Investigation of Hybrid Turbulence Modeling Techniques in the Context of Aeroacoustic Simulation

Aerodynamic noise is produced in air through internal interaction because of turbulent flows or external interaction with solid structures. The increase of traffic volume and its emissions of the aerodynamic noise makes this noise a non-negligible factor detrimen- tal to the human’s health. Consequently, the research regarding the reduction of it has gained significant importance. The computational aero-acoustics (CAA) investigates the aeroacoustic phenomena using computational techniques. In comparison to experimentation, simulation-based methods are in general less expensive and can achieve much more detailed information. As the computing power increases, it can be foreseen that CAA will play a more important role in the future. However, there are still many challenges remaining in this field. A reliable solution of the broad band aeroacoustic problems re- quires accurate solution of the underlying flow problems, which cannot be achieved by the Reynolds averaged Navier-Stokes (RANS) models. The direct numerical simulation (DNS) method and the large eddy simulation (LES) models are not applicable for engineering problems in a foreseeable future due to their high computational costs. The application of hybrid LES/RANS models combine the advantages of both LES and RANS models and have shown very promising results. The objective of this work is to enhance the understanding of the application of hybrid LES/RANS turbulence modeling strategies in aeroacoustic simulations. For this purpose, a new hybrid LES/RANS turbulence model, the limited numerical scales (LNS) model, is first implemented and then validated. The latency factors determining which part uses RANS and which uses LES models are derived. The ζ − f based LNS model yields quite satisfactory results, owing to the introduction of some scales of anisotropy in the turbulence modeling. Thus, the ζ − f based LNS model is applied in the subsequent aeroacoustic simulations. In addition, a new coupling strategy between the flow solver and the acoustic solver is implemented, which enables the application of different computational domains and different space discretizations for the flow and the sound. By doing so, the computational cost for the acoustic simulation is substantially reduced compared with the existing coupling strategy. The LES model resolves the large eddies and implicitly accounts for the small scale structures, indicating that a part of the high frequency noise cannot be reproduced by the LES model. The hybrid LES/RANS model uses RANS models in certain regions to reduce the computational cost. It can be expected that the range of scales obtained from the hybrid turbulence models is further narrowed. In order to predict the noise from the unresolved scales, a synthetic method which is able to reconstruct the unresolved scales artificially is implemented. The verification based on a benchmark channel flow shows that the synthetic method implemented here is able to generate small scale fluctuations and improve the spectral results significantly. With the help of a cylinder test case, it is shown, that the hybrid LNS model delivers more accurate sound spectra than the RANS model. The LNS model reproduces the sound spectrum accurately with only minor mismatches in the region around the fundamental frequency. The use of RANS mode in the shear layer as well as in the wake region leads to the under-prediction of the von Kármán vortex, which in turn results in this mismatch in the spectrum. A test case with the NACA 0012 airfoil is used to compare the LNS model and an existing hybrid turbulence model, the very large eddy simulation (VLES) model. The VLES model provides more detailed information in terms of acoustic sources. On the other hand, the LNS model behaves better at predicting the flow separation. Overall, the ζ − f based LNS model is capable of providing accurate aeroacoustic results at moderate computational cost