Machine learning driven bond performance prediction between FRP bars and coral aggregate concrete

• Published in: Construction & building materials Vol. 442; p. 137684
• Main Authors: Tang, Yixing, Zhou, Chang, Wang, Wenwei, Zhang, Wutong, Cheng, Yi
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 06.09.2024
• Subjects: Bond performance; Coral aggregate concrete (CAC); Fiber reinforced polymer (FRP); Machine learning; Model interpretation; SHAP algorithm; SHAP algorithm; Bond performance; Model interpretation; Fiber reinforced polymer (FRP); Coral aggregate concrete (CAC); Machine learning
• ISSN: 0950-0618
• DOI: 10.1016/j.conbuildmat.2024.137684

This study aims to apply machine learning (ML) methods to predict the failure mode and bond stress between FRP bars and coral aggregate concrete (CAC). The central pull test dataset of 221 FRP bars and CAC was synthesized for training and testing six supervised ML models, including Artificial Neural Network (ANN), Decision Tree, K-Nearest Neighbors (KNN), Random Forest, Support Vector Machine (SVM) and eXtreme Gradient Boosting Trees (XGBoost). The performance of ML algorithms in predicting failure modes was analyzed using confusion matrices, with the random forest model exhibiting the highest classification ability. Regarding the prediction of bond stress, ML algorithms demonstrated superior precision compared to three existing empirical methods, with XGBoost and KNN being identified as the optimal algorithms based on five regression performance metrics. Additionally, the application of the Shapley Additive Explanation (SHAP) method provided insights into the contribution of each input feature to the prediction process of ML models. Notably, variables including the rib height, diameter and bond length of FRP bars significantly influenced failure mode. In contrast, the compressive strength of CAC, the rib height, and the bond length of FRP bars, played crucial roles in predicting bond stress. These findings provide valuable guidance for enhancing the collaborative efficiency between FRP bars and CAC and optimizing structural design practices. •Six supervised machine learning algorithms are employed to predict failure mode and bond stress between FRP bars and CAC.•SHAP algorithm is applied to elucidate the prediction process and variable influence of Random Forest and XGBoost models.•The accuracies of machine learning models in predicting bond stress are contrasted with that of empirical methods.