An Efficient Kriging Surrogate Model for Developing Seismic Fragility Curves of Low-rise Steel Structures Considering Epistemic Uncertainties

Abstract

Seismic fragility curves are essential for assessing structural performance under earthquake loading; however, their development requires addressing both aleatory and epistemic uncertainties, which can be computationally intensive. This study introduces an efficient Kriging-based surrogate modeling framework for deriving collapse fragility curves of steel moment-resisting frames while explicitly incorporating uncertainties in key hysteretic parameters. Incremental Dynamic Analysis (IDA) is conducted on a 5-story steel frame using 22 far-field ground motions and 125 structural parameter scenarios, reflecting probabilistic variability in plastic rotation capacity, post-capping rotation, and cyclic deterioration. The Kriging surrogate is calibrated using maximum likelihood estimation, enabling accurate interpolation of nonlinear structural response behavior. Validation results demonstrate a prediction accuracy of R² = 0.92, outperforming the conventional response surface method (R² = 0.88). Additionally, the Kriging model reduces mean collapse capacity prediction error to 0.5%, compared to 24% in the response surface approach, while maintaining significantly lower computational cost. The proposed methodology provides a computationally efficient and uncertainty-aware framework for collapse fragility curve development, supporting performance-based seismic design and risk assessment.