Fragility analysis of structures under non-synoptic winds using stochastic emulators

The assessment of structures under non-synoptic winds such as tornadoes and downbursts is challenging due to the complexity of both structural dynamics and highly variable wind fields. The difficulty increases in Performance-Based Wind Engineering (PBWE), where fragility models quantify structural vulnerability by linking Engineering Demand Parameters (EDPs) to wind Intensity Measures (IMs). Traditional methods often rely on simplified models or computationally expensive Monte Carlo simulations with limited samples. To overcome these limitations, this work formulates the wind-fragility problem within a stochastic-simulator framework, treating wind-field turbulence as aleatory uncertainty and enabling the use of stochastic emulators. We employ Stochastic Polynomial Chaos Expansions (SPCE) to efficiently map IMs to conditional distributions of EDPs and to generate fragility curves through SPCE post-processing. Although no new SPCE methodology is introduced, we investigate its ability to handle multi-dimensional IMs, compute multi-dimensional fragilities, and achieve accurate and efficient solutions for non-synoptic wind effects. Results show that this non-parametric approach captures detailed response features often missed by traditional parametric fragility models. Validation against high-fidelity simulations indicates high accuracy, with mean square errors around 1% for 2D fragility surfaces using only 125 support points. The method also extends naturally to higher-dimensional IM vectors while maintaining consistent vulnerability representation. Additionally, the SPCE post-processing capability allows multiple fragility surfaces to be generated from a single emulator, underscoring its potential to enhance PBWE studies involving complex wind hazards. • Fragility under turbulent non-synoptic winds via stochastic simulation. • Structural response is emulated using Stochastic Polynomial Chaos Expansion (SPCE). • Few structural analyses yield accurate, complex fragility surfaces. • Fragility surfaces for multiple performance levels require no additional simulations. • Higher-dimensional fragility problems are efficiently solved.