Scholarlink

<p>Machine learning is becoming increasingly popular for designing materials. However, high-quality data only available for specific materials. Therefore, it is a hot research topic to use machine learning for material design in the context of a small amount of insufficient data. In this work, a two-way feedback network structure was designed. Two different machine learning models (C2T and T2C) were trained to predict material composition directly from material property values using a small amount of experimental data. The C2T model could predict the properties of a material using the composition, and the T2C model could predict the composition of a material using the properties. The prediction accuracy of both C2T and T2C models was &gt; 90%. Finally, the material compositions and properties designed by the C2T and T2C models were experimentally verified. The results showed that the experimental and predicted values of the compositions and properties of the material were consistent, demonstrating the feasibility of the inverse design of this research method in the context of small data.</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

Machine Learning Inverse Prediction of Cu-Zn-Al Alloy’s Composition and Target Properties Using a Two-Way Feedback Structure on a Small Data Set

  • 期刊论文

  • 发表日期：

    2025年5月29日

查看全文
  • Shuang Zhou,
  • Qian Lei,
  • Jie Yin,
  • Xiang Yan,
  • Xiukuang Zhang,
  • Wen Liu

摘要

Machine learning is becoming increasingly popular for designing materials. However, high-quality data only available for specific materials. Therefore, it is a hot research topic to use machine learning for material design in the context of a small amount of insufficient data. In this work, a two-way feedback network structure was designed. Two different machine learning models (C2T and T2C) were trained to predict material composition directly from material property values using a small amount of experimental data. The C2T model could predict the properties of a material using the composition, and the T2C model could predict the composition of a material using the properties. The prediction accuracy of both C2T and T2C models was > 90%. Finally, the material compositions and properties designed by the C2T and T2C models were experimentally verified. The results showed that the experimental and predicted values of the compositions and properties of the material were consistent, demonstrating the feasibility of the inverse design of this research method in the context of small data.