A novel attention enhanced deep neural network for hypersonic spatiotemporal turbulence prediction

Physics of Fluids, Journal Year: 2024, Volume and Issue: 36(5)

Published: May 1, 2024

High Reynolds number turbulent flow of hypersonic vehicles exhibits multi-scale structures and non-equilibrium high-frequency characteristics, presenting a significant challenge for accurate prediction. A deep neural network integrated with attention mechanism as reduced order model is proposed, which capable capturing spatiotemporal characteristics from high-dimensional numerical data directly. The leverages encoder–decoder architecture where the encoder captures high-level semantic information input field, Convolutional Long Short-Term Memory learns low-dimensional characteristic evolution, decoder generates pixel-level multi-channel field information. Additionally, skip connection structure introduced at decoding stage to enhance feature fusion while incorporating Dual-Attention-Block that automatically adjusts weights capture spatial imbalances in turbulence distribution. Through evaluating time generalization ability, effectively evolution characteristics. It enables rapid prediction high over reasonable accuracy maintaining excellent computational efficiency.

Language: Английский

Physics-assisted multi-scale convolutional autoencoder for turbulence reduced-order modeling

Physics of Fluids, Journal Year: 2025, Volume and Issue: 37(1)

Published: Jan. 1, 2025

Reduced-order modeling techniques, including the proper orthogonal decomposition and dynamic mode decomposition, have been widely applied in unsteady flow rather than fully developed turbulent flows, but these techniques are faced with challenges simulating turbulence high degrees of freedom complex nonlinear interactions. One possible approach is to utilize a series neural networks, such as autoencoders, reduce dimensionality flows. This study began combining multi-scale convolutional autoencoder block attention module extract main features turbulence. Then, physical constraint terms were added loss function improve accuracy feature extraction. Finally, data was restored potential properties. Forced isotropic Reλ=418 channel Reτ=1000 employed test model's performance, numerical results verified that model can accurately has an excellent ability restore data.

Language: Английский

Prediction of bend-pipeline erosion field based on CFD and Reduced Order Method

Chemical Engineering and Processing - Process Intensification, Journal Year: 2025, Volume and Issue: unknown, P. 110320 - 110320

Published: April 1, 2025

Language: Английский