Scholarlink

<p>Corrosion-induced bond strength degradation between reinforcing bars and concrete significantly compromises the durability and structural performance of reinforced concrete (RC) systems. Accurate prediction of bond behavior under corrosion is critical for assessing residual capacity and informing rehabilitation strategies. This study proposes a data-driven framework for predicting both Ultimate Bond Strength (UBS) and Relative Bond Strength (RBS) of corroded reinforcement using a multi-output machine learning (ML) approach. A comprehensive dataset was compiled from experimental studies, and six ML algorithms, Multi-Layer Perceptron (MLP), Support Vector Regression (SVR), Decision Tree (DT), Random Forest (RF), Gradient Boosting (GBoost), and Extreme Gradient Boosting (XGBoost), were trained to forecast UBS and RBS simultaneously. Among them, XGBoost exhibited the highest predictive performance, achieving <i>R</i><sup>2</sup> values of 0.977 for UBS and 0.966 for RBS, with corresponding mean absolute percentage errors <i>(MAPE</i>) of 9.6% and 7.0%, respectively. Feature importance was evaluated using SHapley Additive exPlanations (SHAP), which revealed that Corrosion Level, Compressive Strength, and Yield Strength were the most influential factors for both targets. The results underscore the potential of explainable ML tools as efficient alternatives to traditional laboratory testing for evaluating bond degradation in corrosion-affected RC structures.</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

ML modeling of ultimate and relative bond strength for corroded rebars based on concrete and steel properties

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年7月23日

查看全文
  • Alireza Hosseinzadeh Kashani,
  • Mansour Ghalehnovi,
  • Hossein Etemadfard

摘要

Corrosion-induced bond strength degradation between reinforcing bars and concrete significantly compromises the durability and structural performance of reinforced concrete (RC) systems. Accurate prediction of bond behavior under corrosion is critical for assessing residual capacity and informing rehabilitation strategies. This study proposes a data-driven framework for predicting both Ultimate Bond Strength (UBS) and Relative Bond Strength (RBS) of corroded reinforcement using a multi-output machine learning (ML) approach. A comprehensive dataset was compiled from experimental studies, and six ML algorithms, Multi-Layer Perceptron (MLP), Support Vector Regression (SVR), Decision Tree (DT), Random Forest (RF), Gradient Boosting (GBoost), and Extreme Gradient Boosting (XGBoost), were trained to forecast UBS and RBS simultaneously. Among them, XGBoost exhibited the highest predictive performance, achieving R2 values of 0.977 for UBS and 0.966 for RBS, with corresponding mean absolute percentage errors (MAPE) of 9.6% and 7.0%, respectively. Feature importance was evaluated using SHapley Additive exPlanations (SHAP), which revealed that Corrosion Level, Compressive Strength, and Yield Strength were the most influential factors for both targets. The results underscore the potential of explainable ML tools as efficient alternatives to traditional laboratory testing for evaluating bond degradation in corrosion-affected RC structures.