Method for automatically constructing digital twin model of bridge prefabricated component under sparse point cloud data

Abstract

The application discloses a method for automatically constructing a digital twin model of a bridge prefabricated component under sparse point cloud data, and steps include: S1, acquiring a sparse point cloud data set of the bridge prefabricated component; S2, improving a region growing algorithm and combining principal component analysis to dynamically search for an optimal growing threshold, thereby realizing automatic segmentation of sparse point cloud data of a component end face; S3, determining a main shaft direction of the component, arranging a slice plane along the main shaft direction, adaptively determining a slice thickness in combination with an end face geometric thickness, and extracting a measured slice contour through boundary point detection; S4, introducing a design section contour as a geometric prior, registering the measured slice contour extracted in S3, constructing a parameterized section contour, and optimizing and updating geometric parameters of the contour; and S5, constructing a digital twin model of the bridge prefabricated component. The method realizes automatic processing of sparse point cloud and robust extraction of key geometric elements, and provides method support for geometric review and construction geometric control of the prefabricated component.

Description

Technical Field

[0001] This invention belongs to the field of intelligent detection technology in civil engineering, and in particular relates to an automatic construction method for digital twin models of precast bridge components based on sparse point cloud data. Background Technology

[0002] With the continuous expansion of infrastructure construction in my country, including roads, bridges, municipal engineering, and rail transit, prefabricated construction methods are becoming increasingly popular. Precast bridge components (such as precast box girders and precast slab girders) have become core components in prefabricated structural systems due to their high standardization, high production efficiency, and easy quality control. However, throughout the entire process of factory production, on-site storage, transportation, hoisting, and operation, the geometric shape of precast components is affected by various factors, including template installation deviations, prestressing tensioning, hoisting deformation, and environmental loads. The actual formed state often deviates to varying degrees from the design model, easily leading to discrepancies between drawings and reality, posing potential risks to structural quality acceptance, safety assessment, and life-cycle management.

[0003] Digital twin technology can provide a new technological path for infrastructure status awareness, performance evaluation, and intelligent operation and maintenance by constructing virtual models that are highly consistent with physical entities and enabling real-time mapping and interaction of multi-source data. In the field of precast bridge components, establishing a high-precision digital twin model that is consistent with the actual geometric state of the component and continuously updating it throughout the component's life cycle will help achieve traceability of component production quality, quantification of construction and installation deviations, visualization of service status, and intelligent operation and maintenance decisions. However, how to efficiently and reliably construct and update digital twin models at the engineering scale remains a key technological bottleneck in current engineering applications.

[0004] In recent years, measurement methods such as 3D laser scanning can acquire point cloud data of precast component surfaces in a relatively short time. However, under factory and field conditions, point clouds often exhibit noise, non-uniform density, and local missing data due to factors such as occlusion, reflection, viewing angle, and sampling distance, and may even appear as sparse point clouds. Automatically constructing twin models for engineering applications based on such point cloud data still faces challenges, mainly in the following aspects:

[0005] (1) Insufficient automation in point cloud processing: Existing point cloud segmentation, registration and feature extraction processes generally rely on manual setting of thresholds and repeated parameter tuning, which are sensitive to operational experience; under sparse or missing point cloud conditions, the stability of the algorithm is further reduced, making it difficult to meet the automation needs of mass production and rapid on-site detection in factories.

[0006] (2) Insufficient accuracy and completeness of key geometric element extraction: Precast components often contain complex structures such as hollow cavities, end reinforcement, and local thickening. Point clouds in key areas such as the inner cavity surface and end face are prone to noise, voids or sparseness due to occlusion and reflection, resulting in discontinuous boundaries and missing features. Traditional methods based on fixed thresholds or rules are difficult to stably recover geometric details under sparse conditions, which can easily cause local distortion or topological errors.

[0007] (3) Difficulty in integrating and updating the design model and the measured point cloud: Complete CAD / BIM design models are common in engineering projects, but there is still a lack of mature and reliable automated technology for efficiently registering, identifying differences, and updating geometric twins between parametric design models and discrete measured sparse point clouds. Existing methods mostly rely on manual comparison, local repair, or simple rigid registration, which makes it difficult to simultaneously consider computational efficiency, geometric accuracy, and the consistency of component topology, and also makes it difficult to accurately characterize non-rigid geometric deviations caused by construction errors, material deformation, etc.

[0008] Therefore, there is an urgent need for an automatic construction method for digital twin models that integrate prior design information for bridge prefabricated components. Summary of the Invention

[0009] The purpose of this invention is to provide an automatic construction method for digital twin models of precast bridge components under sparse point cloud data. This method enables stable execution of end face segmentation, slice contour extraction, registration, and parameterized updates under sparse point cloud conditions. It also utilizes prior information to automatically process measured sparse point clouds and identify key set features, driving continuous and accurate updates of the digital twin model throughout its lifecycle, thereby obtaining a digital twin model that can be used for geometric verification and construction geometric control.

[0010] The technical solution adopted in this invention is an automatic construction method for digital twin models of precast bridge components based on sparse point cloud data, the steps of which include:

[0011] Step S1: Obtain the sparse point cloud dataset of the bridge prefabricated components;

[0012] Step S2: Based on the sparse point cloud dataset, improve the region growing algorithm and combine global and local PCA to adaptively search for the optimal growing threshold to achieve automatic segmentation of sparse point cloud data on the end face of bridge prefabricated components.

[0013] Step S3: Based on the segmentation results, determine the main axis direction of the bridge component, arrange parallel slice planes along the main axis direction, and adaptively determine the slice thickness in combination with the end face geometric thickness. Perform boundary point detection in each slice and extract the measured slice contour.

[0014] Step S4: Introduce the designed cross-sectional profile as a geometric prior, register the measured slice profile extracted in S3, construct the parametric cross-sectional profile, and optimize and update the geometric parameters of the cross-sectional profile.

[0015] Step S5: Based on the slice-by-slice parameterized cross-sectional profile, construct a digital twin model of the bridge prefabricated component.

[0016] Furthermore, the specific steps of S2 are as follows:

[0017] S21. Global PCA analysis is used to locate the end face center of the bridge prefabricated component, and the point closest to the center is set as the seed point of the improved region growing algorithm.

[0018] S22, establish a data index structure for the 3D sparse point cloud of bridge prefabricated components. First, input the 3D sparse point cloud dataset of bridge prefabricated components into the KD-tree construction algorithm for hierarchical partitioning to form a balanced search tree structure; then, analyze the sparse point cloud of bridge prefabricated components... any point p in i Perform a k-nearest neighbor search to traverse the sparse point cloud of precast bridge components. All points in the local neighborhood set N(p) of the bridge prefabricated component are used to obtain the local neighborhood set N(p) of the bridge prefabricated component. i ), then for N(p i Principal component analysis was performed on the sparse point cloud within the area to calculate the local geometric features of the bridge prefabricated components.

[0019] S23, Design the fitness function to obtain the optimal angle between curvature and normal vector, as shown in the following formula:

[0020]

[0021]

[0022] In the formula, For the region growing and segmentation operator, For the curvature to be optimized, Let the angle between the normal vectors be the angle to be optimized. Sparse point cloud for prefabricated bridge components. The sparse point cloud result of the end face obtained by the region growing algorithm segmentation; Sparse point cloud for the design end face of precast bridge components. It is a distance function for evaluating segmentation quality. This represents the function that minimizes the objective function. For the optimal curvature, This is the optimal normal angle.

[0023] Furthermore, the specific steps of S21 are as follows:

[0024] S21a, Calculating sparse point clouds of precast bridge components global covariance matrix Furthermore, the global covariance matrix is ​​decomposed using the following formula:

[0025]

[0026] In the formula, Indicates in Eigenvalues ​​in the direction, Indicates different main directions, , Represents eigenvalues The corresponding feature vector, Represents a 3×3 identity matrix;

[0027] S21b, PCA analysis is performed on the sparse point cloud of the precast bridge components, taking the maximum eigenvalue and its corresponding eigenvector as the main direction of the overall extension trend of the sparse point cloud; then, the extreme points on the end face are searched, and the point cloud of the precast bridge components is analyzed. Projecting all points onto the principal direction vector yields the scalar projection value for each point. The two points with the smallest and largest projection values ​​are then identified and denoted as follows: and The and To correspond to the extreme value positions on the two end faces of the precast bridge components, seed points are finally located on the end faces where the extreme value points are located.

[0028] Furthermore, the local geometric features include the curvature features corresponding to the minimum eigenvalue. The angle between the normal vectors of any two points in the local neighborhood set and the normal vector of the point. The calculation formula is as follows:

[0029]

[0030]

[0031] In the formula, , , These are the eigenvalues ​​of the local neighborhood covariance matrix of the KD-tree structure in the three principal directions. ; For inverse cosine operation, , These represent two sets of direction vectors, Let be the vector magnitude.

[0032] Furthermore, the specific steps of S3 are as follows:

[0033] S31, perform plane fitting on the sparse point clouds of the two end faces of the bridge prefabricated component extracted in S2 to obtain the unit normal vectors of the two end faces. Determine the consistency of their directions by calculating the dot product of the unit normal vectors of the two end faces. Then, take the average and normalize the normal vectors of the two end faces to obtain the principal axis of the bridge prefabricated component along the length direction. With the principal axis direction as the target direction, construct a rotation matrix according to the rotation relationship between the principal direction and the target direction of the original sparse point cloud, and perform a rigid body rotation transformation on the original sparse point cloud as a whole.

[0034] S32, under the guidance of the end face, a unified sparse point cloud slice benchmark is established, the main axis direction is used as the sparse point cloud slice normal, parallel sparse point cloud slice planes are arranged along the main axis direction of the bridge prefabricated component, and boundary point detection is performed in each sparse point cloud slice to extract the corresponding slice contour.

[0035] S33. Utilizing the geometric thickness of the sparse point cloud on the end face of the precast bridge component in the normal direction, the thickness of the sparse point cloud slice of the precast bridge component is adaptively determined. Specifically, plane fitting is performed on the sparse point clouds at both end faces of the precast bridge component, the thickness of the sparse point cloud on the end face in the normal direction of the plane is calculated, and the average of the two end face thicknesses is taken. As the benchmark scale for point cloud slices of precast bridge components, the thickness of the point cloud slice is set to twice the average thickness of the end face.

[0036] S34. Based on the slicing direction and slicing thickness determined in S31~S33, the sparse point cloud of the bridge prefabricated component is processed into parallel slices along the main axis to obtain multiple point cloud slices. The sparse point cloud in each slice is orthogonally projected onto the corresponding cross-sectional plane to form a two-dimensional projection point set. Then, boundary point detection is performed on the two-dimensional projection point set. Finally, all boundary points in the slice are collected to form the measured slice outline.

[0037] Furthermore, the specific steps of S4 are as follows:

[0038] S41: Extract the design section profile from the design drawings of the precast bridge components, select key points on the design section profile as design feature points, perform initial alignment between the design profile and the measured slice profile obtained in S3, complete the preliminary registration, and introduce an adaptive template switching mechanism to detect outliers in the error distribution after preliminary registration. When the error exceeds the normal range, it is determined that the measured slice profile does not match the current standard design section profile, and the system automatically switches to the corresponding web-plate design profile and re-executes coarse registration.

[0039] S42, based on the initial registration, the design contour and the measured slice contour are finely registered. Specifically, in the two-dimensional cross-sectional plane corresponding to the point cloud slice, the correspondence between the points on the measured slice contour and the nearest point of the design edge is established. The vertical distance between the two is used as the registration error measure. The rigid transformation parameters between the two are iteratively updated so that the distance error gradually decreases and converges, and the alignment result of the finely registered contour is obtained.

[0040] S43, based on the fine registration results of S42, constructs a parametric cross-sectional profile model while keeping the design profile topology and construction information unchanged, and achieves adaptive updating of cross-sectional profile geometry through the offset of design feature points;

[0041] S44 uses the geometric update parameter set as the parameters to be optimized to perform optimization and update the parameterized cross-sectional profile.

[0042] Furthermore, the specific steps of S43 are as follows:

[0043] S43a, the design profile is discretized into an ordered set of edge segments under the constraint of design feature points, and the minimum distance mapping relationship from the measured slice profile to the design edge segments is established, specifically as follows:

[0044] First, obtain the design cross-sectional outline and design feature point index from the design drawings. Following the connection order of the design feature points, decompose the design outline into several ordered segments connected end-to-end, forming a set of design segments. ,in This represents the number of outline segments;

[0045] Then, for any contour point among the slice contour points obtained in S3 In the design section profile design edge set Search for the corresponding design edge segment in the middle, calculate the perpendicular distance from the contour point to the corresponding segment, and take the minimum distance value as the contour point. Find the shortest distance to the design contour, and at the same time determine the design segment number that achieves the minimum distance as the segment index to which the point belongs, and establish the mapping relationship between the measured slice contour point and the design contour segment.

[0046] S43b, based on S43a, constructs a parametric cross-sectional profile and measures the fitting error to design the set of edge segments. As the initial contour skeleton, the offsets of the design feature points in the cross-sectional plane coordinate system are used as geometric update variables to form the geometric update parameter set X={x r The positions of the updated feature points are calculated based on the geometric update parameter set X, according to the set r = 1, ..., Z. This yields the parameterized cross-sectional profile characterized by the updated feature points, where x... rZ represents the coordinate offset of the r-th design feature point in the local cross-section plane coordinate system, and Z is the number of design feature point parameters.

[0047] S43c, the offset of the design feature points is limited to a range consistent with the scale of the fine registration residual in step S42, and any contour point of the measured slice contour is calculated. The vertical distance to the parameterized cross-sectional profile determined by the geometric update parameter set X is used as the error metric, and the overall fitting error is defined as the mean absolute deviation, as shown in the following formula:

[0048]

[0049] In the formula, Represents any contour point of the measured slice profile. This represents the number of measured slice contour points. For the parameter set corresponding to the parameterized cross-sectional profile, Represents contour points From The vertical distance between the defined parameterized cross-sectional profiles, F(X), represents the objective function used to characterize the overall fitting error of the parameterized cross-sectional profiles to the geometry of the measured slice.

[0050] Furthermore, the specific process of optimizing and updating the parameterized cross-sectional profile in S44 is as follows:

[0051] Multiple candidate geometric update parameter sets X are randomly generated, and the parameters are updated according to the objective function defined in S43b. For each candidate parameter set, the fitting error of the parameterized cross-sectional profile model is calculated, and the obtained objective function value is used as the fitness evaluation index of the candidate parameter set. The candidate parameter set is iteratively updated and the fitness value is continuously reduced. After reaching the preset maximum number of iterations or meeting the convergence condition, the optimal geometric update parameter X* that minimizes the objective function is output. Based on this, the parameterized cross-sectional profile of the updated point cloud slice and its updated feature point coordinates are generated, forming a sequence of slice-by-slice parameterized cross-sectional profiles arranged in an orderly manner along the beam direction, which is used as the input of the digital twin model.

[0052] Furthermore, the specific steps of S5 are as follows:

[0053] S51 takes the parameterized section profile sequence and its updated feature point coordinates as input, and the design feature point number determined in S41 as the identifier. Under the topological consistency constraints of S43 to S44, the updated feature points with the same number are merged into the same number of updated feature point sequence, the correspondence between adjacent section segments is established, and according to the spatial order of the point cloud slices in the beam direction, cubic spline interpolation is performed on the coordinate sequence of the same number of updated feature points to obtain the continuously changing feature point trajectory along the beam direction.

[0054] S52, using the parameterized cross-sectional profile of each point cloud slice as the control section, and updating the feature points with the same number according to the one-to-one correspondence, the corresponding edge segments between adjacent sections are connected in sequence, and a three-dimensional surface is generated by sweeping / lofting. The reconstruction process is performed on different structural sections such as the main body of the bridge precast component and the thickening of the end web plate to obtain the corresponding local solid model.

[0055] S53 applies positional and tangential continuity constraints to each local entity model at the connection points, and closes the first and last sections to form a complete three-dimensional entity model, resulting in a digital twin model of the bridge prefabricated component that is geometrically consistent with the measured sparse point cloud, continuous along the beam direction, and retains the structural topology information.

[0056] The beneficial effects of this invention are:

[0057] 1. This invention achieves fully automated end-face segmentation, significantly reducing reliance on manual intervention and improving recognition accuracy. By using global PCA to automatically select seed points and designing a priori-driven threshold optimization framework, the key thresholds in region growing are transformed from empirically set parameters into quantifiable and optimizable mathematical problems, enabling fully automated execution of end-face point cloud segmentation. The fully automated execution of end-face segmentation reduces reliance on manual threshold setting; by designing a priori-driven adaptive threshold search, the stability and accuracy of end-face segmentation under sparse and missing point cloud conditions are improved, providing a high-precision initial geometric benchmark for subsequent point cloud slicing and twin modeling. Compared to traditional manual parameter tuning methods, this invention offers significant improvements in both accuracy and stability.

[0058] 2. In this invention, the point cloud slices have a unified orientation and robust contour extraction, which significantly improves the geometric usability under complex working conditions. By constructing a unified beam coordinate system through end face fitting normal, it ensures that all point cloud slices are strictly perpendicular to the end face, avoiding slice orientation drift caused by overall PCA under conditions of axis offset, uneven point density, and complex end structure. At the same time, the edge detection based on the adaptive point cloud slice thickness and multi-scale neighborhood angle analysis based on end face thickness can suppress edge misjudgment caused by sparse point cloud or local missing measurements, and improve the continuity and geometric integrity of the slice contour.

[0059] 3. This invention achieves high accuracy in updating the cross-sectional profile under design prior constraints, accurately reflecting local non-rigid deformations. By utilizing multi-template design priors and topological consistency mapping, a stable correspondence is established between each measured profile point (slice boundary point) and a specific design segment. Then, through constrained offset of design feature points and metaheuristic global optimization, the corresponding updated feature points are obtained, driving the design cross-sectional profile to adaptively evolve towards the true component shape. Error statistics show that after parametric updating, most cross-sectional profile errors can be controlled within the millimeter range. This proves that the method of this invention can accurately characterize non-rigid deformation features such as base plate deflection, flange outward movement, and local web bulging while maintaining the topological structure, providing a highly reliable geometric basis for subsequent load-bearing capacity assessment and health monitoring.

[0060] 4. The modeling process in this invention is highly automated, making it suitable for batch engineering applications and full lifecycle digital management. This invention integrates end face segmentation, point cloud slicing construction, template selection, topology mapping, parameterized updating, and twin reconstruction into a unified automated processing flow. Only sparse point cloud data and corresponding design drawings need to be input to complete the automatic construction and updating of the digital twin model, which greatly reduces manual interaction and subjective intervention. It can directly serve the quality inspection of precast beam factories, deviation analysis before and after erection, geometric status tracking during operation, and automatic generation of finite element analysis models, providing efficient and reliable geometric support for the intelligent manufacturing, digital delivery, and full lifecycle management of bridge and building precast components. Attached Figure Description

[0061] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0062] Figure 1 This is a flowchart of the present invention.

[0063] Figure 2 This refers to the sparse point cloud of the precast beam in this invention.

[0064] Figure 3 This is the flowchart of the adaptive segmentation algorithm for sparse point clouds at the end face, which is an improved region growing algorithm in this invention.

[0065] Figure 4 These are the design drawings and point clouds of the end faces of the precast bridge components in this invention, where (a) is the design end face of the test beam and (b) is the point cloud of the design end face.

[0066] Figure 5These are the convergence curves and times of the end face under different algorithms, where (a) is the convergence curve of end face #1, (b) is the convergence time of end face #1, (c) is the convergence curve of end face #2, and (d) is the convergence time of end face #2.

[0067] Figure 6 This refers to the end-face separation effect in this invention.

[0068] Figure 7 This is a schematic diagram of point cloud slicing and mapping in this invention, where (a) is along the principal axis. (a) shows the arrangement of point cloud slices, and (b) shows the typical projection result of point cloud slices.

[0069] Figure 8 These are edge detection results with different k values ​​in this invention, where (a) k=5, (b) k=10, (c) k=20, and (d) k=30.

[0070] Figure 9 The cross-sectional profiles designed in this invention are as follows: (a) is a standard cross-sectional profile, and (b) is a cross-sectional profile with a crossbeam.

[0071] Figure 10 This is the abnormal point cloud slice identification in this invention.

[0072] Figure 11 This is the ICP registration error distribution in this invention.

[0073] Figure 12 This is a topological mapping diagram of point cloud slices in this invention, where (a) is the topological mapping of point cloud slice #1 and (b) is the topological mapping of point cloud slice #3.

[0074] Figure 13 These are the convergence curves of point cloud slices in this invention, where (a) is the convergence curve of point cloud slice #1 and (b) is the convergence curve of point cloud slice #3.

[0075] Figure 14 These are comparison images of cross-sectional contour updates in this invention, where (a) is the cross-sectional contour update of point cloud slice #1 and (b) is the cross-sectional contour update of point cloud slice #3.

[0076] Figure 15 This is the updated error distribution in this invention.

[0077] Figure 16 These are smoothing effect diagrams in this invention, where (a) is the smoothing effect on the Y coordinate and (b) is the smoothing effect on the Z coordinate.

[0078] Figure 17 This is the error distribution diagram after smoothing in this invention.

[0079] Figure 18This is the result of twin modeling of bridge prefabricated components driven by updated feature points in this invention, where (a) is a partitioned reconstruction twin model and (b) is an overall continuous twin model. Detailed Implementation

[0080] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0081] Embodiments of the present invention provide an automatic method for constructing digital twin models of precast bridge components using sparse point cloud data, the flowchart of which is shown below. Figure 1 As shown, the steps include:

[0082] Step S1: Use 3D laser scanning to scan the precast bridge components and obtain the sparse point cloud of the precast bridge components. ,like Figure 2 As shown, each point in the figure represents a sampled data point. Due to limitations in scanning station location and field of view, at least the bottom slab area of ​​the precast bridge components will have missing point cloud measurements. In the formula, The first point in the point cloud representing prefabricated bridge components One point, This represents the total amount of data in the sparse point cloud of precast bridge components.

[0083] Step S2: Improve the region growing algorithm by optimizing the algorithm (in this embodiment, the Zebra optimization algorithm is selected, but other optimization algorithms can be replaced according to specific needs). Combine global and local PCA (Principal Component Analysis) to dynamically search for the optimal growth threshold, i.e., the optimal angle between curvature and normal vector. The improved region growing algorithm's end-face sparse point cloud adaptive segmentation algorithm flow is as follows: Figure 3 As shown, the specific steps are as follows:

[0084] S21. Global PCA analysis is used to locate the end face center of the bridge precast component, and the point closest to this center is set as the seed point of the improved region growing algorithm to automate the seed point extraction process. The specific steps are as follows:

[0085] S21a, Calculating sparse point clouds of precast bridge components global covariance matrix To reflect the degree of dispersion of the sparse point cloud in various directions, the global covariance matrix is ​​decomposed using the following formula:

[0086]

[0087]

[0088] In the formula, The centroid of the sparse point cloud representing the precast bridge components is the geometric center (arithmetic mean) of the data in the sparse point cloud. Indicates in Eigenvalues ​​in the direction are used to characterize sparse point clouds. The degree of dispersion in direction, These represent different main directions (3 in total). , Represents eigenvalues The corresponding feature vector; The first point in the sparse point cloud representing precast bridge components One point, This represents the total amount of data in the sparse point cloud of precast bridge components. Indicates transpose. This represents a 3×3 identity matrix.

[0089] S21b, PCA analysis is performed on the sparse point cloud of the precast bridge components. The dimensions of the precast components along the longitudinal direction (length) are significantly larger than in other directions, resulting in the largest variance in the sparse point cloud distribution along this direction. The largest eigenvalue and its corresponding eigenvector are taken as the principal direction of the overall extension trend of the sparse point cloud. Then, extreme point search is performed on the end face. All points in the sparse point cloud are projected onto the principal direction vector to obtain the scalar projection values ​​for each point. The two points with the smallest and largest projection values ​​are identified and denoted as follows: and The and This corresponds to the extreme value positions on the two end faces of the precast bridge component; finally, seed point positioning is performed on the end face where the extreme value points are located. The end face where it is located, with Centered on the end face plane, a neighborhood of points with radius R is selected as the candidate point set for the end face. In this embodiment, R is taken as 5 times the average point spacing of the sparse point cloud. The center position of this candidate point set in the end face plane is used as the reference point. The point in the candidate point set that is closest to the reference point is selected as the initial seed point of the end face. The method for determining the initial seed point on the end face and same.

[0090] S22. A data index structure for sparse point clouds of bridge prefabricated components is established based on KD-tree (K-Dimensional Tree) to support fast retrieval of local neighborhoods and acquisition of local geometric features.

[0091] Specifically, firstly, the sparse point cloud dataset of bridge prefabricated components is input into the KD-tree construction algorithm. The sparse point cloud is then recursively divided into hierarchical parts according to its spatial distribution across various coordinate dimensions, forming a balanced search tree structure for Euclidean space nearest neighbor queries, thereby achieving efficient indexing of large-scale sparse point cloud data. Subsequently, the sparse point cloud of bridge prefabricated components is further processed... any point p in i Perform a k-nearest neighbor search based on a KD-tree structure and Euclidean distance to traverse the sparse point cloud of precast bridge components. All points in the local neighborhood set N(p) of the bridge prefabricated component are used to obtain the local neighborhood set N(p) of the bridge prefabricated component. i Then, for this local neighborhood point set N(p) i Principal component analysis is performed on the sparse point cloud within the area to calculate the local geometric features of the precast bridge components. These local geometric features include the curvature feature σ corresponding to the minimum eigenvalue and the angle θ between the normal vectors of any two points in the local neighborhood point set. The calculation formula is as follows:

[0092]

[0093]

[0094] In the formula, , , These are the eigenvalues ​​of the local neighborhood covariance matrix of the KD-tree structure in the three principal directions. ; For inverse cosine operation, , These represent two sets of direction vectors, Let be the vector magnitude.

[0095] S23, design the fitness function. Using the design information of the precast bridge components as geometric priors, the optimal curvature and normal vector angle is obtained through automatic search using the Zebra optimization algorithm, achieving high-precision, fully automatic segmentation of the two end faces. The formula is as follows:

[0096]

[0097]

[0098] In the formula, For the region growing and segmentation operator, For the curvature to be optimized, Let the angle between the normal vectors be the angle to be optimized. Sparse point cloud for prefabricated bridge components. The sparse point cloud result of the end face obtained by the region growing algorithm segmentation; The sparse point cloud of the design end face of the bridge prefabricated component is obtained by extracting the corresponding end face from the CAD design model of the bridge prefabricated component and discretizing it, and its geometric shape is consistent with the design drawings. It is a distance function used to evaluate the segmentation quality and quantify P. seg and Consistency between them; This represents the function that minimizes the objective function. For the optimal curvature, This is the optimal normal angle.

[0099] Step S3: Determine the main axis of the precast bridge component, arrange parallel slicing planes along the main axis, adaptively determine the slice thickness based on the end face geometric thickness, and perform boundary point detection within each slice to extract the measured slice contour. The specific steps are as follows:

[0100] S31. Plane fitting is performed on the sparse point clouds of the two end faces of the bridge prefabricated component extracted in S2 to obtain the unit normal vectors of the two end faces. The consistency of their directions is judged by calculating the dot product of the unit normal vectors of the two end faces. When the dot product is negative, the normal vector of one end is inverted so that the normal vectors of the two end faces point to the same axis, thus making the directions consistent. Then, the normal vectors of the two end faces are averaged and normalized to obtain the principal axis of the bridge prefabricated component along the length direction. Taking the principal axis direction as the target direction, a rotation matrix is ​​constructed according to the rotation relationship between the principal direction of the original sparse point cloud and the target direction. Rigid body rotation transformation is performed on the original sparse point cloud as a whole to align the principal axis of the bridge prefabricated component to a unified direction, thereby realizing the attitude correction and direction unification of the sparse point cloud.

[0101] S32 establishes a unified point cloud slicing benchmark under end-face guidance, uses the main axis direction as the point cloud slicing normal, arranges parallel point cloud slicing planes along the main axis direction of the bridge precast component, and performs boundary point detection in each point cloud slice to extract the corresponding slice contour, thereby effectively suppressing the overall PCA direction drift caused by noise, occlusion and missing measurements, and ensuring the direction consistency and slice contour extraction stability between different point cloud slices.

[0102] S33. The thickness of the point cloud slices on the end faces of precast bridge components is adaptively determined by utilizing the geometric thickness of the point cloud in the normal direction. This avoids the problem of contour distortion that can easily occur when relying on experience to select the point cloud slice thickness. Specifically, plane fitting is performed on the point clouds at both ends of the precast bridge component, and the thickness of the end face point cloud in the normal direction of this plane is calculated. This is the discrete range of the distance from the points on the end face to the fitted plane in the normal direction. The average thickness of the two end face slices is then taken. As the benchmark scale for point cloud slices of bridge precast components, the thickness of the point cloud slice is set to twice the average thickness of the end face (so that the slice thickness expands symmetrically relative to the slice plane).

[0103] The purpose of setting the coefficient 2 is to ensure that the slice thickness expands symmetrically relative to the slice plane. Due to end-face fitting errors, local missing measurements, and possible slight offsets in the slice plane, setting the slice thickness only on one side can easily cause the effective point cloud to be truncated, resulting in incomplete or unstable slice contours. By using a symmetrical slice thickness setting, the effective point cloud on both sides of the slice plane can still be fully included under the above conditions, thus ensuring that the point cloud slice range completely covers the end-face point cloud.

[0104] S34. Based on the slicing direction and thickness determined in S31~S33, the point cloud of the precast bridge component is processed into parallel slices along the main axis to obtain multiple point cloud slices. The point cloud in each slice is orthogonally projected onto the corresponding cross-sectional plane to form a two-dimensional projection point set. Then, boundary point detection is performed on the two-dimensional projection point set to obtain a topologically complete and robust slice profile. The specific process of boundary detection is as follows: taking any projection point in the two-dimensional projection point set as the center, a fixed number of nearest neighbor points k is used to search for its nearest neighbor point set (set according to the point cloud density in the slice, so that the nearest neighbor point set can cover the continuous geometric features of the local profile). Then, the direction angle from the point to each nearest neighbor point is calculated and sorted according to the angle size. The interval between adjacent direction angles is further calculated, and the largest angle interval is taken as the angle interval feature of the point. The angle threshold is set to 180°. When the maximum angle interval obtained by the point under different nearest neighbor scales is greater than the threshold, the point is determined to be a slice boundary point. Finally, all boundary points in the slice are collected to form the measured slice profile.

[0105] In this invention, the two-dimensional boundary obtained by S3 is called the measured slice profile; the cross-sectional boundary of the design drawing is called the design cross-sectional profile; and the updated profile obtained while maintaining the design topological constraints is called the parametric cross-sectional profile.

[0106] Step S4: Introducing the designed cross-sectional profile as a geometric prior, the measured slice profile extracted in S3 is registered from coarse to fine. A parameterized cross-sectional profile is constructed under the constraint of maintaining the design topology unchanged. A metaheuristic optimization algorithm is used to optimize and update the geometric parameters of the cross-sectional profile, achieving adaptive updating of the cross-sectional profile geometry. The specific steps are as follows:

[0107] S41: Extract the design section profile from the design drawings of the precast bridge components, and select key points on the design section profile as design feature points. Use a rigid transformation method based on centroid consistency to initially align the design profile with the measured slice profile obtained in S3, completing the preliminary registration. Introduce an adaptive template switching mechanism and use a box plot method to detect outliers in the error distribution after preliminary registration. Statistically calculate the first quartile Q1, the third quartile Q3, and the interquartile range IQR = Q3 - Q 1，A normal value discrimination interval [Q1-1.5×IQR, Q3+1.5×IQR] is constructed. When the error exceeds the normal interval, it is determined that the measured slice profile does not match the current standard design section profile. The system automatically switches to the corresponding web-plate design profile and re-executes coarse registration to achieve adaptive selection of section profile templates for different structural sections.

[0108] S42. Based on the initial registration, the Iterative Closest Point (ICP) algorithm is used to perform fine registration between the design profile and the measured slice profile. Specifically, in the two-dimensional cross-sectional plane corresponding to the point cloud slice, the correspondence between the points on the measured slice profile and the nearest points of the design edge is established. The vertical distance between the two is used as the registration error metric. The rigid transformation parameters between the two are iteratively updated so that the distance error gradually decreases and converges, and the alignment result of the finely registered profile is obtained, which is used for subsequent parameterized cross-sectional profile updates.

[0109] S43, Based on the fine registration results of step S42, while maintaining the design contour topology and construction information, a parametric cross-sectional contour model is constructed, and the adaptive update of the cross-sectional contour geometry is achieved through the offset of design feature points. The specific steps are as follows:

[0110] S43a involves discretizing the design profile into an ordered set of edge segments under the constraint of design feature points, and establishing a minimum distance mapping relationship between the measured slice profile and the design edge segments to form a stable topological correspondence. Specifically:

[0111] First, obtain the design cross-sectional outline and design feature point index from the design drawings. Following the connection order of the design feature points, decompose the design outline into several ordered segments connected end-to-end, forming a set of design segments. ( (The number of contour segments) serves as the initial geometric skeleton for the parametric cross-sectional contour, ensuring that the segment order is consistent with the construction information (top plate outer edge, web plate edge, bottom plate edge, etc.).

[0112] Then, for any contour point among the slice contour points obtained in S3 In the design section profile design edge set In the process, carefully search for the corresponding design edge segment, calculate the perpendicular distance from the contour point to the corresponding segment, and take the minimum distance value as the contour point. The shortest distance to the design contour is determined, and the design segment number that yields the minimum distance is assigned as the segment index to that point, thus establishing a mapping relationship between the measured slice contour point and the design contour segment. The distance calculation formula is as follows:

[0113]

[0114] in, Point To the edge segment vertical distance, Point gather The minimum perpendicular distance between each side segment of the middle. For the design outline of the first A border segment is used to represent a contour segment formed by connecting adjacent design feature points. This represents the set of design outline segments. Take the minimum value among all the included edge segments.

[0115] Through the above mapping relationship, the parametric cross-sectional profile update process is always constrained by the corresponding design edge segment, effectively avoiding topological errors caused by cross-edge segment fitting.

[0116] S43b, based on S43a, constructs a parametric cross-sectional profile and measures the fitting error to design the set of edge segments. As the initial contour skeleton, the connection order and construction information of the side segments remain unchanged; the offsets of the design feature points in the cross-sectional plane coordinate system are used as geometric update variables to form the geometric update parameter set X={x r |r=1,…,Z} (where x r Z represents the coordinate offset of the r-th design feature point in the local cross-sectional plane coordinate system, and Z is the number of design feature point parameters (used to drive the geometric update of the cross-sectional profile). Without changing the connection relationship of the design feature points, the position of the updated feature point is calculated according to the geometric update parameter set X, so that the cross-sectional profile undergoes continuous deformation as the updated feature point is offset, thereby obtaining the parameterized cross-sectional profile characterized by the updated feature points. This ensures that the cross-sectional profile maintains its closure and geometric continuity, without destroying the topological structure and construction information of the design profile.

[0117] S43c, the design feature point offset (geometric update variable) is limited to a range consistent with the fine registration residual scale in step S42, and any contour point of the measured slice contour is calculated. The vertical distance to the parameterized cross-sectional profile determined by the geometric update parameter set X is used as the error metric, and the overall fitting error (objective function) is defined as the mean absolute deviation, as shown in the following formula:

[0118]

[0119] In the formula, Represents any contour point of the measured slice profile. This represents the number of measured slice contour points. For the parameter set corresponding to the parameterized cross-sectional profile, Represents contour points From The vertical distance between the defined parameterized cross-sectional profiles, F(X), represents the objective function used to characterize the overall fitting error of the parameterized cross-sectional profiles to the geometry of the measured slice.

[0120] The objective function is used to quantify the approximation of the parameterized cross-sectional profile to the measured geometry, providing a unified fitness evaluation basis for subsequent metaheuristic optimization, thereby formalizing the parameterization update process into a constrained nonlinear optimization problem.

[0121] S44, using the geometric update parameter set X constructed in step S43b as the parameters to be optimized, a metaheuristic optimization algorithm (the Zebra optimization algorithm is used in this embodiment, but it can also be replaced with other metaheuristic optimization algorithms according to specific needs) is used to optimize and update the parameterized cross-sectional contour. Specifically, multiple sets of candidate geometric update parameter sets X are randomly generated, and the objective function defined in S43b is used to optimize the parameterized cross-sectional contour. For each candidate parameter set, the fitting error of the parameterized cross-sectional profile model is calculated, and the obtained objective function value is used as the fitness evaluation index of the candidate parameter set. The candidate parameter set is iteratively updated through a metaheuristic optimization algorithm, and the fitness value is continuously reduced until the preset maximum number of iterations is reached or the convergence condition is met. The optimal geometric update parameter X* that minimizes the objective function is output, which is the optimal geometric update parameter X* corresponding to the point cloud slice. Based on this, the updated parameterized cross-sectional profile of the point cloud slice and its updated feature point coordinates are generated, forming a sequence of slice-by-slice parameterized cross-sectional profiles arranged in an orderly manner along the beam direction, which is used as the input of the digital twin model.

[0122] Step S5: Construct a digital twin model of the prefabricated bridge components. The specific steps are as follows:

[0123] S51 takes the parameterized section profile sequence and its updated feature point coordinates as input, and the design feature point number determined in S41 as the identifier. Under the topological consistency constraints of S43 to S44, updated feature points with the same number are merged into a sequence of updated feature points with the same number. The correspondence between adjacent section segments is established so that the segments of adjacent sections correspond one-to-one. According to the spatial order of the point cloud slices in the beam direction, cubic spline interpolation is performed on the coordinate sequence of updated feature points with the same number to obtain the trajectory of feature points that change continuously along the beam direction, so as to eliminate discrete jumps between adjacent slices and ensure geometric continuity.

[0124] S52, using the parameterized cross-sectional profile of each point cloud slice as the control section, and updating the feature points with the same number according to the one-to-one correspondence, the corresponding edge segments between adjacent sections are connected in sequence, and a three-dimensional surface is generated by sweeping / lofting. The reconstruction process is performed on different structural sections such as the main body of the bridge precast component and the thickening of the end web plate to obtain the corresponding local solid model. The reconstruction process is the process of generating a three-dimensional surface.

[0125] S53 applies positional and tangential continuity constraints to the connection points of each local entity model, and closes the first and last sections to form a complete digital twin model. This results in a digital twin model of a precast bridge component that is geometrically consistent with the measured sparse point cloud, continuous along the beam direction, and retains structural topological information. Figure 18 As shown in (b).

[0126] This model can be directly used for dimensional quality inspection and deviation analysis of precast components. By comparing it with the design model, it can achieve quantitative assessment of key cross-sectional dimensions, linear offsets and local structural deformations, providing a reliable basis for factory acceptance, construction verification and quality traceability.

[0127] Experimental verification

[0128] (1) End face segmentation

[0129] Taking the original sparse point cloud of the test beam as an example, global PCA was used to analyze the overall sparse point cloud. The eigenvalue ratio of the three principal directions was approximately λ1:λ2:λ3 = 637:5:1, indicating that the dispersion of the sparse point cloud is most significant in the first principal direction, which is highly consistent with the structural feature of the bridge precast component with the largest longitudinal dimension. Subsequently, all points were projected onto the first principal direction to obtain scalar projection values. By identifying the minimum and maximum values ​​of the projection, the extreme points of the two end faces were automatically located, and the initial seed points of the two end faces were successfully extracted, namely Pseed1 (–178.3440,–167.9680,1.6480) and Pseed2 (–153.6360,–167.5540,1.8430).

[0130] After obtaining the seed points, all design feature points of the design end face are extracted and connected to construct a closed cross-sectional profile. The profile segments are then discretized regularly, and a dense, fully covered 3D sparse point cloud is generated by randomly filling point sets inside the profile. This achieves the mapping from 2D design parameters to 3D prior expression, providing a unified reference for end face segmentation evaluation and registration. For example, the end face of a precast bridge component... Figure 4As shown, (a) is the designed end face of the test beam, and (b) is the point cloud of the designed end face. Subsequently, region growth segmentation is performed on the given threshold combination and plane fitting is performed using RANSAC (Random Sample Consensus). The segmented end face and the designed end face are projected onto the same plane, and registration is achieved by solving the optimal two-dimensional rigid body transformation. The average error after registration is used as the segmentation quality evaluation index (the smaller the fitness value, the more complete the end face and the more consistent it is with the design geometry). Based on this mechanism, multiple metaheuristic algorithms are used to globally optimize the threshold combination. In terms of convergence performance, the Northern Goshawk Optimization (NGO), Harris Hawks Optimization (HHO), and Zebra Optimization Algorithm (ZOA) all achieve low fitting errors. ZOA shows the fastest decrease in error, with the two end-face errors converging to approximately 0.0025 and 0.0043 respectively in the early stages of iteration. Regarding time efficiency, ZOA's average running time is significantly lower than HHO and roughly equivalent to NGO. The convergence curves and time are shown below. Figure 5 As shown, (a) is the convergence curve of end face #1, (b) is the convergence time of end face #1, (c) is the convergence curve of end face #2, and (d) is the convergence time of end face #2. Considering both accuracy and efficiency, this invention selects ZOA as the optimized solver for end face sparse point cloud segmentation, and the segmentation results are as follows. Figure 6 As shown.

[0131] (2) Establishment of end face datum and point cloud slicing system

[0132] After separating the end faces, a plane fitting is performed on the point cloud of the end faces to obtain the equations of the two end faces: Π1: 0.9997x−0.0057y+0.0235z+177.2859=0, Π2: 1.0000x−0.0075y−0.0064z+152.3873=0, indicating that the two end faces are approximately planar and the normal direction is stable. From this, the normal vectors n1=(0.9997,−0.0057,0.0235) and n2=(1.0000,−0.0075,−0.0064) are obtained. The average of the two normal vectors is taken and normalized to obtain the direction of the principal axis of the beam. Based on this, a local coordinate system orthogonal to the global Z-axis is constructed, and the entire beam point cloud is rigidly rotated with the centroid of the two end faces as the rotation center to achieve the unification of the scanning coordinate system to the beam coordinate system. Regarding point cloud slice parameters, based on the average point cloud thickness of approximately 0.035 m at both ends, the point cloud slice thickness was set to twice this value, i.e., 0.070 m. Considering the beam length of approximately 25 m and the average point spacing of 0.023 m, the point cloud slice interval was set to 0.5 m (approximately 1 / 50 of the beam length), and the point cloud slices were densified in the complex structural area at the ends, generating a total of 32 sets of point cloud slices. Figure 7 As shown, (a) is a schematic diagram of the point cloud slice arrangement along the main axis, and (b) is a typical point cloud slice projection result.

[0133] (3) Two-stage approximation of contour extraction and “registration-parametric update”

[0134] Boundary point identification is performed on the slice point set based on the angular interval features of local neighborhoods. Comparing k=10, 20, 30, and 40, it can be seen that: k=10 has insufficient neighborhoods and is easily affected by sparsity / noise, leading to false detections; when k increases to 20, the statistics are more sufficient, and false detections are reduced; when k=30, the boundary is continuous, stable, and exhibits the best integrity and robustness; k=40 yields similar results but the boundary points are sparser and the computational cost increases, such as... Figure 8 As shown, (a) k=5, (b) k=10, (c) k=20, and (d) k=30. Considering all factors, k=30 is chosen as the default parameter. To ensure topological consistency across different structural sections, two types of design profiles—standard cross-sections and cross-sections with beams—are extracted from the drawings and discretized into point cloud priors, as shown below. Figure 9 As shown, (a) is the standard cross-sectional profile, and (b) is the cross-sectional profile with a crossbeam. Coarse registration was completed using bounding box centroid translation, and anomalous point cloud slices (#2, #3, #4, #29, #30, #31) were identified based on the error box plot. The template was then switched, and the anomalous point cloud slices are shown below. Figure 10 As shown in the figure. Subsequently, ICP (Iterative Closest Point) fine registration was performed, and the ICP registration error distribution is as follows. Figure 11 As shown.

[0135] This study statistically analyzed the distance distribution characteristics from the measured contour points to the design contour after ICP fine registration. The results showed that 74.53% of the point cloud errors were <0.010m, 95.92% were <0.020m, and approximately 4% of the points fell within the 0.02–0.07m range, mainly located in areas of geometric abrupt change such as diaphragms and end reinforcements. To further compensate for non-rigid offsets, a topological mapping of "12 design feature points – 12 design edge segments" was constructed, and parametric cross-sectional contour updates were performed. The topological mapping diagram is shown below. Figure 12 As shown, (a) is the topological mapping of point cloud slice #1, and (b) is the topological mapping of point cloud slice #3. Figure 12 Different colors represent the design contour segment numbers to which the measured slice contour points were determined to belong during the topology mapping process. Under shared offset constraints (P5–P8 have unified Y / Z, while P1–P2 and P11–P12 share Y-direction offsets respectively), ZOA optimization is performed with the goal of minimizing the average vertical distance. The optimal fitness of typical point cloud slices is 0.00386 m (#1) and 0.00287 m (#3). The convergence curves of the point cloud slices are shown in the figure. Figure 13 As shown, (a) is the convergence curve of point cloud slice #1, and (b) is the convergence curve of point cloud slice #3. The cross-sectional contour update comparison diagram is shown below. Figure 14 As shown, (a) is the updated profile of point cloud slice #1, and (b) is the updated profile of point cloud slice #3. Statistics for the entire beam show that the error converged significantly after the update; 96.68% of the point cloud errors were <0.010 m, and the proportion of errors >0.02 m decreased to less than 1%. The error distribution after the update is shown in the figure. Figure 15 As shown.

[0136] (4) Beam continuity processing and digital twin reconstruction

[0137] Since independent updates of the slice contour may cause slight jumps in the Y / Z offsets of updated feature points in adjacent slice contours, this invention performs cubic spline smoothing on the updated feature points with the same number along the beam direction sequence: P1–P8 and P11–P12 participate in the fitting, while P9–P10 (slices #2–#4 and #29–#31) located in special structural sections retain the original optimized values. The smoothing effect diagram is shown below. Figure 16 As shown in the figure, (a) represents the smoothing effect on the Y-axis, and (b) represents the smoothing effect on the Z-axis. Statistical results show that the error distribution after smoothing is consistent with that in the update stage, with 95.84% of the errors < 0.01m and 99.02% of the errors < 0.02m. Furthermore, the number of error points above 0.02m did not increase significantly. The error distribution after smoothing is shown in the figure below. Figure 17As shown in the figure, smoothness does not introduce new deviations. Based on this, a 3D twin reconstruction is performed: the smooth section is used as the control contour for lofting / sweeping to form the main body of the bridge precast component. For the thickened section of the end web, a closed transition contour is constructed by combining the optimized values ​​of P9–P10 at special sections with the spline prediction values ​​of the main body segment, achieving a natural and smooth splicing with the main body. Finally, a continuous, smooth, topologically consistent digital twin model is obtained, which highly matches the measured geometry. The updated feature-point driven twin modeling results of the bridge precast component are shown below. Figure 18 As shown, (a) is a partitioned reconstruction twin model, and (b) is a global continuous twin model.

[0138] The various embodiments in this specification are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0139] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of protection of the present invention.

Claims

1. An automatic method for constructing digital twin models of precast bridge components based on sparse point cloud data, characterized by the following steps: include: Step S1: Obtain the sparse point cloud dataset of the bridge prefabricated components; Step S2: Based on the sparse point cloud dataset, improve the region growing algorithm and combine global and local PCA to adaptively search for the optimal growing threshold to achieve automatic segmentation of sparse point cloud data on the end face of bridge prefabricated components. Step S3: Based on the segmentation results, determine the main axis direction of the bridge component, arrange parallel slice planes along the main axis direction, and adaptively determine the slice thickness in combination with the end face geometric thickness. Perform boundary point detection in each slice and extract the measured slice contour. Step S4: Introduce the designed cross-sectional profile as a geometric prior, register the measured slice profile extracted in S3, construct the parametric cross-sectional profile, and optimize and update the geometric parameters of the cross-sectional profile. Step S5: Based on the slice-by-slice parameterized cross-sectional profile, construct a digital twin model of the bridge prefabricated component; The specific steps of S2 are as follows: S21. Global PCA analysis is used to locate the end face center of the bridge prefabricated component, and the point closest to the center is set as the seed point of the improved region growing algorithm. S22, establish a data index structure for the 3D sparse point cloud of bridge prefabricated components. First, input the 3D sparse point cloud dataset of bridge prefabricated components into the KD-tree construction algorithm for hierarchical partitioning to form a balanced search tree structure; then, analyze the sparse point cloud of bridge prefabricated components... any point p in i Perform a k-nearest neighbor search to traverse the sparse point cloud of precast bridge components. All points in the local neighborhood set N(p) of the bridge prefabricated component are used to obtain the local neighborhood set N(p) of the bridge prefabricated component. i ), then for N(p i Principal component analysis was performed on the sparse point cloud within the area to calculate the local geometric features of the bridge prefabricated components. S23, Design the fitness function to obtain the optimal angle between curvature and normal vector, as shown in the following formula: ； ； In the formula, For the region growing and segmentation operator, For the curvature to be optimized, Let the angle between the normal vectors be the angle to be optimized. Sparse point cloud for prefabricated bridge components. The sparse point cloud result of the end face obtained by the region growing algorithm segmentation; Sparse point cloud for the design end face of precast bridge components. It is a distance function for evaluating segmentation quality. This represents the function that minimizes the objective function. For the optimal curvature, The optimal normal angle; The specific steps of S21 are as follows: S21a, Calculating sparse point clouds of precast bridge components global covariance matrix Furthermore, the global covariance matrix is ​​decomposed using the following formula: ； In the formula, Indicates in Eigenvalues ​​in the direction, Indicates different main directions, , Represents eigenvalues The corresponding feature vector, Represents a 3×3 identity matrix; S21b, PCA analysis is performed on the sparse point cloud of the precast bridge components, taking the maximum eigenvalue and its corresponding eigenvector as the main direction of the overall extension trend of the sparse point cloud; then, the extreme points on the end face are searched, and the point cloud of the precast bridge components is analyzed. Projecting all points onto the principal direction vector yields the scalar projection value for each point. The two points with the smallest and largest projection values ​​are then identified and denoted as follows: and The and To correspond to the extreme value positions on the two end faces of the precast bridge components; finally, seed point positioning is performed on the end face where the extreme value point is located. The local geometric features include the curvature features corresponding to the minimum eigenvalue. The angle between the normal vectors of any two points in the local neighborhood set and the normal vector of the point. The calculation formula is as follows: ； ； In the formula, , , These are the eigenvalues ​​of the local neighborhood covariance matrix of the KD-tree structure in the three principal directions. ; For inverse cosine operation, , These represent two sets of direction vectors, Let be the vector magnitude.

2. The method for automatically constructing digital twin models of precast bridge components based on sparse point cloud data according to claim 1, characterized in that, The specific steps of S3 are as follows: S31, perform plane fitting on the sparse point clouds of the two end faces of the bridge prefabricated component extracted in S2 to obtain the unit normal vectors of the two end faces. Determine the consistency of their directions by calculating the dot product of the unit normal vectors of the two end faces. Then, take the average and normalize the normal vectors of the two end faces to obtain the principal axis of the bridge prefabricated component along the length direction. With the principal axis direction as the target direction, construct a rotation matrix according to the rotation relationship between the principal direction and the target direction of the original sparse point cloud, and perform a rigid body rotation transformation on the original sparse point cloud as a whole. S32, under the guidance of the end face, a unified sparse point cloud slice benchmark is established, the main axis direction is used as the sparse point cloud slice normal, parallel sparse point cloud slice planes are arranged along the main axis direction of the bridge prefabricated component, and boundary point detection is performed in each sparse point cloud slice to extract the corresponding slice contour. S33. Utilizing the geometric thickness of the sparse point cloud on the end face of the precast bridge component in the normal direction, the thickness of the sparse point cloud slice of the precast bridge component is adaptively determined. Specifically, plane fitting is performed on the sparse point clouds at both end faces of the precast bridge component, the thickness of the sparse point cloud on the end face in the normal direction of the plane is calculated, and the average of the two end face thicknesses is taken. As the benchmark scale for point cloud slices of precast bridge components, the thickness of the point cloud slice is set to twice the average thickness of the end face. S34. Based on the slicing direction and slicing thickness determined in S31~S33, the sparse point cloud of the bridge prefabricated component is processed into parallel slices along the main axis to obtain multiple point cloud slices. The sparse point cloud in each slice is orthogonally projected onto the corresponding cross-sectional plane to form a two-dimensional projection point set. Then, boundary point detection is performed on the two-dimensional projection point set. Finally, all boundary points in the slice are collected to form the measured slice outline.

3. The method for automatically constructing digital twin models of precast bridge components based on sparse point cloud data according to claim 1, characterized in that, The specific steps of S4 are as follows: S41: Extract the design section profile from the design drawings of the precast bridge components, select key points on the design section profile as design feature points, perform initial alignment between the design profile and the measured slice profile obtained in S3, complete the preliminary registration, and introduce an adaptive template switching mechanism to detect outliers in the error distribution after preliminary registration. When the error exceeds the normal range, it is determined that the measured slice profile does not match the current standard design section profile, and the system automatically switches to the corresponding web-plate design profile and re-executes coarse registration. S42, based on the initial registration, the design contour and the measured slice contour are finely registered. Specifically, in the two-dimensional cross-sectional plane corresponding to the point cloud slice, the correspondence between the points on the measured slice contour and the nearest point of the design edge is established. The vertical distance between the two is used as the registration error measure. The rigid transformation parameters between the two are iteratively updated so that the distance error gradually decreases and converges, and the alignment result of the finely registered contour is obtained. S43, based on the fine registration results of S42, constructs a parametric cross-sectional profile model while keeping the design profile topology and construction information unchanged, and achieves adaptive updating of cross-sectional profile geometry through the offset of design feature points; S44 uses the geometric update parameter set as the parameters to be optimized to perform optimization and update the parameterized cross-sectional profile.

4. The method for automatically constructing digital twin models of precast bridge components based on sparse point cloud data according to claim 3, characterized in that, The specific steps of S43 are as follows: S43a, the design profile is discretized into an ordered set of edge segments under the constraint of design feature points, and the minimum distance mapping relationship from the measured slice profile to the design edge segments is established, specifically as follows: First, obtain the design cross-sectional outline and design feature point index from the design drawings. Following the connection order of the design feature points, decompose the design outline into several ordered segments connected end-to-end, forming a set of design segments. ,in This represents the number of outline segments; Then, for any contour point among the slice contour points obtained in S3 In the design section profile design edge set Search for the corresponding design edge segment in the middle, calculate the perpendicular distance from the contour point to the corresponding segment, and take the minimum distance value as the contour point. Find the shortest distance to the design contour, and at the same time determine the design segment number that achieves the minimum distance as the segment index to which the point belongs, and establish the mapping relationship between the measured slice contour point and the design contour segment. S43b, based on S43a, constructs a parametric cross-sectional profile and measures the fitting error to design the set of edge segments. As the initial contour skeleton, the offsets of the design feature points in the cross-sectional plane coordinate system are used as geometric update variables to form the geometric update parameter set X={x r The positions of the updated feature points are calculated based on the geometric update parameter set X, according to the set r = 1, ..., Z. This yields the parameterized cross-sectional profile characterized by the updated feature points, where x... r Z represents the coordinate offset of the r-th design feature point in the local cross-section plane coordinate system, and Z is the number of design feature point parameters. S43c, the offset of the design feature points is limited to a range consistent with the scale of the fine registration residual in step S42, and any contour point of the measured slice contour is calculated. The vertical distance to the parameterized cross-sectional profile determined by the geometric update parameter set X is used as the error metric, and the overall fitting error is defined as the mean absolute deviation, as shown in the following formula: ； In the formula, Represents any contour point of the measured slice profile. This represents the number of measured slice contour points. For the parameter set corresponding to the parameterized cross-sectional profile, Represents contour points From The vertical distance between the defined parameterized cross-sectional profiles, F(X), represents the objective function used to characterize the overall fitting error of the parameterized cross-sectional profiles to the geometry of the measured slice.

5. The method for automatically constructing digital twin models of precast bridge components based on sparse point cloud data according to claim 3, characterized in that, The specific process of optimizing and updating the parametric cross-sectional profile in S44 is as follows: Multiple candidate geometric update parameter sets X are randomly generated, and the parameters are updated according to the objective function defined in S43b. For each candidate parameter set, the fitting error of the parameterized cross-sectional profile model is calculated, and the obtained objective function value is used as the fitness evaluation index of the candidate parameter set. The candidate parameter set is iteratively updated and the fitness value is continuously reduced. After reaching the preset maximum number of iterations or meeting the convergence condition, the optimal geometric update parameter X* that minimizes the objective function is output. Based on this, the parameterized cross-sectional profile of the updated point cloud slice and its updated feature point coordinates are generated, forming a sequence of slice-by-slice parameterized cross-sectional profiles arranged in an orderly manner along the beam direction, which is used as the input of the digital twin model.

6. The method for automatically constructing digital twin models of precast bridge components based on sparse point cloud data according to claim 3, characterized in that, The specific steps of S5 are as follows: S51 takes the parameterized section profile sequence and its updated feature point coordinates as input, and the design feature point number determined in S41 as the identifier. Under the topological consistency constraints of S43 to S44, the updated feature points with the same number are merged into the same number of updated feature point sequence, the correspondence between adjacent section segments is established, and according to the spatial order of the point cloud slices in the beam direction, cubic spline interpolation is performed on the coordinate sequence of the same number of updated feature points to obtain the continuously changing feature point trajectory along the beam direction. S52, using the parameterized cross-sectional profile of each point cloud slice as the control section, and updating the feature points with the same number according to the one-to-one correspondence, the corresponding edge segments between adjacent sections are connected in sequence, and a three-dimensional surface is generated by sweeping / lofting. The reconstruction process is performed on different structural sections such as the main body of the bridge precast component and the thickening of the end web plate to obtain the corresponding local solid model. S53 applies positional and tangential continuity constraints to each local entity model at the connection points, and closes the first and last sections to form a complete three-dimensional entity model, resulting in a digital twin model of the bridge prefabricated component that is geometrically consistent with the measured sparse point cloud, continuous along the beam direction, and retains the structural topology information.

Images