Real-time life-cycle seismic fragility assessment of coastal bridges: a novel approach using deep temporal neural networks

As global climate change and infrastructure ageing accelerate, particularly in coastal regions, the demand for real-time life-cycle seismic management of bridges and rapid post-earthquake response is increasing. Traditional seismic fragility assessments, based on static models or finite element (FE) analyses at specific time points, are computationally intensive and inefficient, and fail to meet practical decision-making needs. This study proposes a real-time life-cycle (0–100 years) seismic fragility assessment framework for coastal bridges, accounting for time-dependent deterioration caused by chloride corrosion, with a case study of medium- and small-span curved bridges in southeastern coastal China. A corrosion-informed numerical model is developed, and parametric cyclic FE analyses generate seismic fragilities at different service years, forming time-series samples of fragility parameters. Based on these samples, a deep temporal neural network (DTNN) predicts time-varying fragility parameters, and a graphical user interface (GUI) is developed. The results demonstrate that seismic fragility increases with service time and is affected by horizontal curvature. The proposed model accurately captures long-term fragility evolution, achieving a mean absolute percentage error below 5%. The GUI produces fragility curves for any service year within 11.4 seconds, reducing computational time by over 1500 times compared with conventional FE-based methods.