Metamodeling Framework for Multivariate Probabilistic Prediction of Story-Drift Profile of Multistory Buildings

Abstract

This paper proposes a probabilistic metamodeling framework for predicting the joint probability distribution of the interstory drift ratio (IDR) profile of multistory buildings. IDR is an engineering demand parameter that correlates well with structural and nonstructural damages. However, the use of IDR for regional loss estimation has not been practiced due to the high computational cost of such estimations. The metamodel proposed herein efficiently estimates the IDR profile while accounting for the cross-correlations. This facilitates the use of component-level damage and consequence models for regional loss estimations and significantly refines the resulting risk estimates compared with the current state of practice in regional risk analysis, which relies on a single-degree-of-freedom representation of a building. The proposed framework employs multivariate Bayesian regression calibrated using a large data set produced using detailed nonlinear time-history analyses. To showcase the methodology, metamodels are developed for predicting the IDR profile of special steel moment frame buildings of various heights ranging from three to 12 stories. The resulting models are validated against the output of detailed multidegree-of-freedom models at different levels of ground motion intensity. Finally, the proposed metamodel is employed for estimating the seismic loss of a nine-story building, and the results are compared with the loss using two extremes of modeling in terms of refinement and computational effort: the FEMA P-58 approach and the Hazus approach. The results confirm that the proposed metamodel significantly refines the loss estimates compared with those of Hazus with a computational effort that is four orders of magnitude smaller than that required by the FEMA P-58 approach.