A rock mass structural surface network simulation method integrating artificial intelligence and laser scanning

Abstract

The present invention belongs to the field of three-dimensional simulation technology and specifically provides a rock mass structural surface network simulation method that integrates artificial intelligence and laser scanning, comprising: obtaining point cloud data; clustering the point cloud data using a clustering algorithm to obtain original structural surfaces; statistically analyzing the original structural surfaces according to rock mass type and structural partitioning to obtain representative structural surfaces, and fitting a probability density function based on the occurrence parameters of the representative structural surfaces; converting the probability density function into a volume density function based on the development volume density of the representative structural surfaces, and generating random structural surfaces using a Monte Carlo simulation method based on the volume density function; constructing a BIM model of the rock mass structure, and performing three-dimensional spatial simulation of the rock mass structural surface network based on the BIM model and the random structural surfaces to obtain simulation results. The present invention significantly improves the accuracy, comprehensiveness, and authenticity of rock mass structural surface depiction and can quantitatively evaluate the spatial distribution characteristics of the structural surface network.

Description

Technical Field

[0001] The present invention belongs to the technical field of three-dimensional simulation, and in particular relates to a rock mass structural surface network simulation method integrating artificial intelligence and laser scanning. Background Art

[0002] Rock masses encountered in geotechnical engineering typically consist of structural planes and rock blocks. Structural planes refer to various discontinuities within a rock mass, such as joints, bedding planes, faults, and cleavages. The development of these structural planes, including their type, occurrence, scale, density, and connectivity, is crucial for determining the physical and mechanical properties of the rock mass.

[0003] Conventional geotechnical engineering geological exploration typically involves field surveys, drilling, and geophysical exploration to describe and measure rock masses. These methods primarily target large, extended structural planes, such as regional faults and large, weak interlayers. By measuring parameters such as the occurrence, scale, and filling of these major structural planes, combined with laboratory testing and empirical judgment, the integrity and anisotropy of the rock mass can be roughly assessed.

[0004] However, rock masses also commonly harbor numerous, small-scale, limited-extent, and randomly distributed secondary structural planes. While small in size, these planes are numerous and dense, significantly influencing the overall mechanical properties of the rock mass. In particular, in rock masses characterized by high stress, intense weathering, and the superposition of multiple structural stages, the development of these small structural planes is often more complex, significantly impacting engineering stability.

[0005] Traditional geological exploration techniques, limited by scope, resolution, and cost, make it difficult to systematically detect and analyze these small structural surfaces. Drilling surveys, constrained by hole diameter and spacing, struggle to fully capture the spatial distribution of structural surfaces. Geophysical methods, with their limited resolution, struggle to identify small structural surfaces. Field surveys, confined to rock outcrops, struggle to understand the subsurface.

[0006] Therefore, detailed characterization and modeling of rock mass structural surface networks, based on conventional exploration, has become a technically challenging and hot topic in geotechnical engineering. Traditional methods struggle to accurately capture key parameters such as the spatial distribution and connectivity of these structural surface networks, making it difficult to accurately determine the physical and mechanical parameters of the rock mass, leading to significant uncertainty in engineering design and construction.

[0007] These technical deficiencies restrict the scientific and reliable design and construction plans for rock mass excavation and support. In the absence of detailed characterization of structural surfaces, engineering designs often tend to use conservative parameters. However, unpredictable rock mass structural issues may arise during actual construction, leading to design changes, project delays, and increased costs. Summary of the Invention

[0008] The purpose of the present invention is to address the above-mentioned problems existing in the prior art and to provide a rock structure surface network simulation method that integrates artificial intelligence and laser scanning, which greatly improves the accuracy, comprehensiveness and authenticity of rock structure surface characterization, and can quantitatively evaluate the spatial distribution characteristics of the structure surface network, providing a more reliable geological basis for the design and construction of rock excavation, support and other engineering projects.

[0009] To achieve the above objectives, the technical solutions of the present invention are as follows:

[0010] In a first aspect, the present invention provides a rock mass structural surface network simulation method integrating artificial intelligence and laser scanning, comprising:

[0011] Step S1: using a 3D laser scanner and a drone to acquire and fuse point clouds of the rock mass excavation surface to obtain point cloud data;

[0012] Step S2: clustering the point cloud data using a clustering algorithm, analyzing the clusters to obtain the original structural surface, determining the spatial range of the original structural surface, calculating the normal vector of each original structural surface, and obtaining the occurrence parameters of the original structural surface according to the normal vector;

[0013] Step S3: performing statistical analysis on the original structural planes according to the rock mass type and structural partition, selecting typical structural planes based on the analysis results, taking the typical structural planes as representative structural planes of the corresponding structural partitions, and fitting the probability density function of the representative structural planes based on the occurrence parameters of the representative structural planes;

[0014] Step S4: converting the probability density function of the representative structural surface into a volume density function of the representative structural surface according to the development volume density of the representative structural surface, and generating a random structural surface using a Monte Carlo simulation method according to the volume density function of the representative structural surface;

[0015] Step S5: construct a BIM model of the rock mass structure, and perform a three-dimensional spatial simulation on the rock mass structural surface network based on the BIM model and the random structural surface to obtain a simulation result.

[0016] Based on the above solution, preferably, step S1 includes:

[0017] Step S11: Arrange control points around the rock mass excavation surface, and use a total station to measure the three-dimensional coordinates of the control points to generate control point data;

[0018] Step S12: selecting a scanning site, and scanning the rock mass excavation surface at the scanning site using a three-dimensional laser scanner to obtain laser scanning data, wherein the laser scanning data covers all areas of the rock mass excavation surface;

[0019] Step S13: designing a flight route, equipping the drone with a high-resolution camera, photographing the rock excavation surface along the flight route, obtaining drone image data, and acquiring the drone's GPS trajectory data;

[0020] Step S14: pre-processing the laser scanning data and the UAV image data to remove abnormal data and bad images, thereby obtaining pre-processed laser scanning data and UAV image data;

[0021] Step S15: Using the control point data, the laser scanning data of different scanning sites are spliced ​​and registered to obtain a laser point cloud;

[0022] Step S16: Process the drone image data according to the GPS trajectory data of the drone to generate an image point cloud;

[0023] Step S17: Using the control point data, the laser point cloud and the image point cloud are simultaneously fused and completed to obtain complete point cloud data.

[0024] Based on the above solution, preferably, step S17 includes:

[0025] Step S171: Calculate the initial coordinate transformation relationship between the laser point cloud and the image point cloud using the three-dimensional coordinates of the control points. The initial coordinate transformation relationship includes an initial translation vector and an initial rotation matrix. Establish a common coordinate system for the laser point cloud and the image point cloud based on the initial coordinate transformation relationship, and perform a coarse registration of the laser point cloud and the image point cloud. The initial coordinate transformation relationship is expressed as follows:

[0026]

[0027] In the above formula, [xyz 1] is the homogeneous coordinate of the point in the laser point cloud; [x′ y′ z′ 1] is the homogeneous coordinate of the point in the transformed image point cloud; T = (T x , T y , T z ) is the initial translation vector of 3×1, which represents the relative translation relationship between the two coordinate systems; r is the element in the initial rotation matrix R, which represents the relative rotation relationship between the two coordinate systems;

[0028] Step S172: Optimize the coordinate transformation relationship between the laser point cloud and the image point cloud using the ICP algorithm to find the best coordinate transformation relationship and accurately align the laser point cloud and the image point cloud. The optimization goal of the ICP algorithm is:

[0029]

[0030] In the above formula, p nis the nth point in the laser point cloud, N is the number of points in the laser point cloud; q m is the image point cloud with p n The nearest point, M is the number of image point clouds;

[0031] Step S173: Use a fusion and completion algorithm to fuse and complete the registered laser point cloud and image point cloud to generate complete point cloud data.

[0032] Based on the above solution, preferably, the fusion completion algorithm includes:

[0033] Step 1: Initially fuse the registered laser point cloud and image point cloud to generate the initial fused point cloud P fused ;

[0034] Step 2: Detect missing areas in the fused point cloud;

[0035] Step 3: Select a point p in the missing area point cloud missing ;

[0036] Step 4: In the fusion point cloud P fused Search distance point p missing closest and smaller than d max B points, recorded as candidate fusion point set Among them, d max is the maximum search distance;

[0037] Step 5: Calculate point p missing With each candidate fusion point The distance weight w between b :

[0038]

[0039] In the above formula, σ is the parameter that controls the weight decay speed;

[0040] Step 6. Calculate point p missing The completion point p completed :

[0041]

[0042] In the above formula, w completed is the weight of the completion point, which is calculated according to the local density of the completion point;

[0043] Step 7: Completion point p completed To perform smoothing:

[0044] p′ completed =p completed +α×Laplacian(p completed , Pfused );

[0045] In the above formula, p′ completed represents the smoothed completion point; α is the smoothing factor, which is used to control the smoothness between the completion point and the fused point cloud; Laplacian (p completed , P fused ) represents the completion point p completed In the fusion point cloud P fused The Laplace operator in ;

[0046] Step 8: The smoothed completion point p′ completed Add to the completed point cloud;

[0047] Step 9: Repeat steps 3 to 8 until all points in the missing area are processed to obtain the final completed point cloud;

[0048] Step 10: Compare the final completed point cloud with the fused point cloud P fused Perform fusion, update the fused point cloud, and calculate the confidence of the updated fused point cloud;

[0049] Step 11: Repeat steps 2 to 10 until the preset number of iterations is reached or the confidence level of the fused point cloud meets the requirement, and the final fused point cloud, i.e., the complete point cloud data, is obtained.

[0050] Based on the above solution, preferably, step S2 includes:

[0051] Step S21: Determine the number of components k of the clustering model, where one component represents one cluster. Randomly select k points as the initial mean vector, initialize the covariance matrix of each component to the identity matrix, and initialize the weight of each component to 1k.

[0052] Step S22: Calculate the posterior probability that each point in the point cloud data belongs to each component;

[0053] Step S23: updating the parameters of the clustering model, i.e., the mean vector, covariance matrix, and weight, according to the posterior probability;

[0054] Step S24, repeating steps S22 to S23 until the parameters of the clustering model converge or the maximum number of iterations is reached;

[0055] Step S25: assign each point to the component with the largest posterior probability to obtain a clustering result;

[0056] Step S26: For each cluster in the clustering results, calculate the spatial range of the cluster, use the least squares method to fit a plane based on the spatial range of each cluster, minimize the sum of the squares of the distances from the points in the cluster to the plane, and calculate the fitting error. If the fitting error is less than a preset threshold, it is determined that the cluster corresponds to an original structural surface;

[0057] Step S27: Calculate the normal vector of the original structural surface according to the fitting plane, and calculate the attitude parameters of the original structural surface based on the normal vector. The attitude parameters include dip, inclination, spacing and trace length.

[0058] Based on the above solution, preferably, the calculation formula of the posterior probability is:

[0059]

[0060] In the above formula, p i represents the i-th point; γ ij represents the posterior probability that the i-th point belongs to the j-th component; π j represents the weight of the jth component, π j satisfy μ j represents the mean vector of the jth component; ∈ j represents the covariance matrix of the jth component; N(p i |μ j Indicated by μ j is the mean vector, ∈ j The Gaussian probability function of the covariance matrix at point p i The value at In , l traverses from 1 to k, indicating the sum of all components.

[0061] Based on the above solution, preferably, the update formula of the mean vector is:

[0062]

[0063] The updating formula of the covariance matrix is:

[0064]

[0065] The weight update formula is:

[0066]

[0067] In the above formula, A represents the number of points in the point cloud data.

[0068] Based on the above solution, preferably, step S3 includes:

[0069] S31. Divide the rock mass into different types based on the geological characteristics of the rock mass excavation surface. Within each rock mass type, divide the rock mass excavation surface into different structural zones based on the structural characteristics, wherein each structural zone has consistent rock mass type and structural characteristics.

[0070] S32. For each structural partition, count the occurrence parameters of all original structural surfaces in the structural partition, draw a statistical graph of the occurrence parameters of the original structural surfaces, and analyze the spatial distribution pattern of the original structural surfaces;

[0071] S33, performing aggregation analysis on the statistical graph to identify the aggregation area and dominant direction of the original structural surface;

[0072] S34. Based on the aggregation analysis results, select an aggregation area representing the structural characteristics of the original structural surface in the structural partition, define the original structural surface therein as a typical structural surface, and use the typical structural surface as the representative structural surface of the corresponding structural partition;

[0073] S35. For the representative structural surface of each structural partition, estimate the probability distribution of the occurrence parameter of the representative structural surface using the Weibull distribution probability, and construct a probability density function of the occurrence parameter of the representative structural surface based on the estimated probability distribution of the occurrence parameter of the representative structural surface;

[0074] S36. Evaluate the fitting effect of the probability density function;

[0075] S37. Repeat steps S35 to S36 for each structural partition until the fitting effect meets the requirements, and obtain the probability density functions of the representative structural surfaces of all structural partitions.

[0076] The step S4 comprises:

[0077] S41. Counting the number of representative structural surfaces in each structural partition, and calculating the development volume density of the representative structural surfaces;

[0078] S42, multiplying the probability density function representing the structural surface by the development volume density representing the structural surface to obtain a volume density function representing the structural surface;

[0079] S43. Using the Monte Carlo method, R random structural surfaces are generated in each structural partition according to the volume density function representing the structural surface.

[0080] Based on the above solution, preferably, step S5 includes:

[0081] S51. Use 3D modeling software to construct a geometric model of the rock mass structure, add rock mass type, structural partitions, and original structural surfaces to the geometric model, and form a BIM model of the rock mass structure;

[0082] S52. Import the random structural surfaces into the BIM model and identify the structural partition where each random structural surface is located:

[0083] For the u-th random structure surface, determine its center point P u Is it located in the Vth structural partition? If point P u If it is located in the structural partition v, the structural partition identifier of the random structural surface u is set to v;

[0084] S53, integrating the original structural surface and the random structural surface of each structural partition according to the recognition result of the random structural surface to obtain a comprehensive structural surface;

[0085] S54. In the BIM model, for each structural partition, according to the number of comprehensive structural surfaces contained therein, randomly select Y target structural surfaces from the comprehensive structural surfaces, wherein the number of target structural surfaces is the same as the number of original structural surfaces;

[0086] S55. Perform three-dimensional visual simulation on the selected Y target structural surfaces to generate a structural surface network model, i.e., a simulation result.

[0087] In a second aspect, the present invention provides a rock mass structural surface network simulation system that integrates artificial intelligence and laser scanning, including: a point cloud acquisition and fusion module, an original structural surface and its occurrence parameter acquisition module, a representative structural surface and its probability density function acquisition module, a random structural surface acquisition module, and a three-dimensional simulation module;

[0088] Point cloud acquisition and fusion module, used to acquire and fuse point clouds of rock excavation surfaces using 3D laser scanners and drones to obtain point cloud data;

[0089] The module for obtaining the original structural surface and its occurrence parameters is used to cluster the point cloud data using a clustering algorithm, analyze the clusters, obtain the original structural surface, determine the spatial range of the original structural surface, calculate the normal vector of each original structural surface, and obtain the occurrence parameters of the original structural surface based on the normal vector;

[0090] The module for obtaining representative structural surfaces and their probability density functions is used to perform statistical analysis on the original structural surfaces according to the rock mass type and structural partition, select typical structural surfaces based on the analysis results, use the typical structural surfaces as the representative structural surfaces of the corresponding structural partitions, and fit the probability density functions of the representative structural surfaces based on the occurrence parameters of the representative structural surfaces.

[0091] A random structural surface acquisition module is used to convert the probability density function of the representative structural surface into a volume density function of the representative structural surface according to the development volume density of the representative structural surface, and generate a random structural surface using a Monte Carlo simulation method based on the volume density function of the representative structural surface;

[0092] The three-dimensional simulation module is used to construct the BIM model of the rock mass structure, and to perform three-dimensional spatial simulation of the rock mass structural surface network based on the BIM model and random structural surfaces to obtain simulation results.

[0093] Based on the above solution, preferably, the point cloud acquisition and fusion module performs point cloud acquisition and fusion according to the following steps:

[0094] Step S11: Arrange control points around the rock mass excavation surface, and use a total station to measure the three-dimensional coordinates of the control points to generate control point data;

[0095] Step S12: selecting a scanning site, and scanning the rock mass excavation surface at the scanning site using a three-dimensional laser scanner to obtain laser scanning data, wherein the laser scanning data covers all areas of the rock mass excavation surface;

[0096] Step S13: designing a flight route, equipping the drone with a high-resolution camera, photographing the rock excavation surface along the flight route, obtaining drone image data, and acquiring the drone's GPS trajectory data;

[0097] Step S14: pre-processing the laser scanning data and the UAV image data to remove abnormal data and bad images, thereby obtaining pre-processed laser scanning data and UAV image data;

[0098] Step S15: Using the control point data, the laser scanning data of different scanning sites are spliced ​​and registered to obtain a laser point cloud;

[0099] Step S16: Process the drone image data according to the GPS trajectory data of the drone to generate an image point cloud;

[0100] Step S17: Using the control point data, the laser point cloud and the image point cloud are simultaneously fused and completed to obtain complete point cloud data.

[0101] Based on the above solution, preferably, the point cloud acquisition and fusion module simultaneously fuses and completes the laser point cloud and the image point cloud according to the following steps:

[0102] Step S171: Calculate the initial coordinate transformation relationship between the laser point cloud and the image point cloud using the three-dimensional coordinates of the control points. The initial coordinate transformation relationship includes an initial translation vector and an initial rotation matrix. Establish a common coordinate system for the laser point cloud and the image point cloud based on the initial coordinate transformation relationship, and perform a coarse registration of the laser point cloud and the image point cloud. The initial coordinate transformation relationship is expressed as follows:

[0103]

[0104] In the above formula, [xyz 1] is the homogeneous coordinate of the point in the laser point cloud; [x′ y′ z′ 1] is the homogeneous coordinate of the point in the transformed image point cloud; T = (T x , T y , T z ) is the initial translation vector of 3×1, which represents the relative translation relationship between the two coordinate systems; r is the element in the initial rotation matrix R, which represents the relative rotation relationship between the two coordinate systems;

[0105] Step S172: Optimize the coordinate transformation relationship between the laser point cloud and the image point cloud using the ICP algorithm to find the best coordinate transformation relationship and accurately align the laser point cloud and the image point cloud. The optimization goal of the ICP algorithm is:

[0106]

[0107] In the above formula, p n is the nth point in the laser point cloud, N is the number of points in the laser point cloud; q m is the image point cloud with p n The nearest point, M is the number of image point clouds;

[0108] Step S173: Use a fusion and completion algorithm to fuse and complete the registered laser point cloud and image point cloud to generate complete point cloud data.

[0109] Based on the above solution, preferably, the fusion completion algorithm includes:

[0110] Step 1: Initially fuse the registered laser point cloud and image point cloud to generate the initial fused point cloud P fused ;

[0111] Step 2: Detect missing areas in the fused point cloud;

[0112] Step 3: Select a point p in the missing area point cloud missing ;

[0113] Step 4: In the fusion point cloud P fused Search distance point p missing closest and smaller than d max B points, recorded as candidate fusion point set Among them, d max is the maximum search distance;

[0114] Step 5: Calculate point p missing With each candidate fusion point The distance weight w between b :

[0115]

[0116] In the above formula, σ is the parameter that controls the weight decay speed;

[0117] Step 6. Calculate point p missing The completion point p completed :

[0118]

[0119] In the above formula, W completed is the weight of the completion point, which is calculated according to the local density of the completion point;

[0120] Step 7: Completion point p completed To perform smoothing:

[0121] p′ completed =p completed +α×Laplacian(p completed , P fused );

[0122] In the above formula, p′ completed represents the smoothed completion point; α is the smoothing factor, which is used to control the smoothness between the completion point and the fused point cloud; Laplacian (p completed , P fused ) represents the completion point p completed In the fusion point cloud P fused The Laplace operator in ;

[0123] Step 8: The smoothed completion point p′ completed Add to the completed point cloud;

[0124] Step 9: Repeat steps 3 to 8 until all points in the missing area are processed to obtain the final completed point cloud;

[0125] Step 10: Compare the final completed point cloud with the fused point cloud P fused Perform fusion, update the fused point cloud, and calculate the confidence of the updated fused point cloud;

[0126] Step 11: Repeat steps 2 to 10 until the preset number of iterations is reached or the confidence level of the fused point cloud meets the requirement, and the final fused point cloud, i.e., the complete point cloud data, is obtained.

[0127] Based on the above solution, preferably, the original structural surface and its occurrence parameter acquisition module acquires the original structural surface and its occurrence parameters according to the following steps:

[0128] Step S21: Determine the number of components k of the clustering model, where one component represents one cluster, randomly select k points as the initial mean vector, initialize the covariance matrix of each component to the identity matrix, and initialize the weight of each component to 1 / / k;

[0129] Step S22: Calculate the posterior probability that each point in the point cloud data belongs to each component;

[0130] Step S23: updating the parameters of the clustering model, i.e., the mean vector, covariance matrix, and weight, according to the posterior probability;

[0131] Step S24, repeating steps S22 to S23 until the parameters of the clustering model converge or the maximum number of iterations is reached;

[0132] Step S25: assign each point to the component with the largest posterior probability to obtain a clustering result;

[0133] Step S26: For each cluster in the clustering results, calculate the spatial range of the cluster, use the least squares method to fit a plane based on the spatial range of each cluster, minimize the sum of the squares of the distances from the points in the cluster to the plane, and calculate the fitting error. If the fitting error is less than a preset threshold, it is determined that the cluster corresponds to an original structural surface;

[0134] Step S27: Calculate the normal vector of the original structural surface according to the fitting plane, and calculate the attitude parameters of the original structural surface based on the normal vector. The attitude parameters include dip, inclination, spacing and trace length.

[0135] Based on the above solution, preferably, the calculation formula of the posterior probability is:

[0136]

[0137] In the above formula, p i represents the i-th point; γ ij represents the posterior probability that the i-th point belongs to the j-th component; π j represents the weight of the jth component, π j satisfy μ j represents the mean vector of the jth component; ∈ j represents the covariance matrix of the jth component; N(p i |μ j Indicated by μ j is the mean vector, ∈ j The Gaussian probability function of the covariance matrix at point p i The value at In , l traverses from 1 to k, indicating the sum of all components.

[0138] Based on the above solution, preferably, the update formula of the mean vector is:

[0139]

[0140] The update formula of the covariance matrix is:

[0141]

[0142] The weight update formula is:

[0143]

[0144] In the above formula, A represents the number of points in the point cloud data.

[0145] Based on the above solution, preferably, the representative structural surface and its probability density function acquisition module acquires the representative structural surface and its probability density function according to the following steps:

[0146] S31. Divide the rock mass into different types based on the geological characteristics of the rock mass excavation surface. Within each rock mass type, divide the rock mass excavation surface into different structural zones based on the structural characteristics, wherein each structural zone has consistent rock mass type and structural characteristics.

[0147] S32. For each structural partition, count the occurrence parameters of all original structural surfaces in the structural partition, draw a statistical graph of the occurrence parameters of the original structural surfaces, and analyze the spatial distribution pattern of the original structural surfaces;

[0148] S33, performing aggregation analysis on the statistical graph to identify the aggregation area and dominant direction of the original structural surface;

[0149] S34. Based on the aggregation analysis results, select an aggregation area representing the structural characteristics of the original structural surface in the structural partition, define the original structural surface therein as a typical structural surface, and use the typical structural surface as the representative structural surface of the corresponding structural partition;

[0150] S35. For the representative structural surface of each structural partition, estimate the probability distribution of the occurrence parameter of the representative structural surface using the Weibull distribution probability, and construct a probability density function of the occurrence parameter of the representative structural surface based on the estimated probability distribution of the occurrence parameter of the representative structural surface;

[0151] S36. Evaluate the fitting effect of the probability density function;

[0152] S37. Repeat steps S35 to S36 for each structural partition until the fitting effect meets the requirements, and obtain the probability density functions of the representative structural surfaces of all structural partitions.

[0153] Based on the above solution, preferably, the random structural surface acquisition module generates the random structural surface according to the following steps:

[0154] S41. Counting the number of representative structural surfaces in each structural partition, and calculating the development volume density of the representative structural surfaces;

[0155] S42, multiplying the probability density function representing the structural surface by the development volume density representing the structural surface to obtain a volume density function representing the structural surface;

[0156] S43. Using the Monte Carlo method, R random structural surfaces are generated in each structural partition according to the volume density function representing the structural surface.

[0157] Based on the above solution, preferably, the 3D simulation module performs 3D visual simulation according to the following steps to generate simulation results:

[0158] S51. Use 3D modeling software to construct a geometric model of the rock mass structure, add rock mass type, structural partitions, and original structural surfaces to the geometric model, and form a BIM model of the rock mass structure;

[0159] S52. Import the random structural surfaces into the BIM model and identify the structural partition where each random structural surface is located:

[0160] For the u-th random structure surface, determine its center point P u Is it located in the vth structural partition? If point P u If it is located in the structural partition v, the structural partition identifier of the random structural surface u is set to v;

[0161] S53, integrating the original structural surface and the random structural surface of each structural partition according to the recognition result of the random structural surface to obtain a comprehensive structural surface;

[0162] S54. In the BIM model, for each structural partition, according to the number of comprehensive structural surfaces contained therein, randomly select Y target structural surfaces from the comprehensive structural surfaces, wherein the number of target structural surfaces is the same as the number of original structural surfaces;

[0163] S55. Perform three-dimensional visual simulation on the selected Y target structural surfaces to generate a structural surface network model, i.e., a simulation result.

[0164] Compared with the prior art, the present invention has the following beneficial effects:

[0165] 1. The present invention obtains high-precision rock point cloud data through three-dimensional laser scanning and drone imagery, combines artificial intelligence algorithms to accurately identify rock structural surfaces, and considers differences in rock type and structural zoning to perform statistical analysis and probabilistic modeling on parameters such as the occurrence and scale of the structural surfaces. On this basis, the Monte Carlo method is used to simulate and generate random structural surfaces. Finally, three-dimensional visualization simulation of the rock structural surface network is achieved based on the BIM model, which greatly improves the accuracy, comprehensiveness and authenticity of the rock structural surface depiction, and can quantitatively evaluate the spatial distribution characteristics of the structural surface network, providing a more reliable geological basis for the design and construction of rock excavation, support and other engineering projects.

[0166] 2. The present invention proposes a fusion completion algorithm to achieve point cloud completion in the process of point cloud fusion, so that the completion process and the fusion process are carried out alternately, and the fused point cloud information is used to constrain and guide the missing area. It not only aligns the laser point cloud and the image point cloud, unifies the two heterogeneous point clouds into the same coordinate system, and realizes the fusion of multi-source data, but also adopts a completion strategy based on local features, which fully considers the spatial relationship between the points to be completed and the surrounding points. By introducing constraints such as distance weights and point cloud density, the accuracy of the calculation of the completion points is improved. At the same time, the fused point cloud is updated through an iterative optimization mechanism to continuously improve the confidence of the point cloud, making the completion result more stable and reliable; and the Laplace operator is used to smooth the completion points, and a smoothing factor is introduced to control the degree of fusion between the completion points and the original point cloud, so that the newly added points can smoothly transition and naturally merge with the original point cloud, avoiding obvious seams or mutations in the completion area.

[0167] 3. The present invention complements the advantages of the two point clouds, which not only ensures the accuracy of the point cloud data, but also makes full use of the image information to make up for the blind spots of laser scanning. By cross-detecting the missing areas of the laser point cloud and the image point cloud, the information of one point cloud is used to fill the missing areas of the other point cloud, thereby realizing the complementarity of multi-source data.

[0168] 4. The clustering model of the present invention can well fit the multimodal distribution characteristics of point cloud data, divide point clouds of different structural surfaces into different clusters, and make the clustering results more accurate and stable by iteratively optimizing the parameters of the Gaussian mixture model.

[0169] 5. The present invention realizes the three-dimensional visual simulation of the rock mass structural surface network through the BIM model, which not only takes into account the actual distribution characteristics of the original structural surface, but also incorporates the statistical information of the random structural surface. The generated structural surface network model is closer to the actual state of the rock mass. BRIEF DESCRIPTION OF THE DRAWINGS

[0170] Figure 1 The present invention is a flowchart of the method.

[0171] Figure 2This is a structural block diagram of the system of the present invention. DETAILED DESCRIPTION

[0172] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0173] See also Figure 1 The present invention provides a rock mass structural surface network simulation method integrating artificial intelligence and laser scanning, which is carried out in the following steps:

[0174] The rock mass structural surface network simulation method comprises:

[0175] Step S1: using a 3D laser scanner and a drone to acquire and fuse point clouds of the rock mass excavation surface to obtain point cloud data;

[0176] Step S2: clustering the point cloud data using a clustering algorithm, analyzing the clusters to obtain the original structural surface, determining the spatial range of the original structural surface, calculating the normal vector of each original structural surface, and obtaining the occurrence parameters of the original structural surface according to the normal vector;

[0177] Step S3: performing statistical analysis on the original structural planes according to the rock mass type and structural partition, selecting typical structural planes based on the analysis results, taking the typical structural planes as representative structural planes of the corresponding structural partitions, and fitting the probability density function of the representative structural planes based on the occurrence parameters of the representative structural planes;

[0178] Step S4: converting the probability density function of the representative structural surface into a volume density function of the representative structural surface according to the development volume density of the representative structural surface, and generating a random structural surface using a Monte Carlo simulation method according to the volume density function of the representative structural surface;

[0179] Step S5: construct a BIM model of the rock mass structure, and perform a three-dimensional spatial simulation on the rock mass structural surface network based on the BIM model and the random structural surface to obtain a simulation result.

[0180] In one embodiment of the present invention, step S1 includes:

[0181] Step S11: Arrange control points around the rock mass excavation surface, and use a total station to measure the three-dimensional coordinates of the control points to generate control point data;

[0182] Specifically, select appropriate locations around the rock mass excavation surface and arrange several control points. The control points should be distributed as evenly as possible, with a number of no less than four. The distance between the control points should be set according to the size of the excavation surface and the terrain conditions, and can be set between 20 and 50 meters. Use a total station to accurately measure each control point and record the three-dimensional coordinates (X, Y, Z) of the control point. Save the measured control point coordinate data to form a control point data file.

[0183] Step S12: selecting a scanning site, and scanning the rock mass excavation surface at the scanning site using a three-dimensional laser scanner to obtain laser scanning data, wherein the laser scanning data covers all areas of the rock mass excavation surface;

[0184] Specifically, based on the size and complexity of the rock mass excavation surface, appropriate scanning stations are selected, which should be able to cover the entire excavation surface. A 3D laser scanner is set up at each scanning station, and parameters such as the scanner's angle and resolution are adjusted. The scanner is started to perform a full-scale scan of the rock mass excavation surface to obtain high-density point cloud data. The data obtained from the scan at each station is saved to form a laser scanning data file.

[0185] Step S13: designing a flight route, equipping the drone with a high-resolution camera, photographing the rock excavation surface along the flight route, obtaining drone image data, and acquiring the drone's GPS trajectory data;

[0186] Specifically, the optimal UAV flight route is designed based on the shape and size of the rock excavation surface to ensure the integrity and overlap of the image data. A high-resolution camera, such as a SLR camera or a professional aerial camera, is installed on the UAV. The UAV is controlled to fly along the designed route and continuously take photos of the rock excavation surface while recording the UAV's GPS trajectory data. The captured image data and GPS trajectory data are downloaded and saved.

[0187] Step S14: pre-processing the laser scanning data and the UAV image data to remove abnormal data and bad images, thereby obtaining pre-processed laser scanning data and UAV image data;

[0188] Specifically, the laser scanning data is filtered using statistical filters to remove noise and outliers based on the local density and distance threshold of the point cloud, thereby improving the quality of the point cloud. Photos taken by drones are checked to remove blurry, underexposed, overexposed, and other undesirable images. Image enhancement processing is performed on the image data, such as adjusting brightness and contrast, to improve image quality.

[0189] Step S15: Using the control point data, the laser scanning data of different scanning sites are spliced ​​and registered to obtain a laser point cloud;

[0190] Specifically, control points are identified in the laser scanning data of different sites, and the coordinates corresponding to the control points are extracted. Based on the control points, the scanning data of different sites are unified into the same coordinate system through coordinate transformation. The scanning data of different sites are finely aligned using the point cloud registration algorithm to eliminate the deviation between the scanning data. The registered scanning data are merged to generate complete laser point cloud data.

[0191] Step S16: Process the drone image data according to the GPS trajectory data of the drone to generate an image point cloud;

[0192] Specifically, the GPS trajectory data, including flight time, latitude and longitude, altitude, heading and other information, is exported from the UAV flight control system; the shooting position of each image is calculated based on the UAV GPS trajectory data and the shooting timestamp of the image through interpolation algorithms, such as linear interpolation or spline interpolation; the shooting attitude of each image is calculated using the UAV's IMU (inertial measurement unit) data, such as heading, pitch, and roll angles; the image shooting position and attitude information are associated with the corresponding image file to generate the spatial reference information of the image; image matching algorithms, such as SIFT (scale-invariant feature transform) and SURF (speeded up robust features), are used to find the same-name points between adjacent images; and matching is determined by similarity metrics of feature point descriptors, such as Euclidean distance and Hamming distance. Point pairing; using the RANSAC (Random Sample Consensus) algorithm to remove incorrect matching points and improve the reliability of matching results; generating a file of homonymous points between images, recording the pixel coordinates and correspondences of the matching points; using the drone's posture information and the homonymous points between images, the three-dimensional coordinates of the homonymous points in space are calculated through the principle of aerial triangulation; minimizing the reprojection error through iterative optimization and improving the accuracy of the three-dimensional point coordinates; generating an aerial triangulation result file containing the three-dimensional point coordinates and the corresponding image information; using the three-dimensional points obtained by aerial triangulation as the initial value, through multi-view geometric constraints, such as the PMVS algorithm, to extract more matching points on the image; integrating the three-dimensional points on all images into a dense point cloud dataset through depth map fusion and point cloud merging;

[0193] Step S17: Using the control point data, the laser point cloud and the image point cloud are simultaneously fused and completed to obtain complete point cloud data.

[0194] In one embodiment of the present invention, step S17 includes:

[0195] Step S171: Calculate the initial coordinate transformation relationship between the laser point cloud and the image point cloud using the three-dimensional coordinates of the control points. The initial coordinate transformation relationship includes an initial translation vector and an initial rotation matrix. Establish a common coordinate system for the laser point cloud and the image point cloud based on the initial coordinate transformation relationship, and perform a coarse registration of the laser point cloud and the image point cloud. The initial coordinate transformation relationship is expressed as follows:

[0196]

[0197] In the above formula, [xyz 1] is the homogeneous coordinate of the point in the laser point cloud; [x′ y′ z′ 1] is the homogeneous coordinate of the point in the transformed image point cloud; T = (T x , T y , T z ) is the initial translation vector of 3×1, which represents the relative translation relationship between the two coordinate systems; r is the element in the initial rotation matrix R, which represents the relative rotation relationship between the two coordinate systems;

[0198] Specifically, laser point clouds and image point clouds are usually collected in different coordinate systems. The laser point cloud uses the measuring instrument as the origin, while the image point cloud uses the camera as the origin. In order to align them, it is necessary to find the transformation relationship between the two coordinate systems, namely the translation vector and the rotation matrix R. This transformation relationship can be calculated using control points. Control points are feature points that can be identified and matched in both point clouds. Through the three-dimensional coordinates of the control points, the correspondence between the two coordinate systems can be established and the optimal transformation parameters can be solved. Through this transformation relationship, any point in the laser point cloud can be converted to the coordinate system of the image point cloud to achieve alignment of the two point clouds.

[0199] Step S172: Optimize the coordinate transformation relationship between the laser point cloud and the image point cloud using the ICP algorithm to find the best coordinate transformation relationship and accurately align the laser point cloud and the image point cloud. The optimization goal of the ICP algorithm is:

[0200]

[0201] In the above formula, p n is the nth point in the laser point cloud, N is the number of points in the laser point cloud; q m is the image point cloud with p n The nearest point, M is the number of image point clouds;

[0202] Specifically, the number of points in the laser point cloud and the image point cloud may not be the same. This embodiment uses the ICP algorithm to find the optimal transformation matrix between the two point clouds so as to minimize the distance between corresponding points. In each iteration, the algorithm finds the nearest corresponding point in the image point cloud for each point in the laser point cloud, and then optimizes the transformation matrix to minimize the distance between corresponding points. This process is repeated until the convergence condition is met or the maximum number of iterations is reached.

[0203] Step S173: Use a fusion and completion algorithm to fuse and complete the registered laser point cloud and image point cloud to generate complete point cloud data.

[0204] In one embodiment of the present invention, the fusion completion algorithm includes:

[0205] Step 1: Initially fuse the registered laser point cloud and image point cloud to generate the initial fused point cloud P fused ;

[0206] Step 2: Detect missing areas in the fused point cloud;

[0207] Step 3: Select a point p in the missing area point cloud missing ;

[0208] Step 4: In the fusion point cloud P fused Search distance point p missing closest and smaller than d max B points, recorded as candidate fusion point set Among them, d max is the maximum search distance;

[0209] Step 5: Calculate point p missing With each candidate fusion point The distance weight w between b :

[0210]

[0211] In the above formula, σ is the parameter that controls the weight decay speed;

[0212] Step 6. Calculate point p missing The completion point p completed :

[0213]

[0214] In the above formula, w completed is the weight of the completion point, which is calculated according to the local density of the completion point. The local density reflects the integrity and continuity of the completion point cloud in the local area. Specifically, the KNN algorithm is used to calculate the local density of each completion point: density(p)=s / (π×h2 ), where s is a point in a spherical neighborhood with a radius of h and a center of point p; the local density is normalized to w completed , then w completed =density(p) / max(density), max(density) represents the maximum density;

[0215] Step 7: Completion point p completed To perform smoothing:

[0216] p′ completed =p completed +α×Laplacian(p completed , P fused );

[0217] In the above formula, p′ completed represents the smoothed completion point; α is the smoothing factor, which is used to control the smoothness between the completion point and the fused point cloud; Laplacian (p completed , P fused ) represents the completion point p completed In the fusion point cloud P fused The Laplace operator in ;

[0218] Step 8: The smoothed completion point p′ completed Add to the completed point cloud;

[0219] Step 9: Repeat steps 3 to 8 until all points in the missing area are processed to obtain the final completed point cloud;

[0220] Step 10: Compare the final completed point cloud with the fused point cloud P fused Perform fusion, update the fused point cloud, and calculate the confidence of the updated fused point cloud;

[0221] Step 11: Repeat steps 2 to 10 until the preset number of iterations is reached or the confidence level of the fused point cloud meets the requirement, and the final fused point cloud, i.e., the complete point cloud data, is obtained.

[0222] Specifically, when fusing and completing the point cloud, this embodiment adopts an alternating method of the completion process and the fusion process, and uses the geometric information around the fused point cloud to complete the missing area, while using Laplace smoothing to make the completed area more continuous with the original area.

[0223] In the fusion completion algorithm, the quality of fusion is judged according to the confidence of the fused point cloud. The confidence is the degree of trustworthiness of each point, which is used to reflect the accuracy, completeness and consistency of the fused point cloud. The confidence calculation formula is:

[0224]

[0225] Where p i Refers to the newly fused completion point; p j For point p i Neighborhood points; N i For point p i The neighborhood point set of ω; ij is the neighborhood weight, indicating that point p j Point p i The weight of point p j With point p i is inversely proportional to the distance, θ ij Represents point p j With point p i The angle between the normal vectors of , which is calculated by the dot product of the two normal vectors; is the Gaussian weight; σ1 is a constant used to control the width of the Gaussian function.

[0226] In this embodiment, by introducing an inverse weight of the distance, it is ensured that points with closer distances play a greater role in the confidence calculation, so that the confidence can better reflect the consistency of the local environment; the dot product of the normal vector is introduced to ensure that the consistency of the local surface morphology of the point cloud is taken into account in the confidence evaluation. The consistency of the normal vector helps to reduce the mutation of the surface of adjacent points and improve the smoothness and accuracy of the point cloud; Gaussian weighting is used to more smoothly consider the influence of neighboring points to ensure that the influence of points with farther distances on the confidence gradually decreases.

[0227] Specifically, in one embodiment of the present invention, step S2 includes:

[0228] Step S21: Determine the number of components k of the clustering model, where one component represents one cluster, randomly select k points as the initial mean vector, initialize the covariance matrix of each component to the identity matrix, and initialize the weight of each component to 1 / k;

[0229] Step S22: Calculate the posterior probability that each point in the point cloud data belongs to each component;

[0230] Step S23: updating the parameters of the clustering model, i.e., the mean vector, covariance matrix, and weight, according to the posterior probability;

[0231] Step S24, repeating steps S22 to S23 until the parameters of the clustering model converge or the maximum number of iterations is reached;

[0232] Step S25: Assign each point to the component with the largest posterior probability to obtain the clustering result; that is, assign point pi to the component j with the largest posterior probability, then

[0233] Step S26: For each cluster in the clustering result, calculate the spatial range of the cluster, for example, calculate its bounding box, and use the least squares method to fit a plane based on the spatial range of each cluster to obtain the plane equation ax+by+c=0; the formula is:

[0234]

[0235] Minimize the sum of squares of distances from points in the cluster to the plane and calculate the fitting error. If the fitting error is less than a preset threshold, the cluster is determined to correspond to an original structural surface.

[0236] Step S27: Calculate the normal vector of the original structural surface according to the fitting plane to obtain the normal vector n = (a, b, -1); normalize the normal vector to obtain the normalized normal vector The attitude parameters of the original structural surface are calculated based on the normalized normal vector n. The attitude parameters include dip, inclination, spacing and trace length. Dip refers to the direction of the normal vector projection on the horizontal plane, expressed as azimuth; inclination refers to the angle between the normal vector and the horizontal plane; spacing refers to the distance between points in different parallel planes; trace length refers to the length of the projection of the structural plane on the surface in the dip direction.

[0237] Specifically, in one embodiment of the present invention, the calculation formula of the posterior probability is:

[0238]

[0239] In the above formula, p i represents the i-th point; γ ij represents the posterior probability that the i-th point belongs to the j-th component; π j represents the weight of the jth component, π j satisfy μ j represents the mean vector of the jth component; ∈ j represents the covariance matrix of the jth component; N(p i |μ j Indicated by μ j is the mean vector, ∈ j The Gaussian probability function of the covariance matrix at point p i The value at In , l traverses from 1 to k, indicating the sum of all components.

[0240] Specifically, in one embodiment of the present invention, the updating formula of the mean vector is:

[0241]

[0242] The update formula of the covariance matrix is:

[0243]

[0244] The weight update formula is:

[0245]

[0246] In the above formula, A represents the number of points in the point cloud data;

[0247] In this embodiment, when determining the fitting error, the preset threshold adopts a dynamic threshold. During each round of fitting, feedback adjustment is made based on the error distribution after fitting. Specifically, the preset threshold can be adaptively adjusted based on the change in error during each iteration:

[0248] Threshold t+1 =Threshold t -η·ΔE t ;

[0249] In the above formula, Threshold t The preset threshold for the current round; Threshold t+1 is the preset threshold for the next round; η is the learning rate; ΔE t is the change in error.

[0250] Specifically, in one embodiment of the present invention, step S3 includes:

[0251] S31. Divide the rock mass into different types based on the geological characteristics of the rock mass excavation surface. Within each rock mass type, divide the rock mass excavation surface into different structural zones based on the structural characteristics, wherein each structural zone has consistent rock mass type and structural characteristics.

[0252] Specifically, a detailed geological survey is conducted on the excavation surface, including the identification and recording of various rock mass characteristics, such as lithology, degree of weathering, joints, and fissures. Based on the geological survey results, the excavation surface is divided into different rock mass type areas. The rock mass type of each area should be independent and unified, with clear geological characteristics. Within each rock mass type area, different structural zones are further divided. The division of structural zones is based on structural characteristics such as the number, direction, and spacing of the original structural planes, ensuring that the structural characteristics within each zone are consistent.

[0253] S32. For each structural partition, count the occurrence parameters of all original structural surfaces in the structural partition, draw a statistical graph of the occurrence parameters of the original structural surfaces, and analyze the spatial distribution pattern of the original structural surfaces;

[0254] Specifically, within each structural partition, measurement tools such as a compass, laser scanner, or other geological tools are used to record the attitude parameters of all original structural surfaces, including dip, inclination, spacing, and trace length. The collected original structural surface data are organized into a database to ensure data integrity and accuracy. Geological software Dips is used to draw statistical maps of the attitude parameters of the original structural surfaces, such as polar elongation maps. The spatial distribution patterns of the original structural surfaces are analyzed through statistical maps to identify major trends and characteristics.

[0255] S33, performing aggregation analysis on the statistical graph to identify the aggregation area and dominant direction of the original structural surface;

[0256] Specifically, statistical graphs are used to perform clustering analysis. This embodiment uses K-means clustering analysis to identify clustered areas of structural surfaces, analyze the dominant directions of structural surfaces in the clustered areas, and quantitatively describe these directions.

[0257] S34. Based on the aggregation analysis results, select an aggregation area representing the structural characteristics of the original structural surface in the structural partition, define the original structural surface therein as a typical structural surface, and use the typical structural surface as the representative structural surface of the corresponding structural partition;

[0258] Specifically, according to the results of the aggregation analysis, the aggregation area of ​​the original structural surfaces in each structural partition is selected, and several representative structural surfaces are selected in the aggregation area. These structural surfaces are defined as typical structural surfaces to serve as the representative structural surfaces of the structural partition;

[0259] S35. For the representative structural surface of each structural partition, estimate the probability distribution of the occurrence parameter of the representative structural surface using the Weibull distribution probability, and construct a probability density function of the occurrence parameter of the representative structural surface based on the estimated probability distribution of the occurrence parameter of the representative structural surface;

[0260] Specifically, the Weibull distribution is used to fit the occurrence parameters of the structural surface, including dip, inclination, spacing, and trace length. The parameters of the Weibull distribution are estimated using maximum likelihood, and based on the estimated parameters, the probability density function of the Weibull distribution representing the occurrence parameters of the structural surface is established. The process is as follows:

[0261] The probability density function of the Weibull distribution is expressed as:

[0262]

[0263] Where λ is the scale parameter; k is the shape parameter; x represents the observed data, that is, the occurrence parameter of the structural surface;

[0264] Use maximum likelihood estimation (MLE) to estimate the parameters λ and k of the Weibull distribution;

[0265] Given observation data x1, x2, ..., x n , the likelihood function L of the Weibull distribution is:

[0266]

[0267] To simplify the calculation, take the logarithm to get the log-likelihood function ln L:

[0268]

[0269] To find the estimated values ​​of the parameters, we need to take the derivatives with respect to the parameters λ and k and set them to zero. The steps are:

[0270] Step a: Derivative the parameter λ and set it to zero:

[0271]

[0272] Step b: Derivative the parameter k and set it to zero:

[0273]

[0274] Step c: Solve these two equations using the Newton-Raphson method to obtain the estimated values ​​of parameters λ and k.

[0275] According to the estimated value The probability density function of the Weibull distribution representing the occurrence parameter of the structural surface is constructed as follows:

[0276]

[0277] S36. Evaluate the fitting effect of the probability density function; specifically, use the Kolmogorov-Smirnov test (KS test) to evaluate the fitting effect.

[0278] S37. Repeat steps S35 to S36 for each structural partition until the fitting effect meets the requirements, and obtain the probability density functions of the representative structural surfaces of all structural partitions.

[0279] Specifically, in one embodiment of the present invention, step S4 includes:

[0280] S41. Counting the number of representative structural surfaces in each structural partition and calculating the development volume density of the representative structural surfaces; the volume density ρ is defined as the number of representative structural surfaces per unit volume, ρ = N / V, where N is the number of representative structural surfaces in the structural partition and V is the volume of the structural partition;

[0281] S42, multiplying the probability density function representing the structural surface by the development volume density representing the structural surface to obtain a volume density function representing the structural surface;

[0282] Specifically, the volume density ρ is multiplied by the Weibull distribution probability density function representing the occurrence parameter of the structural surface to obtain the volume density function fV(x) representing the structural surface:

[0283]

[0284] S43. Using the Monte Carlo method, generate R random structural surfaces in each structural partition according to the volume density function representing the structural surface. The specific steps are as follows:

[0285] First, determine the boundaries of each partition, that is, the boundaries of the structural partition;

[0286] Then, within each partition, sampling is performed based on the volume density function fV(x) to define the distribution and value range of the structural surface within the partition: points are randomly selected within the volume range of each partition, and at each point, the inverse transform sampling method or acceptance-rejection method is used to determine the specific structural surface properties, such as angle, orientation, size, etc., based on fV(x);

[0287] Repeat the above steps until R structural surfaces are generated.

[0288] Specifically, in one embodiment of the present invention, step S5 includes:

[0289] S51. Use 3D modeling software to construct a geometric model of the rock mass structure, add rock mass type, structural partitions, and original structural surfaces to the geometric model, and form a BIM model of the rock mass structure;

[0290] Import geological exploration data and structural surface data into 3D modeling software, and use modeling tools to draw the geometric shape of the rock mass structure, including the overall outline and internal features; mark different rock mass types, such as bedrock and soil layers, in the geometric model; divide different structural zones and mark them in the model; based on the preliminary exploration data, add original structural surfaces, ensuring that their position and shape are consistent with the actual situation;

[0291] S52. Import the random structural surfaces into the BIM model and identify the structural partition where each random structural surface is located:

[0292] For the u-th random structure surface, determine its center point P u Is it located in the vth structural partition? If point P u If it is located in the structural partition v, the structural partition identifier of the random structural surface u is set to v; the generated random structural surface data is imported into the BIM model in an appropriate format, such as CSV or XML format.

[0293] S53, integrating the original structural surface and the random structural surface of each structural partition according to the recognition result of the random structural surface to obtain a comprehensive structural surface;

[0294] Specifically, for each structural partition, all original structural surfaces and identified random structural surfaces are extracted; these structural surfaces are integrated into a comprehensive model, and the accuracy and rationality of the integrated structural surfaces are verified. If necessary, manual adjustments or fine-tuning using the software's optimization tools are performed;

[0295] S54. In the BIM model, for each structural partition, according to the number of comprehensive structural surfaces contained therein, randomly select Y target structural surfaces from the comprehensive structural surfaces, wherein the number of target structural surfaces is the same as the number of original structural surfaces;

[0296] Count the number of comprehensive structural surfaces in each structural partition, and randomly select Y comprehensive structural surfaces that are equal to the number of original structural surfaces. You can use a random algorithm or a built-in random extraction function in a programming language to mark the selected structural surfaces in the BIM model.

[0297] S55. Perform three-dimensional visual simulation on the selected Y target structural surfaces to generate a structural surface network model, i.e., a simulation result.

[0298] Specifically, use the visualization tools provided by the BIM software to 3D render the selected target structural surface, organize and render the results, and generate a structural surface network model. You can also use extension plug-ins or scripts to further adjust and beautify the model.

[0299] See also Figure 2In another embodiment of the present invention, a rock mass structural surface network simulation system integrating artificial intelligence and laser scanning is provided, including a point cloud acquisition and fusion module, an original structural surface and its attitude parameter acquisition module, a representative structural surface and its probability density function acquisition module, a random structural surface acquisition module, and a three-dimensional simulation module; the point cloud acquisition and fusion module is used to use a three-dimensional laser scanner and an unmanned aerial vehicle to perform point cloud acquisition and fusion on the rock mass excavation surface to obtain point cloud data. Its specific implementation method is the same as steps S11 to S17 and will not be repeated here; the original structural surface and its attitude parameter acquisition module is used to cluster the point cloud data using a clustering algorithm, analyze the cluster clusters, obtain the original structural surface, determine the spatial range of the original structural surface, calculate the normal vector of each original structural surface, and obtain the attitude parameters of the original structural surface according to the normal vector. Its specific implementation method is the same as steps S21 to S27 and will not be repeated here; the representative structural surface and its probability density function acquisition module is used to According to the rock type and structural partition, the original structural surface is statistically analyzed, and a typical structural surface is selected according to the analysis results. The typical structural surface is used as the representative structural surface of its corresponding structural partition, and the probability density function of the representative structural surface is fitted according to the attitude parameters of the representative structural surface. The specific implementation method is the same as steps S31 to S35, which will not be repeated here; the random structural surface acquisition module is used to convert the probability density function of the representative structural surface into the volume density function of the representative structural surface according to the development volume density of the representative structural surface, and generate the random structural surface using the Monte Carlo simulation method according to the volume density function of the representative structural surface. The specific implementation method is the same as steps S41 to S43, which will not be repeated here; the three-dimensional simulation module is used to construct a BIM model of the rock structure, and perform three-dimensional spatial simulation on the rock structural surface network based on the BIM model and the random structural surface to obtain simulation results. The specific implementation method is the same as steps S51 to S53, which will not be repeated here.

[0300] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A rock mass structural surface network simulation method integrating artificial intelligence and laser scanning, characterized by: The rock mass structural surface network simulation method comprises: Step S1: using a 3D laser scanner and a drone to acquire and fuse point clouds of the rock mass excavation surface to obtain point cloud data; Step S2: clustering the point cloud data using a clustering algorithm, analyzing the clusters to obtain the original structural surface, determining the spatial range of the original structural surface, calculating the normal vector of each original structural surface, and obtaining the occurrence parameters of the original structural surface according to the normal vector; Step S3: performing statistical analysis on the original structural planes according to the rock mass type and structural partition, selecting typical structural planes based on the analysis results, taking the typical structural planes as representative structural planes of the corresponding structural partitions, and fitting the probability density function of the representative structural planes based on the occurrence parameters of the representative structural planes; Step S4: converting the probability density function of the representative structural surface into a volume density function of the representative structural surface according to the development volume density of the representative structural surface, and generating a random structural surface using a Monte Carlo simulation method according to the volume density function of the representative structural surface; Step S5: constructing a BIM model of the rock mass structure, and performing a three-dimensional spatial simulation on the rock mass structural surface network based on the BIM model and the random structural surface to obtain a simulation result; The step S1 comprises: Step S11: Arrange control points around the rock mass excavation surface, and use a total station to measure the three-dimensional coordinates of the control points to generate control point data; Step S12: selecting a scanning site, and scanning the rock mass excavation surface at the scanning site using a three-dimensional laser scanner to obtain laser scanning data, wherein the laser scanning data covers all areas of the rock mass excavation surface; Step S13: designing a flight route, equipping the drone with a high-resolution camera, photographing the rock excavation surface along the flight route, obtaining drone image data, and acquiring the drone's GPS trajectory data; Step S14: pre-processing the laser scanning data and the UAV image data to remove abnormal data and bad images, thereby obtaining pre-processed laser scanning data and UAV image data; Step S15: Using the control point data, the laser scanning data of different scanning sites are spliced ​​and registered to obtain a laser point cloud; Step S16: Process the drone image data according to the GPS trajectory data of the drone to generate an image point cloud; Step S17: Using the control point data, the laser point cloud and the image point cloud are simultaneously fused and completed to obtain complete point cloud data; The step S17 includes: Step S171: Calculate the initial coordinate transformation relationship between the laser point cloud and the image point cloud using the three-dimensional coordinates of the control points. The initial coordinate transformation relationship includes an initial translation vector and an initial rotation matrix. Establish a common coordinate system for the laser point cloud and the image point cloud based on the initial coordinate transformation relationship, and perform a coarse registration of the laser point cloud and the image point cloud. The initial coordinate transformation relationship is expressed as follows: ； In the above formula, are the homogeneous coordinates of the points in the laser point cloud; are the homogeneous coordinates of the points in the transformed image point cloud; is the initial translation vector of 3×1, which represents the relative translation relationship between the two coordinate systems; is the initial rotation matrix The elements in the initial rotation matrix Represents the relative rotation relationship between two coordinate systems; Step S172: Optimize the coordinate transformation relationship between the laser point cloud and the image point cloud using the ICP algorithm to find the best coordinate transformation relationship and accurately align the laser point cloud and the image point cloud. The optimization goal of the ICP algorithm is: ； In the above formula, It is the first Points, is the number of points in the laser point cloud; For the image point cloud The nearest point, is the number of image point clouds; Step S173: Using a fusion and completion algorithm to fuse and complete the registered laser point cloud and image point cloud to generate complete point cloud data; The fusion completion algorithm includes: Step 1: Initially fuse the registered laser point cloud and image point cloud to generate the initial fused point cloud ; Step 2: Detect missing areas in the fused point cloud; Step 3: Select a point in the missing area point cloud ; Step 4: Fusion point cloud Search distance point closest and less than of points, recorded as candidate fusion point set ;in, is the maximum search distance; Step 5: Calculate points With each candidate fusion point The distance weight between : ； In the above formula, Parameters that control the speed of weight decay; Step 6: Calculate points Completion point : ； In the above formula, is the weight of the completion point, which is calculated according to the local density of the completion point; Step 7: Completion Points To perform smoothing: ； In the above formula, represents the completed points after smoothing; is the smoothing factor, which is used to control the smoothness between the completed points and the fused point cloud; Indicates completion point In the fusion point cloud The Laplace operator in ; Step 8: Smooth the completed points Add to the completed point cloud; Step 9: Repeat steps 3 to 8 until all points in the missing area are processed to obtain the final completed point cloud; Step 10: Combine the final completed point cloud with the fused point cloud Perform fusion, update the fused point cloud, and calculate the confidence of the updated fused point cloud; Step 11: Repeat steps 2 to 10 until the preset number of iterations is reached or the confidence level of the fused point cloud meets the requirement, and the final fused point cloud, i.e., the complete point cloud data, is obtained.

2. The rock mass structural surface network simulation method integrating artificial intelligence and laser scanning according to claim 1, characterized in that: The step S2 comprises: Step S21: Determine the number of components of the clustering model , a component represents a cluster, randomly selected points as the initial mean vector, and initialize the covariance matrix of each component to the identity matrix, and initialize the weight of each component to ; Step S22: Calculate the posterior probability that each point in the point cloud data belongs to each component; Step S23: updating the parameters of the clustering model, i.e., the mean vector, covariance matrix, and weight, according to the posterior probability; Step S24, repeating steps S22 to S23 until the parameters of the clustering model converge or the maximum number of iterations is reached; Step S25: assign each point to the component with the largest posterior probability to obtain a clustering result; Step S26: For each cluster in the clustering results, calculate the spatial range of the cluster, use the least squares method to fit a plane based on the spatial range of each cluster, minimize the sum of the squares of the distances from the points in the cluster to the plane, and calculate the fitting error. If the fitting error is less than a preset threshold, it is determined that the cluster corresponds to an original structural surface; Step S27: Calculate the normal vector of the original structural surface according to the fitting plane, and calculate the attitude parameters of the original structural surface based on the normal vector. The attitude parameters include dip, inclination, spacing and trace length.

3. The rock mass structural surface network simulation method integrating artificial intelligence and laser scanning according to claim 2, characterized in that: The calculation formula of the posterior probability is: ； In the above formula, Indicates the points; Indicates the The point belongs to The posterior probability of the components; Indicates the The weight of the components, satisfy ; Indicates the The mean vector of the components; Indicates the The covariance matrix of the components; Indicates is the mean vector, The Gaussian probability function with the covariance matrix is ​​at the point The value at middle, From 1 to Traversing means summing all components.

4. The rock mass structural surface network simulation method integrating artificial intelligence and laser scanning according to claim 3, characterized in that: The update formula of the mean vector is: ； The updating formula of the covariance matrix is: ； The weight update formula is: ； In the above formula, Indicates the number of points in the point cloud data.

5. The rock mass structural surface network simulation method integrating artificial intelligence and laser scanning according to claim 1, characterized in that: The step S3 comprises: S31. Divide the rock mass into different types based on the geological characteristics of the rock mass excavation surface. Within each rock mass type, divide the rock mass excavation surface into different structural zones based on the structural characteristics, wherein each structural zone has consistent rock mass type and structural characteristics. S32. For each structural partition, count the occurrence parameters of all original structural surfaces in the structural partition, draw a statistical graph of the occurrence parameters of the original structural surfaces, and analyze the spatial distribution pattern of the original structural surfaces; S33, performing aggregation analysis on the statistical graph to identify the aggregation area and dominant direction of the original structural surface; S34. Based on the aggregation analysis results, select an aggregation area representing the structural characteristics of the original structural surface in the structural partition, define the original structural surface therein as a typical structural surface, and use the typical structural surface as the representative structural surface of the corresponding structural partition; S35. For the representative structural surface of each structural partition, estimate the probability distribution of the occurrence parameter of the representative structural surface using the Weibull distribution probability, and construct a probability density function of the occurrence parameter of the representative structural surface based on the estimated probability distribution of the occurrence parameter of the representative structural surface; S36. Evaluate the fitting effect of the probability density function; S37. Repeat steps S35 to S36 for each structural partition until the fitting effect meets the requirements, and obtain the probability density functions of the representative structural surfaces of all structural partitions.

6. The rock mass structural surface network simulation method integrating artificial intelligence and laser scanning according to claim 1, characterized in that: The step S4 comprises: S41. Counting the number of representative structural surfaces in each structural partition, and calculating the development volume density of the representative structural surfaces; S42, multiplying the probability density function representing the structural surface by the development volume density representing the structural surface to obtain a volume density function representing the structural surface; S43, using the Monte Carlo method, based on the volume density function representing the structural surface, generate A random structure surface.

7. The rock mass structural surface network simulation method integrating artificial intelligence and laser scanning according to claim 1, characterized in that: The step S5 comprises: S51. Use 3D modeling software to construct a geometric model of the rock mass structure, add rock mass type, structural partitions, and original structural surfaces to the geometric model, and form a BIM model of the rock mass structure; S52. Import the random structural surfaces into the BIM model and identify the structural partition where each random structural surface is located: For the Random structure surface, determine its center point Is it located in In a structural partition, if the point Located in the structural partition In the random structure surface The structure partition identifier is set to ; S53, integrating the original structural surface and the random structural surface of each structural partition according to the recognition result of the random structural surface to obtain a comprehensive structural surface; S54. In the BIM model, for each structural partition, according to the number of comprehensive structural surfaces it contains, randomly select target structural surfaces, where the number of target structural surfaces is the same as the number of original structural surfaces; S55、Select A three-dimensional visual simulation is performed on each target structural surface to generate a structural surface network model, which is the simulation result.

Images