Scholarlink

<p>Accurate prediction of the shear strength of steel fiber-reinforced concrete (SFRC) beams is critical for structural design. This study develops a robust machine learning (ML) framework to predict the shear strength of SFRC beams without web reinforcement. A database of&#xa0;450 experimental results&#xa0;was used, with eight key input parameters: beam width (b), effective depth (d), compressive strength (f’c), longitudinal reinforcement ratio (ρ), shear span-to-depth ratio (av/d), fiber factor (F), yield strength (fy), and fiber volume (Vf). A comparative analysis of eight ML algorithms was conducted using a&#xa0;grouped cross-validation strategy&#xa0;to prevent data leakage and ensure generalizability. The Extreme Gradient Boosting (XGBoost) model demonstrated superior performance on the&#xa0;locked test set, achieving an R² of 0.973, a root mean square error (RMSE) of&#xa0;25.74 KN, and a mean absolute error (MAE) of&#xa0;17.18 KN. This represents a significant improvement over predictions from empirical design codes, such as ACI 544.4R and Eurocode 2. An interpretability analysis using SHAP values confirmed the model’s alignment with engineering principles, identifying beam&#xa0;effective depth (d)&#xa0;as the most influential factor. The model was deployed as a user-friendly web application, providing engineers with a practical tool. This work highlights the potential of transparent, rigorously validated ML models to advance structural engineering practice.</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

Advancing shear behavior prediction in SFRC beams: a comparative study of machine learning techniques

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月24日

查看全文
  • Esayies A. Wassie,
  • Leule M. Hailemariam,
  • Habtamu A. Tadesse,
  • Temesgen A. Kassie

摘要

Accurate prediction of the shear strength of steel fiber-reinforced concrete (SFRC) beams is critical for structural design. This study develops a robust machine learning (ML) framework to predict the shear strength of SFRC beams without web reinforcement. A database of 450 experimental results was used, with eight key input parameters: beam width (b), effective depth (d), compressive strength (f’c), longitudinal reinforcement ratio (ρ), shear span-to-depth ratio (av/d), fiber factor (F), yield strength (fy), and fiber volume (Vf). A comparative analysis of eight ML algorithms was conducted using a grouped cross-validation strategy to prevent data leakage and ensure generalizability. The Extreme Gradient Boosting (XGBoost) model demonstrated superior performance on the locked test set, achieving an R² of 0.973, a root mean square error (RMSE) of 25.74 KN, and a mean absolute error (MAE) of 17.18 KN. This represents a significant improvement over predictions from empirical design codes, such as ACI 544.4R and Eurocode 2. An interpretability analysis using SHAP values confirmed the model’s alignment with engineering principles, identifying beam effective depth (d) as the most influential factor. The model was deployed as a user-friendly web application, providing engineers with a practical tool. This work highlights the potential of transparent, rigorously validated ML models to advance structural engineering practice.