Scholarlink

<p>This study proposed an interpretable machine learning (ML) framework for predicting phase and hardness of the AlCrFeNiCuSi high-entropy alloys (HEAs). A comprehensive database of 65 features was constructed, and optimal feature subsets were systematically selected: 9 features for phase prediction and 10 for hardness prediction. Multiple ML models were then applied to predict the phase and hardness, with the optimal model chosen based on comparative performance. Comprehensive evaluation showed that generative adversarial networks-augmented random forest achieved 94.6% phase classification accuracy. The extreme gradient boosting model attained an <i>R</i><sup>2</sup> of 0.955 for hardness prediction. Underlying physical mechanisms were elucidated through Shapley additive explanations (SHAP) and one-way analysis of variance (ANOVA) analysis: valence electron concentration and atomic size difference (<i>δ</i>) were identified as the dominant factors controlling phase formation and hardness, respectively. Guided by the ML framework, the AlCrFeNiCuSi HEAs were screened at low cost and prepared, while the phase and hardness were effectively predicted and verified. This work will advance the intelligent design of HEAs and bridge artificial intelligence with materials science.</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

Phase Identification and Hardness Prediction of Al-Cr-Fe-Ni-Cu-Si High-Entropy Alloys Guided by Explainable Machine Learning

  • 期刊论文

  • 发表日期：

    2025年10月27日

查看全文
  • Tianming Li,
  • Mengdi Zhang,
  • Hanqing Xu,
  • Zhuoyi Wang,
  • Xin Zhao

摘要

This study proposed an interpretable machine learning (ML) framework for predicting phase and hardness of the AlCrFeNiCuSi high-entropy alloys (HEAs). A comprehensive database of 65 features was constructed, and optimal feature subsets were systematically selected: 9 features for phase prediction and 10 for hardness prediction. Multiple ML models were then applied to predict the phase and hardness, with the optimal model chosen based on comparative performance. Comprehensive evaluation showed that generative adversarial networks-augmented random forest achieved 94.6% phase classification accuracy. The extreme gradient boosting model attained an R2 of 0.955 for hardness prediction. Underlying physical mechanisms were elucidated through Shapley additive explanations (SHAP) and one-way analysis of variance (ANOVA) analysis: valence electron concentration and atomic size difference (δ) were identified as the dominant factors controlling phase formation and hardness, respectively. Guided by the ML framework, the AlCrFeNiCuSi HEAs were screened at low cost and prepared, while the phase and hardness were effectively predicted and verified. This work will advance the intelligent design of HEAs and bridge artificial intelligence with materials science.