Scholarlink

<p>In order to meet the urgent demand for high-precision material performance data in advanced structural design, this paper has developed and validated a data-driven machine learning framework. This framework is based on Gaussian process regression, learning from 183 sets of deformation and rapid heat-treatment experimental data, covering a wide process space, and achieving multi-objective collaborative prediction of yield strength, tensile strength, and elongation of low-carbon, low-alloy steel. the impact of different iteration strategies on the generalization ability of the model has been analyzed, and it has been confirmed that, by constructing a more complete global model, combined iteration has better robustness than transfer learning in predicting new processes. Furthermore, customizing covariance functions for different physical properties of mechanical performance is key to achieving high-fidelity modeling. The framework achieves an <i>R</i><sup>2</sup> value greater than 0.5 for new process predictions in forward design. In inverse design, combined with the expected hypervolume optimization algorithm, it successfully expands the strength–ductility Pareto frontier, with experimentally validated inverse design points achieving <i>R</i><sup>2</sup> values above 0.8. This work provides an efficient and reliable closed-loop solution for the rapid development of high-performance steels.</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

Accelerating Steel Design with Multi-objective Machine Learning

  • 期刊论文

  • 发表日期：

    2025年12月22日

查看全文
  • Jiazheng Zhao,
  • Shengen Zhang,
  • Jian Wang,
  • Jun Zhang,
  • Jun Li,
  • Jingge Dai,
  • Fenghua Luo

摘要

In order to meet the urgent demand for high-precision material performance data in advanced structural design, this paper has developed and validated a data-driven machine learning framework. This framework is based on Gaussian process regression, learning from 183 sets of deformation and rapid heat-treatment experimental data, covering a wide process space, and achieving multi-objective collaborative prediction of yield strength, tensile strength, and elongation of low-carbon, low-alloy steel. the impact of different iteration strategies on the generalization ability of the model has been analyzed, and it has been confirmed that, by constructing a more complete global model, combined iteration has better robustness than transfer learning in predicting new processes. Furthermore, customizing covariance functions for different physical properties of mechanical performance is key to achieving high-fidelity modeling. The framework achieves an R2 value greater than 0.5 for new process predictions in forward design. In inverse design, combined with the expected hypervolume optimization algorithm, it successfully expands the strength–ductility Pareto frontier, with experimentally validated inverse design points achieving R2 values above 0.8. This work provides an efficient and reliable closed-loop solution for the rapid development of high-performance steels.