Data-driven prediction of FRP strengthened reinforced concrete beam capacity based on interpretable ensemble learning algorithms

Abstract

Fiber-reinforced polymer (FRP) materials are one of the commonly used materials for strengthening aged reinforced concrete (RC) beams. However, it is still challenging to accurately predict the flexural capacity of an FRP-strengthened RC beam due to the intricate mechanism. To overcome the limitation of mechanical-based models, a comprehensive database of FRP-strengthened RC beam experiments was collected to develop data-driven prediction models. Four different ensemble learning (EL) algorithms, namely random forest, adaptive boosting, gradient boosting decision tree, and extreme gradient boosting were used to realize this model based on this database. To demonstrate their superiority, these models were compared with representative empirical models and the ones based on single machine learning (ML) algorithms. The performances of the EL-based models were significantly better than those of the empirical models and single ML-based models. Thus, the EL-based models proposed in this study demonstrate potential for use in engineering applications. In addition, the Shapley additive explanation (SHAP) was introduced to interpret the importance of input features in the prediction process from local and global perspectives. Finally, reliability analysis was performed to calibrate the reduction coefficient of bearing capacity.