A filling process prediction method and device based on discrete element method and data driving

Abstract

The present invention provides a filling process prediction method and device based on discrete element method and data-driven, which relates to the field of bulk material forming technology. The method combines the physical modeling advantages of discrete element method and the powerful predictive ability of data-driven method. The filling process of granular materials under different process parameters is simulated by discrete element method, and key quality indicators are obtained as input of the data-driven model, while the process parameters are trained as output. This model can not only accurately predict the filling quality, but also adjust the process parameters in real time according to the target quality indicators to achieve intelligent optimization of the filling process. This method greatly improves the computing efficiency, enhances the generalization ability of the model, and reduces the dependence on empirical data. It provides a more efficient and accurate solution for the filling process in the building materials, pharmaceutical, powder metallurgy and other industries, which helps to improve product quality, reduce production costs, and accelerate the research and development and application of new materials and equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of bulk material forming, and in particular to a filling process prediction method and device based on discrete element method and data driving. Background Art

[0002] The filling process for granular materials is a crucial step in numerous industrial production processes. It is widely used in fields such as building materials manufacturing (such as automated filling of blocks and bricks), pharmaceutical engineering (such as tablet compression molding), powder metallurgy (such as metal powder molding), and chemical engineering (such as tower filling). In these fields, the quality of the filling process has a decisive impact on the final product quality. For example, in building materials manufacturing, the accuracy of the filling process directly affects the structural strength and durability of the blocks or bricks; in pharmaceutical engineering, the filling quality of tablets directly affects the stability and consistency of the drug's efficacy; in powder metallurgy, the filling quality of metal powders determines the mechanical properties of the molded parts; and in the chemical industry, the filling quality of tower filling affects the efficiency and safety of chemical reactions.

[0003] However, traditional filling processes face many challenges. On the one hand, the filling process involves complex movements of particles, including flow, collision, and accumulation, which are affected by multiple factors, such as the shape, size, density, and surface characteristics of the particles. On the other hand, there is also the coupling of multiple physical fields in the filling process, such as friction, adhesion, and humidity between particles. These factors interact with each other, making the control of the filling process extremely complex. In addition, the nonlinear effects of process parameters cannot be ignored. Slight changes in parameters such as the moving speed of the feed cart, the rotation speed of the agitator, and the structure of the mold can lead to significant differences in filling quality.

[0004] Existing technologies rely primarily on two approaches for simulating and optimizing filling processes. One is pure discrete element simulation, which solves the equation of motion for each particle. While this method can realistically reflect the micromechanical behavior of particles (contact, friction, and flow), its computational cost is extremely high. For filling systems containing millions of particles in actual industry, a complete simulation can take days or even weeks, which clearly cannot meet the needs of real-time process optimization. The other is empirical models based on statistical regression or simplified physical equations (such as the filling rate-density relationship). While these models are computationally efficient, they have poor generalization capabilities, rely heavily on specific experimental data, and are difficult to adapt to changes in material properties (such as particle size distribution, shape, and wetness) and equipment parameters (such as the speed of the feed truck and the speed of the agitator).

[0005] With the continuous development of intelligent manufacturing technology, the industry's requirements for filling process models are becoming increasingly stringent. An ideal filling process model should possess high precision, high efficiency, and strong generalization capabilities. It should enable virtual trial and error and intelligent optimization of process parameters, thereby reducing the cost of physical experiments; enable real-time prediction of filling states and online quality control to avoid batch defects; and be able to quickly adapt to new materials and equipment, shortening the R&D cycle. However, existing technologies cannot simultaneously meet these requirements, which has significantly limited the digital upgrade and intelligent development of filling processes.

[0006] In view of this, this application is filed. Summary of the Invention

[0007] The present invention provides a filling process prediction method and device based on discrete element method and data driving, which can at least partially improve the above problems.

[0008] To achieve the above object, the present invention adopts the following technical solutions:

[0009] A filling process prediction method based on discrete element method and data-driven, comprising:

[0010] Creating a filling process DEM model, defining material properties and contact mechanics parameters on the filling process DEM model, and performing simulation preprocessing on the filling process DEM model to obtain a simulation result data set, wherein the filling process DEM model includes a particle model and a filling geometry model;

[0011] Extracting multiple filling quality indicators from the simulation result data set, creating an initial data-driven model, and training the initial data-driven model using a preset machine learning algorithm model;

[0012] The preset filling quality target value is input into the trained data-driven model to generate process parameter prediction values, which are compared with the actual values ​​to obtain comparison results. Model training is performed based on the comparison results to obtain the optimal driving model and generate the final prediction results.

[0013] The present invention also provides a filling process prediction device based on discrete element method and data driving, which includes:

[0014] a model building unit, configured to create a filling process DEM model, define material properties and contact mechanics parameters on the filling process DEM model, and perform simulation preprocessing on the filling process DEM model to obtain a simulation result data set, wherein the filling process DEM model includes a particle model and a filling geometry model;

[0015] A training unit, configured to extract a plurality of filling quality indicators from the simulation result data set, create an initial data-driven model, and train the initial data-driven model using a preset machine learning algorithm model;

[0016] The optimization unit is used to input the preset filling quality target value into the trained data-driven model, generate process parameter prediction values, compare the process parameter prediction values ​​with the actual values, obtain comparison results, and perform model training based on the comparison results to obtain the optimal driving model and generate the final prediction results.

[0017] In summary, the described filling process prediction method based on the discrete element method and data-driven approach aims to address many problems existing in traditional filling process simulations, such as insufficient precision, low computational efficiency, difficulty in accurately predicting filling density and uniformity under complex working conditions, and poor generalization due to over-reliance on experience or a single physical model. The core of this method lies in constructing an intelligent optimization system driven by a dual "DEM numerical simulation-data" approach. The discrete element method is used to simulate the filling process of material particles under different process parameters to obtain key quality indicators such as block quality consistency, segregation index, filling volume fraction, and coordination number. These indicators are then used as input parameters for the data-driven model, while the filling process parameters are used as output parameters for model training. Ultimately, a quality-oriented, process-adjustable filling data-driven model is formed, which can achieve real-time online control of the filling process and provide an intelligent decision-making tool for process parameter optimization and quality control. This model not only improves the accuracy and efficiency of the filling process, but also enhances the model's generalization ability and reduces dependence on empirical data. It provides a more efficient and accurate solution for the filling process in industries such as building materials, pharmaceuticals, and powder metallurgy, helping to improve product quality, reduce production costs, and accelerate the research and development and application of new materials and equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 1 is a flow chart of a filling process prediction method based on discrete element method and data-driven according to a first embodiment of the present invention;

[0019] Figure 2 This is a flow chart of a filling process prediction method based on discrete element method and data-driven according to an embodiment of the present invention;

[0020] Figure 3 This is a schematic diagram of the filling process model provided by an embodiment of the present invention. Figure 3 The left side is the filling process DEM model, Figure 3 The right side of is the data-driven model;

[0021] Figure 41 is a schematic diagram of a module of a filling process prediction device based on discrete element method and data driving provided by a second embodiment of the present invention;

[0022] Figure 5 Schematic diagram of predicted values ​​and actual values ​​of a test set provided by an embodiment of the present invention;

[0023] Figure 6 Schematic diagram of actual values ​​and predicted values ​​of the total data set provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0024] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below in conjunction with the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0025] refer to Figures 1 to 3 As shown, the first embodiment of the present invention discloses a filling process prediction method based on discrete element method and data-driven, which can be executed by a filling process prediction device based on discrete element method and data-driven (hereinafter referred to as prediction device), and in particular, executed by one or more processors in the prediction device to implement the following method:

[0026] S1, creating a filling process DEM model, defining material properties and contact mechanics parameters on the filling process DEM model, and performing simulation preprocessing on the filling process DEM model to obtain a simulation result data set, wherein the filling process DEM model includes a particle model and a filling geometry model;

[0027] Specifically, step S1 further includes: constructing a particle model using a multi-sphere method or a superquadric method, constructing a filling geometry model using a three-dimensional modeling software, defining material properties and contact mechanics parameters on the particle model, and defining contact mechanics parameters on the filling geometry model;

[0028] The material properties include density and elastic modulus, and the contact mechanics parameters include friction coefficient, restitution coefficient, and adhesion coefficient. The adhesion coefficient is used to characterize the dryness and wetness of the material.

[0029] Obtain an stl file or mesh file as input for filling a geometric model, and set a filling method, wherein the filling method includes filling process parameters.

[0030] On the basis of the filling process DEM model, different preset process parameters and physical property parameters are input to create multiple filling process DEM models to obtain a multi-objective discrete element filling model;

[0031] A simulation calculation is performed on the multi-objective discrete element filling model to obtain a simulation result data set.

[0032] Preferably, spherical particles and a rolling resistance model can be used instead of the particle model.

[0033] In the present embodiment, first, the creation operation of the filling process DEM model is carried out. The filling process DEM model is used to construct the geometric and physical properties of the granular material and the filling, and simulate the contact mechanics behavior during the filling process. In this link, the construction of the particle model is one of the key steps, and the multi-sphere method or the super quadratic surface method can be used to build the particle model architecture. At the same time, with the help of the power of three-dimensional modeling software, the filling geometry model is carefully constructed. On the particle model, it is necessary to accurately define the material properties, covering key elements such as density and elastic modulus, as well as contact mechanics parameters such as friction coefficient, restitution coefficient, and adhesion coefficient, wherein the adhesion coefficient plays an indispensable role in characterizing the dry and wet properties of the material. On the filling geometry model, the contact mechanics parameters should also be defined. Obtain an stl file or mesh file and use it as the input of the filling geometry model, and then set the filling method, which is the filling process parameter.

[0034] Among them, linear adhesion, JKR and other adhesion models are used to define the adhesion force between particles. Set process parameters, including the speed of the material cart and the speed of the agitator, and set particle physical parameters such as the adhesion coefficient and surface energy. Use a particle shape and size analyzer to measure the particle size distribution of the material; use the weighing method to measure the bulk density of the granular material; use uniaxial compression to test the basic information of the granular material such as Young's modulus and Poisson's ratio. In addition, two fluidity indicators are used, but not limited to, to calibrate the contact parameters of the granular material (friction coefficient, adhesion parameter and resistance coefficient) and verify the accuracy of the parameters. The fluidity indicators specifically include: stacking angle, flow energy, internal friction angle, etc. The material properties are in the particle model, and the contact mechanics parameters include those between particles and between particles and geometric bodies.

[0035] Based on the filling process DEM model, different preset process parameters and physical property parameters are input to create multiple filling process DEM models, forming a multi-objective discrete element filling model. These multi-objective discrete element filling models are simulated and calculated to obtain a simulation result data set. In addition, it is worth mentioning that when constructing the particle model, spherical particles and rolling resistance models can be used to replace the original particle model. This alternative solution can shorten the calculation cycle to a certain extent, improve the operating efficiency of the model, and provide more convenient conditions for subsequent simulation calculations and data analysis. It helps to speed up the construction and optimization process of the entire filling process model, making it more in line with the dual needs of efficiency and accuracy in actual industrial production.

[0036] S2, extracting multiple filling quality indicators from the simulation result data set, creating an initial data-driven model, and training the initial data-driven model using a preset machine learning algorithm model;

[0037] Specifically, step S2 further includes: extracting the quality consistency, filling volume fraction, segregation index, and density between each mold from the simulation result data set, and organizing them into a preset data format to create an initial data-driven model, the formula of which is: , , , , , To ensure the quality consistency between each mold, is the average property of a particle group in the mixture system, is the standard deviation of a particle group in the mixing system, is the number of particle attributes in the current sampling unit, such as the number of particle size groups, is the number of sampling units set, is the concentration of a certain type of particles in the current sampling unit, is the filling volume fraction, is the total volume of the filling particles, is the volume of the mold cavity, is the segregation index, is the mass fraction of fine particles in the pre-set system, For density, is the total number of particles, is the total number of contacts between particles;

[0038] The initial data-driven model is trained using a machine learning algorithm model, wherein the machine learning algorithm model includes a fully connected layer neural network, model number, polynomial regression, KNN, decision tree, and random forest.

[0039] Specifically, we deeply explored the simulation result data set and accurately extracted the four key filling quality indicators: quality consistency between each mold, filling volume fraction, segregation index, and density. These extracted indicator data were standardized and organized according to the preset data format, as shown in Table 1, thereby creating an initial data-driven model. The data-driven model establishes the mapping relationship between process parameters, material properties, and filling quality indicators based on machine learning or deep learning algorithms. Among them, in the calculation formula of quality consistency RSD, is the concentration (mass fraction) of a certain type of particles in the current sampling unit; the filling volume fraction VF represents the relative filling volume; the segregation index SI represents the degree of particle size segregation. In this embodiment, the formula of the segregation index SI is The particle properties represented are fine particles. Generally, the closer the segregation index is to 1, the more uniform the particle size distribution is. The density CN, also known as the coordination number, represents the packing density. For example, in the output parameters, the speed of the feed car is v and the agitator speed in the material truck w are set to five levels, among which v =(0.25, 0.5, 0.75, 1.00, 1.25), unit: m / s; w=(2, 3, 4, 5, 6), unit: r / s, i.e. "revolutions per second"; physical parameters, such as moisture content M , set 3 levels, M=(6,8, 10), unit: %. Total: 75 sets of DEM simulations are required to obtain 75 sets of data, that is, Table 1 has a total of 76 columns.

[0040] Table 1

[0041]

[0042] After successfully creating the initial data-driven model, it is trained using a pre-defined machine learning algorithm. A wide range of machine learning models are available, including fully connected neural networks, model trees, polynomial regression, KNN (nearest neighbor algorithm), decision trees, and random forests. By training the initial data-driven model with these advanced machine learning algorithms, the model can fully learn the complex nonlinear relationships between filling quality indicators and process parameters, providing powerful intelligent decision support for subsequent filling process parameter optimization and quality control. This process not only significantly improves the model's prediction accuracy but also enhances its generalization capabilities, enabling it to better adapt to variations in material properties, equipment parameters, and process conditions, laying a solid foundation for intelligent, efficient, and precise filling process control. The hyperparameters, dataset partitioning criteria, and activation function selection of the relevant algorithm models are determined by the dataset size and test set error and are not specified here. For example, a fully connected neural network was selected for model training on the 75 data sets mentioned above. The model hyperparameters are set as follows: the number of hidden layers is 64, the learning rate is 0.01, the activation function from the input layer to the hidden layer and between the hidden layers is the ReLU function, and the activation function from the hidden layer to the output layer is the Sigmoid function. According to the 8:2 principle, the dataset is divided into a training set and a test set, where the training set is used for model training and the test set is used for model prediction and accuracy adjustment. The training set data volume is 60 and the test set data volume is 15. The data-driven model obtained from the training set is as follows Figure 3 , as shown in the figure on the right.

[0043] S3, input the preset filling quality target value into the trained data-driven model, generate process parameter prediction values, compare the process parameter prediction values ​​with the actual values, obtain comparison results, and perform model training based on the comparison results to obtain the optimal driving model and generate the final prediction results.

[0044] Specifically, step S3 further includes: inputting the preset filling quality target value into the trained data-driven model to generate process parameter prediction values, and arranging them into a preset data format;

[0045] Taking the simulation result data set as the actual value, comparing the actual value with the process parameter prediction value to obtain an error value, and determining whether the error value is greater than a preset threshold;

[0046] If yes, dynamically update the simulation result dataset according to the process parameter prediction value, and retrain the data-driven model based on the updated simulation result dataset, and repeat the above steps;

[0047] If not, the currently trained data-driven model is used as the optimal driving model, and the currently predicted values ​​of the process parameters are used as the final prediction results.

[0048] Preferably, the preset threshold is 5%, and the process parameter prediction values ​​include the material cart movement speed, the agitator speed and the particle physical property parameters.

[0049] For example: select the above 15 test sets and enter RSD , SI , VF , CN , get the predicted value, compare the predicted value with the actual value, the comparison result is as follows Figure 4 As shown in Figure 2, the predicted values ​​are close to the actual values. The actual and predicted values ​​of all sample data sets are compared. Figure 5 As shown in the figure, to further improve the accuracy of the model, the amount of training data samples can be appropriately increased.

[0050] In this embodiment, the core of this step is to input the preset fill quality target value into the trained data-driven model to generate predicted values ​​for the process parameters. These predicted values ​​are then compared with the actual values. The comparison results are then used to further optimize the model, ultimately obtaining the optimal driving model and the most accurate prediction results.

[0051] First, the preset filling quality target value is input into the trained data-driven model. The model outputs the corresponding process parameter prediction value based on its internal learning mechanism and existing data relationships. These process parameter prediction values ​​cover several key aspects such as the material cart movement speed, agitator speed, and particle physical parameters. They play a decisive role in the quality control of the filling process. Next, the result data set previously obtained through DEM simulation is used as the actual value and carefully compared with the process parameter prediction value generated by the data-driven model. The accuracy of the model prediction is evaluated by calculating the error value between the two. At this time, a preset threshold value, such as 5%, is set as the standard for judging whether the error is within an acceptable range. If the error value is greater than this preset threshold value, it means that the prediction accuracy of the current model has not reached the ideal state and needs further optimization.

[0052] To improve the model's prediction accuracy, the simulation result dataset is dynamically updated based on the predicted process parameter values. This update process adjusts the original dataset based on the predicted values ​​to better reflect actual production conditions. The data-driven model is then retrained based on the updated simulation result dataset. Through this iterative training approach, the model continuously learns and adjusts, gradually improving its predictive capabilities. After training is complete, an error comparison is performed again, and the above steps are repeated until the error value drops below a preset threshold.

[0053] When the error value is less than or equal to the preset threshold, it means that the model's prediction accuracy has reached a high level. At this point, the currently trained data-driven model is determined to be the optimal driving model, and the current process parameter prediction value is used as the final prediction result. This final prediction result can provide valuable guidance for the actual filling process, helping to accurately control process parameters and ensure that the filling quality meets the preset target.

[0054] Through this series of steps, this method not only accurately predicts filling process parameters but also dynamically adjusts and optimizes the model based on actual needs, maintaining good predictive performance under varying production conditions. This data-driven dynamic optimization mechanism significantly enhances the intelligence of the filling process, reduces reliance on experience, improves production efficiency, and ensures consistent product quality.

[0055] In summary, the filling process prediction method based on discrete element method and data-driven method aims to solve the problems existing in traditional filling process simulation, such as limited accuracy, low computational efficiency, difficulty in accurately predicting filling density and uniformity under complex working conditions, and poor generalization ability due to over-reliance on experience or a single physical model, by integrating the physical modeling capability of discrete element method (DEM) for particle / block motion and interaction with the powerful learning and prediction capability of data-driven method for the nonlinear relationship between complex process parameters and results. It also realizes real-time online control of filling process guided by filling quality, and provides intelligent decision-making tools for process parameter optimization and quality control.

[0056] Its core lies in building an intelligent optimization system driven by both DEM numerical simulation and data. Using the discrete element method, the flow behavior of material particles during the filling process under different process parameters is simulated, statistically obtaining key quality indicators such as block quality consistency, segregation index, filling volume fraction, and coordination number. These indicators are then used as input parameters for the data-driven model, while the filling process parameters are used as output parameters for model training. Ultimately, a quality-oriented, process-adjustable filling data-driven model is formed, enabling real-time online control of the filling process and providing an intelligent decision-making tool for process parameter optimization and quality control.

[0057] In the specific implementation process, a discrete element model of the filling process is first created, including particle modeling and boundary and operating condition settings. Particle modeling uses the polysphere method or superquadric surfaces to construct a geometric model for non-spherical particles. Alternatively, a spherical particle and rolling resistance model can be used to replace the non-spherical particle modeling to shorten the calculation cycle. Material properties (density, elastic modulus) and contact mechanics parameters (friction coefficient, restitution coefficient, and adhesion coefficient) are defined. The adhesion parameter can be used to characterize the material's wetness. For boundary and operating condition settings, the filling geometry is constructed using 3D modeling software and exported as an "stl" or "mesh" file, which serves as the geometric input for the DEM. The filling method (such as the trolley speed and agitator speed) is also set. Subsequently, a dataset is collected from the DEM simulation results, including the mass consistency (RSD) between molds, the fill volume fraction (VF), the segregation index (SI), and the density (coordination number, CN). This data serves as the source for the data-driven model, which is then trained using machine learning algorithms (such as fully connected neural networks, polynomial regression, and kNN) to construct the initial data-driven model.

[0058] Furthermore, the preset filling quality target value is input into the trained data-driven model to generate process parameter predictions. The simulation result dataset is used as the actual value, and the actual value is compared with the process parameter prediction value to obtain an error value. A determination is then made as to whether the error value is greater than a preset threshold (e.g., 5%). If the error value is greater than the preset threshold, the simulation result dataset is dynamically updated based on the process parameter prediction value, and the data-driven model is retrained based on the updated simulation result dataset, repeating the above steps. If the error value is less than or equal to the preset threshold, the currently trained data-driven model is used as the optimal driving model, and the current process parameter prediction value is used as the final prediction result.

[0059] Compared with the existing technology, the beneficial effects of this method are significant. First, by combining DEM with data-driven methods, high-precision simulation and optimization of the filling process are achieved, significantly improving the prediction accuracy of the filling quality. Second, the model can be dynamically adjusted and optimized to adapt to different production conditions and material properties, enhancing the generalization ability of the model. In addition, through the closed-loop optimization mechanism, the dependence on experience is reduced, the cost of physical experiments is reduced, and production efficiency is improved. Finally, the model can predict the filling state in real time, realize online quality control, avoid batch defects, shorten the R&D cycle, and provide a more efficient and accurate solution for the filling process in industries such as building materials, pharmaceuticals, and powder metallurgy. It helps to improve product quality, reduce production costs, and accelerate the research and development and application of new materials and equipment.

[0060] See also Figure 6 The second embodiment of the present invention provides a filling process prediction device based on discrete element method and data driving, which includes:

[0061] The model building unit 101 is used to create a filling process DEM model, define material properties and contact mechanics parameters on the filling process DEM model, and perform simulation preprocessing on the filling process DEM model to obtain a simulation result data set, wherein the filling process DEM model includes a particle model and a filling geometry model;

[0062] A training unit 102 is configured to extract a plurality of filling quality indicators from the simulation result data set, create an initial data-driven model, and train the initial data-driven model using a preset machine learning algorithm model;

[0063] The optimization unit 103 is used to input the preset filling quality target value into the trained data-driven model, generate process parameter prediction values, compare the process parameter prediction values ​​with the actual values, obtain comparison results, and perform model training based on the comparison results to obtain the optimal driving model and generate the final prediction results.

[0064] The above is a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principles of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A filling process prediction method based on discrete element method and data driven, characterized in that: include: Create a filling process DEM model, define material properties and contact mechanics parameters on the filling process DEM model, and perform simulation preprocessing on the filling process DEM model to obtain a simulation result data set. The filling process DEM model includes a particle model and a filling geometry model, specifically: Use the multi-sphere method or super-quadratic surface method to build the particle model, use the 3D modeling software to build the filling geometry model, and define the material properties and contact mechanics parameters on the particle model, and define the contact mechanics parameters on the filling geometry model; The material properties include density and elastic modulus, and the contact mechanics parameters include friction coefficient, restitution coefficient, and adhesion coefficient. The adhesion coefficient is used to characterize the dryness and wetness of the material. Obtain an STL file or a mesh file as input for filling a geometric model, and set a filling method, wherein the filling method includes filling process parameters; Extract multiple filling quality indicators from the simulation result data set, create an initial data-driven model, and use a preset machine learning algorithm model to train the initial data-driven model, specifically: The quality consistency, filling volume fraction, segregation index, and density between each mold are extracted from the simulation result data set and organized into a preset data format to create an initial data-driven model, whose formula is: , , , , , To ensure the quality consistency between each mold, is the average property of a particle group in the mixture system, is the standard deviation of a particle group in the mixing system, is the number of particle attributes in the current sampling unit, is the number of sampling units set, is the concentration of a certain type of particles in the current sampling unit, is the filling volume fraction, is the total volume of the filling particles, is the volume of the mold cavity, is the segregation index, is the mass fraction of fine particles in the pre-set system, For density, is the total number of particles, is the total number of contacts between particles; The initial data-driven model is trained using a machine learning algorithm model, wherein the machine learning algorithm model includes a fully connected layer neural network, a model tree, polynomial regression, KNN, a decision tree, and a random forest; The preset filling quality target value is input into the trained data-driven model to generate process parameter prediction values, which are compared with the actual values ​​to obtain comparison results. Model training is performed based on the comparison results to obtain the optimal driving model and generate the final prediction results.

2. The filling process prediction method based on discrete element method and data driving according to claim 1 is characterized in that: Spherical particles and a rolling resistance model can be used instead of the particle model.

3. The filling process prediction method based on discrete element method and data driving according to claim 1 is characterized in that: The filling process DEM model is simulated and preprocessed to obtain a simulation result data set, specifically: On the basis of the filling process DEM model, different preset process parameters and physical property parameters are input to create multiple filling process DEM models to obtain a multi-objective discrete element filling model; A simulation calculation is performed on the multi-objective discrete element filling model to obtain a simulation result data set.

4. The filling process prediction method based on discrete element method and data driving according to claim 1 is characterized in that: The preset filling quality target value is input into the trained data-driven model to generate process parameter prediction values. The process parameter prediction values ​​are compared with the actual values ​​to obtain comparison results. The model is trained based on the comparison results to obtain the optimal driving model and generate the final prediction results, specifically: Input the preset filling quality target value into the trained data-driven model to generate process parameter prediction values ​​and organize them into a preset data format; Taking the simulation result data set as the actual value, comparing the actual value with the process parameter prediction value to obtain an error value, and determining whether the error value is greater than a preset threshold; If yes, dynamically update the simulation result dataset according to the process parameter prediction value, and retrain the data-driven model based on the updated simulation result dataset, and repeat the above steps; If not, the currently trained data-driven model is used as the optimal driving model, and the currently predicted values ​​of the process parameters are used as the final prediction results.

5. The filling process prediction method based on discrete element method and data driving according to claim 4 is characterized in that: The preset threshold is 5%, and the process parameter prediction values ​​include the material cart movement speed, the agitator speed and the particle physical property parameters.

6. A filling process prediction device based on discrete element method and data driven, characterized in that: The device applies the filling process prediction method based on discrete element method and data driving according to any one of claims 1 to 5, comprising: a model building unit, configured to create a filling process DEM model, define material properties and contact mechanics parameters on the filling process DEM model, and perform simulation preprocessing on the filling process DEM model to obtain a simulation result data set, wherein the filling process DEM model includes a particle model and a filling geometry model; A training unit, configured to extract a plurality of filling quality indicators from the simulation result data set, create an initial data-driven model, and train the initial data-driven model using a preset machine learning algorithm model; The optimization unit is used to input the preset filling quality target value into the trained data-driven model, generate process parameter prediction values, compare the process parameter prediction values ​​with the actual values, obtain comparison results, and perform model training based on the comparison results to obtain the optimal driving model and generate the final prediction results.

Images