Intelligent detection robot for steel truss bridge and intelligent detection method thereof

Abstract

This invention discloses an intelligent inspection robot and its intelligent inspection method for steel truss bridges. The robot includes a permanent magnet adsorption walking mechanism with ball joints, a multimodal detection module, an edge intelligent processing unit, and a structural adaptive navigation system. The walking mechanism achieves adaptive contact with various structural surfaces such as steel truss bridge members and node plates through four independently driven permanent magnet adsorption track units and ball joint connectors. The multimodal detection module integrates four types of sensors: vision, laser, ultrasound, and acoustic vibration. The edge intelligent processing unit runs a multimodal fusion analysis model. The structural adaptive navigation system achieves tight coupling fusion of IMU and lidar through extended Kalman filtering. The remote communication module supports automatic switching between 5G and BeiDou short message services. This invention addresses the challenges of bridge inspection in complex environments such as large elevation differences, high-intensity earthquake zones, strong winds, large temperature differences, and lack of ground network coverage in western mountainous areas, achieving efficient, accurate, and unmanned intelligent inspection.

Description

Technical Field

[0001] This invention belongs to the field of intelligent inspection equipment technology, specifically relating to an autonomous mobile intelligent robot system for detecting the condition of high-strength bolts at steel truss bridge nodes in complex environments in western mountainous areas, and particularly to an intelligent inspection robot and its intelligent inspection method for steel truss bridges that integrates structural adaptive navigation, multimodal perception and edge intelligent analysis. Background Technology

[0002] Steel truss bridges are one of the mainstream forms of large bridge structures, and their joint connections mainly rely on high-strength bolts. During service, bolts are susceptible to loosening, corrosion, cracking, and even breakage due to alternating loads, environmental corrosion, and vibration relaxation, making them a critical weak point for structural safety. Especially in the mountainous areas of western China, bridges face extreme environmental challenges such as large elevation differences (the height difference between the bridge deck and the valley floor can reach 200-300m), high-intensity earthquake zones (fortification intensity VIII and above), strong winds (above level 6), and large temperature differences (diurnal temperature range exceeding 25℃). The damage and degradation of structural components are accelerated, and the difficulty of inspection and safety risks are significantly increased.

[0003] Currently, bolt inspection of steel trusses mainly relies on manual methods. Inspectors approach bolt nodes by erecting scaffolding, using aerial work platforms, or climbing trusses, and then conduct inspections using visual inspection, tapping, and torque wrenches. Existing technologies have the following problems: First, low inspection efficiency: Steel truss nodes are densely packed, and the number of bolts is enormous; manual inspection is time-consuming and difficult to achieve high-frequency, full-coverage inspections. Second, high safety risks: Inspectors must work at heights and in confined spaces, posing safety hazards such as falls and collisions. The risks are even more pronounced on long-span bridges in western mountainous areas, where variable weather further exacerbates the safety threats. Third, highly subjective inspection results: There is a lack of unified standards for manually judging the degree of bolt loosening and corrosion, and the quality of inspections depends on the experience of the personnel. Fourth, difficulty in data traceability: Inspection results are mostly recorded on paper, making it difficult to form a systematic approach. Fifth, existing inspection robots have poor adaptability: most publicly available wall-climbing robots are suitable for planar or regular curved surface structures, making it difficult to adapt to the complex spatial truss structure, abrupt node changes, and irregular bolt distribution of steel trusses. Furthermore, they lack dedicated inspection modules for bolt conditions. Inspection devices for rail bolts or general bolts cannot be directly applied to the open-air, large-span, multi-node complex environment of steel truss bridges. Sixth, they lack environmental adaptability in western mountainous areas: most existing technologies are designed for plains or single structural forms, failing to consider the harsh conditions of western mountainous areas, such as large elevation differences, strong winds, large temperature differences, and lack of ground network coverage. For example, robot adsorption stability is insufficient in strong wind environments; large temperature differences cause sensor accuracy drift; areas without ground networks cannot transmit inspection data in real time; and the sparse features of large-span truss structures cause traditional SLAM positioning to fail.

[0004] Therefore, there is an urgent need for an intelligent robot and its detection method that can autonomously adapt to the complex environment of steel truss structures in the western mountainous areas and perform multi-dimensional and precise detection of bolts. Summary of the Invention

[0005] Therefore, this invention aims to solve the problems of low detection efficiency, high personnel safety risks, strong subjectivity of detection results, poor adaptability of existing robots, especially inability to adapt to the complex environment of western mountainous areas in the existing steel truss bolt detection methods, and to provide an intelligent detection robot for steel truss bridges with structural adaptive navigation, multimodal perception fusion and edge intelligent analysis capabilities and its intelligent detection method.

[0006] This invention is implemented by constructing an intelligent inspection robot for steel truss bridges, comprising: The walking mechanism consists of four independently driven permanent magnet adsorption track units. Each permanent magnet adsorption track unit is connected to the robot's main frame through a ball joint connector. The ball joint connector provides a swing range of ±30°, which is used to achieve adaptive fitting and stable walking on various structural surfaces such as the members and node plates of the steel truss bridge. The multimodal detection module is installed at the end of the five-degree-of-freedom telescopic robotic arm at the front of the robot. This module integrates a high-definition industrial camera, a laser displacement sensor, an electromagnetic ultrasonic transducer, and an acoustic vibration analysis unit consisting of an electromagnetic striking head and a MEMS microphone. It is used to simultaneously acquire visual images, displacement signals, ultrasonic guided wave signals, and acoustic vibration signals of the bolt. The edge intelligent processing unit is electrically connected to the multimodal detection module. It adopts an ARM+NPU heterogeneous computing architecture to run a multimodal fusion analysis model. It performs time synchronization and feature-level fusion of data from four types of sensors: vision, laser, ultrasound, and acoustic vibration, and outputs bolt health status scores in real time. The structure-adaptive navigation system is connected to the walking mechanism and the edge intelligent processing unit. It integrates lidar point cloud, inertial measurement unit data, encoder mileage information and steel truss bridge BIM model. It achieves multi-source information fusion through extended Kalman filtering. It is used to maintain positioning continuity in areas where the structural features of the steel truss bridge are sparse or visually obstructed, and realizes autonomous positioning, path planning and bolt node identification. The remote communication module is used to upload the test results to the cloud management platform in real time.

[0007] Furthermore, the permanent magnet adsorption track unit includes: The track body has an array of permanent magnets embedded on its surface. Each permanent magnet is 20mm×20mm×10mm in size and is arranged in 2 columns×10 rows. The drive motor is connected to the track body for transmission. A force sensor, integrated at the connection between the permanent magnet adsorption track unit and the main frame, is used to monitor the adsorption force in real time. When the adsorption force is lower than the safety threshold, the safety protection module is triggered. The ball joint connector is a ball joint structure that allows each permanent magnet adsorption track unit to rotate in three degrees of freedom relative to the main frame.

[0008] Furthermore, in the multimodal detection module: High-definition industrial cameras identify bolt locations, types, and rusted areas by improving the YOLOv8 model; A laser displacement sensor measures the relative displacement between the bolt head and the connecting plate, and calculates the loosening angle with an accuracy of 1°. The electromagnetic ultrasonic transducer uses a non-contact method to excite and receive ultrasonic guided waves to detect cracks and axial stress changes in the bolt shank. The acoustic vibration analysis unit excites the bolts by using an electromagnetic tapping head, and the MEMS microphone collects the acoustic vibration signals. After wavelet packet energy spectrum decomposition, the signals are input into the support vector machine classification model to identify the loosening level.

[0009] Furthermore, the multimodal fusion analysis model operated by the edge intelligent processing unit employs a feature-level fusion strategy, including: Convolutional neural network features are used to extract bolt appearance features from visual images; Temporal feature extraction is performed on the laser displacement data to obtain the loosening angle features; Time-frequency domain transformation of ultrasonic guided wave signals is performed to obtain crack and stress characteristics; Wavelet packet energy spectrum decomposition is performed on the acoustic vibration signal to obtain the loosening level characteristics; The above multi-source features are input into the fusion classifier, which outputs a bolt health status score.

[0010] Furthermore, the structure-adaptive navigation system achieves tight coupling and fusion of the IMU and lidar through extended Kalman filtering, wherein: The IMU provides high-frequency relative motion information at 1000Hz to fill the motion estimation gaps between lidar update cycles; LiDAR point cloud data is used for 3D map construction and pose observation and updating; Encoder odometer information is used for speed constraints; In areas where the structural features of the steel truss bridge are sparse, short-term positioning continuity is maintained through IMU dead reckoning.

[0011] Furthermore, the telescopic robotic arm is a five-degree-of-freedom flexible robotic arm with an arm span range of 0–700 mm, used to position the multimodal detection module to the bolt detection station.

[0012] Furthermore, the remote communication module is a 5G communication module or a Beidou short message communication module, used to realize real-time transmission of detection data in areas without terrestrial network coverage in western mountainous areas, and automatically switches to Beidou short message communication when the 5G signal is unavailable.

[0013] Furthermore, the intelligent inspection robot for the steel truss bridge also includes: The energy management system includes a high-energy-density lithium battery pack and a power management module. The high-energy-density lithium battery pack is designed to withstand low temperatures and can discharge normally in an environment of -20°C. The safety protection module includes an emergency stop button, a tilt sensor, and a power failure brake. When the tilt sensor detects an abnormal robot posture or the force sensor detects insufficient suction force, the robot will automatically stop running and activate the brake.

[0014] A method for intelligent inspection of steel truss bridges, using the aforementioned intelligent inspection robot for steel truss bridges, includes the following steps: Step S1: Obtain the BIM model of the steel truss bridge through the structural adaptive navigation system and plan the global detection path; Step S2: The walking mechanism moves autonomously along the detection path, and real-time positioning is achieved through the tight coupling and fusion of lidar and IMU; Step S3: Upon reaching the detection node, the telescopic robotic arm unfolds, and the multimodal detection module simultaneously acquires visual, laser, ultrasonic, and acoustic vibration signals of the component under inspection. Step S4: The edge intelligent processing unit runs a multimodal fusion analysis model to identify the status of bolt components in real time and generate detection results; Step S5: The test results are uploaded to the cloud management platform via the remote communication module and linked to the structural health record.

[0015] Furthermore, in the intelligent detection method for steel truss bridges, step S2, which describes achieving real-time positioning through tight coupling and fusion of lidar and IMU, specifically includes: (1) Motion distortion correction of lidar point cloud is performed using high-frequency data from the IMU; (2) The IMU state prediction and lidar observation update are fused by extended Kalman filtering; In areas where the structural features of steel truss bridges are sparse or visually obstructed, IMU dead reckoning is used to maintain positioning continuity and ensure stable navigation of the robot in long-span truss structures.

[0016] The present invention has the following advantages: (1) Strong structural adaptability: The permanent magnet adsorption track mechanism with ball joint connection enables the robot to walk stably on the surface of complex structures such as steel truss members and node plates, adapting to angle changes of ±30°, solving the problem that existing wall-climbing robots are difficult to adapt to spatial truss structures, and is especially suitable for bridges with large drops and variable cross sections in western mountainous areas. (2) High detection accuracy: The four-modal sensor fusion detection (vision + laser + ultrasound + acoustic vibration) realizes for the first time the synchronous quantitative evaluation of bolt loosening angle (accuracy 1°), corrosion area (accuracy ±2%), crack depth (accuracy 0.1mm) and preload changes, which is far superior to human experience judgment. (3) High level of intelligence: The edge intelligent unit realizes real-time analysis and abnormal warning of detection data, without relying on remote servers, and is suitable for on-site operations in environments without network coverage in western mountainous areas. (4) Strong data traceability: The detection data is automatically associated with the BIM model to form a structural health record, supporting structural status trend analysis and maintenance decision-making. (5) High work efficiency: The number of bolts detected per day can reach more than 3,000, which is more than 10 times more efficient than manual detection. (6) Safe and reliable: Unmanned operation completely eliminates the safety risks of high-altitude operation; multiple safety redundancies (force sensor + IMU + tilt sensor + electromagnetic brake) realize active protection; the permanent magnet adsorption system provides a total adsorption force of ≥12000N with a working air gap of 2mm, with a high safety factor and can resist strong winds of level 6. (7) Strong environmental adaptability: It is specially designed for the harsh conditions of large temperature difference, strong wind and no network coverage in the western mountainous area, including temperature compensation algorithm, low temperature battery, Beidou short message communication, wind-resistant adsorption optimization, etc., filling the existing technology gap. Attached Figure Description

[0017] Figure 1 A schematic diagram of the overall structure of the intelligent detection robot of this invention; Figure 2 Structural diagram of a flexible articulated permanent magnet adsorption track mechanism; Figure 3 Multimodal detection module structure diagram; Figure 4 Edge intelligent processing unit architecture diagram; Figure 5 Flowchart of a structural adaptive navigation system; Figure 6 A schematic diagram of the robot's operation process according to the present invention. Detailed Implementation

[0018] The following will be combined with the appendix Figures 1-6This invention will be described in detail, and the technical solutions in the embodiments of this invention will be clearly and completely described. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0019] The following is an implementation description based on the testing application of a large-span steel truss bridge in a mountainous area in western China (main span 1200m, bridge deck height difference from valley floor 310m, annual wind speed 4-6, maximum diurnal temperature difference 20℃); Example 1: The intelligent inspection robot for steel truss bridges of the present invention includes: a walking mechanism, a multimodal detection module, an edge intelligent processing unit, a structural adaptive navigation system, and a remote communication module.

[0020] See Figure 1 In this embodiment, the intelligent inspection robot has a main frame 100 with dimensions of 600mm × 400mm × 150mm and an overall weight of 18kg. The robot's walking mechanism consists of four independently driven permanent magnet adsorption track units 201. Each track unit is connected to the main frame 100 via a ball joint connector 202. The ball joint provides a ±30° swing range, enabling adaptive contact and stable movement on various structural surfaces such as steel truss bridge members and node plates. This structure is specifically designed for the changing angles of bridge truss members in western mountainous areas, ensuring the robot maintains full track contact even on non-straight structural surfaces. Figure 1 and Figure 2 As shown, the track body 201a of the track unit has a permanent magnet array 203 (using N52 neodymium iron boron permanent magnets) embedded on its surface. The size of a single permanent magnet is 20mm×20mm×10mm, and 20 magnets are arranged in each track unit in 2 columns × 10 rows. Static magnetic field simulation verification shows that with a working air gap of 2mm, the adsorption force of a single track unit reaches 3040N, and the total adsorption force of four track units is 12160N, which is much greater than the robot's own weight and can resist the lateral force generated by a level 6 strong wind (wind speed 13.8m / s).

[0021] The ball joint connector 202 has a ball head diameter of 30mm, a ball-and-socket clearance of 0.02~0.05mm, and is made of 40Cr (ball head) and tin bronze (ball socket). Finite element analysis shows that under a combined load of 1500N, the maximum equivalent stress is 235MPa, the safety factor is greater than 3, and the structural strength meets the requirements.

[0022] The drive motor 201b is connected to the track body 201a in a transmission manner; Force sensor 204 is integrated at the connection between the track unit and the main frame to monitor the adsorption force in real time. When the adsorption force is lower than the safety threshold, the safety protection module is triggered.

[0023] like Figure 1 and Figure 3 As shown, the multimodal detection module 300 is mounted at the end of a five-DOF telescopic robotic arm 301, with an arm span range of 0~400mm. The corresponding detection module body 302 has dimensions of 120mm × 80mm × 60mm; Figure 3 As shown, the following sensors are integrated: High-definition industrial camera 302a: 12 megapixels, 60fps, used to acquire bolt images. It identifies bolt location, type, and rust areas using an improved YOLOv8 model, with a detection accuracy of 98.5%. Laser displacement sensor 302b: accuracy 0.001mm, used to measure the relative displacement between the bolt head and the connecting plate, and to calculate the loosening angle; Electromagnetic ultrasonic transducer 302c: center frequency 5MHz, non-contact detection of cracks and axial stress changes in bolt shanks; Electromagnetic striking head 302d and MEMS microphone 302e: excite the bolts and collect acoustic vibration signals. After wavelet packet energy spectrum decomposition, the signals are input into the support vector machine classification model to identify the loosening level.

[0024] This module has a built-in temperature sensor and compensation algorithm that automatically corrects sensor drift within the range of -10℃ to 40℃ to ensure detection accuracy.

[0025] Therefore, the multimodal detection module 300 integrates a high-definition industrial camera, a laser displacement sensor, an electromagnetic ultrasonic transducer, and an acoustic vibration analysis unit consisting of an electromagnetic striking head and a MEMS microphone, used to simultaneously acquire visual images, displacement signals, ultrasonic guided wave signals, and acoustic vibration signals of the bolt. This module incorporates a temperature compensation design for the large temperature difference environment in western mountainous areas, ensuring stable detection accuracy within the range of -10℃ to 40℃.

[0026] The system includes: a high-definition industrial camera for capturing images of the bolt and surrounding area, and an improved YOLOv8 model for identifying bolt location, type, and rusted areas; a laser displacement sensor for measuring the relative displacement between the bolt head and the connecting plate, calculating the loosening angle with an accuracy of 0.01°; an electromagnetic ultrasonic transducer for non-contact excitation and reception of ultrasonic guided waves to detect cracks and axial stress changes in the bolt shank; and an acoustic vibration analysis unit for exciting the bolt with an electromagnetic striking head, collecting acoustic vibration signals with a MEMS microphone, and inputting the signals into a support vector machine classification model after wavelet packet energy spectrum decomposition to identify the loosening level.

[0027] like Figure 4As shown, the edge intelligent processing unit 400 adopts the NVIDIA Jetson AGX Orin platform (32GB memory, 275 TOPS computing power) and runs the ROS 2 operating system. After the multi-sensor data is aligned by the hardware time synchronization module, it is input into the multimodal fusion model and outputs a bolt health status score (0~100 points). Among them, visual features (appearance, corrosion), laser features (loosening angle), ultrasonic features (cracks, stress), and acoustic vibration features (loosening level) are fused at the feature level and then output by the classifier to output the final score.

[0028] In implementation, the edge intelligent processing unit 400 is electrically connected to the multimodal detection module 300, employing an ARM+NPU heterogeneous computing architecture to run a multimodal fusion analysis model. This model performs time synchronization and feature-level fusion of data from four types of sensors: vision, laser, ultrasound, and acoustic vibration, outputting a bolt health status score (0-100 points) in real time. The unit incorporates a temperature compensation algorithm to automatically correct sensor drift under large temperature differences.

[0029] The multimodal fusion analysis model employs a feature-level fusion strategy, including: Convolutional neural network features are used to extract bolt appearance features from visual images; Temporal feature extraction is performed on the laser displacement data to obtain the loosening angle features; Time-frequency domain transformation of ultrasonic guided wave signals is performed to obtain crack and stress characteristics; Wavelet packet energy spectrum decomposition is performed on the acoustic vibration signal to obtain the loosening level characteristics; The above multi-source features are input into the fusion classifier, which outputs a bolt health status score.

[0030] like Figure 5 As shown, the structure-adaptive navigation system integrates a lidar 501 (360° scanning, 30m range), an inertial measurement unit 502 (gyroscope zero-bias stability 0.5° / h, output frequency 1000Hz), an encoder 503, and a steel truss BIM model 504. Extended Kalman filtering achieves tight coupling and fusion of the IMU (Inertial Measurement Unit) and lidar. In areas with sparse structural features of long-span bridges in western mountainous regions (such as a 50m long, featureless truss section between two nodes), IMU dead reckoning maintains short-term positioning continuity, solving the problem of positioning loss in traditional SLAM in such environments. The path planning module 507 uses an improved A* algorithm, combined with structural constraints, to generate a globally optimal detection path. Upon reaching a bolt node, the node recognition module 508 triggers a detection task.

[0031] During implementation, the structural adaptive navigation system is connected to the walking mechanism and the edge intelligent processing unit, and integrates lidar point cloud, inertial measurement unit data, encoder mileage information and steel truss bridge BIM model. Multi-source information fusion is achieved through extended Kalman filtering, which is used to maintain positioning continuity in areas where the structural features of the steel truss bridge are sparse or visually obstructed, and to achieve autonomous positioning, path planning and bolt node identification.

[0032] Specifically, it includes: a lidar (360° scanning, 30m range) for acquiring 3D point cloud data of the steel truss structure; an inertial measurement unit (gyroscope zero-bias stability 0.5° / h, output frequency 1000Hz) for acquiring the robot's posture and acceleration information, providing high-frequency relative motion information to fill the motion estimation gap between lidar update cycles; an encoder, integrated into the drive motor of the walking mechanism, for acquiring the robot's displacement information; and a path planning module, based on an improved A* algorithm and combined with a prior model of the bridge structure, to generate a globally optimal detection path and automatically trigger detection tasks at bolt nodes.

[0033] The structure-adaptive navigation system achieves tight coupling and fusion of IMU and LiDAR through extended Kalman filtering. Specifically, the IMU provides high-frequency relative motion information at 1000Hz to fill the motion estimation gaps between LiDAR update cycles; LiDAR point cloud data is used for 3D map construction and pose observation updates; encoder odometer information is used for velocity constraints; and in sparse areas of the steel truss bridge structure (such as areas without nodes in the middle of a large-span truss), short-term positioning continuity is maintained through IMU dead reckoning, solving the problem of easy positioning loss in traditional SLAM on large-span bridges in mountainous areas.

[0034] The present invention also includes a remote communication module: This system is used to upload test results to a cloud management platform in real time. It employs either a 5G communication module or a BeiDou short message communication module to enable real-time data transmission in mountainous areas of western China where there is no terrestrial network coverage. When 5G signals are unavailable, it automatically switches to BeiDou short message communication to ensure that test data is not lost.

[0035] This invention also includes an energy management system: It includes a high-energy-density lithium battery pack and a power management module to power various functional modules (providing a stable and reliable power supply to all power-consuming modules on the robot, consisting of a high-energy-density lithium battery pack and an intelligent power management module). The battery pack is designed to withstand low temperatures and can discharge normally in environments as low as -20°C.

[0036] This invention also includes a security protection module: It includes an emergency stop button, a tilt sensor, and a power-off brake, used to automatically stop operation and activate braking when the robot's posture is abnormal or the suction force is insufficient. When the tilt sensor detects an abnormal robot posture (such as a roll angle > 45°) or the force sensor detects that the suction force is below a safe threshold, it automatically stops walking and activates the electromagnetic brake to prevent the robot from falling. It primarily functions on the walking mechanism, but also controls the robotic arm and detection module to prevent falls or collisions.

[0037] The following is a detailed description in conjunction with the accompanying drawings; See Figure 1 As shown ( Figure 1 (This is a schematic diagram of the overall structure): The main frame is a rectangular load-bearing structure, with the tracked units connected at the four corners via ball joints. A retractable robotic arm is mounted on the lower front of the robot, with a multimodal detection module at its end. Internally, it integrates an edge intelligence unit, a navigation system, a communication module, and an energy system. The robot adheres to the steel truss beam node, with the detection module aligned with the bolts.

[0038] Figure 1 The overall structure of the steel truss bolt inspection robot of the present invention is shown. The robot includes a main frame 100, around which four sets of permanent magnet adsorption track units 201 are connected. Each track unit is connected to the main frame via ball joint connectors 202, achieving adaptive contact with the complex surface of the steel truss structure. A retractable robotic arm 301 is installed at the front end of the robot, with a multimodal detection module 300 at its end. The main frame integrates an edge intelligent processing unit 400, a structural adaptive navigation system, a remote communication module 600, and an energy management system 700. The robot adheres to the surface of the steel truss node 800 to perform inspection operations.

[0039] See Figure 2 As shown ( Figure 2 (Structural diagram of flexible articulated permanent magnet adsorption track mechanism): Ball joint connector 202 enables multi-degree-of-freedom rotation. A permanent magnet array is embedded on the surface of the track body, and the inner side is driven by a drive motor through a reducer. A tensioning mechanism maintains track tension. A force sensor is integrated at the connection point for monitoring the adsorption force.

[0040] Figure 2 The internal structure of the track unit in the walking mechanism is shown in detail. The track unit includes a track body 201a, whose surface is embedded with a permanent magnet array 203. The track body is driven by a drive motor 201b through a reducer 201c, and a tensioning mechanism 201d is provided to maintain track tension. The track unit is connected to the main frame connecting seat 205 through a ball joint connector 202, realizing multi-degree-of-freedom rotation. A force sensor 204 is integrated at the connection point to monitor the adsorption force in real time.

[0041] See Figure 3 As shown ( Figure 3(Enlarged view of the multimodal detection module): The sensors are arranged in a compact layout, with the camera in the center and laser, ultrasonic, impact head, and microphone arranged around it. An LED array provides supplementary lighting. The module is connected to the robotic arm via a flange.

[0042] Figure 3 The sensor integration method of the multimodal detection module is shown. The detection module body 302 integrates a high-definition industrial camera 302a, a laser displacement sensor 302b, an electromagnetic ultrasonic transducer 302c, an electromagnetic tapping head 302d, and a MEMS microphone 302e, and is equipped with an illumination LED array 302f. The module is connected to a robotic arm via an end flange 303 to achieve multi-dimensional state detection of bolts.

[0043] See Figure 4 As shown ( Figure 4 Edge Intelligent Processing Unit Architecture Diagram (Block Diagram): This diagram uses a modular block diagram format to illustrate the internal hardware components and data flow.

[0044] Figure 4 The hardware architecture and data processing flow of the edge intelligent processing unit 400 are illustrated. The heterogeneous computing chip 401 receives data from multiple sensors via the data acquisition interface 404. After time synchronization by the multi-sensor synchronization module 405, the multi-modal fusion processing module 406 runs a fusion analysis algorithm, and the detection results are uploaded via the communication interface 407. The memory module 402 and storage module 403 provide computing resources.

[0045] See Figure 5 As shown ( Figure 5 (Flowchart of the structure-adaptive navigation system): The flowchart shows the complete navigation logic from sensor input to control command output, combining BIM model and real-time perception.

[0046] Figure 5 The data flow and processing logic of the robot navigation system are illustrated. Real-time data is collected by the LiDAR 501, inertial measurement unit 502, and encoder 503. After being fused with the bridge BIM model 504, the data is positioned by the 3D map construction module 505 and the positioning fusion module 506. The path planning module 507 generates the detection path, the node recognition module 508 identifies the bolt node positions, and the motion control module 509 outputs control commands for the walking mechanism.

[0047] See Figure 6 As shown ( Figure 6 (Diagram of robot operation process): This shows the robot moving along the truss trajectory and deploying its detection posture at the nodes. Detection data is uploaded to the cloud platform in real time.

[0048] Figure 6The complete operation process of the robot on the steel truss bridge is shown. The robot 900 walks autonomously along the motion trajectory 901. After reaching the steel truss node 800, the robotic arm extends to the detection posture 902 to detect the high-strength bolts 801. The detection results are transmitted to the cloud platform 904 through data upload 903, realizing automated management and analysis of the detection data.

[0049] The workflow of the example is as follows: Taking the bolt inspection of a cross-river steel truss bridge as an example, after the robot is hoisted into place from the end of the truss, it performs the following operations; After the robot is hoisted into place from the end of the truss, it will operate according to the following steps: (1) System initialization, loading the steel truss BIM model; (2) The structure-adaptive navigation system plans the global detection path; (3) The walking mechanism moves autonomously along the path and achieves real-time positioning through tight coupling and fusion of lidar and IMU; (4) After reaching the bolt node, the robotic arm unfolds and the multimodal detection module simultaneously collects visual, laser, ultrasonic, and acoustic vibration signals; (5) The edge intelligent processing unit analyzes and outputs the bolt health status in real time; (6) The test results are uploaded to the cloud management platform via 5G or Beidou short message module and automatically linked to the structural health record; (7) The robot continues to move to the next node and repeats steps 4-6.

[0050] On-site verification results: On a large-span steel truss bridge (main span 1200m, bridge deck elevation difference 310m from valley floor) in a mountainous area in western China, the robot operated continuously for 8 hours a day, inspecting 3200 bolts, achieving 100% inspection coverage. Comparison of inspection results with manual verification showed: loosening detection accuracy 96.5%, corrosion detection accuracy 94.2%, crack detection accuracy 92.8%, and overall inspection accuracy exceeding 95%. Under conditions of strong winds (level 6) and large temperature differences (25℃ diurnal temperature range), the robot operated stably without adsorption failure or abnormal movement. In areas without terrestrial network coverage, BeiDou short message communication successfully transmitted all inspection data, achieving 100% data integrity. The robot operated stably without human intervention throughout the entire process.

[0051] Example 2: An intelligent inspection method for steel truss bridges, implemented based on the intelligent inspection robot for steel truss bridges described in Example 1, includes the following steps: Step S1: Obtain the BIM model of the steel truss bridge through the structural adaptive navigation system and plan the global detection path; Step S2: The walking mechanism moves autonomously along the detection path, and real-time positioning is achieved through the tight coupling and fusion of lidar and IMU; Step S3: Upon reaching the detection node, the telescopic robotic arm unfolds, and the multimodal detection module simultaneously acquires visual, laser, ultrasonic, and acoustic vibration signals of the component under inspection. Step S4: The edge intelligent processing unit runs a multimodal fusion analysis model to identify the status of bolt components in real time and generate detection results; Step S5: The test results are uploaded to the cloud management platform via the remote communication module and linked to the structural health record.

[0052] In this embodiment, the real-time positioning achieved through tight coupling and fusion of lidar and IMU in step S2 specifically includes: (1) Motion distortion correction of lidar point cloud is performed using high-frequency data from the IMU; (2) The IMU state prediction and lidar observation update are fused by extended Kalman filtering; In areas where the structural features of steel truss bridges are sparse or visually obstructed, IMU dead reckoning is used to maintain positioning continuity and ensure stable navigation of the robot in long-span truss structures.

[0053] The above description of the disclosed embodiments enables those skilled in the art to make or use the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An intelligent inspection robot for steel truss bridges, characterized in that, include: The walking mechanism consists of four independently driven permanent magnet adsorption track units. Each permanent magnet adsorption track unit is connected to the robot's main frame through a ball joint connector. The ball joint connector provides a swing range of ±30°, which is used to achieve adaptive fitting and stable walking on various structural surfaces such as the members and node plates of the steel truss bridge. The multimodal detection module is installed at the end of the five-degree-of-freedom telescopic robotic arm at the front of the robot. This module integrates a high-definition industrial camera, a laser displacement sensor, an electromagnetic ultrasonic transducer, and an acoustic vibration analysis unit consisting of an electromagnetic striking head and a MEMS microphone. It is used to simultaneously acquire visual images, displacement signals, ultrasonic guided wave signals, and acoustic vibration signals of the bolt. The edge intelligent processing unit is electrically connected to the multimodal detection module. It adopts an ARM+NPU heterogeneous computing architecture to run a multimodal fusion analysis model. It performs time synchronization and feature-level fusion of data from four types of sensors: vision, laser, ultrasound, and acoustic vibration, and outputs bolt health status scores in real time. The structure-adaptive navigation system is connected to the walking mechanism and the edge intelligent processing unit. It integrates lidar point cloud, inertial measurement unit data, encoder mileage information and steel truss bridge BIM model. It achieves multi-source information fusion through extended Kalman filtering. It is used to maintain positioning continuity in areas where the structural features of the steel truss bridge are sparse or visually obstructed, and realizes autonomous positioning, path planning and bolt node identification. The remote communication module is used to upload the test results to the cloud management platform in real time.

2. The intelligent inspection robot for steel truss bridges according to claim 1, characterized in that: The permanent magnet adsorption track unit includes: The track body has an array of permanent magnets embedded on its surface. Each permanent magnet is 20mm×20mm×10mm in size and is arranged in 2 columns×10 rows. The drive motor is connected to the track body for transmission. A force sensor, integrated at the connection between the permanent magnet adsorption track unit and the main frame, is used to monitor the adsorption force in real time. When the adsorption force is lower than the safety threshold, the safety protection module is triggered. The ball joint connector is a ball joint structure that allows each permanent magnet adsorption track unit to rotate in three degrees of freedom relative to the main frame.

3. The intelligent inspection robot for steel truss bridges according to claim 1, characterized in that: In the multimodal detection module: High-definition industrial cameras identify bolt locations, types, and rusted areas by improving the YOLOv8 model; A laser displacement sensor measures the relative displacement between the bolt head and the connecting plate, and calculates the loosening angle with an accuracy of 1°. The electromagnetic ultrasonic transducer uses a non-contact method to excite and receive ultrasonic guided waves to detect cracks and axial stress changes in the bolt shank. The acoustic vibration analysis unit excites the bolts by using an electromagnetic tapping head, and the MEMS microphone collects the acoustic vibration signals. After wavelet packet energy spectrum decomposition, the signals are input into the support vector machine classification model to identify the loosening level.

4. The intelligent inspection robot for steel truss bridges according to claim 1, characterized in that: The multimodal fusion analysis model run by the edge intelligent processing unit adopts a feature-level fusion strategy, including: Convolutional neural network features are used to extract bolt appearance features from visual images; Temporal feature extraction is performed on the laser displacement data to obtain the loosening angle features; Time-frequency domain transformation of ultrasonic guided wave signals is performed to obtain crack and stress characteristics; Wavelet packet energy spectrum decomposition is performed on the acoustic vibration signal to obtain the loosening level characteristics; The above multi-source features are input into the fusion classifier, which outputs a bolt health status score.

5. The intelligent inspection robot for steel truss bridges according to claim 1, characterized in that: The structure-adaptive navigation system achieves tight coupling and fusion of the IMU and lidar through extended Kalman filtering, wherein: The IMU provides high-frequency relative motion information at 1000Hz to fill the motion estimation gaps between lidar update cycles; LiDAR point cloud data is used for 3D map construction and pose observation and updating; Encoder odometer information is used for speed constraints; In areas where the structural features of the steel truss bridge are sparse, short-term positioning continuity is maintained through IMU dead reckoning.

6. The intelligent inspection robot for steel truss bridges according to claim 1, characterized in that: The telescopic robotic arm is a five-degree-of-freedom flexible robotic arm with an arm span of 0–700 mm, used to position the multimodal detection module at the bolt detection station.

7. The intelligent inspection robot for steel truss bridges according to claim 1, characterized in that, The remote communication module is either a 5G communication module or a BeiDou short message communication module, used to achieve real-time transmission of detection data in mountainous areas of western China without terrestrial network coverage. When the 5G signal is unavailable, it automatically switches to BeiDou short message communication.

8. The intelligent inspection robot for steel truss bridges according to claim 1, characterized in that, Also includes: The energy management system includes a high-energy-density lithium battery pack and a power management module. The high-energy-density lithium battery pack is designed to withstand low temperatures and can discharge normally in an environment of -20°C. The safety protection module includes an emergency stop button, a tilt sensor, and a power failure brake. When the tilt sensor detects an abnormal robot posture or the force sensor detects insufficient suction force, the robot will automatically stop running and activate the brake.

9. An intelligent inspection method for steel truss bridges, applied to the intelligent inspection robot for steel truss bridges as described in any one of claims 1 to 8, characterized in that, Includes the following steps: Step S1: Obtain the BIM model of the steel truss bridge through the structural adaptive navigation system and plan the global detection path; Step S2: The walking mechanism moves autonomously along the detection path, and real-time positioning is achieved through the tight coupling and fusion of lidar and IMU; Step S3: Upon reaching the detection node, the telescopic robotic arm unfolds, and the multimodal detection module simultaneously acquires visual, laser, ultrasonic, and acoustic vibration signals of the component under inspection. Step S4: The edge intelligent processing unit runs a multimodal fusion analysis model to identify the status of bolt components in real time and generate detection results; Step S5: The test results are uploaded to the cloud management platform via the remote communication module and linked to the structural health record.

10. The intelligent detection method for steel truss bridges according to claim 9, characterized in that, Step S2, which describes achieving real-time positioning through tight coupling and fusion of lidar and IMU, specifically includes: (1) Motion distortion correction of lidar point cloud is performed using high-frequency data from the IMU; (2) The IMU state prediction and lidar observation update are fused by extended Kalman filtering; In areas where the structural features of steel truss bridges are sparse or visually obstructed, IMU dead reckoning is used to maintain positioning continuity and ensure stable navigation of the robot in long-span truss structures.

Images