New Papers in Fluid Mechanics
Machine-learning-based simulation of turbulent flows over periodic hills using a hybrid U-Net and Fourier neural operator framework
Author(s): Yunpeng Wang, Huiyu Yang, Zelong Yuan, Zhijie Li, Wenhui Peng, and Jianchun Wang
A machine-learning-based surrogate model is proposed for the large-eddy simulation of three-dimensional turbulent flows over curved boundaries with strong flow separation. The model, termed as hybrid U-Net and Fourier neural operator (HUFNO), is based on an integrated framework of convolutional neural networks and Fourier neural operators, tailored for problems involving mixed periodic and non-periodic boundary conditions. The HUFNO model is validated in the fast prediction of turbulent dynamics of periodic-hill flow, with transferable accuracy to unseen initial conditions, Reynolds numbers, and hill shapes.
[Phys. Rev. Fluids 11, 024601] Published Mon Feb 02, 2026
Pages
