Scholarlink

<p>The path planning of welding robots is a critical step toward achieving high-quality robotic welding. Automatic seam detection and path planning have emerged as a prominent research hotspot. This study proposed a method for automatic robotic welding path planning. First, an efficient deep neural network (denoted as EfficientWeld) is developed for weld seam detection and endpoint localization, with the objectives of achieving gross localization of weld seam regions and guiding subsequent precise scanning. The EfficientWeld model demonstrates accurate extraction of start and end points for butt joints, lap joints, and irregular joints, achieves a 36% improvement in inference speed, and exhibits robust performance in scenarios involving multiple weld seams and varying backgrounds. Subsequently, the identified joint start and end points are utilized to guide structured laser sensors in performing fine scanning of workpieces and extraction of weld seams. To accelerate feature extraction, an enhanced gray gravity method is proposed. Finally, a spatial position and pose model for complex curved weld seams is constructed using 3D point cloud data, enabling high-precision path planning and robotic welding posture planning for complex curved seams. The root mean square error (RMSE) of the generated path is less than 0.25&#xa0;mm, with a maximum lateral error below 0.61&#xa0;mm, satisfying the requirements for practical industrial 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

Weld seam detection based on 2D image and 3D point cloud segmentation: a coarse-to-fine weld path planning method

  • 期刊论文

  • 发表日期：

    2025年11月22日

查看全文
  • Chengyuan Ma,
  • Zheng Meng,
  • Zhaochuan Wang,
  • Shenghong Yan,
  • Bo Chen,
  • Caiwang Tan,
  • Xiaoguo Song

摘要

The path planning of welding robots is a critical step toward achieving high-quality robotic welding. Automatic seam detection and path planning have emerged as a prominent research hotspot. This study proposed a method for automatic robotic welding path planning. First, an efficient deep neural network (denoted as EfficientWeld) is developed for weld seam detection and endpoint localization, with the objectives of achieving gross localization of weld seam regions and guiding subsequent precise scanning. The EfficientWeld model demonstrates accurate extraction of start and end points for butt joints, lap joints, and irregular joints, achieves a 36% improvement in inference speed, and exhibits robust performance in scenarios involving multiple weld seams and varying backgrounds. Subsequently, the identified joint start and end points are utilized to guide structured laser sensors in performing fine scanning of workpieces and extraction of weld seams. To accelerate feature extraction, an enhanced gray gravity method is proposed. Finally, a spatial position and pose model for complex curved weld seams is constructed using 3D point cloud data, enabling high-precision path planning and robotic welding posture planning for complex curved seams. The root mean square error (RMSE) of the generated path is less than 0.25 mm, with a maximum lateral error below 0.61 mm, satisfying the requirements for practical industrial applications.