Scholarlink

<p>Al-Li alloys, while offering low density and high specific strength, exhibit a pronounced trade-off between mechanical strength and solidification cracking resistance, significantly limiting their engineering applications. Traditional trial-and-error approaches prolong development cycles and incur high labor and material costs. Herein, we adopt a data augmentation-assisted machine learning (ML) strategy to accelerate the discovery of high-strength Al-Li alloy with enhanced cracking resistance. Given the scarcity of available data, various ML algorithms were compared, and linear regression (LR) was selected for its high predictive accuracy and robustness against overfitting. Based on this model, data augmentation was performed using SMOGN and CVAE methods. SHAP analysis identified that Cu and Sc are the most influential elements for strength, while cracking resistance is predominantly governed by Sc and Li. Under the constraint of low cracking volume, optimal compositions were predicted using LR, SMOGN and CVAE, followed by experimental validation. The alloy predicted by SMOGN achieves a superior yield strength of 412&#xa0;MPa, exceeding that of current high crack-resistant Al-Li alloys. In contrast, the alloy designed by LR exhibits inferior mechanical performance due to a low Cu/Mg ratio (~ 0.8), which suppresses the precipitation of strengthening T<sub>1</sub>- Al<sub>2</sub>CuLi on the preferred {111}<sub>Al</sub> planes and promotes the formation of coarse, coalesced S′- Al<sub>2</sub>CuMg on {210}<sub>Al</sub> planes. The latter not only provides limited strengthening but also induces local strain concentration at the interfaces, leading to early crack initiation and reduced ductility. The CVAE-predicted alloy exhibits a large discrepancy between predicted and experimental strength, attributed to insufficient Mg content that reduced the nucleation efficiency and stability of T<sub>1</sub>-Al<sub>2</sub>CuLi, resulting in a sparse distribution and diminished strengthening effect. These findings provide valuable insights for the design of high-strength and crack-resistant Al-Li alloys.</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

Data augmentation-assisted machine learning for accelerated discovery of high-strength lightweight alloys with enhanced solidification cracking resistance

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月3日

查看全文
  • Wengang Bu,
  • Yihao Wang,
  • Yue Zhang,
  • Jiamao Hao,
  • Rong Wang,
  • Zhenfeng Hu,
  • Xiubing Liang

摘要

Al-Li alloys, while offering low density and high specific strength, exhibit a pronounced trade-off between mechanical strength and solidification cracking resistance, significantly limiting their engineering applications. Traditional trial-and-error approaches prolong development cycles and incur high labor and material costs. Herein, we adopt a data augmentation-assisted machine learning (ML) strategy to accelerate the discovery of high-strength Al-Li alloy with enhanced cracking resistance. Given the scarcity of available data, various ML algorithms were compared, and linear regression (LR) was selected for its high predictive accuracy and robustness against overfitting. Based on this model, data augmentation was performed using SMOGN and CVAE methods. SHAP analysis identified that Cu and Sc are the most influential elements for strength, while cracking resistance is predominantly governed by Sc and Li. Under the constraint of low cracking volume, optimal compositions were predicted using LR, SMOGN and CVAE, followed by experimental validation. The alloy predicted by SMOGN achieves a superior yield strength of 412 MPa, exceeding that of current high crack-resistant Al-Li alloys. In contrast, the alloy designed by LR exhibits inferior mechanical performance due to a low Cu/Mg ratio (~ 0.8), which suppresses the precipitation of strengthening T1- Al2CuLi on the preferred {111}Al planes and promotes the formation of coarse, coalesced S′- Al2CuMg on {210}Al planes. The latter not only provides limited strengthening but also induces local strain concentration at the interfaces, leading to early crack initiation and reduced ductility. The CVAE-predicted alloy exhibits a large discrepancy between predicted and experimental strength, attributed to insufficient Mg content that reduced the nucleation efficiency and stability of T1-Al2CuLi, resulting in a sparse distribution and diminished strengthening effect. These findings provide valuable insights for the design of high-strength and crack-resistant Al-Li alloys.