Bridge bearing damage identification and three-dimensional reconstruction wall-climbing robot and detection method thereof

Abstract

This invention discloses a bridge bearing damage identification and 3D reconstruction climbing robot and its detection method, belonging to the field of civil engineering structure inspection and robotics technology. The robot includes a negative pressure adsorption chassis, a walking mechanism, a five-degree-of-freedom robotic arm, and an RGB-D vision perception module. The robot crawls on the surface of bridge piers and uses its robotic arm to perform close-range, multi-view observations of the bearings. The onboard computing unit runs a YOLO network to locate the damaged area, performs TSDF voxel fusion on the depth data based on camera pose, generates a dense 3D model of the damaged surface, and calculates quantitative indicators such as crack length, spalling area, and pitting volume on the model. This invention achieves efficient automated inspection of bearing areas, observation of hidden parts, and 3D quantification of damage, solving the problems of high risk and difficulty in data quantification associated with manual inspection.

Description

Technical Field

[0001] This invention relates to the field of civil engineering structure inspection and robotics technology, particularly to a bridge bearing damage identification and 3D reconstruction climbing robot and its inspection method. Background Technology

[0002] As a critical force-transmitting component connecting the superstructure and substructure of a bridge, the health of bridge bearings directly affects the overall safety and durability of the bridge. During long-term service, bearings are highly susceptible to defects such as rubber aging and cracking, steel corrosion, and abnormal displacement; therefore, regular condition inspections are crucial. However, because bearings are typically located in the narrow space between the cap beam and the bridge structure, and are situated at a relatively high position, inspection work faces significant challenges.

[0003] Currently, in engineering practice, close-range visual inspections and measurements are mainly conducted manually using inspection vehicles or scaffolding. This method is not only labor-intensive and inefficient, but also exposes inspectors to a high risk of falls from heights. Furthermore, the limited operating space makes it difficult to effectively observe concealed sides. While drone inspection technology has emerged in recent years, drones suffer from poor flight stability under bridges due to GPS signal obstruction and airflow. They also struggle to penetrate deep into support gaps for close-range imaging, only acquiring distant two-dimensional images. This fails to accurately obtain three-dimensional quantitative information such as the depth and volume of defects, resulting in inspection results that are often only qualitative descriptions, lacking precise data support and failing to meet the demands of modern, refined bridge maintenance. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a bridge bearing damage identification and 3D reconstruction wall-climbing robot and its detection method. By using a negative pressure adsorption chassis to carry a five-degree-of-freedom robotic arm and a visual perception module, and combining the target detection network YOLO damage identification and the truncated symbolic distance field TSDF voxel fusion algorithm, the robot can realize automated close-range observation and high-precision 3D quantization of bearing damage in narrow spaces.

[0005] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows: In a first aspect, the present invention provides a bridge bearing damage identification and three-dimensional reconstruction wall-climbing robot, comprising: Motion module, adsorption module, robotic arm module, vision perception module, and control and communication module; The motion module includes a robot chassis and a walking mechanism disposed on both sides of the robot chassis, used to drive the robot to move on a vertical or inclined structural surface; The adsorption module is located in the middle of the bottom of the robot chassis, and includes a negative pressure source and a negative pressure sealing cavity surrounded by a flexible sealing structure, which is used to form a negative pressure adsorption force on the surface of the structure. The robotic arm module includes a five-degree-of-freedom robotic arm mounted on the robot chassis, and a camera mounting base is provided at the end of the five-degree-of-freedom robotic arm; The visual perception module is mounted on the camera mounting base and includes a depth camera and a supplementary lighting component for acquiring image data and depth data. The visual perception module specifically includes an RGB camera and a depth camera or an integrated RGB-D camera. The control and communication module includes an airborne computing unit, which is electrically connected to the walking mechanism, the negative pressure source, the five-degree-of-freedom robotic arm, and the vision perception module. The airborne computing unit is configured to: identify bridge supports and their damage in the image data based on the target detection network YOLO and determine the region of interest (ROI); perform truncated symbolic distance field (TSDF) voxel fusion on the depth data of the ROI based on the camera pose estimation results to generate a dense 3D model; and calculate damage quantization parameters on the dense 3D model.

[0006] Furthermore, the walking mechanism is a wheeled mechanism or a tracked mechanism, and the walking mechanism includes a drive motor. The onboard computing unit controls the speed difference and direction of the drive motors on both sides to realize the robot's straight-line movement, turning, and turning in place.

[0007] Furthermore, the flexible sealing structure is made of silicone, rubber or polyurethane elastic material, or adopts a composite structure of brush sealing skirt and elastic sealing ring to compensate for the roughness of the working surface and the unevenness of the joint.

[0008] Furthermore, the five-degree-of-freedom robotic arm includes a base rotation joint, a shoulder joint, an elbow joint, a wrist pitch joint, and a wrist yaw joint. The onboard computing unit plans a multi-view scanning pose sequence based on the spatial location of the region of interest (ROI) and controls the execution of the five-degree-of-freedom robotic arm.

[0009] Furthermore, the adsorption module also includes a pressure sensor. The onboard computing unit adjusts the rotation speed or valve opening of the negative pressure source in a closed loop based on the real-time reading of the pressure sensor to maintain the target negative pressure value, and judges the leakage situation based on the pressure change rate to trigger a docking or evacuation strategy.

[0010] In a second aspect, the present invention provides a bridge bearing detection method using a bridge bearing damage identification and 3D reconstruction climbing robot as described in the first aspect, comprising the following steps: Step S1, Wall Attachment and Initialization: Activate the negative pressure source of the adsorption module to attach the robot to the structural surface and complete the calibration of the visual perception module and the robotic arm module; Step S2, Area Search and Localization: Control the movement of the motion module, and the airborne computing unit runs the YOLO target detection network to identify the bridge support area; Step S3, Damage Identification and Region of Interest (ROI) Determination: Damage detection is performed on the support area image, and the damage category and location are output. The area whose confidence level meets the preset conditions is determined as the region of interest (ROI). Step S4, Scan Trajectory Planning: Based on the spatial location of the region of interest (ROI) of damage, generate a camera observation pose sequence and plan the motion trajectory of the five-degree-of-freedom robotic arm; Step S5, Data Acquisition and Pose Estimation: Control the five-degree-of-freedom robotic arm to move along the motion trajectory, acquire multiple frames of RGB images and depth maps, and calculate the camera pose corresponding to each frame; Step S6, 3D Reconstruction: Based on the camera pose, perform truncated symbolic distance field (TSDF) voxel fusion on multiple frames of the depth map, extract isosurfaces to generate a dense 3D model of the region of interest (ROI) of the damage; Step S7, Quantization and Output: Calculate damage quantification parameters on the dense 3D model, generate a detection report and output it.

[0011] Furthermore, in step S3, when the confidence level of damage identification is lower than a preset threshold, the corresponding area is marked as a suspected damage; and in step S4, the number of observation poses is increased or the scanning distance is reduced for the suspected damage area to improve the reconstruction resolution.

[0012] Furthermore, in step S6, the truncated symbolic distance field (TSDF) voxel fusion process includes: adaptively setting the truncation distance according to the ranging range of the depth camera, and performing bilateral filtering and outlier removal on the depth map to reduce the impact of reflection and noise on the fusion result.

[0013] Further, in step S7, the damage quantification parameter includes pitting depth, which is calculated by performing local plane fitting or quadratic surface fitting on the undamaged reference surface in the dense three-dimensional model, calculating the distance from the point cloud of the damaged area to the reference surface, and statistically obtaining the maximum depth and average depth.

[0014] Furthermore, the airborne computing unit is connected to an inertial measurement unit (IMU). In step S5, the data from the IMU, the kinematic data of the robotic arm, and the visual odometry calculation method are combined to perform a joint calculation to obtain a high-precision camera pose.

[0015] The present invention has the following beneficial effects: (1) This invention adopts an integrated design, deeply integrating the negative pressure adsorption walking chassis with a five-degree-of-freedom vision robotic arm, which is like equipping the inspection equipment with a sturdy gecko foot and a flexible probing hand, solving the problems of high-altitude operation risks and difficulty in penetrating narrow spaces in traditional manual inspection. Through the overall control of the onboard computing unit, the robot can not only stably adhere to the surface of vertical or inclined bridge piers, but also use the robotic arm to send the sensing terminal into the narrow support gap, realizing the full-process automation from close observation to damage quantification, significantly reducing inspection costs and improving operational safety.

[0016] (2) In terms of adsorption and motion control, this invention constructs an elastic negative pressure cavity that can adapt to the rough surface of concrete through the synergistic effect of a flexible sealing structure and a negative pressure source. It locks the wall surface tightly like a powerful suction cup, effectively compensating for the risk of air leakage caused by micro-cracks or uneven joints on the bridge surface. Combined with the independent left and right driving walking mechanism, the robot has the ability to move flexibly on complex facades and turn in place, ensuring that it can maintain a stable posture for operation even under wind loads or vibration environments.

[0017] (3) In terms of damage identification and reconstruction, this invention utilizes the multi-view scanning capability of an RGB camera combined with a five-degree-of-freedom robotic arm to overcome occlusion and reflection interference under a single viewpoint. The target detection network YOLO is used to quickly locate the damaged area, and the truncated symbolic distance field (TSDF) voxel fusion algorithm is combined to perform high-precision stitching of depth data, which is like performing a micron-level CT scan on the damaged area. The generated high-fidelity dense three-dimensional model can accurately reflect the three-dimensional morphology of cracks, spalling and corrosion, making up for the deficiency of traditional two-dimensional image detection in obtaining depth information.

[0018] (4) The method provided by this invention constructs a complete detection closed loop from macroscopic positioning to microscopic quantification, transforming subjective visual inspection into objective data measurement. By directly calculating key indicators such as crack length, spalling area and pitting volume on the reconstructed three-dimensional model, and generating a digital inspection report containing spatial location information, it provides a traceable and quantifiable scientific basis for bridge maintenance decisions, and realizes the digital transformation of bridge bearing operation and maintenance management. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the technical route of the present invention; Figure 2 This is a schematic diagram of the overall structure of the wall-climbing robot for bridge bearing damage identification and 3D reconstruction according to the present invention. Figure 3 This is a schematic diagram of the structure of the motion module and the adsorption module of the present invention; Figure 4 This is a schematic diagram of the structure of the five-degree-of-freedom robotic arm module and the vision perception module of the present invention; Figure 5 This is a schematic diagram of the working scenario where the robot of the present invention performs support detection on the surface of a bridge pier.

[0020] The components include: 1. Motion module; 11. Robot chassis; 12. Walking mechanism; 2. Adsorption module; 21. Negative pressure source; 22. Negative pressure sealing cavity; 23. Flexible sealing structure; 3. Robotic arm module; 31. Five-degree-of-freedom robotic arm; 32. End-effector camera mounting base; 4. Visual perception module; 41. Depth camera; 42. Lighting assembly; 5. Control and communication module. Detailed Implementation

[0021] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. Those skilled in the art should understand that adjustments can be made to the specific structures and parameters in the following embodiments without departing from the concept of the present invention, and all such adjustments fall within the protection scope of the present invention.

[0022] In the description of this invention, it should be understood that the terms left side, right side, upper part, lower part, etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. The terms first, second, etc. do not indicate the importance of the components, and therefore should not be construed as limiting the invention.

[0023] like Figures 1 to 3 As shown, the bridge bearing damage identification and 3D reconstruction wall-climbing robot provided in this embodiment mainly includes a motion module 1, an adsorption module 2, a robotic arm module 3, a vision perception module 4, and a control and communication module 5. The motion module 1, adsorption module 2, robotic arm module 3, and vision perception module 4 are integrated and installed on the robot chassis 11 to enable the robot to adhere and walk on the vertical or inclined surface of the bridge structure and perform close-range detection of the bridge bearings.

[0024] like Figure 2 and Figure 3As shown, the motion module 1 includes a robot chassis 11 and walking mechanisms 12 disposed on both sides of the robot chassis 11. In this embodiment, the walking mechanism 12 can be a wheeled mechanism, comprising a drive motor and drive wheels connected to the drive motor, used to drive the robot to move on vertical or inclined structural surfaces such as bridge piers, the sides of cap beams, or the web. The onboard computing unit controls the speed and direction of the drive motors of the left and right walking mechanisms 12 respectively through a motor drive board, realizing the robot's straight-line movement, turning, and in-situ turning to meet the positioning and close-to-work requirements of the support area. In some embodiments, the walking mechanism 12 can also be a tracked mechanism, which also includes a drive motor and drive wheels connected to the drive motor, and its control method is the same as that of the wheeled mechanism.

[0025] like Figure 3 As shown, the adsorption module 2 is located at the bottom center of the robot chassis 11, used to generate negative pressure adsorption force on the surface of the bridge structure, achieving reliable robot adhesion and anti-slip performance. The adsorption module 2 includes a negative pressure source 21, a negative pressure sealing cavity 22, and a flexible sealing structure 23. The negative pressure source 21 can be a vacuum pump, a ducted fan, or a combination of both. During operation, it draws air from the negative pressure sealing cavity 22, making the pressure inside the cavity lower than the external atmospheric pressure, thereby generating adsorption force. The negative pressure sealing cavity 22 and the robot chassis 11 form a closed or quasi-closed space, which is the main structure generating the negative pressure difference. The flexible sealing structure 23 is located around the periphery of the negative pressure sealing cavity 22, used to compensate for the roughness, pores, and unevenness of the concrete surface and joints, improving sealing performance and reducing the risk of air leakage.

[0026] The flexible sealing structure 23 can be made of elastic materials such as silicone, rubber, or polyurethane, or it can be a composite structure of a brush-like sealing skirt and an elastic sealing ring to compensate for the roughness of the working surface and unevenness of the joints, thereby improving its adaptability to complex surfaces. During actual operation, once the robot reaches the target wall, the negative pressure source 21 is activated. The negative pressure sealing cavity 22, in conjunction with the flexible sealing structure 23, forms a sealed area with the wall surface. The robot generates a stable adsorption force and maintains its attachment, providing a safe foundation for subsequent walking and robotic arm inspection.

[0027] In a preferred embodiment, the adsorption module 2 further includes a pressure sensor for real-time monitoring of the pressure value within the negative pressure sealing chamber 22. The onboard computing unit uses the real-time readings from the pressure sensor to adjust the rotational speed of the negative pressure source 21 or the valve opening in a closed loop to maintain the target negative pressure value and ensure the stability of the adsorption force. Furthermore, the onboard computing unit can also determine leakage conditions such as sudden or slow leakage based on the pressure change rate. When an abnormal pressure drop is detected, it can trigger deceleration, docking, or evacuation strategies to prevent the robot from detaching due to insufficient adsorption force, thereby improving system robustness and operational safety. The negative pressure source 21 can also be connected to a negative pressure buffer tank and a one-way valve to improve system stability.

[0028] like Figure 2 and Figure 4 As shown, the robotic arm module 3 includes a five-degree-of-freedom (DOF) robotic arm 31 mounted on the robot chassis 11, with a camera mounting base 32 at its end. The five-DOF robotic arm 31 includes a base rotation joint, a shoulder joint, an elbow joint, a wrist pitch joint, and a wrist yaw joint. The combination of these joints enables spatial pose adjustment and multi-view scanning of the end-effector. The onboard computing unit plans a multi-view scanning pose sequence based on the spatial location of the region of interest (ROI) and controls the five-DOF robotic arm 31 to execute it. In this embodiment, the five-DOF robotic arm 31 can achieve at least the following motion capabilities: base rotation, shoulder pitch, elbow pitch, and wrist pose adjustment. This allows for close-range, multi-view coverage observation of the damaged area of ​​the support even under conditions of limited or obstructed space around the support. The robotic arm module 3 allows the robot to complete multi-directional scanning without frequent chassis pose adjustments, improving detection efficiency and data integrity.

[0029] like Figure 4 As shown, the visual perception module 4 is mounted on the camera mounting base 32 and includes a depth camera 41 and a supplementary lighting component 42 for acquiring image and depth data. Specifically, the visual perception module 4 includes an RGB camera and a depth camera or an integrated RGB-D camera. The depth camera 41 is used to acquire depth information of the support surface and its damaged areas, outputting depth maps or point cloud data. The RGB images are used for support and damage identification. The depth camera 41 can employ any of the following: structured light, time-of-flight (ToF), or binocular depth cameras, to adapt to different operating distances, ambient light, and surface material conditions.

[0030] The supplementary lighting component 42 is used to provide auxiliary illumination in shadowed, low-light, or high-contrast environments at the bottom of the bridge, reducing the impact of image noise and texture loss on detection and reconstruction, thereby improving the usability and consistency of the acquired data. The supplementary lighting component 42 can employ a ring-shaped LED or a strip-shaped LED structure and can adjust its brightness according to the ambient light to avoid depth anomalies caused by overexposure and reflections. In a preferred embodiment, the supplementary lighting component 42 can be synchronously triggered with the camera or adaptively adjusted in brightness based on exposure feedback to reduce the impact of shadows and reflections on damage identification.

[0031] The control and communication module 5 includes an onboard computing unit, which is electrically connected to the walking mechanism 12, the negative pressure source 21, the five-degree-of-freedom robotic arm 31, and the vision perception module 4. The control and communication module 5 also includes a motor drive board, a power supply module, and a wireless communication module. The onboard computing unit controls each actuator through the motor drive board and connects to the ground workstation through the wireless communication module to achieve remote monitoring and data transmission.

[0032] The airborne computing unit is configured to identify bridge bearings and their damage in image data and determine the Region of Interest (ROI) based on the YOLO object detection network. Specifically, the airborne computing unit runs the YOLO-based object detection network to locate and identify bridge bearings and their damage in RGB images, including cracks, spalling, corrosion, missing or loose bolts, slip surface wear or contamination, etc., determine the ROI, and output the damage category, location, and confidence level.

[0033] The onboard computing unit is also configured to perform truncated symbolic distance field (TSDF) voxel fusion on the depth data of the region of interest (ROI) based on camera pose estimation results to generate a dense 3D model. Specifically, based on camera calibration parameters, the kinematics of the five-DOF robotic arm 31, and camera pose estimation results, the onboard computing unit performs truncated symbolic distance field (TSDF) voxel fusion on the depth map corresponding to the ROI to generate a dense 3D model and mesh of the damage surface. TSDF voxel fusion includes performing truncated distance calculation and confidence-weighted update on each frame of the depth map to obtain the TSDF values ​​and weights in the voxel mesh, and extracting isosurfaces from the voxel mesh using the Marching Cubes algorithm to generate a triangular mesh as the dense 3D model.

[0034] The airborne computing unit further calculates damage quantification parameters on the dense 3D model. These parameters include at least one or more of the following: crack length, spalling area, maximum pitting depth, pitting volume, or corrosion coverage. The airborne computing unit correlates the quantification parameters with the support component number, spatial location, and image evidence to generate an inspection report, and also correlates the identification and quantification results with the support's spatial location, outputting the inspection report. The inspection results can be mapped onto the support's 3D model or the bridge component topology diagram and visualized using color or symbols.

[0035] In a preferred embodiment, the airborne computing unit is connected to an inertial measurement unit (IMU). The airborne computing unit runs visual odometry or visual-inertial odometry to estimate the camera pose, which is used for voxel coordinate transformation in TSDF voxel fusion. Specifically, the airborne computing unit uses the kinematic data of the robotic arm, the data from the IMU, and visual odometry to perform joint calculations to obtain a high-precision camera pose, thereby improving the accuracy of 3D reconstruction.

[0036] like Figure 5As shown, in a typical bridge bearing inspection task, the robot in this embodiment can work as follows: After the robot is driven by the motion module 1 to reach the surface of the target bridge component, the adsorption module 2 activates the negative pressure source 21, achieving stable attachment under the action of the negative pressure sealing cavity 22 and the flexible sealing structure 23. Subsequently, the walking mechanism 12 is controlled to approach the bridge bearing area along a preset path. The robotic arm module 3 drives the five-degree-of-freedom robotic arm 31 to move, so that the camera mounting base 32 carries the vision perception module 4 to perform multi-view scanning of the bearing surface. The depth camera 41 collects multiple frames of depth data, and the supplementary lighting component 42 provides stable illumination simultaneously. Based on the multi-view depth data, the damaged area is reconstructed in three dimensions. The depth data of each frame are fused in a unified coordinate system to form a dense three-dimensional model, and the mesh can be further extracted for subsequent geometric quantitative analysis, such as the morphology of the damaged area, the depth of the depression, and the volume of the spalling. Finally, the detection results and the three-dimensional model are used for the visualization display and report output of the bearing damage, providing a basis for bridge maintenance.

[0037] This invention also provides a bridge bearing detection method using the aforementioned bridge bearing damage identification and 3D reconstruction climbing robot, comprising the following steps: Step S1, Wall Attachment and Initialization: Activate the negative pressure source 21 of the adsorption module 2 to attach the robot to the structural surface, and complete the calibration of the vision perception module 4 and the robotic arm module 3. Specifically, activate the negative pressure source 21 of the adsorption module 2 to attach the robot to the bridge structural surface, activate the vision perception module 4, and complete the calibration of the camera's intrinsic and extrinsic parameters, the calibration of the hand-eye relationship between the camera and the five-DOF robotic arm 31, and the zero-position calibration of the five-DOF robotic arm 31.

[0038] Step S2, Region Search and Localization: Control the movement of motion module 1, and the onboard computing unit runs the YOLO object detection network to identify the bridge support area. Specifically, control motion module 1 to crawl along a preset path to acquire continuous RGB images, and the onboard computing unit runs the YOLO object detection network to detect the bridge support area and output the support bounding box.

[0039] Step S3: Damage Identification and Region of Interest (ROI) Determination: Damage detection is performed on the support area image, outputting the damage category and location. Regions with confidence levels meeting preset conditions are determined as ROIs. Specifically, damage detection is performed on the image within the support frame, outputting the damage category, location, and confidence level. Damage frames with confidence levels higher than a preset threshold are determined as ROIs. When the confidence level of damage identification is lower than the preset threshold, the corresponding area is marked as a suspected damage. In step S4, the number of observation poses is increased or the scanning distance is decreased for suspected damage areas to improve reconstruction resolution.

[0040] Step S4, Scan Trajectory Planning: Based on the spatial location of the region of interest (ROI) of damage, a sequence of camera observation poses is generated and the motion trajectory of the five-DOF robotic arm 31 is planned. Specifically, based on the spatial location of the ROI of damage and the estimation of the surface normal, several camera observation poses are generated and the scan trajectory of the five-DOF robotic arm 31 is planned.

[0041] Step S5, Data Acquisition and Pose Estimation: Control the five-DOF robotic arm 31 to move along the motion trajectory, acquire multiple frames of RGB images and depth maps, and calculate the camera pose corresponding to each frame. Specifically, control the five-DOF robotic arm 31 to move along the scanning trajectory and acquire multiple frames of RGB images and depth maps. Calculate the camera pose for each frame using the kinematic data of the five-DOF robotic arm 31, visual odometry or visual-inertial odometry, and hand-eye calibration. In a preferred embodiment, the data from the inertial measurement unit (IMU), the robotic arm kinematic data, and the visual odometry are combined for joint calculation to obtain a high-precision camera pose.

[0042] Step S6, 3D Reconstruction: Based on the camera pose, truncated symbolic distance field (TSDF) voxel fusion is performed on multiple depth maps to extract isosurfaces and generate a dense 3D model of the region of interest (ROI) for the damage. Specifically, TSDF voxel fusion is performed on multiple depth maps corresponding to the ROI in a unified coordinate system to obtain the TSDF values ​​and weights of the voxel mesh. A triangular mesh is then extracted using the Marching Cubes algorithm to generate a dense 3D model. The truncated symbolic distance field (TSDF) voxel fusion process includes adaptively setting the truncation distance according to the ranging range of the depth camera 41, and performing bilateral filtering and outlier removal on the depth map to reduce the impact of reflections and noise on the fusion result. The TSDF truncation distance and voxel resolution can be adaptively set according to the camera ranging range and the target damage scale.

[0043] Step S7, Quantization and Output: Calculate damage quantification parameters on the dense 3D model, generate an inspection report, and output it. Specifically, calculate quantification parameters such as crack length, spalling area, pitting depth, or pitting volume on the dense 3D model, and generate an inspection report with damage category, confidence level, quantification parameters, and support spatial location, then transmit it back to the ground workstation. Damage quantification parameters include pitting depth, which is calculated by performing local planar fitting or quadratic surface fitting on the undamaged reference surface in the dense 3D model, calculating the distance from the damaged area point cloud to the reference surface, and statistically obtaining the maximum and average depths. The inspection results can be mapped to the support 3D model or bridge component topology diagram and visualized using color or symbols.

[0044] This invention integrates the negative pressure adsorption module 2 with the walking mechanism 12, enabling the robot to reliably attach and maneuver to the support area, reducing the risk of personnel working at heights. A five-DOF robotic arm 31 enables multi-view close-range observation in confined spaces, improving the effective imaging rate and coverage of minor damage, such as fine cracks and missing bolts. YOLO is used for rapid positioning of the support and damaged area, and ROI-guided TSDF voxel fusion reconstruction is employed, balancing real-time performance and reconstruction stability. Quantitative indicators such as crack length, spalling area, pitting depth, or pitting volume are output, providing measurable data for maintenance decisions. Pressure closed-loop control and leakage detection strategies enhance adsorption reliability and improve system robustness under complex surface conditions.

[0045] The preferred embodiments of the present invention have been described in detail above, but the present invention is not limited to the specific details of the above embodiments. Within the scope of the inventive concept, various equivalent modifications can be made to the technical solutions of the present invention, and all such equivalent modifications fall within the protection scope of the present invention.

Claims

1. A wall-climbing robot for bridge bearing damage identification and 3D reconstruction, characterized in that, include: Motion module (1), adsorption module (2), robotic arm module (3), vision perception module (4), and control and communication module (5); The motion module (1) includes a robot chassis (11) and walking mechanisms (12) disposed on both sides of the robot chassis (11) for driving the robot to move on a vertical or inclined structural surface; The adsorption module (2) is located at the bottom center of the robot chassis (11), including a negative pressure source (21) and a negative pressure sealing cavity (22) surrounded by a flexible sealing structure (23), which is used to form a negative pressure adsorption force on the surface of the structure. The robotic arm module (3) includes a five-degree-of-freedom robotic arm (31) mounted on the robot chassis (11), and a camera mounting base (32) is provided at the end of the five-degree-of-freedom robotic arm (31). The visual perception module (4) is installed on the camera mounting base (32) and includes a depth camera (41) and a supplementary lighting component (42) for acquiring image data and depth data. The visual perception module (4) specifically includes an RGB camera and a depth camera or an integrated RGB-D camera. The control and communication module (5) includes an airborne computing unit, which is electrically connected to the walking mechanism (12), the negative pressure source (21), the five-degree-of-freedom robotic arm (31), and the visual perception module (4). The airborne computing unit is configured to: identify bridge supports and their damage in the image data based on the target detection network YOLO and determine the region of interest (ROI); perform truncated symbolic distance field (TSDF) voxel fusion on the depth data of the ROI based on the camera pose estimation results to generate a dense 3D model; and calculate damage quantization parameters on the dense 3D model.

2. The bridge bearing damage identification and three-dimensional reconstruction wall-climbing robot according to claim 1, characterized in that, The walking mechanism (12) is a wheeled mechanism or a tracked mechanism. The walking mechanism (12) includes a drive motor. The onboard computing unit controls the speed difference and direction of the drive motors on both sides to realize the robot's straight-line movement, turning and turning in place.

3. The bridge bearing damage identification and three-dimensional reconstruction wall-climbing robot according to claim 1, characterized in that, The flexible sealing structure (23) is made of silicone, rubber or polyurethane elastic material, or a composite structure of brush sealing skirt and elastic sealing ring, to compensate for the roughness of the working surface and the unevenness of the joint.

4. The bridge bearing damage identification and three-dimensional reconstruction wall-climbing robot according to claim 1, characterized in that, The five-degree-of-freedom robotic arm (31) includes a base rotation joint, a shoulder joint, an elbow joint, a wrist pitch joint, and a wrist yaw joint. The onboard computing unit plans a multi-view scanning pose sequence based on the spatial location of the region of interest (ROI) and controls the five-degree-of-freedom robotic arm (31) to execute it.

5. The bridge bearing damage identification and three-dimensional reconstruction wall-climbing robot according to claim 1, characterized in that, The adsorption module (2) also includes a pressure sensor. The airborne computing unit adjusts the rotation speed or valve opening of the negative pressure source (21) in a closed loop based on the real-time reading of the pressure sensor to maintain the target negative pressure value, and judges the leakage situation based on the pressure change rate to trigger the docking or evacuation strategy.

6. A method for bridge bearing detection using the bridge bearing damage identification and three-dimensional reconstruction climbing robot as described in claim 1, characterized in that, Includes the following steps: Step S1, Wall Attachment and Initialization: Start the negative pressure source (21) of the adsorption module (2) to make the robot attach to the structural surface and complete the calibration of the visual perception module (4) and the robotic arm module (3); Step S2, Area Search and Positioning: Control the movement of the motion module (1), and the airborne computing unit runs the YOLO target detection network to identify the bridge support area; Step S3, Damage Identification and Region of Interest (ROI) Determination: Damage detection is performed on the support area image, and the damage category and location are output. The area whose confidence level meets the preset conditions is determined as the region of interest (ROI). Step S4, Scan trajectory planning: Based on the spatial location of the region of interest (ROI) of the damage, generate a camera observation pose sequence and plan the motion trajectory of the five-degree-of-freedom robotic arm (31); Step S5, Data Acquisition and Pose Estimation: Control the five-degree-of-freedom robotic arm (31) to move along the motion trajectory, acquire multiple frames of red, green and blue RGB images and depth maps, and calculate the camera pose corresponding to each frame; Step S6, 3D Reconstruction: Based on the camera pose, perform truncated symbolic distance field (TSDF) voxel fusion on multiple frames of the depth map, extract isosurfaces to generate a dense 3D model of the region of interest (ROI) of the damage; Step S7, Quantization and Output: Calculate damage quantification parameters on the dense 3D model, generate a detection report and output it.

7. The bridge bearing testing method according to claim 6, characterized in that, In step S3, when the confidence level of damage identification is lower than a preset threshold, the corresponding area is marked as a suspected damage; and in step S4, the number of observation poses is increased or the scanning distance is reduced for the suspected damage area to improve the reconstruction resolution.

8. The bridge bearing testing method according to claim 6, characterized in that, In step S6, the truncated symbolic distance field (TSDF) voxel fusion process includes: adaptively setting the truncation distance according to the ranging range of the depth camera (41), and performing bilateral filtering and outlier removal on the depth map to reduce the impact of reflection and noise on the fusion result.

9. The bridge bearing testing method according to claim 6, characterized in that, In step S7, the damage quantification parameters include pitting depth, which is calculated by performing local plane fitting or quadratic surface fitting on the undamaged reference surface in the dense three-dimensional model, calculating the distance from the point cloud of the damaged area to the reference surface, and statistically obtaining the maximum depth and average depth.

10. The bridge bearing testing method according to claim 6, characterized in that, The airborne computing unit is connected to an inertial measurement unit (IMU). In step S5, the data from the IMU, the kinematic data of the robotic arm, and the visual odometry calculation method are combined to perform a joint calculation to obtain the high-precision camera pose.

Images