Scholarlink

<p>Accurate weld seam localization and autonomous guidance remain challenging for robotic welding in complex and unstructured environments. This paper investigates a method for autonomous positioning of welding workpieces based on large-field-of-view (LFoV) 3D vision and point cloud registration, laying the foundation for adaptive planning of welding paths for complex components. Firstly, an LFoV 3D vision robot welding system is established, and a hand-eye calibration method for the LFoV camera is designed to ensure system accuracy meets welding requirements. Secondly, based on scene images, the YOLOv5 (You Only Look Once) network is trained to recognize welding workpieces automatically. Subsequently, a coarse-to-fine point cloud registration strategy combining Super 4PCS and Anderson-accelerated ICP is designed to achieve robust and precise model alignment. Experimental results demonstrate that the proposed method achieves an average positioning error of 0.62&#xa0;mm and an angular error of 0.43°, confirming its effectiveness in enhancing the adaptability and precision of robotic welding in complex industrial scenarios. The proposed method provides a reliable foundation for autonomous weld seam guidance and future intelligent welding applications.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A novel autonomous guidance method for welding robot based on point cloud registration and 3D vision

  • 期刊论文

  • 发表日期：

    2026年3月18日

查看全文
  • Runquan Xiao,
  • Yanling Xu,
  • Lu Lu,
  • Xinghua Wang,
  • Xiaoyang Ma,
  • Qiang Wang,
  • Huajun Zhang

摘要

Accurate weld seam localization and autonomous guidance remain challenging for robotic welding in complex and unstructured environments. This paper investigates a method for autonomous positioning of welding workpieces based on large-field-of-view (LFoV) 3D vision and point cloud registration, laying the foundation for adaptive planning of welding paths for complex components. Firstly, an LFoV 3D vision robot welding system is established, and a hand-eye calibration method for the LFoV camera is designed to ensure system accuracy meets welding requirements. Secondly, based on scene images, the YOLOv5 (You Only Look Once) network is trained to recognize welding workpieces automatically. Subsequently, a coarse-to-fine point cloud registration strategy combining Super 4PCS and Anderson-accelerated ICP is designed to achieve robust and precise model alignment. Experimental results demonstrate that the proposed method achieves an average positioning error of 0.62 mm and an angular error of 0.43°, confirming its effectiveness in enhancing the adaptability and precision of robotic welding in complex industrial scenarios. The proposed method provides a reliable foundation for autonomous weld seam guidance and future intelligent welding applications.