Existing damaged hollow slab girder bridge automatic finite element modeling and performance evaluation method based on visual data

Abstract

The invention discloses a visual data-based automatic finite element modeling and performance evaluation method for an existing damaged hollow slab girder bridge. The method comprises the steps of S1, acquiring complete three-dimensional point cloud data of a multi-beam-slab model; s2, based on the three-dimensional point cloud data, extracting key geometric features of the structure by adopting a structure three-dimensional size automatic measurement method fusing plane projection and point cloud distortion correction; s3, for the damaged area of the multi-beam-slab structure, identifying the boundary of the damaged area and the non-damaged area of the multi-beam-slab structure through a point cloud neighborhood point height difference comparison and clustering analysis method, and carrying out detailed spatial position analysis on the damage; and S4, a damage unit set is automatically retrieved in combination with damage positioning quantitative information, and automatic updating of hinge joint damage of the multi-beam-slab structure is achieved through a life-death unit method failure damage area finite element grid. According to the method, automatic finite element modeling and evaluation of the existing hinge joint damage multi-beam-slab structure are achieved, and an efficient and accurate technical means is provided for health monitoring and performance evaluation of a bridge structure.

Description

Technical Field

[0001] The present invention relates to the technical field of civil engineering structure detection and evaluation, and particularly relates to a method for automatically performing finite element modeling and evaluation on an existing multi-girder slab structure with damaged hinge joints by using visual data. Background Art

[0002] With the increase of service life, concrete bridge structures, especially multi-girder slab structures widely used in medium and small span bridges, are facing increasingly serious disease problems. Among them, damaged hinge joints, as a typical disease form, not only affect the aesthetics of the bridge, but more importantly, will significantly reduce the bearing capacity of the structure, thus directly threatening the safety of the bridge. Therefore, it is particularly important to effectively evaluate multi-girder slab structures, especially structures with damaged hinge joints. However, the existing methods still have the following problems:

[0003] (1) Most existing bridge modeling methods mainly rely on drawing materials and construct the geometric model of the structure by manual drawing or modeling with CAD software. Although this method can reflect the basic shape of the structure to a certain extent, it is difficult to accurately reflect the true three-dimensional information of the structure, especially the detailed damage in complex structures. For example, drawing materials often cannot record the damage on the surface of the structure in detail, resulting in deviations between the established finite element model and the real structure in details, and further affecting the accuracy of the evaluation results.

[0004] (2) Although the damage detection methods based on two-dimensional image data can quickly detect damage, it is difficult to provide accurate three-dimensional damage spatial information. These methods usually use technologies such as convolutional neural networks and image segmentation to process the surface images of the structure and identify the damaged areas. However, due to the limitations of image data itself, such as single perspective and low information dimension, these methods are difficult to accurately describe the three-dimensional shape and spatial position of the damage, restricting their application in finite element model updating.

[0005] (3) For the detection data of existing damage to the structure, the existing methods often make insufficient use of it, and it is difficult to effectively utilize the damage spatial information and quickly update the finite element model. Traditional damage detection methods, such as visual inspection and non-destructive testing, although they can obtain a certain degree of damage spatial information, this information often exists in the form of text, images, etc., and it is difficult to directly integrate it into the finite element model, resulting in a cumbersome and inefficient model updating process.

[0006] In summary, the existing research methods have many deficiencies in bridge structure modeling and evaluation, and it is difficult to meet the requirements of modern bridge engineering for efficient and accurate evaluation. Summary of the Invention

[0007] The object of the present invention is to overcome the defects existing in the above-mentioned prior art and provide a method for finite element modeling and evaluation of multi-girder slab structures with damaged hinge joints, aiming to achieve high-precision and high-efficiency finite element modeling and safety evaluation of existing damaged bridge structures.

[0008] An automatic finite element modeling and performance evaluation method for existing damaged hollow slab beam bridges based on visual data, comprising the following steps:

[0009] S1. Based on three-dimensional laser scanning technology, comprehensively scan the multi-girder slab structure to obtain complete three-dimensional point cloud data of the multi-girder slab model.

[0010] S2. Based on the obtained three-dimensional point cloud data, adopt a structural three-dimensional dimension automatic measurement method that combines plane projection and point cloud distortion correction. By extracting the key geometric features of the structure, provide three-dimensional dimension data of the multi-girder slab structure with damaged hinge joints for subsequent finite element modeling.

[0011] S3. For the damaged area of the multi-girder slab structure, identify the boundary between the damaged area and the undamaged area of the multi-girder slab structure and conduct a detailed spatial position analysis of the damage through the method of comparing the height differences of adjacent points in the point cloud and clustering analysis; use the voxelization-binarization method to convert the complete three-dimensional point cloud data of the multi-girder slab model into a three-dimensional binary matrix, complete the accurate extraction and complete retention of the damage spatial information of the multi-girder slab structure, and provide damage location quantification information for automatic finite element modeling.

[0012] S4. Combine the above damage location quantification information to automatically retrieve the damaged element set, establish an accurate mapping between voxels and finite element mesh elements, invalidate the finite element mesh of the damaged area through the birth and death element method, set the elements corresponding to the damaged area as "dead" elements to simulate the damage in the actual structure; at the same time, keep the elements in the undamaged area as "live" elements to achieve automatic adjustment of the finite element model of the multi-girder slab structure with damaged hinge joints.

[0013] Beneficial effects: The automatic finite element modeling and performance evaluation method for existing damaged hollow slab beam bridges based on visual data of the present invention, based on the collected three-dimensional point cloud data, corrects the point cloud distortion through the rotation matrix and projects it onto the fitting plane to generate a high-precision height map, realizing the accurate acquisition of the three-dimensional dimension information of the multi-girder slab structure with damaged hinge joints; based on the comparison of the height differences of adjacent points in the three-dimensional point cloud data and clustering analysis and voxelization technology, constructs a three-dimensional binary matrix of the damage space, realizes the effective extraction of the damage spatial information of the structure, and provides accurate damage location and quantification information for the damage update of the finite element model; based on the above damage location information, automatically retrieves the damaged element set, and invalidates the finite element mesh of the damaged area through the birth and death element method, realizing the automatic update of the hinge joint damage of the multi-girder slab structure.

[0014] As a further preferred embodiment of the present invention, in step S2, the steps of extracting the key geometric features of the structure by using the method of fusing planar projection and point cloud distortion correction specifically include:

[0015] S201, perform RANSAC segmentation on the three-dimensional point cloud model to obtain six faces of the complete multi-beam slab model;

[0016] S202, use the rotation matrix R to perform a spatial transformation on the three-dimensional point cloud data, and rotate the front or left side of the multi-beam slab model to a coplanar position with the upper surface respectively to ensure the accuracy and consistency of the three-dimensional point cloud data. Let the rotation matrix be R. For the case of 90 degrees around the x-axis and y-axis, the rotation matrix is expressed as:

[0017]

[0018] S203, use the PCA algorithm to perform plane fitting on the segmented three-dimensional point cloud data to obtain the approximate plane equation where the three-dimensional point cloud data is located;

[0019] S204, by calculating the perpendicular projection points of each three-dimensional point cloud data to the fitting plane, project all the three-dimensional point cloud data onto the same plane to generate a height map with high precision; Let the rotated point cloud point set be P = [p1, p2,..., p n , where p i = [x i , y i , z i T represents the three-dimensional coordinates of the i-th point. By calculating the minimum and maximum values of the x and y coordinates, set the boundary range of the height map, that is:

[0020] x min = min(x i ), x max = max(x i )

[0021] y min = min(y i ), y max = max(y i )

[0022] S205, normalize the height values in the height map and map them to the gray value range of 0 - 255, so that the height information is presented in the form of gray levels to obtain the corresponding gray image. The pixel value of the gray image at the position (x i , y i ):

[0023] ​

[0024] Among them,

[0025] H(x i , y i ) = max(z i |(x i , y i , z i ))

[0026] represents the height value of a certain point on the height map, and z represents the Z coordinate of the three-dimensional point cloud data point;

[0027] S206. Through an image processing algorithm, edge detection and contour extraction are performed on the grayscale image to identify the key geometric features of the structure, such as the beam-slab boundary and structural holes, providing three-dimensional dimension data for subsequent finite element modeling.

[0028] Beneficial effects: The present invention adopts the technology of fusing planar projection and point cloud distortion correction, effectively solving the problem of image deformation that may occur when directly generating an image from three-dimensional point cloud data. The RANSAC segmentation is used to accurately separate the six faces of the multi-beam-slab model, and the rotation matrix space transformation is combined to ensure the consistency of the point cloud data. The high-precision and distortion-free height map is generated through PCA plane fitting and vertical projection, and after normalization processing, it is converted into a grayscale image, providing a standardized data basis for subsequent feature recognition.

[0029] As a further preferred solution of the present invention, in step S206, the steps of using an image processing algorithm to calculate the beam-slab boundary specifically include:

[0030] S2061. Apply Gaussian blur processing to the original image, and smooth the image by setting an appropriate standard deviation. The kernel function of Gaussian blur is expressed as:

[0031]

[0032] where σ is the standard deviation of the Gaussian distribution;

[0033] S2062. Use the Canny edge detection algorithm to process the smoothed image. By setting high and low thresholds, extract the contour edges of the target area, and at the same time suppress the response of non-edge points to generate a continuous and complete edge image;

[0034] S2063. Use the Hough gradient method to detect straight lines in the image, identify the straight line segments in the image, traverse all the detected straight lines, extract their endpoint coordinates, and use the minimum area rectangle fitting algorithm to approximate the target rectangle frame; finally, obtain the rectangle frame and the coordinates of its four vertices on the height image;

[0035] S2064. Calculate the distance between two points according to the Euclidean distance formula to provide data support for subsequent dimension measurement. In a two-dimensional plane, for two points P1(x1, y1) and P2(x2, y2), the Euclidean distance d between them pixel is defined as:

[0036]

[0037] S2065. Align the pixel data obtained based on the image to the real boundary data in the height map. During the image processing process, two-dimensional coordinates are usually converted into pixel coordinates to achieve the alignment of the image size:

[0038]

[0039] d real = k·d pixel

[0040] In the formula, P d is the pixel height of the height map, P w is the pixel width of the height map, B y is the coordinate system boundary height of the height map, B[[ID=Z7]] x is the coordinate system boundary width of the height map, k is the conversion alignment scale factor, and d real is the real dimension parameter.

[0041] Beneficial effects: The present invention significantly improves the calculation accuracy and efficiency of the beam-slab boundary dimension parameters by optimizing the image processing process, and provides more reliable dimension data for the construction of the finite element model. The Gaussian blur processing effectively reduces the interference of noise and small miscellaneous points in the image on the edge detection result; in view of the limited accuracy and stability of the traditional corner detection algorithms (Harris corner detection algorithm and Shi-Tomasi corner detection algorithm) in the face of complex situations such as image rotation, and the lack of scale invariance and the ability to accurately predict the number of output corners, the present invention combines the Canny edge detection to accurately extract the target contour, and cooperates with the Hough gradient line detection and the minimum rectangle fitting algorithm to achieve the robust recognition of the beam-slab boundary features; through the Euclidean distance calculation, the pixels are aligned with the real physical coordinates of the height map to complete the three-dimensional dimension measurement, thus forming a complete closed loop from image processing to real space parameter conversion.

[0042] As a further preferred solution of the present invention, in step S206, introducing the Hough circle transform to detect the circular holes in the image to obtain the key dimensions of the holes and locate the holes when calculating the parameters of the structural holes through the image processing algorithm.

[0043] Beneficial effects: The present invention uses the Hough circle transform algorithm to detect circular holes in an image. This algorithm can effectively identify circular features in the image and return the center coordinates and diameter of the circle. At the same time, the distances between adjacent holes and the distance to the edge contour line are calculated using the center coordinates for hole positioning. Finally, the key size and positioning data of the holes are obtained.

[0044] As a further preferred solution of the present invention, in step S3, the step of identifying the boundary between the damaged area and the undamaged area by the method of comparing the height differences of neighboring points in the point cloud and clustering analysis specifically includes:

[0045] S301, use the KD-tree spatial indexing technique to find the set of neighboring points N(P) within a certain radius r around it and calculate the average height of the neighboring points

[0046] N(P) = KDTree.query(P,r)

[0047]

[0048] Among them, KDTree.query(P,r) is a query function in the KD-tree spatial indexing technique. It accepts two parameters: point P and radius r. P represents the point to be queried, that is, the center point of the neighboring points we hope to find; r represents the query radius, which defines a spherical region centered on point P. All points falling within this region will be regarded as the neighboring points of point P; |N(P)| is the number of neighboring points;

[0049] S302, perform point cloud neighboring height difference comparison analysis, compare the height z of the current point i with the average height of the neighboring points If the height z of the current point i is significantly lower than the average height of the neighboring points and exceeds the preset threshold Δz, that is then mark this point as a potential damage point;

[0050] S303, use the DBSCAN clustering algorithm to group the potential damage points. Each cluster in the clustering result represents an independent damaged area, and the points inside it are regarded as damaged points, and finally independent damaged areas are identified.

[0051] Beneficial effects: Through the point cloud neighborhood point height difference comparison and clustering analysis technology, the present invention realizes the accurate identification of the structural damage area. The KD tree spatial indexing technology is used to efficiently extract the neighborhood point cloud, combined with the neighborhood average height comparison method to effectively quantify the local height anomaly, and the potential damage points are determined by the dynamic threshold, significantly improving the sensitivity and anti-interference ability of damage feature extraction. Further, the DBSCAN clustering algorithm is used to distinguish the outliers and the damage points that truly form the area, thereby filtering out false alarms and retaining the effective damage area. This method not only relies on the abnormal change of the height value to identify the damage, but also combines the neighborhood analysis and clustering analysis to enhance the robustness and accuracy of the identification.

[0052] As a further preferred solution of the present invention, in step S3, for the damage area of the multi-beam slab structure, the steps of spatial position analysis by the method of point cloud neighborhood point height difference comparison and clustering analysis specifically include:

[0053] S304, calculate the average value of the three-dimensional coordinates of all points in the identified damage area to obtain the coordinates of the center position of the damage area; assume that there are n points in the damage area, and the three-dimensional coordinates of each point are (x i , y i , z i ), where i = 1, 2,..., n. Then the coordinates of the center position of the damage area are calculated by the following formula:

[0054] S305, based on the obtained coordinates of the center position of the damage area, design a cuboid space area centered on the damage area . The depth of the cuboid, that is, the height direction, is set to d, which is determined according to the depth feature of the damage area and the safety margin Δz set to ensure that the complete damage space information is included. The length (set as l) and width (set as w) of the cuboid are then determined according to the expansion range of the damage area and the additional boundaries (set as Δx and Δy respectively) set to ensure that the complete damage space information is included. Specifically, it is obtained by adding the corresponding additional boundaries (set as Δx and Δy respectively) or the safety margin Δz to the difference between the maximum and minimum values of all points in the damage area in the x-axis, y-axis, and z-axis directions:

[0055]

[0056] S306, extract all the three-dimensional point cloud data in this cuboid space area to form a point cloud subset containing the damage and its surrounding environment.

[0057] Beneficial effects: By conducting a detailed spatial location analysis of the damage, the present invention locates the damage position and accurately extracts the area where the damage is located, thereby significantly reducing the amount of point cloud data to be faced in subsequent processing. By calculating the central coordinates of the damage area, a precise spatial reference is established. Combining with the damage depth, a cuboid spatial area is constructed to ensure complete coverage of the damage and its surrounding environment, and additional boundaries and safety margins are set to effectively adapt to different morphological damage characteristics and avoid information loss. This method not only accurately extracts the damage area but also retains its position information relative to the entire structure (such as the upper surface of the beam and slab), providing an important basis for subsequent damage quantification and analysis.

[0058] As a further preferred solution of the present invention, in step S3, the step of converting the three-dimensional point cloud data into a three-dimensional binary matrix by using the voxelization-binarization method specifically includes:

[0059] S307, rotating the three-dimensional point cloud data of the damage area after detection and segmentation to the XOY plane through the Rodrigues rotation formula. The Rodrigues rotation formula is:

[0060] v rot = vcosθ+(k × v)sinθ + k(k·v)(1 - cosθ)

[0061] where v is the original vector, k is the unit vector of the rotation axis, and θ is the rotation angle;

[0062] S308, performing voxelization processing on the rotated three-dimensional point cloud data to generate a regular three-dimensional grid, that is, voxels. Each voxel represents a fixed volume unit in space;

[0063] S309, constructing a three-dimensional binary matrix according to the voxelized data to realize the quantification of the damage spatial information.

[0064] Beneficial effects: The present invention realizes the three-dimensional spatial representation and precise quantification of the damage spatial information through the voxelization-binarization technology. The three-dimensional point cloud is accurately aligned to the XOY plane by using the Rodrigues rotation formula to eliminate the interference of spatial postures. The voxelization processing converts the discrete point cloud into a regular grid structure, realizes data standardization through the division of fixed volume units, and combines with the construction of a three-dimensional binary matrix to complete the digital spatial mapping of the damage area. This method not only retains the detailed features of the damage but also eliminates the interference of redundant data.

[0065] As a further preferred solution of the present invention, in step S308, the steps of voxelization processing specifically include:

[0066] S3081. For the separated cuboid cross-section, perform complete voxelization using the same grid resolution as in the previous stage, with a resolution of 3 mm, to suit downstream tasks for damage assessment. The voxelized cuboid contains three types of voxels: (1) empty voxels in the damage space, (2) voxels occupied by the solid surface and the damage surface, and (3) voxels not observed due to occlusion.

[0067] S3082. Define the mathematical representation of voxelization using the inclusion–exclusion principle. Let the center coordinates of voxel v j be c j = [c x , c y , c z T . Then, the calculation formula for the number of point clouds in each voxel is:

[0068]

[0069] where N j is the number of point clouds in the j-th voxel, N is the total number of points in the point cloud, ||·||2 represents the Euclidean distance, is an indicator function. When the distance from p i to c j does not exceed , its value is 1; otherwise, it is 0. If N j exceeds the preset threshold, then voxel v j is regarded as an occupied state; otherwise, it is regarded as an unoccupied state. After assigning values to the voxels and marking the unoccupied state, i.e., damaged voxels or unobserved voxels, as 1, and the occupied-state voxels, i.e., undamaged voxels, as 0, each layer (or slice) of the voxelized cuboid along the depth direction serves as a binary image.

[0070] S3083. Update the empty voxels, occupied voxels, and unobserved voxels in the subsequent layers along the depth direction through morphological operations.

[0071] Beneficial effects: Through the refined voxelization processing technology, the present invention realizes the structured analysis and three-dimensional spatial representation of damage space information. Complete voxelization is performed with a resolution of 3 mm to accurately distinguish among three types of states: empty voxels, occupied voxels, and unobserved voxels, providing a refined modeling basis for complex damage morphologies. Based on the inclusion–exclusion principle, the occupancy state of voxels is determined through the point cloud density threshold to generate a binary image, effectively eliminating noise interference and enhancing data consistency. Combining morphological operations to optimize the voxel state update in the depth direction further improves the continuity of spatial information and ensures the complete retention of the damaged area.

[0072] ​As a further preferred solution of the present invention, in step S309, the specific steps of constructing a three-dimensional binary matrix based on the voxelized data include:

[0073] S3091, obtain the boundary information of the voxel grid, and then calculate the size of the grid;

[0074] S3092, initialize a three-dimensional array all with 1 as a blank binary matrix, where 1 represents damaged voxels, including unobserved voxels; 0 represents undamaged voxels;

[0075] S303, traverse each voxel in the voxel grid, and according to its grid index, set the value to 0 at the corresponding position in the binary matrix to mark the positions occupied by the point cloud.

[0076] Beneficial effects: The present invention constructs a three-dimensional binary matrix reflecting the spatial distribution of the point cloud, realizing the digital efficient expression and accurate quantification of the damage spatial information. Based on the voxelized grid boundary information, the matrix dimension is accurately calculated to ensure the integrity of the data space mapping. The all-1 matrix initialization strategy is adopted to default mark potential damage and unobserved areas and dynamically update the occupancy status. This method compresses complex three-dimensional damage features into a three-dimensional binary matrix, which not only retains the damage spatial information but also eliminates redundant information, significantly improving the data storage and calculation efficiency.

[0077] As a further preferred solution of the present invention, in step S4, the specific steps of establishing an accurate mapping between voxels and finite element mesh elements include:

[0078] S401, establish an FE model with voxelized equivalent resolution;

[0079] S402, initialize an empty set of damaged elements to store the finite element elements determined to be damaged subsequently; according to the cuboid space region obtained in step 304, as the divided damage space region, locate this damage space region in the finite element model and determine the finite element mesh elements it contains;

[0080] S403, utilize the voxelization result obtained in step 3, where the voxel value of 1 represents the damaged area and the voxel value of 0 represents the non-damaged area; for each element in the finite element model, if the voxel value corresponding to the element is 1, add the element to the set of damaged elements; through the above mapping process, generate a list containing all damaged elements for accurate simulation of the damage situation in the actual structure based on the birth and death element method subsequently.

[0081] Beneficial effects: By establishing an accurate mapping between voxels and finite element mesh elements, the present invention realizes the adjustment of the damaged area of the finite element model. A finite element model with the same equivalent resolution as the voxels is established to ensure the fine restoration of damage characteristics in numerical analysis and significantly improve the calculation accuracy. Through damage spatial positioning and voxel state discrimination, damaged elements are automatically marked and a damaged element list is generated. Combining with the birth and death element method, the damage situation in the actual structure is accurately simulated.

[0082] Compared with the prior art, the present invention has the following beneficial effects:

[0083] (1) Achieving high-precision automatic measurement of the three-dimensional dimensions of a structure based on point cloud vision data: Aiming at the problem of improving the accuracy of dimension measurement based on three-dimensional point cloud data and the difficulty of obtaining accurate three-dimensional information, the present invention proposes an automatic measurement method for the three-dimensional dimensions of a structure that combines plane projection and point cloud distortion correction. The point cloud is transformed by a rotation matrix and projected onto a fitted plane to generate a height map. Then, using the Canny edge detection and Hough gradient methods, key features such as edge contours and holes in the image are accurately identified, thereby accurately calculating the three-dimensional dimension parameters of the structure and providing more reliable basic data for the construction of the finite element model.

[0084] (2) Achieving effective positioning and quantification of existing damage based on "voxelization - binarization": Aiming at the problem of difficult extraction and quantification of the spatial information of complex-shaped hinge joint damage, the present invention proposes a method for automatically extracting the spatial information of existing hinge joint damage assisted by vision. This method combines neighborhood point height comparison and clustering analysis techniques to effectively identify and locate the damaged area in the three-dimensional point cloud data. Using the "voxelization - binarization" method to generate a three-dimensional binary matrix of the damage space, the accurate quantification and three-dimensional spatial description of the damage space information are realized.

[0085] (3) Establishing a twin model for dynamic adjustment of existing damage to achieve bearing capacity analysis: Realizing automatic adjustment of the damaged area of the finite element model based on vision data: Aiming at the problem that it is difficult to effectively integrate the spatial information of existing damage with the finite element model, a method for adjusting the damaged area of automatic finite element modeling is proposed. Using the obtained three-dimensional dimension parameters and damage space information, an accurate mapping between voxels and finite element mesh elements is established. By using the birth and death element method to invalidate the finite element mesh in the damaged area, the rapid update of the finite element model is realized, improving the efficiency and accuracy of model update and providing strong support for the safety assessment of bridge structures. Description of the Drawings

[0086] Figure 1 Schematic diagram of the automatic finite element modeling and performance evaluation method for existing damaged hollow slab girder bridges based on vision data proposed by the present invention;

[0087] Figure 2Height map generation by point cloud projection for the multi - plate beam structure proposed in the present invention;

[0088] Figure 3 Schematic diagram for extracting three - dimensional dimension parameters of the structure proposed in the present invention;

[0089] Figure 4 Flow chart for obtaining spatial information of point cloud damage proposed in the present invention;

[0090] Figure 5 Principle diagram of the point cloud damage recognition method based on the comparison of height differences and clustering analysis of point cloud neighborhood points proposed in the present invention;

[0091] Figure 6 Schematic diagram of damage location in point cloud processing proposed in the present invention;

[0092] Figure 7 Schematic diagram of separating damaged cuboid slices and spatial voxelization proposed in the present invention;

[0093] Figure 8 Schematic diagram of the existing damage birth - and - death element method proposed in the present invention;

[0094] Figure 9 Schematic diagram of finite element modeling in the intact state: where (a) is the overall finite element model; (b) is the internal steel bar information;

[0095] Figure 10 Test device diagram proposed in the example;

[0096] Figure 11 Sensor layout diagram of the test multi - plate beam structure proposed in the example, where (a) is the sensor layout diagram on the upper part of the plate structure, and (b) is the sensor layout diagram on the lower part of the plate structure;

[0097] Figure 12 Result diagram of the existing damage modeling method proposed in the example, where (a) is the manual modeling combined with the element deletion method, and (b) is the method (birth - and - death element) proposed in the present invention;

[0098] Figure 13 Schematic diagram of the stress simulation results under the finite element modeling method proposed in the example, where (a) is working condition one, (b) is working condition two; (c) is working condition three; (d) is working condition four;

[0099] Figure 14 Comparison diagram of the transverse distribution coefficient of the results proposed in the example, where (a) is working condition one, (b) is working condition two; (c) is working condition three; (d) is working condition four;

[0100] Figure 15 Comparison diagram of the bearing capacity curves of the results proposed in the example, where (a) is the bearing capacity of the existing damaged structure, and (b) is the comparison between the intact structure and the damaged structure. Detailed implementation method

[0101] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0102] Some definitions related to this application are as follows:

[0103] Multi-girder slab structure: A plate-like structure formed by connecting multiple parallel beams through transverse connectors (such as hinge joints), which is widely used in medium and small span bridges.

[0104] Visual data: Three-dimensional point cloud data obtained through devices such as laser scanners, which reflects information such as the shape, size, and surface texture of an object.

[0105] Automatic finite element modeling: Automatically constructing a finite element model of a structure using a computer program, including the definition of nodes, elements, material properties, etc. and mesh generation.

[0106] In view of the problems existing in the aforementioned prior art, this embodiment provides a method for automatic finite element modeling and performance evaluation of existing damaged hollow slab beam bridges based on visual data to achieve automatic construction of a high-fidelity finite element model of the damaged structure and safety assessment. First, aiming at the problem that the traditional height map measurement of structural dimensions is inaccurate, a method for automatically measuring the three-dimensional dimensions of a structure by fusing plane projection and point cloud distortion correction is proposed. Then, aiming at the problem that it is difficult to effectively extract the spatial information of structural hinge joint damage based on three-dimensional point cloud data, a damage identification and spatial positioning method by fusing the height difference comparison of neighboring points in the point cloud and cluster analysis is proposed, and the damage quantification information is obtained by combining the "voxelization - binarization" method. Finally, aiming at the problem that it is difficult to effectively fuse damage detection information with the finite element model, an automatic finite element damage area adjustment method based on visual data is proposed. This method automatically establishes a finite element model using three-dimensional point cloud data and structural drawing information, and realizes automatic update of hinge joint damage by automatically adjusting the finite element mesh corresponding to the damaged part.

[0107] See Figure 1, This method mainly includes three parts: the three-dimensional information extraction method of structural damage based on three-dimensional point cloud data, the damage finite element adjustment method based on the birth and death element method, and the safety performance evaluation of hollow beam-slab structures. The method for extracting the three-dimensional dimension information of the structure based on three-dimensional point cloud data corresponds to steps S1 - S4, the method for extracting the three-dimensional information of structural damage based on three-dimensional point cloud data corresponds to steps S5 - S6, and the damage finite element adjustment and safety performance evaluation of hollow beam-slab structures based on the birth and death element method correspond to steps S7 - S8. The automatic finite element modeling and performance evaluation method for existing damaged hollow slab beam bridges based on visual data includes the following steps:

[0108] Step S1, Use a standing lidar to perform an omnidirectional scan of the existing multi-girder structure with damaged hinge joints. Set up stations on the four sides of the model to obtain three-dimensional point cloud data of the sides; at the same time, select appropriate positions to scan the upper bottom surface. To capture the information of the lower surface, place the lidar scanner at the bottom of the model for scanning. Based on the three-dimensional point cloud data collected from these station layouts, remove noise, background, and supports, and only retain the structural component area; through multi-station stitching of the point cloud, obtain a complete point cloud model of the structure. Perform voxel downsampling on the registered three-dimensional point cloud data to reduce the point cloud density, and perform RANSAC segmentation to obtain the six faces of the complete multi-girder model.

[0109] Step S2, Use the rotation matrix R to perform a spatial transformation on the point cloud model, rotate the front or left side of the multi-girder model to a coplanar position with the upper surface respectively to ensure the accuracy and consistency of the three-dimensional point cloud data. Let the rotation matrix be R. For the case of 90 degrees around the x-axis and y-axis, the rotation matrix Can be expressed as:

[0110]

[0111] Step S3, Adopt a method that combines plane projection and point cloud distortion correction to directly generate a high-precision and distortion-free height image from the three-dimensional point cloud data. This step includes the following sub-steps:

[0112] Step S31, As Figure 2 Shown, for the distortion problem during height map generation, perform plane fitting on the three-dimensional point cloud data through the PCA algorithm. By calculating the perpendicular projection points of each three-dimensional point cloud data to the fitting plane, project all three-dimensional point cloud data onto the same plane to generate a high-precision height map. Let the rotated point cloud point set be P = [p1, p2,..., p n , where p i = [x i , y i , z i T ​Denote the three-dimensional coordinates of the \(i\)-th point. By calculating the minimum and maximum values of the \(x\) and \(y\) coordinates, set the boundary range of the height map, that is:

[0113] x min =\(\min(x\) i ), x\) max =\(\max(x\) i )

[0114] y min =\(\min(y\) i ), y\) max =\(\max(y\) i )

[0115] Step S32: Calculate the image size according to the preset resolutions \(\Delta x\) and \(\Delta y\). During the process of projecting the point cloud onto the height image, traverse all the three-dimensional point cloud data, calculate the row and column positions of each point in the height image according to its \(x\) and \(y\) coordinates, and assign the \(z\) coordinate as the height value to the corresponding pixel point. That is:

[0116]

[0117]

[0118] Step S33: For the case where multiple points are mapped to the same pixel point, adopt the method of comparing the \(z\) values and only keep the point with the highest height to ensure the accuracy of the height map. The generation formula of the height map is as follows:

[0119] H(x\) i , y\) i ) = \(\max(z\) i |(x\) i , y\) i , z\) i ))

[0120] where \(H(x\) i , y\) i ) represents the height value of a certain point on the height map, and \(z\) represents the \(Z\) coordinate of the three-dimensional point cloud data point.

[0121] Step S34: Normalize the height values in the height map and map them to the grayscale value range of 0 - 255, so that the height information is presented in the form of grayscale levels to obtain the corresponding grayscale image. The pixel value of the grayscale image at position \((x\) i , y\) i ):

[0122]

[0123] Step S4, apply the Canny edge detection and Hough gradient methods to identify key features such as edge contours and holes in the grayscale image; use the minimum area rectangle fitting algorithm and the Hough circle transform technique to calculate size parameters, such as Figure 3 as shown. This step includes the following sub-steps:

[0124] Step S41, apply Gaussian blur processing to the original image. By setting appropriate blur radii and standard deviations, smooth the image to effectively reduce the interference of noise and small miscellaneous points in the image on the edge detection results. The kernel function of Gaussian blur can be expressed as:

[0125]

[0126] where σ is the standard deviation of the Gaussian distribution;

[0127] Step S42, use the Canny edge detection algorithm to process the smoothed image. By setting high and low thresholds, accurately extract the contour edges of the target area, while suppressing the response of non-edge points to generate a continuous and complete edge image;

[0128] Step S43, use the Hough gradient method to detect straight lines in the image, identify the straight line segments in the image, and return their endpoint coordinates. By traversing all the detected straight lines, extract their endpoints and use the minimum area rectangle fitting algorithm to approximate the target rectangle. This algorithm can find the minimum area rectangle that contains all the given points, and its direction can be arbitrary. Finally, obtain the rectangle and the coordinates of its four vertices on the height image;

[0129] Step S44, calculate the distance between two points according to the Euclidean distance formula, providing data support for subsequent size measurement. In a two-dimensional plane, for two points p1(x1,y1) and p2(x2,y2), the Euclidean distance d between them pixel is defined as:

[0130]

[0131] Step S45, align the pixel data obtained based on the image to the real boundary data in the height map. During the image processing process, two-dimensional coordinates are usually converted into pixel coordinates. Achieve the alignment of the image size:

[0132]

[0133] d real = k·d pixel

[0134] In the formula, P d is the pixel height of the height map, P w is the pixel width of the height map, By is the boundary height of the coordinate system of the height map, B x is the boundary width of the coordinate system of the height map, k is the proportionality coefficient for conversion alignment, d real is the true dimensional parameter.

[0135] In addition, the dimensional parameter results obtained by the method used in the present invention were compared with the existing Harris corner detection algorithm and Shi-Tomasi corner detection algorithm, as shown in Table 1:

[0136]

[0137] Table 1

[0138] Step S46 introduces the Hough circle transform to detect circular holes in the image, effectively identifying the circular features in the image and returning the center coordinates and diameter of the circle. At the same time, the distances between adjacent holes and the distances to the edge contour line were calculated using the center coordinates for hole positioning; let the center coordinates of two adjacent holes be (x c1 , y c1 ) and (x c2 , y c2 ), then the distance d holes between them is also calculated by the Euclidean distance formula:

[0139]

[0140] Step S5, as Figure 4 shown, for the damaged area of the multi-slab structure, the boundary between the damaged area and the undamaged area was identified through the comparison of the height differences of the neighboring points in the point cloud and clustering analysis, and a detailed spatial position analysis was carried out. This step includes the following sub-steps:

[0141] Step S51, for each point p i = [x i , y i , z i T in the point cloud, the KD-tree spatial indexing technique was used to quickly find the set of neighboring points N(P) within a certain radius r around it. This step aims to construct the local environment of each point for subsequent analysis. The formula is expressed as:

[0142] N(P) = KDTree.query(P, r)

[0143] ​Among them, KDTree.query(P, r) is a query function in the KD-tree spatial indexing technology. It accepts two parameters: the point P and the radius r. P represents the point to be queried, that is, the center point for which the surrounding neighborhood points are expected to be found. r represents the query radius. It defines a spherical region centered at the point P, and all points falling within this region will be regarded as the neighborhood points of point P.

[0144] Step S52, calculate the average height of these neighborhood points The formula is expressed as:

[0145]

[0146] Among them, |N(P)| is the number of neighborhood points. If the height z of the current point i is significantly lower than the average height of the neighborhood points and exceeds the preset threshold Δz, then mark this point as a potential damage point;

[0147] Step S53, use the DBSCAN clustering algorithm to group the potential damage points. Clustering analysis can distinguish outliers and truly formed regional damage points, thus filtering out false alarms and retaining effective damage regions. Each cluster in the clustering result represents an independent damage region, and the points inside it are regarded as damage points. Finally, independent damage regions are identified, as Figure 5 shown;

[0148] Step S53, calculate the average value of the three-dimensional coordinates of all points within the identified damage region to obtain the coordinates of the center position of the damage region. Suppose there are n points within the damage region, and the three-dimensional coordinates of each point are (x i , y i , z i ), where i = 1, 2,..., n. Then the coordinates of the center position of the damage region can be calculated by the following formula:

[0149]

[0150] Step S54, based on the coordinates of the center position of the obtained damage region, design a damage region A cuboid spatial region centered on [the center], where the depth of the cuboid, i.e., the height direction, is set to d, determined according to the depth characteristics of the damage region, and a safety margin Δz set to ensure the inclusion of complete damage space information. The length of the cuboid is set to l and the width is set to w, which are then determined according to the expansion range of the damage region and additional boundaries Δx and Δy set to ensure the inclusion of complete damage space information, respectively. Specifically, it is obtained by adding the corresponding additional boundaries, Δx and Δy or the safety margin Δz, respectively, to the difference between the maximum and minimum values of all points in the damage region in the x-axis, y-axis, and z-axis directions:

[0151]

[0152] Step S55, extract all the three-dimensional point cloud data within this cuboid spatial region to form a point cloud subset containing the damage and its surrounding environment. As Figure 6 shown. Let the entire three-dimensional point cloud data set be P, then the extracted point cloud subset P subset can be expressed as:

[0153]

[0154] Step S6, as Figure 4 shown, use the "voxelization - binarization" method to convert the three-dimensional point cloud data into a three-dimensional binary matrix, completing the accurate extraction and complete retention of the damage space information, providing reliable damage space information for automatic finite element modeling. This step includes the following sub-steps:

[0155] Step S61, rotate the three-dimensional point cloud data of the damage region after detection and segmentation to the XOY plane through the Rodrigues rotation formula for subsequent voxelization processing. The Rodrigues rotation formula is:

[0156] v rot = vcosθ+(k × v)sinθ + k(k · v)(1 - cosθ)

[0157] where v is the original vector, k is the unit vector of the rotation axis, and θ is the rotation angle;

[0158] Step S62, perform voxelization processing on the rotated three-dimensional point cloud data, and perform complete voxelization with the same resolution as the grid resolution in the previous stage. The resolution is 3mm to suit downstream tasks for damage assessment. The voxelized cuboid contains three types of voxels: namely, (1) empty voxels in the damage space, (2) voxels occupied by the solid surface and the damage surface, and (3) voxels not observed due to occlusion. As Figure 7 shown.

[0159] Step S63, define the mathematical representation of voxelization using the inclusion–exclusion principle. Let the center coordinates of voxel v j be c j = [c x , c y , c z T , then the calculation formula for the number of point clouds in each voxel is:

[0160]

[0161] where N j is the number of point clouds in the j-th voxel, N is the total number of points in the point cloud, ||·||2 represents the Euclidean distance, is an indicator function, when the distance from p i to c j does not exceed , its value is 1; otherwise it is 0. By traversing the separated damage space, that is, all the point cloud points in the cuboid, count the number of points located in each voxel v j area. If N j exceeds the preset threshold, then the voxel v j is regarded as the occupied state; otherwise, it is regarded as the unoccupied state. By assigning values to the voxels, marking the unoccupied state, that is, the damaged voxels or unobserved voxels as 1, and the occupied voxels, that is, the undamaged voxels as 0, each layer or slice of the voxelized cuboid along the depth direction can be used as a binary image.

[0162] Step S64, update the empty voxels, occupied voxels, and unobserved voxels in the subsequent layers along the depth direction through morphological operations.

[0163] Step S65, obtain the boundary information of the voxel grid, and then calculate the size of the grid.

[0164] Step S66, initialize a three-dimensional array all filled with 1 as a blank binary matrix, where 1 represents damaged voxels, including unobserved voxels; 0 represents undamaged voxels;

[0165] Step S67, traverse each voxel in the voxel grid, and set the value to 0 at the corresponding position in the binary matrix according to its grid index, marking the positions occupied by the point cloud, so as to construct a three-dimensional binary matrix.

[0166] Step S7, according to the extracted damage space information, based on the birth and death element method, modify the element stiffness matrix of the finite element grid, set the elements corresponding to the damage area as "dead" elements to simulate the damage in the actual structure; at the same time, keep the elements in the undamaged area as "live" elements to realize the automatic adjustment of the finite element model of the multi-slab structure with existing hinge joint damage, such as​Figure 8 As shown. This step includes the following sub-steps:

[0167] Step S71, based on the extracted three-dimensional structural dimension information and drawing information, establish a complete finite element model with voxelized equivalent resolution, as Figure 9 shown;

[0168] Step S72, initialize an empty set of damaged elements. According to the segmented cuboid space region, locate this space region in the finite element model and determine the finite element mesh elements it contains.

[0169] Step S73, using the voxelization assignment result, for each element in the finite element model, if the voxel value corresponding to the element is 1, add the element to the set of damaged elements, establish an exact mapping between voxels and finite element mesh elements, and generate a list containing all damaged elements;

[0170] Step S74, in the finite element model, use the "kill element" technique to simulate the damaged area;

[0171] Step S75, by preselecting the elements or sets of elements in the damaged area in the "model change" setting in the interaction module, perform the element birth and death operation;

[0172] Step S76, multiply the stiffness matrix of the element by a very small coefficient, such as 1e-6, so that its contribution to the overall stiffness matrix can be ignored, and achieve flexible control in the structural damage modeling and evolution analysis.

[0173] Step S77, keep the elements in the undamaged area as "live" elements to ensure that the finite element model can accurately simulate the damage situation in the actual structure and normally participate in the finite element calculation.

[0174] Step S8, compare the finite element model constructed by the birth and death element method with the test data of the actual structure and the finite element model constructed by the manual modeling combined with the element deletion method, simulate and analyze the mechanical properties of the structure under different loads, and according to the simulation results, evaluate the impact of damage on the overall performance of the structure, including changes in aspects such as the stress, lateral distribution coefficient, and bearing capacity of the structure, and finally verify the effectiveness of the finite element model damage area adjustment method proposed by the present invention in simulating the mechanical properties of damaged structures.

[0175] The following illustrates the solution of this embodiment by constructing a test scenario.

[0176] Construct a test model of a multi - beam slab structure with a scale ratio of 1:4, made of C40 - grade concrete and HRB400 and HPB400 - grade steel bars. The joint area is connected by post - cast concrete, and the steel bars are connected by welding. Steel plates are embedded in the loading area and welded to the steel bars to avoid local bearing failure of the concrete during loading. Use a standing lidar to collect three - dimensional point cloud data, and apply the method described in the present invention for automatic finite - element modeling and damage assessment. The experimental device is arranged as Figure 10 shown.

[0177] For each test condition, load is applied in stages at the mid - span of each slab span. Use a jack to apply a 3t load. Record data such as the actual magnitude of the applied load and the reaction forces at each support. The sensors are arranged as Figure 11 shown.

[0178] As Figure 12 shown, it shows a comparison of two common damage - modeling methods: manual modeling combined with the element - deletion method, Figure 12 (a), and the birth - and - death element method proposed in the present invention, Figure 12 (b). The figure shows the mesh division of the same structural model under these two methods. The element - deletion method simulates structural damage by deleting the elements in the damaged area. Figure 12 The mesh in (a) represents that under this method, the elements corresponding to the damaged area are completely removed. Although this method is simple, it has certain limitations in simulating damage propagation and accuracy. Especially in complex damage patterns, it may not accurately reflect the actual response of the structure. In contrast, the method proposed in the present invention, which combines point - cloud information and the birth - and - death element method, has an automated function, Figure 12 (b). It quickly locates damage by combining point - cloud information, automatically locates the damaged elements through an algorithm, and then marks the elements in the damaged area as "birth - and - death elements", that is, allowing the physical properties of these elements to change during the analysis process to simulate the development and evolution of damage. This method can capture the damage process more accurately. Especially when dealing with nonlinear and dynamic responses, compared with the element - deletion method, it can better reflect the actual damage behavior of the structure and provide more accurate simulation results.

[0179] To further verify the effectiveness of the proposed method, load the existing damaged finite - element model. Four loading conditions are set in this test. Apply a 3t load at different beam positions of the multi - beam slab with existing damage. Condition 1: The load acts on beam 2; Condition 2: The load acts on beam 3; Condition 3: The load acts on beam 4; Condition 4: The load acts on beam 5. The experimental loading process uses a hydraulic jack for loading to simulate the stress conditions during the actual use of the bridge. Through finite - element analysis, obtain response parameters such as the stress and lateral distribution coefficient of the structure in the damaged state, and compare them with the experimental results for verification.

[0180] Figure 13 It shows the stress simulation results of the multi - slab beam structure under different loading conditions at a specific load level. In the figure, (a) is the stress nephogram under the condition of intact structure, with uniform stress distribution and no obvious stress concentration phenomenon. (b) to (d) are the stress nephograms under different hinge joint loading conditions respectively. These simulation results reveal the influence of damage on the internal stress distribution of the structure, providing an important basis for understanding the damage mechanism and evaluating the structural safety performance.

[0181] Figure 14 It compares the simulation results and experimental results of the lateral distribution coefficient of the multi - slab beam structure under different loading conditions at a specific load level. Among the four working conditions, the method proposed in the present invention has a higher degree of coincidence with the experimental data than the method of manual modeling combined with the element deletion method. Especially at key positions such as Beam 6, Beam 3, and Beam 5, the method of the present invention shows higher simulation accuracy. The element deletion method has larger errors at the positions of some beams, while the method of the present invention shows smaller errors at all beam positions. In addition, it can be seen that under different loading conditions, the simulation results and experimental results show good consistency in both numerical values and trends. As the degree of damage increases, the lateral distribution coefficient changes, reflecting the influence of damage on the load - sharing ability of the structure. These comparison results verify the accuracy of the finite - element model in simulating the structural load distribution.

[0182] Figure 15 (a) shows the comparison of the experimental results of the bearing capacity of the existing damaged multi - slab beam structure, the results simulated by the method proposed in the present invention, and the results simulated by the method of manual modeling combined with the element deletion method. The results show that the method proposed in the present invention is closer to the experiment at the turning points of the ascending and descending sections of the bearing capacity, while the manual modeling method has deviations at some key points, proving the advantage of the method proposed in the present invention in improving the simulation accuracy. Figure 15 (b) compares the bearing capacity curves of the intact structure and the damaged structure. It can be seen that the bearing capacity curve of the intact structure shows typical characteristics of an ascending section, a stable section, and a descending section, and the ultimate bearing capacity is relatively high. While the bearing capacity curve of the damaged structure shows an obvious downward trend in the ascending section, and the ultimate bearing capacity is significantly reduced. This comparison result fully illustrates the adverse effect of damage on the mechanical properties of the structure, and also verifies the effectiveness of the method proposed in the present invention in simulating the mechanical properties of damaged structures.

[0183] In summary, the specific embodiments verify the feasibility of the proposed solution of the present invention. The experimental results show that the simulation accuracy of the method proposed by the present invention in terms of stress, transverse distribution coefficient, bearing capacity, etc. is significantly higher than that of the traditional manual modeling combined with the element deletion method. Specifically, under different loading conditions, the simulation results of the method of the present invention are highly consistent with the experimental data, demonstrating its superiority and practicality in dealing with complex structural deformation and damage assessment. Therefore, the present invention provides an efficient and accurate technical means for the performance evaluation of existing damaged bridge structures.

[0184] As described above, the above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention can easily think of various equivalent modifications or substitutions, and these modifications or substitutions should be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be subject to the protection scope of the claims.

Claims

1. An automatic finite element modeling and performance evaluation method for existing damaged hollow slab beam bridges based on visual data, characterized in that, Including the following steps: S1. Based on 3D laser scanning technology, comprehensively scan the multi-girder slab structure to obtain complete 3D point cloud data of the multi-girder slab model; S2. Based on the obtained 3D point cloud data, adopt a 3D dimensional automatic measurement method for the structure by fusing plane projection and point cloud distortion correction. By extracting the key geometric features of the structure, provide 3D dimensional data of the multi-girder slab structure with damaged hinge joints for subsequent finite element modeling; S3. For the damaged area of the multi-girder slab structure, identify the boundary between the damaged area and the undamaged area of the multi-girder slab structure and conduct a detailed spatial position analysis of the damage through the method of comparing the height differences of neighboring points in the point cloud and clustering analysis; use the voxelization-binarization method to convert the complete 3D point cloud data of the multi-girder slab model into a 3D binary matrix, complete the accurate extraction and complete retention of the damage spatial information of the multi-girder slab structure, and provide damage location quantification information for automatic finite element modeling; S4. Combine the above damage location quantification information to automatically retrieve the damaged element set, establish an accurate mapping between voxels and finite element mesh elements, deactivate the finite element mesh in the failure damaged area through the birth and death element method, set the elements corresponding to the damaged area as "dead" elements to simulate the damage in the actual structure; at the same time, keep the elements in the undamaged area as "live" elements to achieve automatic adjustment of the finite element model of the multi-girder slab structure with damaged hinge joints.

2. The automatic finite element modeling and performance evaluation method for existing damaged hollow slab girder bridges based on visual data according to claim 1, characterized in that, In step S2, the steps of extracting the key geometric features of the structure by using the method of fusing plane projection and point cloud distortion correction specifically include: S201. Perform RANSAC segmentation on the 3D point cloud model to obtain six faces of the complete multi-girder slab model; S202. Use the rotation matrix R to perform a spatial transformation on the three-dimensional point cloud data, and rotate the front or left side of the multi-beam slab model to a position coplanar with the upper surface respectively to ensure the accuracy and consistency of the three-dimensional point cloud data. Let the rotation matrix be R. For the case of 90 degrees around the x-axis and y-axis, the rotation matrix is expressed as: S203. Use the PCA algorithm to perform plane fitting on the segmented 3D point cloud data to obtain the approximate plane equation where the 3D point cloud data is located; S204. By calculating the vertical projection points of each 3D point cloud data onto the fitted plane, project all the 3D point cloud data onto the same plane to generate a height map with high precision. Let the rotated point cloud point set be P = [p1, p2,..., p n , where p i = [x i , y i , z i T represents the 3D coordinates of the i-th point. By calculating the minimum and maximum values of the x and y coordinates, set the boundary range of the height map, that is:​ x min = min(x i ), x max = max(x i ) y min = min(y i ), y max = max(y i ) S205, normalize the height values in the height map and map them to the grayscale value range of 0 - 255, so that the height information is presented in the form of gray levels, and obtain the corresponding grayscale image. The pixel value of the grayscale image at the position (x i , y i ) is: Among them, H(x i , y i ) = max(z i |(x i , y i , z i )) represents the height value of a certain point on the height map, and z represents the Z coordinate of the three-dimensional point cloud data point; S206. Through image processing algorithms, perform edge detection and contour extraction on the grayscale image to identify the key geometric features of the structure, such as beam-slab boundaries and structural holes, and provide 3D dimensional data for subsequent finite element modeling.

3. The automatic finite element modeling and performance evaluation method for existing damaged hollow slab girder bridges based on visual data according to claim 1, characterized in that, In step S206, the steps of calculating the beam-slab boundary by using image processing algorithms specifically include: S2061. Apply Gaussian blur processing to the original image, smooth the image by setting an appropriate standard deviation, and the kernel function of Gaussian blur is expressed as: where σ is the standard deviation of the Gaussian distribution; S2062. Use the Canny edge detection algorithm to process the smoothed image, extract the contour edges of the target area by setting high and low thresholds, and at the same time suppress the response of non-edge points to generate a continuous and complete edge image; S2063. Use the Hough gradient method to detect straight lines in the image, identify the straight line segments in the image, traverse all the detected straight lines, extract their endpoint coordinates, and use the minimum area rectangle fitting algorithm to approximate the target rectangle frame; finally, obtain the rectangle frame and the coordinates of its four vertices on the height image; S2064. Calculate the distance between two points according to the Euclidean distance formula to provide data support for subsequent dimension measurement. In a two-dimensional plane, for two points P1(x1, y1) and P2(x2, y2), the Euclidean distance d between them pixel is defined as: S2065. Align the pixel data obtained based on the image to the real boundary data in the height map. During the image processing process, two-dimensional coordinates are usually converted into pixel coordinates to achieve the alignment of the image size: d real = k·d pixel Where P d is the pixel height of the height map, P w is the pixel width of the height map, B y is the coordinate system boundary height of the height map, B x is the coordinate system boundary width of the height map, k is the conversion alignment scale factor, d real is the true dimension parameter.

4. The automatic finite element modeling and performance evaluation method for existing damaged hollow slab girder bridges based on visual data according to claim 1, characterized in that In step S206, the parameters of the structural holes are calculated using an image processing algorithm by introducing the Hough circle transform to detect circular holes in the image, so as to obtain the key dimensions of the holes and locate the holes.

5. The automatic finite element modeling and performance evaluation method for existing damaged hollow slab beam bridges based on visual data according to claim 1, characterized in that In step S3, the steps of identifying the boundary between the damaged area and the undamaged area by comparing the height differences of the neighboring points of the point cloud and clustering analysis specifically include: S301, use the KD-tree spatial indexing technology to find the set of neighborhood points N(P) within a certain radius r around it and calculate the average height of the neighborhood points N(P) = KDTree.query(P, r) Among them, KDTree.query(P, r) is a query function in the KD-tree spatial indexing technology. It accepts two parameters: point P and radius r. P represents the point to be queried, that is, the center point of the neighboring points around which it is hoped to find; r represents the query radius, which defines a spherical area centered on point P, and all points falling within this area will be regarded as the neighboring points of point P; |N(P)| is the number of neighboring points; S302, perform point cloud neighborhood height difference comparison analysis, and compare the height z of the current point i with the average height of the neighborhood points If the height z of the current point i is significantly lower than the average height of the neighborhood points and exceeds the preset threshold Δz, that is, z i < then mark this point as a potential damage point; S303, using the DBSCAN clustering algorithm to group the potential damage points. Each cluster in the clustering result represents an independent damaged area, and the points inside it are regarded as damage points, and finally independent damaged areas are identified.

6. The automatic finite element modeling and performance evaluation method for existing damaged hollow slab beam bridges based on visual data according to claim 1, characterized in that In step S3, for the damaged area of the multi-beam slab structure, the steps of performing spatial position analysis by comparing the height differences of the neighboring points of the point cloud and clustering analysis specifically include: S304, calculate the average value of the three-dimensional coordinates of all points within the identified damage area to obtain the coordinates of the center position of the damage area; assume that there are n points within the damage area, and the three-dimensional coordinates of each point are (x i , y i , z i ), where i = 1, 2, …, n, then the coordinates of the center position of the damage area are calculated by the following formula: S305. Based on the coordinates of the center position of the obtained damage area, design a cuboid spatial area centered on the damage area The depth of the cuboid, that is, the height direction, is set to d, which is determined according to the depth characteristics of the damage area and the safety margin Δz set to ensure that complete damage space information is included; the length of the cuboid is set to l and the width is set to w. Then, according to the expansion range of the damage area and the additional boundaries set to ensure that complete damage space information is included, which are set to Δx and Δy respectively, it is determined specifically by adding the corresponding additional boundaries, which are set to Δx and Δy respectively, or the safety margin Δz to the differences between the maximum and minimum values of all points in the damage area in the x-axis, y-axis, and z-axis directions: S306, extracting all the three-dimensional point cloud data within this cuboid spatial area to form a point cloud subset containing the damage and its surrounding environment.

7. The automatic finite element modeling and performance evaluation method for existing damaged hollow slab girder bridges based on visual data according to claim 1, characterized in that, In step S3, the steps of converting the three-dimensional point cloud data into a three-dimensional binary matrix using the voxelization-binarization method specifically include: S307, rotating the three-dimensional point cloud data of the damaged area after detection and segmentation to the XOY plane through the Rodrigues rotation formula. The Rodrigues rotation formula is: v rot = v cosθ + (k × v) sinθ + k(k · v)(1 - cosθ) Among them, v is the original vector, k is the unit vector of the rotation axis, and θ is the rotation angle; S308, performing voxelization processing on the rotated three-dimensional point cloud data to generate a regular three-dimensional grid, that is, a voxel. Each voxel represents a fixed volume unit in space; S309, constructing a three-dimensional binary matrix according to the voxelized data to realize the quantization of the damaged space information.

8. The automatic finite element modeling and performance evaluation method for existing damaged hollow slab girder bridges based on visual data according to claim 7, characterized in that In step S308, the steps of voxelization processing specifically include: S3081, for the separated cuboid cross-section, complete voxelization is performed using the same resolution as the grid resolution in the previous stage. The resolution is 3mm to suit the downstream tasks for damage assessment. The voxelized cuboid contains three types of voxels: namely (1) empty voxels in the damaged space, (2) voxels occupied by the solid surface and the damaged surface, and (3) voxels not observed due to occlusion; S3082, the mathematical representation of voxelization is defined using the inclusion–exclusion principle. Let the center coordinates of voxel v j be c j = [c x , c y , c z T , then the calculation formula for the number of point clouds within each voxel is:​ where N j is the number of points in the j-th voxel, N is the total number of points in the point cloud, ||·||2 represents the Euclidean distance, is an indicator function that has a value of 1 when the distance from p i to c j is no more than ; otherwise, it is 0; if N j exceeds a preset threshold, then the voxel v j is considered to be in an occupied state; otherwise, it is considered to be in an unoccupied state; by assigning values to the voxels, the unoccupied state, i.e., damaged voxels or unobserved voxels, is marked as 1, and the occupied-state voxels, i.e., undamaged voxels, are marked as 0. Then, each layer or slice of the voxelized cuboid along the depth direction serves as a binary image; S3083, the empty voxels, occupied voxels, and unobserved voxels of the subsequent layers along the depth direction are updated through morphological operations.

9. The automatic finite element modeling and performance evaluation method for existing damaged hollow slab girder bridges based on visual data according to claim 8, characterized in that In step S309, the specific steps of constructing a three-dimensional binary matrix according to the voxelized data include: S3091, obtaining the boundary information of the voxel grid, and then calculating the size of the grid; S3092, initializing a three-dimensional array all filled with 1 as a blank binary matrix, where 1 represents damaged voxels, including unobserved voxels; 0 represents undamaged voxels; In S3093, traverse each voxel in the voxel grid, and according to its grid index, set the value to 0 at the corresponding position in the binary matrix to mark the positions occupied by the point cloud.

10. The automatic finite element modeling and performance evaluation method for existing damaged hollow slab girder bridges based on visual data according to claim 6, characterized in that, In step S4, the specific steps of establishing the exact mapping between voxels and finite element mesh elements include: S401, establish an FE model with a voxelized equivalent resolution; S402, initialize an empty set of damaged elements to store the finite element elements determined to be damaged subsequently; according to the cuboid spatial region obtained in step 304, as the segmented damaged spatial region, locate this damaged spatial region in the finite element model and determine the finite element mesh elements it contains; S403, utilize the voxelization result obtained in step 3, where a voxel value of 1 represents the damaged region and a voxel value of 0 represents the non-damaged region; for each element in the finite element model, if the voxel value corresponding to the element is 1, add the element to the set of damaged elements; through the above mapping process, generate a list containing all damaged elements for subsequent accurate simulation of the damage situation in the actual structure based on the birth and death element method.

Images