Non-Parametric Seismic Fragility Curves of SDOF Systems Based on a Clustering Process

Abstract

Most non-parametric methods to extract the fragility curves are based on simulated records which not only can impose additional computational efforts, but also may not represent stochastic nature of real ground motion records. Therefore, a new classification-based procedure is proposed in this study to extract the seismic fragility curves using real ground, motion records. Generally, providing an applicable method for record selection to achieve acceptable non-parametric fragility curves with the aim of minimizing the computational efforts can be the main novelty of this study. The proposed non-parametric method which uses a clustering process on the records based on their intensity measure (IM), can considerably decrease the number of required nonlinear analyses. A large group of real ground motion records concluding 12,580 earthquake records is selected as input dataset for extracting the fragility curves. Subsequently, a classification process based on two effective parameters of source to site distance and magnitude is proposed in the framework of the non-parametric method. Results show that the classification process can lead to more accurate curves rather than the one obtained using the whole input data where mean probability difference between the obtained curve and a benchmark fragility curve can be decreased between 2.05% and 4.38% by using a smaller set of records. As another result, the number of 8–12 clusters can lead to comparatively smooth and acceptable fragility curves for real ground motion records. Generally, a significant reduction (over than half) can be reached in the number of required analyses by selecting an appropriate set of records and a suitable number of clusters.