Investigation of stiffened UHPC-filled steel tube columns: performance, modeling, and machine learning-based capacity prediction

Abstract This study presents an investigation into the axial compression behavior of Ultra-High-Performance Concrete-filled Steel Tube (UHPCFST) columns enhanced with internal stiffening methods, specifically longitudinal stiffeners and binding bars. The research begins with an experimental program focused on stub columns to evaluate the effects of these stiffening techniques on bond performance and ultimate axial capacity. Subsequently, a three-dimensional nonlinear Finite Element (FE) model was developed using ABAQUS and rigorously verified against the experimental results from this study and supplementary data from existing literature on slender columns. The validated FE model demonstrated high fidelity in predicting structural responses, including load-displacement curves and failure modes. Leveraging the verified numerical framework, an extensive parametric study was conducted to systematically vary key geometric and material parameters, thereby generating a robust dataset of 67 simulations. This enriched database, totaling 132 specimens when combined with experimental data, was utilized to train and test three distinct machine learning (ML) algorithms: Random Forest Regressor (RFR), Gradient Boosting Regressor (GBR), and Extreme Gradient Boosting (XGBR). The ML models were developed to accurately predict the ultimate axial capacity of stiffened CFST columns, effectively filling the gaps left by limited experimental data. The results indicate that the developed ML models, particularly GBR, offer exceptional predictive accuracy (R 2 = 0.995), providing a powerful and efficient tool for the design and analysis of advanced composite columns.