Scholarlink

<p>This study investigates the mechanical performance of self-compacting concrete (SCC) incorporating Linz–Donawitz Slag (LDS) as a partial cement replacement, with an emphasis on developing an accurate predictive framework for compressive strength (CS). A total of 147 experimental data points were generated from seven SCC mix designs with varying LDS contents. To model the complex nonlinear relationships between input variables and CS test values, several machine learning (ML) models, including Random Forest (RF), Support Vector Machine (SVM), Artificial Neural Network (ANN), Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and a hybrid CNN–LSTM model were developed to predict CS of LDS based-SCC. Among these, the hybrid CNN–LSTM model demonstrated superior predictive performance, achieving a coefficient of determination (R<sup>2</sup>) of approximately 0.931 and lower error metrics (RMSE = 0.059, MAE = 0.045) as compared to other models. To improve interpretability, Shapley Additive exPlanations (SHAP) and Partial Dependence Plot (PDP) analyses were performed to identify the governing parameters and their interaction-driven effects on CS development. SHAP analysis revealed that cement content and LDS were the most influential variable for predicting the CS of SCC incorporating LDS, while water content was identified as the least significant factor. Partial Dependence Plot (PDP) analysis was employed to assess the impact of each input variable on CS. Consequently, the required amount of each input variable can be determined to meet a specific design strength. Therefore, the CNN-LSTM approach proved to be a reliable and efficient ML model, offering quick and accurate predictions for sustainable SCC performance. However, the present research is limited by the relatively small experimental dataset and the absence of external validation under field conditions. Future research should focus on incorporating larger and more diverse datasets, real-world validation, and multi-objective assessment including durability and long-term performance of LDS-incorporated SCC systems.</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

An experimental and computational intelligence simulation for predicting the strength accuracy of self-compacting concrete using Linz–Donawitz Slag as industrial waste derivatives

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年5月23日

查看全文
  • M. K. Diptikanta Rout,
  • Bibhu Prasad Mishra,
  • Abhilash Gogineni,
  • Regasa Yadeta Sembeta

摘要

This study investigates the mechanical performance of self-compacting concrete (SCC) incorporating Linz–Donawitz Slag (LDS) as a partial cement replacement, with an emphasis on developing an accurate predictive framework for compressive strength (CS). A total of 147 experimental data points were generated from seven SCC mix designs with varying LDS contents. To model the complex nonlinear relationships between input variables and CS test values, several machine learning (ML) models, including Random Forest (RF), Support Vector Machine (SVM), Artificial Neural Network (ANN), Convolutional Neural Network (CNN), Long Short-Term Memory (LSTM), and a hybrid CNN–LSTM model were developed to predict CS of LDS based-SCC. Among these, the hybrid CNN–LSTM model demonstrated superior predictive performance, achieving a coefficient of determination (R2) of approximately 0.931 and lower error metrics (RMSE = 0.059, MAE = 0.045) as compared to other models. To improve interpretability, Shapley Additive exPlanations (SHAP) and Partial Dependence Plot (PDP) analyses were performed to identify the governing parameters and their interaction-driven effects on CS development. SHAP analysis revealed that cement content and LDS were the most influential variable for predicting the CS of SCC incorporating LDS, while water content was identified as the least significant factor. Partial Dependence Plot (PDP) analysis was employed to assess the impact of each input variable on CS. Consequently, the required amount of each input variable can be determined to meet a specific design strength. Therefore, the CNN-LSTM approach proved to be a reliable and efficient ML model, offering quick and accurate predictions for sustainable SCC performance. However, the present research is limited by the relatively small experimental dataset and the absence of external validation under field conditions. Future research should focus on incorporating larger and more diverse datasets, real-world validation, and multi-objective assessment including durability and long-term performance of LDS-incorporated SCC systems.