Scholarlink

<p>The complex and heterogeneous microstructure of aluminum foam poses challenges for accurate mechanical characterization using traditional methods such as finite element analysis and analytical models. In this study, machine learning techniques were employed to develop a data-driven approach for predicting the compressive stress of aluminum foam based on key structural parameters. Experimental data involving porosity, mean pore size, and strain were used as input features, while compressive stress served as the output variable. Five regression models—Random Forest Regression (RFR), Bayesian Regression, Support Vector Machine Regression, Artificial Neural Network Regression, and Recurrent Neural Network Regression—were trained and compared. Among them, the RFR model exhibited the best performance, achieving an <i>R</i><sup>2</sup> of 0.9576, MAE of 0.8810&#xa0;MPa, and RMSE of 1.9615&#xa0;MPa. The model’s predictions showed excellent agreement with experimental results, with an average error of ± 0.0664&#xa0;MPa. This work demonstrates a robust and interpretable data-driven framework for aluminum foam characterization, significantly reducing the reliance on extensive experimental testing.</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

Compression Stress Prediction of Aluminum Foam Based on Machine Learning

  • 期刊论文

  • 发表日期：

    2025年12月15日

查看全文
  • Ruotong Wang,
  • Wenhao Da,
  • Yanli Wang,
  • Lucai Wang,
  • Xiaohong You,
  • Wenzhan Huang,
  • Fang Wang,
  • Yuancong Mi

摘要

The complex and heterogeneous microstructure of aluminum foam poses challenges for accurate mechanical characterization using traditional methods such as finite element analysis and analytical models. In this study, machine learning techniques were employed to develop a data-driven approach for predicting the compressive stress of aluminum foam based on key structural parameters. Experimental data involving porosity, mean pore size, and strain were used as input features, while compressive stress served as the output variable. Five regression models—Random Forest Regression (RFR), Bayesian Regression, Support Vector Machine Regression, Artificial Neural Network Regression, and Recurrent Neural Network Regression—were trained and compared. Among them, the RFR model exhibited the best performance, achieving an R2 of 0.9576, MAE of 0.8810 MPa, and RMSE of 1.9615 MPa. The model’s predictions showed excellent agreement with experimental results, with an average error of ± 0.0664 MPa. This work demonstrates a robust and interpretable data-driven framework for aluminum foam characterization, significantly reducing the reliance on extensive experimental testing.