Exploring Modeling Fidelity for Computationally Efficient Seismic Performance Assessment

Advances in structural modeling, combined with the growth in computational processing power, have motivated the use of more accurate but, at the same time, more computationally intensive finite element models (FEMs) for seismic performance assessment. Though, these models can accurately describe hysteretic structural behavior even for strong levels of earthquake shaking, their use for seismic risk assessment imposes, still, a substantial challenge, since uncertainty propagation in this setting, typically performed using Monte Carlo (MC) approaches, entails a large number of nonlinear response history analyses (NLRHAs). To alleviate this challenge, this dissertation develops methodologies that can leverage lower fidelity models to accommodate higher computational efficiency while still maintaining the accuracy of an original high-fidelity FEM for a structure. The foundation of all developments considered is a new reduced order model (ROM) framework for hysteretic moment-resisting frame structures. The ROM is calibrated by explicitly comparing its nonlinear time-history response to the nonlinear FEM response, under different earthquake excitations. Two different calibration approaches are proposed: i) a deterministic, nonlinear least-squares optimization and ii) a Bayesian model updating implementation. An extension of the ROM framework to describe the behavior of structures with seismic protective devices (SPDs) is also examined. It is shown that the ROM calibration does not need to be updated to accurately assess the effect of the SPD addition. While the ROM can offer a substantial speedup, the corresponding predictions might be biased when compared against the ones established using the higher-fidelity FEM. A multi-fidelity Monte Carlo (MFMC) framework for seismic risk assessment is also proposed to guarantee unbiased predictions. This approach utilizes evaluations from both the high-fidelity nonlinear FEM and the ROM, to established unbiased risk estimates, and balances the number of evaluations between the two models to accommodate the highest computational efficiency (smallest MC variability) within the uncertainty propagation. These methodological advancements supported additional advances across two different earthquake engineering practical applications: a) design of seismic protective devices (SPDs), focusing on the impact of the fidelity of the structural response estimation on the device effectiveness; b) accuracy of seismic performance assessment within a Performance-Based Earthquake Engineering (PBEE) setting, examining the potential vulnerabilities of alternative approaches to achieve computational efficiency by approximating either the uncertainty propagation or the fidelity of the structural response estimation