[Artur]

Hi Foamers,

I am trying to replicate the zero pressure gradient flat plate results by Wieghardt and TIllmann (http://www.grc.nasa.gov/WWW/wind/val...rb/fpturb.html) using LES. I am doing this to improve the results I get for my airfoil simulations.

The plate is 5m long, the domain starts 1.25m upstream of the leading edge, ends 2.5m dowstream and its height is 1.5m (approx. 15 BL thicknesses at the downstream end). I have also tried a 2.5 x 5 x 5m domain and got very similar results. The width is 0.1m, with inflow speed of 33ms-1 (see attachment for an overview of the d domain and locations of the probing stations). My mesh is designed to be y+ < 1 for resolved and y+<30 for wall-modelled meshes, with z+ = 190 and x+ = 600 (600 elements stream-wise and 40 span-wise). I used blockMesh to create the grid, it comprises of 3 regions in the wall-normal direction, one for inner, one for outer BL and one for free-stream.

I have a fixedValue inlet on U, outlet is set as convectiveOutlet. Pressure is zeroGradient at the inlet and fixedValue 0 at the outlet. The span-normal patches are set as cyclic, the rest of the BCs are slip.

I used filteredLinear 1 0 convection scheme and backward time scheme, GAMG solver for p and PBiCG for U, PISO algorithm with 2 inner loops and 2 non-orthogonal correctors.

I have first ran the case using k-omega SST RANS on several grids to choose an appropriate one for wall-modelled and resolved simulations. I then used these to initialise my LES runs (just the U and p fields).

I tried the Smagorinsky model with nutUSpaldingWallFunction applied to nuSgs and zeroGradient for the resolved mesh. Also, I tried the oneEquationEddyViscosity model with the wall function only. All of the setups were ran with both cubeRootVol and vanDriest LES deltas with similar results.

The problem is that I get much smaller boundary layer thickness all over the plate for all setups and my skin friction drag is largely underestimated, which wasn't the case for the RANS runs. I enclose a few plots for the vanDries delta runs (data was first time-averaged for each probe and then span-averaged at each x-location).

Any advice on how I can improve the results will be much appreciated as I'm starting to run out of ideas ... I have also uploaded a tared case for my Smagorinsky with wall-function simulation to show my entire setup.

All the best,

A

Edit: I am currently running Smagorinsky with limitedLinear 0.1 on both resolved and w-f meshes and it seems like they will give similar results to the rest of the simulations.

ZPG_fine_LES_clean.tar.gz

flatPlate_y_vs_Ux.jpg

flatPlate_U+_vs_y+.jpg

flatPlate_domainOverview.jpg

flatPlate_Cf_vs_Rex.jpg

[Artur]

Hi All,

I've re-run the Smagorinsky and one-equation eddy viscosity model simulations on a grid refined down to y+ 25, z+ 125 and x+ 340. I also ran a simulation using a dynamic mixed Smagorinsky model recommended by a friend. I also switched the convection scheme to pure linear hoping this will increase accuracy. The results may be seen below.

My intermediate conclusion is that the Smagorinsky and one equation models somehow treat the flow as laminar and thus yield the incorrect skin friction and boundary layer profiles.

The question is: why is this happening (I've quadruple checked all my settings like inflow speed, nu, etc.) and how can this be overcome? Does anyone think that imposing small velocity perturbations at the inlet to simulate the wind tunnel conditions more closely could solve the issue?

All the best,

A

flatPlate_Cf_vs_Rex_denseMesh.jpg

flatPlate_U+_vs_y+_denseMesh.jpg

flatPlate_y_vs_Ux_denseMesh.jpg