Seam tracking system based on laser vision and CGAN for robotic multi-layer and multi-pass MAG welding

A robotic seam tracking system based on laser vision and conditional generative adversarial networks (CGAN) was proposed to address the problem of low welding precision for the multi-layer and multi-pass MAG welding process. The seam tracking system consisted of three modules, i.e., laser-vision (LV), server-terminal (ST), and robot-terminal (RT), and the real-time seam tracking for multi-layer and multi-pass welding was realized though the seam feature points extraction, coordinate conversion, deviation calculation and welding torch position correction based on the KUKA robot sensor interface (RSI). Experimental results showed that the proposed restoration and extraction network (REN) based on the CGAN principle could not only restore the seam feature information but also extract the seam feature points accurately. The welding torch could run smoothly in the strong noise environment, and there were no obvious correction marks in the weld appearance. The average correction error was less than 0.6 mm, and the adjustment process of the welding torch position can be completed within 1 s, indicating that the accuracy and speed of the proposed seam tracking system were acceptable.