Coal mine underground surrounding rock deformation monitoring method and system based on three-dimensional point cloud

Abstract

This invention provides a method and system for monitoring surrounding rock deformation in underground coal mines based on three-dimensional point clouds, relating to the field of coal mine safety monitoring. The invention includes the following steps: installing a reference target in the monitoring area and collecting point cloud data; preprocessing the point cloud data, performing coarse registration using the reference target and fine registration using the iterative nearest point algorithm, and generating a triangular mesh model using the Poisson surface reconstruction algorithm; parameterizing the triangular mesh model of the monitoring area into a two-dimensional analysis mesh, calculating the three-dimensional displacement vector of each node, and establishing a displacement field matrix; calculating the displacement gradient tensor based on the displacement vector, deriving the Green-Lagrange strain tensor, and obtaining the full-field strain distribution; recording the strain value of each mesh node, and determining whether an instability warning is triggered based on the relationship between the strain value and the warning threshold; this method can achieve high-precision monitoring of surrounding rock deformation in underground coal mines.

Description

Technical Field

[0001] This invention relates to the field of coal mine safety monitoring, specifically to a method and system for monitoring surrounding rock deformation in underground coal mines based on three-dimensional point clouds. Background Technology

[0002] As coal mining depths continue to increase, the stability of the surrounding rock underground becomes an increasingly prominent issue, making surrounding rock deformation monitoring a crucial link in ensuring safe coal mine production. Traditional coal mine surrounding rock deformation monitoring mainly relies on manual measurement and single-point monitoring equipment, such as convergence meters and displacement meters. While these methods are widely used, they suffer from limitations such as limited monitoring points, discrete data, and the inability to obtain comprehensive deformation information, making it difficult to meet the needs of modern coal mine safety production.

[0003] In recent years, with the development of 3D laser scanning technology, rock deformation monitoring methods based on point cloud data have been gradually applied to underground coal mine environments. For example, the prior art disclosed in CN114155245A is a method for monitoring rock deformation in underground coal mines based on 3D point clouds. This method drives a laser radar to move along the tunnel axis, performs multiple measurements at a preset benchmark reset point, collects 3D point cloud data of the tunnel cross-section, and generates a point cloud data matrix of the tunnel end face, thereby achieving accurate monitoring and calculation of the deformation of the surrounding rock in the underground tunnel. Compared with traditional monitoring methods, this method can obtain more comprehensive information about the surrounding rock surface. The prior art disclosed in CN118293811A proposes a mobile high-precision measurement and analysis method for tunnel convergence deformation. This method uses a mobile 3D laser scanner to collect and monitor tunnel point cloud data, automatically extracts the coordinates of the center point of the measurement marker panel, calculates the distance change between monitoring points of the monitoring cross-section deformation, and obtains the convergence deformation data of the tunnel surrounding rock. This method offers flexibility in on-site operation, efficient data acquisition, and minimal impact from construction interference. However, it primarily focuses on distance changes between pre-set monitoring points and fails to fully utilize point cloud data for comprehensive deformation analysis. Existing technology, with publication number CN119321729A, proposes a lidar-based method for monitoring roof deformation in mine working faces. This method constructs a triangular mesh model from two phases of scanned point clouds and extracts the center point of each triangular face in the second phase model to form a second-phase center point cloud. The distance between the second-phase center point cloud and the first-phase triangular mesh model is calculated as the basis for deformation and deformation risk assessment. This method utilizes non-destructive testing technology, reducing labor costs and workload and improving detection safety. However, it mainly targets roof deformation and has limitations in monitoring the overall deformation of the surrounding rock.

[0004] However, existing methods for monitoring surrounding rock deformation based on 3D point clouds still suffer from the following problems: First, the registration accuracy of point cloud data is insufficient, especially in complex underground environments where the lack of effective reference markers leads to large registration errors. Second, existing methods focus primarily on the geometric changes in point cloud data, lacking in-depth analysis of deformation mechanisms and making it difficult to extract key mechanical parameters such as strain fields. Third, the deformation early warning mechanism is incomplete, lacking multi-level early warning standards based on strain and strain rate, making it difficult to detect potential surrounding rock instability risks in a timely manner. Finally, existing methods still have room for improvement in point cloud data processing efficiency, failing to meet the needs of real-time monitoring. These problems limit the application effectiveness of 3D point cloud technology in monitoring surrounding rock deformation in coal mines, hindering accurate assessment and timely early warning of surrounding rock stability, and impacting safe coal mine production.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for monitoring the deformation of surrounding rock in underground coal mines based on three-dimensional point clouds, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: A method for monitoring surrounding rock deformation in underground coal mines based on three-dimensional point clouds, comprising the following steps: S1: Determine the monitoring area by leveling the surrounding rock surface in the coal mine, install a reference target in the monitoring area, set two monitoring times, the first time and the second time, and collect point cloud data of the surrounding rock surface and the reference target in the monitoring area at the two monitoring times respectively, and label them as the first point cloud data and the second point cloud data. Preprocess the collected first point cloud data and the second point cloud data. S2: Coarse registration is performed on the preprocessed first and second point cloud data by referencing the target, fine registration is performed by iterating the nearest point algorithm, and a triangular mesh model of the monitoring area is generated by using the Poisson surface reconstruction algorithm on the registered first and second point cloud data. S3: Parameterize the triangular mesh model of the monitoring area into a two-dimensional analysis mesh, assign three-dimensional coordinates to each mesh node based on the point cloud data in the mesh node's neighborhood, compare the coordinates of the point cloud mesh nodes at the second time step with the coordinates of the point cloud mesh nodes at the first time step, calculate the three-dimensional displacement vector of each node, and establish a complete displacement field matrix. S4: Based on the three-dimensional displacement vector of each grid node, calculate the displacement gradient tensor, derive the Green-Lagrange strain tensor, and obtain the strain distribution across the entire field; S5: Record the strain value of each grid node, and determine whether to trigger an instability warning based on the relationship between the strain value and the set strain warning threshold and strain rate warning threshold.

[0008] Point cloud data of the surrounding rock surface and reference target within the monitoring area were collected using a ground-based laser scanner. The first point cloud data was recorded as [data point 1], and the point cloud data collected at the second time point was recorded as the second point cloud data. The point cloud data included the three-dimensional coordinates of each point. A multi-level filtering algorithm was used to preprocess the collected point cloud data of the first and second time points. The preprocessing first performed statistical outlier filtering and calculated the coordinates of each point to the reference target. The formulas for the mean distance of the nearest neighbors, the global distance mean, and the standard deviation are as follows: in, The global distance mean. The global distance standard deviation, This represents the total number of points in the point cloud. For the first Point to The average distance of the nearest neighbors is used to remove all points that satisfy this condition. The outliers are identified; next, voxel mesh downsampling is performed; the point cloud space is divided into a uniform voxel mesh, the voxel size is set according to the point density, and all points within each voxel are replaced by the centroid: in, Let the coordinates be the centroid coordinates of the voxel. The number of points within a single voxel. Point cloud data points within the same voxel mesh The three-dimensional coordinate vector.

[0009] Furthermore, the coarse registration is performed using a reference target, which is a spherical target. By identifying the reference target in the point cloud at the first and second time moments, the rigid body transformation matrix of the target center is calculated. For the target point set, the spherical equation is fitted using the least squares method, as shown in the following formula: in,( () is the reference target center coordinate. The reference target radius is used as a reference. After obtaining the coordinates of the reference target center, fine registration is performed using the iterative nearest-point algorithm, with the objective function being: in, For the iterative nearest point registration error function, To be the optimal rotation matrix, It is a translation vector. For weighting coefficients, the second point cloud data The first in The coordinates of each point are recorded as follows: , Initial reference point cloud Zhongyu Corresponding points The number of point pairs establishing correspondences in point cloud registration; the optimal transformation is solved using singular value decomposition: in, This is the result of the second point cloud SVD decomposition. It is a left singular vector matrix. It is a singular value diagonal matrix. It is a right singular vector matrix. The average position of all points in the second point cloud data. The average position of all points in the point cloud data at the first time step; define the nearest point convergence threshold for iteration, iterate until the registration error is less than the nearest point convergence threshold, and complete the fine registration; the second point cloud after fine registration. .

[0010] Furthermore, based on the registered first and second point cloud data, a triangular mesh model of the monitoring area is generated using the Poisson surface reconstruction algorithm, and the Poisson equation is solved: By discretizing the space using an octree structure, the continuous equations are transformed into a system of linear equations, as shown in the following formula: in, For the Laplace operator, As a gradient operator, the finely registered point cloud generates a steering field. For is the discrete Laplace matrix, Let be the unknown vector to be solved. For discrete divergence term vectors, based on Reconstruct a continuous indicator function field ,extract The isosurface generates a triangular network model.

[0011] Furthermore, the triangular mesh models of the monitoring area at the first and second time points are parameterized into the same two-dimensional analysis mesh, using a specific mesh spacing. A uniform grid is used, and the grid nodes are determined by surface projection: For each grid node, its nearest point on the triangular grid is found, and its accurate 3D coordinates are obtained by interpolation using the barycentric coordinates. in, In a two-dimensional analysis grid The three-dimensional coordinates of the node. The vertex of the triangle, Let the weights be the centroid coordinates, and satisfy the following conditions: Displacement calculation uses the direct comparison method, for each grid node. The formula for calculating the three-dimensional displacement vector is as follows: in, To analyze the grid spacing, For the first The three-dimensional displacement vector of each node. This represents the displacement component of a point along the X-axis. This represents the displacement component of the point along the Y-axis. This represents the displacement component of the point along the Z-axis. For the second point cloud data Node coordinates For the first point cloud data The coordinates of each node are used to perform error compensation, as follows: in, For the revised first Displacement vectors at each node For the first The three-dimensional displacement vector of each node. For the first The three-dimensional displacement vector of a reference target. The total number of reference targets.

[0012] Furthermore, S5 describes the calculation of the displacement gradient tensor. The formula is as follows: in, For unit tensors, For the first point cloud data Node coordinates This refers to the displacement gradient; the displacement gradient is calculated numerically using the central difference method; the displacement gradient tensor is calculated, and the Green-Lagrange strain tensor is derived, as shown in the following formula: in, For the first The strain tensor at each node is expanded to obtain the strain components, thus yielding the overall strain distribution. The principal strains are obtained by solving the eigenvalue problem. , The eigenvalues ​​are used to obtain the maximum principal strain. and minimum principal strain .

[0013] Furthermore, the strain value of each grid node is recorded, and strain warning thresholds and strain rate warning thresholds are set based on engineering safety standards. The strain warning threshold is... and strain rate warning threshold The formula for calculating strain rate is as follows: in, For the first The strain rate at each node, For the first The maximum principal strain at each node The time interval between the first and second moments; a multi-level dynamic threshold mechanism is adopted: the strain of a single grid point exceeds... This triggers a primary deformation warning; strain at three or more consecutive adjacent grid points exceeds [a certain value]. This triggers a mid-level deformation warning; the strain rate exceeds [a certain threshold]. Or the strain of four or more consecutive adjacent grid points exceeds This triggers an advanced deformation warning.

[0014] This invention also provides a coal mine underground rock deformation monitoring system based on three-dimensional point clouds. The coal mine underground rock deformation monitoring system based on three-dimensional point clouds is used to execute the above-mentioned coal mine underground rock deformation monitoring method based on three-dimensional point clouds, including: Point cloud data acquisition module: used to determine the monitoring area by leveling the surrounding rock surface in the coal mine, install reference targets in the monitoring area, set two monitoring times, the first time and the second time, and collect point cloud data of the surrounding rock surface and the reference target in the monitoring area at the two monitoring times respectively, and label them as the first point cloud data and the second point cloud data. The collected first point cloud data and the second point cloud data are preprocessed. Point cloud registration and reconstruction module: used to perform coarse registration of the preprocessed first and second point cloud data with reference target, fine registration of the iterative nearest point algorithm, and generate a triangular mesh model of the monitoring area using the Poisson surface reconstruction algorithm for the registered first and second point cloud data. Mesh Displacement Field Calculation Module: This module parameterizes the triangular mesh model of the monitoring area into a two-dimensional analysis mesh. Each mesh node is assigned three-dimensional coordinates. The coordinates of the point cloud mesh nodes at the second time step are compared with the coordinates of the point cloud mesh nodes at the first time step. The three-dimensional displacement vector of each node is calculated, and a complete displacement field matrix is ​​established. Full-field strain analysis module: used to calculate the displacement gradient tensor based on the three-dimensional displacement vector of each grid node, derive the Green-Lagrange strain tensor, and obtain the full-field strain distribution; Deformation early warning and identification module: used to record the strain value of each grid node, set strain early warning threshold and strain rate early warning threshold, and use a multi-level dynamic threshold mechanism to trigger instability early warning.

[0015] Compared with the prior art, the beneficial effects of the present invention are: Real-time monitoring of surface deformation of surrounding rock in underground coal mines has been achieved, improving the timeliness of monitoring; high-precision and comprehensive monitoring of surface deformation of surrounding rock has been achieved through three-dimensional point cloud technology, resulting in higher detection accuracy; multi-level filtering algorithms and coarse registration followed by fine registration have been adopted to improve the accuracy and reliability of monitoring data; displacement field and strain field analysis can comprehensively assess the stability of surrounding rock; and a multi-level dynamic threshold early warning mechanism has been established to achieve early warning of surrounding rock instability, thereby improving the level of safe production in coal mines. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall method flow of the present invention; Figure 2 This is a diagram of the displacement field data after node correction in an embodiment of the present invention; Figure 3 This is a diagram showing the grid node coordinates and displacement data in an embodiment of the present invention; Figure 4 This is a schematic block diagram of the coal mine underground surrounding rock deformation monitoring system based on three-dimensional point cloud according to the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0018] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0019] Example: Please see Figures 1 to 3 The present invention provides a technical solution: A method for monitoring surrounding rock deformation in underground coal mines based on three-dimensional point clouds, comprising the following steps: S1: Determine the monitoring area by leveling the surrounding rock surface in the coal mine, install a reference target in the monitoring area, set two monitoring times, the first time and the second time, and collect point cloud data of the surrounding rock surface and the reference target in the monitoring area at the two monitoring times respectively, and label them as the first point cloud data and the second point cloud data. Preprocess the collected first point cloud data and the second point cloud data. In this embodiment, four spherical reference targets are securely installed within the monitoring area. The targets should be distributed in the corners of the monitoring area, and their positions should be kept absolutely fixed between scans. A ground-based laser scanner is used to scan the entire monitoring area, including the surrounding rock surface and the reference targets, at each monitoring moment. The scanning resolution and point density are ensured to meet the accuracy requirements of deformation monitoring. The data collected at the first moment is calibrated as the first point cloud data. The data collected at the second moment is labeled as the second point cloud data. All data is given precise timestamps and stored in a dedicated database for management; the core is to obtain the 3D coordinates of each point to form a point cloud set. and The reference target provides high-precision corresponding feature points for subsequent point cloud registration, overcoming the registration difficulties caused by the indistinct or changing features of the surrounding rock surface. By monitoring the displacement of the target itself, systematic errors caused by slight movement of the scanner can be effectively identified and compensated, which is a key part of this solution to achieve high-precision monitoring.

[0020] In this embodiment, a multi-level filtering strategy combining "statistical outlier filtering" and "voxel grid downsampling" is used to preprocess the point cloud data. First, the first batch of collected point cloud data is processed. With the second point cloud data The point cloud data is preprocessed using a multi-level filtering algorithm. The preprocessing first performs statistical outlier filtering and calculates the distance from each point to the target value. The formulas for the mean distance of the nearest neighbors, the global distance mean, and the standard deviation are as follows: in, The global distance mean. The global distance standard deviation, This represents the total number of points in the point cloud. For the first Point to The average distance of the nearest neighbors is used to remove all points that satisfy this condition. Outliers; in this embodiment, the nearest outlier is searched. The nearest neighbor point, In three-dimensional space, find the distance Recent Calculate points. Calculate the arithmetic mean of the distances to each of its nearest neighbors to obtain... ; should be all in the point cloud Each point was calculated. Then, a set of global statistical characteristics was obtained. ; It is the mean of the average nearest neighbor distances of all points, which represents the average density of the point cloud as a whole. The standard deviation of the average nearest neighbor distance of all points represents the dispersion of the point cloud density distribution. Points far from the main distribution are outliers. A threshold is set, and if... Then consider the point It is an outlier and is removed from the point cloud. Statistical outlier filtering can intelligently identify and remove flying point noise without damaging the true surface features.

[0021] Next, voxel mesh downsampling is performed; the point cloud space is divided into a uniform voxel mesh, the voxel size is set according to the point density, and all points within each voxel are replaced by centroid points: in, Let the coordinates be the centroid coordinates of the voxel. The number of points within a single voxel. Point cloud data points within the same voxel mesh The algorithm iterates through all voxel meshes in the entire point cloud space and calculates a 3D coordinate vector for each voxel. After that, all of these These points constitute a new, downsampled point cloud dataset; this new point cloud set is the preprocessed point cloud data. In this embodiment, the point cloud space is divided into a series of segments with side lengths of... voxels; voxel size Based on the required precision and point density settings, traverse each voxel. If a voxel contains multiple points, calculate the centroid coordinates of these points and replace all the original points within that voxel with this single centroid.

[0022] Using the centroid best represents the average position of points within that small region, thus smoothing and downsampling. Voxel size This is the most critical parameter in this step. The larger the size, the smaller the data volume, but the more details are lost; The smaller the voxel size, the richer the details preserved, but the worse the data compression effect. Voxel downsampling significantly reduces the amount of point cloud data while preserving the original geometry to the maximum extent, thus improving the speed of subsequent processing. The original scanned point cloud contains millions or even tens of millions of points, many of which are redundant. Voxel meshing divides the space into small cubes, and regardless of how many original points are in each voxel, only one is ultimately used. This reduces the computational burden; at the same time, by calculating the centroid of all points within a voxel, random noise can be averaged out to some extent; noise points usually deviate from the true surface, and taking the centroid can make the new points... It is closer to the theoretical actual surface location.

[0023] S2: Coarse registration is performed on the preprocessed first and second point cloud data by referencing the target, fine registration is performed by iterating the nearest point algorithm, and a triangular mesh model of the monitoring area is generated by using the Poisson surface reconstruction algorithm on the registered first and second point cloud data. In this embodiment, coarse registration is performed using a reference target, which is a spherical target. By identifying the reference target in the point cloud at the first and second time moments, the rigid transformation matrix of the target center is calculated. For the target point set, the spherical equation is fitted using the least squares method, as shown in the following formula: in,( () is the reference target center coordinate. For reference target radius; in this embodiment, respectively in the point cloud and The point cloud clusters of all reference targets are identified in the second time point cloud data. The point cloud data of the spherical target is fitted by the least squares method to solve the coordinates of its sphere center in the current scanning coordinate system. Based on the target sphere center coordinates in the first time point cloud coordinate system and the second time point cloud coordinate system of at least three non-collinear targets, a preliminary rotation matrix and translation vector are calculated to roughly align the second time point cloud to the first time point cloud coordinate system, thus completing the coarse registration.

[0024] After obtaining the coordinates of the reference target center, fine registration is performed using the iterative nearest-point algorithm, with the objective function being: in, For the iterative nearest point registration error function, To be the optimal rotation matrix, It is a translation vector. For weighting coefficients, the second point cloud data The first in The coordinates of each point are recorded as follows: , Initial reference point cloud Zhongyu Corresponding points The number of point pairs establishing correspondences in point cloud registration; the optimal transformation is solved using singular value decomposition: in, This is the result of the second point cloud SVD decomposition. It is a left singular vector matrix. It is a singular value diagonal matrix. It is a right singular vector matrix. The average position of all points in the second point cloud data. Let be the average position of all points in the point cloud data at the first moment; define an iterative nearest point convergence threshold, iterating until the registration error is less than the iterative nearest point convergence threshold. In this embodiment, the preset nearest point convergence threshold is . The fine registration was completed at cm, and the second point cloud was obtained after fine registration. SVD decomposition can efficiently find the optimal solution. In essence, it involves passing the second cloud through... Rotate to align the main distribution directions of the two point clouds; through Translate the points so that the centers of the two point clouds coincide; align the centers and directions to complete the fine registration.

[0025] In this embodiment, based on the registered first and second point cloud data, a triangular mesh model of the monitoring area is generated using the Poisson surface reconstruction algorithm, and the Poisson equation is solved: By discretizing the space using an octree structure, the continuous equations are transformed into a system of linear equations, as shown in the following formula: in, For the Laplace operator, As a gradient operator, the finely registered point cloud generates a steering field. For is the discrete Laplace matrix, Let be the unknown vector to be solved. For discrete divergence term vectors, based on Reconstruct a continuous indicator function field ,extract The isosurface is used to generate a triangular network model. In this embodiment, the first time-series point cloud after fine registration is used. Second time point cloud division Instead of using the point cloud as input, adaptively partition the 3D space containing the point cloud and construct an octree. For each node in the octree, calculate a vector field pointing inwards and outwards based on the normal vectors of its neighboring point clouds. Constructing and solving the Poisson equation: The core idea of ​​the Poisson equation is: indicator function gradient field It should be equal to the vector field defined by the point cloud normal vector. divergence Therefore, the equation is established. After discretization using an octree, the partial differential equation is transformed into a large system of linear equations. Solving the system of linear equations yields the indicator function field. The values ​​at each node. Then, using algorithms such as the moving cube algorithm, the values ​​are extracted. The isosurface. This isosurface is the final generated triangular mesh model that can completely wrap around the original point cloud.

[0026] This step converts the registered discrete point cloud into a continuous triangular mesh model, explicitly defining the surface geometry and topological relationships, which serves as the sole basis for subsequent mesh generation in analysis. The algorithm treats the point cloud as an indicator function. By solving the Poisson equation, this implicit surface is reconstructed. Compared to other algorithms, Poisson reconstruction handles point clouds with noise and missing data better, making it ideal for irregular surfaces like surrounding rock. It also generates high-quality, watertight meshes, providing a perfect geometric basis for strain analysis. The most critical parameter is the depth of the octree. Greater depth results in finer spatial partitioning, a larger linear equation system, and slower computation, but also richer model detail, potentially even reconstructing noisy details. Conversely, shallower depth leads to faster computation, but an overly smooth model that may lose realistic geological features. Technicians need to strike a trade-off between point cloud density and the desired level of detail.

[0027] S3: Parameterize the triangular mesh model of the monitoring area into a two-dimensional analysis mesh, assign three-dimensional coordinates to each mesh node based on the point cloud data in the mesh node's neighborhood, compare the coordinates of the point cloud mesh nodes at the second time step with the coordinates of the point cloud mesh nodes at the first time step, calculate the three-dimensional displacement vector of each node, and establish a complete displacement field matrix. The triangular mesh models of the monitoring area at the first and second time points are parameterized into the same two-dimensional analysis mesh, using the mesh spacing. A uniform grid is used, and the grid nodes are determined by surface projection: For each grid node, its nearest point on the triangular grid is found, and its accurate 3D coordinates are obtained by interpolation using the barycentric coordinates. in, In a two-dimensional analysis grid The three-dimensional coordinates of the node. The vertex of the triangle, Let the weights be the centroid coordinates, and satisfy the following conditions: Two 3D mesh models from different time points are parameterized onto the same 2D analysis mesh, thus achieving a one-to-one correspondence between model nodes at different times. Because of this node correspondence, displacement calculation is simplified to simple vector subtraction, resulting in a clear concept and high computational efficiency. Correcting the overall field displacement by referencing the target displacement is a very ingenious design. It identifies and eliminates scanner system errors or overall rigid body displacement, ensuring that the final displacement field purely reflects the deformation of the surrounding rock itself. In this embodiment, a regular 2D mesh is defined on the horizontal projection plane of the monitoring area, with a mesh spacing of [missing information]. This grid will serve as a unified benchmark for every node on the two-dimensional grid. Project it vertically onto the 3D triangular mesh model, find which triangle the projection point falls within, and use the three vertices of that triangle... The coordinates of the projection point are accurately calculated using barycentric coordinate interpolation. This coordinate is then assigned to the 2D mesh node. At this point, each 2D mesh node possesses a precise 3D coordinate.

[0028] Displacement calculation uses the direct comparison method, for each grid node. The formula for calculating the three-dimensional displacement vector is as follows: in, To analyze the grid spacing, For the first The three-dimensional displacement vector of each node. This represents the displacement component of a point along the X-axis. This represents the displacement component of the point along the Y-axis. This represents the displacement component of the point along the Z-axis. For the second point cloud data Node coordinates For the first point cloud data The coordinates of each node are used to perform error compensation, as follows: in, For the revised first Displacement vectors at each node For the first The three-dimensional displacement vector of each node. For the first The three-dimensional displacement vector of a reference target. To establish the total number of reference targets, a complete displacement field matrix is ​​constructed: in, for Directional displacement field matrix, for Directional displacement field matrix, for Directional displacement field matrix, To analyze the number of rows and columns of the grid.

[0029] In this embodiment, for each grid node Using the coordinates at the second moment Subtract the coordinates at the first moment Obtain the original displacement vector Calculate the displacement vector of the center point of all reference targets. Calculate the average value of these target displacements; this average value is considered as the global systematic error for the original displacement of each grid node. Subtracting this systematic error yields the corrected displacement vector that truly represents the deformation of the surrounding rock. In this embodiment, a 10m×10m roadway roof monitoring area was established with a monitoring grid spacing of 0.5m, totaling 441 monitoring points. Fifteen nodes were selected as examples, and the corrected displacement field data are shown in Table 1 below: Table 1: Displacement field data after nodal correction See Figure 2 , Figure 2 , Figure 3 The table clearly displays the coordinates and displacement data of 15 example mesh nodes in Table 1. In Table 1, X represents the X-axis coordinate of the node, and Y represents the Y-axis coordinate of the node. The Z-coordinate at the first moment. The Z-coordinates at the second moment, as shown in Table 1, indicate that all displacements have been corrected using the average displacement of the reference target, eliminating systematic errors; the displacement is mainly settlement in the Z-direction. and The directional displacement is very small.

[0030] S4: Based on the three-dimensional displacement vector of each grid node, calculate the displacement gradient tensor, derive the Green-Lagrange strain tensor, and obtain the strain distribution across the entire field; In this embodiment, S5 describes the calculation of the displacement gradient tensor. The formula is as follows: in, For unit tensors, For the first point cloud data Node coordinates This refers to the displacement gradient; the displacement gradient is calculated numerically using the central difference method; the displacement gradient tensor is calculated, and the Green-Lagrange strain tensor is derived, as shown in the following formula: in, For the first The strain tensor of each node is expanded to obtain the strain components, and the strain distribution of the whole field is obtained. The principal strain is obtained by solving the eigenvalue problem. , The eigenvalues ​​are used to obtain the maximum principal strain. and minimum principal strain This tensor is defined based on the change in the square of the length of the infinitesimal line segment before and after deformation. A major advantage is its independence from rigid body rotation. That is, regardless of the object's rotation, its strain value remains constant. This is crucial for analyzing surrounding rock that has undergone complex deformation. The formula... This is to ensure that the definition is consistent with the engineering strain definition under small strain conditions. The spacing of the analytical mesh is used when calculating the displacement gradient. This is crucial. A mesh that is too coarse will fail to reflect drastic gradient changes and underestimate strain; a mesh that is too fine will be extremely sensitive to noise in the displacement field, resulting in the calculation of unrealistically large strains. An appropriate mesh density needs to be selected based on the smoothness of the displacement field and the noise level. In this example, 15 nodes in the selected area are used as an example, and the corrected displacements are shown in Table 2 below: Table 2: Mesh Node Coordinates and Displacement Data See Figure 3 , Figure 3 The table clearly shows the normal strain components of 15 example nodes. Table 2 shows that all nodes are subjected to compressive strain in the vertical direction, directly corresponding to the roof settlement phenomenon; the larger the value, the more severe the compression; tensile strain appears in the horizontal direction. This indicates that the roof has undergone lateral expansion or bending during the settlement process; except for node J237, the other 14 nodes have triggered the primary warning, indicating that the roof has undergone large-scale micro-deformation. As an early warning indicator, it simplifies the judgment logic.

[0031] S5: Record the strain value of each grid node, and determine whether to trigger an instability warning based on the relationship between the strain value and the set strain warning threshold and strain rate warning threshold.

[0032] In this embodiment, the strain value of each grid node is recorded, and a strain warning threshold and a strain rate warning threshold are set based on engineering safety standards. The strain warning threshold is... and strain rate warning threshold The formula for calculating strain rate is as follows: in, For the first The strain rate at each node, For the first The maximum principal strain at each node The time interval between the first and second moments; a multi-level dynamic threshold mechanism is adopted: the strain of a single grid point exceeds... This triggers a primary deformation warning; strain at three or more consecutive adjacent grid points exceeds [a certain value]. This triggers a mid-level deformation warning; the strain rate exceeds [a certain threshold]. Or the strain of four or more consecutive adjacent grid points exceeds This triggers an advanced deformation warning. In this embodiment, In this embodiment, a single, isolated point exceeding the limit may be noise and will not trigger a high-level warning. Requiring multiple consecutive adjacent points to exceed the limit indicates that deformation has reached a certain scale, significantly increasing reliability. Introducing a strain rate threshold allows for the identification of the more dangerous symptom of "accelerated deformation" before the absolute strain value reaches a dangerous level, enabling early warning. Primary, intermediate, and advanced warnings provide action guidelines. This embodiment performs the above operations based on the data in S4, and includes warning triggering status data as shown in Table 3 below: Table 3: Warning Trigger Status Data Table Please see Figure 4 The present invention also provides a coal mine underground rock deformation monitoring system based on three-dimensional point clouds. This system is used to execute the aforementioned coal mine underground rock deformation monitoring method based on three-dimensional point clouds, and includes: Point cloud data acquisition module: used to determine the monitoring area by leveling the surrounding rock surface in the coal mine, install reference targets in the monitoring area, set two monitoring times, the first time and the second time, and collect point cloud data of the surrounding rock surface and the reference target in the monitoring area at the two monitoring times respectively, and label them as the first point cloud data and the second point cloud data. The collected first point cloud data and the second point cloud data are preprocessed. Point cloud registration and reconstruction module: used to perform coarse registration of the preprocessed first and second point cloud data with reference target, fine registration of the iterative nearest point algorithm, and generate a triangular mesh model of the monitoring area using the Poisson surface reconstruction algorithm for the registered first and second point cloud data. Mesh Displacement Field Calculation Module: This module parameterizes the triangular mesh model of the monitoring area into a two-dimensional analysis mesh. Each mesh node is assigned three-dimensional coordinates. The coordinates of the point cloud mesh nodes at the second time step are compared with the coordinates of the point cloud mesh nodes at the first time step. The three-dimensional displacement vector of each node is calculated, and a complete displacement field matrix is ​​established. Full-field strain analysis module: used to calculate the displacement gradient tensor based on the three-dimensional displacement vector of each grid node, derive the Green-Lagrange strain tensor, and obtain the full-field strain distribution; Deformation early warning and identification module: used to record the strain value of each grid node, set strain early warning threshold and strain rate early warning threshold, and use a multi-level dynamic threshold mechanism to trigger instability early warning.

[0033] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0034] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0035] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0036] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A method for monitoring surrounding rock deformation in underground coal mines based on three-dimensional point clouds, characterized in that, The specific steps include: S1: Determine the monitoring area on the surface of the surrounding rock in the underground coal mine, install a reference target in the monitoring area, set two monitoring times, the first time and the second time, and collect point cloud data of the surrounding rock surface and the reference target in the monitoring area at the two monitoring times respectively, and label them as the first point cloud data and the second point cloud data. Preprocess the collected first point cloud data and the second point cloud data. S2: Coarse registration is performed on the preprocessed first and second point cloud data by referencing the target, fine registration is performed by iterating the nearest point algorithm, and a triangular mesh model of the monitoring area is generated by using the Poisson surface reconstruction algorithm on the registered first and second point cloud data. S3: Parameterize the triangular mesh model of the monitoring area into a two-dimensional analysis mesh, assign three-dimensional coordinates to each mesh node based on the point cloud data in the mesh node's neighborhood, compare the coordinates of the point cloud mesh nodes at the second time step with the coordinates of the point cloud mesh nodes at the first time step, calculate the three-dimensional displacement vector of each node, and establish a complete displacement field matrix. S4: Based on the three-dimensional displacement vector of each grid node, calculate the displacement gradient tensor, derive the Green-Lagrange strain tensor, and obtain the strain distribution across the entire field; S5: Record the strain value of each grid node, and determine whether to trigger an instability warning based on the relationship between the strain value and the set strain warning threshold and strain rate warning threshold.

2. The method for monitoring surrounding rock deformation in underground coal mines based on three-dimensional point clouds according to claim 1, characterized in that: Point cloud data of the surrounding rock surface and reference target within the monitoring area were collected using a ground-based laser scanner. The first point cloud data was recorded as [data point 1], and the point cloud data collected at the second time point was recorded as the second point cloud data. The point cloud data included the three-dimensional coordinates of each point. A multi-level filtering algorithm was used to preprocess the collected point cloud data of the first and second time points. The preprocessing first performed statistical outlier filtering and calculated the coordinates of each point to the reference target. The formulas for the mean distance of the nearest neighbors, the global distance mean, and the standard deviation are as follows: in, The global distance mean. The global distance standard deviation, This represents the total number of points in the point cloud. For the first Point to The average distance of the nearest neighbors is used to remove all points that satisfy this condition. The outliers are identified; next, voxel mesh downsampling is performed; the point cloud space is divided into a uniform voxel mesh, the voxel size is set according to the point density, and all points within each voxel are replaced by the centroid: in, Let the coordinates be the centroid coordinates of the voxel. The number of points within a single voxel. Point cloud data points within the same voxel mesh The three-dimensional coordinate vector.

3. The method for monitoring surrounding rock deformation in underground coal mines based on three-dimensional point clouds according to claim 1, characterized in that: The coarse registration is performed using a reference target, which is a spherical target. By identifying the reference target in the point clouds at the first and second time points, the rigid body transformation matrix of the target center is calculated. For the target point set, the spherical equation is fitted using the least squares method, as shown in the following formula: in,( () is the reference target center coordinate. The reference target radius is used as a reference. After obtaining the coordinates of the reference target center, fine registration is performed using the iterative nearest-point algorithm, with the objective function being: in, For the iterative nearest point registration error function, To be the optimal rotation matrix, It is a translation vector. For weighting coefficients, the second point cloud data The first in The coordinates of each point are recorded as follows: , Initial reference point cloud Zhongyu Corresponding points The number of point pairs establishing correspondences in point cloud registration; the optimal transformation is solved using singular value decomposition: in, This is the result of the second point cloud SVD decomposition. It is a left singular vector matrix. It is a singular value diagonal matrix. It is a right singular vector matrix. The average position of all points in the second point cloud data. The average position of all points in the point cloud data at the first time step; define the nearest point convergence threshold for iteration, iterate until the registration error is less than the nearest point convergence threshold, and complete the fine registration; the second point cloud after fine registration. .

4. The method for monitoring surrounding rock deformation in underground coal mines based on three-dimensional point clouds according to claim 3, characterized in that: Based on the registered first and second point cloud data, a triangular mesh model of the monitoring area is generated using the Poisson surface reconstruction algorithm, and the Poisson equation is solved: By discretizing the space using an octree structure, the continuous equations are transformed into a system of linear equations, as shown in the following formula: in, For the Laplace operator, As a gradient operator, the finely registered point cloud generates a steering field. For is the discrete Laplace matrix, Let be the unknown vector to be solved. For discrete divergence term vectors, based on Reconstruct a continuous indicator function field ,extract The isosurface generates a triangular network model.

5. The method for monitoring surrounding rock deformation in underground coal mines based on three-dimensional point clouds according to claim 4, characterized in that: The triangular mesh models of the monitoring area at the first and second time points are parameterized into the same two-dimensional analysis mesh, using the mesh spacing. A uniform grid is used, and the grid nodes are determined by surface projection: For each grid node, its nearest point on the triangular grid is found, and its accurate 3D coordinates are obtained by interpolation using the barycentric coordinates. in, In a two-dimensional analysis grid The three-dimensional coordinates of the node. The vertex of the triangle, Let the weights be the centroid coordinates, and satisfy the following conditions: Displacement calculation uses the direct comparison method, for each grid node. The formula for calculating the three-dimensional displacement vector is as follows: in, To analyze the grid spacing, For the first The three-dimensional displacement vector of each node. This represents the displacement component of a point along the X-axis. This represents the displacement component of the point along the Y-axis. This represents the displacement component of the point along the Z-axis. For the second point cloud data Node coordinates For the first point cloud data The coordinates of each node are used to perform error compensation, as follows: in, For the revised first Displacement vectors at each node For the first The three-dimensional displacement vector of each node. For the first The three-dimensional displacement vector of a reference target. The total number of reference targets.

6. The method for monitoring surrounding rock deformation in underground coal mines based on three-dimensional point clouds according to claim 5, characterized in that: S5 describes the calculation of the displacement gradient tensor. The formula is as follows: in, For unit tensors, For the first point cloud data Node coordinates This refers to the displacement gradient; the displacement gradient is calculated numerically using the central difference method; the displacement gradient tensor is calculated, and the Green-Lagrange strain tensor is derived, as shown in the following formula: in, For the first The strain tensor at each node is expanded to obtain the strain components, thus yielding the overall strain distribution. The principal strains are obtained by solving the eigenvalue problem. , The eigenvalues ​​are used to obtain the maximum principal strain. and minimum principal strain .

7. The method for monitoring surrounding rock deformation in underground coal mines based on three-dimensional point clouds according to claim 3, characterized in that: Record the strain value of each grid node, and set strain warning thresholds and strain rate warning thresholds based on engineering safety standards. The strain warning threshold is... and strain rate warning threshold The formula for calculating strain rate is as follows: in, For the first The strain rate at each node, For the first The maximum principal strain at each node The time interval between the first and second moments; a multi-level dynamic threshold mechanism is adopted: the strain of a single grid point exceeds... This triggers a primary deformation warning; strain at three or more consecutive adjacent grid points exceeds [a certain value]. This triggers a mid-level deformation warning; the strain rate exceeds [a certain threshold]. Or the strain of four or more consecutive adjacent grid points exceeds This triggers an advanced deformation warning.

8. A coal mine underground rock deformation monitoring system based on three-dimensional point cloud, characterized in that: A coal mine underground rock deformation monitoring system based on three-dimensional point clouds is used to execute the coal mine underground rock deformation monitoring method based on three-dimensional point clouds as described in any one of claims 1-7, comprising: Point cloud data acquisition module: used to determine the monitoring area by leveling the surrounding rock surface in the coal mine, install reference targets in the monitoring area, set two monitoring times, the first time and the second time, and collect point cloud data of the surrounding rock surface and the reference target in the monitoring area at the two monitoring times respectively, and label them as the first point cloud data and the second point cloud data. The collected first point cloud data and the second point cloud data are preprocessed. Point cloud registration and reconstruction module: used to perform coarse registration of the preprocessed first and second point cloud data with reference target, fine registration of the iterative nearest point algorithm, and generate a triangular mesh model of the monitoring area using the Poisson surface reconstruction algorithm for the registered first and second point cloud data. Mesh Displacement Field Calculation Module: This module parameterizes the triangular mesh model of the monitoring area into a two-dimensional analysis mesh. Each mesh node is assigned three-dimensional coordinates. The coordinates of the point cloud mesh nodes at the second time step are compared with the coordinates of the point cloud mesh nodes at the first time step. The three-dimensional displacement vector of each node is calculated, and a complete displacement field matrix is ​​established. Full-field strain analysis module: used to calculate the displacement gradient tensor based on the three-dimensional displacement vector of each grid node, derive the Green-Lagrange strain tensor, and obtain the full-field strain distribution; Deformation early warning and identification module: used to record the strain value of each grid node, set strain early warning threshold and strain rate early warning threshold, and use a multi-level dynamic threshold mechanism to trigger instability early warning.

Images