Scholarlink

The bond development between glass fibre reinforced polymer (GFRP) and concrete is crucial to the design and functioning of RC structures. Therefore, it is essential to predict the bond strength between GFRP and concrete with the highest possible accuracy. The accuracy of bond codes and guidelines is restrained due to inadequate testing data used for bond prediction and simplifying assumptions. This study investigates the ability of data-driven machine learning (ML) models to predict the bond between GFRP bars and concrete. A dataset consisting of 331 experimental results was collected from previous studies. Three ML models, namely the ensemble methods of extreme gradient boosting (XGBoost) and adaptive boosting (AdaBoost), and the supervised learning method of K-nearest neighbors (KNN), were employed for predicting the bond between GFRP bars and concrete. The input features of the ML models are the bar embedment length, bar diameter, compressive strength of the concrete, concrete cover, and GFRP position in the cross section (top or bottom of the cross section), whereas the output feature is the bar stress. To ensure the reliability of the ML models, their prediction accuracy was compared with that of existing code equations. The results revealed that the prediction accuracy of the ML models surpassed that of the code equations. To examine the relative influence of input features on the bond predictions, a feature importance analysis was conducted.

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

Predictive Machine Learning Models for GFRP-Concrete Bond Strength

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Habibi Omid,
  • Abdelrahman Mahmoud,
  • Belal Abdelrahman,
  • Galal Khaled

摘要

The bond development between glass fibre reinforced polymer (GFRP) and concrete is crucial to the design and functioning of RC structures. Therefore, it is essential to predict the bond strength between GFRP and concrete with the highest possible accuracy. The accuracy of bond codes and guidelines is restrained due to inadequate testing data used for bond prediction and simplifying assumptions. This study investigates the ability of data-driven machine learning (ML) models to predict the bond between GFRP bars and concrete. A dataset consisting of 331 experimental results was collected from previous studies. Three ML models, namely the ensemble methods of extreme gradient boosting (XGBoost) and adaptive boosting (AdaBoost), and the supervised learning method of K-nearest neighbors (KNN), were employed for predicting the bond between GFRP bars and concrete. The input features of the ML models are the bar embedment length, bar diameter, compressive strength of the concrete, concrete cover, and GFRP position in the cross section (top or bottom of the cross section), whereas the output feature is the bar stress. To ensure the reliability of the ML models, their prediction accuracy was compared with that of existing code equations. The results revealed that the prediction accuracy of the ML models surpassed that of the code equations. To examine the relative influence of input features on the bond predictions, a feature importance analysis was conducted.