Self-adaptive generation method and system for multi-layer and multi-pass welding path of robot

Abstract

The invention discloses a robot multi-layer and multi-pass welding path self-adaptive generation method and system, and the method comprises the steps: collecting a three-dimensional point cloud of a welding line through a time flight sensor, and carrying out the preprocessing of the collected welding line point cloud; segmenting the preprocessed welding seam point cloud through a random sample consistent welding seam segmentation algorithm based on density clustering so as to extract welding seam features, and performing interpolation fitting on welding seam feature points by adopting a cubic spline interpolation method; generating a plurality of layers of welding paths according to the geometric information of the welding workpiece and the weld joint feature points after interpolation fitting; and on the basis of multiple layers and multiple welding paths, the posture of the tail end of the welding gun is calculated through a normal estimation method of weighted principal component analysis. According to the method, the welding workpiece three-dimensional point cloud containing the complex curved surface can be segmented to extract the welding seam feature points, and then multiple layers and multiple welding paths are generated through the welding seam feature points and geometric information of the welding workpiece so as to solve the problems of low welding precision, low speed and narrow application range of the medium-thickness plate.

Description

Technical Field

[0001] The invention belongs to the technical field of welding, and in particular relates to a method and system for adaptively generating multi-layer and multi-pass welding paths of a robot. Background Art

[0002] With the rapid development of industrial automation technology, the size of robot welding workpieces has gradually transitioned from miniaturization to large-scale. At present, medium and thick plates with a thickness of more than 4 mm have become common materials in many fields such as automobile manufacturing, marine ships, and bridge construction. When welding workpieces composed of medium and thick plates, due to the deep weld and large welding area, multi-layer multi-pass (MLMP) welding method is usually used for welding.

[0003] However, in addition to manual welding, the existing multi-layer and multi-pass welding technology mainly uses the "teach and playback" mode of the welding robot for welding. In this mode, the operator needs to teach each welding workpiece first, that is, manually set the robot's welding path, and then the robot welds layer by layer according to the taught path. Although this method can adapt to workpieces of different shapes and sizes, the teaching process relies on manual intervention, which is time-consuming and prone to errors, especially when dealing with large and complex-shaped workpieces. The teaching workload is even heavier. In addition, the playback welding mode after teaching lacks adaptability and is difficult to adjust according to slight changes in the workpiece morphology or the uncertainty of the welding environment. Therefore, when the workpiece morphology is complex or the welding environment changes, the applicability and effectiveness of this method are limited.

[0004] As the manufacturing industry's requirements for production efficiency and welding quality continue to increase, the traditional "teaching and playback" mode can no longer meet the demand for efficient and automated welding. In terms of dealing with complex-shaped welds and dynamic welding environments, welding systems urgently need more adaptive welding path generation technology. To this end, a welding path generation method based on a visual sensor is proposed. However, the current mainstream weld feature extraction technology is mostly limited to two-dimensional plane information processing, ignoring the three-dimensional spatial geometric characteristics of the weld. Patents CN118809035A and CN117260086B plan multi-layer and multi-pass welding paths through the two-dimensional plane information of the weld, but the description accuracy of the actual shape of the weld is insufficient during the welding path generation process, resulting in the welding effect deviating from the ideal state. More importantly, the existing multi-layer and multi-pass welding path generation methods mostly focus on V-groove welds (such as patent CN113352317A), and when faced with more complex weld shapes or diversified welding workpieces, they show obvious lack of adaptability and cannot meet the diverse industrial production needs. These problems seriously restrict the promotion and application of welding technology in complex manufacturing scenarios. Summary of the invention

[0005] In view of the above problems, the present invention provides a method and system for adaptively generating a multi-layer and multi-pass welding path of a robot, aiming to provide a multi-layer and multi-pass welding path generating method with high precision, high speed and wide applicability.

[0006] A method for adaptively generating a multi-layer and multi-pass welding path of a robot, the method comprising the following steps:

[0007] A time-of-flight sensor is used to collect the three-dimensional point cloud of the weld, and the collected weld point cloud is preprocessed;

[0008] The preprocessed weld point cloud is segmented by a random sample consistent weld segmentation algorithm based on density clustering to extract the weld features, and the weld feature points are interpolated and fitted by a cubic spline interpolation method.

[0009] Generate multi-layer and multi-pass welding paths based on the geometric information of the welding workpiece and the weld feature points after interpolation fitting;

[0010] Based on the multi-layer and multi-pass welding path, the welding gun tip posture is calculated by the normal estimation method of weighted principal component analysis.

[0011] A further technical solution of the present invention is: preprocessing the collected weld point cloud includes using a radius filtering method to remove outliers and using a point cloud downsampling method based on voxel filtering to reduce the amount of point cloud data.

[0012] A further technical solution of the present invention is: a random sample consistent weld segmentation algorithm based on density clustering, specifically comprising the following steps:

[0013] Iterate the following steps until the maximum number of iterations:

[0014] Randomly initialize three non-collinear points from the point cloud collection.

[0015] Calculate the plane equation based on the initialized points;

[0016] Calculate the distances from all points to the plane based on the plane equation, and obtain the number of all internal points that are less than the point-to-plane distance threshold;

[0017] If the number of inliers is greater than a threshold value of the number of inliers, the plane equation represented by each inlier is calculated as input data for extracting weld feature points, and each inlier is removed from the point cloud set;

[0018] After the iteration, if there are still remaining point clouds in the point cloud set, the following steps are performed on the remaining point clouds:

[0019] Take the point cloud as the center of the circle and the radius distance threshold as the radius, and calculate the number of points inside the circle;

[0020] If the number of points in the circle is greater than the threshold of the number of adjacent points, the point cloud as the center of the circle is marked, and another point cloud is randomly selected in the circle until all the remaining point clouds are marked;

[0021] Each point cloud is classified according to the label, and finally the segmented point cloud result is obtained.

[0022] A further technical solution of the present invention is: extracting weld features includes identifying and extracting connecting lines between different surfaces in the weld area from the segmented point cloud, and further analyzing the geometric features. The specific steps include: treating the point cloud on plane P2 that is closest to plane P1 as feature points, and these feature points constitute the weld path of the fillet joint.

[0023] A further technical solution of the present invention is: before extracting weld features, first extract the region of interest from the point cloud, and set a distance threshold from the point to the plane when extracting feature points.

[0024] A further technical solution of the present invention is to generate a multi-layer multi-pass welding path for a fillet weld, specifically comprising the following steps:

[0025] According to the extracted weld features A, B, C, the expression of the number of welding layers L' is:

[0026]

[0027] Wherein, h is the predetermined weld bead thickness, and th is the set error threshold. In order to make the weld bead cover the entire weld seam, the predetermined weld bead thickness h is adaptively adjusted to h′. When the remaining depth is within the allowable error range, the number of weld bead layers is rounded down, otherwise the number of weld bead layers L′=[L]+1, and the predetermined weld bead thickness is adjusted to:

[0028]

[0029] After the initial welding is performed to obtain the new weld features D and E, the first weld is welded from the right side of each layer; the number of welds is determined by the groove width of each layer and the width of the weld. The specific number of welds per layer is P′ m As shown below:

[0030]

[0031] Where w is the predetermined weld width, D m is the cross-sectional width of the mth layer;

[0032] In order to make the weld bead just covered by the weld seam, the predetermined weld bead width w is adaptively adjusted to m -[P m]≤th, th is the set error threshold, indicating that after rounding the width w, there is still a width of th·w left; if the remaining width is within the allowable error range, then P m Round down to make the number of weld passes an integer; if P m -[P m ]>th, indicating that the remaining width is large and an additional weld bead is required, P′ m =[P m ]+1, and adjust the predetermined weld bead width to:

[0033]

[0034] Based on the number of welding layers and welds of each layer, the coordinate position of each weld is further determined, and the weld position on each section is planned respectively. The height between two adjacent layers is the corrected weld thickness h′, and the width between two adjacent welds of the same layer is the corrected weld width w′. Therefore, the coordinate W of the jth weld of the i-th layer is ij for:

[0035]

[0036] A further technical solution of the present invention is to generate a multi-layer multi-pass welding path for a lap weld formed by connecting two workpieces together, specifically comprising the following steps:

[0037] The weld features B and C are obtained according to the extracted weld feature point A and the preset welding height H, and then the number of welding layers is calculated by the predetermined weld thickness h and the error range th. The number of layers L′ is:

[0038]

[0039] The weld thickness h is adjusted to get h′. The number of welds per layer needs to be determined by the cross-sectional width of the layer and the width of the weld. The cross-sectional width can be obtained based on the similarity ratio, and then the number of welds can be calculated based on the error range th:

[0040]

[0041] The weld width w is adaptively adjusted according to the error range, and the adjusted w′ is

[0042] According to the number of welding layers L' and the number of passes P' of each layer m Further determine the starting coordinate position W of each weld bead ij :

[0043]

[0044] For butt welds, generate multi-layer and multi-pass welding paths, including the following steps:

[0045] The number of layers required for welding is calculated by extracting the weld feature connection lines A, B, C, D, the predetermined weld thickness h, and the allowable error range th. The number of layers L′ is:

[0046]

[0047] Adaptive adjustment of weld bead thickness:

[0048]

[0049] After calculating the number of welding layers, it is necessary to calculate the coordinates of the starting point and the end point of the welding path, and then perform weaving welding from the starting point to the end point; the coordinates of the starting point and the end point and for:

[0050]

[0051] A further technical solution of the present invention is: for vertical plate welding, a multi-layer multi-pass welding path is generated, which specifically includes the following steps:

[0052] According to the extracted weld feature points A, B, and C, the feature points are located in the XOY plane, and the CB direction is the positive direction of the X axis;

[0053] The V-groove weld angle θ is:

[0054]

[0055] According to the known weld thickness h and weld angle θ, the number of weld passes P can be calculated as:

[0056]

[0057] Adaptively adjust the weld thickness h′ to:

[0058]

[0059] The coordinates of the starting point, middle point and end point of the welding path are calculated based on the number of welding passes, and then the welding is performed from the starting point to the middle point and then to the end point. and for:

[0060]

[0061] A further technical solution of the present invention is: calculating the welding gun end posture by a normal estimation method of weighted principal component analysis, specifically comprising the following steps:

[0062] For point pi , select the neighborhood point set S around it i ;

[0063] Calculate the neighborhood point set S i Point p in i The Euclidean distance d to the neighborhood center c i , d i For: i =||p i -c||;

[0064] The weights of different points are defined according to the Euclidean distance of the points in the neighborhood from the center. The formula for weight w is:

[0065] in is the average distance between the neighborhood points and the center point;

[0066] The neighborhood point set S i Subtract the coordinates of the center point c from the coordinates of the points in to achieve mean centering of the coordinates, and then minimize their weighted covariance matrix:

[0067] Through the singular value decomposition of the weighted covariance matrix, the eigenvector corresponding to the minimum eigenvalue is obtained as the point p i The normal vector of .

[0068] On the other hand, a robot multi-layer multi-pass welding path adaptive generation system is provided, the system comprising:

[0069] The weld point cloud acquisition module is used to acquire the three-dimensional point cloud of the weld using a time-of-flight sensor and to pre-process the acquired weld point cloud;

[0070] The weld feature point acquisition module is used to segment the preprocessed weld point cloud by using a random sample consistent weld segmentation algorithm based on density clustering to extract the weld features, and to interpolate and fit the weld feature points by using a cubic spline interpolation method;

[0071] A multi-layer and multi-pass welding path generation module is used to generate a multi-layer and multi-pass welding path according to the geometric information of the welding workpiece and the weld feature points after interpolation fitting;

[0072] The welding gun end posture acquisition module is used to calculate the welding gun end posture based on the multi-layer and multi-pass welding path through the normal estimation method of weighted principal component analysis.

[0073] The disclosed embodiments provide a method and system for adaptively generating a robot multi-layer and multi-pass welding path based on a three-dimensional point cloud. The method obtains the three-dimensional point cloud data of the welding surface through a TOF sensor, and combines the geometric feature information of the workpiece to automatically plan the number of layers and passes required for multi-layer and multi-pass welding, and then adaptively generates the welding path. This method not only improves the accuracy and efficiency of welding, but also enhances the system's adaptability to different welding tasks. By adapting to a variety of welding workpiece types, the system can achieve efficient and high-precision automated operations in medium and thick plate welding operations, greatly reducing the need for manual intervention, reducing the occurrence of errors and defects, and thus effectively improving the efficiency and quality of welding. The specific advantages are as follows:

[0074] 1. The present invention proposes a random sample consistent weld segmentation algorithm based on density clustering, which can not only process simple planar structures, but also cope with complex surface structures, and can be effectively applied to the surface segmentation of medium and thick plate welding workpieces.

[0075] 2. The present invention fully considers the diversity and complexity of welding workpieces in industrial production, and designs multi-layer and multi-pass welding paths for a variety of common medium and thick plate welds. It is generally applicable to welding workpieces of various shapes and sizes in industrial production, and can be efficiently applied to multi-layer and multi-pass welding technology.

[0076] 3. The present invention has been tested on a variety of welding workpieces, and the maximum error (ME) and root mean square error (RMSE) of the generated multi-layer and multi-pass welding paths are controlled below 1.103 mm and 0.677 mm respectively. And the time required is greatly reduced compared to the traditional "teaching and playback" mode. An efficient and stable welding path generation process is achieved.

[0077] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0078] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present invention and, together with the description, serve to explain the principles of the present invention.

[0079] Figure 1 1 is a schematic flow chart of a method for adaptively generating a multi-layer and multi-pass welding path of a robot according to an embodiment of the present invention;

[0080] Figure 2 Schematic diagram of the principle of the method for adaptively generating multi-layer and multi-pass welding paths of a robot in an embodiment of the present invention;

[0081] Figure 3 is a point cloud noise image of a V-groove weld in an embodiment of the present invention;

[0082] Figure 4 is a flowchart of point cloud preprocessing in an embodiment of the present invention, Figure 4 (a) is the display image before point cloud processing; Figure 4 (b) Display graph after removing outliers; Figure 4 (c) is the display image after point cloud downsampling;

[0083] Figure 5 is a flow chart of weld segmentation in an embodiment of the present invention, Figure 5 (a) is the original point cloud data rendering of the weld; Figure 5 (b) is the effect diagram of weld segmentation results;

[0084] Figure 6 This is a weld feature extraction effect diagram in an embodiment of the present invention. Figure 6 (a) is the point cloud ROI area map; Figure 6 (b) is the result of ROI region feature extraction;

[0085] Figure 7 is a comparison diagram of weld feature point interpolation in an embodiment of the present invention, Figure 7 (a) The effect of uninterpolated feature points; (b) The interpolation result.

[0086] Figure 8 Schematic diagram of the overall and cross-section of a V-groove weld in an embodiment of the present invention;

[0087] Fig. 9 Schematic diagram of the whole and cross section of the lap weld in the embodiment of the present invention;

[0088] Fig.10 Schematic diagram of the overall and cross-section of an S-type butt weld in an embodiment of the present invention;

[0089] Fig.11 Schematic diagram of the force principle of the molten droplet in the vertical plate welding of the V-groove weld in the embodiment of the present invention. Fig.11 (a) Schematic diagram of the force principle of the middle molten droplet; Fig.11 (b) Schematic diagram of the force principle of the molten droplets at both ends;

[0090] Fig.12 Schematic diagram of the welding path planning of the vertical plate of the V-groove weld in an embodiment of the present invention;

[0091] Fig.13 is a comparison diagram of point cloud normal vector estimation in an embodiment of the present invention, Fig.13 (a) is the PCA point cloud normal vector estimation result; Fig.13 (b) is the weighted PCA point cloud normal vector estimation result;

[0092] Fig.14Schematic diagram of the structure of a robot multi-layer multi-pass welding path adaptive generation system in an embodiment of the present invention;

[0093] Fig.15 : is a diagram showing the effect of generating the MLMP welding path of a V-groove weld in an embodiment of the present invention. Fig.15 (a) is the original data result diagram of the weld; Fig.15 (b) is the weld segmentation result diagram; Fig.15 (c) is the weld feature extraction result diagram; Fig.15 (d) Schematic diagram of MLMP welding path and welding gun end posture;

[0094] Fig.16 This is a diagram showing the effect of generating the MLMP welding path for the lap weld in an embodiment of the present invention. Fig.16 (a) is the original data result diagram of the weld; Fig.16 (b) is the weld segmentation result diagram; Fig.16 (c) is the weld feature extraction result diagram; Fig.16 (d) Schematic diagram of MLMP welding path and welding gun end posture;

[0095] Fig.17 This is a diagram showing the effect of generating the MLMP welding path for an S-type butt weld in an embodiment of the present invention. Fig.17 (a) is the original data result diagram of the weld; Fig.17 (b) Fig.17 (c) is the weld segmentation result diagram; Fig.17 (d) is the weld feature extraction result diagram; Fig.17 (e) Fig.17 (f) is a schematic diagram of the MLMP welding path and the end posture of the welding gun; Fig.17 (g) Fig.17 (h) Fig.17 (i) Fig.17 (d) Fig.17 (e) Fig.17 (f) A partial enlarged view;

[0096] Fig.18 is a comparison diagram between the simulation and actual welding of the V-groove weld in the embodiment of the present invention, Fig.18 (a) is the workpiece image before welding; Fig.18 (b) Schematic diagram of the simulated MLMP welding path; Fig.18 (c) is the workpiece image after welding. DETAILED DESCRIPTION

[0097] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only parts related to the present invention, rather than all structures, are shown in the accompanying drawings.

[0098] It should be mentioned before discussing the exemplary embodiments in more detail that some exemplary embodiments are described as processes or methods depicted as flow charts. Although the flow charts describe the steps as sequential processes, many of the steps therein can be implemented in parallel, concurrently or simultaneously. In addition, the order of the steps can be rearranged. The process can be terminated when its operation is completed, but can also have additional steps not included in the accompanying drawings. The process can correspond to a method, function, procedure, subroutine, subprogram, etc.

[0099] In the field of medium and thick plate welding, multi-layer and multi-pass welding technology with high precision and wide adaptability has attracted much attention. However, the traditional "teaching and playback" mode of welding robots relies on manual teaching, which is not only inefficient, but also difficult to adapt to the real-time dynamic adjustment requirements of complex welds, especially in multi-layer and multi-pass welding. In addition, most of the current weld feature extraction technologies are based on two-dimensional plane information, ignoring the spatial geometric characteristics of the weld, resulting in limited expression of the weld shape during the welding path generation process. At the same time, the existing multi-layer and multi-pass welding path generation method is only for V-groove welds, and has poor adaptability when dealing with complex weld shapes or different types of welding workpieces. In response to the above problems, the embodiment proposes a robot multi-layer and multi-pass welding path adaptive generation method based on three-dimensional point cloud. The method can segment the three-dimensional point cloud of the welding workpiece containing complex surfaces to extract the characteristic points of the weld, and then generate a multi-layer and multi-pass welding path through the weld characteristic points and the geometric information of the welding workpiece to solve the problems of low precision, slow speed and narrow application range of medium and thick plate welding.

[0100] The following embodiments are provided for a method and system for adaptively generating a multi-layer and multi-pass welding path of a robot:

[0101] A method for adaptively generating multi-layer and multi-pass welding paths for robots, such as Figure 1 As shown, the method comprises the following steps:

[0102] S1. Using a time-of-flight sensor to collect the three-dimensional point cloud of the weld, and preprocessing the collected weld point cloud;

[0103] S2. Segment the preprocessed weld point cloud by using a random sample consistent weld segmentation algorithm based on density clustering to extract weld features, and use a cubic spline interpolation method to interpolate and fit the weld feature points;

[0104] S3, generating a multi-layer multi-pass welding path according to the geometric information of the welding workpiece and the weld feature points after interpolation fitting;

[0105] S4. Based on the multi-layer and multi-pass welding path, the welding gun end posture is calculated by the normal estimation method of weighted principal component analysis.

[0106] Specifically, the overall process schematic diagram of the embodiment is as follows: Figure 2 As shown in the figure, a time-of-flight (TOF) sensor is used to collect the three-dimensional point cloud of the weld, and the collected weld point cloud is preprocessed to remove the noise points of the point cloud and reduce the complexity of the calculation. Then, the weld is segmented by a random sample consistent weld segmentation algorithm based on density clustering to extract the features of the weld, and the weld feature points are interpolated and fitted by a cubic spline interpolation method. In addition, a multi-layer and multi-pass welding path is generated based on the geometric information of the welding workpiece and the weld feature points, and finally the end posture of the welding gun is calculated by the normal estimation method of weighted PCA.

[0107] The preprocessing of the collected weld point cloud in S1 includes using a radius filtering method to remove outliers and using a point cloud downsampling method based on voxel filtering to reduce the amount of point cloud data.

[0108] Point cloud preprocessing is a crucial part of the point cloud processing process, which usually includes two main steps: outlier removal and downsampling. These steps help to accurately analyze point cloud data and provide a more reliable basis for subsequent processing and analysis. First of all, outlier removal is to remove noise and outliers caused by sensor accuracy and surrounding environment when extracting the three-dimensional point cloud of the welding workpiece to ensure the accuracy and stability of the subsequent processing process. In order to solve this problem, the radius filtering method can be used to remove these outliers. This method calculates the number of points in the neighborhood of each point in the point cloud. If the number of neighborhood points is less than the set threshold, the point is marked as an outlier or noise point.

[0109] The point cloud downsampling method based on voxel filtering is used to reduce the amount of point cloud data, simplify the computational complexity, and retain sufficient spatial structural features. This method divides the point cloud data into voxel units, calculates the center position of each voxel unit, and selects the point closest to the center as the representative point. This can effectively reduce the amount of data, remove noise, and will not lose important spatial structural information.

[0110] The embodiment directly uses a TOF sensor to obtain a three-dimensional point cloud of the weld. Since the accuracy of the sensor is not high, the extracted point cloud usually contains many noise points. Figure 3 Several examples with obvious noise are marked in the figure. Therefore, the point cloud preprocessing method proposed above is used to denoise and downsample the three-dimensional point cloud of the welding workpiece to enhance the robustness of the proposed method. The specific process principle is as follows Figure 4 shown.

[0111] The random sample consistent weld segmentation algorithm based on density clustering in S2 specifically includes the following steps:

[0112] Iterate the following steps until the maximum number of iterations:

[0113] Randomly initialize three non-collinear points from the point cloud collection.

[0114] Calculate the plane equation based on the initialized points;

[0115] Calculate the distances from all points to the plane based on the plane equation, and obtain the number of all internal points that are less than the point-to-plane distance threshold;

[0116] If the number of inliers is greater than a threshold value of the number of inliers, the plane equation represented by each inlier is calculated as input data for extracting weld feature points, and each inlier is removed from the point cloud set;

[0117] After the iteration, if there are still remaining point clouds in the point cloud set, the following steps are performed on the remaining point clouds:

[0118] Take the point cloud as the center of the circle and the radius distance threshold as the radius, and calculate the number of points inside the circle;

[0119] If the number of points in the circle is greater than the threshold of the number of adjacent points, the point cloud as the center of the circle is marked, and another point cloud is randomly selected in the circle until all the remaining point clouds are marked;

[0120] Each point cloud is classified according to the label, and finally the segmented point cloud result is obtained.

[0121] Specifically, after preprocessing the three-dimensional point cloud of the welding workpiece, it is necessary to segment it into different levels, and then extract the characteristic points of the weld. The random sample consensus (RANSAC) algorithm is a model fitting algorithm with high efficiency, high precision and strong robustness. This method is based on the principle of hypothesis and test to realize the estimation of model parameters. The core idea is to eliminate the influence of outliers through an iterative process when estimating model parameters, so as to obtain a set of data samples that are not interfered by outliers. This method can perform plane fitting on the welding workpiece and then segment it into different planes. However, in a real welding environment, most welding workpieces also contain curved surface structures, and directly using a segmentation algorithm based on RANSAC cannot accurately segment the welding workpiece into different parts. To this end, the embodiment is based on a random sample consensus weld segmentation algorithm (DC-RANSAC) based on density clustering, which is mainly divided into two parts: plane fitting and clustering. The specific process is shown in Algorithm 1:

[0122]

[0123] The proposed 3D point cloud segmentation method can effectively segment the different surfaces of the welding workpiece, and obtain the corresponding plane equation of the plane area as the input data for extracting the weld feature points, such as Figure 5 shown.

[0124] Extracting weld features in S2 includes identifying and extracting connecting lines between different surfaces in the weld area from the segmented point cloud, and further analyzing geometric features. The specific steps include: considering the point cloud on plane P2 that is closest to plane P1 as feature points, and these feature points constitute the weld path of the fillet joint.

[0125] Before extracting weld features in S2, the region of interest is extracted from the point cloud, and the distance threshold from the point to the plane is set when extracting feature points.

[0126] Specifically, the features of a weld can be interpreted as the connecting lines between different surfaces on the weld area within a welded workpiece. Weld connecting lines can be considered as straight lines or curves that connect two adjacent surfaces in a welded workpiece. These connecting lines usually represent the geometry of the weld joint, including its length, shape, and direction. The feature extraction of a weld refers to identifying and extracting these connecting lines from the point cloud and further analyzing their geometric features. In order to achieve robust weld feature extraction, a distance-based weld feature extraction algorithm is proposed. For corner joints, such as Figure 6 As shown, the points on plane P2 closest to plane P1 are regarded as feature points, and these points constitute the weld path of the fillet joint. The distance from the point to the plane is calculated by formula (1).

[0127]

[0128] In order to resist the influence of noise on feature extraction, an appropriate threshold must be selected. The selection of the threshold is crucial to the extraction results. If the threshold is too small, the number of feature points will be reduced and the geometric characteristics of the weld cannot be fully expressed. If the threshold is too large, the feature points obtained may include noise points around the weld, and the calculation of the geometric characteristics of the weld will affect the accuracy of welding path planning.

[0129] Point cloud data usually contains a large number of data points. Directly calculating the distance between these points will greatly increase the amount of calculation. In order to maintain computational efficiency, the point cloud is extracted from the region of interest (ROI). Similar to the ROI extraction in the image, the area around the weld can be set as the ROI area based on the appearance characteristics of the welded workpiece. In this way, the number of points that need to be calculated can be reduced, thereby improving computational efficiency. Figure 6 (b) shows the ROI area of ​​the point cloud and the results of ROI extraction. ROI extraction not only helps to reduce the computational burden, but also focuses on the area of ​​interest, thereby more accurately analyzing and processing weld data.

[0130] After obtaining the weld feature points, it is necessary to interpolate the sparse and discontinuous data to ensure that the transition area between the weld connection lines has continuity and local smoothness. On the basis of minimizing the error between data points, the weld data is made as continuous and smooth as possible to facilitate subsequent welding path planning.

[0131] Cubic spline interpolation is a commonly used interpolation method. It ensures the continuity of the first-order and second-order derivatives while ensuring that the interpolation curve passes through all nodes, providing a smooth curve shape. At the same time, it has good local controllability, can effectively reduce the overall error and avoid drastic fluctuations. Therefore, cubic spline interpolation has the advantages of high precision and good stability. The embodiment chooses to use this method to compensate for the characteristic points of the weld, and the specific principle is as follows.

[0132] The goal of cubic spline interpolation is to find multiple cubic polynomials to connect adjacent data points so that the entire curve is smooth and continuous. Suppose there is a set of data points (x0, y0), (x1, y1), ..., (x n ,y n ), a cubic polynomial is defined between each two adjacent points. Defined as S i (x) = a i x 3 +b i x 2 +c i x+d i , where i = 0, 1, ..., n-1, a total of n polynomials are needed. For each interval [x i , x i+1 ], construct a cubic polynomial:

[0133] S i (x) = a i (xx i ) 3 +b i (xx i ) 2 +c i (xx i )+d i (2)

[0134] Among them, the coefficient a i , b i , c i , d i It is to be determined.

[0135] Each cubic spline interpolation function S i (x) must pass through the two nodes of its definition interval, that is, it must meet the following conditions:

[0136]

[0137] At the same time, in order to ensure the smoothness of the interpolation curve, the derivatives of the cubic polynomial on each interval at the nodes are required to be first-order and second-order continuous:

[0138]

[0139] And the second-order derivatives at both ends are zero, that is, S″0(x0)=0, S″ n-1 (x n )=0.

[0140] Transforming the above conditions into a system of equations yields 4n unknowns. By constructing 4n equations through interpolation conditions and continuity conditions, all coefficients a can be uniquely determined. i , b i , c i , d i .

[0141] For every interval [x i , x i+1 ] can be expressed as a tridiagonal system of equations:

[0142] Am=d (5)

[0143] Where A is a tridiagonal matrix, m is the unknown second-order derivative vector, and by solving this equation, we can get the second-order derivative value of each interval, and then inversely deduce each S i The coefficient of (x).

[0144] The comparison before and after interpolation is Figure 7 As shown. Figure 7 The S-shaped butt weld shown is usually welded by swing welding. Swing welding is a commonly used welding technology. During the swing welding process, the welding robot swings left and right to complete the welding of the workpiece. Therefore, it is necessary to ensure that the number of weld points at both ends of the swing welding process is consistent to minimize unnecessary errors. Therefore, we control the S-shaped butt weld to have the same number of feature points on both sides by interpolation to ensure the smooth swing welding.

[0145] In order to ensure the quality of the weld in the multi-layer and multi-pass welding process, the generation of the welding path must take into account the position and posture of the end of the welding gun. The selection of the position of the end of the welding gun needs to take into account the distance between the weld and the weld joint to ensure that the welding wire can accurately and evenly fill the weld to avoid problems such as leaking or welding slag. The posture of the welding gun can realize the flexible movement of the welding path in three-dimensional space, ensuring that the welding wire is correctly delivered to the weld area, thereby achieving complete filling and good bonding of the weld joint. According to the process parameters in the MLMP welding process, the cross-sectional area of ​​the weld can be calculated as follows:

[0146]

[0147] Where D is the wire diameter, W is f is the wire feeding speed, η is the welding coverage rate, and v is the welding speed.

[0148] Existing multi-layer and multi-pass welding path generation methods are mostly limited to V-groove welds and are difficult to apply to other types of welding workpieces. Therefore, the embodiment proposes a multi-layer and multi-pass welding path generation method based on the geometric shape of the welding workpiece.

[0149] (1) Fillet weld: V-shaped groove is a typical representative of fillet weld. Its weld cross section diagram is as follows: Figure 8 As shown in the figure, the multi-layer multi-pass welding path is planned based on the extracted weld edge features A, B, and C. The number of welding layers is determined by the groove depth and the thickness of each weld. In order to reduce errors and improve the accuracy of path planning, the groove depth is calculated by combining the information of all extracted weld edge feature points. Therefore, the number of welding layers L′ is:

[0150]

[0151] In the formula, z Ai 、z Bi 、z Ci Respectively represent the depth coordinate value of the i-th weld feature point in the weld edge features A, B, and C, n represents the number of weld feature points, h is the predetermined weld thickness, and th is the set error threshold. In order to make the weld just cover the entire weld, it is necessary to adaptively adjust the predetermined weld thickness. When the remaining depth is within the allowable error range, the number of weld layers is rounded down, otherwise the number of weld layers L′=[L]+1, and the predetermined weld thickness is adjusted to:

[0152]

[0153] When welding a V-shaped groove with multiple layers and multiple passes, the initial weld is first performed to obtain the new weld features D and E. Next, the first weld is welded from the right side of each layer. The number of weld passes should be determined by the groove width of the layer and the width of the weld pass. Therefore, the number of weld passes per layer P′ m As shown below:

[0154]

[0155] In the formula, z Ai 、z Bi 、z Di 、z Ei Respectively represent the depth coordinate value of the i-th weld feature point in the weld edge features A, B, D, and E, y Ai ,y Bi ,y Di,y Ei represents the width coordinate value of the ith weld feature point in the weld edge features A, B, D, and E, respectively. w is the predetermined weld width, m is the number of layers currently welded, and D m is the width of the mth layer section. In order to ensure that the weld is exactly covered by the weld, the predetermined weld width needs to be adjusted adaptively. m -[P m ]≤th, th is the set error threshold, indicating that after rounding the width w, there is still a width of th·w left. If the remaining width is within the allowable error range, then P m Round down to make the number of weld passes an integer. m -[P m ]>th, indicating that the remaining width is large and an additional weld bead is required, P′ m =[P m ]+1, and adjust the predetermined weld bead width to:

[0156]

[0157] After calculating the number of welding layers and passes for each layer, the coordinate position of each pass can be further determined, and the weld pass position on each section can be planned separately. The height between two adjacent layers is the corrected weld pass thickness h', and the width between two adjacent passes in the same layer is the corrected weld pass width w'. Therefore, the coordinate W of the jth pass in the i-th layer is ij for:

[0158]

[0159] In the formula Respectively represent the average width coordinate values ​​of weld features B and E, Represent the average depth coordinate values ​​of weld features B and E respectively.

[0160] (2) Lap weld: A lap weld is formed by joining two workpieces together. In a lap weld, one side of the welded part (usually called the cover part) partially covers the other side (usually called the base part). A cross-sectional diagram of a lap weld is shown in Figure 1. Fig. 9 shown.

[0161] The weld features B and C are obtained according to the extracted weld feature point A and the preset welding height H. Then the number of welding layers is calculated by the predetermined weld thickness h and the error range th. The number of layers L′ is:

[0162]

[0163] The thickness of the weld is adjusted to obtain h'. The number of welds per layer needs to be determined by the cross-sectional width of the layer and the width of the weld. m is the number of layers currently welded, and Dm is the cross-sectional width of the mth layer. The cross-sectional width can be obtained based on the similarity ratio and then the number of weld passes can be calculated based on the error range th:

[0164]

[0165] Similarly, the width of the weld needs to be adaptively adjusted according to the error range, and the adjusted w′ is shown in formula (12). Then, the number of weld layers L′ and the number of passes P′ can be calculated based on the number of weld layers L′ and the number of passes P′. m Further determine the starting coordinate position W of the jth track of the i-th layer ij :

[0166]

[0167] In the formula, Indicates the average width coordinate value of weld feature A, Represents the average depth coordinate value of weld feature A.

[0168] (3) Butt weld: For welds with butt interfaces and irregular shapes, weaving welding is used. Weaving welding is a commonly used welding technology. During the weaving welding process, the welding robot periodically swings left and right at a specific angle along the welding direction to control the formation of the welding pool and weld bead. Weaving welding technology has the following advantages: First, it is suitable for various weld shapes and sizes, whether it is a straight line, curve or irregular contour, it can be flexibly dealt with; second, by controlling welding parameters and operating techniques, the stability and consistency of the welding process can be achieved to ensure the welding quality; in addition, weaving welding can also improve the uniformity and aesthetics of the weld and reduce welding residual stress and deformation. The schematic diagram of a butt weld with an S-shaped curve is shown in the figure below. Fig.10 shown.

[0169] The number of layers required for welding is calculated by extracting the weld feature connection lines A, B, C, D, the predetermined weld thickness h, and the allowable error range th. The number of layers L′ is:

[0170]

[0171] In the formula, z Ai 、z Bi 、z Ci 、z Di They represent the depth coordinate value of the i-th weld feature point in the weld edge features A, B, C, and D respectively, and n represents the number of weld feature points.

[0172] Adaptive adjustment of weld bead thickness:

[0173]

[0174] After calculating the number of welding layers, it is necessary to calculate the coordinates of the starting point and the end point of the welding path, and then perform weaving welding from the starting point to the end point. and the end point coordinates for:

[0175]

[0176] In the formula, y Cj ,y Dj represents the width coordinate value of the jth weld feature point in weld features C and D, respectively, Cj 、z Dj They respectively represent the depth coordinate values ​​of the j-th weld feature point in weld features C and D.

[0177] (4) Vertical plate welding: During the welding process, liquid metal droplets are formed at the end of the welding wire. This molten liquid metal is usually called a molten droplet. The phenomenon of it falling from the end of the welding wire to the workpiece metal under the influence of surface tension and gravity is called molten droplet dripping. In actual welding, for a vertically placed welding workpiece, its molten droplets are more likely to drip under the action of gravity. This phenomenon will change the shape of the workpiece surface, causing the actual welding path to deviate from the predetermined path, affecting the welding accuracy.

[0178] The stress analysis of the molten droplets generated during the vertical plate welding process is as follows: Fig.11 shown. Fig.11 (a) shows the force on the molten droplet in the middle of the V-groove weld, which is mainly affected by gravity G, the surface tension F of the V-groove workpiece, and the arc force P of the welding gun.

[0179] The direction of the gravity G on the molten droplet is vertically downward, which will aggravate the phenomenon of molten droplet dripping. The surface tension F on the molten droplet is mainly manifested as the restraining effect of the side wall of the V-groove workpiece and the condensed metal solder on it, which can alleviate the phenomenon of molten droplet dripping to a certain extent. The direction of the arc force of the welding gun on the molten droplet is controlled by the inclination angle of the welding gun. When the welding gun is tilted upward, the direction of the arc force can be decomposed into vertical upward and horizontal components. The vertical upward component can alleviate the phenomenon of molten droplet dripping to a certain extent. The magnitude of the arc force is determined by the welding speed of the welding gun. When the welding speed is too slow, the molten droplet will overflow from the weld and destroy the workpiece structure. Too fast welding speed will make the weld cladding uneven and affect the forming quality of the weld.

[0180] Fig.11(b) shows the force on the molten droplet at both ends of the V-groove weld. In addition to the gravity G, the surface tension F of the V-groove workpiece, and the arc force P of the welding gun, it is also affected by the support force N of the lower end of the workpiece on the molten droplet. Although in order to avoid the collision between the welding gun and the welding workpiece, the arc force of the welding gun must be tilted downward to aggravate the molten droplet dripping phenomenon, the direction of the support force N is always vertically upward, so that the resultant force on the molten droplet in the vertical direction is 0, thereby solving the molten droplet dripping phenomenon caused by gravity.

[0181] In the actual welding process, the welding speed often needs to be set according to the materials of the workpiece and the solder, the welding coverage rate and other parameters. Fig.11 The molten droplet in the middle of the V-groove weld in (a) cannot be solved by simply controlling the welding speed to solve the molten droplet dripping caused by gravity. Fig.11 The stress condition of the molten droplet (b) is designed as follows Fig.12 The welding path shown in the figure solves the problem of molten droplet dripping. Fig.12 The weld feature points A, B, and C are shown. The feature points are located in the XOY plane, and the CB direction is the positive direction of the X axis.

[0182] The V-groove weld angle θ is:

[0183]

[0184] According to the known weld thickness h and weld angle θ, the number of weld passes P can be calculated as:

[0185]

[0186] In the formula, y Ai ,y Bi ,y Ci They represent the width coordinate value of the i-th weld feature point in weld features A, B, and C respectively, n represents the number of weld feature points, and th is the set error threshold.

[0187] Adaptively adjust the weld bead thickness h′ to:

[0188]

[0189] After calculating the number of weld passes, you need to calculate the coordinates of the starting point, middle point, and end point of the welding path, and then perform weaving welding from the starting point to the middle point and then to the end point. and for:

[0190]

[0191] In the formula, x Aj 、x Bj、x Cj Respectively represent the length coordinate value of the jth weld feature point in weld features A, B, and C, They represent the average width coordinate values ​​of weld features A, B, and C. Using this welding path to weld from bottom to top can greatly reduce the droplet transfer phenomenon caused by gravity factors.

[0192] The welding gun posture mainly refers to the inclination angle of the welding gun, which needs to be perpendicular to the plane of the welding path, that is, the normal vector of the plane where the welding path is located. The common method for calculating the plane normal vector is the principal component analysis (PCA) algorithm, which estimates the local plane through the least squares method for the points in the neighborhood of the current point, and regards the main direction of the plane as the normal vector of the point.

[0193] The traditional PCA method has certain limitations. The weights of points in the neighborhood of normal estimation are the same. However, when there are many noise points and points that do not belong to the plane in the neighborhood (such as the junction of the welding groove and the weld), the estimation accuracy of the normal vector is not high. Fig.13 Therefore, in order to improve the accuracy of normal vector estimation, it is necessary to i Point p in i Assign different weights. As the distance between the neighborhood points and the center point increases, the importance of the tangent plane projection estimation gradually decreases. The specific algorithm is as follows:

[0194] (1) For each point p i , select the neighborhood point set S around it i .

[0195] (2) Calculate the neighborhood point set S i Point p in i The Euclidean distance d to the neighborhood center c i , d i for:

[0196] d i =||p i -c|| (26)

[0197] (3) The weights of different points are defined according to the Euclidean distance of the points in the neighborhood from the center. The formula for the weight w is:

[0198]

[0199] Where d is the average distance between the neighborhood points and the center point. th is an artificially set distance threshold, which decreases as the density of points in the neighborhood increases.

[0200] (4) The neighborhood point set S iSubtract the coordinates of the center point c from the coordinates of the points in to achieve mean centering of the coordinates and then minimize their weighted covariance matrix.

[0201]

[0202] Where N represents the number of points in the neighborhood, w i is the weight of the i-th point, p i represents the spatial coordinates of the i-th point, c represents the spatial coordinates of the neighborhood center point, and n is the normal vector. Through the singular value decomposition (SVD) of the weighted covariance matrix, the eigenvector corresponding to the minimum eigenvalue is obtained as the normal vector of the point.

[0203] Another embodiment is used to illustrate a robot multi-layer multi-pass welding path adaptive generation system 140, such as Fig.14 As shown, the system 140 includes:

[0204] The weld point cloud acquisition module 141 is used to acquire the three-dimensional point cloud of the weld using a time-of-flight sensor and to pre-process the acquired weld point cloud;

[0205] The weld feature point acquisition module 142 is used to segment the preprocessed weld point cloud by using a random sample consistent weld segmentation algorithm based on density clustering to extract the weld features, and to interpolate and fit the weld feature points by using a cubic spline interpolation method;

[0206] A multi-layer multi-pass welding path generation module 143 is used to generate a multi-layer multi-pass welding path according to the geometric information of the welding workpiece and the weld feature points after interpolation fitting;

[0207] The welding gun end posture acquisition module 144 is used to calculate the welding gun end posture based on the multi-layer and multi-pass welding path by using the normal estimation method of weighted principal component analysis.

[0208] In addition to the upper module, the system 140 may also include other components. However, since these components are irrelevant to the content of the embodiment of the present disclosure, their illustration and description are omitted here.

[0209] The other specific working processes of the robot multi-layer multi-pass welding path adaptive generation system 140 refer to the description of the above-mentioned robot multi-layer multi-pass welding path adaptive generation method embodiment, which will not be repeated here.

[0210] In order to verify the effectiveness of the method of the present invention, a welding robot system was built using a commercially available welding robot. The welding robot system mainly includes two parts: an industrial robot system and a sensor system. The robot system includes a robotic arm, a welding controller, and a workpiece. The sensor system includes a TOF sensor and a visual controller.

[0211] To verify the effectiveness of the proposed algorithm, a series of experiments were conducted, focusing on verifying the accuracy and efficiency of its generated multi-layer and multi-pass welding paths in different types of welds. These experiments included tests on fillet, lap, and butt welds.

[0212] Welding path generation

[0213] (1) Simulation experiment: Three-dimensional simulation experiments were carried out on three typical weld types (V-groove weld, lap weld and S-butt weld). First, the welded workpiece was modeled using three-dimensional modeling technology to obtain their 3D point cloud data. These point cloud data contain detailed information about the weld and its surrounding areas, but are also accompanied by a certain amount of noise. To ensure the accuracy of subsequent analysis, a denoising method based on radius filtering was used to remove isolated noise points in the point cloud, and voxel filtering was combined to perform point cloud downsampling to further reduce data redundancy. Next, the DC-RANSAC algorithm was used to accurately segment the weld point cloud. The algorithm divides the weld workpiece point cloud into different geometric regions, providing clear input data for subsequent weld feature point extraction. In the segmentation experiment of the V-groove weld, the DC-RANSAC algorithm showed good stability and robustness, and was able to effectively separate the weld area from the non-weld area. The segmentation results of the V-groove weld are shown in Figure 2. Fig.15 (b) As shown. In the stage of weld feature point extraction, based on the distribution characteristics of weld geometry in point cloud data, a distance-based feature extraction method is used to accurately locate weld feature points. In order to improve the continuity and smoothness of the weld trajectory, a cubic spline interpolation method is further used to generate a continuous trajectory curve of the weld. This curve can fully reflect the spatial direction of the weld and provide an accurate reference for subsequent welding path planning. The V-groove weld feature point extraction results are shown in Figure 2. Fig.15 (c) As shown. Finally, based on the multi-layer multi-pass welding path generation algorithm, the multi-layer multi-pass welding path and the end posture of the welding gun of the V-shaped groove weld are calculated. The multi-layer multi-pass welding path and the end posture of the welding gun are shown in Fig.15 (d) as shown.

[0214] The experimental results of lap welds and S-type butt welds are as follows: Fig.16 and Fig.17 As shown in the figure, it can be seen that the DC-RANSAC algorithm can not only accurately segment the point cloud of V-groove welds, but also be applicable to the segmentation of lap welds and complex-shaped S-shaped butt welds. The experimental results show that this method has accurate segmentation effect and high robustness, and can flexibly adapt to complex working conditions of different weld shapes.

[0215] (2) Physical welding: In order to verify the practical application effect of the proposed path generation method, a V-groove weld was selected for physical welding experiments. During the experiment, the welding process parameters were set as follows: the wire diameter D was 1 mm, the wire feeding speed W f The welding speed is 6m / min, the welding coverage rate η is 0.92, and the welding speed v is 6mm / s. Based on these parameters, the cross-sectional area of ​​the weld can be obtained as 12.04mm2 according to formula (7). In addition, the bottom width of the V-groove welding workpiece is 0.9mm. The groove height is 35mm, the number of welding layers is 7, the number of welding passes is 36, and the set allowable error th is 1.5mm. The final welding result is as follows Fig.18 (c) as shown. Fig.18 It is a comparison chart of the simulated welding path of the V-groove weld and the actual welding. It can be seen from the figure that the MLMP welding path generated by the simulation can be effectively applied to the actual welding, and the trajectory generated is highly consistent with the actual weld morphology, which can meet the path accuracy requirements of MLMP welding. In addition, from the perspective of the weld filling effect, the weld shape after actual welding is uniform and continuous, the surface is flat, and there are no obvious defects, which can meet the requirements of industrial welding for weld quality. In addition, the consistency of the simulation path and the actual welding results verifies the effectiveness and robustness of the simulation model, proving that the method proposed in the present invention is not only suitable for path planning in an experimental environment, but can also be successfully promoted to actual industrial scenarios, providing an important basis for the development of intelligent welding systems.

[0216] Overall, it can be seen from the experimental results that the robot multi-layer multi-pass welding path adaptive generation method based on three-dimensional point cloud has significant adaptability, accuracy and stability. This method can adaptively estimate the number of layers and passes required for welding according to the characteristics of different weld shapes, and generate accurate welding paths and welding gun end postures for each layer of welds. The experiment also showed that the algorithm can show good adaptability and robustness on different weld types such as V-groove welds, lap welds and S-butt welds, providing technical support for the stable application of complex welding processes. The experiment shows that the robot multi-layer multi-pass welding path adaptive generation method based on three-dimensional point cloud provides a reliable welding path generation solution for the engineering application of multi-layer multi-pass welding processes, which can effectively support various welding requirements in actual production.

[0217] Accuracy and stability verification

[0218] In order to further verify the accuracy and stability of the multi-layer and multi-pass welding path generation method proposed in this invention, three different types of welds were modeled in three dimensions, including V-groove welds, lap welds, and S-butt welds. Then, these three-dimensional models were sampled to obtain their three-dimensional point cloud data. Based on the point cloud data, Gaussian noise was added to simulate the complex conditions in the actual welding environment, thereby constructing a variety of representative point cloud data sets.

[0219] Next, the multi-layer and multi-pass welding path adaptive generation method of the robot based on three-dimensional point cloud is used to generate multi-layer and multi-pass welding paths for the point cloud with Gaussian noise added and the original point cloud. On the generated welding path, the maximum error (ME) and root mean square error (RMSE) in the X, Y, and Z directions and their average values ​​are measured as performance evaluation indicators to quantitatively evaluate the accuracy and stability of the algorithm. These error metrics can intuitively reflect the degree of deviation between the generated path and the ideal path, thereby helping us evaluate the performance of the algorithm. The specific ME and RMSE are shown in Table 3. The error is the difference between the welding path generated by the point cloud with Gaussian noise and the welding path generated by the original point cloud. The experimental results show that the welding path generated by the proposed method is highly accurate and has strong robustness.

[0220] Table 3 MLMP welding path errors of V-groove welds, lap welds and S-butt welds

[0221]

[0222] It can be seen from Table 3 that the maximum error ME of the welding path is generally no more than 1.103 mm, and the maximum root mean square error RMSE is 0.677 mm, indicating that the path generation method has high accuracy in practical applications. In terms of error distribution, the error in the X direction is generally smaller than the error in the Y and Z directions. This may be because the mean of the y and z coordinates of the weld feature points is used to estimate the y and z coordinates of the welding path in the path calculation, and there may be a certain number of outliers in the weld feature point data. The sensitivity of these outliers to the mean causes the calculation results to deviate from the true value, thereby increasing the errors in the Y and Z directions.

[0223] At the same time, the performance of the proposed method was tested under different noise intensities, and the error between the welding path generated by the algorithm in the V-groove weld and the expected welding position was specifically analyzed after adding Gaussian noise with different variances. The error results are shown in Table 4.

[0224] It can be seen from the data in Table 4 that the maximum error (ME) and root mean square error (RMSE) of this method under different intensity Gaussian noise environments are controlled within 0.194 mm and 0.028 mm respectively, and the error fluctuation range is small. This shows that the proposed method can maintain good stability and accuracy under complex noise conditions, can reliably extract weld features and generate high-quality welding paths. Even in the case of strong noise, the algorithm can still effectively cope with environmental interference, and the generated path maintains a high consistency with the expected path.

[0225] Table 4 Experimental results of MLMP welding path generation under different noise intensities

[0226]

[0227] This excellent performance is mainly due to the design features of the algorithm. First, in the point cloud denoising and segmentation stage, the algorithm comprehensively considers the local density characteristics of the point cloud data, and combines a robust outlier processing strategy to effectively filter out the interference of noise points on weld feature extraction. Secondly, in the multi-layer and multi-pass welding path generation stage, the algorithm improves the tolerance to noise-induced deviations by dynamically adjusting the path planning parameters, making the path generation process more robust to point cloud errors.

[0228] In addition, the experimental results also show that even in the case of high noise intensity, the generated welding path can still meet the accuracy requirements of industrial production. This further verifies the potential and practicality of the method proposed in this invention in real industrial application scenarios. In summary, the stability and noise resistance of this method provide a reliable solution to the path planning problem in complex welding environments, and lay a solid technical foundation for the development and application of intelligent welding systems.

[0229] Algorithm efficiency analysis

[0230] In addition, the efficiency of the algorithm is verified in detail, and the running time of each part of the proposed algorithm is analyzed and calculated to comprehensively evaluate its efficiency performance in practical applications. To this end, experiments are carried out specifically for V-groove welds, and the experimental results are shown in Table 5.

[0231] At present, industrial applications of multi-layer and multi-pass welding usually rely on the "teach and playback" mode to set the welding path. In this mode, the operator needs to manually teach the welding path and record the weld information. This process takes about 3-5 minutes to complete, especially in the case of complex welds. It will take more time. In contrast, the robot multi-layer and multi-pass welding path adaptive generation method based on 3D point cloud can realize the automatic generation of multi-layer and multi-pass welding paths through algorithms, thereby significantly reducing manual intervention.

[0232] The efficiency of each step of the proposed algorithm is analyzed, including the main steps of point cloud preprocessing, weld segmentation, weld feature extraction, weld feature point interpolation, MLMP welding path generation and normal vector estimation, and the running time of each step is recorded in Table 5. From the results in Table 5, it can be seen that the calculation time of point cloud segmentation is relatively long, which is mainly because the DC-RANSAC algorithm needs to iteratively calculate the distance from each point to the plane equation multiple times when segmenting the point cloud to determine whether the point belongs to the plane where the weld is located. When the amount of point cloud data is large, the computational burden of the process increases significantly, resulting in more time consumption.

[0233] Table 5 Running time of each part of the algorithm

[0234]

[0235] Nevertheless, compared with the overall time consumption of the traditional "teach and playback" mode, the calculation time of this algorithm is still significantly reduced, especially in the automatic generation of welding paths and the rapid planning of welding postures. Through automated point cloud segmentation processing, the shape and position of the weld can be more accurately identified, so that the welding path can be accurately generated. This method not only greatly shortens the time for welding path teaching, but also significantly reduces labor costs, and is particularly suitable for mass production in multi-layer and multi-pass welding scenarios.

[0236] In general, the adaptive generation method of multi-layer and multi-pass welding paths for robots based on three-dimensional point clouds effectively solves the cumbersome teaching problem in traditional methods and shows significant advantages in path generation efficiency. Experimental results show that the algorithm not only ensures the accuracy of welding path generation, but also greatly improves production efficiency. It can well meet the efficiency requirements of industrial production for multi-layer and multi-pass welding processes, and provides strong support for the intelligent and automated development of robot welding technology.

[0237] Comparison with existing welding path generation methods

[0238] In order to further verify the advantages of the proposed method in the generation of multi-layer and multi-pass welding paths, two existing welding path generation methods were selected for comparison. These two methods represent typical path generation strategies based on three-dimensional reconstruction technology and two-dimensional feature point extraction, respectively. Method 1 is a welding path generation method based on three-dimensional point cloud reconstruction. It uses digital fringe projection technology to perform three-dimensional reconstruction of the welding workpiece, obtains three-dimensional point cloud data through the 3D structure of the workpiece, and then uses the RANSAC algorithm to segment and process the point cloud data, extracts the features of the weld and generates the welding path. Method 2 is a welding path generation method based on two-dimensional feature point extraction. First, the key feature points of the weld on the two-dimensional plane are extracted by image processing technology, and then the two-dimensional feature points are mapped to three-dimensional space by laser plane calibration technology to plan the multi-layer and multi-pass welding path. Since Method 2 is only applicable to the path generation of V-groove welds, V-groove welds are selected as the comparative experimental objects to ensure the comparability and fairness of the experiment. The error data of the welding paths generated by the three methods are shown in Table 6.

[0239] From the data in Table 6, it can be seen that method 2 performs better than method 1 in terms of the accuracy of the generated path. The maximum values ​​of its ME and RMSE are 1.092 mm and 0.667 mm, respectively, which are significantly lower than the ME and RMSE of method 1. This shows that the accuracy and stability of method 2 are relatively high. However, the applicability of method 2 is relatively limited. It is mainly designed for V-groove welds and is difficult to extend to other weld types commonly found in industrial environments, such as welded workpieces with complex shapes including lap welds and butt welds. Therefore, the scope of application of method 2 is greatly limited.

[0240] Table 6 Comparative experimental results of MLMP welding path generation

[0241]

[0242] The method proposed in this paper performs well in terms of accuracy, with its maximum ME and RMSE being 1.103 mm and 0.663 mm, respectively, which are basically the same as the accuracy of Method 2. At the same time, the average RMSE value of the proposed method in the XYZ direction is lower than that of Method 2, showing stronger robustness. This means that the proposed method can effectively reduce the influence of noise points and outliers on the welding path generation process, and significantly reduce the error between the actual generated path and the ideal welding position. In addition, the proposed method is not only suitable for V-groove welds, but can also be used for various weld types such as lap joints and butt joints, basically covering the common weld forms in industrial applications, and has a wider applicability.

[0243] From the above experimental results, it can be seen that the proposed method for adaptive generation of multi-layer and multi-pass welding paths based on three-dimensional point clouds has significant advantages over existing methods in terms of accuracy and applicability. While maintaining high welding path accuracy, this method can adapt to a variety of weld shapes, showing excellent robustness and wide application potential.

[0244] Based on the technical solutions provided by the above embodiments, a method and system for adaptively generating a robot multi-layer and multi-pass welding path based on a three-dimensional point cloud is provided. The method obtains the three-dimensional point cloud data of the welding surface through a TOF sensor, and combines the geometric feature information of the workpiece to automatically plan the number of layers and passes required for multi-layer and multi-pass welding, and then adaptively generates the welding path. This method not only improves the accuracy and efficiency of welding, but also enhances the system's adaptability to different welding tasks. By adapting to a variety of welding workpiece types, the system can achieve efficient and high-precision automated operations in medium and thick plate welding operations, greatly reducing the need for manual intervention, reducing the occurrence of errors and defects, and thus effectively improving the efficiency and quality of welding.

[0245] The specific advantages are as follows:

[0246] 1. The present invention proposes a random sample consistent weld segmentation algorithm based on density clustering, which can not only process simple planar structures, but also cope with complex surface structures, and can be effectively applied to the surface segmentation of medium and thick plate welding workpieces.

[0247] 2. The present invention fully considers the diversity and complexity of welding workpieces in industrial production, and designs multi-layer and multi-pass welding paths for a variety of common medium and thick plate welds. It is generally applicable to welding workpieces of various shapes and sizes in industrial production, and can be efficiently applied to multi-layer and multi-pass welding technology.

[0248] 3. The present invention has been tested on a variety of welding workpieces, and the maximum error (ME) and root mean square error (RMSE) of the generated multi-layer and multi-pass welding paths are controlled below 1.103 mm and 0.677 mm respectively. And the time required is greatly reduced compared to the traditional "teaching and playback" mode. An efficient and stable welding path generation process is achieved.

[0249] In this document, the terms "comprises," "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, such that a step or method that includes a series of elements includes not only those elements, but also includes other elements not explicitly listed, or also includes elements inherent to such step or method.

[0250] The above contents are further detailed descriptions of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is limited to these descriptions. For ordinary technicians in the technical field to which the present invention belongs, several simple deductions or substitutions can be made without departing from the concept of the present invention, which should be regarded as falling within the protection scope of the present invention.

Claims

1. A method for adaptively generating multi-layer and multi-pass welding paths of a robot, characterized in that: The method comprises the following steps: A time-of-flight sensor is used to collect the three-dimensional point cloud of the weld, and the collected weld point cloud is preprocessed; The preprocessed weld point cloud is segmented by a random sample consistent weld segmentation algorithm based on density clustering to extract the weld features, and the weld feature points are interpolated and fitted by a cubic spline interpolation method. Generate multi-layer and multi-pass welding paths based on the geometric information of the welding workpiece and the weld feature points after interpolation fitting; Based on the multi-layer and multi-pass welding path, the welding gun tip posture is calculated by the normal estimation method of weighted principal component analysis.

2. The method for adaptively generating multi-layer and multi-pass welding paths of a robot according to claim 1, characterized in that: The preprocessing of the collected weld point cloud includes using a radius filtering method to remove outliers and using a point cloud downsampling method based on voxel filtering to reduce the amount of point cloud data.

3. The method for adaptively generating multi-layer and multi-pass welding paths of a robot according to claim 1, characterized in that: The random sample consistent weld segmentation algorithm based on density clustering specifically includes the following steps: Iterate the following steps until the maximum number of iterations: Randomly initialize three non-collinear points from the point cloud collection. Calculate the plane equation based on the initialized points; Calculate the distances from all points to the plane based on the plane equation, and obtain the number of all internal points that are less than the point-to-plane distance threshold; If the number of inliers is greater than a threshold value of the number of inliers, the plane equation represented by each inlier is calculated as input data for extracting weld feature points, and each inlier is removed from the point cloud set; After the iteration, if there are still remaining point clouds in the point cloud set, the following steps are performed on the remaining point clouds: Take the point cloud as the center of the circle and the radius distance threshold as the radius, and calculate the number of points inside the circle; If the number of points in the circle is greater than the threshold of the number of adjacent points, the point cloud as the center of the circle is marked, and another point cloud is randomly selected in the circle until all the remaining point clouds are marked; Each point cloud is classified according to the label, and finally the segmented point cloud result is obtained.

4. The method for adaptively generating a multi-layer and multi-pass welding path of a robot according to claim 1, characterized in that: Extracting weld features includes identifying and extracting connecting lines between different surfaces in the weld area from the segmented point cloud, and further analyzing the geometric features. The specific steps include: considering the point cloud on plane P2 that is closest to plane P1 as feature points, and these feature points constitute the weld path of the fillet joint.

5. The method for adaptively generating a multi-layer and multi-pass welding path of a robot according to claim 4, characterized in that: Before extracting weld features, the region of interest is extracted from the point cloud, and the distance threshold from the point to the plane is set when extracting feature points.

6. The method for adaptively generating multi-layer and multi-pass welding paths of a robot according to claim 1, characterized in that: For fillet welds, generate multi-layer and multi-pass welding paths, including the following steps: According to the extracted weld features A, B, C, the expression of the number of welding layers L' is: Among them, z Ai 、z Bi 、z Ci They represent the depth coordinate value of the i-th weld feature point in the weld edge features A, B, and C, respectively, n represents the number of weld feature points, h is the predetermined weld thickness, and th is the set error threshold; in order to make the weld just cover the entire weld, the predetermined weld thickness h is adaptively adjusted to h′, when the remaining depth is within the allowable error range, the number of weld layers is rounded down, otherwise the number of weld layers L′=[L]+1, and the predetermined weld thickness is adjusted to: After the initial welding is performed to obtain the new weld features D and E, the first weld is welded from the right side of each layer; the number of welds is determined by the groove width of each layer and the width of the weld. The specific number of welds per layer is P′ m As shown below: In the formula, z Ai 、z Bi 、z Di 、z Ei Respectively represent the depth coordinate value of the i-th weld feature point in the weld edge features A, B, D, and E, y Ai ,y Bi ,y Di ,y Ei represents the width coordinate value of the ith weld feature point in the weld edge features A, B, D, and E, respectively, w is the predetermined weld width, D m is the width of the mth layer section, m is the number of layers currently welded; In order to make the weld bead just covered by the weld seam, the predetermined weld bead width w is adaptively adjusted to ω′. m -[P m ]≤th, th is the set error threshold, indicating that after rounding the width w, there is still a width of th·w left; if the remaining width is within the allowable error range, then P m Round down to make the number of weld passes an integer; if P m -[P m ]>th, indicating that the remaining width is large and an additional weld bead is required, P′ m =[P m ]+1, and adjust the predetermined weld bead width to: Based on the number of welding layers and welds of each layer, the coordinate position of each weld is further determined, and the weld position on each section is planned respectively. The height between two adjacent layers is the corrected weld thickness h′, and the width between two adjacent welds of the same layer is the corrected weld width w′. Therefore, the coordinate W of the jth weld of the i-th layer is ij for: in Respectively represent the average width coordinate values ​​of weld features B and E, Represent the average depth coordinate values ​​of weld features B and E respectively.

7. The method for adaptively generating a multi-layer and multi-pass welding path of a robot according to claim 1, characterized in that: Generate a multi-layer, multi-pass weld path for a lap weld that joins two workpieces together, including the following steps: The weld features B and C are obtained according to the extracted weld feature point A and the preset welding height H, and then the number of welding layers is calculated by the predetermined weld thickness h and the error range th. The number of layers L′ is: The weld thickness h is adjusted to get h′. The number of welds per layer needs to be determined by the cross-sectional width of the layer and the width of the weld. The cross-sectional width can be obtained based on the similarity ratio, and then the number of welds can be calculated based on the error range th: m is the number of layers currently welded, D m is the cross-sectional width of the mth layer, and the weld width w is adaptively adjusted according to the error range. The adjusted w′ is According to the number of welding layers L' and the number of passes P' of each layer m Further determine the starting coordinate position W of the jth track of the i-th layer ij : Represents the average width coordinate value of weld feature A, Represents the average depth coordinate value of weld feature A. For butt welds, a multi-layer multi-pass welding path is generated, which specifically includes the following steps: The number of layers required for welding is calculated by extracting the weld feature connection lines A, B, C, D, the predetermined weld thickness h, and the allowable error range th. The number of layers L′ is: z Ai 、z Bi 、z Ci 、z Di Respectively represent the depth coordinate value of the i-th weld feature point in the weld edge features A, B, C, and D, n represents the number of weld feature points, and the weld thickness is adaptively adjusted: After calculating the number of welding layers, it is necessary to calculate the coordinates of the starting point and the end point of the welding path, and then perform weaving welding from the starting point to the end point; the coordinates of the starting point and the end point and for: where y Cj ,y Dj represents the width coordinate value of the jth weld feature point in weld features C and D, respectively, Cj 、z Dj They respectively represent the depth coordinate values ​​of the j-th weld feature point in weld features C and D.

8. The method for adaptively generating multi-layer and multi-pass welding paths of a robot according to claim 1, characterized in that: For vertical plate welding, generate multi-layer multi-pass welding paths, including the following steps: According to the extracted weld feature points A, B, and C, the feature points are located in the XOY plane, and the CB direction is the positive direction of the X axis; The V-groove weld angle θ is: According to the known weld thickness h and weld angle θ, the number of weld passes P can be calculated as: y Ai ,y Bi ,y Ci They represent the width coordinate value of the i-th weld feature point in weld features A, B, and C respectively, n represents the number of weld feature points, th is the set error threshold, and the adaptive adjustment of weld thickness h′ is: The coordinates of the starting point, middle point and end point of the welding path are calculated based on the number of welding passes, and then the welding is performed from the starting point to the middle point and then to the end point. and for: x Aj 、x Bj 、x Cj Respectively represent the length coordinate value of the jth weld feature point in weld features A, B, and C, Represent the average width coordinate values ​​of weld features A, B, and C respectively.

9. The method for adaptively generating multi-layer and multi-pass welding paths of a robot according to claim 1, characterized in that: The welding gun tip posture is calculated by the normal estimation method of weighted principal component analysis, which specifically includes the following steps: For point p i , select the neighborhood point set S around it i ; Calculate the neighborhood point set S i Point p in i The Euclidean distance d to the neighborhood center c i , d i For: i =||p i -c||; The weights of different points are defined according to the Euclidean distance of the points in the neighborhood from the center. The formula for weight w is: in is the average distance between the neighborhood points and the center point, and th is the manually set distance threshold; The neighborhood point set S i Subtract the coordinates of the center point c from the coordinates of the points in to achieve mean centering of the coordinates, and then minimize their weighted covariance matrix: N represents the number of points in the neighborhood, w i is the weight of the i-th point, p i represents the spatial coordinates of the i-th point, c represents the spatial coordinates of the neighborhood center point, and n is the normal vector; Through the singular value decomposition of the weighted covariance matrix, the eigenvector corresponding to the minimum eigenvalue is obtained as the point p i The normal vector of .

10. A robot multi-layer multi-pass welding path adaptive generation system, characterized in that: The system comprises: The weld point cloud acquisition module is used to acquire the three-dimensional point cloud of the weld using a time-of-flight sensor and to pre-process the acquired weld point cloud; The weld feature point acquisition module is used to segment the preprocessed weld point cloud by using a random sample consistent weld segmentation algorithm based on density clustering to extract the weld features, and to interpolate and fit the weld feature points by using a cubic spline interpolation method; A multi-layer and multi-pass welding path generation module is used to generate a multi-layer and multi-pass welding path according to the geometric information of the welding workpiece and the weld feature points after interpolation fitting; The welding gun end posture acquisition module is used to calculate the welding gun end posture based on the multi-layer and multi-pass welding path through the normal estimation method of weighted principal component analysis.

Images