Scholarlink

Abstract <p>Uniform mechanical properties in medium and thick steel plates are essential for structural reliability in critical engineering applications. However, achieving stable and controllable mechanical performance during hot rolling remains challenging due to the complex coupling between process parameters and microstructural evolution. Conventional approaches based on macroscopic process optimization often fail to capture the essential role of microstructural dynamics, resulting in persistent variability and limited process transparency. To address this issue, we propose a Feature-enhanced Theory-Guided Data-Driven (FTGDD) modeling framework that incorporates theoretically derived microstructural descriptors alongside empirical industrial data to construct a robust multiscale dataset. Leveraging multiscale feature fusion and perturbation-based sensitivity analysis, the FTGDD framework demonstrates superior accuracy, robustness, and generalization capability in mechanical property prediction when compared to conventional data-driven models. In addition, the framework enables interpretable identification of latent sources of property variation, providing targeted guidance for process refinement. By integrating domain knowledge with data-driven insights, the FTGDD method offers a practical and scalable framework for reliable prediction and control of mechanical properties in hot rolling, contributing to the development of robust and interpretable modeling strategies for consistent quality assurance in steel plate production.</p> Graphical abstract <p></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

Physically Informed Multiscale Learning for Property Prediction and Process Optimization in Hot Rolling

  • 期刊论文

  • 发表日期：

    2025年6月19日

查看全文
  • Pengju Xu,
  • Yifan Yan,
  • Zhimin Lv,
  • Zhiyan Zhang,
  • Yiquan An,
  • Keming Hu,
  • Ri Jin,
  • Ziqing Zhu

摘要

Abstract

Uniform mechanical properties in medium and thick steel plates are essential for structural reliability in critical engineering applications. However, achieving stable and controllable mechanical performance during hot rolling remains challenging due to the complex coupling between process parameters and microstructural evolution. Conventional approaches based on macroscopic process optimization often fail to capture the essential role of microstructural dynamics, resulting in persistent variability and limited process transparency. To address this issue, we propose a Feature-enhanced Theory-Guided Data-Driven (FTGDD) modeling framework that incorporates theoretically derived microstructural descriptors alongside empirical industrial data to construct a robust multiscale dataset. Leveraging multiscale feature fusion and perturbation-based sensitivity analysis, the FTGDD framework demonstrates superior accuracy, robustness, and generalization capability in mechanical property prediction when compared to conventional data-driven models. In addition, the framework enables interpretable identification of latent sources of property variation, providing targeted guidance for process refinement. By integrating domain knowledge with data-driven insights, the FTGDD method offers a practical and scalable framework for reliable prediction and control of mechanical properties in hot rolling, contributing to the development of robust and interpretable modeling strategies for consistent quality assurance in steel plate production.

Graphical abstract