Hierarchical and mixed uncertainty quantification for simulation-based structural seismic fragility analysis

Abstract

Seismic fragility assessments (SFA) encompass significant uncertainties, including ground motions, materials, geometry, boundaries, and modeling, which impart inherent uncertainties to the fragility curve. Quantifying these uncertainties is essential for informed decision-making in earthquake disaster mitigation. This paper presents a hierarchical uncertainty quantification (UQ) framework for simulation-based SFA, which can address mixed aleatory and epistemic uncertainties. The framework comprises two levels: At Level 1, ground motion uncertainties are accounted for through record-to-record variation, while structural parameter uncertainties are captured using well-defined probability distributions. The quantification of uncertainty in seismic fragility is accomplished using mean and quantile fragility curves, representing the average outcome and the dispersion of structural fragility, respectively. Fuzziness is also introduced to enhance the objectivity of failure determination. Level 2 addresses uncertainties in the distribution hyper-parameters of structural variables through evidence theory, facilitating an understanding of their influence on the seismic fragility curve and quantifying uncertainty in mean and quantile fragility curves. This UQ framework builds upon the multivariate seismic fragility analysis method, integrating Kriging regression for seismic demand prediction and logistic regression for generating multivariate seismic fragility functions. The proposed framework is validated numerically on an existing three-span reinforced concrete continuous girder bridge, affirming its practical utility and reliability.