Durability assessment of FRP-to-concrete bonded connections under moisture condition using data-driven machine learning-based approaches

Abstract

The prediction of long-term performance is a crucial step in the effective structural design of adhesively bonded connections. A lack of knowledge on the durability of the FRP-to-concrete bonded connections compromises structural safety and requires employing higher safety factors in design. This study examines the capability of machine learning techniques in studying the long-term performance of FRP-to-concrete bonded connections under moisture conditions. A comprehensive database of 429 durability test results is built and used to develop various machine learning models. A Bayesian optimisation method is employed during the model development to tune the machine learning hyperparameters. Based on the detailed evaluation of prediction results, artificial neural networks (ANN) and ensemble models are proposed to study the bond strength and the failure mode of the connections. The ANN model is used to evaluate the impact of various moisture conditioning regimes on bond strength. The contribution of environmental factors in identifying the failure mode is studied by the ensemble method. Moreover, a practical equation is proposed by the M5P model tree method to predict the bond strength. The prediction results in this study indicate the capability of machine learning in predicting the durability of FRP-bonded concrete connections under moisture conditions.