Advanced computational framework for fragility analysis of elevated steel tanks using hybrid and ensemble machine learning techniques

This research examines the elevated steel tanks’ seismic fragility with emphasis on the impact of significant structural and ground motion characteristics on the parameters of the fragility curve. Focusing on prediction accuracy by sophisticated machine learning techniques and the underlying variable importance in driving the fragility response, the research introduces meaningful insights toward seismic design and retrofitting. The fragility curve is developed for several Engineering Demand Parameters (EDPs) containing Base Shear $\left(V_b\right)$, Overturning Moment $\left(M_b\right)$, Tower Displacement $\left(X_s\right)$, vertical liquid surface displacement $\left(d_v\right)$, meridional tank wall stress (Sigma), Elephant Foot Buckling $\left(f_{\mathrm{pb}}\right)$, and base isolation displacement $\left(U_b\right)$. A detailed sensitivity analysis is conducted using the Fourier Amplitude Spectrum Technique (FAST) to assess the impact of input characteristics on the median $\left(\theta \right)$ and dispersion $\left(\beta \right)$ parameters of the fragility curve. For tanks that are not isolated, geometric characteristics like slenderness (S) and liquid height (H) play greater roles than material characteristics like the density of steel $\left({\rho }_{\mathrm{steel}}\right)$, emphasizing the structural superiority of geometry to material weight in seismic behavior. Likewise, environmental and ground motion inputs like peak ground acceleration (PGA) and site conditions (Field) show different types and magnitudes of impacts across the EDPs. For isolated tanks, sensitivity patterns change, with the damping ratio $\left({\xi }_b\right)$, isolation base period $\left(T_b\right)$, and viscous behavior (ν), demonstrating significant impact on some, like $U_b$ and Sigma, while others like H and R show minimal impact. These results highlight the variable-specific nature of fragility behavior, providing a granular understanding of the design characteristics that play crucial roles in determining seismic vulnerability. Results enhance model interpretability and provide actionable insights that support the engineer's prioritization of the most impactful design parameters. The feature-based understanding enhances the synergy between data-driven modeling and domain-specific knowledge, leading to performance-based design of elevated steel tanks.