A deep learning-based method for full-bridge flutter analysis considering aerodynamic and geometric nonlinearities

Abstract

The nonlinearity effect grows with the span length of the bridge. Traditional linear flutter analysis methods are gradually unable to meet research and design needs. This study proposed a time-domain nonlinear method for the 3D full-bridge flutter analysis based on deep learning. It considered multiple effects including aerostatic effect, as well as aerodynamic and geometric nonlinearities of the structure. Firstly, a nonlinear reduced-order aerodynamic model was constructed for the bridge deck section based on the long short-term memory (LSTM) network. The section displacement was the input, and the aerodynamic force was the output. The aerodynamic data were obtained from CFD numerical simulation of forced vibration. Next, the 3D full-bridge flutter analysis framework was constructed. The aerodynamic forces predicted by the reduced-order model were imposed as concentrated forces on the deck nodes of bridge finite element model. The restart technique was implemented to achieve the dynamic iterative calculation. The influence of the initial excitation caused by the static wind effect on the vibration mode was eliminated by changing Rayleigh damping to make the bridge reaches the static wind equilibrium state in advance. Finally, a case study on a triple-tower suspension bridge was performed. The nonlinear flutter phenomenon observed in the full-bridge model wind tunnel test was reproduced using the proposed method. The influence of geometric nonlinearity on the flutter phenomenon was analyzed. This study provides an important reference for accurately evaluating the nonlinear flutter performance of bridges and further improving their anti-wind performance.