Prediction of FRP RC columns using AI and machine learning

In recent years, fiber-reinforced polymers (FRP) have been gaining attention as a replacement for steel bars in concrete columns. Like steel reinforcement, FRP contributes to the axial load-carrying capacity. Multiple equations were proposed to understand the load-carrying capacity of FRP-reinforced concrete columns. However, existing design codes limit the use of FRP bars in columns since comprehensive predictive models are lacking. To address this gap, artificial intelligence (AI) and machine learning (ML) methods provide a powerful alternative by capturing nonlinear relationships between key structural parameters and column capacity. The study aims to check the reliability of the most well-known physical models that predict the effect and contribution of FRP bars to the overall capacity of columns at the ultimate limit state. The key parameters included in the data to be considered in this study are the column’s cross-sectional area, the column length, the compressive strength of the concrete, the elastic modulus of GFRP bars, and both longitudinal and transverse GFRP reinforcement ratios. A comprehensive dataset of tested FRP-RC was collected from existing literature to train, validate, and test machine learning models. Ann and Extreme Gradient Boosting ML algorithms are explored to determine the most accurate predictive model. Feature Sensitivity analysis is implemented using Shapley Additive explanations (SHAP) method to understand the machine learning models in the study, with focus on which variables influence the ML models the most or the least. The outcome of this research will contribute to ongoing discussions on using FRP reinforcement in compression members, where existing gaps in design methodologies will be addressed and potentially influencing future structural codes.