Collaborative filtering-based collapse fragility assessment

Abstract

This paper presents a computationally efficient collapse fragility assessment approach that leverages the collaborative filtering technique. The proposed method randomly selects and runs a subset of the nonlinear response history analyses (NRHAs) required for the truncated incremental dynamic analysis. The building responses in the unselected scenarios are then estimated using collaborative filtering, which requires significantly less time than the nonlinear structural analysis. The NRHA-based and collaborative filtering-generated building responses are then assembled to determine the empirical probability of collapse at each intensity, which is later fitted with a lognormal distribution to obtain the median collapse capacity and associated dispersion. The approach is applied to a three-story building to demonstrate its efficacy. It is shown to reduce the required number of NRHAs by 50% while providing a reliable collapse fragility estimation.