A DEM simulation model automatic calibration method and system based on parallel computing

Abstract

This invention relates to the field of simulation model calibration, and provides an automatic calibration method for DEM simulation models based on parallel computing. The method includes acquiring physical macroscopic features, generating multiple sets of initial simulation microscopic parameters, performing parallel simulation and data processing, acquiring simulation macroscopic features, and establishing an intelligent model between the simulation macroscopic features and physical macroscopic features. The physical macroscopic features are input into the constructed intelligent model to acquire the corresponding simulation microscopic parameters as initial values. Parallel simulation, data analysis, and comparison of the initial values ​​are performed, and the microscopic parameter corresponding to the simulation macroscopic feature with the smallest error is selected as the initial value for the next round. This process of parallel simulation, data analysis, and comparison of the initial values ​​is repeated until the relative error between the simulation macroscopic features and physical macroscopic features in the current round is lower than a set threshold or the number of corrections reaches a set upper limit, thus completing the calibration. This invention, based on parallel computing, can significantly improve the calibration efficiency of simulation parameters.

Description

Technical Field

[0001] This invention relates to the field of simulation model calibration, and more specifically, to an automatic calibration method for DEM simulation models based on parallel computing. Background Technology

[0002] With the rapid development of my country's economy, rotary kilns, dryers, coating machines, and other rotary drum equipment play a crucial role in industries such as cement, metallurgy, and pharmaceuticals. The core functions of these machines, such as material mixing, drying, and coating, have a vital impact on improving product quality and production efficiency. However, due to the complex properties of particulate materials, which combine solid and fluid characteristics, the mechanism of material movement within the drum remains unclear.

[0003] With the rapid development of advanced computing technologies, the Discrete Element Method (DEM) has become an important technique for analyzing the complex physical processes of particulate materials. By utilizing DEM, the movement and interactions of a large number of particles can be simulated under specific conditions, thus providing effective support for understanding and improving material handling processes in rotary drum equipment. Based on DEM simulations, researchers can explore the impact of key operating conditions and material properties on mixing efficiency without relying on actual production experiments. This largely helps avoid the high costs and economic losses that may result from traditional experimental research.

[0004] To ensure the accuracy of DEM model simulation results, parameter calibration is necessary. This involves continuously adjusting the simulation input parameters to match the acquired macroscopic features from the physical experiments. However, current DEM model parameter calibration largely relies on manual operation, which is tedious, inefficient, and lacks accuracy. Our team proposed a data-driven parameter calibration method in Chinese Invention Patent Application No. CN202310828095.5, which describes a method, apparatus, device, and storage medium for calibrating simulation parameters of particulate materials. However, this method only performs a set of parameter simulations at a time to obtain the simulated macroscopic features, which are then matched with the macroscopic features from the physical experiments. If the relative error is less than a set value, calibration stops; otherwise, parameter adjustments are made, and the simulation is repeated until the stopping condition is met. This method typically requires multiple rounds of parameter tuning and simulation to complete the DEM model parameter calibration, making it a serial parameter calibration method whose efficiency needs further improvement.

[0005] To address the current problem of low calibration efficiency of DEM models, there is a need for an automatic calibration method for DEM simulation models based on parallel computing that can significantly improve the calibration efficiency of simulation parameters. Summary of the Invention

[0006] The problem addressed by this invention is how to provide an automatic calibration method for DEM simulation models based on parallel computing, which can significantly improve the calibration efficiency of simulation parameters.

[0007] To address the aforementioned issues, this invention provides an automatic calibration method for DEM simulation models based on parallel computing, comprising: acquiring physical macroscopic features; generating multiple sets of initial simulation microscopic parameters when the number of calibrations count = 0; performing parallel simulation and data processing on the multiple sets of initial simulation microscopic parameters to acquire the simulation macroscopic features corresponding to the initial simulation microscopic parameters; sending the multiple sets of simulation macroscopic features and physical macroscopic features into a comparison and selection module to determine whether the relative error between the simulation macroscopic features and physical macroscopic features is lower than a set threshold; if so, calibration is completed; if not, based on the simulation data from the parallel simulation of the initial simulation microscopic parameters, an intelligent model between the simulation macroscopic features and physical macroscopic features is established.

[0008] When the number of corrections count = 1, the physical macroscopic features are input into the constructed intelligent model to obtain the corresponding simulation microscopic parameters as initial values. Parallel simulation, data analysis, and comparison of the initial values ​​are performed. The microscopic parameters corresponding to the simulation macroscopic feature with the smallest error are obtained as the initial values ​​for the next round. Parallel simulation, data analysis, and comparison of the initial values ​​are performed repeatedly until the relative error between the simulation macroscopic feature and the physical macroscopic feature in this round is lower than the set threshold or the number of corrections count reaches the set upper limit, and the calibration is completed.

[0009] Furthermore, the acquisition of physical macroscopic features includes:

[0010] Images of the movement of granular materials inside the rotating drum are collected by the physical experiment module and sent to the data processing and analysis module for filtering and binarization. The angle of repose data at different times are extracted, and a second-order model is used to fit the acquired angle of repose data to obtain the dynamic angle of repose, which serves as a physical macroscopic feature.

[0011] Furthermore, when the number of corrections count = 0, generating multiple sets of initial simulation micro-parameters and performing parallel simulation and data processing on these multiple sets of initial simulation micro-parameters includes:

[0012] Using a parallel simulation parameter generation module, parameters between 0 and 1 are generated with a step size of Δμ. b Input parameter μ b (1) ,…,μ b (k) ,μ b (k+1) ,…,μ b (n) , where μ b (1) =Δμ b,μ b (k+1) =μ b (k) +Δμ b (k>0), And create the corresponding simulation folders in batches;

[0013] Parallel simulation is performed using the DEM simulation module to collect simulation data;

[0014] The simulation data is sent to the data processing and analysis module to obtain the simulation macroscopic features corresponding to each input parameter.

[0015] Furthermore, the establishment of the intelligent model between simulated macroscopic features and physical macroscopic features includes:

[0016] Establish a simulation database that connects the initial simulation micro-parameters with their corresponding simulation macro-characteristics;

[0017] Perform data preprocessing, remove outliers, and clean the data;

[0018] The least squares method is used as the core fitting algorithm, and the polynomial function is used as the fitting model to construct an intelligent model between simulated macroscopic features and physical macroscopic features.

[0019] Furthermore, the outlier removal includes obtaining the smoothed value of the simulation macroscopic features output by using the moving average method, determining whether the distance between the simulation macroscopic features and the smoothed value in each group of simulation data is greater than a set value, and if so, removing the group of simulation data; otherwise, retaining it.

[0020] Furthermore, the data processing includes rearranging the simulation data after outlier removal, taking the simulation macroscopic features as input and the simulation microscopic parameters as output.

[0021] Furthermore, the parallel simulation, data analysis, and comparison of the initial values ​​include:

[0022] The initial value of this cycle is μ. b0 As input, with μ b0 Centered on, with a step size of Δμ b Generate equally spaced microscopic parameters to be simulated: μ b (1) ,…,μ b (i) ,…,μ b (k+1) ,…,μ b (n) , where: μ b (k+1) =μ b0 ,μ b (i)=μ b0 +(ik-1)Δμ b (i = 1, ..., n), n = 2k + 1;

[0023] Based on the microscopic parameters to be simulated, create corresponding simulation folders in batches;

[0024] Using the microscopic parameters to be simulated, DEM simulation is performed in parallel to generate simulation data;

[0025] Input the simulation data into the data processing and analysis module to obtain the macroscopic characteristics of each simulation in this round;

[0026] Determine whether the relative error between the simulated macroscopic features and the physical macroscopic features in this round is lower than the set threshold. If yes, end the calibration; otherwise, select the microscopic parameter corresponding to the simulated macroscopic feature with the smallest error as the initial value for the next round.

[0027] Furthermore, the parallel execution of DEM simulation includes:

[0028] Open n DEM simulation software interfaces, and run the simulation of the microscopic parameter μ in each interface. b (1) ,…,μ b (n) The corresponding DEM simulation code is stored, along with the corresponding simulation data, images, and other relevant parameters.

[0029] Furthermore, when completing the calibration, if the calibration termination condition is that the relative error between the simulated macroscopic features and the physical macroscopic features of this round is lower than the set threshold, then the microscopic parameter corresponding to the smallest relative error is selected as the final calibration value. If the calibration termination condition is that the number of corrections reaches the set upper limit, then the microscopic parameter corresponding to the smallest relative error among all rounds is selected as the final calibration value.

[0030] This invention also provides an automatic calibration system for DEM simulation models based on parallel computing, used to execute the above-described automatic calibration method for DEM simulation models based on parallel computing, comprising:

[0031] The physics experiment module is used to acquire physics experiment data;

[0032] The data processing and analysis module is used to obtain physical macroscopic features based on physical experimental data, obtain simulation macroscopic features based on simulation data, and build intelligent models.

[0033] The parallel simulation parameter generation module is used to generate corresponding simulation micro parameters based on the number of calibrations.

[0034] The DEM simulation module is used to perform parallel DEM simulation processing on multiple sets of simulation micro parameters generated by the parallel simulation parameter generation module.

[0035] It also includes a comparison and selection module, which is used to determine the calibration completion status and output the final calibration value.

[0036] Compared with the prior art, the beneficial effects of the present invention are:

[0037] This invention generates multiple sets of equally spaced simulation micro-parameters initially, performs parallel simulations, and establishes an intelligent model. Simultaneously, during subsequent calibration, multiple sets of simulation parameters are generated with the initial value as the center and the same timing, and parallel simulation, data processing, and comparison are performed in a loop to output the final calibration value. Through the above-mentioned parallel data processing method, the problems of cumbersome complexity and long time consumption of traditional calibration methods can be effectively solved. Attached Figure Description

[0038] Figure 1 This is a schematic diagram of the overall principle and structure of the present invention;

[0039] Figure 2 This is a table of physical parameters for parallel simulation in an embodiment of the present invention;

[0040] Figure 3 This is a simulation database table diagram of the first embodiment of the present invention when the number of corrections is 0;

[0041] Figure 4 This is a comparison table of the intelligent model simulation database and physical experiment results of the first embodiment of the present invention;

[0042] Figure 5 This is a table diagram of the intelligent model simulation database of the first embodiment of the present invention;

[0043] Figure 6 This is a schematic diagram of the fitting results of the intelligent model in the first embodiment of the present invention;

[0044] Figure 7 This is a table of parallel simulation parameters when the number of corrections is 1 in the first embodiment of the present invention;

[0045] Figure 8 This is a table of simulated macroscopic parameters when the number of corrections is 1 in the first embodiment of the present invention;

[0046] Figure 9 This is a comparison table of parallel simulation results and physical experiment results when the number of corrections is 1 in the first embodiment of the present invention;

[0047] Figure 10 This is a simulation database table diagram of the second embodiment of the present invention;

[0048] Figure 11 This is a table comparing the parallel simulation results and physical experiment results of the second embodiment of the present invention. Detailed Implementation

[0049] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0050] In the description of this invention, it should be noted that, unless otherwise explicitly specified and limited, the terms "set," "install," "connect," and "link" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal communication between two components. Those skilled in the art can understand the specific meaning of the above terms in this invention based on the specific circumstances.

[0051] In the description of this specification, references to terms such as "embodiment," "one embodiment," and "one implementation" indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or implementation is included in at least one embodiment or illustrative implementation of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or implementation. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or implementations.

[0052] like Figure 1 As shown, this invention provides an automatic calibration system for DEM simulation models based on parallel computing, mainly comprising five modules: physical experiment, data processing and analysis, parallel simulation parameter generation, DEM simulation, and comparative optimization. The main steps of its automatic calibration method are as follows:

[0053] Step 1: Use physical experimental apparatus to acquire experimental data, images and other relevant parameters of the calibration object, and perform experimental data processing and analysis to obtain the macroscopic characteristics fexp of the physical experiment.

[0054] Step 2: When the number of calibrations count = 0, use the parallel simulation parameter generation module to generate parameters between 0 and 1, with a step size of Δμ. b Input parameter μ b (1) ,…,μ b (k) ,μ b (k+1) ,…,μ b (n) , where μ b (1) =Δμ b ,μ b (k+1) =μ b (k) +Δμb (k>0), And create the corresponding simulation folders in batches;

[0055] Next, parallel simulations are performed using the DEM simulation module to collect simulation data; the data is then sent to the data processing and analysis module to obtain the macroscopic features f corresponding to each simulation. s1 f s2 , ..., f sn The simulation macroscopic features and the physical experiment macroscopic features f obtained in Step 1 will be combined. exp The data is fed into the comparison and selection module to determine whether the relative error e between the simulated macroscopic features and the physical features is lower than the set threshold E%. If it is lower than the set threshold, the calibration ends directly, and the microscopic parameter corresponding to the minimum relative error e is output as the calibration value; otherwise, based on the data obtained from the simulation when count=0, an intelligent model between the dynamic angle of repose and the friction coefficient is established, and the calibration continues in the order of Step2 to Step5.

[0056] If count = 1, the obtained macroscopic physical characteristics f exp The input is fed into the constructed intelligent model to predict the corresponding initial values ​​μ of the simulation micro-parameters. b0 Otherwise, continue.

[0057] The initial value μ b0 As input, it is fed into the parallel simulation parameter generation module. This module uses μ... b0 Centered on, with a step size of Δμ b Generate equally spaced microscopic parameters to be simulated: μ b (1) ,…,μ b (i) ,…,μ b (k+1) ,…,μ b (n) , where: μ b (k+1) =μ b0 ,μ b (i) =μ b0 +(ik-1)Δμ b (i = 1, ..., n), n = 2k + 1;

[0058] Based on the generated simulation parameters, corresponding simulation folders are created in batches. These folders contain the main program code required for the simulation and folders for saving simulation data.

[0059] Step 3: Using the simulation parameters generated in Step 2, execute DEM simulations in parallel. Specifically, the calibration platform opens n DEM simulation software interfaces, and each interface runs μ... b (1) ,…,μ b (n) The corresponding DEM simulation code is stored, along with the corresponding simulation data, images, and other relevant parameters.

[0060] Step 4: Input the simulation data generated in Step 3 into the data processing and analysis module to extract the macroscopic features of each simulation.

[0061] Step 5: Calculate the simulated macroscopic features f obtained in Step 4. s1 f s2 , ..., f sn And the physical macroscopic features f obtained in Step 1 exp The data is sent to the comparison and selection module. It determines whether the relative error *e* between the simulated macroscopic features and the physical macroscopic features in the current round is lower than a set threshold *E*. If the minimum relative error *e* is less than *E*, the microscopic parameter corresponding to the minimum relative error *e* is selected as the final calibration value, and the calibration process is terminated. If the minimum relative error *e* in the current round is still greater than *E*, it determines whether the calibration round count has reached a set upper limit *C*. If yes, the calibration is terminated, and the microscopic parameter with the smallest error in all previous simulations is selected as the final calibration value; otherwise, the microscopic parameter corresponding to the minimum relative error *e* is used as the new initial input value *μ*. b0 Run Step 2 through Step 5 again to continue calibration.

[0062] The above process will continue iteratively until either of the following stopping conditions is met: the relative error e is lower than the set threshold E%, or the number of calibration rounds count reaches the set upper limit C. The microscopic parameter with the smallest final output error is taken as the calibration result. The above steps are repeated until the stopping condition is reached, completing the model calibration.

[0063] Furthermore, two embodiments with different particle sizes are selected for detailed explanation, such as... Figure 2 As shown, in this embodiment, the cylinder length and diameter are 67mm and 250mm respectively, and the cylinder rotation speed is 5.91rpm; the particulate material is acrylic spherical particles, and the calibrated micro-parameter is the friction coefficient μ of the particulate material. b The macroscopic feature used is the dynamic rest angle f.

[0064] Example 1:

[0065] Step 1: First, images of the movement of granular material inside the rotating drum are acquired through the physical experiment module and sent to the data processing and analysis module for filtering, binarization, and other processing to extract the angle of repose data at different times [10,11]. Then, a second-order model is used to fit the acquired angle of repose data to obtain the dynamic angle of repose f. exp =15.4°.

[0066] Step 2: At this point, count = 0. Using the parallel simulation parameter generation module, generate a step size of Δμ. b =0.05, n=19, 19 groups of friction coefficients μ with equal intervals between 0 and 1 b and the corresponding simulation folder, such as Figure 3 As shown, parallel simulation is performed using the DEM simulation module.

[0067] 1) Extraction of macroscopic features. The images obtained from 19 simulations were processed and analyzed, following the same steps as the method for extracting physical macroscopic features. The simulation input microscopic parameters (friction coefficient μ) were established. b ) and output macroscopic characteristics (dynamic repose angle f) s Simulation databases between ) such as Figure 3 As shown;

[0068] The dynamic repose angle f obtained from calculation and simulation s With physical dynamic rest angle f exp The relative error e between them is calculated using the following formula:

[0069]

[0070] Some results can be found in Figure 4 As shown, the relative errors e all failed to fall below the set threshold of 5%, with the minimum relative error being e = 10.64%, which does not meet the conditions for ending the calibration.

[0071] 2) Data preprocessing

[0072] a) Outlier Removal. First, a 5-point moving average smoothing filter is used to process the dynamic rest angle of the simulation output when it approaches a stable state. Then, it is determined whether the distance between the simulated output dynamic rest angle and the smoothed value is greater than a set value H (H = 1° in this example). If yes, the set of simulation data pairs is removed to ensure the reliability of the modeling data; otherwise, it is retained. For example, in this case... Figure 3 The dynamic rest angle and smoothing value obtained from the simulation in group 6 are 29.92° and 24.512°, respectively, with a distance greater than 1° between the two values; the dynamic rest angle and smoothing value obtained from the simulation in group 12 are 27.84° and 29.144°, respectively, with a distance greater than 1° between the two values. Therefore, these two sets of data are discarded.

[0073] b) Data preparation. After outlier removal, the macroscopic feature fs is used as input, and the microscopic parameter μ is... b As output, it is rearranged for use in building intelligent models, such as Figure 5 As shown.

[0074] This embodiment uses the Least Squares Method (LSM) as the core fitting algorithm and a polynomial function as the fitting model to construct a dynamic rest angle f. s coefficient of friction with particles μ b The intelligent model between them. The simulation database used for modeling, such as... Figure 5 As shown.

[0075] Different orders can be selected for polynomial models. To ensure that the fitting results have both high accuracy and avoid negative values, the polynomial order that performs best in both aspects was finally selected.

[0076] The fitting results are as follows Figure 6 As shown, its mathematical expression is:

[0077] μ b =0.0000012656f s 5 -0.0000805962f s 4 +0.0017596225f s 3 -0.0156296032f s 2 +0.0617626772f s ;

[0078] Step 2: count = 1, and the physical angle of repose f collected by the physics experiment module is counted. exp =15.4° is used as input to the intelligent model to obtain the initial predicted value of the micro-friction coefficient. Using a parallel simulation parameter generation module, to Centered on, Δμ b =0.02 is the step size, and five simulation input friction coefficients μ are generated using the method described in Step 3. b (n=5), such as Figure 7 As shown:

[0079] To automate the execution of this algorithm, this paper builds a DEM simulation parameter generation module based on the LabVIEW platform. This module can generate DEM simulation parameters based on the initial friction coefficient. Combined with the set step size Δμ bThe value n will automatically create a corresponding number of simulation directories, simulation_i (i = 1, ..., n, n = 5). Each directory contains the input files and script code necessary to run the DEM simulation, such as the main program main.dat, subroutine functions function.dat, etc., and has the ability to run independently.

[0080] Step 3: The calibration platform will simultaneously open 5 DEM simulation software interfaces and call the main program main.dat in the generated simulation directories simulation_1, simulation_2, ... to perform simulations.

[0081] Step 4: Subsequently, the images and particle information obtained from the simulation are transmitted to the MATLAB environment through the built LabVIEW automation platform for further image processing and feature extraction, thereby realizing the identification and extraction of the dynamic repose angle of the simulated macroscopic features.

[0082] The simulation images obtained from the five sets of simulations were processed and analyzed to obtain the corresponding simulation repose angle data, as follows: Figure 8 As shown.

[0083] Step 5: Calculate the dynamic repose angle f for each simulation group. s With physical dynamic rest angle f exp (f exp The relative error e of (=15.4°) is as follows: Figure 9 As shown.

[0084] according to Figure 9 It can be seen that when the simulated friction coefficients are 0.19 and 0.23, the calculated relative errors are both less than the set value of 5%, therefore the calibration process is terminated here. After comparison, the relative error is smallest when the simulated friction coefficient is 0.23, therefore μ is determined. b =0.23 was used as the final calibration value.

[0085] Example 2:

[0086] Step 1: Using the physical experiment module and data processing and analysis module, obtain the dynamic angle of repose f of the granular material with a particle diameter of 4.8 mm in Example 2. exp =29.24°.

[0087] Step 2: At this point, count = 0. Use the parallel simulation parameter generation module to generate parallel simulation parameters (the variable is the friction coefficient μ). b The simulation data includes the corresponding simulation folder; parallel simulation is performed using the DEM simulation module; then, the simulation data is sent to the data processing and analysis module to obtain the macroscopic features (dynamic repose angle f) corresponding to each simulation. s ), specifically as Figure 10 As shown.

[0088] Calculate the dynamic rest angle f for each simulation group. s With physical dynamic rest angle f exp The relative error e, such as Figure 11 As shown.

[0089] according to Figure 11 It can be seen that the error between the simulated dynamic angle of repose and the physical dynamic angle of repose is less than the set value of 5%. Therefore, the calibration is completed, and the friction coefficient of 0.80 corresponding to the minimum relative error of 0.31% is the calibration value.

[0090] While the disclosure is as stated above, its scope of protection is not limited thereto. Those skilled in the art can make various changes and modifications without departing from the spirit and scope of this disclosure, and all such changes and modifications will fall within the protection scope of this invention.

Claims

1. An automatic calibration method for DEM simulation models based on parallel computing, characterized in that, include: When the number of corrections count = 0, multiple sets of initial simulation micro-parameters are generated. Parallel simulation and data processing are performed on these multiple sets of initial simulation micro-parameters to obtain the simulation macro-parameters corresponding to the initial simulation micro-parameters. The multiple sets of simulation macro-parameters and physical macro-parameters are then sent to a comparison and selection module to determine whether the relative error between the simulation macro-parameters and physical macro-parameters is lower than a set threshold. If so, the calibration is completed. If not, an intelligent model between the simulation micro-parameters and physical macro-parameters is established based on the simulation data from the parallel simulation of the initial simulation micro-parameters. When the number of corrections count = 1, the physical macroscopic features are input into the constructed intelligent model to obtain the corresponding simulation microscopic parameters as initial values. Parallel simulation, data analysis, and comparison of the initial values ​​are performed. The microscopic parameters corresponding to the simulation macroscopic feature with the smallest error are obtained as the initial values ​​for the next round. Parallel simulation, data analysis, and comparison of the initial values ​​are performed repeatedly until the relative error between the simulation macroscopic feature and the physical macroscopic feature in this round is lower than the set threshold or the number of corrections count reaches the set upper limit, and the calibration is completed.

2. The automatic calibration method for DEM simulation models based on parallel computing according to claim 1, characterized in that, The acquisition of physical macroscopic features includes: Images of the movement of granular materials inside the rotating drum are collected by the physical experiment module and sent to the data processing and analysis module for filtering and binarization. The angle of repose data at different times are extracted, and a second-order model is used to fit the acquired angle of repose data to obtain the dynamic angle of repose, which serves as a physical macroscopic feature.

3. The automatic calibration method for DEM simulation models based on parallel computing according to claim 1, characterized in that, The step of generating multiple sets of initial simulation micro-parameters when the number of corrections count = 0, and performing parallel simulation and data processing on multiple sets of initial simulation micro-parameters includes: Using a parallel simulation parameter generation module, parameters between 0 and 1 are generated with a step size of [value missing]. Input parameters , …, , , …, ,in, = , = ( ), = , And create the corresponding simulation folders in batches; Parallel simulation is performed using the DEM simulation module to collect simulation data; The simulation data is sent to the data processing and analysis module to obtain the simulation macroscopic features corresponding to each input parameter.

4. The automatic calibration method for DEM simulation models based on parallel computing according to claim 1, characterized in that, The establishment of the intelligent model between simulation microscopic parameters and physical macroscopic characteristics includes: Establish a simulation database that connects the initial simulation micro-parameters with their corresponding simulation macro-characteristics; Perform data preprocessing, remove outliers, and clean the data; The least squares method is used as the core fitting algorithm, and the polynomial function is used as the fitting model to construct an intelligent model between the simulation microscopic parameters and the physical macroscopic characteristics.

5. The automatic calibration method for DEM simulation models based on parallel computing according to claim 4, characterized in that, The outlier removal process involves using a moving average method to obtain a smoothed value of the simulation macroscopic features in the simulation output, determining whether the distance between the simulation macroscopic features and the smoothed value in each group of simulation data is greater than a set value, and if so, removing the group of simulation data; otherwise, retaining it.

6. The automatic calibration method for DEM simulation models based on parallel computing according to claim 5, characterized in that, The data processing involves rearranging the simulation data after outlier removal, taking the simulation macroscopic features as input and the simulation microscopic parameters as output.

7. The automatic calibration method for DEM simulation models based on parallel computing according to claim 1, characterized in that, The parallel simulation, data analysis, and comparison of the initial values ​​include: Set the initial value for this round As input, Centered on, step size is Generate equally spaced microscopic parameters to be simulated: , …, , …, , …, , in: = , (i=1, ..., n), n=2k+1; Based on the microscopic parameters to be simulated, create corresponding simulation folders in batches; Using the microscopic parameters to be simulated, DEM simulation is performed in parallel to generate simulation data; Input the simulation data into the data processing and analysis module to obtain the macroscopic characteristics of each simulation in this round; Determine whether the relative error between the simulated macroscopic features and the physical macroscopic features in this round is lower than the set threshold. If yes, end the calibration; otherwise, select the microscopic parameter corresponding to the simulated macroscopic feature with the smallest error as the initial value for the next round.

8. The automatic calibration method for DEM simulation models based on parallel computing according to claim 7, characterized in that, The parallel execution of DEM simulation includes: Open n DEM simulation software interfaces, and run the simulation of the microscopic parameters in each interface. , …, The corresponding DEM simulation code is stored, along with the corresponding simulation data, images, and other relevant parameters.

9. The automatic calibration method for DEM simulation models based on parallel computing according to claim 1, characterized in that, When calibration is completed, if the calibration termination condition is that the relative error between the simulated macroscopic features and the physical macroscopic features of this round is lower than the set threshold, then the microscopic parameter corresponding to the smallest relative error is selected as the final calibration value. If the calibration termination condition is that the number of corrections reaches the set upper limit, then the microscopic parameter corresponding to the smallest relative error among all rounds is selected as the final calibration value.

10. An automatic calibration system for a DEM simulation model based on parallel computing, used to execute the automatic calibration method for a DEM simulation model based on parallel computing as described in any one of claims 1-9, characterized in that, include: The physics experiment module is used to acquire physics experiment data; The data processing and analysis module is used to obtain physical macroscopic features based on physical experimental data, obtain simulation macroscopic features based on simulation data, and build intelligent models. The parallel simulation parameter generation module is used to generate corresponding simulation micro parameters based on the number of calibrations. The DEM simulation module is used to perform parallel DEM simulation processing on multiple sets of simulation micro parameters generated by the parallel simulation parameter generation module. It also includes a comparison and selection module, which is used to determine the calibration completion status and output the final calibration value.

Images