High-precision steel structure geometric dimension measuring and positioning method

Abstract

The invention relates to the technical field of steel structure measurement, discloses a high-precision steel structure geometric dimension measuring and positioning method, and aims to solve the problems that an existing measuring means is low in efficiency, limited in coverage range, poor in key feature extraction robustness, difficult to adapt to a complex field environment and lack of a real-time comparison feedback mechanism. The method comprises the following steps: deploying a multi-modal sensing system, and synchronously acquiring laser point cloud and binocular vision data; adaptive denoising and multi-view point cloud splicing are executed; fusing the visual edge and the point cloud curvature features to generate a high-confidence geometric feature set; reconstructing a local fine three-dimensional model based on a moving least square method; dynamically comparing the actual measurement model with the BIM design model to generate a full-surface deviation distribution diagram; and identifying the out-of-tolerance area and outputting a three-dimensional deviation correction instruction. By the adoption of the technical scheme, non-contact measurement and closed-loop correction with millimeter-level precision can be achieved, and the one-time success rate and the intelligent construction level of steel structure installation are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of steel structure measurement technology, specifically to a high-precision method for measuring and positioning the geometric dimensions of steel structures. Background Technology

[0002] With the rapid development of large-scale infrastructure and super high-rise buildings, steel structures are widely used in modern construction engineering due to their high strength, high rigidity, and industrialized assembly advantages. The geometrical accuracy and spatial positioning accuracy of steel structure components directly affect the overall structural safety, construction efficiency, and subsequent operation and maintenance costs. In actual engineering projects, factors such as component manufacturing deviations, transportation deformation, and on-site installation errors often lead to assembly difficulties and even major safety accidents. Therefore, high-precision and high-efficiency geometrical measurement and three-dimensional spatial positioning of steel structures have become a key link in ensuring project quality and construction progress.

[0003] Among them, the high-precision steel structure geometric dimension measurement and positioning method aims to achieve rapid acquisition and calibration of the outline, key control point coordinates, and relative pose of complex steel components through non-contact sensing and intelligent data processing technology. This technology needs to integrate multi-source sensing information and overcome adverse factors such as obstruction, lighting changes, and vibration interference in complex construction site environments to meet the measurement accuracy requirements at the millimeter or even sub-millimeter level.

[0004] However, existing measurement methods mostly rely on single-point scanning equipment such as total stations or laser trackers, which suffer from low measurement efficiency, limited coverage, and difficulty in obtaining complete surface information. Meanwhile, traditional image or point cloud processing algorithms lack robustness in extracting key features such as steel structure edges, welds, and connection nodes, and are easily affected by surface reflections, corrosion, or background clutter, leading to significant positioning deviations. Furthermore, existing systems lack real-time dynamic comparison and error feedback mechanisms between measurement data and design models, failing to provide timely guidance and correction during construction, thus hindering the improvement of intelligent steel structure construction. Therefore, a high-precision method for measuring and positioning the geometric dimensions of steel structures that integrates multimodal perception and intelligent analysis is urgently needed. Summary of the Invention

[0005] The purpose of this invention is to provide a high-precision method for measuring and positioning the geometric dimensions of steel structures, which can effectively solve the problems mentioned in the background art, such as low efficiency, limited coverage, poor robustness of key feature extraction, difficulty in adapting to complex field environments, and lack of real-time comparison and feedback mechanisms of existing measurement methods.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A high-precision method for measuring and positioning the geometric dimensions of steel structures, including the following specific steps: Step 1: Deploy a multimodal sensing system. At least three high-resolution 3D laser scanners and two sets of binocular vision cameras are deployed around the steel structure to be measured. Each device achieves nanosecond-level clock alignment through a time synchronization protocol. The single-point ranging accuracy of the scanner is better than 0.5 mm, the field of view coverage is not less than 270 degrees, and the spatial resolution of the binocular cameras reaches 0.3 mm per meter, which together form a surround non-contact sensing network. Step 2: Collect multi-source spatial data. Start synchronous scanning while the structural components are stationary to obtain raw laser point cloud data and stereo image sequences. The point cloud density is no less than 5,000 points per square meter, the image resolution is 4,096 pixels multiplied by 2,160 pixels, and the frame rate is stable at 30 frames per second. All data are initially registered according to a unified spatiotemporal coordinate system and transmitted to the central processing unit. Step 3: Perform point cloud enhancement processing. Use an adaptive denoising algorithm based on local plane fitting to eliminate outliers. The neighborhood search radius is dynamically set from 10 mm to 30 mm. Combine the normal consistency judgment criterion to remove non-structured surface noise. Then, use the ICP iterative nearest point algorithm accelerated by KD tree to accurately stitch together the multi-view point clouds. The registration error is controlled within 0.3 mm. Step 4: Fuse visual and point cloud features, extract edge segments and corner information from the stereo image, detect weld contours and connection plate boundaries through sub-pixel level Canny operator and Hough transform, back-project the extracted two-dimensional features to three-dimensional space and cross-validate with the point cloud curvature abrupt change region to generate a high-confidence geometric feature set. Step 5: Construct a local fine model. Based on the enhanced point cloud and fusion features, the moving least squares method is used to accurately reconstruct the continuous geometric shape of the component surface to reflect the actual situation of the component. Local fine modeling is implemented for key areas such as beam and column ends and node domains. The mesh resolution is increased to 5 mm to form a field-measured 3D model with millimeter-level detail restoration capability. Step 6: Perform dynamic comparison of the design model, import the theoretical design geometric data in the Building Information Model (BIM), convert it into a coordinate reference consistent with the local fine model, and use an octree-based spatial hash matching strategy to construct a deviation distribution map. The maximum matching time is no more than 8 seconds. Step 7: Identify dimensional deviations and pose errors, extract out-of-tolerance areas based on the deviation distribution map, and detect the actual values ​​of geometric parameters; the geometric parameters include component length, cross-sectional height, torsion angle, and installation offset, which provide key basis for subsequent output of positioning correction commands. When any dimensional deviation exceeds 5 mm or the relative angle deviation is greater than 0.15 degrees, an alarm signal is triggered. Step 8: Output positioning and correction commands, visualize the detection results of geometric parameters and overlay them on the digital twin platform of the construction site to generate a three-dimensional guidance map containing the deviation vector direction, correction suggestion path and tool guidance coordinates. Push the map to the construction terminal equipment through the wireless communication module to drive the hydraulic adjustment device or robot-assisted system to perform automatic correction.

[0007] Preferably, in step 1, the 3D laser scanner uses the phase difference ranging principle, with a modulation frequency of 100 MHz, a ranging range of 0.5 meters to 120 meters, a horizontal scanning step angle of 0.009 degrees, and an adjustable vertical tilt angle range of -10 degrees to +30 degrees. Let the modulation frequency be... Distance measurement Satisfy the formula Where c is the speed of light. This represents the phase difference.

[0008] Preferably, in step 2, the binocular vision camera is equipped with a global shutter CMOS sensor, a baseline length of 250 mm, a focal length of 12 mm, and a lens coated with an anti-glare and anti-reflection coating. It can stably image within an illumination range of 50 lux to 100,000 lux. For depth calculation in binocular vision, depth... Satisfy the formula ,in and These are the x-coordinates of the corresponding pixels in the left and right images, respectively.

[0009] Preferably, in step 3, the adaptive denoising algorithm dynamically adjusts the size of the filtering window according to the local curvature change rate of the point cloud. A larger radius is used for smoothing in planar areas, while the search range is reduced around edges and holes to preserve geometric sharpness. The overall data simplification rate is controlled within 40%.

[0010] Preferably, the cross-validation process in step 4 introduces a weighted scoring mechanism, assigning a weight of 0.7 to areas with significant point cloud curvature and a weight of 0.6 to areas with good image edge continuity, and increasing the confidence of the intersection of the two to above 0.9, which is used as the key constraint point set for subsequent model reconstruction.

[0011] Preferably, in step 5, the support radius of the moving least squares method is set to 15 mm, the polynomial fitting order is 2, and the parameterized template is automatically matched to the H-beam or box column to constrain the reconstruction process.

[0012] Preferably, in step 6, the building information model (BIM) conversion retains the component hierarchy and material attribute labels, and the initial coarse registration is completed by extracting the principal axis direction and centroid coordinates.

[0013] Preferably, in step 7, the deviation statistics process distinguishes between manufacturing errors and installation errors, establishes a database of mean and standard deviation for similar components produced in batches, and determines an abnormal component and marks it with a traceability number when the individual deviation exceeds the mean plus twice the standard deviation.

[0014] Preferably, in step 8, the three-dimensional guidance map displays the degree of deviation using a color heatmap. Red indicates out-of-tolerance areas with deviations greater than 5 mm, yellow indicates near-threshold areas with deviations between 3 mm and 5 mm, and green indicates qualified areas with deviations not exceeding 3 mm. Specific values ​​and units are also indicated.

[0015] Preferably, it also includes: continuously collecting dynamic point cloud sequences during the steel structure hoisting process, using a motion estimation algorithm based on Kalman filtering to track component pose changes, predicting the placement posture and adjusting the sling tension in advance, and realizing closed-loop monitoring of the entire process.

[0016] Preferably, it also includes: continuously collecting dynamic point cloud sequences during the steel structure hoisting process, using a motion estimation algorithm based on Kalman filtering to track component pose changes, predict the placement posture, and adjust the sling tension in advance; or establishing a measurement data traceability archive, with each record associated with a timestamp, weather conditions, equipment status, and operator number, supporting queries by project location, component number, or multi-dimensional condition combination, with a storage period of no less than 10 years.

[0017] Compared with the prior art, the beneficial effects achieved by the present invention are: This invention achieves efficient and high-precision non-contact measurement of the geometric dimensions and spatial pose of large steel structures by constructing a multimodal sensing fusion system. It overcomes the limitations of single sensors under conditions of occlusion, lighting fluctuations, and vibration interference, significantly improving the reliability of key feature extraction in complex environments. Adaptive point cloud enhancement and multi-source feature cross-validation mechanisms effectively suppress the effects of surface corrosion, reflection, and background noise, ensuring the stability of millimeter-level measurement accuracy. By introducing a real-time dynamic comparison process between the on-site measured model and the BIM design model, a complete closed loop is established from data acquisition to deviation identification and correction guidance, enabling construction errors to be detected and corrected immediately, significantly improving the first-time assembly success rate. This method supports automated data processing and visual feedback, reducing reliance on highly skilled surveyors and enhancing the intelligent construction level of steel structure engineering. It is applicable to various major engineering scenarios such as bridges, towers, stadiums, and super high-rise buildings, demonstrating significant technological advancement and engineering application value. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the overall technical solution architecture of the method of the present invention; Figure 2This is a schematic diagram of the core principle framework of multimodal sensing fusion and feature cross-validation in this invention; Figure 3 This is a logical flowchart of the dynamic comparison and deviation identification between the on-site measured model and the BIM design model in this invention. Figure 4 This is a schematic diagram of the data flow of measurement-feedback-correction closed-loop control and digital twin visualization interaction in this invention. Detailed Implementation

[0019] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0020] Example 1 Currently, with the rapid development of large-scale infrastructure and super high-rise buildings, steel structures are widely used in modern construction engineering due to their high strength, high rigidity, and industrialized assembly advantages. The geometric dimensional accuracy and spatial positioning accuracy of steel structure components directly affect the overall structural safety, construction efficiency, and subsequent operation and maintenance costs. In actual engineering projects, factors such as component manufacturing deviations, transportation deformation, and on-site installation errors often lead to assembly difficulties and even major safety accidents. Therefore, high-precision and high-efficiency geometric dimensional measurement and three-dimensional spatial positioning of steel structures have become a key link in ensuring project quality and construction progress. To address the above technical problems, this invention proposes to construct a multimodal sensing fusion system to achieve efficient and high-precision non-contact measurement of the geometric dimensions and spatial pose of large steel structures. This overcomes the limitations of single sensors under conditions of occlusion, light fluctuations, and vibration interference, significantly improves the reliability of key feature extraction in complex environments, and establishes a complete closed loop from data acquisition to deviation identification and correction guidance. This technical solution is applied to a high-precision steel structure geometric dimensional measurement and positioning method.

[0021] Reference Appendix Figure 1 The overall technical architecture of the high-precision steel structure geometric dimension measurement and positioning method proposed in this invention includes core functional modules such as multimodal sensing system deployment, multi-source spatial data acquisition, point cloud enhancement processing, visual and point cloud feature fusion, local fine model construction, dynamic comparison of design models, identification of dimensional deviations and pose errors, and positioning correction command output. The entire system is centered on a central processing unit, which coordinates and controls various sensing devices and actuators to form a closed-loop feedback mechanism.

[0022] In the aforementioned high-precision steel structure geometric dimension measurement and positioning method, step 1 involves deploying a multimodal sensing system. At least three high-resolution 3D laser scanners and two sets of binocular vision cameras are arranged around the steel structure to be measured. Each device achieves nanosecond-level clock alignment through a time synchronization protocol. The scanner's single-point ranging accuracy is better than 0.5 mm, and its field of view coverage is not less than 270 degrees. The binocular cameras achieve a spatial resolution of 0.3 mm per meter, collectively forming a surround-type non-contact sensing network. Specifically, the 3D laser scanner uses the phase difference ranging principle, with a modulation frequency of 100 MHz, a ranging range of 0.5 meters to 120 meters, a horizontal scanning step angle of 0.009 degrees, and an adjustable vertical tilt angle range of -10 degrees to +30 degrees. The modulation frequency is set to... Distance measurement Satisfy the formula ,in At the speed of light, To mitigate phase difference, accurate long-distance capture of the high-rise steel frame is ensured. The binocular vision camera is equipped with a global shutter CMOS sensor, a baseline length of 250 mm, a focal length of 12 mm, and a lens coated with an anti-glare and anti-reflection coating. It can stably image within an illumination range of 50 lux to 100,000 lux. For depth calculation using binocular vision, depth... Satisfy the formula ,in and The left and right images are respectively the pixel coordinates of corresponding points, effectively suppressing overexposure caused by direct sunlight. Each device achieves hardware-level clock synchronization via the IEEE 1588 precision time protocol, with synchronization jitter less than 10 nanoseconds, ensuring strict consistency of multi-source data in the spatiotemporal dimension. The sensor deployment strategy is optimized based on the geometric complexity of the steel structure under test and the distribution of obstacles in the environment, adopting a design principle of coverage redundancy of no less than 1.5 times to ensure that any key feature point is simultaneously observed by at least two scanners and one set of binocular cameras, thereby eliminating data loss due to occlusion. The equipment support is made of carbon fiber composite material, possessing high rigidity and a low coefficient of thermal expansion, avoiding measurement reference drift caused by temperature changes or wind loads.

[0023] In the aforementioned high-precision steel structure geometric dimension measurement and positioning method, step 2 involves acquiring multi-source spatial data, initiating synchronous scanning while the structural components are stationary, obtaining raw laser point cloud data and stereo image sequences, with a point cloud density of no less than 5000 points per square meter, an image resolution of 4096 pixels multiplied by 2160 pixels, and a frame rate stable at 30 frames per second. All data is initially registered according to a unified spatiotemporal coordinate system and transmitted to the central processing unit. Specifically, before scanning begins, the system first verifies the status of each sensor, including laser power stability, camera exposure consistency, and synchronization signal integrity. During scanning, the 3D laser scanner emits a modulated laser beam at a rate of 2 million points per second. After the receiver captures the reflected signal, it calculates the target point distance based on the phase difference and combines the horizontal and vertical encoder angle information to calculate the 3D coordinates. The binocular vision cameras are synchronously triggered, with the left and right cameras capturing stereo image pairs of the same scene respectively. Depth information is recovered through epipolar constraints and parallax calculations. The raw point cloud data is stored in binary format, containing the X, Y, and Z coordinates, intensity value, and timestamp for each point; the stereo image sequence is saved in a lossless compression format, preserving complete color and texture information. All data streams are transmitted in real time to the central processing unit via gigabit Ethernet, and based on the equipment's factory calibration parameters and field calibration board data, the coordinate systems of each sensor are transformed to the preset global engineering coordinate system, completing the initial registration. This process uses a quaternion-based rigid body transformation matrix to ensure the numerical stability of the coordinate transformation.

[0024] In the aforementioned high-precision steel structure geometric dimension measurement and positioning method, step 3 involves performing point cloud enhancement processing. An adaptive denoising algorithm based on local plane fitting is used to eliminate outliers. The neighborhood search radius is dynamically set from 10 mm to 30 mm. Combined with the normal consistency judgment criterion, non-structural surface noise points are eliminated. Subsequently, the multi-view point cloud is precisely stitched together using the ICP iterative nearest-neighbor algorithm accelerated by KD trees, with the registration error controlled within 0.3 mm. Specifically, the adaptive denoising algorithm first calculates the local curvature change rate of each point. This index is obtained by fitting the covariance matrix of its k nearest neighbors (k=30) and analyzing the eigenvalues. In planar regions where the curvature change rate is below the threshold of 0.01, the neighborhood search radius is expanded to 30 mm to achieve smoother noise suppression; in edges or around holes where the curvature change rate is above 0.1, the search radius is reduced to 10 mm to preserve geometric sharpness. The normal consistency judgment criterion requires that the angle between the normal vector of a candidate point and the average normal vector of most points in its neighborhood is less than 15 degrees; otherwise, it is considered an outlier and eliminated. After denoising, the point cloud data simplification rate was controlled within 40%, effectively balancing data volume and feature fidelity. In the multi-view point cloud stitching stage, initial coarse registration was first performed by extracting ISS keypoints from each viewpoint and calculating FPFH descriptors; subsequently, a KD-tree-accelerated ICP algorithm was used for fine registration, improving the corresponding point search efficiency by more than an order of magnitude. The ICP iteration termination condition was set at a root mean square error change of less than 0.05 mm over three consecutive iterations, and the final stitching registration error was strictly controlled within 0.3 mm.

[0025] In the aforementioned high-precision steel structure geometric dimension measurement and positioning method, step 4 involves fusing visual and point cloud features to extract edge segments and corner information from the stereo image. The weld contour and connecting plate boundary are detected using a sub-pixel-level Canny operator and Hough transform. The extracted two-dimensional features are then back-projected into three-dimensional space and cross-validated with the point cloud curvature abrupt change region to generate a high-confidence geometric feature set. Specifically, the sub-pixel-level Canny operator first applies Gaussian filtering (σ=1.0) to the stereo image, calculates the gradient magnitude and direction, then determines edge candidate points through non-maximum suppression and double threshold detection, and finally uses interpolation to improve the edge positioning accuracy to the 0.1 pixel level. The Hough transform is used to cluster discrete edge points into continuous straight line segments, with a parameter space resolution set to ρ=0.5 pixels and θ=0.5 degrees. The detected weld contour and connecting plate boundary, as two-dimensional features, are back-projected into three-dimensional space using the depth map from a binocular camera to obtain the corresponding three-dimensional line segment and corner point set. Meanwhile, the point cloud data is used to identify regions of abrupt curvature changes (curvature values ​​greater than 0.05) by calculating the curvature tensor of each point. A weighted scoring mechanism is introduced into the feature cross-validation process: regions with significant point cloud curvature are assigned a weight of 0.7, and regions with good image edge continuity (line segment length greater than 50 pixels) are assigned a weight of 0.6. The confidence level of the spatially overlapping intersection of these two regions is increased to above 0.9. This high-confidence geometric feature set serves as a key constraint point for subsequent model reconstruction, significantly improving the robustness of feature extraction and effectively overcoming interference from surface corrosion, reflections, and background clutter.

[0026] In the aforementioned high-precision steel structure geometric dimension measurement and positioning method, step 5 involves constructing a local fine model. Based on the enhanced point cloud and fusion features, the moving least squares method is used to reconstruct the continuous geometric shape of the component surface. Local densification modeling is implemented for key areas such as beam and column ends and node regions, increasing the mesh resolution to 5 mm, forming a field-measured 3D model with millimeter-level detail restoration capability. Specifically, the support radius of the moving least squares method is set to 15 mm, and the polynomial fitting order is 2. This parameter combination achieves the best balance between smoothness and detail preservation. During the reconstruction process, the system automatically identifies known standard section types, such as H-beams and box columns, and constrains the geometric shape during the reconstruction process by matching pre-stored parametric templates, avoiding model distortion caused by missing local data. For key areas such as beam and column ends, bolt hole groups, and welded node regions, the system automatically delineates local densification areas based on the fusion feature set, increasing the mesh resolution from the global 10 mm to 5 mm to ensure accurate restoration of millimeter-level details. The reconstructed surface model is stored in the form of a triangular mesh in OBJ format, containing vertex coordinates, normal vectors, and texture coordinates. The mesh topology meets the watertightness requirements and has no holes or self-intersections.

[0027] In the aforementioned high-precision steel structure geometric dimension measurement and positioning method, step 6 involves performing dynamic comparison of the design model, importing theoretical design geometric data from the Building Information Model (BIM), converting it into a coordinate reference consistent with the measured model, employing an octree-based spatial hash matching strategy, calculating the bidirectional Euclidean distance field voxel by voxel, and generating a full-surface deviation distribution map. The maximum matching time does not exceed 8 seconds. (See attached reference.) Figure 3 Specifically, the BIM model is imported in IFC format, and the system parses its geometric entities and attribute information, preserving the component hierarchy and material attribute labels. Coordinate datum transformation first extracts the principal axis directions (through PCA principal component analysis) and centroid coordinates of the measured and theoretical models, calculating the initial coarse registration transformation matrix to reduce the number of iterations for subsequent fine matching. Then, the two models are embedded into a unified octree spatial index structure, with the voxel size set to 2 mm. A spatial hash matching strategy quickly locates potential corresponding voxels using a hash table, avoiding global traversal. The formula for calculating the bidirectional Euclidean distance field is as follows: in, For designing the model, This is a measured model. The points represent the measured points on the actual model. This formula considers both the distance from the measured point to the design surface and the distance from the design surface to the measured point, effectively avoiding misjudgments of the unidirectional distance field in concave regions. The calculation results are mapped onto the surface of the measured model in chromatographic form, generating a full-surface deviation distribution map. The entire matching process, accelerated by GPU, takes no more than 8 seconds at most.

[0028] In the above-mentioned high-precision steel structure geometric dimension measurement and positioning method, step (7) identifies dimensional deviations and pose errors, extracts out-of-tolerance areas based on the deviation distribution map, and statistically analyzes the actual values ​​of geometric parameters such as component length, cross-sectional height, torsion angle, and installation offset, comparing them with the design allowable tolerance zone. When any dimensional deviation exceeds 5 mm or the relative angle deviation is greater than 0.15 degrees, an alarm signal is triggered. Specifically, the out-of-tolerance area is defined as a continuous surface area with an absolute deviation greater than 5 mm. The system automatically extracts the boundaries of these areas and calculates their area, volume, and spatial position. The geometric parameter statistical process distinguishes between manufacturing errors and installation errors: manufacturing errors refer to the deviation between the component's own geometric dimensions (such as length and cross-sectional height) and the design value; installation errors refer to the deviation between the component's pose in space (such as torsion angle and installation offset) and the design position. For similar components produced in batches, the system establishes a mean and standard deviation database to record the statistical characteristics of historical measurement data. When the deviation of an individual component exceeds the mean plus twice the standard deviation, the system determines it to be an abnormal part and automatically marks it with a unique traceability number for easy quality traceability. Alarm signals are triggered through both audible and visual prompts and software interface pop-ups to ensure timely response from operators.

[0029] In the above-mentioned high-precision steel structure geometric dimension measurement and positioning method, step 8, outputting positioning correction commands, refers to the attached... Figure 4 The detection results are visualized and overlaid on a digital twin platform at the construction site, generating a 3D guidance map that includes the deviation vector direction, suggested correction path, and tool guidance coordinates. This map is then pushed to the construction terminal equipment via a wireless communication module, driving the hydraulic adjustment device or robot-assisted system to perform automatic correction. Specifically, the 3D guidance map displays the degree of deviation as a color heatmap: red indicates out-of-tolerance areas (deviation > 5 mm), yellow indicates areas near the threshold (3 mm < deviation ≤ 5 mm), and green indicates acceptable areas (deviation ≤ 3 mm), with specific values ​​and units (millimeters or degrees) marked at key locations. The deviation vector direction is visually indicated by arrows, the suggested correction path is generated based on the shortest path algorithm and on-site obstacle avoidance strategies, and the tool guidance coordinates precisely specify the target point of action for the hydraulic jack or robot end effector. All information is pushed in real time to construction terminal equipment such as tablets and AR glasses via a 5G wireless communication module. In automated scenarios, commands can directly drive the hydraulic adjustment device or ABB IRB 1200 industrial robot to perform automatic correction, forming a closed-loop control system without human intervention.

[0030] To further enhance the practicality of the method, this invention also includes continuously acquiring dynamic point cloud sequences during the steel structure hoisting process, employing a Kalman filter-based motion estimation algorithm to track component pose changes, predict the placement attitude, and adjust the sling tension in advance, achieving closed-loop monitoring throughout the entire process. The dynamic point cloud acquisition frequency is increased to 10 Hz, and the Kalman filter state vector includes six degrees of freedom parameters such as position, velocity, and angular velocity. The covariance matrix of process noise and observation noise is dynamically adjusted according to the on-site wind speed and hoisting speed. The error in predicting the placement attitude converges to within 2 mm within 100 milliseconds, providing a reliable basis for the advance adjustment of sling tension.

[0031] In addition, the present invention also includes the establishment of a measurement data traceability archive, with each record associated with a timestamp, weather conditions (temperature, humidity, wind speed), equipment status (battery power, laser power) and operator number, supporting queries by project location, component number, deviation type or multi-dimensional condition combination, and all data is encrypted and stored on a cloud server for a storage period of no less than 10 years, meeting the quality management needs of the entire life cycle of the project.

[0032] The method described in this invention was successfully applied to the installation of the core steel frame of a 300-meter super high-rise building. A single site deployment covered eight floors (approximately 32 meters in height), using four 3D laser scanners and three sets of binocular vision cameras. The total measurement time for the entire building was reduced by 70% compared to traditional total station operations, the consistency of repeated measurements reached 0.4 mm, 12 out-of-tolerance installation points were successfully identified and corrected, and the first-time assembly success rate increased to 98.5%.

[0033] Example 2 Based on Example 1, the present invention also provides an alternative implementation method suitable for the assembly of steel truss segments for long-span bridges. (See attached document.) Figure 2 The core of this implementation lies in the targeted optimization of multimodal sensing fusion and feature cross-validation mechanisms.

[0034] In this scenario, the object under test is a 30-meter-long steel truss segment, characterized by numerous slender members connected to complex node plates, and a surface with dense welds and temporary support fixtures. To accommodate this type of structure, step 1 employs a linear array layout for sensor deployment, with five 3D laser scanners evenly spaced 6 meters apart along the truss length to ensure full coverage of the slender members; the number of binocular vision cameras is increased to four sets, focusing on covering the node areas where the upper and lower chords intersect with the web members. The vertical tilt angle of the scanners is adjusted from -15 degrees to +45 degrees to capture the hidden areas at the bottom of the truss.

[0035] In the feature fusion stage of step 4, to address the issue of image blurring caused by vibration in slender rods, the system introduces an LSTM-based temporal image stabilization algorithm. This algorithm uses a sequence of five consecutive stereo images to predict and compensate for pixel displacement caused by micro-vibrations, improving the stability of edge detection. Simultaneously, the point cloud curvature calculation employs an anisotropic kernel function to differentially weight the curvature changes along the axial and radial directions of the rod, more accurately identifying the rod's centerline.

[0036] In the model reconstruction step 5, the system automatically identifies the standard member types (round tube, square tube, H-beam) and calls the corresponding parametric cross-section template. The reconstruction process not only outputs the surface mesh but also generates a member centerline skeleton model, facilitating subsequent accurate calculations of length and straightness. The mesh resolution is set to 8 mm on the member surface and refined to 4 mm in the node plate area.

[0037] In the model comparison in step 6, the axis information of the members is specifically preserved during the BIM model conversion. Deviation calculation not only considers surface distances but also calculates the Hausdorff distance between the measured centerline and the design axis, using the following formula: in, This is the measured centerline. This is for designing the axis. This indicator can more comprehensively reflect the overall deformation of slender components.

[0038] In the correction command of step 8, considering the need for multi-point synchronous adjustment of truss segments, the 3D guidance diagram clearly marks the coordinated action sequence and force distribution of multiple hydraulic jacks to ensure the stability of the segments during translation and rotation. This implementation method was applied in a Yangtze River Bridge project, where the measurement and correction time for a single segment was controlled within 15 minutes, and the assembly accuracy met the stringent requirement of 2 mm.

[0039] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A high-precision method for measuring and positioning the geometric dimensions of steel structures, characterized in that: The specific steps include the following: Step 1: Deploy a multimodal sensing system, placing at least 3 high-resolution 3D laser scanners and 2 sets of binocular vision cameras around the steel structure to be tested; Step 2: Collect multi-source spatial data. Start synchronous scanning while the steel structure components are stationary to obtain raw laser point cloud data and stereo image sequences. Step 3: Perform point cloud enhancement processing, using an adaptive denoising algorithm based on local plane fitting to eliminate outliers in the original laser point cloud data; Step 4: Fuse visual and point cloud features to extract edge segments and corner information from the stereo image, and detect weld contours and connection plate boundaries using sub-pixel level Canny operator and Hough transform; Step 5: Construct a local fine model. Based on the enhanced point cloud and fused features, use the moving least squares method to accurately reconstruct the continuous geometric shape of the component surface to reflect the actual situation of the component. Step 6: Perform dynamic comparison of the design model, import the theoretical design geometric data from the Building Information Model (BIM), convert it into a coordinate reference consistent with the local fine model, and construct a deviation distribution map using a spatial hash matching strategy based on octree partitioning. Step 7: Identify dimensional deviations and pose errors, extract out-of-tolerance areas based on the deviation distribution map, and detect the actual values ​​of geometric parameters; the geometric parameters include component length, cross-sectional height, torsion angle, and installation offset, providing key basis for subsequent output of positioning correction commands; Step 8: Output positioning and correction commands, visualize the detection results of geometric parameters and overlay them on the construction site digital twin platform to generate a 3D guidance map containing the deviation vector direction, correction suggestion path and tool guidance coordinates.

2. The high-precision steel structure geometric dimension measurement and positioning method according to claim 1, characterized in that: The 3D laser scanner employs the phase difference ranging principle, with a modulation frequency of 100 MHz, a ranging range of 0.5 meters to 120 meters, a horizontal scanning step angle of 0.009 degrees, and an adjustable vertical tilt angle ranging from -10 degrees to +30 degrees. The modulation frequency is set to... Distance measurement Satisfy the formula ,in At the speed of light, This represents the phase difference.

3. The high-precision steel structure geometric dimension measurement and positioning method according to claim 1, characterized in that: The binocular vision camera is equipped with a global shutter CMOS sensor, a baseline length of 250 mm, a focal length of 12 mm, and a lens coated with an anti-glare and anti-reflection coating. It can stably image within an illumination range of 50 lux to 100,000 lux. For depth calculation in binocular vision, depth... Satisfy the formula ,in and These are the x-coordinates of the corresponding pixels in the left and right images, respectively.

4. The high-precision steel structure geometric dimension measurement and positioning method according to claim 1, characterized in that: The adaptive denoising algorithm dynamically adjusts the neighborhood search radius based on the local curvature change rate of the point cloud, using a radius of 30 mm in planar areas and a radius of 10 mm around edges and holes, keeping the overall data simplification rate below 40%.

5. The high-precision steel structure geometric dimension measurement and positioning method according to claim 1, characterized in that: The feature cross-validation process introduces a weighted scoring mechanism, assigning a weight of 0.7 to areas with significant point cloud curvature and a weight of 0.6 to areas with good image edge continuity, increasing the confidence of the intersection of the two to over 0.

9.

6. The high-precision steel structure geometric dimension measurement and positioning method according to claim 1, characterized in that: The moving least squares method has a support radius of 15 mm, a polynomial fitting order of 2, and automatically matches parametric templates to H-beams or box columns to constrain the reconstruction process.

7. The high-precision steel structure geometric dimension measurement and positioning method according to claim 1, characterized in that: The building information model (BIM) conversion retains the component hierarchy and material attribute labels, and the initial coarse registration is completed by extracting the principal axis direction and centroid coordinates.

8. The high-precision steel structure geometric dimension measurement and positioning method according to claim 1, characterized in that: The deviation statistics process distinguishes between manufacturing errors and installation errors. A database of mean and standard deviation is established for similar components produced in batches. When an individual deviation exceeds the mean plus twice the standard deviation, it is judged as an abnormal part and marked with a traceability number.

9. The high-precision steel structure geometric dimension measurement and positioning method according to claim 1, characterized in that: The three-dimensional guidance diagram uses a color heatmap to display the degree of deviation. Red indicates out-of-tolerance areas with deviations greater than 5 mm, yellow indicates near-threshold areas with deviations between 3 mm and 5 mm, and green indicates qualified areas with deviations not exceeding 3 mm. Specific values ​​and units are also indicated.

10. The high-precision steel structure geometric dimension measurement and positioning method according to claim 1, characterized in that: It also includes continuously collecting dynamic point cloud sequences during the steel structure hoisting process, using a motion estimation algorithm based on Kalman filtering to track component pose changes, predict placement posture, and adjust sling tension in advance; or establishing a measurement data traceability archive, with each record associated with a timestamp, weather conditions, equipment status, and operator number, supporting queries by project location, component number, or multi-dimensional condition combination, with a storage period of no less than 10 years.

Images