Development of machine learning based seismic retrofit scheme for AFRP retrofitted RC column

Abstract

A fiber reinforced polymer (FRP) column jacketing system can provide additional confining pressure on existing RC columns and enhance their lateral resisting capacities. This paper aims to develop machine learning-based fast running model predicting seismic performance of aramid FRP-retrofitted RC columns, and the fast-running model was utilized to establish optimum retrofit schemes with respect to confinement- and stiffness-related parameters. To develop a physics-based dataset, the loading and retrofit parameters were implemented to the finite element column models validated with the experimental study. The various machine-learning models were trained, validated and tested by using the dataset, and the best-fit model was selected based on each model’s performance. The best-fit model was utilized to predict the seismic performance varying the loading and output parameters, and the optimum retrofit scheme for confinement ratio, stiffness ratio and effective bond length was proposed. The optimum retrofit details need to be higher than the retrofitted length-to-total height of 0.20 and the confinement ratio of 0.15 regardless of the stiffness ratio.