Pose compensation method for mounting overhead line system cantilever based on 3D visual guidance and automatic mounting system

Abstract

The invention discloses a position and posture compensation method for guiding installation of a contact network cantilever based on 3D vision and an automatic installation system. The position and posture compensation method comprises the steps that the spatial transformation relation between a 3D vision sensor coordinate system and a robot coordinate system is determined; the robot is controlled to move, and the three-dimensional point cloud data of the to-be-installed cantilever component and the three-dimensional point cloud data of the field installation foundation are collected; the collected three-dimensional point cloud data is registered with a corresponding preset CAD model, and a first real-time pose of the cantilever component in a robot coordinate system and a second real-time pose of the installation foundation in the robot coordinate system are obtained through calculation; calculating a pose compensation transformation quantity based on a pre-stored theoretical mounting pose of the cantilever, the first real-time pose and the second real-time pose; and according to the pose compensation transformation amount, planning and executing a motion track of the robot, and completing butt joint installation. The method is high in compensation precision, robustness and adaptability and high in construction efficiency, and the problem of uncertainty in automatic installation is fundamentally solved.

Description

Technical Field

[0001] This invention relates to the fields of electrified railway construction automation and machine vision precision measurement technology, and in particular to a pose compensation method and automated installation system for catenary cantilever installation guided by 3D vision. Background Technology

[0002] The overhead contact system of electrified railways is a key infrastructure for ensuring stable and reliable power supply to electric locomotives. As the core supporting component of the overhead contact system, the accuracy of the installation posture (i.e., spatial position and attitude) of the cantilever arm directly determines key geometric parameters such as the conductor height and pull-out value of the contact wire, which in turn affects the sliding quality, wear life, and even train operation safety of the pantograph.

[0003] Currently, manual labor remains the mainstream construction method for cantilever arm installation both domestically and internationally. The typical process is as follows: First, based on the design drawings, construction workers use traditional measuring tools such as optical theodolites, levels, and steel tape measures to manually mark the approximate installation position of the cantilever arm base on the support column. Then, a crane is used to hoist the cantilever arm to a height, where several workers on an aerial work platform repeatedly and roughly adjust its position using methods such as visual inspection, prying with crowbars, and plumb bobs, ultimately achieving connection and fastening with the support column.

[0004] The traditional manual operation mode has the following inherent defects: ① Difficulty in guaranteeing accuracy: The limitations of human visual resolution and manual tools result in installation errors typically at the centimeter level, making it difficult to meet the stringent millimeter-level installation accuracy requirements of high-speed railways. Even minute positional deviations, amplified by the contact network system, can cause significant changes in the contact wire position, creating safety hazards. ② Low work efficiency: Measurement, marking, hoisting, adjustment, re-measurement, and fastening are all interconnected and highly dependent on worker coordination, making the entire process cumbersome and time-consuming. Especially in large hub stations, where a large number of cantilever arms need to be installed, manual methods severely restrict construction progress. ③ Significant safety risks: The high-altitude, heavy-load, and multi-trade working environment poses a continuous threat to the personal safety of construction workers. ④ Poor quality consistency: Installation quality relies excessively on the technical level and experience of construction workers, leading to significant fluctuations in installation results across different shifts and time periods, making standardization and digital management difficult. ⑤ High labor intensity: This is a complex labor activity requiring heavy physical labor and high technical skills, facing pressure from labor shortages and rising costs.

[0005] In recent years, with the development of industrial robots and BIM technology, automated solutions using robotic arms for cantilever installation have emerged. However, these solutions face a core bottleneck: the mismatch between the theoretical model and the physical reality of the position. Specifically, this manifests as: ① Manufacturing and installation errors of the support column: During the prefabrication and on-site installation of concrete or steel columns, tolerances inevitably exist in their spatial position, verticality, and torsion angle. ② Foundation construction errors: There are construction tolerances in the position and elevation of the support column foundation. ③ Environmental and load deformation: Temperature, wind load, and long-term operation may cause minor settlement or deformation of the foundation.

[0006] These factors result in a significant six-degree-of-freedom pose deviation between the actual installation interface of the cantilever arm (i.e., the connecting flange or bolt holes on the support column) and its theoretical design position in the BIM model. If the robotic arm can only move "blindly" according to the theoretical coordinate trajectory, it will lead to the cantilever arm failing to connect, mechanical interference, or generating huge assembly stress, seriously affecting structural safety and the automation process. Therefore, developing a method that can accurately sense and intelligently compensate for this pose deviation in real time is an indispensable core technology for achieving high-quality, fully automated installation of the cantilever arm. Summary of the Invention

[0007] To overcome the aforementioned problems in the prior art, this invention proposes a pose compensation method and automated installation system for the installation of catenary arms guided by 3D vision.

[0008] The technical solution adopted by this invention to solve its technical problem is: a pose compensation method for the installation of a catenary arm based on 3D vision guidance, comprising the following steps: Step 1, Calibration and Preparation: Obtain the transformation matrix from the 3D vision sensor coordinate system to the robotic arm end effector coordinate system; at the same time, determine the mapping relationship between the robotic arm base coordinate system and the global BIM model coordinate system; Step 2, 3D scene data acquisition: Perform 3D scanning on the cantilever arm to be installed, which is stationary on the tool rack, and the support foundation on site, and simultaneously acquire high-resolution point clouds of the cantilever arm. and installation of basic point cloud ; Step 3, Point Cloud Preprocessing and Pose Calculation: The point cloud obtained in Step 2 is processed... and installation of basic point cloud Preprocessing is performed using a strategy that combines feature-based coarse registration with improved iterative nearest-point algorithm for fine registration, to process the carpal point cloud after step 2. and the processed installation base point cloud With the CAD model of the wrist arm and basic CAD model Perform registration to determine the actual pose of the wrist arm in the current coordinate system. The actual position of the installation foundation ; Step 4, Compensation Calculation: Using the transformation relationship calibrated in Step 1, the... and The coordinates are uniformly converted to the robot arm's base coordinate system, and the theoretical installation pose is read from the BIM database. Substitute the compensation amount to calculate the pose compensation transformation matrix ΔT that the robotic arm end needs to perform; Step 5, Motion trajectory planning and compensation execution: Based on the pose compensation transformation matrix ΔT obtained in Step 4, plan and execute the robot's motion trajectory to accurately complete the installation and docking.

[0009] The pose compensation method for the installation of the contact wire cantilever arm based on 3D vision guidance described above, specifically step 1 is as follows: Step 1.1: Fix a standard calibration plate in the workspace of the robotic arm. Control the end effector of the robotic arm to move the calibration plate to multiple different poses. In each pose, the 3D scanner scans the calibration plate and obtains its point cloud. By solving the hand-eye calibration equation, calculate the transformation matrix from the 3D scanner coordinate system to the robotic arm base coordinate system. ; Step 1.2: Measure the on-site control points using a total station to establish the transformation relationship between the robotic arm's base coordinate system and the global BIM model coordinate system. .

[0010] In the aforementioned pose compensation method for the installation of a catenary arm guided by 3D vision, step 3 specifically comprises: Step 3.1, perform wrist-arm point cloud formation. and installation of basic point cloud Statistical filtering algorithms are applied to remove isolated noise points; then, voxel grid filters are used for downsampling to reduce the amount of data while preserving shape features. Step 3.2: Using the sample consistency initial registration algorithm, based on the fast point feature histogram features, the processed carpal point cloud is processed. and wrist CAD model Initial pose estimation is performed to obtain a preliminary transformation matrix, which is then applied to the processed installation foundation point cloud. and basic CAD model Perform initial pose estimation to obtain a preliminary transformation matrix; Step 3.3: Using the preliminary transformation matrix obtained in Step 3.2 as the initial value, execute the ICP algorithm. Iteratively find the nearest neighbor pair between two point clouds and minimize their distance error, converging to a high-precision transformation matrix. and .

[0011] The above-mentioned pose compensation method for the installation of a catenary arm based on 3D vision guidance, wherein the transformation matrix in step 3.3... and The solution process is the same, the above The specific solution process is as follows: The installation foundation point cloud obtained in step 3.1... For each point, transform it to the underlying CAD model using the current transformation matrix. coordinate system, in Find the nearest point in the middle and record the corresponding point pair; Based on distance threshold Threshold of the angle between the normal vector and the normal vector Filter the point pairs and calculate the weight of each point pair; Construct a weighted point-to-surface error objective function and use a robust kernel function to reduce the impact of outliers; Linearize the objective function, construct the Jacobian matrix and residual vector for each point pair, and form a linear system; Solve the linear system to obtain the transformation increment, and update the transformation matrix. If the norm of the transformation increment is less than a threshold, stop the iteration and obtain the transformation matrix.

[0012] The pose compensation method for the installation of a catenary arm based on 3D vision guidance described above, wherein the formula for calculating the pose compensation transformation matrix ΔT in step 4 is as follows: Here, inv(·) represents the inverse of the matrix.

[0013] The pose compensation method for catenary arm installation based on 3D vision guidance described above, specifically step 5, involves: parsing the pose compensation transformation matrix ΔT into a three-dimensional translation compensation vector [Δx, Δy, Δz] and a three-dimensional rotation compensation vector at the end of the robotic arm in the base coordinate system; generating a Cartesian space motion trajectory from the robot's current position to the target position based on the three-dimensional translation compensation vector and the three-dimensional rotation compensation vector; and controlling the robot to move along the motion trajectory to complete the precise installation of the catenary arm.

[0014] The aforementioned pose compensation method for the installation of a contact wire cantilever arm based on 3D vision guidance includes an installation verification step after step 5. This installation verification step specifically involves: re-scanning the installed cantilever arm using a 3D vision sensor, obtaining the actual installation pose through point cloud registration, and comparing it with the theoretical installation pose. The comparison is performed to generate an installation quality report.

[0015] An automated wrist arm mounting system for implementing the method described above, the system comprising: an industrial robot having a dedicated actuator at its end for gripping the wrist arm; A 3D vision sensor is fixedly installed at the end of the robot or at a fixed position within the robot's workspace; Central control unit, the central control unit comprising: The storage module is used to store the CAD model, BIM data, and theoretical installation pose of the cantilever arm and its mounting foundation. ; The point cloud processing module is used to perform point cloud registration and real-time pose calculation; The compensation calculation module is used to calculate the pose compensation transformation amount ΔT; The motion control module is used to plan the robot's motion trajectory and drive its movement.

[0016] In the aforementioned system, the central control unit is also communicatively connected to a BIM data server and / or a monitoring system at the construction site to achieve data synchronization and interaction.

[0017] The beneficial effects of this invention are: ① This invention achieves a true "perception-decision-action" closed loop: using 3D vision as the perception front end, BIM as the decision-making brain, and robots as the execution end, a complete intelligent installation system is formed, fundamentally solving the uncertainty problem in automated installation. ② High compensation accuracy: Based on high-precision 3D point clouds and robust registration algorithms, the pose measurement accuracy can reach sub-millimeter level, far surpassing manual and traditional measurement methods, ensuring the smoothness of the contact network system. ③ Significantly improved automation and efficiency: Full-process automation liberates workers from heavy and dangerous high-altitude operations, shortening the single installation cycle to the minute level, greatly improving construction efficiency. ④ Strong robustness and adaptability: It can automatically adapt to random pose deviations caused by various factors on site, enabling the automated system to cope with complex on-site conditions. ⑤ Empowering digital construction and operation and maintenance: This method naturally generates accurate as-built data for each cantilever installation, providing valuable data assets for subsequent digital delivery, intelligent operation and maintenance, and reverse modeling. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the process of this invention; Figure 2 This is a schematic diagram of the automated installation system of the present invention; Figure 3 This is a coordinate system transformation diagram of the pose compensation calculation principle of this invention; Figure 4 This is a schematic diagram of the point cloud registration process of the present invention; Figure 5 This is a schematic diagram of the trajectory of the robotic arm performing the compensation motion according to the present invention. Detailed Implementation

[0019] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0020] like Figure 1 As shown, this embodiment discloses a pose compensation method for the installation of a catenary arm based on 3D vision guidance, including the following steps: Step 1, Calibration and Preparation: Obtain the transformation matrix from the 3D vision sensor coordinate system to the robotic arm end effector coordinate system; at the same time, determine the mapping relationship between the robotic arm base coordinate system and the global BIM model coordinate system.

[0021] An "eye-to-hand" calibration method is employed. A standard calibration plate is fixed within the workspace of the robotic arm. The robotic arm's end effector moves the calibration plate to multiple different poses. In each pose, a 3D scanner scans the calibration plate and obtains its point cloud. The hand-eye calibration equation is solved as follows: AX = XB In the formula: A For the motion transformation of the robot end effector, it represents the transformation of the end effector from pose 1 to pose 2; B The motion transformation observed by the camera represents the pose transformation of the same fixed calibration plate observed when the camera moves from pose 2 to pose 1. X Let be the "hand-eye transformation matrix" to be solved, which represents the rigid body transformation from the camera coordinate system to the robot end effector coordinate system; A , B , X All are homogeneous transformation matrices.

[0022] Accurately calculate the transformation matrix from the 3D scanner coordinate system to the robotic arm base coordinate system. Simultaneously, using precision surveying instruments such as a total station, a few control points on site were measured to establish the transformation relationship between the robotic arm's base coordinate system and the global BIM model coordinate system. .

[0023] Step 2, 3D scene data acquisition: Perform 3D scanning on the cantilever arm to be installed, which is stationary on the tool rack, and the support foundation on site, and simultaneously acquire high-resolution point clouds of the cantilever arm. and installation of basic point cloud .

[0024] The wrist arm is pre-positioned on the tool holder in a known approximate pose. The robotic arm first moves to a predefined "wrist arm scanning pose" capable of fully scanning the wrist arm, triggering the 3D scanner to acquire a complete point cloud containing all key connecting components of the wrist arm. P armSubsequently, the robotic arm moves to a "foundation scanning pose" that allows for a clear scan of the support mounting base (such as the mounting surface of the cantilever arm base and the bolt group), triggering a scan again to acquire point clouds. P base .

[0025] Step 3, Point Cloud Preprocessing and Pose Calculation: The point cloud obtained in Step 2 is processed... and installation of basic point cloud Preprocessing is performed using a strategy that combines feature-based coarse registration with improved iterative nearest-point algorithm for fine registration, to process the carpal point cloud after step 2. and the processed installation base point cloud With the CAD model of the wrist arm and basic CAD model Perform registration to determine the actual pose of the wrist arm in the current coordinate system. The actual position of the installation foundation The specific process is as follows: ① Preprocessing. For P arm and P base Statistical filtering algorithms are applied to remove isolated noise points; then, voxel grid filters are used for downsampling to reduce the amount of data while preserving shape features.

[0026] ② Feature extraction and coarse registration. For the carpal point cloud, its CAD model is known. M arm First, using the SAC-IA (Sample Consistency Initial Registration) algorithm, based on FPFH (Fast Point Feature Histogram) features, the data is processed... P arm and M arm Initial pose estimation is performed to obtain a preliminary transformation matrix. T coarse This step effectively reduces the pose deviation between the point cloud and the model to within the convergence region of the ICP algorithm.

[0027] ③ Precise registration. (The previous step...) T coarse Using these as initial values, the ICP algorithm is executed. This algorithm iteratively finds the nearest neighbor pair between two point clouds and minimizes their distance error, ultimately converging to a high-precision transformation matrix. For installing basic point cloud P base Using the same process as M base Registration was performed to obtain .

[0028] The registration solution process in this embodiment is as follows: Figure 4As shown, specifically, it is necessary to solve for both the actual pose of the wrist arm and the actual pose of the mounting base, but usually the pose of the wrist arm relative to the mounting base is of greater concern. In practice, the point clouds of the wrist arm and the mounting base may be registered with their CAD models separately, and then the relative poses are obtained through coordinate system transformation.

[0029] For simplicity, we assume that the transformation relationship between the camera and the robotic arm has been obtained through hand-eye calibration, and that the point cloud can be transformed into the robotic arm's base coordinate system.

[0030] Specific steps: Let the arm and arm point cloud be... Install basic point cloud for Their CAD model point clouds are respectively and ; The point clouds of the wrist arm and the mounting base are registered with their respective CAD models to obtain their poses in their respective model coordinate systems. These poses are then transformed to the robot arm's base coordinate system. The pose of the wrist arm relative to the mounting base is calculated using these two poses.

[0031] However, in this embodiment, a more direct method is to directly register the carpal point cloud and the mounting base point cloud, or to register the carpal point cloud to the CAD model of the mounting base point cloud (because the mounting base is fixed and its CAD model is known). This embodiment uses the method of registering the carpal point cloud (source point cloud) to the mounting base point cloud (target point cloud).

[0032] The detailed registration process is as follows: Initialization: Set the initial transformation T 0 can be the identity matrix, or it can be set based on the robot arm's initial pose. This sets the maximum number of iterations. max_iter and convergence threshold epsilon For each iteration k = 0,1,..., max_iter- 1.

[0033] a. Using the current transformation T k Transform the source point cloud (arm point cloud) into the coordinate system of the target point cloud (mounting base point cloud): P arm-transformed = T k * P arm .

[0034] b. For each point after transformation p i exist P arm-transformed Find the nearest point in the target point cloud.q j Calculate the distance between point pairs and the angle between their normal vectors.

[0035] c. Filter valid point pairs based on distance threshold and normal vector angle threshold.

[0036] d. Calculate the weights for each valid point pair, including distance weight, normal vector weight, and curvature weight.

[0037] e. Construct a weighted point-to-surface error objective function and use a robust kernel function (such as the Huber kernel) to reduce the impact of outliers.

[0038] f. Linearize the objective function, construct the Jacobian matrix and residual vector, and form a linear system.

[0039] g. Solve the linear system to obtain the transformation increment Δ T .

[0040] h. Update Transformation: .

[0041] i. Check the convergence condition: If the norm of the transformation increment is less than the threshold. epsilon If the iteration stops, then stop.

[0042] Final transformation T This refers to the position of the wrist arm relative to the mounting base.

[0043] In this embodiment, it may also be necessary to obtain the actual pose of the installation foundation, as the installation foundation may deviate from the theoretical model due to construction errors. Therefore, it is also necessary to register the point cloud of the installation foundation with its CAD model to obtain the actual pose of the installation foundation.

[0044] Therefore, two registrations are actually required: Registration 1: Installing the basic point cloud With installation base CAD model The actual position of the installation foundation is obtained. T base .

[0045] Registration 2: Wrist-Arm Dot Cloud With the CAD model of the wrist arm To obtain the actual position of the wrist and arm T arm Then, the pose of the wrist arm relative to the mounting base is: .

[0046] However, by directly using the mounting base point cloud as the target point cloud and registering the carpal point cloud to the mounting base point cloud, the pose of the carpal arm relative to the mounting base can be obtained directly. However, this method leaves the discrepancy between the actual pose of the mounting foundation and the theoretical model unknown. Therefore, the invention employs a separate registration method.

[0047] The detailed processes for the two registrations are given below (taking the installation base registration as an example; the same applies to the carpal-arm registration).

[0048] Install basic point cloud and CAD model registration: Input: Install basic point cloud Install the basic CAD model Initial pose T base0 (This can be obtained from the current position of the robotic arm, or by using a coarse registration method); Output: The actual pose of the mounting base. T base .

[0049] The specific steps are as follows: Preprocessing: for and Perform downsampling and noise reduction, and calculate the normal vector.

[0050] Coarse registration: If the initial pose T base0 If the initial pose is not accurate enough, use feature matching (such as FPFH) and RANSAC for coarse registration to obtain a better initial pose.

[0051] Fine registration: Using an improved ICP algorithm. Improved ICP algorithm steps: Initialization: T = T base0； for k = 0 to max_iter -1.

[0052] a. Recent point search: For For each point in the array, use the current transformation. T Transform to the model coordinate system, and then... Find the nearest point in the middle and record the corresponding point pair.

[0053] b. Filter point pairs: based on distance threshold Threshold of the angle between the normal vector and the normal vector Filter point pairs.

[0054] Distance threshold is expressed as The angle threshold is expressed as .in d max , d min These are the maximum and minimum distances, respectively. It is the attenuation constant. , These are the maximum and minimum included angles, k This represents the number of iterations.

[0055] c. Calculate the weight of each pair of points: Where: distance weight Normal vector weight , curvature weight . here It is a local curvature characteristic value.

[0056] d. Construct the objective function: weight the point-to-surface error and use the Huber kernel function: In the formula: For an effective set of corresponding points, , For Huber kernel function, ,otherwise, .

[0057] e. Linearization: Compute the Jacobian matrix for each pair of points. (6 dimensions, the first 3 dimensions are about rotation, the last 3 dimensions are about translation), residual .

[0058] f. Constructing a linear system: , .

[0059] g. Solving linear systems: .

[0060] h. Update Transformation: (Here, exp is the Lie group exponent mapping, which maps Lie algebras) (Convert to a transformation matrix).

[0061] i. If If the condition is met, the loop will exit. Return. T base = T .

[0062] Similarly, by registering the wrist and arm point clouds, we obtain... T arm .

[0063] Step 4, Compensation Calculation: Using the transformation relationship calibrated in Step 1, the... and Transform to the robot arm's base coordinate system: Retrieves theoretical installation pose from BIM database Substitute the compensation amount into the formula to calculate the pose compensation transformation matrix ΔT required by the robotic arm's end effector. The specific calculation formula is as follows: in inv (·) denotes the inverse of the matrix. The calculated Δ T It is a 4x4 homogeneous transformation matrix.

[0064] Step 5, Motion Trajectory Planning and Compensation Execution: Based on the pose compensation transformation matrix ΔT obtained in Step 4, plan and execute the robot's motion trajectory to accurately complete the installation and docking. The coordinate system transformation relationship for pose compensation calculation is as follows: Figure 3 As shown. The trajectory of the robotic arm performing the compensation motion is as follows. Figure 5 As shown.

[0065] The control system analyzes ΔT. For example, it decomposes it into a translation vector [dx, dy, dz] = [ΔT (1,4), ΔT (2,4), ΔT (3,4)] and ZYX Euler angles [dA,dB, dC] obtained by the rotation matrix ΔT (1:3, 1:3). Based on the current pose of the robotic arm and the compensated target pose (current pose * ΔT), the motion planner plans a smooth and safe Cartesian linear motion trajectory. The robotic arm strictly follows this trajectory, driving the wrist arm to move precisely, ensuring perfect alignment of the wrist arm's connection hole with the bolt holes of the support foundation. Finally, the bolts are tightened manually or using automated tooling.

[0066] Step 6, Effect Verification: After tightening the bolts, the robotic arm moves again to the "Verification Scan Pose" position to perform a final scan of the installed wrist arm. The actual completed pose (Tas-built) is obtained through point cloud registration and compared with Tdesign to generate a digital acceptance report containing pose deviation data.

[0067] like Figure 2 As shown, this embodiment also discloses an automated wrist arm installation system, including: an industrial robot with a dedicated actuator for gripping the wrist arm at its end; A 3D vision sensor is fixedly installed at a fixed position on the end effector of the robot or within the robot's workspace; the 3D vision sensor is one of a lidar, a structured light 3D camera, or a binocular stereo vision system.

[0068] Central control unit, the central control unit comprising: The storage module is used to store the CAD model, BIM data, and theoretical installation pose of the cantilever arm and its mounting foundation. ; The point cloud processing module is used to perform point cloud registration and real-time pose calculation; The compensation calculation module is used to calculate the pose compensation transformation amount ΔT; The motion control module is used to plan the robot's motion trajectory and drive its movement.

[0069] The central control unit is also communicatively connected to a BIM data server and / or a monitoring system at the construction site to enable data synchronization and interaction.

[0070] The above embodiments are merely exemplary embodiments of the present invention and are not intended to limit the present invention. Those skilled in the art can make various modifications or equivalent substitutions to the present invention within its scope and spirit, and such modifications or equivalent substitutions should also be considered to fall within the scope of protection of the present invention.

Claims

1. A pose compensation method for the installation of a catenary arm based on 3D vision guidance, characterized in that, Includes the following steps: Step 1, Calibration and Preparation: Obtain the transformation matrix from the 3D vision sensor coordinate system to the robotic arm end effector coordinate system; at the same time, determine the mapping relationship between the robotic arm base coordinate system and the global BIM model coordinate system; Step 2, 3D scene data acquisition: Perform 3D scanning on the cantilever arm to be installed, which is stationary on the tool rack, and the support foundation on site, and simultaneously acquire high-resolution point clouds of the cantilever arm. and installation of basic point cloud ; Step 3, Point Cloud Preprocessing and Pose Calculation: The point cloud obtained in Step 2 is processed... and installation of basic point cloud Preprocessing is performed using a strategy that combines feature-based coarse registration with improved iterative nearest-point algorithm for fine registration, to process the carpal point cloud after step 2. and the processed installation base point cloud With the CAD model of the wrist arm and basic CAD model Perform registration to determine the actual pose of the wrist arm in the current coordinate system. The actual position of the installation foundation ; Step 4, Compensation Calculation: Using the transformation relationship calibrated in Step 1, the... and The coordinates are uniformly converted to the robot arm's base coordinate system, and the theoretical installation pose is read from the BIM database. Substitute the compensation amount to calculate the pose compensation transformation matrix ΔT that the robotic arm end needs to perform; Step 5, Motion trajectory planning and compensation execution: Based on the pose compensation transformation matrix ΔT obtained in Step 4, plan and execute the robot's motion trajectory to accurately complete the installation and docking.

2. The pose compensation method for the installation of a catenary arm based on 3D vision guidance as described in claim 1, characterized in that, Step 1 specifically involves: Step 1.1: Fix a standard calibration plate in the workspace of the robotic arm. Control the end effector of the robotic arm to move the calibration plate to multiple different poses. In each pose, the 3D scanner scans the calibration plate and obtains its point cloud. By solving the hand-eye calibration equation, calculate the transformation matrix from the 3D scanner coordinate system to the robotic arm base coordinate system. ; Step 1.2: Measure the on-site control points using a total station to establish the transformation relationship between the robotic arm's base coordinate system and the global BIM model coordinate system. .

3. The pose compensation method for the installation of a contact wire cantilever arm based on 3D vision guidance according to claim 1, characterized in that, Step 3 specifically involves: Step 3.1, perform wrist-arm point cloud formation. and installation of basic point cloud Statistical filtering algorithms are applied to remove isolated noise points; then, voxel grid filters are used for downsampling to reduce the amount of data while preserving shape features. Step 3.2: Using the sample consistency initial registration algorithm, based on the fast point feature histogram features, the processed carpal point cloud is processed. and wrist CAD model Initial pose estimation is performed to obtain a preliminary transformation matrix, which is then applied to the processed installation foundation point cloud. and basic CAD model Perform initial pose estimation to obtain a preliminary transformation matrix; Step 3.3: Using the preliminary transformation matrix obtained in Step 3.2 as the initial value, execute the ICP algorithm. Iteratively find the nearest neighbor pair between two point clouds and minimize their distance error, converging to a high-precision transformation matrix. and .

4. The pose compensation method for the installation of a contact wire cantilever arm based on 3D vision guidance according to claim 3, characterized in that, The transformation matrix in step 3.3 and The solution process is the same, the above The specific solution process is as follows: The installation foundation point cloud obtained in step 3.1... For each point, transform it to the underlying CAD model using the current transformation matrix. coordinate system, in Find the nearest point in the middle and record the corresponding point pair; Based on distance threshold Threshold of the angle between the normal vector and the normal vector Filter the point pairs and calculate the weight of each point pair; Construct a weighted point-to-surface error objective function and use a robust kernel function to reduce the impact of outliers; Linearize the objective function, construct the Jacobian matrix and residual vector for each point pair, and form a linear system; Solve the linear system to obtain the transformation increment, and update the transformation matrix. If the norm of the transformation increment is less than a threshold, stop the iteration and obtain the transformation matrix.

5. The pose compensation method for the installation of a contact wire cantilever arm based on 3D vision guidance according to claim 1, characterized in that, The formula for calculating the pose compensation transformation matrix ΔT in step 4 is as follows: Here, inv(·) represents the inverse of the matrix.

6. The pose compensation method for the installation of a catenary arm based on 3D vision guidance according to claim 1, characterized in that, Step 5 specifically involves: resolving the pose compensation transformation matrix ΔT into a three-dimensional translation compensation vector [Δx, Δy, Δz] and a three-dimensional rotation compensation vector at the end of the robotic arm in the base coordinate system; generating a Cartesian space motion trajectory from the robot's current position to the target position based on the three-dimensional translation compensation vector and the three-dimensional rotation compensation vector; and controlling the robot to move along the motion trajectory to complete the precise installation of the wrist arm.

7. The pose compensation method for the installation of a catenary arm based on 3D vision guidance according to claim 1, characterized in that, Following step 5, an installation verification step is included. This step specifically involves: rescanning the installed wrist arm using a 3D vision sensor, obtaining the actual installation pose through point cloud registration, and comparing it with the theoretical installation pose. The comparison is performed to generate an installation quality report.

8. An automated carpal tunnel assembly system for implementing the method as described in any one of claims 1-7, characterized in that, The system includes: an industrial robot with a dedicated actuator at its end for gripping a wrist arm; A 3D vision sensor is fixedly installed at the end of the robot or at a fixed position within the robot's workspace; Central control unit, the central control unit comprising: The storage module is used to store the CAD model, BIM data, and theoretical installation pose of the cantilever arm and its mounting foundation. ; The point cloud processing module is used to perform point cloud registration and real-time pose calculation; The compensation calculation module is used to calculate the pose compensation transformation amount ΔT; The motion control module is used to plan the robot's motion trajectory and drive its movement.

9. The system according to claim 8, characterized in that, The central control unit is also communicatively connected to a BIM data server and / or a monitoring system at the construction site to achieve data synchronization and interaction.

Images