Ensemble learning based approach for FRP-concrete bond strength prediction

Nowadays, externally bonding fiber reinforced polymer (FRP) plates or sheets have become a major maintenance approach for aged reinforced concrete flexure structures. However, the capacity of strengthend structure cannot be precisely estimated as a result of the critical FRP-concrete interfacial (FCI) bond strength unpredictable. In order to solve this issue, many experimental studies have been carried out with corresponding emipirical models proposed. Due to limited experiment samples, these models were found more or less lacking the generalization ability. Under this circumstance, in this study, an ensemble learning algorithm “gradient boosted regression trees” (GBRT) was employed to develop a prediction model for FCI bond strength prediction based on a collected comprehensive database containing 520 tested samples. The model’s performance has been thoroughly compared with the representative empirical models and the common utilized machine learning algorithms. The rationality of this model has also been discussed through feature importance analysis. The results showed that the model in this study exhibits the highest accuracy and is proven to be feasible for predicting FCI bond strength in actual practice.