Seismic fragility analysis of elevated RC tanks based on IDA and machine learning

Abstract

Elevated reinforced concrete (RC) water tanks are critical lifeline structures whose seismic performance is governed by fluid–structure interaction (FSI) and staging systems. Conventional fragility curves developed through incremental dynamic analysis (IDA) provide probabilistic insights but demand extensive nonlinear time‐history analyses, limiting their practical use. This study introduces a hybrid IDA–machine learning (ML) framework that couples IDA with support vector regression (SVR) and a physics-informed neural network (PINN) surrogate to accelerate fragility curve generation for three elevated water tanks (75 m³, 320 m³, 1008 m³). Finite element (FE) models in SAP2000 embed Housner’s added mass to capture hydrodynamic effects. IDA under 22 far-field ground motions produces 738 nonlinear response samples of peak inter-story drift ratio (IDR) across spectral acceleration (Sa), peak ground velocity (PGV), and geometric inputs. SVR and PINN models are trained on this dataset, with Bayesian hyperparameter tuning and Shapley additive explanations (SHAP) interpretability. PINN outperforms SVR (R2 = 0.99 vs 0.95; RMSE = 0.0008 vs 0.0021), sustaining errors below 5% at collapse prevention (CP) thresholds while delivering millisecond-scale inference. ML-derived fragility curves align with IDA baselines for immediate occupancy (IO), life safety (LS), and CP states within 0.05 g medians. Global sensitivity and input uncertainty analysis via Saltelli quasi-Monte Carlo highlight standard deviation (SD) as the principal driver of IDR variance (> 55%) and define a 5%–95% IDR band of 0.005–0.045. The proposed approach cuts computational time by orders of magnitude while preserving probabilistic rigor, enabling rapid, code-compliant seismic risk assessment of elevated RC tanks.