ADVANCED CFD TECHNIQUES FOR PREDICTION OF UNSTEADY FLOW DISTORTION IN AERO ENGINE INTAKES

Advanced computational techniques, such as Delayed Detached Eddy Simulation (DDES) and Large Eddy Simulation (LES) methods, have demonstrated the complex unsteady aerodynamics associated with S-duct intakes, but they are still too computationally expensive. The promising substitute Lattice-Boltzmann Method (LBM) was selected in this project. The commercial solver XFlow was used to perform the simulation of an S-duct intake operating in the incompressible regime, with an inlet Mach of 0.27. The impact of sub-grid scale modeling and lattice structure on LB unsteady flow distortion predictions at the S-duct aerodynamic interface plane has been addressed. A validation against Particle Image Velocimetry (PIV) measurement has been then conducted. The main outcome of this work is that the grid density is paramount to predict the unsteady distortion events. An ever-increasing secondary flow extent arises as a consequence of the lattice refinement, directly affecting swirl distortion descriptors. On the other side, the sub-grid scale modeling or turbulence model shows a second-order impact on unsteady flow distortion prediction capability. The spectral analysis has revealed a maximization of the unsteady fluctuations at 𝑆𝑡 ∈ [0.4, 0.6], independently of the turbulence model or grid size. Concerning the validation against PIV measurements, LB detects the typical low-velocity steady regions, even though the time-averaged flow field is not fully captured. On the other side, LB simulations indicated notably lower levels of unsteadiness across the aerodynamic interface plane (AIP) than time-resolved (TR) PIV data, especially for the in-plane velocity components. Finally, the swirl distortion extent is severely underpredicted by LB, especially at inner radii, where the highest distortion takes place, as a direct consequence of its impossibility to forecast the adequate azimuthal flows development. Despite this, the main swirl distortion descriptors distributions are well predicted.