Scholarlink

<p>Ni-based single crystal superalloys play a vital role in critical areas such as aerospace and gas turbines due to their superior high-temperature strength. However, accurately predicting the creep rupture life of these alloys has been a challenge. In this study, an artificial neural network-based prediction model was developed to effectively improve the accuracy of creep life prediction for Ni-based single crystal superalloys by incorporating 15 new descriptors. The R<sup>2</sup> for the test set was 0.8595. Further, the SHAP value results guided the design of new low-cost, high-performance alloys, among which the new designed alloy (5.91 Cr, 6.21 Co, 1.62 Mo, 6.37 W, 5.64 Al, 7.22 Ta, 1.45 Re, 0.52 Ti and Ni balance, wt%) showed a higher creep life than the existing alloy CMSX-4, while having a Re content &lt; 1.5 wt%. The results not only provide new tools for superalloy design, but also confirm the practical value of machine learning in materials science.</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

Integrating Machine Learning and Thermodynamic Descriptors for Enhanced Ni-Based Single Crystal Superalloys Creep Life Prediction and Alloy Design

  • 期刊论文

  • 发表日期：

    2025年3月22日

查看全文
  • Jian Yao,
  • Wanchan Yu,
  • Juncheng Wang,
  • Longfei Zhang,
  • Feng Liu,
  • Weifu Li,
  • Liming Tan,
  • Lan Huang,
  • Yong Liu

摘要

Ni-based single crystal superalloys play a vital role in critical areas such as aerospace and gas turbines due to their superior high-temperature strength. However, accurately predicting the creep rupture life of these alloys has been a challenge. In this study, an artificial neural network-based prediction model was developed to effectively improve the accuracy of creep life prediction for Ni-based single crystal superalloys by incorporating 15 new descriptors. The R2 for the test set was 0.8595. Further, the SHAP value results guided the design of new low-cost, high-performance alloys, among which the new designed alloy (5.91 Cr, 6.21 Co, 1.62 Mo, 6.37 W, 5.64 Al, 7.22 Ta, 1.45 Re, 0.52 Ti and Ni balance, wt%) showed a higher creep life than the existing alloy CMSX-4, while having a Re content < 1.5 wt%. The results not only provide new tools for superalloy design, but also confirm the practical value of machine learning in materials science.

Graphical Abstract