Deep learning-based framework for time-dependent reliability analysis of a cable-stayed bridge with corroded PSC box girders under gravity loads.

Proper maintenance of civil infrastructure, such as cable-stayed bridges, is critical to ensuring safe and efficient logistics and transportation. Prestressed concrete (PSC) box girders in such bridges are particularly vulnerable to corrosion-induced strand deterioration. Existing methods often overlook system-level behavior or rely on computationally intensive simulations. This study proposes a deep learning-based framework to evaluate the time-dependent reliability of cable-stayed bridges with corroded PSC box girders. We develop a finite element model of the Hwayang-Jobal Bridge to simulate progressive structural degradation over time. A deep neural network (DNN) is then trained to predict the flexural characteristics of PSC box girders under varying corrosion levels and cross-sectional properties. The DNN outputs are then integrated into the global bridge model to enable efficient, Monte-Carlo-based system reliability analysis. The proposed framework incorporates various sources of uncertainty, including corrosive environmental factors, bridge geometry, and load redistribution within the structural system. Numerical investigations demonstrate the impact of corrosion on structural performance and load redistribution mechanisms. This comprehensive and efficient approach offers a robust tool for evaluating the time-variant reliability of aging bridge systems.