Scholarlink

<p>Plastic deformation governs the performance and failure of ductile materials and is often mediated by dislocation slip. In polycrystals, slip in one grain can trigger slip in neighboring grains, known as slip transmission, linking grain-scale events to bulk plasticity. We pose a fundamental question: if the location of an initial slip event is known, can we predict its transmission pathway through a grain network? Using in situ high-energy diffraction microscopy, we measure slip transmission in a 3D grain network. These measurements inform the development of a topology-guided graph neural network for slip transmission (TGNN-ST). TGNN-ST integrates grain, grain-boundary, and network-topology features and operates on local grain-neighborhood subgraphs, explicitly capturing the surrounding microstructural context of candidate transmission events. This approach outperforms phenomenological criteria, conventional deep learning models, and node-based GNNs. Beyond prediction, the model reveals that transmission is dominated by first-order neighbors and governed by local network topology, providing new insight into plastic deformation.</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

Predicting slip transmission pathways using graph neural networks

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月8日

查看全文
  • Yuefeng Jin,
  • Peiyu Zhang,
  • Xiongye Xiao,
  • Ezra Mengiste,
  • Wenxi Li,
  • Amlan Das,
  • Katherine Shanks,
  • Matthew Kasemer,
  • Paul Bogdan,
  • Ashley Bucsek

摘要

Plastic deformation governs the performance and failure of ductile materials and is often mediated by dislocation slip. In polycrystals, slip in one grain can trigger slip in neighboring grains, known as slip transmission, linking grain-scale events to bulk plasticity. We pose a fundamental question: if the location of an initial slip event is known, can we predict its transmission pathway through a grain network? Using in situ high-energy diffraction microscopy, we measure slip transmission in a 3D grain network. These measurements inform the development of a topology-guided graph neural network for slip transmission (TGNN-ST). TGNN-ST integrates grain, grain-boundary, and network-topology features and operates on local grain-neighborhood subgraphs, explicitly capturing the surrounding microstructural context of candidate transmission events. This approach outperforms phenomenological criteria, conventional deep learning models, and node-based GNNs. Beyond prediction, the model reveals that transmission is dominated by first-order neighbors and governed by local network topology, providing new insight into plastic deformation.