Groove pipe butt joint device and pipe clamping butt joint method

Abstract

The application provides a trench pipeline butt joint device and a pipeline clamping butt joint method, walking mechanisms are symmetrically arranged on both sides of a vehicle frame, and a pipe grabbing butt joint device for performing core actions is hung in the middle of the vehicle frame. In order to realize high-precision automatic butt joint, a multi-dimensional sensing system is arranged on the device: a first camera and a second laser radar are symmetrically arranged above fixed jaws of the pipe grabbing butt joint device close to butt joint claws, and are used for accurately sensing a pipeline mouth butt joint state and distance at a short distance; a second camera is symmetrically arranged at the bottom of the rear side of the vehicle frame, and provides a global auxiliary visual angle of the butt joint process; a sixth camera is arranged on the side of the vehicle frame, and is used for recognizing a lateral pipeline position to guide automatic grabbing. Through deep integration of a multi-source sensor layout and a mechanical structure, the application solves the problems of a large visual blind area and difficult manual command cooperation in traditional trench construction, and realizes full-process unmanned operation of the pipeline from automatic pipe searching, stable hoisting to high-precision butt joint in a deep foundation pit.

Description

Technical Field

[0001] This invention relates to the field of pipe connection, and in particular to a trench pipe connection device and a pipe clamping and connection method. Background Technology

[0002] With the acceleration of urbanization, the demand for underground integrated pipe corridors and urban drainage pipe networks is increasing, and the laying of large-diameter concrete or composite material pipes has become a core part of municipal engineering. In the traditional trench pipe construction process, the lowering and connection of pipes mainly relies on excavators and manual assistance. The usual operation mode is: the excavator uses slings to lift the pipe section into the trench, and the construction workers at the bottom of the trench use hand gestures or walkie-talkies to direct the excavator operator to adjust the position and posture of the pipe section. The assistants use tools such as crowbars to make fine adjustments, and finally complete the spigot and socket connection.

[0003] However, this traditional construction method has many significant technical defects and safety hazards. First, the trench environment is narrow and complex. Due to the large blind spots between the excavator cab and the trench bottom, the operator finds it difficult to accurately judge the relative positions of pipe sections, trench walls, and installed pipes, making collisions highly likely, leading to pipe damage or trench collapse. Second, manual command and coordination are inefficient and lack precision, especially when performing millimeter-level socket alignment, which often requires repeated lifting and probing. This not only prolongs the construction period but also easily damages the pipe seals due to uneven stress, causing leakage risks during subsequent operation. In addition, the trench bottom is usually accompanied by water, silt, and dust, creating a harsh working environment. Construction workers face extremely high personal safety risks when working in deep pits for extended periods.

[0004] While some existing automated construction equipment possesses basic lifting capabilities, most lack environmental perception and intelligent decision-making systems. They cannot automatically identify the three-dimensional terrain of trenches, cannot actively suppress pipe section swaying, and lack a closed-loop feedback mechanism during the docking process, making them ill-suited for unstructured outdoor environments with significant light variations and strong dust interference. To address these issues, developing a trench pipe docking device and its control method that integrates environmental perception, intelligent planning, and visual servo control has become a crucial technology urgently needing breakthroughs in the field of municipal construction equipment. Summary of the Invention

[0005] The main objective of this invention is to provide a trench pipe docking device and a pipe clamping docking method. The main technical problem to be solved by this application is that the existing large-diameter pipe trenching construction relies on manual command of excavators, which has significant safety hazards, low centering accuracy, limited visual blind spots, and lack of anti-interference automated closed-loop feedback control, resulting in unstable construction quality and low efficiency.

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a trench pipe docking device, including a gantry frame, with a traveling mechanism symmetrically arranged on both sides of the frame, and a pipe gripping docking device provided in the middle of the gantry frame; The pipe-gripping docking device has a first camera symmetrically positioned on both sides above the fixed claw near the docking claw position, and a second lidar is positioned above the first camera. The first camera and the second lidar face the pipe docking position. A second camera is symmetrically installed on both sides of the bottom rear side of the frame, with the second camera facing the pipe connection point; A sixth camera is also installed on the side of the frame. The sixth camera is used to detect the position of the side pipe and to use the fixing claw of the pipe-grabbing docking device to grab the pipe.

[0007] In the preferred embodiment, the front end of the vehicle frame is equipped with two first lidars and two first cameras arranged symmetrically, and the bottom beam is equipped with a second camera facing the bottom of the ditch; The top of the middle section of the frame mast is equipped with a fifth camera on both sides, with the fifth camera facing the location where the pipe is installed on the pipe-grabbing docking device.

[0008] In the preferred embodiment, the pipe clamping and docking device includes a pipe clamping beam, which is sleeved with a telescopic beam. A second telescopic cylinder is provided inside the pipe clamping beam, and the second telescopic cylinder drives the telescopic beam to extend and retract. The telescopic beam is provided with a docking claw at the end, and two fixed claws are symmetrically provided at both ends of the clamping crossbeam. The fixed claws are slidably connected to the clamping crossbeam, and the first telescopic cylinder is hinged on the clamping crossbeam. The telescopic end of the first telescopic cylinder is hinged to the fixed claw. Two first telescopic cylinders work together to drive two fixed claws to slide on the clamping crossbeam.

[0009] In the preferred embodiment, the fixed claw includes a claw frame, an inwardly inverted sliding frame at the top, first slide rails on both sides of the clamping tube crossbeam, the sliding frame being slidably connected to the first slide rails, two symmetrically arranged movable frames at the bottom of the claw frame, a claw plate with an arc-shaped structure at the bottom of the movable frame, and movable sliding frames on both sides of the upper part of the movable frame being slidably connected to the sliding strips on both sides of the lower part of the claw frame. The gripper frame is equipped with at least two third telescopic cylinders inside, and the bottom of the gripper frame is also equipped with a waist-shaped hole. The telescopic end of the third telescopic cylinder passes through the waist-shaped hole and connects to the movable frame. The gripper frame also has an arc-shaped top plate with an arc structure at the bottom center, and both the top arc plate and the gripper plate are equipped with anti-slip pads.

[0010] In the preferred embodiment, the frame structure is as follows: multiple legs at the bottom of the portal frame are connected to the bottom beam of the rectangular frame structure, and a single walking mechanism is connected to a single leg; The structure of the walking mechanism is as follows: one end of the support connecting frame is hinged to the outrigger, the other end of the support connecting frame is connected to the lifting outrigger, a fourth telescopic cylinder is hinged on the outrigger, and the end of the fourth telescopic cylinder is hinged to the middle of the lifting outrigger. The lifting outrigger is equipped with a fifth telescopic cylinder. The lower telescopic end of the lifting outrigger is connected to the rotary drive device. The rotary drive device uses a hydraulically driven worm gear to drive the worm wheel. The worm wheel is connected to the inverted U-shaped frame. The two sides of the groove at the bottom of the U-shaped frame are connected to the track wheels.

[0011] A pipe clamping and docking method for a trench pipe docking device, the method comprising: S1. Use the walking mechanism to move the vehicle frame to the working position, collect data through the first lidar and the first camera, construct a 3D model of the trench environment and store the depth data; S2. Based on the visual recognition data of the sixth camera, calculate the spatial coordinates of the first pipe, drive the pipe gripping and docking device to move, and use the fixed claw and docking claw to complete the gripping and lifting of the first pipe and place it in the preset installation position in the trench. S3. Using the second camera and the first camera to monitor the relative position of the first pipe and the installed second pipe in real time, drive the first telescopic cylinder and the second telescopic cylinder to make fine adjustments until the tail of the first pipe is aligned with the head of the second pipe. S4. Use the second laser radar to measure the distance between the nozzle and the nozzle. Once docking is complete, release the fixing claw and the docking claw, and repeat steps S1 to S4.

[0012] In the preferred embodiment, in step S1: By pre-establishing a joint calibration extrinsic parameter matrix for the first lidar and the first camera, the point cloud data in the first lidar coordinate system is mapped to the first camera pixel coordinate system through a rigid body transformation matrix to achieve spatial alignment between the point cloud and the image pixels. A voxel grid filter is used to downsample the original point cloud, and a statistical outlier removal algorithm is used to remove noise data caused by dust or reflection. The random sampling consensus algorithm is used to perform plane fitting on the processed point cloud data, segmenting the planar models of the two side walls of the trench and the planar model of the bottom of the trench, and calculating the equation of the central axis of the trench and the depth parameters. By fusing close-up texture images captured by the second camera at the bottom beam, the micro-topography at the bottom of the trench is reconstructed using the structure-of-motion-reconstruction algorithm. Areas with flatness variance exceeding the threshold are marked in the 3D model, generating a raster map of drivable and placeable areas.

[0013] In the preferred embodiment, in step S2: An improved YOLO series deep learning object detection network is deployed in the image processing unit of the sixth camera to select the region of interest in the first pipe in the image and output the two-dimensional pixel bounding box and confidence score of the pipe. Combining the prior geometric dimensions of the first pipe with the bounding box pixel coordinates, the perspective n-point pose calculation algorithm is used to calculate the rotation matrix and translation vector of the first pipe relative to the sixth camera, and determine the 6-DOF pose of the pipe center point in the vehicle frame coordinate system. The motion planning module generates a smooth motion trajectory using a fifth-order polynomial interpolation method based on the current position of the first pipe and the target placement position, and superimposes the input shaping algorithm to suppress residual vibrations of the pipe clamping beam and the first pipe during the hoisting process. During the closing process of the fixed claw, the controller collects pressure sensor data of the hydraulic circuit in real time and uses a fuzzy PID control algorithm to adjust the output pressure of the third telescopic cylinder to ensure that the clamping force is maintained within the set range that neither damages the surface of the first pipe nor causes slippage.

[0014] In the preferred embodiment, in step S3: Gaussian filtering is applied to the images captured by the second and first cameras to remove noise. The Canny operator is used to extract the image edges, and the Hough circle transform algorithm is used to accurately locate the center and radius of the pipe opening contours of the first and second pipes. Construct an image-based visual servo control law, and calculate the image Jacobian matrix between the image feature deviations of the nozzle center distance deviation and radius deviation and the motion speed of the robot end effector including the docking claw and the fixed claw; The calculated six-dimensional correction velocity vector is decomposed into the differential drive command of the first telescopic cylinder and the axial feed command of the second telescopic cylinder. A strategy combining coarse and fine adjustment is adopted. When the distance is greater than the set threshold, positional control is used to quickly approach the target. When the distance is less than the set threshold, incremental PI control based on image features is switched until the coaxiality deviation and axial clearance of the pipe end meet the assembly tolerance requirements.

[0015] In the preferred embodiment, the method deploys a layered computer software system on an embedded industrial computer and a PLC controller, deploys a robot operating system middleware in a Linux environment, loads the first LiDAR driver node and the driver nodes of the first and second cameras, collects high-resolution image data through the GigE interface, collects point cloud data through the Ethernet interface, and publishes it as a topic message. A CUDA-accelerated TensorRT deep neural network inference engine is deployed in the GPU unit of the industrial control computer to run object detection algorithms and point cloud processing algorithms in real time. The state machine management program runs in the industrial control computer and is responsible for scheduling the work process from S1 to S4 and monitoring the health status of each subsystem in real time. When an abnormality is detected, the emergency stop logic is triggered through the watchdog mechanism. Establish real-time industrial Ethernet communication between the industrial computer and the PLC controller using Modbus-TCP or EtherCAT. The industrial computer packages and sends the calculated joint speed and position commands to the PLC. The PLC parses the commands into pulse signals or analog signals to drive each hydraulic valve group and servo motor driver. The system background runs a data log module that uploads the final position data, torque data, and image evidence from each docking to the cloud database and synchronously updates the virtual model's posture in the digital twin interface.

[0016] This invention provides a trench pipe docking device and a pipe clamping docking method. This application uses multi-source sensor fusion and environmental perception algorithms, combined with joint calibration technology and random sampling consistency algorithm to construct a high-precision three-dimensional trench model. Combined with statistical outlier removal algorithm, it effectively overcomes the interference of on-site dust and light changes, solves the visual blind spot problem of traditional mechanical operation, provides reliable environmental constraints for path planning, and significantly improves the safety of operation under complex working conditions.

[0017] This application combines improved deep learning target detection, perspective n-point pose calculation and input shaping algorithms to achieve automatic identification, positioning and anti-sway stable hoisting of pipes; with the addition of adaptive clamping control based on fuzzy PID, it ensures that the fixed claw does not slip or cause surface damage when gripping heavy pipes, thus realizing the automated and precise transfer and stable lowering of heavy load pipes.

[0018] This application utilizes visual servo control technology based on image Jacobian matrix, employing a multi-level approximation strategy combining coarse and fine adjustments to achieve millimeter-level coaxiality alignment of the pipe orifice, effectively preventing damage to the sealing ring caused by forceful insertion. Simultaneously, the layered hardware and software deployment, coupled with CUDA acceleration and real-time industrial Ethernet communication, ensures millisecond-level response from sensing to execution, realizing the digitalization and intelligentization of the construction process. Attached Figure Description

[0019] The present invention will be further described below with reference to the accompanying drawings and embodiments: Figure 1 This is a front view of the trench pipe docking device of the present invention. Figure 2 This is a structural diagram of the trench pipe docking device of the present invention being hoisted into the trench; Figure 3 This is a diagram of the lifting structure of the trench pipe docking device of the present invention; Figure 4This is a side view of the lifting structure of the trench pipe docking device of the present invention; Figure 5 This is a main view of the hoisting structure of the trench pipe docking device of the present invention; Figure 6 This is a side view of the hoisting structure of the trench pipe docking device of the present invention; Figure 7 This is a front view structural diagram of the trench pipe docking device of the present invention; Figure 8 This is a side view of the fixing claw structure of the present invention; Figure 9 This is a front view structural diagram of the fixing claw of the present invention; Figure 10 This is a diagram of the overall structure of the vehicle frame of this invention; Figure 11 This is a structural diagram of the running mechanism of the vehicle frame of the present invention.

[0020] In the diagram: frame 1; bottom beam 101; lifting frame 102; outrigger 103; Pipe gripping and docking device 2; docking claw 21; pipe clamping beam 22; second telescopic cylinder 2201; telescopic beam 23; first telescopic cylinder 24; first slide rail 25; fixed claw 26; sliding frame 2601; claw frame 2602; sliding strip 2603; third telescopic cylinder 2604; waist-shaped hole 2605; moving frame 2606; claw plate 2607; moving sliding frame 2608; top arc plate 2609; 27 bearing beam; 2701 sliding beam; 2702 sliding beam; 2703 first rack; 2704 first motor; 28 first traveling motor; 29 first electric crane; Walking mechanism 3; support connecting frame 301; fourth telescopic cylinder 302; lifting outrigger 303; fifth telescopic cylinder 304; rotary drive device 305; worm gear 306; worm 307; U-shaped frame 308; track wheel 309; First camera 4; Second LiDAR 401; Second camera 5; Third camera 6; Fourth camera 7; Control center 8; Fifth camera 9; First pipe 10; Second pipe 11; First LiDAR 12; Sixth camera 13. Detailed Implementation

[0021] Example 1 like Figure 1-11 As shown, a trench pipe docking device includes a gantry frame 1 with symmetrically arranged traveling mechanisms 3 on both sides of the frame 1, and a pipe-gripping docking device 2 is provided in the middle of the gantry frame 1. The pipe-gripping docking device 2 has a first camera 4 symmetrically arranged on both sides above the fixed claw 26 near the docking claw 21. A second laser radar 401 is arranged above the first camera 4. The first camera 4 and the second laser radar 401 face the pipe docking position. The bottom rear side of the frame 1 is symmetrically equipped with second cameras 5, which face the pipe docking position; The side of the frame 1 is also equipped with a sixth camera 13, which is used to detect the position of the side pipe and to use the fixing claw 26 of the pipe-grabbing docking device 2 to grab the pipe.

[0022] This embodiment proposes a trench pipe docking device, the main structure of which adopts a gantry-type frame 1. This gantry design allows the frame 1 to span across the trench, with the supporting structures on both sides located on the ground on either side of the trench, thus avoiding direct pressure from the weight of the equipment on the soil at the trench edge and reducing the risk of collapse. Symmetrically arranged on both sides of the frame 1 are traveling mechanisms 3, which drive the entire device to move longitudinally along the trench, achieving coarse position adjustments. A pipe-gripping docking device 2 is located in the middle of the gantry-type frame 1. This device is the core actuator for performing pipe gripping, handling, and high-precision docking. Located within the internal space of the frame 1, the pipe-gripping docking device 2 can flexibly adjust the pipe's posture in three-dimensional space by utilizing the height and width advantages of the gantry.

[0023] In the preferred embodiment, the front end of the frame 1 is provided with two first lidars 12 and two first cameras 6 arranged symmetrically, and the bottom beam 101 is provided with a second camera 7 facing the bottom of the ditch; The top of the middle section of the frame 1 mast is equipped with a fifth camera 9 on both sides, and the fifth camera 9 faces the position where the pipe is installed on the pipe-grabbing docking device 2.

[0024] A second camera 7, facing the bottom of the trench, is installed on the bottom beam 101 of the chassis 1. This design primarily addresses the blind spot problem in the vertical viewing angle. Because the sensor at the front of the chassis 1 has a large downward angle when observing the bottom of the trench, shadowed areas or blind spots deep within the trench are difficult to completely cover. The second camera 7, with its vertically downward view, can directly acquire an orthophoto of the trench bottom. Its advantage lies in clearly monitoring the flatness of the trench bottom and whether there is accumulated water, silt, or leftover tools. Before the pipe is placed in the trench, the system analyzes the image from the second camera 7 to confirm whether the flatness of the laid subgrade meets the process standards, preventing pipe breakage due to uneven stress caused by an uneven trench bottom. This design ensures the construction quality of concealed works and avoids the rework risk caused by blindly laying pipes.

[0025] A fifth camera 9 is installed on both sides of the top center of the mast of the frame 1, facing the location where the pipe-gripping docking device 2 is installed. This set of cameras constitutes the system's global monitoring perspective, i.e., an omniscient view. Unlike the close-up shots mounted on the robotic arm, the fifth camera 9 has a wide field of view, covering the entire working space of the pipe-gripping docking device 2. Its advantage lies in providing macroscopic motion interference detection capabilities. When the pipe-gripping docking device 2 performs large-scale gripping, lifting, and rotating movements, the control system monitors the relative positions of the robotic arm with the frame 1 column, hydraulic lines, and surrounding personnel in real time through the fifth camera 9. Once it detects that the movement trajectory of the pipe-gripping docking device 2 may collide with the frame 1, or that personnel have accidentally entered the work area, the system can immediately trigger a safety brake. In addition, the top-down view provided by the fifth camera 9 also provides the most intuitive reference image for remote monitoring by the operator, reducing the cognitive load of remote operation.

[0026] In the preferred embodiment, the pipe clamping and docking device 2 includes a pipe clamping beam 22, which is sleeved with a telescopic beam 23. The pipe clamping beam 22 is provided with a second telescopic cylinder 2201 inside, which drives the telescopic beam 23 to extend and retract. The telescopic beam 23 is provided with a docking claw 21 at the end, and the clamping crossbeam 22 is provided with two fixed claws 26 symmetrically at both ends. The fixed claws 26 are slidably connected to the clamping crossbeam 22. The first telescopic cylinder 24 is hinged on the clamping crossbeam 22, and the telescopic end of the first telescopic cylinder 24 is hinged to the fixed claw 26. The two first telescopic cylinders 24 work together to drive the two fixed claws 26 to slide on the clamping crossbeam 22.

[0027] The clamping beam 22 not only bears the weight of the pipe but also serves as a mounting base for other moving parts. To enable adjustment and extension along the length of the device, the clamping beam 22 and the telescopic beam 23 are connected by a sleeve connection, meaning the telescopic beam 23 is inserted inside or outside the clamping beam 22, allowing for relative axial displacement. This relative displacement is powered by a second telescopic cylinder 2201 located inside the clamping beam 22. One end of the second telescopic cylinder 2201 is fixed to the clamping beam 22, and the other end is connected to the telescopic beam 23. The cylinder's extension and retraction are driven by hydraulic or pneumatic pressure, directly causing the telescopic beam 23 to extend and retract on the clamping beam 22, thereby changing the overall working length of the pipe-gripping and docking device 2 to accommodate pipe docking requirements of different distances.

[0028] When performing specific gripping and docking tasks, docking claws 21 are installed at the ends of the telescopic beam 23. Since the docking claws 21 are located on the telescopic beam 23, they can extend into the depths of the trench or reach adjacent installed pipes as the telescopic beam 23 extends, serving to assist in positioning or tightening the docking. Simultaneously, two fixed claws 26 are symmetrically arranged at both ends of the pipe clamping beam 22. These two fixed claws 26 are mainly used to clamp the pipe to be installed. The fixed claws 26 are not rigidly fixed to the pipe clamping beam 22, but rather through a sliding connection, meaning that the fixed claws 26 can move along the guide rails or surface of the pipe clamping beam 22. To control this movement, a first telescopic cylinder 24 is hinged to the pipe clamping beam 22, and the telescopic end of the first telescopic cylinder 24 is directly hinged to the fixed claws 26.

[0029] The coordinated action of the two first telescopic cylinders 24 drives the two fixed claws 26 to slide on the pipe clamping beam 22. This design allows the two fixed claws 26 to move inward or outward synchronously, thereby realizing the clamping and releasing operation of the pipe body, or adjusting the distance between the two fixed claws 26 to adapt to pipes of different lengths and specifications, ensuring a stable center of gravity for clamping.

[0030] In the preferred embodiment, the fixed claw 26 includes a claw frame 2602, with an inwardly inverted sliding frame 2601 on the upper part. The clamping tube crossbeam 22 has first slide rails 25 on both sides. The sliding frame 2601 is slidably connected to the first slide rails 25. The claw frame 2602 has two symmetrically arranged movable frames 2606 below it. The lower end of the movable frame 2606 has an arc-shaped claw plate 2607. The movable sliding frames 2608 on both sides of the upper part of the movable frame 2606 are slidably connected to the sliding strips 2603 on both sides below the claw frame 2602. The gripper frame 2602 is provided with at least two third telescopic cylinders 2604 inside, and the bottom of the gripper frame 2602 is also provided with a waist-shaped hole 2605. The telescopic end of the third telescopic cylinder 2604 passes through the waist-shaped hole 2605 and is connected to the movable frame 2606. The gripper frame 2602 is also provided with a top arc plate 2609 with an arc structure at the bottom center. Both the top arc plate 2609 and the gripper plate 2607 are provided with anti-slip pads.

[0031] To achieve stable movement and positioning of the fixed claw 26 on the pipe clamping beam 22, the upper part of the claw frame 2602 is designed with an inwardly inverted sliding frame 2601. Correspondingly, first slide rails 25 are provided on both sides of the pipe clamping beam 22, and the sliding frame 2601 is snapped onto the first slide rails 25, forming a slidable connection. This inverted sliding fit structure not only ensures that the fixed claw 26 can move smoothly along the beam axial direction to adjust its clamping position, but also restricts the vertical freedom of the fixed claw 26 through its inverted geometry, ensuring structural safety when hoisting heavy pipes and preventing derailment.

[0032] Below the gripper frame 2602, two symmetrically arranged movable frames 2606 for performing opening and closing actions are provided. Each movable frame 2606 has a curved gripper plate 2607 fixed to its lower end. This curved structure is designed to conform to the shape of the pipe's outer wall, increasing the contact area. To guide the movable frames 2606 in precise opening and closing movements, movable sliding frames 2608 are provided on both sides of the upper part of the movable frames 2606, while side sliding strips 2603 are correspondingly provided on both sides below the gripper frame 2602. The movable sliding frames 2608 are fitted or embedded in the side sliding strips 2603, forming a stable guiding mechanism. The power source driving this opening and closing mechanism comes from at least two third telescopic cylinders 2604 installed inside the gripper frame 2602. An oblong hole 2605 is provided at the bottom of the gripper frame 2602. The telescopic ends of the third telescopic cylinders 2604 pass downward through this oblong hole 2605 and are rigidly connected to the movable frames 2606 below.

[0033] In addition, to further enhance clamping stability and protect the tube body, a top arc-shaped plate 2609 with an arc structure is fixed at the center of the lower part of the gripper frame 2602. When the two side gripper plates 2607 retract inward to clamp the tube body, the top outer wall of the tube body will abut against the top arc-shaped plate 2609, thus forming a three-point or multi-point surrounding support structure. At the same time, the surfaces of the top arc-shaped plate 2609 and the two side gripper plates 2607 are covered with anti-slip pads.

[0034] The cooperation between the side sliding strip 2603 and the movable sliding frame 2608 ensures that the movable frame 2606 maintains stable linear movement under load, preventing the clamps from tilting or jamming due to the weight of the pipe. Furthermore, the top arc-shaped plate 2609 provides a rigid top positioning reference, which, together with the symmetrically contracting movable frame 2606 on both sides, enables automatic centering of the pipe, automatically correcting its posture to align its axis during clamping. Finally, the application of anti-slip pads not only increases the friction between the mechanical claws and the pipe, preventing slippage during lifting, but also effectively prevents the metal claws from directly scratching the anti-corrosion layer of the pipe's outer wall, ensuring the quality of pipe construction.

[0035] In the preferred embodiment, the structure of the frame 1 is as follows: multiple legs 103 at the bottom of the portal frame are connected to the bottom beam 101 of the rectangular frame structure, and a single walking mechanism 3 is connected to a single leg 103. The structure of the walking mechanism 3 is as follows: one end of the support connecting frame 301 is hinged to the outrigger 103, the other end of the support connecting frame 301 is connected to the lifting outrigger 303, and a fourth telescopic cylinder 302 is hinged on the outrigger 103. The end of the fourth telescopic cylinder 302 is hinged to the middle of the lifting outrigger 303. The lifting outrigger 303 is equipped with a fifth telescopic cylinder 304. The lower telescopic end of the lifting outrigger 303 is connected to the rotary drive device 305. The rotary drive device 305 uses a hydraulic drive worm 307 to drive the worm wheel 306. The worm wheel 306 is connected to the inverted U-shaped frame 308. The two sides of the groove at the bottom of the U-shaped frame 308 are connected to the track wheels 309.

[0036] Multiple outriggers 103 at the lower part of the frame 1 are securely connected to a bottom beam 101 with a rectangular frame structure. The bottom beam 101 not only enhances the overall rigidity of the frame 1 but also serves as the mounting base for the material spreading and compaction equipment. To enable the mobile operation of the device, a single traveling mechanism 3 is mounted on a single outrigger 103. The connection between the traveling mechanism 3 and the outrigger 103 is hinged. Specifically, one end of the support connecting frame 301 is hinged to the outrigger 103 via a pin, while the other end is rigidly connected or hinged to the lifting outrigger 303. At the same time, a fourth telescopic cylinder 302 is also hinged to the outrigger 103, and the output end of the fourth telescopic cylinder 302 is hinged to the middle of the lifting outrigger 303. This structure, consisting of the outrigger 103, the support connecting frame 301, the fourth telescopic cylinder 302, and the lifting outrigger 303, allows the span between the left and right traveling mechanisms 3 to be adjusted by controlling the extension length of the fourth telescopic cylinder 302, thereby changing the unfolding angle or lateral distance of the lifting outrigger 303 relative to the frame 1.

[0037] In terms of vertical adjustment and steering control, the lifting outrigger 303 integrates a fifth telescopic cylinder 304. The fifth telescopic cylinder 304 serves as the power source for vertical lifting, driving the lower telescopic end of the lifting outrigger 303 to move up and down, thereby adjusting the ground clearance of the chassis 1 or leveling the vehicle body. A rotary drive device 305 is connected to the lower end of the lifting outrigger 303; this device is a key component for achieving travel and steering. The rotary drive device 305 is hydraulically driven, using a hydraulic motor to rotate the worm gear 307, which in turn drives the worm wheel 306 meshing with the worm gear 307. An inverted U-shaped frame 308 is connected to the lower part of the worm wheel 306, straddling the traveling components. Track wheels 309 are mounted on both sides of its lower groove. When the worm wheel 306 rotates, it drives the inverted U-shaped frame 308 and the track wheels 309 to rotate as a whole, thereby changing the direction of travel of the track wheels 309.

[0038] The cooperation between the fourth telescopic cylinder 302 and the articulated structure enables variable adjustment of the walking mechanism's span. This allows the construction device to adapt to trench operations of varying widths or to flexibly adjust the support position when ground conditions on both sides of the trench are inconsistent, greatly improving the equipment's adaptability to different working conditions. Secondly, the lifting outriggers 303 are equipped with a fifth telescopic cylinder 304, enabling independent lifting of the four outriggers. When the ground at the construction site is uneven or has a slope, the height of each outrigger can be adjusted individually to maintain the horizontal state of the main body of the chassis 1, ensuring the accuracy of pipe clamping and compaction operations. Furthermore, the slewing drive device 305 adopts a worm gear transmission mechanism, which not only provides a large steering torque to overcome ground resistance when the tracks turn, but also utilizes the reverse self-locking characteristic of the worm gear mechanism to ensure that the track wheels 309 will not unexpectedly deflect due to external forces when traveling in a straight line or stopping, significantly enhancing the stability and safety of the entire machine's movement. Finally, the application of track wheels 309 effectively reduces the ground pressure, preventing the risk of the equipment getting stuck or collapsing the trench wall when operating on soft soil at the edge of the trench.

[0039] Example 3 Further explanation in conjunction with Example 1, such as Figure 1-11 The structure shown illustrates a pipe clamping and docking method for a trench pipe docking device, the method comprising: S1. Use the walking mechanism 3 to move the vehicle frame 1 to the working position, collect data through the first lidar 12 and the first camera 6, construct a 3D model of the trench environment and store the depth data; S2. Based on the visual recognition data of the sixth camera 13, calculate the spatial coordinates of the first pipe 10, drive the pipe gripping and docking device 2 to move, and use the fixed claw 26 and docking claw 21 to complete the gripping and lifting of the first pipe 10 and place it in the preset installation position in the trench. S3. Using the second camera 5 and the first camera 4 to monitor the relative position of the first pipe 10 and the installed second pipe 11 in real time, drive the first telescopic cylinder 24 and the second telescopic cylinder 2201 to make fine adjustments until the tail of the first pipe 10 is aligned with the head of the second pipe 11. S4. Use the second laser radar 401 to measure the distance between the nozzle and the pipe opening. After determining that the docking is completed, release the fixing claw 26 and the docking claw 21, and repeat steps S1 to S4.

[0040] This embodiment provides a pipe clamping and docking method for a trench pipe docking device. Based on the aforementioned hardware, this method achieves fully automated operation from environmental perception and automatic material handling to high-precision docking through multi-sensor information fusion and a phased control strategy. The method mainly includes the following four core steps: Step S1: Environmental Perception and Initial Positioning: This step is fundamental to the entire operation. First, the operator or the upper-level dispatch system issues an instruction to drive the walking mechanism 3, which in turn moves the gantry frame 1 along the trench. During movement or after stopping at the designated work position, the system activates the front-end perception unit. The first lidar 12 performs a high-frequency scan of the terrain in front of the trench, acquiring point cloud data containing distance information; simultaneously, the first camera 6 collects color image data of the environment. The control system performs spatiotemporal alignment and fusion of the two sets of data to construct a high-precision 3D environmental model of the current work area. This model not only includes the trench's depth, width, and edge slope data but also marks the flatness of the trench bottom. The system stores this depth data and geometric information in the onboard database as an "electronic map" for subsequent path planning, ensuring that the pipe-grabbing docking device 2 can actively avoid trench walls and obstacles during movement, preventing collisions.

[0041] Step S2: Intelligent Pipe Locator and Automated Lifting: This step automates the material flow. Once the chassis 1 is positioned, the sixth camera 13 on the side is activated, visually scanning the pipes stacked on the side of the trench. Using a target detection algorithm running on the industrial computer, the system can identify the first pipe 10 to be installed from a complex background and calculate its center coordinates and axis orientation in the world coordinate system. Based on this coordinate information, the control system generates the motion trajectory of the pipe-grabbing docking device 2, driving it to move directly above the first pipe 10. Subsequently, the fixing claw 26 and the docking claw 21 work together to firmly grasp the pipe. During lifting, the system plans the optimal lowering path based on the 3D model established in step S1, controlling the first pipe 10 to smoothly cross the edge of the trench and descend to the preset installation position inside the trench. This process completely replaces manual hooking and command, eliminating the safety hazards of personnel working under heavy loads.

[0042] Step S3: Visual Servo Fine-Tuning: This step is crucial for ensuring project quality. When the first pipe 10 is lowered close to the already installed second pipe 11, the system enters fine-tuning mode. At this time, the second camera 5, mounted on the rear of the chassis, provides macroscopic relative position monitoring to ensure that the axes of the two pipes are roughly aligned; simultaneously, the first camera 4, located on the gripper, provides a microscopic close-up view of the interface, capturing minute deviations at the pipe opening edges in real time. Based on the visual error signals fed back by these two cameras, the control system constructs a closed-loop control circuit and calculates the required correction amount. Subsequently, the system precisely drives the first telescopic cylinder 24 to adjust the horizontal and vertical positions of the pipe body, and drives the second telescopic cylinder 2201 to adjust the axial telescopic position of the pipe body. Through this multi-dimensional coordinated fine-tuning, the tail of the first pipe 10 and the head of the second pipe 11 are forced to achieve millimeter-level precise alignment in space, effectively avoiding damage to the sealing ring or pipe opening caused by forceful insertion.

[0043] Step S4: Closed-Loop Verification and Continuous Operation: This step confirms the docking quality and establishes an operational cycle. Simultaneously with the docking action, a second lidar 401 installed near the docking claws measures the absolute distance change between the pipe section and the sensor in real time. When the measured distance value stabilizes within a preset standard threshold range, it indicates that the first pipe 10 has been fully inserted and is securely connected. Based on this, the control system determines that the docking is complete and immediately issues a command to release the fixing claw 26 and the docking claw 21, separating the device from the pipe body. Afterward, the pipe-gripping docking device 2 resets, the frame 1 moves forward one pipe section, and the system returns to step S1 to begin the installation cycle for the next pipe. This acceptance mechanism based on objective sensor data avoids subjective errors from human visual judgment and ensures the consistency of the entire pipeline construction quality.

[0044] In summary, this method decomposes the complex trench pipeline construction process into quantifiable and controllable standardized steps through the orderly connection of S1 to S4. Combined with multi-source sensing and feedback control, it significantly improves construction efficiency and docking accuracy.

[0045] In the preferred embodiment, in step S1: By pre-establishing a joint calibration extrinsic parameter matrix for the first lidar 12 and the first camera 6, the point cloud data in the coordinate system of the first lidar 12 is mapped to the pixel coordinate system of the first camera 6 through a rigid body transformation matrix to achieve spatial alignment between the point cloud and the image pixels. A voxel grid filter is used to downsample the original point cloud, and a statistical outlier removal algorithm is used to remove noise data caused by dust or reflection. The random sampling consensus algorithm is used to perform plane fitting on the processed point cloud data, segmenting the planar models of the two side walls of the trench and the planar model of the bottom of the trench, and calculating the equation of the central axis of the trench and the depth parameters. By integrating the close-up texture images captured by the second camera 7 at the bottom beam 101, the micro-topography of the trench bottom is reconstructed using the structure of motion restoration algorithm. Areas with flatness variance exceeding the threshold are marked in the 3D model, and a grid map of the drivable and placeable areas is generated.

[0046] In step S1 of this embodiment, to address the limited sensing capability of a single sensor in complex trench environments, a high-precision environmental modeling method based on multi-source information fusion is employed. First, considering the spatial differences between the first lidar 12 and the first camera 6, the system establishes a rigid connection between them through a pre-executed joint calibration process. At the data processing level, the 3D point cloud data acquired by the first lidar 12 is based on the lidar coordinate system, while the image data acquired by the first camera 6 is based on the camera coordinate system. To achieve precise correspondence between point clouds and image pixels, the system constructs a joint calibration extrinsic parameter matrix, which includes a rotation matrix and a translation vector. The mathematical model for mapping any point in the lidar coordinate system to the pixel coordinate system of the first camera 6 is shown in the following equation: ; In the above formula, The three-dimensional coordinate vector of a point in space in the first lidar 12 coordinate system is usually represented as: . This represents the rotation matrix that transforms the coordinate system from the LiDAR coordinate system to the coordinate system of the first camera (6 cameras). It is an orthogonal matrix that describes the relative attitude between the two sensors. This represents the translation vector from the origin of the lidar coordinate system to the optical center of the first camera 6. The intrinsic parameter matrix of the first camera 6 includes inherent optical parameters such as focal length, principal point coordinates, and pixel tilt factor. This is the depth value of the point in the camera coordinate system. These are the pixel coordinates projected onto the image plane. Using this algorithm, the system can assign depth information from the LiDAR to corresponding image pixels, or map the color and texture information of the image onto a 3D point cloud, thereby generating a true-color point cloud with color information. The beneficial effect of this fusion technology is that it enables the control system not only to perceive the distance to obstacles, but also to identify the material properties of obstacles through color and texture, such as distinguishing between concrete walls and soft soil, thus significantly improving the accuracy of environmental semantic understanding.

[0047] After acquiring the raw point cloud data, considering the significant dust, fluff, and reflective interference from metal surfaces at the construction site, directly using the raw data would lead to decreased modeling accuracy or even the creation of false obstacles. Therefore, this embodiment employs a two-stage filtering strategy. First, a voxel mesh filter is used to downsample the point cloud. This algorithm divides the 3D space into a series of tiny cubic meshes, i.e., voxels, and calculates the geometric centroid of all points within each voxel, using this centroid to represent the entire set of points within the voxel. This process significantly reduces the amount of data while preserving the macroscopic geometric features of the environment, improving the real-time performance of subsequent algorithms. Subsequently, a statistical outlier removal algorithm is used for fine denoising. This algorithm traverses every point in the point cloud, calculating its distance to its nearest neighbor... The average distance to each neighboring point is calculated, and the mean of the average distance distribution of all points is calculated. and standard deviation The system sets a standard deviation multiple threshold; any value with an average distance greater than a certain threshold will be flagged. All points were identified as noise points and removed. The beneficial effect of this step is that it can effectively filter out floating noise points caused by dust diffuse reflection and outliers caused by strong light, ensuring that the constructed 3D model is pure and reliable, and preventing the walking mechanism 3 from being triggered to stop suddenly due to false obstacles.

[0048] After point cloud preprocessing, the system employs a random sampling consensus algorithm to extract structured features of grooves from the disordered point cloud. This algorithm iteratively selects a subset of points cloud data to fit a planar model, the plane equation of which is typically expressed as... The algorithm calculates the Euclidean distance from all data points to the fitted plane and counts the number of inliers whose distance is less than a preset threshold. After multiple iterations and optimizations, the system can robustly segment the left and right sidewall planes and the bottom plane of the trench. Based on the segmented plane model, the system further solves the intersection or midline of the left and right sidewall planes as the equation of the trench's central axis through geometric calculations, and calculates the vertical distance from the bottom plane to the ground as the depth parameter. The beneficial effect of this technology is that it transforms unstructured environmental data into mathematical parameters that can be used for mechanical control, enabling the pipe-gripping and docking device 2 to automatically center the trench and plan the descent path of the pipe fittings according to the depth parameters, thus realizing intelligent guidance of the operation process.

[0049] Finally, addressing the issue that the flatness of the trench bottom directly affects the quality of pipe installation, this embodiment integrates close-range texture images captured by the second camera 7 at the bottom beam 101. Utilizing a motion reconstruction algorithm, the system continuously acquires multiple frames of bottom images as the chassis 1 moves, reconstructing the microscopic three-dimensional terrain of the trench bottom through feature point matching and triangulation principles. After obtaining the microscopic terrain, the system divides the bottom area into several grid cells and calculates the variance of the height data within each cell. If the variance value of a cell exceeds a set flatness threshold, it is determined that the area contains protruding gravel or pits, failing to meet pipe laying requirements. Based on this, the system generates a grid map containing drivable and placement areas and marks abnormal areas on the map. The beneficial effect of this design is that it provides an automated method for inspecting the quality of concealed engineering works, ensuring that pipes are placed on a flat and solid base layer, effectively preventing the risk of uneven stress or breakage of the pipe body due to uneven bottom, and also providing navigation guidance for the walking mechanism 3 to avoid ground potholes.

[0050] In the preferred embodiment, in step S2: An improved YOLO series deep learning object detection network is deployed in the image processing unit of the sixth camera 13 to select the region of interest of the first pipe 10 in the image and output the two-dimensional pixel bounding box and confidence score of the pipe body. Combining the prior geometric dimensions of the first pipe 10 with the pixel coordinates of the bounding box, the perspective n-point pose calculation algorithm is used to calculate the rotation matrix and translation vector of the first pipe 10 relative to the sixth camera 13, and the 6-DOF pose of the pipe center point in the frame 1 coordinate system is determined. The motion planning module generates a smooth motion trajectory using a fifth-order polynomial interpolation method based on the current position of the first pipe 10 and the target placement position, and superimposes the input shaping algorithm to suppress the residual vibration of the pipe clamping beam 22 and the first pipe 10 during the hoisting process. During the closing process of the fixed claw 26, the controller collects pressure sensor data of the hydraulic circuit in real time and uses a fuzzy PID control algorithm to adjust the output pressure of the third telescopic cylinder 2604 to ensure that the clamping force is maintained within the set range that neither damages the surface of the first pipe 10 nor causes slippage.

[0051] In step S2 of this embodiment, to achieve automated grasping and stable lifting of unstructured stacked pipes, an intelligent operation system integrating perception, decision-making, and control is constructed. First, in the visual perception stage, the system utilizes a sixth camera 13 mounted on the side of the chassis 1 as an image acquisition terminal. To accurately extract the target pipe from complex backgrounds such as soil, gravel, and weeds, an improved YOLO series deep learning object detection network is deployed within the image processing unit. This network employs a deep convolutional neural network as a feature extractor and introduces a feature pyramid structure to enhance the detection capability for pipes of different scales. During algorithm execution, the input video stream is analyzed frame by frame, quickly selecting the region of interest (ROI) of the first pipe 10 in the image, and outputting a two-dimensional pixel bounding box containing the coordinates of the upper left and lower right corners, as well as a probability score representing the confidence level. The beneficial effect of this process is that it endows the device with "machine vision," enabling it to quickly locate the work object like a skilled worker, unaffected by changes in lighting or cluttered backgrounds, completely eliminating the safety hazards of manual hooking.

[0052] After obtaining the two-dimensional pixel information of the pipe, in order to guide the robotic arm to accurately reach the grasping position, the system needs to convert the two-dimensional image coordinates into a three-dimensional spatial pose. At this time, the system combines the pre-stored prior geometric dimensions of the first pipe 10, such as pipe diameter and length, with the detected bounding box pixel coordinates, and uses the perspective n-point pose calculation algorithm PnP to calculate. The core of this algorithm is to solve for the rotation matrix and translation vector of the target object coordinate system relative to the camera coordinate system. Its mathematical projection model is shown in the following equation: ; In the above formula, is the scale factor, representing depth information. These are the pixel coordinates of the feature points on the image plane. Let be the intrinsic parameter matrix of the sixth camera 13, containing the focal length and principal point offset. Let [R | t] be the extrinsic parameter matrix to be solved, where ... yes The rotation matrix describes the attitude angles of the first pipe 10 relative to the camera: roll, pitch, and yaw. yes The translation vector describes the spatial distance between the center point of the first pipe 10 and the camera. The known three-dimensional coordinates of the feature points of the first pipe 10 in the world coordinate system are given. By solving this equation, the system determines the precise 6-DOF pose of the pipe center point in the frame 1 coordinate system. The beneficial effect of this technology is that it achieves a leap from "seeing" to "positioning", ensuring that the pipe gripping and docking device 2 can approach the pipe in the correct posture, avoiding gripping failure or collision due to positional deviation.

[0053] After determining the target and the current position of the robotic arm, the motion planning module generates an optimal motion path from the current point to the target point. To prevent severe swaying of the heavy-duty pipe during lifting and movement, this embodiment abandons the traditional trapezoidal velocity planning and instead employs a fifth-order polynomial interpolation method. This method constrains position, velocity, acceleration, and jerk accelerometer to ensure the continuity and smoothness of the motion curve. Its trajectory equation is shown below: ; In the above formula, Indicates time The displacement of the joint at any moment, to The coefficients are polynomials, calculated from the position, velocity, and acceleration boundary conditions at the starting and ending points. Based on this, the system further superimposes an input shaping algorithm. This algorithm decomposes the original drive command into a series of pulse sequences with specific time delays and amplitudes, using the vibration generated by the subsequent pulse to cancel out the residual vibration generated by the previous pulse. The beneficial effect of this technology is that it suppresses the swaying effect of flexible loads from the control source, allowing the tens-of-tons-weight pipe to come to an instantaneous stop after high-speed movement, greatly improving operational efficiency and positioning accuracy, and preventing the pipe from swaying and impacting the trench wall.

[0054] Finally, during the process of the fixing claw 26 contacting and gripping the pipe body, in order to resolve the contradiction between "excessive clamping force damaging the pipe" and "insufficient clamping force causing slippage," the controller uses a fuzzy PID control algorithm to adjust the output pressure of the third telescopic cylinder 2604. The system collects the values ​​of the pressure sensors in the hydraulic circuit in real time as feedback signals to calculate the current pressure deviation. and its rate of change The fuzzy controller dynamically adjusts the proportional gain of the PID controller based on a preset fuzzy rule table. Integral coefficient and differential coefficients The output of the PID controller As shown in the following formula: ; In the above formula, This is an electrical signal used to control the opening of a hydraulic proportional valve. When the deviation is large, the fuzzy logic increases. To accelerate response speed; when the deviation is small, increase To eliminate static error; when pressure changes drastically, increase This technology helps suppress overshoot. Its beneficial effect lies in giving the hydraulic system adaptive capabilities, enabling it to automatically match the optimal clamping force according to different materials such as concrete, steel pipes, PE pipes, and pipe diameters, achieving "flexible gripping" and effectively ensuring the integrity of the pipe and the safety of construction.

[0055] In the preferred embodiment, in step S3: The images captured by the second camera 5 and the first camera 4 are subjected to Gaussian filtering for noise reduction. The Canny operator is used to extract the image edges, and the Hough circle transform algorithm is used to accurately locate the center and radius of the pipe opening contours of the first pipe 10 and the second pipe 11. Construct an image-based visual servo control law, and calculate the image Jacobian matrix between the image feature deviation of the nozzle center distance deviation and radius deviation and the motion speed of the robot end effector including the docking claw 21 and the fixed claw 26. The calculated six-dimensional correction velocity vector is decomposed into the differential drive command of the first telescopic cylinder 24 and the axial feed command of the second telescopic cylinder 2201. A strategy combining coarse and fine adjustment is adopted. When the distance is greater than the set threshold, positional control is used to quickly approach the target. When the distance is less than the set threshold, incremental PI control based on image features is switched until the coaxiality deviation and axial clearance of the pipe end meet the assembly tolerance requirements.

[0056] In step S3 of this embodiment, to address the issues of poor alignment accuracy and repeated trial and error caused by reliance on visual observation in traditional manual docking processes, a high-precision visual servo closed-loop control system based on images is constructed. First, the system preprocesses the raw video streams acquired by the second camera 5 and the first camera 4. Due to vibration and light interference at the construction site, images often contain Gaussian noise. The system first uses a Gaussian filter to perform convolution operations on the images, effectively suppressing high-frequency noise by utilizing the smoothing properties of the Gaussian kernel function. Subsequently, the Canny operator is used to perform edge detection on the denoised image. The Canny operator, by calculating the magnitude and direction of the image gradient and utilizing non-maximum suppression and a double threshold algorithm, can accurately extract the single-pixel edge contour of the pipe end face. Based on this, the Hough circle transform algorithm is used to process the edge image. This algorithm performs voting statistics in the parameter space, enabling robust fitting of the pipe opening contours of the first pipe 10 and the second pipe 11 from discontinuous or partially occluded edge information, and accurately calculating the center coordinates of both. and radius The beneficial effect of this processing flow is that it can stably extract geometric features in complex unstructured backgrounds, providing reliable sub-pixel-level measurement data for subsequent deviation calculations, and overcoming the impact of lighting changes and pipe surface stains on recognition accuracy.

[0057] After obtaining the image features of the pipe opening, in order to transform the feature deviations on the two-dimensional image plane into mechanical motion commands in three-dimensional space, the system constructs an image-based visual servo control law. Its core lies in calculating the image Jacobian matrix, which describes the mapping relationship between the motion velocity of the robot's end effector, i.e., the pipe-gripping and docking device 2, in Cartesian space and the motion velocity of feature points on the image plane. Let the image feature vector... The corresponding image plane motion speed is The velocity of the camera relative to the target is... The relationship between the two is described by the following formula: ; In the above formula, This is the image Jacobian matrix. For a point on the image plane... When the depth is At that time, its Jacobian matrix The specific form is as follows: ; In the formula, This represents the depth value of the nozzle feature point in the camera coordinate system, which is estimated from the LiDAR data or the PnP algorithm in the first step. and This represents the normalized image plane coordinates. Based on this matrix, the control system calculates the required correction velocity vector. Control laws are typically designed as follows: ,in To control the gain, The pseudo-inverse of the image Jacobian matrix is... The current feature coordinates, The desired feature coordinates are the coordinates in the centering state. The beneficial effect of this algorithm is that it establishes a mathematical bridge between "what the vision sees" and "how the hand should move," enabling the control system to directly adjust the six-degree-of-freedom posture of the robotic arm in real time based on image errors, achieving robust control without the need for precise hand-eye calibration.

[0058] The calculated six-dimensional correction velocity vector The process needs to be further decomposed into drive commands for specific actuators. In this embodiment, the actuator is a hydraulic telescopic cylinder, thus requiring inverse kinematics analysis. The system maps the axial translational velocity component in the velocity vector to the axial feed command of the second telescopic cylinder 2201, controlling the insertion and retraction of the pipe. Simultaneously, the horizontal translational velocity component and the yaw angular velocity component around the vertical axis are mapped to the differential drive commands of the two symmetrically arranged first telescopic cylinders 24. That is, by controlling the extension of the left first telescopic cylinder and the retraction of the right first telescopic cylinder, or vice versa, the fixed claw 26 drives the first pipe 10 to perform left-right translation or yaw adjustment. The beneficial effect of this step is that it achieves decoupled control of multi-degree-of-freedom motion, transforming complex spatial attitude adjustment into a simple single-axis hydraulic cylinder telescopic action, greatly reducing the computational complexity of the control system and improving the system's response speed.

[0059] Finally, to balance docking efficiency and safety, the system employs a segmented control strategy combining coarse and fine adjustments. In the initial docking phase, when the distance between the two pipe openings exceeds a set threshold (e.g., 50 cm), the system uses a position-based control mode to drive the chassis 1 and the pipe-gripping docking device 2 to quickly approach the target position at a higher speed, minimizing non-operational time. When the distance falls below the set threshold and the system enters the final docking stage, it automatically switches to an incremental PI control mode based on image features. In this mode, the controller adjusts the control based on the current image feature error... Calculate control quantity The formula is as follows: ; In the above formula, This is the proportionality coefficient. The integral coefficient is... The error at the current moment, This represents the error from the previous moment. Incremental PI control not only eliminates static errors but also avoids overshoot caused by integral saturation. Through this strategy, the device achieves a "soft landing" docking while ensuring that the pipe end coaxiality deviation is typically less than 2 mm and the axial clearance meets stringent assembly tolerance requirements. The beneficial effects of this technology are that it effectively prevents inertial collisions caused by excessive speed, protects the fragile concrete pipe ends and rubber seals, and ensures the sealing performance and service life of the pipe connection.

[0060] In the preferred embodiment, the method deploys a layered computer software system on an embedded industrial computer and a PLC controller, deploys a robot operating system middleware in a Linux environment, loads the driver nodes of the first LiDAR 12 and the first camera 6 and the second camera 5, collects high-resolution image data through the GigE interface, collects point cloud data through the Ethernet interface, and publishes it as a topic message. A CUDA-accelerated TensorRT deep neural network inference engine is deployed in the GPU unit of the industrial control computer to run object detection algorithms and point cloud processing algorithms in real time. The state machine management program runs in the industrial control computer and is responsible for scheduling the work process from S1 to S4 and monitoring the health status of each subsystem in real time. When an abnormality is detected, the emergency stop logic is triggered through the watchdog mechanism. Establish real-time industrial Ethernet communication between the industrial computer and the PLC controller using Modbus-TCP or EtherCAT. The industrial computer packages and sends the calculated joint speed and position commands to the PLC. The PLC parses the commands into pulse signals or analog signals to drive each hydraulic valve group and servo motor driver. The system background runs a data log module that uploads the final position data, torque data, and image evidence from each docking to the cloud database and synchronously updates the virtual model's posture in the digital twin interface.

[0061] This embodiment employs a layered computer software system, achieving decoupling and efficient collaboration of complex perception, decision-making, and execution functions. The system is deployed on a dual-core hardware architecture of an embedded industrial computer and a programmable logic controller (PLC). For the industrial computer's software environment, the real-time and open-source Linux operating system was chosen as the underlying platform, upon which a robot operating system middleware was deployed. The advantage of this middleware architecture lies in its modular node communication mechanism. The system achieves standardized encapsulation of hardware interfaces by loading the driver nodes for the first LiDAR 12 and other cameras such as the first camera 6 and the second camera 5. Each sensor node acquires high-resolution image data through a Gigabit Ethernet interface (GigE interface) and a large amount of point cloud data through a standard Ethernet interface. This heterogeneous data is then uniformly encapsulated into topic messages and published on the system bus. This publish-subscribe model decouples data acquisition from data processing, greatly improving the system's scalability. When a new sensor needs to be added, only the corresponding driver node needs to be started and the topic subscribed to; there is no need to refactor the underlying code.

[0062] To meet the real-time requirements of processing massive amounts of visual data in industrial settings, this embodiment deploys a CUDA parallel computing architecture in the GPU unit of the industrial control computer. Leveraging the high concurrency of the CUDA architecture, the system runs the TensorRT deep neural network inference engine. This engine features operator fusion and memory optimization tailored to specific GPU hardware, specifically designed to accelerate the inference process of object detection and point cloud processing algorithms. In traditional CPU computing mode, processing a single frame of high-resolution point cloud or image typically takes hundreds of milliseconds, which is insufficient for closed-loop control requirements. However, using the hardware acceleration scheme described in this embodiment, the processing latency for each frame of data can be reduced to less than tens of milliseconds. The advantage of this technology is that it ensures low-latency visual feedback, enabling the pipe-grabbing docking device 2 to respond to environmental changes in real time during high-speed movement, achieving true dynamic servo control.

[0063] At the logic control level, a state machine management program runs on the industrial control computer. This program employs a finite state machine model, rigorously defining the transition logic and triggering conditions between the four operational processes: S1 environmental perception, S2 automatic pipe finding, S3 visual servo docking, and S4 closed-loop verification. The state machine management program not only handles task scheduling but also assumes the responsibility of system health monitoring. It polls the status heartbeat packets of each subsystem in real time, and once it detects fault signals such as sensor disconnection, communication timeout, or algorithm anomaly, the watchdog mechanism is immediately triggered. The watchdog mechanism, through hardware-level timer reset logic, forcibly cuts off the control output and triggers emergency stop logic in the event of software deadlock or anomaly. The beneficial effect of this design is that it endows the system with deterministic behavior patterns and fault-tolerant capabilities, preventing robotic arm malfunctions caused by software logic chaos or program crashes, and ensuring absolute safety at the construction site.

[0064] In the execution-level communication architecture, this embodiment establishes a high-speed communication link between the industrial control computer (ICC) and the PLC controller. The system adopts Modbus-TCP or EtherCAT real-time industrial Ethernet communication protocols. The ICC, acting as the host computer, is primarily responsible for complex trajectory planning and kinematic calculations, packaging and sending the calculated joint speed and position commands to the PLC. The PLC, acting as the slave computer, utilizes its highly reliable real-time kernel to parse the received digital commands into high-frequency pulse signals or analog voltage signals, directly driving the opening of the hydraulic valve group and the rotational speed of the servo motor driver. The advantage of using the EtherCAT protocol lies in its distributed clock synchronization mechanism, which ensures that the drivers of multiple axes operate synchronously within microsecond-level time errors. This is particularly crucial for the synchronous control of the dual-side traveling mechanism of the gantry frame, effectively preventing frame twisting or jamming caused by inconsistent speeds on both sides.

[0065] Finally, the system runs a data logging module in the background, enabling digital management of the construction process. This module automatically records key data for each docking process, including the final alignment coordinates, the torque curve of the hydraulic system, and image evidence of the docking completion moment, and uploads it to a cloud database via 4G or 5G networks. Simultaneously, the system supports digital twin technology, mapping on-site sensor data into a virtual 3D model in real time, and synchronously updating the equipment's attitude and movements on the remote monitoring interface. The beneficial effects of this technology are that it not only provides tamper-proof data evidence for project quality traceability but also allows managers to intuitively grasp the operational details within hidden trenches from a remote monitoring center, achieving transparency and intelligence in construction management.

[0066] Example 3 Further explanation in conjunction with Example 1, such as Figure 1-11 The structure shown uses a walking mechanism 3 to drive the entire frame 1 to move above the trench, positioning the frame 1 above the middle of the trench. The first lidar 12 and the first camera 6 work together to scan the front of the trench. The first lidar 12 and the first camera 6 jointly scan and create a 3D data map, storing the position coordinates and depth data of the trench.

[0067] First, using the visual data from the first LiDAR 12 and the first camera 6, along with the second camera 7, 3D shape data of the bottom surface of the trench is constructed. Then, pipe gripping is performed. The engineering vehicle has pre-positioned multiple pipes in the trench according to their installation sequence, maintaining a certain distance between the pipes and the traveling mechanism 3. The pipe gripping and docking device 2 operates according to the following logic: the sixth camera 13 on the side of the chassis 1 visually identifies the position of the first pipe 10; the first motor 2704 drives the sliding beam 2702 to move towards the side with the pipe; once the sliding beam 2702 reaches the critical position, the two first traveling motors 28 drive the pipe clamping beam 22 to move above the first pipe 10; then, the two first electric cranes... The machine 29 drives the pipe clamping beam 22 to descend, and the two fixed claws 26 on the pipe clamping beam 22 open through the third telescopic cylinder 2604 to clamp the first pipe body 10. At the same time, the docking claw 21 also opens and is located at the tail of the first pipe body 10, but it is not clamped at this time. Then, the two first electric cranes 29 drive the pipe clamping beam 22 to rise and lift the first pipe body 10. The first motor 2704 and the first traveling motor 28 drive the pipe clamping beam 22 to return to the position above the middle of the trench. Finally, combined with the 3D data map established by the first laser radar 12 and the first camera 6, the first electric crane 29 drives the pipe clamping beam 22 to descend, so that the first pipe body 10 reaches the trench installation position and is placed in the trench.

[0068] The first tube 10 is docked with the already installed second tube 11. The specific steps include: using a second camera 5 to visually monitor the position of the second tube 11 opening, with the docking claw 21 in an open state, the second telescopic cylinder 2201 drives the telescopic beam 23 to extend so that the docking claw 21 reaches the position of the second tube 11, and then the docking claw 21 clamps the second tube 11; after the openings of the first tube 10 and the second tube 11 are aligned end to end, the two symmetrically installed first telescopic cylinders 24 begin to drive, with one telescopic cylinder pushing and the other... The telescopic cylinder retracts, driving the two fixed claws 26 to dock the tail of the first tube 10 with the head opening of the second tube 11. During this process, the first camera 4 monitors the docking process between the first tube 10 and the second tube 11. After docking, the second laser radar 401 measures the distance between itself and the opening of the second tube 11. When the distance reaches the standard, it indicates that the docking is complete. After docking, the two fixed claws 26 and the docking claw 21 open, and the steps of placing and docking the tubes in S1 to S3 are repeated.

[0069] The above embodiments are merely preferred technical solutions of the present invention and should not be considered as limitations on the present invention. The scope of protection of the present invention should be limited to the technical solutions described in the claims, including equivalent substitutions of the technical features described in the claims. That is, equivalent substitutions and improvements within this scope are also within the scope of protection of the present invention.

Claims

1. A trench pipe connection device, characterized in that: The vehicle includes a gantry frame (1), with a traveling mechanism (3) symmetrically arranged on both sides of the frame (1), and a pipe-gripping docking device (2) in the middle of the gantry frame (1). The pipe-gripping docking device (2) has a first camera (4) symmetrically arranged on both sides above the fixed claw (26) near the docking claw (21), and a second laser radar (401) is arranged above the first camera (4). The first camera (4) and the second laser radar (401) face the pipe docking position. The bottom rear side of the frame (1) is symmetrically equipped with second cameras (5), which face the pipe connection position; The side of the frame (1) is also equipped with a sixth camera (13), which is used to detect the position of the side pipe and to use the fixing claw (26) of the pipe grabbing and docking device (2) to grab the pipe.

2. The trench pipe docking device according to claim 1, characterized in that: The front end of the frame (1) is equipped with two first lidars (12) and two first cameras (6) arranged symmetrically, and the bottom beam (101) is equipped with a second camera (7) facing the bottom of the ditch. The frame (1) has a fifth camera (9) on both sides of the top of the middle part of the gantry. The fifth camera (9) faces the pipe-grabbing docking device (2) where the pipe is installed.

3. The apparatus of claim 1 wherein the gripping means comprises a pair of jaws. The pipe docking device (2) includes a pipe clamping beam (22), which is sleeved with a telescopic beam (23). The pipe clamping beam (22) is equipped with a second telescopic cylinder (2201) inside, which drives the telescopic beam (23) to extend and retract. The telescopic beam (23) is provided with a docking claw (21) at the end, and two fixed claws (26) are symmetrically provided at both ends of the clamping crossbeam (22). The fixed claws (26) are slidably connected to the clamping crossbeam (22), and the first telescopic cylinder (24) is hinged on the clamping crossbeam (22). The telescopic end of the first telescopic cylinder (24) is hinged to the fixed claws (26). Two first telescopic cylinders (24) work together to drive two fixed claws (26) to slide on the clamping crossbeam (22).

4. The apparatus of claim 3 wherein: The fixed claw (26) includes a claw frame (2602), with an inwardly inverted sliding frame (2601) on the upper part. The clamping tube beam (22) has first slide rails (25) on both sides. The sliding frame (2601) is slidably connected to the first slide rails (25). The claw frame (2602) has two symmetrically arranged movable frames (2606) below it. The movable frames (2606) have an arc-shaped claw plate (2607) at the lower end. The movable sliding frames (2608) on both sides of the upper part of the movable frames (2606) are slidably connected to the sliding strips (2603) on both sides below the claw frame (2602). The gripper frame (2602) is provided with at least two third telescopic cylinders (2604) inside. The bottom of the gripper frame (2602) is also provided with a waist-shaped hole (2605). The telescopic end of the third telescopic cylinder (2604) passes through the waist-shaped hole (2605) and is connected to the movable frame (2606). The gripper frame (2602) is also provided with an arc-shaped top plate (2609) with an arc structure in the middle of the bottom. The surfaces of the top arc plate (2609) and the gripper plate (2607) are both provided with anti-slip pads.

5. The trench pipe docking device according to claim 1, characterized in that: The structure of the frame (1) is as follows: multiple legs (103) at the bottom of the portal frame are connected to the bottom beam (101) of the rectangular frame structure, and a single walking mechanism (3) is connected to a single leg (103); The structure of the walking mechanism (3) is as follows: one end of the support connecting frame (301) is hinged to the outrigger (103), the other end of the support connecting frame (301) is connected to the lifting outrigger (303), and a fourth telescopic cylinder (302) is hinged on the outrigger (103). The end of the fourth telescopic cylinder (302) is hinged to the middle of the lifting outrigger (303). The lifting outrigger (303) is equipped with a fifth telescopic cylinder (304). The lower telescopic end of the lifting outrigger (303) is connected to the rotary drive device (305). The rotary drive device (305) uses a hydraulic drive worm (307) to drive the worm wheel (306). The worm wheel (306) is connected to the inverted U-shaped frame (308). The two sides of the groove at the bottom of the U-shaped frame (308) are connected to the track wheels (309).

6. The pipe clamping butt joining method of the trench pipe butt joining apparatus according to any one of claims 1 to 5, characterized by: The method includes: S1. Use the walking mechanism (3) to move the vehicle frame (1) to the working position, collect data through the first laser radar (12) and the first camera (6), construct a 3D model of the trench environment and store the depth data; S2. Based on the visual recognition data of the sixth camera (13), calculate the spatial coordinates of the first pipe (10), drive the pipe grabbing and docking device (2) to move and use the fixed claw (26) and docking claw (21) to grab and lift the first pipe (10) and place it in the preset installation position in the trench. S3. Using the second camera (5) and the first camera (4) to monitor the relative position of the first pipe (10) and the installed second pipe (11) in real time, drive the first telescopic cylinder (24) and the second telescopic cylinder (2201) to make fine adjustments until the tail of the first pipe (10) is aligned with the head of the second pipe (11); S4. Use the second laser radar (401) to measure the distance between the nozzle and the nozzle. After the docking is completed, release the fixing claw (26) and the docking claw (21) and repeat steps S1 to S4.

7. A method of clamping and butting pipes for a trench pipe butt joint device according to claim 6, Its characteristic is: In step S1: By pre-establishing the joint calibration extrinsic parameter matrix of the first lidar (12) and the first camera (6), the point cloud data in the coordinate system of the first lidar (12) is mapped to the pixel coordinate system of the first camera (6) through the rigid body transformation matrix to achieve spatial alignment between the point cloud and the image pixels. A voxel grid filter is used to downsample the original point cloud, and a statistical outlier removal algorithm is used to remove noise data caused by dust or reflection. The random sampling consensus algorithm is used to perform plane fitting on the processed point cloud data, segmenting the planar models of the two side walls of the trench and the planar model of the bottom of the trench, and calculating the equation of the central axis of the trench and the depth parameters. By integrating the close-up texture images captured by the second camera (7) at the bottom beam (101), the micro-topography at the bottom of the trench is reconstructed using the motion recovery structure algorithm. Areas with flatness variance exceeding the threshold are marked in the 3D model, and a grid map of the drivable and placeable areas is generated.

8. The pipe clamping and docking method of the trench pipe docking device according to claim 6, Its characteristic is: In step S2: An improved YOLO series deep learning object detection network is deployed in the image processing unit of the sixth camera (13) to select the region of interest of the first pipe (10) in the image and output the two-dimensional pixel bounding box and confidence score of the pipe. Combining the prior geometric dimensions of the first pipe (10) with the pixel coordinates of the bounding box, the perspective n-point pose calculation algorithm is used to calculate the rotation matrix and translation vector of the first pipe (10) relative to the sixth camera (13), and the 6-DOF pose of the pipe center point in the frame (1) coordinate system is determined. The motion planning module generates a smooth motion trajectory using a fifth-order polynomial interpolation method based on the current position of the first pipe (10) and the target placement position, and superimposes the input shaping algorithm to suppress the residual vibration of the clamping beam (22) and the first pipe (10) during the hoisting process; During the closing process of the fixed claw (26), the controller collects the pressure sensor data of the hydraulic circuit in real time and uses the fuzzy PID control algorithm to adjust the output pressure of the third telescopic cylinder (2604) to ensure that the clamping force is maintained within the set range that neither damages the surface of the first pipe (10) nor causes slippage.

9. The pipe clamping and docking method of the trench pipe docking device according to claim 6, characterized in that: In step S3: Gaussian filtering is applied to the images captured by the second camera (5) and the first camera (4) to remove noise. The Canny operator is used to extract the image edges. The Hough circle transform algorithm is used to accurately locate the center and radius of the pipe opening contours of the first pipe (10) and the second pipe (11). Construct an image-based visual servo control law, and calculate the image Jacobian matrix between the image feature deviation of the nozzle center distance deviation and radius deviation and the motion speed of the robot end effector containing the docking claw (21) and the fixed claw (26); The calculated six-dimensional correction velocity vector is decomposed into the differential drive command of the first telescopic cylinder (24) and the axial feed command of the second telescopic cylinder (2201); A strategy combining coarse and fine adjustment is adopted. When the distance is greater than the set threshold, positional control is used to quickly approach the target. When the distance is less than the set threshold, incremental PI control based on image features is switched until the coaxiality deviation and axial clearance of the pipe end meet the assembly tolerance requirements.

10. The method of claim 6-9, wherein the method further comprises: This method deploys a layered computer software system in an embedded industrial computer and a PLC controller, deploys a robot operating system middleware in a Linux environment, loads the first laser radar (12) driver node and the first camera (6) and the second camera (5) and other camera driver nodes, collects high-resolution image data through the GigE interface, collects point cloud data through the Ethernet interface, and publishes it as a topic message; A CUDA-accelerated TensorRT deep neural network inference engine is deployed in the GPU unit of the industrial control computer to run object detection algorithms and point cloud processing algorithms in real time. The state machine management program runs in the industrial control computer and is responsible for scheduling the work process from S1 to S4 and monitoring the health status of each subsystem in real time. When an abnormality is detected, the emergency stop logic is triggered through the watchdog mechanism. Establish real-time industrial Ethernet communication between the industrial computer and the PLC controller using Modbus-TCP or EtherCAT. The industrial computer packages and sends the calculated joint speed and position commands to the PLC. The PLC parses the commands into pulse signals or analog signals to drive each hydraulic valve group and servo motor driver. The system background runs a data log module that uploads the final position data, torque data, and image evidence from each docking to the cloud database and synchronously updates the virtual model's posture in the digital twin interface.

Images