Research on the recurrent neural network-based fatigue damage model of asphalt binder and the finite element analysis development

Abstract The evaluation of the fatigue performances of asphalt binders is a difficult engineering problem due to the complexity of their polymer compositions. In the present work, a fatigue damage model of asphalt binders based on a recurrent neural network (RNN) is introduced. The proposed model was validated by the finite element method (FEM). It was found that the RNN could accurately predict the fatigue behaviors of asphalt under cyclic loading, and the loading frequency and the strain ranged from 0.1 to 10 Hz and from 0.005 to 1, respectively. Traditional constitutive models have difficulty operating under such large loading ranges. The average error of all the test specimens achieved by the proposed method was 0.078. As the RNN was purely linear in the FEM model, the FEM converged quickly.