Epistemic and Aleatory Modeling Uncertainty in Seismic Collapse Capacity of Reinforced Concrete Components, in Influence of Epistemic and Aleatory Modeling Uncertainty on the Seismic Collapse Capacity of Reinforced Concrete Components

Assessment of extreme limit states in structures due to earthquake loading may be associated with significant uncertainty due to the randomness in structural and material properties (aleatory uncertainty), and the bias induced by numerical simulation of the strength and stiffness deterioration mechanisms (epistemic uncertainty). We evaluate the influence of epistemic and aleatory modeling uncertainties on estimating seismic collapse capacities of reinforced concrete columns using models of different fidelities, which span lumped-plasticity (LP) models and distributed-plasticity (DP) models, which can capture the propagation of structural damage in a mesh-objective manner. We benchmark the performance of different models against laboratory observations, and analyze the structural fragilities estimated by each model for different limit states, and under the influence of individual earthquakes or earthquake sequences. Our analysis identifies systematic differences between the seismic collapse capacities estimated by the DP and LP models. Our findings suggest that conventional LP models could overestimate the collapse capacity of near-fault structures and the residual capacity of structures subjected to mainshock-aftershock sequences. An extensive suite of Monte Carlo simulations of representative columns demonstrates that the dispersion in the collapse capacity due to aleatory modeling uncertainty is, in most cases, substantially smaller than the epistemic uncertainty (or bias) associated with using the alternative modeling methods, and than the dispersion due to the ground motion record-to-record variability, especially for DP models. Finally, we identify the modeling methods that offer a reasonable performance trade-off between sensitivity to aleatory uncertainty and the computational effort required to simulate collapse limit states.