiple sets of discrete parameters;

[0017] Step 3: determining non-microscopic parameters, and simulating the multiple sets of discrete parameters based on the non-microscopic parameters and the initial simulation time step to obtain multiple elastic modulus values and multiple failure strain values, and determining maximum and minimum thresholds of the elastic modulus and the failure strain based on the multiple elastic modulus values and the multiple failure strain values;

[0018] Step 4: Obtain the elastic modulus value and failure strain value of the reference uniaxial compression experiment, and judge the elastic modulus value and failure strain value of the reference uniaxial compression experiment based on the maximum threshold value and the minimum threshold value. When the elastic modulus value and failure strain value of the reference uniaxial compression experiment are not between the maximum threshold value and the minimum threshold value, repeat steps 2 to 3 until the elastic modulus value and failure strain value of the reference uniaxial compression experiment are between the maximum threshold value and the minimum threshold value. When the elastic modulus value and failure strain value of the reference uniaxial compression experiment are between the maximum threshold value and the minimum threshold value, use the initial value range as the value range of the microscopic parameter.

[0019] In a possible implementation, determining the simulation time step includes:

[0020] Presetting a simulation time step based on the Rayleigh time step;

[0021] Adjust the preset simulation time step to obtain multiple sets of discrete element simulation time steps;

[0022] Based on the non-microscopic parameters, microscopic parameters and multiple sets of discrete element simulation time steps, multiple failure strain values are obtained through multiple simulations;

[0023] A failure strain error threshold is preset, and a failure strain error is calculated based on the failure strain value. When the failure strain error is less than the failure strain error threshold, a simulation time step is determined according to a maximum time step criterion, wherein the failure strain error is an indicator for evaluating the correctness of the simulation.

[0024] In a possible implementation, obtaining multiple sets of stress-strain data based on the value range, the simulation time step, and the significance parameter includes:

[0025] Adjusting the significance parameter based on the value range to obtain multiple groups of discrete parameters of the significance parameter;

[0026] Determine non-significant parameters, simulate multiple groups of discrete parameters of the significant parameters based on the non-microscopic parameters, the non-significant parameters and the simulation time step, and obtain multiple groups of stress and strain data.

[0027] In one possible implementation, the training of the initial neural network model using the multiple sets of stress-strain data to obtain a fully trained neural network model includes:

[0028] Establishing an initial neural network model, wherein the initial neural network model includes an encoder and a decoder, the encoder includes a gated logic unit and an attention mechanism, the decoder includes a fully connected neural network, the input of the initial neural network model is the multiple sets of stress and strain data, and the output of the initial neural network model is the significance parameter;

[0029] Dividing the multiple sets of stress-strain data into a training set and a test set;

[0030] Using the multiple sets of stress-strain data to train the initial neural network model to obtain the coefficient of determination of the training set and the coefficient of determination of the test set;

[0031] When the coefficient of determination of the training set and the coefficient of determination of the test set are both greater than the coefficient threshold, a fully trained neural network model is obtained.

[0032] In a possible implementation, the calculation formula of the gating logic unit is:

[0033] Z t =σ(W (Z) x t+U (Z) h t-1 ),

[0034] r t =σ(W (r) x t +U (r) h t-1 ),

[0035]

[0036]

[0037] Among them, h t is the output of the t-th gate logic unit, x t is the input of the t-th gate logic unit, h t-1 is the output of the t-1th gated logic unit, σ is the Sigmoid activation function, tanh is the Tanh activation function, z t is the intermediate variable, r t is an intermediate variable, is the intermediate variable, U (z) 、W (z) 、U (r) 、W (r) , U and W are trainable parameter matrices;

[0038] The calculation formula of the attention mechanism is:

[0039]

[0040] Among them, y is the output of the attention mechanism, h t represents the output of the tth gated logic unit, α t is the influence of the output of the t-th gated logic unit on the next layer, and n is the number of gated logic units;

[0041] The calculation formula of the determination coefficient is:

[0042]

[0043] Among them, R 2 is the coefficient of determination, y true is the actual value of the dependent variable, y pre is the predicted value of the dependent variable, is the mean of the actual values of the dependent variable.

[0044] On the other hand, the present invention also provides a microscopic parameter calibration device for brittle materials, comprising:

[0045] a microscopic parameter determination module, configured to establish a bonded particle model of the brittle material and determine a plurality of microscopic parameters based on the bonded particle model;

[0046] A range and time step determination module, used to determine the value range of the microscopic parameters and the simulation time step;

[0047] A significance parameter acquisition module, used for performing dimension reduction on the microscopic parameters using the Plackett-Burman method to obtain significance parameters;

[0048] A stress and strain data acquisition module, configured to obtain multiple sets of stress and strain data based on the value range, the simulation time step, and the significance parameter;

[0049] A neural network model training module is used to train the initial neural network model using the multiple sets of stress and strain data to obtain a fully trained neural network model;

[0050] The significance parameter calibration module is used to input the stress-strain curve to be calibrated into the neural network model to obtain the value of the significance parameter.

[0051] On the other hand, the present invention also provides an electronic device, comprising: a processor and a memory;

[0052] The memory stores a computer-readable program executable by the processor;

[0053] When the processor executes the computer-readable program, the steps in the microscopic parameter calibration method of the brittle material as described above are implemented.

[0054] On the other hand, the present invention also provides a computer-readable storage medium, which stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps in the micro-parameter calibration method of brittle materials as described above.

[0055] The beneficial effects of the present invention are as follows: determining a simulation time step, proposing a simulation time step determination method, which can simultaneously take into account the efficiency and accuracy of parameter calibration, performing dimensionality reduction on microscopic parameters through a Plackett-Burman experimental design method, obtaining significant parameters, greatly reducing the number of simulations, thereby reducing computing costs and improving calibration efficiency, simulating significant parameters based on the value range of microscopic parameters and the simulation time step, obtaining corresponding multiple groups of stress and strain data, providing data support for a neural network model, establishing a neural network model, using multiple groups of stress and strain data to train the neural network model to obtain a fully trained neural network model, establishing a mapping relationship between macroscopic characteristics and microscopic parameters of a geometric model, using the fully trained neural network model to predict a stress-strain curve to be calibrated, obtaining the value of the significant parameter, predicting the significant parameter of the microscopic parameter based on the established mapping relationship, completing the calibration of the microscopic parameter, and improving the efficiency and accuracy of the microscopic parameter calibration. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 A flow chart of an embodiment of a method for calibrating microscopic parameters of brittle materials provided by the present invention;

[0057] Figure 2 This is a diagram of the architecture of a neural network model for the microscopic parameter calibration method of brittle materials provided by the present invention;

[0058] Figure 3 A schematic diagram comparing the predicted results of the microscopic parameter calibration method for brittle materials provided by the present invention with the stress-strain curve of a reference experiment;

[0059] Figure 4 A schematic structural diagram of an embodiment of a device for calibrating microscopic parameters of brittle materials provided by the present invention;

[0060] Figure 5 This is a schematic structural diagram of an embodiment of an electronic device provided by the present invention. DETAILED DESCRIPTION

[0061] The preferred embodiments of the present invention will be described in detail below in conjunction with the accompanying drawings, wherein the accompanying drawings constitute a part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, and are not used to limit the scope of the present invention.

[0062] The present invention discloses a method, device, electronic device, and storage medium for calibrating microscopic parameters of brittle materials, which can be used in a computer. The method, device, or computer-readable storage medium involved in the present invention can be integrated with the above-mentioned device or can be relatively independent.

[0063] A specific embodiment of the present invention discloses a method for calibrating microscopic parameters of brittle materials, which can be executed by a computer, specifically by one or more processors of the computer. Figure 1 This is a flow chart of the microscopic parameter calibration method of brittle materials provided by the embodiment of the present invention. Figure 1 , the microscopic parameter calibration methods of brittle materials include:

[0064] S101, establishing a bonded particle model of brittle materials, and determining multiple microscopic parameters based on the bonded particle model;

[0065] S102, determining the value range of microscopic parameters and the simulation time step;

[0066] S103, using the Plackett-Burman method to reduce the dimension of micro parameters and obtain significant parameters;

[0067] S104, obtaining multiple sets of stress and strain data based on the value range, simulation time step, and significance parameters;

[0068] S105, using multiple sets of stress and strain data to train the initial neural network model to obtain a fully trained neural network model;

[0069] S106: Input the stress-strain curve to be calibrated into the neural network model to obtain the value of the significance parameter.

[0070] During implementation, first, a bonded particle model of brittle materials is established based on the uniaxial compression experiment. Based on the bonded particle model, multiple microscopic parameters are determined, and non-microscopic parameters are specified. The initial simulation time step is set, where the initial simulation time step is a preset multiple of the Rayleigh time step to ensure the correctness of the simulation. Based on the initial simulation time step, non-microscopic parameters, and multiple microscopic parameters, the value range of the microscopic parameters is determined; the simulation time step is determined to take into account both the correctness of the simulation and the simulation speed; secondly, the Plackett-Burman method is used to reduce the dimensionality of the microscopic parameters to obtain the significance parameters and Non-significant parameters are adjusted based on the value range of micro parameters and the simulation time step, and multiple sets of stress-strain data are obtained through multiple simulations; finally, a neural network model based on gated logic units and attention mechanism is established, and multiple sets of stress-strain data are divided into training sets and test sets. The neural network model is trained using multiple sets of stress-strain data to obtain a fully trained neural network model; the stress-strain curve to be calibrated is input into the fully trained neural network model to obtain the value of the significant parameter, complete the calibration of the micro parameters, and improve the efficiency and accuracy of the micro parameter calibration.

[0071] Compared with the prior art, the micro-parameter calibration method for brittle materials provided in this embodiment establishes a bonded particle model of brittle materials based on a uniaxial compression experiment, determines multiple micro-parameters based on the bonded particle model, specifies non-micro-parameters of the bonded particle model, sets an initial simulation time step to ensure simulation accuracy, obtains multiple elastic moduli and multiple failure strains through simulation based on the initial time step and multiple micro-parameters, determines the maximum threshold and minimum threshold of the elastic modulus and failure strain, judges the elastic modulus value and failure strain value of the reference uniaxial compression experiment based on the maximum threshold and minimum threshold, and determines the value range of the micro-parameters; determines the simulation time step to take into account both simulation accuracy and simulation speed, and adopts Plackett-Burman The method reduces the dimensionality of micro parameters, obtains the significant parameters and non-significant parameters of the micro parameters, and determines the value of the non-significant parameters. The significant parameters are calibrated by the subsequent neural network, which reduces the number of simulations, thereby reducing the computational cost and improving the calibration efficiency. Based on the value range of the micro parameters, the significant parameters of the micro parameters are adjusted. Multiple sets of stress and strain data are obtained through simulation to provide data support for the neural network model. A neural network model based on gated logic units and attention mechanism is established. The neural network model is trained until the coefficient of determination on the training set and the test set is greater than the coefficient threshold. The stress-strain curve to be calibrated is input into the trained neural network model to obtain a set of significant parameters, completing the calibration of the micro parameters, and improving the efficiency and accuracy of the micro parameter calibration.

[0072] In some embodiments, in step S101, a bonded particle model of brittle material is established based on a reference uniaxial compression experiment, multiple micro parameters are determined based on the bonded particle model, and non-micro parameters are specified, wherein the non-micro parameters are parameters that can be directly determined, and the micro parameters are parameters that cannot be directly determined, wherein the micro parameters include three contact parameters and four parallel bond parameters, the contact parameters include the particle-particle collision recovery coefficient, the particle-particle static friction coefficient, and the particle-particle rolling friction coefficient, and the parallel bond parameters include the normal stiffness per unit area of the parallel bond, the tangential stiffness per unit area of the parallel bond, the critical normal stress of the parallel bond, and the critical shear stress of the parallel bond.

[0073] In some embodiments, in step S102, the value ranges of seven microscopic parameters are determined, and the specific steps are as follows:

[0074] Step 1. Set the initial simulation time step. Set the initial simulation time step to a preset multiple of the Rayleigh time step. The simulation time step is the time required for each step of the simulation. The simulation time step affects the simulation speed and the accuracy of the simulation. The simulation step needs to be lower than a certain threshold to ensure the accuracy of the simulation. However, reducing the time step without a lower limit will result in a very slow simulation speed. When determining the value range of microscopic parameters, the number of simulations used is relatively small, so the simulation speed can be ignored first, and only the accuracy of the simulation can be focused on. Therefore, the initial simulation time step is set to 0.01 times the Rayleigh time step. 0.01 times the Rayleigh time step is a very small time step, which is far below the threshold and is sufficient to ensure the accuracy of the simulation.

[0075] Step 2: Preset the initial value range of the micro-parameters, adjust the micro-parameters based on the initial value range, and obtain multiple sets of discrete parameters. First, the value range of the micro-parameters is proposed, please refer to Table 1,

[0076] Table 1

[0077] parameter Numerical <![CDATA[Normal stiffness per unit area of parallel keys (×10 11 N / m 3 )]]> 4-8 <![CDATA[Tangential stiffness per unit area of the parallel key (×10 11 N / m 3 )]]> 2-4 <![CDATA[Critical normal stress of parallel keys (×10 7 Pa)]]> 2-4 <![CDATA[Critical shear stress of parallel keys (×10 6 Pa)]]> 2-4 Coefficient of restitution for particle-particle collisions 0.085-0.115 Particle-particle static friction coefficient 0.16-0.24 Particle-particle rolling friction coefficient 0.005-0.015

[0078] Secondly, the micro parameters are adjusted within the proposed micro parameter value range to obtain multiple sets of discrete parameters;

[0079] Step 3: Determine non-microscopic parameters. Simulate multiple sets of discrete parameters based on the non-microscopic parameters and the initial simulation time step. Through 12 sets of simulations, obtain multiple elastic modulus values and multiple failure strain values. Based on the multiple elastic modulus values and multiple failure strain values, determine the maximum and minimum thresholds of the elastic modulus and failure strain. For each simulation, the non-microscopic parameters and microscopic parameters need to be input into the model for simulation.

[0080] Step 4. Obtain the elastic modulus and failure strain value of the reference uniaxial compression experiment, and judge the elastic modulus and failure strain of the reference uniaxial compression experiment based on the maximum threshold and the minimum threshold. When the elastic modulus and failure strain value of the reference uniaxial compression experiment are not between the maximum threshold and the minimum threshold, repeat steps 2 to 3 until the elastic modulus and failure strain value of the reference uniaxial compression experiment are between the maximum threshold and the minimum threshold. When the elastic modulus and failure strain value are between the maximum threshold and the minimum threshold, the initial value range is used as the value range of the microscopic parameter; the maximum value of the elastic modulus is 0.645 GPa, and the minimum value of the elastic modulus is It is 0.487GPa, the maximum value of the failure strain is 3.40%, and the minimum value of the failure strain is 1.10%. Only when the elastic modulus and failure strain of the reference uniaxial compression experiment meet the conditions, the proposed micro-parameter value range is considered to be the final micro-parameter value range, that is, the elastic modulus value of the reference uniaxial compression experiment is within the range of the maximum and minimum values of the elastic modulus obtained after the discrete element simulation of the micro-parameters, and the failure strain value of the reference uniaxial compression experiment is within the range of the maximum and minimum values of the failure strain obtained after the discrete element simulation of the micro-parameters, then the proposed micro-parameter value range is considered to be the final micro-parameter value range.

[0081] The elastic modulus obtained from the reference uniaxial compression test is 0.560 GPa and the failure strain is 2.45%, which satisfies the following conditions:

[0082] E min <E exp <E max

[0083] σ min <σ exp <σ max ,

[0084] Among them, E min is the minimum value of elastic modulus, E max is the maximum value of elastic modulus, E exp is the elastic modulus value obtained experimentally, σ max is the maximum value of the failure strain, σ min Minimum value of failure strain, σ exp is the value of the failure strain obtained from the experiment.

[0085] In some embodiments, the simulation time step is determined. The simulation time step affects the simulation speed and the accuracy of the simulation. The simulation step needs to be lower than a certain threshold to ensure the accuracy of the simulation. However, reducing the time step without a lower limit will result in a very slow simulation speed. Many simulations will be performed later, so it is necessary to take into account both the accuracy and speed of the simulation. Therefore, a maximum time step should be selected from among the many time steps that can ensure the accuracy of the simulation. The specific steps are as follows: based on the Rayleigh time step, a simulation time step is preset, and the preset simulation time step is set to iT Rayleigh , i=0.01, 0.02, ... 0.07, where T Rayleigh is the Rayleigh time step; adjust the preset simulation time step, that is, adjust iT by changing the value of variable i Rayleigh Based on the non-microscopic parameters, microscopic parameters and multiple discrete element simulation time steps, multiple failure strain values σ are obtained through multiple simulations. i , the failure strain error threshold is preset to 0.01, and the failure strain error is calculated based on the failure strain value. When the failure strain error is less than the failure strain error threshold, the final simulation time step is determined according to the maximum time step criterion; seven time steps are tried for simulation, Δ i It is an indicator for evaluating the correctness of simulation, which refers to the simulation result error at different time steps. If Δ i Satisfied the condition Δ i <0.01, it is considered that the simulation correctness is met, and i at this time is recorded as i critical , which means that we only need to ensure that the time step is less than i critical T Rayleigh , the simulation accuracy can be guaranteed, and the failure strain error Δ i The calculation formula is:

[0086]

[0087] The calculation formula of the maximum time step criterion is:

[0088] T final critical T Rayleigh ,

[0089] Among them, T final is the final simulation time step, i critical is the critical value of variable i;

[0090] Calculate Δ i <0.01 corresponds to a maximum i value of 6, so i critical is 6, and the final simulation time step is T final Select 3e -7 s, satisfying the maximum time step criterion condition.​

[0091] In some embodiments, in step S103, the Plackett-Burman experimental design method is used to perform micro-parameter dimensionality reduction on seven parameters, including the normal stiffness per unit area of the parallel key, the tangential stiffness per unit area of the parallel key, the critical normal stress of the parallel key, the critical shear stress of the parallel key, the particle-particle collision recovery coefficient, the particle-particle static friction coefficient, and the particle-particle rolling friction coefficient. The Plackett-Burman experiment uses elastic modulus and failure strain as characteristic response indicators, and the significance level α is selected as 0.05. The micro-parameters are divided into significant parameters and non-significant parameters according to the Pareto chart. For the elastic modulus, the unit area normal stiffness and unit area tangential stiffness of the parallel bond are significant parameters. For the failure strain, the unit area tangential stiffness and the critical shear stress of the parallel bond are significant parameters. Therefore, it is believed that the unit area normal stiffness, unit area tangential stiffness and the critical shear stress of the parallel bond are significant parameters. The non-significant parameters include the particle-particle collision recovery coefficient, the particle-particle static friction coefficient, the particle-particle rolling friction coefficient and the critical normal stress of the parallel bond. The values of the non-significant parameters are set to the median of the value range of the micro parameters, and the significant parameters are calibrated by the neural network.

[0092] In some embodiments, in step S104, after determining the simulation time step, the significance parameter is adjusted based on the value range of the microscopic parameter to obtain multiple sets of discrete parameters of the significance parameter, and the multiple sets of discrete parameters of the significance parameter are simulated based on the non-microscopic parameter, the non-significant parameter and the simulation time step, 60 discrete element simulations are run, and 60 sets of stress and strain data corresponding to the significance parameter are extracted, and the 60 sets of stress and strain data and the multiple sets of discrete parameters of the significance parameter are used as a data set, which is used for neural network training.

[0093] In some embodiments, in step S105, a neural network model based on a gated logic unit and an attention mechanism is established. The input of the neural network model is 60 sets of stress and strain data, and the output is a significance parameter. For the architecture diagram of the neural network model, please refer to Figure 2 , its neural network model consists of an encoder and a decoder. The encoder consists of a gating logic unit and an attention mechanism. Its gating logic unit is used to analyze sequence data and capture implicit information in the sequence data. The output of the t-th gating logic unit is calculated using the following formula:

[0094] Z t =σ(W (Z) x t +U (Z) h t-1 ),

[0095] rt =σ(W (r) x t +U (r) h t-1 ),

[0096]

[0097]

[0098] Among them, h t is the output of the t-th gate logic unit, x t is the input of the t-th gate logic unit, h t-1 is the output of the t-1th gated logic unit, σ is the Sigmoid activation function, tanh is the Tanh activation function, z t is the intermediate variable, r t is an intermediate variable, is the intermediate variable, U (z) 、W (z) 、U (r) 、W (r) , U and W are trainable parameter matrices.

[0099] The attention mechanism is used to solve the long-range dependency problem of gated logic units when processing sequence data. The calculation formula of the attention mechanism is:

[0100]

[0101] Among them, y is the output of the attention mechanism, h t represents the output of the tth gated logic unit, α t is the influence of the output of the t-th gated logic unit on the next layer, and n is the number of gated logic units;

[0102] The decoder is composed of a fully connected neural network, and the data of the i-th layer is calculated by the following formula:

[0103] y i =σ(Linear(y i-1 )),

[0104] Among them, σ is the sigmoid activation function, Linear is the linear transformation, y i-1 The data of the previous layer.

[0105] The stress-strain data are divided into a training set and a test set. Multiple sets of stress-strain data are used to train the neural network model. The neural network model is trained and the determination coefficient of the neural network on the training set and the test set is recorded to obtain the determination coefficient of the training set and the determination coefficient of the test set. The determination coefficient changes with the change of the training generation number. The calculation formula of the determination coefficient is:

[0106]

[0107] Among them, R 2 is the coefficient of determination, y true is the actual value of the dependent variable, y pre is the predicted value of the dependent variable, is the mean of the actual values of the dependent variable.

[0108] When the coefficient of determination of the training set and the coefficient of determination of the test set are both greater than the coefficient threshold of 0.998, it is considered that the trained neural network model can be used to predict significant parameters to obtain a fully trained neural network model.

[0109] In some embodiments, in step S106, the stress-strain curve of the uniaxial compression experiment is obtained, that is, the stress-strain curve to be calibrated, and the stress-strain curve of the uniaxial compression experiment is input into the trained neural network model to predict the unit area normal stiffness of the parallel key to be 4.8543×10 11 N / m 3 The tangential stiffness per unit area of the parallel key is 2.6154×10 11 N / m 3 , the critical shear stress of the parallel key is 3.4354×10 6 Pa, to obtain a set of significance parameter values. At this point, the calibration of the significance parameters is completed.

[0110] The stress-strain curves are obtained by running discrete element simulations with this set of parameters. The comparison diagram of the predicted results with the stress-strain curves of the reference experiment is shown in Fig. Figure 3 .

[0111] The micro-parameter calibration method of brittle materials is applicable to the micro-parameter calibration problems of bonded particle models of various brittle materials and quasi-brittle materials, and has a wider scope of application.

[0112] In order to better implement the microscopic parameter calibration method of the brittle material in the embodiment of the present invention, based on the microscopic parameter calibration method of the brittle material, correspondingly, Figure 4 As shown, an embodiment of the present invention further provides a microscopic parameter calibration device for brittle materials. The microscopic parameter calibration device 400 for brittle materials includes:

[0113] A micro-parameter determination module 401 is used to establish a bonded particle model of a brittle material and determine a plurality of micro-parameters based on the bonded particle model;

[0114] Range and time step determination module 402, used to determine the value range of microscopic parameters and simulation time step;

[0115] The significance parameter acquisition module 403 is used to reduce the dimension of the microscopic parameters using the Plackett-Burman method to obtain the significance parameters;

[0116] A stress and strain data acquisition module 404 is used to obtain multiple sets of stress and strain data based on a value range, a simulation time step, and a significance parameter;

[0117] A neural network model training module 405 is used to train the initial neural network model using multiple sets of stress and strain data to obtain a fully trained neural network model;

[0118] The significance parameter calibration module 406 is used to input the stress-strain curve to be calibrated into the neural network model to obtain the value of the significance parameter.

[0119] like Figure 5 As shown, based on the microscopic parameter calibration method of brittle materials, the present invention also provides an electronic device 500, which can be a computing device such as a mobile terminal, desktop computer, notebook, PDA, server, etc. The electronic device 500 includes a processor 501, a memory 502, and a display 503. Figure 5 Only some of the components of the electronic device 500 are shown, but it should be understood that implementation of all of the shown components is not required, and more or fewer components may be implemented instead.

[0120] In some embodiments, the memory 502 may be an internal storage unit of the electronic device 500, such as a hard disk or memory of the electronic device 500. In other embodiments, the memory 502 may also be an external storage device of the electronic device 500, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the electronic device 500. Furthermore, the memory 502 may include both an internal storage unit of the electronic device 500 and an external storage device. The memory 502 is used to store application software installed in the electronic device 500 and various types of data, such as program code installed in the electronic device 500. The memory 502 may also be used to temporarily store data that has been output or is about to be output. In one embodiment, the memory 502 stores a micro-parameter calibration program for brittle materials, which can be executed by the processor 501, thereby implementing the micro-parameter calibration method for brittle materials according to various embodiments of the present invention.

[0121] In some embodiments, the processor 501 may be a central processing unit (CPU), a microprocessor, or other data processing chip, configured to execute program codes or process data stored in the memory 502, such as a microscopic parameter calibration method for brittle materials.

[0122] In some embodiments, display 503 can be an LED display, a liquid crystal display, a touch-sensitive liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. Display 503 is used to display identification information for the microscopic parameter calibration program for brittle materials and to display a visual user interface. Components 501-503 of electronic device 500 communicate with each other via a system bus.

[0123] In some embodiments, when the processor 501 executes the micro-parameter calibration program of the brittle material in the memory 502, the various steps in the micro-parameter calibration method of the brittle material described in the above embodiments are implemented. Since the micro-parameter calibration method of the brittle material has been described in detail above, it will not be repeated here.

[0124] In summary, the present invention provides a method, device, equipment, and medium for calibrating microscopic parameters of brittle materials. First, a bonded particle model of the brittle material is established, and multiple microscopic parameters are determined based on the bonded particle model; the value range of the microscopic parameters and the simulation time step are determined; secondly, the Plackett-Burman method is used to reduce the dimension of the microscopic parameters to obtain significance parameters; multiple sets of stress-strain data are obtained based on the value range, simulation time step, and significance parameters; finally, an initial neural network model is trained using the multiple sets of stress-strain data to obtain a fully trained neural network model; the stress-strain curve to be calibrated is input into the neural network model to obtain the value of the significance parameter, and the calibration of the microscopic parameters is completed, thereby improving the efficiency and accuracy of the microscopic parameter calibration.

[0125] Those skilled in the art will appreciate that all or part of the process steps of the above-described embodiments can be implemented by instructing related hardware through a computer program, and the program can be stored in a computer-readable storage medium, such as a magnetic disk, an optical disk, a read-only memory, or a random access memory.

[0126] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by any technician familiar with this technical field within the technical scope disclosed by the present invention should be covered by the scope of protection of the present invention.

Claims

1. A method for calibrating microscopic parameters of brittle materials, characterized in that: include: establishing a cohesive particle model of the brittle material, and determining a plurality of microscopic parameters based on the cohesive particle model; Determining the value range of the microscopic parameters and the simulation time step; Plackett-Burman method is used to reduce the dimension of the microscopic parameters to obtain significant parameters; Obtaining multiple sets of stress and strain data based on the value range, the simulation time step, and the significance parameter; Using the multiple sets of stress-strain data to train the initial neural network model to obtain a fully trained neural network model; The stress-strain curve to be calibrated is input into the neural network model to obtain the value of the significance parameter.

2. The microscopic parameter calibration method of brittle materials according to claim 1, characterized in that: The microscopic parameters include contact parameters and parallel bond parameters. The contact parameters include particle-particle collision recovery coefficient, particle-particle static friction coefficient and particle-particle rolling friction coefficient. The parallel bond parameters include unit area normal stiffness of the parallel bond, unit area tangential stiffness of the parallel bond, critical normal stress of the parallel bond and critical shear stress of the parallel bond.

3. The microscopic parameter calibration method of brittle materials according to claim 1, characterized in that: Determining the value range of the microscopic parameter includes: Step 1: Setting an initial simulation time step, wherein the initial simulation time step is a preset multiple of the Rayleigh time step; Step 2: presetting an initial value range of the micro-parameters, and adjusting the micro-parameters based on the initial value range to obtain multiple groups of discrete parameters; Step 3: determining non-microscopic parameters, and simulating the multiple sets of discrete parameters based on the non-microscopic parameters and the initial simulation time step to obtain multiple elastic modulus values and multiple failure strain values, and determining maximum and minimum thresholds of the elastic modulus and the failure strain based on the multiple elastic modulus values and the multiple failure strain values; Step 4: Obtain the elastic modulus value and failure strain value of the reference uniaxial compression experiment, and judge the elastic modulus value and failure strain value of the reference uniaxial compression experiment based on the maximum threshold value and the minimum threshold value. When the elastic modulus value and failure strain value of the reference uniaxial compression experiment are not between the maximum threshold value and the minimum threshold value, repeat steps 2 to 3 until the elastic modulus value and failure strain value of the reference uniaxial compression experiment are between the maximum threshold value and the minimum threshold value. When the elastic modulus value and failure strain value of the reference uniaxial compression experiment are between the maximum threshold value and the minimum threshold value, use the initial value range as the value range of the microscopic parameter.

4. The microscopic parameter calibration method of brittle materials according to claim 3, characterized in that: Determining the simulation time step includes: Presetting a simulation time step based on the Rayleigh time step; Adjust the preset simulation time step to obtain multiple sets of discrete element simulation time steps; Based on the non-microscopic parameters, microscopic parameters and multiple sets of discrete element simulation time steps, multiple failure strain values are obtained through multiple simulations; A failure strain error threshold is preset, and a failure strain error is calculated based on the failure strain value. When the failure strain error is less than the failure strain error threshold, a simulation time step is determined according to a maximum time step criterion, wherein the failure strain error is an indicator for evaluating the correctness of the simulation.

5. The microscopic parameter calibration method of brittle materials according to claim 4, characterized in that: The obtaining of multiple sets of stress-strain data based on the value range, the simulation time step, and the significance parameter includes: Adjusting the significance parameter based on the value range to obtain multiple groups of discrete parameters of the significance parameter; Determine non-significant parameters, simulate multiple groups of discrete parameters of the significant parameters based on the non-microscopic parameters, the non-significant parameters and the simulation time step, and obtain multiple groups of stress and strain data.

6. The microscopic parameter calibration method of brittle materials according to claim 5, characterized in that: The initial neural network model is trained using the multiple sets of stress and strain data to obtain a fully trained neural network model, including: Establishing an initial neural network model, wherein the initial neural network model includes an encoder and a decoder, the encoder includes a gated logic unit and an attention mechanism, the decoder includes a fully connected neural network, the input of the initial neural network model is the multiple sets of stress and strain data, and the output of the initial neural network model is the significance parameter; Dividing the multiple sets of stress-strain data into a training set and a test set; Using the multiple sets of stress-strain data to train the initial neural network model to obtain the coefficient of determination of the training set and the coefficient of determination of the test set; When the coefficient of determination of the training set and the coefficient of determination of the test set are both greater than the coefficient threshold, a fully trained neural network model is obtained.

7. The microscopic parameter calibration method of brittle materials according to claim 6, characterized in that: The calculation formula of the gate control logic unit is: Z t =σ(W (Z) x t +U (Z) h t-1 ), r t =σ(W (r) x t +U (r) h t-1 ), Among them, h t is the output of the t-th gate logic unit, x t is the input of the t-th gate logic unit, h t-1 is the output of the t-1th gated logic unit, σ is the Sigmoid activation function, tanh is the Tanh activation function, z t is the intermediate variable, r t is the intermediate variable, is the intermediate variable, U (z) 、W (z) 、U (r) 、W (r) , U and W are trainable parameter matrices; The calculation formula of the attention mechanism is: Among them, y is the output of the attention mechanism, h t represents the output of the tth gated logic unit, α t is the influence of the output of the t-th gated logic unit on the next layer, and n is the number of gated logic units; The calculation formula of the determination coefficient is: Among them, R 2 is the coefficient of determination, y true is the actual value of the dependent variable, y pre is the predicted value of the dependent variable, is the mean of the actual values of the dependent variable.

8. A microscopic parameter calibration device for brittle materials, characterized in that: include: a microscopic parameter determination module, configured to establish a bonded particle model of the brittle material and determine a plurality of microscopic parameters based on the bonded particle model; A range and time step determination module, used to determine the value range of the microscopic parameters and the simulation time step; A significance parameter acquisition module, used for performing dimension reduction on the microscopic parameters using the Plackett-Burman method to obtain significance parameters; A stress and strain data acquisition module, configured to obtain multiple sets of stress and strain data based on the value range, the simulation time step, and the significance parameter; A neural network model training module is used to train the initial neural network model using the multiple sets of stress and strain data to obtain a fully trained neural network model; The significance parameter calibration module is used to input the stress-strain curve to be calibrated into the neural network model to obtain the value of the significance parameter.

9. An electronic device, characterized in that: including memory and processor; The memory stores a computer-readable program executable by the processor; When the processor executes the computer-readable program, the processor implements the steps of the micro-parameter calibration method of brittle materials according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps in the micro-parameter calibration method of brittle materials according to any one of claims 1 to 7.

Images