A method and device for predicting mixture quality based on discrete element method and data driving

Abstract

The present invention provides a method and device for predicting mixing quality based on discrete element method and data-driven, which relates to the technical field of granular material mixing. The method uses discrete element method to construct a physical simulation model of the mixing process, which can accurately simulate the mixing behavior of granular materials in different mixing equipment, covering a variety of process parameters. In addition, data-driven technology is introduced, and with the help of machine learning algorithms, a large amount of high-fidelity data generated by DEM simulation is used to train the prediction model, thereby constructing a hybrid model that integrates physical mechanism and data intelligence. This model can not only quickly and accurately predict the mixing effect indicators based on the input mixing process parameters, but also reversely predict the corresponding optimal process parameters, effectively improving production efficiency. It aims to solve the problems existing in the existing mixing process, such as optimization dependence on experience, low efficiency, high cost, and difficulty in accurately predicting the mixing effect.

Description

Technical Field

[0001] The present invention relates to the technical field of granular material mixing, and in particular to a method and device for predicting mixing quality based on discrete element method and data driving. Background Art

[0002] In numerous industrial sectors, such as construction, pharmaceuticals, chemicals, and food, the mixing process for granular materials is a critical step in the production process. The quality of these processes directly impacts product quality, performance, and production efficiency. However, existing mixing processes face numerous challenges and limitations in practical application.

[0003] Traditional mixing process optimization relies primarily on engineers' experience and repeated trial and error adjustments. While this experience-driven approach can meet production needs to a certain extent, it has significant shortcomings. First, it lacks systematicity and scientificity. The setting of process parameters (filling level, rotation speed, material moisture content, etc.) relies on the engineers' experience, resulting in high trial and error costs. It also makes it difficult to accurately describe and predict the behavior of particulate materials and the mixing effect during the mixing process. Second, when faced with changes in material properties (such as particle size, shape, moisture content, etc.), a large amount of trial and error is required to readjust the process parameters, which is not only time-consuming and labor-intensive, but also costly. Furthermore, empirical methods make it difficult to achieve automated and intelligent control of mixing processes, and are unable to meet the requirements of modern industrial production for efficient, precise, and stable production.

[0004] With the development of computer technology, the discrete element method (DEM) has been introduced into the study of mixing processes. DEM can simulate the movement of particulate materials, providing a new perspective for understanding the mixing process. However, DEM also has some limitations in practical applications. On the one hand, the computational cost of DEM is extremely high. For complex mixing processes and large-scale particle systems, simulation calculations may take hours or even days. This inefficient computing speed makes it difficult for DEM to guide the optimization of mixing processes in real time in actual production. On the other hand, DEM simulation results are very sensitive to initial conditions and parameter settings. Any slight change may lead to significant differences in simulation results, which increases the difficulty and uncertainty of using DEM for mixing process optimization.

[0005] In recent years, data-driven methods have been widely used in various fields. In the mixing process, data-driven models (such as neural networks) attempt to learn the relationship between mixing process parameters and mixing quality through a large amount of experimental data. However, this approach also has some problems. First, obtaining sufficient quantity and quality of experimental data is a huge challenge. In actual production, due to factors such as production costs, time constraints, and operational complexity, it is difficult to collect experimental data that comprehensively covers various process parameters and material properties. Secondly, data-driven models lack an in-depth understanding of the physical mechanisms of the mixing process, and their prediction results are often difficult to interpret. In addition, when faced with new materials or process conditions, the model's generalization ability is weak, and prediction deviations are prone to occur.

[0006] In general, existing mixing process optimization methods, whether relying on experience, traditional DEM simulation or data-driven models, are unable to meet the needs of modern industrial production for efficient, precise and intelligent mixing processes.

[0007] In view of this, this application is filed. Summary of the Invention

[0008] The present invention provides a mixture quality prediction method and device based on discrete element method and data driving, which can at least partially improve the above problems.

[0009] To achieve the above object, the present invention adopts the following technical solutions:

[0010] A mixture quality prediction method based on discrete element method and data-driven, comprising:

[0011] Creating a particle dynamics simulation model of the mixing process, obtaining an input parameter set, and simulating the particle dynamics simulation model according to the input parameter set to obtain a discrete element simulation result data set, wherein the particle dynamics simulation model includes a non-spherical particle geometry model and a mixing process geometry model;

[0012] Extracting multiple mixture quality indicators from the discrete element simulation result data set, creating an initial data-driven model, and training the initial data-driven model using a preset machine learning algorithm model;

[0013] The preset mixing quality target value is input into the trained data-driven model to generate the predicted value of the mixing process parameter. The predicted value of the process parameter is compared with the actual value to obtain the comparison result. The model training is performed or the optimal driving model is obtained based on the comparison result to generate the final prediction result. According to the prediction result, the operating parameters of the mixing equipment are automatically adjusted.

[0014] The present invention also provides a mixture quality prediction device based on discrete element method and data driving, which includes:

[0015] a simulation unit, configured to create a particle dynamics simulation model of the mixing process, obtain an input parameter set, and simulate the particle dynamics simulation model according to the input parameter set to obtain a discrete element simulation result data set, wherein the particle dynamics simulation model includes a non-spherical particle geometry model and a mixing process geometry model;

[0016] A training unit, configured to extract a plurality of mixture quality indicators from the discrete element simulation result data set, create an initial data-driven model, and train the initial data-driven model using a preset machine learning algorithm model;

[0017] The feedback unit is used to input the preset mixing quality target value into the trained data-driven model, generate the predicted value of the mixing process parameter, compare the predicted value of the process parameter with the actual value, obtain the comparison result, and perform model training or obtain the optimal driving model based on the comparison result to generate the final prediction result, and automatically adjust the operating parameters of the mixing equipment based on the prediction result.

[0018] In summary, the described method for predicting mixing quality based on the discrete element method and data-driven mixing quality prediction aims to address the problems existing in existing mixing processes, such as reliance on experience for optimization, low efficiency, high cost, and difficulty in accurately predicting mixing effects. This method first uses the discrete element method to construct a physical simulation model of the mixing process, which can accurately simulate the mixing behavior of particulate materials in different mixing equipment, covering a variety of process parameters such as rotation speed, filling rate, mixing time, and material properties. On this basis, data-driven technology is introduced. With the help of machine learning algorithms, a large amount of high-fidelity data generated by DEM simulation, such as particle velocity, position, contact force, mixing uniformity, etc., is used to train the prediction model, thereby constructing a hybrid model that integrates physical mechanisms and data intelligence. This model can not only quickly and accurately predict mixing effect indicators such as mixing uniformity, segregation degree, mixing time, etc. based on the input mixing process parameters, but also reversely predict the corresponding optimal process parameters, effectively improving production efficiency. This method significantly improves the efficiency and accuracy of mixing process parameter optimization by combining the physical mechanism advantages of DEM with the data-driven efficient prediction capabilities, providing a powerful theoretical model and technical support for the intelligent design of mixing equipment, precise control of process parameters, and stable assurance of product quality. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 1 is a flow chart of a mixture quality prediction method based on discrete element method and data-driven method provided by the first embodiment of the present invention;

[0020] Figure 2 This is a technical principle diagram of a mixture quality prediction method based on discrete element method and data-driven according to an embodiment of the present invention;

[0021] Figure 3 This is a schematic diagram of a mixing process model provided by an embodiment of the present invention. Figure 3 The left side is the particle dynamics simulation model, and the right side is the data-driven network model;

[0022] Figure 4 Schematic diagram of a mixing quality index statistical unit provided by an embodiment of the present invention;

[0023] Figure 5 3 is a module diagram of a mixture quality prediction device based on discrete element method and data driving provided in the second 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 mixture quality prediction method based on discrete element method and data-driven, which can be executed by a mixture quality 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 particle dynamics simulation model of a mixing process, obtaining an input parameter set, and simulating the particle dynamics simulation model according to the input parameter set to obtain a discrete element simulation result data set, wherein the particle dynamics simulation model includes a non-spherical particle geometry model and a mixing process geometry model;

[0027] Specifically, step S1 further includes: constructing a non-spherical particle geometric model using a multi-sphere method or a super-quadratic surface method, constructing a mixing process geometric model using a three-dimensional modeling software, and obtaining an stl file as an input of the mixing process geometric model;

[0028] Based on the particle dynamics simulation model, different preset process parameters and physical property parameters are input to create multiple particle dynamics simulation models, and a multi-objective discrete element mixture model is obtained to simulate the particle movement under the same filling level, rotation speed, and particle physical property conditions;

[0029] An input parameter set is obtained, and a DEM simulation process is performed on the multi-objective discrete element mixture model according to the input parameter set. A particle motion dataset and a mixture quality index are output to obtain a discrete element simulation result set, wherein the particle motion dataset includes particle position, velocity, and contact force, and the mixture quality index includes uniformity index, segregation index, and mixing time.

[0030] Preferably, a spherical particle and rolling resistance model can be used instead of a non-spherical particle geometric model.

[0031] Preferably, the input parameter set includes filling level, stirrer speed, and particle physical property parameters, and the particle physical property parameters include moisture content, particle size distribution, and density.

[0032] In this embodiment, a particle dynamics simulation model of the mixing process is first created to simulate particle motion under different filling levels (e.g., 40% to 60%), rotation speeds (e.g., 30 to 90 rpm), and particle properties (e.g., moisture content 5% to 10%). Specifically, the multi-sphere method or the superquadratic surface method is used to construct a geometric model of non-spherical particles. These two methods can more accurately reflect the true shape of the particles, thereby improving the accuracy of the simulation. At the same time, a three-dimensional modeling software is used to construct a geometric model of the mixing process, and it is exported as a ".stl" format file as input for the geometric model of the mixing process. The advantage of this is that it can ensure that the model is highly consistent with the geometric structure of the actual mixing equipment, providing an accurate physical environment for subsequent simulations.

[0033] On the basis of constructing the particle dynamics simulation model, different preset process parameters and physical property parameters are further input to create multiple particle dynamics simulation models and form a multi-objective discrete element mixing model. These parameters include filling level, agitator speed and physical property parameters of the particles. The physical property parameters include at least one of moisture content, particle size distribution and density, and the number is not limited. In this way, the particle movement under different filling levels, rotation speeds and particle physical properties can be simulated, providing rich and diverse data samples for subsequent data-driven models. For example, in a specific embodiment, the filling level can be set to 40%, 50% and 60%, the agitator speed can be set to 30 rpm, 60 rpm and 90 rpm respectively, and the particle moisture content can be considered at 5%, 7.5% and 10%. In this way, simulation models under different conditions can be generated, thereby more comprehensively covering the various possibilities of the mixing process.

[0034] In this embodiment, a particle size analyzer was used to measure the particle size distribution of the material; a weighing method was used to measure the bulk density of the granular material; and uniaxial compression was used to determine basic information about the granular material, such as Young's modulus and Poisson's ratio. Two flowability indices, but not limited to them, were used to calibrate the granular material's contact parameters (friction coefficient, adhesion parameter, and drag coefficient) and verify the accuracy of these parameters. Specifically, these flowability indices included using flow energy similar to that used in mixing conditions to calibrate discrete element contact parameters and using the internal friction angle to verify the calibrated contact parameter values.

[0035] Next, the input parameter set is obtained and a DEM simulation is performed on the multi-objective discrete element mixture model based on this parameter set. This process outputs a particle motion dataset and mixture quality indicators, forming a discrete element simulation result set. The particle motion dataset includes information such as particle position, velocity, and contact force, which can provide a detailed description of the dynamic behavior of particles during the mixing process. Mixture quality indicators include key parameters such as uniformity index, segregation index, and mixing time, which directly reflect the quality of the mixing. For example, the uniformity index can be calculated by calculating the ratio of the standard deviation to the mean of the mass distribution of particles of a certain size group in various regions within the mixing tank. The segregation index can be measured by comparing the mass fraction of fine particles in the system with a pre-set value. The mixing time is the minimum duration required to achieve a uniform mixture. These detailed simulation results provide high-quality data support for subsequent data-driven model training, thereby improving the model's predictive accuracy.

[0036] During implementation, some preferred solutions can also be adopted according to actual conditions. For example, when there are high requirements for computational efficiency, spherical particles and a rolling resistance model can be used to replace the non-spherical particle geometric model. Although the spherical particle model has certain differences in shape from the actual particles, the introduction of the rolling resistance model can compensate for this difference to a certain extent, significantly shorten the calculation cycle, and improve the efficiency of the simulation. This flexible model selection method enables the present invention to better adapt to different application scenarios and needs.

[0037] Through this step, this method effectively combines the physical mechanism advantages of the discrete element method with the efficient data-driven prediction capabilities, providing a new approach for optimizing mixing processes. This method not only improves the efficiency and accuracy of optimizing mixing process parameters, but also provides strong technical support for the intelligent design of mixing equipment, the precise control of process parameters, and the stable assurance of product quality.

[0038] See also Figure 4 , S2, extracting multiple mixture quality indicators from the discrete element simulation result data set, creating an initial data-driven model, and training the initial data-driven model using a preset machine learning algorithm model;

[0039] Specifically, step S2 further includes: extracting the particle mass distribution uniformity, segregation index, and mixing time T of a certain particle size group in each area of ​​the mixing tank from the discrete element simulation result data set, and arranging them into a preset data format to create an initial data-driven model, the formula of which is: , , , The uniformity of particle mass distribution of a certain particle size group in each area inside the mixing tank, is the standard deviation of a particle group in the mixing system, is the average property of a particle group in the mixture 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 segregation index, is the mass fraction of fine particles in the preset system;

[0040] The initial data-driven model is trained using a machine learning algorithm model to establish a mutual mapping relationship between process parameters and quality indicators, wherein the machine learning algorithm model includes XGBoost, MLP, KNN or Polynomial.

[0041] In this embodiment, the key mixing quality indicators extracted from the discrete element simulation result data set include the various regions inside the mixing tank ( Figure 4 The uniformity of the mass distribution of particles in a certain particle size group (sampling unit shown), the segregation index, and the mixing time T. These indicators can fully reflect the effect of the mixing process and provide key data support for subsequent data-driven models.

[0042] Specifically, the uniformity of the mass distribution of particles of a certain particle size group in each area inside the mixing tank is obtained by calculation. represents the average property of a particle group in the mixture system, such as mass fraction; Indicates the concentration of a certain type of particles in the current sampling unit, such as mass fraction. The segregation index SI indicates the degree of particle size segregation. In this embodiment, the mass fraction of fine particles in the pre-set system represents particle properties of a particular size group, such as the mass fraction of fine particles. Generally, the closer the segregation index is to 1, the more uniform the particle size distribution. Mixing time T represents the minimum mixing time required to achieve a uniform mixture. The extraction and calculation of these metrics laid the foundation for creating the initial data-driven model.

[0043] Based on the extracted mixing quality indicators, they are organized into a preset data format, and then an initial data-driven model is created. The purpose of establishing this model is to explore the intrinsic relationship between mixing process parameters and mixing quality indicators through a data-driven approach, providing a model foundation for predicting mixing quality. In this process, selecting an appropriate machine learning algorithm model to train the initial data-driven model is crucial. The machine learning algorithm models that can be used in the present invention include XGBoost, MLP (Multi-Layer Perceptron), KNN (K Nearest Neighbor), or Polynomial (Polynomial Regression). The specific algorithm can be selected based on the dataset type, model error, and generalization ability. The hyperparameters, dataset partitioning criteria, and activation function selection of the relevant algorithm models all depend on the dataset size and test set error, and are not limited here. These algorithms each have their own characteristics and can learn and fit data from different perspectives, thereby establishing a mutual mapping relationship between process parameters and quality indicators.

[0044] Take XGBoost, for example. It's a gradient-boosting ensemble learning algorithm that combines multiple weak learners to build powerful predictive models. It boasts high efficiency and accuracy, making it particularly well-suited for processing large datasets. During training, XGBoost automatically handles missing values ​​in the data and is robust to outliers. Through training, the XGBoost model learns the complex nonlinear relationships between mixing process parameters and mixing quality indicators, providing strong support for accurate prediction of mixing quality.

[0045] Another example is the MLP (Multi-Layer Perceptron), a feedforward neural network composed of multiple layers of neurons that can learn the mapping relationship between input and output data. MLPs have strong fitting capabilities and can handle complex nonlinear problems. In the present invention, by training the MLP model, a model can be obtained that can accurately predict mixture quality indicators, thereby providing guidance for optimizing the mixing process.

[0046] The KNN (K Nearest Neighbor) algorithm is an instance-based learning method. Its core concept is to find the K samples in the training data that are most similar to the sample to be predicted and then classify or regress the unknown sample based on the labels of these neighbors. The KNN algorithm is simple to implement and has good prediction results for small datasets. In the case of mixture quality prediction, the KNN algorithm can quickly find the K samples that are most similar to the sample to be predicted based on existing mixture process parameters and quality indicator data, thereby predicting the mixture quality.

[0047] Polynomial regression is an extension of linear regression, incorporating polynomial features to fit nonlinear relationships in data. While the polynomial regression model remains linear in form, by adding polynomial combinations of input features, it can better capture complex patterns in the data. In the present invention, the polynomial regression model establishes a polynomial regression equation based on the relationship between mixing process parameters and mixing quality indicators, thereby predicting mixing quality.

[0048] By using the above-mentioned machine learning algorithm model to train the initial data-driven model, not only can the mutual mapping relationship between process parameters and quality indicators be established, but also the rapid and accurate prediction of mixing quality can be achieved. This process effectively combines the physical mechanism advantages of the discrete element method with the efficient prediction capabilities of data-driven, significantly improving the efficiency and accuracy of mixing process parameter optimization. For example, in practical applications, through the trained data-driven model, mixing quality indicators such as mixing uniformity, degree of segregation, mixing time, etc. can be quickly predicted given the mixing process parameters. This enables engineers to evaluate the mixing effect under different process parameter combinations in a short period of time, thereby quickly finding the optimal process parameter combination, improving production efficiency, and reducing production costs. At the same time, this data-driven prediction method can also provide support for the intelligent design of mixing equipment, realize the automated control and optimization of the mixing process, and further improve the stability and consistency of product quality.

[0049] S3, input the preset mixing quality target value into the trained data-driven model, generate the predicted value of the mixing process parameter, compare the predicted value of the process parameter with the actual value, obtain the comparison result, and perform model training or obtain the optimal driving model based on the comparison result to generate the final prediction result, and automatically adjust the operating parameters of the mixing equipment based on the prediction result.

[0050] Specifically, step S3 further includes: inputting the preset mixing quality target value into the trained data-driven model to generate the mixing process parameter prediction value, and arranging it into a preset data format;

[0051] Taking the discrete element simulation result data set as the actual value, comparing the actual value with the predicted value of the mixing process parameter to obtain an error value, and determining whether the error value is greater than a preset threshold;

[0052] If yes, dynamically update the discrete element simulation result dataset according to the process parameter prediction value, and retrain the data-driven model based on the updated discrete element simulation result dataset, and repeat the above steps;

[0053] If not, the currently trained data-driven model is used as the optimal driving model, and the currently predicted values ​​of the mixing process parameters are used as the final prediction results.

[0054] Preferably, the preset threshold is 5%, and the mixing process parameter prediction values ​​include a filling level prediction value V, an agitator speed prediction value w, and a particle property parameter prediction value M. In this embodiment, the preset mixing quality target value is input into the trained data-driven model. These target values ​​are set according to actual production needs, such as the desired mixing uniformity and segregation degree. The model generates corresponding mixing process parameter prediction values ​​based on these target values, such as the filling level prediction value V, which represents the percentage of the total volume of the particles in the mixing tank volume, the agitator speed prediction value w, the particle property parameter prediction value M (such as different adhesion parameters characterizing materials with different moisture contents), etc., and organizes these prediction values ​​into a preset data format for subsequent comparison and analysis, as shown in Table 1.

[0055] Table 1

[0056]

[0057] Next, the discrete element simulation result data set is used as the actual value and compared with the predicted value of the mixing process parameter generated by the data-driven model. By calculating the error value between the two, the accuracy of the model prediction can be intuitively evaluated. Specifically, for each mixing process parameter, the difference between its predicted value and the actual value is calculated, and it is determined whether the error value is greater than a preset threshold. In this embodiment, the preferred preset threshold is 5%. The setting of this threshold is based on the accuracy requirements for mixing quality control in actual production and the efficiency considerations of model optimization. If the error value is greater than 5%, it means that the current data-driven model has a large deviation in predicting the parameter and needs further optimization and adjustment.

[0058] In this case, the discrete element simulation result dataset is dynamically updated based on the predicted values ​​of the process parameters. Specifically, the predicted values ​​are used as new input parameters, and the discrete element simulation is re-performed to generate a new simulation result dataset. Then, based on this updated discrete element simulation result dataset, the data-driven model is retrained. This process actually uses new and more accurate data to adjust and optimize the model parameters to improve the model's prediction accuracy. After training is completed, the preset mixture quality target value is again input into the updated data-driven model to generate new mixture process parameter prediction values, and the above-mentioned comparison and judgment steps are repeated. Through this iterative optimization method, the gap between the predicted value and the actual value is gradually narrowed until the error value is less than or equal to the preset threshold.

[0059] When the error is less than or equal to 5%, the trained data-driven model has achieved high prediction accuracy and can accurately reflect the relationship between mixing process parameters and mixing quality. At this point, the trained data-driven model is considered the optimal driving model, and the current predicted mixing process parameters are used as the final prediction results. Based on these prediction results, the optimal process parameter combination is reversely recommended (for example, a material moisture content of 8%, a fill level of 50%, and a rotation speed of 60 rpm provide optimal mixing quality). These final prediction results can be directly applied to the actual mixing production process. Based on the predicted optimal process parameters, the operating parameters of the mixing equipment can be automatically adjusted, such as adjusting the agitator speed or changing the fill level, thereby achieving automated control and optimization of the mixing process.

[0060] It should be noted that this embodiment only uses the filling level and rotation speed for explanation, but the mixing process in real life is by no means limited to these two. All related processes such as vibration frequency and amplitude are included. The discrete element method and data-driven mixing quality prediction method can handle them, and no limitation is made here.

[0061] This process not only improves the efficiency and accuracy of mixing process parameter optimization, but also reduces reliance on manual experience and lowers trial-and-error costs. Automatically adjusting mixing equipment operating parameters based on real-time predictions ensures the mixing process remains optimal, improving the consistency and stability of product quality. Furthermore, this data-driven model-based optimization approach is highly adaptable, enabling rapid response to changes in material properties and adjustments to production requirements, providing strong technical support for the intelligent upgrade of mixing processes.

[0062] In summary, the DEM-based and data-driven mixing quality prediction method, by constructing a physical simulation model of the mixing process and integrating it with a data-driven model, achieves high-precision, real-time prediction of everything from process parameters (filling level, rotational speed, material moisture content) to quality indicators (uniformity index, segregation index, mixing time). This overcomes the limitations of traditional methods, which struggle to balance efficiency and accuracy. This approach aims to address existing mixing process issues such as reliance on experience, low efficiency, high costs, and difficulty in accurately predicting mixing results.

[0063] Specifically, the proposed method for predicting mixing quality based on the discrete element method (DEM) and data-driven methods first utilizes the discrete element method to construct a particle dynamics simulation model of the mixing process. This model accurately simulates the mixing behavior of granular materials within various mixing equipment (such as double-cone mixers, V-type mixers, and spiral mixers), encompassing a wide range of process parameters such as rotational speed, fill rate, mixing time, and material properties. Furthermore, data-driven technology is introduced. Using machine learning algorithms, the prediction model is trained using a large amount of high-fidelity data (such as particle velocity, position, contact force, and mixing uniformity) generated by DEM simulations. This results in a hybrid model that integrates physical mechanisms with data intelligence. This model not only quickly and accurately predicts mixing performance indicators (such as mixing uniformity, segregation, and mixing time) based on input mixing process parameters (such as rotational speed, fill rate, mixing time, and material properties), but also reversely predicts the corresponding optimal process parameters, effectively improving production efficiency.

[0064] Simply put, this method significantly improves the efficiency and accuracy of mixing process parameter optimization by combining the physical mechanism advantages of DEM with the efficient data-driven prediction capabilities, providing an effective theoretical model and technical support for the intelligent design of mixing equipment, precise control of process parameters, and stable assurance of product quality.

[0065] Compared with existing technologies, this method offers significant benefits. First, improved accuracy: The physical mechanism ensures the model's reliable prediction (error <5%) for complex parameters that are difficult to obtain experimentally, such as segregation index and mixing index. Second, a leap in efficiency: The data-driven model's prediction efficiency is reduced from hours to seconds, supporting online parameter optimization. Finally, improved generalization: Dynamic adaptation of physical properties (moisture content) to different materials reduces the number of experimental calibrations by over 90%.

[0066] See also Figure 5 The second embodiment of the present invention provides a mixture quality prediction device based on discrete element method and data driving, which includes:

[0067] A simulation unit 101 is configured to create a particle dynamics simulation model of a mixing process, obtain an input parameter set, and simulate the particle dynamics simulation model according to the input parameter set to obtain a discrete element simulation result data set, wherein the particle dynamics simulation model includes a non-spherical particle geometry model and a mixing process geometry model;

[0068] A training unit 102 is configured to extract a plurality of mixture quality indicators from the discrete element simulation result data set, create an initial data-driven model, and train the initial data-driven model using a preset machine learning algorithm model;

[0069] The feedback unit 103 is used to input the preset mixing quality target value into the trained data-driven model, generate the predicted value of the mixing process parameter, compare the predicted value of the process parameter with the actual value, obtain the comparison result, and perform model training or obtain the optimal driving model based on the comparison result to generate the final prediction result, and automatically adjust the operating parameters of the mixing equipment based on the prediction result.

[0070] 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 mixture quality prediction method based on discrete element method and data driven, characterized in that: include: Create a particle dynamics simulation model of the mixing process, obtain an input parameter set, simulate the particle dynamics simulation model according to the input parameter set, and obtain a discrete element simulation result data set, wherein the particle dynamics simulation model includes a non-spherical particle geometry model and a mixing process geometry model, specifically: The geometric model of non-spherical particles is constructed using the multi-sphere method or the super-quadratic surface method, and the geometric model of the mixing process is constructed using 3D modeling software, and the stl file is obtained as the input of the geometric model of the mixing process; Based on the particle dynamics simulation model, different preset process parameters and physical property parameters are input to create multiple particle dynamics simulation models, and a multi-objective discrete element mixture model is obtained to simulate the particle movement under the same filling level, rotation speed, and particle physical property conditions; Obtaining an input parameter set, performing DEM simulation processing on a multi-objective discrete element mixture model according to the input parameter set, outputting a particle motion data set and a mixture quality index, and obtaining a discrete element simulation result set, wherein the particle motion data set includes particle position, velocity, and contact force, and the mixture quality index includes a uniformity index, a segregation index, and a mixing time; Extract multiple mixture quality indicators from the discrete element 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 particle mass distribution uniformity, segregation index, and mixing time T of a certain particle size group in each area of ​​the mixing tank are extracted from the discrete element simulation result data set, and the data are organized into a preset data format to create an initial data-driven model, whose formula is: , , , The uniformity of particle mass distribution of a certain particle size group in each area inside the mixing tank, is the standard deviation of a particle group in the mixing system, is the average property of a particle group in the mixture 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 segregation index, is the mass fraction of fine particles in the preset system; Using a machine learning algorithm model to train the initial data-driven model to establish a mutual mapping relationship between process parameters and quality indicators, wherein the machine learning algorithm model includes XGBoost, MLP, KNN or Polynomial; The preset mixing quality target value is input into the trained data-driven model to generate the predicted value of the mixing process parameter. The predicted value of the process parameter is compared with the actual value to obtain the comparison result. The model training is performed or the optimal driving model is obtained based on the comparison result to generate the final prediction result. According to the prediction result, the operating parameters of the mixing equipment are automatically adjusted.

2. The mixture quality prediction method based on discrete element method and data-driven according to claim 1 is characterized in that: Spherical particles and a rolling resistance model can be used instead of the non-spherical particle geometry model.

3. The mixture quality prediction method based on discrete element method and data-driven according to claim 1 is characterized in that: The input parameter set includes filling level, stirrer speed, and particle physical property parameters, and the particle physical property parameters include moisture content, particle size distribution, and density.

4. The method for predicting mixture quality based on discrete element method and data-driven according to claim 1, characterized in that: Input the preset mixing quality target value into the trained data-driven model to generate the predicted value of the mixing process parameter. Compare the predicted value of the process parameter with the actual value to obtain the comparison result. Based on the comparison result, perform model training or obtain the optimal driving model to generate the final prediction result, specifically: Input the preset mixture quality target value into the trained data-driven model to generate the mixture process parameter prediction value and organize it into the preset data format; Taking the discrete element simulation result data set as the actual value, comparing the actual value with the predicted value of the mixing process parameter to obtain an error value, and determining whether the error value is greater than a preset threshold; If yes, dynamically update the discrete element simulation result dataset according to the process parameter prediction value, and retrain the data-driven model based on the updated discrete element 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 mixing process parameters are used as the final prediction results.

5. The method for predicting mixture quality based on discrete element method and data-driven according to claim 4, characterized in that: The preset threshold is 5%, and the mixing process parameter prediction values ​​include a filling level prediction value V, a stirrer speed prediction value w, and a particle parameter prediction value M.

6. A mixture quality prediction device based on discrete element method and data driven, characterized in that: The device applies the mixture quality prediction method based on discrete element method and data driving according to any one of claims 1 to 5. The device is integrated into an industrial control platform and includes: a simulation unit, configured to create a particle dynamics simulation model of the mixing process, obtain an input parameter set, and simulate the particle dynamics simulation model according to the input parameter set to obtain a discrete element simulation result data set, wherein the particle dynamics simulation model includes a non-spherical particle geometry model and a mixing process geometry model; A training unit, configured to extract a plurality of mixture quality indicators from the discrete element 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 feedback unit is used to input the preset mixing quality target value into the trained data-driven model, generate the predicted value of the mixing process parameter, compare the predicted value of the process parameter with the actual value, obtain the comparison result, and perform model training or obtain the optimal driving model based on the comparison result to generate the final prediction result, and automatically adjust the operating parameters of the mixing equipment based on the prediction result.

Images