Virtual pre-assembly quality evaluation method for assembled bridge based on laser point cloud data

Abstract

The application discloses a kind of virtual pre-assembly quality evaluation methods of assembly type bridge based on laser point cloud data.The method steps:1. carry out scanning scheme design, clear component parameter and precision requirement, survey site and lay out scanning station and magnetic attraction type target, optimize scanning resolution;2. adopt ground three-dimensional laser scanner to collect bridge component point cloud data, record key information simultaneously and complete data backup;3. complete point cloud preprocessing by filtering denoising, voxel grid method extraction, clustering algorithm block;4. with target as constraint, construct three-dimensional grid model and simulate field condition to carry out virtual pre-assembly, realize the quantitative calculation of geometric error by feature point comparison and section analysis;5. based on relevant specification, construct quality evaluation index system, complete index fusion combined with cloud theory, obtain overall quality evaluation result and output visual evaluation report.The application does not need to carry out physical test, can effectively save site resource, shorten pre-assembly period.

Description

Technical Field

[0001] This invention relates to the field of bridge engineering inspection and quality assessment technology. Specifically, it is a method for virtual pre-assembly and quality assessment of prefabricated bridges based on laser point cloud data. It is applicable to the construction and acceptance stage of steel structure bridges such as prefabricated steel truss bridges and steel box girder bridges. It can realize high-precision virtual pre-assembly and full-dimensional quality assessment of bridge components, providing technical support for bridge construction quality control and full life cycle performance management. Background Technology

[0002] Prefabricated bridges, with their advantages of factory-produced components and rapid on-site assembly, are widely used in transportation infrastructure construction. Their construction quality directly determines the structural load-bearing capacity, service durability, and operational safety. Traditional prefabricated bridge pre-assembly relies on physical trial assembly, which requires a large amount of space, has high labor costs, and a long construction period, making it difficult to meet the pre-assembly requirements of large and complex bridge components. At the same time, geometric accuracy testing relies on discrete point measurement equipment such as total stations, which is inefficient and cannot capture the overall shape of the components, easily overlooking cumulative errors such as segmental assembly misalignment and alignment deviation. Quality assessment often uses deterministic evaluation methods, which are difficult to quantify the uncertainties brought about by factors such as measurement noise and environmental interference, resulting in highly subjective and inaccurate evaluation results. Summary of the Invention

[0003] The purpose of this invention is to provide a method for virtual pre-assembly and quality assessment of prefabricated bridges based on laser point cloud data. Through standardized point cloud data processing, high-precision virtual pre-assembly, and multi-dimensional quality assessment, the method achieves refined control over the construction quality of prefabricated bridges.

[0004] The specific technical solution of the present invention includes the following steps:

[0005] Step 1: First, review the prefabricated bridge design drawings to clarify the component types, dimensional parameters, and assembly accuracy requirements; then conduct a site survey to record the terrain, obstacle distribution, and environmental conditions; based on the survey results, set up scanning stations and magnetic targets to ensure that the target overlap rate of adjacent scanning stations meets the registration requirements; perform pre-scanning to obtain low-precision point cloud data, optimize the scanning resolution and scanning range, and determine the final scanning scheme.

[0006] Step 2: Using a terrestrial 3D laser scanner, the entire surface of the manufactured bridge components is scanned according to the scanning scheme to obtain high-precision point cloud data covering the key assembly parts of the components. The key parts include the splicing surface and weld area. The point cloud density of a single component must not be lower than the specified number requirement. During the scanning process, the component number, scanning time and scanning parameters are recorded to ensure the consistency between the point cloud data and the corresponding component.

[0007] Step 3: Preprocess the point cloud data obtained in Step 2, including noise point filtering, point cloud thinning and point cloud block processing, and form multiple point cloud sub-modules containing complete component surface information and assembly boundary information through the block processing.

[0008] Step 4: Using the spatial coordinates of the magnetic target as the global constraint benchmark, perform multi-station registration processing on the point cloud sub-modules to ensure that the point clouds of each sub-module are in a unified coordinate system; on this basis, construct a three-dimensional mesh model of the component, and perform virtual pre-assembly according to the component assembly sequence and assembly constraints to generate a virtual pre-assembly model of the prefabricated bridge; align the virtual pre-assembly model with the design BIM model, compare and extract the three-dimensional coordinate deviations of key assembly feature points, and construct an assembly error field to characterize the spatial distribution characteristics of assembly errors on the component surface and at assembly nodes.

[0009] Furthermore, the 3D coordinate deviations of feature points between the virtual pre-assembled model and the design BIM model are mapped according to their spatial positions on the component surfaces and assembly nodes, forming an assembly error field to characterize the spatial distribution of assembly errors. This assembly error field describes the concentration, distribution range, and directional characteristics of assembly errors in different component areas, providing a spatial distribution basis for subsequent quality assessment.

[0010] Step 5: Construct a multi-dimensional index system of "macro-geometric accuracy + micro-defect correlation". Based on the assembly error field, construct a multi-dimensional quality assessment index system that includes macro-geometric accuracy index and micro-assembly defect index, and introduce error spatial distribution characteristics as an index weight adjustment factor; adopt a cloud model-based evaluation method to perform uncertainty modeling on the quality assessment index, calculate the expected value, entropy value and hyper-entropy value of each index, thereby realizing a comprehensive evaluation of the virtual pre-assembly quality of the prefabricated bridge.

[0011] Furthermore, in the method of the present invention, the specific content of step 1 is as follows:

[0012] Study the prefabricated bridge design drawings, compile a component list, and mark key assembly points. Clarify the material, dimensional parameters, and assembly accuracy requirements for various components such as chords, web members, and node plates. Mark key areas (e.g., weld bevel angles, flange surface flatness) to provide a basis for subsequent scanning range determination and accuracy verification. During construction, avoid rainy days, strong winds, and extreme temperatures; prioritize sunny, light-wind periods to minimize the impact of environmental factors on scanning accuracy. Based on site conditions, use a total station to mark key control points (e.g., bridge bearing centers, temporary support foundation locations), record the location, size, and height of obstacles (e.g., construction machinery, temporary material storage), assess obstruction of the scanning line of sight, and mark areas where obstacles need to be temporarily removed or avoided.

[0013] In terms of site deployment, the principle of "fewer sites, full coverage, and high overlap" is followed, with the goal of being able to completely scan single or multiple adjacent components without scanning blind spots, combined with the unobstructed areas determined by surveying.

[0014] The locations of scanning stations are determined using measuring instruments, and magnetic targets are deployed on non-critical stress areas of the components or on fixed structures on site to ensure a target overlap rate between adjacent scanning stations that meets the requirements for multi-station registration. Each station must ensure that at least three subsequently deployed targets can be observed. Targets are deployed on non-critical stress areas of the components (such as the non-spliced ​​side of the chord members and the edge of the node plate) and on fixed structures on the construction site to ensure that the target overlap rate between adjacent stations is ≥30%, meeting the requirements for subsequent multi-station registration. The three-dimensional coordinates (X, Y, Z) of each target are measured using a total station, and the target number and corresponding coordinates are recorded to establish a target coordinate database, which serves as the constraint benchmark for subsequent point cloud registration.

[0015] Then, a terrestrial 3D laser scanner is used with low scanning resolution and low scanning quality to pre-scan the planned scanning area, acquiring low-precision point cloud data and generating a preliminary point cloud model. The pre-scanned point cloud model is loaded into professional point cloud processing software to check for any scanning blind spots (such as narrow gaps between components or edge areas not covered by the target). If blind spots exist, the scanning station positions are adjusted or auxiliary scanning stations are added to ensure that all critical parts of the components (joint surfaces, weld areas) are scanned and covered.

[0016] Finally, the locations of the scanning sites, target layout diagrams, scanning resolution, scanning sequence, and environmental parameter monitoring records were compiled to form a formal laser scanning plan, which was then implemented after approval by the project's technical leader.

[0017] Furthermore, in the method of the present invention, step 2 specifically includes:

[0018] Before laser scanning, the distance and angle accuracy of the terrestrial 3D laser scanner are calibrated, and the correspondence between point cloud reflection intensity and component surface characteristics is established through reflectivity calibration to improve the reliability of point cloud data. The specific implementation is as follows:

[0019] Before scanning, a visual inspection of the terrestrial 3D laser scanner is performed to confirm that the lens is free of stains and scratches, the device battery has sufficient power, and the data storage hard drive has enough remaining space based on the number of sites.

[0020] Using the standard distance calibration plate provided with the scanner, the distance from the scanner to the calibration plate was measured in an open area. The deviation between the measured value and the standard value was compared, and the distance was corrected using the equipment calibration software until the deviation was ≤ ±0.1mm. Using the angle calibration tool provided with the equipment, the measurement accuracy of the scanner's horizontal and vertical angles was adjusted to ensure that the horizontal angle deviation was ≤ ±5″ and the vertical angle deviation was ≤ ±5″. Using a standard reflectivity plate, under the same distance and lighting conditions, the point cloud reflectance intensity values ​​of different reflectivity plates were measured to establish the correspondence between reflectance intensity and actual reflectance. This is used to distinguish noise points (such as low-reflectance dust points) from effective component points during subsequent point cloud denoising.

[0021] Remove temporary obstacles identified in the preliminary survey. For large obstacles that cannot be removed, adjust the scanner angle during scanning to avoid obstructed areas, or add auxiliary scanning stations to ensure that there are no blind spots in the scanning of the components.

[0022] Following the scanning sequence determined by the previously prepared laser scanning plan, the scanner is placed at the preset station, leveled using a tripod, and the target coordinate database is accessed. The scanner is then manually aimed at and identified at least three targets around the station to complete the station initialization.

[0023] During the scanning process, the point cloud generation on the scanning software interface is monitored in real time to ensure that the target is always effectively identified. If the target is lost, the scanning is paused, and the target is re-aimed before continuing. At critical parts of the component (such as splicing holes and welds), the point cloud density is observed using the software's magnification function. If the density is insufficient, the scanning is paused, and the local scanning resolution of that area is increased before re-scanning.

[0024] Furthermore, in the method of the present invention, step 3 specifically includes:

[0025] Noise points are identified and removed from the point cloud based on the statistical characteristics of the point cloud's neighborhood distance to reduce the impact of non-component points on the accuracy of subsequent modeling. Single-site point cloud data is loaded into the point cloud processing software, and filtering parameters are set to calculate the distance between each point and its neighbors. The mean (μ) and standard deviation (σ) of all distances are calculated. If the average distance between a point and its neighbors is greater than μ+3σ or less than μ-3σ, the point is identified as an isolated noise point (such as dust reflections in the air or accidental noise points appearing during scanning), and is marked and removed. After denoising, the point cloud data of component edges and open areas is observed in the software. If significant isolated noise points still exist, the number of neighboring points or the mean multiple is adjusted, and the above operation is repeated until the isolated noise points are basically eliminated.

[0026] The denoised point cloud data is thinned using a voxel grid method to reduce redundancy while preserving the geometric features of the components. In the software, the voxel grid thinning function is enabled, the denoised point cloud data is imported, and corresponding voxel sizes are set for the point clouds of the main truss and secondary components. The software automatically retains a representative point within each voxel, completing the thinning process. Comparing the point cloud data volume before and after thinning ensures significant data compression, thereby reducing the workload of subsequent processing.

[0027] Based on the curvature variation characteristics and spatial adjacency relationships of the point cloud, the thinned point cloud is segmented, automatically identifying component splicing boundaries and forming multiple point cloud sub-modules. Each point cloud sub-module contains complete surface information and assembly boundary information of the corresponding component or component region, providing basic data for subsequent virtual pre-assembly and error analysis. The specific implementation is as follows: Import the thinned overall point cloud data, run a clustering algorithm. The algorithm automatically identifies component connection boundaries based on curvature differences and, combined with distance thresholds, segments the overall point cloud into independent sub-modules such as chord segments, web segments, and node regions. Each sub-module automatically generates a unique number. Each sub-module point cloud is loaded one by one in the software, and compared with the component dimensions in the design drawings to confirm that the sub-module point cloud covers the complete surface of the component (e.g., the chord segment point cloud includes both splicing surfaces and the entire side), with no missing areas.

[0028] Furthermore, in the method of the present invention, step 4 specifically includes:

[0029] The registration algorithm is run, and the software automatically adjusts the spatial position of each station's point cloud according to the target constraint points, realizing the stitching of multiple station point clouds to generate an overall point cloud model. The deviation between the actual coordinates (extracted from the overall point cloud model) of each target's center point after registration and the design coordinates is calculated. The 3D coordinate deviations (ΔX, ΔY, ΔZ) of all targets are statistically analyzed to ensure that the maximum deviation is within the allowable range. If the deviation exceeds the limit, it is checked whether the target has shifted during the scanning process, and the target coordinates are remeasured and re-registered.

[0030] Import the registered overall point cloud model (or the segmented sub-module point cloud) into professional modeling software, such as Geomagic Wrap. Utilize the software's smoothing function to eliminate minor noise and preserve the true geometry of the components, addressing any small bumps or depressions on the point cloud surface. Create a mesh from the point cloud, merging non-manifold edges into normal manifold edges to ensure a reasonable mesh topology. Automatically identify and delete overlapping mesh faces, retaining only valid faces. Measure key dimensions of the mesh model in the software (such as chord cross-section width and height, and web length) and compare them with the dimensions on the design drawings to confirm that the mesh has no non-manifold edges, overlapping faces, or isolated mesh cells, ensuring a reasonable model topology and no distortions or overlaps.

[0031] Finally, referring to the actual assembly process at the construction site and considering the stress characteristics of the bridge structure, a virtual assembly sequence was determined, such as "lower chord first → rear web members → upper chord then → node plate last," to ensure that the components are subjected to reasonable stress and there is no collision or interference during the assembly process. A component geometric error comparison table was created, including component number, error type (characteristic point deviation, cross-sectional dimension deviation, axis alignment deviation), maximum error value, allowable error value, and whether it is qualified (meeting the allowable error value is qualified, otherwise it is unqualified), providing a basis for subsequent quality assessment and rectification.

[0032] Furthermore, the accuracy of the virtual pre-assembled model and the pre-designed BIM model was compared, as follows:

[0033] Extract the three-dimensional coordinate information of the center point of the splicing hole, the end point of the component, and the axis control point, and calculate the coordinate deviation of the corresponding feature points in three spatial directions; based on the spatial distribution of the coordinate deviation on the surface of the component and the assembly node area, construct an assembly error field to characterize the concentration, distribution range and directional characteristics of the assembly error in different areas, thereby revealing the spatial correlation between component processing error and assembly error.

[0034] Furthermore, in the method of the present invention, step 5 performs uncertainty modeling on the quality assessment indicators based on the cloud model, including: calculating the expected value, entropy value and hyperentropy value corresponding to each quality assessment indicator to characterize the central tendency, dispersion and uncertainty characteristics of the indicator values; matching the cloud model corresponding to the component or virtual pre-assembled whole with the preset multi-level quality standard cloud model for similarity, and determining the quality level of the component or virtual pre-assembled whole based on the similarity analysis, so that the quality assessment results can simultaneously reflect the magnitude of geometric deviation and its spatial distribution uncertainty.

[0035] Based on relevant standards such as the "Code for Design of Steel Structures" (GB 50017-2017), the "Standard for Evaluation of Technical Condition of Highways" (JTG / TH21-2021), and the "Standard for Acceptance of Construction Quality of Steel Structures" (GB 50205-2020), evaluation indicators related to the virtual pre-assembly quality of prefabricated bridges are extracted, such as component size deviation, assembly alignment deviation, axis alignment deviation, and weld appearance quality (if combined with ultrasonic testing, internal weld defect indicators can be added).

[0036] Compared with the prior art, the present invention has the following advantages:

[0037] (1) It replaces traditional physical trial assembly, saves space, shortens the pre-assembly cycle, and quantifies the impact of uncertainty in cloud theory evaluation, avoiding the subjective bias of traditional deterministic evaluation.

[0038] (2) It is applicable to prefabricated steel structure bridges of different spans and types. It has strong scalability and can be integrated with technologies such as ultrasonic phased array and UAVs to achieve integrated detection of "macro + micro" and "ground + air", providing technical support for the quality control of the entire life cycle of bridges. Attached Figure Description

[0039] Figure 1 This is a schematic diagram of the virtual assembly of the side arch and central arch segments on one side of the present invention. Figure I ;

[0040] Figure 2 This is a schematic diagram of the virtual assembly of the side arch and central arch segments on one side of the present invention. Figure II ;

[0041] Figure 3 This is a schematic diagram of the virtual assembly result of the side arch and central arch segments of the present invention;

[0042] Figure 4 This is a schematic diagram showing the assembly error results of the side arch and central arch segments of the present invention;

[0043] Figure 5 This is a schematic diagram illustrating the dimensional error of the arch foot section of the side arch in this invention;

[0044] Figure 6 This is a schematic diagram showing the dimensional error of the arch foot section at the transition between the arch and the side arch in this invention;

[0045] Figure 7 This is a schematic diagram of the lateral dimensional error at the side arch in this invention;

[0046] Figure 8 This is a schematic diagram of the process of the present invention.

[0047] Figure 9 is a flowchart of the three-dimensional laser scanning measurement process of the present invention. Detailed Implementation

[0048] To make the technical problems, technical solutions and advantages of the present invention clearer, a detailed description will be given below in conjunction with the accompanying drawings and specific embodiments.

[0049] like Figure 8 As shown, the method for evaluating the quality of virtual pre-assembly of prefabricated bridges based on laser point cloud data includes the following steps:

[0050] Step 1: First, review the prefabricated bridge design drawings to clarify the component types (such as chords, web members, and node plates), dimensional parameters, and assembly accuracy requirements; then conduct a site survey, recording the terrain, obstacle distribution, and environmental conditions (light, temperature, and airflow); based on the survey results, set up scanning stations and magnetic targets to ensure that the overlap rate of targets at adjacent stations meets the registration requirements; perform pre-scanning to obtain low-precision point cloud data, optimize the scanning resolution and scanning range, and determine the final scanning scheme.

[0051] Step 2: Use a terrestrial 3D laser scanner, such as... Figure 9 As shown, the bridge components (including factory-processed segmental components and on-site assembled nodes) are scanned in full according to the design scheme. During the data acquisition process, the number of each component and the scanning time are recorded to ensure that the point cloud data covers key parts such as component splicing surfaces and weld areas. The point cloud density of a single component scan cannot be lower than a certain number of requirements.

[0052] Step 3: Preprocess the point cloud data using a three-step method: filtering, thinning, and block segmentation. Filtering removes isolated noise points caused by dust reflection and equipment errors. It also eliminates edge blurring points caused by excessive scanning angles, improving the signal-to-noise ratio of the point cloud. Thinning, based on the voxel grid method, dynamically adjusts voxel sizes and secondary components according to component accuracy requirements. While preserving key geometric features (such as splicing holes and weld edges), it compresses the point cloud data volume, reducing the processing load for subsequent steps. Block segmentation utilizes a curvature and distance-based clustering algorithm, using node centers as the reference, to divide the overall point cloud into independent sub-modules such as chord segments, web segments, and node domains. Each sub-module contains complete surface information and splicing boundary data, facilitating subsequent regional accuracy analysis.

[0053] Step 4: Using a magnetic target as a constraint benchmark, perform multi-station registration on the pre-processed point clouds of each sub-module. Import the registered point clouds into professional software and construct a 3D mesh model of the component through operations such as trimming, encapsulation, and repair. Ensure that the model's topology is reasonable and free from distortion and overlap. Then, virtually assemble the constructed 3D mesh model of the component according to the assembly sequence (e.g., lower chord first, then web members, then upper chord), simulating the on-site assembly conditions and setting assembly constraints (e.g., alignment accuracy of splicing holes, weld gap requirements), generating a virtual pre-assembled overall model. By introducing the virtual pre-assembly process, the quality assessment is expanded from the traditional single-component geometric accuracy inspection to a holistic assessment method considering the assembly relationship between components, thus avoiding misjudgments caused by judging assembly quality solely based on the inspection results of a single component. Finally, extract the key feature points (e.g., splicing hole centers, component endpoints) between the virtual pre-assembled model and the design BIM model, and calculate the 3D coordinate deviation of the feature points. Record the maximum error value and the error distribution area.

[0054] Step 5: Based on standards such as the "Code for Design of Steel Structures" (GB 50017-2017) and the "Standard for Assessment of Technical Condition of Highways" (JTG / T H21-2021), a multi-dimensional index system of "macro-geometric accuracy + micro-defect correlation" is constructed. A cloud theory evaluation model designed using the weighted average method is employed to calculate the expected values, entropy, and hyper-entropy of each index, and finally, the overall bridge quality is assessed.

[0055] In step 1, the prefabricated bridge design drawings are studied, the component list is compiled, and the material, dimensional parameters, and assembly accuracy requirements of various components such as chords, web members, and node plates are clarified. Key control points (such as weld bevel angles and flange surface flatness) are marked to provide a basis for subsequent scanning range determination and accuracy verification. During construction, rainy days, strong winds, high temperatures, or low temperatures are avoided; sunny, light-wind periods are preferred to reduce the impact of environmental factors on scanning accuracy. A total station is used to mark key control points on the construction site (such as the center of bridge bearings and the location of temporary support foundations), record the location, size, and height of obstacles (such as construction machinery and temporarily stored materials), assess the obstruction of the scanning line of sight, and mark areas where obstacles need to be temporarily removed or avoided. In terms of station layout, the principle of "fewer stations, full coverage, and high overlap" is followed, aiming to completely scan one or more adjacent components without scanning blind spots. Based on the unobstructed areas determined by the survey, the total station is used to initially determine the scanning station locations. Each station must ensure that at least three targets to be subsequently deployed can be observed. Targets were placed on non-critical load-bearing parts of the components (such as the non-spliced ​​ends of chord members and the edges of node plates) and on fixed structures at the construction site, ensuring that the overlap rate of targets at adjacent sites was ≥30% to meet the requirements of subsequent multi-station registration. The three-dimensional coordinates (X, Y, Z) of each target were measured using a total station, and the target number and corresponding coordinates were recorded to establish a target coordinate database, which served as the constraint benchmark for subsequent point cloud registration. Figure 9 As shown, a terrestrial 3D laser scanner was then used with low scanning resolution and low scanning quality to pre-scan the planned scanning area, acquiring low-precision point cloud data and generating a preliminary point cloud model. The pre-scanned point cloud model was then loaded into professional point cloud processing software to check for any scanning blind spots (such as narrow gaps between components or edge areas not covered by the target). If blind spots were found, the scanning station positions were adjusted or auxiliary scanning stations were added to ensure that all critical parts of the components (joint surfaces, weld areas) were scanned and covered. Finally, the scanning station positions, target layout diagram, scanning resolution, scanning sequence, and environmental parameter monitoring records were compiled to form a formal laser scanning plan, which was then implemented after approval by the project's technical lead.

[0056] In step 2, before scanning, a visual inspection of the terrestrial 3D laser scanner is performed to ensure the lens is free of stains and scratches, the device battery has sufficient power, and the data storage hard drive has enough remaining space based on the number of sites. Using the scanner's built-in standard distance calibration plate, the distance from the scanner to the calibration plate is measured in an open area. The deviation between the measured value and the standard value is compared, and the distance is corrected using the device calibration software until the deviation is ≤ ±0.1mm. Using the device's angle calibration tool, the measurement accuracy of the scanner's horizontal and vertical angles is adjusted to ensure that the horizontal angle deviation is ≤ ±5″ and the vertical angle deviation is ≤ ±5″. Using a standard reflectivity plate, under the same distance and lighting conditions, the point cloud reflectance intensity values ​​of different reflectivity plates are measured to establish the correspondence between reflectance intensity and actual reflectance. This is used to distinguish noise points (such as dust points with low reflectance) from effective points of the component during subsequent point cloud denoising. Temporary obstacles identified in the preliminary survey are removed. For large obstacles that cannot be removed, the scanner angle is adjusted during scanning to avoid obstructed areas, or auxiliary scanning sites are added to ensure that there are no blind spots in the scanning of the component. Following the scanning sequence determined in the previously prepared laser scanning plan, the scanner was placed at the preset station, leveled using a tripod, and the target coordinate database was accessed. The scanner was manually aimed at and identified at least three targets around the station to complete station initialization. During the scanning process, the point cloud generation on the scanning software interface was monitored in real time to ensure that the targets were always effectively identified. If a target was lost, the scanning was paused, re-aimed at the target, and then resumed. At critical parts of the component (such as splicing holes and welds), the point cloud density was observed using the software's magnification function. If the density was insufficient, the scanning was paused, the local scanning resolution of that area was increased, and the scanning was restarted.

[0057] In step 3, single-site point cloud data is loaded into the point cloud processing software. Filtering parameters are set to calculate the distance between each point and its neighboring points, and the mean μ and standard deviation σ of all distances are calculated. If the average distance between a point and its neighboring points is >μ+3σ or <μ-3σ, the point is identified as an isolated noise point (such as dust reflection points in the air or random noise points that appear during scanning), and this type of noise is marked and removed. After denoising, the point cloud data of the component edges and open areas is observed in the software. If obvious isolated noise points still exist, the number of neighboring points or the mean multiple is adjusted appropriately, and the above operation is repeated until the isolated noise points are basically eliminated. In the software, the voxel raster thinning function is enabled, and the denoised point cloud data is imported. The corresponding voxel sizes are set for the point clouds of the main truss components and secondary components, respectively. The software automatically retains a representative point within each voxel to complete the thinning. The amount of point cloud data before and after thinning is compared to ensure that the data volume is significantly compressed, thereby reducing the processing load. Import the thinned overall point cloud data and run a clustering algorithm. The algorithm automatically identifies the component connection boundaries based on curvature differences and, combined with distance thresholds, divides the overall point cloud into independent sub-modules such as chord segments, web segments, and node domains. Each sub-module is automatically assigned a unique number. Load the point cloud of each sub-module one by one in the software and compare it with the component dimensions in the design drawings to confirm that the sub-module point cloud covers the entire surface of the component (e.g., the chord segment point cloud includes the splicing surfaces at both ends and the entire side), with no missing areas.

[0058] In step 4, the registration algorithm is run to adjust the spatial position of each site's point cloud according to the target constraint points, achieving the stitching of multiple site point clouds to generate an overall point cloud model. The deviation between the actual coordinates (extracted from the overall point cloud model) of each target's center point after registration and the design coordinates is calculated. The 3D coordinate deviations (ΔX, ΔY, ΔZ) of all targets are statistically analyzed to ensure the maximum deviation is within the allowable range. If the deviation exceeds the limit, it is checked whether the target has shifted during the scanning process; the target coordinates are remeasured and registration is repeated. The successfully registered overall point cloud model (or the segmented sub-module point cloud) is imported into professional modeling software. For any minor protrusions or depressions on the point cloud surface, the software's smoothing function is enabled to eliminate minor noise and preserve the true geometric shape of the components. A mesh is created from the point cloud, merging non-manifold edges into normal manifold edges to ensure a reasonable mesh topology. Overlapping mesh surfaces are automatically identified and deleted, retaining only a single valid surface. In the software, key dimensions of the mesh model (such as chord cross-section width and height, and web member length) are measured and compared with the dimensions in the design drawings to confirm that the mesh has no non-manifold edges, overlapping surfaces, or isolated mesh cells, ensuring that the model's topology is reasonable and free from distortion and overlap. Finally, referring to the actual assembly process at the construction site and considering the stress characteristics of the bridge structure, the virtual assembly sequence is determined, such as "lower chord first → then web member → then upper chord → finally node plate," ensuring that the components are subjected to reasonable stress and there is no collision interference during assembly. A component geometric error comparison table is created, including component number, error type (feature point deviation, cross-sectional dimension deviation, axis alignment deviation), maximum error value, allowable error value, and whether it is qualified (meeting the allowable error value is qualified, otherwise it is unqualified), providing a basis for subsequent quality assessment and rectification. Furthermore, during the assembly process, alignment constraints for splicing hole axes, splicing surface gap control constraints, and component spatial interference constraints are set. These constraints limit the relative spatial positional relationship between components, enabling the virtual assembly process to truly reflect the geometric matching state between components under actual construction conditions, thereby avoiding the problem of ignoring the assembly relationship between components based solely on the geometric accuracy assessment of a single component.

[0059] In step 5, for each secondary index of each component, the parameters of the cloud model are calculated, and cloud droplets for that index are generated. Based on relevant standards such as the *Code for Design of Steel Structures* (GB 50017-2017), the *Standard for Evaluation of Highway Technical Condition* (JTG / T H21-2021), and the *Standard for Acceptance of Construction Quality of Steel Structures* (GB 50205-2020), evaluation indicators related to the virtual pre-assembly quality of prefabricated bridges are extracted, such as component dimensional deviation, assembly alignment deviation, axis alignment deviation, and weld appearance quality (if combined with ultrasonic testing, internal weld defect indicators can be added).

[0060] Based on actual engineering needs, the indicators are divided into two main categories: "macro-geometric accuracy indicators" and "micro-defect correlation indicators." Macro-geometric accuracy indicators (60% weighting) include component dimensional deviation (0.25 weighting), assembly alignment deviation (0.2 weighting), and axis alignment deviation (0.15 weighting). Micro-defect correlation indicators (40% weighting) include component surface flatness deviation (0.2 weighting) and surface defects (such as scratches and dents, 0.2 weighting). It should be noted that the above weighting is an implementation example. In practical applications, the weights of macro-geometric accuracy indicators and micro-defect correlation indicators can be dynamically adjusted based on the spatial distribution characteristics of the error field reflected in the assembly error field, thereby enhancing the sensitivity of the quality assessment results to key assembly components.

[0061] Based on the importance of each indicator to the pre-assembly quality of the bridge, a judgment matrix is ​​constructed, such as the judgment matrix for the correlation between macroscopic geometric accuracy indicators and microscopic defect indicators. Let U be a universe of discourse, and C be a fuzzy set on U. For any element x, the membership degree to C is a stably inclined random number μ(x). The distribution of membership degrees on U is called the membership cloud, denoted as cloud C(x), and (x, μ(x)) is called a cloud droplet. The above cloud model quantifies the uncertainty of quality indicators through expected value, entropy, and hyperentropy, enabling the quality assessment results to comprehensively reflect the magnitude of geometric deviations and their randomness and fuzziness. The numerical characteristics of the cloud are determined by the expected value... ,entropy and hyperentropy These three parameters represent, among which, the expected value Entropy represents the expected value of the cloud droplet distribution in the universe of discourse; It is a measure of ambiguity; hyperentropy. It reflects the degree of cloud droplet cohesion.

[0062] For a given valid domain , and These represent the lower and upper bounds of the effective universe of discourse, respectively, and the expected value of the cloud model. See equation (1).

[0063] (1)

[0064] In the formula: This is a parameter, typically taken as 0.5. For any cloud droplet... ,because Follows a normal distribution ,therefore The value of should satisfy the 3σ criterion, as shown in equation (2).

[0065] (2)

[0066] In the formula: ≥1. Consider it as its corresponding cloud and adjacent cloud The average value. And from Follows a normal distribution N( , ),therefore The value of should also satisfy the 3σ criterion, as shown in equation (3).

[0067] (3)

[0068] In the formula: ≥1, to avoid premature overlap of state clouds, generally Take 1, Take 2.

[0069] For each secondary indicator of each component, calculate the parameters of the cloud model and generate cloud droplets for that indicator.

[0070] Referring to the "Highway Technical Condition Assessment Standard" (JTG / T H21-2021) and in conjunction with the construction and acceptance requirements of prefabricated bridges, the quality level is divided into 4 categories, and the index score value [0, 100] is divided into 5 intervals, with interval limits (U1, U2, U3, U4) of (95, 80, 60, 40) respectively. Since this project is a new construction project, a score of 95 or above is required to pass the acceptance test. Therefore, the normal cloud model parameters of this standard are (100, 4.17, 0.417).

[0071] For each component, the similarity between its fused overall cloud model and the four standard cloud models is calculated. The cloud model similarity calculation method (such as the Euclidean distance method based on cloud droplet membership) is used to calculate the average membership distance between the component cloud model and each standard cloud model. The smaller the distance, the higher the similarity.

[0072] The quality level corresponding to the standard cloud model with the highest similarity is selected as the quality level of the component. If the overall standardization score of the component is ≥95 points (corresponding to a Class 1 standard cloud model), it is judged as a Class 1 component and meets the acceptance requirements; if the score is <60 points, it is judged as a Class 4 component and requires major rectification (such as component rework or reprocessing) or scrapping.

[0073] Furthermore, as shown in Table 1, the virtual pre-assembly technology for prefabricated bridges based on laser point clouds of the present invention is compared with the traditional method. It can be clearly seen from Table 1 that the present invention has significant improvements in pre-assembly cycle, key point accuracy, and overall cost. Moreover, the on-site risk occurrence rate is almost zero through the method of the present invention.

[0074] Table 1. Comparison between virtual pre-assembly technology for assembled bridges based on laser point clouds and traditional methods.

[0075] Comparison Dimensions Traditional on-site pre-assembly This patent Pre-assembly cycle Including site setup and component transportation, it takes 7-14 days. Point cloud processing and virtual computing require 2-3 days. Key point accuracy 3-5mm depends on manual measurement Target constraint can achieve a thickness of approximately 1mm. Comprehensive cost Including venue and labor, it costs over 100,000. Within 100,000 (point cloud devices) On-site risk incidence rate There are risks of hoisting collisions and working at heights. The entire process is a virtual simulation with no physical operation.

[0076] Furthermore, such as Figure 5-7 The figures shown are schematic diagrams of dimensional errors at the arch foot section of the side arch, the arch foot section at the transition between the central arch and the side arch, and the lateral dimensional error at the side arch. It is clear from the data that the error value of this invention is negligible.

[0077] Furthermore, such as Figures 1-3 The image shows a partial assembly process and result. After the assembly process is completed, the assembled parts of the model are magnified. The magnified result is shown below. Figure 4 As shown in (a), the error results are as follows: Figure 4 As shown in (b), software was used to calculate and measure the error at the assembly point. The results showed that there was a certain offset at the assembly point of the side arch and the central arch segments along the height and width directions of the arch, with the offset occurring in the height direction (…). Figure 4 (a) As shown by the red line in the middle, the offset relative to the width direction of the arch ( Figure 4 The value shown by the green line in (a) appears slightly larger. Measurements indicate that the assembly error in the height direction of the arch is approximately 1.883 mm (as shown in [reference]). Figure 4 As shown in (d), it accounts for 0.063% of the arch height. The assembly error in the width direction of the arch is approximately 0.291 mm. Figure 4 As shown in (e), it accounts for 0.022% of the arch width. The results show that the assembly process can be well realized in the subsequent actual assembly process and meets the specification requirements.

Claims

1. A method for evaluating the quality of virtual pre-assembly of prefabricated bridges based on laser point cloud data, characterized in that, Includes the following steps: Step 1: First, review the prefabricated bridge design drawings to clarify the component types, dimensional parameters, and assembly accuracy requirements; Subsequently, a site survey was conducted to record the terrain, obstacle distribution, and environmental conditions. Based on the survey results, scanning stations and magnetic targets were deployed to ensure that the overlap rate of targets at adjacent scanning stations could meet the registration requirements. Pre-scanning was performed to obtain low-precision point cloud data, and the scanning resolution and scanning range were optimized to determine the final scanning scheme. Step 2: Using a terrestrial 3D laser scanner, the entire surface of the manufactured bridge components is scanned according to the scanning scheme to obtain high-precision point cloud data covering the key assembly parts of the components. The key parts include the splicing surface and weld area. During the scanning process, the component number, scanning time and scanning parameters are recorded to ensure the consistency between the point cloud data and the corresponding components. Step 3: Preprocess the point cloud data obtained in Step 2, including noise point filtering, point cloud thinning and point cloud block processing, and form multiple point cloud sub-modules containing complete component surface information and assembly boundary information through the block processing. Step 4: Using the spatial coordinates of the magnetic target as the global constraint benchmark, perform multi-station registration processing on the point cloud sub-modules to ensure that the point clouds of each sub-module are in a unified coordinate system; on this basis, construct a three-dimensional mesh model of the component, and perform virtual pre-assembly according to the component assembly sequence and assembly constraints to generate a virtual pre-assembly model of the prefabricated bridge; align the virtual pre-assembly model with the design BIM model, compare and extract the three-dimensional coordinate deviations of key assembly feature points, and construct an assembly error field to characterize the spatial distribution characteristics of assembly errors on the component surface and at assembly nodes; Step 5: Based on the assembly error field, construct a multi-dimensional quality assessment index system that includes macroscopic geometric accuracy indicators and microscopic assembly defect indicators, and introduce error spatial distribution characteristics as an indicator weight adjustment factor; adopt a cloud model-based evaluation method to perform uncertainty modeling on the quality assessment index, calculate the expected value, entropy value and hyperentropy value of each index, thereby realizing a comprehensive evaluation of the virtual pre-assembly quality of prefabricated bridges.

2. The method for evaluating the quality of virtual pre-assembly of prefabricated bridges based on laser point cloud data according to claim 1, characterized in that, In step 1, the prefabricated bridge design drawings are studied, the component list is sorted out and the key assembly parts are marked; combined with the construction site conditions, the scanning station positions are determined by measuring instruments, and magnetic targets are set up in the non-critical stress areas of the components or on fixed structures on site to ensure that the target overlap rate between adjacent scanning stations meets the requirements of multi-station registration.

3. The method for evaluating the quality of virtual pre-assembly of prefabricated bridges based on laser point cloud data according to claim 1, characterized in that, The three-dimensional coordinates (X, Y, Z) of each target were measured using a total station, and the target number and corresponding coordinates were recorded to establish a target coordinate database, which served as a constraint benchmark for subsequent point cloud registration. Then, a terrestrial 3D laser scanner was used with low scanning resolution and low scanning quality to pre-scan the planned scanning area, obtain low-precision point cloud data, and directly generate a preliminary point cloud model. The preliminary point cloud model was loaded into professional point cloud processing software to check for any set scanning blind spots. If blind spots existed, the scanning station positions were adjusted or auxiliary scanning stations were added to ensure that all key parts of the components were scanned and covered. Finally, the scanning station positions, target layout diagram, scanning resolution, scanning sequence, and environmental parameter monitoring records were compiled to form a formal laser scanning plan.

4. The method for evaluating the quality of virtual pre-assembly of prefabricated bridges based on laser point cloud data according to claim 3, characterized in that, Before laser scanning, the distance and angle accuracy of the ground-based 3D laser scanner are calibrated, and the correspondence between the point cloud reflection intensity and the surface characteristics of the component is established through reflectivity calibration to improve the reliability of the point cloud data.

5. The method for evaluating the quality of virtual pre-assembly of prefabricated bridges based on laser point cloud data according to claim 4, characterized in that, In step 3, the preprocessing of the point cloud data includes: identifying and removing noisy points in the point cloud based on the statistical characteristics of the point cloud neighborhood distance to reduce the impact of non-component points on the accuracy of subsequent modeling; thinning the denoised point cloud data based on the voxel grid method to reduce the redundancy of the point cloud data while ensuring that the geometric features of the components are not distorted; segmenting the thinned point cloud based on the curvature change characteristics and spatial adjacency relationship of the point cloud, automatically identifying the component splicing boundary and forming multiple point cloud sub-modules, wherein each point cloud sub-module contains complete surface information and splicing boundary information of the corresponding component or component region, so as to provide basic data for subsequent virtual pre-assembly and error analysis.

6. The method for virtual pre-assembly and quality assessment of prefabricated bridges based on laser point cloud data according to claim 5, characterized in that, Step 4 is as follows: First, multi-station registration is performed on the preprocessed point cloud data of each sub-module; the spatial coordinates of the magnetic target are used as the global constraint benchmark, and the spatial position of the sub-module point clouds collected from different stations is calibrated using a professional point cloud registration algorithm; During the registration process, the system automatically identifies the feature information of the target in the point cloud data of each station. Using the three-dimensional coordinates of the target center point as the anchor point, it iteratively optimizes the spatial pose and position parameters of the point cloud of each sub-module, eliminates the coordinate deviation caused by multi-station scanning, and realizes the accurate stitching of the point cloud of each sub-module under a unified coordinate system, ensuring the overall continuity and geometric consistency of the stitched point cloud. After registration, the integrated point cloud data is imported into professional reverse engineering software to construct a 3D mesh model of the component. Specific operations include: accurately trimming redundant data based on the spatial distribution characteristics of the point cloud, retaining the main point cloud of the component; converting discrete point clouds into continuous polygonal mesh surfaces to initially form the 3D outline of the component; using a surface fitting algorithm to generate supplementary meshes for the missing areas of the 3D mesh model, ultimately constructing a 3D mesh model of the component with a reasonable topological structure, no distortion or overlap, and a geometric shape consistent with the actual component; subsequently, based on the constructed 3D mesh model, virtual pre-assembly is carried out; referring to the actual assembly process at the construction site and combining the stress characteristics of the bridge structure, the assembly sequence is determined to ensure a reasonable force transmission path between components during assembly, without collision or interference; during the assembly process, alignment constraints of splicing hole axes, splicing surface gap control constraints, and component spatial interference constraints are set. These constraints limit the relative spatial positional relationship between components, enabling the virtual assembly process to truly reflect the geometric matching state between components under actual construction conditions, thereby avoiding the problem of ignoring the assembly relationship between components based solely on the geometric accuracy assessment of a single component.

7. The method for virtual pre-assembly and quality assessment of prefabricated bridges based on laser point cloud data according to claim 6, characterized in that, The accuracy of the virtual pre-assembled model and the pre-designed BIM model were compared as follows: Extract the three-dimensional coordinate information of the center point of the splicing hole, the end point of the component, and the axis control point, and calculate the coordinate deviation of the corresponding feature points in three spatial directions; based on the spatial distribution of the coordinate deviation on the surface of the component and the assembly node area, construct an assembly error field to characterize the concentration, distribution range and directional characteristics of the assembly error in different areas, thereby revealing the spatial correlation between component processing error and assembly error.

8. The method for virtual pre-assembly and quality assessment of prefabricated bridges based on laser point cloud data according to claim 7, characterized in that, In step 5, uncertainty modeling is performed on the quality assessment indicators based on the cloud model, including: calculating the expected value, entropy value and hyperentropy value for each quality assessment indicator to characterize the central tendency, dispersion and uncertainty characteristics of the indicator values; matching the cloud model corresponding to the component or virtual pre-assembled whole with the preset multi-level quality standard cloud model for similarity, and determining the quality level of the component or virtual pre-assembled whole based on the similarity analysis, so that the quality assessment results can simultaneously reflect the magnitude of geometric deviation and its spatial distribution uncertainty.

Images