Scholarlink

<p>Addressing the challenges of weld bead feature extraction in thick steel plate multi-layer multi-pass automated welding using a structured laser vision sensing method, this study developed a hybrid measurement system integrating a single-stripe laser generator and a high-resolution camera to acquire weld bead profiles during fine-wire submerged arc welding. A novel method combining adaptive region growing segmentation with the YOLOv8 deep learning algorithm is proposed for the accurate extraction of weld bead features in each layer. Furthermore, YOLOv8 is employed to obtain pixel coordinates of key feature points within regions of interest (ROI) in the multi-layer multi-pass weld bead images, enabling the reconstruction and planning of the welding seam tracking path. Experimental results demonstrate that the proposed segmentation algorithm effectively extracted characteristics from both laser stripe images and multi-layer multi-pass weld seams. The trained deep learning model achieved a root mean square error (RMSE) of 0.0551&#xa0;mm and a feature point extraction accuracy of approximately 98.41%, confirming its high accuracy and robustness.</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

Advanced feature extraction and path planning for multi-layer multi-pass welding using region growing segmentation and deep learning integration

  • 期刊论文

  • 发表日期：

    2025年11月1日

查看全文
  • Gang Zhang,
  • Zhimin Du,
  • Mengyu Jiao,
  • Ming Zhu,
  • Yu Shi

摘要

Addressing the challenges of weld bead feature extraction in thick steel plate multi-layer multi-pass automated welding using a structured laser vision sensing method, this study developed a hybrid measurement system integrating a single-stripe laser generator and a high-resolution camera to acquire weld bead profiles during fine-wire submerged arc welding. A novel method combining adaptive region growing segmentation with the YOLOv8 deep learning algorithm is proposed for the accurate extraction of weld bead features in each layer. Furthermore, YOLOv8 is employed to obtain pixel coordinates of key feature points within regions of interest (ROI) in the multi-layer multi-pass weld bead images, enabling the reconstruction and planning of the welding seam tracking path. Experimental results demonstrate that the proposed segmentation algorithm effectively extracted characteristics from both laser stripe images and multi-layer multi-pass weld seams. The trained deep learning model achieved a root mean square error (RMSE) of 0.0551 mm and a feature point extraction accuracy of approximately 98.41%, confirming its high accuracy and robustness.