Search

CN-122023861-A - Welding track planning method, system, storage medium and computer equipment based on physical modeling drive

CN122023861ACN 122023861 ACN122023861 ACN 122023861ACN-122023861-A

Abstract

The invention discloses a welding track planning method based on physical modeling driving, which comprises the following steps of S1, constructing a physical model for a welding seam of a workpiece based on a welding seam characteristic point track and a workpiece groove cross-section shape, adopting a data augmentation method to generate a point cloud database of the welding seam, S2, adopting a laser vision sensor to collect point clouds on the surface of the welding workpiece, S3, constructing a point cloud segmentation model, training and optimizing the point cloud segmentation model, deducing the point clouds by utilizing the optimized point cloud segmentation model to obtain an interesting area of the welding seam point cloud, S4, determining the welding seam point cloud by adopting a local orientation boundary box and a local concave-convex degree, S5, carrying out noise filtering and characteristic point extraction on the welding seam point cloud, and S6, planning the track and the gesture of the extracted welding seam characteristic points to generate a track sequence executed by a welding robot. The method solves the problems of high cost and low efficiency of manually collecting and labeling point cloud data, and remarkably improves the robustness of weld feature extraction and the accuracy of track planning.

Inventors

  • ZHANG TIE
  • WANG XINJIE
  • ZOU YANBIAO

Assignees

  • 华南理工大学

Dates

Publication Date
20260512
Application Date
20251128

Claims (10)

  1. 1. The welding track planning method based on the physical modeling driving is characterized by comprising the following steps of: S1, constructing a physical model for a workpiece welding seam based on a welding seam characteristic point track and a workpiece groove cross-section shape, wherein the physical model adopts a data augmentation method to generate a point cloud database of the welding seam; S2, acquiring point clouds on the surface of a welding workpiece by using a laser vision sensor; S3, constructing a point cloud segmentation model, performing training optimization on the point cloud segmentation model by adopting a point cloud database, and deducing the point cloud acquired in the S2 by using the optimized point cloud segmentation model to obtain an interesting region of the weld point cloud; s4, determining weld joint point cloud by adopting a local orientation boundary box and local concave-convex degree; s5, carrying out noise filtration and feature point extraction on the weld point cloud; s6, performing track and gesture planning on the extracted weld characteristic points to generate a track sequence executed by the welding robot.
  2. 2. The welding track planning method based on physical modeling driving as claimed in claim 1, wherein the step S1 includes the following specific steps: s11, sampling in a basic coordinate system plane based on the cross section shape of the groove to obtain a cross section point set S base of the two-dimensional groove; S12, constructing a discrete weld characteristic point sequence by utilizing a weld characteristic point track, taking the discrete weld characteristic point sequence as input, and generating a three-dimensional characteristic point track Q feature through NURBS curve interpolation and discretization; s13, calculating a tangent vector E of each point in the point set of the characteristic point track Q feature by using a neighboring point difference method; s14, gaussian noise is added to the cross section point set S base according to the extraction error of the laser stripe skeleton, and the method is obtained: S′ base ＝S base +N(0,σ 2 ) Where N represents a gaussian distribution and σ 2 represents variance; S15, establishing a local coordinate system along a characteristic point track Q feature , transforming a cross section point set S' base added with Gaussian noise into each local coordinate system, and splicing the cross section point sets into a three-dimensional weld point cloud; S16, performing data enhancement processing on the three-dimensional weld point cloud by adopting a data enhancement method so as to generate a point cloud database.
  3. 3. The welding track planning method based on physical modeling driving as claimed in claim 1, wherein the step S3 includes the following specific steps: s31, dividing point cloud data in a point cloud database to obtain a training set, a verification set and a test set; s32, constructing a point cloud segmentation model, and using a training set, a verification set and a test set for training the point cloud segmentation model to perform training optimization on the point cloud segmentation model and obtain training weights; s33, taking the point cloud acquired in the S2 as input of an optimized point cloud segmentation model, loading training weights for inference, and obtaining an interesting region of the workpiece point cloud.
  4. 4. The welding track planning method based on physical modeling driving as claimed in claim 1, wherein the step S4 includes the following specific steps: s41, enabling the point cloud acquired in the S2 to be O, wherein O= { O 1 ,o 2 ,…,o N },o i ∈R 3 ,o i is a query point, R 3 represents each point in the set O, and O i is a three-dimensional vector; Constructing a local orientation bounding box for each query point o i , and testing the flatness of the local orientation bounding box: Wherein lambda 1 、λ 2 and lambda 3 are the covariance matrixes obtained by calculating the neighborhood with the radius r corresponding to each query point from large to small characteristic values; S42, clustering is carried out based on the flatness to extract points with high flatness; S43, extracting the points with high flatness obtained by extraction to determine local concave-convex degree scores, and screening weld point clouds P initial .
  5. 5. The welding track planning method based on physical modeling driving as claimed in claim 1, wherein the step S5 includes the steps of: s51, denoising the weld joint point cloud by using a DBSCAN clustering method to obtain a point cloud data set; S52, extracting characteristic points of the welding line from the point cloud data set by means of a Mean-Shift method.
  6. 6. The welding track planning method based on physical modeling driving as claimed in claim 5, wherein the step S52 includes the steps of: S521, selection function K (x): where h represents a bandwidth parameter and phi (u) represents a gaussian kernel function; s522, calculating a Mean-Shift vector based on the data point x i in the point cloud data set: Wherein N (x i ) represents the set of adjacent points of x i within a specified radius r; Causing the sliding of adjacent point sets to follow An iteration is performed, wherein, And The coordinates of this point at the t and t +1 iterations are shown, Is a Mean-shift vector obtained by calculation of the t-th iteration point x i ; m (x i ) until the modulus is below a specified threshold, the algorithm converges; after convergence, the data points are aggregated, and each data point is assigned to its nearest aggregation center, and these aggregation centers are regarded as characteristic points of the weld joint, and are denoted as p= { P 1 ,P 2 ,…,P n }.
  7. 7. The welding track planning method based on physical modeling driving as claimed in claim 1, wherein the step S6 includes the steps of: S61, performing main direction projection sequencing on the extracted feature points: Wherein P is a feature point set, M is a feature point sequence obtained by normalizing the feature point set P by taking the geometric center of the point set as a reference, v 1 is a main direction vector and s i is a feature point sequence; S62, carrying out curve interpolation and uniform discretization on the obtained set M by adopting NURBS curves so as to obtain smooth characteristic point tracks; s63, establishing a local coordinate system based on the characteristic point track, and planning the gesture of the robot to generate a track sequence executed by the welding robot.
  8. 8. A welding system based on a physical modeling driver, characterized in that it implements the welding track planning method based on a physical modeling driver according to any one of claims 1 to 7, comprising: The data acquisition module is used for acquiring point cloud data of the surface of the welding workpiece; The data processing module is used for processing the collected point cloud data and generating a welding track; And the execution module is used for executing the welding task based on the welding track.
  9. 9. A storage medium storing a program, wherein the program, when executed by a processor, implements the welding track planning method based on a physical modeling driver according to any one of claims 1 to 7.
  10. 10. Computer device comprising a processor and a memory for storing a program executable by the processor, characterized in that the processor, when executing the program stored in the memory, implements the welding track planning method based on the physical modeling driver according to any one of claims 1-7.

Description

Welding track planning method, system, storage medium and computer equipment based on physical modeling drive Technical Field The invention relates to a welding technology, in particular to a welding track planning method, a system, a storage medium and computer equipment based on physical modeling driving. Background Current industrial robot welding relies mainly on a "teaching-reproduction" mode which performs well in the face of standardized production tasks, but does not allow for adaptive adjustment of the position and geometry of the actual weld when dealing with small and medium batches of non-standardized production tasks. In order to improve the intelligent degree of robot welding, three-dimensional vision sensors represented by RGB-D sensors and laser structure light sensors are widely applied to the accurate positioning of workpiece welding seams. The three-dimensional vision sensor has the characteristics of non-contact and high efficiency, can directly obtain the point cloud on the surface of the workpiece, and plays an important role in automatic programming of welding. However, the point cloud obtained by the sensor often contains a large amount of redundant environmental information, and cannot directly reflect the position and the groove shape of the weld, and the area where the point cloud of the weld is located needs to be determined by a point cloud segmentation algorithm. When the three-dimensional point cloud segmentation network is trained to perform point cloud segmentation, the workpiece point cloud is usually required to be manually collected and the area where the weld is located is marked, which is very time-consuming. Therefore, when facing production tasks with different groove shapes and weld positions, how to efficiently manufacture a training set of a point cloud segmentation network and realize accurate positioning of a welding track by utilizing the weld point cloud is a problem to be solved. Disclosure of Invention The first object of the present invention is to overcome the above disadvantages of the prior art, and to provide a welding track planning method based on a physical modeling drive. The welding track planning method can greatly improve the production efficiency of the welding robot and the intelligent degree of the welding of the robot. A second object of the present invention is to provide a welding system based on a physical modeling drive. A third object of the present invention is to provide a storage medium. A fourth object of the invention is to provide a computer device. The invention aims at realizing the technical scheme that the welding track planning method based on the physical modeling drive comprises the following steps: S1, constructing a physical model for a workpiece welding seam based on a welding seam characteristic point track and a workpiece groove cross-section shape, wherein the physical model adopts a data augmentation method to generate a point cloud database of the welding seam; S2, acquiring point clouds on the surface of a welding workpiece by using a laser vision sensor; S3, constructing a point cloud segmentation model, performing training optimization on the point cloud segmentation model by adopting a point cloud database, and deducing the point cloud acquired in the S2 by using the optimized point cloud segmentation model to obtain an interesting region of the weld point cloud; s4, determining weld joint point cloud by adopting a local orientation boundary box and local concave-convex degree; s5, carrying out noise filtration and feature point extraction on the weld point cloud; s6, performing track and gesture planning on the extracted weld characteristic points to generate a track sequence executed by the welding robot. Step S1 comprises the following specific steps: s11, sampling in a basic coordinate system plane based on the cross section shape of the groove to obtain a cross section point set S base of the two-dimensional groove; S12, constructing a discrete weld characteristic point sequence by utilizing a weld characteristic point track, taking the discrete weld characteristic point sequence as input, and generating a three-dimensional characteristic point track Q feature through NURBS curve interpolation and discretization; s13, calculating a tangent vector E of each point in the point set of the characteristic point track Q feature by using a neighboring point difference method; s14, gaussian noise is added to the cross section point set S base according to the extraction error of the laser stripe skeleton, and the method is obtained: S′base＝Sbase+N(0,σ2) Where N represents a gaussian distribution and σ 2 represents variance; S15, establishing a local coordinate system along a characteristic point track Q feature, transforming a cross section point set S' base added with Gaussian noise into each local coordinate system, and splicing the cross section point sets into a three-dimensional weld point cloud; S16, performing