Scholarlink

<p>Wire Arc Additive Manufacturing (WAAM) offers high deposition rates and low feedstock cost, but process variability can cause defects, residual stresses, and geometric errors. No single sensor captures the full process state, motivating multimodal monitoring. This review traces progress from vision- or thermal-only monitoring to machine-learning approaches for vision–thermal fusion. We summarise sensing hardware and measurable parameters; outline traditional image/thermal analyses; review recent ML methods for unimodal data (classification, detection, transfer learning, physics-informed learning); and discuss fusion architectures (early/mid/late), including their synchronisation assumptions and practical guidance for trigger-based versus learned alignment. We then connect intelligent monitoring to control strategies (inter-layer, intra-layer, and hybrid), highlighting opportunities for model-predictive and adaptive control. Open challenges include robust sensing in arc environments, sim-to-real fidelity, data scarcity, and generalisation across materials and setups. Overall, integrating complementary sensors with ML can improve process interpretation and control, supporting the transition toward more autonomous WAAM.</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

Visual and thermal monitoring techniques for Wire Arc Additive Manufacturing: from conventional to machine learning approaches

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年4月28日

查看全文
  • Maedeh Gourakani,
  • Carmelo Mineo,
  • Donatella Cerniglia,
  • Andrea Bertazzo,
  • Federica Tiso,
  • Giovanni Avallone

摘要

Wire Arc Additive Manufacturing (WAAM) offers high deposition rates and low feedstock cost, but process variability can cause defects, residual stresses, and geometric errors. No single sensor captures the full process state, motivating multimodal monitoring. This review traces progress from vision- or thermal-only monitoring to machine-learning approaches for vision–thermal fusion. We summarise sensing hardware and measurable parameters; outline traditional image/thermal analyses; review recent ML methods for unimodal data (classification, detection, transfer learning, physics-informed learning); and discuss fusion architectures (early/mid/late), including their synchronisation assumptions and practical guidance for trigger-based versus learned alignment. We then connect intelligent monitoring to control strategies (inter-layer, intra-layer, and hybrid), highlighting opportunities for model-predictive and adaptive control. Open challenges include robust sensing in arc environments, sim-to-real fidelity, data scarcity, and generalisation across materials and setups. Overall, integrating complementary sensors with ML can improve process interpretation and control, supporting the transition toward more autonomous WAAM.