Data-driven machine learning methodology for designing slender FRP-RC columns

Abstract

Current design guidelines for reinforced concrete (RC) with fiber-reinforced polymers (FRP) bars lack provisions for designing slender columns. Limited attempts were made to modify the moment magnification procedure to accommodate FRP-RC columns. This study proposes a novel approach for designing FRP-RC slender columns based on suggesting a simplified slenderness reduction factor to account for the slenderness (ϕslender). The novel reduction factor has been developed using data-driven machine learning, where the available experimental database of short and slender FRP-RC columns has been employed to train a robust generalized artificial neural network (ANN) model. The ANN model is then utilized to generate reduction-factor curves, facilitating a straightforward approach to designing slender FRP-RC columns. For design practice purposes, genetic expression programming (GEP) was used to generate a mathematical equation for calculating ϕslender. The proposed ϕslender is a function of concrete compressive strength, reinforcement ratio, eccentricity, and column slenderness ratio. Statistical correlation analysis indicated that implementing the ϕslender eliminates the axial capacity correlation to the slenderness ratio, indicating a very good representation of the slenderness effect on the FRP-RC columns. The proposed approach revealed higher accuracy and consistent conservatism compared to the moment magnification procedure in predicting the strength of the slender FRP-RC columns.