Artificial Neural Network model to predict the flutter velocity of suspension bridges

This paper discusses the implementation of an artificial neural network (ANN) for predicting the critical flutter velocity of suspension bridges with closed box deck sections. Deck chord length, bridge weight and structural damping were varied. The ANN model was derived and trained using a dataset of critical flutter velocities, calculated using flutter derivatives (FDs) from experiments and by varying geometrical and mechanical parameters. The ANN model was derived by training and comparing two different, preliminary ANNs. The first one was based on thirty sets of experimental FDs. This first set was subsequently used to calibrate the second model, based on surrogate FDs obtained by curve fitting of the experimental data. The surrogate FD dataset was subsequently expanded by Nataf-model Monte Carlo (MC) and Polynomial Chaos Expansion (PCE)-model MC simulation. Finally, the ANN was employed to synthetically generate a larger dataset of critical flutter velocities and estimate the corresponding probability distribution. © 2020 Elsevier Ltd