Material parameter calibration method based on discrete element method

Abstract

The present invention discloses a material parameter calibration method based on discrete element method, comprising the following steps: constructing a discrete element constitutive model according to material physical property parameters; setting parameters and ranges to be calibrated; performing LHS sampling according to the parameters and ranges to be calibrated; importing the sampling results of LHS sampling as input data into the discrete element constitutive model for discrete element simulation; establishing a training set according to the LHS sampling results and the output data of the discrete element simulation; performing neural network training according to the training set to obtain an approximate model between the parameters to be calibrated and the output data of the discrete element simulation; verifying whether the approximate model meets the first preset condition, if not, expanding the sample size of LHS sampling; performing genetic algorithm optimization on the approximate model; and calibrating the material parameters according to the output results of the genetic algorithm. The material parameter calibration method based on the discrete element method of the present invention improves the efficiency of parameter calibration, verifies the parameter calibration results in a variety of ways, and improves the accuracy of parameter calibration.

Description

Technical Field

[0001] The invention relates to the technical field of material processing, and in particular to a material parameter calibration method based on discrete element method. Background Art

[0002] Discrete element method (DEM) is one of the important methods for numerical simulation of powder and bulk materials. It has been widely used in many fields. This method has also been gradually applied to the simulation of particle motion behavior in mineral processing with its advantages of high efficiency and high precision, providing an effective means for optimizing mineral processing process parameters and improving process technical indicators. The particle physical properties parameters, which are the basis of numerical simulation technology, are the basic basis for numerical simulation research and play a vital role in numerical simulation research. Accurate and reasonable particle model parameters are necessary to ensure the high precision of numerical simulation results.

[0003] In the current numerical simulation research on particle motion behavior, the particle parameters that are easier to measure are generally obtained by conventional measurement methods. For example, particle density and particle size can be measured by a specific gravity bottle and a laser particle size analyzer, respectively. However, contact coefficients such as the friction coefficient between particles and the elastic recovery coefficient between particles are difficult to obtain through conventional physical test methods. Even if they are measured by complex methods, there are large deviations in the accuracy of the measurement results, which directly affects the accuracy of the numerical simulation results and cannot provide scientific and effective guidance for the setting and optimization of process parameters.

[0004] A method for calibrating material parameters using response surface method is disclosed in the related art, such as Fig.12 As shown, however, when using the response surface method to calibrate parameters, the number of factor codes is limited and the selected data samples are not broadly representative. Summary of the invention

[0005] The present invention is based on the inventor's discovery and understanding of the following facts and problems:

[0006] In the related art, a method for calibrating the parameters of a mineral particle model based on the discrete element method is provided. The constitutive model of the parameters of a mineral particle model based on the discrete element method is mainly constructed through physical test determination and preliminary proposed methods. The constitutive model is combined with the angle of repose determination to conduct physical and numerical simulation bidirectional tests. On the basis of the preliminary verification results, the orthogonal numerical test is conducted with the mineral particle model parameters as the independent variable, and the regression analysis is conducted with the angle of repose determination value as the evaluation index. The physical test measurement of the angle of repose is used as the target value to establish a regression model, determine the optimal combination of the mineral particle model parameters, and verify the established optimal combination through numerical tests again, and finally determine the parameters of the mineral particle model based on the discrete element method. However, the inventor of the present application has found that there is a lack of extensive experimental verification in the related art, and the influence of the calibrated contact parameters on the stacking height is not considered in the study. At the same time, the same angle of repose value and stacking height can often be obtained by different contact parameter combinations. Therefore, this technology lacks reliability and accuracy in material parameter calibration.

[0007] In the related art, a BP neural network parameter calibration method based on LHS is proposed. The related art includes establishing a finite element model; parameter sensitivity analysis; determining the parameters to be calibrated and their ranges; LHS sampling; establishing a training set; training the BP neural network; verifying the BP neural network; and soil mechanics parameter calibration. The present invention utilizes the characteristics of LHS sampling that can construct a small number of highly representative samples and the powerful nonlinear mapping ability of the BP neural network, bringing a new model to the soil mechanics parameter calibration. This calibration method not only avoids the disadvantage of the gradient optimization algorithm that is prone to fall into the local optimum, but also makes up for the deficiency of the random optimization algorithm that requires a large number of finite element calculations. It can be well applied to the soil mechanics parameter calibration and can be extended to other fields that require parameter calibration. However, the inventors of the present application found that the parameter calibration results of the related technology have not been physically verified, which affects the accuracy of the virtual simulation results.

[0008] The present invention aims to solve one of the technical problems in the related art to at least a certain extent. To this end, the present invention proposes a material neural network model parameter calibration method based on discrete element method, which can improve the efficiency of parameter calibration and verify the parameter calibration results in multiple ways to improve the accuracy of parameter calibration.

[0009] According to an embodiment of the present invention, a material parameter calibration method based on the discrete element method includes the following steps: constructing a discrete element constitutive model according to the physical properties of the material; setting the parameters and ranges to be calibrated; performing LHS sampling according to the parameters and ranges to be calibrated; importing the sampling results of the LHS sampling as input data into the discrete element constitutive model for discrete element simulation; establishing a training set according to the LHS sampling results and the output data of the discrete element simulation; performing neural network training according to the training set to obtain an approximate model between the parameters to be calibrated and the output data of the discrete element simulation; verifying whether the approximate model meets the first preset condition, and if not, expanding the sample size of the LHS sampling; optimizing the approximate model using a genetic algorithm; and calibrating the material parameters according to the output results of the genetic algorithm.

[0010] The material parameter calibration method based on the discrete element method according to the embodiment of the present invention can improve the efficiency of parameter calibration, and verify the parameter calibration results in various ways to improve the accuracy of parameter calibration.

[0011] In some embodiments, the LHS sampling includes extracting multiple samples from each of the parameters to be calibrated, and the multiple samples extracted from each of the parameters to be calibrated are divided into multiple sample groups.

[0012] In some embodiments, the output data of the discrete element simulation are the macroscopic static repose angle S and the stacking height H.

[0013] In some embodiments, the macro-static angle of repose is detected according to an image recognition algorithm, and the image recognition algorithm includes: intercepting the macro-static angle of repose image, grayscale or binarization processing the macro-static angle of repose image, extracting the macro-static angle of repose edge, fitting the macro-static angle of repose edge and obtaining the macro-static angle of repose value.

[0014] Macroscopic static repose angle In some embodiments, the LHS sampling results are used as the input layer of the neural network training, and the output data of the discrete element simulation are used as the output layer of the neural network training.

[0015] In some embodiments, the first preset condition is R 2 ≥0.9, and R 2 is the residual sum of squares.

[0016] In some embodiments, performing genetic algorithm optimization on the approximate model includes: verifying whether the output result of the neural network meets a second preset condition, and if not, adjusting the parameters of the genetic algorithm.

[0017] In some embodiments, the second preset condition is St-1°≤S≤St+1° and Ht-2≤H≤Ht+2, and △=|S-St|+|H-Ht|, wherein St is the macroscopic static repose angle of the material measured according to the stacking physical test, Ht is the material stacking height Ht measured according to the stacking physical test, and △ is the minimum value of the deviation and.

[0018] In some embodiments, before calibrating the material parameters according to the output results of the genetic algorithm, the method also includes: verifying the output results of the genetic algorithm, and the verification of the output results of the genetic algorithm includes: selecting multiple groups of output results from the output results of the genetic algorithm as verification data, dividing the multiple groups of output results into multiple groups of verification data, and the multiple groups of output results correspond one-to-one to the multiple groups of verification data, importing the multiple groups of verification data into the discrete element constitutive model for discrete element to obtain multiple groups of verification solutions, if the error between the multiple groups of verification solutions and the multiple groups of output results does not meet the third preset condition, then expanding the sample size of the LHS sampling, and the third preset condition is E1≤5%, where E1 is the error between the multiple groups of verification solutions and the multiple groups of output results.

[0019] In some embodiments, the verification of the output result of the genetic algorithm also includes: performing a unloading simulation test based on the output result of the genetic algorithm, and performing a unloading physical test on the material to be calibrated. If the error between the output result of the unloading simulation test and the output result of the unloading physical test does not meet a fourth preset condition, then expanding the sample size of the LHS sampling. The fourth preset condition is whether the material flow pattern in the unloading simulation test is consistent with the material flow pattern in the unloading physical test, and / or whether the error between the material mass flow rate in the unloading simulation test and the material mass flow rate in the unloading physical test gradually decreases. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a flowchart of a material parameter calibration method based on discrete element method in an embodiment of the present invention.

[0021] Figure 2 It is a flow chart of a material parameter calibration method based on discrete element method according to an embodiment of the present invention.

[0022] Figure 3 It is a neural network structure diagram in the material parameter calibration method based on discrete element method in an embodiment of the present invention.

[0023] Figure 4 It is a schematic diagram of a funnel blanking model according to an embodiment of the present invention.

[0024] Figure 5 It is a flow chart of detecting macroscopic static repose angle by an image recognition algorithm according to an embodiment of the present invention.

[0025] Figure 6 Schematic diagram of the macroscopic static angle of repose obtained by discrete element simulation in an embodiment of the present invention.

[0026] Figure 7 Schematic diagram of the extracted macroscopic static repose angle edge according to an embodiment of the present invention.

[0027] Figure 8 Schematic diagram of the fitting degree of the macroscopic static repose angle approximation model of the embodiment of the present invention.

[0028] Fig. 9 Schematic diagram of the stacking height approximation model fitting degree of an embodiment of the present invention.

[0029] Fig.10 It is a schematic diagram comparing the material flow pattern in the unloading simulation test and the material flow pattern in the unloading physical test according to the embodiment of the present invention.

[0030] Fig.11 It is a schematic diagram of an error curve between the material mass flow rate in the unloading simulation test and the material mass flow rate in the unloading physical test according to an embodiment of the present invention.

[0031] Fig.12 It is a sampling table of calibration parameters of response surface method in related technology. DETAILED DESCRIPTION

[0032] Embodiments of the present invention are described in detail below, and examples of the embodiments are shown in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to be used to explain the present invention, but should not be understood as limiting the present invention.

[0033] like Figures 1 to 7 As shown, the material parameter calibration method based on discrete element method according to an embodiment of the present invention includes the following steps:

[0034] S1: Construct a discrete element constitutive model based on the physical properties of the material.

[0035] It should be noted that the physical property parameters of the material are elastic modulus, Poisson's ratio, shear modulus, density and particle size of the material, etc. It is understandable that other physical property parameter components of the material can also be selected as discrete element constitutive models.

[0036] S2: Set the parameters and range to be calibrated.

[0037] It should be noted that the parameters to be calibrated are the material-material recovery coefficient e_pp, and 0.1≤e_pp≤0.35, the material-material static friction coefficient sf_pp, and 0.25≤sf_pp≤0.65, the material-material dynamic friction coefficient rf_pp, and 0.1≤rf_pp≤0.35, the material-equipment recovery coefficient e_pw, and 0.1≤e_pw≤0.35, the material-equipment static friction coefficient sf_pw, and 0.2≤sf_pw≤0.65, and the material-equipment dynamic friction coefficient rf_pw, and 0.1≤rf_pw≤0.35.

[0038] S3: Perform LHS sampling according to the parameters and ranges to be calibrated.

[0039] It should be noted that Latin hypercube sampling randomly selects 150 samples from each parameter to be calibrated, and then divides the sampled samples into 150 sample groups, that is, each sample group contains any one of the 150 data of each parameter to be calibrated.

[0040] S4: Importing the sampling result of the LHS sampling as input data into the discrete element constitutive model for discrete element simulation.

[0041] Specifically, 150 sample groups formed by Latin hypercube sampling were used as input data of the discrete element constitutive model for discrete element simulation.

[0042] S5: Establish a training set based on the LHS sampling results and the output data of discrete element simulation.

[0043] It should be noted that not every one of the 150 sample groups has data results. Some sample groups cannot form an angle of repose in discrete element simulation due to parameter reasons. In this embodiment, there are a total of 123 groups of data that can form a stable stacking effect. The 123 sample groups and the discrete element simulation output results corresponding to each sample group are used to establish a training set.

[0044] S6: Perform neural network training based on the training set to obtain an approximate model between the parameters to be calibrated and the output data of the discrete element simulation.

[0045] It should be noted that the 123 sample groups with stable stacking effect are used as the input layer in the neural network training, and the discrete element simulation output results corresponding to the 123 sample groups are used as the output layer in the neural network training, and an approximate model between the parameters to be calibrated and the output data of the discrete element simulation is obtained according to the training neural network model. Specifically, the neural network model is a radial basis function neural network model.

[0046] Specifically, Figure 8As shown, H-RBFNN is the predicted value of the pile height by the radial basis function neural network model; S-RBFNN is the predicted value of the repose angle by the radial basis function neural network model; H-DEM is the pile height value obtained in the discrete element simulation; S-DEM is the repose angle value obtained in the discrete element simulation.

[0047] S7: Verify whether the approximate model obtained by the neural network satisfies the first preset condition. If not, increase the sample size of LHS sampling.

[0048] It should be noted that the first preset condition is R 2 ≥0.9. Among them, R 2 is the residual sum of squares, that is, the residual sum of squares R of the approximate model generated by neural network training 2 When it is less than 0.9, the sample size of LHS sampling is increased.

[0049] S8: Use genetic algorithm to optimize the approximate model.

[0050] It should be noted that the approximate model is a functional relationship model generated between the 123 sample groups and the discrete element simulation output results corresponding to the 123 sample groups, and then the approximate model is optimized by a genetic algorithm.

[0051] S9: Calibrate material parameters according to the output results of the genetic algorithm.

[0052] It should be noted that the output result of the genetic algorithm is one or more sets of numerical solutions that are the most optimized among multiple approximate models.

[0053] According to the material parameter calibration method based on the discrete element method in the embodiment of the present invention, an approximate model between the parameters to be calibrated and the output data of the discrete element can be obtained, and the optimal solution set can be selected through a genetic algorithm to improve the accuracy and efficiency of the parameter calibration. The parameter calibration results are verified in a variety of ways to improve the accuracy of the parameter calibration. It also avoids the disadvantage of the traditional response surface method falling into local optimality in parameter calibration, and makes up for the disadvantage of insufficient sampling caused by random sampling.

[0054] In some embodiments, LHS sampling includes extracting multiple samples from each parameter to be calibrated, and dividing the multiple samples extracted from each parameter to be calibrated into multiple sample groups.

[0055] Specifically, LHS sampling extracts 150 samples from each parameter to be calibrated, and randomly divides the 150 samples extracted from each parameter to be calibrated into 150 sample groups, that is, each sample group contains any one of the 150 data of each parameter to be calibrated.

[0056] In some embodiments, the output data of the discrete element simulation are the macroscopic static repose angle S and the stacking height H.

[0057] It should be noted that if Figure 8 As shown, the first preset condition R corresponding to the macroscopic static repose angle S 2 =95%, the first preset condition R corresponding to the stacking height H 2 =96%. It should be noted that the output data of discrete element simulation can also be other material parameters. Specifically, the discrete element simulation model uses the funnel drop model. In the discrete element simulation, the contact model between particles and the contact model between particles and geometric bodies are both Hertz-Mindlin non-slip contact models, also known as the "elastic-damping-friction" contact force model. This model introduces springs, dampers, and friction devices to describe the deformation of particles when they collide and contact, and can handle static and multi-particle interactions.

[0058] In some embodiments, the macro-static angle of repose is detected according to an image recognition algorithm, and the image recognition algorithm includes: intercepting a macro-static angle of repose image, grayscale or binarization processing the macro-static angle of repose image, extracting the macro-static angle of repose edge, fitting the macro-static angle of repose edge and obtaining the macro-static angle of repose value shown.

[0059] It should be noted that the macro static repose angle can be intercepted by the screenshot tool provided by the discrete element analysis software, or by other screenshot tools. The screenshot macro static repose angle image is grayed or binarized by Python program code, and the boundary is extracted and the macro static repose angle edge is fitted to finally obtain the macro static repose angle. The stacking height of the material is obtained by measuring the change in the particle coordinate value in the height direction in the discrete element software.

[0060] In some embodiments, the LHS sampling results are used as the input layer of the neural network training, and the output data of the discrete element simulation is used as the output layer of the neural network training.

[0061] It should be noted that the 123 sample groups that can form a stable stacking effect are used as the input layer in the neural network training, and the discrete element simulation output results corresponding to the 123 sample groups are used as the output layer in the neural network training. According to the trained neural network model, an approximate model between the parameters to be calibrated and the output data of the discrete element simulation is obtained.

[0062] Specifically, Figure 2 As shown, S is the macroscopic static repose angle S, H is the stacking height H, and the macroscopic static repose angle and stacking height are used as the output layer.

[0063] In some embodiments, performing genetic algorithm optimization according to the output result of the neural network includes: verifying whether the output result of the neural network meets the second preset condition, and if not, adjusting the parameters of the genetic algorithm.

[0064] Specifically, the second preset condition is St-1°≤S≤St+1° and Ht-2≤H≤Ht+2, and △=|S-St|+|H-Ht|, wherein St is the macroscopic static repose angle of the material measured according to the stacking physics test, Ht is the material stacking height measured according to the stacking physics test, and △ is the minimum value of the sum of deviations. It should be noted that the sum of deviations is the sum of the deviations between the macroscopic static repose angle St of the material measured according to the stacking physics test and the macroscopic static repose angle S of the discrete element simulation output data, and the deviations between the material stacking height Ht measured according to the stacking physics test and the stacking height H of the discrete element simulation output data.

[0065] For example, in the actual stacking test, St was measured to be 29° and Ht was 48mm, so the second preset condition is set to 28°≤S≤30°, and 47≤H≤49, then the minimum macro static angle of repose is 28°, the minimum stacking height H is 47, and the minimum deviation and △ is 2.

[0066] In some embodiments, before calibrating the material parameters according to the output results of the genetic algorithm, the material parameter calibration method based on the discrete element method also includes: verifying the output results of the genetic algorithm, verifying the output results of the genetic algorithm includes: selecting multiple groups of output results from the output results of the genetic algorithm as verification data, dividing the multiple groups of output results into multiple groups of verification data, and the multiple groups of output results correspond one-to-one to the multiple groups of verification data, importing the multiple groups of verification data into the discrete element constitutive model for discrete element to obtain multiple groups of verification solutions, if the error between the multiple groups of verification solutions and the multiple groups of output results does not meet the third preset condition, then expanding the sample size of LHS sampling, the third preset condition is E1≤5%, where E1 is the error between the multiple groups of verification solutions and the multiple groups of output results.

[0067] It should be noted that if the second preset condition is met, it is proved that the output result of the neural network after genetic algorithm optimization can be used as the output result of genetic algorithm optimization, and the output result after genetic algorithm optimization is verified again.

[0068] Specifically, multiple groups of output results are selected from the output results of the genetic algorithm, and these multiple groups of output results are imported into the discrete element constitutive model as input data for discrete element simulation and obtaining multiple groups of verification solutions. The multiple groups of verification solutions correspond one-to-one to the multiple groups of output results of the extracted genetic algorithm, and the multiple groups of verification solutions are compared one-to-one with the multiple groups of output results of the extracted genetic algorithm. If the error E1 between each group of verification solutions and the output results of the genetic algorithm is less than or equal to 5%, it proves that the output results of the genetic algorithm meet the third preset condition.

[0069] On the contrary, if the output result of the genetic algorithm does not meet the third preset condition, the sample size of LHS sampling is increased.

[0070] It should be noted that in this embodiment, 5 groups of output results are selected as verification data. The specific 5 groups of verification data and the results of 5 groups of verification solutions corresponding to the 5 groups of data are shown in Table 1:

[0071] Table 1: Verification data and discrete element simulation results

[0072]

[0073] It should be noted that: S-MIGA in Table 1 is the repose angle obtained by the multi-island genetic algorithm; S-DEM is the repose angle obtained by DEM simulation; H-MIGA is the accumulation height obtained by the multi-island genetic algorithm; H-DEM is the accumulation height obtained by DEM simulation.

[0074] The material parameter calibration method based on the discrete element method in the embodiment of the present invention verifies the output result of the genetic algorithm optimization, thereby improving the accuracy and precision of the material parameter calibration.

[0075] In some embodiments, verifying the output results of the genetic algorithm also includes: performing a unloading simulation test based on the output results of the genetic algorithm, performing a unloading physical test on the material to be calibrated, and if the error between the output result of the unloading simulation test and the output result of the unloading physical test does not meet the fourth preset condition, then expanding the sample size of the LHS sampling, and the fourth preset condition is whether the material flow pattern in the unloading simulation test is consistent with the material flow pattern in the unloading physical test, and / or whether the error between the material mass flow rate in the unloading simulation test and the material mass flow rate in the unloading physical test gradually decreases.

[0076] It should be noted that when the output result of the genetic algorithm meets the third preset condition, it is verified whether the output result of the genetic algorithm meets the fourth preset condition. If the output result of the genetic algorithm also meets the fourth preset condition, the output result is used as the calibration of the material parameters.

[0077] Specifically, the fourth preset condition may only be whether the material flow pattern in the unloading simulation test is consistent with the material flow pattern in the unloading physical test, or whether the error between the material mass flow rate in the unloading simulation test and the material mass flow rate in the unloading physical test is gradually reduced.

[0078] The fourth preset condition may also be whether the material flow pattern in the unloading simulation test is consistent with the material flow pattern in the unloading physical test and whether the error between the material mass flow rate in the unloading simulation test and the material mass flow rate in the unloading physical test is gradually reduced.

[0079] It should be noted that the physical test of unloading adopts the silo unloading experiment. The silo unloading experiment uses a box made of acrylic glass as the silo, the width of the discharge port is 50mm, and the length, width and height of the silo are 200mm, 100mm and 300mm respectively. In the experiment, the particle laying height in the silo is 105mm. The model with the same size as the physical test of unloading is used in the unloading simulation test, and the fifth group of parameters in Table 1 are selected as the particle input contact parameters.

[0080] Specifically, Fig.11 As shown, Fig.11 The three groups of pictures at the top are the material flow patterns in the unloading physical test. Fig.11 The three groups of pictures at the bottom are the material flow patterns in the unloading simulation test. The process of the unloading physical test was recorded using a high-speed camera with a camera frame rate of 500FPS. The material flow patterns at three time points, t=0.19s, 0.31s, and 0.53s, were selected to compare with the material flow patterns in the unloading simulation test. Fig.11 It can be seen that the material experiments at the three time points are basically consistent with the simulated flow morphology.

[0081] It should be noted that whether the error between the material mass flow rate in the unloading simulation test and the material mass flow rate in the unloading physical test is gradually reduced in the fourth preset condition, the mass flow rate of the unloading physical test and the unloading simulation test process is recorded. In the unloading physical test, the force measurement process can be converted into mass flow with the help of a mechanical testing machine. The mechanical testing machine has a high sampling rate of 500Hz and can record the changes in mass data during the unloading process.

[0082] During the unloading physical test, the silo is fixed on the chuck of the mechanical testing machine. Specifically, Fig.12As shown in the figure, at the same particle laying height, the total initial material mass of the unloading physical test is greater than the total material mass of the material in the unloading simulation test. It should be noted here that this is because the material particle size distribution range used in the unloading physical test is wide, and the particles filled in the silo have a smaller porosity, while the material particles set in the unloading simulation test are only a few representative sizes, and the porosity is relatively larger after the particles are stacked, which leads to quality differences between the experiment and the simulation.

[0083] After the unloading begins, the material mass flow curve in the unloading physical test has obvious fluctuations in the initial stage. It should be noted here that this is mainly due to the fact that the equilibrium static state of the particle system is instantly broken after the baffle at the bottom of the silo is evacuated, which causes vibration to the silo. At this time, the overall mass of the silo fixed together with the chuck changes, and the mass of the chuck and the silo will act on the sensor and be converted into force changes through the data acquisition software. Because the gravity acceleration g is 9.8N / kg, that is, a 1kg object is subject to a gravity of 9.8N. Therefore, after the sensor collects the gravity change value of the measured object, it can be converted into mass change through the gravity acceleration g.

[0084] from Fig.12 It can be seen that the error between the material mass flow curve in the physical test of unloading and the material mass flow curve in the unloading simulation test is gradually decreasing with the progress of the unloading process, and the mass flow change trend is gradually approaching consistency. Therefore, by observing the error between the material mass flow curve in the physical test of unloading and the material mass flow curve in the unloading simulation test, the effectiveness of the output results of the genetic algorithm can be further proved.

[0085] The material parameter calibration method based on the discrete element method in the embodiment of the present invention can verify the output result of the genetic algorithm optimization by setting the third preset condition and the fourth preset condition, thereby improving the accuracy and precision of the material parameter calibration result.

[0086] Combine the following Figures 1 to 12 Some specific examples of the material parameter calibration method based on the discrete element method according to the embodiment of the present invention are described.

[0087] The discrete element analysis software is used to establish the discrete element constitutive model of the material, and the parameters and ranges to be calibrated of the material are determined. The specific parameters and ranges to be calibrated are shown in Table 2:

[0088] Table 2: Parameters and ranges to be calibrated

[0089]

[0090] LHS sampling is performed according to the parameters and ranges to be calibrated. Latin hypersquare sampling randomly selects 150 samples from each parameter to be calibrated, and then divides the sampled samples into 150 sample groups. The specific LHS sampling results are shown in Table 3:

[0091] Table 3: LHS sampling results

[0092]

[0093]

[0094]

[0095]

[0096]

[0097]

[0098] The sampled data of LHS sampling in Table 3 are randomly divided into 150 groups, each group includes 6 parameters to be calibrated, and the values ​​of the 6 parameters to be calibrated are any one of the 150 samples, and the 150 groups of data are imported as input data into the discrete element constitutive model for discrete element simulation. The results after discrete element simulation are shown in Table 4:

[0099] Table 4: Discrete element simulation results

[0100]

[0101]

[0102]

[0103]

[0104] It should be noted that S in Table 4 is the static repose angle, and simulation results that do not meet the requirements have been eliminated in Table 4, such as the case where the repose angle is 0. The simulation results that meet the requirements are trained with a neural network. The neural network obtains an approximate model between the 6 parameters to be calibrated and the output data of the discrete element simulation. When the approximate model meets the first preset condition and the fourth preset condition, the approximate model is optimized by a genetic algorithm. The results of the genetic algorithm optimization are shown in Table 5:

[0105] Table 5: Genetic algorithm output results

[0106]

[0107]

[0108]

[0109]

[0110]

[0111] Specifically, as shown in Table 5, after 1000 iterations, a total of 110 available solutions are generated, forming a solution set, and each solution corresponds to a different number of iteration steps.

[0112] The embodiment of the present invention can obtain an approximate model between the parameters to be calibrated and the output data of the discrete element, and select the optimal approximate solution set through a genetic algorithm, thereby improving the accuracy and efficiency of parameter calibration, and verify the parameter calibration results in a variety of ways to improve the accuracy of parameter calibration. It also avoids the disadvantage of the traditional response surface method falling into local optimality in parameter calibration, and makes up for the disadvantage of insufficient sampling caused by random sampling.

[0113] In the description of the present invention, it is to be understood that the terms “center”, “longitudinal”, “lateral”, “length”, “width”, “thickness”, “up”, “down”, “front”, “back”, “left”, “right”, “vertical”, “horizontal”, “top”, “bottom”, “inside”, “outside”, “clockwise”, “counterclockwise”, “axial”, “radial”, “circumferential”, etc., indicating orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the referred device or element must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be understood as limiting the present invention.

[0114] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0115] In the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral one; it can be a mechanical connection, an electrical connection, or communication with each other; it can be a direct connection, or an indirect connection through an intermediate medium, it can be the internal connection of two elements or the interaction relationship between two elements, unless otherwise clearly defined. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0116] In the present invention, unless otherwise clearly specified and limited, a first feature being "above" or "below" a second feature may mean that the first and second features are in direct contact, or the first and second features are in indirect contact through an intermediate medium. Moreover, a first feature being "above", "above" or "above" a second feature may mean that the first feature is directly above or obliquely above the second feature, or simply means that the first feature is higher in level than the second feature. A first feature being "below", "below" or "below" a second feature may mean that the first feature is directly below or obliquely below the second feature, or simply means that the first feature is lower in level than the second feature.

[0117] In the present invention, the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" etc. mean that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the described specific features, structures, materials or characteristics may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0118] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A material parameter calibration method based on discrete element method, characterized in that: The steps include: Construct discrete element constitutive model according to the material's physical property parameters; Set the parameters and ranges to be calibrated; Perform LHS sampling according to the parameters and ranges to be calibrated; Importing the sampling result of the LHS sampling as input data into the discrete element constitutive model for discrete element simulation; Establishing a training set according to the LHS sampling results and the output data of the discrete element simulation, wherein the output data of the discrete element simulation is a macroscopic static repose angle S and a stacking height H; Performing neural network training according to the training set to obtain an approximate model between the parameters to be calibrated and the output data of discrete element simulation; Verify whether the approximate model satisfies the first preset condition, and if not, expand the sample size of the LHS sampling; Performing genetic algorithm optimization on the approximate model; Calibrate material parameters according to the output results of the genetic algorithm; Before calibrating the material parameters according to the output results of the genetic algorithm, the method further includes: verifying the output results of the genetic algorithm, wherein verifying the output results of the genetic algorithm includes: selecting multiple groups of output results from the output results of the genetic algorithm as verification data, dividing the multiple groups of output results into multiple groups of verification data, and the multiple groups of output results correspond to the multiple groups of verification data one by one, importing the multiple groups of verification data into the discrete element constitutive model for discrete element simulation to obtain multiple groups of verification solutions, if the error between the multiple groups of verification solutions and the multiple groups of output results does not meet the third preset condition, then expanding the sample size of the LHS sampling, wherein the third preset condition is E1≤5%, wherein E1 is the error between the multiple groups of verification solutions and the multiple groups of output results; The verification of the output result of the genetic algorithm also includes: performing a unloading simulation test according to the output result of the genetic algorithm, and performing a unloading physical test on the material to be calibrated. If the error between the output result of the unloading simulation test and the output result of the unloading physical test does not meet the fourth preset condition, then expanding the sample size of the LHS sampling. The fourth preset condition is whether the material flow pattern in the unloading simulation test is consistent with the material flow pattern in the unloading physical test, and / or whether the error between the material mass flow rate in the unloading simulation test and the material mass flow rate in the unloading physical test gradually decreases.

2. The material parameter calibration method based on discrete element method according to claim 1 is characterized in that: The LHS sampling includes extracting a plurality of samples from each of the parameters to be calibrated, and the plurality of samples extracted from each of the parameters to be calibrated are divided into a plurality of sample groups.

3. The material parameter calibration method based on discrete element method according to claim 1 is characterized in that: The macro-static angle of repose is detected according to an image recognition algorithm, and the image recognition algorithm includes: intercepting the macro-static angle of repose image, grayscale or binarization processing the macro-static angle of repose image, extracting the macro-static angle of repose edge, fitting the macro-static angle of repose edge and obtaining the macro-static angle of repose value.

4. The material parameter calibration method based on discrete element method according to claim 1, characterized in that: The LHS sampling result is used as the input layer of the neural network training, and the output data of the discrete element simulation is used as the output layer of the neural network training.

5. The material parameter calibration method based on discrete element method according to claim 1, characterized in that: The first preset condition is R 2 ≥0.9, and R 2 is the residual sum of squares.

6. The material parameter calibration method based on discrete element method according to claim 1, characterized in that: Performing genetic algorithm optimization on the approximate model includes: verifying whether the output result of the neural network meets the second preset condition, and if not, adjusting the parameters of the genetic algorithm.

7. The material parameter calibration method based on discrete element method according to claim 6, characterized in that: The second preset condition is St-1°≤S≤St+1° and Ht-2≤H≤Ht+2, and △=|S-St|+|H-Ht|, wherein St is the macroscopic static repose angle of the material measured according to the stacking physical test, Ht is the stacking height of the material measured according to the stacking physical test, and △ is the minimum value of the deviation and.

Images