Machine learning-based method to maximize allowable axial load for green remodeling or vertical extension of existing reinforced concrete moment frame buildings

초록

Green remodeling or vertical extension of existing reinforced concrete (RC) building structures is one of green building strategies to reduce the carbon emissions rather than constructing new buildings. The amplified gravity loads from the remodeling measures can lead to the seismic vulnerability of existing building frames. This paper proposes a machine-learning (ML) based method that rapidly estimates allowable axial loads on the building frames using brief information and establishes stiffness- and ductility-based retrofit schemes to maximize the allowable loads. To accomplish this goal, a sequential approach with three learning models (identification model of failure modes, and prediction models of seismic performance for as-built and massamplified conditions) was adopted as follows: (1) predicting seismic performance for as-built conditions, (2) estimating additional axial loading by varying the structural masses within the target performance, and (3) maximizing additional axial loading by varying the retrofit-related variables. Through the ML-based method, the retrofit schemes maximizing the allowable axial loads on the existing RC frames were derived for various combinations of stiffness and confinement ranges. The increase in the allowable axial loads by increasing column stiffness was limited within the low level of confinement, while the allowable axial load was continuously increased by additional stiffness withing the high level of confinement. Therefore, combined retrofit schemes ensuring adequate ductility capacities were needed to maximize the axial-load carrying capacity on the existing RC building frames.

키워드