Efficient sensitivity analysis for structural seismic fragility assessment based on surrogate models

Abstract

The variability in seismic fragility due to structural parameter uncertainty highlights the necessity of sensitivity analysis (SA) to identify critical parameters. However, the computational intensity of fragility analysis, especially global SA, limits its feasibility for complex structural systems. This study aims to overcome this challenge by developing a surrogate model-based framework for both global and local SA in seismic fragility assessment. The proposed methodology includes establishing sensitivity measures to assess deviations between fragility curves, with the global measure accounting for the full distribution of input variables. A multivariate seismic fragility analysis method, utilizing Gaussian process regression as a surrogate model, is introduced to efficiently generate both unconditional and conditional mean fragility curves. Integrating SA with this method significantly reduces the computational burden associated with nonlinear time history analysis. Additionally, a pooled sensitivity algorithm is proposed to address both uncertain and deterministic parameters. The methodology is validated through two case studies, demonstrating that the proposed approach effectively ranks the importance of structural parameters and distinguishes between deterministic and uncertain variables. The results confirm the efficiency, precision, and practical applicability of the proposed framework.