Scholarlink

<p>The traditional design of single-crystal superalloys relies heavily on trial-and-error experimentation, which is time-consuming and costly. Here, we present an intelligent alloy design strategy that integrates natural language processing (NLP) and machine learning (ML). A domain-specific NLP model was developed to automatically extract γ′ solvus temperature data from scientific literature, enabling the construction of a high-quality database. Machine learning models trained on this data accurately predict both γ′ solvus temperature and creep life. Guided by these models, we screened over 340000 virtual compositions and successfully designed a new low-cost alloy, CSU-S1. Experimental validation shows that CSU-S1 achieves a γ′ solvus temperature near 1300 °C and a creep life of 224.7 h at 1100 °C/137 MPa, comparable to third-generation single-crystal superalloys, while using only 3.1 wt% Re and costing just 121 USD/kg. This work not only delivers a high-performance, cost-effective superalloy but also demonstrates a generalizable “knowledge-to-innovation” design paradigm, offering a powerful new route to accelerate the development of advanced engineering materials.</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

Alloy design integrating natural language processing and machine learning: breakthrough development of low-cost, high-performance Ni-based single-crystal superalloys

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年12月19日

查看全文
  • Jian Yao,
  • Zi Wang,
  • Juncheng Wang,
  • Wanchan Yu,
  • Yuxuan Chen,
  • Weifu Li,
  • Jianhui Wei,
  • Yunxing Zhao,
  • Yan Wang,
  • Li Wang,
  • Liming Tan,
  • Lan Huang,
  • Feng Liu,
  • Yong Liu

摘要

The traditional design of single-crystal superalloys relies heavily on trial-and-error experimentation, which is time-consuming and costly. Here, we present an intelligent alloy design strategy that integrates natural language processing (NLP) and machine learning (ML). A domain-specific NLP model was developed to automatically extract γ′ solvus temperature data from scientific literature, enabling the construction of a high-quality database. Machine learning models trained on this data accurately predict both γ′ solvus temperature and creep life. Guided by these models, we screened over 340000 virtual compositions and successfully designed a new low-cost alloy, CSU-S1. Experimental validation shows that CSU-S1 achieves a γ′ solvus temperature near 1300 °C and a creep life of 224.7 h at 1100 °C/137 MPa, comparable to third-generation single-crystal superalloys, while using only 3.1 wt% Re and costing just 121 USD/kg. This work not only delivers a high-performance, cost-effective superalloy but also demonstrates a generalizable “knowledge-to-innovation” design paradigm, offering a powerful new route to accelerate the development of advanced engineering materials.