Contact parameter calibration method and device for discrete element method

Abstract

This invention discloses a discrete element method (DEM) for contact parameter calibration and an apparatus. The method includes: obtaining experimental values ​​of the angle of repose of a target material; simulating the target material based on a pre-established geometric model and the range of contact parameter values ​​to obtain a simulation dataset; training a random forest model based on the simulation dataset and iteratively updating the population according to the obtained prediction error to obtain a trained random forest model; predicting the contact parameter combinations in the simulation dataset based on the trained random forest model to obtain predicted values ​​of the angle of repose corresponding to each contact parameter combination; and generating an optimal contact parameter combination if the predicted value is close to the experimental value. This invention enables automatic inversion and efficient calibration of contact parameters, improving the accuracy and efficiency of parameter calibration.

Description

Technical Field

[0001] This invention relates to the field of machine learning technology, and in particular to a method and apparatus for calibrating contact parameters using the discrete element method. Background Technology

[0002] Hydrogen-based shaft furnaces operate under high-temperature conditions, and the microscopic physical processes such as the flow state, collision behavior, and frictional characteristics of particles within the furnace are difficult to obtain directly using conventional sensors or visualization methods. In current engineering practice, the operating status of the shaft furnace is typically judged indirectly through macroscopic parameters such as outlet gas composition, temperature, and pressure. However, these parameters only reflect the overall operating conditions and cannot accurately characterize the actual motion patterns and local response characteristics of particles within the furnace, thus limiting the refined control of the shaft furnace operation and the effective identification of abnormal operating conditions. To obtain microscopic behavioral information of the particle system, the discrete element method (DEM) is usually used to numerically simulate particle motion and contact processes; however, the computational accuracy of this method is highly dependent on the reasonable setting of contact parameters.

[0003] In existing technologies, the contact parameters of the discrete element method (DEM) are typically calibrated using the angle of repose experiment as a macroscopic characterization quantity. For example, the technical solution with publication number CN120651706A proposes a calibration method based on the dynamic angle of repose. This method uses a rotating cylinder device to create a dynamic accumulation state of the granular material during cylinder rotation. When the granular pile reaches its maximum stable inclination angle without slippage, the angle between the slope of the pile and the horizontal plane is obtained as the dynamic angle of repose. Subsequently, using the dynamic angle of repose as the calibration target value, the experimental process is simulated through discrete element simulation, and the contact parameters are adjusted so that the simulation results gradually approximate the experimentally measured angle of repose, thereby achieving the calibration of the contact parameters.

[0004] However, the above methods still rely heavily on manual trial and error in the contact parameter calibration process. This involves repeatedly adjusting the contact parameters and running discrete element simulations to gradually approximate the experimental results, leading to repeated simulation calculations, high computational costs, and low calibration efficiency. Furthermore, this method struggles to guarantee the uniqueness and stability of the obtained parameter results, especially when dealing with irregularly shaped particles or multi-component mixtures. These problems become more pronounced, making it difficult to meet the efficiency and accuracy requirements for parameter calibration in industrial applications.

[0005] This section is intended to provide background or context for the embodiments of the invention set forth in the claims. The description herein is not an admission that it is prior art simply because it is included in this section. Summary of the Invention

[0006] This invention provides a discrete element method for contact parameter calibration, which enables automatic inversion and efficient calibration of contact parameters, thereby improving the accuracy and efficiency of parameter calibration.

[0007] Discrete element method contact parameter calibration methods include: Obtain the experimental value of the angle of repose of the target material; The target material is simulated based on a pre-established geometric model and the range of contact parameters to obtain a simulation dataset. The random forest model is trained based on the simulation dataset, and the population is iteratively updated according to the obtained prediction error to obtain the trained random forest model. Based on the trained random forest model, the contact parameter combinations in the simulation dataset are predicted to obtain the predicted values ​​of the repose angles corresponding to each contact parameter combination. If the predicted value is close to the experimental value, a preferred combination of contact parameters is generated.

[0008] In some embodiments, obtaining the experimental value of the angle of repose of the target material includes: The target material is filled into a pre-set experimental chamber; the ratio of the volume of the target material to the volume of the experimental chamber is greater than or equal to 85%. Remove the side door of the experimental chamber to allow the target material to flow out of the experimental chamber, forming a slope and acquiring an image of the slope; The slope image is processed to obtain the experimental values.

[0009] In some embodiments, the image processing of the slope image to obtain the experimental values ​​includes: The slope image is processed by grayscale and binarization to separate the target material region; Based on the preset image size and number of intervals, the target material area is divided into multiple intervals; The target pixel and its pixel coordinates within each interval are determined based on the ordinate of the pixel within each interval. Linear regression fitting is performed on the pixel coordinates to determine the angle between the fitted line and the horizontal axis, thus obtaining the experimental value of the rest angle.

[0010] In some embodiments, the simulation of the target material based on a pre-established geometric model of the target material and the range of contact parameters to obtain a simulation dataset includes: The geometric model is established using discrete element simulation software; Multiple combinations of contact parameters are determined based on the preset value range of the contact parameters; Discrete element simulations are performed based on the geometric model and each combination of contact parameters to generate the simulation dataset. The simulation dataset includes: multiple sets of contact parameter combinations and corresponding repose angle data pairs; the contact parameters include: coefficient of restitution, static friction coefficient and rolling friction coefficient.

[0011] In some embodiments, the geometric model includes a spherical particle model or a multi-sphere model; establishing the geometric model in discrete element simulation software includes: If the target material is a spherical particle, select the spherical particle model in the discrete element simulation software; If the target material is an irregularly shaped particle, a multi-sphere model is constructed based on the target material.

[0012] In some embodiments, constructing a multi-sphere model based on the target material includes: The surface of the target material is scanned using a 3D laser scanner to obtain point cloud data; wherein, the point cloud data is a set of coordinate points on the surface; The point cloud data is reconstructed to obtain a closed triangular network; Feature points are extracted from the closed triangular mesh, and clustering is performed on the feature points to obtain a multi-sphere model.

[0013] In some embodiments, training the random forest model based on the simulation dataset and iteratively updating the population according to the obtained prediction error to obtain the trained random forest model includes: Set the range of values ​​for the hyperparameters of the random forest model, and initialize the population based on the range of values; A corresponding random forest model is constructed for each individual chromosome in the population, and each random forest model is trained based on the simulation dataset to obtain the prediction error; Based on the prediction errors described above, the population is updated through selection, crossover, and mutation operations; The above steps are executed iteratively until the preset number of iterations is reached or the prediction error converges, thus obtaining the optimal individual chromosome. Based on the hyperparameters corresponding to the selected individual chromosomes, the corresponding random forest model is retrained to obtain the trained random forest model.

[0014] In some embodiments, the step of constructing a corresponding random forest model for each individual chromosome in the population, and training each random forest model based on the simulation dataset to obtain the prediction error includes: The simulation dataset is randomly sampled to generate multiple sub-datasets; Decision trees are constructed based on the aforementioned subsets of data to generate a random forest model; The random forest model is trained based on the hyperparameters corresponding to the individual chromosomes and the simulation dataset to obtain prediction results; The prediction error is determined based on the prediction results and the simulated values ​​of the angle of repose in the simulation dataset.

[0015] This invention also provides a discrete element method contact parameter calibration device to achieve automatic inversion and efficient calibration of contact parameters, thereby improving the accuracy and efficiency of parameter calibration.

[0016] The discrete element method contact parameter calibration device includes: The angle of repose acquisition module is used to obtain the experimental value of the angle of repose of the target material; The simulation data construction module is used to simulate the target material based on the pre-established geometric model of the target material and the range of values ​​of the contact parameters, and obtain a simulation dataset. The model training and optimization module is used to train the random forest model based on the simulation dataset and perform iterative updates of the population according to the obtained prediction error to obtain the trained random forest model. The parameter prediction module is used to predict the combination of contact parameters in the simulation dataset based on the trained random forest model, and obtain the predicted value of the angle of repose corresponding to each combination of contact parameters. The parameter filtering module is used to generate a preferred combination of contact parameters if the predicted value is close to the experimental value.

[0017] This invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the above-described discrete element method contact parameter calibration method.

[0018] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described discrete element method contact parameter calibration method.

[0019] This invention also provides a computer program product, which includes a computer program that, when executed by a processor, implements the above-described discrete element method contact parameter calibration method.

[0020] The discrete element method (DEM) contact parameter calibration method and apparatus provided in this invention combine experimental data on the angle of repose with DEM simulation results, introduce a random forest model to establish a mapping relationship between contact parameters and the angle of repose, and optimize the random forest model based on prediction errors using a population iterative update mechanism. This achieves rapid prediction and selection of contact parameter combinations, thus replacing the traditional parameter calibration method that relies on repeated manual calculations. Compared with existing technologies, this invention can significantly reduce the number of calculations in the DEM simulation while ensuring calibration accuracy, improve the efficiency and automation of contact parameter calibration, and enhance the consistency and stability of contact parameter combination results. It is particularly suitable for parameter calibration scenarios of irregular particles or multi-component systems. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings: Figure 1 This is a flowchart illustrating a contact parameter calibration method using the discrete element method in one embodiment of the present invention. Figure 2 This is a flowchart illustrating the contact parameter calibration method of the discrete element method in another embodiment of the present invention; Figure 3 This is a schematic diagram of a ore pile experiment in an embodiment of the present invention; Figure 4 This is a flowchart illustrating the contact parameter calibration method of the discrete element method in another embodiment of the present invention; Figure 5A , Figure 5B and Figure 5C These are the ramp diagram and curve fitting diagram in the embodiments of the present invention; Figure 6 This is a flowchart illustrating the contact parameter calibration method of the discrete element method in another embodiment of the present invention; Figure 7 This is a flowchart illustrating the contact parameter calibration method of the discrete element method in another embodiment of the present invention; Figure 8 This is a flowchart illustrating the contact parameter calibration method of the discrete element method in another embodiment of the present invention; Figure 9 This is a process diagram of the 3D laser scanner acquiring the actual shape of particles in an embodiment of the present invention; Figure 10 This is a modeling diagram of ore morphology in an embodiment of the present invention; Figure 11This is a flowchart illustrating the contact parameter calibration method of the discrete element method in another embodiment of the present invention; Figure 12 This is a flowchart illustrating the contact parameter calibration method of the discrete element method in another embodiment of the present invention; Figure 13A , Figure 13B , Figure 13C and Figure 13D This is a diagram showing the relationship between contact parameters and the angle of repose in an embodiment of the present invention; Figure 14 This is a schematic diagram of the contact parameter calibration device for the discrete element method in an embodiment of the present invention; Figure 15 This is a schematic diagram of the physical structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. Here, the illustrative embodiments and their descriptions are used to explain the present invention, but are not intended to limit the present invention. It should be noted that, unless otherwise specified, the embodiments and features in the embodiments of this application can be arbitrarily combined with each other. The acquisition, storage, use, and processing of data in the technical solutions of this application all comply with relevant laws and regulations. The user information in the embodiments of this application is obtained through legal and compliant means, and the acquisition, storage, use, and processing of user information have been authorized and agreed upon by the customer.

[0023] To facilitate understanding of the technical solution provided in this application, the relevant content of the technical solution in this application will be explained below.

[0024] To address the problems of high computational costs and difficulty in guaranteeing parameter uniqueness and accuracy in existing Discrete Element Method (DEM) contact parameter calibration processes, which rely on repeated manual calculations, this application provides a DEM contact parameter calibration method. This method obtains experimental values ​​of the angle of repose of the target material and constructs a simulation dataset relating contact parameters to the angle of repose using DEM simulation. A random forest model is then introduced to train the simulation dataset, establishing a mapping relationship between contact parameters and the angle of repose. Based on the prediction error, a genetic algorithm is used to optimize the hyperparameters of the random forest model. On this basis, the trained random forest model is used to predict and filter contact parameter combinations, thereby achieving automatic inversion calibration of contact parameters, improving the efficiency and accuracy of parameter calibration, and reducing reliance on manual experience.

[0025] This invention provides a method for calibrating contact parameters using the discrete element method. For example... Figure 1As shown, the contact parameter calibration method includes steps 101 to 105: Step 101: Obtain the experimental value of the angle of repose of the target material.

[0026] Step 102: Simulate the target material based on the pre-established geometric model of the target material and the range of contact parameters to obtain a simulation dataset.

[0027] Step 103: Train the random forest model based on the simulation dataset, and perform iterative updates of the population according to the obtained prediction error to obtain the trained random forest model.

[0028] Step 104: Based on the trained random forest model, predict the combination of contact parameters in the simulation dataset to obtain the predicted value of the angle of repose corresponding to each combination of contact parameters.

[0029] Step 105: If the predicted value is close to the experimental value, generate the preferred combination of contact parameters.

[0030] According to the above embodiments, by combining experimental data on the angle of repose with discrete element simulation results, a random forest model is introduced to establish a mapping relationship between contact parameters and the angle of repose. Based on the prediction error, a population iterative update mechanism is used to optimize the random forest model, achieving rapid prediction and selection of contact parameter combinations. This replaces the traditional parameter calibration method that relies on repeated manual calculations. Compared to existing technologies, this invention significantly reduces the number of calculations in discrete element simulations while ensuring calibration accuracy, improving the efficiency and automation of contact parameter calibration. It also enhances the consistency and stability of contact parameter combination results, making it particularly suitable for parameter calibration scenarios involving irregular particles or multi-component systems.

[0031] In some embodiments, such as Figure 2 As shown, step 101 includes steps 201 to 203.

[0032] Step 201: Fill the pre-set experimental chamber with the target material. The ratio of the volume of the target material to the volume of the experimental chamber is greater than or equal to 85%.

[0033] Step 202: Remove the side door of the test chamber to allow the target material to flow out of the test chamber, forming a ramp and acquiring an image of the ramp.

[0034] Step 203: Perform image processing on the slope image to obtain experimental values.

[0035] In this embodiment of the invention, pellet ore is selected as the target material, but lump ore can also be used as the target material; the invention is not limited thereto.

[0036] The self-flow angle experiment simulates the accumulation behavior of the target material in both individual and mixed states. Methods for measuring the self-flow angle include injection, discharge, and tilting methods; this invention employs the tilting method.

[0037] like Figure 3 As shown, pellets are selected as the target material for illustration, but lump ore can also be used; this invention is not limited to this. The packing characteristics of the target material in single and mixed states are characterized by a free-flow angle experiment. Methods for measuring the free-flow angle include the injection method, discharge method, and tilting method; this embodiment preferably uses the tilting method for measurement.

[0038] like Figure 3 As shown, the experimental setup is an experimental chamber with dimensions of 0.4 m × 0.4 m × 0.4 m. The chamber has detachable magnetic gates on both sides. The entire chamber is constructed of a 6063-T5 aluminum alloy frame, with an open top, a magnesium-silicon-aluminum alloy bottom, and transparent acrylic sidewalls to facilitate observation of particle packing morphology.

[0039] During the experiment, the target material was filled into the experimental chamber, ensuring that its volume occupied more than 85% of the chamber's capacity. Figure 3 As shown in (a). After the target material reaches a static state inside the experimental chamber, the magnetic gate is removed, allowing the granular material to flow out from the side of the chamber under gravity and undergo rearrangement. When the particles stop flowing and form a stable accumulation structure, a stable ramp extending from the bottom to the front is formed inside the experimental chamber, as shown in (a). Figure 3 As shown in (b).

[0040] Under stable accumulation conditions, images of the slope were acquired using a camera device, and the angle between the ore pile slope and the horizontal reference plane was calculated based on image processing methods to obtain the angle of repose of the target material. To improve the reliability of the measurement results, the above-mentioned free-flow angle experiment was repeated multiple times, and the results of each measurement were statistically averaged to obtain an average angle of repose of 30.8° for the pellets.

[0041] According to the above embodiments, by standardizing the filling, controlling the release, and obtaining the angle of repose experimental value through image processing of the target material, a quantitative characterization of particle packing behavior is achieved. Filling the experimental chamber to more than 85% capacity helps ensure the consistency of the initial packing state and reduces measurement errors caused by filling differences. Removing the side door allows the material to flow out naturally under gravity and form a stable slope, accurately reflecting the natural packing characteristics of the particle system. Combining image processing methods to extract and calculate the slope profile improves the accuracy and objectivity of the angle of repose measurement compared to manual measurement, reducing the influence of human interference. Simultaneously, this method exhibits good repeatability and stability, providing reliable experimental benchmark data for subsequent discrete element simulation parameter calibration, thereby improving the accuracy and reliability of the overall parameter calibration process.

[0042] In some embodiments, such as Figure 4 As shown, step 203 includes steps 401 to 404.

[0043] Step 401: Perform grayscale and binarization processing on the slope image to separate the target material region.

[0044] Step 402: Based on the preset image size and number of intervals, divide the target material area into multiple intervals.

[0045] Step 403: Determine the target pixel and its pixel coordinates in each interval based on the ordinate of the pixel in each interval.

[0046] Step 404: Perform linear regression fitting on the pixel coordinates to determine the angle between the fitted line and the horizontal axis, and obtain the experimental value of the rest angle. .

[0047] In this embodiment of the invention, image processing technology is used to process the slope images obtained from each group of experiments in order to obtain the angle of repose of the target material.

[0048] Specifically, the slope image is first preprocessed to separate the ore pile area, and the image resolution is uniformly adjusted to 400×400 pixels, such as... Figure 5A As shown. Subsequently, the slope image is converted into a grayscale image and binarized based on a preset threshold to obtain a binary image of the ore pile outline.

[0049] After determining the image size, the physical dimensions of the image were set to 400 mm × 400 mm, and the ore heap area was divided into 20 sections horizontally. Within each section, the black pixel with the largest ordinate value was identified, and its corresponding pixel coordinates (i.e., Y-coordinate) were recorded, such as... Figure 5B As shown. Further, the pixel coordinates are converted into corresponding physical coordinates to extract the ore pile contour point set.

[0050] Based on the above contour point set, a linear regression method is used to fit a straight line, and the angle between the fitted line and the horizontal reference axis is calculated (e.g., ...). Figure 5C As shown in the figure, the included angle is the angle of repose of the target material.

[0051] To improve the stability and reliability of the measurement results, the experiments were repeated multiple times, and the obtained angles of repose were averaged to obtain the average angles of repose under different ore ratios.

[0052] According to the above embodiments, by standardizing the slope image through grayscale and binarization, effective separation of the ore pile area from the background is achieved. By partitioning the target area according to a preset image size and number of intervals, and selecting representative target pixels within each interval, the impact of local noise and irregular particle accumulation on the overall contour extraction is effectively reduced. Linear regression fitting using pixel coordinates transforms discrete boundary points into continuous slope features, thereby improving the stability and accuracy of the angle of repose calculation. Compared to traditional manual reading or single-point measurement methods, this scheme enables automated processing and batch calculation, significantly improving measurement efficiency while reducing human error and enhancing the consistency and repeatability of experimental results.

[0053] In some embodiments, such as Figure 6 As shown, step 102 includes steps 601 to 603.

[0054] Step 601: Establish a geometric model in the discrete element simulation software.

[0055] Step 602: Determine multiple combinations of contact parameters based on the preset range of contact parameter values.

[0056] Step 603: Perform discrete element simulations based on the geometric model and each combination of contact parameters to generate a simulation dataset. The simulation dataset includes multiple sets of data pairs of contact parameter combinations and their corresponding angles of repose. Contact parameters include: coefficient of restitution e, static friction coefficient... and rolling friction coefficient .

[0057] According to the above embodiments, by establishing a geometric model matching the actual particle morphology in discrete element simulation software and constructing multiple parameter combinations within a preset range of contact parameter values, multi-condition simulation calculations can be systematically carried out to obtain the correspondence data between contact parameters and the angle of repose. Compared with the traditional trial-and-error method that relies on adjusting a single parameter, this scheme generates a structured dataset through batch simulation, which not only improves the coverage and representativeness of the parameter space but also significantly reduces the computational cost of repeated simulations, providing a high-quality data foundation for subsequent machine learning model training, thereby improving the efficiency and accuracy of contact parameter calibration.

[0058] In some embodiments, the geometric model includes: a spherical particle model or a multi-sphere model. For example... Figure 7 As shown, step 601 includes steps 701 to 702.

[0059] Step 701: If the target material is a spherical particle, select the spherical particle model in the discrete element simulation software.

[0060] Step 702: If the target material is an irregularly shaped particle, construct a multi-sphere model based on the target material.

[0061] In this embodiment of the invention, different particle modeling methods are employed in the Discrete Element Method (DEM) simulation based on the different shapes of the target material particles. When the target material particles are approximately spherical, a spherical particle model is used for modeling in the DEM simulation software. When the target material particles are irregularly shaped, a non-spherical particle modeling method is used for simulation. Non-spherical particle modeling methods include: ideal polyhedron model, smooth polyhedron model, multi-sphere model, and hyperellipsoidal model, etc. In this embodiment, a multi-sphere model is preferably used to model the blocky ore particles, and the irregular particle geometry is approximated by multiple interconnected spheres.

[0062] According to the above embodiments, by adaptively selecting the particle modeling method based on the shape characteristics of the target material particles, a spherical particle model can be directly used for near-spherical particles, which can reduce model complexity and improve simulation efficiency. For irregularly shaped particles, a multi-sphere model is used for approximate modeling. While ensuring the accuracy of particle geometric feature representation, this avoids the computational burden brought by complex polyhedral models. Thus, while ensuring simulation accuracy, computational efficiency and stability are improved, making discrete element simulation closer to the physical behavior of actual particle systems and enhancing the applicability and reliability of the model.

[0063] In some embodiments, such as Figure 8 As shown, step 702 includes steps 801 to 803.

[0064] Step 801: Use a 3D laser scanner to scan the surface of the target material to obtain point cloud data. The point cloud data is a set of coordinate points on the surface.

[0065] Step 802: Reconstruct the point cloud data to obtain a closed triangular network.

[0066] Step 803: Extract feature points from the closed triangular mesh and perform clustering on the feature points to obtain a multi-sphere model.

[0067] In embodiments of the present invention, such as Figure 9 As shown, a 3D laser scanner is used to scan blocky ore particles. This scanning process employs a point measurement method, using a laser beam to scan the particle surface point by point. A sensor records the spatial coordinate information of the particle surface, thereby acquiring high-precision point cloud data. During the scanning process, to improve modeling efficiency, the microscopic roughness and minute pores on the particle surface are appropriately ignored, and the overall geometric contour features of the particles are obtained using a 3D laser scanner with a minimum resolution of 0.02 mm.

[0068] The obtained point cloud data is first preprocessed, including denoising, outlier removal, and point cloud smoothing to improve data quality. Then, the point cloud data undergoes coordinate unification and density equalization, and is converted into Standard Triangulated Language (STL) format using a triangulation algorithm (such as Poisson reconstruction or Delaunay triangulation) to obtain a closed triangulated mesh model (watertight mesh) used to characterize the shape contour of the particles.

[0069] Based on this, key feature points are extracted from the closed triangular mesh, and clustering algorithms (such as K-means clustering) are used to group these feature points to determine several representative sphere center positions. The sphere radii are then determined based on the sphere center positions and local geometric features, thus constructing a multi-sphere model (clump model) composed of multiple spheres. By adjusting the number and distribution of spheres, the model complexity is reduced while maintaining the accuracy of particle shape approximation, thus balancing simulation accuracy and computational efficiency.

[0070] Finally, the constructed multi-sphere model is imported into the discrete element method (DEM) simulation software for subsequent numerical simulation of the particle system's motion and contact behavior.

[0071] like Figure 10As shown in the embodiment of the present invention, the geometric morphology of blocky ore particles in different particle size ranges is obtained by three-dimensional laser scanning technology, and two to three representative particle shapes are selected for modeling and analysis for each particle size range. Figure 10 In the diagram, A1 to C1 represent the actual shape of the blocky ore particles, A2 to C2 represent the geometric model reconstructed based on three-dimensional laser scanning data, A3 to C3 represent the multi-sphere model constructed with a higher number of spheres, and A4 to C4 represent the multi-sphere model constructed with a lower number of spheres.

[0072] Specifically, in the process of multi-sphere modeling, Figure 10 In the model, A3, B3, and C3 are composed of 100, 100, and 50 spheres, respectively, to approximate the actual geometry of the particles with high accuracy. A4, B4, and C4 are constructed using 40, 20, and 15 spheres, respectively, reducing the number of spheres while maintaining the overall outline characteristics of the particles, thereby reducing the complexity of the model.

[0073] According to the above embodiments, by reducing the number of spheres in the multi-sphere model, the computational scale of discrete element simulation can be significantly reduced, and the computational efficiency can be improved. This is especially true in the simulation of large-scale multi-particle systems, where the advantages are more obvious. At the same time, an effective balance can be achieved between computational efficiency and shape representation accuracy.

[0074] In some embodiments, step 602 specifically includes: setting the range of contact parameters for the pellet ore in the discrete element method (DEM) simulation software. The contact parameters include: coefficient of restitution. static friction coefficient and rolling friction coefficient Among them, the coefficient of recovery The value range is 0.2~0.8, and the static friction coefficient is... The value range is 0.2~0.8, and the rolling friction coefficient is... The value range is 0.1 to 0.7.

[0075] Based on this, the range of contact parameters is discretized into four levels to form a parameter level table (as shown in Table 1).

[0076] Table 1

[0077] Subsequently, the first three columns of the L16 (4^3) orthogonal experimental table were selected to analyze the coefficients of restitution. static friction coefficient and rolling friction coefficient A combined design was performed, resulting in 16 representative combinations of contact parameters (as shown in Table 2). By employing orthogonal experimental design, the number of experiments was reduced while achieving uniform sampling of the multi-parameter space, thus improving the representativeness and coverage of the parameter combinations.

[0078] Table 2

[0079] In some embodiments, step 603 specifically includes: performing repose angle simulation calculations in discrete element simulation software for each of the 16 sets of contact parameter combinations, recording the repose angle results corresponding to each set of contact parameters, thereby constructing a simulation dataset consisting of contact parameter combinations and repose angles (as shown in Table 3). This simulation dataset can be used for subsequent training of machine learning models to establish the mapping relationship between contact parameters and repose angles.

[0080] Table 3

[0081] In some embodiments, such as Figure 11 As shown, step 103 includes steps 1101 to 1105.

[0082] Step 1101: Set the range of hyperparameter values ​​for the random forest model and initialize the population based on the range. Each individual chromosome corresponds to a set of hyperparameter combinations.

[0083] Step 1102: Construct a corresponding random forest model for each individual chromosome in the population, and train each random forest model based on the simulation dataset to obtain the prediction error of each random forest model.

[0084] Step 1103: Based on each prediction error, update the population through selection, crossover, and mutation operations.

[0085] Step 1104: Iterate through the above steps until the preset number of iterations is reached or the prediction error converges to obtain the optimal individual chromosome.

[0086] Step 1105: Based on the hyperparameter combination corresponding to the optimal individual chromosome, retrain the random forest model to obtain the trained random forest model.

[0087] In this embodiment of the invention, the obtained simulation dataset is input into a random forest model for training. The simulation dataset includes: input features and output labels. The input features are combinations of contact parameters. The output label is the simulated value of the angle of repose corresponding to the contact parameter combination. .

[0088] Before model training, the simulation dataset is randomly divided into training and test sets in an 8:2 ratio, and a random seed (random_state) is set to ensure the repeatability of the partitioning process.

[0089] During model construction, the hyperparameter values ​​for the random forest model are set, including: the number of decision trees (n_estimators, ranging from 10 to 200), the decision tree depth (max_depth, ranging from 1 to 20), and whether to enable out-of-bag (oob_score) evaluation. The hyperparameters of the random forest model are encoded as individual chromosomes. Each individual chromosome corresponds to a set of random forest model hyperparameter combinations. The population is initialized according to the preset hyperparameter values, and the population size is set to 10.

[0090] In some embodiments, such as Figure 12 As shown, step 1102 includes steps 1201 to 1204.

[0091] Step 1201: Randomly sample the simulation dataset to generate multiple sub-datasets.

[0092] Step 1202: Construct decision trees based on the subsets of data to generate a random forest model.

[0093] Step 1203: Train the random forest model based on the hyperparameters corresponding to individual chromosomes and the simulation dataset to obtain prediction results.

[0094] Step 1204: Determine the prediction error based on the prediction results and the simulated values ​​of the rest angle in the simulation dataset.

[0095] In this embodiment of the invention, a corresponding random forest model is constructed based on each hyperparameter combination. Each random forest model uses a bootstrap sampling method to sample with replacement from the training set, generating multiple training subsets. A decision tree is then constructed based on each training subset, and each decision tree is denoted as […]. A random forest model, composed of multiple decision trees, is used to establish a nonlinear mapping relationship between contact parameters and the angle of repose.

[0096] The corresponding random forest models are trained based on the training set and various hyperparameter combinations to obtain the prediction errors of each random forest model.

[0097] Specifically, each training subset The inputs are fed into the corresponding decision trees in the random forest model to obtain the prediction results of each decision tree. The random forest model obtains the final predicted value by averaging the predictions of all decision trees.

[0098] in, For the number of decision trees, This is the predicted value of the angle of repose. The prediction results for each decision tree.

[0099] For example, the evaluation function for prediction error can be the mean squared error (MSE), or other evaluation functions.

[0100] The predicted values ​​of the random forest model and the corresponding simulated value of the angle of repose. Substitute into the following formula (2) to calculate the mean squared error (MSE) of the random forest model.

[0101]

[0102] in, The predicted value of the angle of repose. is the simulated value of the angle of repose, and n is the number of samples.

[0103] Based on the prediction error, selection, crossover, and mutation operations are performed on the population to achieve iterative updates until a preset number of iterations or fitness convergence condition is met, thereby obtaining the optimal hyperparameter combination. The random forest model is then retrained based on the optimal hyperparameters to obtain the trained random forest model.

[0104] According to the above embodiments, by introducing a genetic algorithm to adaptively optimize the hyperparameters of the random forest model, automatic optimization of the model structure parameters is achieved. Compared with traditional methods that rely on manual experience or fixed parameter settings, this approach can efficiently search for optimal solutions within a larger parameter space, improving model performance and stability. Iterative updates of the population based on prediction errors enable the model training process to have global optimization capabilities, reducing the risk of getting trapped in local optima. Retraining the random forest model after obtaining the optimal hyperparameters further improves the fitting accuracy of the mapping relationship between contact parameters and the angle of repose, thereby enhancing the accuracy and reliability of subsequent contact parameter inversion calibration.

[0105] In some embodiments, during the model evaluation phase, the trained random forest model is externally validated using a test set, and evaluation metrics are calculated based on the predicted values ​​from the test set and the corresponding simulated rest angle values. These evaluation metrics include: Mean Squared Error (MSE), Mean Absolute Error (MAE), and Coefficient of Determination (R²). 2 This is used to evaluate the model's prediction accuracy and fitting ability.

[0106] Simultaneously, an out-of-bag (OOB) evaluation mechanism is enabled during model training. This mechanism utilizes samples not sampled during the bootstrap sampling process as validation data to evaluate the prediction performance of the corresponding decision tree and statistically analyzes the out-of-bag prediction results to calculate the OOB error. By combining the test set evaluation results with the OOB evaluation results, the generalization ability of the model can be effectively evaluated without the need for additional datasets, thus improving the reliability of the random forest model evaluation.

[0107] In some embodiments, step 104 specifically includes: using the trained random forest model to process each combination of contact parameters in the simulation dataset. Make a prediction to obtain the corresponding predicted value of the angle of repose. This establishes the correspondence between the combination of contact parameters and the predicted value of the angle of repose.

[0108] In some embodiments, step 105 specifically includes: predicting the angle of repose. With the corresponding experimental value of the angle of repose Compare the prediction errors. When the value is less than a preset threshold, the random forest model is considered to have high prediction accuracy.

[0109] If the deviation between the predicted value and the experimental value is greater than a preset threshold, the random forest model will be iteratively optimized.

[0110] Specifically, steps 102 to 104 can be repeated to gradually improve the ability of the random forest model to fit the relationship between the contact parameters and the angle of repose by adjusting the range of contact parameter values, updating the simulation dataset, or retraining the random forest model, until the prediction results of the random forest model meet the preset accuracy requirements.

[0111] In some embodiments, Figure 13A , Figure 13B , Figure 13C and Figure 13D Showing the combination of contact parameters (coefficient of restitution) static friction coefficient and rolling friction coefficient The relationship between the angle of repose and the angle of rest.

[0112] Specifically, such as Figure 13A As shown, the coefficient of restitution The negative correlation between the angle of repose and the particle's collision recovery capability (correlation coefficient r = -0.30) indicates that as the collision recovery capability of particles increases, the energy dissipation of the particle system decreases, the packing stability decreases, and thus the angle of repose decreases. Figure 13B and Figure 13CAs shown, the static friction coefficient and rolling friction coefficient Both are positively correlated with the angle of repose (correlation coefficients are r = 0.66 and r = 0.43, respectively), indicating that the increase in interparticle frictional resistance will inhibit particle slippage and rolling, which is conducive to the formation of a larger accumulation slope.

[0113] Furthermore, such as Figure 13B As shown, when the rolling friction coefficient At that time, the predicted value of the angle of repose was closer to the experimental average of 40.8°. For example... Figure 13C As shown, when the static friction coefficient At that time, the predicted value of the angle of repose was closer to the experimental average of 40.8°.

[0114] Figure 13D A comparison chart of the prediction results based on the random forest model and the DEM simulation results is shown. The predicted values ​​and simulated values ​​are basically distributed near the diagonal, showing high consistency. The evaluation indicators are: mean squared error (MSE) of 0.3744, root mean square error (RMSE) of 0.6119, mean absolute percentage error (MAPE) of 1.17%, and goodness of fit (R²). 2 The accuracy reached 0.95, indicating that the established random forest model can accurately characterize the nonlinear mapping relationship between contact parameters and the angle of repose, and has high prediction accuracy and generalization ability, thus verifying the effectiveness and reliability of the method of the present invention in contact parameter calibration.

[0115] This application provides a discrete element method (DIM) contact parameter calibration device, applied to the aforementioned DIM contact parameter calibration method. This DIM contact parameter calibration device and the DIM contact parameter calibration method in one embodiment of this application are based on the same inventive concept and have similar problem-solving principles. Therefore, the implementation of the DIM contact parameter calibration device is the same as the implementation of the DIM contact parameter calibration method in one embodiment of this application, and repeated details will not be described again. As used below, the terms "unit" or "module" can refer to a combination of software and / or hardware that performs a predetermined function. Although the system described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0116] like Figure 14 As shown, the present invention provides a discrete element method contact parameter calibration device 1400, comprising: The angle of repose acquisition module 1401 is used to acquire the experimental value of the angle of repose of the target material.

[0117] The simulation data construction module 1402 is used to simulate the target material based on the pre-established geometric model of the target material and the range of values ​​of the contact parameters, and to obtain a simulation dataset.

[0118] The model training and optimization module 1403 is used to train the random forest model based on the simulation dataset and perform iterative updates of the population according to the obtained prediction error to obtain the trained random forest model.

[0119] The parameter prediction module 1404 is used to predict the combination of contact parameters in the simulation dataset based on the trained random forest model, and obtain the predicted value of the angle of repose corresponding to each combination of contact parameters.

[0120] The parameter filtering module 1405 is used to generate an optimal combination of contact parameters if the predicted value is close to the experimental value.

[0121] In some embodiments, the angle of repose acquisition module includes: The material filling submodule is used to fill the target material into a pre-set experimental chamber. The ratio of the target material volume to the experimental chamber volume is greater than or equal to 85%.

[0122] The ramp forming submodule is used to remove the side door of the experimental chamber to allow the target material to flow out of the experimental chamber, forming a ramp and acquiring a ramp image.

[0123] The image processing submodule is used to process the slope image to obtain experimental values.

[0124] In some embodiments, the image processing submodule includes: The image preprocessing unit is used to perform grayscale and binarization processing on the slope image to separate the target material region.

[0125] The region division unit is used to divide the target material region into multiple intervals based on the preset image size and number of intervals.

[0126] The feature extraction unit is used to determine the target pixel and its pixel coordinates in each interval based on the ordinate of the pixel in each interval.

[0127] The angle calculation unit is used to perform linear regression fitting on the pixel coordinates, determine the angle between the fitted line and the horizontal axis, and obtain the experimental value of the rest angle.

[0128] In some embodiments, the simulation data construction module includes: The geometric modeling submodule is used to create geometric models in discrete element simulation software.

[0129] The parameter combination generation submodule is used to determine multiple sets of contact parameter combinations based on the preset value range of the contact parameters.

[0130] The simulation execution submodule is used to perform discrete element simulations based on the geometric model and various combinations of contact parameters, generating simulation datasets. These datasets include multiple sets of data pairs of contact parameter combinations and their corresponding angles of repose. The contact parameters include the coefficient of restitution, the static friction coefficient, and the rolling friction coefficient.

[0131] In some embodiments, the geometric model includes a spherical particle model or a multi-sphere model. The geometric modeling submodule includes: The model selection unit is used to select a spherical particle model in the discrete element simulation software if the target material is a spherical particle.

[0132] The multi-sphere modeling unit is used to construct a multi-sphere model based on the target material if the target material is an irregularly shaped particle.

[0133] In some embodiments, the multi-sphere modeling unit includes: The point cloud acquisition unit is used to scan the surface of the target material using a 3D laser scanner to obtain point cloud data. The point cloud data is a set of coordinate points on the surface.

[0134] The mesh reconstruction unit is used to reconstruct point cloud data to obtain a closed triangular network.

[0135] Feature clustering units are used to extract feature points from closed triangular meshes and perform clustering on the feature points to obtain a multi-sphere model.

[0136] In some embodiments, the model training optimization module includes: The population initialization submodule is used to set the range of values ​​for the hyperparameters of the random forest model and initialize the population based on the range of values.

[0137] The model building and training submodule is used to build corresponding random forest models for each individual chromosome in the population, and train each random forest model based on the simulation dataset to obtain the prediction error.

[0138] The population update submodule is used to update the population based on each prediction error through selection, crossover, and mutation operations.

[0139] The iterative optimization submodule is used to iteratively execute the above steps until the preset number of iterations is reached or the prediction error converges, thereby obtaining the optimal individual chromosome.

[0140] The final model generation submodule is used to retrain the corresponding random forest model based on the hyperparameters corresponding to the optimal individual chromosome, thus obtaining the trained random forest model.

[0141] In some embodiments, the model building and training submodule includes: The data sampling unit is used to randomly sample the simulation dataset to generate multiple sub-datasets.

[0142] The decision tree building unit is used to build decision trees based on subsets of the dataset to generate a random forest model.

[0143] The model training unit is used to train the random forest model based on the hyperparameters corresponding to individual chromosomes and the simulation dataset to obtain prediction results.

[0144] The error calculation unit is used to determine the prediction error based on the prediction results and the simulated values ​​of the rest angles in the simulation dataset.

[0145] Figure 15 This is a schematic diagram of the physical structure of a computer device provided in an embodiment of the present invention, such as... Figure 15 As shown, the computer device includes: a processor 1501, a memory 1502, and a bus 1503.

[0146] The processor 1501 and the memory 1502 communicate with each other via the bus 1503.

[0147] The processor 1501 is used to call program instructions in the memory 1502 to execute the methods provided in the above-described method embodiments.

[0148] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described discrete element method contact parameter calibration method.

[0149] This invention also provides a computer program product, which includes a computer program that, when executed by a processor, implements the above-described discrete element method contact parameter calibration method.

[0150] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0151] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0152] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0153] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0154] In the description of this specification, the references to terms such as "an embodiment," "a specific embodiment," "some embodiments," "for example," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0155] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A discrete element method contact parameter calibration method, characterized in that, include: Obtain the experimental value of the angle of repose of the target material; The target material is simulated based on a pre-established geometric model and the range of contact parameters to obtain a simulation dataset. The random forest model is trained based on the simulation dataset, and the population is iteratively updated according to the obtained prediction error to obtain the trained random forest model. Based on the trained random forest model, the contact parameter combinations in the simulation dataset are predicted to obtain the predicted values ​​of the repose angles corresponding to each contact parameter combination. If the predicted value is close to the experimental value, a preferred combination of contact parameters is generated.

2. The method of claim 1, wherein, The experimental values ​​for obtaining the angle of repose of the target material include: The target material is filled into a pre-set experimental chamber; the ratio of the volume of the target material to the volume of the experimental chamber is greater than or equal to 85%. Remove the side door of the experimental chamber to allow the target material to flow out of the experimental chamber, forming a slope and acquiring an image of the slope; The slope image is processed to obtain the experimental values.

3. The method of claim 2, wherein, The process of image processing of the slope image to obtain the experimental values ​​includes: The slope image is processed by grayscale and binarization to separate the target material region; Based on the preset image size and number of intervals, the target material area is divided into multiple intervals; The target pixel and its pixel coordinates within each interval are determined based on the ordinate of the pixel within each interval. Linear regression fitting is performed on the pixel coordinates to determine the angle between the fitted line and the horizontal axis, thus obtaining the experimental value of the rest angle.

4. The method of claim 1, wherein, The simulation of the target material is performed based on a pre-established geometric model of the target material and the range of contact parameters to obtain a simulation dataset, including: The geometric model is established using discrete element simulation software; Multiple combinations of contact parameters are determined based on the preset value range of the contact parameters; Discrete element simulations are performed based on the geometric model and each combination of contact parameters to generate the simulation dataset. The simulation dataset includes: multiple sets of contact parameter combinations and corresponding repose angle data pairs; the contact parameters include: coefficient of restitution, static friction coefficient and rolling friction coefficient.

5. The method of claim 4, wherein, The geometric model includes a spherical particle model or a multi-sphere model; the establishment of the geometric model in discrete element simulation software includes: If the target material is a spherical particle, select the spherical particle model in the discrete element simulation software; If the target material is an irregularly shaped particle, a multi-sphere model is constructed based on the target material.

6. The method according to claim 5, characterized in that, The construction of the multi-sphere model based on the target material includes: The surface of the target material is scanned using a 3D laser scanner to obtain point cloud data; wherein, the point cloud data is a set of coordinate points on the surface; The point cloud data is reconstructed to obtain a closed triangular network; Feature points are extracted from the closed triangular mesh, and the feature points are clustered to obtain a multi-sphere model.

7. The method of claim 1, wherein, The process of training a random forest model based on the simulation dataset and iteratively updating the population according to the obtained prediction error to obtain the trained random forest model includes: Set the range of values ​​for the hyperparameters of the random forest model, and initialize the population based on the range of values; A corresponding random forest model is constructed for each individual chromosome in the population, and each random forest model is trained based on the simulation dataset to obtain the prediction error; Based on the prediction errors described above, the population is updated through selection, crossover, and mutation operations; The above steps are executed iteratively until the preset number of iterations is reached or the prediction error converges, thus obtaining the optimal individual chromosome. Based on the hyperparameters corresponding to the selected individual chromosomes, the corresponding random forest model is retrained to obtain the trained random forest model.

8. The method of claim 7, wherein, The step involves constructing a corresponding random forest model for each individual chromosome in the population, and training each random forest model based on the simulation dataset to obtain the prediction error, including: The simulation dataset is randomly sampled to generate multiple sub-datasets; Decision trees are constructed based on the aforementioned subsets of data to generate a random forest model; The random forest model is trained based on the hyperparameters corresponding to the individual chromosomes and the simulation dataset to obtain prediction results; The prediction error is determined based on the prediction results and the simulated values ​​of the angle of repose in the simulation dataset.

9. A discrete element method contact parameter calibration device, characterized by, include: The angle of repose acquisition module is used to obtain the experimental value of the angle of repose of the target material; The simulation data construction module is used to simulate the target material based on the pre-established geometric model of the target material and the range of values ​​of the contact parameters, and obtain a simulation dataset. The model training and optimization module is used to train the random forest model based on the simulation dataset and perform iterative updates of the population according to the obtained prediction error to obtain the trained random forest model. The parameter prediction module is used to predict the combination of contact parameters in the simulation dataset based on the trained random forest model, and obtain the predicted value of the angle of repose corresponding to each combination of contact parameters. The parameter filtering module is used to generate a preferred combination of contact parameters if the predicted value is close to the experimental value.

10. A computer device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 8.

11. A computer readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the method of any one of claims 1 to 8.

12. A computer program product, characterised in that, The computer program product includes a computer program that, when executed by a processor, implements the method of any one of claims 1 to 8.

Images