Scholarlink

<p>This study presents a machine learning-based framework for optimizing Al-Mg-Si alloy extrusions, where achieving both high strength and ductility remains a persistent challenge. Using 263 industrial instances with 31 process parameters, six models were developed to predict tensile properties. Among them, XGBoost achieved the highest accuracy, with test-set R<sup>2</sup> values of 0.89 for tensile strength, 0.90 for yield strength, and 0.76 for elongation. Feature importance analysis using Shapley values confirmed that the model captured metallurgically meaningful factors such as Mg, Si, homogenization temperature, and aging time. Based on this interpretable model, an inverse design approach was implemented. Out of 100,000 virtual samples, 12,083 satisfied the development targets (≥300&#xa0;MPa strength and ≥ 8.0% elongation). Two top candidates were fabricated, and their experimental results showed deviations within ± 5&#xa0;MPa for strength and ± 1.2% for elongation from model predictions. Unlike previous studies limited to alloy composition or aging, our framework captures the full manufacturing chain from casting through aging using real production data. This approach provides a reliable and scalable solution for designing high-performance aluminum extrusions and reveals interpretable process–property relationships to accelerate materials development.</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

Process Optimization of Al-Mg-Si Alloy Extrusions via Machine Learning and Inverse Design

  • 期刊论文

  • 发表日期：

    2026年1月29日

查看全文
  • Yoshikazu Hayashi,
  • Kenjiro Sugio,
  • Gen Sasaki,
  • Masahiko Shioi,
  • Masato Okuno

摘要

This study presents a machine learning-based framework for optimizing Al-Mg-Si alloy extrusions, where achieving both high strength and ductility remains a persistent challenge. Using 263 industrial instances with 31 process parameters, six models were developed to predict tensile properties. Among them, XGBoost achieved the highest accuracy, with test-set R2 values of 0.89 for tensile strength, 0.90 for yield strength, and 0.76 for elongation. Feature importance analysis using Shapley values confirmed that the model captured metallurgically meaningful factors such as Mg, Si, homogenization temperature, and aging time. Based on this interpretable model, an inverse design approach was implemented. Out of 100,000 virtual samples, 12,083 satisfied the development targets (≥300 MPa strength and ≥ 8.0% elongation). Two top candidates were fabricated, and their experimental results showed deviations within ± 5 MPa for strength and ± 1.2% for elongation from model predictions. Unlike previous studies limited to alloy composition or aging, our framework captures the full manufacturing chain from casting through aging using real production data. This approach provides a reliable and scalable solution for designing high-performance aluminum extrusions and reveals interpretable process–property relationships to accelerate materials development.