Scholarlink

<p>Alkali-activated concrete (AAC) is a sustainable alternative to Portland cement, offering superior thermal resistance. However, predicting the residual compressive strength of AAC after high-temperature exposure remains a complex challenge. This study addresses this gap by using machine learning (ML) to model AAC performance. A comprehensive database of 371 samples was compiled from the experimental analysis, detailing mix proportions and residual compressive strength after exposure to temperatures from 30&#xa0;°C to 1000&#xa0;°C. Five supervised machine learning models (Decision Tree, Bagging Regressor, AdaBoost, Random Forest, and XGBoost) were developed and evaluated. The XG Boost (XGB) model demonstrated the highest predictive accuracy, achieving a coefficient of determination (R²) of 0.95 and the lowest root mean square error (RMSE) of 2.50. A feature correlation analysis identified curing temperature, curing duration, and alkali activator concentration as the most influential parameters. This study provides a validated ML model for accurately predicting the residual strength of AAC, offering a reliable tool for designing fire-resistant, sustainable concrete mixtures.</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

Assessment of thermally induced strength loss in alkali-activated concrete through ensemble regression models

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年3月21日

查看全文
  • Yellanki Deepti,
  • Sanjay Kumar,
  • Atrayee Bandyopadhyay,
  • Pramod Kumar,
  • Regasa Yadeta Sembeta

摘要

Alkali-activated concrete (AAC) is a sustainable alternative to Portland cement, offering superior thermal resistance. However, predicting the residual compressive strength of AAC after high-temperature exposure remains a complex challenge. This study addresses this gap by using machine learning (ML) to model AAC performance. A comprehensive database of 371 samples was compiled from the experimental analysis, detailing mix proportions and residual compressive strength after exposure to temperatures from 30 °C to 1000 °C. Five supervised machine learning models (Decision Tree, Bagging Regressor, AdaBoost, Random Forest, and XGBoost) were developed and evaluated. The XG Boost (XGB) model demonstrated the highest predictive accuracy, achieving a coefficient of determination (R²) of 0.95 and the lowest root mean square error (RMSE) of 2.50. A feature correlation analysis identified curing temperature, curing duration, and alkali activator concentration as the most influential parameters. This study provides a validated ML model for accurately predicting the residual strength of AAC, offering a reliable tool for designing fire-resistant, sustainable concrete mixtures.