Fast prediction of compressor flow field based on a deep attention symmetrical neural network DOI
Yue Wu, Dun Ba, Juan Du

и другие.

Physics of Fluids, Год журнала: 2024, Номер 36(11)

Опубликована: Ноя. 1, 2024

Accurate and rapid prediction of compressor performance key flow characteristics is critical for digital design, twin modeling, virtual–real interaction. However, the traditional methods obtaining field parameters by solving Navier–Stokes equations are computationally intensive time-consuming. To establish a model in transonic three-stage axial compressor, this study proposes novel data-driven deep attention symmetric neural network fast reconstruction at different blade rows spanwise positions. The integrates vision transformer (ViT) convolutional (SCNN). ViT extracts geometric features from passages. SCNN used deeper extraction input such as boundary conditions coordinates, enabling precise predictions. Results indicate that trained can efficiently accurately reconstruct internal 0.5 s, capturing phenomena separation wake. Compared with numerical simulations, current offers significant advantages computational speed, delivering three-order magnitude speedup compared to fluid dynamics simulations. It shows strong potential engineering applications provides robust support building models turbomachinery fields.

Язык: Английский

Fast prediction of compressor flow field based on a deep attention symmetrical neural network DOI
Yue Wu, Dun Ba, Juan Du

и другие.

Physics of Fluids, Год журнала: 2024, Номер 36(11)

Опубликована: Ноя. 1, 2024

Accurate and rapid prediction of compressor performance key flow characteristics is critical for digital design, twin modeling, virtual–real interaction. However, the traditional methods obtaining field parameters by solving Navier–Stokes equations are computationally intensive time-consuming. To establish a model in transonic three-stage axial compressor, this study proposes novel data-driven deep attention symmetric neural network fast reconstruction at different blade rows spanwise positions. The integrates vision transformer (ViT) convolutional (SCNN). ViT extracts geometric features from passages. SCNN used deeper extraction input such as boundary conditions coordinates, enabling precise predictions. Results indicate that trained can efficiently accurately reconstruct internal 0.5 s, capturing phenomena separation wake. Compared with numerical simulations, current offers significant advantages computational speed, delivering three-order magnitude speedup compared to fluid dynamics simulations. It shows strong potential engineering applications provides robust support building models turbomachinery fields.

Язык: Английский

Процитировано

1