Interior inspection device for steel box girder

Abstract

The invention discloses a steel box girder interior inspection device, belongs to the technical field of bridge detection equipment, and solves the problems that the existing steel box girder interior detection is low in efficiency, insufficient in precision and difficult to adapt to a complex structure. According to the technical scheme, flexible movement of a movable chassis in a box body is achieved through a crawler-type walking mechanism and an independently-controlled electromagnetic brake; the navigation module is fused with a binocular vision sensor and a laser radar to construct a three-dimensional grid map, and a path is planned by combining environmental data of the distance between diaphragm plates and the height of stiffening ribs; the detection module cooperatively detects surface defects and internal damage through a multispectral imaging unit and an ultrasonic flaw detection unit, wherein the contact pressure of a probe array is dynamically adjusted by a spring mechanism; the positioning module realizes high-precision real-time positioning through the ultra-wideband base station and the inertial measurement unit; the control module optimizes motion control based on map data and attitude feedback. The device is mainly used for efficient, accurate and automatic inspection of the internal structure of the steel box girder, and bridge safety is guaranteed.

Description

Technical Field

[0001] The present invention relates to the field of bridge detection equipment, and more particularly to a steel box girder internal inspection device. Background Art

[0002] As the core load-bearing component of long-span bridges, the internal structure of steel box girders is subject to long-term environmental corrosion, load fatigue, and other factors, making them prone to hidden dangers such as weld cracking, rust, and stiffener deformation. Traditional manual inspections rely on technicians entering the box for visual inspections or using handheld devices for localized detection. This method suffers from low efficiency, limited coverage, and high risks associated with working at height and in confined spaces. In the complex environment of densely packed diaphragms and interlaced stiffening ribs within the box, comprehensive inspections are difficult to achieve manually, and inspection accuracy is significantly affected by subjective experience.

[0003] Existing automated inspection equipment often uses wheeled or tracked mobile platforms. However, the narrow interior space of steel box girders and the large variations in floor inclination make traditional wheeled chassis prone to slipping or jamming. Tracked solutions require pre-installed guide rails, which are inflexible and costly. Furthermore, the navigation systems of existing equipment often rely on a single sensor (such as lidar or vision). This can lead to map construction errors in conditions such as low light levels and complex reflective surfaces within the box, resulting in path planning failures. For example, if minor design deviations in the spacing between diaphragms are not corrected in real time, they can cause equipment collisions or redundant detours.

[0004] In terms of detection technology, conventional methods such as single-spectrum imaging or ultrasonic testing alone struggle to simultaneously identify both surface defects (such as rust spots) and internal damage (such as cracks). Insufficient wavelength switching frequency in multispectral imaging units can easily lead to asynchronous data collection, while unstable contact pressure of ultrasonic probes can significantly reduce detection accuracy. Furthermore, existing equipment's positioning systems typically rely on inertial navigation, which can easily accumulate errors due to gyroscope drift over long periods of operation, requiring frequent manual corrections and affecting detection continuity.

[0005] The root of these problems lies in the fact that existing technologies lack mobile mechanisms adapted to the complex structure of steel box girders, environmental perception systems that integrate multiple sensors, robust positioning methods, and collaborative detection mechanisms. Key technical bottlenecks hindering automated internal inspections of steel box girders include ensuring stable movement in narrow, unstructured environments, integrating multidimensional data to build precise maps and dynamically correct paths, and balancing the accuracy and efficiency of multimodal detection. Summary of the Invention

[0006] An object of the present invention is to solve at least the above problems and to provide at least the advantages which will be described hereinafter.

[0007] Another object of the present invention is to provide an internal inspection device for steel box girders, which significantly improves the steering flexibility and obstacle-crossing ability in narrow spaces through the independent control of the crawler mobile chassis and dual electromagnetic brakes; the multimodal environmental perception system that integrates binocular vision and lidar can accurately construct a three-dimensional map including the spacing between diaphragms and the height of stiffening ribs, providing high-precision spatial data for path planning; the collaborative detection mechanism of multispectral imaging and ultrasonic flaw detection takes into account the simultaneous identification of surface defects and internal damage, and the spring mechanism dynamically adjusts the probe contact pressure to ensure the stability of the detection results; the fusion correction of ultra-wideband positioning and inertial navigation effectively suppresses the cumulative error of long-term operation and ensures the positioning accuracy of equipment in complex environments.

[0008] In order to achieve these objects and other advantages according to the present invention, a steel box girder internal inspection device is provided, comprising a mobile chassis, a navigation module, a detection module, a positioning module and a control module; The mobile chassis includes a crawler-type traveling mechanism, a drive motor and a steering mechanism, and the mobile chassis moves inside the box body of the steel box girder. The steering mechanism includes two independently controlled electromagnetic brakes, which act on the crawler wheel shafts on the left and right sides of the crawler-type traveling mechanism respectively; The navigation module includes a binocular vision sensor, a laser radar, and an environmental data processing unit. The binocular vision sensor is installed at a height of 400 mm to 600 mm and has a field of view of 120 to 150 degrees. The scanning plane of the laser radar is at an elevation angle of 5 to 15 degrees to the top plate inside the steel box girder. The environmental data processing unit establishes a three-dimensional grid map including the spacing between diaphragms and the height of stiffeners. The detection module includes a multispectral imaging unit and an ultrasonic flaw detection unit. The multispectral imaging unit includes a filter array with a wavelength range of 380nm to 1700nm. The probe array spacing of the ultrasonic flaw detection unit is 50mm to 100mm. The contact pressure between the probe array and the inner wall of the steel box girder is maintained at 5N to 10N by a spring mechanism. The positioning module includes an inertial measurement unit and an ultra-wideband positioning base station. The sampling frequency of the inertial measurement unit is 100Hz to 200Hz, the tag arrangement spacing of the ultra-wideband positioning base station is 3m to 5m, and the tag coordinates are mapped to the three-dimensional grid map. The control module includes a path planning unit and a motion control unit. The path planning unit calculates the shortest path avoiding the diaphragm according to the three-dimensional grid map, and the motion control unit adjusts the braking torque of the electromagnetic brake according to the posture data of the inertial measurement unit.

[0009] Preferably, the output end of the environmental data processing unit is connected to the input end of the path planning unit, the output end of the motion control unit is connected to the input end of the drive motor and the electromagnetic brake, and the output end of the ultra-wideband positioning base station is connected to the correction interface of the inertial measurement unit.

[0010] Preferably, the three-dimensional grid map output by the environmental data processing unit to the path planning unit includes the deviation between the measured value and the design value of the diaphragm spacing and the stiffener height data, and the path planning unit generates a turning angle threshold of the detour path according to the diaphragm spacing deviation; The control signal sent by the motion control unit to the drive motor and electromagnetic brake includes a braking torque gradient parameter and a motor speed compensation coefficient calculated based on a steering angle threshold, wherein the braking torque gradient parameter is linearly related to the angular velocity of the track wheel shaft, and the motor speed compensation coefficient is dynamically adjusted according to the inclination angle of the steel box girder bottom plate; The correction data output by the ultra-wideband positioning base station to the inertial measurement unit includes the mapping error value between the tag coordinates and the three-dimensional grid map, and the tag coordinates are attached with a timestamp synchronized with the lidar scanning cycle. The inertial measurement unit performs segmented compensation for gyroscope drift based on the mapping error value and timestamp.

[0011] Preferably, the output shaft of the drive motor is mechanically connected to the track wheel shaft through a planetary reduction gear set, and the transmission ratio of the planetary reduction gear set is 10:1 to 15:1; An incremental encoder is installed at the end of the rotor shaft of the drive motor. The pulse signal output end of the incremental encoder is connected to the speed feedback interface of the control module. The control module generates a pulse width modulation signal according to the difference between the encoder pulse signal and the preset speed. The pulse width modulation signal drives the drive motor through the H-bridge circuit.

[0012] Preferably, within the angle range of 30 to 45 degrees between the optical axis of the multispectral imaging unit and the detection plane of the ultrasonic flaw detection unit, the overlap rate between the field of view coverage area of ​​the multispectral imaging unit and the detection area of ​​the ultrasonic flaw detection unit is 30% to 50%; The switching cycle of the filter array is 0.5s to 1.5s, and the time of maintaining a fixed wavelength after each switching accounts for 60% to 80% of a single cycle. The probe trigger interval of the ultrasonic flaw detection unit is an integer multiple of the filter switching cycle; The wavelength range of the multispectral imaging unit includes the visible light band of 380nm-780nm and the short-wave infrared band of 900nm-1700nm, and the detection plane of the ultrasonic flaw detection unit is arranged with an array of calibration reference points corresponding to the overlapping area of ​​the field of view of the multispectral imaging unit; The spacing of the calibration reference point array is 1.2 to 1.5 times the spacing of the ultrasonic probe array, and the difference in reflectivity of each calibration reference point at each wavelength of the multispectral imaging unit does not exceed 15%.

[0013] Preferably, the environmental data processing unit creates a three-dimensional grid map by fusing the point cloud data of the laser radar with the stereo matching data of the binocular vision sensor. The point cloud density of a single scan of the laser radar is 2000 points / m 2 Up to 5000 points / m 2 , the number of feature point matches of the binocular vision sensor is 50 to 80 per frame image; The calculation of the diaphragm spacing is based on the geometric projection relationship between the laser radar scanning plane and the roof elevation angle. The spacing of the reflection point clouds of adjacent diaphragms is extracted as the measured value, and the difference operation is performed with the design value. The difference operation result is stored as an attribute field of the raster map; The height of the stiffening rib is measured by calculating the disparity map of the binocular vision sensor. The depth resolution of the disparity map reaches 0.1mm to 0.3mm, and during the measurement, at least three feature points are ensured to fall on the vertical section of a single stiffening rib. The dynamic update frequency of the three-dimensional grid map is 0.5Hz to 1Hz. During each update, the map origin is drift-corrected using the tag coordinates of the ultra-wideband positioning base station. The drift correction thresholds are set to 3mm to 5mm for lateral deviation and 2mm to 4mm for height deviation. The feature matching algorithm built into the environmental data processing unit aligns the spatial coordinates of the diaphragm edge contour extracted by the lidar with the stiffening rib corner points identified by binocular vision, and the alignment error does not exceed 1.5 times the lidar point cloud spacing.

[0014] Preferably, a height adjustment rail is provided at the bottom of the mounting bracket of the binocular vision sensor, the scale accuracy of the height adjustment rail is 1 mm, and the baseline length of the binocular vision sensor is set to 0.3 to 0.5 times the mounting height; When the field of view angle of the binocular vision sensor is 120 degrees to 150 degrees, its lens distortion correction adopts a quintic polynomial model, and the coefficients of the quintic polynomial model are obtained by fitting 20 to 30 groups of pose data of a checkerboard calibration plate, where the side length of a single grid of the checkerboard calibration plate is 30 mm to 50 mm; When the installation height of the binocular vision sensor is 400mm to 600mm, the angle between its optical axis and the bottom plate of the steel box girder is 70 degrees to 80 degrees, and the vertical distance between the lowest field of view boundary and the ground contact point of the mobile chassis crawler track is 100mm to 150mm; The exposure time of the binocular vision sensor is dynamically adjusted according to the light intensity inside the steel box girder. The light intensity is detected by the photoresistor built into the sensor. The exposure time adjustment range is 1ms to 10ms, and the exposure time change rate of adjacent frames does not exceed 20%.

[0015] Preferably, the height adjustment guide rail adopts a rack-type lifting column; the base of the rack-type lifting column is fixed to the mobile chassis by bolts, and the binocular vision sensor bracket is connected to the slider of the rack-type lifting column by a flange.

[0016] The present invention has at least the following beneficial effects: First, through the independent control of the tracked mobile chassis and dual electromagnetic brakes, the steering flexibility and obstacle-crossing ability in narrow spaces are significantly improved; the multimodal environmental perception system that integrates binocular vision and lidar can accurately construct a three-dimensional map including the spacing between diaphragms and the height of stiffeners, providing high-precision spatial data for path planning; the collaborative detection mechanism of multispectral imaging and ultrasonic flaw detection takes into account the simultaneous identification of surface defects and internal damage, and the spring mechanism dynamically adjusts the probe contact pressure to ensure the stability of the detection results; the fusion correction of ultra-wideband positioning and inertial navigation effectively suppresses the accumulated errors in long-term operation and ensures the positioning accuracy of equipment in complex environments.

[0017] Second, the directional connection of data interfaces between modules enables closed-loop coordination of environmental perception, path planning, motion control, and positioning correction, shortening signal transmission delays. Ultra-wideband positioning data directly corrects the inertial measurement unit, compensating for gyroscope drift in real time, avoiding the interruption of traditional manual calibration and significantly improving detection continuity and system robustness.

[0018] Third, real-time feedback on the deviation in the spacing between the diaphragms enables path planning to dynamically correct the detour strategy, and the introduction of a steering angle threshold avoids redundant paths. The dynamic adjustment mechanism of the braking torque gradient parameter and the speed compensation coefficient enables the equipment to adapt to changes in the bottom plate inclination angle, improving climbing and steering stability. The synchronized timestamps of the tag coordinates and the lidar scanning cycle enable segmented compensation of inertial navigation errors, further reducing positioning drift errors.

[0019] Fourth, the high transmission ratio design of the planetary reduction gear set increases the output torque, ensuring the traction of the track on slippery or inclined surfaces; the real-time speed feedback of the incremental encoder combined with the precise drive of the H-bridge circuit further reduces the motor speed error, effectively solving the problem of path deviation caused by slippage in traditional wheeled equipment.

[0020] Fifth, the overlapping layout of the multispectral and ultrasonic detection areas enables spatial correlation analysis of surface rust and internal cracks; the integer multiple relationship between the filter switching cycle and the probe trigger ensures the temporal synchronization of multimodal data, and the reflectivity consistency design of the calibration reference point array eliminates cross-sensor data alignment errors and improves the accuracy of defect recognition.

[0021] Sixth, the cross-modal data fusion of lidar and binocular vision improves the absolute accuracy of three-dimensional maps; the combination of cross-partition spacing difference calculation and stiffener height parallax measurement realizes the quantitative assessment of box structure deformation; the linkage mechanism of dynamic map update and ultra-wideband drift correction ensures navigation reliability in complex environments.

[0022] Seventh, the precise scale of the height adjustment rail reduces the depth measurement error of binocular vision; the quintic polynomial distortion correction model combined with multiple sets of calibration data reduces image stitching errors.

[0023] Other advantages, objectives and features of the present invention will be reflected in part from the following description and will be understood by those skilled in the art through study and practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 This is a schematic structural diagram of a steel box girder internal inspection device according to one of the technical solutions of the present invention; Figure 2 This is a structural schematic diagram of the spring mechanism described in one of the technical solutions of the present invention.

[0025] Figure numerals: 1-probe array; 2-spring mechanism; 3-steel box girder; 4-movable chassis; 5-drive motor; 6-electromagnetic brake; 7-mounting plate; 8-spring; 9-limiting sleeve; 10-planetary reduction gear set; 11-mounting platform; 12-guide column; 13-mounting frame; 14-adjusting part. DETAILED DESCRIPTION

[0026] The present invention will be described in further detail below in conjunction with the accompanying drawings so that those skilled in the art can implement the invention with reference to the description.

[0027] It should be understood that terms such as “having”, “including” and “comprising” used herein do not preclude the existence or addition of one or more other elements or combinations thereof.

[0028] like Figures 1 and 2 As shown, the present invention provides a steel box girder internal inspection device, including a mobile chassis 4, a navigation module, a detection module, a positioning module and a control module; The mobile chassis 4 includes a crawler-type traveling mechanism, a drive motor 5 and a steering mechanism, and the mobile chassis 4 moves inside the box body of the steel box girder 3. The steering mechanism includes two independently controlled electromagnetic brakes 6, which act on the crawler wheel shafts on the left and right sides of the crawler-type traveling mechanism respectively; The navigation module includes a binocular vision sensor, a laser radar, and an environmental data processing unit. The binocular vision sensor is installed at a height of 400 mm to 600 mm and a field of view of 120 to 150 degrees. The scanning plane of the laser radar is at an elevation angle of 5 to 15 degrees to the top plate of the box body of the steel box girder 3. The environmental data processing unit establishes a three-dimensional grid map including the spacing between the diaphragms and the height of the stiffeners. The detection module includes a multispectral imaging unit and an ultrasonic flaw detection unit. The multispectral imaging unit includes a filter array with a wavelength range of 380nm to 1700nm. The probe array 1 of the ultrasonic flaw detection unit has a spacing of 50mm to 100mm. The probe array 1 is in elastic contact with the wall surface inside the box body of the steel box girder 3 through a spring mechanism 2. The positioning module includes an inertial measurement unit and an ultra-wideband positioning base station. The sampling frequency of the inertial measurement unit is 100Hz to 200Hz, the tag arrangement spacing of the ultra-wideband positioning base station is 3m to 5m, and the tag coordinates are mapped to the three-dimensional grid map. The control module includes a path planning unit and a motion control unit. The path planning unit calculates the shortest path to avoid the diaphragm according to the three-dimensional grid map, and the motion control unit adjusts the braking torque of the electromagnetic brake 6 according to the posture data of the inertial measurement unit. In the above technical solution, the internal inspection device of the steel box girder 3 includes multiple parts, as follows: The mobile chassis 4 utilizes a lightweight aluminum alloy mounting platform 11, with a crawler-type running mechanism mounted on its underside. The crawler tracks are designed with anti-slip grooves to enhance adhesion to the inner wall of the steel box girder. The drive motors 5 can be two brushless DC motors (rated power of 200W), each independently driving the left and right crawler axles. The motors have built-in encoders for speed feedback. The steering mechanism utilizes two independently controlled electromagnetic brakes (e.g., EMB-15, with adjustable braking torque of 0-5 N·m) to achieve differential steering. When turning left, the right electromagnetic brake applies torque to reduce the right crawler speed, and vice versa. The maximum climbing angle of the mobile chassis 4 can be designed to be 30° according to actual circumstances to adapt to the internal slope of the steel box girder and the environment of the weld protrusion; the mounting platform 11 provides mounting points for the installation of corresponding components. For example, the drive motor 5 can be installed on the mounting platform 11, and its driving end is coaxially connected to the track wheel shaft of the crawler walking structure to drive the crawler walking mechanism to move. The electromagnetic brake 6 can be respectively installed at the track wheel shafts on the left and right sides of the crawler walking mechanism; the drive motor 5 drives the mobile chassis 4 to move and the electromagnetic brake 6 brakes the mobile chassis 4. Both are existing very mature technologies and are not limited to the methods listed above, so I will not go into details here. Navigation module part: The binocular vision sensor (not shown in the figure) can use a global shutter CMOS camera (resolution 1280×960), the horizontal spacing between the two cameras is 200mm, the installation height is 500mm, and the field of view angle is 135°. The depth information of the stiffeners and diaphragms in the steel box girder is extracted through the stereo matching algorithm. The laser radar can use a 16-line laser radar (scanning frequency 10Hz), installed at an elevation angle of 10°, and the scanning plane covers the junction area between the top plate and the side wall, which is used to detect the top plate welds and rusted areas. The environmental data processing unit is equipped with an embedded GPU (NVIDIA Jetson TX2), which fuses the visual point cloud and the laser point cloud based on the SLAM algorithm to generate a three-dimensional grid map (grid resolution 20mm) including the diaphragm spacing (marked with ±10mm error) and the stiffener height (marked with ±5mm error). The binocular vision sensor can be installed at a height of 400mm, 500mm, or 600mm, with a field of view of 120, 130, 140, or 150 degrees. The elevation angle between the laser radar scanning plane and the interior roof of the steel box girder 3 can be 5, 10, or 15 degrees. The binocular vision sensor is installed at a suitable position above the mobile chassis 4. The laser radar is mounted on the mobile chassis 4 with its scanning direction facing the interior space of the steel box girder 3. The environmental data processing unit is installed inside the mobile chassis 4.

[0029] Detection Module: The multispectral imaging unit can include a filter array ranging from 380nm to 1700nm. Combined with a high-sensitivity CMOS sensor, it can detect paint peeling (abrupt reflectivity changes in the 450nm band), rust (absorption characteristics in the 1450nm band), and hidden cracks (near-infrared scattering characteristics at 1600nm). The ultrasonic flaw detection unit's probe array can be arranged in a 5×5 matrix (80mm spacing) with a center frequency of 5MHz. Each probe maintains a contact force of 3-5N with the wall surface via a spring mechanism (with a spring constant of 50N / m), ensuring uniform coverage of the coupling agent. The filter array switching cycle of the multispectral imaging unit can be selected from 0.5s, 1s, and 1.5s. The percentage of time a fixed wavelength is maintained after each switch can be selected from 60%, 70%, and 80% of the cycle. The ultrasonic flaw detection unit's probe trigger interval is an integer multiple of the filter switching cycle. Both the multispectral imaging unit (not shown) and the ultrasonic flaw detection unit are mounted on the mounting platform 11 of the mobile chassis 4, with the detection direction facing the inner wall of the steel box girder 3.

[0030] For the positioning module, the inertial measurement unit (IMU) can use a MEMS six-axis sensor (gyroscope range ±2000° / s, accelerometer range ±16g), with a sampling frequency of 150Hz. It uses a Kalman filter to fuse angular velocity and acceleration data to calculate the chassis' real-time position. Ultra-wideband (UWB) positioning base stations can be arranged with UWB tags (Decawave DW1000 chips) every 4 meters inside the steel box girder. Communication latency between the base station and the tag is <1ms, and positioning accuracy is ±30m. The IMU is installed inside the mobile chassis 4, and the tags of the UWB positioning base stations are installed in appropriate locations inside the steel box girder 3 so that their coordinates can be mapped to the 3D grid map.

[0031] Control module part: The path planning unit and the motion control unit can be functional modules integrated on a control circuit board. The path planning unit calculates the shortest path to avoid the diaphragm based on the three-dimensional grid map, and the motion control unit adjusts the braking torque of the electromagnetic brake 6 based on the posture data of the inertial measurement unit. The control module is installed inside the mobile chassis 4 and is connected to other modules through lines. When the inspection device is working, the mobile chassis 4 moves inside the steel box girder 3, and the binocular vision sensor and lidar of the navigation module collect environmental data, which is processed by the environmental data processing unit to establish a three-dimensional grid map. The detection module detects the wall surface of the steel box girder 3, the positioning module determines the position of the device, and the control module performs path planning and motion control based on this information. This device can realize autonomous inspection of the interior of the steel box girder 3, effectively detect the internal condition of the steel box girder 3, and detect problems in a timely manner.

[0032] Operational Process: During operation, the mobile chassis 4, powered by a crawler-type travel mechanism, moves within the steel box girder 3. Drive motors 5 provide power, and electromagnetic brakes 6 in the steering mechanism brake the left and right track axles of the crawler travel mechanism, enabling flexible steering. The navigation module begins operation, with binocular vision sensors and a lidar radar simultaneously collecting environmental data from within the steel box girder 3. This data is transmitted to the environmental data processing unit, which then creates a three-dimensional grid map containing information such as diaphragm spacing and stiffener height. The detection module operates simultaneously. The multispectral imaging unit uses a filter array to image the steel box girder 3 wall at different wavelengths. The ultrasonic flaw detection unit's probe array 1, guided by a spring mechanism 2, adheres to the wall surface with an appropriate contact pressure for flaw detection. The positioning module's inertial measurement unit measures the device's attitude data in real time at a sampling rate of 100 to 200 Hz. The ultra-wideband positioning base station determines the device's position using tags placed within the steel box girder 3 at intervals of 3 to 5 meters, mapping the tag coordinates to the three-dimensional grid map. The control module receives data from each module, the path planning unit calculates the shortest path to avoid the diaphragm based on the three-dimensional grid map, and the motion control unit adjusts the braking torque of the electromagnetic brake 6 according to the posture data of the inertial measurement unit, and controls the mobile chassis 4 to move stably along the planned path, thereby realizing a comprehensive inspection of the interior of the steel box girder 3.

[0033] Technical Effect: This inspection device enables automated and precise inspection of the interior of the steel box girder 3. It effectively captures structural information such as the spacing between diaphragms and stiffener height within the steel box girder 3. It also uses multispectral imaging and ultrasonic testing to inspect the wall surface of the steel box girder 3, promptly identifying potential defects. The positioning module and navigation module work together to enable the device to accurately navigate and position itself within the complex internal environment of the steel box girder 3, improving inspection efficiency and reliability and ensuring the safe operation of the steel box girder 3.

[0034] In another technical solution, Figure 2As shown, the spring mechanism 2 includes multiple groups of adjustment components, and the multiple groups of adjustment components correspond one to one with the multiple probes in the probe array 1. Each adjustment component includes a spring 8, a guide column 12, a limiting sleeve 9, a mounting plate 7, an adjustment member 14, and a mounting bracket 13. The adjustment member 14 is mounted on the mounting platform 11, and the adjustment end of the adjustment member is connected to the mounting bracket 13. The limiting sleeve 9 is mounted on the mounting bracket 113. The guide column 12 is slidingly and coaxially sleeved in the limiting sleeve 9, and the guide column 12 slides along the limiting sleeve 9. Specifically, the limiting sleeve 9 is provided with a sliding groove along its axial direction, and the guide column 12 is provided with a slider adapted to the sliding groove. When the slider slides along the sliding groove, the guide column 12 moves along the The limiting sleeve 9 limits sliding, and a mounting plate 7 is coaxially provided on one end of the guide column 12 extending out of the limiting sleeve 9. A corresponding probe is provided on the mounting plate, and the spring 8 is sleeved on the column body of the guide column 12 located between the mounting plate 7 and the limiting sleeve 9. Specifically, the diameter of the spring 8 is larger than the inner diameter of the limiting sleeve 9 and smaller than the outer diameter of the limiting sleeve 9 so that the spring 8 can be supported on the end face of the limiting sleeve 9. The diameter of the spring 8 is smaller than the diameter of the mounting plate 7 so that the spring 8 can be sleeved on the guide column 12 and supported between the mounting plate 7 and the limiting sleeve 9; the adjusting member 14 is a driving telescopic rod for adjusting the horizontal displacement of the mounting frame 13. The adjusting member 14 has various forms, such as an electric push rod, a hydraulic rod, etc. In this technical solution, when in use, the detection end of the probe array 1 is brought into contact with the wall of the steel box girder 3. By adjusting the extension length of the adjustment member 14, the contact force between the probe array 1 and the wall can be adjusted. When the probe array 1 passes through a concave or convex area, the stretching and retraction properties imparted by the spring 8 enable the probe array 1 to better pass through the concave or convex area. In actual use, the adjustment member 14 can also be connected to the control module to flexibly control the contact force between the probe array 1 and the wall. It should be noted that Figure 1 and Figure 2 All of them are structural schematic diagrams.

[0035] In another technical solution, the output end of the environmental data processing unit is connected to the input end of the path planning unit, the output end of the motion control unit is connected to the input end of the drive motor 5 and the electromagnetic brake 6, and the output end of the ultra-wideband positioning base station is connected to the correction interface of the inertial measurement unit; In the above technical solution, the output end of the environmental data processing unit is connected to the input end of the path planning unit through an RJ45 interface. The data transmission rate can be set to 100Mbps, and the core cross-sectional area of ​​the shielded twisted pair is 0.5mm. 2, the outer layer is covered with PVC insulation material. The environmental data processing unit can be an embedded industrial computer, installed in the central control box of the mobile chassis 4 and fixed by bolts; the path planning unit is located in the adjacent slot of the same box, and the two are connected by a 30cm long cable. The ends of the cable are gold-plated terminals to reduce contact resistance. The output end of the motion control unit is connected to the input end of the drive motor 5 and the electromagnetic brake 6 through an aviation plug. The control signal line can be made of high-temperature resistant silicone wire, which can withstand temperatures of -40°C to 150°C and has a wire core cross-sectional area of ​​1.0mm. 2 The motion control unit can use a controller with PWM output, which is installed on the control panel on the side of the mobile chassis 4. The interface of the drive motor 5 is located on the rear side of the motor housing. The terminal box of the electromagnetic brake 6 is fixed to the end of the track wheel shaft. The length of the connecting line does not exceed 80cm.

[0036] The output of the UWB base station is connected to the calibration interface of the inertial measurement unit via the CAN bus. The communication rate can be set to 250 kbps. The coaxial cable has a 50Ω impedance and is coated with polytetrafluoroethylene. The UWB base station is mounted on the antenna mounting plate at the rear of mobile chassis 4 using a bracket. The inertial measurement unit is fixed to a vibration-damping platform in the center of the chassis. The two are connected via a 25 cm coaxial cable, with a rubber waterproof seal installed at the interface.

[0037] Working process: The environmental data processing unit transmits the fused 3D grid map (resolution 50 mm × 50 mm × 50 mm) to the path planning unit via an RJ45 interface. The map data includes the deviation between the diaphragms (threshold 50 mm) and the height of the stiffeners. The path planning unit generates a detour path based on these deviations, with a steering angle threshold set to ±20 degrees. Using an internal algorithm, the path planning unit outputs a sequence of path coordinates to the motion control unit. Upon receiving these instructions, the motion control unit adjusts the speed of the drive motor 5 using a PWM signal (for example, a duty cycle of 55% for the left motor and 45% for the right motor) and sends braking torque parameters (for example, 10 N·m for the left motor and 8 N·m for the right motor) to the electromagnetic brake 6. The incremental encoder of the drive motor 5 provides real-time speed feedback to the control unit, forming a closed loop. At the same time, the ultra-wideband positioning base station transmits the tag coordinates (synchronization error ≤ 1ms) to the inertial measurement unit via the CAN bus at a rate of 250kbps. The latter uses the Kalman filter algorithm to correct the gyroscope drift (the drift amount is reduced from 0.05° / s to 0.01° / s). When the map origin drift exceeds the lateral threshold of 4mm or the height threshold of 3mm, real-time coordinate correction is triggered to ensure that the positioning error is ≤ 5mm.

[0038] Technical effect: Through standardized interfaces and high-reliability connection design, the stability of data transmission between environmental data processing, path planning, motion control and positioning modules is guaranteed; shielded cables and anti-interference interfaces reduce the impact of electromagnetic interference inside the steel box girder 3 on the signal; closed-loop control and real-time correction mechanism improve motion precision and positioning accuracy, providing a stable mobile platform and data synchronization basis for the detection module. The overall system achieves efficient collaborative operation in complex environments.

[0039] In another technical solution, the three-dimensional grid map output by the environmental data processing unit to the path planning unit includes the deviation between the measured value and the design value of the diaphragm spacing and the stiffener height data. The path planning unit generates a turning angle threshold for the detour path based on the diaphragm spacing deviation. The control signal sent by the motion control unit to the drive motor 5 and the electromagnetic brake 6 includes a braking torque gradient parameter and a motor speed compensation coefficient calculated based on the steering angle threshold, wherein the braking torque gradient parameter is linearly related to the angular velocity of the track wheel shaft, and the motor speed compensation coefficient is dynamically adjusted according to the inclination angle of the bottom plate of the steel box girder 3; The correction data output by the ultra-wideband positioning base station to the inertial measurement unit includes a mapping error value between the tag coordinates and the three-dimensional grid map, and the tag coordinates are attached with a timestamp synchronized with the lidar scanning cycle. The inertial measurement unit performs segmented compensation for gyroscope drift based on the mapping error value and timestamp; In the above technical solution, the three-dimensional grid map output by the environmental data processing unit to the path planning unit includes the deviation between the measured and designed values ​​of the diaphragm spacing (in mm) and the stiffener height data (in mm). The path planning unit generates a steering angle threshold for the detour path based on the diaphragm spacing deviation. This threshold range is positively correlated with the deviation. For example, when the deviation exceeds 10 mm, the steering angle threshold is set to 15 to 30 degrees. The environmental data processing unit can be an industrial-grade embedded computer, integrated into the control module, and installed in the center of the mobile chassis 4 to facilitate the reception of sensor data from the navigation module. The path planning unit can be an off-the-shelf path planning algorithm module, connected to the environmental data processing unit via a data bus to obtain grid map attributes in real time.

[0040] The control signals sent by the motion control unit to the drive motor 5 and electromagnetic brake 6 include a braking torque gradient parameter (in N·m / rad / s) calculated based on the steering angle threshold and a motor speed compensation coefficient (dimensionless). The braking torque gradient parameter is linearly related to the angular velocity of the track wheel axle. For example, for every 1 rad / s increase in angular velocity, the braking torque gradient parameter increases by 0.5 N·m / rad / s. The motor speed compensation coefficient is dynamically adjusted based on the inclination angle (in degrees) of the bottom plate of the steel box girder 3. When the inclination exceeds 5 degrees, the compensation coefficient is adjusted between 0.9 and 1.1. The drive motor 5 can be a DC servo motor mounted on the end of the track wheel axle of the mobile chassis 4. The electromagnetic brake 6 can be an electromagnetic power-off brake, fixed to the outer sides of each left and right track wheel axle. The motion control unit can be a dedicated control circuit board connected to the drive motor 5 and electromagnetic brake 6 via a cable to output control signals in real time.

[0041] The correction data output by the UWB base station to the IMU includes the mapping error between the tag coordinates and the 3D grid map (in millimeters), along with a timestamp (in milliseconds) synchronized with the lidar scanning cycle. The IMU compensates for gyroscope drift in segments based on the mapping error and timestamp. For example, correction data is collected every 100 milliseconds, and gyroscope drift error is adjusted segment by segment according to the timestamp sequence. The UWB base station can be a commercial UWB positioning device, with tags placed at intervals of 3 to 5 meters on the top and side walls of the inner wall of the steel box girder 3. The IMU can be a combined module containing a gyroscope and accelerometer, mounted at the center of gravity of the mobile chassis 4 to ensure the accuracy of attitude data collection.

[0042] Working Process: When the inspection device is operating, the environmental data processing unit first fuses data from the LiDAR and binocular vision sensors to create a three-dimensional grid map that includes diaphragm spacing and stiffener heights. It then calculates the deviation between the measured diaphragm spacing and the designed value and transmits this information as map attributes to the path planning unit. Based on this deviation, the path planning unit generates a corresponding steering angle threshold. For example, if the diaphragm spacing at a certain section is 20mm narrower than the designed value, a 25-degree steering angle threshold is automatically generated to ensure the detour avoids obstacles.

[0043] The motion control unit receives real-time attitude data from the inertial measurement unit and, combined with the steering angle threshold from the path planning unit, calculates the braking torque gradient parameter and motor speed compensation coefficient. When the mobile chassis 4 turns, the braking torque of the electromagnetic brake 6 is linearly adjusted based on the angular velocity of the track wheel axle. Simultaneously, the speed of the drive motor 5 is dynamically corrected based on the inclination angle of the baseplate. For example, when the chassis turns left, the left electromagnetic brake 6 applies a greater braking torque, and the speed compensation coefficient of the right drive motor 5 is adjusted to 1.05 to ensure smooth steering.

[0044] The ultra-wideband positioning base station collects tag coordinates in real time, compares them with the corresponding coordinates in the 3D grid map, calculates the mapping error, and sends it to the inertial measurement unit (IMU) with a timestamp. The IMU processes the correction data in chronological segments. For example, after receiving five timestamps of error values, it compensates for gyroscope drift in segments. This reduces positioning errors caused by long-term operation and ensures the position accuracy of the inspection device.

[0045] Technical effect: By integrating the diaphragm spacing deviation and stiffener height data into the three-dimensional grid map, more accurate environmental parameters are provided for path planning, so that the steering angle of the detour path is more in line with the actual structural changes. The dynamic adjustment mechanism of the motion control signal, combined with the bottom plate inclination and the axle angular velocity, improves the steering stability and speed consistency of the mobile chassis 4 in complex terrain. The segmented compensation method of ultra-wideband positioning data and inertial measurement unit effectively reduces the positioning error caused by gyroscope drift, ensuring that the inspection device maintains a high position accuracy during long-distance inspections. These designs jointly ensure the rationality of path planning, motion control accuracy and positioning stability of the inspection device in the complex environment inside the steel box girder 3, and provide technical support for the efficient completion of inspection tasks.

[0046] In another technical solution, the output shaft of the drive motor 5 is mechanically connected to the track wheel shaft through a planetary reduction gear set 10, and the transmission ratio of the planetary reduction gear set 10 is 10:1 to 15:1; An incremental encoder is installed at the end of the rotor shaft of the drive motor 5. The pulse signal output end of the incremental encoder is connected to the speed feedback interface of the control module. The control module generates a pulse width modulation signal according to the difference between the encoder pulse signal and the preset speed. The pulse width modulation signal drives the drive motor 5 through the H-bridge circuit. In the above technical solution, the planetary reduction gear set 10 can be mounted on the mounting platform 11 via a bracket (not shown). The output shaft of the drive motor 5 is mechanically connected to the track wheel shaft via the planetary reduction gear set 10. The transmission ratio of the planetary reduction gear set 10 can be set to 12:1 or 14:1, the gear module can be selected from 2.0 or 2.5, and the gears can be made of 20CrMnTi alloy steel, with a surface carburized and quenched treatment to a hardness of HRC 58-62. The housing of the planetary reduction gear set 10 can be made of ADC12 aluminum alloy and is bolted to the output flange of the drive motor 5. The input spline of the track wheel shaft meshes with the spline of the gear set output shaft. The spline teeth can be set to 10 or 12, with a clearance of no more than 0.05 mm. The rotor shaft end of the drive motor 5 is connected to the gear set input shaft via a keyway. The keyway dimensions can be selected from 6 mm × 6 mm × 20 mm, and the material is 45 steel.

[0047] The incremental encoder is installed at the end of the rotor shaft of the drive motor 5. It can be selected with a resolution of 500PPR or 1000PPR. The output signal type can be selected as A / B phase quadrature pulse or differential signal. The mounting flange of the encoder can be made of aluminum alloy and fixed to the motor shaft end by a screw. The coaxial error with the keyway of the shaft end shall not exceed 0.02mm. The signal line can be a twisted shielded pair with a core cross-sectional area of ​​0.5mm. 2 The outer layer is covered with a PVC sheath, and the shield layer is grounded at one end. The encoder's pulse signal output is connected to the speed feedback interface of the control module via a DB9 interface. The interface card can be equipped with an optocoupler isolation circuit with an isolation voltage of no less than 2500Vrms.

[0048] The control module generates a pulse-width modulation signal based on the difference between the encoder pulse signal and the preset speed. The PWM frequency can be set to 10kHz or 20kHz, and the duty cycle adjustment accuracy can be set to 1% or 0.5%. The H-bridge circuit can use a MOSFET module with an on-resistance of no more than 5mΩ and a withstand voltage of no less than 60V. The heat sink material can be 6063 aluminum alloy with an anodized surface. The H-bridge circuit is connected to the power terminal of the drive motor 5 via a copper busbar. The cross-sectional area of ​​the copper busbar can be selected to be 10mm×2mm, and the contact surface is tin-plated to reduce resistance. The PWM signal output terminal of the control module is connected to the drive port of the H-bridge circuit via a cable. The cable length should not exceed 20cm and the outer layer is coated with an anti-interference magnetic ring.

[0049] Operation: The output shaft of drive motor 5 rotates at 2500 rpm, which is reduced to 208 rpm by planetary reduction gear set 10 (gear ratio 12:1). This speed drives the track wheel shaft, driving the crawler mechanism at a speed of 0.3 m / s. An incremental encoder rotates synchronously with the motor shaft, outputting 500 A / B-phase quadrature pulse signals per revolution. These pulse signals are transmitted to the control module via twisted-pair shielded cable. The control module calculates the real-time speed based on the pulse frequency. If the actual speed is 200 rpm (lower than the preset 208 rpm), it generates a 65% duty cycle PWM signal (frequency 10 kHz) and outputs it to drive motor 5 via an H-bridge circuit. The MOSFET modules in the H-bridge switch on and off according to the PWM signal, outputting an average voltage of 24 V (peak voltage 48 V), thereby increasing the motor speed to the target value. Simultaneously, the heat sink controls the surface temperature of the MOSFET modules below 60°C through natural convection, and the temperature rise at the copper busbar connection is monitored by infrared temperature measurement to not exceed 40°C. The closed-loop feedback of the encoder and the PWM regulation form a dynamic balance to ensure the stable walking speed of the crawler.

[0050] Technical effect: The high transmission ratio design of the planetary reduction gear set 10 improves the output torque of the drive motor 5 and meets the load requirements of the crawler walking mechanism; the closed-loop feedback mechanism of the incremental encoder improves the speed control accuracy and reduces the speed difference caused by road resistance or load changes; the low conduction loss and heat dissipation optimization of the H-bridge circuit ensure long-term operation stability. The overall transmission system realizes efficient and reliable power transmission and speed control in the complex environment inside the steel box girder 3.

[0051] In another technical solution, within the angle range of 30 to 45 degrees between the optical axis of the multispectral imaging unit and the detection plane of the ultrasonic flaw detection unit, the overlap rate between the field of view coverage area of ​​the multispectral imaging unit and the detection area of ​​the ultrasonic flaw detection unit is 30% to 50%; The switching cycle of the filter array is 0.5s to 1.5s, and the time of maintaining a fixed wavelength after each switching accounts for 60% to 80% of a single cycle. The probe trigger interval of the ultrasonic flaw detection unit is an integer multiple of the filter switching cycle; The wavelength range of the multispectral imaging unit includes the visible light band of 380nm-780nm and the short-wave infrared band of 900nm-1700nm, and the detection plane of the ultrasonic flaw detection unit is arranged with an array of calibration reference points corresponding to the overlapping area of ​​the field of view of the multispectral imaging unit; The spacing of the calibration reference point array is 1.2 to 1.5 times the spacing of the ultrasonic probe array 1, and the difference in reflectivity of each calibration reference point at each wavelength of the multispectral imaging unit does not exceed 15%; In the above technical solution, the optical axis of the multispectral imaging unit forms an angle of 30 to 45 degrees with the detection plane of the ultrasonic flaw detection unit, and the overlap rate of their field of view is 30 to 50%. The multispectral imaging unit can be an industrial-grade multispectral camera equipped with a switchable filter array, mounted above the detection module bracket, with the optical axis facing the wall of the steel box girder 3. The ultrasonic flaw detection unit can be a phased array ultrasonic probe array 1, fixed below the detection module bracket, with the detection plane parallel to the wall. The two are rigidly connected by an aluminum alloy bracket to ensure a stable angle.

[0052] The filter array switching cycle is 0.5 to 1.5 seconds, and the fixed wavelength is maintained for 60% to 80% of the cycle after each switch. For example, a specific wavelength is maintained for 0.6 to 0.8 seconds within a 1-second cycle. The ultrasonic flaw detection unit's probe trigger interval is an integer multiple of the filter switching cycle. For example, if the switching cycle is 1 second, the probe trigger interval is set to 0.5 or 1 second. The filter array can be a motorized rotating filter wheel integrated into the front end of the multispectral camera's lens. The ultrasonic probe's trigger control circuit can be an off-the-shelf pulse generator, connected to the control module via a synchronization signal line.

[0053] The multispectral imaging unit's wavelength range covers the visible light band of 380nm-780nm and the shortwave infrared band of 900nm-1700nm. The filter material is a glass-based multilayer film structure. An array of calibration reference points is arranged within the ultrasonic flaw detection unit's detection plane, with a spacing of 1.2 to 1.5 times the spacing of the ultrasonic probe array 1 (50mm to 100mm). For example, with an 80mm probe spacing, the reference point spacing is set to 96mm to 120mm. The reflectivity of each calibration reference point should not differ by more than 15% across all multispectral wavelengths. The material can be an aluminum disc with a uniform diffuse reflective coating sprayed on the surface, fixed to the edge of the detection plane of the ultrasonic probe array 1.

[0054] Working process: When the inspection module of the patrol device is in close proximity to the wall of steel box girder 3, the multispectral imaging unit and ultrasonic flaw detection unit operate synchronously. The multispectral camera uses a filter wheel to sequentially switch wavelengths between 380nm and 1700nm, maintaining each wavelength for 0.6 to 0.8 seconds while capturing wall images. The ultrasonic probe array 1 emits ultrasonic waves at intervals that are integer multiples of the filter switching cycle (e.g., 1 second or 0.5 seconds) to detect internal defects on the wall. The fields of view of the two units overlap by 30% to 50% within an angle range of 30 to 45 degrees, ensuring that the image and ultrasonic data of the same area correspond.

[0055] The calibration reference point array serves as a spatial reference, aligning multispectral images and ultrasonic detection data within overlapping areas. For example, when the multispectral camera captures an image containing reference points, the reference point locations are identified and combined with the physical coordinates of the reference points on the ultrasonic probe to establish a spatial mapping between the two, correcting for positional deviations caused by installation angles. The control module coordinates the filter switching cycle with the probe trigger interval to ensure data acquisition timing synchronization and avoid signal interference.

[0056] Technical effect: By limiting the spatial angle and field of view overlap rate between the multispectral imaging unit and the ultrasonic flaw detection unit, spatial correspondence between visual imaging and ultrasonic detection data of the same area is achieved, which facilitates the subsequent cross-verification of defect locations. The coordinated design of the filter switching cycle and the probe trigger interval reduces the timing conflicts between sensors and improves data acquisition efficiency. The parameter design and material selection of the calibration reference point array ensure the precise alignment of data from different detection modalities and reduce the calibration error caused by reflectivity differences. These designs jointly ensure the consistency and complementarity of multimodal detection data, provide a reliable fusion data basis for the identification of apparent defects and internal structural flaw detection of the wall of the steel box girder 3, and improve the accuracy and reliability of the detection results.

[0057] In another technical solution, the environmental data processing unit creates a three-dimensional grid map by fusing the point cloud data of the laser radar with the stereo matching data of the binocular vision sensor. The point cloud density of a single scan of the laser radar is 2000 points / m 2 Up to 5000 points / m 2 , the number of feature point matches of the binocular vision sensor is 50 to 80 per frame image; The calculation of the diaphragm spacing is based on the geometric projection relationship between the laser radar scanning plane and the roof elevation angle. The spacing of the reflection point clouds of adjacent diaphragms is extracted as the measured value, and the difference operation is performed with the design value. The difference operation result is stored as an attribute field of the raster map; The height of the stiffening rib is measured by calculating the disparity map of the binocular vision sensor. The depth resolution of the disparity map reaches 0.1mm to 0.3mm, and during the measurement, at least three feature points are ensured to fall on the vertical section of a single stiffening rib. The dynamic update frequency of the three-dimensional grid map is 0.5Hz to 1Hz. During each update, the map origin is drift-corrected using the tag coordinates of the ultra-wideband positioning base station. The drift correction thresholds are set to 3mm to 5mm for lateral deviation and 2mm to 4mm for height deviation. The feature matching algorithm built into the environmental data processing unit aligns the edge contour of the diaphragm extracted by the lidar with the corner points of the stiffener identified by binocular vision in spatial coordinates, with the alignment error not exceeding 1.5 times the spacing of the lidar point cloud; In the above technical solution, the 3D grid map is established by fusing the LiDAR point cloud data with the binocular vision stereo matching data. The point cloud density of a single scan of the LiDAR can be 2000 points / m 2 , 3000 points / m 2 , 4000 points / m 2 or 5000 points / m 2The binocular vision sensor matches 50, 60, 70, or 80 feature points per frame. The laser radar can be a mechanical rotating laser radar or a solid-state laser radar, and the binocular vision sensor can be an industrial-grade binocular camera. Both are mounted above the front of the mobile chassis 4. The laser radar's scanning plane is at an elevation angle of 5 to 15 degrees relative to the top plate of the steel box girder 3, and the optical axis of the binocular vision sensor faces straight ahead. The environmental data processing unit can be an embedded industrial computer integrated into the control module to receive and process sensor data. Parameter setting is based on the internal spatial dimensions of the steel box girder 3. Calibration experiments are performed to determine the sensor installation angle and data fusion algorithm parameters to ensure spatial consistency between the point cloud and image features.

[0058] The calculation of diaphragm spacing is based on the geometric projection of the LiDAR scanning plane. The spacing between adjacent diaphragm reflection point clouds is extracted as the measured value, and the difference between the measured and designed values ​​is stored as a raster map attribute. The stiffener height is calculated using a binocular disparity map with a depth resolution of 0.1mm, 0.2mm, or 0.3mm. During measurement, at least three feature points must fall within the vertical cross-section of a single stiffener. The LiDAR point cloud data processing software can be an off-the-shelf point cloud processing library, and the binocular disparity calculation module can be a commercially available vision processing chip. During assembly, the LiDAR scan direction covers the top plate and side walls of the steel box girder 3. The baseline length of the binocular vision sensor is adjusted according to the installation height to ensure accurate disparity calculation. The difference between the measured and designed values ​​is calculated using a pre-set geometric model and stored in each cell attribute of the raster map.

[0059] The dynamic update frequency of the three-dimensional grid map is 0.5Hz or 1Hz. Each update uses the tag coordinates of the ultra-wideband positioning base station to correct the drift of the map origin. The correction threshold is lateral deviation of 3mm, 4mm or 5mm, and height deviation of 2mm, 3mm or 4mm.

[0060] The feature matching algorithm built into the environmental data processing unit aligns the spatial coordinates of the diaphragm edge contours extracted by the lidar with the stiffener corner points identified by binocular vision. The alignment error does not exceed 1.5 times the lidar point cloud spacing. The ultra-wideband positioning base station tag is placed on the inner wall of the steel box girder 3, and the inertial measurement unit is installed at the center of gravity of the mobile chassis 4 to ensure that the positioning data is consistent with the sensor coordinate system. When updating the map, the map origin offset is first corrected using ultra-wideband positioning data. Newly collected lidar and vision data are then integrated to update attribute fields such as diaphragm spacing and stiffener height. The feature matching algorithm is implemented using existing algorithms such as iterative closest point (ICP), and the parameters are pre-calibrated using a calibration plate.

[0061] Working Process: As the inspection device moves, a LiDAR scans the interior of the steel box girder 3 at an elevation angle of 5 to 15 degrees, generating point cloud data. A binocular vision sensor simultaneously captures images and extracts feature points, generating a disparity map through stereo matching. The environmental data processing unit fuses the LiDAR point cloud with the visual feature points to create a three-dimensional grid map that includes diaphragm spacing and stiffener heights. The LiDAR point cloud is geometrically projected to calculate the measured diaphragm spacing, which is then compared with the designed value and stored as a map attribute. The binocular vision disparity map calculates stiffener heights through feature point matching, ensuring that at least three feature points cover a single stiffener section. The map is updated at a frequency of 0.5 to 1 Hz. Before each update, map origin drift is corrected using tag coordinates from an ultra-wideband positioning base station. Correction is triggered when lateral deviation exceeds 3 to 5 mm or height deviation exceeds 2 to 4 mm. Simultaneously, a feature matching algorithm aligns the diaphragm edges with the stiffener corner coordinates, ensuring spatial consistency of structural features in the map and providing an accurate environmental model for path planning.

[0062] Technical Results: By fusing LiDAR and binocular vision data, high-precision 3D modeling of the internal structure of steel box girder 3 is achieved. Measured values ​​for diaphragm spacing and stiffener height are compared and stored with designed values, providing detailed environmental parameters for path planning. Dynamic updates and drift correction ensure the real-time and accuracy of the map. A feature matching algorithm improves the spatial consistency of data from different sensors, enabling the inspection device to accurately perceive the internal structure of steel box girder 3. This provides a reliable environmental data foundation for obstacle avoidance path planning and precise positioning of the detection module, ensuring the efficiency and reliability of inspection operations.

[0063] In another technical solution, a height adjustment rail is provided at the bottom of the mounting bracket of the binocular vision sensor, the scale accuracy of the height adjustment rail is 1 mm, and the baseline length of the binocular vision sensor is set to 0.3 to 0.5 times the mounting height; When the field of view angle of the binocular vision sensor is 120 degrees to 150 degrees, its lens distortion correction adopts a quintic polynomial model, and the coefficients of the quintic polynomial model are obtained by fitting 20 to 30 groups of pose data of a checkerboard calibration plate, where the side length of a single grid of the checkerboard calibration plate is 30 mm to 50 mm; When the installation height of the binocular vision sensor is 400mm to 600mm, the angle between its optical axis and the bottom plate of the steel box girder 3 is 70 degrees to 80 degrees, and the vertical distance between the lowest field of view boundary and the ground contact point of the crawler of the mobile chassis 4 is 100mm to 150mm; The exposure time of the binocular vision sensor is dynamically adjusted according to the light intensity inside the steel box girder 3. The light intensity is detected by the built-in photoresistor of the sensor. The exposure time adjustment range is 1ms to 10ms, and the exposure time change rate of adjacent frames does not exceed 20%; In the above technical solution, the binocular vision sensor is mounted on a mounting bracket, the bottom of which is provided with a height adjustment rail, which is mounted on the mobile chassis 4 (e.g., the mounting platform 11). The height adjustment rail has a scale accuracy of 1 mm, and the baseline length can be set to 0.3 to 0.5 times the installation height (400 mm to 600 mm). For example, for an installation height of 500 mm, the baseline length can be 150 mm to 250 mm. The height adjustment rail can be a rack-type or screw-type lifting structure, which is a very mature structure and will not be described in detail here. The height adjustment rail is preferably mounted on the front upper portion of the mobile chassis 4 to facilitate vertical height adjustment.

[0064] When the binocular vision sensor has a field of view of 120 to 150 degrees, lens distortion correction uses a quintic polynomial model. The model coefficients are obtained by fitting 20 to 30 sets of pose data from a checkerboard calibration plate. The calibration plate can have a single grid length of 30 mm, 40 mm, or 50 mm. When installed at a height of 400 to 600 mm, the angle between the optical axis and the bottom plate of the steel box girder 3 is 70 to 80 degrees, and the vertical distance from the lowest field of view edge to the track contact point is 100 to 150 mm to ensure that the sensor can capture bottom details. The binocular vision sensor can be an industrial-grade binocular camera. During installation, the height is adjusted using guide rails to ensure that the optical axis angle meets the design requirements. The bracket material can be aluminum alloy with an anti-corrosion surface treatment.

[0065] The binocular vision sensor's dynamic exposure time adjustment mechanism detects light intensity via a built-in photoresistor. The exposure time is adjustable from 1ms to 10ms, with the rate of change between frames not exceeding 20%. The photoresistor can be a silicon-based photosensor integrated around the binocular vision sensor lens. The control module calculates and adjusts the exposure time in real time based on light intensity, for example, extending the exposure time to 8ms in dim light and shortening it to 2ms in bright light, ensuring a clear, unblurred image.

[0066] Working Process: After the inspection device is activated, first adjust the binocular vision sensor's mounting height to 400mm to 600mm using the height adjustment rail. Based on actual requirements, set the baseline length to 0.3 to 0.5 times the mounting height. For example, a 500mm height corresponds to a 200mm baseline. During installation, ensure that the angle between the optical axis and the baseplate is 70 to 80 degrees, and the minimum field of view is approximately 120mm from the track contact point, covering the bottom and sidewall areas of the steel box girder.

[0067] The sensor uses a checkerboard calibration plate for distortion correction. It collects 20 to 30 sets of images at different poses and fits the coefficients of a quintic polynomial model to eliminate the effects of lens distortion. During operation, a photoresistor monitors the internal light intensity in real time. The control module dynamically adjusts the exposure time based on this value. For example, when light intensity decreases, the exposure time is gradually increased, with each adjustment rate not exceeding 20% ​​to avoid sudden changes in image brightness. If light intensity is stable, the exposure time is maintained within an appropriate range, ensuring clear image features and providing high-quality data for binocular stereo matching and environmental modeling.

[0068] Technical Effect: Through the height-adjustable mounting structure and baseline setting, the binocular vision sensor can adapt to the inspection requirements at different heights inside the steel box girder 3, covering the key areas from the bottom to the side walls. Lens distortion correction and precise mounting angle design improve the accuracy of image feature point extraction, ensuring the accuracy of parallax calculation and depth measurement. The dynamic exposure adjustment mechanism adapts to the complex lighting environment inside the steel box girder 3, avoiding overexposure or underexposure, and ensuring the stability and reliability of visual data. These designs jointly enhance the navigation module's ability to perceive the environment, provide high-precision visual information for the establishment of three-dimensional grid maps and path planning, and ensure the stable operation of the inspection device within complex structures.

[0069] In another technical solution, the height adjustment guide rail adopts a rack-type lifting column; the base of the rack-type lifting column is fixed to the mobile chassis 4 by bolts, and the binocular vision sensor bracket is connected to the slider of the rack-type lifting column through a flange; In the above technical solution, the rack-type lifting column is an existing product, such as the optical slide model ZKB20-300-210. The rack module of the rack-type lifting column can be 1.5 or 2.0, and the effective stroke can be 400mm, 500mm, or 600mm, meeting the binocular vision sensor's installation height adjustment requirement of 400mm to 600mm. The rack can be made of steel with a hardened surface for enhanced wear resistance, and the base and slider can be made of aluminum alloy, balancing strength and lightweight. The base (i.e., bottom) of the rack-type lifting column is bolted to the mobile chassis 4. The binocular vision sensor bracket is connected to the lifting column's slider via a flange. Bolts are used to tighten the flange and slider to ensure that the sensor is perpendicular to the axis of the lifting column after installation.

[0070] Working process: When the binocular vision sensor's mounting height needs to be adjusted, the operator operates the rack-type lifting column to change the sensor's height. Once adjusted, the slider is locked in position to prevent height changes caused by vibration during inspection.

[0071] Technical Effect: The rack-type lifting column structure design enables manual and precise adjustment of the binocular vision sensor height, meeting the coverage requirements of detection areas at different heights inside the steel box girder 3. This improves the environmental adaptability of the vision sensor, provides a stable hardware foundation for the establishment of three-dimensional grid maps and obstacle recognition, and ensures the visual perception accuracy of the inspection device within complex structures. The number of devices and processing scales described here are used to simplify the description of the present invention. Applications, modifications and variations of the internal inspection device for steel box beams of the present invention will be obvious to those skilled in the art.

[0072] Although the embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the description and implementation methods. They can be fully applied to various fields suitable for the present invention. For those familiar with the art, additional modifications can be easily implemented. Therefore, without departing from the general concept defined by the claims and the scope of equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.

Claims

1. The internal inspection device of the steel box girder is characterized by: It includes mobile chassis, navigation module, detection module, positioning module and control module; The mobile chassis includes a crawler-type traveling mechanism, a drive motor and a steering mechanism, and the mobile chassis moves inside the box body of the steel box girder. The steering mechanism includes two independently controlled electromagnetic brakes, which act on the crawler wheel shafts on the left and right sides of the crawler-type traveling mechanism respectively; The navigation module includes a binocular vision sensor, a laser radar, and an environmental data processing unit. The binocular vision sensor is installed at a height of 400 mm to 600 mm and has a field of view of 120 to 150 degrees. The scanning plane of the laser radar is at an elevation angle of 5 to 15 degrees to the top plate inside the steel box girder. The environmental data processing unit establishes a three-dimensional grid map including the spacing between diaphragms and the height of stiffeners. The detection module includes a multispectral imaging unit and an ultrasonic flaw detection unit. The multispectral imaging unit includes a filter array with a wavelength range of 380nm to 1700nm. The probe array of the ultrasonic flaw detection unit has a spacing of 50mm to 100mm. The probe array is in elastic contact with the wall surface inside the steel box girder through a spring mechanism. The positioning module includes an inertial measurement unit and an ultra-wideband positioning base station. The sampling frequency of the inertial measurement unit is 100Hz to 200Hz, the tag arrangement spacing of the ultra-wideband positioning base station is 3m to 5m, and the tag coordinates are mapped to the three-dimensional grid map. The control module includes a path planning unit and a motion control unit. The path planning unit calculates the shortest path avoiding the diaphragm according to the three-dimensional grid map, and the motion control unit adjusts the braking torque of the electromagnetic brake according to the posture data of the inertial measurement unit.

2. The internal inspection device of the steel box girder according to claim 1, characterized in that: The output end of the environmental data processing unit is connected to the input end of the path planning unit, the output end of the motion control unit is connected to the input end of the drive motor and the electromagnetic brake, and the output end of the ultra-wideband positioning base station is connected to the correction interface of the inertial measurement unit.

3. The internal inspection device of the steel box girder according to claim 1, characterized in that: The three-dimensional grid map output by the environmental data processing unit to the path planning unit includes the deviation between the measured value and the designed value of the diaphragm spacing and the stiffener height data, and the path planning unit generates a turning angle threshold for the detour path based on the deviation of the diaphragm spacing; The control signal sent by the motion control unit to the drive motor and electromagnetic brake includes a braking torque gradient parameter and a motor speed compensation coefficient calculated based on a steering angle threshold, wherein the braking torque gradient parameter is linearly related to the angular velocity of the track wheel shaft, and the motor speed compensation coefficient is dynamically adjusted according to the inclination angle of the steel box girder bottom plate; The correction data output by the ultra-wideband positioning base station to the inertial measurement unit includes the mapping error value between the tag coordinates and the three-dimensional grid map, and the tag coordinates are attached with a timestamp synchronized with the lidar scanning cycle. The inertial measurement unit performs segmented compensation for gyroscope drift based on the mapping error value and timestamp.

4. The internal inspection device of the steel box girder according to claim 3, characterized in that: The output shaft of the drive motor is mechanically connected to the track wheel shaft through a planetary reduction gear set, and the transmission ratio of the planetary reduction gear set is 10:1 to 15:1; An incremental encoder is installed at the end of the rotor shaft of the drive motor. The pulse signal output end of the incremental encoder is connected to the speed feedback interface of the control module. The control module generates a pulse width modulation signal according to the difference between the encoder pulse signal and the preset speed. The pulse width modulation signal drives the drive motor through the H-bridge circuit.

5. The internal inspection device of the steel box girder according to claim 4, characterized in that: Within the angle range of 30 to 45 degrees between the optical axis of the multispectral imaging unit and the detection plane of the ultrasonic flaw detection unit, the overlap rate between the field of view coverage area of ​​the multispectral imaging unit and the detection area of ​​the ultrasonic flaw detection unit is 30% to 50%; The switching cycle of the filter array is 0.5s to 1.5s, and the time of maintaining a fixed wavelength after each switching accounts for 60% to 80% of a single cycle. The probe trigger interval of the ultrasonic flaw detection unit is an integer multiple of the filter switching cycle; The wavelength range of the multispectral imaging unit includes the visible light band of 380nm-780nm and the short-wave infrared band of 900nm-1700nm, and the detection plane of the ultrasonic flaw detection unit is arranged with an array of calibration reference points corresponding to the overlapping area of ​​the field of view of the multispectral imaging unit; The spacing of the calibration reference point array is 1.2 to 1.5 times the spacing of the ultrasonic probe array, and the difference in reflectivity of each calibration reference point at each wavelength of the multispectral imaging unit does not exceed 15%.

6. The internal inspection device of the steel box girder according to claim 1, characterized in that: The environmental data processing unit creates a three-dimensional grid map by fusing the point cloud data of the laser radar with the stereo matching data of the binocular vision sensor. The point cloud density of a single scan of the laser radar is 2000 points / m 2 Up to 5000 points / m 2 , the number of feature point matches of the binocular vision sensor is 50 to 80 per frame image; The calculation of the diaphragm spacing is based on the geometric projection relationship between the laser radar scanning plane and the roof elevation angle. The spacing of the reflection point clouds of adjacent diaphragms is extracted as the measured value, and the difference operation is performed with the design value. The difference operation result is stored as an attribute field of the raster map; The height of the stiffening rib is measured by calculating the disparity map of the binocular vision sensor. The depth resolution of the disparity map reaches 0.1mm to 0.3mm, and during the measurement, at least three feature points are ensured to fall on the vertical section of a single stiffening rib. The dynamic update frequency of the three-dimensional grid map is 0.5Hz to 1Hz. During each update, the map origin is drift-corrected using the tag coordinates of the ultra-wideband positioning base station. The drift correction thresholds are set to 3mm to 5mm for lateral deviation and 2mm to 4mm for height deviation. The feature matching algorithm built into the environmental data processing unit aligns the spatial coordinates of the diaphragm edge contour extracted by the lidar with the stiffening rib corner points identified by binocular vision, and the alignment error does not exceed 1.5 times the lidar point cloud spacing.

7. The internal inspection device of the steel box girder according to claim 1, characterized in that: A height adjustment rail is provided at the bottom of the mounting bracket of the binocular vision sensor. The scale accuracy of the height adjustment rail is 1 mm, and the baseline length of the binocular vision sensor is set to 0.3 to 0.5 times the mounting height; When the field of view angle of the binocular vision sensor is 120 degrees to 150 degrees, its lens distortion correction adopts a quintic polynomial model, and the coefficients of the quintic polynomial model are obtained by fitting 20 to 30 groups of pose data of a checkerboard calibration plate, where the side length of a single grid of the checkerboard calibration plate is 30 mm to 50 mm; When the installation height of the binocular vision sensor is 400mm to 600mm, the angle between its optical axis and the bottom plate of the steel box girder is 70 degrees to 80 degrees, and the vertical distance between the lowest field of view boundary and the ground contact point of the mobile chassis crawler track is 100mm to 150mm; The exposure time of the binocular vision sensor is dynamically adjusted according to the light intensity inside the steel box girder. The light intensity is detected by the photoresistor built into the sensor. The exposure time adjustment range is 1ms to 10ms, and the exposure time change rate of adjacent frames does not exceed 20%.

8. The internal inspection device for steel box beams according to claim 7, characterized in that: The height adjustment guide rail adopts a rack-type lifting column; the base of the rack-type lifting column is fixed to the mobile chassis by bolts, and the binocular vision sensor bracket is connected to the slider of the rack-type lifting column by a flange.

Images