Efficient seismic fragility functions through sequential selection

Abstract

Fragility functions enable the assessment of a structural system for a given hazard scenario. Specifically, the fragility function provides the probability of an undesirable structural state conditioned on the occurrence of a specific hazard level. Multiple sources of uncertainty are present when estimating fragility functions, e.g., record-to-record variation, uncertain material and geometric properties, model assumptions, and limited data to characterize the hazard. The objective of this study is to develop a methodology that will accelerate the process of fragility function estimation under limitations in computational resources and data. The approach used in the methodology is as follows. The stochastic map between hazard level and structural response is first constructed using Bayesian inference for a finite number of simulations. The Bayesian approach enables the quantification of the epistemic uncertainty due to a limited number of simulations. This epistemic uncertainty is exploited to sequentially select subsequent simulations that accelerate learning based on up to two different earthquake intensity measures, peak ground velocity and spectral velocity. The methodology is applied to a benchmark model of a twenty-story nonlinear building. Simulations are performed using a set of synthetic ground motions obtained from scenario earthquakes in California. Through this case study the methodology developed here is demonstrated. Additionally, the case study highlights the ability of the methodology to achieve lower levels of epistemic uncertainty than traditional techniques using the same number of simulations. This approach is expected to enable more efficient fragility function determination.