[saintandsinnerz]

Good evening CFD online,

I am currently an MEng student who is conducting his research project. The goal is to use a machine-learning algorithm to discover something about the form of the anisotropy term of the algebraic stress model. The inputs will be the variables of the flow field from high fidelity simulations of different flow cases (external, internal aerodynamics). The current focus is on obtaining a validated high-fidelity simulation of the flow over an Ahmed Body. An LES simulation has been chosen with the WALE (Wall-Adapting-Local Eddy-Viscosity) subgrid-scale modeling.

The simulation will be performed in OpenFOAM and the mesh generated in STAR-CCM+ (v2019.2 Build 14.04.011). The mesh will then be converted using the ccm26ToFoam function. I am running an Intel(R) Core i7-7500U CPU with 4 cores.

I have done some googling and have found a recommended grid spacing of (∆x+, ∆y+, ∆z+). From literature, the grid spacing for a wall-resolved LES is the following:

50 ≤ ∆x + ≤ 150, ∆y + wall < 1, 15 ≤ ∆z + ≤ 40

My case setup is based on the motorbike LES tutorial in OpenFOAM and since wall functions are used, I changed the range of ∆y +wall to be 30 ≤ ∆y +wall ≤ 300 instead. This value of ∆y is also applied throughout the entire domain. As the formulas of ∆x + and ∆z + are similar ∆y+wall, I simply rescaled the ∆x and ∆z as needed to obtain the correct range.

I have set up a computational domain following the ERCOFTAC guidelines (8L x 5W x 5H) with the Ahmed geometry situated at 2L from the inlet. I am using a trimmer mesh to reduce the amount of non-orthogonality and I can control the grid spacing by defining the anisotropic sizes of the trimmer cells as part of the Volumetric Controls. i.e. The above LES grid spacing is set only for a volume of refinement that covers the Ahmed Body and some length of the domain behind it.

I have attempted to mesh the domain and it took me hours before the software froze during the “optimizing mesh” portion of the operation. The mesh had a density of 62 million cells in total.

My questions are:
• Is my methodology even correct? I have read alternative methods involving the creation of field functions to obtain the ratio of Resolved Kinetic Energy/Total Kinetic Energy and performing grid refinements to ensure that it is >0.8 throughout the domain (a grid spacing that allows at least 80% of the kinetic energy to be resolved by the LES instead of modelled by its subgrid-scale model)
• If the LES grid spacing specified above is a correct methodology, does it have to be applied throughout the entire domain or can be it confined to volumes of refinement around the geometry and in the wake and have a coarse mesh further outfield. What is the effect of either of these configurations?
• Is the trimmer mesh the only mesh where the anisotropic size can be specified? Are there better mesh settings to be chosen for an LES

I have attached screenshots of my computational domain in STAR-CCM+ as well as my selected mesh settings. Note the mesh is very coarse as the mesh operation with the LES grid spacings above did not complete.

Sources:
• For grid spacing
https://www.researchgate.net/publica...uture_Research