Scholarlink

<p>For companies that produce aluminium components, creating aluminium alloys (AAs) with desirable properties is a major task. The traditional method of making and assessing alloys to ascertain their mechanical properties is costly and time-consuming. In this study, the mechanical properties (elongation E, tensile strength Uts, and yield strength Yts) of aluminium alloys (AAs) are predicted using a hybrid machine learning framework. For validation on datasets containing alloy composition and processing methods, the proposed framework consists of Artificial Neural Networks (ANN) and Multiple Linear Regression (MLR). While MLR can capture linear dependencies, ANN greatly lowers Mean Squared Error (MSE) and enhances R-squared (R²) values to more precisely estimate interaction effects. Loss curves from residual analysis show that the suggested framework performs better over time and in a wider range of situations. The results demonstrate the promising potential of the suggested framework in materials science as a scalable and effective substitute for conventional experimental techniques in alloy property prediction.</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

MLR-ANN: hybrid framework for data driven prediction of properties of aluminium alloys

  • 期刊论文

  • 发表日期：

    2025年11月11日

查看全文
  • R. Shoba,
  • Ajeet Bhagwat Bhane,
  • U. Srilakshmi,
  • Gulbir Singh,
  • Ayush Gandhi,
  • Rajeev Sharma,
  • Y. J. Nagendra Kumar

摘要

For companies that produce aluminium components, creating aluminium alloys (AAs) with desirable properties is a major task. The traditional method of making and assessing alloys to ascertain their mechanical properties is costly and time-consuming. In this study, the mechanical properties (elongation E, tensile strength Uts, and yield strength Yts) of aluminium alloys (AAs) are predicted using a hybrid machine learning framework. For validation on datasets containing alloy composition and processing methods, the proposed framework consists of Artificial Neural Networks (ANN) and Multiple Linear Regression (MLR). While MLR can capture linear dependencies, ANN greatly lowers Mean Squared Error (MSE) and enhances R-squared (R²) values to more precisely estimate interaction effects. Loss curves from residual analysis show that the suggested framework performs better over time and in a wider range of situations. The results demonstrate the promising potential of the suggested framework in materials science as a scalable and effective substitute for conventional experimental techniques in alloy property prediction.