Scholarlink

<p>Concrete compressive strength (CS) is a critical parameter in the design and performance assessment of concrete structures. Accurate CS prediction reduces design costs, time, and material waste during mixture testing. In addressing these challenges, Deep Learning (DL) methodologies have proven effective in structural engineering applications, particularly for CS prediction. This study employed an interpretable Deep-Neural-Network (DNN) model to advance CS prediction, leveraging a dataset of 1030 experimental CS values obtained from previous research. Bayesian Optimization was utilized to optimize the model’s performance and tune its hyperparameters. A comprehensive evaluation of the model’s performance was conducted using both visual and quantitative methods. To further interpret the model, Shapley-Additive-exPlanations (SHAP) and Partial-Dependence-Plot (PDP) analyses were applied to examine the contribution of each input variable to the CS predictions. Results revealed that the optimal architecture of the DNN model, comprising an 8-1936-1680-16-16-16-1 configuration, achieved a high determination-coefficient (R²) of 0.964 during training and 0.946 during testing, outpacing other models from similar research. SHAP and PDP results highlighted the significant influence of curing age on prediction accuracy, with cement content ranking as the second most impactful factor. To support real-world civil engineering design and decision-making, a user-friendly Graphical User Interface (GUI) was developed, enabling practitioners to quickly and efficiently estimate CS.</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

Interpretable Deep Neural Network Deployment for Concrete Compressive Strength Prediction

  • 期刊论文

  • 发表日期：

    2026年1月27日

查看全文
  • Abdelrahman Kamal Hamed,
  • Mostafa M. Alsaadawi,
  • Mohamed Kamel Elshaarawy

摘要

Concrete compressive strength (CS) is a critical parameter in the design and performance assessment of concrete structures. Accurate CS prediction reduces design costs, time, and material waste during mixture testing. In addressing these challenges, Deep Learning (DL) methodologies have proven effective in structural engineering applications, particularly for CS prediction. This study employed an interpretable Deep-Neural-Network (DNN) model to advance CS prediction, leveraging a dataset of 1030 experimental CS values obtained from previous research. Bayesian Optimization was utilized to optimize the model’s performance and tune its hyperparameters. A comprehensive evaluation of the model’s performance was conducted using both visual and quantitative methods. To further interpret the model, Shapley-Additive-exPlanations (SHAP) and Partial-Dependence-Plot (PDP) analyses were applied to examine the contribution of each input variable to the CS predictions. Results revealed that the optimal architecture of the DNN model, comprising an 8-1936-1680-16-16-16-1 configuration, achieved a high determination-coefficient (R²) of 0.964 during training and 0.946 during testing, outpacing other models from similar research. SHAP and PDP results highlighted the significant influence of curing age on prediction accuracy, with cement content ranking as the second most impactful factor. To support real-world civil engineering design and decision-making, a user-friendly Graphical User Interface (GUI) was developed, enabling practitioners to quickly and efficiently estimate CS.