Optimal intensity measures for fragility assessment of bridge components under mainshock–aftershock seismic sequences in subduction zones

Bridges in seismically active regions are exposed to mainshock–aftershock (MS–AS) sequences, yet most fragility models either neglect AS effects or rely on intensity measures (IMs) developed for single events, resulting in overly optimistic predictions. This study introduces a framework that identifies optimal IMs for bridge components under MS–AS sequences, addressing a gap largely untested in subduction environments. The methodology combines seismic disaggregation to define site-specific scenarios with a stochastic AS model tailored to subduction zones, generating 10,000 physically consistent MS–AS sequences for each MS scenario. Conditional IM vectors are generated using a recurrent neural network-based ground motion model, calibrated for the Chilean subduction context, which preserves observed MS–AS spectral correlation. Ground motions are selected through multi-objective optimization to ensure hazard consistency. The framework is demonstrated on the Águila Norte Bridge in Chile through 12,000 nonlinear dynamic analyses capturing structural response under MS–AS excitation. Engineering demand parameters for elastomeric bearings, abutments, and columns are defined using both traditional indices and residual-capacity-based measures. Eleven IMs are evaluated using four complementary metrics—efficiency, practicality, proficiency, and sufficiency. Results consistently rank Housner Intensity (HI), average spectral acceleration (Sa_avg(T₁)), and orientation-independent spectral acceleration (Sa_RotD50(T₁)) ahead of peak ground acceleration (PGA) and spectral acceleration at the fundamental period (Sa(T₁)), with R² improving by 8–14% and selected optimal IMs reducing dispersion β by more than 50% relative to PGA for the elastomeric-bearing response. Logistic exceedance probability surfaces further confirm that neglecting ASs systematically underestimates component-level damage exceedance probabilities.