CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   OpenFOAM (http://www.cfd-online.com/Forums/openfoam/)
-   -   Ahmed Body (http://www.cfd-online.com/Forums/openfoam/78673-ahmed-body.html)

andrea.pasquali July 29, 2010 08:09

Ahmed Body
 
Hello,
I'd like to study many turbulence models RAS and LES (and discretization schemes) with Ahmed Body 25 slant.

My mesh from ANSA (y+ not so good min 14 max 28 average 21, I need to improve it).

My first test is steady analysis with k omega SST; my BCs are
U: inlet fixedValue (40 0 0); outlet inletOutlet; ahmed fixedValue (0 0 0); tunnel slip; floor fixedValue (40 0 0)
p: inlet zeroGradient; outlet fixedValue 0; ahmed zeroGradient; tunnel slip; floor zeroGradient
k: inlet fixedValue 6; outlet inletOutlet; ahmed kqRWallFunction 0; tunnel slip; floor kqRWallFunction 0
omega: inlet fixedValue 77; outlet inletOutlet; ahmed omegaWallFunction 0; tunnel slip; floor omegaWallFunction 0

Solvers
p: GAMG
U, k, omega: smoothSolver

Schemes
gradScheme Gauss linear
div(phi,U) Gauss upwind
div(phi,k) Gauss upwind
div(phi,omega) Gauss upwind
interpolationSchemes lienar
snGradSchemes limited 0.333

Reference Area = 0.115 m^2 (body + legs).
Cd calculated from forceCoeffs is 0.45 (VS experimental 0.29)

I think is too high, this could be because of my y+ or my BCs or schemes?


Thanks for any replies

Andrea

scott July 29, 2010 09:56

Hi Andrea,

Y+ is important, but it is also very important in simulations such as this with adverse pressure gradient and separated flow to have a fine mesh resolution within the wake. You could post some images of you mesh and we could comment. You need this fine mesh to capture the vortical flow and also separation.

Are you comparing to data with a moving ground or fixed ground? This will have an impact on the result.

Are you running a Symmetry model or full model and are you results steady state or Statistical Averaged Transient?

I know from my experience that you will get vastly different results from the different turbulence models for this type of geometry, so keep trying the others and you will get a feel for what is best.

I am only starting out with Openfoam having come from a FLUENT background, so I am not the best person to comment on your settings.

Regards,

Scott

andrea.pasquali July 29, 2010 11:34

2 Attachment(s)
Hi Scott,

you can see my mesh in the pictures attached

for the ground I also tried fixed ground, moving ground and slip condition but with similar results

my model is 3d and the first analysis I made are steady

I also tried k epsilon, RNG k epsilon, realizable k epsilon but with similar results


Thanks

Andrea

sega July 30, 2010 03:05

Sorry to say that. This mesh looks like a mess.

Did you try to do a mesh with hexahedral cells?
Maybe you can get a good mesh with snappyHexMesh?

scott July 30, 2010 03:49

Hi Andrea,

I too feel that you would need to generate a better quality mesh for better results. You have way too coarse a mesh upstream of the body, and on the area that you have zoomed in on, it appears that you have boundary layer type mesh on the curved section, but as soon as it goes flat on top the mesh becomes quite coarse again.

I would try creating a mesh in snappyHexMesh from an STL file and I would also put a boundary layer on the ground.

The idea of the wake is good, but might need to be slightly "cleaner".

A mesh that looks good is a mesh that will solve well!

Regards,

Scott

allensoo November 16, 2010 14:21

Hello, Does anyone can provide Gambit Tutorial file for the Ahmed Body 3d case? I am struggling with the 3d case as it is sort of my first attempt on 3d case.

Thank you.

Lukasz November 16, 2010 15:11

FYI

Together with Ensys we run OF AhmedBody3D test (2.2 Million cells) on multi-gpu system and reached the following acceleration on two multi-gpu systems: 1. 4xTesla C2050 and 2. GTX295+GTX460. On system 1. we had x27 speed-up and on system 2. x4 speed-up comparing to Intel Quad Core (measured for 10 iterations).

vkrastev November 17, 2010 05:41

Quote:

Originally Posted by andrea.pasquali (Post 269357)
My mesh from ANSA (y+ not so good min 14 max 28 average 21, I need to improve it).

The problem is not in the y+: if you are using wall functions, values above 30 are preferable (wall functions are based on equilibrium assumptions which are reasonable only in the logarithmic region of the b. l., e. g. y+>30). The problem is that the overall mesh quality seems to be quite horrible (ANSA can do far better jobs). Can you post your checkMesh report?

Regards

V.

andrea.pasquali November 17, 2010 05:56

Hi Vesselin,
I fixed my problems some mounths ago (improving mesh with ANSA and using HEXA cells).
My last result Ahmed 25 slant steady RAS k-omega is:
Cd 0.299886 (VS Exp. 0.299)
Cl 0.360877 (VS Exp. 0.345)
You can see: http://www.cfd-online.com/Forums/ope...ynamics-5.html

As soon as I found "free time" I'd like to test unsteady analysis with both RAS and LES

Thanks however

Best Regards

Andrea

vkrastev November 17, 2010 07:57

Quote:

Originally Posted by andrea.pasquali (Post 283752)
Hi Vesselin,
I fixed my problems some mounths ago (improving mesh with ANSA and using HEXA cells).
My last result Ahmed 25 slant steady RAS k-omega is:
Cd 0.299886 (VS Exp. 0.299)
Cl 0.360877 (VS Exp. 0.345)
You can see: http://www.cfd-online.com/Forums/ope...ynamics-5.html

As soon as I found "free time" I'd like to test unsteady analysis with both RAS and LES

Thanks however

Best Regards

Andrea

Sorry, I haven't noticed that actually this tread was started so far ago...:D
However, nice results! Now I'm working on the Ahmed body too, but with tetra/prisms unstructured mesh, and I have some troubles with using higher order discretization schemes (I mean higher than the upwind choice) for the divSchemes (div(phi,U), div(phi,k) and so on). What is your fvSchemes setting (I mean the one that gave you the results posted above with the hex mesh)?

Thank you

Best regards

V.


All times are GMT -4. The time now is 20:09.