A tailings particle classification method based on multi-dimensional morphological feature compression and clustering

Abstract

This invention provides a tailings particle classification method based on multi-dimensional morphological feature compression and clustering, belonging to the fields of geotechnical engineering, mineral processing, and particle classification technology. The method includes: S1-S6: acquiring an STL file database of tailings particles; S7: calculating and extracting morphological parameters to form an original parameter dataset; S8-S10: obtaining a three-dimensional morphological parameter set retaining the original morphological feature information; S11-S12: determining the optimal number of clusters using the elbow rule, silhouette coefficient, and Gap statistic, and using the K-means algorithm to cluster the three-dimensional morphological parameters into particle morphology groups; S13: using the discrete element method to conduct a stacking test simulation on the clustered particle categories, analyzing the stacking density and average coordination number of different particle categories, and verifying the physical meaning of the classification results. This tailings particle classification method can systematically and efficiently integrate multi-dimensional particle morphological information and directly correlate this information with macroscopic physical and mechanical properties. It is data-driven, has a clear mechanism, and the results are interpretable.

Description

Technical Field

[0001] This invention relates to the fields of geotechnical engineering, mineral processing and particle classification technology, and specifically to a tailings particle classification method based on multi-dimensional morphological feature compression and clustering. Background Technology

[0002] Tailings, as waste products from mining, require safe storage and comprehensive utilization, which is crucial for developing green mining and ensuring safety throughout the entire life cycle of mines. The stability and permeability of tailings dams, as well as their performance during comprehensive utilization, largely depend on the morphological characteristics of the tailings particles themselves. Existing particle classification methods have significant limitations. Traditional methods often rely on single-dimensional characteristics, such as classifying particles into large, coarse, and fine particles solely based on size; or classifying particles roughly into disc-shaped, spherical, blade-shaped, and rod-shaped categories based solely on elongation and flattening. These methods fail to systematically and accurately integrate key morphological information such as the actual shape and surface angularity of particles in three-dimensional space, making it difficult to fully reveal the intrinsic relationship between morphological characteristics and macroscopic physical and mechanical behavior.

[0003] With advancements in image processing and 3D reconstruction technologies, existing research has begun to incorporate more precise morphological parameters such as sphericity and 3D roundness. However, these multidimensional parameters are numerous and exhibit complex nonlinear correlations. Directly applying them all to particle classification analysis inevitably leads to the "curse of dimensionality," resulting in inefficient and unstable classification models that are difficult to implement in practice and fail to clearly explain the specific impact mechanisms of different morphological features on macroscopic performance. Therefore, we propose a tailings particle classification method based on multidimensional morphological feature compression and clustering. Summary of the Invention

[0004] The purpose of this invention is to provide a tailings particle classification method based on multi-dimensional morphological feature compression and clustering, so as to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] This invention provides a tailings particle classification method based on multi-dimensional morphological feature compression and clustering, comprising the following steps:

[0007] S1: Use a CT scanning system to obtain a sequence of continuous two-dimensional slice images of tailings particle samples;

[0008] S2: The heterogeneous filtering algorithm is used to perform noise suppression and image enhancement processing on the two-dimensional slice image sequence;

[0009] S3: Perform three-dimensional reconstruction on the two-dimensional slice image sequence processed in step S2;

[0010] S4: Use an interactive threshold segmentation method to separate tailings particles, and use a watershed algorithm to segment connected particles;

[0011] S5: Perform tag analysis based on the segmented particles;

[0012] S6: Based on the tag attributes attached to the particles, separate individual particles from the whole and generate STL files, and establish an STL file database;

[0013] S7: Based on the STL file database established in step S6, calculate and extract the morphological parameters of each tailings particle to form the original parameter dataset;

[0014] S8: Divide the original parameter dataset from step S7 into a training set, a validation set, and a test set according to a preset ratio;

[0015] S9: Fit a normalizer using the training set data and standardize all datasets;

[0016] S10: Input the original parameter dataset processed in step S9 into the autoencoder model to obtain a three-dimensional morphological parameter set that retains the original morphological feature information;

[0017] S11: Determine the optimal number of clusters using a multi-method fusion strategy;

[0018] S12: Based on the three-dimensional morphological parameter set obtained in step S10 and the optimal number of clusters determined in step S11, the tailings particles are clustered by particle morphology.

[0019] S13: Use the discrete element method to physically verify the particle morphology clustering grouping results of step S12.

[0020] Furthermore, in step S7, the morphological parameters of the tailings particles include size dimension parameters, shape dimension parameters, and roundness dimension parameters.

[0021] Furthermore, the dimensional parameters include the volume equivalent diameter. Volume to surface area ratio ;

[0022] The volume equivalent diameter The calculation formula is:

[0023] ,

[0024] in, Indicates the volume of tailings particles;

[0025] The volume of the tailings particles The calculation formula is:

[0026] ,

[0027] in, , , These represent the first and second tailings particles in the STL file. The coordinates of the three vertices of a surface triangular facet; Represents the centroid coordinates of the tailings particles;

[0028] The volume to surface area ratio The calculation formula is:

[0029] ,

[0030] in, This indicates the surface area of ​​the tailings particles.

[0031] The surface area of ​​the tailings particles The calculation formula is:

[0032] .

[0033] Furthermore, the shape dimension parameters include elongation. ,flat rate sphericity and convexity ;

[0034] The elongation The calculation formula is:

[0035] ,

[0036] in, This indicates the length of the longest axis of the tailings particles; Indicates the length of the central axis of the tailings particles;

[0037] The flatness The calculation formula is:

[0038] ,

[0039] in, This indicates the length of the shortest axis of the tailings particles;

[0040] sphericity The calculation formula is:

[0041] ,

[0042] The convexity The calculation formula is:

[0043] ,

[0044] in, This represents the volume of the smallest convex hull containing the tailings particles.

[0045] Furthermore, the roundness dimension parameter includes local roundness. and global roundness ;

[0046] Local roundness The calculation formula is:

[0047] ,

[0048] in, This represents the first tailings particle in the STL file. The area of ​​a surface triangular patch that is identified as a corner or ridge region; This indicates its corresponding curvature; This represents the curvature of the largest inscribed sphere of the tailings particle;

[0049] global roundness The calculation formula is:

[0050] ,

[0051] in, , , These represent the tailings particles in the STL file. The unit position vector, unit normal vector, and area of ​​a surface triangular patch.

[0052] Furthermore, the autoencoder model in step S10 includes an encoder and a decoder. The encoder contains an input layer and a hidden layer. The number of neurons in the input layer is the same as the number of morphological parameters in step S7. The number of neurons in the hidden layer decreases layer by layer, ultimately compressing the data into a three-dimensional latent space feature representation. The decoder adopts a symmetrical structure.

[0053] Furthermore, the autoencoder model uses LeakyReLU as the activation function, is trained using the Adam optimizer, employs mean squared error as the loss function, uses an early stopping mechanism to prevent overfitting, and adopts the ReduceLROnPlateau learning rate scheduling strategy.

[0054] Furthermore, the specific process of step S11 includes: analyzing the inflection point position of the sum of squares within clusters as a function of the number of clusters using the elbow rule; calculating the silhouette coefficient under different numbers of clusters to evaluate the compactness and separation of the clustering results; and using the Gap statistic method to generate a reference dataset through Monte Carlo simulation to compare the differences in clustering effects between the actual data and the random data.

[0055] When the number of clusters given by the elbow rule, silhouette coefficient, and gap statistic are not completely consistent, a voting mechanism is used to select the number of clusters that is the same given by at least two methods as the final optimal number of clusters.

[0056] Furthermore, the specific process of step S12 includes: performing unsupervised clustering analysis using the K-means clustering algorithm, measuring sample similarity through Euclidean distance, iteratively optimizing the cluster center position, and automatically grouping tailings particles with similar morphological features into the same category, thereby achieving automatic particle grouping based on morphological features.

[0057] Furthermore, the specific process of step S13 includes: selecting different confidence levels for each tailings particle category obtained after clustering and grouping in step S12. Three-dimensional packing tests were conducted on the tailings particles to simulate the natural packing process under the action of gravity. The same normal stiffness, tangential stiffness and friction coefficient parameters were set. After reaching static equilibrium, the packing density and average coordination number of each tailings particle pack were calculated. The packing density and coordination number of different types of particles were analyzed to verify the physical meaning of the classification results.

[0058] The confidence level The calculation formula is:

[0059] ,

[0060] in, This represents the shortest distance between tailings particles and the cluster center; This represents the average of the shortest distances from all tailings particles to the cluster center.

[0061] Compared with the prior art, the present invention has the following technical effects:

[0062] 1. Multi-dimensional information fusion: For the first time, eight parameters from three dimensions of tailings particles—size, shape, and roundness—are systematically incorporated into the classification system. Starting from the true three-dimensional morphology, the information is more comprehensive and more accurate.

[0063] 2. Feature compression: Innovatively using deep learning autoencoders for nonlinear dimensionality reduction, which can effectively overcome the "curse of dimensionality" while preserving core information and eliminating multicollinearity among parameters, providing a stable and reliable feature representation for subsequent clustering analysis.

[0064] 3. Scientific determination of the number of clusters: Combining the elbow rule, silhouette coefficient and gap statistic, three mainstream evaluation methods, the optimal number of clusters is determined through a voting mechanism, resulting in a more objective and reliable outcome, avoiding the bias that may be caused by a single method.

[0065] 4. Clear physical meaning: Through discrete element packing experiments, mathematical classification is directly linked to the macroscopic physical properties (packing density, coordination number) of the particle system, giving the classification results clear interpretability and application value. Attached Figure Description

[0066] Figure 1 This is a flowchart of the tailings particle classification method according to an embodiment of the present invention;

[0067] Figure 2 This is a diagram illustrating the elbow rule for determining the optimal number of clusters in an embodiment of the present invention.

[0068] Figure 3 This is a silhouette coefficient diagram for determining the optimal number of clusters in an embodiment of the present invention;

[0069] Figure 4 This is a Gap statistic diagram for determining the optimal number of clusters in an embodiment of the present invention.

[0070] Figure 5 This is a three-dimensional display of the K-means clustering results in the three-dimensional latent space according to an embodiment of the present invention;

[0071] Figure 6 This is a two-dimensional projection of the K-means clustering results in the three-dimensional latent space of this invention. Figure 1 ;

[0072] Figure 7 This is a two-dimensional projection of the K-means clustering results in the three-dimensional latent space of this invention. Figure 2 ;

[0073] Figure 8 This is a two-dimensional projection of the K-means clustering results in the three-dimensional latent space of this invention. Figure 3 ;

[0074] Figure 9 This is a comparison diagram of the bulk density of different types of tailings particles in an embodiment of the present invention.

[0075] Figure 10This is a comparison chart of the average coordination number of different types of tailings particles in an embodiment of the present invention. Detailed Implementation

[0076] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the protection scope of the present invention.

[0077] In this article, terms such as "left," "right," "up," "down," "front," and "back" are established based on the positional relationships shown in the attached drawings. Depending on the attached drawings, the corresponding positional relationships may also change. Therefore, they should not be interpreted as an absolute limitation on the scope of protection.

[0078] Please see Figure 1 This embodiment provides a tailings particle classification method based on multi-dimensional morphological feature compression and clustering, including the following steps:

[0079] S1: Use a CT scanning system to obtain a sequence of continuous two-dimensional slice images of tailings particle samples.

[0080] S2: The heterogeneous filtering algorithm is used to perform noise suppression and image enhancement on the two-dimensional slice image sequence, which can effectively preserve the particle boundary features while eliminating scanning artifacts.

[0081] S3: Perform three-dimensional reconstruction on the two-dimensional slice image sequence processed in step S2.

[0082] S4: Use an interactive threshold segmentation method to separate tailings particles, and use a watershed algorithm to segment connected particles.

[0083] S5: Perform tag analysis based on the segmented particles.

[0084] S6: Based on the tag attributes attached to the particles, separate individual particles from the whole and generate STL files, and establish an STL file database.

[0085] S7: Based on the STL file database established in step S6, calculate and extract eight morphological parameters for each tailings particle in three dimensions to form the original parameter dataset. The morphological parameters of the tailings particles include size dimension parameters, shape dimension parameters, and roundness dimension parameters. The size dimension parameters include the volume equivalent diameter. Volume to surface area ratio Shape dimension parameters include elongation. ,flat rate sphericity and convexity The roundness dimension parameter includes local roundness. and global roundness .

[0086] Specifically, the volume equivalent diameter The calculation formula is:

[0087] ,

[0088] in, Indicates the volume of tailings particles;

[0089] The volume of the tailings particles The calculation formula is:

[0090] ,

[0091] in, , , These represent the first and second tailings particles in the STL file. The coordinates of the three vertices of a surface triangular facet; Represents the centroid coordinates of the tailings particles;

[0092] The volume to surface area ratio The calculation formula is:

[0093] ,

[0094] in, This indicates the surface area of ​​the tailings particles.

[0095] The surface area of ​​the tailings particles The calculation formula is:

[0096] .

[0097] The elongation The calculation formula is:

[0098] ,

[0099] in, This indicates the length of the longest axis of the tailings particles; Indicates the length of the central axis of the tailings particles;

[0100] The flatness The calculation formula is:

[0101] ,

[0102] in, This indicates the length of the shortest axis of the tailings particles;

[0103] sphericity The calculation formula is:

[0104] ,

[0105] The convexity The calculation formula is:

[0106] ,

[0107] in, This represents the volume of the smallest convex hull containing the tailings particles.

[0108] Local roundness The calculation formula is:

[0109] ,

[0110] in, This represents the first tailings particle in the STL file. The area of ​​a surface triangular patch that is identified as a corner or ridge region; This indicates its corresponding curvature; This represents the curvature of the largest inscribed sphere of the tailings particles.

[0111] global roundness The calculation formula is:

[0112] ,

[0113] in, , , These represent the tailings particles in the STL file. The unit position vector, unit normal vector, and area of ​​a surface triangular patch.

[0114] S8: Divide the original parameter dataset from step S7 into a training set, a validation set, and a test set according to a preset ratio (set according to actual needs).

[0115] S9: Fit a normalizer using the training set data and perform Z-score normalization on all datasets to eliminate the influence of units.

[0116] Specifically, the mean and standard deviation of the dataset are calculated using the following formula:

[0117] ,

[0118] in, This represents the standardized numerical value; Represents the original data value; This represents the mean of each feature parameter in the training set; This represents the standard deviation of each feature parameter in the training set.

[0119] S10: Input the original parameter dataset processed in step S9 into the autoencoder model. The autoencoder model performs nonlinear dimensionality reduction on the original parameter dataset to obtain a three-dimensional morphological parameter set that retains the original morphological feature information.

[0120] Specifically, the autoencoder model in step S10 includes an encoder and a decoder. The encoder contains an input layer and hidden layers, and the number of neurons in the input layer is the same as the number of morphological parameters in step S7. The number of neurons in the hidden layers decreases layer by layer, ultimately compressing the data into a three-dimensional latent space feature representation; the decoder adopts a symmetric structure. The autoencoder model uses LeakyReLU as the activation function, is trained using the Adam optimizer, uses mean squared error as the loss function, employs an early stopping mechanism to prevent overfitting, and uses the ReduceLROnPlateau learning rate scheduling strategy.

[0121] S11: Use a multi-method fusion strategy to determine the optimal number of clusters.

[0122] Specifically, step S11 includes: analyzing the inflection point of the sum of squares within clusters as a function of the number of clusters using the elbow rule; calculating the silhouette coefficients for different numbers of clusters to evaluate the compactness and separation of the clustering results; and using the Gap statistic method to generate a reference dataset through Monte Carlo simulation to compare the differences in clustering effects between the actual data and the random data.

[0123] When the number of clusters given by the elbow rule, silhouette coefficient, and gap statistic are not completely consistent, a voting mechanism is used to select the number of clusters that is the same given by at least two methods as the final optimal number of clusters.

[0124] S12: Based on the three-dimensional morphological parameter set obtained in step S10 and the optimal number of clusters determined in step S11, the tailings particles are grouped by particle morphology clustering.

[0125] Specifically, step S12 includes: using the K-means clustering algorithm for unsupervised clustering analysis, measuring sample similarity through Euclidean distance, iteratively optimizing the cluster center position, and automatically grouping tailings particles with similar morphological features into the same category, thereby achieving automatic particle grouping based on morphological features.

[0126] S13: Use the discrete element method to physically verify the particle morphology clustering grouping results of step S12.

[0127] Specifically, step S13 includes: selecting different confidence levels for each tailings particle category obtained after clustering in step S12. Three-dimensional stacking tests were conducted on the tailings particles to simulate the natural stacking process under the action of gravity. The normal stiffness, tangential stiffness and friction coefficient parameters were set. After reaching static equilibrium, the packing density and average coordination number of each tailings particle stack were calculated. The packing density and coordination number of different types of particles were analyzed to verify the physical meaning of the classification results.

[0128] Confidence The calculation formula is:

[0129] ,

[0130] in, This represents the shortest distance between tailings particles and the cluster center; This represents the average of the shortest distances from all tailings particles to the cluster center.

[0131] Specifically, this method, for the first time, systematically incorporates eight parameters across three dimensions—size, shape, and roundness—of tailings particles into the classification system simultaneously. Starting from the true three-dimensional morphology, the information is more comprehensive and the accuracy is higher. It innovatively uses a deep learning autoencoder for nonlinear dimensionality reduction, effectively overcoming the "curse of dimensionality" and eliminating multicollinearity among parameters while preserving core information, providing stable and reliable feature representations for subsequent cluster analysis. Combining three mainstream evaluation methods—elbow rule, silhouette coefficient, and Gap statistic—a voting mechanism is used to jointly determine the optimal number of clusters, resulting in more objective and reliable results and avoiding biases that may arise from a single method. Through discrete element packing experiments, mathematical classification is directly correlated with the macroscopic physical properties of the particle system (packing density, coordination number), giving the classification results clear interpretability and application value.

[0132] To further illustrate this tailings particle classification method, the following section uses gold mine tailings as the research object and describes the tailings particles using this method.

[0133] A sequence of continuous two-dimensional slice images of tailings particle samples was acquired using a CT scanning system. Specific scanning parameters were set as follows: X-ray source operating voltage 70 kV, operating current 86 μA, and projection exposure time 32 seconds, maintaining a spatial resolution of 2.45 µm during 360° rotation scanning. Subsequently, Avizo software was used to perform three-dimensional reconstruction of the scanned images. This process first employed a nonlocal mean filtering algorithm for noise suppression and image enhancement, effectively preserving particle boundary features while eliminating scanning artifacts. Then, interactive thresholding with a threshold set to 7500 was used to separate tailings particles from the original images, and a watershed algorithm was applied to segment connected particles. Based on the segmented particles, label analysis was performed. Finally, Python code was written to automate the process, separating individual particles from the whole based on their attached label attributes and generating STL files, creating an STL file database containing 4177 particles.

[0134] Based on the obtained STL file database, dedicated processing code was written using open-source third-party libraries such as numpy, pandas, trimesh, and scipy to automatically calculate and extract eight key morphological parameters for each tailings particle in three dimensions. The morphological parameters will not be repeated here. The extracted parameters form the original parameter dataset.

[0135] The original parameter dataset was divided into training, validation, and test sets in a 70%:15%:15% ratio. The training set contained 2923 samples, the validation set contained 627 samples, and the test set contained 627 samples. The StandardScaler normalizer was fitted using the training set data, and Z-score normalization was applied to all datasets to eliminate the influence of units.

[0136] Subsequently, an autoencoder model was constructed, with its input layer having 8 neurons to match the number of morphological parameters. The encoder's hidden layers adopted a hierarchical structure of 128→64→32→16→3, with each layer including batch normalization and the LeakyReLU activation function (with a negative slope set to 0.05), ultimately compressing the data into a 3D latent space feature representation. The decoder employed a symmetric structure for data reconstruction. Model training used the Adam optimizer with a learning rate of 0.005 and a weight decay of 1e-5, using mean squared error as the loss function and employing an early stopping mechanism to prevent overfitting, along with the ReduceLROnPlateau learning rate scheduling strategy. After training, the model terminated after 580 epochs due to the early stopping mechanism, with a final training loss of 0.067800, a validation loss of 0.067744, and a test loss of 0.068345. The standardized original parameter dataset was then input into the trained autoencoder model, which performed dimensionality reduction to obtain a 3D morphological parameter set that preserved the original morphological feature information.

[0137] like Figure 2 As shown, the elbow rule is used to analyze the inflection point of the sum of squares within a cluster as a function of the number of clusters. The inflection point of the elbow rule is defined as the critical point where the rate of decrease of the sum of squares within a cluster slows down. It is found that the inflection point occurs optimally at K=4. Figure 3 As shown, the silhouette coefficients for different numbers of clusters were calculated to evaluate the compactness and separation of the clustering results. The results show that the optimal K=3. Figure 4 As shown, the Gap statistic method was used to generate a reference dataset through Monte Carlo simulation. The clustering effect difference between the actual data and the random data was compared to determine the optimal K=3. Based on the results of the three methods, a voting mechanism was adopted to select the same number of clusters given by the silhouette coefficient method and the Gap statistic method as the final optimal number of clusters, and the optimal number of clusters was finally determined to be 3.

[0138] Based on a three-dimensional morphological parameter set and the optimal number of clusters, the K-means clustering algorithm is used for unsupervised clustering analysis. Euclidean distance is used to measure sample similarity, and the cluster center positions are iteratively optimized. Tailings particles with similar morphological features are automatically grouped into the same category, achieving automatic particle grouping based on morphological characteristics. The three-dimensional display of the clustering results is shown in the figure below. Figure 5 As shown. Figures 6 to 8 These are two-dimensional projections of a three-dimensional display image in different dimensions. Figure 6 This is a two-dimensional planar projection of a three-dimensional image onto the plane formed by dimension 1 and dimension 2. Figure 7 This is a two-dimensional planar projection of a three-dimensional image onto the plane formed by dimension 1 and dimension 3. Figure 8 This is a two-dimensional planar projection of a three-dimensional image onto a plane composed of dimensions 2 and 3.

[0139] The discrete element method was used to physically verify each tailings particle category obtained after clustering, and the categories were then ranked according to confidence level. Five representative particles of varying sizes were selected for three-dimensional packing simulation tests. The particle size of the selected tailings particles was magnified 50 times. A linear contact model was used in PFC3D software, and a uniform normal stiffness of 5.4 × 10⁻⁶ was set. N / m, tangential stiffness 3.6× Using parameters of N / m and a friction coefficient of 0.5, the natural accumulation process of tailings particles in a 0.3m × 0.3m × 1m box with a height 5 times their particle size was simulated under a gravitational field. A total of 3000 particles were collected. After reaching static equilibrium, the packing density and average coordination number of each tailings particle accumulation were calculated, revealing significant differences. (The comparison of packing density is shown below.) Figure 9 As shown, the average coordination number pairs are as follows: Figure 10 As shown.

[0140] The above embodiments merely illustrate the basic principles and characteristics of the present invention, but are not limited to the above implementation schemes. It should be understood that those skilled in the art can make various changes and modifications to the present invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A tailings particle classification method based on multi-dimensional morphological feature compression and clustering, characterized in that, Includes the following steps: S1: Use a CT scanning system to obtain a sequence of continuous two-dimensional slice images of tailings particle samples; S2: The heterogeneous filtering algorithm is used to perform noise suppression and image enhancement processing on the two-dimensional slice image sequence; S3: Perform three-dimensional reconstruction on the two-dimensional slice image sequence processed in step S2; S4: Use an interactive threshold segmentation method to separate tailings particles, and use a watershed algorithm to segment connected particles; S5: Perform tag analysis based on the segmented particles; S6: Based on the tag attributes attached to the particles, separate individual particles from the whole and generate STL files, and establish an STL file database; S7: Based on the STL file database established in step S6, calculate and extract the morphological parameters of each tailings particle to form the original parameter dataset; S8: Divide the original parameter dataset from step S7 into a training set, a validation set, and a test set according to a preset ratio; S9: Fit a normalizer using the training set data and standardize all datasets; S10: Input the original parameter dataset processed in step S9 into the autoencoder model to obtain a three-dimensional morphological parameter set that retains the original morphological feature information; S11: Determine the optimal number of clusters using a multi-method fusion strategy; S12: Perform particle morphology clustering based on the three-dimensional morphological parameter set obtained in step S10 and the optimal number of clusters determined in step S11; S13: Use the discrete element method to physically verify the particle morphology clustering grouping results of step S12.

2. The tailings particle classification method based on multi-dimensional morphological feature compression and clustering according to claim 1, characterized in that, In step S7, the morphological parameters of the tailings particles include size dimension parameters, shape dimension parameters, and roundness dimension parameters.

3. The tailings particle classification method based on multi-dimensional morphological feature compression and clustering according to claim 2, characterized in that, The dimensional parameters include the volume equivalent diameter. Volume to surface area ratio ; The volume equivalent diameter The calculation formula is: ， in, Indicates the volume of tailings particles; The volume of the tailings particles The calculation formula is: ， in, , , These represent the first and second tailings particles in the STL file. The coordinates of the three vertices of a surface triangular facet; Represents the centroid coordinates of the tailings particles; The volume to surface area ratio The calculation formula is: ， in, Indicates the surface area of ​​tailings particles; The surface area of ​​the tailings particles The calculation formula is: 。 4. The tailings particle classification method based on multi-dimensional morphological feature compression and clustering according to claim 2, characterized in that, The shape dimension parameters include elongation. ,flat rate sphericity and convexity ; The elongation The calculation formula is: ， in, This indicates the length of the longest axis of the tailings particles; Indicates the length of the central axis of the tailings particles; The flatness The calculation formula is: ， in, This indicates the length of the shortest axis of the tailings particles; sphericity The calculation formula is: ， The convexity The calculation formula is: ， in, This represents the volume of the smallest convex hull containing the tailings particles.

5. The tailings particle classification method based on multi-dimensional morphological feature compression and clustering according to claim 2, characterized in that, The roundness dimension parameter includes local roundness. and global roundness ; Local roundness The calculation formula is: ， in, This represents the first tailings particle in the STL file. The area of ​​a surface triangular patch that is identified as a corner or ridge region; This indicates its corresponding curvature; This represents the curvature of the largest inscribed sphere of the tailings particle; global roundness The calculation formula is: ， in, , , These represent the tailings particles in the STL file. The unit position vector, unit normal vector, and area of ​​a surface triangular patch.

6. The tailings particle classification method based on multi-dimensional morphological feature compression and clustering according to claim 1, characterized in that, The autoencoder model in step S10 includes an encoder and a decoder. The encoder contains an input layer and a hidden layer. The number of neurons in the input layer is the same as the number of morphological parameters in step S7. The number of neurons in the hidden layer decreases layer by layer, compressing the data into a three-dimensional latent space feature representation. The decoder adopts a symmetrical structure.

7. The tailings particle classification method based on multi-dimensional morphological feature compression and clustering according to claim 6, characterized in that, The autoencoder model uses LeakyReLU as the activation function, is trained using the Adam optimizer, uses mean squared error as the loss function, employs an early stopping mechanism to prevent overfitting, and uses the ReduceLROnPlateau learning rate scheduling strategy.

8. The tailings particle classification method based on multi-dimensional morphological feature compression and clustering according to claim 1, characterized in that, The specific process of step S11 includes: analyzing the inflection point of the sum of squares within clusters as a function of the number of clusters using the elbow rule; calculating the silhouette coefficient under different numbers of clusters to evaluate the compactness and separation of the clustering results; and using the Gap statistic method to generate a reference dataset through Monte Carlo simulation to compare the differences in clustering effects between actual data and random data. When the number of clusters given by the elbow rule, silhouette coefficient, and gap statistic are not completely consistent, a voting mechanism is used to select the number of clusters that is the same given by at least two methods as the final optimal number of clusters.

9. The tailings particle classification method based on multi-dimensional morphological feature compression and clustering according to claim 8, characterized in that, The specific process of step S12 includes: using the K-means clustering algorithm to perform unsupervised clustering analysis, measuring sample similarity through Euclidean distance, iteratively optimizing the cluster center position, and automatically grouping tailings particles with similar morphological features into the same category, thereby realizing automatic particle grouping based on morphological features.

10. The tailings particle classification method based on multi-dimensional morphological feature compression and clustering according to claim 1, characterized in that, The specific process of step S13 includes: selecting different confidence levels for each tailings particle category obtained after clustering and grouping in step S12. Three-dimensional packing tests were conducted on the tailings particles to simulate the natural packing process under the action of gravity. The same normal stiffness, tangential stiffness and friction coefficient parameters were set. After reaching static equilibrium, the packing density and average coordination number of each tailings particle pack were calculated. The packing density and coordination number of different types of particles were analyzed to verify the physical meaning of the classification results. The confidence level The calculation formula is: ， in, This represents the shortest distance between tailings particles and the cluster center; This represents the average of the shortest distances from all tailings particles to the cluster center.

Images