Digital pre-assembly manufacturing control method for long-span arch bridge

Abstract

The application discloses a large-span arch bridge digital pre-assembly manufacturing control method, relates to the technical field of bridge engineering digital manufacturing, and takes the semantic importance of components, the distribution density of point clouds and the design connection relationship as cores, constructs semantic weight, distribution weight and connection relationship weight, eliminates overall deviation through weighted centroid calculation and decentralization, fuses the weight to construct a covariance matrix and solves an optimal rigid transformation through SVD decomposition, realizes global optimization in combination with an improved graph cut algorithm, and finally outputs aligned point clouds containing manufacturing errors. The public segment is taken as a benchmark to calibrate a reference line shape, data support is provided for pre-assembly target line shape calculation of conventional segments, the alignment priority of the core area is effectively strengthened, sparse point cloud and local deviation interference is avoided, the overall splicing line shape consistency of arch ribs is ensured, and accurate technical basis is provided for segment splicing construction.

Description

Technical Field

[0001] This invention relates to the field of digital manufacturing technology for bridge engineering, and more specifically to a digital pre-assembly manufacturing control method for long-span arch bridges. Background Technology

[0002] Long-span arch bridges have become a core bridge type for crossing rivers and canyons due to their advantages such as reasonable stress distribution, high stiffness, and strong adaptability to mountainous areas, with their spans gradually breaking through to the 500-meter and even 700-meter levels. To ensure on-site installation accuracy and avoid assembly problems in advance, the pre-assembly of arch rib segments is a key step in the manufacturing of long-span arch bridges. Among them, the N+1 multi-round pre-assembly method is widely used in actual projects because it is suitable for the manufacturing needs of multiple segments.

[0003] With the development of digital twin and 3D scanning technologies, digital pre-assembly is gradually replacing traditional physical pre-assembly, becoming the core direction for improving pre-assembly efficiency and accuracy. In existing technologies, point cloud data of the arch rib is typically collected using a 3D laser scanner and matched with a theoretical BIM model to achieve arch rib posture adjustment and pre-assembly. However, in practical applications, traditional digital pre-assembly often uses spherical targets for multi-point cloud stitching, but the laser reflection angle changes drastically at the edge of the sphere, easily generating point cloud noise, leading to sphere center fitting deviation, and thus affecting the matching accuracy between the actual arch rib and the theoretical BIM model. Simultaneously, the point cloud matching process does not consider the differences in component importance and point cloud distribution characteristics; manufacturing errors of non-critical components can easily be transmitted to core components such as the main chord, further reducing alignment accuracy. Traditional arch rib posture adjustment relies on repeated manual adjustments of the jig height, combined with multiple calibrations using large lifting equipment. This is not only time-consuming, labor-intensive, and costly, but also difficult to accurately match the target posture, unable to quickly obtain the jig's 3D coordinates based on point cloud features, and difficult to achieve posture positioning in one step through quantitative parameters.

[0004] Therefore, there is an urgent need for a digital pre-assembly manufacturing control method that takes into account high precision, high efficiency, and strong adaptability, in order to solve the above-mentioned technical pain points and ensure the manufacturing quality and installation efficiency of long-span arch bridges. Summary of the Invention

[0005] In view of this, the present invention provides a digital pre-assembly manufacturing control method for long-span arch bridges to solve the problems existing in the background art.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A digital pre-assembly manufacturing control method for long-span arch bridges includes: S1. Plan the scanning stations, set up rotating triangular pyramid targets, and collect point cloud data of each station of the arch rib using a laser scanner; preprocess the point cloud data, calculate the covariance matrix of each point, take the eigenvector corresponding to the smallest eigenvalue as the normal vector, and obtain a preprocessed point cloud containing six-dimensional information of position and normal vector. S2, based on the random sampling consensus algorithm, fit the equations of the three cone surfaces of the rotating triangular pyramid target, calculate the target feature point set, and perform multi-site point cloud stitching on the preprocessed point cloud by coarse registration using the fast four-point consensus set algorithm combined with the ICP algorithm for fine registration; according to the component semantic weight, point cloud distribution weight, and connection relationship weight, perform weight matching and alignment with the theoretical BIM model to obtain the aligned point cloud model; S3. Based on the aligned point cloud model, calculate the optimal pre-assembly target line shape with the smallest deviation between the regular segment and the theoretical line shape; for the common segment of multiple rounds of pre-assembly, restore the actual pre-assembly error of the common segment of the previous round through coordinate transformation relationship, so that the line shape error of the common segment of the next round is consistent with that of the previous round, and obtain the point cloud of the target line shape of the pre-assembly target line shape of the whole round. S4, extract the axis features of the target linear point cloud of the full round of pre-assembly to obtain the center line of the arch rib; calculate the three-dimensional coordinates of the frame according to the center line, and place the arch rib according to the three-dimensional coordinates of the frame using the adjustment bracket to obtain the coarsely adjusted arch rib posture data. S5, use the laser scanner to perform a second scan on the coarsely adjusted arch rib to obtain actual posture data; generate target posture data based on the pre-assembled target linear point cloud, calculate the transformation parameters between the actual posture data and the target posture data, solve the elongation of the three-dimensional jack in each direction, and obtain the finely adjusted arch rib posture. S6. Based on the finely adjusted arch rib posture, install the flange adjustment wall, fix the flange to the vertical track, and complete the welding after maintaining the vertical posture to obtain the pre-assembled arch rib. Repeat S1-S5 and use the line error consistency control strategy of the common segment to achieve precise connection of the arch rib line of the whole bridge.

[0007] Preferably, the process of obtaining the preprocessed point cloud containing six-dimensional information including position and normal vector specifically includes adjusting the distance from the acquisition device to the arch rib and the sparsity of the point cloud, acquiring the effective nearest neighbors of each point, and processing each point... Given the set of effective nearest neighbors, calculate its three-dimensional centroid. Then construct the covariance matrix. Eigenvalue decomposition of the covariance matrix C yields three eigenvalues. and the corresponding feature vectors , , Take the smallest eigenvalue Corresponding feature vector , as a point normal vector Integrate the 3D position information (x, y, z) and normal vector information of each point. This forms a preprocessed point cloud containing six-dimensional information.

[0008] Preferably, performing multi-site point cloud stitching on the preprocessed point cloud specifically includes: extracting a subset of the point cloud from each rotating triangular pyramid target. The random sample consensus algorithm is used to... Perform repeated random sampling at 3 points, and calculate the plane equation after each sampling. parameters The probability density distribution of each parameter is statistically analyzed, and the peak value of the probability density is taken as the reliable parameter of the cone surface; each target corresponds to 3 sets of planar parameters. Given j=1,2,3, process the point cloud subsets of all targets at all stations; calculate the unique intersection point for the three sets of cone surface parameters of each target to obtain the three-dimensional coordinates of the target feature points at each station; the target feature point sets of adjacent stations A and B are respectively and k≥2; The fast four-point consensus set algorithm is used to analyze the target feature point set. and Perform coarse registration to obtain the initial rotation matrix. and the initial translation matrix ; Constructing an error function using the ICP algorithm The constraints are determinant ,in, For rotation matrix, Given a translation matrix, find the optimal rotation matrix that minimizes the error function through iterative optimization. Translation matrix Repeat this process to complete the multi-site cloud splicing.

[0009] Preferably, obtaining the aligned point cloud model specifically includes calculating the triple-weighted centroids of the stitched point cloud and the theoretical point cloud respectively: Weighted centroid of the stitched point cloud ; Theoretical point cloud weighted centroid ; Centralized processing ; in, Let N be the 3D coordinates of the i-th point in the stitched point cloud, and N be the total number of points in the stitched point cloud. The overall weight for a single point; Let M be the 3D coordinates of the j-th point in the theoretical point cloud, and M be the total number of points in the theoretical point cloud. For the comprehensive weight of individual points in the theoretical point cloud, Let i be the semantic weight of the component containing the i-th point. Let be the distribution weight of the i-th point. The connection weights for the i-th point are... The coordinates of the decentralized, stitched point cloud. The coordinates of the decentralized theoretical point cloud; Construct a triple-weighted covariance matrix ,in, The matching point set between the theoretical point cloud and the stitched point cloud; singular value decomposition of M yields... Where U and V are orthogonal matrices, Given a diagonal matrix; solve for the optimal rotation matrix using the least squares method. Translation matrix The stitched point cloud is then processed through... A rigid transformation is performed, and then an energy function is used for global optimal alignment to obtain an aligned point cloud model.

[0010] Preferably, the calculation of the optimal pre-stitched target alignment that minimizes the deviation between the conventional segment and the theoretical alignment specifically includes: based on the aligned point cloud model, extracting the point cloud data of the conventional segment, combining it with the theoretical BIM point cloud model, calculating the point-to-point Euclidean distance between the point cloud after the conventional segment stitching and the theoretical point cloud, and defining a deviation function. ;in, The coordinates of the stitched point cloud are aligned. Let ΔL be the theoretical point cloud coordinates and N be the total number of point clouds. By minimizing ΔL using the least squares method, the optimal pre-assembly target pose of the regular segment is obtained, and this pose is taken as the optimal pre-assembly target line shape.

[0011] Preferably, obtaining the centerline of the arch rib specifically includes: adjusting the spatial attitude of the pre-assembled target linear point cloud for all rounds, obtaining the projection slope of the arch rib on the xy plane through linear fitting, and calculating the angle with the x-axis. Rotate clockwise around the Z-axis This makes the arch rib point cloud parallel to the XZ plane, where, The slope of the linear fit; Constructing the axis fitting equation for the point cloud of the main chord of the arch rib , The parameters to be fitted are: [parameters]; with step size [...]. Divide the point cloud into m segments to obtain a curved cylindrical point cloud with a step size of d; where... The maximum coordinate value of the linear point cloud of the arch rib pre-assembly target in the x-axis direction. The minimum coordinate value of the target linear point cloud in the x-axis direction for pre-assembling the arch rib; Project the point cloud of each curved cylindrical segment onto the xy plane, and fit the projected circle using the iterative least squares method to solve for the coordinates of the circle center. The three-dimensional coordinates of the micro-segment control points are calculated by combining the z-coordinates of the point cloud. All micro-segment control points are restored to their original spatial positions according to the scanning sequence, and the complete arch rib centerline is generated by cubic spline interpolation.

[0012] Preferably, the calculation of the elongation of the three-dimensional jack in each direction specifically includes: rotating the matrix based on the transformation parameters. and transformation parameter translation matrix A mapping relationship between the three-dimensional jacks and the arch rib posture was established. Each adjustment bracket was equipped with four three-dimensional jacks, and the elongation per unit X, Y, and Z directions was calculated. Influence Matrix and the influence matrix on T' ; Construct a system of equations for the elongation of the jack ; in, , , These represent the longitudinal, transverse, and vertical elongation of jack n, respectively. , The precise elongation of each jack is obtained by solving the problem.

[0013] Preferably, obtaining the pre-assembled arch rib specifically includes: using the flange welding area as the constraint boundary, creating a three-dimensional mesh state space along the flange circumference and radial direction, dividing the circumference into angle intervals, and the radial direction into distance intervals, combined with the welding torch height, to form discrete states, each state using a triplet. express, Let r be the circumferential angle, r be the radial distance, and h be the welding torch height; the action space includes 5 types of welding torch operations: circumferential clockwise movement, circumferential counterclockwise movement, radial feed, radial retraction, and welding torch height fine-tuning; initialize the Q-value table as a zero matrix of discrete states × welding torch operations; set the learning rate. Discount rate Greed value Design a multi-objective weighted reward function: ;in, To constrain the satisfaction of rewards, To ensure uniformity of the weld, As an efficiency bonus; taking the initial welding point of the flange as the initial state and the endpoint as the overlapping state after one circumferential revolution of the initial state, according to... -Greedy strategy selects action: with Probably select the action with the highest current Q value, in order to Probabilistic random selection of actions to explore new paths; after each action, switching to a new state and calculating an immediate reward based on a reward function. and according to the formula Update the Q-value table; where, This represents the dynamic scene state at time t during the welding process. For the dynamic scene state based on time t The welding torch operation instructions are executed; when the preset iteration threshold is reached, the optimal action for each state in the Q-value matrix is ​​extracted to form a continuous welding path.

[0014] As can be seen from the above technical solution, compared with the prior art, the present invention discloses a digital pre-assembly manufacturing control method for long-span arch bridges. Through the collaborative weighting and weighted alignment calculation of semantic, distribution, and connection relationship weights, it not only strengthens the alignment priority of the core components of the arch rib, dense point cloud, and connection nodes, but also effectively eliminates local matching deviations and cumulative errors through decentralized processing, SVD optimal transformation, and improved global graph cut optimization. This significantly improves the alignment accuracy between the actual point cloud and the theoretical BIM point cloud, provides high-fidelity data support for the calibration of the baseline shape of common segments and the calculation of personalized pre-assembly targets for conventional segments, greatly reduces the construction error and rework risk of arch rib segment splicing, and is suitable for engineering-level high-precision pre-assembly requirements. Attached Figure Description

[0015] 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 embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0016] Figure 1 The method steps provided by this invention are illustrated. Detailed Implementation

[0017] 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.

[0018] This invention discloses a digital pre-assembly manufacturing control method for long-span arch bridges, such as... Figure 1 As shown, it includes: S1. Plan the scanning stations, set up rotating triangular pyramid targets, and collect point cloud data of each station on the arch rib using a laser scanner; preprocess the point cloud data, calculate the covariance matrix of each point, take the eigenvector corresponding to the smallest eigenvalue as the normal vector, and obtain a preprocessed point cloud containing six-dimensional information of position and normal vector. S2. Based on the random sampling consensus algorithm, fit the equations of the three cone surfaces of the rotating triangular pyramid target, calculate the target feature point set, and perform multi-site point cloud stitching on the preprocessed point cloud by coarse registration using the fast four-point consensus set algorithm combined with the ICP algorithm for fine registration. According to the component semantic weight, point cloud distribution weight, and connection relationship weight, perform weight matching and alignment with the theoretical BIM model to obtain the aligned point cloud model. S3, based on the aligned point cloud model, calculates the optimal pre-assembly target line shape with the smallest deviation between the regular segment and the theoretical line shape; for the common segment of multiple rounds of pre-assembly, the actual pre-assembly error of the previous round of common segment is restored through coordinate transformation relationship, so that the line shape error of the common segment of the next round is consistent with that of the previous round, and the point cloud of the target line shape of the whole round of pre-assembly is obtained. S4. Extract the axis features of the target linear point cloud of the whole round of pre-assembly to obtain the center line of the arch rib; calculate the three-dimensional coordinates of the frame according to the center line; use the adjustment bracket to place the arch rib according to the three-dimensional coordinates of the frame and obtain the coarsely adjusted arch rib attitude data. S5. Use a laser scanner to perform a second scan on the coarsely adjusted arch rib to obtain actual posture data; generate target posture data based on the pre-stitched target linear point cloud, calculate the transformation parameters between the actual posture data and the target posture data, solve the elongation of the three-dimensional jack in each direction, and obtain the finely adjusted arch rib posture. S6. Based on the finely adjusted arch rib posture, install the flange adjustment wall, fix the flange to the vertical track, and complete the welding after maintaining the vertical posture to obtain the pre-assembled arch rib. Repeat S1-S5 and use the line error consistency control strategy of common segments to achieve precise connection of the arch rib line of the whole bridge.

[0019] Common segments are the benchmark segments for arch rib splicing, requiring the highest precision. They serve as a unified reference for the splicing line shape of all segments and are extremely rare. Regular segments are the main splicing segments of the arch rib, custom-made and prefabricated segment by segment based on the common segments. They constitute the vast majority of the segments and form the main structure of the arch rib.

[0020] In one specific embodiment, obtaining a preprocessed point cloud containing six-dimensional information including position and normal vectors specifically includes adjusting the point cloud based on the distance from the acquisition device to the arch rib and the sparsity of the point cloud, acquiring the effective nearest neighbors of each point, and processing each point... Given the set of effective nearest neighbors, calculate its three-dimensional centroid. Then construct the covariance matrix. Eigenvalue decomposition of the covariance matrix C yields three eigenvalues. and the corresponding feature vectors , , Take the smallest eigenvalue Corresponding feature vector , as a point normal vector Integrate the 3D position information (x, y, z) and normal vector information of each point. This forms a preprocessed point cloud containing six-dimensional information.

[0021] In one specific embodiment, performing multi-site point cloud stitching on the preprocessed point cloud specifically includes: extracting a subset of the point cloud for each rotating triangular pyramid target. The random sample consensus algorithm is used to... Perform repeated random sampling at 3 points, and calculate the plane equation after each sampling. parameters The probability density distribution of each parameter is statistically analyzed, and the peak value of the probability density is taken as the reliable parameter of the cone surface; each target corresponds to 3 sets of planar parameters. Given j=1,2,3, process the point cloud subsets of all targets at all stations; calculate the unique intersection point for the three sets of cone surface parameters of each target to obtain the three-dimensional coordinates of the target feature points at each station; the target feature point sets of adjacent stations A and B are respectively and k≥2; The fast four-point consensus set algorithm is used to analyze the target feature point set. and Perform coarse registration to obtain the initial rotation matrix. and the initial translation matrix ; Constructing an error function using the ICP algorithm The constraints are determinant ,in, For rotation matrix, Given a translation matrix, find the optimal rotation matrix that minimizes the error function through iterative optimization. Translation matrix Repeat this process to complete the multi-site cloud splicing.

[0022] In the initial stage of S2, semantic segmentation of the stitched point cloud and theoretical point cloud was completed. Based on the structural importance of the arch rib and the high-precision requirements of pre-assembly, fixed weights were assigned to point clouds of different component types to ensure that the alignment of core component point clouds had the highest priority. Local density statistics of the stitched point cloud and theoretical point cloud were also completed in the initial stage of S2. The point cloud was divided into dense, uniform, and sparse regions according to density, and weights were dynamically assigned to avoid interference from scanning blind spots and sparse point clouds at the edges. Based on the semantic segmentation results of S2, the inherent component connection relationships of the arch rib were extracted, and the correlation between the point cloud location and the designed connection nodes was calculated. Weights were assigned to these connection relationships to ensure that point clouds conforming to the design connection logic had a higher proportion in the alignment. The calculation rules are as follows: Connectivity node extraction: Extract the set of inherent connectivity nodes for the arch rib design from the S2 semantic segmentation results. For any point in the stitched point cloud Calculate its distance to the nearest connected node Euclidean distance Define the degree of correlation ,in The influence radius of the connecting nodes is determined; if the radius exceeds this radius, the correlation degree is 0. Weight conversion , This serves as a basic safety net coefficient for the weighting of connection relationships. This is the correlation adjustment coefficient for the weight of the connection relationship. That is, connecting points near the node. but ; In one specific embodiment, obtaining the aligned point cloud model specifically includes calculating the triple-weighted centroids of both the stitched point cloud and the theoretical point cloud: Weighted centroid of the stitched point cloud ; Theoretical point cloud weighted centroid ; Centralized processing ; in, Let N be the 3D coordinates of the i-th point in the stitched point cloud, and N be the total number of points in the stitched point cloud. The overall weight for a single point; Let M be the 3D coordinates of the j-th point in the theoretical point cloud, and M be the total number of points in the theoretical point cloud. For the comprehensive weight of individual points in the theoretical point cloud, Let i be the semantic weight of the component containing the i-th point. Let be the distribution weight of the i-th point. The connection weights for the i-th point are... The coordinates of the decentralized, stitched point cloud. The coordinates of the decentralized theoretical point cloud; Construct a triple-weighted covariance matrix ,in, The matching point set between the theoretical point cloud and the stitched point cloud; singular value decomposition of M yields... Where U and V are orthogonal matrices, Given a diagonal matrix; solve for the optimal rotation matrix using the least squares method. Translation matrix The stitched point cloud is then processed through... A rigid transformation is performed, and then an energy function is used for global optimal alignment to obtain an aligned point cloud model.

[0023] A triple-weighted energy function is constructed, and global optimal alignment is achieved by minimizing the energy function, thus eliminating local biases. The specific steps are as follows: The initially aligned and stitched point cloud and the theoretical point cloud are regarded as an undirected graph G=(V, E), where the point set V is all the matching points of the two sets of point clouds, and the edge set E is the point cloud neighborhood connection constructed based on kd-tree, ensuring that the point cloud in the neighborhood is spatially continuous.

[0024] The energy function includes a univariate term (point cloud residual penalty) and a binary term (neighborhood point consistency penalty), both incorporating triple weights to avoid overfitting and interference from noise points. The function form is as follows: ; in, This is a unary term that measures the actual alignment residual of a single point, incorporating connection relationship weights. Points with higher connection rationality have lower residual penalty weights. The formula is as follows: ,in For the initial alignment point coordinates, These are the coordinates of the theoretical point; It is a binary term that measures the alignment consistency of neighboring points. The Potts model is used, and the indicator function is... The value is 1 if the neighboring point labels are inconsistent, and 0 if they are consistent; the value is also incorporated into the product of the comprehensive weights of the neighboring points. The consistency penalty in the core point cloud neighborhood is higher; The smoothing coefficient is a fixed penalty coefficient that balances the contributions of univariate and binary terms.

[0025] In one specific embodiment, calculating the optimal pre-assembly target alignment that minimizes the deviation between the conventional segment and the theoretical alignment specifically includes: extracting the point cloud data of the conventional segment based on the aligned point cloud model, combining it with the theoretical BIM point cloud model, calculating the point-to-point Euclidean distance between the point cloud after the conventional segment is assembled and the theoretical point cloud, and defining the deviation function. ;in, The coordinates of the stitched point cloud are aligned. Let ΔL be the theoretical point cloud coordinates and N be the total number of point clouds. By minimizing ΔL using the least squares method, the optimal pre-assembly target pose of the regular segment is obtained, and this pose is taken as the optimal pre-assembly target line shape.

[0026] In one specific embodiment, obtaining the centerline of the arch rib includes: adjusting the spatial attitude of the pre-assembled target linear point cloud for all rounds, obtaining the projection slope of the arch rib on the xy plane through linear fitting, and calculating the angle with the x-axis. Rotate clockwise around the Z-axis This makes the arch rib point cloud parallel to the XZ plane, where, The slope of the linear fit; Constructing the axis fitting equation for the point cloud of the main chord of the arch rib , The parameters to be fitted are: [parameters]; with step size [...]. Divide the point cloud into m segments to obtain a curved cylindrical point cloud with a step size of d; where... The maximum coordinate value of the linear point cloud of the arch rib pre-assembly target in the x-axis direction. The minimum coordinate value of the target linear point cloud in the x-axis direction for pre-assembling the arch rib; Project the point cloud of each curved cylindrical segment onto the xy plane, and fit the projected circle using the iterative least squares method to solve for the coordinates of the circle center. The three-dimensional coordinates of the micro-segment control points are calculated by combining the z-coordinates of the point cloud. All micro-segment control points are restored to their original spatial positions according to the scanning sequence, and the complete arch rib centerline is generated by cubic spline interpolation.

[0027] In one specific embodiment, solving for the elongation of the three-dimensional jack in each direction specifically includes: rotating the matrix based on the transformation parameters. and transformation parameter translation matrix A mapping relationship between the three-dimensional jacks and the arch rib posture was established. Each adjustment bracket was equipped with four three-dimensional jacks, and the elongation per unit X, Y, and Z directions was calculated. Influence Matrix and the influence matrix on T' ; Construct a system of equations for the elongation of the jack ; in, , , These represent the longitudinal, transverse, and vertical elongation of jack n, respectively. , The precise elongation of each jack is obtained by solving the problem.

[0028] In one specific embodiment, obtaining the pre-assembled arch rib product specifically includes: using the flange welding area as the constraint boundary, creating a three-dimensional mesh state space along the flange circumference and radial direction. The circumference is divided into angle intervals, and the radial direction is divided into distance intervals. Combined with the welding torch height, discrete states are formed. Each state is represented by a triplet. express, Let r be the circumferential angle, r be the radial distance, and h be the welding torch height; the action space includes 5 types of welding torch operations: circumferential clockwise movement, circumferential counterclockwise movement, radial feed, radial retraction, and welding torch height fine-tuning; initialize the Q-value table as a zero matrix of discrete states × welding torch operations; set the learning rate. Discount rate Greed value Design a multi-objective weighted reward function: ;in, To constrain the satisfaction of rewards, To ensure uniformity of the weld, As an efficiency bonus; taking the initial welding point of the flange as the initial state and the endpoint as the overlapping state after one circumferential revolution of the initial state, according to... -Greedy strategy selects action: with Probably select the action with the highest current Q value, in order to Probabilistic random selection of actions to explore new paths; after each action, switching to a new state and calculating an immediate reward based on a reward function. and according to the formula Update the Q-value table; where, This represents the dynamic scene state at time t during the welding process. For the dynamic scene state based on time t The welding torch operation instructions are executed; when the preset iteration threshold is reached, the optimal action for each state in the Q-value matrix is ​​extracted to form a continuous welding path.

[0029] In Example 1

[0030] In S1, based on the spatial distribution of the arch rib segments, the scanning stations for each arch rib segment are divided into four basic stations: the left station at the small mileage, the right station at the small mileage, the left station at the large mileage, and the right station at the large mileage. All stations are located on the side away from the beam to avoid the beam obstructing the scanning line of sight. The straight-line distance dz between the control station and the corner point of the arch rib is controlled to satisfy 10m ≤ dz ≤ 20m to ensure scanning accuracy and coverage. The angle α between the line connecting the station and the corner point of the arch rib and the direction of the arch rib axis is strictly controlled to be 40° ≤ α ≤ 60° to ensure uniform acquisition of point cloud data on the surface of the arch rib and reduce scanning blind spots.

[0031] The distance parameters for the laser scanner were set to 120m, and the target object resolution was set to 3.1mm / 10m to ensure that high-precision point cloud data could still be acquired when scanning at a long distance.

[0032] The rotating triangular pyramid target consists of a base, a rotatable support, and a triangular pyramid body. The unique intersection of the three planes of the pyramid is the target feature point, located on the rotation axis of the rotating support, which is perpendicular to the base. Flexible alignment with the scanner can be achieved by rotating the pyramid surface. At least two rotating triangular pyramid targets should be placed between two adjacent scanning stations, and the targets should be positioned where they are visible from both adjacent stations. The axial projection dy of the target relative to the center of the arch rib should satisfy 10m ≤ dy ≤ 25m, and the lateral projection dx should satisfy 5m ≤ dx ≤ 10m. Furthermore, the sum of the lateral projections of adjacent targets, dx1 + dx2 ≥ 8m, ensures accurate identification of the target feature point. Before scanning, adjust the target's rotating support so that one facet of the pyramid faces the scanner, reducing the impact of scanning noise on feature point extraction.

[0033] Start the laser scanner and scan the arch ribs sequentially according to the preset station order. During each scan, ensure that the target and the entire surface of the arch rib segment are fully covered without significant obstruction. Record the acquisition time, target position, and other relevant information for each station during the scanning process to facilitate subsequent point cloud stitching. After acquisition, export the raw point cloud data for each station. The data format includes the three-dimensional coordinates (x, y, z) of each point and the scan intensity information.

[0034] S2 collects design data for long-span arch bridges, including the structural dimensions of the arch ribs (section dimensions, segment lengths, flange positions, etc.), material parameters (elastic modulus, density, etc.), and construction specifications. Based on BIM modeling software, a theoretical BIM model is established according to the vertical orientation of the flange. The model must contain the precise geometric information of all components, including the main chord, web members, and horizontal bracing.

[0035] Import the theoretical BIM model into the point cloud processing software, set the sampling density, and perform high-density sampling on the model. The point cloud data format includes the three-dimensional coordinates (x, y, z) of each point to ensure that the density magnitude of the theoretical point cloud is consistent with that of the actual scanned point cloud, which facilitates subsequent matching and alignment.

[0036] The calculation of the optimal pre-assembly target alignment that minimizes the deviation between the conventional segment and the theoretical alignment in S3 specifically includes: extracting the point cloud data of the conventional segment based on the aligned point cloud model, combining it with the theoretical BIM point cloud model, calculating the point-to-point Euclidean distance between the point cloud after the conventional segment is assembled and the theoretical point cloud, and defining the deviation function. ;in, The coordinates of the stitched point cloud are aligned. Let N be the theoretical point cloud coordinates and N be the total number of point clouds. By minimizing ΔL using the least squares method, the optimal pre-assembly target pose of the regular segment is obtained and used as the optimal pre-assembly target line shape. Extract the pre-assembly data of the common segments from the previous round, including: the transformation parameters (rotation matrix R) between the theoretical coordinate system P1 and the overall theoretical coordinate system P0 from the previous round. 01 Translation matrix t 01 ), the actual pre-assembled alignment of the common segment D1 (including the first-end error δ) 11 Tail end error δ 12 This data comes from the calculation results of step 3 in the previous round and the on-site measurement records.

[0037] Derivation of coordinate transformation relationship: Obtain the transformation parameters (rotation matrix R) between the next round theoretical coordinate system P2 and the overall theoretical coordinate system P0. 02 Translation matrix t 02 Based on the transitivity of coordinate transformation, the transformation relationship from the previous coordinate system P1 to the next coordinate system P2 is derived: Rotation matrix ; Translation matrix ; Substitute the actual pre-assembled line shape D1 of the previous round of common segments into the conversion formula. This yields the pre-assembled target alignment D2 for the next round of common segments, ensuring the accuracy of the initial error of D2. Tail end error This enables consistent continuity of linear errors across multiple rounds.

[0038] Integrate the optimal pre-segmented target alignment of regular segments with the target alignment of common segment connections to construct a full-round pre-segmented target point cloud model, maintaining the point cloud density consistent with the theoretical point cloud density in step 2. Calculate the axis smoothness and cross-sectional alignment at the connection points of adjacent segments; if not satisfied, re-optimize the transformation parameters of common segments or the target pose of regular segments until the alignment is continuously smooth throughout the entire round. Finally, output the full-round pre-segmented target alignment point cloud, providing basic data for axis feature extraction in S4.

[0039] In S4, based on the length and stiffness distribution of the arch rib, a formwork is arranged every 3-5m along the centerline. Each formwork corresponds to two three-dimensional jacks, symmetrically arranged on both sides of the arch rib. The lateral spacing QY of the jacks is equal to the axial distance D between the centers of the lower chord tubes of the arch rib in the pre-assembly coordinate system. 实 .

[0040] Taking adjustment bracket 1 as an example, after determining its mileage position, the three-dimensional coordinates (x1, y1, z1) of the center line of the arch rib at that mileage are extracted. 1轴 Based on the diameter D1 of the arch rib section, calculate the coordinates (x1, y1, z) of the contact point between the jack and the arch rib. 1轴 -D1 / 2); Vertical height of the jack QZ=z 1轴 -D1 / 2-250mm-H1; H1 is the fixed height of the support frame.

[0041] Import all three-dimensional coordinates of the jigs into the BIM software and perform collision detection with the target linear point cloud of the full-round pre-assembly of S3. If the deviation between the jig and the arch rib contact point is >2mm, readjust the interpolation accuracy of the control points or the jig arrangement spacing until the requirements are met.

[0042] Install and adjust the support frame according to the calculated three-dimensional coordinates of the jig. The support frame consists of a support frame, a three-dimensional jack, and a crescent plate. The spacing between the bases of the support frame is consistent with the center distance of the arch rib segments and is fixed to the ground by the column feet. The radius of the crescent plate matches the radius of the steel pipe of the arch rib to be assembled, ensuring a tight fit with the arch rib.

[0043] The arch rib segment is lifted to the top of the jig using a gantry crane, initially aligned according to the center line markings, and slowly lowered onto the crescent plate. The horizontal position is adjusted based on QY and the vertical height is adjusted based on QZ using three-dimensional jacks, so that the deviation between the center line of the arch rib and the planned center line of the jig is ≤5mm.

[0044] The three-dimensional coordinates of the key sections of the arch rib (both ends, midpoint and frame position) are measured using a laser rangefinder. If the preset conditions are not met, the elongation of the corresponding jacks is readjusted until the coarse adjustment accuracy requirements are met, and the coarse adjustment arch rib posture data is obtained.

[0045] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. The methods disclosed in the embodiments are described simply because they correspond to the methods disclosed in the embodiments; relevant parts can be found in the method section.

[0046] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A digital pre-assembly manufacturing control method for long-span arch bridges, characterized in that, include: S1, plan the scanning stations, set up rotating triangular pyramid targets, and collect point cloud data of each station on the arch rib using a laser scanner; The point cloud data is preprocessed to calculate the covariance matrix of each point, and the eigenvector corresponding to the smallest eigenvalue is taken as the normal vector to obtain a preprocessed point cloud containing six-dimensional information of position and normal vector. S2, based on the random sampling consensus algorithm, fit the equations of the three cone surfaces of the rotating triangular pyramid target, calculate the target feature point set, and perform multi-site point cloud stitching on the preprocessed point cloud by coarse registration using the fast four-point consensus set algorithm combined with the ICP algorithm for fine registration; according to the component semantic weight, point cloud distribution weight, and connection relationship weight, perform weight matching and alignment with the theoretical BIM model to obtain the aligned point cloud model; S3. Based on the aligned point cloud model, calculate the optimal pre-assembly target line shape with the smallest deviation between the regular segment and the theoretical line shape; for the common segment of multiple rounds of pre-assembly, restore the actual pre-assembly error of the common segment of the previous round through coordinate transformation relationship, so that the line shape error of the common segment of the next round is consistent with that of the previous round, and obtain the point cloud of the target line shape of the pre-assembly target line shape of the whole round. S4, extract the axis features of the target linear point cloud of the full round of pre-assembly to obtain the center line of the arch rib; calculate the three-dimensional coordinates of the frame according to the center line, and place the arch rib according to the three-dimensional coordinates of the frame using the adjustment bracket to obtain the coarsely adjusted arch rib posture data. S5, use the laser scanner to perform a second scan on the coarsely adjusted arch rib to obtain actual posture data; generate target posture data based on the pre-assembled target linear point cloud, calculate the transformation parameters between the actual posture data and the target posture data, solve the elongation of the three-dimensional jack in each direction, and obtain the finely adjusted arch rib posture. S6. Based on the finely adjusted arch rib posture, install the flange adjustment wall, fix the flange to the vertical track, and complete the welding after maintaining the vertical posture to obtain the pre-assembled arch rib. Repeat S1-S5 and use the line error consistency control strategy of the common segment to achieve precise connection of the arch rib line of the whole bridge.

2. The digital pre-assembly manufacturing control method for a long-span arch bridge according to claim 1, characterized in that, The specific steps for obtaining the preprocessed point cloud containing six-dimensional information including position and normal vectors include adjusting the distance from the acquisition device to the arch rib and the sparsity of the point cloud, acquiring the effective nearest neighbors of each point, and processing each point... Given the set of effective nearest neighbors, calculate its three-dimensional centroid. Then construct the covariance matrix. ; Eigenvalue decomposition of the covariance matrix C yields three eigenvalues. and the corresponding feature vectors , , Take the smallest eigenvalue corresponding feature vector , as a point normal vector Integrate the 3D position information (x, y, z) and normal vector information of each point. This forms a preprocessed point cloud containing six-dimensional information.

3. The digital pre-assembly manufacturing control method for a long-span arch bridge according to claim 1, characterized in that, The process of stitching together the preprocessed point cloud into multiple points specifically includes: extracting a subset of the point cloud from each rotating triangular pyramid target. The random sample consensus algorithm is used to... Perform repeated random sampling at 3 points, and calculate the plane equation after each sampling. parameters The probability density distribution of each parameter is statistically analyzed, and the peak value of the probability density is taken as the reliable parameter of the cone surface; each target corresponds to 3 sets of planar parameters. Given j=1,2,3, process the point cloud subsets of all targets at all stations; calculate the unique intersection point for the three sets of cone surface parameters of each target to obtain the three-dimensional coordinates of the target feature points at each station; the target feature point sets of adjacent stations A and B are respectively and k≥2; The fast four-point consensus set algorithm is used to analyze the target feature point set. and Perform coarse registration to obtain the initial rotation matrix. and the initial translation matrix ; Constructing an error function using the ICP algorithm The constraints are determinant ,in, For rotation matrix, Given a translation matrix, find the optimal rotation matrix that minimizes the error function through iterative optimization. Translation matrix Repeat this process to complete the multi-site cloud splicing.

4. The digital pre-assembly manufacturing control method for a long-span arch bridge according to claim 1, characterized in that, The acquisition of the aligned point cloud model specifically includes calculating the triple-weighted centroids for both the stitched point cloud and the theoretical point cloud: Weighted centroid of the stitched point cloud ; Theoretical point cloud weighted centroid ; Centralized processing ; in, Let represent the 3D coordinates of the i-th point in the stitched point cloud, and N represent the total number of points in the stitched point cloud. The overall weight for a single point; Let M be the 3D coordinates of the j-th point in the theoretical point cloud, and M be the total number of points in the theoretical point cloud. For the comprehensive weight of individual points in the theoretical point cloud, Let i be the semantic weight of the component containing the i-th point. Let be the distribution weight of the i-th point. The connection weights for the i-th point are... The coordinates of the decentralized, stitched point cloud. The coordinates of the decentralized theoretical point cloud; Construct a triple-weighted covariance matrix ,in, The matching point set between the theoretical point cloud and the stitched point cloud; singular value decomposition of M yields... Where U and V are orthogonal matrices, Given a diagonal matrix; solve for the optimal rotation matrix using the least squares method. Translation matrix The stitched point cloud is then processed through... A rigid transformation is performed, and then an energy function is used for global optimal alignment to obtain an aligned point cloud model.

5. The digital pre-assembly manufacturing control method for a long-span arch bridge according to claim 4, characterized in that, The calculation of the optimal pre-stitched target alignment that minimizes the deviation between the conventional segment and the theoretical alignment specifically includes: based on the aligned point cloud model, extracting the point cloud data of the conventional segment, combining it with the theoretical BIM point cloud model, calculating the point-to-point Euclidean distance between the point cloud after the conventional segment stitching and the theoretical point cloud, and defining a deviation function. ;in, The coordinates of the stitched point cloud are aligned. Let ΔL be the theoretical point cloud coordinates and N be the total number of point clouds. By minimizing ΔL using the least squares method, the optimal pre-assembly target pose of the regular segment is obtained, and this pose is taken as the optimal pre-assembly target line shape.

6. The digital pre-assembly manufacturing control method for a long-span arch bridge according to claim 1, characterized in that, The specific steps for obtaining the centerline of the arch rib include: adjusting the spatial attitude of the pre-assembled target linear point cloud from all rounds of assembly, obtaining the projection slope of the arch rib on the xy plane through linear fitting, and calculating the angle with the x-axis. Rotate clockwise around the Z-axis This makes the arch rib point cloud parallel to the XZ plane, where, The slope of the linear fit; Constructing the axis fitting equation for the point cloud of the main chord of the arch rib , The parameters to be fitted are: [parameters]; with step size [...]. Divide the point cloud into m segments to obtain a curved cylindrical point cloud with a step size of d; where... The maximum coordinate value of the linear point cloud of the arch rib pre-assembly target in the x-axis direction. The minimum coordinate value of the target linear point cloud in the x-axis direction for pre-assembling the arch rib; Project the point cloud of each curved cylindrical segment onto the xy plane, and fit the projected circle using the iterative least squares method to solve for the coordinates of the circle center. The three-dimensional coordinates of the micro-segment control points are calculated by combining the z-coordinates of the point cloud. All micro-segment control points are restored to their original spatial positions according to the scanning sequence, and the complete arch rib centerline is generated by cubic spline interpolation.

7. The digital pre-assembly manufacturing control method for a long-span arch bridge according to claim 1, characterized in that, The specific steps for calculating the elongation of the three-dimensional jack in each direction include: based on the rotation matrix of the transformation parameters. and transformation parameter translation matrix A mapping relationship between the three-dimensional jacks and the arch rib posture was established. Each adjustment bracket was equipped with four three-dimensional jacks, and the elongation per unit X, Y, and Z directions was calculated. Influence Matrix and the influence matrix on T' ; Construct a system of equations for the elongation of the jack ; in, , , These represent the longitudinal, transverse, and vertical elongation of jack n, respectively. , The precise elongation of each jack is obtained by solving the problem.

8. The digital pre-assembly manufacturing control method for a long-span arch bridge according to claim 1, characterized in that, The process of obtaining the pre-assembled arch rib specifically includes: using the flange welding area as the constraint boundary, creating a three-dimensional mesh state space along the flange's circumference and radial direction. The circumference is divided into angle intervals, and the radial direction is divided into distance intervals. Combined with the welding torch height, this forms discrete states, each state represented by a triplet. express, Let r be the circumferential angle, r be the radial distance, and h be the welding torch height; the action space includes 5 types of welding torch operations: circumferential clockwise movement, circumferential counterclockwise movement, radial feed, radial retraction, and welding torch height fine-tuning; initialize the Q-value table as a zero matrix of discrete states × welding torch operations; set the learning rate. Discount rate Greed value Design a multi-objective weighted reward function: ;in, To constrain the satisfaction of rewards, To ensure uniformity of the weld, As an efficiency bonus; taking the initial welding point of the flange as the initial state and the endpoint as the overlapping state after one circumferential revolution of the initial state, according to... -Greedy strategy selects action: with Probably select the action with the highest current Q value, in order to Probabilistic random selection of actions to explore new paths; after each action, switching to a new state and calculating an immediate reward based on a reward function. and according to the formula Update the Q-value table; where, This represents the dynamic scene state at time t during the welding process. For the dynamic scene state based on time t The welding torch operation instructions are executed; when the preset iteration threshold is reached, the optimal action for each state in the Q-value matrix is ​​extracted to form a continuous welding path.

Images