Automatic high-strength bolt tightening machine based on machine vision positioning and tightening method

Abstract

The invention relates to steel box girder high-strength bolt operation, in particular to a high-strength bolt automatic tightening machine based on machine vision positioning and a tightening method. The automatic tightening machine for the high-strength bolts is used for automatically tightening the bolts of the steel box girder and comprises a multi-shaft moving support, a tightening assembly and a control unit. The multi-axis moving support achieves plane movement of a steel box girder top plate through a servo-driven moving part, a multi-axis support provides three-dimensional space freedom, and the tightening assembly is driven to accurately adjust the posture. The tightening assembly adopts a double-station design, a first tightening head and a second tightening head which are symmetrically arranged on the two sides synchronously move through a door frame type lifting shaft, the image obtaining set is combined to collect position and posture data of a nut and a screw in real time, and a three-dimensional motion path is generated to guide centering. The control unit coordinates multi-axis motion and torque output based on machine vision feedback, full-process automation of the bolt from recognition, positioning to tightening is achieved, the problems that manual operation is low in efficiency and large in error are solved, and screwing precision and structural reliability are ensured.

Description

Technical Field

[0001] The present invention relates to the operation of high-strength bolts for steel box girders, and particularly to an automatic tightening machine for high-strength bolts based on machine vision positioning and a tightening method thereof. Background Art

[0002] During the construction of steel box girder bridges, the operation of high-strength bolt connections between the diaphragms and webs on the box girders still mainly relies on manual labor, which is likely to cause the following problems: 1. The working conditions are poor and the risk coefficient of manual operation is high. During the tightening process of high-strength bolts, workers often have to rely on simple hanging baskets or hanging nets for high-altitude operations, which will lead to great potential safety hazards; 2. The manual operation intensity is high and the tightening efficiency is low. For a highway steel box girder section with a length, width and span of about 40m, the number of high-strength bolts used only for connecting the diaphragms is as high as nearly ten thousand. Coupled with high-altitude operations, the manual operation intensity is extremely high, and every time the installation work station is moved, workers have to re-place hanging baskets, etc., which also makes the manual tightening efficiency relatively low; 3. The stability of manual tightening is poor, and the consistency and reliability of high-strength bolt tightening cannot be effectively guaranteed, which directly affects the overall construction quality of the steel box girder bridge; 4. It is difficult to realize torque measurement and data collection. At present, it is difficult to implement aspects such as torque control accuracy, torque data change conditions, and re-measurement of final tightening torque data during the tightening process of high-strength bolts, and it is impossible to cover all bolt detections.

[0003] Therefore, aiming at the tightening working conditions and process requirements of high-strength bolts for steel box girders, designing and developing an automatic tightening device and a supporting method for high-strength bolts for steel box girders has extremely high engineering application value and broad market prospects for reducing the risk and intensity of manual operations, improving the tightening efficiency of high-strength bolts, and ensuring the reliability and stability of bolt tightening. Summary of the Invention

[0004] In view of the above technical problems, the present invention provides an automatic tightening machine for high-strength bolts based on machine vision positioning and a tightening method thereof, which are mainly used for the automatic tightening and tightening path planning of high-strength bolts connecting the diaphragms and webs of steel box girders, so as to solve the problems of high manual operation intensity, low efficiency, high requirements for workers' tightening skills, and poor operation level stability in the current high-strength bolt operation of steel box girders.

[0005] On the one hand, the present invention provides an automatic tightening machine for high-strength bolts based on machine vision positioning. The automatic tightening machine for high-strength bolts is used to tighten the high-strength bolts on the steel box girder. The automatic tightening machine for high-strength bolts includes a multi-axis moving support, a tightening assembly, and a control unit;

[0006] The multi-axis moving support has a moving part for enabling the high-strength bolt automatic tightening machine to move on the top plate of the steel box girder; the above-mentioned multi-axis moving support further includes a multi-axis bracket arranged on the moving part; the tightening assembly is used to butt or clamp the high-strength bolts on the steel box girder, and the power part configured according to the tightening assembly drives the high-strength bolts to rotate; the tightening assembly is arranged on the above-mentioned multi-axis bracket, and a driving assembly is configured on the multi-axis bracket for driving the tightening assembly to be movable on multiple spatial axes; the control unit is connected to the tightening assembly and the multi-axis moving support for controlling the tightening assembly and the multi-axis moving support according to the input instructions; wherein, the above-mentioned tightening assembly includes a tightening frame for clamping the steel box girder from both sides of the steel box girder and an image acquisition group for identifying the position of the bolt, and the control unit is connected to the image acquisition group for generating a three-dimensional motion path through the bolt position or pose data collected by the image acquisition group, and the above-mentioned tightening assembly tightens the high-strength bolts on the steel box girder in sequence according to the three-dimensional motion path; first screw heads for adapting to and fixing or screwing the nuts of the high-strength bolts and second screw heads for adapting to and fixing or screwing the high-strength screw rods are respectively arranged at both ends of the clamping of the above-mentioned tightening frame. The three-dimensional motion path here can be a virtual channel with a width in the XZ plane, that is, there is a certain adjustable motion margin or amplitude in the channel width direction.

[0007] By adopting an automatic high-strength bolt tightening machine based on machine vision positioning of the present application, this solution realizes three-dimensional positioning and automatic tightening of high-strength bolts of the steel box girder through the collaborative design of the multi-axis moving support and the machine vision system. Specifically, the multi-axis moving support realizes planar movement on the top plate of the steel box girder through the moving part (such as an X / Y axis platform driven by a servo), and its multi-axis bracket (such as a main cross beam, a vertical frame and a lifting shaft can be set) provides three-dimensional space degrees of freedom, driving the tightening assembly to accurately adjust the position and pose to adapt to the distribution of the bolt group; the tightening assembly adopts a two-station design, the first screw head (adapting to the nut end) and the second screw head (adapting to the screw rod end) are respectively fixed on both sides of the gantry-type lifting shaft, and are symmetrically arranged in cooperation with the image acquisition group to synchronously collect the position data of the nut and the screw rod of the bolt, and a three-dimensional motion path can be generated to guide the centering. This structural design solves the problems of low efficiency and large error of traditional manual positioning, and provides a rigid support basis for automatic screwing.

[0008] In some embodiments, the above-mentioned multi-axis bracket includes a main cross beam and a vertical frame; the main cross beam is arranged on the moving part; the vertical frame is movably arranged on the cross beam, and the tightening assembly is movably arranged on the vertical frame; two photoelectric limit switches located on both sides of the vertical frame are arranged on the above-mentioned main cross beam for respectively detecting the distances between the vertical frame and the two photoelectric limit switches;

[0009] The above vertical frame includes a secondary crossbeam and a gantry; the secondary crossbeam is movably arranged on the main crossbeam and is used to selectively move along the length direction of the main crossbeam; the gantry includes lifting shafts that are respectively vertically movably connected to both ends of the secondary crossbeam, and the upper ends of the two lifting shafts are connected by a connecting rod to form a door frame structure, and the above-mentioned tightening assembly is arranged at the lower end of the lifting shaft; the above-mentioned first screwing head and the second screwing head are respectively located at the lower ends of the two lifting shafts and are arranged oppositely.

[0010] The photoelectric limit switch on the main crossbeam real-time detects the moving range of the vertical frame to ensure the positioning accuracy. The gantry structure realizes the adaptive adjustment of the tightening assembly in the Z-axis direction through the combination of the secondary crossbeam and the lifting shaft, covering complex working surfaces such as the web and diaphragm of the steel box girder.

[0011] Specifically, this solution realizes the three-dimensional space positioning and synchronous tightening functions through the multi-stage motion design of the main crossbeam, vertical frame and gantry. The multi-axis support is composed of a main crossbeam (fixed to the moving part) and a vertical frame. The vertical frame horizontally moves along the length direction of the main crossbeam (X-axis) through the secondary crossbeam, and realizes height adjustment through the vertical movement (Z-axis) of the two side lifting shafts, forming the motion ability of two degrees of freedom in X / Z. The gantry structure forms a door frame by connecting the upper ends of the two lifting shafts with a connecting rod to ensure the synchronous movement of the two side lifting shafts. The first screwing head (fitting nut) and the second screwing head (fitting screw) are respectively installed at the lower ends of them, and realize the bilateral clamping tightening through the relative layout. Photoelectric limit switches are arranged on both sides of the main crossbeam to real-time detect the moving range of the vertical frame (such as detecting the distance between the vertical frame and the limit switch), prevent over-travel and calibrate the positioning accuracy.

[0012] The overall structure of this solution improves the adaptability to bolts of different specifications through modular design (such as replaceable screwing heads, rigid connection of the gantry), and at the same time covers the bolt group operation requirements of the complex curved surface of the steel box girder through a multi-stage motion mechanism (horizontal drive of the main crossbeam + vertical drive of the lifting shaft).

[0013] In some embodiments, the above-mentioned tightening frame includes a tightening seat, a linear module, and a torque sensor; the tightening seat is used to support the second screwing head or the first screwing head; the tightening seat is arranged on the linear module and is used to selectively move on the linear module; the torque sensor is arranged on the tightening seat, the above-mentioned torque sensor is connected to the second screwing head and is used to collect the torque data of the second screwing head; the torque sensor is connected to the control unit, and the control unit is used to real-time receive the dynamic signal of the torque sensor and adjust the output parameters. The above-mentioned torque sensor can be installed between the tightening seat and the second screwing head for detecting torque.

[0014] The dynamic adjustment and precise control of the tightening assembly are achieved through the integrated design of the linear module and the torque sensor in this tightening frame. Specifically, the tightening seat is horizontally or vertically moved along a preset path through a linear module (such as a servo-driven ball screw mechanism), thereby adjusting the relative position between the second tightening head and the bolt; the torque sensor is directly integrated on the tightening seat to collect the torque data of the second tightening head in real time, and the dynamic signal is analyzed through a control unit (such as a PLC or an industrial PC), and then the output parameters such as the rotation speed and torque threshold of the servo motor are adjusted. Among them, the rigid transmission characteristics of the linear module and the closed-loop feedback mechanism of the torque sensor work together to ensure the stability of the tightening quality under complex working conditions. This structural design solves the problem of over-tightening / under-tightening caused by the lack of real-time force feedback in traditional tightening equipment.

[0015] In some embodiments, the above image acquisition group includes: a first camera and a second camera; the first camera is arranged inside the first tightening head or on the tightening seat adjacent to the first tightening head; the second camera is arranged inside the second tightening head or on the tightening seat adjacent to the second tightening head; the above first camera and the second camera are arranged opposite to each other and the imaging directions are on the same straight line in space.

[0016] In some embodiments, at least two ultrasonic rangefinders are arranged on the above tightening frame, and the two ultrasonic rangefinders are arranged with their orientations parallel; the ultrasonic rangefinders located on both sides of the steel box girder are arranged facing each other; the measurement directions of the above ultrasonic rangefinders are parallel to the telescopic direction of the first tightening head or the second tightening head.

[0017] In some embodiments, an electromagnetic suction seat is arranged on the above multi-axis bracket or the moving part, and the electromagnetic suction seat is connected to the control unit and is used to perform opening and closing operations on the electromagnetic suction seat according to the instructions received by the control unit.

[0018] On the other hand, the present invention provides a method for tightening high-strength bolts of a steel box girder based on machine vision positioning, including using the high-strength bolt automatic tightening machine in the above-mentioned one aspect; the tightening method includes the following operations:

[0019] Move the tightening assembly to a preset position, make the image acquisition group face the target area, and make the image acquisition group scan the structural objects of the steel box girder in the current target area to identify whether there are bolts;

[0020] a. If there are bolts, move the tightening assembly according to the bolt position so that the first tightening head or the second tightening head of the tightening assembly corresponds to the nut or the screw head of the high-strength bolt;

[0021] b. If there are no bolts, move the tightening assembly horizontally or vertically according to a preset program until a bolt is identified or an unrecognized signal is issued after the number of moving steps exceeds the preset maximum number of steps;

[0022] Among them, when a bolt is recognized, the first screwing head and the second screwing head are rotated and moved towards the nut and the screw head until the first screwing head is docked with the nut and the second screwing head is docked with the screw head;

[0023] After the docking is completed, the first screwing head keeps rotating and the second screwing head stops rotating, or the first screwing head and the second screwing head continue to rotate. When the tightening process reaches the preset torque / angle threshold, the tightening assembly is triggered to stop tightening and lock.

[0024] This solution realizes automatic tightening control through bolt recognition guided by machine vision and adaptive path planning. Specifically, first, the tightening assembly is moved to the preset work station, and the image acquisition group (such as a dual camera) is started to scan the target area. Based on the target detection algorithm (such as YOLO or template matching), the existence of the bolt is recognized. If the bolt is detected, a three-dimensional motion path of the first screwing head and the second screwing head is generated according to its position coordinates, and the multi-axis bracket is driven to move to the docking position, and the nut and the screw are initially centered by the synchronous rotation of the dual screwing heads. If the bolt is not detected, it moves horizontally or vertically according to the preset step length (a rasterized search path can be used) until the bolt is found or an unrecognized signal is triggered. After the centering is completed, for example, the first screwing head keeps rotating to lock the nut, and the second screwing head dynamically adjusts the rotation speed according to the preset torque / angle threshold until the threshold is reached to trigger shutdown, ensuring the consistency of the pre-tightening force. This method shortens the single-bolt tightening time through the cooperation of visual positioning error compensation and path search logic, and solves the problems of low efficiency of traditional manual operations and path dependence on experience.

[0025] In some embodiments, the method for tightening high-strength bolts of steel box girders based on machine vision positioning further includes the following operations:

[0026] S100. Use the target detection algorithm to identify whether there is a bolt in the target area. If there is, obtain the local image of the bolt group where the bolt is located;

[0027] S200. Perform feature point matching and homography matrix calculation on adjacent local images above, splice and generate a panoramic view of the bolt group, and establish a global physical coordinate system;

[0028] S300. Perform edge detection and Hough circle transformation on the panoramic view above to identify the center positions of all bolts in the panoramic view to obtain the pixel coordinates of the bolt centers;

[0029] S400. Convert the above pixel coordinates of the bolt centers into global physical coordinates to generate a bolt position set;

[0030] S500. Calculate the convex hull outer contour of the bolt group according to the above bolt position set, and determine the geometric center of the outer contour based on the centroid method or the minimum bounding rectangle method;

[0031] S600. Generate a tightening path from the inside out based on the geometric center of the outer contour, and the tightening assembly tightens the bolts in a single global physical coordinate system according to the tightening path.

[0032] In some embodiments, the tightening assembly tightening the bolts in a single global physical coordinate system according to the tightening path further includes: screening the bolt nodes on or near the tightening path by a spatial proximity threshold to generate a bolt tightening priority sequence, and the tightening assembly tightens the bolts according to the priority sequence.

[0033] In some embodiments, the method for applying torque to high-strength bolts of steel box girders based on machine vision positioning further includes selecting a spiral path or a radial path based on the symmetry index of the bolt group and the evaluation result of the contour complexity, including:

[0034] Calculate the ratio R of the convex hull area to the minimum bounding rectangle area of the bolt group and the local density variation coefficient CV;

[0035] If R < R th or CV > CV th , generate a spiral path; R th is a preset area ratio threshold, and CV th is a preset local density variation coefficient threshold;

[0036] When R ≥ R th and CV ≤ CV th , obtain the symmetry S of the convex hull outer contour and the radial error σr of the bolt group;

[0037] If S ≤ S th and σr < σ th , generate a radial path; σ th is the radial error threshold; S th is the preset symmetry threshold of the convex hull outer contour;

[0038] If S ≤ S th and σr ≥ σ th , generate a spiral path;

[0039] If S > S th , generate a radial or path;

[0040] Among them, the radial path is to apply torque layer by layer from the geometric center outward in concentric circles; the spiral path is to apply torque continuously in ascending order of the distance from the geometric center.

[0041] In some embodiments, the tightening assembly tightens each bolt one by one in a concentric circle stratification order from the geometric center outwards, or the tightening assembly tightens each bolt one by one in an ascending order of the distance from the center, including: driving the tightening assembly to move to the target bolt position along a planned path, detecting the tightening torque in real time through a torque sensor, and performing closed-loop control with a preset threshold; recording the final tightening torque and coordinate data of each bolt, and sending a warning message and marking the re-inspection site when the tolerance is exceeded.

[0042] In some embodiments, the above-mentioned tightening assembly includes a first rotation mode and a second rotation mode; when the first wrench head and the second wrench head move to the position to be docked with the target bolt after obtaining the image information through the image acquisition group, the tightening assembly enters the first rotation mode. In this first rotation mode, the self-rotation speed of the first wrench head is v m1 , and the self-rotation speed of the second wrench head is v g1 , and the torque of the high-speed wrench head is T s ; when the wrench head rotates, the radial force fluctuation is detected through a torque sensor. If the fluctuation value < the preset fluctuation value, it is determined that the centering is completed; when the centering is completed, the tightening assembly switches to the second rotation mode. In this second rotation mode, the self-rotation speed of the first wrench head is v m2 , and the self-rotation speed of the second wrench head is v g2 , and the torque of the high-torque wrench head in this second rotation mode is T b , where v m2 > v m1 , v g2 > v g1 , T b > T s .

[0043] In some embodiments, the distances from at least two ultrasonic rangefinders on the tightening frame to the surfaces on both sides of the steel box girder are detected. If the difference between the distances measured by the ultrasonic rangefinders on the same side is less than the preset threshold, the tightening assembly enters the first rotation mode to start docking the bolts; if the difference between the distances measured by the ultrasonic rangefinders on the same side is greater than the preset threshold, a warning message is sent and a request to execute tightening is generated, and the bolt position coordinate data in the current area is recorded.

[0044] In some embodiments, before the tightening assembly enters the first rotation mode, the nut position of the bolt is identified by the above-mentioned first camera and mapped to the second camera coordinate system to generate a theoretical projection point; the position of the screw head is actually detected by the above-mentioned second camera, and the deviation distance from the above-mentioned theoretical projection point is calculated; if the deviation distance exceeds the preset threshold, it is determined that the bolt is installed obliquely, the coordinate information is recorded, and a warning message is issued and a request for whether to perform tightening is generated. Through this solution, automatic determination replaces manual visual inspection, shortening the single-bolt detection time, solving the pain points of low efficiency and high missed inspection rate in traditional manual visual inspection, and enabling the risk of oblique installation of the bolt to be judged again before bolt tightening (checking for obliqueness during the initial tightening of high-strength bolts manually installed). BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 FIG. is a schematic structural diagram of a high-strength bolt automatic tightening machine for explaining an embodiment;

[0046] Figure 2 FIG. is a schematic structural diagram of a high-strength bolt automatic tightening machine for explaining an embodiment;

[0047] Figure 3 FIG. is a schematic tightening sequence diagram for explaining the tightening method in an embodiment when using a radiation or concentric circle sequence operation;

[0048] Figure 4 FIG. is a schematic tightening sequence diagram for explaining the tightening method in an embodiment when using a spiral trajectory sequence operation;

[0049] Figure 5 FIG. is a schematic diagram for explaining that the tightening method in an embodiment determines that the bolt is not oblique;

[0050] Figure 6 FIG. is a schematic diagram for explaining that the tightening method in an embodiment determines that the bolt is oblique;

[0051] Figure 7 FIG. is a schematic diagram for explaining that the tightening method in an embodiment prompts that the device posture needs to be adjusted;

[0052] 1 - High-strength bolt; 100 - Multi-axis moving support; 110 - Moving part; 120 - Multi-axis bracket; 121 - Main crossbeam; 122 - Vertical frame; 124 - Auxiliary crossbeam; 125 - Gantry; 126 - Lifting shaft; 127 - Connecting rod; 200 - Tightening assembly; 210 - Tightening frame; 211 - First screwdriver head; 212 - Second screwdriver head; 213 - Tightening seat; 214 - Linear module; 300 - Photoelectric limit switch; 410 - First camera; 420 - Second camera; 500 - Ultrasonic rangefinder; 600 - Electromagnetic suction seat; 700 - Electric push cylinder. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0053] As Figure 1 and Figure 2 , an automatic high-strength bolt tightening machine based on machine vision positioning, which is used to tighten the high-strength bolts on the steel box girder. The automatic high-strength bolt tightening machine includes a multi-axis moving support 100, a tightening assembly 200, and a control unit;

[0054] The multi-axis moving support 100 has a moving part 110 for enabling the automatic high-strength bolt tightening machine to move on the top plate of the steel box girder; the above-mentioned multi-axis moving support 100 further includes a multi-axis bracket 120 arranged on the moving part 110; the tightening assembly 200 is used to dock or clamp the high-strength bolts on the steel box girder, and a power part configured according to the tightening assembly 200 drives the high-strength bolts to rotate; the tightening assembly 200 is arranged on the above-mentioned multi-axis bracket 120, and a driving assembly is configured on the multi-axis bracket 120 for driving the tightening assembly 200 to be movable on multiple spatial axes; the control unit is connected to the tightening assembly 200 and the multi-axis moving support 100 for controlling the tightening assembly 200 and the multi-axis moving support 100 according to input instructions; wherein, the above-mentioned tightening assembly 200 includes a tightening frame 210 for clamping the steel box girder from both sides of the steel box girder and an image acquisition group for identifying the bolt position, and the control unit is connected to the image acquisition group for generating a three-dimensional motion path through the bolt position or pose data collected by the image acquisition group, and the tightening assembly 200 sequentially performs bolt tightening work according to the three-dimensional motion path; both ends of the clamping of the above-mentioned tightening frame 210 are respectively provided with a first screwdriver head 211 for adapting and fixing or screwing the nut of the high-strength bolt and a second screwdriver head 212 for adapting and fixing or screwing the high-strength screw rod.

[0055] This multi-axis moving support 100 is the core support and moving platform of the automatic high-strength bolt tightening machine, and its function is to enable the entire machine to move and position flexibly and accurately on the top plate of the steel box girder. This multi-axis moving support 100 can be an XYZ three-axis motion platform, driven by a servo motor and a ball screw to achieve high-precision three-dimensional movement, or it can be a robot arm structure with a multi-joint design, providing a larger moving range and flexibility. As long as it can ensure that the machine accurately reaches the specified position on the top plate of the steel box girder.

[0056] The above-mentioned moving part 110 can be a part of the multi-axis moving support 100, responsible for driving the physical displacement of the entire machine on the top plate of the steel box girder to ensure that the entire device can move smoothly on the top surface of the steel box girder. Here, the moving part 110 can be a wheeled chassis equipped with a motor and rubber wheels, moving on the top plate of the steel box girder by electric drive, or a suspended moving system, sliding on the top plate of the steel box girder fixed by a slide rail or a guide rail, etc.

[0057] The multi-axis support 120 can be used as a part of the multi-axis mobile support 100 to support the tightening assembly 200 and provide multi-axis motion capability. It can be a gantry 125 structure to provide linear motion capability of the X, Y, and Z axes. As long as it can support the tightening assembly 200 and achieve precise movement, it can also be a multi-joint robotic arm, which can drive the tightening assembly 200 to move in three-dimensional space through joint motors and flexibly adjust the posture.

[0058] The above-mentioned first screw head 211 and second screw head 212 are devices that directly interact with high-strength bolts and are responsible for docking, clamping and rotating the bolts. The tightening action is completed by the drive of the power unit. Its appearance can be an electric wrench that rotates the bolts through a motor drive, or it can be a pneumatic tightening tool that extends in the radial direction like a clamp and uses compressed air to provide high torque output. It can also use smart wrenches and socket wrenches with force feedback, etc., which adapt to the shape and size of the nut and automatically adjust the torque and speed through sensors.

[0059] The tightening frame 210 is designed as a frame mechanical structure, on which an electric push cylinder 700 or a hydraulic cylinder, etc. can be configured to drive the first screwing head 211 and the second screwing head 212 to move in the axial direction.

[0060] The control unit can be an industrial computer running dedicated control software to process input instructions and sensor data, or a PLC (programmable logic controller) or embedded system, as long as it can meet the real-time control requirements. The image acquisition group can include multiple high-definition cameras installed on the side of the tightening frame 210 facing the steel box girder to capture the two-dimensional image of the bolt, or a laser scanner to generate three-dimensional point cloud data of the bolt, and an infrared sensor can be configured to assist in identifying the bolt in a low-light environment, as long as it can adapt to different environmental conditions.

[0061] The above-mentioned power unit can be a servo motor, a pneumatic motor or a hydraulic system, as long as it can provide precise rotational force. The driving assembly is used to drive the tightening assembly 200 to move on multiple spatial axes to ensure that it can reach the bolts at different positions on the steel box girder. It can include a linear motor and a guide rail to achieve the linear movement of the tightening assembly 200. The three-dimensional motion path in this scheme is a motion trajectory generated by the control unit based on the bolt position or posture data collected by the image acquisition group, which is used to guide the movement of the tightening assembly 200. It can be generated by an algorithm. The three-dimensional motion path can also be a pre-programmed range channel trajectory, which is set according to the bolt distribution law. The bolts close to or covering the range are tightened in sequence. The path can be dynamically adjusted through real-time visual feedback to adapt to changes in the bolt position.

[0062] This embodiment proposes an automatic tightening machine for high-strength bolts based on machine vision positioning. Through the collaborative design of a multi-axis moving bracket and an advanced machine vision system, three-dimensional spatial precise positioning and efficient automated tightening operations of high-strength bolts for steel box girders are achieved.

[0063] The multi-axis moving support 100 can adopt an X / Y-axis motion platform driven by a servo motor, which can achieve high-precision two-dimensional planar movement on the surface of the steel box girder top plate, ensuring the flexible deployment of the tightening machine on complex working surfaces. At the same time, the multi-axis bracket 120 provides multi-degree-of-freedom (X / Y / Z-axis) motion capabilities in three-dimensional space through its carefully designed structure (such as the combination of the main cross beam 121, vertical support frames, and lifting shaft 126), enabling the tightening assembly 200 to dynamically adjust its posture according to the spatial distribution of the bolt group to meet the requirements of precise alignment. The tightening assembly 200 adopts an innovative dual-station design. Among them, the first wrench head 211 adapts to and fixes or turns the nut end of the high-strength bolt, and the second wrench head 212 is for the screw end. The two are respectively installed on both sides of the gantry 125-type lifting shaft 126 to form a stable clamping and tightening system. In addition, the image acquisition group equipped in the tightening assembly 200 is arranged in a symmetric layout, synchronously collecting the spatial position data of the bolt nut and screw through high-resolution vision sensors, and generating an accurate three-dimensional motion path in combination with image processing algorithms, thereby guiding the tightening assembly 200 to complete high-precision centering operations. This structural design effectively overcomes the limitations of the traditional manual positioning method, such as low efficiency, large errors, and difficult-to-guarantee consistency, provides a solid rigid support foundation for the automated tightening of high-strength bolts for steel box girders, and significantly improves the construction efficiency at the same time.

[0064] Embodiment Two:

[0065] Based on the above Embodiment One, the above multi-axis bracket 120 includes a main cross beam 121 and a vertical frame 122; the main cross beam 121 is arranged on the moving part 110; the vertical frame 122 is movably arranged on the main cross beam 121, and the tightening assembly 200 is movably arranged on the vertical frame 122; two photoelectric limit switches 300 are arranged on the main cross beam 121 on both sides of the vertical frame 122, used to respectively detect the distances between the vertical frame 122 and the two photoelectric limit switches 300;

[0066] The above vertical frame 122 includes a secondary cross beam 124 and a gantry 125; the secondary cross beam 124 is movably arranged on the main cross beam 121, and is used to selectively move along the length direction of the main cross beam 121; the gantry 125 includes lifting shafts 126 respectively vertically movably connected to both ends of the secondary cross beam 124, and the upper ends of the two lifting shafts 126 are connected by a connecting rod 127 to form a door frame structure, and the above tightening assembly 200 is arranged at the lower end of the lifting shaft 126; the first wrench head 211 and the second wrench head 212 are respectively located at the lower ends of the two lifting shafts 126 and are arranged oppositely.

[0067] Through the designed multi-stage motion mechanism and detection system, the present invention significantly improves the positioning accuracy and operation adaptability. The photoelectric limit switch 300 configured on the main crossbeam 121 can monitor the moving range of the vertical frame 122 in real time, accurately detect the distance between the vertical frame 122 and the limit switch through highly sensitive photoelectric induction technology, ensure that the positioning accuracy is controlled within millimeters, thus effectively avoiding over-travel phenomena and providing a reliable calibration basis for the system. At the same time, through the coordinated combination of the secondary crossbeam 124 and the lifting shaft 126 in the gantry 125 structure, the height adaptive adjustment of the tightening assembly 200 in the Z-axis direction is realized, and its motion range is sufficient to cover the diverse requirements of complex working surfaces such as the web and diaphragm of the steel box girder.

[0068] In terms of structural design, this solution adopts a multi-stage motion system composed of the main crossbeam 121, the vertical frame 122 and the gantry 125, realizing the functions of high-efficiency positioning and synchronous tightening in three-dimensional space. The main crossbeam 121 can be welded with high-strength steel (such as Q345B steel) or made of lightweight and high-rigidity aluminum alloy profiles (such as 6061-T6) through precision machining. It is fixed on the moving part 110 to provide stable horizontal support for the entire system. The vertical frame 122 realizes horizontal movement along the length direction (X-axis) of the main crossbeam 121 through the secondary crossbeam 124. The secondary crossbeam 124 can be equipped with a ball screw drive mechanism or a linear motor drive to ensure smooth and high-precision displacement control. In addition, the lifting shafts 126 on both sides of the vertical frame 122 realize vertical movement (Z-axis) through a gear-rack system or a hydraulic cylinder driven by a servo motor, forming a motion ability with two degrees of freedom in the X / Z directions. The gantry 125 structure consists of the upper ends of two lifting shafts 126 forming an integral door frame through high-strength connecting rods 127 (such as carbon steel welded parts or integrally formed aluminum alloy rods), ensuring the synchronism and rigidity of the two sides of the lifting shafts 126. The first wrench head 211 (fitting the nut end) and the second wrench head 212 (fitting the screw end) are respectively installed at the lower ends, and bilateral clamping tightening operations are realized through symmetric layout. The photoelectric limit switches 300 on both sides of the main crossbeam 121 adopt industrial-grade photoelectric sensors (such as infrared or laser types) to monitor the dynamic position of the vertical frame 122 in real time, and the detection accuracy can reach ±0.1 mm, effectively preventing over-travel of the motion and optimizing positioning calibration.

[0069] The overall structure of this solution adopts a modular design. For example, the first wrench head 211 and the second wrench head 212 can be designed as quickly replaceable tool heads to support the adaptation requirements of different specifications of bolts (such as M20, M24); the rigid connection of the gantry 125 is fixed by high-strength bolts or pins to ensure structural stability. At the same time, the combination of the horizontal drive of the main crossbeam 121 and the vertical drive of the lifting shaft 126 can flexibly meet the requirements of the bolt group distribution on the complex curved surface of the steel box girder (such as the junction of the web and the diaphragm). In this way, not only the full coverage of the working surface is realized, but also the tightening efficiency is significantly improved through the high-rigidity structure and precise motion control.

[0070] Embodiment 3:

[0071] Based on the above embodiments, the tightening frame 210 includes a tightening seat 213, a linear module 214, and a torque sensor; the tightening seat 213 is used to support the second screwing head 212 or the first screwing head 211; the tightening seat 213 is arranged on the linear module 214 and is used to move on the linear module 214 selectively; the torque sensor is arranged on the tightening seat 213, and the torque sensor is connected to the second screwing head 212 and is used to collect the torque data of the second screwing head 212; the torque sensor is connected to the control unit, and the control unit is used to receive the dynamic signal of the torque sensor in real time and adjust the output parameters.

[0072] In this embodiment, the tightening seat 213 is used as the core load-bearing component, and precise displacement control along the preset path is realized relying on the linear module 214. The linear module 214 can adopt a ball screw mechanism driven by a servo motor, and its high-rigidity transmission characteristics ensure the stable movement of the tightening seat 213 in the horizontal or vertical direction, so as to accurately adjust the relative position between the second screwing head 212 and the target bolt; or an optional slide rail system driven by a linear motor can be configured to provide a higher dynamic response speed and positioning accuracy, up to ±0.01 mm. At the same time, the torque sensor is directly embedded in the force-bearing node of the tightening seat 213, and can collect the torque data of the second screwing head 212 during the tightening process in real time, with an accuracy of up to ±0.1 N·m. The control unit can select an industrial-grade PLC or a high-performance industrial PC, and perform real-time analysis on the collected data through the built-in high-speed signal processing algorithm, and then dynamically adjust the operating parameters of the servo motor, such as the rotation speed, torque threshold (supporting multi-level setting, such as the initial tightening of 10 N·m and the final tightening of 50 N·m), etc., to meet the tightening requirements under different working conditions.

[0073] Through the above design, the tightening frame 210 realizes excellent performance in dynamic adjustment and high-precision control of the tightening assembly 200 through the high integration of the linear module 214 and the torque sensor.

[0074] Embodiment 4:

[0075] Based on the above embodiments, the image acquisition group includes: a first camera 410 and a second camera 420; the first camera 410 is arranged inside the first screwing head 211 or on the tightening seat 213 adjacent to the first screwing head 211; the second camera 420 is arranged inside the second screwing head 212 or on the tightening seat 213 adjacent to the second screwing head 212; the above-mentioned first camera 410 and the second camera 420 are arranged opposite to each other and the imaging directions are on the same straight line in space.

[0076] The first camera 410 and the second camera 420 can be coaxially embedded, such as integrating a micro industrial camera inside the screwing head or symmetrically externally mounted, to ensure that the optical axes of the two cameras are collinear and coincide with the axis of the bolt, establishing a spatial pose reference line. Here, a ring structured light can be integrated around the camera to project a coded pattern onto the end face of the bolt, enhancing the texture information of the nut / screw through feature points (such as hexagonal corners or thread profiles). For example, using a high-precision calibration plate (checkerboard / concentric circles) simultaneously appears in the fields of view of both cameras, which can ensure feature extraction in low-contrast environments.

[0077] After the first camera 410 captures the nut image, it extracts the hexagonal contour based on Canny edge detection, locates the center of the circle through Hough transform, and establishes a local coordinate system using the SIFT feature descriptor.

[0078] The second camera 420 analyzes the thread inclination angle in combination with the current structured light stripe technology, aligns the feature points of the two views through region correlation matching, generates a disparity map and calculates the depth information. Based on the Zhang calibration method, the coordinate systems of the two cameras are mapped to the global coordinate system, and the lens distortion is eliminated using the homography matrix.

[0079] In this solution, according to the three-dimensional coordinates of the center of the nut (X 1 , Y 1 , Z 1 ) and the center of the end face of the screw (X 2 , Y 2 , Z 2 ), the bolt axis equation can be calculated, and the path parameters for the screwing head to move along the axis can be generated;

[0080] Among them, the displacement ΔL = , and the inclination angle θ = arctan[(Z 2 - Z 1 ) / ΔL].

[0081] The two cameras synchronously scan the target area, quickly lock the approximate position of the bolt based on template matching, and trigger the tightening frame 210 to move to the preset working position. The first camera 410 extracts the center of the nut, analyzes the plane normal vector in combination with the structured light stripe, and calculates the end pose of the nut (yaw angle α, pitch angle β). The second camera 420 uses the region growing algorithm to segment the thread area and fits the axis direction of the screw by the least squares method. The control unit receives the pose data at both ends, adjusts the position of the screwing head until the optical axes of the two cameras coincide with the actual axis of the bolt (error < ±0.2 mm), and completes the alignment. If the alignment time exceeds the threshold or the error is too large, it is determined that there is a bolt skew situation, a warning message is issued, and the coordinate position of the bolt is calibrated.

[0082] This camera system transmits images to the industrial control computer via the GigE Vision protocol. After being processed by the vision algorithm, motion commands are generated and sent to the servo driver via the EtherCAT bus.

[0083] The human-machine interface configured for the camera terminal can develop a visual control panel based on ROS (Robot Operating System) to display the bolt pose, tightening torque curve, and abnormal alarm information in real time. The lightweight YOLOv5s-T network is deployed on the embedded GPU.

[0084] Example Five:

[0085] Based on the above embodiment, two ultrasonic rangefinders 500 are provided on the tightening frame 210 (a total of four, two on each side), and the two ultrasonic rangefinders 500 are arranged with their orientations parallel; the ultrasonic rangefinders 500 located on both sides of the steel box girder are arranged facing each other; the measuring direction of the ultrasonic rangefinder 500 is parallel to the telescopic direction of the first screwing head 211 or the second screwing head 212. The core function of the ultrasonic rangefinder 500 is to dynamically calibrate the parallelism between the translation direction of the tightening frame 210 and the steel box girder plate surface through bilateral distance synchronous detection. Specifically, the two ultrasonic rangefinders 500 are installed facing each other on the tightening frames 210 on both sides of the steel box girder, and their measuring directions are strictly parallel to the telescopic direction of the screwing head (Z-axis). By measuring the distance difference between the two sides of the rangefinders to the steel box girder plate surface in real time (ΔS = |S 1 −S 2 |), S 1 is the real-time measured distance from the ultrasonic rangefinder 500 installed on one side of the steel box girder to the plate surface of this side, and S 2 is the real-time measured distance from the ultrasonic rangefinder 500 installed on the other side of the steel box girder to the plate surface of the opposite side. Determine whether the moving direction of the tightening frame 210 is consistent with the normal direction of the plate surface: If ΔS exceeds the preset threshold, it indicates that the tightening frame 210 is tilted or offset from the steel box girder plate surface. At this time, the control unit can immediately adjust the drive parameters of the multi-axis bracket 120 (such as X / Y-axis compensation movement or Z-axis lifting correction) to ensure that the first screwing head 211 and the second screwing head 212 are always centered with the bolt axis during the screwing process, or perform manual intervention according to the feedback warning information to verify whether it is a sudden change in the parallelism of the steel box girder plate surface or an offset of the tightening frame 210, thereby eliminating the risk of tightening failure caused by uneven plate surface or installation deviation.

[0086] Example Six:

[0087] Based on the above embodiments, an electromagnetic suction seat 600 is provided on the multi-axis bracket 120 or the moving part 110. The electromagnetic suction seat 600 is connected to the control unit and is used to perform opening and closing operations on the electromagnetic suction seat 600 according to the instructions received by the control unit. In this way, the rapid switching of the adsorption state can be realized through the control unit instructions, providing stable connection and positioning switching for the multi-axis bracket 120 or the moving part 110. Specifically, the electromagnetic suction seat 600 is integrated at the bottom of the multi-axis bracket 120 or the moving part 110. After the control unit receives the operation instruction, it triggers the electromagnet to conduct and cut off the power: when powered on, the electromagnetic suction seat 600 generates a strong magnetic field and adsorbs on the surface of the steel box girder top plate, forming an adsorption force to ensure the rigid fixation of the equipment and the workpiece during the processing; when powered off, the magnetic field disappears and the adsorption force is released, facilitating the rapid transfer of the equipment to the working station. By precisely controlling the on-off of the current, the instantaneous establishment and elimination of the adsorption force are realized, avoiding the positioning time-consuming problem of traditional mechanical clamps.

[0088] Embodiment Seven:

[0089] A method for tightening high-strength bolts of a steel box girder based on machine vision positioning includes using the high-strength bolt automatic tightening machine of the above-mentioned one aspect; the tightening method includes the following operations:

[0090] Move the tightening assembly 200 to a preset position, make the image acquisition group face the target area, and make the image acquisition group scan the structural objects of the steel box girder in the current target area to identify whether there are bolts;

[0091] a. If there are bolts, move the tightening assembly 200 according to the bolt position so that the first screwing head 211 or the second screwing head 212 of the tightening assembly 200 corresponds to the nut or the screw head of the high-strength bolt;

[0092] b. If there are no bolts, move the tightening assembly 200 horizontally or vertically according to the preset program until a bolt is identified or an unrecognized signal is sent after the number of moving steps exceeds the preset maximum number of steps;

[0093] Among them, when a bolt is identified, make both the first screwing head 211 and the second screwing head 212 rotate and move towards the nut and the screw head until the first screwing head 211 docks with the nut and the second screwing head 212 docks with the screw head;

[0094] After the docking is completed, the first screwing head 211 keeps rotating, the second screwing head 212 stops rotating, or the first screwing head 211 and the second screwing head 212 continue to keep rotating, and when the tightening process reaches the preset torque / angle threshold, trigger the tightening assembly 200 to stop tightening and lock.

[0095] This solution realizes automatic tightening control through bolt recognition guided by machine vision and adaptive path planning. Specifically, first, the tightening assembly 200 is moved to a preset station, and the image acquisition group (such as a dual camera) is started to scan the target area. Based on the target detection algorithm (such as YOLO or template matching), the existence of bolts is recognized. If a bolt is detected, a three-dimensional motion path of the first wrench head 211 and the second wrench head 212 is generated according to its position coordinates, and the multi-axis bracket 120 is driven to move to the docking position. The nut and the screw are initially centered by the synchronous rotation of the dual wrench heads. If no bolt is detected, it moves horizontally or vertically according to a preset step length (a rasterized search path can be used) until a bolt is found or an unrecognized signal is triggered. After centering, for example, the first wrench head 211 keeps rotating to lock the nut, and the second wrench head 212 dynamically adjusts the rotation speed according to the preset torque / rotation angle threshold until the threshold is reached to trigger shutdown, ensuring the consistency of the pre-tightening force. This method shortens the tightening time of a single bolt through the cooperation of visual positioning error compensation and path search logic, and solves the problems of low efficiency of traditional manual operations and path dependence on experience.

[0096] Embodiment Eight:

[0097] Based on the above Embodiment Seven, the tightening method further includes the following operations:

[0098] S100. Use the target detection algorithm to identify whether there is a bolt in the target area. If there is, obtain the local image of the bolt group where the bolt is located;

[0099] S200. Perform feature point matching and homography matrix calculation on adjacent local images above, splice and generate a panoramic image of the bolt group, and establish a global physical coordinate system;

[0100] S300. Perform edge detection and Hough circle transformation on the above panoramic image to identify the center positions of all bolts in the panoramic image to obtain the pixel coordinates of the bolt centers;

[0101] S400. Convert the above pixel coordinates of the bolt centers to global physical coordinates to generate a bolt position set;

[0102] S500. Calculate the convex hull outer contour of the above bolt group according to the above bolt position set, and determine the geometric center of the outer contour based on the centroid method or the minimum bounding rectangle method;

[0103] S600. Generate a tightening path from the inside to the outside according to the above geometric center of the outer contour, and the tightening assembly 200 tightens the bolts in a single global physical coordinate system in sequence according to the above tightening path.

[0104] The tightening assembly sequentially tightens the bolts in a single global physical coordinate system according to the above tightening path, and further includes: screening the bolt nodes on or near the tightening path according to the spatial proximity threshold to generate a bolt tightening priority sequence, and the tightening assembly tightens the bolts in sequence according to this priority sequence.

[0105] This method aims at the problems of strong manual dependence, low positioning accuracy, and unscientific path planning in the tightening operation of high-strength bolts for steel box girders. Through the object detection algorithm and image stitching technology of machine vision, the global recognition and panoramic modeling of bolt groups are realized, and the outer contour graph of the current stud group array can be completely obtained, which makes technical preparations for a more scientific and efficient tightening path. Based on the convex hull outer contour analysis and geometric center path planning, a specific tightening path from the inside to the outside is generated. In this way, not only the path is automatically planned according to the mechanical properties (from the middle to the surrounding, from the large stiffness to the small stiffness), avoiding stress concentration, but also the efficiency of the automated process is ensured, so that the re-tightening and final tightening paths are strictly consistent, reducing the risk of structural deformation, and significantly improving the reliability of the steel box girder connection and the construction standardization level.

[0106] This method also uses machine vision to perform object detection and feature matching on bolt groups. Through the Hough circle transform or deep learning model (the lightweight network YOLOv5s-tiny can be used), the center position of the bolt is accurately identified, that is, the pixel coordinates of the bolt center are obtained, and a panoramic image is stitched based on the homography matrix to establish a global coordinate system. In the path planning stage, a tightening path radiating from the center to the outside is generated by combining the convex hull algorithm and the centroid method, and at the same time, the torque closed-loop control is used to ensure the consistency of the pre-tightening force. During the tightening process, the visual servo technology is used to dynamically correct the pose of the robotic arm, and the reverse torque is offset by the torque release device. Finally, the tightening-in-place detection is integrated to form a closed-loop process of "recognition - positioning - tightening - verification", with an efficiency improvement of more than 50% compared to manual work, solving the problems of high risk of high-altitude operation and poor torque consistency in traditional processes. The camera is installed on a mobile pan-tilt head, and the scanning path is controlled by a servo motor to cover the target area on the surface of the steel box girder.

[0107] Specifically, an industrial-grade global shutter camera with a resolution ≥ 20 million pixels and a frame rate ≥ 30fps can be used, and it is equipped with a ring-shaped LED light source. The lightweight YOLOv5s-tiny model is used, and the training data set contains positive and negative samples of steel box girder bolts (the positions of nuts and screw heads are marked). After the bolts are detected, the camera is triggered to take high-resolution local images (size 1024×1024 pixels), and the overlapping rate of adjacent images ≥ 30%.

[0108] In step S200, ORB feature points are extracted from adjacent images (the number ≥ 1000 points per image), and the mismatched points are removed through the RANSAC algorithm, and the homography matrix H is solved:

[0109]

[0110] H: A homography matrix, a 3×3 projective transformation matrix, used to describe the mapping relationship between two planes. Project the point (u i , v i ) in the source image into the coordinate system of the target image to obtain (u i ′, v i ′).

[0111] Adopt a weighted fusion algorithm to eliminate the stitching seam and generate a seamless panoramic image (resolution ≥ 8000×6000 pixels).

[0112] Pre-set reference marking points on the surface of the steel box girder (fixed value taken between 100 - 500 mm), calculate the pixel - physical coordinate transformation matrix by the least squares method, and the calibration error ≤ 0.2 mm / m for global coordinate system calibration.

[0113] In step S300, perform adaptive histogram equalization on the panoramic image, use Canny edge detection (low threshold 50, high threshold 150) to extract the bolt contour. Based on the prior of the bolt diameter (such as the diameter of M24 bolt is 24 mm), set the radius range r ∈ [0.9D, 1.1D] (D represents the nominal diameter of the bolt, that is, the standard design diameter of the bolt), reduce the calculation amount, adopt gradient direction weighted Hough voting, and improve the center positioning accuracy to ±0.3 pixels to complete the optimization of the Hough circle transformation.

[0114] In step S400, perform global physical coordinate mapping, and the coordinate transformation formula is:

[0115] X i = k x *(u i − u 0 ) + X offset , Z i = k Z *(v i − v 0 ) + Z offset

[0116] k x , k Z : Pixel resolution (calculated by the calibration plate, unit mm / pixel);

[0117] (u 0 , v 0 ): Pixel coordinates of the center of the panoramic image;

[0118] X offset , Z offset : Offset of the origin of the global coordinate system.

[0119] Output result, generate a set of bolt positions {(X 1 , Z 1 ),...,(X n , Z n )}, and store it in JSON format for saving.

[0120] In step S500, calculate the outer contour and geometric center of the convex hull, including: generating the convex hull, using the two-dimensional convex hull algorithm (Andrew's Monotone Chain algorithm), with a time complexity of O(nlogn), and outputting the sequence of convex hull vertices. Calculate the convex hull area A convex and the area A rect of the minimum bounding rectangle. Then determine the geometric center:

[0121] Centroid method:

[0122]

[0123] C x and C z are the X coordinate and Z coordinate of the centroid; n is the total number of points; X i and Z i are the x coordinate and z coordinate of the i-th point.

[0124] Center of the minimum bounding rectangle:

[0125]

[0126] C rect is the center coordinate of the minimum bounding rectangle; X min and X max are the minimum and maximum values of the x coordinates among all points; Z min and Z max are the minimum and maximum values of the z coordinates among all points.

[0127] In step S600, the specific settings of the path planning strategy can be referred to as follows:

[0128] Figure 3 and Figure 4 where P represents a bolt position, and O is the bolt position where the path starts, that is, the bolt position where the geometric center of the identified bolt group contour is located or close to. Combining Figure 3 , radial path (symmetric structure): Stratify by concentric circles (layer spacing ΔR = 50 - 150 mm), sort each layer by angle, and the tightening sequence is from the inside out. Figure 3 where the dashed line M cc represents the concentric circle stratification marking line.

[0129] Path equation:

[0130]

[0131] Path is the finally generated path set, which contains the positions of all bolts to be tightened; k is the index of the concentric circle layer (k = 1, 2,..., m); m is the total number of concentric circle layers; R k is the theoretical radius of the k-th concentric circle, which can be dynamically adjusted; θ j is the angle of the j-th sector; j is the sector index (j = 0, 1,..., n sector −1); n sector is the number of sectors divided in each layer (8 sectors).

[0132] Combined with Figure 4 , the spiral path (irregular contour): arranged in ascending order of the distance from the bolt to the center, forming a continuous spiral trajectory. When tightening one by one, the bolt closest to the trajectory line on the trajectory is operated. That is, when matching and correcting the path for each bolt, each bolt is matched to the nearest target point on the spiral path. If the bolt is not on the theoretical spiral line, a continuous trajectory is formed through spline interpolation or polyline connection. Figure 4 In, M if The dashed line represents the outer contour of the identified bolt group convex hull.

[0133] Dynamic pitch adjustment:

[0134] Δr = α * r prev (α ≤ 1.2)

[0135] Torque closed-loop control: Using a servo motor + harmonic reducer, the torque control accuracy is ±1.5%;

[0136] PID adjustment formula:

[0137]

[0138] where e(t) = T set −T actual

[0139] u(t) is the output signal of the controller (such as motor torque, valve opening); K p is the proportional gain coefficient, which determines the response intensity to the current error; K i is the integral gain coefficient, which determines the correction strength for the accumulation of historical errors; K d is the differential gain coefficient, which determines the suppression ability for the error change rate; T set is the preset torque, T actual is the actual torque; e(t) = preset torque − actual torque; τ is the integral variable, representing the time history integral interval. In this way, the controller calculates the motor adjustment amount u(t) through PID and dynamically adjusts the output torque.

[0140] Embodiment Nine:

[0141] Based on the above embodiments, the method for applying high-strength bolts to steel box girders based on machine vision positioning further includes selecting a spiral path or a radial path based on the symmetry index of the bolt group and the evaluation result of the contour complexity, including:

[0142] Calculating the ratio R of the convex hull area to the minimum bounding rectangle area of the bolt group and the local density variation coefficient CV;

[0143] If R < R th or CV > CV th , generate a spiral path; R th is the preset area ratio threshold, and CV th is the preset local density variation coefficient threshold;

[0144] When R ≥ R th and CV ≤ CV th , obtain the symmetry degree S of the convex hull outer contour and the radial error σr of the bolt group;

[0145] If S ≤ S th and σr < σ th , generate a radial path; σ th is the radial error threshold; S th is the preset symmetry degree threshold of the convex hull outer contour;

[0146] If S ≤ S th and σr ≥ σ th , generate a spiral path;

[0147] If S > S th , generate a radial or path;

[0148] Among them, the radial path is to apply the torque layer by layer from the geometric center outward in concentric circles; the spiral path is to apply the torque continuously in ascending order of the distance from the geometric center.

[0149] Currently, the operation of applying high-strength bolts includes initial tightening, re-tightening, and final tightening. After the initial tightening, all the bolts at the joint are re-tightened, and the re-tightening value is the same as the initial tightening value. Similar to the initial tightening, the re-tightening needs to be carried out radially from the middle of the joint connection to the surrounding. The bolts on the same connection surface are reinforced from the middle of the joint to both ends or from the part with greater stiffness to the part with smaller stiffness.

[0150] Specifically, the traditional tightening of high-strength bolts in steel box girders usually starts from the center of the connection node and proceeds radially or in a star pattern outward, for example, gradually tightening in quadrant order. This method is common in bridge construction and aims to ensure uniform distribution of pre-tightening force and reduce stress concentration, especially in steel box girders under high torsional loads. According to current standards, the tightening process is carried out in stages, first preliminarily tightened and then the final torque is applied, expanding from the center to the edge to avoid slippage at the connection.

[0151] However, directly tightening randomly from the center outward without distinguishing the bolt group situation may lead to uneven distribution of pre-tightening force, elastic interaction between bolts, some bolts being over-tightened while others being loose, increasing the risk of stress concentration. Especially under high loads in steel box girders, it is prone to fatigue cracks or connection failures, raising safety risks in bridge applications. For example, slippage of key connections may lead to serious accidents.

[0152] In contrast, this solution uses machine vision to identify the outer contour of the bolt group, evaluates symmetry, radial error, area ratio, and density change, and generates a specific tightening path. According to symmetry conditions and radial error situations, a radial or concentric circle path, or a spiral path is adopted. Tightening is carried out from the center outward; if the area ratio is low or the density change is large, a spiral path is selected and tightened in ascending order of distance. This radial / concentric circle path is suitable for layouts with small radial errors, ensuring uniform pre-tightening force, reducing stress concentration, and being highly efficient. It is suitable for the connections of regularly distributed steel box girders, while the spiral path is suitable for complex layouts with large radial errors, adapting to scattered distributions, reducing the risk of deformation, and dynamically adjusting to improve construction flexibility.

[0153] This method ensures uniform pre-tightening force, reduces stress concentration, and reduces the risk of deformation. The technical effects include an approximately 50% increase in efficiency, a reduction in safety risks during high-altitude operations, and ensuring consistency through torque closed-loop control. This can significantly improve the reliability and standardization level of steel box girder connections.

[0154] This solution sets multiple judgment logics.

[0155] When the symmetry < threshold and the radial error < threshold, the path adopted is a radial or multi-layer concentric circle path. That is, low symmetry indicates an irregular bolt group shape, but a small radial error means that the bolts are relatively evenly distributed radially. The radial path is suitable for this regular distribution, ensuring uniform pre-tightening force and reducing stress concentration, especially suitable for the stability requirements of steel box girder connection points.

[0156] When the symmetry < threshold and the radial error ≥ threshold, a spiral path is adopted. Low symmetry and large radial error mean that the bolt distribution is irregular and scattered. The radial path may lead to uneven pre-tightening force due to its inability to adapt to the scattered layout, while the spiral path tightens in ascending order of the distance from the center, which can gradually apply the load and reduce the interference of early tightening on subsequent bolts.

[0157] When the symmetry degree > threshold value and the radial error < threshold value, a radial or multi-layer concentric circle path is adopted. A high symmetry indicates a regular shape of the bolt group, and a small radial error means a uniform distribution. This situation is closest to the ideal layout. The radial path is tightened layer by layer from the center outwards, which can effectively apply the load and maintain the structural stability.

[0158] When the symmetry degree > threshold value and the radial error ≥ threshold value, a spiral path is adopted. The symmetry is high but the radial error is large. Although the bolt distribution is symmetric, it is scattered, which may be caused by installation errors or design deviations. The radial path may fail due to the scattered distribution, while the spiral path is tightened in the order of distance, which can adapt to this deviation, gradually apply the load, and prevent local stress concentration.

[0159] Specifically as follows in the table:

[0160] During specific implementation, it can be carried out with reference to the following method:

[0161] Set parameters:

[0162] Symmetry degree S (range 0 - 1, the smaller the value, the more symmetric);

[0163] Radial error σ r (Degree of dispersion of the bolt from the theoretical layer);

[0164] Convex hull area ratio R = A convex / A rect ;

[0165] Coefficient of variation of density CV (standard deviation of local density / mean value).

[0166] Path selection rules:

[0167] Rule 1: If S ≤ S th and σ r <σ th , adopt a radial path;

[0168] Rule 2: If S ≤ St h but σ r ≥σ th , adopt a spiral path;

[0169] Rule 3: If S > S th but R < R th or CV > CV th , adopt a spiral path;

[0170] Rule 4: In other cases, the default is a radial path.

[0171] Threshold example: S th=0.1, σ th =0.05R k , R th =0.7, CV th =0.3

[0172] Calculation of symmetry degree S. S measures the uniformity of the angular distribution of the bolt group, and the formula is expressed as:

[0173]

[0174] m: Number of polar coordinate sectors (can be taken as 8);

[0175] Nj: Number of bolts in the j-th sector;

[0176] n: Total number of bolts.

[0177] Calculation of radial error σr, which is the degree of dispersion of the bolts from the theoretical layer radius. The formula is expressed as:

[0178]

[0179] R k : Theoretical radius of the k-th layer (e.g., Rk = k * 150mm);

[0180] r i : Actual distance from the bolt to the center, ; C x ,C z (is the center coordinate of the bolt group)

[0181] n k : Total number of bolts in the k-th layer. For example, if there are 8 bolts in the 2nd layer, then n k = 8.

[0182] Calculation of convex hull area ratio R, the formula is:

[0183] R = A convex / A rect

[0184] A convex : Convex hull area (calculated by Graham scan method);

[0185] A rect : Area of the minimum bounding rectangle (calculated by rotating calipers algorithm).

[0186] Calculation of coefficient of variation of density CV, the formula is:

[0187]

[0188] ρ ij : Bolt density of the grid cell;

[0189] : Average density.

[0190] Radial path steps:

[0191] For example, with the geometric center C as the origin, concentric circles are divided by ΔR = 150 mm; sectors are divided, and each layer is divided into 8 sectors by θ j = j * 2π / 8 (j = 0, 1,..., 7);

[0192] The bolts in each layer are arranged in ascending order of angle, and the tightening sequence is carried out layer by layer from the inside to the outside.

[0193] Path equation:

[0194]

[0195] Spiral path steps: Calculate the distance di of the bolt to the center = ; Arrange in ascending order of di, and the adjacent bolt spacing is adjusted adaptively (Δr ≤ 1.2d prev ); Generate a continuous spiral trajectory to avoid sudden stops and turns of the robotic arm.

[0196] Path equation:

[0197] Path = {(X i , Z i ) ∣ d1 ≤ d2 ≤ ⋯ ≤ dn};

[0198] For example, parameter calculation: S = 0.08 (symmetric), σ r = 2.1 mm (theoretical layer R k = 300 mm, σ th = 15 mm); R = 0.62, CV = 0.45, then the spiral path is selected. The total length of the spiral path is 12.5 m, which is shorter than the radial path (14.2 m).

[0199] In this embodiment, the tightening assembly 200 tightens each bolt one by one in the order of concentric circles from the geometric center outwards, or the tightening assembly 200 moves in ascending order of the distance from the center to tighten each bolt one by one, including: driving the tightening assembly 200 to move to the target bolt position according to the planned path, detecting the tightening torque in real time through a torque sensor, and performing closed-loop control with a preset threshold; recording the final tightening torque and coordinate data of each bolt, and sending a warning message and marking the re-inspection site when it exceeds the tolerance.

[0200] By adopting a scientific tightening sequence of concentric circle stratification or ascending distance, combined with the real-time torque detection of the torque sensor and the closed-loop control system, it ensures that the pre-tightening force of each bolt precisely reaches the design requirements, thereby significantly improving the assembly quality of complex structures such as steel box girders. This method optimizes the stress distribution, reduces the risk of local stress concentration and structural deformation, and at the same time reduces human errors through automated operations, greatly improving the assembly efficiency. In addition, this solution also effectively reduces the risk of bolt loosening and meets the structural requirements of complex geometric shapes; the system records the final tightening torque and coordinate data of each bolt, issues a warning when out of tolerance and marks the re-inspection site, reducing the rework cost and providing traceability.

[0201] This embodiment can be further optimized. The above-mentioned tightening assembly 200 includes a first rotation mode and a second rotation mode;

[0202] After the first screwdriver head 211 and the second screwdriver head 212 move to the target bolt waiting-to-be-docked position after obtaining the image information through the image acquisition group, the tightening assembly 200 enters the first rotation mode. In this first rotation mode, the self-rotation speed of the first screwdriver head 211 is v m1 , and the self-rotation speed of the second screwdriver head 212 is v g1 , and the torque of the high-speed screwdriver head is T s ; when the screwdriver head rotates, the radial force fluctuation is detected through the torque sensor. If the fluctuation value < the preset fluctuation value, it is determined that the centering is completed;

[0203] After the centering is completed, the tightening assembly 200 switches to the second rotation mode. In this second rotation mode, the self-rotation speed of the first screwdriver head 211 is v m2 , and the self-rotation speed of the second screwdriver head 212 is v g2 , and the torque of the high-torque screwdriver head in this second rotation mode is T b , where v m2 > v m1 , v g2 > v g1 , T b > T s .

[0204] This ensures the assembly quality of the steel box girder. That is, in the first rotation mode, when the screwdriver head rotates, the radial force fluctuation is detected in real time through the torque sensor. When the fluctuation value is less than the preset value, it is determined that the bolt centering is completed. This process uses low-speed rotation and force feedback to ensure that the bolt and the screwdriver head can be reliably and precisely docked while reducing the intensity of rigid collision, greatly reducing the risk of assembly failure or structural damage caused by inaccurate centering. After the centering is completed, the system automatically switches to the second rotation mode to achieve fast and efficient tightening, greatly improving the assembly efficiency.

[0205] At least two ultrasonic rangefinders 500 on the tightening frame 210 detect the distances from the ultrasonic rangefinders 500 to the surfaces on both sides of the steel box girder. If the difference between the distances measured by the ultrasonic rangefinders 500 on the same side is less than a preset threshold, the tightening assembly 200 enters the first rotation mode to start docking the bolts; if the difference between the distances measured by the ultrasonic rangefinders 500 on the same side is greater than the preset threshold, a warning message is sent, a request for whether to perform tightening is generated, and the bolt position coordinate data in the current area is recorded.

[0206] By detecting the distances to the surfaces on both sides of the steel box girder through at least two ultrasonic rangefinders 500 on the tightening frame 210, it is ensured that the surface distances on both sides of the steel box girder before docking the bolts meet the preset standards. And if the difference between the distances measured by the ultrasonic rangefinders 500 on the same side is less than the preset threshold, the tightening assembly 200 will enter the first rotation mode to start docking the bolts, improving the accuracy and safety of docking the bolts and effectively avoiding assembly failures and structural damages caused by distance deviations.

[0207] Combined Figure 5 with Figure 6 , before the tightening assembly 200 enters the first rotation mode, the nut position of the bolt is identified by the above-mentioned first camera 410 and mapped to the coordinate system of the second camera 420 to generate a theoretical projection point; the position of the screw head is actually detected by the above-mentioned second camera 420, and the deviation distance from the theoretical projection point is calculated; if the deviation distance exceeds the preset threshold, it is determined that the bolt is installed skew, the coordinate information is recorded, and at the same time a warning message is sent and a request for whether to perform tightening is generated. Through this solution, automated determination replaces manual visual inspection, shortening the single-bolt detection time, solving the pain points of low efficiency and high missed inspection rate of traditional manual visual inspection, and enabling the skew risk of bolt installation to be judged again before bolt tightening (checking for skew during the initial tightening of high-strength bolts installed manually).

[0208] Specifically, the following methods can be used for deployment:

[0209] The optical axes of the two cameras are strictly collinear through precise mechanical calibration (deviation ≤ 0.05°) to form a spatial straight-line reference. The cameras are encapsulated with shock and dust protection, the lens surfaces are coated with anti-reflection films, and the installation brackets are reserved with an angle adjustment range of ±5° to be compatible with different bolt inclinations. A high-precision calibration board (checkerboard / concentric circles) appears in the fields of view of both cameras at the same time;

[0210] The internal parameters (focal length, distortion coefficient) and external parameters (relative pose) of the two cameras are obtained through the Zhang Zhengyou calibration method; a unified global coordinate system is established to ensure that the coordinate system conversion error between the two cameras ≤ 0.1 mm.

[0211] Trigger the synchronous image acquisition of the dual cameras using a hardware signal (such as GPIO), with a time difference ≤ 1 ms to avoid motion blur and dynamic errors. After the first camera 410 acquires an image, the nut contour is recognized through the YOLOv5 model, and the central point pixel coordinates (u 1 , v 1 ) are extracted. After the second camera 420 acquires an image, the center coordinates of the screw head are located using Canny edge detection + Hough circle transformation, (u 2 , v 2 ). Convert (u 1 , v 1 ) to the coordinate system of the second camera 420 to generate the theoretical projection point (u 2 ′, v 2 ′). The formula is:

[0212]

[0213] where H is the homography matrix, which is pre-calculated during the calibration phase. Deviation calculation and decision: Calculate the Euclidean distance d = between the actual screw head position and the theoretical projection point. If d > the threshold (such as 8%, 10% of the bolt diameter, etc.), it is determined that the bolt is installed skewed.

[0214] Based on the pose data of the dual cameras, generate the three-dimensional motion trajectory of the screwdriver head. Starting from the center of the nut and ending at the center of the screw head, the path equation can be expressed as:

[0215] P(t) = P n + t * (P b - P n )(t ∈ [0, 1])

[0216] where Pn and Pb are the three-dimensional coordinates of the nut and the screw head respectively.

[0217] If a deviation is detected, control the electric push cylinder 700 to drive the screwdriver head to adjust the displacement in the normal direction:

[0218] Δx = k * (u 2 ′ - u 2 ), Δz = k * (v 2 ′ - v 2 ), where k is the proportionality coefficient (such as set to 0.8) until d ≤ the threshold.

[0219] For example, for the installation of M30 bolts on a steel box girder (diameter 30 mm, allowable deviation ≤ 2.4 mm), after the dual cameras are calibrated, the measured coordinate system conversion error is 0.08 mm; the screw head is manually offset by 3 mm, and the system detects d = 3.1 mm, triggers an alarm and generates an instruction on whether to perform manual operation, and marks the bolt.

[0220] Combination Figure 7 This solution can distinguish between bolt skew and overall tightening machine skew by fusing ultrasonic ranging data with visual baseline deviation:

[0221] If the ultrasonic distance measurement difference ΔS>ΔS th , but the actual position of the bolt deviates from the reference line by d≤d th . The moving direction of the tightening frame is not parallel to the steel box girder surface, and a warning message is issued to adjust the multi-axis bracket posture. When correcting, the deflection angle θ=arctan(ΔS / L) can be calculated according to ΔS (L is the distance between the distance meters); the X / Y axis of the multi-axis bracket is driven to move in compensation or the Z axis is lifted to restore parallelism.

[0222] If the bolt position deviation d>d th , but the ultrasonic distance measurement difference ΔS≤ΔS th , the bolt itself is installed crookedly, which has nothing to do with the tightening machine posture. Mark the bolt coordinates and generate a maintenance work order;

[0223] If ΔS>ΔS th And d>d th , then the tightening machine is skewed and the bolt is skewed at the same time, or the bolt is installed correctly but the tightening machine is skewed as a whole. When correcting, calibrate the tightening machine posture first (restore ΔS≤ΔS th ); Re-measure the bolt position. If it is still out of tolerance, it is determined that the bolt is skewed.

Claims

1. High-strength bolt automatic tightening machine based on machine vision positioning, characterized by: The high-strength bolt automatic tightening machine is used to tighten the high-strength bolts on the steel box girder. The high-strength bolt automatic tightening machine includes: A multi-axis movable support, the multi-axis movable support having a movable part for enabling the high-strength bolt automatic tightening machine to move on the top plate of the steel box beam; the multi-axis movable support also includes a multi-axis bracket arranged on the movable part; A tightening assembly, which is used to dock or clamp the high-strength bolts on the steel box girder, and is used to drive the high-strength bolts to rotate according to the power unit configured by the tightening assembly; the tightening assembly is arranged on the multi-axis bracket, and the multi-axis bracket is equipped with a driving assembly, which is used to drive the tightening assembly to move on multiple spatial axes; A control unit, which is connected to the tightening assembly and the multi-axis movable support and is used to control the tightening assembly and the multi-axis movable support according to input instructions; Among them, the tightening assembly includes a tightening frame for clamping the steel box girder from both sides of the steel box girder and an image acquisition group for identifying the bolt position. The control unit is connected to the image acquisition group and is used to generate a three-dimensional motion path through the bolt position or posture data collected by the image acquisition group. The tightening assembly tightens the high-strength bolts on the steel box girder in sequence according to the three-dimensional motion path; the two ends of the tightening frame are respectively provided with a first screw head for adapting and fixing or tightening the high-strength bolt nut and a second screw head for adapting and fixing or tightening the high-strength screw.

2. The high-strength bolt automatic tightening machine based on machine vision positioning according to claim 1 is characterized in that: The multi-axis support comprises: A main beam, which is arranged on the moving part; A vertical frame, the vertical frame is movably arranged on the crossbeam, and the tightening assembly is movably arranged on the vertical frame; The main crossbeam is provided with two photoelectric limit switches located on both sides of the vertical frame, which are used to respectively detect the distance between the vertical frame and the two photoelectric limit switches; The vertical frame comprises: A secondary crossbeam, the secondary crossbeam is movably arranged on the main crossbeam, and is used to selectively move along the length direction of the main crossbeam; The door frame includes lifting shafts which are respectively connected to the two ends of the secondary crossbeam for vertical movement, the upper ends of the two lifting shafts are connected by connecting rods to form a door frame structure, and the tightening assembly is arranged at the lower end of the lifting shaft; the first screw head and the second screw head are respectively located at the lower ends of the two lifting shafts and are arranged opposite to each other.

3. The high-strength bolt automatic tightening machine based on machine vision positioning according to claim 1 or 2, characterized in that: The tightening frame comprises: A tightening seat, the tightening seat is used to support the second screw head or the first screw head; A linear module, wherein the tightening seat is arranged on the linear module and is used for selectively moving on the linear module; A torque sensor is arranged on the tightening seat, and the torque sensor is connected to the second screwing head for collecting torque data of the second screwing head; the torque sensor is connected to a control unit, and the control unit is used to receive the dynamic signal of the torque sensor in real time and adjust the output parameters.

4. The high-strength bolt automatic tightening machine based on machine vision positioning according to claim 3 is characterized in that: The image acquisition group includes: A first camera, which is disposed in the first screw head or on a tightening seat adjacent to the first screw head; A second camera, which is disposed in the second screwing head or on a tightening seat adjacent to the second screwing head; The first camera and the second camera are arranged opposite to each other and their shooting directions are located on the same straight line in space.

5. A method for tightening high-strength bolts of steel box girders based on machine vision positioning, characterized in that: The invention comprises using the automatic high-strength bolt tightening machine as described in any one of claims 1 to 4; the tightening method comprises the following operations: Move the tightening assembly to a preset position, direct the image acquisition group toward the target area, and enable the image acquisition group to scan the structural object of the steel box girder in the current target area to identify whether there are bolts; a. If there is a bolt, then according to the position of the bolt, move the tightening assembly so that the first screw head or the second screw head of the tightening assembly corresponds to the nut or screw head of the high-strength bolt; b. If there is no bolt, the tightening component is moved horizontally or vertically according to the preset program until a bolt is identified or the number of moving steps exceeds the preset maximum number of steps and a non-identification signal is issued; Wherein, after the bolt is identified, the first screwing head and the second screwing head are both rotated and moved toward the nut and the screw head until the first screwing head is connected to the nut and the second screwing head is connected to the screw head; After the docking is completed, the first screw head keeps rotating and the second screw head stops rotating, or the first screw head and the second screw head continue to rotate. When the tightening process reaches a preset torque / angle threshold, the tightening assembly is triggered to stop tightening and lock.

6. The method for tightening high-strength bolts for steel box girders based on machine vision positioning according to claim 5, characterized in that: The following operations are also included: S100. Using a target detection algorithm to identify whether there is a bolt in the target area, and if so, obtaining a local image of the bolt group where the bolt is located; S200. Perform feature point matching and homography matrix calculation on adjacent local images, stitch them together to generate a panoramic image of the bolt group, and establish a global physical coordinate system; S300. Perform edge detection and Hough circle transform on the panoramic image, identify the center positions of all bolts in the panoramic image to obtain pixel coordinates of the center of the bolts; S400. Convert the pixel coordinates of the center of the bolt circle into global physical coordinates to generate a bolt position set; S500. Calculate the convex hull outer contour of the bolt group according to the bolt position set, and determine the geometric center of the outer contour based on the centroid method or the minimum enclosing rectangle method; S600. Generate a tightening path from inside to outside according to the geometric center of the outer contour, and the tightening component tightens the bolts in a single global physical coordinate system in sequence according to the tightening path.

7. The method for tightening high-strength bolts for steel box girders based on machine vision positioning according to claim 6, characterized in that: The tightening component tightens the bolts in a single global physical coordinate system in sequence according to the tightening path, and further includes: screening bolt nodes on or near the tightening path by a spatial proximity threshold, generating a bolt tightening priority sequence, and the tightening component tightens the bolts in sequence according to the priority sequence.

8. The method for tightening high-strength bolts of steel box girders based on machine vision positioning according to claim 6, characterized in that: It also includes the selection of spiral or radial paths based on the bolt group's symmetry indicators and profile complexity evaluation results, including: Calculate the ratio R of the convex hull area of ​​the bolt group to the minimum enclosing rectangle area and the local density variation coefficient CV; If R <R th or CV>CV th , generating a spiral path; R th is the preset area ratio threshold, CV th To preset the local density coefficient of variation threshold; When R ≥ R th And CV≤CV th When , the symmetry S of the convex hull outer contour and the radial error σr of the bolt group are obtained; If S≤S th And σr<σ th , generating radial paths; σ th is the radial error threshold; S th is the symmetry threshold of the preset convex hull outer contour; If S≤S th And σr≥σ th , generating a spiral path; If S>S th , generating radials or paths; Among them, the radial path is to tighten layer by layer from the geometric center to the outside according to concentric circles; the spiral path is to tighten continuously in ascending order of distance from the geometric center.

9. The method for tightening high-strength bolts of steel box girders based on machine vision positioning according to claim 8, characterized in that: The tightening assembly tightens the bolts one by one in a concentric circle layered order from the geometric center outward, or the tightening assembly moves in ascending order of distance from the center to tighten the bolts one by one, including: The drive tightening component moves to the target bolt position along the planned path, and the tightening torque is detected in real time by the torque sensor, and closed-loop control is performed with the preset threshold. The final tightening torque and coordinate data of each bolt are recorded, and a warning message is issued and the re-inspection location is marked when the tolerance is exceeded.

10. The method for tightening high-strength bolts of steel box girders based on machine vision positioning according to claim 9, characterized in that: The tightening assembly includes a first rotation mode and a second rotation mode; When the first screwing head and the second screwing head move to the target bolt docking position after the image information acquired by the image acquisition group, the tightening assembly enters the first rotation mode. In the first rotation mode, the rotation speed of the first screwing head is v m1 , the second screw head rotation speed is v g1 , high-speed screwing torque is T s When the screw head rotates, the radial force fluctuation is detected by the torque sensor. If the fluctuation value is less than the preset fluctuation value, it is determined that the centering is completed. After the alignment is completed, the tightening assembly switches to the second rotation mode, in which the first screw head rotates at a speed of v m2 , the second screw head rotation speed is v g2 In the second rotation mode, the high torque screwing torque is T b , where v m2 >v m1 , v g2 >v g1 , T b >T s .

Images