Hazard-based seismic fragility functions for steel moment-resisting frame buildings through data-driven damage state identification

Abstract

This paper employs supervised and unsupervised learning methods to present hazard-based seismic fragility functions for Steel Moment-Resisting Frame (SMRF) buildings. The database supporting this research comprises structural responses of over 12,000 time history analyses for 100 SMRF buildings lumped into three categories: low-, mid-, and high-rise. The ground motions have been selected to represent three hazard levels, resulting in Service Level Earthquake (SLE), Design Basis Earthquake (DBE), and Maximum Considered Earthquake (MCE). Considering the primary period of each building and the target response spectra, a set of ground motions is selected, and the peak story drift ratios are extracted. Subsequently, unsupervised clustering techniques are employed to identify drift thresholds that distinguish between different damage states across various hazard levels, thereby refining the fixed boundaries recommended in existing codes and guidelines. Supervised learning techniques, on the other hand, are employed to predict the maximum drift ratio using features from ground motions and structural periods. The resulting drift ratio serves as an Engineering Demand Parameter (EDP), which, along with the hazard-informed drift threshold, is used to generate a machine learning-based fragility function. The proposed approach enables damage state identification of SMRF buildings under a specific ground motion, using only structural periods and signal features, without requiring detailed structural response data. The results of this study provide a set of site-specific hazard-based fragility curves, supporting seismic risk and loss assessment across different earthquake intensities.