Scholarlink

<p>The electrical and thermal properties of cement-based composites are garnering increasing attention due to their vast potential for applications in self-sensing, self-heating, and efficient thermal energy management. Although finite element simulations and machine learning have advanced performance prediction, current studies fail to capture special distribution of fine aggregate and interphase connectivity between aggregate and cement paste in microscopic images, limiting high-precision predictions. This study generates 1,000 sets of representative volume element (RVE) microscopic structural images along with their corresponding physical parameters through finite element simulations. Seven machine learning models—Support Vector Regression (SVR), Gaussian Process Regression (GPR), Convolutional Neural Networks (CNN), Multilayer Perceptron (MLP), Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), and Transformer—are utilized to predict the electrical and thermal properties, and a comprehensive evaluation is conducted on the performance of each model in terms of prediction accuracy and applicability. The results show that the GPR model performs best in predicting electrical conductivity and thermal conductivity, with <i>R</i><sup>2</sup> values exceeding 0.9985, MAE values of 0.0002% and 0.62%, and RMSE values of 0.0003% and 0.79%, respectively. The CNN and MLP models follow, with <i>R</i><sup>2</sup> values exceeding 0.9948, MAE values for electrical and thermal conductivity predictions below 1.29%, and RMSE values below 1.69%. In comparison, the RNN, LSTM, and Transformer models exhibit larger prediction errors, with RNN performing the worst. In this case, <i>R</i><sup>2</sup> drops to 0.9610, MAE rises to 3.58%, and RMSE climbs to 4.49%. This study significantly enhances the prediction accuracy of the thermoelectric properties of cement-based composites while simultaneously reducing computational cost, thereby facilitating rapid material design and optimization.</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

Prediction of electrical and thermal properties of cement-based composites driven by microscopic structural data

  • 期刊论文

  • 发表日期：

    2026年1月21日

查看全文
  • Bo Liu,
  • Weisu Weng,
  • Song Zhang,
  • Huaibing Xu,
  • Zhihui Wu,
  • Xiaolong Zhou,
  • Chengyu Jia,
  • Yong Liu

摘要

The electrical and thermal properties of cement-based composites are garnering increasing attention due to their vast potential for applications in self-sensing, self-heating, and efficient thermal energy management. Although finite element simulations and machine learning have advanced performance prediction, current studies fail to capture special distribution of fine aggregate and interphase connectivity between aggregate and cement paste in microscopic images, limiting high-precision predictions. This study generates 1,000 sets of representative volume element (RVE) microscopic structural images along with their corresponding physical parameters through finite element simulations. Seven machine learning models—Support Vector Regression (SVR), Gaussian Process Regression (GPR), Convolutional Neural Networks (CNN), Multilayer Perceptron (MLP), Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM), and Transformer—are utilized to predict the electrical and thermal properties, and a comprehensive evaluation is conducted on the performance of each model in terms of prediction accuracy and applicability. The results show that the GPR model performs best in predicting electrical conductivity and thermal conductivity, with R2 values exceeding 0.9985, MAE values of 0.0002% and 0.62%, and RMSE values of 0.0003% and 0.79%, respectively. The CNN and MLP models follow, with R2 values exceeding 0.9948, MAE values for electrical and thermal conductivity predictions below 1.29%, and RMSE values below 1.69%. In comparison, the RNN, LSTM, and Transformer models exhibit larger prediction errors, with RNN performing the worst. In this case, R2 drops to 0.9610, MAE rises to 3.58%, and RMSE climbs to 4.49%. This study significantly enhances the prediction accuracy of the thermoelectric properties of cement-based composites while simultaneously reducing computational cost, thereby facilitating rapid material design and optimization.