Stochastic Response Assessment of Cross-Sea Bridges under Correlated Wind and Waves via Machine Learning

The stochastic response of cross-sea bridges is susceptible to the significant effects of wind and waves. In this study, an efficient probabilistic assessment framework for cross-sea bridges was developed by combining a wind–wave bridge (WWB) model with machine learning methods. The WWB model was first proposed based on finite element analysis (FEA) where the wind and wave parameters were obtained by structural health monitoring (SHM) and then correlated using copula models. The coupling effects in the wind–bridge and the wave–bridge were solved using the Newmark-β method. Taking a cable-stayed bridge as an example to illustrate the accuracy and efficiency of the proposed method, the WWB model was established and then performed to compute the dynamic response at different positions on the bridge. To deal with the time-consuming issues, a learning machine including support vector regression (SVR) and Latin hypercube sampling (LHS) was implemented to substitute further finite element calculations. The WWB model was simplified parametrically as response surfaces for stochastic wind and wave variables, and probabilistic simulations with a large number of samples were performed. The results show that the wind load controlled the displacement response of the girder, while the wave load dominated the base shear response of the foundation. The bridge response, considering when wind and waves were correlated, was 6%–25% lower than that when wind and waves were independent. Further response contour analysis demonstrated a direct relationship between the environmental parameters and the structural response to quickly estimate the bridge's maximum response in different return periods.