Integrated framework for seismic fragility assessment of cable-stayed bridges using deep learning neural networks

The increasing intensity of strong earthquakes has a large impact on the seismic safety of bridges worldwide. As the key component in the transportation network, the cable-stayed bridge should cope with the increasing future hazards to improve seismic safety. Seismic fragility analysis can assist the resilience assessment under different levels of seismic intensity. However, such an analysis is computationally intensive, especially when considering various random factors. The present paper implemented the deep learning neural networks that are integrated into the performance-based earthquake engineering framework to predict fragility functions and associated resilience index of cable-stayed bridges under seismic hazards to improve the computational efficiency while having sufficient accuracy. In the proposed framework, the Latin hypercube sampling was improved with additional uniformity to enhance the training process of the neural network. The well-trained neural network was then applied in a probabilistic simulation process to derive different component fragilities of the cable-stayed bridge. The estimated fragility functions were combined with the Monte Carlo simulations to predict system resilience. The proposed integrated framework in this study was demonstrated on an existing single-pylon cable-stayed bridge in China. Results reveal that this integrated framework yields accurate predictions of fragility functions for the cable-stayed bridge and has reasonable accuracy compared with the conventional methods.