Real-Time Seam Tracking without Spatial Lag using Reversed Electrode Image in Robotic GTAW — Part I: Seam Tracking of Variable Curvature Weld

This paper provides a new perspective on seam tracking in the welding process that has no spatial lag and does not require additional auxiliary light sources. The method is helpful for weld tracking under complex working conditions. The proposed seam tracking method is RST-REI (robotic seam tracking by REI), which is based on weld pool reversed electrode images (REIs) in the GTAW process. By using the passive weld pool image of the welding process and the relationship of REI and the welding torch pose, RST-REI achieved high-precision weld seam tracking and correction. RST-REI consists of two parts: first, a weld tracking model based on REI, which can calculate the error between the welding torch and the position to be welded through the tungsten electrode, REI, and passive image information and correct the robot pose in the welding process; second, an efficient and robust image processing algorithm, which uses the segment anything model to extract the electrode and REI foreground image in the weld pool image in real time. With the help of quadratic curve fitting, it could accurately extract the required parameters for calculating the welding torch pose in the RST-REI model. Furthermore, the experimental results of straight-line, right-angle, and S-shaped weld seams showed significant performance with the tracking error within 30 pixels, which was about 0.5 mm in our experiments. The tracking results met the weld tracking requirements of the actual welding process.