The importance of hazard-consistency when estimating seismic residual drifts in steel moment frames

Abstract

Residual drifts are routinely used to assess the post-earthquake safety of buildings. Despite their importance, studies on the probabilistic assessment of residual drifts in multi-storey buildings with Steel Moment Resisting Frames (SMRF) are far less common than those dealing with their transient peak drift counterpart. More seriously, although residual drift prediction models have been developed with a particular broad ground-motion type or soil condition in mind, all models proposed to date remain oblivious to the central issue of hazard consistency. Hence, missing the causal connection between the seismic hazard level and the ground-motion suite used and potentially compromising the meaningfulness of their results. This oversight is expected to introduce a bias in the estimation of residual drifts but its magnitude has not yet been properly evaluated. In this paper, we evaluate and quantify the significance of these effects. To this end, we use the Conditional Scenario Spectra framework, which provides a set of realistic earthquake spectra with assigned rates of occurrences based on their spectral shape and intensity, thus preserving the critical relationship of hazard consistency. Nonlinear Response History Analyses (NRHA) of 24 deteriorating SMRFs under 816 ground-motion records are performed to construct a database of residual drift demands. These NRHA results are used to examine the residual drift trends and to construct benchmark residual drift hazard curves. An extensive feature selection process employing several Machine Learning (ML) algorithms precedes the development of hybrid data-driven predictive models. Finally, we compare our hazard-consistent predictions with currently available models and quantify the massive under- and over-estimations associated with previously proposed non-hazard-consistent assumptions.