Robot self-adaptive impeller welding control method and system fusing multi-modal data

Abstract

The invention belongs to the field of robot control, and particularly relates to a robot self-adaptive impeller welding control method and system fused with multi-modal data, and the method comprises the steps: obtaining an impeller workpiece image and three-dimensional point cloud data through an industrial vision subsystem, and extracting a three-dimensional welding seam position point set representing a welding seam track; generating a candidate welding path set based on welding seam geometric feature analysis; in combination with welding seam morphology and process requirements, optimal welding process parameters are predicted through a regression model, and a welding gun attitude angle is calculated through quaternion conversion according to a curved surface normal vector; in the welding process, the relative position deviation of a robot and a workpiece and the parameter deviation of welding current / voltage and a planned value are monitored in real time, when the deviation exceeds a preset tolerance interval, a PID control algorithm is adopted to dynamically adjust robot motion parameters or welding process parameters, and closed-loop control over the welding process is achieved; according to the method, precise planning of the welding path and self-adaptive regulation and control of the process are achieved, and the welding quality and the forming consistency are remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of robot control, and in particular relates to a robot adaptive impeller welding control method and system that integrates multimodal data. Background Technology

[0002] In the energy and aerospace fields, the weld seams of complex curved impellers are often located between narrow blades with varying curvatures, posing a significant challenge to the accuracy and adaptability of robotic welding. Currently widely used traditional technologies mainly include pre-programming based on 3D CAD models (offline), path reproduction based on manual teaching, and weld seam tracking systems using 2D laser vision. However, these existing technologies all have significant limitations: offline programming cannot compensate for positional deviations during actual workpiece processing and assembly, leading to easy collisions between the welding torch and blades or deviations from the bevel; the welding parameters and welding torch postures set by the teaching-and-reproducibility method are fixed values, making it difficult to dynamically adjust with surface curvature and bevel shape, resulting in unstable welding quality; while 2D vision systems can achieve a certain degree of lateral tracking, they lack 3D shape perception capabilities, and process parameters still heavily rely on manual experience for configuration, resulting in long changeover and debugging cycles and insufficient production line flexibility; more importantly, most existing systems operate in an open-loop execution state, lacking real-time monitoring and compensation mechanisms for dynamic disturbances such as welding thermal deformation and robot positioning drift, making it difficult to ensure process consistency and finished product qualification rates in mass production.

[0003] The existing technologies described above have the following problems: Existing welding technologies often rely on preset trajectories and fixed parameters, and are limited to offline programming of ideal workpiece models. However, for impeller-like parts with complex curved surfaces and spatial welds, it is difficult to achieve high-quality automated welding if only the geometric model is relied upon and the dynamic changes of actual working conditions are ignored. Especially in actual production, impellers are subject to various interferences such as positional errors, machining errors, and welding thermal deformation. Therefore, this invention provides an intelligent welding system and method based on machine vision and adaptive control. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention proposes a robot adaptive impeller welding control method and system that integrates multimodal data. The method acquires impeller workpiece images and 3D point cloud data through an industrial vision subsystem, extracting a set of 3D weld position points representing the weld trajectory. Based on weld geometric features, it analyzes weld direction, bevel angle, and accessibility to generate a set of candidate welding paths. Combining weld morphology and process requirements, it predicts optimal welding process parameters using a regression model and calculates the welding torch attitude angle using quaternion transformation based on surface normal vectors. During welding, it monitors the relative positional deviation between the robot and the workpiece, as well as the parameter deviation between the welding current / voltage and the planned values ​​in real time. When the deviation exceeds a preset tolerance range, a PID control algorithm dynamically adjusts the robot motion parameters or welding process parameters, achieving closed-loop control of the welding process. This invention solves the problems of inaccurate trajectory positioning, adaptive matching of process parameters, and real-time dynamic compensation in impeller welding, achieving precise welding path planning and adaptive process control, significantly improving welding quality and consistency.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A robot adaptive impeller welding control method integrating multimodal data includes:

[0007] The configured industrial vision subsystem is used to acquire images and 3D point cloud data of the impeller workpiece surface, and to extract a set of 3D weld position points that characterize the impeller weld trajectory.

[0008] Based on the three-dimensional weld location point set and combined with surface geometric feature analysis, the weld direction, bevel angle and welding accessibility are analyzed and judged to generate a set of candidate welding paths.

[0009] Based on the weld morphology features and welding process requirements corresponding to the candidate welding path set, the optimal welding process parameter set is obtained by parameter prediction based on the regression model. At the same time, based on the surface normal vector corresponding to the candidate welding path set, combined with quaternion transformation, the welding gun attitude angle is obtained.

[0010] The system responds to preset welding commands containing optimal welding process parameters and welding torch attitude angles, and monitors in real time the relative positional deviation between the end effector of the six-axis industrial robot and the weld seam trajectory of the impeller workpiece, as well as the parameter deviations between the actual welding current or voltage, welding speed and process planning values.

[0011] When the relative position deviation or parameter deviation does not meet the corresponding preset process tolerance range, the six-axis motion parameters or welding process parameters are dynamically adjusted according to the type and magnitude of the deviation, so that the welding process is always within the preset process tolerance range.

[0012] Specifically, the process of extracting the three-dimensional weld location point set representing the impeller weld trajectory includes:

[0013] S105. Collect the left-right view image pair sequence of the impeller workpiece to be welded at the same time and the three-dimensional point cloud data with the same timestamp. Combine Gaussian convolution and contrast-limited adaptive histogram equalization algorithm to preprocess noise and contrast and perform timestamp alignment to obtain the enhanced left-right view impeller workpiece image pair sequence.

[0014] S106. Based on the enhanced left-right view impeller workpiece image pair sequence, the Sobel operator is used to calculate the gradient components of the image in the horizontal and vertical directions respectively, so as to obtain the gradient magnitude and gradient direction of each pixel in each left-right view impeller workpiece image pair.

[0015] S107. Based on the gradient magnitude and gradient direction, traverse each pixel in the image, compare the gradient magnitude of the pixel with its neighboring pixels along its gradient direction, retain the pixels with local maxima in the gradient direction, and obtain the edge pixel candidate image.

[0016] S108. Based on the edge pixel candidate image and the preset filtering rules, the edge pixel candidate image is judged and filtered. All pixels marked as the final edge are set as the foreground color and all non-edge pixels are set as the background color. A binary contour image representing the edge of the impeller weld is generated from the left and right perspectives.

[0017] Specifically, the process of extracting the set of three-dimensional weld position points characterizing the impeller weld trajectory also includes:

[0018] S109. Based on the binarized contour map, slide an evaluation window of a preset size on the image, calculate the amount of grayscale change caused by the evaluation window moving a preset length along the horizontal and vertical directions at each pixel position on the contour map, and calculate the corner response function value of each pixel based on the amount of grayscale change.

[0019] S110. The calculated corner response function value is compared with a preset filtering threshold, and all pixels whose corner response function value exceeds the threshold are selected as candidate corner points in the left-right view.

[0020] S111. Based on the candidate corner point set under the left-right perspective and the non-maximum suppression process, obtain the feature point set of the left image and the feature point set of the right image under the left-right perspective.

[0021] S112. Based on the feature point set of the left image and the feature point set of the right image, the feature point matching algorithm is used to obtain a set of feature point matching pairs from the left and right perspectives.

[0022] S113. Based on the feature point matching pair set under the left and right viewpoints, combined with the internal parameter matrix corresponding to the left and right viewpoints, the lens distortion coefficient set, and the relative spatial pose relationship between the cameras under the left and right viewpoints, the three-dimensional spatial coordinates of each feature point in the industrial vision subsystem coordinate system are obtained by calculation using the triangulation principle.

[0023] Specifically, the process of extracting the set of three-dimensional weld position points characterizing the impeller weld trajectory also includes:

[0024] S114. Based on the three-dimensional spatial coordinates of each feature point in the industrial vision subsystem coordinate system, the first three-dimensional coordinates of each feature point in the robot base coordinate system are obtained through coordinate transformation calculation, which serves as the image three-dimensional weld position point set representing the impeller weld trajectory.

[0025] S115. Based on the timestamp-aligned 3D point cloud data, calculate the second 3D coordinates of each feature point in the 3D point cloud data in the robot base coordinate system, and use it as the point cloud 3D weld position point set representing the impeller weld trajectory.

[0026] S116. Based on the point cloud three-dimensional weld position point set and the image three-dimensional weld position point set, calculate the position point deviation to obtain the image-point cloud feature position point deviation sequence.

[0027] S117. When at least one image-point cloud feature position point deviation is greater than a preset deviation threshold, the mapped image three-dimensional weld position point set and the point cloud three-dimensional weld position point set are adjusted in real time so that the image-point cloud feature position point deviation is less than or equal to the preset deviation threshold in real time, and a second-optimized three-dimensional weld position point set is obtained.

[0028] Specifically, the process of generating a set of candidate welding paths includes:

[0029] S201. Based on the three-dimensional weld location point set after secondary optimization, the K-nearest neighbor algorithm is used to establish a local neighborhood point cloud for each feature point. At the same time, the surface normal vector and curvature features of each local neighborhood are calculated by principal component analysis.

[0030] S202. Based on the surface normal vector, curvature features, weld direction and bevel angle, the local surface is fitted by moving least squares method, and the Gaussian curvature and average curvature at each feature point are calculated. The absolute value of the average curvature is used as the curvature feature value of that point.

[0031] S203. Obtain the type of base material and welding material of the current impeller workpiece and the heat input sensitivity coefficient from the preset impeller base material and welding material process database, and determine the welding curvature threshold range under different material combinations and bevel shapes by combining the preset material-curvature-heat input correlation model and bevel angle.

[0032] S204. Compare the curvature feature value with the welding curvature threshold range, and select feature points whose curvature feature value does not exceed the preset maximum allowable welding curvature threshold as preliminary path candidate points.

[0033] Specifically, the process of generating a set of candidate welding paths also includes:

[0034] S205. Based on the obtained three-dimensional coordinates of the preliminary path candidate points, spatial clustering analysis is performed using the DBSCAN clustering algorithm to remove noise points that do not meet the preset minimum number of neighborhood points, so that the distribution density of the preliminary path candidate points in the corresponding local surface region meets the preset path point distribution density threshold.

[0035] S206. Based on the curvature characteristic value, material heat input sensitivity coefficient, bevel angle, and path point distribution density of each local curved surface region obtained in S205, calculate the comprehensive welding accessibility score of each preliminary path candidate point using a linear weighted algorithm.

[0036] S207. Based on the comprehensive welding accessibility scores of each preliminary path candidate point obtained in S206, a sorting and filtering algorithm is used to select the top N feature points with the highest scores to form the final candidate welding path set.

[0037] Specifically, the process of obtaining the welding torch attitude angle includes:

[0038] S301. Based on the candidate welding path set, obtain the surface normal vector of each candidate welding path point calculated in S201, and calculate the target posture of the welding gun by combining the preset welding gun working angle and travel angle rules.

[0039] S302. Convert the target posture into XYZ axis rotation angles in the robot base coordinate system to obtain the welding gun posture angle.

[0040] Specifically, the process of obtaining the optimal welding process parameter set includes:

[0041] S304. Based on the candidate welding path set, obtain the curvature feature value of each candidate welding path point calculated in S201.

[0042] S305. Based on the preset formula for calculating the base welding heat input, and combined with the weldability parameters of the base material obtained from the impeller base material and welding material process database, calculate the base heat input value.

[0043] S306. Based on the curvature feature value, determine the heat input coefficient of each candidate welding path point through a preset curvature-heat input coefficient mapping relationship;

[0044] S307. Multiply the reference heat input value by the heat input coefficient to obtain the optimal heat input value for each candidate welding path point;

[0045] S308. Based on the optimal heat input value, the corresponding welding current, voltage and welding speed parameters are calculated through a preset welding parameter mapping model to obtain the optimal welding process parameter set.

[0046] Specifically, the response to the preset welding command, which includes optimal welding process parameters and welding torch attitude angle, includes:

[0047] S401. Based on the optimal welding process parameter set output by S308 and the welding gun attitude angle output by S302, generate a robot welding action instruction sequence containing the target space coordinates, welding gun attitude angle and optimal welding process parameters corresponding to the candidate welding path.

[0048] S402. Convert the target space coordinates and attitude parameters in the motion instructions generated in S401 into executable instructions in the robot base coordinate system.

[0049] S403. Based on the executable instructions in the transformed robot base coordinate system, the target angles of each joint axis of the robot are calculated through the robot inverse kinematics algorithm, and a smooth motion trajectory is generated.

[0050] S404. During the welding process, the following monitoring tasks shall be performed simultaneously:

[0051] S404-1. The relative position between the robot end effector and the impeller workpiece is collected in real time at a preset sampling frequency based on the industrial vision subsystem, and the position deviation is calculated.

[0052] S404-2. Based on the welding power supply system, the actual welding current and voltage are collected in real time at a preset sampling rate, and the parameter deviation from the optimal welding process parameters obtained in S308 is calculated.

[0053] S405. When the detected position deviation exceeds a preset position deviation threshold or the parameter deviation exceeds a preset parameter deviation threshold, a corresponding compensation mechanism is triggered, including:

[0054] S405-1. For positional deviation, calculate the motion correction amount of each axis of the robot based on the visual positioning deviation compensation algorithm, and output it to the robot controller to adjust the motion trajectory.

[0055] S405-2. For parameter deviations, calculate the welding parameter adjustment amount based on the process parameter compensation algorithm, and adjust the welding current and voltage through the welding power supply system.

[0056] S406. After performing compensation, re-execute the monitoring task of S404 to verify whether the position deviation and parameter deviation have returned to the preset deviation range.

[0057] S407. When both the position deviation and parameter deviation are stable within the preset deviation range, and the current welding path segment is completed, the current welding stage is determined to be successfully executed, and preparations are made to execute the welding command for the next welding path segment.

[0058] The robot adaptive impeller welding control system, which integrates multimodal data, includes: a vision module, an analysis module, an attitude parameter module, a response module, and a discrimination and adjustment module.

[0059] The vision module uses the configured industrial vision subsystem to acquire images and three-dimensional point cloud data of the impeller workpiece surface, and extracts a set of three-dimensional weld position points that characterize the impeller weld trajectory.

[0060] The analysis module, based on a three-dimensional weld location point set and combined with surface geometric feature analysis, analyzes and judges the weld direction, bevel angle and welding accessibility, and generates a set of candidate welding paths.

[0061] The attitude parameter module obtains the optimal welding process parameter set by parameter prediction based on regression model, based on the weld morphology features and welding process requirements corresponding to the candidate welding path set. At the same time, it obtains the welding gun attitude angle by combining the surface normal vector corresponding to the candidate welding path set with quaternion transformation.

[0062] The response module responds to preset welding commands containing optimal welding process parameters and welding torch attitude angles, and monitors in real time the relative positional deviation between the end effector of the six-axis industrial robot and the weld seam trajectory of the impeller workpiece, as well as the parameter deviations between the actual welding current or voltage, welding speed and process planning values.

[0063] When the relative position deviation or parameter deviation does not meet the corresponding preset process tolerance range, the discrimination and adjustment module dynamically adjusts the six-axis motion parameters or welding process parameters according to the type and magnitude of the deviation, so that the welding process is always within the preset process tolerance range.

[0064] Compared with the prior art, the beneficial effects of the present invention are:

[0065] This invention addresses the shortcomings of existing technologies by configuring a dual-camera vision subsystem and employing multi-source data fusion technology to accurately acquire the three-dimensional weld seam location set of the impeller weld, achieving sub-millimeter-level precise positioning. Based on surface geometric feature analysis, it automatically generates the optimal welding path, and combines material properties and curvature characteristics to predict welding process parameters through a regression model. Furthermore, it plans the welding torch attitude angle based on the surface normal vector, achieving adaptive matching of welding parameters. During welding execution, it monitors the relative position deviation and parameter deviation between the robot and the workpiece in real time, and uses a PID control algorithm to dynamically adjust motion and process parameters, forming a closed-loop control that effectively compensates for interference such as thermal deformation. This invention significantly improves welding accuracy and quality consistency, greatly reduces debugging time, enhances production line flexibility, and realizes full-process automation and intelligence in welding complex curved surface impellers. Attached Figure Description

[0066] Figure 1 This is a flowchart of the robot adaptive impeller welding control method that integrates multimodal data according to the present invention;

[0067] Figure 2 This is a block diagram of the robot adaptive impeller welding control system that integrates multimodal data according to the present invention. Detailed Implementation

[0068] Example 1

[0069] Please see Figure 1 The present invention provides an embodiment of a robot adaptive impeller welding control method that integrates multimodal data, comprising the following steps:

[0070] S1. Use the configured industrial vision subsystem to acquire images and 3D point cloud data of the impeller workpiece surface, and extract the 3D weld position point set that characterizes the impeller weld trajectory.

[0071] S2. Based on the three-dimensional weld location point set and combined with surface geometric feature analysis, the weld direction, bevel angle and welding accessibility are analyzed and judged to generate a set of candidate welding paths.

[0072] S3. Based on the weld morphology features and welding process requirements corresponding to the candidate welding path set, the optimal welding process parameter set is obtained through parameter prediction based on the regression model. At the same time, based on the surface normal vector corresponding to the candidate welding path set, combined with quaternion transformation, the welding torch attitude angle is obtained. The weld morphology features in this embodiment mainly include: the spatial three-dimensional coordinates of the weld trajectory, the tangent direction of the weld direction, the bevel angle, the bevel depth, the local surface curvature feature value, the surface normal vector direction, the weld cross-sectional area, and the welding accessibility between blades. The welding process requirements in this embodiment mainly include: the type of base material and welding material and the heat input sensitivity coefficient, the recommended heat input range per unit length, the deposited metal quantity requirement, the welding material melting efficiency, the heat input adjustment coefficient, the matching range of welding current and voltage, the welding speed constraint, the interpass temperature control requirements, and the quality standard of weld formation.

[0073] S4. Respond to the preset welding command containing the optimal welding process parameters and welding torch attitude angle, and monitor in real time the relative position deviation between the end of the six-axis industrial robot and the weld seam trajectory of the impeller workpiece, as well as the parameter deviation between the actual welding current or voltage, welding speed and the process planning value.

[0074] S5. When the relative position deviation or parameter deviation does not meet the corresponding preset process tolerance range, the six-axis motion parameters or welding process parameters are dynamically adjusted according to the type and magnitude of the deviation, so that the welding process is always within the preset process tolerance range.

[0075] It should be further explained that the industrial vision subsystem in this embodiment includes two 12-megapixel industrial cameras, mounted on both sides of the workstation frame, with a 60° shooting angle, covering the workpiece operation area; it should also be explained that the process of extracting the three-dimensional weld position point set representing the impeller weld trajectory in this embodiment includes:

[0076] S101. Obtain image sequences of a calibration board containing a known two-dimensional planar pattern in multiple different spatial poses. Based on the correspondence between two-dimensional image points and three-dimensional spatial points extracted from the image sequences, calculate the internal parameter matrix and lens distortion coefficient set of each camera using a nonlinear optimization algorithm. The preferred nonlinear optimization algorithm is the Levenberg-Marquardt algorithm.

[0077] S102. Based on the internal parameters of each camera obtained in S101, the industrial vision subsystem synchronously acquires stereo image pairs of the same calibration board under multiple different spatial poses. Based on the matching feature points extracted from each stereo image pair, the relative spatial pose relationship between the two cameras is calculated, and a complete spatial transformation model between the left and right camera coordinate systems is established. The relative spatial pose relationship includes a three-dimensional spatial rotation matrix, a three-dimensional translation vector, and the physical baseline distance between the optical centers of the left and right cameras. The physical baseline distance between the optical centers of the left and right cameras is calculated based on the three-dimensional translation vector combined with the Euclidean norm formula.

[0078] S103. Control the robot's end effector to carry the calibration plate in a fixed position, and perform at least three non-coplanar spatial pose transformations within the robot's workspace, simultaneously completing the following operations in each pose:

[0079] A complete image of the calibration board is acquired through the industrial vision subsystem;

[0080] Read and record the precise pose of the end effector in the robot's base coordinate system from the robot controller;

[0081] S104. Based on the multiple sets of synchronized data obtained in S103, each set containing the end effector pose in the robot base coordinate system and the calibration board pose in the industrial vision subsystem coordinate system calculated by corresponding to the calibration board image acquired at that moment, a hand-eye calibration equation is established. The fixed homogeneous transformation matrix from the industrial vision subsystem coordinate system to the robot base coordinate system is solved using a least-squares optimization algorithm. It should be further noted that in this embodiment, the industrial vision subsystem coordinate system is constructed with the optical center of the left camera as the origin, the Z-axis along the optical axis of the left camera, and the X-axis horizontally... The robot's base coordinate system is constructed with the robot's mounting base center as the origin, the Z-axis pointing vertically upwards, the X-axis pointing horizontally in front of the robot, and the Y-axis determined by the right-hand rule. The two coordinate systems establish a precise spatial mapping relationship through the hand-eye calibration equation obtained in S104. This relationship is obtained by performing non-coplanar pose transformation on the calibration plate carried by the robot's end effector in S103, and by solving the least squares optimization algorithm in S104 based on the binocular vision system model established in S102.

[0082] S105. Collect the left-right view image pairs sequence and the three-dimensional point cloud data with the same timestamp of the impeller workpiece to be welded at the same time. Combine Gaussian convolution and contrast-limited adaptive histogram equalization algorithm to preprocess noise and contrast and perform timestamp alignment to obtain the enhanced left-right view impeller workpiece image pair sequence and timestamp aligned three-dimensional point cloud data.

[0083] S106. Based on the enhanced left-right view impeller workpiece image pair sequence, the Sobel operator is used to calculate the gradient components of the image in the horizontal and vertical directions respectively, so as to obtain the gradient magnitude and gradient direction of each pixel in each left-right view impeller workpiece image pair.

[0084] S107. Based on the gradient magnitude and gradient direction, traverse each pixel in the image, compare the gradient magnitude of the pixel with its neighboring pixels along its gradient direction, retain the pixels with local maxima in the gradient direction, and obtain the edge pixel candidate image.

[0085] S108. Based on the candidate edge pixel image and a preset filtering rule, perform discrimination and filtering, setting all pixels marked as final edges to the foreground color and all non-edge pixels to the background color, generating binary contour images representing the impeller weld edge from left and right perspectives respectively; wherein the filtering rule includes:

[0086] Pixels with gradient magnitudes higher than a preset high threshold are identified as strong edge pixels and directly marked as final edges.

[0087] Pixels with gradient magnitudes below a preset low threshold are identified as non-edge pixels and are removed.

[0088] Pixels with gradient magnitudes between the high and low thresholds are identified as weak edge pixels, and their connectivity is used for further discrimination: if a weak edge pixel is connected to any strong edge pixel, it is marked as a final edge; otherwise, it is discarded. In this embodiment, a high threshold and a low threshold are set based on the statistical distribution of gradient magnitudes in the candidate edge pixel image. The high threshold is determined based on a preset high quantile of the gradient magnitude, and the low threshold is determined based on the product of the high threshold and a preset scaling factor. The high quantile and scaling factor are configured according to the image noise level and edge detection sensitivity requirements.

[0089] S109. Based on the binarized contour map, slide an evaluation window of a preset size on the image, calculate the amount of grayscale change caused by the evaluation window moving a preset length along the horizontal and vertical directions at each pixel position on the contour map, and calculate the corner response function value of each pixel based on the amount of grayscale change; it should be further explained that this embodiment detects points in the image with drastic grayscale changes by calculating the corner response function value. These points correspond to points of abrupt curvature change on the surface, such as sharp edges, corners or edges of positioning reference holes, and corners of process grooves.

[0090] S110. The calculated corner response function value is compared with a preset filtering threshold, and all pixels whose corner response function value exceeds the threshold are selected as candidate corner points in the left-right view.

[0091] S111. Based on the candidate corner point set under the left-right perspective and combined with the non-maximum suppression process, the left image feature point set and the right image feature point set under the left-right perspective are obtained. In this embodiment, the candidate corner point set is extracted through the non-maximum suppression process, and the left image feature point set and the right image feature point set are output. The left image feature point set and the right image feature point set directly contain the pixel coordinates of key geometric features such as the leading and trailing edges of the impeller blades, the junction of the blade and the hub, and the start and end points of the weld. Each feature point in these left image feature point sets and right image feature point sets corresponds to specific key geometric features on the impeller workpiece surface, including: the curvature change point of the leading and trailing edges of the blades: represented by the local maxima of the corner point response function value; the positioning reference hole on the impeller: represented by the corner point or center feature point of the hole edge; and the process heat dissipation groove on the blade surface: represented by the corner point or contour change point of the groove edge.

[0092] It should be further explained that the specific steps of the nonmaximum suppression process in this embodiment include:

[0093] c1. For each candidate corner point in the candidate corner point set, define a rectangular neighborhood of a preset size centered on it;

[0094] c2. Iterate through each candidate corner point and compare its corner response function value with the corner response function values ​​of all other candidate corner points in its defined neighborhood;

[0095] c3. If the corner response function value of the current candidate corner point is the maximum value in its defined neighborhood, then retain it; otherwise, suppress and remove it from the candidate corner point set.

[0096] c4. After traversing all candidate corner points and executing steps c2 and c3, output the final set of retained candidate corner points as the refined corner point set;

[0097] S112. Based on the feature point set of the left image and the feature point set of the right image, a stereo vision matching algorithm is used to process the data to obtain a set of feature point matching pairs from the left and right perspectives; wherein, the specific process of processing with the stereo vision matching algorithm includes:

[0098] S1121. Based on the spatial transformation relationship between the left and right camera image coordinate systems established in S102, stereo correction is performed on the original image containing the feature points so that the corresponding epipolar lines of the left and right images are horizontally aligned.

[0099] S1122. For each feature point in the feature point set of the left image and the feature point set of the right image, calculate a feature descriptor representing local texture features based on the gray-level distribution of its neighboring pixels; the construction process of the feature descriptor includes:

[0100] S1122a: For each feature point in the left / right image feature point set obtained in S111, extract a neighborhood window of a preset size centered on the feature point coordinates;

[0101] S1122b: Based on the gradient magnitude and gradient direction obtained in S106, calculate the gradient direction histogram within the neighborhood window, and divide the 360-degree direction into a preset number of direction intervals.

[0102] S1122c. Based on the edge distribution features in the binarized contour map, calculate three texture feature parameters: edge density, edge direction consistency, and local contrast within the neighborhood.

[0103] S1122d: Combine the corner response function values ​​obtained in S109 to encode the feature point type information into the descriptor;

[0104] S1122e: Normalize the extracted gradient direction histogram and texture feature parameters to eliminate the influence of illumination changes;

[0105] S1122f: Convert the normalized feature vector into a compact binary descriptor format;

[0106] S1122g, based on the neighborhood size retained by feature points during the non-maximum suppression process of S111, adaptively adjusts the scale parameter of the descriptor;

[0107] S1123. For each feature point in the feature point set of the left image, within a preset search range on the same scan line of the corrected right image, the candidate matching feature point with the highest similarity is obtained through traversal based on the similarity measure of the feature descriptor.

[0108] S1124. Perform a bidirectional consistency check on the initially matched feature point pairs, eliminate mismatched pairs that do not meet the uniqueness constraint, and obtain a set of feature point matching pairs.

[0109] S113. Based on the feature point matching pair set under the left and right views, combined with the internal parameter matrix corresponding to the left and right views, the lens distortion coefficient set, and the relative spatial pose relationship between cameras under the left and right views, the three-dimensional spatial coordinates of each feature point in the industrial vision subsystem coordinate system are calculated using the triangulation principle. The specific process includes:

[0110] S1131. For each feature point matching pair, calculate the difference in its horizontal coordinates in the image pair after correction in S1121, and obtain the disparity value of the feature point.

[0111] S1132. Based on the obtained disparity value, the camera intrinsic parameter matrix obtained in S101 and the physical baseline distance obtained in S102, the three-dimensional spatial coordinates of the feature points in the industrial vision subsystem coordinate system are calculated using the triangulation principle.

[0112] S1133. Based on the lens distortion coefficient obtained in S101, the three-dimensional coordinates calculated in S1132 are distorted to obtain the three-dimensional spatial coordinates of the feature points in the industrial vision subsystem coordinate system.

[0113] S114. Based on the three-dimensional spatial coordinates of each feature point in the industrial vision subsystem coordinate system, and combined with the fixed homogeneous transformation matrix from the industrial vision subsystem coordinate system to the robot base coordinate system, the first three-dimensional coordinates of each feature point in the robot base coordinate system are obtained through coordinate transformation operation, which serves as the image three-dimensional weld position point set representing the impeller weld trajectory.

[0114] The specific process of coordinate transformation operations includes:

[0115] S1141. For the three-dimensional spatial coordinates of each feature point in the coordinate system of the industrial vision subsystem, apply the fixed homogeneous transformation matrix obtained in S104 to perform coordinate transformation operation, and calculate the preliminary three-dimensional coordinates of each feature point in the robot base coordinate system.

[0116] S1142. Verify the validity of the preliminary three-dimensional coordinates of all feature points obtained through coordinate transformation, eliminate abnormal coordinate points that exceed the robot's workspace, and integrate the three-dimensional coordinates of all verified feature points into a set of feature point three-dimensional coordinates in the robot's base coordinate system.

[0117] S1143. Perform spatial distribution optimization processing on the three-dimensional coordinate set of feature points in the robot base coordinate system, remove redundant feature points and ensure the uniform spatial distribution of key position points, and obtain the optimized three-dimensional weld position point set representing the impeller weld trajectory.

[0118] S115. Based on the timestamp-aligned 3D point cloud data, coordinate transformation operation is performed by combining the fixed homogeneous transformation matrix obtained in S104 to calculate the second 3D coordinates of each feature point in the 3D point cloud data in the robot base coordinate system, which is used as the point cloud 3D weld position point set representing the impeller weld trajectory.

[0119] S116. Based on the point cloud three-dimensional weld position point set and the image three-dimensional weld position point set, calculate the position point deviation to obtain the image-point cloud feature position point deviation sequence.

[0120] S117. When at least one image-point cloud feature position point deviation is greater than a preset deviation threshold, the mapped image three-dimensional weld position point set and the point cloud three-dimensional weld position point set are adjusted in real time so that the image-point cloud feature position point deviation is less than or equal to the preset deviation threshold in real time, and a second-optimized three-dimensional weld position point set is obtained.

[0121] This method for acquiring 3D weld trajectory point sets achieves sub-millimeter-level positioning of complex curved surface welds on impellers by constructing a high-precision visual measurement model and a multi-source information fusion mechanism. Specifically, it first employs the Levenberg-Marquardt algorithm for high-precision in-camera non-coplanar hand-eye calibration, laying a solid foundation for accurate measurement throughout the system. Secondly, it utilizes adaptive dual-threshold edge detection and non-maximum suppression corner extraction based on gradient magnitude statistics, which sensitively captures curvature abrupt changes in features such as blade edges and weld bevels, effectively suppressing noise interference. Furthermore, at the 3D reconstruction and fusion level, it ensures robustness of stereo matching through SIFT-like binary descriptors and bidirectional consistency checks, combined with triangulation and lens distortion correction. A high-confidence 3D point set of images was obtained. On the other hand, by aligning with timestamps, the 2D image features and 3D point cloud data were mapped and biased in the robot's base coordinate system. The absolute spatial accuracy of the point cloud was used to cross-validate and compensate the image reconstruction results. This fusion strategy effectively overcomes the limitations of single vision technology, which is susceptible to the influence of workpiece surface reflection and simple texture. Finally, through spatial distribution optimization and multi-source data closed-loop adjustment, the generated secondary optimized 3D weld position point set is not only better in single-point accuracy, but also more complete and uniform in covering the spatial direction of the impeller weld. This provides reliable spatial topology information and quality assurance for the adaptive planning of subsequent welding paths and process parameters, and significantly improves the success rate of welding complex curved surfaces and the weld formation quality.

[0122] It should be further explained that the process of generating the candidate welding path set in this embodiment includes:

[0123] S201. Based on the three-dimensional weld location point set after secondary optimization, the K-nearest neighbor algorithm is used to establish a local neighborhood point cloud for each feature point. At the same time, the surface normal vector and curvature features of each local neighborhood are calculated by principal component analysis. The specific process of obtaining the surface normal vector and curvature features of each local neighborhood in this embodiment includes:

[0124] S2011. Based on the robot base coordinate system obtained by S116, the 3D weld position point set after secondary optimization is used to search for a preset number of nearest neighbor feature points for each feature point and construct a local neighborhood point cloud.

[0125] S2012. Based on the three-dimensional coordinates of the local neighborhood point cloud, calculate the covariance matrix of each local neighborhood;

[0126] S2013. Perform eigenvalue decomposition on the covariance matrix to obtain three eigenvalues ​​and their corresponding eigenvectors; wherein: the largest eigenvalue represents the degree of dispersion of the point cloud distribution in the principal direction of the tangent plane, the middle eigenvalue represents the degree of dispersion of the distribution in the secondary direction of the tangent plane, and the smallest eigenvalue and its corresponding eigenvector together determine the surface geometric properties. The eigenvector defines the direction of the surface normal vector, and the ratio of the smallest eigenvalue to the sum of the eigenvalues ​​quantifies the local curvature feature of the point.

[0127] S2014. The eigenvector corresponding to the minimum eigenvalue is taken as the surface normal vector of the feature point, and the ratio of the minimum eigenvalue to the sum of the three eigenvalues ​​is taken as the local curvature feature of the point.

[0128] In this embodiment, based on the surface normal vector and local neighborhood point cloud obtained by S201, geometric feature analysis is performed on the weld direction and bevel angle. The specific process includes:

[0129] S201A1. Based on the optimized three-dimensional weld location point set, the centerline of the weld is fitted using the master curve extraction algorithm, and the tangent direction at each feature point is calculated along the centerline. This tangent direction represents the local orientation of the weld at that point.

[0130] S201A2. In the normal plane of the weld centerline, perform cross-sectional analysis on the local neighborhood point cloud constructed in S2011. By identifying gradient abrupt change points in the cross-sectional point cloud, locate the upper and lower edge points of the bevel.

[0131] S201A3. Based on the identified upper and lower edge points of the bevel, construct two fitting lines in the normal plane to represent the slopes on both sides of the bevel respectively, and calculate the angle between the fitting lines representing the slopes on both sides of the bevel to obtain the bevel angle value at the feature point.

[0132] S202. Based on the surface normal vector, curvature features, and the weld orientation and bevel angle analyzed in S201A, the local surface is fitted using the moving least squares method, and the Gaussian curvature and average curvature at each feature point are calculated. The absolute value of the average curvature is taken as the curvature feature value of that point. It should be further noted that the process of obtaining the curvature feature value in this embodiment includes:

[0133] S2021. Based on the surface normal vector and local neighborhood point cloud obtained from S201, and combined with the weld direction and bevel angle features, a weighted covariance matrix is ​​constructed using the moving least squares method to fit the local quadratic surface equation; where the weld direction is used to determine the main direction of the surface fitting, and the bevel angle features are used to constrain the boundary conditions of the fitted surface.

[0134] S2022. Based on the local quadratic surface equation, calculate the first and second basic form coefficients of the surface; during the calculation, the weld direction is taken as one of the main directions of the surface to improve the curvature calculation accuracy in the weld extension direction.

[0135] S2023. Based on the first and second fundamental form coefficients, two principal curvatures are calculated through eigenvalue solving. Gaussian curvature and mean curvature are then calculated based on these principal curvatures. The direction perpendicular to the weld direction is defined as the preferred direction for calculating the principal curvatures, ensuring that the curvature characteristic values ​​accurately reflect the geometric changes in the weld cross-sectional direction. It should be further noted that the first and second fundamental form coefficients in this embodiment are differential geometric quantities calculated from the first and second partial derivatives of a local quadratic surface fitted using the moving least squares method. The first fundamental form coefficient characterizes the local intrinsic metric properties of the surface, including length and area elements, while the second fundamental form coefficient characterizes the external curvature of the surface in three-dimensional space. Together, they constitute the mathematical basis for calculating the principal curvatures, Gaussian curvature, and mean curvature. The calculation of the curvature characteristic values ​​pays particular attention to the curvature changes in the weld cross-sectional direction, which is perpendicular to the weld direction and effectively characterizes the geometric features of the bevel region, providing an accurate geometric basis for determining subsequent welding process parameters.

[0136] S203. Obtain the type of base material and welding material of the current impeller workpiece and the heat input sensitivity coefficient from the preset impeller base material and welding material process database, and determine the welding curvature threshold range under different material combinations and bevel shapes by combining the preset material-curvature-heat input correlation model and bevel angle.

[0137] It should be further explained that the process of determining the welding curvature threshold range under different material combinations and bevel morphologies in this embodiment includes:

[0138] S2031. From the preset impeller base material and welding material process database, by querying the data record associated with the current impeller workpiece identification code, obtain its base material type, welding material type and corresponding heat input sensitivity coefficient; the heat input sensitivity coefficient is divided into three levels: low sensitivity is 0.8, medium sensitivity is 1.0, and high sensitivity is 1.2; at the same time, obtain the calculated bevel angle value at the current feature point;

[0139] S2032. Based on the obtained base material type and welding material type, query the reference welding curvature threshold mapping table in the impeller base material and welding material process database to obtain the corresponding reference curvature threshold; the reference curvature threshold is set in the range of 0.05 per millimeter to 0.15 per millimeter according to the different material combinations.

[0140] S2033. When the heat input sensitivity coefficient is low, keep the reference curvature threshold unchanged; when the heat input sensitivity coefficient is medium, reduce the reference curvature threshold by 10%; when the heat input sensitivity coefficient is high, reduce the reference curvature threshold by 20%.

[0141] S2034. Using 60 degrees as the base bevel angle, when the actual bevel angle is less than 60 degrees, the curvature threshold is reduced by 5% for every 10-degree decrease; when the actual bevel angle is greater than 60 degrees, the curvature threshold is increased by 5% for every 10-degree increase; the correction range of the bevel angle is limited to between 45 degrees and 75 degrees.

[0142] S2035. The final curvature threshold obtained after two corrections is set as the upper limit of the recommended welding curvature threshold at the current feature point; simultaneously, the lower limit of the threshold is fixed to 0, thus constituting the welding curvature threshold range for that point. It should be further noted that the specific values ​​of the reference curvature thresholds in the impeller base material and welding material process database of this embodiment are specifically set by those skilled in the art according to the corresponding material types; all correction coefficients and reference parameters are stored in the system configuration file in fixed numerical form.

[0143] S204. The curvature feature value is compared with the welding curvature threshold range, and feature points whose curvature feature value does not exceed the preset maximum allowable welding curvature threshold are selected as preliminary path candidate points. It should be further explained that the preset maximum allowable welding curvature threshold in this embodiment is the final upper limit of the curvature threshold determined by obtaining the benchmark value by querying the material process database and then correcting it twice according to the heat input sensitivity coefficient and bevel angle of the specific workpiece.

[0144] S205. Based on the obtained three-dimensional coordinates of the preliminary path candidate points, spatial clustering analysis is performed using the DBSCAN clustering algorithm to remove noise points that do not meet the preset minimum number of neighborhood points, so that the distribution density of the preliminary path candidate points in the corresponding local surface region meets the preset path point distribution density threshold.

[0145] It should be further explained that the specific process of spatial clustering analysis using the DBSCAN clustering algorithm in this embodiment includes:

[0146] S2051. Based on the three-dimensional coordinates of the preliminary path candidate points obtained in S204, set the preset neighborhood search radius and minimum number of neighborhood points as clustering parameters.

[0147] S2052. Traverse all preliminary path candidate points, calculate the number of adjacent points of each point within its neighborhood radius, and mark the points that meet the minimum number of neighborhood points requirement as core points.

[0148] S2053. Starting from the core point, recursively expand the clustering region through the density reachability principle, and classify all points that satisfy density connectivity into the same cluster.

[0149] S2054. Mark points that are not assigned to any cluster region as noise points and remove them from the path candidate point set;

[0150] S2055. Calculate the number of candidate path points per unit surface area within each cluster region to ensure that the distribution density of all regions is not lower than the preset path point distribution density threshold.

[0151] S206. Based on the curvature characteristic value, material heat input sensitivity coefficient, bevel angle, and the distribution density of path points in each local curved surface area obtained in S205, a linear weighted algorithm is used to calculate the comprehensive welding accessibility score of each preliminary path candidate point, where the weight coefficient is preset according to the welding quality and accessibility requirements.

[0152] S207. Based on the comprehensive welding accessibility scores of each preliminary path candidate point obtained in S206, the sorting and filtering algorithm is used to select the top N feature points with the highest scores to form the final candidate welding path set. The value of N is dynamically calculated based on the product of the diagonal length of the three-dimensional bounding box of the impeller workpiece and the preset path point density parameter.

[0153] Furthermore, the process of obtaining the welding torch attitude angle in this embodiment includes:

[0154] S301. Based on the candidate welding path set, obtain the surface normal vector of each candidate welding path point calculated in S201, and calculate the target posture of the welding gun by combining the preset welding gun working angle and travel angle rules.

[0155] S302. Convert the target posture into XYZ axis rotation angles in the robot base coordinate system to obtain the welding gun posture angle; it should be further noted that the process of obtaining the welding gun posture angle in this embodiment includes:

[0156] S3021. Based on the complete target pose of the robot end effector in the robot base coordinate system obtained in S3024, extract the 3×3 rotation matrix components.

[0157] S3022. Convert the rotation matrix components into Euler angle representation of the XYZ fixed-axis rotation sequence in the robot base coordinate system;

[0158] S3023. Adjust the converted Euler angles to the angle range specified by the robot control system to ensure that the rotation angles of each axis meet the robot's kinematic constraints.

[0159] S3024. Verify that the rotation angle accuracy of the converted XYZ axes reaches the preset angle deviation threshold.

[0160] S3025. The standardized X-axis rotation angle, Y-axis rotation angle and Z-axis rotation angle are output as the final welding gun attitude angle to the robot control system.

[0161] S303. Associate and store the optimal welding process parameters of each candidate welding path point with the welding torch attitude angle, and output a set of welding parameters including the position of the candidate welding path point, the optimal welding process parameters and the welding torch attitude angle.

[0162] Furthermore, the process of obtaining the optimal welding process parameter set in this embodiment includes:

[0163] S304. Based on the candidate welding path set obtained in S207, obtain the curvature feature value of each candidate welding path point calculated in S201.

[0164] S305. Based on the preset welding heat input base calculation formula, and combined with the base material weldability parameters obtained from the impeller base material and welding material process database, the reference heat input value is calculated; it should be further explained that the process of obtaining the reference heat input value in this embodiment includes:

[0165] S3051. Obtain the recommended unit length heat input range for the current impeller base material and welding material combination from the preset welding impeller base material and welding material process database;

[0166] S3052. Based on the cross-sectional morphology of the weld, the amount of deposited metal per unit length of weld is determined by calculating the cross-sectional area of ​​the weld.

[0167] S3053. Calculate the basic heat input value that meets the welding forming requirements based on the amount of deposited metal and the melting efficiency of the welding material;

[0168] S3054. Based on the thermal sensitivity of the base material and the welding position requirements, obtain the corresponding heat input adjustment coefficient from the process parameter table;

[0169] S3055. Multiply the basic heat input value by the heat input adjustment coefficient to obtain the final reference heat input value;

[0170] S306. Based on the curvature feature value, the heat input coefficient of each candidate welding path point is determined through a preset curvature-heat input coefficient mapping relationship; wherein the larger the curvature feature value, the smaller the corresponding heat input coefficient, and the value range of the heat input coefficient is 0.8-1.2; the method for obtaining the curvature-heat input coefficient mapping relationship is as follows: weld formation quality data under different curvature feature values ​​are obtained through welding process experiments, a correspondence model between curvature feature value and heat input coefficient is established, and the relationship model is stored in the control system in the form of a data mapping table, wherein the larger the curvature feature value, the smaller the corresponding heat input coefficient;

[0171] S307. Multiply the reference heat input value by the heat input coefficient to obtain the optimal heat input value for each candidate welding path point;

[0172] S308. Based on the optimal heat input value, the corresponding welding current, voltage, and welding speed parameters are calculated through a preset welding parameter mapping model to obtain the optimal welding process parameter set. It should be further explained that the construction method of the welding parameter mapping model in this embodiment is as follows: Based on welding process test data, by collecting process parameters and weld quality data of different material combinations under various working conditions, a multiple regression model is established with heat input value as the input variable and welding current, voltage, and welding speed as the output variables; firstly, the experimental data is standardized and outliers are removed; then, the quantitative relationship between each parameter is obtained by fitting using the least squares method; finally, the determined model parameters are stored in the impeller base material and welding material process database for real-time retrieval, realizing the function of accurately back-deriving the execution layer process parameters based on the optimal heat input value.

[0173] It should be further explained that the response to the preset welding command in this embodiment includes the optimal welding process parameters and the welding torch attitude angle, including:

[0174] S401. Based on the optimal welding process parameter set output by S308 and the welding torch attitude angle output by S3025, a robot welding action command sequence is generated, containing the target spatial coordinates corresponding to the candidate welding paths, the welding torch attitude angle, and the optimal welding process parameters. The welding action commands in this embodiment include: target point position command, used to determine the spatial coordinates of the welding torch in the robot base coordinate system; welding torch attitude angle command, used to control the spatial orientation of the welding torch; welding process parameter command, used to adjust the welding current, voltage, and welding speed; motion trajectory command, used to plan the continuous motion path of each joint axis; real-time monitoring command, used to collect the actual values ​​of position and welding parameters; dynamic compensation command, used to generate parameter correction amounts; status feedback command, used to verify the system status; and welding completion command, used to confirm the termination of the task. These commands establish a collaborative working mechanism through the transformation relationship between the industrial vision subsystem coordinate system and the robot base coordinate system, forming a complete welding control closed loop. In this embodiment, the instructions are generated by combining the corresponding acquired control parameter features with a PID control algorithm. For example, the method for generating the welding current control instruction in the welding process parameter instructions in this embodiment is as follows: based on the real-time collected welding current deviation and the curvature feature value of the candidate welding path point, the current deviation is processed by the PID control algorithm, and the PID parameter adjustment coefficient is output by combining the preset control rules. The current adjustment amount is calculated by the PID control algorithm, and finally the welding current control instruction is generated.

[0175] S402. The target space coordinates and attitude parameters in the motion instructions generated in S401 are uniformly transformed into executable instructions in the robot base coordinate system using the fixed homogeneous transformation matrix from the industrial vision subsystem coordinate system to the robot base coordinate system obtained in S104.

[0176] S403. Based on the executable instructions in the transformed robot base coordinate system, the target angles of each joint axis of the robot are calculated through the robot inverse kinematics algorithm, and a smooth motion trajectory is generated.

[0177] It should be further explained that the process of calculating the target angles of each joint axis of the robot and generating a smooth motion trajectory using the robot inverse kinematics algorithm in this embodiment includes:

[0178] S4031. Based on the link parameters and joint constraints of a six-axis industrial robot, establish a robot kinematic model including DH parameters.

[0179] S4032. Based on the complete target pose of the robot end effector in the robot base coordinate system obtained in S3024, extract the position vector and rotation matrix;

[0180] S4033. Based on the position vector and rotation matrix, solve the robot's inverse kinematics equations using analytical or numerical methods, and calculate all possible joint angle combinations that satisfy the target pose.

[0181] S4034. Based on joint limit constraints and motion smoothness requirements, the optimal joint angle combination is selected from all possible joint angle combinations.

[0182] S4035. Based on the current joint angle and the optimal joint angle obtained in S4034, a fifth-order polynomial interpolation algorithm is used to generate smooth motion trajectories for each joint axis.

[0183] S4036. Perform singularity detection and interference check on the generated joint trajectory. When the trajectory is detected to pass through a singular configuration or interfere with the impeller blade, replan the trajectory.

[0184] S4037. Output the final determined target angle sequence of each joint axis to the robot control system.

[0185] S404. During the welding process, the following monitoring tasks shall be performed simultaneously:

[0186] S404-1. The relative position between the robot end effector and the impeller workpiece is collected in real time at a preset sampling frequency based on the industrial vision subsystem, and the position deviation is calculated.

[0187] S404-2. Based on the welding power supply system, the actual welding current and voltage are collected in real time at a preset sampling rate, and the parameter deviation from the optimal welding process parameters obtained in S308 is calculated.

[0188] S405. When the detected position deviation exceeds a preset position deviation threshold or the parameter deviation exceeds a preset parameter deviation threshold, a corresponding compensation mechanism is triggered, including:

[0189] S405-1. For positional deviation, the motion correction amount of each axis of the robot is calculated based on the visual positioning deviation compensation algorithm, and output to the robot controller to adjust the motion trajectory; further, the process of calculating the motion correction amount of each axis of the robot based on the visual positioning deviation compensation algorithm in this embodiment includes:

[0190] S405-1a: Based on the real-time position deviation between the robot end effector and the impeller workpiece obtained from S404-1, a three-dimensional position deviation vector is extracted in the coordinate system of the industrial vision subsystem.

[0191] S405-1b: Based on the fixed homogeneous transformation matrix from the industrial vision subsystem coordinate system to the robot base coordinate system obtained in S104, the three-dimensional position deviation vector is transformed to the robot base coordinate system.

[0192] S405-1c: Calculate the Jacobian matrix of the robot in the current configuration based on the current joint angles of the robot.

[0193] S405-1d: Based on the transformed position deviation vector and the Jacobian matrix, the angle correction of each joint axis is solved by the inverse Jacobian algorithm or the damped least squares method.

[0194] S405-1e. Perform kinematic constraint checks on the obtained joint axis angle corrections to ensure that the corrections are within the limits of joint velocity and acceleration.

[0195] S405-1f converts the constrained angle correction amount into motion parameter correction instructions for each axis of the robot and outputs them to the robot controller.

[0196] S405-2. For parameter deviations, the welding parameter adjustment amount is calculated based on the process parameter compensation algorithm, and the welding current and voltage are adjusted through the welding power supply system; further, the specific process of calculating the welding parameter adjustment amount based on the process parameter compensation algorithm in this embodiment includes:

[0197] S405-2a. Based on the real-time parameter deviation between the actual welding current / voltage obtained from S404-2 and the optimal welding process parameters in S308, obtain the parameter deviation value.

[0198] S405-2b: Based on the parameter deviation value, the initial welding parameter adjustment amount is calculated using a PID control algorithm;

[0199] S405-2c: Based on the curvature characteristic value of the current welding path point calculated in S202, adaptive gain adjustment is performed on the initial welding parameter adjustment amount;

[0200] S405-2d converts the adjusted welding parameters into welding power control signals, which include a precise voltage-current correspondence.

[0201] S405-2e. Limit the amplitude and rate of change of the welding power supply control signal to ensure that the welding parameters are adjusted within the allowable range of the process.

[0202] S405-2f: The constrained welding power control signal is sent to the welding power system in real time;

[0203] S405-2g, based on subsequent sampling data from the welding monitoring system, verifies whether the actual welding parameters have returned to the preset deviation range of the optimal welding process parameters in S308.

[0204] S406. After compensation, the monitoring task of S404 is re-executed to verify whether the position deviation and parameter deviation have returned to the preset deviation range. In this embodiment, the preset deviation range is set based on the welding process quality standard and the robot system positioning accuracy. The position deviation range is set by those skilled in the art based on the weld seam tracking accuracy requirements and the robot repeatability positioning accuracy. In this embodiment, the parameter deviation range is set according to the welding quality specification.

[0205] S407. When both the position deviation and parameter deviation are stable within the preset deviation range, and the current welding path segment is completed, the current welding stage is determined to be successfully executed, and preparations are made to execute the welding command for the next welding path segment.

[0206] This application first establishes a high-precision measurement foundation by configuring a dual-industry-camera vision subsystem and employing a multi-stage calibration method. First, nonlinear optimization algorithms are used to accurately calculate the camera's intrinsic and extrinsic parameters. Then, a robot carrying a calibration board completes non-coplanar hand-eye calibration, establishing a precise coordinate mapping relationship between the vision system and the robot. In the weld trajectory extraction stage, the system combines image and point cloud data, employing multi-level image processing techniques: including gradient-statistic-based adaptive edge detection, non-maximum suppression corner extraction, and a stereo matching algorithm that fuses texture and geometric features. This achieves sub-millimeter-level positioning of key features such as blade edges and weld bevels. Furthermore, it innovatively overcomes interference from workpiece surface reflection and monotonous textures through multi-source data fusion and cross-validation mechanisms. Second, based on a precise 3D point set, a local geometric model is constructed using the K-nearest neighbor algorithm. Principal component analysis is combined to calculate surface normal vectors and curvature features. Weld direction analysis and bevel angle detection are introduced, and the moving least squares method is used to accurately fit the local surface geometry. By establishing a multi-parameter correlation model of material-curvature-heat input, and comprehensively considering the coupling effects of material thermal sensitivity, bevel morphology, and surface curvature, the welding process threshold is dynamically determined. Then, through DBSCAN clustering optimization and comprehensive reachability scoring, an optimized welding path that satisfies both geometric constraints and process requirements is generated. Finally, based on the material database and weld cross-sectional features, the baseline heat input is calculated. Dynamic adaptive adjustment of parameters is achieved through the curvature-heat input mapping relationship, ensuring precise matching of heat input to surface geometric changes. The welding torch attitude angle is obtained based on the aforementioned normal vector through coordinate transformation and kinematic constraint detection, ensuring precise contact between the welding torch and the weld surface. During execution, the system monitors position and parameter deviations in real time, uses a visual compensation algorithm to calculate robot joint corrections, and dynamically adjusts the welding current and voltage in conjunction with a process parameter compensation algorithm, forming a fully closed-loop control. Finally, a smooth motion trajectory is generated through fifth-order polynomial interpolation, ensuring that the welding process is always in the optimal process state.

[0207] Example 2

[0208] Please see Figure 2Another embodiment of the present invention provides a robot adaptive impeller welding control system that integrates multimodal data, comprising: a vision module, an analysis module, an attitude parameter module, a response module, and a discrimination and adjustment module;

[0209] The vision module uses the configured industrial vision subsystem to acquire images and 3D point cloud data of the impeller workpiece surface, and extracts a set of 3D weld position points that characterize the impeller weld trajectory.

[0210] The analysis module, based on a three-dimensional weld location point set and combined with surface geometric feature analysis, analyzes and judges the weld direction, bevel angle and welding accessibility, and generates a set of candidate welding paths.

[0211] The attitude parameter module obtains the optimal welding process parameter set by predicting parameters based on regression models, based on the weld morphology features and welding process requirements corresponding to the candidate welding path set. At the same time, it obtains the welding gun attitude angle by combining the surface normal vector corresponding to the candidate welding path set with quaternion transformation.

[0212] The response module responds to preset welding commands containing optimal welding process parameters and welding torch attitude angles, and monitors in real time the relative positional deviation between the end effector of the six-axis industrial robot and the weld seam trajectory of the impeller workpiece, as well as the parameter deviations between the actual welding current or voltage, welding speed and process planning values.

[0213] The discrimination and adjustment module dynamically adjusts the six-axis motion parameters or welding process parameters according to the type and magnitude of the deviation when the relative position deviation or parameter deviation does not meet the corresponding preset process tolerance range, so that the welding process is always within the preset process tolerance range.

[0214] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments under the guidance of the present invention without departing from the spirit and scope of the present invention. All of these variations are within the protection scope of the present invention.

Claims

1. A method of robot adaptive impeller welding control fusing multi-modal data, characterized in that, The method comprises the following steps: acquiring images and three-dimensional point cloud data of the surface of the impeller workpiece by using a configured industrial vision subsystem, and extracting a three-dimensional weld position point set representing a weld trajectory of the impeller; based on the three-dimensional weld position point set, combining with the analysis of the geometric characteristics of the curved surface, analyzing and distinguishing the weld orientation, the groove angle and the weld accessibility, and generating a candidate weld path set; based on the weld appearance characteristics corresponding to the candidate weld path set and the welding process requirements, obtaining an optimal welding process parameter set through parameter prediction based on a regression model, and simultaneously obtaining a welding torch posture angle based on the normal vector of the curved surface corresponding to the candidate weld path set and combining with quaternion conversion; responding to a preset welding instruction containing the optimal welding process parameters and the welding torch posture angle, and monitoring the relative position deviation between the end of the six-axis industrial robot and the weld trajectory of the impeller workpiece, as well as the parameter deviation between the actual welding current or voltage, welding speed and process planning value in the welding process; when the relative position deviation or the parameter deviation does not meet the corresponding preset process tolerance interval, dynamically adjusting the six-axis motion parameters or the welding process parameters according to the type and size of the deviation by using a PID control algorithm, so that the welding process is always within the preset process tolerance interval.

2. The robot adaptive impeller welding control method fusing multi-modal data of claim 1, wherein, The process of extracting the three-dimensional weld position point set representing the weld trajectory of the impeller comprises the following steps: S105, acquiring a left-right view image pair sequence of the same time of the impeller workpiece to be welded and three-dimensional point cloud data with the same timestamp, and performing noise and contrast preprocessing and timestamp alignment by combining Gaussian convolution and limited contrast self-adaptive histogram equalization algorithm, to obtain an enhanced left-right view impeller workpiece image pair sequence; S106, based on the enhanced left-right view impeller workpiece image pair sequence, the gradient components of the images in the horizontal direction and the vertical direction are calculated respectively by combining the Sobel operator, to obtain the gradient amplitude and gradient direction of each pixel point in each left-right view impeller workpiece image pair; S107, based on the gradient amplitude and gradient direction, each pixel point in the image is traversed, the gradient amplitudes of the pixel point and its adjacent pixel points are compared along the gradient direction of the pixel point, and the pixel points with local maximum value in the gradient direction are retained, to obtain an edge pixel candidate map; S108, based on the edge pixel candidate map, the preset screening rule is used for discrimination and screening, all pixel points marked as final edges are set as foreground color, and all non-edge pixel points are set as background color, to generate a binary contour map representing the weld edge of the impeller in left-right view.

3. The robot adaptive impeller welding control method fusing multi-modal data of claim 2, wherein, The process of extracting the three-dimensional weld position point set representing the weld trajectory of the impeller further comprises the following steps: S109, based on the binary contour map, a preset size evaluation window is slid on the image, the gray scale variation caused by moving the evaluation window in the horizontal and vertical directions by a preset length at each pixel position on the contour map is calculated, and the corner point response function value of each pixel is calculated based on the gray scale variation; S110, the calculated corner point response function value is compared with a preset screening threshold value, and all pixel points with a corner point response function value exceeding the threshold value are selected as a candidate corner point set in left-right view. S111, based on the candidate corner point set under left-right view angles, a non-maximum suppression process is performed to obtain a left image feature point set and a right image feature point set under left-right view angles; S112, based on the left image feature point set and the right image feature point set, a stereo vision matching algorithm is used for processing to obtain a feature point matching pair set under left-right view angles; S113, based on the feature point matching pair set under left-right view angles, an internal parameter matrix corresponding to left-right view angles, a lens distortion coefficient set and a relative spatial pose relationship between cameras under left-right view angles are combined, and a three-dimensional space coordinate of each feature point in an industrial vision subsystem coordinate system is obtained through a triangulation principle.

4. The robot adaptive impeller welding control method fusing multi-modal data of claim 3, wherein, The process of extracting the three-dimensional weld position point set representing the weld trajectory of the impeller further includes: S114, based on the three-dimensional space coordinates of each feature point in the industrial vision subsystem coordinate system, a first three-dimensional coordinate of each feature point in the robot base coordinate system is obtained through coordinate transformation operation, which is used as the image three-dimensional weld position point set representing the weld trajectory of the impeller; S115, based on the timestamp-aligned three-dimensional point cloud data, a second three-dimensional coordinate of each feature point in the robot base coordinate system in the three-dimensional point cloud data is calculated, which is used as the point cloud three-dimensional weld position point set representing the weld trajectory of the impeller; S116, based on the point cloud three-dimensional weld position point set and the image three-dimensional weld position point set, a position point deviation calculation is performed to obtain an image-point cloud feature position point deviation sequence; S117, when at least one image-point cloud feature position point deviation is greater than a preset deviation threshold, the mapped image three-dimensional weld position point set and the point cloud three-dimensional weld position point set are adjusted in real time, so that the image-point cloud feature position point deviation is less than or equal to the preset deviation threshold in real time, and a secondary optimized three-dimensional weld position point set is obtained.

5. The robot adaptive impeller welding control method fusing multi-modal data of claim 4, wherein, The process of generating the candidate weld path set includes: S201, based on the secondary optimized three-dimensional weld position point set, a K-nearest neighbor algorithm is used to establish a local neighborhood point cloud for each feature point, and a principal component analysis method is used to calculate the surface normal vector and the curvature feature of each local neighborhood; S202, based on the surface normal vector, the curvature feature, and the weld orientation and the bevel angle, a local surface is fitted by a moving least squares method, and the absolute value of the average curvature is taken as the curvature feature value of the point; S203, the base material and welding material type, heat input sensitivity coefficient of the current impeller workpiece are obtained from the preset impeller base material and welding material impeller base material and welding material process database, and combined with the preset material-curvature-heat input correlation model and the bevel angle, the welding curvature threshold range under different material combinations and bevel shapes is determined; S204, the curvature feature value is compared with the welding curvature threshold range, and the feature points with curvature feature values not exceeding the preset maximum allowed welding curvature threshold are selected as the preliminary path candidate points.

6. The robot adaptive impeller welding control method fusing multi-modal data of claim 5, wherein, The process of generating the candidate weld path set further includes: S205, based on the obtained three-dimensional coordinates of the preliminary path candidate points, spatial clustering analysis is performed in combination with a DBSCAN clustering algorithm, noise points that do not satisfy a preset minimum number of neighbor points are removed, and the distribution density of the preliminary path candidate points in the corresponding local surface area satisfies a preset path point distribution density threshold; S206, based on the curvature characteristic value, the material heat input sensitivity coefficient, the bevel angle, and the path point distribution density of each local surface area obtained in S205, a linear weighting algorithm is used to calculate a comprehensive weld accessibility score of each preliminary path candidate point; S207, based on the comprehensive weld accessibility score of each preliminary path candidate point obtained in S206, a sorting and screening algorithm is used to select the top N feature points with the highest scores to form a final candidate weld path set.

7. The robot adaptive impeller welding control method fusing multi-modal data of claim 6, wherein, The process of obtaining the welding gun attitude angle includes: S301, based on the candidate weld path set, the surface normal vector calculated in S201 for each candidate weld path point is obtained, and the target attitude of the welding gun is calculated in combination with preset welding gun working angle and walking angle rules; S302, the target attitude is converted into X-Y-Z axis rotation angles in the robot base coordinate system to obtain the welding gun attitude angle.

8. The robot adaptive impeller welding control method fusing multi-modal data of claim 7, wherein, The process of obtaining the optimal weld process parameter set includes: S304, based on the candidate weld path set, the curvature characteristic value calculated in S201 for each candidate weld path point is obtained; S305, based on a preset welding heat input base number calculation formula, in combination with the base material weldability parameters obtained through the impeller base material and welding material process database, a reference heat input value is calculated; S306, based on the curvature characteristic value, the heat input coefficient of each candidate weld path point is determined through a preset curvature-heat input coefficient mapping relationship; S307, the reference heat input value is multiplied by the heat input coefficient to obtain the optimal heat input value of each candidate weld path point; S308, based on the optimal heat input value, the corresponding welding current, voltage, and welding speed parameters are calculated through a preset welding parameter mapping model to obtain the optimal weld process parameter set.

9. The robot adaptive impeller welding control method fusing multi-modal data of claim 8, wherein, The response includes a preset welding instruction containing the optimal weld process parameters and the welding gun attitude angle, which includes: S401, based on the optimal weld process parameter set output in S308 and the welding gun attitude angle output in S302, a robot welding action instruction sequence containing the target space coordinates corresponding to the candidate weld path, the welding gun attitude angle, and the optimal weld process parameters is generated; S402, the target space coordinates and attitude parameters in the action instruction generated in S401 are uniformly converted into executable instructions in the robot base coordinate system; S403, based on the converted executable instructions in the robot base coordinate system, the target angles of each joint axis of the robot are calculated through a robot inverse kinematics algorithm, and a smooth motion trajectory is generated; S404, during the welding process, the following monitoring tasks are performed synchronously: S404-1, based on the industrial vision subsystem, the relative position of the robot end and the impeller workpiece is collected in real time at a preset sampling frequency, and the position deviation is calculated; S404-2, based on the welding power supply system, real-time acquisition of actual welding current and voltage at a preset sampling rate, calculation of parameter deviation of the optimal welding process parameters obtained in S308; S405, when the monitored position deviation exceeds the preset position deviation threshold or the parameter deviation exceeds the preset parameter deviation threshold, trigger the corresponding compensation mechanism, including: S405-1, for position deviation, based on visual positioning deviation compensation algorithm to calculate the motion correction amount of each axis of the robot, output to the robot controller to adjust the motion trajectory; S405-2, for parameter deviation, based on process parameter compensation algorithm to calculate the welding parameter adjustment amount, and adjust the welding current and voltage through the welding power supply system; S406, after executing the compensation, re-execute the monitoring task of S404 to verify whether the position deviation and parameter deviation have returned to the preset deviation interval; S407, when the position deviation and parameter deviation are both stable in the preset deviation interval, and the current welding path segment is completed, it is determined that the current welding stage is executed successfully, and the welding instruction for the next welding path segment is prepared for execution.

10. A robot adaptive impeller welding control system fusing multi-modal data, implemented based on the robot adaptive impeller welding control method fusing multi-modal data of any one of claims 1-9, characterized in that, Including: vision module, analysis module, attitude parameter module, response module, discrimination and adjustment module; The vision module uses the configured industrial vision subsystem to obtain the image and three-dimensional point cloud data of the impeller workpiece surface, and extracts a set of three-dimensional weld position points representing the impeller weld seam track; The analysis module, based on the three-dimensional weld position point set, combines the curved surface geometry feature analysis, analyzes and discriminates the weld seam trend, groove angle and weld accessibility, and generates a candidate weld path set; The attitude parameter module, based on the weld seam topographic features corresponding to the candidate weld path set and the welding process requirements, obtains the optimal welding process parameter set through parameter prediction based on the regression model, and obtains the welding gun attitude angle based on the curved surface normal vector corresponding to the candidate weld path set and the quaternion conversion; The response module responds to the preset welding instruction containing the optimal welding process parameters and the welding gun attitude angle, and monitors the relative position deviation between the six-axis industrial robot end and the impeller workpiece weld seam track in the welding process, as well as the parameter deviation of the actual welding current or voltage, welding speed and process planning value; The discrimination and adjustment module, when the relative position deviation or parameter deviation does not meet the corresponding preset process tolerance interval, adjusts the six-axis motion parameters or welding process parameters dynamically according to the deviation type and size using the PID control algorithm, so that the welding process is always within the preset process tolerance interval.

Images