Complex steel structure construction quality detection method and system based on three-dimensional laser scanning

Abstract

The invention relates to the technical field of quality inspection, in particular to a complex steel structure construction quality detection method and system based on three-dimensional laser scanning, and the method comprises the steps: obtaining three-dimensional laser scanning data and corresponding design data of a steel member in a machining quality detection stage, a virtual pre-assembly stage and an on-site actual assembly stage of the steel member; preprocessing the obtained three-dimensional laser scanning data to obtain point cloud data corresponding to a real object, constructing the obtained design data into a BIM model and converting the BIM model into point cloud data corresponding to the model, carrying out point cloud registration on the two parts of data through a registration algorithm, obtaining an error between the two parts of data through comparison, and obtaining the point cloud data of the real object according to the obtained error information. And determining the actual quality of the steel member.

Description

Technical Field

[0001] The present invention relates to the technical field of quality inspection, and particularly to a method and system for detecting the construction quality of complex steel structures based on three-dimensional laser scanning. Background Technique

[0002] The statements in this part only provide background technical information related to the present invention and do not necessarily constitute prior art.

[0003] During the processing and assembly of steel components, errors may occur in each link, which has become an issue that cannot be ignored during the construction of steel structure buildings.

[0004] During the processing stage, the measurement results obtained by traditional detection methods such as visual inspection or ruler measurement often deviate significantly from the actual values of steel components, making it difficult to ensure the processing accuracy of complex steel components.

[0005] During the assembly stage, total station is usually used to set points and measure data, and the data is compared with the CAD drawing data to evaluate the quality of components. However, due to the complex operation of the total station, it is easily affected by external conditions, and coupled with the large amount of workload in later data analysis, this method is difficult to be effectively applied in the actual quality inspection of steel components.

[0006] Therefore, when the steel structures in construction projects are becoming more and more complex, the traditional steel structure quality inspection methods have obvious limitations. Summary of the Invention

[0007] To solve the technical problems existing in the above background technique, the present invention provides a method and system for detecting the construction quality of complex steel structures based on three-dimensional laser scanning. During different stages of steel structure construction, three-dimensional laser scanning data and corresponding design data of steel components are obtained. After preprocessing and format conversion, the two parts of data are compared to obtain the error information between the design data and the physical data of the steel structure at different construction stages, so as to use the error information as the basis for construction quality inspection.

[0008] To achieve the above object, the present invention adopts the following technical solutions:

[0009] The first aspect of the present invention provides a method for detecting the construction quality of complex steel structures based on three-dimensional laser scanning, including the following steps:

[0010] Obtain the three-dimensional laser scanning data and corresponding design data of steel components respectively during the processing quality inspection stage, virtual pre-assembly stage, and on-site actual assembly stage of steel components;

[0011] The obtained three-dimensional laser scanning data is preprocessed to obtain the point cloud data corresponding to the physical object, which serves as the source point cloud; the obtained design data is constructed into a BIM model and converted into the point cloud data corresponding to the model, which serves as the target point cloud; the two parts of data are subjected to point cloud registration through a registration algorithm, and the error between the two parts of data is obtained through comparison. According to the obtained error information, the actual quality of the steel component is determined.

[0012] Among them, the registration algorithm includes a coarse registration stage and a fine registration stage.

[0013] In the coarse registration stage, by randomly sampling point pairs in the source point cloud, the best rigid transformation corresponding to the points in the target point cloud is found; the target point cloud is divided into multiple small voxels, and the point cloud within each voxel is regarded as a Gaussian distribution, and the best rigid transformation is found by optimizing the overlapping degree of the Gaussian distributions between the source point cloud and the target point cloud.

[0014] In the fine registration stage, the nearest point on the surface of the target point cloud for each point in the source point cloud is found through nearest neighbor search; the stop timing of the registration process is determined by setting an error convergence threshold; by setting a normal vector angle threshold, it is determined whether the two found points are valid matching points; the rotation and translation matrices that minimize the error between the source point cloud and the target point cloud are found to form error information.

[0015] Furthermore, obtaining the three-dimensional laser scanning data of the steel component includes: according to the size and shape of the steel component to be scanned, arranging scanning stations, using a three-dimensional laser scanner to obtain the point cloud data of the steel component, and converting it into a general format.

[0016] Furthermore, obtaining the three-dimensional laser scanning data of the steel component also includes: using the target paper posted on the surface of the steel component as a scanning reference point, and splicing multiple groups of original three-dimensional laser point cloud data into the three-dimensional point cloud data corresponding to the complete steel component based on the software supporting the three-dimensional laser scanner.

[0017] Furthermore, obtaining the design data corresponding to the three-dimensional laser scanning data of the steel component specifically means: extracting the design information in the drawing file corresponding to the steel component and constructing a BIM model, and performing point cloud processing on the obtained BIM model.

[0018] Furthermore, in the coarse registration stage, the approximate alignment between the source point cloud and the target point cloud is found using the SAC-IA algorithm; by randomly sampling point pairs in the source point cloud, the best rigid transformation corresponding to the points in the target point cloud is found; in each iteration, the matching degree between the source point cloud and the target point cloud under the current transformation is evaluated, and the transformation with the highest matching degree is selected as the result of coarse registration.

[0019] The NDT algorithm is used to further refine the rough registration result, and a more accurate alignment is achieved by calculating the probability distribution of the point cloud in the local coordinate system. The target point cloud is divided into multiple small voxels, and the point cloud within each voxel is regarded as a Gaussian distribution. The best rigid transformation is found by optimizing the overlap degree of the Gaussian distributions between the source point cloud and the target point cloud.

[0020] Furthermore, during the fine registration process, the KD-tree nearest neighbor search is used to find the nearest point on the surface of the target point cloud for each point in the source point cloud, which accelerates the registration process. An error convergence threshold is set to determine whether the registration process stops. When matching point pairs, if the included angle is less than the set threshold, the two points are considered valid matching points. In each iteration, the best rigid transformation matrix between the source point cloud and the target point cloud is calculated using SVD decomposition.

[0021] The second aspect of the present invention provides a complex steel structure construction quality inspection system based on 3D laser scanning, including:

[0022] A data acquisition module, configured to: acquire the 3D laser scanning data of the steel member and the corresponding design data respectively during the processing quality inspection stage, virtual pre-assembly stage, and on-site actual assembly stage of the steel member;

[0023] A registration and comparison module, configured to: the obtained 3D laser scanning data is preprocessed to obtain the point cloud data corresponding to the physical object, the obtained design data is constructed into a BIM model and converted into the point cloud data corresponding to the model. The two parts of data are subjected to point cloud registration through a registration algorithm, and the error between the two parts of data is obtained through comparison. According to the obtained error information, the actual quality of the steel member is determined;

[0024] Among them, the registration algorithm includes a rough registration stage and a fine registration stage;

[0025] In the rough registration stage, the initial transformation matrix between the point cloud data corresponding to the physical object and the point cloud data corresponding to the model is estimated, and the accurate transformation matrix is obtained by using the pre-divided grid and coordinate system transformation;

[0026] In the fine registration stage, based on the nearest neighbor search, a set number of points are determined and constructed into a local plane. The orthogonal distance from the points in the source point cloud to this local plane is used as the error term. By minimizing the error term and using the normal vector included angle and SVD decomposition as constraint conditions, the optimal transformation matrix is iteratively calculated.

[0027] Furthermore, it also has a result interaction platform, which is used to receive and save the point cloud data corresponding to the physical object, the point cloud data corresponding to the model, and the corresponding comparison results during the processing quality inspection stage, virtual pre-assembly stage, and on-site actual assembly stage.

[0028] The third aspect of the present invention provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, the steps in the above-mentioned method for detecting the construction quality of complex steel structures based on 3D laser scanning are implemented.

[0029] The fourth aspect of the present invention provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, the steps in the above-mentioned method for detecting the construction quality of complex steel structures based on 3D laser scanning are implemented.

[0030] Compared with the prior art, the above one or more technical solutions have the following beneficial effects:

[0031] By comparing, the error information between the 3D laser scanning data of the steel members and the corresponding design data is obtained as the basis for steel structure quality inspection. The registration algorithm during data comparison is improved to reduce the interference of human factors and improve the accuracy and efficiency of comparison. During the point cloud registration process, the purpose of improving the registration algorithm is to significantly improve the registration efficiency on the premise of ensuring the registration accuracy, reduce the registration processing time and the use of computing resources, and reduce manual intervention through automated processing. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The accompanying drawings forming a part of this invention are used to provide a further understanding of the invention. The schematic embodiments and descriptions thereof of the invention are used to explain the invention and do not constitute an improper limitation of the invention.

[0033] Figure 1 is a schematic diagram of the process for detecting the construction quality of complex steel structures based on 3D laser scanning provided by one or more embodiments of the present invention;

[0034] Figure 2 is a schematic diagram of the error information obtained by the improved registration algorithm provided by one or more embodiments of the present invention;

[0035] Figure 3 is the error information obtained by using software provided by one or more embodiments of the present invention;

[0036] Figure 4 is a schematic diagram of the layout of 3D scanning stations provided by one or more embodiments of the present invention;

[0037] Figure 5 is a schematic diagram of the virtual pre-assembly of a single steel beam and a single steel column provided by one or more embodiments of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0038] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0039] It should be noted that the following detailed description is exemplary and is intended to provide further illustration of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention pertains.

[0040] It should be noted that the terms herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular forms are also intended to include the plural forms. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0041] Term Explanation:

[0042] The ICP algorithm, (Iterative Closest Point) algorithm, is a commonly used point cloud registration algorithm. Its core idea is to iteratively optimize the transformation from the source point cloud to the target point cloud to minimize the error between the source point cloud and the target point cloud.

[0043] The NDT algorithm, whose full name is The Normal Distributions Transform, is a classic point cloud registration algorithm.

[0044] The SAC-IA algorithm, whose full name is Sample Consensus Initial Alignment, is a point cloud rough registration algorithm based on local feature descriptors.

[0045] The following embodiments provide a method and system for detecting the construction quality of complex steel structures based on 3D laser scanning. At different stages of steel structure construction, 3D laser scanning data of steel components and corresponding design data are obtained. After preprocessing and format conversion, the two parts of data are compared to obtain the error information between the design data and the physical data at different construction stages of the steel structure, so as to use the error information as the basis for construction quality detection.

[0046] Embodiment 1:

[0047] A method for detecting the construction quality of complex steel structures based on 3D laser scanning includes the following steps:

[0048] Obtain the 3D laser scanning data of the steel component and the corresponding design data respectively at the processing quality inspection stage, virtual pre-assembly stage, and on-site actual assembly stage of the steel component;

[0049] The obtained 3D laser scanning data is preprocessed to obtain point cloud data, and the obtained design data is constructed into a BIM model. The point cloud data and the BIM model are subjected to point cloud registration through a registration algorithm, and the error between the two parts of data is obtained through comparison;

[0050] According to the obtained error information, determine the actual quality of the steel members;

[0051] Among them, the registration algorithm includes a rough registration stage and a fine registration stage;

[0052] In the rough registration stage, the SAC-IA algorithm is adopted to quickly find an approximate alignment between the source point cloud and the target point cloud. By randomly sampling point pairs in the source point cloud and attempting to find the best rigid transformation corresponding to the points in the target point cloud. And combined with KD-tree for nearest neighbor search, and at the same time set a distance constraint condition to improve the accuracy of the initial alignment. Then, the NDT algorithm is used for enhanced rough registration. A more accurate alignment is achieved by calculating the probability distribution of the point cloud in the local coordinate system. The target point cloud is divided into multiple small voxels, and the point cloud within each voxel is regarded as a Gaussian distribution. Then, by optimizing the overlap degree of these Gaussian distributions between the source point cloud and the target point cloud, the best rigid transformation is found. Compared with the SAC-IA algorithm, the NDT algorithm has stronger robustness and can further optimize the initial registration result obtained by the SAC-IA algorithm to improve the registration accuracy.

[0053] In the fine registration stage, after the rough registration by the SAC-IA algorithm and the NDT algorithm, the point cloud has obtained a good pose. At this time, the point-to-plane ICP algorithm is adopted for fine registration, and at the same time, optimization measures are added to the ICP algorithm: increasing KD-tree nearest neighbor search, setting an error convergence threshold, setting a threshold for the angle between normal vectors, and using SVD decomposition to find the optimal matrix, making it more accurate and efficient in the fine registration process.

[0054] The 3D laser scanning file (pcd format) of the scanned steel members and the BIM point cloud file (pcd format) are registered and compared through a development tool (such as Visual Studio 2019), combined with the improved registration algorithm to obtain error information, and judge the processing quality, pre-assembly quality, and on-site actual assembly quality of the steel members.

[0055] Use error comparison software to compare the geometric information in the BIM model with the 3D coordinates in the point cloud data to obtain error information, which is mutually verified with the error information obtained by the improved algorithm. As Figure 3As shown, through integration with modeling software (such as Geomagic series software), the three-dimensional laser scanning files (pcd format) of the scanned steel members and the BIM point cloud files (pcd format) are imported into the software to analyze the construction errors. These errors can be reflected on the three planes of X, Y, and Z, thereby obtaining the magnitude of the construction errors. The magnitude of the errors directly reflects the construction quality. If the errors are very small, it indicates high construction quality and a good fit between the construction result and the design model; if the errors are relatively large, it may indicate problems in the construction process and further inspection and correction are required.

[0056] The specific process is as Figure 1 shown.

[0057] S1. According to the size and shape complexity of the steel members to be scanned, reasonably arrange the scanning stations, and use a three-dimensional laser scanner to scan the complex steel members to ensure that all information of the complex steel members is captured completely. After the scanning is completed, convert the collected point cloud to a common point cloud data format.

[0058] S2. Denoise and downsample the point cloud data obtained in S1 through the software supporting the three-dimensional laser scanner, point cloud processing software, and point cloud processing algorithms to obtain the preprocessed point cloud data.

[0059] S3. Use the CAD drawings of the steel members and apply model creation software (such as Tekla or Revit) to construct a BIM model. At the same time, perform point cloud processing on the BIM model. The BIM model point cloud processing specifically means: convert the BIM model from the rvt format to the obj format, and then convert it to the pcd point cloud format. By using the registration algorithm and comparison software for the preprocessed point cloud data and the steel member BIM model, compare the errors between the scanned point cloud data and the BIM point cloud data to inspect the actual quality of the steel members.

[0060] In this embodiment, the algorithm used is an independently programmed algorithm, which is a registration algorithm based on the improvement of the ICP algorithm. This algorithm uses the SAC-IA and NDT algorithms for rough registration, and then uses the point-to-plane ICP algorithm for fine registration.

[0061] S4. Upload the point cloud data and the analysis results at each stage of the complex steel structure construction to the result interaction platform, and through web service data sharing, realize the browsing, export, and printing of the point cloud data analysis report.

[0062] Specifically, the design information entry design is the comparison benchmark for the construction quality inspection of complex steel members. The design information of complex steel members needs to be entered and parsed according to industry specifications, including basic information, geometric information, material information, mechanical information, connection and attachment information, process and manufacturing information.

[0063] In S1, the scanning area needs to be planned in advance before scanning, and the station positions are reasonably set according to the area to be scanned, in order to minimize the external obstruction of the scanning area to be processed and maximize the scanning accuracy. The typical duration of the station scanning may be around 2 minutes to 50 minutes depending on the equipment. In this embodiment, the three-dimensional scanning station is arranged as follows Figure 4 shown.

[0064] Original point cloud data acquisition: The original point cloud after scanning is still stored in the device's memory card (usually an inserted SD card). The original point cloud needs to be exported to the post-processing software for application analysis. Different scanning devices provide different point cloud export methods, including WiFi wireless connection, wired connection, direct plug-in and read SD card, etc. Point cloud data is generally obtained very quickly, with a typical time of about 30 seconds to 2 minutes.

[0065] Point cloud format conversion. Different equipment manufacturers generally have their own encrypted and private point cloud formats, but they can also support the export of various common standard formats. Typical common formats include LAS, LAZ, E57, PTS, XYZ, PCD, etc. It is necessary to ensure that the unique format is converted into a common point cloud format.

[0066] In S2, the point cloud data is preprocessed, the point cloud is registered and the coordinates are converted. This step is required for scanning devices that do not have the station setting capability.

[0067] The point cloud scanned by each station is in a coordinate system with its own station center as the origin. The data scanned by multiple stations need to be registered in the same coordinate system, and then converted to the actual engineering coordinate system through the engineering coordinates of feature points such as target balls or target paper. This step is usually handled by the software that comes with the device. Because in the process of multi-station cloud registration, due to the scanning accuracy and point density problems at the target ball, the registration accuracy may be insufficient and the error may be too large.

[0068] Point cloud denoising: The construction environment around complex steel components is poor. There are various factors, such as vehicles and personnel in construction, ventilation and drainage, and construction material obstruction, which will cause a large amount of unnecessary point cloud data and noise in the collected data. Both manual and automatic methods can filter out unnecessary point clouds and noise, depending on the ability of the point cloud processing software. Ultimately, the point cloud processed by the software will provide reliable support for subsequent analysis; point cloud downsampling: The original point cloud data collected by the scanning equipment is extremely large in magnitude, and there are many redundant points, which hinder the subsequent analysis efficiency, processing smoothness, and transmission failure. Therefore, it is necessary to remove the redundant points in the point cloud data without affecting the point cloud characteristics of the measured subject. This step is automatically implemented by the algorithm. The denoising algorithm uses a statistical filtering algorithm and voxel downsampling.

[0069] In S3, a BIM model is constructed based on the design information of complex steel members. The design information mainly includes the geometric information of complex steel members. Through the comparison between the point cloud and the BIM model, multi-dimensional data analysis of points, lines, surfaces, and volumes of complex steel members is realized. Depending on the software used and the performance of the computer, the data preprocessing and data analysis processes can be completed on-site or at a designated location, with a typical time of 10 minutes to 2 hours.

[0070] In S4, the point cloud data and analysis results can be uploaded to the platform through an interface, and the data is shared with project stakeholders through web services to achieve online browsing, report export, and printing of the point cloud and various result analysis reports. The analysis results are statistically summarized according to the occurrence time of the processing, virtual pre-assembly, and actual assembly processes of complex steel members; the point cloud and results can be directly viewed and managed through the web page to intuitively understand the project progress and the quality information of the construction process of complex steel members; detailed report data of results such as error information can be viewed on the page. The platform can quickly view and manage the above original data and results through a computer or a mobile phone.

[0071] Furthermore, the physical structure is a complex steel member. The construction quality inspection of the complex steel member consists of three key stages: processing quality inspection, virtual pre-assembly, and on-site real-time monitoring. An improved registration algorithm is used in each stage to achieve the precise matching of the BIM point cloud model and the three-dimensional scanned point cloud model. The actual deviation information of the steel member is analyzed.

[0072] Furthermore, after achieving the precise registration of the point cloud model and the BIM model, the processing quality of the point cloud data obtained by three-dimensional scanning and the BIM model data is further compared. Monitoring points are selected from the complex steel member, and Geomagic software is used for deviation analysis. If the deviation is within the controllable range, the subsequent work is continued; if the deviation is too large and exceeds the allowable error limit, the steel member is sent back to the factory for repair.

[0073] In this embodiment, the deviation analysis results obtained by using Geomagic software are as Figure 3 shown.

[0074] Furthermore, after the complex steel member passes the processing quality inspection and meets the quality inspection standards, virtual pre-assembly of the complex steel member is carried out to check its splicing quality and prepare for subsequent actual assembly.

[0075] Furthermore, after the complex steel member completes virtual pre-assembly and is confirmed to meet the installation standards, it is immediately transported to the construction site for precise actual assembly. After the actual assembly is completed, the real-time point cloud data of the steel member is obtained by using three-dimensional laser scanning technology, and then these data are compared in detail with the BIM design model through Geomagic software to verify whether the assembly quality of the steel structure strictly meets the design requirements and quality standards.

[0076] During the point cloud registration process, the set of point cloud data selected as the reference or benchmark is the target point cloud. In this embodiment, the target point cloud is the point cloud data model after BIM point cloud conversion. The BIM point cloud conversion model is predefined and has high precision and high integrity. During the registration process, the target point cloud provides positioning and shape reference for the source point cloud. During the point cloud registration process, the set of point cloud data that needs to be transformed to align with the target point cloud is the source point cloud. In this embodiment, the source point cloud is the original point cloud data of the steel member measured by a 3D laser scanner. During the registration process, the source point cloud will undergo a series of coordinate transformations (such as rotation, translation, etc.) until its shape, position, and orientation match those of the target point cloud.

[0077] The ICP algorithm is a commonly used fine registration algorithm. Its core lies in calculating the optimal rigid transformation between two sets of point clouds to achieve the best match between the source point cloud and the target point cloud. However, the traditional ICP algorithm has some limitations during the registration process, such as being prone to falling into local optimal solutions and having low computational efficiency, which directly affect the accuracy and efficiency of the registration results. To solve these problems, the traditional ICP algorithm is improved by adopting the point-to-plane ICP method, adding KD-Tree nearest neighbor search, setting an error convergence threshold, setting a threshold for the normal vector angle, and using SVD decomposition to find the optimal matrix.

[0078] In the coarse registration stage, the SAC-IA algorithm is adopted and combined with KD-tree for nearest neighbor search. At the same time, distance constraint conditions are set to improve the accuracy of the initial alignment. Then, the NDT algorithm is used to further enhance the effect of coarse registration.

[0079] In the coarse registration stage of point cloud registration, the SAC-IA algorithm (i.e., one of the component algorithms of the improved registration algorithm based on the ICP algorithm) is first used to quickly find an approximate alignment between the source point cloud and the target point cloud. By randomly sampling point pairs in the source point cloud and trying to find the optimal rigid transformation with the corresponding points in the target point cloud.

[0080] In this process, there will be multiple iterations. Each iteration will evaluate the matching degree between the source point cloud and the target point cloud under the current transformation. Finally, the transformation with the highest matching degree is selected as the coarse registration result. At the same time, optimization measures are added to the SAC-IA algorithm to make it more accurate and efficient during the registration process.

[0081] (1) Combining KD-tree: To accelerate the nearest neighbor search, the KD-tree structure is used to quickly locate the nearest neighbor points of each point in the source point cloud in the target point cloud, thereby improving the efficiency of the SAC-IA algorithm.

[0082] (2) Distance constraint: When finding corresponding points, a certain distance threshold is set, and only those point pairs with relatively close distances will be considered to reduce incorrect matches.

[0083] After the SAC-IA algorithm is completed, the NDT algorithm is used to further refine the rough registration result, and a more accurate alignment is achieved by calculating the probability distribution of the point cloud in the local coordinate system. The target point cloud is divided into multiple small voxels, and the point cloud within each voxel is regarded as a Gaussian distribution. Then, the best rigid transformation is found by optimizing the overlap degree of these Gaussian distributions between the source point cloud and the target point cloud.

[0084] After rough registration by the SAC-IA algorithm and the NDT algorithm, the point cloud has obtained a good pose. Considering the closest distance from a point to the surface of the target point cloud, the point-to-plane ICP algorithm is used for fine registration, which can more accurately reflect the true alignment of the point cloud. At the same time, optimization measures are added to the ICP algorithm to make it more accurate and efficient during the registration process.

[0085] (1) KD-tree nearest neighbor search: During the fine registration process, the KD-tree is continued to be used to quickly find the nearest point on the surface of the target point cloud for each point in the source point cloud, thereby accelerating the registration process.

[0086] (2) Error convergence threshold: During the fine registration process, an error convergence threshold is set to determine whether the registration process should stop. When the error change after several consecutive iterations is less than this threshold, it is considered that the registration has converged and the iteration stops.

[0087] (3) Normal vector angle threshold: When matching point pairs, the angle between their normal vectors is considered, and only when the angle is less than the set threshold are these two points considered valid matching points. This helps to exclude abnormal points caused by noise or incorrect matches.

[0088] (4) SVD decomposition to find the optimal matrix: In each iteration, SVD decomposition is used to calculate the best rigid transformation matrix between the source point cloud and the target point cloud. SVD decomposition can find the rotation and translation matrices that minimize the error between the source point cloud and the target point cloud.

[0089] In the point cloud registration process of steel components, the purpose of improving the registration algorithm is to significantly improve the registration efficiency while ensuring the registration accuracy, reduce the registration processing time and the use of computing resources, and reduce manual intervention through automated processing.

[0090] (1) Reduce computing time: Through multi-stage registration (initial rough registration → enhanced rough registration → fine registration), efficient data structures (KD-tree), and fast algorithms (SVD decomposition), the overall computational complexity is reduced.

[0091] (2) Avoid redundant calculations: Filter out invalid point pairs through constraint conditions (such as distance constraints and normal vector angle thresholds) to reduce invalid iterations.

[0092] (3) Automatic parameter setting: Preset reasonable constraint thresholds, such as the normal vector angle threshold, without the need for repeated manual adjustment.

[0093] (4) Full-process automation: From feature extraction to the output of the final transformation matrix, it is all automatically completed, avoiding subjective operations such as manual marking of corresponding points.

[0094] Specific implementation steps:

[0095] 1. Multi-stage registration strategy

[0096] (1) SAC-IA: Quickly obtain the initial transformation matrix through random sampling and feature matching, and preliminarily align the source point cloud P to the target point cloud Q to reduce the number of iterations in subsequent fine registration.

[0097] (2) NDT: Further adjust the transformation parameters through probability density optimization, narrow the search space for fine registration, and avoid ICP falling into local optima.

[0098] (3) Point-to-plane ICP: Based on the result of coarse registration, directly start iterations at a position close to the optimal initial position to shorten the convergence time.

[0099] 2. KD-tree nearest neighbor search

[0100] Search for candidate point pairs (FPFH feature similar points) in the SAC-IA algorithm. Search for the nearest neighbor points and normal vectors in the ICP algorithm. It can significantly reduce the matching time.

[0101] 3. Filter invalid point pairs with constraint conditions

[0102] (1) Distance constraint (SAC-IA stage): Set the maximum matching distance to filter out incorrect matches across regions and reduce the calculation of invalid transformation matrices.

[0103] (2) Normal vector angle threshold (ICP stage): Set the normal vector angle threshold to avoid the interference of incorrect geometric relationships and reduce redundant calculations during the iteration process.

[0104] 4. SVD decomposition to directly solve the optimal transformation

[0105] Calculate the initial transformation matrix through corresponding point pairs in the SAC-IA algorithm. Solve the incremental transformation for iterative optimization through point-to-plane error in the ICP algorithm. SVD is more efficient and stable than traditional iterative optimization, especially with significant advantages when dealing with large-scale point clouds.

[0106] After being processed by the improved registration algorithm, the actual deviation information of steel components is analyzed together with the design and processing information of steel components to effectively detect their processing quality.

[0107] By performing virtual pre-assembly on steel components, such as a single steel column and a single steel beam, assembly errors are obtained, such as Figure 2 As shown, the error is the error of the corresponding bolt hole positions in the point cloud models of the single steel column and the single steel beam after virtual pre-assembly, and the final result is 0.665 mm. By judging the welding or bolt hole installation errors, it is verified whether they meet the actual assembly standards and whether the hole positions can be aligned, so as to prepare for the actual on-site assembly. After the on-site assembly of steel components is completed, three-dimensional laser scanning technology is used to obtain the real-time point cloud data of complex steel components and compare it with the BIM model data of complex steel components. Through the improved registration algorithm based on the ICP algorithm, comparative analysis is carried out to verify whether the actual assembly quality meets the standards. The compared error data are statistically analyzed, including the root mean square error (RMSE), etc., to quantify the construction quality and compare it with industry standards to determine whether the construction meets the requirements. A detailed visualization report is generated to display the error distribution and the construction quality assessment results, and the results are uploaded to the platform so that technicians and non-technicians can understand and evaluate the construction quality.

[0108] Using error comparison software, the geometric information in the BIM model is compared with the three-dimensional coordinates in the point cloud data to obtain error information, which corroborates the error information obtained by the improved algorithm. As Figure 3 shown, through the integration of Geomagic series software, the three-dimensional laser scanning file (pcd format) of the scanned steel component and the BIM point cloud file (pcd format) are imported into the software to analyze and obtain the construction errors. These errors can be reflected on the X, Y, and Z plane surfaces, thereby obtaining the magnitude of the construction errors. The magnitude of the errors directly reflects the construction quality. If the errors are small, it indicates high construction quality and a good fit between the construction results and the design model; if the errors are large, it may indicate problems in the construction process and further inspection and correction are required.

[0109] In the above process, set up stations for three-dimensional laser scanning, obtain the corresponding station data by posting target papers, without the cooperation of a total station, and obtain the absolute coordinates of the point cloud; fast and accurate, automatically level, complete the field scanning in 2 minutes and the accuracy is within 2 mm; instant analysis: the point cloud is automatically and evenly thinned, automatically filtered, and the result data can be obtained immediately within 15 minutes from the start of setting up the instrument and setting the scanning parameters, without any additional in-field work;

[0110] By forming the assets of the complex steel structure construction process with the full amount of data, laser scanning technology can cover the key links of the complex steel structure construction, and the relevant point clouds and real-scene photos can be retained as important construction process assets;

[0111] Ensure quality control with real data. In all aspects of complex steel structure construction, various indicators related to physical dimensions, shapes, and volumes can be detected through laser scanning technology, including some critical indicators. The laser scanning technology determines the characteristics of the massive data, making it essentially difficult to edit or forge the data. The authenticity of the data will ensure the effectiveness of the detection, thus providing reliable support for quality control.

[0112] Promote construction safety. Three-dimensional laser scanning technology can be applied to the quality monitoring of complex steel structure construction under certain conditions. During the processing of complex steel components, virtual pre-assembly process, and actual assembly process, as a key technology, it provides a useful supplement to the improvement of quality detection technical means.

[0113] Cost control. The processing and rework of complex steel components are key factors in the profit and loss of steel structure building construction. The application of laser scanning technology in the field of quality detection will obtain accurate geometric information data of complex steel components. On the one hand, it can promote the adjustment of subsequent complex steel component processing design and the improvement of processing quality; on the other hand, it can force the on-site operation team to pay attention to cost losses from the management level, ensure quality, and promote safety. Whether a technical means that can additionally provide cost savings for the construction unit is an important reason for its willingness to accept and actively promote laser scanning technology.

[0114] Improve efficiency, reduce rework, and promote the orderly progress of the project. The application of laser scanning technology to the quality detection of complex steel components can achieve on-site actual measurement. While obtaining full-scale data, the efficiency of both in-field and out-field work is significantly higher than that of traditional total stations or visual inspection methods. The proper application of laser scanning technology does not occupy additional construction time and does not require additional technical personnel. It not only does not affect the construction progress but also promotes the early discovery and solution of problems. The improvement of detection technical means will also, in turn, promote the improvement of construction technology and the enhancement of construction skills.

[0115] The laser scanning point cloud data can be well combined with the steel structure BIM design to realize the implementation of steel structure BIM applications and become the basis for the construction of steel structure digital twins.

[0116] Example Two:

[0117] A complex steel structure construction quality detection system based on three-dimensional laser scanning, including:

[0118] A data acquisition module, configured to: obtain the three-dimensional laser scanning data of the steel component and the corresponding design data respectively during the processing quality detection stage, virtual pre-assembly stage, and on-site actual assembly stage of the steel component;

[0119] The registration and comparison module is configured to: the obtained three-dimensional laser scanning data is preprocessed to obtain the point cloud data corresponding to the physical object, the obtained design data is constructed into a BIM model and converted into the point cloud data corresponding to the model, the two parts of data are subjected to point cloud registration through a registration algorithm, and the error between the two parts of data is obtained through comparison, and the actual quality of the steel component is determined according to the obtained error information;

[0120] Among them, the registration algorithm includes a rough registration stage and a fine registration stage;

[0121] In the rough registration stage, the initial transformation matrix between the point cloud data corresponding to the physical object and the point cloud data corresponding to the model is estimated, and the accurate transformation matrix is obtained by using the pre-divided grid and coordinate system transformation;

[0122] In the fine registration stage, based on the nearest neighbor search, a set number of points are determined and constructed into a local plane, the orthogonal distance from the points in the source point cloud to the local plane is used as the error term, by minimizing the error term, and using the normal vector angle and SVD decomposition as the constraint conditions, the optimal transformation matrix is obtained through iterative calculation.

[0123] Embodiment 3:

[0124] This embodiment provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, it implements the steps in the method for detecting the construction quality of complex steel structures based on three-dimensional laser scanning as described in Embodiment 2 above.

[0125] Embodiment 4:

[0126] This embodiment provides a computer device, including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the steps in the method for detecting the construction quality of complex steel structures based on three-dimensional laser scanning as described in Embodiment 2 above.

[0127] The steps involved in Embodiments 2 to 4 above correspond to those in Embodiment 1, and the specific implementation manners can refer to the relevant description part of Embodiment 1. The term "computer-readable storage medium" should be understood to include a single medium or multiple media including one or more instruction sets; it should also be understood to include any medium that can store, encode, or carry an instruction set for execution by a processor and enable the processor to execute any method in the present invention.

[0128] The above are only the preferred embodiments of the present invention and are not used to limit the present invention. For those skilled in the art, the present invention can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A method for detecting the construction quality of complex steel structures based on three-dimensional laser scanning, characterized in that, Including the following steps: Obtain the three-dimensional laser scanning data and corresponding design data of the steel members respectively in the processing quality inspection stage, virtual pre-assembly stage and on-site actual assembly stage of the steel members; The obtained three-dimensional laser scanning data is preprocessed to obtain the point cloud data corresponding to the physical object, and the obtained design data is constructed into a BIM model and converted into the point cloud data corresponding to the model. The two parts of data are registered by a registration algorithm, and the error between the two parts of data is obtained by comparison. According to the obtained error information, the actual quality of the steel members is determined; Among them, the registration algorithm includes a coarse registration stage and a fine registration stage; In the coarse registration stage, find the best rigid transformation corresponding to the points in the target point cloud by randomly sampling the points in the source point cloud; divide the target point cloud into multiple small voxels, and the point cloud in each voxel is regarded as a Gaussian distribution. Find the best rigid transformation by optimizing the overlap degree of the Gaussian distributions between the source point cloud and the target point cloud; In the fine registration stage, find the nearest point on the surface of the target point cloud for each point in the source point cloud through nearest neighbor search; determine the stop timing of the registration process by setting an error convergence threshold; determine whether the two points found are valid matching points by setting a normal vector angle threshold; find the rotation and translation matrices that minimize the error between the source point cloud and the target point cloud to form error information.

2. The method for detecting the construction quality of complex steel structures based on 3D laser scanning according to claim 1, characterized in that Obtain the three-dimensional laser scanning data of the steel members, including: arranging scanning stations according to the size and shape of the steel members to be scanned, using a three-dimensional laser scanner to obtain the point cloud data of the steel members, and converting it into a general format.

3. The method for detecting the construction quality of complex steel structures based on three-dimensional laser scanning according to claim 1, characterized in that, Obtaining the three-dimensional laser scanning data of the steel members further includes: using the target paper posted on the surface of the steel members as a scanning reference point, and splicing multiple groups of original three-dimensional laser point cloud data into the three-dimensional point cloud data corresponding to the complete steel members based on the software supporting the three-dimensional laser scanner.

4. The method for detecting the construction quality of complex steel structures based on 3D laser scanning according to claim 1, wherein Obtain the design data corresponding to the three-dimensional laser scanning data of the steel members, specifically: extract the design information from the drawing file corresponding to the steel members and construct a BIM model, and perform point cloud processing on the obtained BIM model.

5. The method for detecting the construction quality of complex steel structures based on 3D laser scanning according to claim 1, characterized in that In the coarse registration stage, use the SAC-IA algorithm to find the approximate alignment between the source point cloud and the target point cloud; find the best rigid transformation corresponding to the points in the target point cloud by randomly sampling the points in the source point cloud; in each iteration, evaluate the matching degree between the source point cloud and the target point cloud under the current transformation, and select the transformation with the highest matching degree as the coarse registration result; Use the NDT algorithm to further refine the coarse registration result, and achieve more accurate alignment by calculating the probability distribution of the point cloud in the local coordinate system; divide the target point cloud into multiple small voxels, and the point cloud in each voxel is regarded as a Gaussian distribution. Find the best rigid transformation by optimizing the overlap degree of the Gaussian distributions between the source point cloud and the target point cloud.

6. The method for detecting the construction quality of complex steel structures based on 3D laser scanning according to claim 1, wherein, In the fine registration process, use KD-tree nearest neighbor search to find the nearest point on the surface of the target point cloud for each point in the source point cloud to accelerate the registration process; set an error convergence threshold to judge whether the registration process stops; When matching point pairs, when the included angle is less than the set threshold, it is considered that the two points are valid matching points; In each iteration, the optimal rigid transformation matrix between the source point cloud and the target point cloud is calculated using SVD decomposition.

7. A complex steel structure construction quality inspection system based on 3D laser scanning, characterized in that, It includes: A data acquisition module configured to obtain the three-dimensional laser scanning data and the corresponding design data of the steel component respectively in the processing quality inspection stage, virtual pre-assembly stage, and on-site actual assembly stage of the steel component. A registration and comparison module configured to: the obtained three-dimensional laser scanning data is preprocessed to obtain the point cloud data corresponding to the physical object, the obtained design data is constructed into a BIM model and converted into the point cloud data corresponding to the model, the two parts of data are subjected to point cloud registration through a registration algorithm, and the error between the two parts of data is obtained through comparison. According to the obtained error information, the actual quality of the steel component is determined. Among them, the registration algorithm includes a rough registration stage and a fine registration stage. In the rough registration stage, the best rigid transformation corresponding to the points in the target point cloud is found by randomly sampling points in the source point cloud; the target point cloud is divided into multiple small voxels, and the point cloud within each voxel is regarded as a Gaussian distribution, and the best rigid transformation is found by optimizing the overlapping degree of the Gaussian distributions between the source point cloud and the target point cloud. In the fine registration stage, the nearest point on the surface of the target point cloud for each point in the source point cloud is found through nearest neighbor search; the stop timing of the registration process is determined by setting an error convergence threshold; the normal vector angle threshold is set to determine whether the two points found are valid matching points; the rotation and translation matrices that minimize the error between the source point cloud and the target point cloud are found to form error information.

8. The complex steel structure construction quality inspection system based on 3D laser scanning according to claim 7, characterized in that It also has a result interaction platform for receiving and saving the point cloud data corresponding to the physical object, the point cloud data corresponding to the model, and the corresponding comparison results in the processing quality inspection stage, virtual pre-assembly stage, and on-site actual assembly stage.

9. A computer-readable storage medium, characterized in that, Stored thereon is a computer program which, when executed by a processor, implements the steps in the method for detecting the construction quality of a complex steel structure based on three-dimensional laser scanning according to any one of claims 1-7.

10. A computer device, characterized in that, It includes a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the program, it implements the steps in the method for detecting the construction quality of a complex steel structure based on three-dimensional laser scanning according to any one of claims 1-7.

Images