Machine learning for the prediction of the axial load-carrying capacity of FRP reinforced hollow concrete column

Abstract

Fiber reinforced polymer (FRP) has emerged as a significant advancement in construction, with design provisions outlined by codes such as GB/T 30022-2013, CSA S806-12 (R2017), and ACI 440:2015. While the use of FRP bars as alternatives to conventional reinforcement in columns has been extensively studied, their application in hollow concrete columns (HCCs) remains underexplored. This study investigates the behavior of FRP-reinforced HCCs using advanced machine learning (ML) models, focusing on the prediction of two critical outputs: first peak load (Y1) and failure load (Y2), based on eight input parameters. Models evaluated include extreme gradient boosting (XGB), light gradient boosting (LGB), and categorical gradient boosting (CGB). A rigorous comparative analysis demonstrated that all models achieved high predictive accuracy, with deviations within ±10% of actual results, validating their reliability. Among the models, CGB exhibited superior generalization and robustness, emerging as the most reliable predictor for FRP-reinforced HCC behavior. To enhance practicality, a user-friendly graphical user interface was developed to allow engineers to input design parameters and instantly obtain predictions for Y1 and Y2. This study not only advances understanding of FRP-reinforced HCCs but also bridges the gap between computational predictions and real-world applications, contributing a robust predictive tool to structural engineering design.