Scholarlink

<p>High-performance concrete (HPC) is widely used in infrastructure development due to its superior mechanical strength and durability. However, its long-term performance can deteriorate under adverse environmental conditions, particularly from chloride exposure and temperature fluctuations. Traditional empirical models struggle to predict such degradation accurately, as they rely on linear assumptions and treat inputs independently. This study proposes a multivariate machine learning (ML) framework to forecast two key durability parameters—compressive strength and chloride ion penetration—using a combination of mix design, curing conditions, and environmental exposure data. Seven ML models were evaluated: Multiple Linear Regression (MLR), Artificial Neural Network (ANN), Decision Tree Regression (DTR), Random Forest Regression (RFR), Support Vector Regression (SVR), Extreme Gradient Boosting (XGBoost), and Long Short-Term Memory (LSTM). Performance was assessed using R², RMSE, MAE, MAPE, IOA, and a20 accuracy metrics. Among these, LSTM and XGBoost outperformed traditional models, achieving R² values of 0.965 and 0.942, respectively. Key predictors included water/cement ratio, silica fume content, and exposure type. The findings demonstrate the potential of time-aware deep learning models, particularly LSTM, in enabling predictive maintenance and service-life forecasting of HPC structures exposed to hostile 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

Multivariate forecasting of durability in high-performance concrete using machine learning: a comparative evaluation of time-aware and ensemble models under adverse environmental conditions

  • 期刊论文

  • 发表日期：

    2026年3月3日

查看全文
  • Vishal Noroji,
  • Mukesh Pandey,
  • Rakesh Gupta

摘要

High-performance concrete (HPC) is widely used in infrastructure development due to its superior mechanical strength and durability. However, its long-term performance can deteriorate under adverse environmental conditions, particularly from chloride exposure and temperature fluctuations. Traditional empirical models struggle to predict such degradation accurately, as they rely on linear assumptions and treat inputs independently. This study proposes a multivariate machine learning (ML) framework to forecast two key durability parameters—compressive strength and chloride ion penetration—using a combination of mix design, curing conditions, and environmental exposure data. Seven ML models were evaluated: Multiple Linear Regression (MLR), Artificial Neural Network (ANN), Decision Tree Regression (DTR), Random Forest Regression (RFR), Support Vector Regression (SVR), Extreme Gradient Boosting (XGBoost), and Long Short-Term Memory (LSTM). Performance was assessed using R², RMSE, MAE, MAPE, IOA, and a20 accuracy metrics. Among these, LSTM and XGBoost outperformed traditional models, achieving R² values of 0.965 and 0.942, respectively. Key predictors included water/cement ratio, silica fume content, and exposure type. The findings demonstrate the potential of time-aware deep learning models, particularly LSTM, in enabling predictive maintenance and service-life forecasting of HPC structures exposed to hostile environments.