A method, device, and system for monitoring mine slopes based on 3D visualization management.

Abstract

The application discloses a mine slope monitoring method, device and system based on three-dimensional visualization management, belongs to the technical field of mine monitoring, and comprises the following steps: collecting original data and processing to obtain a three-dimensional slope model, performing regional division on the three-dimensional slope model, and obtaining the slope surface deformation trend of each region through calculation; based on the above data, internal stress modeling of the region is completed, and the region is scored in terms of risk. The mine slope monitoring method, device and system based on three-dimensional visualization management, through joint modeling of historical monitoring data and current dynamic changes, establish a prediction mechanism for the internal risk evolution process of the slope, and realize cross-level monitoring closed loop from surface perception to deep understanding.

Description

Technical Field

[0001] This invention belongs to the field of mine monitoring technology, and in particular relates to a method, device and system for monitoring mine slopes based on three-dimensional visualization management. Background Technology

[0002] Mine slope stability monitoring has always been a crucial aspect of mine geological disaster early warning and safety management. With the continuous advancement of large-scale open-pit mining projects, the complexity of slope structures, the frequency of operational disturbances, and the coupled influence of external environmental conditions (such as rainfall, earthquakes, and blasting operations) have resulted in significant dynamic evolution characteristics of slopes in both time and space, posing a significant challenge to traditional monitoring and management methods. Currently, mainstream slope monitoring technologies primarily rely on distributed sensing devices such as displacement gauges, inclinometers, GNSS, and surface crack gauges, combined with manual inspections and two-dimensional chart analysis for data interpretation.

[0003] However, these methods generally suffer from limited monitoring area coverage, inability to fully reflect spatial structure, and delayed response of early warning mechanisms. To address the disconnect between spatial structure and dynamic data, spatial modeling technologies such as 3D laser scanning, UAV aerial surveying, and remote sensing imagery have been gradually introduced into slope monitoring in recent years, forming a preliminary foundation of 3D model data. However, current 3D visualization remains primarily static, lacking a linkage mechanism with real-time monitoring data, making it difficult to automatically identify and structurally manage abnormal slope changes. Furthermore, existing 3D models still rely on manual division for regional structure recognition, failing to set differentiated monitoring strategies based on the geomorphological structure, material composition, and strain characteristics of different areas; moreover, they lack systematic modeling of the causal relationship between "surface deformation—deep stress evolution" of slopes, making it impossible for the system to accurately identify potential slip paths or disaster-inducing mechanisms. Therefore, under the current technological framework, 3D visualization serves only as an auxiliary presentation method, failing to establish an integrated closed loop of "intelligent perception—structural identification—dynamic analysis—3D management," thus failing to meet the needs of efficient, accurate, and safe monitoring and management in complex mine slope environments.

[0004] To address these issues, we propose a method, device, and system for monitoring mine slopes based on three-dimensional visualization management. Summary of the Invention

[0005] The purpose of this invention is to solve the problem that existing technologies cannot automatically identify and manage slope conditions, and to propose a method, device and system for monitoring mine slopes based on three-dimensional visualization management.

[0006] To achieve the above objectives, the present invention adopts the following technical solution:

[0007] A method for monitoring mine slopes based on three-dimensional visualization management includes:

[0008] Collect raw data from multiple sources, perform data alignment and registration on the raw data, and transform the processed data into a three-dimensional mesh model through a triangulation algorithm. Assign corresponding physical attribute vectors to each mesh node of the three-dimensional mesh model, and finally obtain a three-dimensional slope model.

[0009] Based on a 3D slope model, geometric cluster labels and physical cluster labels are obtained through geometric feature clustering and physical attribute vector clustering, respectively. Cross-matching is then performed, and structured regions are constructed on grid points with consistent labels. The structured region set is integrated and updated to the slope model to obtain a regionalized 3D slope model. The structured region contains a set of grids, and each grid has a location and a corresponding regional physical attribute.

[0010] In a structured region, a regional average slope normal vector is constructed. This regional average slope normal vector is obtained by weighted averaging of the normal vectors of all triangular facets in the region, with the weight being the facet area. For all matching points in the structured region, the projection residuals along the direction of the regional average slope normal vector are calculated, and the deformation heat value of the corresponding point is calculated through the projection residuals. A dynamic weighting mechanism is introduced, and different residual enhancement strategies are adopted based on the different physical characteristics of different regions to obtain the deformation map value of each point. Finally, the residual heat map of each region is aggregated and output.

[0011] Input the residual thermal spectrum, regional physical properties, and spatial structure information of the three-dimensional slope model for each region to establish a deep stress distribution model for each region. The deep stress distribution model is constructed by using the residual thermal spectrum as a supervision signal within the region to build a mapping function and complete the fitting modeling of the stress distribution of the regional state terms.

[0012] Input the residual thermal map, deep stress distribution model and physical properties of each region, complete the regional risk score modeling based on the regionalized three-dimensional slope model, and map the risk score to the visualization platform; wherein, the regional risk score is calculated by assigning the importance of variables through empirical weights, and the linear weighted sum is compressed to achieve smooth normalization to obtain the risk score.

[0013] Preferably, the physical attribute vector is based on known physical data and uses an interpolation method to assign physical attributes to each grid point. The physical attributes include soil density, friction coefficient, and porosity.

[0014] Preferably, the geometric features include a slope value and a roughness, wherein the slope value is calculated from the angle between the grid surface normal vector and the vertical direction of the ground, and the roughness is calculated as the average value of the difference between the normal vectors of the grid and its adjacent grids.

[0015] Preferably, a neighborhood smoothing difference regularization term is introduced when calculating the deformation heat value, and the real anomalies are distinguished from isolated errors through neighborhood consistency analysis.

[0016] Preferably, the deep stress distribution model adopts a multi-layer sensing grid structure, and is trained separately in each region. The training data comes from multiple time sequence data of the same region during the monitoring period, and small data training is achieved by extracting samples through a sliding window.

[0017] Preferably, mapping risk scores to the visualization platform involves mapping each risk score to an attribute field of the corresponding area in the 3D slope model, specifically as follows:

[0018] Add a risk score to each corresponding area in the three-dimensional slope model; synchronously record the scoring time and mark the source of risk.

[0019] An electronic device, characterized in that it includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, it implements the method as described in any of the preceding claims.

[0020] A mine slope monitoring system based on three-dimensional visualization management includes:

[0021] The data acquisition module is configured to collect raw data from multiple sources, perform data alignment and registration on the raw data, and transform the processed data into a three-dimensional mesh model through a triangulation algorithm. Each mesh node of the three-dimensional mesh model is assigned a corresponding physical attribute vector, and a three-dimensional slope model is obtained.

[0022] The region division module is set to be based on the three-dimensional slope model. After obtaining geometric cluster labels and physical cluster labels through geometric feature clustering and physical attribute vector clustering respectively, cross matching is performed to construct structured regions on grid points with consistent labels. The structured region set is integrated and output and updated to the slope model to obtain the regionalized three-dimensional slope model.

[0023] The deformation recognition module is configured to construct a regional average slope normal vector in a structured region. The regional average slope normal vector is obtained by weighted averaging of the normal vectors of all triangular facets in the region, with the weight being the facet area. The module calculates the projection residual of all matching points in the structured region along the direction of the regional average slope normal vector, and calculates the deformation heat value of the corresponding point through the projection residual. A dynamic weighting mechanism is introduced, and different residual enhancement strategies are adopted based on the different physical characteristics of different regions to obtain the deformation map value of each point. The module then aggregates and outputs the residual heat map spectrum of each region.

[0024] The stress inversion module is set to take into account the residual thermal spectrum, regional physical properties, and spatial structure information of the three-dimensional slope model for each region, and to establish a deep stress distribution model for each region. The deep stress distribution model is constructed by using the residual thermal spectrum as a supervision signal within the region to build a mapping function and complete the fitting modeling of the stress distribution of the regional state terms.

[0025] The scoring and output module is set to take into account the residual thermal map, deep stress distribution model and physical properties of each region. Based on the regionalized three-dimensional slope model, it completes the regional risk score modeling and maps the risk score to the visualization platform. The regional risk score is calculated by assigning the importance of variables with empirical weights, and the linear weighted sum is compressed to achieve smooth normalization to obtain the risk score.

[0026] In summary, the technical effects and advantages of this invention are as follows: This mine slope monitoring method, device, and system based on three-dimensional visualization management constructs an adaptive, partitioned slope structure model, supports the formulation of differentiated monitoring strategies for different areas, and can capture minute trend changes in slope deformation in real time, generating deformation response maps with regional characteristics. Through joint modeling of historical monitoring data and current dynamic changes, a prediction mechanism for the evolution of risks within the slope is established, realizing a cross-level monitoring closed loop from surface perception to deep understanding. This not only significantly improves the accuracy of identifying local anomalies but also enhances the response efficiency of the three-dimensional system in assisting early warning decision-making and operational intervention. Attached Figure Description

[0027] Figure 1 This is a flowchart of the steps in this invention;

[0028] Figure 2 This is a schematic diagram of the system structure in this invention. Detailed Implementation

[0029] 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.

[0030] As shown in the figure, a method for monitoring mine slopes based on three-dimensional visualization management includes:

[0031] Collect raw data from multiple sources, perform data alignment and registration on the raw data, and transform the processed data into a three-dimensional mesh model through a triangulation algorithm. Assign corresponding physical attribute vectors to each mesh node of the three-dimensional mesh model, and finally obtain a three-dimensional slope model.

[0032] Based on a 3D slope model, geometric cluster labels and physical cluster labels are obtained through geometric feature clustering and physical attribute vector clustering, respectively. Cross-matching is then performed, and structured regions are constructed on grid points with consistent labels. The structured region set is integrated and updated to the slope model to obtain a regionalized 3D slope model. The structured region contains a set of grids, and each grid has a location and a corresponding regional physical attribute.

[0033] In a structured region, a regional average slope normal vector is constructed. This regional average slope normal vector is obtained by weighted averaging of the normal vectors of all triangular facets in the region, with the weight being the facet area. For all matching points in the structured region, the projection residuals along the direction of the regional average slope normal vector are calculated, and the deformation heat value of the corresponding point is calculated through the projection residuals. A dynamic weighting mechanism is introduced, and different residual enhancement strategies are adopted based on the different physical characteristics of different regions to obtain the deformation map value of each point. Finally, the residual heat map of each region is aggregated and output.

[0034] Input the residual thermal spectrum, regional physical properties, and spatial structure information of the three-dimensional slope model for each region to establish a deep stress distribution model for each region. The deep stress distribution model is constructed by using the residual thermal spectrum as a supervision signal within the region to build a mapping function and complete the fitting modeling of the stress distribution of the regional state terms.

[0035] Input the residual thermal map, deep stress distribution model and physical properties of each region, complete the regional risk score modeling based on the regionalized three-dimensional slope model, and map the risk score to the visualization platform; wherein, the regional risk score is calculated by assigning the importance of variables through empirical weights, and the linear weighted sum is compressed to achieve smooth normalization to obtain the risk score.

[0036] The specific steps are as follows:

[0037] Step 1: Integration of multi-source heterogeneous data and structured 3D slope modeling

[0038] Multi-source data acquisition and preprocessing

[0039] In mine slope monitoring, we need to collect data from multiple sensors and devices. Common data sources include LiDAR point cloud data, UAV aerial images, and remote sensing imagery. These data sources provide different types of information. To integrate this data from different sources, we first need to align and register it. Since each type of data may use different coordinate systems and spatial resolutions, we need to align the data in the following ways:

[0040] Data Registration: LiDAR and remote sensing imagery data typically use different coordinate systems. We first select a set of common control points, which are easily identifiable geological features in the mining area, such as rock strata, ridges, or prominent rock structures. When using these control points for registration, the spatial differences between the two sets of data can be calculated and corrected using the least squares method. For example, for LiDAR point cloud data... and remote sensing image data Assume the registration transformation matrix is The goal of the registration process is to minimize the registration error:

[0041] ;

[0042] here, and These represent the first and second parts of the lidar point cloud and remote sensing image data, respectively. The spatial coordinates of the points, and It is the transformation matrix that needs to be calculated.

[0043] Data Interpolation: After registration, the LiDAR data may be sparse in some areas, especially in areas of surface detail (such as vegetation or slope details). To compensate for these omissions, we use nearest neighbor interpolation to supplement the point cloud data. For each missing point... We interpolate based on its neighboring known points. Let the neighboring points be... The interpolation formula is:

[0044] ;

[0045] This method can fill in sparse areas, ensuring the integrity and coherence of point cloud data.

[0046] Construction of a 3D slope model

[0047] After data alignment and interpolation, the next step is to convert the processed point cloud data into a 3D slope model. The Delaunay triangulation algorithm is used to convert the point cloud data into a 3D mesh model. A key feature of the Delaunay algorithm is that the generated triangular network has the property of being an "empty circle," meaning that no other points lie within the circumcircle of the triangle, ensuring that the generated mesh is as uniform as possible.

[0048] For example, given point cloud data We use the Delaunay triangulation algorithm to generate triangular meshes, where the edges of each triangle are formed by connecting points in the point cloud. These triangles will represent the surface structure of the slope and provide the basis for assigning physical properties and risk assessment in subsequent steps.

[0049] Attribute assignment and physical feature integration

[0050] The physical properties of a slope (such as soil density, friction coefficient, and porosity) are crucial factors affecting slope stability. Therefore, in addition to geometric modeling, assigning corresponding physical properties to each grid node is an important part of this step. Through on-site geological surveys, sensor data, and remote sensing analysis, each grid point is assigned its corresponding geological information.

[0051] Assuming the geological feature we use is density coefficient of friction These properties can be obtained from geological survey data. For example, in the rock strata of a mine, we assign a higher coefficient of friction (e.g., ...) to the grid points. In loose soil areas, it imparts a lower coefficient of friction (e.g. For density, a value can be assigned based on the lithology of different regions; for example, the density in hard rock regions is... Loose soil area .

[0052] Using interpolation methods (such as Kriging interpolation), we can generalize these physical properties from measurement points to the entire region, thereby assigning accurate physical characteristics to each grid point. In the interpolation process, we utilize known geological point data to extrapolate the properties of unknown points based on their spatial location. For example, for a given region... A grid point within Assuming the point is known The attribute value is The interpolation formula is as follows:

[0053] ;

[0054] in, The weights are calculated based on spatial distance, and are usually calculated using the inverse distance weighting (IDW) method or the Kriging interpolation method.

[0055] Output: A structured 3D model

[0056] Finally, after the above steps, we obtain the three-dimensional model. It includes the slope's geometry and the physical properties of each region. This information will provide the foundational data for subsequent zoning, deformation detection, and risk assessment. The model can be represented as:

[0057] ;

[0058] in:

[0059] The 3D point cloud data, after registration and interpolation processing, contains the geometric information of the slope.

[0060] The physical property data of each grid point, such as rock density, friction coefficient, and porosity, are obtained through sensors, geological surveys, and interpolation methods.

[0061] Step 2: Semantic Region Delineation and Modeling of Slopes with Structural Heterogeneity

[0062] The purpose of this step is to build upon the three-dimensional slope model constructed in step 1. The entire slope area was divided into several sub-regions with consistent geological properties and similar morphological characteristics. It serves as an independent analysis unit for subsequent deformation monitoring and stress inversion. It is 3D point cloud data, including the spatial coordinates of each grid. ; It is a set of attributes, including soil density, friction coefficient, porosity, etc., with each grid corresponding to a physical attribute vector. .

[0063] This step addresses the characteristic of "coexistence of multiple geological types" on mine slopes, considering the following real-world scenario: Within the same slope, there may be areas with exposed rock strata, highly water-bearing loose soil, and artificial deposits, each with distinct deformation response mechanisms and monitoring requirements. Indiscriminate analysis without differentiation can easily lead to erroneous judgments. Therefore, this step achieves spatial difference modeling through regional division, improving the system's adaptability and analytical accuracy under different geomorphic structures.

[0064] The region division begins with geometric features. We utilize the triangular mesh information in the slope model to extract slope values ​​from each triangular facet of the mesh. and roughness As a geometric feature, the slope is calculated from the angle between the grid surface normal vector and the vertical direction of the ground, and the roughness is determined by the ratio of the grid surface normal vector to its adjacent surface roughness. The average value is calculated from the differences in the normal vectors of each grid cell. We construct the following geometric eigenvectors. Preliminary clustering is performed on all grid points.

[0065] This clustering process uses the K-means algorithm, where the number of clusters... Automatically selected using the "elbow rule". Each cluster center corresponds to a terrain type, such as "gentle slope area", "steep slope area", "crack boundary", etc. The input to the K-means algorithm is all... The output is the initial geometry category label for each mesh.

[0066] After completing the geometric clustering, we introduce physical attributes for further refinement. We employ a publicly available Gaussian Mixture Model (GMM) to refine the physical attribute vectors obtained in step 1. Perform probabilistic modeling and classification. Attributes of each grid point. , respectively corresponding to soil and rock density (Unit: g / cm³), coefficient of friction (No unit), porosity (Percentage). We model these attributes using GMM and assign the physical class of grid points based on the maximum a posteriori probability.

[0067] To avoid significant geometric and physical inconsistencies in the region boundaries, we design the following regularization term to embed the objective function of the GMM, which is used to optimize the final region consistency:

[0068] ;

[0069] in:

[0070] This represents two spatially adjacent grid points;

[0071] This is the slope value corresponding to the grid.

[0072] It is the corresponding physical property vector;

[0073] The first part of the regularization term constrains the boundary slope of the region, and the second part constrains the attribute.

[0074] These are weighting factors, and their values ​​in the experiment are as follows: , (Adjustable).

[0075] This regularization term is iteratively optimized during model training by the loss function we embed in the GMM to ensure that the divided regions are geometrically and physically consistent.

[0076] After obtaining the geometric cluster labels and physical cluster labels, we perform "cross-matching," which means using both geometric and physical clustering as the basis for partitioning, and constructing the final region only on grid points where the labels of both are consistent. Each region contains:

[0077] Grid ID set;

[0078] Mean attribute vector (as shown below):

[0079] ;

[0080] in Indicates the region The number of grid points included. It is the first The physical attributes of each point are derived from the physical assignment results in step 1.

[0081] The output is a set of structured regions. The updated slope model is represented as follows:

[0082] ;

[0083] Each region possesses physical consistency and geometric coherence, serving as the smallest spatial unit for subsequent processes such as deformation residual map calculation and deep stress inversion. We will independently model and predict within each region, thereby achieving the core capabilities of "regional monitoring" and "regional early warning."

[0084] Step 3: Identification of slope deformation trends based on semantic attributes (directional residual map)

[0085] This step aims to build upon the regionalized 3D slope model output in step 2. This enables high-precision identification of minute deformation trends in multi-temporal point cloud data of slope surfaces and generates residual maps with geological orientation sensitivity and regional structural consistency. This serves as the input basis for subsequent stress inversion and risk assessment.

[0086] Compared to traditional point cloud analysis methods based on full-slope difference, this step focuses on "structural units". The temporal changes of the "internal" are identified by regional normal vector projection to realize deformation recognition under spatial constraints. On this basis, a neighborhood smoothing difference regularization term is introduced to improve the adaptability and anti-interference ability in real complex landforms in mines (such as accumulation bodies, fracture boundaries, rock-soil transition zones, etc.).

[0087] The input consists of point cloud data from two consecutive time frames. (Registered to a unified coordinate system, sourced from lidar or structured light scanning equipment), and region division results. Each region It contains a set of grids, and each grid has a position. and the corresponding regional physical properties (Including density, coefficient of friction, porosity, etc., derived from step 2).

[0088] We first in each Internally constructed region average slope normal vector It is used to guide the identification of deformation direction. The weighted average of the normal vectors of all triangular facets in the region is used to enhance the stability of the representation of the dominant slope aspect of the region.

[0089] Next, we will... All matching point pairs in the region Calculate its position in the principal direction The projection residual on is defined as follows:

[0090] ;

[0091] in:

[0092] For point-to-point in the region The difference in normal projection over the surface, in meters;

[0093] yes The average normal vector, the unit vector;

[0094] They are the first At time 1 point and 3D coordinates;

[0095] Point pairs are obtained through indexing or nearest neighbor matching after point cloud registration, using a fixed radius. The search ensures the accuracy of the matching points.

[0096] This projection method has three advantages: (1) reducing illegal directional disturbance error; (2) automatically adapting to the main deformation direction of different regions; and (3) preserving the interpretability of the project (such as "large outward normal residual" as a slip risk signal).

[0097] To further enhance the identification of true outliers, we design a residual consistency weighting mechanism that considers not only the displacement of a point itself, but also its consistency with other points in its neighborhood, thereby suppressing single-point errors and enhancing the structural response. The mechanism is defined as follows:

[0098] ;

[0099] in:

[0100] for The Middle The deformation heat value at each point;

[0101] This is the normal deformation value at that point;

[0102] This is the set of neighborhood points of this point in a 3D mesh structure (using a 1-ring neighborhood based on topological connections).

[0103] The first measure is absolute deformation, and the second measure is local consistency deviation.

[0104] To integrate weights, the default is... .

[0105] This design has clear engineering semantics: isolated residuals are reduced, and mass slip is amplified, which can effectively identify local early deformation zones (such as the initial appearance of cracks, the slip front of the deposited layer, etc.) in the context of large-scale stability of the slope.

[0106] To address the challenges posed by the dense noise points and complex deformation patterns in typical "mixed geological structures" of mining areas, we specifically introduced a method related to regional physical properties. The combined dynamic weighting mechanism allows different residual enhancement strategies to be adopted for different regions based on their physical characteristics. The definition is as follows:

[0107] ;

[0108] This item As a regional risk amplification factor, embedded into The final output is used to enhance the expressive power of subtle deformations in high-risk areas. Specifically:

[0109] The density of the region (in g / cm³) is from step 2;

[0110] The regional porosity (percentage) is derived from step 2;

[0111] Set coefficients for experience, such as , ;

[0112] When the regional structure is loose (low density, high porosity), To enhance its deformation sensitivity.

[0113] The final deformed image value for each point is: This enables attribute-driven risk visualization enhancement. The output is for each region. Residual thermogram It not only has clear spatial structural boundaries, but can also be directly mapped onto a 3D visual platform to display deformation trends. It can also be used as a surface boundary condition in deep stress modeling to further improve the accuracy of physical modeling.

[0114] Step 4: Intelligent Inversion of Deep Stress Field with Regional Structural Constraints

[0115] This step involves the regional deformation trend map provided in step 3. Based on this, combined with the regional physical attributes extracted in step 2 and three-dimensional structural model Spatial structural information to establish each semantic region Deep stress distribution model The goal of this stress field modeling is to deduce the internal stress state of the slope, serving as a foundational input for subsequent risk assessment and response mechanisms. All modeling work is confined to the region. Internally, it does not involve cross-regional propagation or strategy response. Considering the highly non-uniform and nonlinear mechanical characteristics of actual mine slopes, traditional continuous medium finite element modeling or simplified homogeneous elastic body models are difficult to adapt. Therefore, we propose a structural constraint-based intelligent modeling method that combines regional physical properties, within the regional unit. Internally, using directional residual maps As a monitoring signal, a mapping function is constructed. This completes the fitting modeling from the regional state to the stress distribution.

[0116] We define this inversion task as:

[0117] ;

[0118] in:

[0119] It is a region Output the set of stress estimates (in MPa) for all meshes within the array;

[0120] The regional normal residual map obtained in step 3 is used as surface deformation observation data;

[0121] It is a vector of regional physical properties, including density. coefficient of friction Porosity wait;

[0122] It is an adjacency diagram of the grid within the region (constructed by the connection relationship of triangular faceted grids), used to reflect the spatial structure of the region;

[0123] It is a trainable prediction function with parameters. This is the stress inversion model learned in this step.

[0124] To ensure the physical rationality and structural consistency of the modeling, we The training objective function is designed with two constraints:

[0125] ;

[0126] Indicates the prediction result Compared with known deformation trends The fitting loss between them. Considering and Instead of a one-to-one mapping, we fit the stress gradient trend in the dominant deformation direction of the region by using a local average constraint weighted by point cloud density.

[0127] The spatial structure consistency loss, used to penalize drastic stress jumps between adjacent meshes, is defined as:

[0128] ;

[0129] in:

[0130] Indicates the region Inner adjacent grid point pairs;

[0131] The predicted points With point The stress value.

[0132] This structural regularization term ensures the spatial continuity of the predicted stress field, conforming to the "local rigid continuum" assumption in geotechnical mechanics, and avoids the model outputting physically unreliable extreme values ​​under under-constraint conditions.

[0133] In terms of model structure, Input can be: Multilayer sensing network structure:

[0134] in For the first The three-dimensional spatial coordinates of each point are derived from the model. Grid information in;

[0135] It is a region constant and can be incorporated into the input features of each point.

[0136] The model in each region It can be trained independently without cross-regional coupling. The training samples come from multiple time-series data of the same region in the historical monitoring period. Small data training is achieved by extracting sample pairs through a sliding window.

[0137] The output is for each Regional stress field distribution ,in This indicates the estimated stress value (in MPa) at this grid location.

[0138] Step 5: Dynamic response generation and 3D management visualization based on regional risk overlay modeling

[0139] This step is based on the slope deformation trend map output in step 3. Step 4: Inversion of the regional stress distribution and each area in step 2 physical properties In the structured slope model The regional risk scoring model was completed and mapped to a 3D visualization management platform for dynamic display and management of information overlay.

[0140] The task of this step is to integrate and model the three types of information—"structure + state + attributes"—to generate a score reflecting the current risk status of each region. Based on this, the attributes of the three-dimensional slope model are enhanced through visualization, thus realizing the basis for spatial perception and management linkage after intelligent risk identification.

[0141] First, we construct a region-oriented... The risk-overlapping scoring model integrates input variables from three sources:

[0142] : Predictive map of deep stress within the region;

[0143] : Normal deformation residual map;

[0144] Regional physical properties, including density (Unit: g / cm³), porosity (Percentage), coefficient of friction (Dimensionless).

[0145] The risk scoring function is designed as follows:

[0146] ;

[0147] in:

[0148] :area The risk score, with a normalized range, is calculated to... ;

[0149] :area The average stress value of all grids within the unit, in MPa;

[0150] The variance of the residual spectrum within the region represents the degree of dispersion of slope deformation;

[0151] The porosity of this region reflects the looseness of its structure;

[0152] Experience weight, example set to: , , ;

[0153] The Sigmoid function is used to limit the output range and smooth the response.

[0154] The design of this model takes into account the internal stress accumulation trend, the degree of uneven surface deformation, and material fragility, forming a comprehensive scoring framework that is highly integrated and interpretable.

[0155] After the risk scoring model is completed, the system will assign each Mapping to a 3D slope model Corresponding area The attribute fields form a "risk status layer" in the structured data. This layer does not change the geometric structure; it only serves as a data augmentation part of the 3D model. Specific mapping methods include:

[0156] exist In the data structure for each Add a field risk_score = S_i;

[0157] Synchronously record the scoring timestamp t to support multi-time model switching;

[0158] Mark the primary cause of the risk (e.g., source_type = 'deformation-dominated' or 'stress-dominated') to facilitate quick interpretation by managers later.

[0159] To support the platform's visual management and user interaction, we define the following spatial mapping strategy, which will... Explicitly presented in the 3D visualization platform:

[0160] Color encoding strategy:

[0161] Green (safe);

[0162] Yellow (Requires attention);

[0163] Orange (Warning);

[0164] Red (high risk); Color mapping is achieved through binding region mesh shader properties, and is rendered dynamically.

[0165] Transparency adjustment:

[0166] High-risk areas automatically have their transparency reduced (e.g., set to 40%~60%) to allow observation of their internal structural details;

[0167] The stable areas remain opaque to highlight the contrast;

[0168] Layer overlay design:

[0169] The risk layer is an independent overlay layer, decoupled from the basic slope model layer, and users can choose to turn it on or off.

[0170] The layer supports a "timeline" control, allowing users to view past data. The historical evolution of risk status is displayed hourly / daily, enabling time-series visual analysis.

[0171] Information card interaction:

[0172] User clicks on any area The platform pops up a data card, displaying the data for that area:

[0173] Current risk score ;

[0174] The past three moments Trend of change (arrow indicator + numerical value);

[0175] The physical parameters included ( );

[0176] Residual spectrum summary (e.g., maximum) );

[0177] Auxiliary indicators such as mean stress and proportion of high-stress mesh.

[0178] The technical solutions described in the embodiments of this application have at least the following technical effects or advantages: Using a 3D visualization platform as the core carrier, an adaptive, partitioned slope structure model is constructed, supporting differentiated monitoring strategies for different areas and enabling real-time capture of minute trend changes in slope deformation, generating deformation response maps with regional characteristics. Furthermore, through joint modeling of historical monitoring data and current dynamic changes, a prediction mechanism for the evolution of risks within the slope is established, achieving a cross-level monitoring closed loop from surface perception to deep understanding. This invention not only significantly improves the accuracy of identifying local anomalies but also enhances the response efficiency of the 3D system in assisting early warning decision-making and operational intervention.

[0179] This application also provides a mine slope monitoring system based on three-dimensional visualization management, such as... Figure 2 As shown, including;

[0180] The data acquisition module is configured to collect raw data from multiple sources, perform data alignment and registration on the raw data, and transform the processed data into a three-dimensional mesh model through a triangulation algorithm. Each mesh node of the three-dimensional mesh model is assigned a corresponding physical attribute vector, and a three-dimensional slope model is obtained.

[0181] The region division module is set to be based on the three-dimensional slope model. After obtaining geometric cluster labels and physical cluster labels through geometric feature clustering and physical attribute vector clustering respectively, cross matching is performed to construct structured regions on grid points with consistent labels. The structured region set is integrated and output and updated to the slope model to obtain the regionalized three-dimensional slope model.

[0182] The deformation recognition module is configured to construct a regional average slope normal vector in a structured region. The regional average slope normal vector is obtained by weighted averaging of the normal vectors of all triangular facets in the region, with the weight being the facet area. The module calculates the projection residual of all matching points in the structured region along the direction of the regional average slope normal vector, and calculates the deformation heat value of the corresponding point through the projection residual. A dynamic weighting mechanism is introduced, and different residual enhancement strategies are adopted based on the different physical characteristics of different regions to obtain the deformation map value of each point. The module then aggregates and outputs the residual heat map spectrum of each region.

[0183] The stress inversion module is set to take into account the residual thermal spectrum, regional physical properties, and spatial structure information of the three-dimensional slope model for each region, and to establish a deep stress distribution model for each region. The deep stress distribution model is constructed by using the residual thermal spectrum as a supervision signal within the region to build a mapping function and complete the fitting modeling of the stress distribution of the regional state terms.

[0184] The scoring and output module is set to take into account the residual thermal map, deep stress distribution model and physical properties of each region. Based on the regionalized three-dimensional slope model, it completes the regional risk score modeling and maps the risk score to the visualization platform. The regional risk score is calculated by assigning the importance of variables with empirical weights, and the linear weighted sum is compressed to achieve smooth normalization to obtain the risk score.

[0185] The working principle is as follows: Using a 3D visualization platform as the core carrier, an adaptive partitioned slope structure model is constructed. Through joint modeling of historical monitoring data and current dynamic changes, a prediction mechanism for the evolution of risks within the slope is established.

[0186] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A mine slope monitoring method based on three-dimensional visualization management, characterized in that, The method comprises the following steps: Collecting multi-source original data, aligning and registering the original data, and converting the processed data into a three-dimensional grid model through a triangulation algorithm, and assigning a corresponding physical attribute vector to each grid node of the three-dimensional grid model to obtain a three-dimensional slope model; Based on the three-dimensional slope model, geometric feature clustering and physical attribute vector clustering are performed to obtain geometric clustering labels and physical clustering labels, respectively, cross matching is performed, and a structured region is constructed at the grid points where the labels are consistent, the structured region set is integrated and updated to the slope model to obtain a regionalized three-dimensional slope model; wherein the structured region comprises a grid set, and each grid has a position and a corresponding regional physical attribute; A regional average slope normal vector is constructed in the structured region, and the regional average slope normal vector is obtained by weighted average of all triangular facet normal vectors in the region, and the weight is the facet area; the projection residual of all matching points in the structured region in the direction of the regional average slope normal vector is calculated, and the deformation heat value of the corresponding point is calculated through the projection residual, a dynamic weighting mechanism is introduced, different residual enhancement strategies are adopted based on different physical characteristics of different regions, and a deformation map value of each point is obtained, and finally a residual heat map of each region is aggregated and output; The residual heat map of each region, the regional physical attribute, and the spatial structure information of the three-dimensional slope model are input to establish a deep stress distribution model for each region, wherein the deep stress distribution model is constructed by taking the residual heat map as a supervision signal in the region to construct a mapping function, and the fitting modeling of the stress distribution from the regional state item is completed; The residual heat map of each region, the deep stress distribution model, and the physical attribute are input, the regional risk score modeling is completed based on the regionalized three-dimensional slope model, and the risk score is mapped to a visualization platform; wherein the regional risk score is obtained by calculating the linear weighted sum through the experience weight allocation variable importance, and the linear weighted sum is compressed to realize smooth normalization to obtain the risk score.

2. The mine slope monitoring method based on three-dimensional visualization management according to claim 1, characterized in that, The physical attribute vector assigns a physical attribute to each grid point based on known physical data through an interpolation method, and the physical attribute includes rock-soil density, friction coefficient, and porosity.

3. The mine slope monitoring method based on three-dimensional visualization management according to claim 1, characterized in that, The geometric features include a slope value and roughness, the slope value is calculated from the included angle between the grid facet normal vector and the vertical direction of the ground, and the roughness is calculated by averaging the normal vector difference between the grid and its adjacent grid.

4. The mine slope monitoring method based on three-dimensional visualization management according to claim 1, characterized in that, The deformation heat value calculation introduces a neighborhood smoothing difference regularization term, and through neighborhood consistency analysis, the true anomaly and isolated error are distinguished.

5. The mine slope monitoring method based on three-dimensional visualization management according to claim 1, characterized in that, The deep stress distribution model adopts a multi-layer perception grid structure, which is trained separately in each region, and the training data comes from multiple time series data of the same region in a monitoring period, and the sample is extracted through a sliding window to realize small data training.

6. The mine slope monitoring method based on three-dimensional visualization management according to claim 1, characterized in that, The risk score is mapped to the visualization platform by mapping each risk score to the attribute field of the corresponding region in the three-dimensional slope model, specifically: A risk score is added to each corresponding region in the three-dimensional slope model; the scoring time is recorded synchronously and the risk source is marked.

7. An electronic device, comprising: A computer program product, comprising a memory and a processor, wherein the processor executes the program to implement the method according to any one of claims 1-6.

8. A mine slope monitoring system based on three-dimensional visualization management, characterized in that, Comprise; The data acquisition module is configured to acquire multi-source original data, perform data alignment and registration on the original data, and convert the processed data into a three-dimensional grid model through a triangulation algorithm, assign a corresponding physical attribute vector to each grid node of the three-dimensional grid model, and obtain a three-dimensional slope model; The region division module is configured to obtain geometric clustering labels and physical clustering labels through geometric feature clustering and physical attribute vector clustering based on the three-dimensional slope model, perform cross matching, construct a structured region on the grid points with consistent labels, integrate and output a structured region set, and update the structured region set to the slope model to obtain a regionalized three-dimensional slope model; The deformation identification module is configured to construct a regional average slope normal vector in the structured region, the regional average slope normal vector being obtained by weighted average of all triangular facet normal vectors in the region, and the weight being a facet area; calculate the projection residual of all matching points in the structured region in the direction of the regional average slope normal vector, and calculate the deformation heat value of the corresponding points through the projection residual; introduce a dynamic weighting mechanism, adopt different residual enhancement strategies based on different physical characteristics of different regions, obtain the deformation map value of each point, and aggregate and output the residual heat map of each region; The stress inversion module is configured to input the residual heat map of each region, the regional physical attribute, and the spatial structure information of the three-dimensional slope model, establish a deep stress distribution model of each region, wherein the deep stress distribution model is constructed by taking the residual heat map as a supervision signal in the region to construct a mapping function, and complete the fitting modeling of the stress distribution from the regional state item; The scoring and output module is configured to input the residual heat map of each region, the deep stress distribution model, and the physical attribute, complete the regional risk score modeling based on the regionalized three-dimensional slope model, and map the risk score to a visualization platform; wherein the regional risk score assigns importance to variables through an empirical weight, calculates a linear weighted sum, and compresses the linear weighted sum to realize smooth normalization to obtain the risk score.

Images