Mainshock-aftershock fragility surfaces: A machine learning approach using direct successive simulation

This study presents a machine learning (ML)-based methodology for developing fragility surfaces under mainshock–aftershock (MS–AS) seismic sequences, extending a previously validated framework originally developed for earthquake–tsunami (EQ–TS) scenarios. Using a two-phase nonlinear time history analysis approach and a dataset of 30 real MS–AS ground motion pairs, fragility surfaces are developed for ductile and non-ductile reinforced concrete (RC) frame archetypes. The ML model predicts joint exceedance probabilities across paired intensity measures by leveraging the cumulative probability form of single-event fragility curves, whose statistical characteristics remain consistent across different hazard pairings. This hazard-agnostic formulation enables the model to generalize from EQ–TS to MS–AS sequences without retraining. Results show excellent agreement with simulation-based benchmarks (Coefficient of Determination, R 2 > 0.97; Root Mean Square Error, RMSE < 0.015), confirming the model’s predictive accuracy, generalizability, and computational efficiency. The synthesized fragility surfaces are formatted for integration into platforms such as IN-CORE (Interdependent Networked Community Resilience Modeling Environment), supporting scalable risk and resilience assessment for building portfolios exposed to sequential seismic hazards.