Scholarlink

<p>Grain boundary (GB) energy has a significant influence on the mechanical properties and behavior of polycrystalline materials. At present, the prediction model of GB energy and the relationship between GB energy and structure still need further investigation. Herein, we combine atomistic simulations with machine learning (ML) methods to establish a prediction model for GB energy and examine the correlation between GB energy and GB parameters. A large GB energy dataset that contains 100000 aluminum GBs of five macroscopic degrees of freedom (DOFs) was created by conducting a series of molecular dynamics (MD) simulations. Except for GB energy, the excess volume and excess centrosymmetry of GB were also calculated using the cut-off sphere model by MD to investigate their correlations with GB energy. The results indicate that the machine learning models are capable of precisely predicting the GB energy as a function of five GB DOFs, which is otherwise challenging for the conventional analytical approaches. Furthermore, the ML models using excess volume or excess centrosymmetry as input, respectively, have 10.6% and 12.9% higher prediction scores than the corresponding well-accepted theoretical linear models, indicating the high predictive accuracy of ML techniques. In addition, the effect of input features on the prediction performance of ML models, the feature importance analysis, and the relations between GB parameters (five GB DOFs, GB excess volume, and excess centrosymmetry) and GB energy are systematically discussed. The MD- and ML-based investigation in this study provides a deeper understanding of the relationship between GB energy-structure and offers a convenient and accurate method for predicting GB energy from GB parameters.</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

Prediction of grain boundary energy and understanding of grain boundary structure-energy relationship by machine learning

  • 期刊论文

  • 发表日期：

    2026年7月28日

查看全文
  • Songjiang Lu

摘要

Grain boundary (GB) energy has a significant influence on the mechanical properties and behavior of polycrystalline materials. At present, the prediction model of GB energy and the relationship between GB energy and structure still need further investigation. Herein, we combine atomistic simulations with machine learning (ML) methods to establish a prediction model for GB energy and examine the correlation between GB energy and GB parameters. A large GB energy dataset that contains 100000 aluminum GBs of five macroscopic degrees of freedom (DOFs) was created by conducting a series of molecular dynamics (MD) simulations. Except for GB energy, the excess volume and excess centrosymmetry of GB were also calculated using the cut-off sphere model by MD to investigate their correlations with GB energy. The results indicate that the machine learning models are capable of precisely predicting the GB energy as a function of five GB DOFs, which is otherwise challenging for the conventional analytical approaches. Furthermore, the ML models using excess volume or excess centrosymmetry as input, respectively, have 10.6% and 12.9% higher prediction scores than the corresponding well-accepted theoretical linear models, indicating the high predictive accuracy of ML techniques. In addition, the effect of input features on the prediction performance of ML models, the feature importance analysis, and the relations between GB parameters (five GB DOFs, GB excess volume, and excess centrosymmetry) and GB energy are systematically discussed. The MD- and ML-based investigation in this study provides a deeper understanding of the relationship between GB energy-structure and offers a convenient and accurate method for predicting GB energy from GB parameters.