Robotic Pipe Welding Weld Recognition Based on Machine Vision

This study addresses the issue of insufficient accuracy and real-time performance in weld seam recognition in pipeline welding automation. A machine vision based method for pipeline welding seam recognition is proposed, and a collaborative system of “visual perception data solution motion execution” is constructed. By using the Zhang Zhengyou calibration method and the Tsai Lenz hand eye calibration algorithm, an accurate mapping between the camera coordinate system and the robot base coordinate system is established. The hand eye conversion error is controlled within 0.5 mm, and the reprojection error is 0.1575 pixels. The image processing module integrates the improved geometric center method and YOLOv5s deep learning model to achieve single pixel skeleton extraction of laser stripes (positioning error of 0.38 pixels) and complex weld seam feature recognition (accuracy of 98.2%). The 3D reconstruction is based on robot motion simulation binocular vision, combined with the principle of triangulation and epipolar constraints, with a depth error of ${\lt}1.2 \mathrm{~mm}$. The motion control module uses fifth degree polynomial interpolation and PID dynamic compensation to achieve realtime trajectory tracking error of $\lt0.32 \mathrm{~mm}$ in serrated strip mode, improve welding efficiency to $250 \mathrm{~mm} / \mathrm{min}$, and significantly optimize forming indicators such as melt width uniformity (standard deviation 0.7 mm) and residual height fluctuation ($\pm 0.3 \mathrm{~mm}$). The experimental results show that the system has an accuracy rate of over 95% in identifying butt/corner welds (with a groove angle of 60° and a gap of 23 mm), verifying its industrial applicability. Research suggests that in the future, it is necessary to further enhance system robustness and engineering adaptability through online self calibration algorithms, multi brand robot compatibility protocols, model lightweighting, and multi-sensor fusion technology.