A surrogate model approach for uncertainty treatment in performance-based engineering

Uncertainty quantification and propagation (UQ&P) is a key component of performancebased engineering (PBE).The PEER framework, widely accepted as the standard approach for PBE under earthquake hazards (PBEE), considers the probabilistic characterization of output variables of each of the four sequential analysis blocks (i.e., probabilistic seismic hazard analysis or PSHA, probabilistic seismic demand analysis or PSDA, probabilistic seismic fragility analysis or PSFA, and probabilistic seismic loss analysis or PSLA).Therefore, uncertainty at each stage must be adequately characterized and propagated to the following stages to establish a meaningful risk assessment.One significant challenge in UQ&P is the high computational cost of extensive simulation performed on structure-level finite element (FE) models.To mitigate this cost, researchers have adopted machine and deep learning surrogate modeling techniques, in which a reduced set of data is used to train a cheap-to-evaluate prediction model.In this area, different approaches have been taken to replace massive FE simulations.These include structure-specific modeling of responses, direct modeling of fragility function (FF) parameters, and portfoliooriented/regional-scale models.Each approach alludes to a certain scope of application and its associated level of detail modeling.For instance, while portfolio-oriented or regional-scale approaches are useful towards improving disaster management policies, structure-specific models might retain greater detail in their focus on structural dynamics and mechanics.This work addresses the challenge at a structure-specific scale by proposing a surrogate model-based approach for UQ&P in PBEE which considers (i) probabilistic seismic hazard modeling and consistent ground motion selection, (ii) probabilistic structural modeling by considering parameter uncertainty in nonlinear material and components, (iii) development of surrogate models for model responses of interest, (iv) probabilistic seismic demand modeling including aleatoric and epistemic uncertainty arising from seismic and model parameter variability, as well as from the surrogate models themselves, (v) fragility analysis considering