Real-time laser weld point seam tracking system for robotic welding

Abstract

This paper presents a real-time vision-based laser weld seam tracking system optimized for robotic welding. The core innovation is a computer vision algorithm that processes video data to detect welding points with high precision, achieving an average absolute error of ±0.23 mm, with varying precision for different joint types (butt joint: ±0.63 mm, lap joint: ±0.04 mm, circular lap joint: ±0.02 mm). Integrated into an NVIDIA Jetson Nano, the system demonstrates robust real-time performance, processing video at 30 FPS for 720p resolution. By leveraging HSV color space analysis, morphological operations, and contour detection, the system effectively isolates and tracks laser points under dynamic conditions. Experimental results across butt, lap, and circular lap joints highlight its adaptability, with errors varying due to joint geometry and environmental factors. Comparative analysis shows superior accuracy over existing methods, such as a 0.31 mm error in laser-structured systems. By integrating directly with robotic welding tools, the system enables precise real-time adjustments, reducing human intervention and improving weld quality. This study provides a significant contribution to advancing automated welding technology.