Method for mechanically grinding coal gangue to enhance activation

Abstract

The application discloses a coal gangue mechanical grinding enhancement activation method, comprising the following steps: obtaining multi-dimensional characterization data of coal gangue, performing inhomogeneous modeling on coal gangue particles through a patch algorithm based on the multi-dimensional characterization data, calculating the crushing energy consumption of each patch according to the grinding parameters of the mineral phase, learning the feature mapping of the grinding parameters and the lattice state through a double-branch attention network for high-activity patches, combining the grinding parameters as a strategy set, solving the optimization objective through a non-dominated sorting genetic algorithm using the strategy set, obtaining an optimal solution set of the grinding parameters, screening a target grinding parameter based on the optimal solution set through the mineral phase of an application scenario, and performing targeted grinding on the coal gangue to obtain activated and enhanced coal gangue grinding particles. The method performs inhomogeneous modeling on the multi-dimensional characterization data of the coal gangue, and screens a target grinding parameter according to an application scenario, thereby improving the pertinence of the coal gangue grinding effect, and reducing the grinding energy consumption and processing cost.

Description

A method for mechanical grinding and activation of coal gangue Technical Field

[0001] This invention relates to the field of coal gangue activation, and more particularly to a method for enhancing the activation of coal gangue through mechanical grinding. Background Technology

[0002] Coal gangue, a major industrial solid waste generated during coal mining and washing in my country, has accumulated in enormous quantities. It not only occupies vast amounts of land resources, but its leaching wastewater and dust also cause water, soil, and air pollution, becoming a key bottleneck restricting the green transformation of the coal industry. At the same time, coal gangue is rich in minerals such as silicon, aluminum, and calcium, possessing the potential to be transformed into high-value-added products such as building material admixtures, wastewater treatment adsorbents, and soil conditioners. Its efficient activation and resource utilization have become important requirements for implementing carbon emission reduction and promoting solid waste reduction. Currently, coal gangue activation technologies mainly include three categories: chemical activation, thermal activation, and mechanical grinding activation. Chemical activation requires the addition of acid and alkali reagents, resulting in secondary pollution and high costs. Thermal activation relies on high-temperature calcination, consuming large amounts of energy and requiring high equipment investment, limiting its large-scale application. Traditional mechanical grinding activation achieves activation by physically crushing particles to increase their specific surface area, and has become the mainstream approach due to its simple process and lack of chemical additives.

[0003] Existing mechanical grinding activation technologies employ a homogenization crushing approach, neglecting the heterogeneous distribution characteristics of mineral phases such as kaolinite and quartz within coal gangue. This makes it impossible to distinguish and crush highly active mineral regions, resulting in wasted grinding energy, insufficient exposure of active sites after grinding, and low activation efficiency. The activation mechanism relies on adjusting parameters based on macroscopic experimental data, failing to clarify the activation effect from the perspectives of lattice defect generation and fine particle crushing. The research and development cycle is long and lacks precision, often exhibiting the contradiction of high activity accompanied by high energy consumption or low energy consumption leading to substandard activity. Furthermore, the technology has poor adaptability to the compositional fluctuations of coal gangue from different mining areas. Therefore, how to grind and activate various types of coal gangue to achieve precise activation, sufficient mechanistic support, and optimal multi-objective balance has become an urgent problem to be solved. Summary of the Invention

[0004] The purpose of this invention is to provide a method for mechanically grinding and enhancing the activation of coal gangue.

[0005] To achieve the above objectives, the present invention is implemented according to the following technical solution:

[0006] This invention provides a method for mechanically grinding and enhancing the activation of coal gangue, comprising:

[0007] Multidimensional characterization data of coal gangue is obtained, and the multidimensional characterization data is preprocessed, including three-dimensional particle size, mineral phase and reference activity.

[0008] Based on the multi-dimensional characterization data, a patch algorithm is used to model the heterogeneity of coal gangue particles, and the lattice structure of the mineral phase is used to assign fracture strength and elastic modulus to each patch.

[0009] The energy consumption for crushing each patch is calculated based on the grinding parameters of the mineral phase. For highly active patches, a dual-branch attention network is used to learn the feature mapping of grinding parameters and lattice state to obtain the combination of grinding parameters and activation effect.

[0010] The grinding parameters are combined as a strategy set, and the activation effect, energy consumption cost and crushing energy consumption are taken as optimization objectives. The optimization objectives are solved by a non-dominated sorting genetic algorithm using the strategy set to obtain the optimal solution set of grinding parameters.

[0011] Based on the optimal solution set, the target grinding parameters are selected by mineral phase screening in the application scenario to perform targeted grinding of coal gangue, thereby obtaining activated and enhanced coal gangue grinding particles.

[0012] Furthermore, the method for obtaining the patch includes:

[0013] Based on the distribution data of three-dimensional particle size in multi-dimensional characterization data, the irregular three-dimensional geometric shape of coal gangue particles is reconstructed by the discrete element method. The particle outline of the irregular three-dimensional geometric shape is represented as a Fourier series. The reconstructed model is voxelized and the voxel resolution is set to 1 / 50 to 1 / 100 of the minimum feature size of the particle. The relative position coordinates and volume ratio of each voxel are recorded according to the coordinates between voxels.

[0014] Based on the spatial distribution data of mineral phases, the main mineral phases are spatially assigned in the voxel model using multi-point geostatistical methods. The same mineral phase type and a difference in elastic modulus of less than 10% between voxels are used as the physical property similarity condition. Based on the volume fraction of mineral phases and the physical property similarity condition, voxel aggregation is performed using a region growth algorithm. If the difference in mineral phase composition after aggregation of adjacent voxels is less than 15%, a heterogeneous patch is obtained. The kaolinite patch in the heterogeneous patch is used as a high-activity region, and the quartz patch is used as a high-hardness region to obtain mineral phase labels.

[0015] The interface region between adjacent patches is obtained. Based on the interface region, a three-layer coupled structure of patch, interface and patch is generated through cohesive unit according to the relationship between the interface bonding energy of mineral phase and crystallographic orientation. This results in the heterogeneous modeling of coal gangue particles. The main mineral phases include quartz, kaolinite and illite.

[0016] Furthermore, the method for obtaining the fracture strength and elastic modulus includes:

[0017] Based on the lattice structure of the patches and mineral phases, the interplanar spacing, lattice shear modulus, surface energy, dislocation density, and Poisson's ratio are obtained, and the initial fracture strength of each patch is calculated. The formula for calculating the initial fracture strength is as follows:

[0018] ;

[0019] in For patch Initial fracture strength, in MPa. This refers to the lattice shear modulus, expressed in GPa. Interplanar spacing, in nm. For surface energy, For dislocation density, Boltzmann's constant, This is the equivalent thermodynamic temperature during grinding, obtained from the grinding parameters of the mineral phase. The patch feature size is obtained from the voxel volume;

[0020] The elastic modulus is calculated using the continuum mechanics formula based on the lattice shear modulus and Poisson's ratio. Each patch is then assigned a value using the elastic modulus and initial fracture strength. The fracture strength and elastic modulus of each patch are randomly perturbed using a Weibull distribution to ensure that the difference between the simulated strength dispersion coefficient and the measured value from a single-particle compression experiment is less than 5%. The spatiotemporal evolution of the damage variable is simulated and calculated using a phase-field algorithm based on the initial fracture strength. The formula for the phase-field algorithm is as follows:

[0021] ;

[0022] in The value is a damage variable, with 0 representing a complete lattice and 1 representing complete damage. For time, For phase field mobility, For phase field free energy functional;

[0023] The dynamic fracture strength is calculated based on the updated damage variables, and the formula for calculating the dynamic fracture strength is as follows:

[0024] ;

[0025] in For patch At any moment Dynamic fracture strength, The characteristic width of the damaged area. The Laplace operator for phase field variables.

[0026] Furthermore, the method for reducing energy consumption during crushing includes:

[0027] Based on the patch, dynamic fracture strength, damage variables, patch volume and patch mass are obtained. Damage residual variables are obtained based on the damage variables. The 80th percentile of the equivalent diameter is used as the initial equivalent particle size based on the patch volume. The initial equivalent particle size is multiplied by the damage residual variables to obtain the predicted particle size after breakage.

[0028] Based on the mineral phase, the corresponding grinding parameters, such as rotation speed, ball-to-particle ratio, and grinding time, are obtained. The Bond work index is calibrated using the grinding parameters. Based on the Bond work index, initial equivalent particle size, and predicted particle size after crushing, the basic energy consumption of the patch is calculated using the Bond third crushing theory. The ratio of the dynamic fracture strength to the preset benchmark strength is calculated. The product of the ratio, the remaining damage variable, and the basic energy consumption is taken as the real-time crushing energy consumption. The real-time crushing energy consumption of the patch is integrated over time during the grinding period to obtain the total cumulative energy consumption of the patch. The total cumulative energy consumption is divided by the mass of the patch to obtain the crushing energy consumption per unit mass.

[0029] Furthermore, the method for obtaining the aforementioned grinding parameter combination and activation effect includes:

[0030] Grinding parameters and lattice states are obtained based on kaolinite patches. The sequence of grinding parameters and lattice states are used as input samples. Grinding process features and lattice structure features are extracted through a dual-branch attention network. The baseline activity is used as the ground truth label for the activation effect. Based on the grinding process features and lattice structure features, the predicted value of the activation effect and the lattice state vector are output through the first output branch of a multilayer perceptron. The activation effect is calculated using the following formula:

[0031] ;

[0032] in This is a predicted value for the activation effect. It is a multilayer perceptron. The total number of feature pairs. For feature pair index, Grinding parameter characteristics Features of lattice states mutual information, It is the sum of the mutual information of all feature pairs. This is the weight matrix for the grinding parameter branch. This is the weight matrix for the lattice state branches. For element-wise product;

[0033] The grinding parameter combination is output based on the second output branch of the multilayer perceptron, and the grinding parameter combination corresponds to the output of the first output branch.

[0034] Furthermore, the method for obtaining the mutual information includes:

[0035] A dual-branch input feature matrix is ​​constructed based on the input samples. Features are extracted through a parallel feature extraction network based on the dual-branch input feature matrix. The grinding parameter branch is nonlinearly mapped through a 3-layer fully connected network to obtain the hidden features of rotation speed, time and energy consumption.

[0036] The spatiotemporal features of the lattice state branches are extracted using a convolutional neural network with skip connections to obtain the spatiotemporal evolution hidden features of the damage variables. Based on the hidden features of the two branches, the joint probability distribution and marginal distribution are calculated using a probability distribution estimation algorithm, and mutual information is obtained based on the joint probability distribution and marginal distribution.

[0037] The grinding parameter branch features of the dual-branch input feature matrix include grinding parameters, crushing energy consumption, and equivalent thermodynamic temperature, while the lattice state branch features include lattice structure parameters, damage variables, and dynamic fracture strength.

[0038] Furthermore, the method for obtaining the optimal solution set includes:

[0039] The grinding parameter combination, activation effect, crushing energy consumption and patch quantity are obtained. The energy consumption cost is calculated based on the crushing energy consumption per unit mass and the number of patches. The optimization objectives are to maximize the predicted value of activation effect, minimize energy consumption cost and minimize crushing energy consumption per unit mass. The reference activity and the maximum value of crushing energy consumption are used as constraints. Solutions that are lower than the reference activity or exceed the upper limit of crushing energy consumption are constrained and marked.

[0040] Based on the optimization objective number, a reference point set is generated using the simplex lattice method. The grinding parameter combination is used as the initial strategy set. The population is iteratively optimized using a multi-objective evolutionary algorithm based on the initial strategy set. The calculation formula for the multi-objective evolutionary algorithm is as follows:

[0041] ;

[0042] in The objective function value for multi-objective optimization. Let be the decision variable, representing the combination of grinding parameters. Represents the set of reference points Each reference point in Find the minimum value. For the first One objective function, For reference point The One portion, For the first The range of the objective function For the first The penalty weight of each constraint, For the first Constraint functions, For the first The threshold of each constraint;

[0043] The Pareto optimal solution set is obtained by using the fast non-dominated sorting algorithm based on the objective function value, thus obtaining the optimal solution set for the grinding parameters.

[0044] Furthermore, the method for obtaining the Pareto optimal solution set includes:

[0045] Pareto rank is constructed using the fast non-dominated sorting algorithm based on the objective function value. The vertical distance from individuals of the same rank to the reference point is calculated sequentially based on the population, and individuals are associated with the nearest reference point. If a reference point is associated with multiple individuals, the individual closer to the reference point is retained first. If a reference point is not associated with any individuals, the closest individual is selected from the population individuals associated with other reference points for re-association, thus completing the environment selection.

[0046] Individuals that violate the constraints are marked as having a constraint dominance level and their selection probability is reduced in the environmental selection process. Individuals that satisfy all constraints are marked as feasible solutions and given priority to enter the next generation of the population. The best individual in each generation is directly copied to the offspring.

[0047] Genetic operations are performed on the selected parent individuals using simulated binary crossover and polynomial mutation operators to generate offspring populations with crossover probabilities greater than or equal to 0.9 and mutation probabilities greater than or equal to 0.1. The offspring are merged with the parents and iterated again until the rate of change of the frontier hypervolume index is less than 0.1%, at which point the Pareto optimal solution set is obtained. The Pareto optimal solution set is sorted using a reference point distance minimization function. Based on the sorting, a comprehensive score is calculated for feasible solutions to the normalized ideal point, and the top 3 to 5 grinding parameter combinations are selected as the optimal solution set.

[0048] Further, a method for obtaining the coal gangue grinding particles includes:

[0049] Based on the application scenario, the mineral phase of the target coal gangue is obtained. The optimal solution set is screened according to the benchmark activity of the mineral phase in the target coal gangue to obtain the targeted grinding parameters of the target coal gangue. The coal gangue is then targeted-ground according to the rotation speed, ball-to-material ratio and grinding time of the targeted grinding parameters to obtain activated and enhanced coal gangue grinding particles.

[0050] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects:

[0051] This invention utilizes multi-dimensional characterization data of coal gangue for heterogeneous modeling, employs a patching algorithm to simulate mineral distribution and mechanical properties, provides a data foundation for activation effects through the microscopic damage evolution of coal gangue particles, performs feature mapping between grinding parameters and patch lattice states using a dual-branch attention network, strengthens feature associations based on mutual information to obtain predicted values ​​of grinding parameter combinations and activation effects, optimizes activation effects, energy consumption costs, and crushing energy consumption through a non-dominated sorting genetic algorithm, and selects targeted grinding parameters according to application scenarios to improve decision-making efficiency and accuracy, thereby enhancing the specificity of coal gangue grinding effects and reducing grinding energy consumption and processing costs. Attached Figure Description

[0052] Figure 1 is a flowchart of the steps of a method for mechanical grinding and enhancing activation of coal gangue according to an embodiment of the present invention. Detailed Implementation

[0053] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0054] Referring to Figure 1, the present invention provides a method for mechanically grinding and enhancing the activation of coal gangue, comprising:

[0055] Multidimensional characterization data of coal gangue is obtained, and the multidimensional characterization data is preprocessed, including three-dimensional particle size, mineral phase and reference activity.

[0056] In the actual assessment, coal gangue from a mining area was selected as the research object. The mining area produces approximately 960,000 tons of coal gangue annually, with a clear mineral phase composition and an urgent need for resource utilization. Fresh coal gangue from underground was selected, and after removing visible impurities, it was crushed to a particle size ≤50mm. The sample was then reduced to 1kg using the quartering method and used as a characterization sample. Three parallel samples were tested using a laser particle size analyzer, and the average value of the results was taken to obtain the particle size distribution as D. 10 =12.4μm, D 50 =48.7μm, D 90=186.3μm, of which the particle size distribution is as follows: 0~20μm particles account for 15.7%, 20~100μm particles account for 58.3%, 100~200μm particles account for 22.1%, and particles larger than 200μm account for 3.9%. This is consistent with the natural gradation characteristics of coal gangue. The proportion of large particles is low, so excessive crushing is not necessary. Quantitative analysis of the coal gangue sample was performed using the Rietveld refinement method with XRD diffraction. The scanning range was 5°~80°, the step size was 0.02°, and the scanning speed was 4° / min. The mass fractions were obtained, including kaolinite 82.4%, quartz 18.7%, illite 3.4%, pyrite 1.7%, siderite 0.3%, talc 0.7%, and the interplanar spacing d of kaolinite. 001 =7.14Å, interplanar spacing of quartz crystals d 101 =3.34Å;

[0057] Using the strength index method with 42.5 grade ordinary Portland cement as the benchmark, coal gangue was ground into stone powder to prepare benchmark mortar and its compressive strength was tested. The compressive strength of the benchmark mortar after 28 days was 48.6 MPa, and the compressive strength of the test mortar was 33.1 MPa. The benchmark activity index was calculated to be 68.3%, which is greater than the minimum activity requirement of 65% for cement admixtures. The loss on ignition was 5.8%, and the SO3 content was 1.2%. Multi-dimensional characterization data of coal gangue were obtained.

[0058] Based on the multi-dimensional characterization data, a patch algorithm is used to model the heterogeneity of coal gangue particles, and the lattice structure of the mineral phase is used to assign fracture strength and elastic modulus to each patch.

[0059] In practical evaluation, based on three-dimensional granularity D 10 =12.4μm, D 50 =48.7μm, D 90 =186.3μm, irregular coal gangue particles were reconstructed using EDEM software. The Fourier series fitting accuracy of the first 10 harmonics was >95%. The reconstructed model was voxelized with a voxel resolution of 7.14pm, which is 1 / 100 of the kaolinite crystal layer thickness. The generated voxel model contained 2.3×106 voxels, with kaolinite accounting for 82.4% and quartz 18.7%. The deviation from the XRD quantitative results was <2%. Mineral phases were assigned using a multi-point geostatistical method, where kaolinite was distributed in layers and quartz was massively embedded. Voxels were aggregated using a region growing algorithm, and the average size of the generated kaolinite patch was 20×15×8μm. 3 The average size of the generated quartz patches is 30×25×12μm. 3 The interface region is constructed with a three-layer coupled structure consisting of a patch, an interface, and another patch, using cohesive units. The interfacial bonding energy between the kaolinite and quartz interfaces in this coupled structure is 0.8 J / m. 2 ;

[0060] Based on the lattice structure of the patches and mineral phases, the interplanar spacing, lattice shear modulus, surface energy, dislocation density, and Poisson's ratio were obtained, and the initial fracture strength of each patch was calculated. The kaolinite patch had an interplanar spacing of 7.14 Å, a lattice shear modulus of 15 GPa, and a surface energy of 1.2 J / m². 2 Dislocation density 1.2 × 10 12 m -2 With a Poisson's ratio of 0.25, the initial fracture strength was 120 MPa and the elastic modulus was 100 GPa. The interplanar spacing of quartz was 3.34 and the lattice shear modulus was 30 GPa, resulting in an initial fracture strength of 350 MPa and an elastic modulus of 105 GPa. The differences between these values ​​and the measured values ​​from the single-particle compression experiment were 118 ± 5 MPa for kaolinite and 345 ± 10 MPa for quartz, with differences of less than 5%. The spatiotemporal evolution of damage variables was simulated using the phase-field algorithm. After 30 minutes of grinding, the damage variables of the kaolinite patch were 0.2 and those of the quartz patch were 0.05. Substituting these values ​​into the calculation of the dynamic fracture strength, the dynamic fracture strength of kaolinite at this time was 100 MPa and that of quartz was 320 MPa.

[0061] The energy consumption for crushing each patch is calculated based on the grinding parameters of the mineral phase. For highly active patches, a dual-branch attention network is used to learn the feature mapping of grinding parameters and lattice state to obtain the combination of grinding parameters and activation effect.

[0062] In practical assessments, dynamic fracture strength, damage variables, patch volume, and patch mass are obtained based on the patch. Damage residual variables are obtained from the damage variables. The 80th percentile of the equivalent diameter is used as the initial equivalent grain size based on the patch volume. Therefore, the kaolinite patch volume V... patch =20×15×8=2400μm 3 The predicted particle size after fragmentation is obtained by multiplying the initial equivalent particle size by the residual damage variable, where the initial equivalent particle size P of the kaolinite patch is... 80,in =24.5μm, P after fragmentation 80,out =12.8μm, based on the mineral phase, the corresponding grinding parameters, such as rotation speed, ball-to-particle ratio, and grinding time, are obtained. The Bond work index energy consumption is calibrated using the grinding parameters, then E Bond =6.9kWh / t, E after real-time correction break =110.4kWh / t, energy consumption per unit mass is 55.2kWh / t, and the initial equivalent particle size P of the quartz patch is... 80,in =30μm, P after fragmentation 80,out =18μm, corresponding to E break =235.6kWh / t, which is 4.26 times that of kaolinite;

[0063] Grinding parameters and lattice states were obtained based on kaolinite patches. The sequence of grinding parameters and lattice states were used as input samples. The grinding parameter sequence consisted of a rotation speed of 100–300 r / min with an interval of 20 r / min, a ball-to-material ratio of 1:1–5:1 with an interval of 1, and a grinding time of 10–120 min with an interval of 10 min, generating a total of 660 sets of parameters. In the lattice state, the dynamic fracture strength of the kaolinite patch was 80–120 MPa, the damage variable was 0–0.5, and the interplanar spacing was 7.14 Å, generating 660 sets of corresponding data. True values ​​were generated based on a baseline activity range of 68.3%–92.5%. Grinding process features and lattice structure features were extracted using a dual-branch attention network, where the grinding parameter branch was a 3-layer fully connected network. A CNN with 28-256-128 neurons was activated using the ReLU function. 128-dimensional hidden features of rotational speed, time, and energy consumption were extracted. The lattice state branch was a CNN with skip connections, including 3 convolutional layers and 2 pooling layers. 64-dimensional spatiotemporal features of damage variables and interplanar spacing were extracted. Kernel density estimation was used to calculate the mutual information of feature pairs. In kaolinite parameters, the MI for lattice pairs was 0.85, and for quartz pairs, it was 0.32. The baseline activity was used as the ground truth label for the activation effect. Based on the grinding process features and lattice structure features, the first output branch of a multilayer perceptron outputs the predicted value of the activation effect and the lattice state vector. The second output branch of the multilayer perceptron outputs the grinding parameter combination. After 30 training epochs, the MSE decreased to 0.012. The test set R... 2 =0.92, the optimal parameters are rotation speed 240 r / min, ball-to-material ratio 3:1, and grinding time 60 min, the predicted activation effect is 91.2%, and the measured value is 89.7%;

[0064] The grinding parameters are combined as a strategy set, and the activation effect, energy consumption cost and crushing energy consumption are taken as optimization objectives. The optimization objectives are solved by a non-dominated sorting genetic algorithm using the strategy set to obtain the optimal solution set of grinding parameters.

[0065] In practical evaluation, the grinding parameter combination, activation effect, crushing energy consumption, and patch quantity are obtained. Energy cost is calculated based on the crushing energy consumption per unit mass and the number of patches. Due to the complex composition of coal gangue, activation requires more refined modeling and grinding parameters, resulting in higher energy costs. Maximizing the predicted activation effect, minimizing energy cost, and minimizing crushing energy consumption per unit mass are taken as optimization objectives, where f1(x) = activation effect (maximization, target ≥ 90%), f2(x) = energy cost (minimization, target ≥ 90%), and f2(x) = energy cost (minimization, target ≥ 90%). (Target ≤ Patch Modeling Fineness Threshold), f3(x) = Crushing Energy Consumption (Minimize, Target ≤ 60kWh / t), Constraints: Baseline Activity ≥ 68.3%, Crushing Energy Consumption ≤ 80kWh / t, Rotation Speed ​​≤ 300r / min. Based on the number of optimization objectives, a set of reference points is generated using the simplex lattice method. The initial population is 200 individuals, with 15 reference points. The Pareto optimal solution set is solved using the fast non-dominated sorting algorithm. The population is divided into 3 layers of Pareto fronts. Individuals in the first layer have no dominance relationship. The vertical distance from an individual to a reference point is calculated. The average distance is <0.1. In environmental selection, 200 optimal individuals are retained for the next generation. Binary crossover is simulated to generate offspring with a crossover probability of 0.9 and a mutation probability of 0.1. After 100 iterations, the rate of change of the leading hypervolume index is <0.1%, indicating algorithm convergence. Based on the Pareto optimal solution set, the top three grinding parameter combinations are selected as the optimal solution set. Combination 1 consists of a rotation speed of 240 r / min, a ball-to-material ratio of 3:1, and a grinding time of 60 min, corresponding to an activation effect of 91.2% and an energy cost of 78.5. (The last part about crushing is incomplete and likely refers to a separate process.) The first combination has an energy consumption of 55.2 kWh / t, a high degree of activation, and is suitable for high-activity demand scenarios. Combination 2 has a rotation speed of 260 r / min, a ball-to-material ratio of 2:1, and a grinding time of 70 min, with an activation effect of 88.7% and an energy cost of 72.3. The crushing energy consumption is 52.1 kWh / t, which is the lowest energy cost. Combination 3 has a rotation speed of 220 r / min, a ball-to-material ratio of 4:1, and a grinding time of 50 min, with an activation effect of 77.5% and an energy cost of 62.3. The crushing energy consumption is 48.3 kWh / t, which is the lowest energy consumption and is suitable for environmentally friendly scenarios.

[0066] Based on the optimal solution set, the target grinding parameters are selected by mineral phase screening in the application scenario to perform targeted grinding of coal gangue, thereby obtaining activated and enhanced coal gangue grinding particles.

[0067] In practical evaluations, within the cement admixture scenario, the target coal gangue's mineral phases are 80.2% kaolinite and 16.7% quartz, with a baseline activity of 78.3%. This scenario requires high pozzolanic activity and a suitable cement particle size distribution. The grinding equipment is a Φ2.2×7m ball mill with steel ball gradation. With a mass ratio of 1:2:3, a rotation speed of 240 r / min and a ball-to-material ratio of 3:1 were selected for combination 1. The grinding time was 60 min and the throughput was 15 t / h. Activated and enhanced coal gangue grinding particles were obtained. XRD analysis showed that the characteristic peak of kaolinite in the coal gangue grinding particles was 2θ=12.3°, the strength decreased by 35%, indicating lattice enhancement and activation. The absence of significant changes in the characteristic peak of quartz indicated that the particles were not over-crushed. The first-grade product had an activity index of 90.5% after 28 days. Compared with coarse grinding, the energy consumption was reduced by 22 kWh / t, and the media loss was reduced by 1.2 kg / t.

[0068] In this embodiment, the method for obtaining the patch includes:

[0069] Based on the distribution data of three-dimensional particle size in multi-dimensional characterization data, the irregular three-dimensional geometric shape of coal gangue particles is reconstructed by the discrete element method. The particle outline of the irregular three-dimensional geometric shape is represented as a Fourier series. The reconstructed model is voxelized and the voxel resolution is set to 1 / 50 to 1 / 100 of the minimum feature size of the particle. The relative position coordinates and volume ratio of each voxel are recorded according to the coordinates between voxels.

[0070] Based on the spatial distribution data of mineral phases, the main mineral phases are spatially assigned in the voxel model using multi-point geostatistical methods. The same mineral phase type and a difference in elastic modulus of less than 10% between voxels are used as the physical property similarity condition. Based on the volume fraction of mineral phases and the physical property similarity condition, voxel aggregation is performed using a region growth algorithm. If the difference in mineral phase composition after aggregation of adjacent voxels is less than 15%, a heterogeneous patch is obtained. The kaolinite patch in the heterogeneous patch is used as a high-activity region, and the quartz patch is used as a high-hardness region to obtain mineral phase labels.

[0071] The interface region between adjacent patches is obtained. Based on the interface region, a three-layer coupled structure of patch, interface and patch is generated through cohesive unit according to the relationship between the interface bonding energy of mineral phase and crystallographic orientation. This results in the heterogeneous modeling of coal gangue particles. The main mineral phases include quartz, kaolinite and illite.

[0072] In this embodiment, the method for obtaining the fracture strength and elastic modulus includes:

[0073] Based on the lattice structure of the patches and mineral phases, the interplanar spacing, lattice shear modulus, surface energy, dislocation density, and Poisson's ratio are obtained, and the initial fracture strength of each patch is calculated. The formula for calculating the initial fracture strength is as follows:

[0074] ;

[0075] in For patch Initial fracture strength, in MPa. This refers to the lattice shear modulus, expressed in GPa. Interplanar spacing, in nm. For surface energy, For dislocation density, Boltzmann's constant, This is the equivalent thermodynamic temperature during grinding, obtained from the grinding parameters of the mineral phase. The patch feature size is obtained from the voxel volume;

[0076] The elastic modulus is calculated using the continuum mechanics formula based on the lattice shear modulus and Poisson's ratio. Each patch is then assigned a value using the elastic modulus and initial fracture strength. The fracture strength and elastic modulus of each patch are randomly perturbed using a Weibull distribution to ensure that the difference between the simulated strength dispersion coefficient and the measured value from a single-particle compression experiment is less than 5%. The spatiotemporal evolution of the damage variable is simulated and calculated using a phase-field algorithm based on the initial fracture strength. The formula for the phase-field algorithm is as follows:

[0077] ;

[0078] in The value is a damage variable, with 0 representing a complete lattice and 1 representing complete damage. For time, For phase field mobility, For phase field free energy functional;

[0079] The dynamic fracture strength is calculated based on the updated damage variables, and the formula for calculating the dynamic fracture strength is as follows:

[0080] ;

[0081] in For patch At any moment Dynamic fracture strength, The characteristic width of the damaged area. The Laplace operator for phase field variables.

[0082] In this embodiment, the method for reducing crushing energy consumption includes:

[0083] Based on the patch, dynamic fracture strength, damage variables, patch volume and patch mass are obtained. Damage residual variables are obtained based on the damage variables. The 80th percentile of the equivalent diameter is used as the initial equivalent particle size based on the patch volume. The initial equivalent particle size is multiplied by the damage residual variables to obtain the predicted particle size after breakage.

[0084] Based on the mineral phase, the corresponding grinding parameters, such as rotation speed, ball-to-particle ratio, and grinding time, are obtained. The Bond work index is calibrated using the grinding parameters. Based on the Bond work index, initial equivalent particle size, and predicted particle size after crushing, the basic energy consumption of the patch is calculated using the Bond third crushing theory. The ratio of the dynamic fracture strength to the preset benchmark strength is calculated. The product of the ratio, the remaining damage variable, and the basic energy consumption is taken as the real-time crushing energy consumption. The real-time crushing energy consumption of the patch is integrated over time during the grinding period to obtain the total cumulative energy consumption of the patch. The total cumulative energy consumption is divided by the mass of the patch to obtain the crushing energy consumption per unit mass.

[0085] In this embodiment, the method for obtaining the grinding parameter combination and activation effect includes:

[0086] Grinding parameters and lattice states are obtained based on kaolinite patches. The sequence of grinding parameters and lattice states are used as input samples. Grinding process features and lattice structure features are extracted through a dual-branch attention network. The baseline activity is used as the ground truth label for the activation effect. Based on the grinding process features and lattice structure features, the predicted value of the activation effect and the lattice state vector are output through the first output branch of a multilayer perceptron. The activation effect is calculated using the following formula:

[0087] ;

[0088] in This is a predicted value for the activation effect. It is a multilayer perceptron. The total number of feature pairs. For feature pair index, Grinding parameter characteristics Features of lattice states mutual information, It is the sum of the mutual information of all feature pairs. This is the weight matrix for the grinding parameter branch. This is the weight matrix for the lattice state branches. For element-wise product;

[0089] The grinding parameter combination is output based on the second output branch of the multilayer perceptron, and the grinding parameter combination corresponds to the output of the first output branch.

[0090] In this embodiment, the method for obtaining the mutual information includes:

[0091] A dual-branch input feature matrix is ​​constructed based on the input samples. Features are extracted through a parallel feature extraction network based on the dual-branch input feature matrix. The grinding parameter branch is nonlinearly mapped through a 3-layer fully connected network to obtain the hidden features of rotation speed, time and energy consumption.

[0092] The spatiotemporal features of the lattice state branches are extracted using a convolutional neural network with skip connections to obtain the spatiotemporal evolution hidden features of the damage variables. Based on the hidden features of the two branches, the joint probability distribution and marginal distribution are calculated using a probability distribution estimation algorithm, and mutual information is obtained based on the joint probability distribution and marginal distribution.

[0093] The grinding parameter branch features of the dual-branch input feature matrix include grinding parameters, crushing energy consumption, and equivalent thermodynamic temperature, while the lattice state branch features include lattice structure parameters, damage variables, and dynamic fracture strength.

[0094] In this embodiment, the method for obtaining the optimal solution set includes:

[0095] The grinding parameter combination, activation effect, crushing energy consumption and patch quantity are obtained. The energy consumption cost is calculated based on the crushing energy consumption per unit mass and the number of patches. The optimization objectives are to maximize the predicted value of activation effect, minimize energy consumption cost and minimize crushing energy consumption per unit mass. The reference activity and the maximum value of crushing energy consumption are used as constraints. Solutions that are lower than the reference activity or exceed the upper limit of crushing energy consumption are constrained and marked.

[0096] Based on the optimization objective number, a reference point set is generated using the simplex lattice method. The grinding parameter combination is used as the initial strategy set. The population is iteratively optimized using a multi-objective evolutionary algorithm based on the initial strategy set. The calculation formula for the multi-objective evolutionary algorithm is as follows:

[0097] ;

[0098] in The objective function value for multi-objective optimization. Let be the decision variable, representing the combination of grinding parameters. Represents the set of reference points Each reference point in Find the minimum value. For the first One objective function, For reference point The One portion, For the first The range of the objective function For the first The penalty weight of each constraint, For the first Constraint functions, For the first The threshold of each constraint;

[0099] The Pareto optimal solution set is obtained by using the fast non-dominated sorting algorithm based on the objective function value, thus obtaining the optimal solution set for the grinding parameters.

[0100] In this embodiment, the method for obtaining the Pareto optimal solution set includes:

[0101] Pareto rank is constructed using the fast non-dominated sorting algorithm based on the objective function value. The vertical distance from individuals of the same rank to the reference point is calculated sequentially based on the population, and individuals are associated with the nearest reference point. If a reference point is associated with multiple individuals, the individual closer to the reference point is retained first. If a reference point is not associated with any individuals, the closest individual is selected from the population individuals associated with other reference points for re-association, thus completing the environment selection.

[0102] Individuals that violate the constraints are marked as having a constraint dominance level and their selection probability is reduced in the environmental selection process. Individuals that satisfy all constraints are marked as feasible solutions and given priority to enter the next generation of the population. The best individual in each generation is directly copied to the offspring.

[0103] Genetic operations are performed on the selected parent individuals using simulated binary crossover and polynomial mutation operators to generate offspring populations with crossover probabilities greater than or equal to 0.9 and mutation probabilities greater than or equal to 0.1. The offspring are merged with the parents and iterated again until the rate of change of the frontier hypervolume index is less than 0.1%, at which point the Pareto optimal solution set is obtained. The Pareto optimal solution set is sorted using a reference point distance minimization function. Based on the sorting, a comprehensive score is calculated for feasible solutions to the normalized ideal point, and the top 3 to 5 grinding parameter combinations are selected as the optimal solution set.

[0104] In this embodiment, the method for obtaining the coal gangue grinding particles includes:

[0105] Based on the application scenario, the mineral phase of the target coal gangue is obtained. The optimal solution set is screened according to the benchmark activity of the mineral phase in the target coal gangue to obtain the targeted grinding parameters of the target coal gangue. The coal gangue is then targeted-ground according to the rotation speed, ball-to-material ratio and grinding time of the targeted grinding parameters to obtain activated and enhanced coal gangue grinding particles.

[0106] The above description is merely an example and illustration of the structure of the present invention. Those skilled in the art can make various modifications or additions to the specific embodiments described, or use similar methods to replace them, as long as they do not deviate from the structure of the invention or exceed the scope defined in the claims, all of which should fall within the protection scope of the present invention.

Claims

1. A method for mechanically grinding and enhancing the activation of coal gangue, characterized in that, Includes the following steps: Multidimensional characterization data of coal gangue is obtained, and the multidimensional characterization data is preprocessed, including three-dimensional particle size, mineral phase and reference activity. Based on the distribution data of three-dimensional particle size in multi-dimensional characterization data, the irregular three-dimensional geometric shape of coal gangue particles is reconstructed using the discrete element method. The particle outline of the irregular three-dimensional geometric shape is represented as a Fourier series. The reconstructed model is voxelized, and the voxel resolution is set to 1 / 50 to 1 / 100 of the smallest feature size of the particles. The relative position coordinates and volume percentage of each voxel are recorded according to the coordinates between voxels. Based on the spatial distribution data of mineral phases, the main mineral phases are spatially assigned in the voxel model using multi-point geostatistical methods. The same mineral phase type and an elastic modulus difference of less than 10% between voxels are used as the physical property similarity condition. Based on the volume fraction of mineral phases and the physical property similarity condition, voxel aggregation is performed using a region growing algorithm. If the difference in mineral phase composition after aggregation of adjacent voxels is less than 15%, a heterogeneous patch is obtained. The kaolinite patch in the heterogeneous patch is designated as a high-activity region, and the quartz patch is designated as a high-hardness region, and mineral phase labels are obtained. In the interface region between patches, a three-layer coupled structure of patches, interfaces, and patches is generated through cohesive units based on the interface bonding energy and crystallographic orientation relationship of the mineral phases. This obtains a heterogeneous model of coal gangue particles. The main mineral phases include quartz, kaolinite, and illite. The fracture strength and elastic modulus of each patch are assigned using the crystal structure of the mineral phases. The crushing energy consumption of each patch is calculated based on the grinding parameters of the mineral phases. For highly active patches, a dual-branch attention network is used to learn the feature mapping of grinding parameters and crystal states to obtain the combination of grinding parameters and activation effect. The combination of grinding parameters is used as a strategy set, and the activation effect, energy consumption cost, and crushing energy consumption are used as optimization objectives. The strategy set is used to solve the optimization objectives through a non-dominated sorting genetic algorithm to obtain the optimal solution set of grinding parameters. Based on the optimal solution set, targeted grinding parameters are selected according to the mineral phase of the application scenario, and the coal gangue is targeted to be ground to obtain activated and enhanced coal gangue grinding particles.

2. The method for mechanical grinding and activation of coal gangue according to claim 1, characterized in that, The method for obtaining the fracture strength and elastic modulus includes: obtaining the interplanar spacing, lattice shear modulus, surface energy, dislocation density, and Poisson's ratio based on the lattice structure of the patch and mineral phase, and calculating the initial fracture strength of each patch, wherein the initial fracture strength is calculated using the following formula: ;in For patch Initial fracture strength, in MPa. This refers to the lattice shear modulus, expressed in GPa. Interplanar spacing, in nm. For surface energy, For dislocation density, Boltzmann's constant, This is the equivalent thermodynamic temperature during grinding, obtained from the grinding parameters of the mineral phase. The patch feature size is obtained from the voxel volume. The elastic modulus is calculated using the continuum mechanics formula based on the lattice shear modulus and Poisson's ratio. Each patch is assigned a value using the elastic modulus and initial fracture strength. The fracture strength and elastic modulus of each patch are randomly perturbed using a Weibull distribution to ensure that the difference between the simulated strength dispersion coefficient and the measured value from the single-particle compression experiment is less than 5%. The spatiotemporal evolution of the damage variable is simulated and calculated using the phase-field algorithm based on the initial fracture strength. The formula for the phase-field algorithm is: ;in The value is a damage variable, with 0 representing a complete lattice and 1 representing complete damage. For time, For phase field mobility, The phase field free energy functional is used; the dynamic fracture strength is calculated based on the updated damage variables, and the formula for calculating the dynamic fracture strength is: ;in For patch At any moment Dynamic fracture strength, The characteristic width of the damaged area. The Laplace operator for phase field variables.

3. The method for mechanically grinding and enhancing the activation of coal gangue according to claim 1, characterized in that, The method for calculating crushing energy consumption includes: obtaining dynamic fracture strength, damage variables, patch volume, and patch mass based on the patch; obtaining residual damage variables based on the damage variables; using the 80th percentile of the equivalent diameter as the initial equivalent particle size based on the patch volume; multiplying the initial equivalent particle size by the residual damage variables to obtain the predicted particle size after crushing; obtaining the rotational speed, ball-to-particle ratio, and grinding time of the corresponding grinding parameters based on the mineral phase; calibrating the Bond work index using the grinding parameters; calculating the basic energy consumption of the patch using the Bond work index, the initial equivalent particle size, and the predicted particle size after crushing through the Bond third crushing theory; calculating the ratio of the dynamic fracture strength to the preset benchmark strength; multiplying the ratio by the residual damage variables and the basic energy consumption as the real-time crushing energy consumption; integrating the real-time crushing energy consumption of the patch over the grinding time to obtain the total cumulative energy consumption of the patch; and dividing the total cumulative energy consumption by the patch mass to obtain the crushing energy consumption per unit mass.

4. The method for mechanical grinding and enhancing activation of coal gangue according to claim 1, characterized in that, The method for obtaining the grinding parameter combination and activation effect includes: acquiring grinding parameters and lattice states based on kaolinite patches; using the sequence of grinding parameters and lattice states as input samples; extracting grinding process features and lattice structure features through a dual-branch attention network; using the baseline activity as the ground truth label for the activation effect; and outputting the predicted value of the activation effect and the lattice state vector through the first output branch of a multilayer perceptron based on the grinding process features and lattice structure features. The activation effect is calculated using the following formula: ;in This is a predicted value for the activation effect. It is a multilayer perceptron. The total number of feature pairs. For feature pair index, Grinding parameter characteristics Features of lattice states mutual information, It is the sum of the mutual information of all feature pairs. This is the weight matrix for the grinding parameter branch. This is the weight matrix for the lattice state branches. The process involves element-wise multiplication; the grinding parameter combination is output based on the second output branch of the multilayer perceptron, and the grinding parameter combination corresponds to the output of the first output branch.

5. The method for mechanically grinding and enhancing the activation of coal gangue according to claim 4, characterized in that, The method for obtaining the mutual information includes: constructing a dual-branch input feature matrix based on the input samples; extracting features from the dual-branch input feature matrix using a parallel feature extraction network; performing nonlinear mapping on the grinding parameter branch using a 3-layer fully connected network to obtain hidden features of rotational speed, time, and energy consumption; extracting spatiotemporal features from the lattice state branch using a convolutional neural network with skip connections to obtain hidden features of the spatiotemporal evolution of damage variables; calculating the joint probability distribution and marginal distribution based on the hidden features of the dual branches using a probability distribution estimation algorithm; and calculating the mutual information based on the joint probability distribution and marginal distribution. The grinding parameter branch features of the dual-branch input feature matrix include grinding parameters, crushing energy consumption, and equivalent thermodynamic temperature; the lattice state branch features include lattice structure parameters, damage variables, and dynamic fracture strength.

6. The method for mechanically grinding and enhancing the activation of coal gangue according to claim 1, characterized in that, The method for obtaining the optimal solution set includes: acquiring the grinding parameter combination, activation effect, crushing energy consumption, and patch quantity; calculating the energy consumption cost based on the crushing energy consumption per unit mass and the number of patches; taking the maximization of the predicted activation effect, the minimization of energy consumption cost, and the minimization of crushing energy consumption per unit mass as optimization objectives; using the baseline activity and the maximum crushing energy consumption as constraints; and marking constraints on solutions that are below the baseline activity or exceed the upper limit of crushing energy consumption. Based on the number of optimization objectives, a reference point set is generated using the simplex lattice method; the grinding parameter combination is used as the initial strategy set; and the population is iteratively optimized using a multi-objective evolutionary algorithm based on the initial strategy set. The multi-objective evolutionary algorithm calculation formula is as follows: ;in The objective function value for multi-objective optimization. Let be the decision variable, representing the combination of grinding parameters. Represents the set of reference points Each reference point in Find the minimum value. For the first One objective function, For reference point The One portion, For the first The range of the objective function For the first The penalty weight of each constraint, For the first Constraint functions, For the first The threshold of each constraint is determined; the Pareto optimal solution set is obtained by using the fast non-dominated sorting algorithm based on the objective function value, thus obtaining the optimal solution set of grinding parameters.

7. The method for mechanical grinding and activation of coal gangue according to claim 6, characterized in that, The method for obtaining the Pareto optimal solution set includes: constructing Pareto ranks using a fast non-dominated sorting algorithm based on the objective function value; calculating the vertical distance from individuals of the same rank to a reference point based on the population, and associating individuals with the nearest reference point; if a reference point is associated with multiple individuals, prioritizing the retention of individuals closer to the reference point; if a reference point is not associated with any individuals, selecting the closest individuals from the population individuals associated with other reference points for re-association, thus completing the environment selection; marking individuals that violate constraints as having a constraint dominance rank and reducing their selection probability in the environment selection; and marking individuals that satisfy all constraints as having a dominance rank. The optimal solution is selected and prioritized for the next generation. The best individual in each generation is directly copied to the offspring. The selected parent individuals are genetically modified using a simulated binary crossover operator and a polynomial mutation operator to generate an offspring population with a crossover probability greater than or equal to 0.9 and a mutation probability greater than or equal to 0.

1. The offspring and parent individuals are merged and iterated again until the rate of change of the frontier hypervolume index is less than 0.1%, at which point the Pareto optimal solution set is obtained. The Pareto optimal solution set is sorted using a reference point distance minimization function. Based on the sorting, a comprehensive score from the feasible solution to the normalized ideal point is calculated, and the top 3 to 5 grinding parameter combinations are selected as the optimal solution set.

8. The method for mechanical grinding and activation of coal gangue according to claim 1, characterized in that, The method for obtaining the coal gangue grinding particles includes: obtaining the mineral phase of the target coal gangue based on the application scenario; screening the optimal solution set according to the benchmark activity of the mineral phase in the target coal gangue; obtaining the targeted grinding parameters of the target coal gangue; and performing targeted grinding of the coal gangue according to the rotation speed, ball-to-material ratio and grinding time of the targeted grinding parameters to obtain activated and enhanced coal gangue grinding particles.

Images