Scholarlink

<p>To mitigate undercut and humping defects induced by arc instability during high-speed coarse-wire gas metal arc welding (MAG), this study proposes a lightweight multi-frame spatio-temporal convolutional network for adaptive defect control. The network takes consecutive multi-frame images as input and employs an ordered cross-entropy loss with reward–penalty weighting to improve defect recognition accuracy. Model lightweighting is achieved via spatio-temporal decomposed convolutions, a depthwise-separable bottleneck, and global adaptive pooling, resulting in a 61% reduction in parameter count. Experiments under different groove configurations and welding angles show that the overall performance is optimal with a five-frame input, achieving an accuracy of 96.1%. For the four welding states, the model attains a macro-averaged F1 score of 0.8315 and an area under the curve (AUC) of 0.965, with a recall of 0.82 for critical defects (undercut/humping), and an average network forward inference latency of 1.5 ms. Furthermore, in proof-of-concept closed-loop tests, the controller exhibits stable behavior and improved weld bead formation, indicating real-time feasibility and deployment potential on the current prototype. This work provides practical guidance for the design of future adaptive welding systems targeting high-speed coarse-wire MAG processes.</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

High-speed monitoring and control of forming defects in coarse-wire MAG welding based on multi-frame spatio-temporal neural networks

  • 期刊论文

  • 发表日期：

    2026年6月15日

查看全文
  • Hao Yuan,
  • Shibo Wei,
  • Jiamin Yi,
  • Chaoke Li,
  • Ziran Wang,
  • Zekun Yang,
  • Zhanli Yang

摘要

To mitigate undercut and humping defects induced by arc instability during high-speed coarse-wire gas metal arc welding (MAG), this study proposes a lightweight multi-frame spatio-temporal convolutional network for adaptive defect control. The network takes consecutive multi-frame images as input and employs an ordered cross-entropy loss with reward–penalty weighting to improve defect recognition accuracy. Model lightweighting is achieved via spatio-temporal decomposed convolutions, a depthwise-separable bottleneck, and global adaptive pooling, resulting in a 61% reduction in parameter count. Experiments under different groove configurations and welding angles show that the overall performance is optimal with a five-frame input, achieving an accuracy of 96.1%. For the four welding states, the model attains a macro-averaged F1 score of 0.8315 and an area under the curve (AUC) of 0.965, with a recall of 0.82 for critical defects (undercut/humping), and an average network forward inference latency of 1.5 ms. Furthermore, in proof-of-concept closed-loop tests, the controller exhibits stable behavior and improved weld bead formation, indicating real-time feasibility and deployment potential on the current prototype. This work provides practical guidance for the design of future adaptive welding systems targeting high-speed coarse-wire MAG processes.