Getting too high Drag Coefficients
I've been running some test cases on simple geometry before jumping into more high detail modes, but have been getting wrong values for Cd on every case.
So fat i've tested the following cases:
Ahmed Body 12,5º Slant - RAS Spalart-Allmaras
Ahmed Body 12,5º Slant - RAS K-Epsilon
Ahmed Body 12,5º Slant - RAS K-Omega
Ahmed Body 12,5º Slant - RAS K-OmegaSST
Ahmed Body 12,5º Slant - LES oneEqEddy
Ahmed Body 12,5º Slant - LES Smagorinsky
All of these tests have been done with different mesh qualities, with and without surface Layers, with different domain sizes, with Hexahedra and tetrahedra meshes.
All of these cases resulted in Cd values between 70 and 130% higher than they should be.
I've also ran several cases with NACA0009 profiles at 0º , 5º , 7.5º and 10º attack angle, With and without Surface layers, With and without dense wake mesh, with different domain sizes and different mesh qualities, all using Spalart-Allmaras.
In all cases, Lift (Cl) errors where between 2% and 10%, however Drag (cd) errors where between 120% and 300%.
Has anyone ever encountered these problems? Am i doing something wrong? Where could the problem be?
The Ahmed body has been studied succesfully with OpenFOAM by a lot of people. You can have a look at this thread:
There you can also see some settings that people have used for their RAS computations. I am not sure about LES. I want to dive into that as well, but have not found the time to really do it well. But their are plenty of threads available that could help you with LES. Just search the forum.
i am having the same problem. I am using k-epsilon.
What I observe is that the y+ values has a big influence on the drag. which is also obvious, due to the skin friction.
With a proper numerical setup, you can easily obtain better results (for the Ahmed model) than the ones reported by you, at least for the more stable flow configurations (not too close to the critical slant angle of about 30°). Try for instance to use a prism-tetra mesh, with 3-5 layers around the body and with an y+ distribution matching (approximately) the range 30-150. Then run it with the simpleFoam solver, using the realizable k-epsilon model and standard wall functions for k, epsilon and nut. If nothing gets better this way, then you probably have to play a bit with the numerics (fvSchemes and fvSolution files): in any case, be sure that at least on the div(phi,U) convenction term you are using something better than the "brutal" first order Gauss upwind scheme, otherwise your solution will be far too diffusive and with too high drag values (resulting also in smaller wake structures behind the body).
Hope this helps
Some of my attempts with K-Epsilon, K-Omega and SpalartAllmaras had surface layers with proper Y+ values. For Ke and Ko i had Y+ in the 30-ish range, and for SA closte to 1.
I also run all my cases with SimpleFoam and use wall functions.
I didn't really know about realizable K-E, i just used plain K-Epsilon. Is it better?
As for the schemes, i was using Gauss upwind because i was adviced to use it to avoid boundings on epsilon and omega. What schemes should i use for better results?
Also, regarding initial and boundary conditions, what would be proper values for K and omega?
im using a turbulence intensity of 0.5% and a mean velocity of 70m/s but i dont know how to obtain the characteristic lenght of the free stream to better estimate omega.
About the div(phi,U) scheme, try to use this:
Gauss linearUpwindV cellMDLimited Gauss linear 1
About the characteristic turbulent lenght scale, you can estimate it as about 10% of the Body's characteristic lenght (for instance Body's height).
Thanks for the help. I ran a new set of tests. 4 meshes with diferent qualities running Spalart Allmaras, KoSST using upwind and KoSST with the above suggested div schemes. All of them running an Ahmed Body with 12.5º slant angle
Mesh1= 77.000 cells
Mesh2= 320.000 cells
Mesh3= 1.000.000 cells
Mesh4= 2.850.000 cells
All mesh are tetrahedral meshes with no surface layers
Kw-SST = K omega SST with upwind scheme
Kw-SSTv2 = K omega SST with Gauss linearUpwindV cellMDLimited Gauss linear 1 scheme
SaM = Spalart Allmaras with upwind scheme
Here are the results. (Experimental drag Coeff = 0.23)
For some reason, Kw-SSTv2 Mesh 2 and Mesh 4 failed with excessive continuity errors. Dont know why.
The closest i got to the experimental value of 0.23 was 0.249 with the Kw-SSTv2 Mesh 3 run with an error of 8.26%. Not bad considering the lack of layers.
Kw-SST and Spalart allmaras (with upwind scheme) runs where all over the place and pretty far away from the experimental value.
Should i stick with Kw-SST and Gauss linearUpwindV cellMDLimited Gauss linear 1 schemes for these sort of tests? Will surface Layers help me improve my results?
i wonder if you finally got the rigth results. i'm currently facing the same problem as you did some time ago, although i am using the 0º back slant model. anyway, i would like to know if you got well results or gave up the case, and if you could help me.
Thanks in advance!
After some testing on different models, mesh types and schemes, the best results i got for a threedimensional 12.5º slant "floating" ahmed body where obteined with:
> K-Omega SST model
> Gauss linearUpwindV cellMDLimited Gauss linear 1 scheme for div(phi,U)
> aprox 3KK tetrahedral cells, giving a fair share to the wake of the vehicle
Errors around 6-8%
Hope it helps.
firstly, i recommend you check your Aref, Uref, and so on.
it maybe occur because of these wrong parameter.
Hi. Please you you tell me what the bottom axis of the graph represents...
|All times are GMT -4. The time now is 07:27.|