Ranking of epistemic uncertainties in scenario-based seismic risk evaluations

Abstract

In the scope of a scenario-based risk analysis, this study aims to quantify and rank various types of epistemic uncertainties that enter into the derivation of fragility functions for common buildings. Using a numerical model of a test structure (a reinforced concrete five-story building with infill panels on the first two floors), a first type of uncertainty is introduced, consisting of the mechanical properties of the materials (i.e.Young's modulus and compressive strength for concrete, andYoung's modulus and yield strength for steel). The area of longitudinal reinforcement is also modified in the model, to generate various damage mechanisms for the same structure, depending on which floor first experiences failure. Finally, another source of epistemic uncertainty is studied, by comparing different types of fragility models: fragility curves derived from dynamic analyses and fragility functions generated from a capacity spectrum approach (i.e. use of a set of natural response spectra to identify a series of performance points from the capacity curve). To this end, a ranking of the importance of different sources of uncertainty in the vulnerability analysis (i.e. mechanical properties, structural models and fragility models) is conducted by computing, for each uncertainty source, the Sobol' indices (i.e. the main effects and total effects of each source of uncertainty). This variancebased sensitivity technique presents the appealing features of both exploring the influence of input parameters over their whole range of variation and fully accounting for possible interactions between them. Nevertheless, addressing the issue of sensitivity to model uncertainty implies paying special attention to the appropriate treatment of different types of input parameters, i.e. continuous for mechanical properties or categorical in the case of fragility models. This is achieved by relying on recent advances in functional variance decomposition. For all these types of models, a uncertainty analysis in terms of the predicted number of damage buildings is carried out for a series of hypothetical scenarios. © 2013 Taylor & Francis Group, London.