Assembly type building prefabricated component concealed engineering detection robot and detection method thereof

Abstract

The application provides an assembled building prefabricated component concealed engineering detection robot and a detection method thereof, which comprises a track frame of a wide frame structure, transverse track beams are arranged between the two sides of the top of the track frame, a mold table for placing a detected object is arranged at the lower part of the track frame, the two ends of the transverse track beams are slidably connected with the two sides of the top of the track frame, at least one vertical lifting frame is arranged on the transverse track beam, a two-dimensional holder is arranged at the end of the vertical lifting frame, and a structured light camera is arranged on the two-dimensional holder; the detected object is arranged on the mold table, and the two-dimensional holder performs visual detection on the detected object. The two-dimensional holder and the structured light camera are carried by the truss type robot to scan the concealed engineering structure of the prefabricated component, and the characteristics of the structure are extracted by using point cloud data and compared with a BIM model.

Description

Technical Field

[0001] This invention relates to the field of prefabricated component inspection for prefabricated buildings, and in particular to a robot and method for inspecting concealed works of prefabricated components for prefabricated buildings. Background Technology

[0002] In the production process of prefabricated components for prefabricated buildings, the concealed works inspection process precedes the concrete pouring process. This mainly involves checking whether embedded parts, reinforcing bars, junction boxes, and other concealed structural components are missing or incorrectly positioned. Furthermore, as prefabricated components are the core building material, the quality of their concealed works directly affects the overall construction quality; therefore, the concealed works inspection process is crucial. Currently, the inspection of concealed works for prefabricated components typically relies on manual methods. Checking for missing components requires workers to verify them against the drawings, and checking installation dimensions requires manual measurement with a measuring tape. Each workstation alone requires at least two people. Manual inspection is highly dependent on the experience and skill level of workers, making it prone to missed or false positives, resulting in low accuracy and quality. Manual inspection requires workers to be on-site at production lines, leading to high labor intensity and potential safety hazards. Furthermore, the process is cumbersome and inefficient, making it difficult to meet the fixed production rhythm of a production line and unsuitable for the testing needs of large-scale prefabricated building component production. Additionally, manual inspection typically only records the results, failing to reconstruct the specific characteristics of the inspected object, and making data traceability virtually impossible even if problems arise later.

[0003] Therefore, to improve the accuracy, efficiency, and quality of inspection of concealed works in prefabricated components and reduce the labor intensity of workers, a robot for inspecting concealed works of prefabricated building components is proposed. This robot, a truss-type robot equipped with a 2D gimbal and a structured light camera, scans the concealed structural elements of prefabricated components. It extracts structural features from point cloud data and compares them with a BIM model, achieving automated inspection of concealed works and thus improving production quality and efficiency. By storing the point cloud data and inspection results on a cloud platform, quality and data traceability can be achieved, improving the level of digital project management. Summary of the Invention

[0004] The main objective of this invention is to provide a robot and method for inspecting concealed works of prefabricated building components, which solves the problems of low inspection accuracy and quality, high labor intensity and safety hazards, low inspection efficiency and unsuitability for large-scale production, and difficulty in data traceability in the current manual inspection of concealed works of prefabricated building components. This invention lays the foundation for subsequent robot solutions to address these problems.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a robot for inspecting concealed works of prefabricated building components, including a wide-frame structure track frame, a transverse track beam between the two sides of the top of the track frame, a mold platform for placing the object to be inspected at the bottom of the track frame, the two ends of the transverse track beam being slidably connected to the two sides of the top of the track frame, at least one vertical lifting frame on the transverse track beam, a two-dimensional gimbal at the end of the vertical lifting frame, and a structured light camera on the two-dimensional gimbal; The object to be detected is placed on the platform, and a 2D gimbal performs visual detection on the object.

[0006] In the preferred embodiment, one side of the transverse track beam is slidably connected to the transverse sleeve, and a transverse servo motor is provided on one side of the transverse sleeve to drive the transverse sleeve to slide on one side of the transverse track beam. The vertical lifting frame is slidably connected to the transverse sleeve up and down. A vertical servo motor is provided on the other side of the horizontal moving sleeve, which drives the vertical lifting frame to move up and down.

[0007] In the preferred embodiment, a transverse track is provided on the crossbeam of the transverse track beam, the transverse sleeve is slidably connected to the transverse track, a transverse rack is provided on the crossbeam, the gear at the output end of the transverse servo motor meshes with the transverse rack, and the transverse servo motor drives the transverse sleeve to slide on the crossbeam.

[0008] In the preferred embodiment, the middle part of the transverse sleeve is a raised hollow rectangular frame structure, the vertical rod of the vertical lifting frame is provided with a vertical rail, the slider on the inner side of the transverse sleeve is slidably connected to the vertical rail, the vertical rod is also provided with a vertical rack, the gear on the vertical servo motor meshes with the vertical rack, and the vertical servo motor drives the vertical rod to slide up and down in the transverse sleeve.

[0009] In the preferred embodiment, the lower end of the vertical rod is connected to the base of the two-dimensional gimbal, the rotating ends on both sides of the two-dimensional gimbal body are connected to the two ends of the rotating frame, and the lower end of the rotating frame is connected to the structured light camera.

[0010] In the preferred embodiment, longitudinal sliders are provided at both ends of the transverse track beam. The longitudinal sliders are slidably connected to the longitudinal track on the longitudinal beam of the track frame. A longitudinal rack is provided on the longitudinal track. Both ends of the transverse track beam are provided with longitudinal servo motors. The gears at the output ends of the longitudinal servo motors mesh with the longitudinal racks. The two longitudinal servo motors drive the transverse track beam to slide on the longitudinal beam.

[0011] In the preferred embodiment, the longitudinal beam is supported by multiple columns and is positioned between the lower parts of the columns; Longitudinal railings are installed on both sides of the longitudinal beam; An electrical control cabinet is installed on the upper part of the transverse track beam.

[0012] The preferred solution also includes a cloud-based production collaborative control system. The cloud platform interfaces with the host computer of the production line and the host computer of the robot. The cloud platform has built-in BIM models of all prefabricated components and retrieves the corresponding BIM model according to the current inspection requirements and transmits it to the host computer of the robot. The robot's host computer includes a vision module and an equipment control module. The vision module is used for controlling the 2D gimbal, acquiring and processing point cloud data, and outputting quality inspection reports. The equipment control module is used to control longitudinal, lateral, and vertical movement.

[0013] The method includes: S1. Precast components are placed on the mold platform. The production line host computer sends a signal back to the cloud platform. After receiving the signal, the cloud platform sends a signal to the robot host computer to start the detection and transmits the BIM model of the precast components on the mold platform to the vision module of the robot host computer. S2. The robot's host computer's device control module sends signals to the longitudinal servo motor, the transverse servo motor, and the vertical servo motor. At the same time, the vision module sends control signals to the two-dimensional gimbal to reset the position of the structured light camera to the zero point O. S3. The equipment control module sends a signal to the longitudinal servo motor, driving the transverse track beam to move along the longitudinal beam from point O to shooting point A and stop, and sends the signal of reaching point A to the vision module. S4. The vision module sends acquisition signals to the 2D pan-tilt unit and structured light camera to acquire point cloud data of area A. After acquisition is completed, it sends an acquisition end signal to the device control module. S5. The equipment control module drives the transverse track beam to move from point A to the shooting point B and stop. The vision module controls the two-dimensional pan-tilt unit and structured light camera to collect point cloud data of the area at point B. After the collection is completed, a collection end signal is fed back. S6. The equipment control module drives the transverse track beam to move from point B to the shooting point C and stop. The vision module controls the completion of point cloud data acquisition in the area of ​​point C and sends back an end signal. S7. The equipment control module drives the transverse track beam to move from point C to the shooting point D and stop. The vision module controls the completion of point cloud data acquisition in the area of ​​point D and sends back an end signal. S8. The equipment control module drives the transverse track beam to move from point D to the shooting point E and stop. The vision module controls the completion of point cloud data acquisition in the area of ​​point E and sends back an end signal. S9. The equipment control module drives the transverse track beam from point E back to point O and stops, sending a signal to the cloud platform indicating the end of data acquisition. S10: The vision module performs post-processing on the collected point cloud data, extracts structural features and builds a model, compares the model with the BIM model, generates a quality inspection report and feeds it back to the cloud platform. S11. If the quality inspection result is qualified, the mold table will enter the next station; if it is unqualified, the worker will handle it according to the quality inspection report and the mold table will enter the next station. S12. The cloud platform stores the point cloud data and quality inspection report of this test.

[0014] In the preferred embodiment, the detection method's algorithm steps include: A1. Import the original point cloud data of the prefabricated component concealed works acquired by the structured light camera into the algorithm system. Through coordinate normalization processing, map the point cloud data to a preset three-dimensional coordinate system to eliminate data deviations caused by differences in coordinate systems at different shooting points. At the same time, record the acquisition time of each point cloud. Corresponding shooting locations and initial coordinates This serves as a foundation for subsequent data tracing and analysis; A2. Based on the local density distribution characteristics of point cloud data, the K nearest neighbor search algorithm is used to calculate the average distance of the K nearest neighbor points around each point cloud. 1.2-1.5 times the average distance is used as the local density threshold to achieve dynamic adaptation of the point cloud density threshold in different regions. A3. Compare the local density of each point cloud with the corresponding dynamic density threshold. If the local density of a point cloud is lower than the threshold, it is determined to be a noise point. Noise points include invalid point clouds caused by environmental dust and light reflection interference. Through the neighborhood point cloud collaborative verification mechanism, the density of the three neighboring points around the suspected noise point is verified a second time. After confirmation, the noise point is removed to avoid the accidental deletion of effective edge point clouds. A4. Using a curvature-based raster sampling method, the preprocessed point cloud data is divided into three-dimensional raster units with a raster size of 5-10 mm. Each raster is denoted as... u, v, w are the indices of the raster along the x, y, and z axes. Calculate the curvature of each point cloud within the raster. Curvature calculation requires first solving the covariance matrix of the point cloud neighborhood. The formula is: ; in , where is the number of neighborhood points selected when calculating curvature. The average coordinates of the neighboring points. Similarly, for the y and z axes, the covariance matrix... Perform eigenvalue decomposition to obtain eigenvalues. Point cloud curvature The formula is: ; The point cloud with the largest curvature in each grid is retained, and the rest of the redundant point clouds are deleted. While ensuring the integrity of the structural features, the amount of point cloud data is reduced by 30%-50%, which improves the efficiency of subsequent processing. For point clouds The covariance matrix of the neighborhood; This refers to the number of neighborhood points used to calculate curvature. The average coordinates of the neighboring points; Covariance matrix eigenvalues; For point clouds The curvature; Calculate the curvature value of the point cloud within each grid cell, retain the point cloud with the largest curvature, and delete the remaining redundant point clouds. While ensuring the integrity of the structural features, the amount of point cloud data is reduced by 30%-50%, thereby improving the efficiency of subsequent processing. A5. Perform multi-scale sampling on the processed point cloud data to construct feature layers at scales of 10mm, 20mm, and 30mm. The small-scale layer is used to extract detailed features of embedded parts and small structures such as junction boxes; the medium-scale layer is used to extract the arrangement features of reinforcing bars; and the large-scale layer is used to extract the overall outline features of the structure, forming a multi-scale feature pyramid. The point cloud collections of the scale layers are then analyzed. satisfy: ; For the point cloud set at the k-th scale layer; Let be the sampling radius of the k-th scale layer; For post-processing point clouds 3D coordinates; A6. In each feature layer at each scale, the Fast Point Feature Histogram algorithm is used to calculate the local feature descriptor for each point cloud, focusing on capturing the normal vector, curvature, and neighborhood point distribution information of the point cloud. For linear structures with reinforced steel bars, the direction vector feature is calculated in addition, which is used for subsequent identification of steel bar direction and spacing. The FPFH algorithm is a Fast Point Feature Histogram algorithm. Fast point feature histogram algorithm calculates each point cloud Local feature descriptors The core is to capture the normal vector, curvature, and neighborhood point distribution information of the point cloud; for point clouds and its neighboring points Define the unit vector of the line connecting two points. , Representing the magnitude of a vector, point cloud unit normal vector Through principal component analysis (PCA), the three key angular features in the fast point feature histogram algorithm are: ; Will The data was divided into 11, 11, and 5 intervals for binning and statistical analysis, resulting in a 125-dimensional feature vector generated by the fast point feature histogram algorithm. For linear structures with reinforced steel bars, additional direction vectors are calculated. , To calculate the number of neighborhood points when calculating the direction vector and supplement it into the feature vector, the recognition accuracy of steel-reinforced structures is improved. For point clouds and The unit vector of the connecting line; Point clouds The unit normal vector; These are the three angular features of the FPFH algorithm; For point clouds 125-dimensional FPFH feature descriptor; To calculate the number of neighborhood points when calculating the direction vector of the reinforcing bars; Point cloud for steel reinforcement The direction vector; A7. Preliminary classification of structures: Based on the preset feature library of concealed engineering structures, including the standard feature parameters of embedded parts, reinforcing bars, and junction boxes, the support vector machine (SVM) classifier is used to perform preliminary classification of the point clouds of feature layers at each scale. The point clouds are divided into embedded parts, reinforcing bars, junction boxes, and background. The classification accuracy threshold is set to 95%, and point clouds below the threshold are marked as unverified. A8. Cross-scale feature fusion and secondary recognition: For the point cloud of the class to be verified in the preliminary classification, extract its feature information at different scale levels, give higher weight to key scale features through the attention mechanism, perform cross-scale feature fusion, and then use the K nearest neighbor classifier for secondary recognition to correct the preliminary classification error and finally determine the point cloud of each structure. A9. Calculation of structural parameters: For the classified point cloud of the structure, the minimum bounding rectangle algorithm is used to calculate the length, width, height and center coordinates of the embedded parts and junction boxes. The Hough transform algorithm is used to detect the axis equation of the reinforcing steel bars and calculate the diameter, length, spacing and orientation angle of the reinforcing steel bars to form a set of structural feature parameters. A10. Retrieve the BIM model of the corresponding prefabricated component from the cloud platform and convert it into a coordinate system consistent with the measured point cloud data. The conversion formula is as follows: ; R represents the center coordinates of the k-th structure in the transformed BIM model; R is a 3×3 rotation matrix. It is a 3×1 translation vector; Extract standard feature parameters of hidden engineering structures from the BIM model and construct a standard feature parameter library; A11. Using the sample consistency initial registration algorithm, 100-200 feature points are randomly sampled from the measured structure point cloud and the BIM model point cloud respectively. Through FPFH feature descriptor matching, 30-50 pairs of feature points with the highest matching degree are selected. The initial transformation matrix is ​​calculated to achieve the initial alignment between the measured model and the BIM model. The registration error is controlled within 5mm. A12. Based on the initial transformation matrix, an improved iterative nearest point algorithm is used for fine registration. A weighting factor is introduced to give higher weight to the key feature points of the structure. During the iteration process, the distance from the point to the surface is used as the error measurement standard. An iteration termination condition is set to finally achieve high-precision registration between the measured model and the BIM model, with the registration error controlled within 2mm. A13. Deviation Calculation and Result Judgment: For the registered measured model and BIM model, deviation calculations are performed on the characteristic parameters of the corresponding structures, including positional deviation, dimensional deviation, and layout deviation. Each deviation value is compared with the preset quality acceptance threshold. If all deviation values ​​are within the threshold range, the inspection is deemed qualified. If any deviation value exceeds the threshold, it is marked as unqualified and the specific value and location of the deviation are recorded. A14. Convert historical quality inspection reports stored on the cloud platform into structured data, build a quality problem database, and encode the types of non-conformities; A15. The improved Apriori algorithm is used to mine frequent itemsets in the quality problem database. The minimum support is set to 5% and the minimum confidence is set to 70%. Redundant candidate sets are removed by pruning strategy to mine frequently occurring quality problem combinations. A16. Association rule generation and analysis: Based on the mined frequent itemsets, generate association rules for quality issues, calculate the lift of association rules, select strong association rules with a lift ≥ 1.2, and analyze the intrinsic associations between quality issues; A17. Quality improvement suggestion generation: Combining strong association rules and construction process flow, the key causes of quality problems are identified. The effectiveness of historical treatment measures is evaluated using decision tree algorithm, effective treatment measures are selected, and targeted quality improvement suggestions are generated and fed back to the construction stage to achieve continuous optimization of the production quality of precast components.

[0015] This invention provides a robot and method for inspecting concealed works of prefabricated building components. The invention uses a truss-type robot equipped with a two-dimensional gimbal and a structured light camera to scan the concealed structural elements of the prefabricated components. It extracts structural features from point cloud data and compares them with a BIM model to obtain the inspection results. This invention offers the following advantages: (1) Using robots to inspect concealed works reduces the labor intensity of workers and reduces the safety risks of manual operation in complex sites. In addition, automated inspection is not affected by factors such as human fatigue, and the inspection process can be carried out continuously, ensuring the stability and sustainability of the inspection work.

[0016] (2) The use of robots and high-precision structured light cameras for automatic inspection of concealed works avoids the influence of human factors and improves the inspection accuracy, inspection efficiency and inspection quality.

[0017] (3) The gantry robot combined with the two-dimensional gimbal can perform all-round inspection of the concealed works of prefabricated components of different sizes and positions, and has strong adaptability. (4) The point cloud data and the test results are stored in the cloud platform, which can be stored and traced for a long time, which facilitates the subsequent tracking and analysis of the quality of precast components, provides reliable data support for engineering quality assessment and problem investigation, and lays the database foundation for quality improvement. (5) The test results are more intuitive and accurate, making it easier to integrate with the whole life cycle management of buildings and improving the digital management level of prefabricated buildings. The application of complete sets of technologies has improved the intelligence level of prefabricated buildings. Attached Figure Description

[0018] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is an isometric drawing of the robot for inspecting concealed works of prefabricated building components according to the present invention. Figure 2 This is another isometric view of the robot for inspecting concealed works of prefabricated building components according to the present invention; Figure 3 This is a front view of the robot for inspecting concealed works of prefabricated building components according to the present invention; Figure 4 This is a side view of the robot for inspecting concealed works of prefabricated building components according to the present invention; Figure 5 This is a schematic diagram of the longitudinal imaging coverage and operation points of the robot for inspecting concealed works of prefabricated building components according to the present invention. Figure 6 This is an isometric view of the transverse track beam in the robot for inspecting concealed works of prefabricated building components of the present invention; Figure 7This is an assembly structure diagram of the horizontal sleeve and vertical lifting frame in the prefabricated building component concealed engineering inspection robot of the present invention; Figure 8 This is an isometric view of the two-dimensional gimbal in the robot for inspecting concealed works of prefabricated building components of the present invention; Figure 9 This is a detailed diagram of the longitudinal imaging coverage and operation points of the robot for inspecting concealed works of prefabricated building components according to the present invention. Figure 10 This is an isometric view of the coverage area captured by the robot for inspecting concealed works of prefabricated building components according to the present invention; Figure 11 This is a production collaborative control system architecture diagram of the robot for inspecting concealed works of prefabricated building components according to the present invention. Figure 12 This is a flowchart of the operation of the robot for inspecting concealed works of prefabricated building components according to the present invention.

[0019] In the diagram: 1. Track frame; 2. Horizontal track beam; 3. Horizontal sliding sleeve; 4. Vertical lifting frame; 5. Two-dimensional pan-tilt head; 6. Structured light camera; 7. Electrical control cabinet; 8. Mold table; 9. Precast component; Column 101; Longitudinal beam 102; Longitudinal track 103; Longitudinal rack 104; Ladder 105; Longitudinal walking platform 106; Longitudinal fence 107; 201. Crossbeam; 202. Longitudinal servo motor; 203. Longitudinal slider; 204. Transverse track; 205. Transverse rack; 206. Transverse walking platform; 207. Transverse fence. Sleeve 301; Horizontal slider 302; Horizontal servo motor 303; Vertical slider 304; Vertical servo motor 305; Vertical rod 401; Vertical track 402; Vertical rack 403; Base 501; gimbal body 502; rotating frame 503. Detailed Implementation

[0020] Example 1 like Figures 1-12 As shown, a robot for inspecting concealed works of prefabricated building components includes a wide-frame structure track frame 1, a transverse track beam 2 between the top two sides of the track frame 1, a mold platform 8 for placing the object to be inspected at the bottom of the track frame 1, the two ends of the transverse track beam 2 being slidably connected to the top two sides of the track frame 1, at least one vertical lifting frame 4 being provided on the transverse track beam 2, a two-dimensional gimbal 5 being provided at the end of the vertical lifting frame 4, and a structured light camera 6 being provided on the two-dimensional gimbal 5. The object to be detected is set on the model platform 8, and the two-dimensional gimbal 5 performs visual detection on the object to be detected.

[0021] The track frame 1 with a wide frame structure is used as the base support. The top two sides of the track frame 1 support the transverse track beams 2. The two are slidably connected so that the transverse track beams 2 can move along the top two sides of the track frame 1. The formwork 8 at the bottom of the track frame 1 is used to place the prefabricated building components to be inspected. The vertical lifting frame 4 mounted on the transverse track beams 2 can realize the vertical position adjustment. The two-dimensional gimbal 5 at its end can drive the structure light camera 6 to rotate at multiple angles. Finally, the structure light camera 6 completes the visual inspection of the hidden works of the object to be inspected on the formwork 8.

[0022] First, the prefabricated components to be inspected are placed stably on the formwork platform 8 at the bottom of the track frame 1 to ensure the stability of the component position. Then, according to the inspection requirements of the component's concealed works, the horizontal track beam 2 is controlled to slide along both sides of the top of the track frame 1 and adjusted to a suitable horizontal inspection position. Next, the vertical lifting frame 4 is moved vertically to adjust the two-dimensional pan-tilt head 5 and the structured light camera 6 to a height suitable for the inspection area. Then, the two-dimensional pan-tilt head 5 is activated to drive the structured light camera 6 to rotate at multiple angles, covering the inspection area of ​​the component's concealed works. The structured light camera 6 collects visual data of the component's concealed works to complete the visual inspection of the object being inspected. During the inspection process, the horizontal position of the horizontal track beam 2, the height of the vertical lifting frame 4, and the angle of the two-dimensional pan-tilt head 5 can be repeatedly adjusted as needed to ensure that no part of the inspection is missed.

[0023] Through the sliding engagement of the track frame 1 and the transverse track beam 2, the height adjustment of the vertical lifting frame 4, and the multi-angle rotation of the two-dimensional gimbal 5, the structured light camera 6 can comprehensively cover the object being inspected on the mold table 8, avoiding the limitations of manual inspection and improving the comprehensiveness of hidden engineering inspection. Visual inspection using the structured light camera 6 replaces traditional manual verification and ruler measurement, reducing the impact of human experience and skill level on the inspection results, lowering the probability of missed or false inspections, and improving inspection accuracy and quality. It eliminates the need for workers to work at close range in the workshop, significantly reducing the labor intensity of workers and avoiding on-site safety hazards. The transverse, vertical, and angle adjustments during the inspection process can all respond quickly, significantly improving inspection efficiency compared to the cumbersome manual inspection process, and adapting to the pace of large-scale production of prefabricated components for prefabricated buildings. The visual data collected by the structured light camera 6 can be stored and recorded, facilitating the traceability of inspection results later, providing data support for the investigation and evaluation of component quality problems, and assisting in the digital management of projects.

[0024] In the preferred embodiment, one side of the transverse track beam 2 is slidably connected to the transverse sleeve 3, and a transverse servo motor 303 is provided on one side of the transverse sleeve 3 to drive the transverse sleeve 3 to slide on one side of the transverse track beam 2. The vertical lifting frame 4 is slidably connected to the transverse sleeve 3 up and down. The other side of the horizontal sleeve 3 is equipped with a vertical servo motor 305, which drives the vertical lifting frame 4 to move up and down.

[0025] One side of the transverse track beam 2 is slidably connected to the transverse sleeve 3. The transverse servo motor 303 configured on one side of the transverse sleeve 3 is the power source, which can drive the transverse sleeve 3 to slide laterally along one side of the transverse track beam 2, thereby adjusting the transverse position of the transverse sleeve 3 and subsequent components. Meanwhile, the vertical lifting frame 4 and the horizontal sliding sleeve 3 form an up-and-down sliding connection. The vertical servo motor 305 set on the other side of the horizontal sliding sleeve 3 can provide vertical driving force, drive the vertical lifting frame 4 to move up and down along the horizontal sliding sleeve 3, thereby adjusting the height of the two-dimensional gimbal 5 and the structure light camera 6 at the end of the vertical lifting frame 4, and finally cooperate with the structure light camera 6 to complete the inspection of the hidden works of the prefabricated components on the mold table 8.

[0026] First, place the prefabricated building component to be inspected on the formwork platform 8 under the track frame 1 to ensure the component's position is stable. According to the inspection area of ​​the component's concealed works, start the horizontal servo motor 303 on one side of the horizontal sleeve 3 to drive the horizontal sleeve 3 to slide along one side of the horizontal track beam 2, adjusting the horizontal sleeve 3 and the connected vertical lifting frame 4, two-dimensional pan-tilt head 5, and structured light camera 6 to the target horizontal inspection position. Then, start the vertical servo motor 305 on the other side of the horizontal sleeve 3 to drive the vertical lifting frame 4 to move up and down along the horizontal sleeve 3, adjusting the structured light camera 6 to a height suitable for the inspection area. Subsequently, the structured light camera 6 is rotated at multiple angles by the two-dimensional pan-tilt head 5 to collect visual data of the component's concealed works, completing the inspection. If the inspection area needs to be adjusted, the horizontal servo motor 303 and the vertical servo motor 305 can be repeatedly started to adjust the horizontal position and height respectively, ensuring that all inspection parts are covered.

[0027] By driving the transverse sleeve 3 to slide along the transverse track beam 2 with the transverse servo motor 303, and driving the vertical lifting frame 4 to move up and down with the vertical servo motor 305, the transverse and vertical positions of the structure light camera 6 can be precisely adjusted. Compared with manual adjustment, it is more stable and effectively improves the adaptability of detection for concealed works of prefabricated components of different sizes and positions.

[0028] In the preferred embodiment, a transverse track 204 is provided on the transverse beam 201 of the transverse track beam 2, the transverse sleeve 3 is slidably connected to the transverse track 204, a transverse rack 205 is provided on the transverse beam 201, the gear at the output end of the transverse servo motor 303 meshes with the transverse rack 205, and the transverse servo motor 303 drives the transverse sleeve 3 to slide on the transverse beam 201.

[0029] The core component of the transverse track beam 2, the transverse beam 201, is equipped with a transverse track 204 and a transverse rack 205. The transverse sleeve 3 is slidably connected to the transverse track 204 to obtain a guiding foundation for moving along the transverse beam 201. The gear at the output end of the transverse servo motor 303 forms a meshing transmission structure with the transverse rack 205. When the transverse servo motor 303 is started, its output gear will roll along the transverse rack 205, thereby driving the transverse sleeve 3 associated with it to slide stably along the transverse track 204. Finally, the transverse sleeve 3 and the subsequently connected vertical lifting frame 4, two-dimensional gimbal 5, and structured light camera 6 are adjusted in the transverse position along the transverse beam 201, providing transverse movement support for the structured light camera 6 to cover the concealed engineering inspection area of ​​the prefabricated components on the mold table 8.

[0030] First, the prefabricated building component to be inspected is placed stably on the formwork platform 8 under the track frame 1, ensuring that the component is fixed in position and without deviation. According to the lateral distribution of the component's concealed works, the control system starts the lateral servo motor 303, so that the gear at the output end of the lateral servo motor 303 meshes and drives the lateral rack 205 on the crossbeam 201, while simultaneously driving the lateral sleeve 3 to slide along the lateral track 204, adjusting the lateral sleeve 3 and the connected vertical lifting frame 4, two-dimensional pan-tilt head 5, and structured light camera 6 to the first target lateral inspection point. After the lateral position is determined, the concealed works data of this point is collected by the structured light camera 6, combined with the height adjustment of the vertical lifting frame 4 and the angle adjustment of the two-dimensional pan-tilt head 5. If it is necessary to inspect other lateral areas of the component, the lateral servo motor 303 is started again, controlling the lateral sleeve 3 to slide along the lateral track 204 to the new target point, until the concealed works inspection of all lateral areas of the component is completed.

[0031] In the preferred embodiment, the transverse sleeve 3 has a raised hollow rectangular frame structure in the middle. The vertical rod 401 of the vertical lifting frame 4 is provided with a vertical rail 402. The slider on the inner side of the transverse sleeve 3 is slidably connected to the vertical rail 402. The vertical rod 401 is also provided with a vertical rack 403. The gear on the vertical servo motor 305 meshes with the vertical rack 403. The vertical servo motor 305 drives the vertical rod 401 to slide up and down in the transverse sleeve 3.

[0032] The transverse sleeve 3 is designed with a raised hollow rectangular frame structure in the middle, which provides installation and accommodation space for the movement of the vertical lifting frame 4. The vertical rod 401 of the vertical lifting frame 4 is provided with a vertical rail 402 and a vertical rack 403 in parallel. The slider on the inner side of the transverse sleeve 3 is slidably connected to the vertical rail 402, which provides stable guidance for the vertical movement of the vertical lifting frame 4. The gear on the vertical servo motor 305 meshes with the vertical rack 403. When the vertical servo motor 305 is started, its gear will roll along the vertical rack 403, thereby driving the vertical rod 401 and the entire vertical lifting frame 4 to slide up and down along the cooperation structure of the slider on the inner side of the transverse sleeve 3 and the vertical rail 402. Finally, the height of the two-dimensional gimbal 5 and the structured light camera 6 at the end of the vertical lifting frame 4 can be adjusted to meet the inspection needs of concealed works of prefabricated components of different heights on the mold table 8.

[0033] First, place the prefabricated building component to be inspected on the formwork 8 at the bottom of the track frame 1 to ensure that the component is stable and does not shift. According to the height position of the component's concealed works, start the vertical servo motor 305 through the control system. The gear on the vertical servo motor 305 meshes with the vertical rack 403 of the vertical rod 401 of the vertical lifting frame 4. At the same time, it drives the vertical rod 401 to slide up and down along the sliding block on the inner side of the horizontal sleeve 3 and the mating structure with the vertical track 402. Adjust the two-dimensional gimbal 5 and the structured light camera 6 at the end of the vertical lifting frame 4 to a height that matches the inspection area. If the horizontal inspection position needs to be adjusted, the horizontal sleeve 3 can be driven by the horizontal servo motor 303 to slide along the crossbeam 201 of the horizontal track beam 2 to the target horizontal point, and then repeat the above height adjustment steps. Finally, start the two-dimensional gimbal 5 to drive the structured light camera 6 to rotate at multiple angles to complete the visual data acquisition of the component's concealed works. If the height of the inspection area needs to be adjusted a second time, the vertical servo motor 305 can be started again to fine-tune the vertical position to ensure full inspection coverage.

[0034] In the preferred embodiment, the lower end of the vertical rod 401 is connected to the base 501 of the two-dimensional gimbal 5, the rotating ends on both sides of the gimbal body 502 of the two-dimensional gimbal 5 are connected to the two ends of the rotating frame 503, and the lower end of the rotating frame 503 is connected to the structured light camera 6.

[0035] The lower end of the vertical rod 401 of the vertical lifting frame 4 is directly connected to the base 501 of the two-dimensional gimbal 5. The base 501 provides stable support for the two-dimensional gimbal 5, so that the height of the two-dimensional gimbal 5 can be adjusted synchronously with the up and down movement of the vertical rod 401. The gimbal body 502 of the two-dimensional gimbal 5 has rotating ends on both sides, which are correspondingly connected to the two ends of the rotating frame 503 to form a rotatable fit. The lower end of the rotating frame 503 is fixedly connected to the structure light camera 6. By rotating the gimbal body 502 around the vertical axis and rotating the rotating frame 503 along its own assembly axis, the structure light camera 6 can be driven to achieve multi-angle rotation. Finally, the structure light camera 6 completes the all-round visual inspection of the concealed works of the prefabricated components on the mold table 8.

[0036] In the preferred embodiment, longitudinal sliders 203 are provided at both ends of the transverse track beam 2. The longitudinal sliders 203 are slidably connected to the longitudinal track 103 on the longitudinal beam 102 of the track frame 1. The longitudinal track 103 is provided with a longitudinal rack 104. Both ends of the transverse track beam 2 are provided with longitudinal servo motors 202. The gears at the output end of the longitudinal servo motors 202 mesh with the longitudinal rack 104. The two longitudinal servo motors 202 drive the transverse track beam 2 to slide on the longitudinal beam 102.

[0037] Both ends of the transverse track beam 2 are equipped with longitudinal sliders 203, which form a sliding connection with the longitudinal rail 103 on the longitudinal beam 102 of the track frame 1, providing a guiding foundation for the longitudinal movement of the transverse track beam 2. At the same time, a longitudinal rack 104 is provided next to the longitudinal rail 103 on the longitudinal beam 102. Both ends of the transverse track beam 2 are equipped with longitudinal servo motors 202, and the gears at their output ends mesh with the longitudinal rack 104. When the two longitudinal servo motors 202 start synchronously, the output gears will roll stably along the longitudinal rack 104, thereby driving the transverse track beam 2 to slide on the longitudinal beam 102 along the longitudinal rail 103 via the longitudinal sliders 203. Finally, the longitudinal position adjustment of the transverse track beam 2 and the subsequently connected transverse sleeve 3, vertical lifting frame 4, two-dimensional gimbal 5, and structured light camera 6 is realized to cover the detection area of ​​the concealed works of the prefabricated components on the mold table 8 distributed longitudinally.

[0038] The sliding engagement between the longitudinal sliders 203 at both ends of the transverse track beam 2 and the longitudinal track 103 on the longitudinal beam 102, combined with the meshing transmission between the longitudinal servo motors 202 at both ends and the longitudinal rack 104, achieves bidirectional symmetrical drive of the transverse track beam 2. Compared with single-sided drive, this ensures smoother longitudinal movement and prevents the transverse track beam 2 from shifting or jamming due to uneven force, thus improving the stability of longitudinal position adjustment. The two longitudinal servo motors 202 are driven synchronously, and the speed and direction can be precisely controlled by the electronic control system, thereby accurately adjusting the longitudinal movement distance and speed of the transverse track beam 2. This ensures that each longitudinal inspection point of the concealed precast component can be accurately located. It reduces missed inspections caused by longitudinal position deviations and improves inspection accuracy; it eliminates the need for manual pushing or pulling of the transverse track beam 2 to adjust the longitudinal position, significantly reducing the labor intensity of workers and avoiding close contact with equipment and components during manual operation, thus reducing safety hazards in workshop workstations; the meshing transmission efficiency of the longitudinal servo motor 202 and the longitudinal rack 104 is high and the response speed is fast, which can quickly complete the longitudinal position switching of the transverse track beam 2, shorten the adjustment time between different longitudinal inspection points, and further improve the overall inspection efficiency of concealed works of prefabricated components in conjunction with transverse, vertical and angle adjustments, adapting to the production and inspection rhythm of prefabricated components for large-scale prefabricated buildings.

[0039] In the preferred embodiment, the longitudinal beam 102 is supported by multiple columns 101 and is arranged between the lower parts of the multiple columns 101; Longitudinal railings 107 are provided on both sides of the longitudinal beam 102; An electrical control cabinet 7 is installed on the upper part of the crossbeam 201 of the transverse track beam 2.

[0040] The longitudinal beam 102 of the track frame 1 is supported by multiple columns 101, and the longitudinal beam 102 is set between the lower parts of the multiple columns 101 to form a stable wide frame support structure, providing a solid foundation for the longitudinal beam 102 and subsequent assembled transverse track beam 2 and other components; longitudinal railings 107 are installed on both sides of the longitudinal beam 102 to protect the upper area of ​​the longitudinal beam 102; at the same time, an electrical control cabinet 7 is placed on the upper part of the transverse beam 201 of the transverse track beam 2. The electrical control cabinet 7 provides electrical control support for the operation of the entire inspection robot, including drive control of the longitudinal servo motor 202, the transverse servo motor 303, and the vertical servo motor 305, as well as operation control of the two-dimensional gimbal 5 and the structured light camera 6.

[0041] The longitudinal beam 102 is supported by multiple columns 101 and is positioned between the lower parts of the columns 101, forming a wide-frame structure that provides stable support and effectively bears the weight of components such as the transverse track beam 2 and the electrical control cabinet 7. This prevents equipment swaying during testing due to unstable foundations, ensuring testing accuracy. The longitudinal railings 107 on both sides of the longitudinal beam 102 prevent personnel from accidentally contacting the transmission components on the upper part of the longitudinal beam 102 or preventing foreign objects from falling onto the components on the mold table 8, improving the safety of equipment operation and personnel handling. The electrical control cabinet 7 is located on the upper part of the transverse beam 201, close to the components that need to be controlled, shortening the time required for operation. By controlling the distance of the control line, reducing signal transmission delay, and improving control response speed, the centralized electrical control design facilitates the operation and maintenance of the electrical control cabinet 7 by staff on the transverse walking platform 206 of the transverse track beam 2, eliminating the need for frequent position changes and reducing operational difficulty. The combination of a stable support structure, safe protective design, and convenient electrical control operation not only ensures the long-term stable operation of the inspection robot but also further enhances the safety and ease of operation of the inspection process, indirectly contributing to the improvement of inspection efficiency and adapting to the inspection needs of large-scale production of prefabricated components for prefabricated buildings.

[0042] The preferred solution also includes a cloud-based production collaborative control system. The cloud platform interfaces with the host computer of the production line and the host computer of the robot. The cloud platform has built-in BIM models of all prefabricated components and retrieves the corresponding BIM model according to the current inspection requirements and transmits it to the host computer of the robot. The robot's host computer includes a vision module and an equipment control module. The vision module is used for controlling the 2D gimbal 5, acquiring and processing point cloud data, and outputting quality inspection reports. The equipment control module is used to control longitudinal, lateral, and vertical movement.

[0043] The inspection robot is equipped with a cloud-based production collaborative control system. The cloud platform interfaces with the host computer of the production line and the robot's host computer. The cloud platform pre-stores BIM models of all prefabricated components. It can retrieve the corresponding BIM model according to the type of prefabricated component being inspected and transmit it to the robot's host computer. The robot's host computer is further divided into a vision module and an equipment control module. The vision module is responsible for controlling the rotation angle of the 2D gimbal 5, collecting and post-processing point cloud data through the structured light camera 6, and generating and outputting a quality inspection report based on the processing results. The equipment control module is specifically used to control the longitudinal movement of the transverse track beam 2 along the longitudinal beam 102, the transverse movement sleeve 3 along the transverse beam 201, and the vertical movement of the vertical lifting frame 4 along the transverse movement sleeve 3. Through the collaboration between the cloud platform and the robot's host computer, the entire inspection process is automated.

[0044] When the mold platform 8 transports the prefabricated component to be inspected to the inspection station, the production line's host computer transmits the mold platform positioning signal to the cloud platform. Upon receiving the signal, the cloud platform retrieves the corresponding BIM model from its built-in BIM model library based on the model of the component to be inspected. Simultaneously, it sends a start inspection command to the robot's host computer and displays the retrieved BIM model. The model is synchronously transmitted to the vision module of the robot's host computer. The device control module of the robot's host computer then starts, sending control signals to the longitudinal servo motor 202, the transverse servo motor 303, and the vertical servo motor 305, respectively, to drive the transverse track beam 2 to slide longitudinally along the longitudinal beam 102, the transverse sleeve 3 to slide laterally along the transverse beam 201, and the vertical lifting frame 4 to move up and down, adjusting the 2D gimbal 5 and the structured light camera 6 to the initial detection position. Then, the vision module sends an angle control signal to the 2D gimbal 5 to adjust the shooting angle of the structured light camera 6, and simultaneously controls the structured light camera 6 to collect point cloud data of the component's hidden works. After the data collection is completed, the vision module processes the original point cloud data, extracts the structural features of the component's hidden works, and constructs a measured model. It then compares and analyzes the measured model with the received BIM model to generate a quality inspection report containing deviation results. Finally, the vision module feeds back the quality inspection report to the cloud platform. If the inspection is qualified, the cloud platform sends a signal to the host computer on the production line to control the model 8 to enter the next workstation. If the inspection is unqualified, the staff processes the component according to the quality inspection report and repeats the above process for re-inspection.

[0045] The cloud platform's two-way interface with the production line and the robot's host computer enables efficient transmission of inspection commands, BIM models, and quality inspection results. This eliminates the need for manual data transfer or model retrieval, reducing human intervention and avoiding data transmission errors. Simultaneously, the cloud platform has built-in BIM models of all prefabricated components, allowing for rapid matching of inspection needs and improving inspection preparation efficiency. The robot's host computer has a clear division of labor between its vision module and equipment control module. The vision module focuses on inspection data processing and gimbal and camera control, while the equipment control module precisely manages three-dimensional movement. Together, they automate the inspection process, significantly improving efficiency compared to manual inspection and ensuring the stability and consistency of inspection results, unaffected by human fatigue or experience differences. The vision module processes point cloud data and integrates with BIM... The model comparison generates quality inspection reports, which not only intuitively presents the detection deviations, but also retains data through the cloud platform, facilitating the traceability of component quality in the later stages and providing a reliable basis for project quality assessment and problem investigation. The digital collaborative design of the entire system combines the inspection process with the digital management system of prefabricated buildings, improving the level of intelligence in the production and inspection of prefabricated components, adapting to the needs of large-scale, standardized prefabricated building production, and helping to improve the overall digital management level of the project.

[0046] Example 2 Further explanation in conjunction with Example 1, such as Figure 1-12As shown in the structure, S1, a prefabricated component 9 is placed on the mold table 8. The production line host computer feeds back the signal to the cloud platform. After receiving the signal, the cloud platform sends a signal to the robot host computer to start the detection and transmits the BIM model of the prefabricated component on the mold table 8 to the vision module of the robot host computer. S2. The robot host computer’s device control module sends signals to the longitudinal servo motor 202, the transverse servo motor 303, and the vertical servo motor 305. At the same time, the vision module sends control signals to the two-dimensional gimbal 5 to reset the position of the structured light camera 6 to the zero point O. S3. The equipment control module sends a signal to the longitudinal servo motor 202, driving the transverse track beam 2 to move along the longitudinal beam 102 from point O to shooting point A and stop, and sends the signal of reaching point A to the vision module. S4. The vision module sends acquisition signals to the 2D pan-tilt unit 5 and the structured light camera 6 to acquire point cloud data of area A. After acquisition is completed, it sends an acquisition end signal to the device control module. S5. The equipment control module drives the transverse track beam 2 to move from point A to the shooting point B and stop. The vision module controls the two-dimensional pan-tilt unit 5 and the structured light camera 6 to collect point cloud data of the area at point B. After the collection is completed, a collection end signal is fed back. S6. The equipment control module drives the transverse track beam 2 to move from point B to the shooting point C and stop. The vision module controls the completion of point cloud data acquisition in the area of ​​point C and sends back an end signal. S7. The equipment control module drives the transverse track beam 2 to move from point C to the shooting point D and stop. The vision module controls the completion of point cloud data acquisition in the area of ​​point D and sends back an end signal. S8. The equipment control module drives the transverse track beam 2 to move from point D to the shooting point E and stop. The vision module controls the completion of point cloud data acquisition in the area of ​​point E and sends back an end signal. S9. The equipment control module drives the transverse track beam 2 from point E back to point O and stops, sending a signal to the cloud platform indicating the end of data acquisition. S10: The vision module performs post-processing on the collected point cloud data, extracts structural features and builds a model, compares the model with the BIM model, generates a quality inspection report and feeds it back to the cloud platform. S11. If the quality inspection result is qualified, the mold table will enter the next station; if it is unqualified, the worker will handle it according to the quality inspection report and the mold table will enter the next station. S12. The cloud platform stores the point cloud data and quality inspection report of this test.

[0047] The inspection process of this prefabricated building component concealed works inspection robot revolves around "signal transmission - equipment reset - point acquisition - data processing - result feedback - data storage": First, the prefabricated component 9 is placed on the mold table 8. The production line host computer transmits a signal to the cloud platform, which then sends an inspection command to the robot host computer and transmits the corresponding prefabricated component's BIM model to the vision module. Next, the robot host computer's equipment control module regulates the longitudinal servo motor 202, the transverse servo motor 303, and the vertical servo motor 305, while the vision module controls the 2D gimbal 5 to reset the structured light camera 6 to the zero position O. Subsequently, the equipment control module drives the transverse track beam 2 to move sequentially along the longitudinal beam 102 to five shooting points A, B, C, D, and E. At each point, the vision module controls the 2D gimbal 5 and the structured light camera 6 to collect point cloud data, and sends a feedback signal after the data acquisition is completed. After the data acquisition is completed, the transverse track beam 2 returns to point O and informs the cloud platform. The vision module then performs post-processing of the data and integrates it with the BIM model. The model comparison generates a quality inspection report. If the model passes the test, it proceeds to the next workstation. If the model fails the test, it is processed and then retested. Finally, the cloud platform stores the point cloud data and quality inspection report of this test.

[0048] The method of using this solution is as follows: First, the precast component 9 to be inspected is placed stably on the mold table 8 to ensure that the component position is fixed and without deviation. After the mold table 8 is transported to the inspection station, the production line host computer automatically feeds back the positioning signal to the cloud platform. After receiving the signal, the cloud platform immediately sends a signal to the robot host computer to start the inspection and retrieves the BIM model corresponding to the precast component 9 from the built-in model library and transmits it to the vision module of the robot host computer. The robot host computer responds to the command, and the equipment control module sends control signals to the longitudinal servo motor 202, the transverse servo motor 303, and the vertical servo motor 305. At the same time, the vision module sends a command to the two-dimensional gimbal 5 to coordinate the precise reset of the structured light camera 6 to the zero point O on the longitudinal beam 102. After the reset is completed, the equipment control module sends a signal to the longitudinal servo motor 202 to drive the transverse track beam 2 to slide along the longitudinal beam 102 from point O to point A. The camera stops at the designated shooting point and transmits the arrival signal to the vision module. Upon receiving the signal, the vision module sends acquisition commands to the 2D pan-tilt unit 5 and the structured light camera 6. The 2D pan-tilt unit 5 adjusts its angle, and the structured light camera 6 acquires point cloud data of the concealed works of precast component 9 in area A. After acquisition, it sends an end signal to the equipment control module. The above point movement and acquisition process is then repeated. The equipment control module sequentially drives the transverse track beam 2 from point A to point B, point B to point C, point C to point D, and point D to point E. Data acquisition and feedback are completed at each point under the control of the vision module. After acquisition at point E, the equipment control module drives the transverse track beam 2 from point E back to point O and stops, sending an end signal to the cloud platform. The vision module then performs post-processing such as noise reduction and feature extraction on all acquired point cloud data to construct a measured model and compare it with the received BIM. The model compares deviations and generates a quality inspection report containing whether the product is qualified, the location of the deviation, and the value, which is then fed back to the cloud platform. If the quality inspection result is qualified, the cloud platform sends a signal to the production line, and the mold 8 carries the precast component 9 to the next production station. If the result is unqualified, the workers, based on the problem description in the quality inspection report, address the hidden engineering defects of the precast component 9, such as the deviation of the embedded parts position or the omission of steel bars. After the processing is completed, the above inspection process is restarted until the inspection is qualified. After each inspection, the cloud platform automatically stores the point cloud data and quality inspection report for subsequent traceability and query.

[0049] The entire inspection process is automated from signal transmission and data acquisition to result feedback through the collaborative control of the cloud platform, the production line host computer, and the robot host computer. This eliminates the need for manual instruction or equipment operation, significantly reducing worker workload and avoiding errors caused by manual operation, thus improving the standardization and stability of the inspection process. The transverse track beam 2 sequentially covers five AE shooting points along the longitudinal beam 102. Combined with the multi-angle acquisition by the structured light camera 6 driven by the 2D pan-tilt head 5, it can comprehensively cover the inspection area of ​​the concealed works of the precast components 9. This, combined with point cloud data and BIM... The accurate comparison of the model effectively reduces missed and false detections, significantly improving detection accuracy and quality. The entire process, from the positioning of the mold platform 8 to data storage, is tightly integrated, with each link responding quickly through signal feedback. The efficient coordination between point movement and data acquisition significantly shortens the detection time compared to traditional manual inspection, adapting to the pace of large-scale production of prefabricated components for prefabricated buildings. The cloud platform stores the point cloud data and quality inspection reports for each inspection, facilitating the traceability of the quality of prefabricated components 9 in the future. If quality problems occur later, historical data can be quickly retrieved to investigate the cause, providing a reliable basis for project quality assessment and improvement, while also enhancing the level of digital management of the project.

[0050] Example 3 Further explanation in conjunction with Example 2, such as Figure 1-12 The detection algorithm steps for the structure shown include: A1. Import the original point cloud data of the prefabricated component concealed works acquired by the structured light camera 6 into the algorithm system. Through coordinate normalization processing, the point cloud data is uniformly mapped to a preset three-dimensional coordinate system to eliminate data deviations caused by differences in coordinate systems at different shooting points. At the same time, the acquisition time of each point cloud is recorded. Corresponding shooting locations and initial coordinates This serves as a foundation for subsequent data tracing and analysis; A2. Based on the local density distribution characteristics of point cloud data, the K nearest neighbor search algorithm is used to calculate the average distance of the K nearest neighbor points around each point cloud. 1.2-1.5 times the average distance is used as the local density threshold to achieve dynamic adaptation of the point cloud density threshold in different regions. A3. Compare the local density of each point cloud with the corresponding dynamic density threshold. If the local density of a point cloud is lower than the threshold, it is determined to be a noise point. Noise points include invalid point clouds caused by environmental dust and light reflection interference. Through the neighborhood point cloud collaborative verification mechanism, the density of the three neighboring points around the suspected noise point is verified a second time. After confirmation, the noise point is removed to avoid the accidental deletion of effective edge point clouds. A4. Using a curvature-based raster sampling method, the preprocessed point cloud data is divided into three-dimensional raster units with a raster size of 5-10 mm. Each raster is denoted as... u, v, w are the indices of the raster along the x, y, and z axes. Calculate the curvature of each point cloud within the raster. Curvature calculation requires first solving the covariance matrix of the point cloud neighborhood. The formula is: ; in , where is the number of neighborhood points selected when calculating curvature. The average coordinates of the neighboring points. Similarly, for the y and z axes, the covariance matrix... Perform eigenvalue decomposition to obtain eigenvalues. Point cloud curvature The formula is: ; The point cloud with the largest curvature in each grid is retained, and the rest of the redundant point clouds are deleted. While ensuring the integrity of the structural features, the amount of point cloud data is reduced by 30%-50%, which improves the efficiency of subsequent processing. For point clouds The covariance matrix of the neighborhood; This refers to the number of neighborhood points used to calculate curvature. The average coordinates of the neighboring points; Covariance matrix eigenvalues; For point clouds The curvature; Calculate the curvature value of the point cloud within each grid cell, retain the point cloud with the largest curvature, and delete the remaining redundant point clouds. While ensuring the integrity of the structural features, the amount of point cloud data is reduced by 30%-50%, thereby improving the efficiency of subsequent processing. A5. Perform multi-scale sampling on the processed point cloud data to construct feature layers at scales of 10mm, 20mm, and 30mm. The small-scale layer (10mm) is used to extract detailed features of embedded parts and small structures such as junction boxes; the medium-scale layer (20mm) is used to extract the arrangement features of reinforcing bars; and the large-scale layer (30mm) is used to extract the overall outline features of the structure, forming a multi-scale feature pyramid. The point cloud collections of the scale layers are then analyzed. satisfy: ; For the point cloud set at the k-th scale layer; Let be the sampling radius of the k-th scale layer; For processed point clouds 3D coordinates; A6. In each feature layer at each scale, the Fast Point Feature Histogram algorithm is used to calculate the local feature descriptor for each point cloud, focusing on capturing the normal vector, curvature, and neighborhood point distribution information of the point cloud. For linear structures with reinforced steel bars, the direction vector feature is calculated in addition, which is used for subsequent identification of steel bar direction and spacing. The FPFH algorithm is a Fast Point Feature Histogram algorithm. Fast point feature histogram algorithm calculates each point cloud Local feature descriptors The core is to capture the normal vector, curvature, and neighborhood point distribution information of the point cloud; for point clouds and its neighboring points Define the unit vector of the line connecting two points. , Representing the magnitude of a vector, point cloud unit normal vector Through principal component analysis (PCA), the three key angular features in the fast point feature histogram algorithm are: ; Will The data was divided into 11, 11, and 5 intervals for binning and statistical analysis, resulting in a 125-dimensional feature vector generated by the fast point feature histogram algorithm. For linear structures with reinforced steel bars, additional direction vectors are calculated. , To calculate the number of neighborhood points when calculating the direction vector and add them to the feature vector, thereby improving the recognition accuracy of steel-reinforced structures; For point clouds and The unit vector of the connecting line; Point clouds The unit normal vector; These are the three angular features of the FPFH algorithm; For point clouds 125-dimensional FPFH feature descriptor; To calculate the number of neighborhood points when calculating the direction vector of the reinforcing bars; Point cloud for steel reinforcement The direction vector; A7. Preliminary classification of structures: Based on the preset feature library of concealed engineering structures, including the standard feature parameters of embedded parts, reinforcing bars, and junction boxes, the support vector machine (SVM) classifier is used to perform preliminary classification of the point clouds of feature layers at each scale. The point clouds are divided into embedded parts, reinforcing bars, junction boxes, and background. The classification accuracy threshold is set to 95%, and point clouds below the threshold are marked as unverified. A8. Cross-scale feature fusion and secondary recognition: For the point cloud of the class to be verified in the preliminary classification, extract its feature information at different scale levels, give higher weight to key scale features through the attention mechanism, perform cross-scale feature fusion, and then use the K nearest neighbor classifier for secondary recognition to correct the preliminary classification error and finally determine the point cloud of each structure. A9. Calculation of structural parameters: For the classified point cloud of the structure, the minimum bounding rectangle algorithm is used to calculate the length, width, height and center coordinates of the embedded parts and junction boxes. The Hough transform algorithm is used to detect the axis equation of the reinforcing steel bars and calculate the diameter, length, spacing and orientation angle of the reinforcing steel bars to form a set of structural feature parameters. A10. Retrieve the BIM model of the corresponding prefabricated component from the cloud platform and convert it into a coordinate system consistent with the measured point cloud data. The conversion formula is as follows: ; R represents the center coordinates of the k-th structure in the transformed BIM model; R is a 3×3 rotation matrix. It is a 3×1 translation vector; Extract standard feature parameters of hidden engineering structures from the BIM model and construct a standard feature parameter library; A11. The SAC-IA algorithm for initial registration is adopted. 100-200 feature points are randomly sampled from the point cloud of the measured structure and the point cloud of the BIM model. Through FPFH feature descriptor matching, 30-50 pairs of feature points with the highest matching degree are selected. The initial transformation matrix is ​​calculated to achieve the initial alignment between the measured model and the BIM model. The registration error is controlled within 5mm. A12. Based on the initial transformation matrix, an improved iterative nearest point (ICP) algorithm is used for fine registration. A weighting factor is introduced to give higher weight to the key feature points of the structure. During the iteration process, the distance from the point to the surface is used as the error measurement standard. An iteration termination condition is set to finally achieve high-precision registration between the measured model and the BIM model, with the registration error controlled within 2mm. A13. Deviation Calculation and Result Judgment: For the registered measured model and BIM model, deviation calculations are performed on the characteristic parameters of the corresponding structures, including positional deviation, dimensional deviation, and layout deviation. Each deviation value is compared with the preset quality acceptance threshold. If all deviation values ​​are within the threshold range, the inspection is deemed qualified. If any deviation value exceeds the threshold, it is marked as unqualified and the specific value and location of the deviation are recorded. A14. Convert historical quality inspection reports stored on the cloud platform into structured data, build a quality problem database, and encode the types of non-conformities; A15. The improved Apriori algorithm is used to mine frequent itemsets in the quality problem database. The minimum support is set to 5% and the minimum confidence is set to 70%. Redundant candidate sets are removed by pruning strategy to mine frequently occurring quality problem combinations. A16. Association rule generation and analysis: Based on the mined frequent itemsets, generate association rules for quality issues, calculate the lift of association rules, select strong association rules with a lift ≥ 1.2, and analyze the intrinsic associations between quality issues; A17. Quality improvement suggestion generation: Combining strong association rules and construction process flow, the key causes of quality problems are identified. The effectiveness of historical treatment measures is evaluated using decision tree algorithm, effective treatment measures are selected, and targeted quality improvement suggestions are generated and fed back to the construction stage to achieve continuous optimization of the production quality of precast components.

[0051] The detailed steps are as follows: A1: Import and Preprocessing of Raw Point Cloud Data: The raw point cloud data set of the prefabricated component concealed works acquired by the structured light camera 6 is denoted as... ,in Let represent the 3D coordinates of the i-th point cloud, and n be the total number of point clouds. To eliminate data bias caused by differences in coordinate systems between different shooting points, the original point cloud is normalized. The normalization formula is: ; These are the normalized point cloud coordinates; These are the minimum and maximum x-axis coordinates of the original point cloud, respectively.

[0052] Simultaneously, the acquisition time for each point cloud was recorded. Corresponding shooting locations and initial coordinates This serves as a foundation for subsequent data tracing and analysis.

[0053] A2: Dynamic Density Threshold Calculation: Based on the local density distribution characteristics of point cloud data, the K-nearest neighbor search algorithm is used, with K set to 15-20, denoted as... It can adaptively adjust according to the structural complexity of the prefabricated components. For each point cloud... Search for it The set of nearest neighbors By calculating point clouds To its The average Euclidean distance between the nearest neighbors The local density of the point cloud is measured by the following formula: ; It is denoted by a factor of 1.2-1.5 times the average distance. As the dynamic density threshold of the point cloud To achieve dynamic adaptation of point cloud density thresholds for different regions, the formula is: ; For point clouds To its The average Euclidean distance between the nearest neighbors; For point clouds The three-dimensional coordinates of the j-th nearest neighbor; For point clouds The corresponding dynamic density threshold.

[0054] A3: Noise Point Identification and Removal: Defining Point Clouds Local density Distance Less than the dynamic density threshold The formula for the number of neighboring points is: ; in For indicator functions: if the condition inside the parentheses is true, ;otherwise .like If the number of neighboring points is less than 3, it is determined to be an isolated noise point, such as an invalid point cloud caused by environmental dust or light reflection interference. The neighborhood verification mechanism is then activated: the average density of the three neighboring points of the point cloud is calculated. ,like If the point cloud is identified as a noise point, it will be removed to avoid accidental deletion of valid edge point clouds.

[0055] For point clouds Local density; For point clouds The average local density of the three nearest neighbors; For point clouds The k-th neighboring point.

[0056] A4: Redundant Point Cloud Reduction: A curvature-based raster sampling method is adopted to divide the preprocessed point cloud data into three-dimensional raster units with a raster size of 5-10mm. Each raster is denoted as... u, v, w are the indices of the raster along the x, y, and z axes. Calculate the curvature of the point cloud within each raster. Curvature calculation requires first solving the covariance matrix of the point cloud neighborhood. The formula is: ; in The number of neighborhood points selected when calculating curvature. The average coordinates of the neighboring points. The same applies to the y and z axes. For the covariance matrix... Perform eigenvalue decomposition to obtain eigenvalues. Point cloud curvature The formula is: ; The point cloud with the largest curvature in each grid cell is retained. The larger the curvature value, the closer it is to key features such as the edge and corner of the structure. The remaining redundant point clouds are deleted. While ensuring the integrity of the structure features, the amount of point cloud data is reduced by 30%-50%, which improves the efficiency of subsequent processing.

[0057] For point clouds The covariance matrix of the neighborhood; This refers to the number of neighborhood points used to calculate curvature. The average coordinates of the neighboring points; Covariance matrix eigenvalues; For point clouds The curvature.

[0058] A5: Construction of Multi-Scale Feature Pyramids The point cloud set after noise reduction and redundancy removal To determine the total number of point clouds after processing, multi-scale feature layers are constructed using different sampling radii: the small-scale layer is 10 mm, denoted as... Used to extract detailed features of small structures such as embedded parts and junction boxes, the medium-scale layer is 20mm, denoted as... Used to extract the arrangement characteristics of reinforcing steel bars, with a large-scale layer of 30mm, denoted as Used to extract the overall contour features of a structure. k-th Point cloud collection at scale level satisfy: ; Let k be the point cloud set at the k-th scale level; Let be the sampling radius of the k-th scale layer; For processed point clouds The three-dimensional coordinates.

[0059] A6: Local Feature Descriptor Calculation: The Fast Point Feature Histogram (FPFH) algorithm is used to calculate each point cloud. Local feature descriptors The core is to capture information such as the normal vector, curvature, and neighborhood point distribution of a point cloud. For point clouds... and its neighboring points Define the unit vector of the line connecting two points. , Representing the magnitude of a vector, point cloud unit normal vector Through principal component analysis (PCA), the three key angular features in the FPFH algorithm are: ; Will The data was divided into 11, 11, and 5 intervals for binning and statistical analysis, resulting in a 125-dimensional FPFH feature vector. For linear structures with reinforced steel bars, additional direction vectors are calculated. , To calculate the number of neighborhood points when calculating the direction vector and add them to the feature vector, the recognition accuracy of steel-reinforced structures is improved.

[0060] For point clouds and The unit vector of the connecting line; Point clouds The unit normal vector; These are the three angular features of the FPFH algorithm; For point clouds 125-dimensional FPFH feature descriptor; To calculate the number of neighborhood points when calculating the direction vector of the reinforcing bars; Point cloud for steel reinforcement The direction vector.

[0061] A7: Preliminary Classification of Structures: Constructing a Standard Feature Library for Hidden Engineering Structures ,in For the standard FPFH feature vector of embedded parts, To enhance the eigenvectors of standard FPFH steel reinforcement, The point cloud features are the standard FPFH feature vectors for wire boxes. A Support Vector Machine (SVM) classifier is used to perform preliminary classification of the point clouds at each scale feature layer. The objective function of the SVM classifier is: ; The constraints are: , ; w is the weight vector of the SVM classifier; b is the bias term. To allow a small amount of classification error as a slack variable, we avoid overfitting; C is the penalty coefficient, set to 10, to balance classification accuracy and generalization ability. Category tags for point clouds 1 = Embedded part, 2 = Reinforcing steel bar, 3 = Junction box, 0 = Background / To be verified; For point clouds FPFH feature descriptor; Represents vector w and The inner product of.

[0062] Calculate classification accuracy If the classification accuracy of a certain point cloud is If so, mark the point cloud as a class to be verified. .

[0063] A8. Cross-scale feature fusion and secondary recognition: For point clouds of classes to be verified in the initial classification. The feature information at the small, medium, and large scales is extracted, and a higher weight is assigned to key scale features through an attention mechanism: for small wire box-like structures, the features at the small scale are assigned higher weights. Mesoscale layer weights Large-scale layer weights For large embedded component structures, assign large-scale layer feature weights. Mesoscale layer weights Small-scale layer weights For reinforced concrete structures, assign mid-scale layer characteristic weights. Small-scale layer weights Large-scale layer weights The fused feature vector The formula is: ; Point cloud to be verified Cross-scale fused feature vectors; Weights of features at small, medium, and large scales, respectively. ; Point clouds FPFH feature vectors at small, medium, and large scale layers.

[0064] Using the K-nearest neighbor classifier The fused feature vectors are then used for secondary identification to correct the initial classification error and finally determine the point cloud affiliation of each structure.

[0065] A9: Structural Parameter Calculation: For the classified structural point cloud, the minimum bounding rectangle algorithm is used to calculate the geometric parameters of embedded parts and junction boxes: Let the point cloud set of embedded parts be... The maximum and minimum coordinates of its minimum bounding rectangle in the x-axis direction are respectively The length of the embedded part Similarly, the width can be obtained. ,high Center coordinates of embedded parts The formula is: ; For the point cloud of reinforced steel bars, the Hough transform algorithm is used to detect the equation of the steel bar axis: Let the parametric equation of the steel bar axis in three-dimensional space be... ,in Let be a point on the axis. Let be the axis direction vector. By accumulating the voting values ​​of spatial lines through Hough transform, the line with the largest voting value is found as the reinforcing bar axis. Combined with the minimum circumscribed cylinder diameter of the reinforcing bar point cloud, the reinforcing bar diameter is obtained. Calculate the shortest distance between the axes of two adjacent reinforcing bars to obtain the reinforcing bar spacing. The angle of the rebar direction is obtained by using the angle between the axial direction vector and the coordinate axes of the coordinate system. The final set of structural characteristic parameters is formed. .

[0066] For embedded parts point cloud collection; The maximum and minimum coordinates of the point cloud of the embedded part along the x-axis; These are the length, width, and height of the embedded parts, respectively. The coordinates of the embedded part's center; The direction vector of the reinforcing bar axis; The diameter of the reinforcing bar; This refers to the spacing between the reinforcing bars; The angle of the reinforcing bar direction; This is a set of characteristic parameters of the measured structure.

[0067] A10: BIM Model Preprocessing: Retrieve the BIM model of the corresponding prefabricated component from the cloud platform and extract the standard feature parameter set of the hidden engineering structures in the model. ,in Let x be the standard center x-coordinate of the k-th structure. For standard lengths, the same applies to the rest. Transform the BIM model coordinates to a coordinate system consistent with the measured point cloud. The transformation formula is: ; R represents the center coordinates of the kth structure in the converted BIM model; R is a 3×3 rotation matrix, initially set as the identity matrix, used to correct the angular deviation between the BIM model and the measured coordinate system. It is a 3×1 translation vector, initially set as the zero vector, used to correct the deviation of the coordinate system origin; The coordinates are the standard center coordinates of the k-th structure in the BIM model.

[0068] A standard feature parameter library for BIM models will be constructed to provide a benchmark for subsequent registration and comparison.

[0069] A11: Initial Registration SAC-IA Algorithm: From the measured structural point cloud P' to the BIM model point cloud In the middle, each random sample For each feature point, the matching degree between the measured point cloud feature vector and the BIM model point cloud feature vector is calculated using cosine similarity, with the formula:

[0070] For measured point cloud With BIM model point cloud The closer the feature matching score is to 1, the higher the matching score. For measured point cloud FPFH eigenvectors; Point cloud for BIM model FPFH eigenvectors; It is the inner product of two eigenvectors; These are the magnitudes of the two eigenvectors, respectively.

[0071] Filter matching degree Point pair ,common Yes, the initial transformation matrix is ​​calculated using Singular Value Decomposition (SVD). To minimize the point-to-point error, the objective function is: ; This is the initial transformation matrix. It includes the initial rotation matrix. With the initial translation vector ; For the k-th pair of matching points, there is a measured point; For the k-th pair of matching points, there is a BIM model point; It is the squared magnitude of the vector, i.e., the square of the Euclidean distance.

[0072] Initial registration error The error margin needs to be controlled within 5mm to ensure the basic accuracy of subsequent fine registration.

[0073] A12: Improved ICP algorithm for fine registration: Introducing weighting factors to enhance the registration accuracy of key feature points. Assign weights to key structural features such as corner points of embedded parts and ends of reinforcing bars. Assign weights to ordinary surface points Based on the initial transformation matrix Iterative optimization of the transformation matrix The objective function is: ; For the first The transformation matrix of the next iteration contains the iterative rotation matrix. With iterative translation vector ; For measured point cloud The weights; For point clouds The nearest neighbor in the BIM model is obtained through K-nearest neighbor search.

[0074] Set the iteration termination condition: the change in error between two consecutive iterations. , For the first The registration error of the next iteration, or the number of iterations. Final fine registration error The accuracy must be controlled within 2mm to achieve high-precision alignment between the measured model and the BIM model.

[0075] A13: Deviation Calculation and Result Judgment: For the registered measured model and BIM model, deviation calculations are performed on the characteristic parameters of the structure one by one: 1. Positional Deviation: The positional deviations of the k-th structure in the x, y, and z axes are respectively: ; 2. Dimensional Deviations: The deviations in length, width, height, or rebar diameter of the k-th structure are as follows: ; 3. Layout deviation: Deviation in the spacing of reinforcing bars. directional angle deviation ; Here are the measured center coordinates of the k-th structure; For the corresponding standard center coordinates; The positional deviations in the x, y, and z axes; These are the actual measured dimensions; Standard size; This is for dimensional deviation; For actual measurement of the spacing and orientation angle of the reinforcing bars; Standard reinforcement spacing and orientation angle; This is due to deviations in the arrangement of reinforcing bars.

[0076] Quality acceptance thresholds are set according to the prefabricated building construction specifications: position deviation threshold. Size deviation threshold Rebar spacing deviation threshold directional angle deviation threshold If all deviation values ​​are within the corresponding threshold range, the concealed works inspection of the precast component is deemed qualified; if any deviation value exceeds the threshold, it is marked as unqualified, and the specific value and location of the deviation are recorded to form a deviation list.

[0077] A14: Quality Inspection Report Data Structure: Historical quality inspection reports stored on the cloud platform, including inspection time, precast component number, non-conformity type, deviation value, and handling measures, are converted into structured data to build a quality issue database. Non-conformity types are coded: For example, missing embedded parts are... The positional deviation of the embedded parts is Rebar spacing exceeding the standard The deviation of the reinforcing bar direction is The junction box was missing. The position deviation of the junction box is Etc. Assume the database contains... There are a total of N detection records, each record Encode the j-th quality problem in the i-th record.

[0078] A15: Frequent Itemset Mining: This section describes the use of an improved Apriori algorithm to mine frequent itemsets from a quality problem database. First, it defines an itemset X as consisting of one or more quality problem codes, such as... Given "rebar spacing exceeding standards + junction box position deviation", calculate the support of itemset X. The frequency of itemset X in all records is given by the formula: ; Let X be the support of itemset X; N is the total number of inspection records in the quality problem database.

[0079] Set minimum support If a certain type of quality issue or combination of issues occurs in ≥5% of the total number of inspections, a pruning strategy is used to remove subsets of itemsets that do not meet the minimum support requirement, thereby reducing the number of candidate itemsets and selecting frequent itemsets. ,like Excessive spacing of reinforcing bars occurs frequently. Excessive spacing of reinforcing bars and deviation in the position of junction boxes often occur simultaneously.

[0080] A16: Association Rule Generation and Analysis: Frequent Itemsets A is the antecedent of the association rule and B is the consequent. Generate the association rule. like This indicates that "if the rebar spacing exceeds the standard, there is a high probability of a junction box position deviation." Calculate the confidence level of the association rule. The probability and lift of the occurrence of the consequent B given the occurrence of the antecedent A. The degree to which the antecedent A promotes the consequent B is expressed by the formula: ; For association rules Confidence level; The elevation degree of the association rule; For itemsets Support level; The support of itemset A; Let B be the support of itemset B.

[0081] Set minimum confidence level Minimum lift Filter out strong association rules such as , Analyzing the inherent relationships between quality problems, such as how deviations in the arrangement of reinforcing bars may lead to limited space for junction box installation, which in turn may cause deviations in the position of the junction box.

[0082] A17: Quality Improvement Suggestion Generation: Combining strong association rules with the precast component production process, identifying the key causes of quality problems: targeting strong association rules... Tracing the construction process revealed that insufficient rebar tying precision was the main cause of excessive rebar spacing. Excessive rebar spacing compresses the installation space for junction boxes, leading to misalignment of the boxes. A decision tree algorithm was used to evaluate the effectiveness of historical handling measures: Let the set of handling measures be... To "add a second check after the rebar is tied", To "optimize the installation sequence of junction boxes", To calculate the decrease in the incidence of quality problems after implementing each measure, such as "replacing with high-precision binding tools". After implementation The incidence rate decreased from 12% to 7.2%, a significant drop. Effective measures with a reduction of ≥30% were selected, and targeted quality improvement suggestions were generated, such as "adding a dedicated person for secondary verification after the rebar tying process can reduce the rate of excessive rebar spacing by 40%, thereby reducing the occurrence of junction box position deviations." These suggestions were then fed back to the construction phase to achieve continuous optimization of the production quality of precast components. Simultaneously, the effective measures and corresponding strong correlation rules were stored in the cloud platform knowledge base to provide reference for subsequent similar projects.

[0083] M represents the set of measures for handling quality problems; For specific handling measures; This represents the decrease in the incidence of quality problems after the implementation of corrective measures.

[0084] The above embodiments are merely preferred technical solutions of the present invention and should not be considered as limitations on the present invention. The scope of protection of the present invention should be limited to the technical solutions described in the claims, including equivalent substitutions of the technical features described in the claims. That is, equivalent substitutions and improvements within this scope are also within the scope of protection of the present invention.

Claims

1. A detection method for a robot used to inspect concealed works of prefabricated building components, characterized by: The method includes: S1. Precast components (9) are placed on the mold table (8). The production line host computer feeds back the signal to the cloud platform. After receiving the signal, the cloud platform sends a signal to the robot host computer to start the detection and transmits the BIM model of the precast components on the mold table (8) to the vision module of the robot host computer. S2, The robot host computer’s equipment control module sends signals to the longitudinal servo motor (202), the transverse servo motor (303), and the vertical servo motor (305), while the vision module sends control signals to the two-dimensional gimbal (5) to reset the position of the structured light camera (6) to the zero point O. S3. The equipment control module sends a signal to the longitudinal servo motor (202) to drive the transverse track beam (2) to move along the longitudinal beam (102) from point O to the shooting point A and stop, and sends the signal of reaching point A to the vision module. S4. The vision module sends acquisition signals to the two-dimensional pan-tilt unit (5) and the structured light camera (6) to acquire point cloud data of the A point area. After the acquisition is completed, it sends an acquisition end signal to the device control module. S5. The equipment control module drives the transverse track beam (2) to move from point A to the shooting point B and stop. The vision module controls the two-dimensional gimbal (5) and the structured light camera (6) to collect point cloud data of the area at point B. After the collection is completed, a signal indicating the end of collection is fed back. S6. The equipment control module drives the transverse track beam (2) to move from point B to the shooting point C and stop. The vision module controls the completion of point cloud data acquisition in the area of ​​point C and sends back the end signal. S7. The equipment control module drives the transverse track beam (2) to move from point C to the shooting point D and stop. The vision module controls the completion of point cloud data acquisition in the area of ​​point D and sends back the end signal. S8. The equipment control module drives the transverse track beam (2) to move from point D to the shooting point E and stop. The vision module controls the completion of point cloud data acquisition in the area of ​​point E and sends back the end signal. S9. The equipment control module drives the transverse track beam (2) to return from point E to point O and stop, and sends a signal to the cloud platform to indicate the end of data collection. S10: The vision module performs post-processing on the collected point cloud data, extracts structural features and builds a model, compares the model with the BIM model, generates a quality inspection report and feeds it back to the cloud platform. S11. If the quality inspection result is qualified, the mold table will enter the next station; if it is unqualified, the worker will handle it according to the quality inspection report and the mold table will enter the next station. S12. The cloud platform stores the point cloud data and quality inspection report of this test; The algorithm steps of the detection method include: A1. Import the original point cloud data of the concealed project of the prefabricated component collected by the structured light camera (6) into the algorithm system. Through coordinate normalization processing, the point cloud data is uniformly mapped to the preset three-dimensional coordinate system to eliminate the data deviation caused by the difference in coordinate systems of different shooting points. At the same time, the acquisition time of each point cloud is recorded. Corresponding shooting locations and initial coordinates This serves as a foundation for subsequent data tracing and analysis; A2. Based on the local density distribution characteristics of point cloud data, the K-nearest neighbor search algorithm is used to calculate the average distance of the K nearest neighbor points around each point cloud. 1.2-1.5 times the average distance is used as the local density threshold to achieve dynamic adaptation of the point cloud density threshold in different regions. A3. Compare the local density of each point cloud with the corresponding dynamic density threshold. If the local density of a point cloud is lower than the threshold, it is determined to be a noise point. Noise points include invalid point clouds caused by environmental dust and light reflection interference. Through the neighborhood point cloud collaborative verification mechanism, the density of the three neighboring points around the suspected noise point is verified a second time. After confirmation, the noise point is removed to avoid the accidental deletion of effective edge point clouds. A4. Using a curvature-based raster sampling method, the preprocessed point cloud data is divided into three-dimensional raster units with a raster size of 5-10 mm. Each raster is denoted as... u, v, w are the indices of the raster along the x, y, and z axes. Calculate the curvature of each point cloud within the raster. Curvature calculation requires first solving the covariance matrix of the point cloud neighborhood. The formula is: ； in , where is the number of neighborhood points selected when calculating curvature. The average coordinates of the neighboring points. Similarly, for the y and z axes, the covariance matrix... Perform eigenvalue decomposition to obtain eigenvalues. Point cloud curvature The formula is: ； A5. Perform multi-scale sampling on the processed point cloud data to construct feature layers at scales of 10mm, 20mm, and 30mm. The 10mm small-scale layer is used to extract detailed features of embedded parts and small structures such as junction boxes; the 20mm medium-scale layer is used to extract the arrangement features of reinforcing bars; and the 30mm large-scale layer is used to extract the overall outline features of the structure, forming a multi-scale feature pyramid. The point cloud collections of the scale layers are then analyzed. satisfy: ； For the point cloud set at the k-th scale layer; Let be the sampling radius of the k-th scale layer; For post-processing point clouds The three-dimensional coordinates.

2. The detection method of the robot for detecting concealed works of prefabricated building components according to claim 1, characterized in that: The algorithm steps of the detection method include: A6. In each feature layer at each scale, the Fast Point Feature Histogram algorithm is used to calculate the local feature descriptor for each point cloud, focusing on capturing the normal vector, curvature, and neighborhood point distribution information of the point cloud. For linear structures with reinforced steel bars, the direction vector feature is calculated in addition, which is used for subsequent identification of steel bar direction and spacing. The FPFH algorithm is a Fast Point Feature Histogram algorithm. Fast point feature histogram algorithm calculates each point cloud Local feature descriptors The core is to capture the normal vector, curvature, and neighborhood point distribution information of the point cloud; for point clouds and its neighboring points Define the unit vector of the line connecting two points. , Representing the magnitude of a vector, point cloud unit normal vector Through principal component analysis (PCA), the three key angular features in the fast point feature histogram algorithm are: ; Will The data was divided into 11, 11, and 5 intervals for binning and statistical analysis, resulting in a 125-dimensional feature vector generated by the fast point feature histogram algorithm. For linear structures with reinforced steel bars, additional direction vectors are calculated. , To calculate the number of neighborhood points when calculating the direction vector and supplement it into the feature vector, the recognition accuracy of steel-reinforced structures is improved. Point clouds The unit normal vector; Point cloud for steel reinforcement The direction vector; A7. Preliminary classification of structures: Based on the preset feature library of concealed engineering structures, including the standard feature parameters of embedded parts, reinforcing bars, and junction boxes, the support vector machine (SVM) classifier is used to perform preliminary classification of the point clouds of feature layers at each scale. The point clouds are divided into embedded parts, reinforcing bars, junction boxes, and background. The classification accuracy threshold is set to 95%, and point clouds below the threshold are marked as unverified. A8. Cross-scale feature fusion and secondary recognition: For the point cloud of the class to be verified in the preliminary classification, extract its feature information at different scale levels, give higher weight to key scale features through the attention mechanism, perform cross-scale feature fusion, and then use the K-nearest neighbor classifier for secondary recognition to correct the preliminary classification error and finally determine the point cloud of each structure. A9. Calculation of structural parameters: For the classified point cloud of the structure, the minimum bounding rectangle algorithm is used to calculate the length, width, height and center coordinates of the embedded parts and junction boxes. The Hough transform algorithm is used to detect the axis equation of the reinforcing steel bars and calculate the diameter, length, spacing and orientation angle of the reinforcing steel bars to form a set of structural feature parameters. A10. Retrieve the BIM model of the corresponding prefabricated component from the cloud platform and convert it into a coordinate system consistent with the measured point cloud data. The conversion formula is as follows: ； R represents the center coordinates of the k-th structure in the transformed BIM model; R is a 3×3 rotation matrix. It is a 3×1 translation vector; Extract standard feature parameters of hidden engineering structures from the BIM model and construct a standard feature parameter library; A11. Using the Sample Consistency Initial Registration (SAC-IA) algorithm, 100-200 feature points are randomly sampled from the measured structure point cloud and the BIM model point cloud respectively. Through FPFH feature descriptor matching, 30-50 pairs of feature points with the highest matching degree are selected. The initial transformation matrix is ​​calculated to achieve the initial alignment between the measured model and the BIM model. The registration error is controlled within 5mm. A12. Based on the initial transformation matrix, an improved Iterative Closest Point (ICP) algorithm is used for fine registration. A weighting factor is introduced to give higher weight to the key feature points of the structure. During the iteration process, the distance from the point to the surface is used as the error measurement standard. An iteration termination condition is set to finally achieve high-precision registration between the measured model and the BIM model, with the registration error controlled within 2mm. A13. Deviation Calculation and Result Judgment: For the registered measured model and BIM model, deviation calculations are performed on the characteristic parameters of the corresponding structures, including positional deviation, dimensional deviation, and layout deviation. Each deviation value is compared with the preset quality acceptance threshold. If all deviation values ​​are within the threshold range, the inspection is deemed qualified. If any deviation value exceeds the threshold, it is marked as unqualified and the specific value and location of the deviation are recorded. A14. Convert historical quality inspection reports stored on the cloud platform into structured data, build a quality problem database, and encode the types of non-conformities; A15. The improved Apriori algorithm is used to mine frequent itemsets in the quality problem database. The minimum support is set to 5% and the minimum confidence is set to 70%. Redundant candidate sets are removed by pruning strategy to mine frequently occurring quality problem combinations. A16. Association rule generation and analysis: Based on the mined frequent itemsets, generate association rules for quality issues, calculate the lift of association rules, select strong association rules with a lift ≥ 1.2, and analyze the intrinsic associations between quality issues; A17. Quality improvement suggestion generation: Combining strong association rules and construction process flow, the key causes of quality problems are identified. The effectiveness of historical treatment measures is evaluated using decision tree algorithm, effective treatment measures are selected, and targeted quality improvement suggestions are generated and fed back to the construction stage to achieve continuous optimization of the production quality of precast components.

3. The detection method of the robot for detecting concealed works of prefabricated building components according to claim 1, characterized in that: The track frame (1) includes a frame structure, a transverse track beam (2) between the top two sides of the track frame (1), a template (8) for placing the object to be detected is provided at the bottom of the track frame (1), the two ends of the transverse track beam (2) are slidably connected to the top two sides of the track frame (1), at least one vertical lifting frame (4) is provided on the transverse track beam (2), a two-dimensional gimbal (5) is provided at the end of the vertical lifting frame (4), and a structured light camera (6) is provided on the two-dimensional gimbal (5). The object to be detected is set on the mold platform (8), and the structured light camera (6) performs visual detection on the object to be detected.

4. The detection method of the robot for detecting concealed works of prefabricated building components according to claim 1, characterized in that: The transverse track beam (2) is slidably connected to the transverse sleeve (3) on one side. A transverse servo motor (303) is provided on one side of the transverse sleeve (3) to drive the transverse sleeve (3) to slide on one side of the transverse track beam (2). The vertical lifting frame (4) is slidably connected to the transverse sleeve (3) up and down. The other side of the horizontal sleeve (3) is equipped with a vertical servo motor (305), which drives the vertical lifting frame (4) to move up and down.

5. The detection method of the robot for detecting concealed works of prefabricated building components according to claim 4, characterized in that: A transverse track (204) is provided on the transverse beam (201) of the transverse track beam (2). The transverse sleeve (3) is slidably connected to the transverse track (204). A transverse rack (205) is provided on the transverse beam (201). The gear at the output end of the transverse servo motor (303) meshes with the transverse rack (205). The transverse servo motor (303) drives the transverse sleeve (3) to slide on the transverse beam (201).

6. The detection method of the robot for detecting concealed works of prefabricated building components according to claim 4, characterized in that: The middle part of the transverse sleeve (3) is a hollow rectangular frame structure with a protrusion. The vertical rod (401) of the vertical lifting frame (4) is provided with a vertical rail (402). The slider on the inner side of the transverse sleeve (3) is slidably connected to the vertical rail (402). The vertical rod (401) is also provided with a vertical rack (403). The gear on the vertical servo motor (305) meshes with the vertical rack (403). The vertical servo motor (305) drives the vertical rod (401) to slide up and down in the transverse sleeve (3).

7. The detection method of the robot for detecting concealed works of prefabricated building components according to claim 6, characterized in that: The lower end of the vertical rod (401) is connected to the base (501) of the two-dimensional gimbal (5), the rotating ends on both sides of the gimbal body (502) of the two-dimensional gimbal (5) are connected to the two ends of the rotating frame (503), and the lower end of the rotating frame (503) is connected to the structured light camera (6).

8. The detection method of the robot for detecting concealed works of prefabricated building components according to claim 4, characterized in that: The transverse track beam (2) is provided with longitudinal sliders (203) at both ends. The longitudinal sliders (203) are slidably connected to the longitudinal rail (103) on the longitudinal beam (102) of the track frame (1). The longitudinal rail (103) is provided with a longitudinal rack (104). The transverse track beam (2) is provided with longitudinal servo motors (202) at both ends. The gear at the output end of the longitudinal servo motor (202) meshes with the longitudinal rack (104). The two longitudinal servo motors (202) drive the transverse track beam (2) to slide on the longitudinal beam (102).

9. The detection method of the robot for detecting concealed works of prefabricated building components according to claim 4, characterized in that: The longitudinal beam (102) is supported by multiple columns (101) and is set between the lower parts of the multiple columns (101); Longitudinal railings (107) are provided on both sides of the longitudinal beam (102); An electrical control cabinet (7) is installed on the upper part of the crossbeam (201) of the transverse track beam (2).

10. The detection method of the robot for detecting concealed works of prefabricated building components according to claim 3, characterized in that: It also includes a cloud-based production collaborative control system. The cloud platform interfaces with the host computer of the production line and the host computer of the robot. The cloud platform has built-in BIM models of all prefabricated components and retrieves the corresponding BIM model according to the current inspection requirements and transmits it to the host computer of the robot. The robot's host computer includes a vision module and an equipment control module. The vision module is used for controlling the two-dimensional gimbal (5), acquiring and processing point cloud data, and outputting quality inspection reports. The equipment control module is used to control longitudinal, lateral, and vertical movement.

Images