Scholarlink

<p>Irradiation hardening of reactor pressure vessel (RPV) steels is a critical factor affecting the long-term safety of nuclear reactors. Previous machine learning approaches for predicting hardening and embrittlement have been primarily data-driven, which has restricted their extrapolation capability due to insufficient consideration of underlying physical mechanisms. Existing attempts to introduce physical constraints have yielded only limited improvements in prediction accuracy. This study develops a physics-informed machine learning strategy to predict irradiation hardening of RPV steels. Known mechanisms of irradiation hardening revealed by existing experimental studies and steel composition properties constructed using materials data mining methods are incorporated into feature engineering, enabling the integration of physically motivated features into the machine learning model. This approach achieves high prediction accuracy within the training data distribution and substantially improves extrapolation performance beyond the training range. To further quantify prediction uncertainty, a model ensemble with multiple initializations is constructed. Comprehensive evaluations on a hardening dataset demonstrate that the proposed approach not only enhances predictive reliability but also provides insights into the influence of individual variables on irradiation hardening, thereby confirming the effectiveness of incorporating physical knowledge into data-driven models.</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

A Study on Irradiation Hardening in Reactor Pressure Vessel Steels Using Physics-Data Fusion Machine Learning Methods

  • 期刊论文

  • 发表日期：

    2026年5月18日

查看全文
  • Yiming Tian,
  • Wei Guo,
  • Yanxiang Liang,
  • Jian Li,
  • Xicheng Huang,
  • Qiang Wan

摘要

Irradiation hardening of reactor pressure vessel (RPV) steels is a critical factor affecting the long-term safety of nuclear reactors. Previous machine learning approaches for predicting hardening and embrittlement have been primarily data-driven, which has restricted their extrapolation capability due to insufficient consideration of underlying physical mechanisms. Existing attempts to introduce physical constraints have yielded only limited improvements in prediction accuracy. This study develops a physics-informed machine learning strategy to predict irradiation hardening of RPV steels. Known mechanisms of irradiation hardening revealed by existing experimental studies and steel composition properties constructed using materials data mining methods are incorporated into feature engineering, enabling the integration of physically motivated features into the machine learning model. This approach achieves high prediction accuracy within the training data distribution and substantially improves extrapolation performance beyond the training range. To further quantify prediction uncertainty, a model ensemble with multiple initializations is constructed. Comprehensive evaluations on a hardening dataset demonstrate that the proposed approach not only enhances predictive reliability but also provides insights into the influence of individual variables on irradiation hardening, thereby confirming the effectiveness of incorporating physical knowledge into data-driven models.