Data-driven predicting of bond strength in corroded BFRP concrete structures

Abstract

The mechanical properties of basalt fiber-reinforced polymer (BFRP) concrete structures in corrosive environments are largely dependent on the bonding performance between the BFRP reinforcement and concrete. An accurate and convenient method for calculating the bonding strength is crucial for engineering design. However, experimental methods and existing empirical models are insufficient to encompass the intricate interdependencies among the bonding factors. This study proposed a machine learning (ML)-based bonding strength prediction method, which utilizes random forests (RF) and adaptive boosting (AdaBoost) algorithms to predict the interfacial bond strength of BFRP reinforcement in corroded concrete. The model was trained on a dataset of 355 samples, effectively quantifying the relationships between the BFRP reinforcement parameters, concrete parameters, corrosion factors, and bonding strength. The performance of the two algorithms was evaluated using R2, RMSE, and MAE indicators, and the prediction results were explained using the SHAP method. The results show that the predicted results of the machine learning model are highly consistent with the experimental results, with the AdaBoost model achieving an R2 of 0.925 on the test set, demonstrating high predictive performance. Among the various factors, the corrosion factor has the most significant impact on bonding strength, followed by concrete compressive strength and BFRP bar yield strength. Ultimately, the model was juxtaposed against traditional empirical formulas, affirming its efficacy and dependability. The research presents a novel approach to predicting bond strength in corroded BFRP bar-concrete systems.

Keywords

• Bond strength
• BFRP
• Data-driven
• Adaptive Boosting
• Corrosion
• SHAP method