Scholarlink

<p>Alkali-activated materials (AAMs) have emerged as promising low-carbon alternatives to ordinary Portland cement for infrastructure exposed to aggressive environments. However, accurately predicting their durability under acid and sulfate attack remains challenging due to the complex interactions among mixture constituents and exposure conditions. This study developed and evaluated four machine-learning (ML) models, namely decision tree, random forest, AdaBoost, and gradient boosting, to predict the compressive strength loss of AAMs subjected to acid and sulfate exposure. A comprehensive database comprising key mixture parameters, including fly ash, hydrated lime, sodium silicate, sodium hydroxide, sand-to-binder ratio, water-to-binder ratio, and exposure duration, was utilized for model development and validation. Model performance was assessed using statistical metrics, including R², R, MAPE, MAE, MSE, and RMSE. All models demonstrated good predictive capability, achieving R² greater than 0.85. For sulfate attack, gradient boosting exhibited the highest prediction accuracy, whereas decision tree and AdaBoost showed superior performance under acid exposure. In addition, response surface methodology (RSM) was employed to quantify the influence of individual variables and their interactions on durability performance. The analysis identified fly ash content as the most influential parameter governing strength loss in sulfate attack, followed by hydrated lime content, and sand-to-binder ratio. The developed ML models provide an efficient and reliable tool for predicting durability-related strength degradation of AAMs, while the RSM analysis offers valuable insights for mix-design optimization. The findings contribute to the development of durable and sustainable AAM for applications in aggressive service environments.</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 and response surface modeling for predicting strength loss of alkali-activated materials exposed to aggressive environment

  • 期刊论文

  • 发表日期：

    2026年7月14日

查看全文
  • Munir Iqbal,
  • Sohaib Nazar,
  • Muhammad Hasham Kashif,
  • Muhammad Ashraf,
  • Haji Sami Ullah

摘要

Alkali-activated materials (AAMs) have emerged as promising low-carbon alternatives to ordinary Portland cement for infrastructure exposed to aggressive environments. However, accurately predicting their durability under acid and sulfate attack remains challenging due to the complex interactions among mixture constituents and exposure conditions. This study developed and evaluated four machine-learning (ML) models, namely decision tree, random forest, AdaBoost, and gradient boosting, to predict the compressive strength loss of AAMs subjected to acid and sulfate exposure. A comprehensive database comprising key mixture parameters, including fly ash, hydrated lime, sodium silicate, sodium hydroxide, sand-to-binder ratio, water-to-binder ratio, and exposure duration, was utilized for model development and validation. Model performance was assessed using statistical metrics, including R², R, MAPE, MAE, MSE, and RMSE. All models demonstrated good predictive capability, achieving R² greater than 0.85. For sulfate attack, gradient boosting exhibited the highest prediction accuracy, whereas decision tree and AdaBoost showed superior performance under acid exposure. In addition, response surface methodology (RSM) was employed to quantify the influence of individual variables and their interactions on durability performance. The analysis identified fly ash content as the most influential parameter governing strength loss in sulfate attack, followed by hydrated lime content, and sand-to-binder ratio. The developed ML models provide an efficient and reliable tool for predicting durability-related strength degradation of AAMs, while the RSM analysis offers valuable insights for mix-design optimization. The findings contribute to the development of durable and sustainable AAM for applications in aggressive service environments.