Geometric digital twin reverse modeling method and system for subway tunnel, electronic equipment and storage medium

Abstract

The invention discloses a subway tunnel geometric digital twin reverse modeling method and system, electronic equipment and a storage medium, and the modeling method comprises the steps: carrying out the preprocessing of tunnel point cloud data, and carrying out the voxel downsampling, denoising processing and curved surface separation of the tunnel point cloud data; extracting a tunnel axis based on a particle swarm optimization algorithm, generating a plurality of sections of linear fitting axes by fitting a plurality of axis points, and accurately describing a complex tunnel curve; performing multi-dimensional semantic segmentation on the tunnel point cloud data, extracting tunnel components and defects through local feature change analysis, and quantifying leakage defects; reverse modeling is carried out on the extracted components based on a modeling strategy from bottom to top, ring-level and tunnel-level models are constructed step by step, and reconstruction of an overall three-dimensional model of the tunnel is achieved; and performing geometric detail enhancement on the reconstructed tunnel geometric digital twin model to ensure model precision and multi-dimensional semantic expression. According to the method, the automation degree, adaptability and modeling efficiency of tunnel twin modeling can be improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of tunnel modeling, and relates to a tunnel modeling method, in particular to a reverse modeling method, system, electronic device and storage medium for the geometric digital twin of subway tunnels. Background Art

[0002] As an important part of the urban transportation network, the structural state of tunnels is directly related to the safety and efficiency of the transportation system. In the design, construction and operation and maintenance stages of the tunnel's entire life cycle, digital twin technology has become one of the key means. Through the construction of a geometric digital twin model (gDT), accurate modeling of the tunnel can be achieved, which provides important technical support for the intelligent operation and maintenance of the tunnel. The generation of the tunnel geometric digital twin model mainly depends on the processing of point cloud data. By preprocessing, axis extraction, semantic segmentation and reverse modeling of the tunnel point cloud data, a high-fidelity, semantically rich three-dimensional digital model can be constructed. However, existing technical methods still have many limitations in practical applications and are difficult to meet the requirements of complex tunnel twin modeling and rapid automated modeling.

[0003] Currently, most tunnel digital twin modeling methods rely on predefined component libraries or parametric design tools, which require modelers to have rich professional knowledge and need to fully interpret the tunnel design drawings. This method is not only complex and inefficient in operation, but also difficult to accurately reflect the actual geometric state of the tunnel due to errors or deformations during actual tunnel construction. In addition, although the introduction of deep learning models has improved the ability of semantic segmentation and component recognition to a certain extent, this method highly depends on large-scale labeled data and high-performance computing devices and is limited in actual tunnel scenarios with scarce data. At the same time, deep learning models have a long training time and high cost, and it is difficult to maintain good generalization ability under changing tunnel conditions. In addition, geometric modeling of tunnels with complex curvatures and long distances also poses great challenges to existing methods. Traditional tunnel axis extraction methods, such as the RANSAC algorithm and the least squares method based on the cylindrical model hypothesis, are easily interfered by noise points and outliers and have insufficient fitting accuracy when dealing with tunnels with variable curvatures. At the same time, the recognition and classification of tunnel components mainly focus on lining blocks, and the semantic extraction of other components such as cables, tracks, and sleepers is not comprehensive enough to meet the needs of the tunnel digital twin model to express multi-dimensional semantic information at different levels.

[0004] Therefore, existing modeling methods have deficiencies in dependence, efficiency, accuracy and adaptability, and are difficult to meet the modeling requirements in complex tunnel scenarios.

[0005] In view of this, there is an urgent need to design a new tunnel modeling method to overcome at least some of the above defects existing in existing tunnel modeling methods. Summary of the Invention

[0006] The present invention provides a method, system, electronic device and storage medium for reverse modeling of geometric digital twins of subway tunnels, which can improve the automation, adaptability and modeling efficiency of tunnel twin modeling.

[0007] To solve the above technical problems, according to one aspect of the present invention, the following technical solutions are adopted:

[0008] A method for reverse modeling of geometric digital twins of subway tunnels, the modeling method includes:

[0009] Step S1, preprocess the tunnel point cloud data, including voxel downsampling, denoising processing and surface separation of the tunnel point cloud data;

[0010] Step S2, extract the tunnel axis based on the particle swarm optimization algorithm, generate multiple segments of linearly fitted axes by fitting multiple axis points, and accurately describe complex tunnel curves;

[0011] Step S3, perform multi-dimensional semantic segmentation on the tunnel point cloud data, extract tunnel components and defects through local feature change analysis, including lining blocks, cables, tracks, sleepers and leakage defects, and quantify the leakage defects;

[0012] Step S4, perform reverse modeling on the extracted components based on a bottom-up modeling strategy, gradually construct ring-level and tunnel-level models, and realize the reconstruction of the overall three-dimensional model of the tunnel;

[0013] Step S5, enhance the geometric details of the reconstructed geometric digital twin model of the tunnel to ensure the model accuracy and multi-dimensional semantic expression.

[0014] As an implementation manner of the present invention, in step S1, the preprocessing of the tunnel point cloud data includes:

[0015] Step S11, adopt the voxel downsampling method, select the centroid points of each voxel unit to reduce the data volume while retaining key geometric information;

[0016] Step S12, use the Gaussian statistical filtering method, calculate the average distance between each point in the tunnel point cloud data and its neighborhood points, and remove the outliers;

[0017] Step S13, separate the plane and surface parts of the tunnel point cloud through the plane RANSAC algorithm to provide support for subsequent tunnel axis extraction.

[0018] In step S2, the steps of extracting the tunnel axis based on the particle swarm optimization algorithm include:

[0019] Step S21, define the objective function, and minimize the distance variance between the tunnel point cloud data and the axis;

[0020] Step S22: Generate axis points through the particle swarm optimization algorithm, where the axis points satisfy the sorting constraint and the spatial range constraint;

[0021] Step S23: Connect the generated axis points into a polyline to adapt to the complex tunnel curvature.

[0022] In step S3, the semantic segmentation of the tunnel point cloud data includes:

[0023] Step S311: Analyze the change of normal vectors in the tunnel point cloud data, segment the lining blocks and identify the component boundaries;

[0024] Step S312: Use a sliding window to extract the local features of the cable and the track, and separate different components based on the change of geometric height;

[0025] Step S313: Combine the strength value and the change of local point density to detect leakage defects and calculate the defect area.

[0026] The detection of leakage defects includes:

[0027] Step S321: Perform threshold filtering on the point intensity value in the local area to extract potential defect points with low intensity;

[0028] Step S322: Use the Alpha Shape algorithm to calculate the defect boundary and estimate the defect area.

[0029] In step S4, the specific steps adopted in reverse modeling include:

[0030] Step S41: Generate a single-ring geometric model through Poisson reconstruction, spherical projection, and Delaunay triangulation at the ring level;

[0031] Step S42: Assemble the ring-level model into an overall tunnel model using the spatial pose matrix.

[0032] In step S5, the model detail optimization includes:

[0033] Step S51: Optimize the geometric features of the lining block model through normal adjustment and thickness enhancement;

[0034] Step S52: Integrate the defect information with the geometric model to generate a three-dimensional model with rich semantics.

[0035] As an implementation manner of the present invention, in step S1, the tunnel point cloud data is optimized through voxel downsampling and noise point filtering, and the curved surface is separated through the RANSAC algorithm, which specifically includes:

[0036] Step S11: Use the voxel downsampling technique to compress the point cloud data to reduce the data volume and retain the geometric features of the tunnel structure;

[0037]

[0038] Among them, P v is the coordinate (geometric center) of the downsampled point, and P i is the point coordinate in the original point cloud belonging to the same voxel, and N is the total number of points belonging to the voxel;

[0039] Step S12: Use the Gaussian filtering method to remove the noise points in the point cloud and optimize the data quality;

[0040]

[0041] Among them, d i is the local average distance of point P i for measuring noise; P i is the coordinate of the current point; P j is the coordinate of the k points closest to P i ; k is the number of points in the neighborhood;

[0042] Step S13: Use the RANSAC algorithm to separate the surface part, extract the main data of the tunnel lining, and remove the redundant interference points;

[0043] In step S2, the tunnel axis is extracted by the particle swarm optimization algorithm, specifically including:

[0044] Step S21: Define the objective function and calculate the minimum distance variance from the point cloud to the axis;

[0045]

[0046] d i = ||p i - p'||

[0047]

[0048] Among them, d i represents the minimum distance from the i-th point in the point cloud to the axis; represents the average value of the distances from all points to the axis; N represents the total number of points in the point cloud; P i represents the coordinate of the i-th point in the point cloud; P s represents the coordinate of a point on the axis; V represents the direction vector of the axis; P' represents the projection point of point P i on the axis;

[0049] Step S22: Initialize the position and velocity of the particle swarm and iteratively optimize the position of the axis points;

[0050] X = [x 1 , y 1 , z 1,…,x k+1 ,y k+1 ,z k+1

[0051] x 1 <x 2 <…<x k+1

[0052] X min <X<X max

[0053] Among them, X represents the position vector of the particle and represents the set of tunnel axis points; x 1 <x 2 <…<x k+1 means that the x - coordinate of the particle increases successively in the tunnel direction; X min <X<X max represents the position constraint range of the particle;

[0054] Step S23: After generating the axis points, use the polyline fitting technology to construct the complete tunnel axis curve for subsequent semantic segmentation and reverse modeling.

[0055] As an implementation manner of the present invention, in step S3, the process of semantic segmentation of the tunnel point cloud data includes:

[0056] Step S31: Analyze the local feature changes of the point cloud through a sliding window and extract the geometric features of the tunnel components;

[0057]

[0058] Among them, μ i represents the mean value of the local feature changes of point P i and is used to describe the geometric features of the tunnel components; |P i | represents the number of points in the local neighborhood of P i ; P i represents the neighborhood set of point i in the point cloud; p j represents a point belonging to the neighborhood of point P i ; β θ,j represents the eigenvalue (such as curvature or gradient) of point p j in a specific direction θ;

[0059] Step S32: Detect tunnel defects (such as leakage) based on the intensity value and geometric feature changes, and use the AlphaShape algorithm to extract the contour and quantify the area of the defect area;

[0060]

[0061] Among them, A represents the area of the defect area; (x k ,y​k ) represents the coordinates of the k-th point on the Alpha Shape boundary; n represents the total number of boundary points;

[0062] Step S33: Classify the point cloud using multi-dimensional semantic information and label the components and defect locations to provide semantic support for subsequent reverse modeling;

[0063] In step S4, the reverse modeling process of the tunnel includes the following steps:

[0064] Step S41: In the ring-level component reconstruction, use Poisson surface reconstruction, spherical projection, and Delaunay triangulation methods to generate geometric models of components such as tunnel lining blocks;

[0065] Step S42: Enhance the surface thickness of the generated components;

[0066] Step S43: According to the spatial pose of the components and the thickness enhancement algorithm, construct the overall three-dimensional model of the tunnel and perform geometric detail optimization;

[0067]

[0068] Among them, T represents the spatial transformation matrix of the geometric model; s x , s y , s z represents the scaling factor; R ij represents the elements of the rotation matrix; t x , t y , t z represents the elements of the translation vector.

[0069] A reverse modeling system for the geometric digital twin of a subway tunnel, the modeling system includes:

[0070] A data preprocessing module for preprocessing tunnel point cloud data, including voxel downsampling, denoising, and surface separation of the tunnel point cloud data;

[0071] A tunnel curve description module for extracting the tunnel axis based on the particle swarm optimization algorithm and generating multiple linear fitting axes by fitting multiple axis points to accurately describe complex tunnel curves;

[0072] A leakage defect quantification module for performing multi-dimensional semantic segmentation on the tunnel point cloud data, extracting tunnel components and defects, including lining blocks, cables, tracks, sleepers, and leakage defects, through local feature change analysis, and quantifying the leakage defects;

[0073] A three-dimensional model reconstruction module for reverse modeling the extracted components based on a bottom-up modeling strategy, gradually constructing ring-level and tunnel-level models, and realizing the reconstruction of the overall three-dimensional model of the tunnel;

[0074] A geometric detail enhancement module is used to enhance the geometric details of the reconstructed tunnel geometric digital twin model to ensure model accuracy and multi-dimensional semantic expression.

[0075] As an implementation manner of the present invention, the preprocessing process of the data preprocessing module for tunnel point cloud data includes:

[0076] Step S11: Adopt a voxel downsampling method to select the centroid points of each voxel unit to reduce the data volume while retaining key geometric information;

[0077] Step S12: Use the Gaussian statistical filtering method to calculate the average distance between each point in the tunnel point cloud data and its neighboring points and remove the outlier points;

[0078] Step S13: Separate the plane and curved surface parts of the tunnel point cloud through the plane RANSAC algorithm to provide support for subsequent tunnel axis extraction.

[0079] The process of the tunnel curve description module extracting the tunnel axis based on the particle swarm optimization algorithm includes:

[0080] Step S21: Define the objective function to minimize the distance variance between the tunnel point cloud data and the axis;

[0081] Step S22: Generate axis points through the particle swarm optimization algorithm, and the axis points satisfy the sorting constraint and the spatial range constraint;

[0082] Step S23: Connect the generated axis points into a polyline to adapt to the complex tunnel curvature.

[0083] The semantic segmentation process of the leakage defect quantification module for tunnel point cloud data includes:

[0084] Step S311: Analyze the change of the normal vector in the tunnel point cloud data, segment the lining blocks and identify the component boundaries;

[0085] Step S312: Use a sliding window to extract the local features of the cable and the track, and separate different components based on the change of geometric height;

[0086] Step S313: Combine the strength value and the change of local point density to detect leakage defects and calculate the defect area.

[0087] The detection process of the leakage defect quantification module for leakage defects includes:

[0088] Step S321: Perform threshold filtering on the point intensity values in the local area to extract potential defect points with low intensity;

[0089] Step S322: Use the Alpha Shape algorithm to calculate the defect boundary and estimate the defect area.

[0090] The reverse modeling process adopted by the three-dimensional model reconstruction module includes:

[0091] Step S41: Generate a single-loop geometric model through Poisson reconstruction, spherical projection, and Delaunay triangulation at the loop level;

[0092] Step S42: Assemble the loop-level model into an overall tunnel model using a spatial pose matrix.

[0093] The model detail optimization of the geometric detail enhancement module includes:

[0094] Step S51: Optimize the geometric features of the lining block model through normal adjustment and thickness enhancement;

[0095] Step S52: Integrate the defect information with the geometric model to generate a three-dimensional model with rich semantics.

[0096] As an implementation manner of the present invention, the data preprocessing module optimizes the tunnel point cloud data through voxel downsampling and noise point filtering, and separates the surface through the RANSAC algorithm, specifically including:

[0097] Step S11: Compress the point cloud data using voxel downsampling technology to reduce the data volume and retain the geometric features of the tunnel structure;

[0098]

[0099] Among them, P v is the coordinate (geometric center) of the downsampled point, P i is the point coordinate in the original point cloud belonging to the same voxel, and N is the total number of points belonging to the voxel;

[0100] Step S12: Use the Gaussian filtering method to remove the noise points in the point cloud and optimize the data quality;

[0101]

[0102] Among them, d i is the local average distance of point P i used to measure the noise; P i is the coordinate of the current point; P j is the coordinate of the k points closest to P i ; k is the number of points in the neighborhood;

[0103] Step S13: Use the RANSAC algorithm to separate the surface part, extract the main data of the tunnel lining, and remove the redundant interference points;

[0104] The tunnel curve description module extracts the tunnel axis through the particle swarm optimization algorithm, specifically including:

[0105] Step S21: Define the objective function and calculate the minimum distance variance from the point cloud to the axis.

[0106]

[0107] d i = ||p i - p'||

[0108]

[0109] where d i represents the minimum distance from the i-th point in the point cloud to the axis; represents the average value of the distances from all points to the axis; N represents the total number of points in the point cloud; P i represents the coordinates of the i-th point in the point cloud; P s represents the coordinates of a certain point on the axis; V represents the direction vector of the axis; P' represents the projection point of point P i on the axis.

[0110] Step S22: Initialize the positions and velocities of the particle swarm and iteratively optimize the positions of the axis points.

[0111] X = [x 1 , y 1 , z 1 , …, x k+1 , y k+1 , z k+1

[0112] x 1 < x 2 < … < x k+1

[0113] X min < X < X max

[0114] where X represents the position vector of the particle, representing the set of tunnel axis points; x 1 < x 2 < … < x k+1 means that the x-coordinates of the particles increase sequentially in the tunnel direction; X min < X < X max represents the position constraint range of the particles;

[0115] Step S23: After generating the axis points, use the polyline fitting technology to construct the complete tunnel axis curve for subsequent semantic segmentation and reverse modeling.

[0116] ​As an implementation mode of the present invention, the process of the leakage defect quantification module performing semantic segmentation on the tunnel point cloud data includes:

[0117] Step S31: Analyze the local feature changes of the point cloud through a sliding window, and extract the geometric features of the tunnel components;

[0118]

[0119] Among them, μ i represents the mean value of the local feature changes of point P i and is used to describe the geometric features of the tunnel components; |P i | represents the number of points in the local neighborhood of P i ; P i represents the neighborhood set of point i in the point cloud; p j represents the point belonging to the neighborhood of point P i ; β θ,j represents the eigenvalue (such as curvature or gradient) of point p j in a specific direction θ;

[0120] Step S32: Detect tunnel defects (such as leakage) based on the intensity value and geometric feature changes, and use the AlphaShape algorithm to extract the contour and quantify the area of the defect area;

[0121]

[0122] Among them, A represents the area of the defect area; (x k , y k ) represents the coordinates of the kth point on the Alpha Shape boundary; n represents the total number of boundary points;

[0123] Step S33: Classify the point cloud using multi-dimensional semantic information and label the positions of the components and defects, providing semantic support for subsequent reverse modeling;

[0124] The tunnel reverse modeling process of the three-dimensional model reconstruction module includes the following steps:

[0125] Step S41: In the ring-level component reconstruction, use the Poisson surface reconstruction, spherical projection, and Delaunay triangulation methods to generate the geometric models of components such as tunnel lining blocks;

[0126] Step S42: Enhance the surface thickness of the generated components;

[0127] Step S43: According to the spatial pose of the components and the thickness enhancement algorithm, construct the overall three-dimensional model of the tunnel and perform geometric detail optimization;

[0128]

[0129] Among them, T represents the spatial transformation matrix of the geometric model; s x , s y , s z represents the scaling factor; R ij represents the elements of the rotation matrix; t x , t y , t z represents the elements of the translation vector.

[0130] According to another aspect of the present invention, the following technical solution is adopted: An electronic device includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the steps of the above method are implemented.

[0131] According to another aspect of the present invention, the following technical solution is adopted: A storage medium stores computer program instructions thereon. When the computer program instructions are executed by a processor, the steps of the above method are implemented.

[0132] The beneficial effects of the present invention are as follows: The subway tunnel geometric digital twin reverse modeling method, system, electronic device, and storage medium proposed by the present invention can improve the automation degree, adaptability, and modeling efficiency of tunnel twin modeling.

[0133] (1) The present invention does not need to rely on a predefined component library, parametric design tool, or neural network, and realizes 3D model reconstruction completely based on point cloud data, significantly improving the adaptability and flexibility of modeling;

[0134] (2) Through the particle swarm optimization algorithm and polyline fitting method, the problem of insufficient axis extraction accuracy of traditional methods under complex tunnel geometric shapes is effectively solved;

[0135] (3) Semantic segmentation is realized by local feature change analysis, and components and defects are identified by combining multi-dimensional semantic features, improving the semantic expression ability of the model;

[0136] (4) Through a bottom-up reverse modeling strategy, the rapid construction of a fine geometric model from a rough geometry is realized, meeting the digital twin modeling requirements of different LoD levels;

[0137] (5) The full-process automated processing significantly reduces the modeling cost, improves the construction efficiency of the tunnel geometric digital twin model, and provides technical support for the intelligent operation and maintenance of the tunnel. Description of the Drawings

[0138] Figure 1 is a flowchart of the subway tunnel geometric digital twin reverse modeling method in an embodiment of the present invention.

[0139] Figure 2Schematic diagram of the composition of the subway tunnel geometric digital twin reverse modeling system in an embodiment of the present invention.

[0140] Figure 3 Schematic diagram of the bottom-up tunnel geometric twin model construction process in an embodiment of the present invention.

[0141] Figure 4 Schematic diagram of the final tunnel geometric digital twin model (partial) in an embodiment of the present invention.

[0142] Figure 5 Schematic diagram of the explosion of the final tunnel geometric digital twin model (partial) in an embodiment of the present invention.

[0143] Figure 6 Schematic diagram of the composition of the electronic device in an embodiment of the present invention. Detailed implementation manners

[0144] The preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0145] To further understand the present invention, the preferred implementation manners of the present invention will be described below in conjunction with embodiments. However, it should be understood that these descriptions are only for further explaining the features and advantages of the present invention, rather than limiting the claims of the present invention.

[0146] The description of this part only targets several typical embodiments, and the present invention is not limited to the scope described in the embodiments. The mutual replacement of the same or similar prior art means and some technical features in the embodiments is also within the scope of the description and protection of the present invention.

[0147] The expressions of the steps in each embodiment in the specification are only for convenience of description, and the implementation manner of the present application is not limited by the order of step implementation.

[0148] "Connection" in the specification includes both direct connection and indirect connection.

[0149] The present invention discloses a subway tunnel geometric digital twin reverse modeling method. Figure 1 Flowchart of the subway tunnel geometric digital twin reverse modeling method in an embodiment of the present invention; please refer to Figure 1 , the modeling method includes:

[0150]

Step S1

[0151]

Step S2

[0152]

Step S3

[0153]

Step S4

[0154]

Step S5

[0155] In an embodiment of the present invention, in step S1, the preprocessing of the tunnel point cloud data includes:

[0156] Step S11: Adopt a voxel downsampling method, select the centroid point of each voxel unit to reduce the data volume while retaining key geometric information;

[0157] Step S12: Use a Gaussian statistical filtering method to calculate the average distance between each point in the tunnel point cloud data and its neighboring points and remove outliers;

[0158] Step S13: Separate the plane and curved surface parts of the tunnel point cloud through the plane RANSAC algorithm to provide support for subsequent tunnel axis extraction.

[0159] In step S2, the steps for extracting the tunnel axis based on the particle swarm optimization algorithm include:

[0160] Step S21: Define the objective function to minimize the distance variance between the tunnel point cloud data and the axis;

[0161] Step S22: Generate axis points through the particle swarm optimization algorithm, and the axis points satisfy sorting constraints and spatial range constraints;

[0162] Step S23: Connect the generated axis points into a polyline to adapt to the complex tunnel curvature.

[0163] In step S3, the semantic segmentation of the tunnel point cloud data includes:

[0164] Step S311: Analyze the change in the normal vector in the tunnel point cloud data, segment the lining blocks, and identify the component boundaries;

[0165] Step S312: Use a sliding window to extract the local features of the cables and tracks, and separate different components based on the change in geometric height;

[0166] Step S313: Combine the intensity value and the change in local point density to detect leakage defects and calculate the defect area.

[0167] The detection of leakage defects includes:

[0168] Step S321: Perform threshold filtering on the point intensity values in the local area to extract potential defect points with low intensity.

[0169] Step S322: Use the Alpha Shape algorithm to calculate the defect boundary and estimate the defect area.

[0170] The specific steps adopted in the reverse modeling in Step S4 include:

[0171] Step S41: Generate a single-ring geometric model through Poisson reconstruction, spherical projection, and Delaunay triangulation at the ring level.

[0172] Step S42: Assemble the ring-level model into the overall tunnel model using the spatial pose matrix.

[0173] The model detail optimization in Step S5 includes:

[0174] Step S51: Optimize the geometric features of the lining block model through normal adjustment and thickness enhancement.

[0175] Step S52: Integrate the defect information with the geometric model to generate a three-dimensional model with rich semantics.

[0176] In an embodiment of the present invention, in Step S1, the tunnel point cloud data is optimized through voxel downsampling and noise point filtering, and the curved surface is separated through the RANSAC algorithm, which specifically includes:

[0177] Step S11: Use the voxel downsampling technique to compress the point cloud data to reduce the data volume and retain the geometric features of the tunnel structure.

[0178]

[0179] where P v is the coordinate (geometric center) of the downsampled point, P i is the point coordinate in the original point cloud belonging to the same voxel, and N is the total number of points belonging to the voxel.

[0180] Step S12: Use the Gaussian filtering method to remove the noise points in the point cloud and optimize the data quality.

[0181]

[0182] where d i is the local average distance of point P i used to measure the noise; P i is the coordinate of the current point; P j is the same as P iCoordinates of the k closest points; k is the number of points in the neighborhood;

[0183] Step S13: Use the RANSAC algorithm to separate the surface part, extract the main data of the tunnel lining, and remove redundant interference points;

[0184] In step S2, the tunnel axis is extracted by the particle swarm optimization algorithm, specifically including:

[0185] Step S21: Define the objective function and calculate the minimum distance variance from the point cloud to the axis;

[0186]

[0187] d i =||p i -p’||

[0188]

[0189] where d i represents the minimum distance from the i-th point in the point cloud to the axis; represents the average value of the distances from all points to the axis; N represents the total number of points in the point cloud; P i represents the coordinates of the i-th point in the point cloud; P s represents the coordinates of a point on the axis; V represents the direction vector of the axis; P' represents the projection point of point P i on the axis;

[0190] Step S22: Initialize the positions and velocities of the particle swarm and iteratively optimize the positions of the axis points;

[0191] X=[x 1 ,y 1 ,z 1 ,…,x k+1 ,y k+1 ,z k+1

[0192] x 1 <x 2 <…<x k+1

[0193] X min <X<X max

[0194] where X represents the position vector of the particle and represents the set of tunnel axis points; x 1 <x 2 <…<x k+1 means that the x coordinates of the particles increase sequentially in the tunnel direction; X min <X<X max represents the position constraint range of the particle;​

[0195] After generating the axis points in step S23, use the polyline fitting technique to construct the complete tunnel axis curve for subsequent semantic segmentation and reverse modeling.

[0196] In step S3, the process of semantic segmentation of the tunnel point cloud data includes:

[0197] Step S31: Analyze the local feature changes of the point cloud through a sliding window to extract the geometric features of the tunnel components;

[0198]

[0199] Among them, μ i represents the mean value of the local feature changes of point P i and is used to describe the geometric features of the tunnel components; |P i | represents the number of points in the local neighborhood of P i ; P i represents the neighborhood set of point i in the point cloud; p j represents the point belonging to the neighborhood of point P i ; β θ,j represents the eigenvalue (such as curvature or gradient) of point p j in a specific direction θ;

[0200] Step S32: Detect tunnel defects (such as leakage) based on the intensity value and geometric feature changes, and use the AlphaShape algorithm to extract the contour and quantify the area of the defect area;

[0201]

[0202] Among them, A represents the area of the defect area; (x k , y k ) represents the coordinates of the kth point on the Alpha Shape boundary; n represents the total number of boundary points;

[0203] Step S33: Classify the point cloud using multi-dimensional semantic information and label the positions of the components and defects to provide semantic support for subsequent reverse modeling;

[0204] In step S4, the reverse modeling process of the tunnel includes the following steps:

[0205] Step S41: In the ring-level component reconstruction, use the Poisson surface reconstruction, spherical projection, and Delaunay triangulation methods to generate the geometric models of components such as tunnel lining blocks;

[0206] Step S42: Enhance the surface thickness of the generated components;

[0207] Step S43: Construct an overall three-dimensional model of the tunnel according to the spatial pose and thickness enhancement algorithm of the components, and perform geometric detail optimization;

[0208]

[0209] Among them, T represents the spatial transformation matrix of the geometric model; s x , s y , s z represents the scaling factor; R ij represents the elements of the rotation matrix; t x , t y , t z represents the elements of the translation vector.

[0210] The present invention further discloses a subway tunnel geometric digital twin reverse modeling system, and the modeling system includes: a data preprocessing module 1, a tunnel curve description module 2, a leakage defect quantification module 3, a three-dimensional model reconstruction module 4, and a geometric detail enhancement module 5.

[0211] The data preprocessing module 1 is used to preprocess the tunnel point cloud data, including voxel downsampling, denoising processing, and surface separation of the tunnel point cloud data.

[0212] The tunnel curve description module 2 is used to extract the tunnel axis based on the particle swarm optimization algorithm, generate multiple linear fitting axes by fitting multiple axis points, and accurately describe complex tunnel curves.

[0213] The leakage defect quantification module 3 is used to perform multi-dimensional semantic segmentation on the tunnel point cloud data, extract tunnel components and defects through local feature change analysis, including lining blocks, cables, tracks, sleepers, and leakage defects, and quantify the leakage defects.

[0214] The three-dimensional model reconstruction module 4 is used to perform reverse modeling on the extracted components based on a bottom-up modeling strategy, gradually construct ring-level and tunnel-level models, and realize the reconstruction of the overall three-dimensional model of the tunnel.

[0215] The geometric detail enhancement module 5 is used to enhance the geometric details of the reconstructed tunnel geometric digital twin model to ensure model accuracy and multi-dimensional semantic expression.

[0216] In an embodiment of the present invention, the preprocessing process of the data preprocessing module 1 for the tunnel point cloud data includes:

[0217] Step S11: Adopt a voxel downsampling method, select the centroid points of each voxel unit to reduce the data volume while retaining key geometric information;

[0218] Step S12: Use the Gaussian statistical filtering method to calculate the average distance between each point in the tunnel point cloud data and its neighborhood points, and remove the outlier points;

[0219] Step S13: Separate the planar and curved surface parts of the tunnel point cloud through the plane RANSAC algorithm to provide support for subsequent tunnel axis extraction.

[0220] The process of the tunnel curve description module 2 extracting the tunnel axis based on the particle swarm optimization algorithm includes:

[0221] Step S21: Define the objective function to minimize the distance variance between the tunnel point cloud data and the axis;

[0222] Step S22: Generate axis points through the particle swarm optimization algorithm, and the axis points satisfy the sorting constraint and the spatial range constraint;

[0223] Step S23: Connect the generated axis points into a polyline to adapt to the complex tunnel curvature.

[0224] The semantic segmentation process of the leakage defect quantification module 3 for the tunnel point cloud data includes:

[0225] Step S311: Analyze the change of the normal vector in the tunnel point cloud data, segment the lining blocks and identify the component boundaries;

[0226] Step S312: Use a sliding window to extract the local features of the cable and the track, and separate different components based on the change of geometric height;

[0227] Step S313: Combine the strength value and the change of local point density to detect leakage defects and calculate the defect area.

[0228] The detection process of the leakage defect quantification module 3 for leakage defects includes:

[0229] Step S321: Perform threshold filtering on the point intensity value in the local area to extract potential defect points with low intensity;

[0230] Step S322: Use the Alpha Shape algorithm to calculate the defect boundary and estimate the defect area.

[0231] The reverse modeling process adopted by the 3D model reconstruction module 4 includes:

[0232] Step S41: Generate a single-ring geometric model through Poisson reconstruction, spherical projection, and Delaunay triangulation at the ring level;

[0233] Step S42: Assemble the ring-level models into a tunnel overall model using the spatial pose matrix.

[0234] The model detail optimization of the geometric detail enhancement module 5 includes:

[0235] Step S51: Optimize the geometric features of the lining block model through normal adjustment and thickness enhancement;

[0236] Step S52: Integrate the defect information with the geometric model to generate a three-dimensional model with rich semantics.

[0237] In an embodiment of the present invention, the data preprocessing module 1 optimizes the tunnel point cloud data through voxel downsampling and noise point filtering, and separates the surface through the RANSAC algorithm, specifically including:

[0238] Step S11: Use the voxel downsampling technique to compress the point cloud data to reduce the data volume and retain the geometric features of the tunnel structure;

[0239]

[0240] Among them, P v is the coordinate (geometric center) of the downsampled point, P i is the point coordinate in the original point cloud belonging to the same voxel, and N is the total number of points belonging to the voxel;

[0241] Step S12: Use the Gaussian filtering method to remove the noise points in the point cloud and optimize the data quality;

[0242]

[0243] Among them, d i is the local average distance of point P i for measuring noise; P i is the coordinate of the current point; P j is the coordinate of the k points closest to P i ; k is the number of points in the neighborhood.

[0244] Step S13: Use the RANSAC algorithm to separate the surface part, extract the main data of the tunnel lining, and remove the redundant interference points.

[0245] The tunnel curve description module 2 extracts the tunnel axis through the particle swarm optimization algorithm, specifically including:

[0246] Step S21: Define the objective function and calculate the minimum distance variance from the point cloud to the axis;

[0247]

[0248] d i = ||p i - p’||

[0249]

[0250] Among them, d i represents the minimum distance from the i-th point in the point cloud to the axis; represents the average value of the distances from all points to the axis; N represents the total number of points in the point cloud; P i represents the coordinates of the i-th point in the point cloud; P s represents the coordinates of a certain point on the axis; V represents the direction vector of the axis; P' represents the projection point of point P i on the axis.

[0251] Step S22: Initialize the positions and velocities of the particle swarm, and iteratively optimize the positions of the axis points;

[0252] X = [x 1 , y 1 , z 1 , …, x k+1 , y k+1 , z k+1

[0253]

[0254] X min < X < X max

[0255] Among them, X represents the position vector of the particle, representing the set of tunnel axis points; x 1 < x 2 < … < x k+1 means that the x coordinates of the particles increase sequentially in the tunnel direction; X min < X < X max represents the position constraint range of the particle.

[0256] Step S23: After generating the axis points, use the polyline fitting technology to construct the complete tunnel axis curve for subsequent semantic segmentation and reverse modeling.

[0257] The process of the leakage defect quantification module 3 performing semantic segmentation on the tunnel point cloud data includes:

[0258] Step S31: Analyze the local feature changes of the point cloud through a sliding window to extract the geometric features of the tunnel components;

[0259]

[0260] Among them, μ i represents the mean value of the local feature changes of point P i , used to describe the geometric features of the tunnel components; |P i | represents the number of points in the local neighborhood of P i ; P i represents the neighborhood set of point i in the point cloud; p​j Points belonging to point P i in the neighborhood; β θ,j represents point p j The eigenvalue (such as curvature or gradient) in a specific direction θ.

[0261] Step S32: Detect tunnel defects (such as leakage) based on the intensity value and geometric feature changes, and use the AlphaShape algorithm to extract the contour and quantify the area of the defect region;

[0262]

[0263] where A represents the area of the defect region; (x k , y k ) represents the coordinates of the k-th point on the Alpha Shape boundary; n represents the total number of boundary points.

[0264] Step S33: Classify the point cloud using multi-dimensional semantic information and label the component and defect positions to provide semantic support for subsequent reverse modeling.

[0265] The tunnel reverse modeling process of the three-dimensional model reconstruction module 4 includes the following steps:

[0266] Step S41: In the ring-level component reconstruction, use Poisson surface reconstruction, spherical projection, and Delaunay triangulation methods to generate the geometric models of components such as tunnel lining blocks;

[0267] Step S42: Enhance the surface thickness of the generated components;

[0268] Step S43: According to the spatial pose and thickness enhancement algorithm of the components, construct the overall three-dimensional model of the tunnel and optimize the geometric details;

[0269]

[0270] where T represents the spatial transformation matrix of the geometric model; s x , s y , s z represents the scaling factor; R ij represents the elements of the rotation matrix; t x , t y , t z represents the elements of the translation vector.

[0271] Figure 3 is a schematic diagram of the bottom-up tunnel geometric twin model construction process, showing the modeling steps from ring-level components to the overall tunnel model. Figure 4 is a schematic diagram (partial) of the final tunnel geometric digital twin model, showing the multi-dimensional semantic information and geometric detail optimization results of the model.Figure 5 It is an exploded view (partial) of the final tunnel geometric digital twin model, showing the different component compositions of the model.

[0272] To verify the implementation effect and technical advantages of the present invention, the following will be described in combination with specific experiments. In this embodiment, a section of the Shanghai Metro tunnel in China with significant curvature is selected as the research object, and the effects of tunnel axis fitting and scene segmentation are systematically evaluated. At the same time, compared with other commonly used methods, the accuracy and robustness of the present invention are verified.

[0273] In terms of experimental data and computing environment, the total length of the tunnel section tested this time is about 1 kilometer, and the total amount of original point cloud data is 20.5GB. The equipment used in the experiment is a 13th generation Intel(R) Core(TM) i5-13500H CPU (2.60GHz), 16GB of RAM, and an NVIDIA GeForce RTX 4050 GPU (5921MB VRAM). Without enabling CUDA acceleration, the reconstruction of the tunnel geometric digital twin model takes about 2 hours. By enabling CUDA acceleration to improve the efficiency of steps such as the particle swarm optimization (PSO) algorithm and local feature calculation, the model reconstruction time is significantly shortened to 40 minutes. The final generated tunnel three-dimensional model file size is about 500MB and is displayed through a visualization platform based on Autodesk Platform Services (APS). The platform not only supports the overall visualization of the tunnel geometric structure but also can highlight defect information such as leakage areas, providing an interactive inspection tool for operation and maintenance personnel. Figures 3 to 5 It shows the subway tunnel geometric digital twin model generated by the method of the present invention, including the overall view, component decomposition view, and local close-up.

[0274] In the evaluation of tunnel axis fitting, the polyline fitting method based on PSO proposed by the present invention was compared with the traditional least squares (LS) fitting, the RANSAC method, and the single-linear fitting variant of PSO. The evaluation index was the variance of the squared distance from the point cloud to the fitted axis. The experimental results showed that the PSO polyline fitting method performed the best, with a distance variance from points to the axis of 0.0243, much lower than other methods. In contrast, although the high-order polynomial fitting (such as cubic polynomial) had higher accuracy (variance of 0.0251), it lacked robustness in regions with sharp curvature changes. The RANSAC method and linear fitting were unstable when dealing with high-noise point cloud data, with variances of 0.5731 and 0.5850 respectively. Although the PSO single-line fitting was better than the former two, it was still difficult to adapt to the geometric features of tunnels with complex curvatures. In contrast, the PSO polyline fitting flexibly adapted to the curvature changes of the tunnel geometry by dividing the tunnel axis into multiple connected line segments, significantly improving the fitting accuracy and robustness, especially suitable for tunnel scenarios with complex geometry and high noise.

[0275] In the scene segmentation task, the method of the present invention selected the Seg2Tunnel dataset as the benchmark and selected the Tunnel-3 data for experimental evaluation, covering two types of tasks: semantic segmentation (Level III) and instance segmentation (Level IV). In semantic segmentation, the evaluation indexes were overall accuracy (OA) and mean intersection over union (mIoU). The experimental results showed that the overall accuracy of the method of the present invention was 0.811 and the mean intersection over union was 0.624, showing a performance close to that of the neural network-based method. However, the difference in mIoU was mainly due to different classification criteria. For example, the Seg2Tunnel dataset classified the circumferential joint as a non-lining member, while the present method classified it into the lining ring. This difference in criteria produced false positives and false negatives in the circumferential joint area. Nevertheless, the method of the present invention still performed well in overall performance.

[0276] In the instance segmentation task, the evaluation index was mean average precision (mAP), including evaluation criteria for multiple IoU thresholds. The results showed that the mAP of the method of the present invention was 0.300, significantly higher than that of Mask3D (0.019) and TD3D (0.270). Especially under strict IoU thresholds (such as mAP@50), the present method led far ahead of Mask3D with a score of 0.582 and TD3D with a score of 0.336, indicating that the present method had higher robustness and accuracy in complex point cloud scenarios. The improvement in this performance was mainly attributed to the in-depth analysis of geometric features and the robust segmentation strategy of the present invention, which could effectively process complex point cloud structures and achieve reliable instance discrimination and accurate component segmentation.

[0277] The experimental results show that the present invention has significant advantages in tunnel geometric twin modeling. It can not only efficiently adapt to complex geometries and large-scale point clouds, but also demonstrate excellent performance in semantic segmentation and instance segmentation tasks, providing a reliable and efficient solution for the construction of tunnel geometric digital twins. This further verifies the practical application value of the present invention in the digital management of the tunnel's entire life cycle.

[0278] The present invention also discloses an electronic device. Figure 6 It is a schematic diagram of the composition of the electronic device in an embodiment of the present invention; please refer to Figure 6 , at the hardware level, the electronic device includes a memory, a processor, and at least one network interface; the processor can be a microprocessor, and the memory can include a memory, such as a random access memory (RAM), and can also include a non-volatile memory, etc. Of course, the electronic device can also be provided with other hardware as needed.

[0279] The processor, network interface, and memory can be interconnected through an internal bus, which can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect Standard) bus, or an EISA (Extended Industry Standard Architecture) bus, etc.; the bus can include an address bus, a data bus, a control bus, etc. The memory is used to store programs (which can include an operating system program and application programs); the programs can include program codes, and the program codes can include computer operation instructions. The memory can include a memory and a non-volatile memory, and provide instructions and data to the processor.

[0280] In one embodiment, the processor can read the corresponding program from the non-volatile memory into the memory and then run; the processor can execute the programs stored in the memory and is specifically used to perform the following operations (as Figure 1 shown):

[0281] Step S1: Preprocess the tunnel point cloud data, including voxel downsampling, denoising, and surface separation of the tunnel point cloud data to ensure data quality and integrity;

[0282] Step S2: Extract the tunnel axis based on the particle swarm optimization algorithm, generate multiple segments of linearly fitted axes by fitting multiple axis points, and accurately describe complex tunnel curves;

[0283] Step S3: Perform multi-dimensional semantic segmentation on the tunnel point cloud data, extract tunnel components and defects through local feature change analysis, including lining blocks, cables, tracks, sleepers, and leakage defects, and quantify the leakage defects;

[0284] Step S4: Based on the bottom-up modeling strategy, perform reverse modeling on the extracted components, gradually construct the ring-level and tunnel-level models, and realize the reconstruction of the overall three-dimensional model of the tunnel;

[0285] Step S5: Enhance the geometric details of the reconstructed tunnel geometric digital twin model to ensure model accuracy and multi-dimensional semantic expression.

[0286] The present invention further discloses a storage medium, on which computer program instructions are stored. When the computer program instructions are executed by a processor, the following steps of the method of the present invention are implemented (as Figure 1 shown):

[0287] Step S1: Preprocess the tunnel point cloud data, including voxel downsampling, denoising, and surface separation of the tunnel point cloud data to ensure data quality and integrity;

[0288] Step S2: Extract the tunnel axis based on the particle swarm optimization algorithm, generate multiple linear fitting axes by fitting multiple axis points, and accurately describe the complex tunnel curve;

[0289] Step S3: Perform multi-dimensional semantic segmentation on the tunnel point cloud data, extract tunnel components and defects through local feature change analysis, including lining blocks, cables, tracks, sleepers, and leakage defects, and quantify the leakage defects;

[0290] Step S4: Based on the bottom-up modeling strategy, perform reverse modeling on the extracted components, gradually construct the ring-level and tunnel-level models, and realize the reconstruction of the overall three-dimensional model of the tunnel;

[0291] Step S5: Enhance the geometric details of the reconstructed tunnel geometric digital twin model to ensure model accuracy and multi-dimensional semantic expression.

[0292] In summary, the subway tunnel geometric digital twin reverse modeling method, system, electronic device, and storage medium proposed by the present invention can improve the automation degree, adaptability, and modeling efficiency of tunnel twin modeling.

[0293] (1) The present invention does not need to rely on a predefined component library, parametric design tool, or neural network, and realizes three-dimensional model reconstruction completely based on point cloud data, significantly improving the adaptability and flexibility of modeling;

[0294] (2) Through the particle swarm optimization algorithm and polyline fitting method, the problem of insufficient axis extraction accuracy of traditional methods under complex tunnel geometric forms is effectively solved;

[0295] (3) Using local feature change analysis to achieve semantic segmentation, combined with multi-dimensional semantic features to identify components and defects, improving the semantic expression ability of the model;

[0296] (4) Through a bottom-up reverse modeling strategy, the rapid construction of a fine geometric model from a rough geometric model is achieved, meeting the digital twin modeling requirements of different LoD levels;

[0297] (5) The full-process automated processing significantly reduces the modeling cost and improves the construction efficiency of the tunnel geometric digital twin model, providing technical support for the intelligent operation and maintenance of the tunnel.

[0298] It should be noted that the present application can be implemented in software and / or a combination of software and hardware; for example, it can be implemented using an application-specific integrated circuit (ASIC), a general-purpose computer, or any other similar hardware device. In some embodiments, the software program of the present application can be executed by a processor to implement the above steps or functions. Similarly, the software program of the present application (including related data structures) can be stored in a computer-readable recording medium; for example, a RAM memory, a magnetic or optical drive, or a floppy disk and similar devices. Additionally, some steps or functions of the present application can be implemented using hardware; for example, as a circuit that cooperates with a processor to execute each step or function.

[0299] The technical features of the above-described embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0300] The description and application of the present invention here are illustrative and do not intend to limit the scope of the present invention to the above embodiments. The effects or advantages involved in the embodiments may not be reflected in the embodiments due to various factors. The description of the effects or advantages is not used to limit the embodiments. The deformations and changes of the embodiments disclosed here are possible, and the substitutions and equivalent components of the embodiments are well known to those of ordinary skill in the art. Those skilled in the art should clearly understand that the present invention can be implemented in other forms, structures, arrangements, proportions, and with other components, materials, and parts without departing from the spirit or essential characteristics of the present invention. Other deformations and changes can be made to the embodiments disclosed here without departing from the scope and spirit of the present invention.

Claims

1. A method for reverse modeling of digital twin geometry of subway tunnels, characterized in that: The modeling method comprises: Step S1, preprocessing the tunnel point cloud data, including voxel downsampling, denoising and surface separation of the tunnel point cloud data; Step S2, extracting the tunnel axis based on the particle swarm optimization algorithm, generating multiple linear fitting axes by fitting multiple axis points, and accurately describing the complex tunnel curve; Step S3, performing multi-dimensional semantic segmentation on the tunnel point cloud data, extracting tunnel components and defects, including lining blocks, cables, tracks, sleepers and leakage defects, through local feature change analysis, and quantifying the leakage defects; Step S4: reverse modeling the extracted components based on a bottom-up modeling strategy, gradually constructing ring-level and tunnel-level models, and reconstructing the overall three-dimensional model of the tunnel; Step S5: Enhance the geometric details of the reconstructed tunnel geometry digital twin model to ensure model accuracy and multi-dimensional semantic expression.

2. The subway tunnel geometry digital twin reverse modeling method according to claim 1 is characterized by: In step S1, the preprocessing of the tunnel point cloud data includes: Step S11, using a voxel downsampling method to select the centroid point of each voxel unit to reduce the amount of data while retaining key geometric information; Step S12: using the Gaussian statistical filtering method, calculating the average distance between each point in the tunnel point cloud data and its neighboring points and removing outliers; Step S13, separating the plane and curved surface parts of the tunnel point cloud by using the planar RANSAC algorithm to provide support for subsequent tunnel axis extraction; In step S2, the step of extracting the tunnel axis based on the particle swarm optimization algorithm includes: Step S21, defining an objective function to minimize the distance variance from the tunnel point cloud data to the axis; Step S22, generating axis points by particle swarm optimization algorithm, wherein the axis points satisfy the sorting constraint and the spatial range constraint; Step S23, connecting the generated axis points into multiple lines to adapt to the complex tunnel curvature; In step S3, the semantic segmentation of the tunnel point cloud data includes: Step S311, analyzing the normal vector changes in the tunnel point cloud data, segmenting the lining blocks and identifying the component boundaries; Step S312: extract local features of cables and tracks using a sliding window, and separate different components based on geometric height changes; Step S313, detecting leakage defects by combining the intensity value and the local point density change, and calculating the defect area; Leakage defect detection includes: Step S321, performing threshold filtering of point intensity values ​​on the local area to extract potential defect points with low intensity; Step S322, using the Alpha Shape algorithm to calculate the defect boundary and estimate the defect area; The specific steps of reverse modeling in step S4 include: Step S41, generating a single ring geometric model at the ring level through Poisson reconstruction, spherical projection and Delaunay triangulation; Step S42, assembling the ring-level models into an overall tunnel model using a spatial pose matrix; The model detail optimization in step S5 includes: Step S51, optimizing the geometric features of the lining block model by normal adjustment and thickness enhancement; Step S52: Fusing the defect information with the geometric model to generate a semantically rich three-dimensional model.

3. The subway tunnel geometry digital twin reverse modeling method according to claim 1 is characterized by: In the step S1, the tunnel point cloud data is optimized by voxel downsampling and noise point filtering, and the surface is separated by the RANSAC algorithm, which specifically includes: Step S11, compressing the point cloud data using voxel downsampling technology to reduce the data volume and retain the geometric features of the tunnel structure; Among them, P v is the coordinate of the downsampled point, P i The coordinates of the points in the original point cloud that belong to the same voxel, N is the total number of points belonging to the voxel; Step S12: Use Gaussian filtering method to remove noise points in the point cloud to optimize data quality; Among them, d i It's point P i The local average distance is used to measure the noise; P i is the coordinate of the current point; P j is with P i The coordinates of the k nearest points; k is the number of points in the neighborhood; Step S13, using the RANSAC algorithm to separate the surface part, extract the main data of the tunnel lining, and remove unnecessary interference points; In step S2, the tunnel axis is extracted by using a particle swarm optimization algorithm, which specifically includes: Step S21, define the objective function, and calculate the minimum distance variance from the point cloud to the axis; d i =||p i -p’|| Among them, d i Indicates the minimum distance from the i-th point in the point cloud to the axis; represents the average distance from all points to the axis; N represents the total number of points in the point cloud; P i represents the coordinates of the i-th point in the point cloud; P s represents the coordinates of a point on the axis; V represents the direction vector of the axis; P′ represents point P i The projection point on the axis; Step S22, initializing the position and velocity of the particle swarm, and iteratively optimizing the position of the axis point; X=[x1,y1,z1,...,x k+1 y k+1 ,z k+1 ] x1 <x2<…<x k+1 X min <X<X max Among them, X represents the position vector of the particle, which represents the set of tunnel axis points; x1 <x2<…<x k+1 The x coordinates of the particles increase in the tunnel direction; min <X <X max Indicates the position constraint range of the particle; Step S23: After the axis points are generated, a complete tunnel axis curve is constructed using polyline fitting technology for subsequent semantic segmentation and reverse modeling.

4. The subway tunnel geometry digital twin reverse modeling method according to claim 1 is characterized by: In step S3, the process of semantically segmenting the tunnel point cloud data includes: Step S31, analyzing the local feature changes of the point cloud through a sliding window to extract the geometric features of the tunnel components; Among them, μ i Indicates point P i The mean value of the local characteristic variation is used to describe the geometric characteristics of the tunnel components; IP i | indicates P i The number of points in the local neighborhood; P i represents the neighborhood set of point i in the point cloud; p j Indicates that it belongs to point P i Neighborhood points; β θ，j Represents point p j The eigenvalue in a specific direction θ; Step S32: detecting tunnel defects (such as leakage) based on intensity values ​​and geometric feature changes, and performing contour extraction and area quantification on the defective area using the Alpha Shape algorithm; Where A represents the area of ​​the defect region; (x k ,y k ) represents the coordinates of the kth point on the boundary of the Alpha Shape; n represents the total number of boundary points; Step S33: using multi-dimensional semantic information to classify the point cloud and mark the components and defect locations, providing semantic support for subsequent reverse modeling; In step S4, the reverse modeling process of the tunnel includes the following steps: Step S41, in the reconstruction of the ring-level components, the geometric model of the components such as the tunnel lining blocks is generated by using Poisson surface reconstruction, spherical projection and Delaunay triangulation method; Step S42, enhancing the surface thickness of the generated component; Step S43: construct an overall three-dimensional model of the tunnel according to the spatial posture and thickness enhancement algorithm of the components, and optimize the geometric details; Where T represents the spatial transformation matrix of the geometric model; s x ,s y ,s z represents the scaling factor; R ij Represents the elements of the rotation matrix; t x ,t y ,t z Represents the elements of the translation vector.

5. A subway tunnel geometry digital twin reverse modeling system, characterized in that: The modeling system comprises: A data preprocessing module is used to preprocess the tunnel point cloud data, including voxel downsampling, denoising and surface separation of the tunnel point cloud data; Tunnel curve description module, which is used to extract the tunnel axis based on the particle swarm optimization algorithm, generate multiple linear fitting axes by fitting multiple axis points, and accurately describe the complex tunnel curve; The leakage defect quantification module is used to perform multi-dimensional semantic segmentation on tunnel point cloud data, extract tunnel components and defects through local feature change analysis, including lining blocks, cables, tracks, sleepers and leakage defects, and quantify leakage defects; The 3D model reconstruction module is used to reverse model the extracted components based on the bottom-up modeling strategy, gradually build the ring-level and tunnel-level models, and realize the reconstruction of the overall 3D model of the tunnel; The geometric detail enhancement module is used to enhance the geometric details of the reconstructed tunnel geometry digital twin model to ensure model accuracy and multi-dimensional semantic expression.

6. The subway tunnel geometry digital twin reverse modeling system according to claim 5 is characterized by: The data preprocessing module preprocesses the tunnel point cloud data including: Step S11, using a voxel downsampling method to select the centroid point of each voxel unit to reduce the amount of data while retaining key geometric information; Step S12: using the Gaussian statistical filtering method, calculating the average distance between each point in the tunnel point cloud data and its neighboring points and removing outliers; Step S13, separating the plane and curved surface parts of the tunnel point cloud by using the planar RANSAC algorithm to provide support for subsequent tunnel axis extraction; The process of extracting the tunnel axis by the tunnel curve description module based on the particle swarm optimization algorithm includes: Step S21, defining an objective function to minimize the distance variance from the tunnel point cloud data to the axis; Step S22, generating axis points by particle swarm optimization algorithm, wherein the axis points satisfy the sorting constraint and the spatial range constraint; Step S23, connecting the generated axis points into multiple lines to adapt to the complex tunnel curvature; The semantic segmentation process of the leakage defect quantification module for tunnel point cloud data includes: Step S311, analyzing the normal vector changes in the tunnel point cloud data, segmenting the lining blocks and identifying the component boundaries; Step S312: extract local features of cables and tracks using a sliding window, and separate different components based on geometric height changes; Step S313, detecting leakage defects by combining the intensity value and the local point density change, and calculating the defect area; The leakage defect quantification module detects leakage defects by: Step S321, performing threshold filtering of point intensity values ​​on the local area to extract potential defect points with low intensity; Step S322, using the Alpha Shape algorithm to calculate the defect boundary and estimate the defect area; The reverse modeling process of the three-dimensional model reconstruction module includes: Step S41, generating a single ring geometric model at the ring level through Poisson reconstruction, spherical projection and Delaunay triangulation; Step S42, assembling the ring-level models into an overall tunnel model using a spatial pose matrix; The model detail optimization of the geometric detail enhancement module includes: Step S51, optimizing the geometric features of the lining block model by normal adjustment and thickness enhancement; Step S52: Fusing the defect information with the geometric model to generate a semantically rich three-dimensional model.

7. The subway tunnel geometry digital twin reverse modeling system according to claim 6, characterized in that: The data preprocessing module optimizes the tunnel point cloud data by voxel downsampling and noise point filtering, and separates the surface by RANSAC algorithm, specifically including: Step S11, compressing the point cloud data using voxel downsampling technology to reduce the data volume and retain the geometric features of the tunnel structure; Among them, P v is the coordinate of the downsampled point, P i The coordinates of the points in the original point cloud that belong to the same voxel, N is the total number of points belonging to the voxel; Step S12: Use Gaussian filtering method to remove noise points in the point cloud to optimize data quality; Among them, d i It's point P i The local average distance is used to measure the noise; P i is the coordinate of the current point; P j is with P i The coordinates of the k nearest points; k is the number of points in the neighborhood; Step S13, using the RANSAC algorithm to separate the surface part, extract the main data of the tunnel lining, and remove unnecessary interference points; The tunnel curve description module extracts the tunnel axis by using a particle swarm optimization algorithm, specifically including: Step S21, define the objective function, and calculate the minimum distance variance from the point cloud to the axis; d i =||p i -p’|| Among them, d i Indicates the minimum distance from the i-th point in the point cloud to the axis; represents the average distance from all points to the axis; N represents the total number of points in the point cloud; P i represents the coordinates of the i-th point in the point cloud; P s represents the coordinates of a point on the axis; V represents the direction vector of the axis; P′ represents point P i The projection point on the axis; Step S22, initializing the position and velocity of the particle swarm, and iteratively optimizing the position of the axis point; X=[x1,y1,z1,...,x k+1 y k+1 ,z k+1 ] x1 <x2<…<x k+1 X min <X<X max Among them, X represents the position vector of the particle, which represents the set of tunnel axis points; x1 <x2<…<x k+1 The x coordinates of the particles increase in the tunnel direction; min <X <X max Indicates the position constraint range of the particle; Step S23: After the axis points are generated, a complete tunnel axis curve is constructed using polyline fitting technology for subsequent semantic segmentation and reverse modeling.

8. The subway tunnel geometry digital twin reverse modeling system according to claim 6, characterized in that: The process of the leakage defect quantification module performing semantic segmentation on the tunnel point cloud data includes: Step S31, analyzing the local feature changes of the point cloud through a sliding window to extract the geometric features of the tunnel components; Among them, μ i Indicates point P i The local characteristic variation mean is used to describe the geometric characteristics of tunnel components; |P i | indicates P i The number of points in the local neighborhood; P i represents the neighborhood set of point i in the point cloud; p j Indicates that it belongs to point P i Neighborhood points; β θ,j Represents point p j The eigenvalue in a specific direction θ; Step S32: detecting tunnel defects (such as leakage) based on intensity values ​​and geometric feature changes, and performing contour extraction and area quantification on the defective area using the Alpha Shape algorithm; Where A represents the area of ​​the defect region; (x k ,y k ) represents the coordinates of the kth point on the boundary of the Alpha Shape; n represents the total number of boundary points; Step S33: using multi-dimensional semantic information to classify the point cloud and mark the components and defect locations, providing semantic support for subsequent reverse modeling; The tunnel reverse modeling process of the three-dimensional model reconstruction module includes the following steps: Step S41, in the reconstruction of the ring-level components, the geometric model of the components such as the tunnel lining blocks is generated by using Poisson surface reconstruction, spherical projection and Delaunay triangulation method; Step S42, enhancing the surface thickness of the generated component; Step S43: construct an overall three-dimensional model of the tunnel according to the spatial posture and thickness enhancement algorithm of the components, and optimize the geometric details; Where T represents the spatial transformation matrix of the geometric model; s x ,s y ,s z represents the scaling factor; R ij Represents the elements of the rotation matrix; t x ,t y ,t z Represents the elements of the translation vector.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 4 are implemented.

10. A storage medium having computer program instructions stored thereon, characterized in that: When the computer program instructions are executed by a processor, the steps of the method according to any one of claims 1 to 4 are implemented.

Images