Scholarlink

<p>Accurate prediction of glass-forming ability (GFA) and optimization of alloy composition are crucial for advancing the development of bulk metallic glasses (BMGs). This study systematically analyzes the key factors affecting GFA. The comprehensive adaptive mutual information selector (CAMI) and sparse autoencoder (SAE) are employed to extract core parameters from complex datasets, while six data balancing methods address sample size imbalance. The proposed enhanced attention gradient ExtraTrees ensemble model (EAGET) integrates efficient channel attention and a cross-channel attention module, enhancing prediction performance through gradient-boosted residual learning. EAGET dynamically allocates weights based on the importance of individual decision trees, addressing limitations of traditional fixed-weight ensemble models. On independent test data, EAGET achieves an exceptional <i>R</i><sup>2</sup> = 0.86, with prediction accuracy 26% higher than conventional models. Using Shapley Additive Explanations (SHAP), critical parameters influencing GFA are identified and successfully guided the design of new alloys in the Zr–Cu–Al–Ag and Pd–Si–Ag–Cu systems, verifying practical utility of model in real-world BMGs development.</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

Prediction of glass-forming ability and alloys design via enhanced attention gradient ExtraTrees ensemble model

  • 期刊论文

  • 发表日期：

    2025年8月4日

查看全文
  • Zhi-han Zhang,
  • Yong-chao Liang,
  • Duan-jie Cheng,
  • Yun-jun Ruan,
  • Qian Chen

摘要

Accurate prediction of glass-forming ability (GFA) and optimization of alloy composition are crucial for advancing the development of bulk metallic glasses (BMGs). This study systematically analyzes the key factors affecting GFA. The comprehensive adaptive mutual information selector (CAMI) and sparse autoencoder (SAE) are employed to extract core parameters from complex datasets, while six data balancing methods address sample size imbalance. The proposed enhanced attention gradient ExtraTrees ensemble model (EAGET) integrates efficient channel attention and a cross-channel attention module, enhancing prediction performance through gradient-boosted residual learning. EAGET dynamically allocates weights based on the importance of individual decision trees, addressing limitations of traditional fixed-weight ensemble models. On independent test data, EAGET achieves an exceptional R2 = 0.86, with prediction accuracy 26% higher than conventional models. Using Shapley Additive Explanations (SHAP), critical parameters influencing GFA are identified and successfully guided the design of new alloys in the Zr–Cu–Al–Ag and Pd–Si–Ag–Cu systems, verifying practical utility of model in real-world BMGs development.

Graphical abstract