A cross-process welding penetration status prediction algorithm based on unsupervised domain adaptation in laser and TIG welding

论文概要

研究领域: 计算机视觉 作者: Sen Li, Haichao Cui, Chendong Shao 发布时间: 2026-06-25 arXiv: 2606.19223

中文摘要

监督深度学习已广泛用于焊接熔透状态分类；然而，其性能在域偏移下通常显著下降，例如在具有不同物理机制的焊接工艺之间迁移模型时：例如从电弧主导的钨极惰性气体（TIG）焊接到基于匙孔的激光焊接。为克服这一限制，我们提出了一种结合渐进源域扩展（GSDE）策略的无监督域自适应（UDA）框架。在专用TIG和激光焊接数据集上评估，我们的方法在同工艺和跨工艺迁移任务中都实现了高准确率。具体而言，在同工艺设置中，它在TIGFH上达到90.65%、在LSPS上达到90.72%的平均准确率，分别超越监督基线35.83%和38.87%。更值得注意的是，在跨工艺场景中，TIG到激光达到80.48%、激光到TIG达到81.13%，比基线提高43.39%和43.40%。UMAP可视化验证模型学习了域不变特征同时保持判别性类别边界。这种方法显著降低了新焊接工艺的重标注成本，增强了不同焊接系统间智能监控的通用性。

原文摘要

Supervised deep learning has been widely used for weld penetration state classification; however, its performance often degrades significantly under domain shift, such as when transferring models between welding processes with distinct physical mechanisms:for instance, from arc-dominated tungsten inert gas (TIG) welding to keyhole-based laser welding. To overcome this limitation, we propose an unsupervised domain adaptation (UDA) framework integrated with a gradual source domain expansion (GSDE) strategy. Evaluated on dedicated TIG and laser welding datasets, our approach achieves high accuracy in both same-process and cross-process transfer tasks. Specifically, it attains average accuracies of 90.65% on TIGFH and 90.72% on LSPS in same-process settings, surpassing a supervised baseline by...

--- *自动采集于 2026-06-26*

#论文 #arXiv #计算机视觉 #小凯