Duration-dependent Fragility Functions for Seismic Damage and Loss Assessment of High-rise Buildings

Abstract

The availability of large repositories of recorded and synthetic ground motions coupled with structural response simulation data in recent years has led to the increased popularity of data-driven models for seismic damage and loss assessment of buildings. This paper explores the benefits of using bivariate fragility functions to estimate earthquake-induced damage and economic loss in high-rise buildings. The dataset used in this study contains 15,000 simulations of modern high-rise reinforced concrete shear wall buildings which are subjected to ground motion records at five different intensity levels ranging from 100-yr to 4975-yr return periods. The proposed functions are conditioned on average spectral accelerations and ground motion significant duration. The results indicate that bivariate fragility functions improve damage state prediction success by 15% relative to conventional univariate functions (standard of practice). To develop bivariate functions, nominal and ordinal probit regression models are fit to the dataset. Although both models improve the predictive performance considerably, ordinal functions can lead to a 10% reduction in misclassified collapse instances, i.e., the minority class. Univariate functions tend to overestimate seismic losses at lower ground motion intensity levels while underestimating them at higher intensities.