Seismic Fragility of Railway Bridge Classes: Methods, Models, and Comparison with the State of the Art

This study proposes an approach for developing seismic fragility models based on elastic net regularized logistic regression and applies it to two railway bridge classes typical to the central and southeastern United States (CSUS). Railway bridge class fragilities are not available in the literature despite recorded evidence of earthquake damage to railway bridges. The introduction of elastic net regularization helps in selecting the best set of predictor variables for fragility modeling even if they are mutually correlated. The proposed fragility models are compared to their corresponding highway bridge counterparts, given that current practice in regional risk assessment recommends adopting these as proxies for railway bridge fragility. The analysis reveals that multispan simply supported steel girder railway bridges, the most common bridge class, show lower fragility than their corresponding highway bridge counterparts. However, multispan simply supported steel through plate girder railway bridges, the other common bridge class, show comparatively higher seismic fragility. The proposed fragility models serve as key inputs in a broader framework of quantifying seismic resilience of railway networks, as well as providing a practical baseline for seismic loss assessment and retrofit prioritization. (C) 2019 American Society of Civil Engineers.