A mine semi-autogenous grinding process simulation method and system based on discrete elements

Abstract

The application discloses a mine semi-autogenous grinding process simulation method and system based on discrete elements, relates to the field of simulation and emulation, and comprises the following steps: according to a crushing characteristic label and a parameter mapping matrix, fusing Internet of Things sensor data, constructing a granularity distribution prediction model and an energy consumption optimization scheduling strategy, obtaining an energy-saving operation plan, comparing simulation data sets with field production data, adjusting multi-phase coupling particle model parameters, generating a calibration parameter set, based on the calibration parameter set, using a control variable method to study influence laws of semi-autogenous mill operation parameters, obtaining a performance index database and an analysis report, fusing multi-sensor data and laser radar data, constructing a grinding process sensing mechanism and optimization control logic, and generating a visual interface; and the application comprehensively improves simulation accuracy and process adaptability through multi-phase modeling and data-driven optimization, and provides an efficient and economic solution for mine semi-autogenous grinding processes.

Description

Technical Field

[0001] This invention relates to the field of simulation, and in particular to a simulation method and system for semi-autogenous grinding processes in mines based on discrete element method. Background Technology

[0002] Semi-autogenous grinding (SAG) technology, a core technology in mineral processing, is widely used in the crushing and grinding of metallic and non-metallic ores. Its efficiency directly affects the energy consumption and output of the mineral processing flow. In recent years, with the improvement of computing power and the development of numerical simulation technology, the discrete element method (DEM) has become an important tool for simulating particle motion and crushing behavior within SAG mills. Traditional DEM simulation establishes a particle geometric model and combines it with a contact mechanics model to simulate collisions and energy transfer between particles, generating motion trajectories and crushing data to support the optimization of mill operating parameters. Simultaneously, industrial CT scanning and 3D reconstruction technologies are gradually being applied to the analysis of ore microstructures. By acquiring fracture networks and mineral distribution characteristics, they provide a geometric basis for particle models. These technological advancements have driven the digital transformation of SAG processes, laying the foundation for improving fine particle yield, reducing energy consumption, and extending equipment life.

[0003] However, the simulation accuracy and optimization capabilities of existing technologies are still limited by the assumption of a single particle model, static operating parameter settings, and limited real-time monitoring methods. This makes it difficult to fully adapt to complex ore types and dynamic operating conditions. First, traditional discrete element simulations primarily use uniform particle models, neglecting the heterogeneity of the microscopic fracture network, bedding distribution, and mineral embedding characteristics within the ore. This leads to significant deviations between the simulation results and actual crushing behavior, making it difficult to accurately predict the crushing characteristics of different mineral regions. Furthermore, existing methods largely rely on static empirical formulas or single-condition analysis for operating parameter optimization, lacking a comprehensive consideration of dynamic adaptability to multiple operating conditions and energy consumption optimization. This limits the effectiveness of improving fine particle yield and reducing energy consumption. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a simulation method and system for semi-autogenous grinding processes in mines based on discrete element method, which solves the problem of limiting the improvement of fine particle yield and the reduction of energy consumption.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a simulation method for semi-autogenous grinding processes in mines based on discrete element method, comprising:

[0008] Microscopic fracture networks, bedding distribution, and mineral embedding characteristics of ore samples were collected. After generating a digital twin library of ore using a three-dimensional reconstruction algorithm, the irregular particle prototypes were segmented and assigned non-uniform physical property parameters to construct a multiphase coupled particle model.

[0009] Based on the multiphase coupled particle model, the dynamic variable boundary surface is updated in the discrete element simulation framework to track the particle motion trajectory and collision mode in the semi-autogenous mill. After generating the simulation dataset, the deep learning algorithm is used to mine the topology of the interparticle contact force network and the energy transfer path to obtain the crushing characteristic label and generate the parameter mapping matrix.

[0010] Based on the crushing characteristic labels and parameter mapping matrix, and by integrating IoT sensor data, a particle size distribution prediction model and an energy consumption optimization scheduling strategy are constructed to obtain an energy-saving operation plan. The simulation dataset is compared with the on-site production data to adjust the parameters of the multiphase coupled particle model and generate a calibration parameter set.

[0011] Based on the calibration parameter set, the influence law of semi-autogenous mill operating parameters is studied using the control variable method. A performance index database and analysis report are obtained. Multi-sensor data and lidar data are integrated to construct a grinding process perception mechanism and optimized control logic, and a visual interface is generated.

[0012] Based on a visual interface, the control logic is verified and optimized on-site, the multiphase coupled particle model and the optimized control logic are iteratively updated, and an implementation report is generated.

[0013] As a preferred embodiment of the discrete element method-based simulation method for semi-autogenous grinding processes in mines described in this invention, the following steps are taken: The microscopic fracture network, bedding distribution, and mineral embedding characteristics of the collected ore samples are used to generate a digital twin library of the ore using a three-dimensional reconstruction algorithm. Then, the irregularly shaped particle prototypes are segmented and assigned non-uniform physical property parameters to construct a multiphase coupled particle model. Specifically:

[0014] Representative ore samples from the mine site were obtained and subjected to three-dimensional scanning to collect projected image data containing microscopic cracks, bedding structures, and mineral phase boundaries.

[0015] The projected image data is imported into 3D reconstruction software to generate 3D volume data, and through preprocessing, a 3D digital model is formed.

[0016] Local regions with irregular shapes and internal structural features are extracted from the three-dimensional digital model as prototypes of irregular particles, separated into independent particle models, and classified and stored as a digital twin library of ores.

[0017] Based on the independent particle model in the ore digital twin library, parameters are assigned to different mineral regions, and the contact relationship of normal stiffness, tangential stiffness and friction coefficient parameters is defined to generate a multiphase coupled particle model.

[0018] As a preferred embodiment of the discrete element method-based simulation method for semi-autogenous grinding processes in mines according to the present invention, the method involves updating dynamically variable boundary surfaces within the discrete element simulation framework based on a multiphase coupled particle model, tracking particle motion trajectories and collision modes within the semi-autogenous grinding mill, and generating a simulation dataset. Specifically:

[0019] Import the multiphase coupled particle model into the discrete element simulation software, load the three-dimensional geometric models of the semi-autogenous mill cylinder, liner, and inlet / outlet, and set the operating parameters;

[0020] Enable the Bonded-Particle-Model contact model, set a weak bonding surface, update the contact state, and generate a sequence of particle motion trajectory data.

[0021] Based on the particle motion trajectory data sequence, the normal and tangential components of the contact force between particles are calculated using the Hertz-Mindlin contact model. The collision frequency, peak collision energy, and total energy dissipation are statistically analyzed to generate a simulation dataset.

[0022] As a preferred embodiment of the discrete element method-based simulation method for semi-autogenous grinding processes in mines described in this invention, the method for obtaining crushing characteristic labels specifically includes:

[0023] Based on the simulation dataset, a dynamic graph structure data is constructed with particles as nodes and contact forces as edges;

[0024] A graph neural network model is used to process dynamic graph structure data, extract the contact force propagation path, and combine it with cluster analysis to generate a set of fracture characteristic labels related to different mineral regions.

[0025] As a preferred embodiment of the discrete element method-based simulation method for semi-autogenous grinding processes in mines according to the present invention, the generation of the parameter mapping matrix specifically comprises:

[0026] Based on the simulation dataset and the fragmentation characteristic label set, the operating parameters and performance indicators are extracted;

[0027] The random forest regression algorithm is used to fit the mapping relationship between running parameters and performance indicators, and a parameter mapping matrix is ​​generated.

[0028] As a preferred embodiment of the discrete element method-based simulation method for semi-autogenous grinding processes in mines described in this invention, the construction of the particle size distribution prediction model specifically includes:

[0029] Deploy IoT sensors to collect operational data, match it with a set of crushing characteristic labels, and generate a joint dataset;

[0030] Based on the joint dataset, a long short-term memory network model is trained to generate a product granularity distribution curve, forming a granularity distribution prediction model to predict the fine particle yield under different combinations of operating parameters.

[0031] As a preferred embodiment of the discrete element method-based simulation method for semi-autogenous grinding processes in mines described in this invention, the method for obtaining an energy-saving operation plan specifically includes:

[0032] Obtain time-of-use electricity price data from the power grid, establish a unit energy consumption cost sequence, and calculate the unit processing cost by combining the operating parameter combinations of the parameter mapping matrix and the fine-grained yield under different operating parameter combinations.

[0033] A genetic algorithm is used to search for the optimal combination of operating parameters to maximize the yield of fine particles and minimize the unit processing cost. The setpoint sequence of mill speed, filling rate and feed rate in different time periods is generated to form an energy consumption optimization scheduling strategy, which is then converted into a control command sequence to generate an energy-saving operation plan.

[0034] As a preferred embodiment of the discrete element method-based simulation method for semi-autogenous grinding processes in mines according to the present invention, the generation of the calibration parameter set specifically includes:

[0035] Implement energy-saving operation plans and collect actual operation data;

[0036] By comparing the actual operating data with the simulation dataset, calculating the deviation value sequence, and adjusting the normal stiffness, tangential stiffness, and bond fracture energy parameters of the multiphase coupled particle model until the deviation meets the process performance requirements, a calibration parameter set is generated.

[0037] As a preferred embodiment of the discrete element method-based simulation method for semi-autogenous grinding processes in mines described in this invention, the construction of the grinding process perception mechanism and optimization control logic, and the generation of a visual interface, specifically includes:

[0038] Based on the calibration parameter set, a performance index database is generated using the control variable method, and an analysis report is generated through regression analysis.

[0039] By integrating point cloud data from vibration sensors, acoustic emission sensors, and lidar, a grinding process perception mechanism is constructed to generate estimated filling rates and early warning status of rock quantity. This is then matched with a performance index database to form optimized control logic and generate a visual interface.

[0040] Secondly, the present invention provides a simulation system for semi-autogenous grinding processes in mines based on discrete element method (DEM) methods, comprising:

[0041] The module collects the microscopic fracture network, bedding distribution and mineral embedding characteristics of ore samples, generates a digital twin library of ore through a three-dimensional reconstruction algorithm, segments the irregular particle prototypes and assigns non-uniform physical property parameters to construct a multiphase coupled particle model.

[0042] The mapping module, based on the multiphase coupled particle model, updates the dynamic variable boundary surface in the discrete element simulation framework, tracks the particle motion trajectory and collision mode in the semi-autogenous mill, generates a simulation dataset, and then uses deep learning algorithms to mine the topology of the interparticle contact force network and energy transfer path, obtains the crushing characteristic label, and generates a parameter mapping matrix.

[0043] The calibration module, based on crushing characteristic labels and parameter mapping matrices, integrates IoT sensor data to construct a particle size distribution prediction model and energy consumption optimization scheduling strategy, obtains an energy-saving operation plan, and compares simulation data with on-site production data to adjust the parameters of the multiphase coupled particle model and generate a calibration parameter set.

[0044] The analysis module, based on the calibration parameter set, uses the controlled variable method to study the influence of semi-autogenous mill operating parameters, obtains a performance index database and analysis report, and integrates multi-sensor data and lidar data to construct a grinding process perception mechanism and optimized control logic, generating a visual interface;

[0045] The summary module, based on a visual interface, verifies and optimizes the control logic on-site, iteratively updates the multiphase coupled particle model and the optimized control logic, and generates an implementation report.

[0046] The beneficial effects of this invention are as follows: By integrating IoT sensor data with crushing characteristic tag sets to generate a joint dataset, a particle size distribution prediction model is trained using a long short-term memory network model, and by combining the time-of-use electricity price of the power grid and the parameter mapping matrix, operating parameters are optimized through a genetic algorithm to generate an energy-saving operation plan. This overcomes the limitations of static operating parameter settings, realizes parameter optimization under dynamic operating conditions, significantly improves fine particle yield and energy efficiency, and reduces equipment operating costs. Through multiphase modeling and data-driven optimization, the simulation accuracy and process adaptability are comprehensively improved, providing an efficient and economical solution for semi-autogenous grinding processes in mines. Attached Figure Description

[0047] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 This is a flowchart of a simulation method for semi-autogenous grinding processes in mines based on discrete element method.

[0049] Figure 2 Flowchart for constructing a multiphase coupled particle model.

[0050] Figure 3The flowchart is for simulation and optimization.

[0051] Figure 4 This is a flowchart of the sensing mechanism in the grinding process. Detailed Implementation

[0052] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0053] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0054] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0055] Reference Figure 1 This is one embodiment of the present invention, which provides a simulation method for a semi-autogenous grinding process in a mine based on discrete element method, comprising the following steps:

[0056] S1: Collect the microscopic fracture network, bedding distribution and mineral embedding characteristics of ore samples, generate a digital twin library of ore through a three-dimensional reconstruction algorithm, segment the irregular particle prototypes and assign non-uniform physical property parameters to construct a multiphase coupled particle model.

[0057] Please see Figure 2 Specifically, it includes the following steps:

[0058] S1.1: Obtain representative ore samples from the mine site, cut and clean the ore samples to prepare ore specimens suitable for industrial CT scanning; use an industrial CT scanner to perform three-dimensional scanning on the ore specimens and acquire projected image data including micro-cracks, layering structure and mineral phase boundaries.

[0059] S1.2: Import the projected image data into the 3D reconstruction software to generate the 3D volume data of the ore sample.

[0060] Median filtering is applied to the 3D volume data to reduce imaging noise and preserve the grayscale data of the internal structure. Based on the gradient difference of the grayscale data, a region growing segmentation algorithm is used to segment the 3D volume data and identify fracture regions, bedding planes and regions with different mineral compositions.

[0061] The Marching-Cubes algorithm is applied to mesh the segmented 3D volume data to generate a triangular facet model containing internal structural information.

[0062] Smoothing and simplification operations are performed on the triangular facet model to remove isolated noise points while preserving the geometric features of the fracture network, bedding planes, and mineral boundaries.

[0063] S1.3: Save the processed triangular facet model as an STL file to form a three-dimensional digital model of the ore.

[0064] Local regions containing irregular shapes and internal structural features are extracted from the three-dimensional digital model as prototypes for heterogeneous particles.

[0065] Using the subset extraction function of 3D reconstruction software, the irregular particle prototype is separated into independent particle models. Surface detail preservation processing is performed on each independent particle model to maintain the original surface roughness morphology. All independent particle models are classified and stored to establish a digital twin library of ores.

[0066] S1.4: Based on the independent particle models in the ore digital twin library, according to the mineral composition distribution within each independent particle model, the hardness parameter is assigned to the quartz enrichment region, sulfide region, and clay region; based on the grayscale data in the three-dimensional volume data, a mapping relationship between grayscale data and mineral density is established, and the mineral density is assigned to the corresponding mineral region of the independent particle model; according to the mineral type and fracture region characteristics, the fracture energy parameter is assigned to the quartz region, sulfide region, and fracture surface region of the independent particle model; the independent particle models with assigned hardness, density, and fracture energy parameters are then imported into the discrete element simulation software.

[0067] By using the material property configuration function of the discrete element simulation software, hardness parameters, density parameters, and fracture energy parameters are bound to the corresponding geometric regions of the independent particle model; the contact relationship between different mineral regions inside the independent particle model is defined based on the normal stiffness parameters, tangential stiffness parameters, and friction coefficient parameters interpolated by local physical properties; the configured independent particle model and property parameters are saved to form a multiphase coupled particle model that can be used for simulation.

[0068] To further explain, the gradual construction from projected image data to 3D volume data, ore digital twin library, and multiphase coupled particle model ensures smooth data transfer, laying a solid foundation for subsequent discrete element simulation, contact force analysis, and process optimization. This helps improve the efficiency of semi-autogenous grinding process, reduce energy consumption, and extend equipment life.

[0069] S2: Based on the multiphase coupled particle model, the dynamic variable boundary surface is updated in the discrete element simulation framework to track the particle motion trajectory and collision mode in the semi-autogenous mill. After generating the simulation dataset, the deep learning algorithm is used to mine the topology of the contact force network between particles and the energy transfer path to obtain the crushing characteristic label and parameter mapping matrix.

[0070] Please see Figure 3 Specifically, it includes the following steps:

[0071] S2.1: Import the multiphase coupled particle model into the discrete element simulation software, load the three-dimensional geometric models of the semi-autogenous mill cylinder, liner, and inlet / outlet to form the simulation calculation domain; set the semi-autogenous mill operating parameters in the discrete element simulation software, including preset speed, preset filling rate, and steel ball size according to the actual process ratio, for example, the speed is 75% of the critical speed, the filling rate is 30%, and the steel ball size is imported according to the actual process ratio; randomly place the multiphase coupled particle model into the semi-autogenous mill geometric model according to the actual ore and steel ball ratio to form the initial particle filling state.

[0072] Based on the initial particle filling state, the Bonded-Particle-Model contact model is enabled in the simulation domain. The cracked and bedding regions inside the multiphase coupled particle model are set as weak bonding surfaces, and the bond fracture energy threshold is defined. The simulation is started, and the simulation process is advanced with a preset time step using an explicit time integration method. The contact state between the multiphase coupled particle model and the cylinder, liner, and particles is updated in the simulation domain. It is determined whether a collision occurs and the contact force is calculated. The position, velocity, acceleration, and force vector of each particle in the simulation domain are recorded to generate a sequence of particle motion trajectory data. The Bonded-Particle-Model contact model is a physical mechanics model based on the principles of Newtonian mechanics and contact mechanics and does not require training.

[0073] Based on the particle motion trajectory data sequence, the normal and tangential components of the interparticle contact force are extracted, and the collision frequency, peak collision energy, and total energy dissipation are statistically analyzed. The process is as follows: Based on the particle motion trajectory data sequence, the contact point position and relative velocity of each pair of particles in the simulation computational domain are extracted. Using the Hertz-Mindlin contact model, the normal contact force is calculated based on the position difference of the contact points. Based on the relative velocity of the particles at the contact points, the tangential component is extracted. Using Mindlin-Deresiewicz theory, tangential stiffness and friction coefficient, combined with tangential relative velocity, the tangential contact force is calculated, generating the normal and tangential components of the interparticle contact force. Based on the normal and tangential components of the interparticle contact force, the number of collisions within each time step is counted, and the collision frequency is calculated. Based on the normal and tangential components of the interparticle contact force, Newton's second law and explicit time integration are used to update the particle velocities before and after the collision. The kinetic energy change is obtained by calculating the squared difference of particle mass and velocity, where particle mass is derived from the density parameter of the multiphase coupled particle model. The maximum kinetic energy change is extracted as the peak collision energy, and the kinetic energy loss of all collisions is accumulated to generate the total energy dissipation. Sub-particle generation is performed on the broken particles to update the geometric shape and physical property binding relationship of the broken particles. The simulation is continuously run until the simulation computational domain reaches a stable condition, and particle motion trajectory and collision mode data are collected within the complete cycle. The particle position, velocity, contact force and breakage event data are exported as a structured time series dataset to form the simulation dataset. The Hertz-Mindlin contact model is a deterministic model based on mechanical formulas and does not require training.

[0074] S2.2: Based on the simulation dataset, a Python data analysis environment was used to construct a dynamic graph structure data with particles as nodes and contact forces as edges. The process was as follows: The position, velocity, and inter-particle contact forces of the particles were extracted at each time step to generate a particle attribute dataset. Each particle was defined as a node in the dynamic graph structure data, and the relationships involving contact forces between particles were defined as edges, generating an initial graph structure. The vector sum of the normal and tangential components of the inter-particle contact forces was calculated as edge weights to update the initial graph structure, generating the dynamic graph structure data. The particle position and velocity in the dynamic graph structure data were used as node features, and the contact force magnitude as edge features. A graph neural network model was used for training, and multi-layer graph convolution operations were used to update the node features. The contact force information of adjacent particles was aggregated, and the propagation path and aggregation pattern of the contact force in the dynamic graph structure data were extracted.

[0075] Cluster analysis is performed on the output features of the graph neural network to classify particles with similar contact force propagation behavior into the same category. Based on the clustering results, crushing behavior category labels for different ore types are generated, forming a crushing characteristic label set, specifically:

[0076] Based on the output features of the graph neural network model, the contact force propagation path and aggregation pattern features of each particle are extracted to generate a set of particle feature vectors. Based on this set, the K-Means clustering algorithm is used for cluster analysis: Using the particle feature vector set output by the graph neural network model, the contact force propagation path and aggregation pattern features of each particle are used as input data to initialize multiple cluster centers. By iteratively calculating the Euclidean distance between each particle's feature vector and the cluster center, particles are assigned to the nearest cluster center. The cluster center is updated to the mean of the assigned particle's feature vector. This iteration is repeated until the cluster centers stabilize or a preset number of iterations is reached. Based on the similarity of contact force propagation behavior, particles are divided into multiple categories, generating particle cluster categories. The ore type distribution of particles in each particle cluster category is analyzed, and crushing behavior features related to quartz-rich areas, sulfide areas, and clay areas are extracted to generate crushing behavior category labels for different ore types. All crushing behavior category labels are categorized and organized, and saved as a structured label dataset, forming a crushing characteristic label set.

[0077] S2.3: Classify the simulation data under different combinations of operating parameters, establish the correspondence between the semi-autogenous mill operating parameters and the crushing characteristic label set, and statistically analyze the fine particle generation rate, crushing energy consumption, and proportion of stubborn rocks under each group of semi-autogenous mill operating parameters to form a multi-dimensional response index set; use the semi-autogenous mill operating parameters as input variables and the multi-dimensional response index set as output variables to fit the nonlinear mapping relationship between parameters and performance; save the mapping relationship model and coefficient configuration to form a parameter mapping matrix, specifically:

[0078] (1) Based on the operating parameters of the semi-autogenous mill, the rotational speed, filling rate and steel ball size are extracted as input variables to generate an operating parameter dataset. Based on the multidimensional response index set, the fine particle generation rate, crushing energy consumption and stubborn rock ratio are extracted as output variables to generate a response index dataset.

[0079] (2) Based on the operating parameter dataset and response index dataset, the operating parameters of the semi-autogenous mill, such as rotational speed, filling rate, and steel ball size, as well as the fine particle generation rate, crushing energy consumption, and rock ratio of the multidimensional response index set, are normalized to generate standardized operating parameter datasets and standardized response index datasets. These datasets are then divided into training datasets and validation datasets. The training dataset is used for constructing decision trees for the random forest regression algorithm, while the validation dataset is used to evaluate the performance of the mapping relationship model. Based on the training dataset, the random forest regression algorithm is used to construct multiple decision trees by randomly sampling subsets of the semi-autogenous mill's operating parameters. Each decision tree is constructed based on the rotational speed, filling rate, steel ball size, fine particle generation rate, crushing energy consumption, and rock ratio of the multidimensional response index set. The filling rate and steel ball size predict the fine particle generation rate, crushing energy consumption, and the proportion of stubborn rocks. After generating a decision tree set, the prediction results of each decision tree are averaged and integrated to generate the prediction output of the random forest regression algorithm. The nonlinear mapping relationship between the semi-autogenous mill operating parameters and the multidimensional response index set is fitted to generate a mapping relationship model. Based on the validation dataset, the error between the prediction output of the mapping relationship model and the actual multidimensional response index set is calculated. The number of trees and the feature selection ratio of the decision tree set are adjusted to optimize the prediction accuracy of the mapping relationship model. Based on the optimized mapping relationship model, the prediction error on the validation dataset is verified to meet the preset accuracy requirements. The final decision tree set and weight configuration are saved to generate the mapping relationship model.

[0080] (3) Based on the mapping relationship model, extract the tree structure weights and feature importance coefficients of the random forest regression algorithm, generate the coefficient configuration dataset, and organize the weights and coefficients of the mapping relationship model into structured data and save them as parameter mapping matrix.

[0081] To further explain, by simulating the crushing behavior of different ore types and optimizing operating parameters, a set of crushing characteristic labels and parameter mapping matrices for mineral regions such as quartz, sulfides, and clay are generated. This can adapt to various ore types and process conditions, optimize the fine particle generation rate, reduce crushing energy consumption, and reduce the proportion of stubborn rocks, thereby improving the efficiency of the semi-autogenous grinding process, reducing energy consumption, and extending the service life of the equipment.

[0082] S3: Based on the crushing characteristic labels and parameter mapping matrix, integrate IoT sensor data to construct a particle size distribution prediction model and energy consumption optimization scheduling strategy, obtain an energy-saving operation plan, and compare the simulation dataset with the on-site production data to adjust the parameters of the multiphase coupled particle model and generate a calibration parameter set.

[0083] Specifically, the following steps are included:

[0084] S3.1: Deploy IoT sensors at the semi-autogenous mill site, including power sensors, vibration sensors, slurry concentration meters, and feed weighing devices, to collect operational data in real time; synchronize the operational data collected by the IoT sensors to the industrial data platform using timestamp alignment to form a continuous time-series stream of operational parameters; perform spatiotemporal matching of the operational parameter time-series stream with the crushing characteristic tag set to construct a joint dataset containing operating condition features and crushing behavior. The process is as follows: Based on the operational parameter time-series stream, extract the semi-autogenous mill operating parameters within each time step, including rotational speed, filling rate, feed rate, and slurry concentration, to form an operational parameter feature sequence; based on the crushing characteristic tag set, extract crushing behavior category tags corresponding to the timestamps of the operational parameter time-series stream for different ore types to form a crushing behavior tag sequence; based on the timestamps, align and match the operational parameter feature sequence with the crushing behavior tag sequence to generate time-synchronized operating condition-behavior data pairs; integrate the semi-autogenous mill operating parameters and crushing behavior category tags through the time-synchronized operating condition-behavior data pairs to generate a joint dataset containing operating condition features and crushing behavior.

[0085] S3.1.1: A long short-term memory network model is trained based on a joint dataset. A multi-layer network structure is used to process time-series data, predicting the product granularity distribution curve. After optimizing the model weights, a granularity distribution prediction model is formed, predicting fine-grained yields under different combinations of operating parameters. Specifically:

[0086] (1) Based on the joint dataset, extract the time-synchronized semi-autogenous mill operating parameters and crushing behavior category labels to construct serialized input data containing time steps; based on the serialized input data, use a long short-term memory network model for training, set the input layer to receive semi-autogenous mill operating parameters and crushing behavior category labels, the hidden layer to process time series features, and the output layer to generate product particle size distribution curve; based on the predicted particle size distribution curve of the long short-term memory network model and the laboratory screening results, calculate the mean square error as the prediction error to measure the deviation of the particle size distribution curve, and use the gradient descent method to adjust the weight of the long short-term memory network model. When the mean square error is lower than the preset error threshold, it indicates convergence and generates a particle size distribution prediction model. The error threshold is customized based on the required prediction accuracy.

[0087] (2) Based on the particle size distribution prediction model, input different combinations of semi-autogenous mill operating parameters to predict the corresponding product particle size distribution curve, extract the particle proportion of the target particle size range from it, and generate fine particle yield.

[0088] S3.1.2: Obtain time-of-use electricity price data from the power grid, divide the peak, flat, and valley periods, establish unit energy consumption cost sequences for different periods, and combine the unit energy consumption cost sequences with the fine-grained yield of the granular distribution prediction model to form an energy-saving operation plan.

[0089] The operating parameter combinations in the parameter mapping matrix are associated with the corresponding energy consumption indicators, combined with the unit energy cost sequence and the predicted fine particle yield. A genetic algorithm is used to search for the optimal operating parameter combination within the parameter mapping matrix to maximize the fine particle yield and minimize the unit processing cost. A time-segmented setpoint sequence of mill speed, filling rate, and feed rate is generated to form an energy consumption optimization scheduling strategy. This strategy is then transformed into an executable control command sequence to form an energy-saving operation plan, specifically:

[0090] (1) Based on the parameter mapping matrix, extract the combination of semi-autogenous mill operating parameters, including rotational speed, filling rate and feed rate, as well as the corresponding energy consumption index, and generate an operating parameter-energy consumption dataset. Based on the operating parameter-energy consumption dataset and the unit energy consumption cost sequence, calculate the unit processing cost of each operating parameter combination. Based on the particle size distribution prediction model, extract the predicted fine particle yield. Combined with the unit processing cost, construct a weighted fitness function, assign positive weight to the fine particle yield and negative weight to the unit processing cost, and comprehensively evaluate the advantages and disadvantages of the operating parameter combination. Based on the weighted fitness function, use a genetic algorithm to initialize a population containing multiple operating parameter combinations. Iteratively update the population through selection, crossover and mutation operations, calculate the fitness value of each operating parameter combination, and select the operating parameter combination with the highest fine particle yield and the lowest unit processing cost. When the fitness value converges, output the optimal operating parameter combination.

[0091] (2) Based on the optimal combination of operating parameters obtained by the genetic algorithm search, the setpoint sequence of mill speed, filling rate and feed rate for each time period is generated according to the peak, flat and valley periods of the grid time-of-use electricity price, forming an energy consumption optimization scheduling strategy. The setpoint sequence for each time period is then converted into a control command sequence that can be recognized by the semi-autogenous mill control equipment according to the energy consumption optimization scheduling strategy, forming an energy-saving operation plan.

[0092] S3.2: Implement an energy-saving operation plan in the mine production center, and simultaneously collect data on particle size distribution, power consumption, and liner wear during actual operation to form on-site production data. Compare the on-site production data with the simulation results of the corresponding working conditions in the simulation dataset item by item, and calculate the deviation value sequence of fine particle yield, peak power, and rock ratio. The process is as follows: Based on the on-site production data, extract the fine particle yield, peak power, and rock ratio under the actual operating conditions to generate an on-site performance index sequence; Based on the simulation dataset, extract the simulation working conditions with the same combination of operating parameters as the on-site production data, obtain the corresponding fine particle yield, peak power, and rock ratio, and generate a simulation performance index sequence; Through the on-site performance index sequence and the simulation performance index sequence, calculate the absolute difference of fine particle yield, peak power, and rock ratio for each time step to generate individual deviation values, sort them by time step, integrate the deviations of fine particle yield, peak power, and rock ratio, and generate a time-seriesd deviation value sequence.

[0093] Based on the statistical trend of the deviation value sequence, the regular errors of the normal stiffness, tangential stiffness, and bond fracture energy parameters in the multiphase coupled particle model are identified. The normal stiffness, tangential stiffness, and bond fracture energy parameters are adjusted sequentially in the discrete element simulation software, and the simulation under the corresponding working conditions is run again. The simulation results after parameter adjustment are compared with the field production data again to verify the convergence of the deviation value sequence.

[0094] When the deviations of fine particle yield, peak power, and rock ratio are all below the parameter thresholds, the current parameter configuration is saved. The parameter thresholds are set based on the performance requirements of the semi-autogenous mill process. The saved combination of normal stiffness, tangential stiffness, and bond fracture energy parameters is defined as the calibration parameter set.

[0095] Preferably, by combining datasets, particle size distribution prediction models, and energy consumption optimization scheduling strategies, the system can adapt to different ore types (such as quartz and sulfides) and operating conditions (such as peak and off-peak electricity prices) to optimize fine particle yield, reduce energy consumption, and decrease liner wear. By calibrating parameter sets, the system ensures that the simulation is consistent with actual production, thereby reducing equipment overload and wear. Compared with existing optimization methods that target a single operating condition, the multi-condition adaptation capability of step S3 significantly improves the operating efficiency and service life of the semi-autogenous mill.

[0096] S4: Based on the calibration parameter set, the influence law of semi-autogenous mill operating parameters is studied using the control variable method. A performance index database and analysis report are obtained. Multi-sensor data and lidar data are integrated to construct a grinding process perception mechanism and optimize control logic, and generate a visualization interface.

[0097] Specifically, the following steps are included:

[0098] S4.1: Based on the calibration parameter set, the influence of operating parameters on the semi-autogenous mill is studied using the controlled variable method, generating a performance index database and analysis report. Specifically:

[0099] The calibration parameter set was loaded into the discrete element simulation software as the baseline model parameters for the simulation experiment. The semi-autogenous mill speed, filling rate, steel ball ratio, and liner lifting angle were set as variables in sequence. For each variable, multiple different values ​​were set, while keeping other parameters constant. The simulation was run, and the fine particle output, unit energy consumption, liner wear rate, and amount of rock accumulation were recorded under each working condition. The recorded data from all working conditions were summarized to establish a performance index database containing the correspondence between operating parameters and performance indicators. Trend analysis was performed on the data in the performance index database to generate an analysis report on the influence of semi-autogenous mill operating parameters. The trend analysis process involves: extracting data pairs between the semi-autogenous mill's operating parameters (including rotational speed, filling rate, steel ball ratio, and liner lift angle) and corresponding performance indicators (including fine particle output, unit energy consumption, liner wear rate, and rock accumulation) based on the performance index database; performing regression analysis on fine particle output, unit energy consumption, liner wear rate, and rock accumulation for each operating parameter to fit the functional relationship between each operating parameter and the performance indicators; and generating continuous data points using interpolation methods based on the fitted functional relationship to plot trend curves of rotational speed, filling rate, steel ball ratio, and liner lift angle with each performance indicator.

[0100] S4.2: Based on a performance index database and multi-sensor data, and by fusing lidar data, a grinding process perception mechanism and optimized control logic are constructed, generating a visual interface. Please refer to [link / reference]. Figure 4 Specifically:

[0101] (1) Install vibration sensors and acoustic emission sensors on the outside of the semi-autogenous mill cylinder, and arrange laser radar scanners at the inlet and outlet ends to synchronously collect vibration signals, acoustic signals and laser radar point cloud data. The laser radar point cloud data reflects the spatial position and trajectory of the material.

[0102] The vibration and acoustic signals are transformed by time and frequency to extract the energy amplitude of the characteristic frequency bands. The lidar point cloud data is reconstructed in three dimensions to generate a dynamic image of the material distribution and motion state inside the mill. The energy amplitude of the characteristic frequency bands is correlated with the material filling height and falling trajectory in the dynamic image to establish an empirical correspondence between the vibration energy amplitude and the material filling rate, and between the acoustic signal characteristics and the amount of rock accumulation. This correspondence is encoded into state recognition rules to form a perception mechanism for the grinding process.

[0103] The specific process of conducting correlation analysis to form a perception mechanism for the grinding process is as follows: Correlation analysis is used to calculate the correlation coefficient between the energy amplitude of characteristic frequency bands and the material filling height and drop trajectory. This process involves: based on time-domain signal data collected by vibration and acoustic emission sensors, the signals are converted to the frequency domain using a fast Fourier transform, and the energy amplitude of the characteristic frequency bands is extracted to generate a sequence of characteristic frequency band energy amplitudes. Based on lidar point cloud data, the geometric distribution of materials within the semi-autogenous mill is reconstructed, and the time-series data of the material filling height and drop trajectory are calculated to generate material filling height sequences and drop trajectory sequences. The Pearson correlation coefficient method was used to calculate the correlation coefficients between the characteristic frequency band energy amplitude sequence and the material filling height sequence and the falling trajectory sequence, respectively, to quantify the linear correlation strength between vibration energy amplitude and material filling height and falling trajectory. Based on the correlation analysis results, regression analysis was used to fit the functional relationship between vibration energy amplitude and material filling rate, and between acoustic signal characteristics and rock accumulation amount, to establish an empirical correspondence. The empirical correspondence was transformed into conditional judgment rules, setting threshold ranges for vibration energy amplitude and acoustic signal characteristics, corresponding to the state intervals of material filling rate and rock accumulation amount, and encoded as state recognition rules. Based on the state recognition rules, a grinding process sensing mechanism was constructed to identify the material filling rate and rock accumulation amount status in real time during the operation of the semi-autogenous mill.

[0104] To transform empirical correspondences into conditional judgment rules, threshold ranges for vibration energy amplitude and acoustic signal characteristics are set, corresponding to the state intervals of material filling rate and rock accumulation. An example of encoding these into state recognition rules is as follows:

[0105] Based on empirical correlations, higher vibration energy amplitude corresponds to higher material filling rate, and higher intensity of a specific frequency band of the acoustic signal corresponds to a greater amount of rock accumulation. Therefore, the following conditional judgment rules are established: if the vibration energy amplitude is greater than the vibration characteristic threshold, the material filling rate is determined to be high; otherwise, it is determined to be low. If the intensity of a specific frequency band of the acoustic signal is greater than the acoustic characteristic threshold, the rock accumulation is determined to be high; otherwise, it is determined to be low. These conditional judgment rules are encoded as state recognition rules. For example, when the vibration energy amplitude exceeds the vibration characteristic threshold and the intensity of a specific frequency band of the acoustic signal is lower than the acoustic characteristic threshold, the operating state is identified as "high material filling rate, low rock accumulation"; when the vibration energy amplitude is lower than the vibration characteristic threshold and the intensity of a specific frequency band of the acoustic signal exceeds the acoustic characteristic threshold, the operating state is identified as "low material filling rate, high rock accumulation". The vibration characteristic threshold range and the acoustic characteristic threshold range are set based on statistical analysis of empirical correspondence. Based on the vibration characteristic threshold range and the acoustic characteristic threshold range, state recognition rules are generated to identify the material filling rate and rock accumulation status in the semi-autogenous mill during operation in real time.

[0106] (2) Match the optimal working condition combination in the performance index database with the state recognition rules of the grinding process sensing mechanism. Set the response action of increasing the feed rate and adjusting the rotation speed when the grinding process sensing mechanism recognizes that the filling rate is too low and the amount of stubborn stone increases. Set the response action of reducing the filling rate or adjusting the steel ball ratio when the grinding process sensing mechanism recognizes that the energy consumption is abnormally high. Combine the response actions into a control command sequence to form the optimized control logic.

[0107] The human-machine interface software imports the mill's 3D model and real-time sensor data, and displays material movement status images, filling rate estimates, rock quantity warning status, and recommended control commands in real time. The control commands are then connected to the field programmable logic controller to enable operation suggestion push and remote parameter setting, generating a visual interface.

[0108] To further explain, the process of establishing empirical correspondences between vibration energy amplitude and material filling rate, and between acoustic signal characteristics and rock accumulation, and encoding these correspondences into state recognition rules to construct a grinding process perception mechanism, is significantly more accurate and real-time than existing process perception methods that often rely on single sensors or static monitoring. This invention uses multi-source data fusion (vibration, acoustic waves, and lidar point clouds) and dynamic three-dimensional reconstruction to comprehensively characterize the material movement state inside the mill, thereby significantly improving the accuracy and real-time performance of operational status perception.

[0109] S5: Based on a visual interface, verify and optimize the control logic on-site, iteratively update the multiphase coupled particle model and optimize the control logic, and generate an implementation report.

[0110] Specifically, the steps include the following:

[0111] Using the visualization interface generated by S4.2, the material movement status image, filling rate estimate, rock quantity warning status, and recommended control commands of the semi-autogenous mill are displayed in real time. The filling rate estimate and rock quantity warning status are generated by the state recognition rules of the grinding process sensing mechanism, and inferred based on the actual operating data collected by vibration sensors, acoustic emission sensors, and lidar scanners. The control command sequence of the optimized control logic is executed on-site in the mine, and the actual operating data, including fine particle output, unit energy consumption, and liner wear rate, are collected. The fine particle output, unit energy consumption, and liner wear rate in the actual operating data are compared with the filling rate estimate and rock quantity warning status displayed on the visualization interface to calculate the deviation. Based on the deviation, the physical property parameters of the multiphase coupled particle model (e.g., normal stiffness, tangential stiffness, and bond fracture energy) and the response actions of the optimized control logic (e.g., feed rate or speed adjustment) are adjusted. On-site verification and parameter adjustment are repeated until the deviation meets the process performance requirements of the semi-autogenous mill.

[0112] Summarize the verification results, parameter adjustment records, and updated control logic optimization plans, generate an implementation report, and record the process optimization effects and operating parameter configurations.

[0113] This embodiment also provides a simulation system for semi-autogenous grinding processes in mines based on discrete element method (DEM) methods, including:

[0114] The module collects the microscopic fracture network, bedding distribution, and mineral embedding characteristics of ore samples. After generating a digital twin library of ore through a three-dimensional reconstruction algorithm, it segments the irregular particle prototypes and assigns non-uniform physical property parameters to construct a multiphase coupled particle model.

[0115] The mapping module, based on a multiphase coupled particle model, updates the dynamically variable boundary surface in the discrete element simulation framework, tracks the particle motion trajectory and collision mode in the semi-autogenous mill, generates a simulation dataset, and then uses deep learning algorithms to mine the topology of the interparticle contact force network and energy transfer path, obtains the crushing characteristic label, and generates a parameter mapping matrix.

[0116] The calibration module, based on crushing characteristic labels and parameter mapping matrices, integrates IoT sensor data to construct a particle size distribution prediction model and energy consumption optimization scheduling strategy, obtains an energy-saving operation plan, and compares simulation data with on-site production data to adjust the parameters of the multiphase coupled particle model and generate a calibration parameter set.

[0117] The analysis module, based on the calibration parameter set, uses the controlled variable method to study the influence of semi-autogenous mill operating parameters, obtains a performance index database and analysis report, and integrates multi-sensor data and lidar data to construct a grinding process perception mechanism and optimize control logic, generating a visualization interface.

[0118] The summary module, based on a visual interface, verifies and optimizes the control logic on-site, iteratively updates the multiphase coupled particle model and the optimized control logic, and generates an implementation report.

[0119] This embodiment also provides a computer device, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the simulation method for semi-autogenous grinding process in mines based on discrete element method proposed in the above embodiment.

[0120] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0121] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the discrete element method-based simulation method for semi-autogenous grinding processes in mines as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Red-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0122] In summary, this invention generates a joint dataset by integrating IoT sensor data with crushing characteristic tag sets, trains a particle size distribution prediction model using a long short-term memory network model, combines time-of-use electricity pricing and parameter mapping matrices, and optimizes operating parameters through a genetic algorithm to generate an energy-saving operation plan. This overcomes the limitations of static operating parameter settings, achieves parameter optimization under dynamic operating conditions, significantly improves fine particle yield and energy efficiency, reduces equipment operating costs, and comprehensively enhances simulation accuracy and process adaptability through multiphase modeling and data-driven optimization, providing an efficient and economical solution for semi-autogenous grinding processes in mines.

[0123] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A simulation method for semi-autogenous grinding processes in mines based on discrete element method, characterized in that: include: Microscopic fracture networks, bedding distribution, and mineral embedding characteristics of ore samples were collected. After generating a digital twin library of the ore using a 3D reconstruction algorithm, irregularly shaped particle prototypes were segmented and assigned non-uniform physical property parameters to construct a multiphase coupled particle model. Specifically: Representative ore samples from the mine site were obtained and subjected to three-dimensional scanning to collect projected image data containing microscopic cracks, bedding structures, and mineral phase boundaries. The projected image data is imported into 3D reconstruction software to generate 3D volume data, and through preprocessing, a 3D digital model is formed. Local regions with irregular shapes and internal structural features are extracted from the three-dimensional digital model as prototypes of irregular particles, separated into independent particle models, and classified and stored as a digital twin library of ores. Based on the independent particle models in the ore digital twin library, parameters are assigned to different mineral regions, and the contact relationships of normal stiffness, tangential stiffness, and friction coefficient parameters are defined to generate a multiphase coupled particle model, specifically: By using the material property configuration function of discrete element simulation software, hardness parameters, density parameters, and fracture energy parameters are bound to the corresponding geometric regions of independent particle models. The contact relationship between different mineral regions within an independent particle model is defined based on the normal stiffness parameter, tangential stiffness parameter, and friction coefficient parameter interpolated from local physical properties. Save the configured independent particle model and its attribute parameters to form a multiphase coupled particle model that can be used for simulation; Based on the multiphase coupled particle model, the dynamic variable boundary surface is updated in the discrete element simulation framework to track the particle motion trajectory and collision mode in the semi-autogenous mill. After generating the simulation dataset, the deep learning algorithm is used to mine the topology of the interparticle contact force network and the energy transfer path, obtain the crushing characteristic label and generate the parameter mapping matrix. The obtained breakage characteristic label is specifically as follows: Based on the simulation dataset, a dynamic graph structure data is constructed with particles as nodes and contact forces as edges; A graph neural network model is used to process dynamic graph structure data, extract contact force propagation paths, and combine this with cluster analysis to generate a set of fracture characteristic labels related to different mineral regions. Specifically: Cluster analysis is performed on the output features of the graph neural network to classify particles with similar contact force propagation behavior into the same category; Based on the clustering results, crushing behavior category labels for different ore types are generated, forming a set of crushing characteristic labels; The generated parameter mapping matrix is ​​specifically as follows: Based on the simulation dataset and crushing characteristic label set, the operating parameters and performance indicators of the semi-autogenous mill are extracted; The random forest regression algorithm was used to fit the mapping relationship between the operating parameters and performance indicators of the semi-autogenous mill, and a parameter mapping matrix was generated. Based on the crushing characteristic labels and parameter mapping matrix, and by integrating IoT sensor data, a particle size distribution prediction model and an energy consumption optimization scheduling strategy are constructed to obtain an energy-saving operation plan. The simulation dataset is compared with the on-site production data to adjust the parameters of the multiphase coupled particle model and generate a calibration parameter set. The construction of the granularity distribution prediction model is specifically as follows: Deploy IoT sensors to collect operational data, match it with a set of crushing characteristic labels, and generate a joint dataset; Based on the joint dataset, a long short-term memory network model is trained to generate a product granularity distribution curve, forming a granularity distribution prediction model to predict the fine particle yield under different combinations of operating parameters. The energy-saving operation plan obtained is as follows: Obtain time-of-use electricity price data from the power grid, establish a unit energy consumption cost sequence, and calculate the unit processing cost by combining the operating parameter combinations of the parameter mapping matrix and the fine-grained yield under different operating parameter combinations. A genetic algorithm is used to search for the optimal combination of operating parameters to maximize the fine particle yield and minimize the unit processing cost. A time-segmented sequence of mill speed, filling rate and feed rate setpoints is generated to form an energy consumption optimization scheduling strategy, which is then converted into a control command sequence to generate an energy-saving operation plan. Based on the calibration parameter set, the influence law of semi-autogenous mill operating parameters is studied using the control variable method. A performance index database and analysis report are obtained. Multi-sensor data and lidar data are integrated to construct a grinding process perception mechanism and optimized control logic, and a visualization interface is generated. The construction of the grinding process perception mechanism and optimization control logic, and the generation of a visual interface, are specifically as follows: Based on the calibration parameter set, a performance index database is generated using the control variable method, and an analysis report is generated through regression analysis. By integrating vibration sensor, acoustic emission sensor and lidar point cloud data, a grinding process perception mechanism is constructed to generate filling rate estimate and rock quantity warning status, and a performance index database is matched to form optimized control logic and generate a visual interface. Based on a visual interface, the optimized control logic is verified on-site, the multiphase coupled particle model and optimized control logic are iteratively updated, an implementation report is generated, and the process optimization effect and operating parameter configuration are recorded.

2. The simulation method for semi-autogenous grinding processes in mines based on discrete element method as described in claim 1, characterized in that: The method based on the multiphase coupled particle model updates the dynamically variable boundary surface within the discrete element simulation framework, tracks the particle motion trajectory and collision mode within the semi-autogenous grinder, and generates a simulation dataset, specifically as follows: Import the multiphase coupled particle model into the discrete element simulation software, load the three-dimensional geometric models of the semi-autogenous mill cylinder, liner, and inlet / outlet, and set the operating parameters; Enable the Bonded-Particle-Model contact model, set a weak bonding surface, update the contact state, and generate a sequence of particle motion trajectory data. Based on the particle motion trajectory data sequence, the normal and tangential components of the contact force between particles are calculated using the Hertz-Mindlin contact model. The collision frequency, peak collision energy, and total energy dissipation are statistically analyzed to generate a simulation dataset.

3. The simulation method for semi-autogenous grinding processes in mines based on discrete element method as described in claim 2, characterized in that: The generation of the calibration parameter set specifically includes: Implement energy-saving operation plans and collect actual operation data; By comparing the actual operating data with the simulation dataset, calculating the deviation value sequence, and adjusting the normal stiffness, tangential stiffness, and bond fracture energy parameters of the multiphase coupled particle model until the deviation meets the process performance requirements, a calibration parameter set is generated.

4. A simulation system for semi-autogenous grinding processes in mines based on discrete element method, based on the simulation method for semi-autogenous grinding processes in mines based on discrete element method as described in any one of claims 1 to 3, characterized in that: include: The module collects the microscopic fracture network, bedding distribution and mineral embedding characteristics of ore samples, generates a digital twin library of ore through a three-dimensional reconstruction algorithm, segments the irregular particle prototypes and assigns non-uniform physical property parameters to construct a multiphase coupled particle model. The mapping module, based on the multiphase coupled particle model, updates the dynamic variable boundary surface in the discrete element simulation framework, tracks the particle motion trajectory and collision mode in the semi-autogenous mill, generates a simulation dataset, and then uses deep learning algorithms to mine the topology of the interparticle contact force network and energy transfer path, obtains the crushing characteristic label, and generates a parameter mapping matrix. The calibration module, based on crushing characteristic labels and parameter mapping matrices, integrates IoT sensor data to construct a particle size distribution prediction model and energy consumption optimization scheduling strategy, obtains an energy-saving operation plan, and compares simulation data with on-site production data to adjust the parameters of the multiphase coupled particle model and generate a calibration parameter set. The analysis module, based on the calibration parameter set, uses the controlled variable method to study the influence of semi-autogenous mill operating parameters, obtains a performance index database and analysis report, and integrates multi-sensor data and lidar data to construct a grinding process perception mechanism and optimized control logic, generating a visual interface; The summary module, based on a visual interface, verifies and optimizes the control logic on-site, iteratively updates the multiphase coupled particle model and the optimized control logic, and generates an implementation report.

Images