Seismic Fragility Analysis of RC Continuous Girder Bridges Using Artificial Neural Network

This research aims to develop seismic fragility curves for small- and medium-sized concrete bridges. Fragility curves were generated as a function of the probability of reaching or exceeding a specific limit state in terms of the peak ground acceleration (PGA) and acceleration spectral intensity (ASI). To this end, a hybrid dataset of the seismic performances of bridges was prepared by combining the results of numerical analyses and neural predictions. Three-dimensional finite-element models for 1032 bridge-earthquake cases were created, considering the nonlinear behavior of critical bridge components. In addition, multilayer perceptron (MLP) neural networks were employed to simulate artificial earthquake-bridge performance scenarios. The yield stress of reinforcing bars (Fy), the bridge height (H) as well as PGA and ASI, were considered as the input vectors of the artificial neural networks (ANN). The results of this study revealed that MLP neural networks are capable of simulating the seismic performances of bridges appropriately. It was also shown that providing a hybrid dataset of numerical results and neural predictions could lead to the fragility curves of higher correlation coefficients. The results also presented that the PGA-based fragility curves had better correlation coefficients comparing to ASI-based ones.