A High-Accuracy and Robust Seam Tracking System Based on Adversarial Learning

Abstract

In the seam tracking process based on laser vision, noise interference such as arc and splash makes it difficult to accurately extract a weld location from a welding image. To solve this problem, although many algorithms have been proposed, they could not achieve good tracking effect when confronted with continuous strong noise interference. Therefore, a two-stage seam tracking method named Heatmap method combining welding image inpainting and feature point locating from a heatmap is proposed in this paper. Firstly, a welding image inpainting model is constructed based on adversarial learning strategy, which could effectively resist the strong noise interference. Secondly, a heatmap generating model is built to map the global feature information from a welding image to the heatmap of feature point, and realize the seam feature point locating. Furthermore, by integrating the welding image inpainting model and the heatmap generating model, a robust seam tracking system is constructed. Finally, a series of seam locating comparative tests and welding experiments are carried out. The results show that the average locating error of the proposed Heatmap seam tracking method is within ±0.15 mm, demonstrating that the proposed method could well resist interference of continuous strong noise and further improve the welding accuracy and robustness of a seam tracking system.