A dynamic rest angle prediction method based on mapping of particle shape features and equivalent contact parameters

Abstract

This invention relates to the field of particulate material flowability analysis technology, and discloses a dynamic angle of repose prediction method based on the mapping of particle shape features and equivalent contact parameters. The method includes: obtaining measured values ​​and images of the dynamic angle of repose of various particles through experiments to construct a first dataset; extracting two-dimensional and three-dimensional shape feature parameters of the particles; constructing a discrete element simulation model, using a rolling resistance model to equivalently characterize the influence of non-spherical particle shape; establishing a mapping relationship between shape features and equivalent contact parameters through parameter calibration; generating a multi-condition simulation dataset based on this mapping relationship; and training a machine learning model to achieve prediction. This invention solves the problems of traditional methods, such as difficulty in quantifying shape influence, high DEM computation cost, and poor generalization ability of data-driven models, achieving efficient and accurate dynamic angle of repose prediction.

Description

Technical Field

[0001] This invention relates to the field of particulate material flowability analysis technology, and more specifically, to a dynamic angle of repose prediction method based on the mapping of particle shape characteristics and equivalent contact parameters. Background Technology

[0002] The dynamic angle of repose is an important macroscopic parameter characterizing the flow properties and packing behavior of particulate materials, and it is widely used in mining, chemical, metallurgical, agricultural, pharmaceutical, and particle engineering fields. In rotary drums, conveying devices, mixing equipment, and particle processing units, the shape, particle size, density, moisture content, and operating conditions of different particles significantly affect their dynamic angle of repose. Therefore, establishing a dynamic angle of repose prediction method applicable to various particulate materials and operating conditions is of significant engineering importance.

[0003] In existing technologies, the dynamic angle of repose is typically obtained through experimental measurement or discrete element method (DEM) simulation. For example, Chinese patent application CN119000426A discloses an experimental characterization method for the flowability and mixing degree of slender, flexible biomass particles. This method uses a rotating drum experimental platform combined with image processing to measure the dynamic angle of repose. However, this method relies solely on experiments, requiring repeated experiments for multiple particles and various working conditions, resulting in a large workload, low efficiency, and high cost. Another example is Chinese patent application CN120651706A, which discloses a discrete element method contact parameter calibration device and method based on the dynamic angle of repose. This method calibrates contact parameters using the dynamic angle of repose as the target value, but it calibrates for a single material and does not establish a correlation between shape features and parameters. Recalibration is required when the particle shape changes. In addition, Chinese patent application CN121562330A discloses a method for predicting the dynamic repose angle of adhesive particles based on a discrete element-machine learning coupled model. The method generates data through discrete element simulation and trains a machine learning model for prediction. However, its input features do not include the three-dimensional shape features of the particles, which limits the generalization ability of the model.

[0004] On the other hand, most existing machine learning prediction methods take process parameters, physical property parameters, or some contact parameters as inputs, while the introduction of particle shape factors usually remains at the level of a few two-dimensional geometric indicators, such as the sphericity disclosed in CN119607991A. This is insufficient to fully express the influence of complex particle morphology on flowability, resulting in insufficient model generalization ability. Chinese patent applications with publication numbers CN120510978A and CN120510980A disclose a method combining discrete element simulation and machine learning, which is used to predict material moisture content and mixing quality, respectively. However, their prediction objects are not dynamic angles of repose, and they do not involve the quantification and mapping of particle shape features.

[0005] Therefore, how to characterize particle shape in a multidimensional and interpretable way, and further establish its mapping relationship with equivalent contact parameters, especially by introducing three-dimensional shape features into the prediction model, to achieve efficient and accurate prediction of dynamic angle of repose while taking into account physical consistency and computational efficiency, is a technical problem that urgently needs to be solved in this field. This invention aims to solve the above problems by establishing a mapping relationship between the multidimensional shape feature vector of particles and the rolling friction coefficient, combined with multi-condition discrete element simulation and machine learning. Summary of the Invention

[0006] The purpose of this invention is to overcome the problems in the prior art, such as the difficulty in quantifying stable particle shape, the complexity of DEM modeling for non-spherical particles, and the insufficient generalization ability of pure data-driven prediction. This invention provides a dynamic angle of repose prediction method and system based on particle shape characteristics. This method, based on the mapping of particle shape characteristics to equivalent contact parameters, quantifies the particle shape and introduces it as a key parameter into the prediction model. Combined with discrete element simulation and machine learning, it achieves efficient prediction of the dynamic angle of repose for particles of different shapes under different physical and process parameters.

[0007] To achieve the above objectives, the present invention adopts the following technical solution:

[0008] A dynamic repose angle prediction method based on the mapping of particle shape features and equivalent contact parameters includes:

[0009] S1. Through experiments, obtain the measured values ​​of dynamic angle of repose of various granular materials with different shape characteristics under the preset dynamic angle of repose measurement conditions, and collect the corresponding particle images to construct a first dataset containing the measured values ​​of dynamic angle of repose and particle images.

[0010] S2. Perform image processing and analysis on the particle images in the first dataset to extract the two-dimensional shape feature parameters of the particles; for particles with corresponding three-dimensional mesh models, further extract the three-dimensional shape feature parameters through three-dimensional geometric analysis, and fuse the two-dimensional and three-dimensional features to form the shape feature vector of the particles.

[0011] S3. Based on the discrete element method, construct a discrete element simulation model corresponding to the preset dynamic angle of repose measurement condition, and use an equivalent simplified particle model to characterize the shape influence of non-spherical particles.

[0012] S4. For at least some of the particulate materials in the first dataset, use discrete element simulation to calibrate the parameters. With the goal of minimizing the error between the simulated dynamic angle of repose and the experimental dynamic angle of repose, determine the equivalent contact parameters of the corresponding particulate materials and establish the mapping relationship between the particle shape characteristic parameters and the equivalent contact parameters.

[0013] S5. Based on the mapping relationship established in step S4, for the particle shape characteristic parameters to be predicted, determine the corresponding equivalent contact parameters, and combine the physical property parameters and target process parameters to carry out multi-condition discrete element simulation, generate dynamic repose angle sample data, and construct the second dataset.

[0014] S6. Construct a machine learning prediction model, using the shape feature parameters, physical property parameters and process parameters in the second dataset as input features, and the dynamic angle of repose as the output label, to train the model;

[0015] S7. Deploy the trained machine learning prediction model in the prediction system to output the dynamic repose angle prediction value in real time based on the input target particle shape characteristics and working conditions.

[0016] Preferably, the two-dimensional shape feature parameters include at least two of aspect ratio, sphericity, convexity, and angularity.

[0017] Preferably, the three-dimensional shape feature parameters are obtained by: reading the STL format three-dimensional mesh model of the particles and extracting vertex information and face information; performing principal component analysis on the vertex set and rotating the three-dimensional mesh model to the principal axis coordinate system; calculating the dimensional parameters of the model in the three principal axis directions in the principal axis coordinate system to obtain the length L, width W and height H, and calculating at least one of the aspect ratio, flatness ratio and elongation ratio accordingly.

[0018] Preferably, the three-dimensional shape feature parameters further include at least one of the following: sphericity calculated based on the surface area S and volume V of the three-dimensional mesh model; convexity calculated based on the ratio of particle volume to convex hull volume; and edge angle calculated based on the angle distribution of the normal vectors of adjacent facets or the change in surface curvature.

[0019] Preferably, for particles that simultaneously possess two-dimensional shape feature parameters and three-dimensional shape feature parameters, a feature splicing method is used to form the particle shape feature vector.

[0020] Preferably, the equivalent contact parameter includes at least the rolling friction coefficient, and further includes at least one of the static friction coefficient and the recovery coefficient.

[0021] Preferably, the calibration of the equivalent contact parameters includes: setting an initial range of the equivalent contact parameters under simulation conditions consistent with the experimental conditions; conducting multiple discrete element simulations to obtain the simulation dynamic angle of repose; comparing the simulation dynamic angle of repose with the experimental dynamic angle of repose, and determining the optimal equivalent contact parameters for the corresponding particulate material with the goal of minimizing the error.

[0022] Preferably, the mapping relationship between the particle shape feature vector and the equivalent contact parameter is a mapping relationship in which the rolling friction coefficient is determined by the particle shape feature vector; when establishing the mapping relationship, the static friction coefficient and the recovery coefficient can be set as empirical values ​​or parameters within a preset range, and the rolling friction coefficient is used as the main calibration parameter.

[0023] Preferably, the physical property parameters include at least one of particle density and particle size, and the process parameters include at least one of cylinder rotation speed and filling rate.

[0024] Preferably, the multi-condition discrete element simulation includes: interpolating and expanding the particle shape feature vector within the range of calibrated particle shape features, and combining and sampling the physical property parameters and process parameters to generate multiple sets of simulation input parameters.

[0025] Preferably, the combined sampling employs one of Latin hypercube sampling, orthogonal experimental design, or grid sampling.

[0026] Preferably, the machine learning model is a random forest model, a gradient boosting tree model, or a neural network model.

[0027] Compared with the prior art, the present invention has the following beneficial effects:

[0028] This invention introduces particle shape factors into the discrete element modeling process by establishing a mapping relationship between particle shape feature parameters and equivalent contact parameters, thereby reducing the need to repeatedly construct complex geometric models for different non-spherical particles and helping to reduce modeling complexity.

[0029] The present invention determines the equivalent contact parameters based on the mapping relationship, and performs multi-condition discrete element simulation in combination with physical property parameters and process parameters, which can obtain dynamic repose angle sample data for predictive model training.

[0030] The present invention constructs a dynamic angle of repose prediction model based on the dynamic angle of repose sample data, which is beneficial to improving the efficiency of dynamic angle of repose prediction. Attached Figure Description

[0031] Figure 1 This is a schematic diagram of the overall process of the dynamic repose angle prediction method based on the mapping of particle shape features and equivalent contact parameters provided in the embodiments of the present invention.

[0032] Figure 2 This is a schematic diagram of the overall process for extracting particle shape feature parameters in an embodiment of the present invention, including two-dimensional image processing, three-dimensional STL model analysis, and feature fusion;

[0033] Figure 3 This is a simplified overall process diagram of particle DEM simulation based on the rolling friction model in an embodiment of the present invention;

[0034] Figure 4 This is a schematic diagram of the dynamic repose angle experimental device in an embodiment of the present invention;

[0035] Figure 5 This is a schematic diagram illustrating the overall process of establishing the mapping relationship between particle shape feature parameters and equivalent contact parameters in an embodiment of the present invention.

[0036] Figure 6 This is a schematic diagram of the two-dimensional particle image feature extraction sub-process in an embodiment of the present invention;

[0037] Figure 7 This is a schematic diagram of the particle shape feature extraction sub-process based on a three-dimensional mesh model in an embodiment of the present invention;

[0038] Figure 8 This is a schematic diagram of the equivalent contact parameter calibration and error minimization sub-process in an embodiment of the present invention;

[0039] Figure 9 This is a schematic diagram of the sub-process for generating multi-condition discrete element simulation samples in an embodiment of the present invention;

[0040] Figure 10 This is a schematic diagram of the sub-processes for training, validating, and deploying machine learning prediction models in an embodiment of the present invention. Detailed Implementation

[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are all within the scope of protection of the present invention.

[0042] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but this should not be construed as limiting the scope of protection of the present invention.

[0043] like Figure 1 As shown, this embodiment provides a dynamic angle of repose prediction method based on the mapping between particle shape features and equivalent contact parameters. The method generally includes the following steps: experimental data acquisition, particle shape feature extraction, discrete element simulation model construction, equivalent contact parameter calibration, establishment of the mapping relationship between particle shape features and equivalent contact parameters, generation of multi-condition simulation samples, training of machine learning prediction model, and dynamic angle of repose prediction output. Figure 1 It is mainly used to demonstrate the overall technical route of the method of the present invention, so as to make the relationship between each step clearer.

[0044] In this embodiment, experimental data acquisition is performed first. Combined with... Figure 4The illustrated dynamic angle of repose experimental device structure can conduct dynamic angle of repose measurement experiments on various granular materials with different shape characteristics. The experimental device may include a rotating cylinder, a drive mechanism, a support structure, an image acquisition unit, and a control unit. Preferably, the inner diameter of the rotating cylinder is 150 mm, the length is 300 mm, the cylinder rotation speed is set to 12 rpm, and the filling rate is set to 40%. The experiment is repeated 5 times for each type of granular material, and the average value is taken as the corresponding measured dynamic angle of repose value. Simultaneously, top-view and side-view images of the particles are acquired using an industrial camera, and STL format 3D mesh models are further obtained for some particles. This constructs a first dataset containing measured dynamic angle of repose values, 2D particle images, and 3D model data. Figure 4 It is mainly used to explain the apparatus structure and data source of the dynamic angle of repose experiment.

[0045] After obtaining the first dataset, the particle shape feature extraction stage begins. Figure 2 The overall process of extracting particle shape feature parameters is shown, which includes three parts: two-dimensional image feature extraction, three-dimensional STL model feature extraction, and two-dimensional and three-dimensional feature fusion.

[0046] Among them, such as Figure 6 As shown, the two-dimensional particle image feature extraction sub-process may include steps such as image grayscale conversion, threshold segmentation, denoising, edge detection, contour extraction, and two-dimensional feature calculation. Based on the particle projection contour, features such as aspect ratio, two-dimensional sphericity, two-dimensional convexity, and two-dimensional edge angle can be extracted. Preferably, multiple particle samples can be randomly selected for each type of particle material, and each two-dimensional shape feature can be calculated separately, with the average value taken as the two-dimensional shape feature parameter of the particle material. Figure 6 It is mainly used to elaborate on the process of two-dimensional image processing and two-dimensional geometric feature extraction.

[0047] For particles with a three-dimensional mesh model, such as Figure 7 As shown, the STL file can be read, vertex and triangular facet information extracted, and principal component analysis performed on the vertex set to rotate the particle model to the principal axis coordinate system. Subsequently, the dimensional parameters of the model in the three principal axis directions are calculated to obtain length, width, and height, and further dimensional features such as flatness and elongation are calculated. Simultaneously, the 3D sphericity can be calculated based on the surface area and volume of the 3D mesh model, the 3D convexity can be calculated based on the ratio of particle volume to convex hull volume, and the 3D edge angles can be calculated based on the angle distribution of the normal vectors of adjacent facets or changes in surface curvature. Figure 7 It is mainly used to elaborate on the process of three-dimensional geometric analysis and three-dimensional shape feature extraction.

[0048] exist Figure 2 In the overall process shown, after completing Figure 6 and Figure 7 After extracting the corresponding two-dimensional and three-dimensional features, the obtained two-dimensional and three-dimensional features are further fused to form the shape feature vector of the particle. Preferably, the shape feature vector may include nine dimensions: aspect ratio, two-dimensional sphericity, two-dimensional convexity, two-dimensional edge angle, flattening ratio, elongation ratio, three-dimensional sphericity, three-dimensional convexity, and three-dimensional edge angle. For particles with only two-dimensional images, default value filling, missing markers, or other compatible processing methods can be used for their three-dimensional feature parts.

[0049] After constructing the particle shape feature vector, a discrete element simulation model corresponding to the experimental conditions is established. For example... Figure 3 As shown, this embodiment constructs a simplified overall process for particle DEM simulation based on a rolling friction model. Specifically, according to... Figure 4 The structural parameters of the experimental setup are established using a rotating cylindrical geometric model in discrete element simulation software, with the cylinder dimensions consistent with the experimental setup. The particle geometry model preferably uses an equivalent particle model, and the particle radius can be taken as the average equivalent radius of actual particles. The contact model preferably uses the Hertz-Mindlin model, and the static friction coefficient, rolling friction coefficient, and coefficient of restitution between particles and between particles and geometry are set as parameters to be calibrated. Figure 3 It is mainly used to illustrate the process of establishing, running and simulating dynamic repose angle output of a simplified particle DEM simulation model.

[0050] After completing the DEM simulation model construction, the equivalent contact parameters are further calibrated and mapping relationships are established. For example... Figure 5 As shown, the overall process for establishing the mapping relationship between particle shape feature parameters and equivalent contact parameters includes: obtaining shape feature vectors for different particle samples, setting simulation conditions consistent with the experiment, searching and calibrating the contact parameters, obtaining the optimal equivalent contact parameters corresponding to each particle sample, and then establishing a mapping model based on the correspondence between the shape features and equivalent contact parameters of multiple samples. Figure 5 It is mainly used to explain the process of establishing an overall mapping relationship based on the calibration results of multiple samples.

[0051] Furthermore, such as Figure 8 As shown, in the parameter calibration process for a single particle sample or a single type of particle, the search range of the rolling friction coefficient can be set first according to the particle shape characteristic parameters, and the static friction coefficient and the coefficient of restitution can be set as empirical values ​​or preset values. Then, multiple sets of candidate contact parameters are generated using central composite design, Latin hypercube sampling or other sampling methods, and DEM simulation is carried out to obtain the corresponding simulated dynamic angle of repose. The simulated dynamic angle of repose is then compared with the experimental dynamic angle of repose, and the optimal equivalent contact parameters corresponding to the particle sample are determined with the minimum error as the objective. Figure 8Primarily used to detail the parameter calibration and error minimization process. This is performed repeatedly based on multiple particle samples. Figure 8 After the process shown, the establishment can be obtained. Figure 5 The sample pairs required for the mapping relationship shown.

[0052] After obtaining the mapping relationship between particle shape characteristic parameters and equivalent contact parameters, further multi-condition discrete element simulations are conducted to generate sample data for training the prediction model. For example... Figure 9 As shown, the multi-condition discrete element simulation sample generation sub-process includes: first, interpolating and expanding the shape feature vector of the calibration sample to generate multiple virtual shape feature samples; then, combining and sampling physical property parameters such as particle density, particle size, cylinder rotation speed and filling rate with process parameters; next, determining the equivalent contact parameters corresponding to each group of shape feature samples according to the mapping model, and inputting the shape feature parameters, equivalent contact parameters, physical property parameters and process parameters into the DEM simulation model to obtain the corresponding dynamic repose angle simulation results; finally, summarizing to form the second dataset. Figure 9 It is mainly used to explain the process of generating dynamic rest angle sample data under multiple working conditions for machine learning training.

[0053] After obtaining the second dataset, a machine learning prediction model is built. For example... Figure 10 As shown, the training, validation, and deployment process of the machine learning prediction model includes: dividing the second dataset into a training set, a validation set, and a test set; training the machine learning model using shape feature parameters, physical property parameters, and process parameters as input features and the dynamic angle of repose as the output label; adjusting the model structure and parameters based on the validation set results; evaluating the model's generalization ability on the test set; and saving and deploying the trained model after its performance meets preset requirements. The machine learning model can be a random forest model, a gradient boosting tree model, or a neural network model. In a preferred embodiment, a TCN-GRU hybrid neural network model is used for dynamic angle of repose prediction. Figure 10 It is mainly used to illustrate the entire process of predictive models from training and validation to deployment and application.

[0054] In this embodiment, Figure 10 The trained dynamic angle of repose prediction model is deployed in the prediction system. For the particle to be predicted, its two-dimensional image is first acquired, and its three-dimensional mesh model is obtained when conditions permit; then, according to... Figure 2 The overall process shown, combined with Figure 6 and Figure 7 The sub-process shown extracts particle shape feature parameters and constructs the corresponding particle shape feature vector; then, it combines particle density, particle size, and target process parameters, and, if necessary, uses... Figure 5The mapping model shown determines the corresponding equivalent contact parameters, and the pre-trained prediction model outputs the predicted dynamic angle of repose under the target working condition. This completes the entire technical loop from particle characterization, parameter mapping, data generation to dynamic angle of repose prediction.

[0055] In a preferred embodiment, the method of the present invention can be used not only for predicting the dynamic angle of repose between different particle shapes, but also for predicting the dynamic angle of repose under different physical property parameters and different process parameters. By introducing two-dimensional and three-dimensional particle shape features and combining them with the equivalent contact parameter mapping relationship, the dependence on modeling and calibrating complex non-spherical particles one by one can be reduced, thereby improving prediction efficiency and enhancing the model's adaptability to different particle morphologies and operating conditions.

[0056] In conclusion, Figure 1 Used to demonstrate the overall process of the present invention. Figure 2 This is used to demonstrate the overall workflow for extracting particle shape features. Figure 3 This is used to demonstrate the simplified overall process of particle DEM simulation. Figure 4 Used to demonstrate the structure of the dynamic angle of repose experimental device. Figure 5 This is used to demonstrate the overall process of establishing the mapping relationship between shape feature parameters and equivalent contact parameters. Figure 6 This is used to demonstrate the sub-process of two-dimensional image feature extraction. Figure 7 This is used to demonstrate the sub-process of 3D STL feature extraction. Figure 8 This is used to demonstrate the sub-process of equivalent contact parameter calibration and error minimization. Figure 9 This is used to demonstrate the sub-process for generating simulation samples under multiple operating conditions. Figure 10 This diagram illustrates the sub-processes of training, validating, and deploying machine learning models. The accompanying figures complement each other to form a complete description of the technical solution of this invention.

[0057] The above description is merely a preferred embodiment of the present invention and is 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 dynamic repose angle prediction method based on the mapping of particle shape features and equivalent contact parameters, characterized in that, Includes the following steps: S1. Through experiments, obtain the measured values ​​of dynamic angle of repose of various granular materials with different shape characteristics under the preset dynamic angle of repose measurement conditions, and collect the corresponding particle images to construct a first dataset containing the measured values ​​of dynamic angle of repose and particle images. S2. Perform image processing and analysis on the particle images in the first dataset to extract the two-dimensional shape feature parameters of the particles; for particles with corresponding three-dimensional mesh models, further extract three-dimensional shape feature parameters through three-dimensional geometric analysis. The three-dimensional shape feature parameters include at least three of the following calculated based on the three-dimensional mesh model: aspect ratio, flatness, elongation, sphericity, convexity, and edge angle. The two-dimensional and three-dimensional features are fused to form a multi-dimensional shape feature vector of the particles. S3. Construct a discrete element simulation model corresponding to the preset dynamic rest angle measurement condition, and use an equivalent simplified particle model of spherical particles combined with rolling friction model to characterize the shape influence of non-spherical particles. S4. For at least some particulate materials, use discrete element simulation to calibrate parameters, aiming to minimize the error between the simulated dynamic angle of repose and the experimental dynamic angle of repose, determine the equivalent contact parameters of the corresponding particulate materials, the equivalent contact parameters include at least the rolling friction coefficient, and establish a mapping relationship between the multidimensional shape feature vector of the particles and the rolling friction coefficient. S5. Based on the mapping relationship, determine the corresponding rolling friction coefficient for the particle shape feature vector to be predicted, and combine the particle physical property parameters and target process parameters to interpolate and extend the particle shape feature vector within the range of calibrated particle shape features. Combine the physical property parameters and process parameters for sampling, carry out multi-condition discrete element simulation, generate dynamic repose angle sample data covering multiple working conditions, and construct the second dataset. S6. Construct a machine learning prediction model, using the multidimensional shape feature parameters, physical property parameters and process parameters in the second dataset as input features, and the dynamic angle of repose as the output label, to train the model; S7. Deploy the trained machine learning prediction model in the prediction system to output the dynamic repose angle prediction value in real time based on the input target particle multidimensional shape feature vector and working conditions.

2. The dynamic angle of repose prediction method based on the mapping of particle shape features and equivalent contact parameters according to claim 1, characterized in that, In step S4, the calibration of the equivalent contact parameters includes: setting an initial range for the equivalent contact parameters under simulation conditions consistent with the experimental conditions; conducting multiple discrete element simulations to obtain the simulated dynamic angle of repose; comparing the simulated dynamic angle of repose with the experimental dynamic angle of repose, and determining the optimal equivalent contact parameters for the corresponding particulate material with the goal of minimizing the error; and, when establishing the mapping relationship, setting the static friction coefficient and the coefficient of restitution to fixed values, and using the rolling friction coefficient as the only variable calibration parameter to determine its mapping relationship with the particle shape feature vector.

3. The dynamic angle of repose prediction method based on the mapping of particle shape features and equivalent contact parameters according to claim 1, characterized in that, In step S5, the physical property parameters include at least one of particle density and particle size, and the process parameters include at least one of cylinder rotation speed and filling rate; the combined sampling adopts one of Latin hypercube sampling, orthogonal experimental design or grid sampling.

4. The dynamic angle of repose prediction method based on the mapping of particle shape features and equivalent contact parameters according to claim 1, characterized in that, In step S2, the three-dimensional shape feature parameters include at least two of the following: three-dimensional sphericity calculated based on the surface area S and volume V of the three-dimensional mesh model, three-dimensional convexity calculated based on the ratio of particle volume to convex hull volume, and three-dimensional edge angles calculated based on the angle distribution of the normal vectors of adjacent facets or the change in surface curvature.

5. The dynamic angle of repose prediction method based on the mapping of particle shape features and equivalent contact parameters according to claim 1, characterized in that, In step S6, the machine learning model is a random forest model, a gradient boosting tree model, or a neural network model.

Images