Cost-efficient seismic retrofit strategy for AFRP-retrofitted RC piloti frames using ML-driven plastic hinge modeling

Abstract

This research introduces a fast-executing machine learning (ML) framework designed to estimate the seismic behavior of piloti-type reinforced concrete (RC) structures strengthened using aramid fiber-reinforced polymer (AFRP) jacketing. The training data were derived from a physics-informed approach based on a plastic hinge model (PHM), which was validated through both experimental testing and finite element (FE) analysis. The surrogate model predicts the inter-story drift ratio (IDR), which represents the seismic performance of retrofitted RC structures. The training data were generated by varying the confinement ratio (CR) and the retrofit height ratio (RH) across a wide range to capture diverse nonlinear structural responses. Three ML algorithms, Support Vector Machines (SVM), Least Squares Boosting (LSBoost), and Extreme Gradient Boosting (XGBoost), were trained and tested. Among them, XGBoost demonstrated the highest prediction accuracy and was selected as the final model. The trained model was then used to evaluate seismic performance and identify optimal retrofit configurations. Results indicate that the most effective retrofit schemes correspond to CR values between 0.75 and 1.25 and RH values between 0.5 and 0.625, where seismic drift is significantly reduced while maintaining cost efficiency. The proposed ML framework enables rapid assessment of retrofit strategies, offering a practical tool for performance-based seismic design of AFRP-retrofitted RC structures.