Resumen
A safety analysis process for an immersed bridge is proposed, in which material nonlinear properties, added mass, soil spring, plastic hinge length and property, nonlinear pushover analysis, and time history analysis are included. Outcomes of interest in a bridge analysis, such as displacement ductility, is not an easy task if the authentic model is adopted. The least-squares support-vector machine (LSSVM) is therefore proposed for replacing the analysis model. Water-immersed bridges under varied water depths, scouring depth, and seismic intensity are investigated. The feasibility of using machine learning as a surrogate model is discussed. Results indicate that the trends of fragility curves derived from LSSVM are very similar to those of the authentic model. Therefore, LSSVM is a suitable tool to reduce the computational burden. Analyses also show that stream velocity and pier length are two important factors for the safety of an immersed bridge, and the immersed effect should not be ignored if pier length is long.