Laser welding seam feature point identification method and device based on deep learning

Abstract

The invention discloses a laser welding seam feature point identification method and device based on deep learning, and the method comprises the steps: obtaining the position of a feature point from a sampling image, classifying the types of welding seams, carrying out the preprocessing of the image, and obtaining a welding seam contour line, calculating a laser incident angle reference vector by combining the position of the feature point, the contour type and the shape of the contour line; coordinates of the feature points in the sampled image and a laser incident angle reference vector are sent to a robot control system; and the robot control system executes space mapping and trajectory planning operation, converts the image feature points and the laser incident angle reference vector into control quantities capable of controlling the robot to move, outputs the control quantities to the robot and then controls the robot to move. The welding seam types are classified through YOLOv4, the positions of the contour feature points of the welding seam are recognized, then the reference vector of the laser incident angle is calculated, and high recognition and classification accuracy and high robustness are achieved in the welding seam tracking process of the three-dimensional special-shaped welding seam.

Description

technical field [0001] The invention belongs to the technical field of robot welding, in particular to a method and device for identifying feature points of laser welding seams based on deep learning. Background technique [0002] In recent years, laser welding has become more and more popular in the industry because of its advantages such as high power density, small residual stress, easy beam control, and high welding precision. However, because the energy of the laser is concentrated at the focal point, it puts forward relatively high requirements for the consistency of the workpiece. Therefore, in the actual welding production application, it is often necessary to prepare a large number of fixtures to fix the position of the weldment during mass production, which increases production. Cost, and lack of flexibility, in addition, if the structural shape of the workpiece changes, these special fixtures will be useless. At present, the mainstream solutions for such problems...

AI Technical Summary

Problems solved by technology

[0013] Among them, the above-mentioned research paper on laser welding seam recognition and trajectory generation based on deep learning is the research content of this research group. This application is based on the further improvement and optimization of the paper. Although the paper mentioned the YOLOv4 model According to the characteristics of the analysis, it is feasible to use YOLOv4 for weld recognition and detection in this paper, but there is no specific plan for using YOLOv4 for feature point extraction.

Images