GPU CABARET for Dynamic Stall Simulations: Verification and Validation Study

A series of Wall-Modelled Large Eddy Simulations (WMLES) using the high-resolution CABARET method accelerated on GPU are performed to investigate the unsteady aerodynamic and aeroacoustic behaviour of a pitching NACA 0012 aerofoil under dynamic stall conditions. First, the numerical sensitivity study is conducted for a symmetrically pitching aerofoil corresponding to a chord length of c = 0.4 m and a wind speed of U∞ = 100 m/s for a range of reduced frequencies k = 0.013–0.63. The results of this study have been used to verify the minimum grid resolution and span-wise length of the computational domain required to resolve three-dimensional flow structures responsible for the aerodynamic response, including the lift and drag hysteresis. The range of numerical grid resolution and domain sizes leading to consistent predictions of aerodynamic parameters across several reduced frequencies with a transition from quasi-steady to strongly unsteady behaviour, characterised by delayed stall onset, increased lift overshoot, and widening hysteresis loops is established. Secondly, GPU CABARET solutions are obtained for the set of parameters for the conditions of the recent pitching aerofoil experiment conducted at the University of Bristol. This experiment corresponds to a smaller chord length aerofoil c = 0.3 m, a lower wind speed U∞ = 20 m/s, and a moderate reduced frequency k ≈ 0.094. For mid to higher frequencies, the wall surface pressure spectra obtained from the LES and the far-field noise spectra obtained from the LES solution coupled with the Ffowcs Williams - Hawkings permeable integration surface method are compared with the experimental measurements. In most cases, good agreement between the LES predictions and the experiment is demonstrated for frequencies ranging from 40 to 2000 Hz. In addition, the time-frequency analysis of the LES dataset is performed using the spectrogram method. The results show that the unsteady response is governed by phase-dependent vortex dynamics, with stronger contributions from the upstroke phase at higher mean angles of attack.