Fourier neural operator-based asphalt concrete pavement response modeling

Mechanical response modeling of asphalt pavements is essential for both pavement design and maintenance. However, neural network-based approaches face challenges in acquiring high-fidelity data and ensuring model generalization. To address these issues, this study proposes BT-FNO, a surrogate model for predicting asphalt pavement mechanical responses by integrating the Boltzmann superposition principle, Fourier Neural Operator and transfer learning. First, based on the Boltzmann superposition principle, the viscoelastic response modeling is simplified, reducing the time cost of high-fidelity data acquisition via 3D finite element method by approximately 60%, while unifying conditions with identical load shapes to make layer parameters the sole input to the neural network. Then, the Fourier Neural Operator is employed to approximate the mapping from pavement layer parameters to displacement responses, achieving higher accuracy and better generalization than U-Net while being four orders of magnitude faster than finite element method. Finally, transfer learning improves model’s adaptability to varying load shapes, significantly reducing the data requirements for network training. These results demonstrate that BT-FNO is a reliable and efficient surrogate model for accelerating asphalt pavement response modeling.