The neural network predicts flow fields based on four inputs: three distance functions (defining proximity to walls, secondary inlets, and outlets) and the specific secondary velocity value. The model was validated against numerical simulations performed in Ansys Fluent at an inlet velocity of 20 m/s with varying secondary velocities (0–40 m/s). Comparisons of velocity and pressure fields in both vertical and horizontal cross-sections demonstrate high agreement between the CNN predictions and traditional CFD computations. This validated 3D steady-state prediction serves as the foundation for future flutter instability studies.
Figure 1a – Comparison of CNN prediction (top) and CFD simulation (bottom images) – velocity field
Figure 1b – Comparison of CNN prediction (top) and CFD simulation (bottom images) – pressure field