Bayesian rational function models for aerodynamic derivatives of parametrized bridge decks

A Bayesian surrogate-modeling framework based on Gaussian process regression is presented for predicting frequency-dependent aerodynamic derivatives of parametrized bridge decks from limited experimental data or simulations. The method builds on the classical rational-function formulation of self-excited forces and extends it to provide smooth predictions across deck parameters such as height or angle. It is computationally efficient and explicitly accounts for experimental noise and its propagation to the predictions. The approach is first validated on a synthetic numerical example, where the underlying aerodynamic derivatives are recovered within their uncertainty bounds. An experimental validation is also shown using data from wind tunnel tests of five bridge decks with varying heights. Finally, these experimental results are applied to the aeroelastic shape optimization of a suspension bridge to find the deck design with the best resistance to flutter.