Predicting Bond Strength between FRP Plates and Concrete Sub-strate: Applications of GMDH and MNLR Approaches

Abstract

Debonding of the fiber-reinforced polymer (FRP) reinforcement due to shear stresses is a very significant issue in design of concrete structures. Several experimental and theoretical investigations have been carried out to produce a relationship between the shear bond strength and the governing variables. However, existing empirical models do not provide an accurate prediction due to the complexity of the debonding process. In the present study, group method of data handling (GMDH) network as a novel machine learning approach was employed to predict the externally bond strength between FRP composites and concrete structures. The GMDH model was developed based on a reliable database including 342 experimental tests obtained from literature. The GMDH results were compared to the most common existing equations and also to the regression approaches developed in this study through statistical error parameters. Furthermore, some correction factors for four well-known equations were suggested based on regression approaches to improve their accuracy. Results indicated that the developed GMDH model outperformed the existing equations and also the developed regression-based equations in terms of both accuracy and safety aspects. Finally, parametric and sensitivity analyses were performed for further verification of the developed GMDH model in capturing the underlying physical behaviors of bond strength.