Strength models of near-surface mounted (NSM) fibre-reinforced polymer (FRP) shear-strengthened RC beams based on machine learning approaches

Abstract

The shear strengthening of reinforced concrete (RC) beams using near-surface mounted (NSM) fibre-reinforced polymer (FRP) bars/strips has gained substantial research attention worldwide. However, owing to the complex failure mechanisms and many influencing parameters, the shear capacities of NSM FRP shear-strengthened beams are difficult to predict. Accordingly, this study adopted machine learning approaches to predict the shear capacity of strengthened beams. An experimental database was constructed comprising 130 rectangular/T-shaped beams and their 15 parameters, collected from the existing literature. Subsequently, a genetic-algorithm-improved back propagation neural network (GA-BPNN) trained with a Bayesian regularisation (BR) algorithm was employed, which was capable of giving accurate predictions on shear capacities of strengthened beams and own good generalisation ability. Furthermore, the GA-BPNN was used for parametric studies to investigate the parameter effects on the contributions of concrete, steel stirrups, and NSM FRP to the shear capacity. Finally, with reference to the GA-BPNN parametric analyses and existing models, a design-oriented strength model for calculating the shear capacities of NSM FRP shear-strengthened beams was proposed and optimised using the genetic algorithm. A comparison with existing models proved the higher prediction accuracy of the proposed strength model.