Scholarlink

<p>This study systematically examines how microalloying with La influences the hot deformation behavior and microstructure evolution of Cu-Ti alloys. Isothermal compression tests on Cu-3Ti and Cu-3Ti-0.1La alloys were performed using a Gleeble simulator at 500-800&#xa0;°C and strain rates of 0.01-10&#xa0;s<sup>−1</sup>. True stress–strain curves were obtained and used to establish a Zener–Hollomon constitutive equation. Results indicate that La addition refines grains, promotes Ti-rich phase precipitation, lowers hot deformation activation energy from 574.1 to 551.2&#xa0;kJ/mol, and improves hot workability and deformation stability. A backpropagation neural network (BPNN) model was also developed to predict flow stress. Comparison between the Levenberg–Marquardt and Bayesian regularization (BR) algorithms shows that a single-hidden-layer BPNN using BR achieves the highest accuracy (R = 0.99985), avoids overfitting, and offers superior generalization and reliability. This work provides theoretical support for hot processing design of high-performance Cu-Ti alloys and validates the effectiveness of machine learning in materials constitutive modeling.</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

Study on the Hot Deformation Behavior of Cu-Ti Alloy Based on Machine Learning Algorithms and Microalloying

  • 期刊论文

  • 发表日期：

    2026年4月27日

查看全文
  • Mengxiao Zhang,
  • Wei Guo,
  • Yang Gao,
  • Yan Jia,
  • DaYong Chen,
  • Huan Liu,
  • Hongwu Song,
  • Ming Cheng,
  • Zeming Ding

摘要

This study systematically examines how microalloying with La influences the hot deformation behavior and microstructure evolution of Cu-Ti alloys. Isothermal compression tests on Cu-3Ti and Cu-3Ti-0.1La alloys were performed using a Gleeble simulator at 500-800 °C and strain rates of 0.01-10 s−1. True stress–strain curves were obtained and used to establish a Zener–Hollomon constitutive equation. Results indicate that La addition refines grains, promotes Ti-rich phase precipitation, lowers hot deformation activation energy from 574.1 to 551.2 kJ/mol, and improves hot workability and deformation stability. A backpropagation neural network (BPNN) model was also developed to predict flow stress. Comparison between the Levenberg–Marquardt and Bayesian regularization (BR) algorithms shows that a single-hidden-layer BPNN using BR achieves the highest accuracy (R = 0.99985), avoids overfitting, and offers superior generalization and reliability. This work provides theoretical support for hot processing design of high-performance Cu-Ti alloys and validates the effectiveness of machine learning in materials constitutive modeling.