Empirical fragility curves of RC buildings in Chile using a Bayesian cumulative link model

Abstract

Chile is known as one of the most seismic countries in the world showing approximately one event above magnitude 8 every ten years. Unexpected damage occurred in reinforced concrete (RC) buildings during the 2010 Mw 8.8 Maule earthquake. Calculating the risk for RC buildings to exceed a given damage state during potential future seismic events is thus of paramount importance. It is assumed that the risk can be calculated from a combination of an exposure model, an intensity hazard distribution, and the vulnerability definition of RC buildings. Construction of fragility curves is part of the process for vulnerability definition. We propose here in an innovative methodology to build empirical fragility curves with the collected data of Chilean RC buildings. After the 2010 Maule earthquake, a database of the damage states of Chilean RC buildings was built by the assessment of structural engineers from companies or municipalities. Damage was classified in five states: no damage, slight, moderate, severe and complete damage states. The database is composed of observations made in six Chilean cities exposed to different seismic intensity. An interpolation of a 2010 shakemap provides the intensity measure occurring at the location of each building. Different generalized linear models were tested and it was shown that the cumulative logit link model provides the best fit of fragility curves. Cumulative link models for ordinal data ensure that fragility curves do not cross. The number of buildings floors and the year of construction are considered as covariates of our model. A Bayesian approach is adopted to obtain estimates and distribution of the parameters describing the statistical model. The resulting fragility curves for RC buildings are compared with results available in the literature. The provided fragility curves for Chilean RC buildings is the major contribution of this study. Finally, the usability of the curves is investigated from a risk perspective, and the need of more information to improve the fitting procedure is discussed.