Non-parametric fragility curves for bridges using recorded ground motions

Seismic fragility curves are commonly used to assess the vulnerability of structures to earthquakes by representing the probability of structural failure as a function of an earthquake intensity measure. The classical approach for computing fragility curves assumes the curves to have a lognormal shape. This approach is therefore parametric. Since fragility curves play an important role in the framework of performance-based earthquake engineering, it is of utmost importance either to validate the accuracy of the lognormal assumption or to propose an assumption-free approach to compute these curves. In a recent work, the authors validated the lognormal assumption for a linear steel frame. However, in a more realistic case study of this steel frame with non-linear material behavior, the lognormal assumption showed insufficient accuracy. In this paper, we consider a different type of structure, i.e. a typical reinforced-concrete bridge column subject to recorded ground motions. We compute fragility curves for the column, first with the classical parametric approach and subsequently with a non-parametric method based on kernel density estimation. Two different intensity measures are considered, namely the peak ground acceleration and the pseudo-spectral acceleration. The results show the limitations of the classical lognormal approach for this type of structure and prove kernel density estimation to be a promising tool for establishing seismic fragility curves.