[vega]

Vincent, BastiL,

I just sent you the whole model via yousendit.com.
Just go to the link that you will get on email accounts in the profile and download the file.

thanks

--
Rajneesh

[vinz]

Thanks a lot Rajneesh, I received it.

I just had a look at your mesh and it's totaly different from mine, so it will be very interesting.

I'm going to run it to see what I get, and I'll post my results.

Regards,

Vincent

[bastil]

I can not try 1.4 at the moment, I am using simpleFoam 1.4.1 and I get:

Starting time loop

Time = 1

DILUPBiCG: Solving for Ux, Initial residual = 1, Final residual = 0.0254472, No Iterations 1
DILUPBiCG: Solving for Uy, Initial residual = 1, Final residual = 0.00714854, No Iterations 1
DILUPBiCG: Solving for Uz, Initial residual = 1, Final residual = 0.0269422, No Iterations 1
DICPCG: Solving for p, Initial residual = 1, Final residual = 0.00938556, No Iterations 13
time step continuity errors : sum local = 38.8664, global = -1.26956, cumulative = -1.26956
DILUPBiCG: Solving for epsilon, Initial residual = 0.999636, Final residual = 0.0488436, No Iterations 1
bounding epsilon, min: -316.208 max: 3.94271e+06 average: 149.224
DILUPBiCG: Solving for k, Initial residual = 1, Final residual = 7.80029e-05, No Iterations 1
ExecutionTime = 215.08 s ClockTime = 215 s

...

Time = 4

DILUPBiCG: Solving for Ux, Initial residual = 0.123077, Final residual = 0.00151918, No Iterations 1
DILUPBiCG: Solving for Uy, Initial residual = 0.291113, Final residual = 0.00156297, No Iterations 1
DILUPBiCG: Solving for Uz, Initial residual = 0.539444, Final residual = 0.000117586, No Iterations 1
DICPCG: Solving for p, Initial residual = 0.646448, Final residual = 0.00622016, No Iterations 59
time step continuity errors : sum local = 17.5265, global = 0.715884, cumulative = 0.36078
DILUPBiCG: Solving for epsilon, Initial residual = 0.0334807, Final residual = 4.11484e-05, No Iterations 1
DILUPBiCG: Solving for k, Initial residual = 0.233191, Final residual = 0.000210396, No Iterations 1
ExecutionTime = 870.11 s ClockTime = 870 s

....

Time = 9

DILUPBiCG: Solving for Ux, Initial residual = 0.147892, Final residual = 1.23908e-07, No Iterations 1
DILUPBiCG: Solving for Uy, Initial residual = 0.414654, Final residual = 1.23514e-06, No Iterations 1
DILUPBiCG: Solving for Uz, Initial residual = 0.51769, Final residual = 1.05291e-06, No Iterations 1
DICPCG: Solving for p, Initial residual = 0.597334, Final residual = 0.00588535, No Iterations 23
time step continuity errors : sum local = 7.93123e+08, global = 3.50323, cumulative = 12.0621
DILUPBiCG: Solving for epsilon, Initial residual = 0.0486521, Final residual = 0.00112071, No Iterations 1
bounding epsilon, min: -1.14166e+14 max: 1.22753e+25 average: 2.34251e+19
DILUPBiCG: Solving for k, Initial residual = 0.611983, Final residual = 9.01935e-06, No Iterations 1
bounding k, min: -3.57866e+09 max: 3.03393e+19 average: 3.13484e+13
ExecutionTime = 1790.34 s ClockTime = 1791 s

...

Time = 12

DILUPBiCG: Solving for Ux, Initial residual = 0.00277457, Final residual = 3.62285e-05, No Iterations 1
DILUPBiCG: Solving for Uy, Initial residual = 0.00162991, Final residual = 4.10471e-05, No Iterations 1
DILUPBiCG: Solving for Uz, Initial residual = 0.00264718, Final residual = 4.61501e-05, No Iterations 1
#0 Foam::error::printStack(Foam:stream&) in "/home/brblo/OpenFOAM/OpenFOAM-1.4.1/lib/linux64GccDPOpt/libOpenFOAM.so"
#1 Foam::sigFpe::sigFpeHandler(int) in "/home/brblo/OpenFOAM/OpenFOAM-1.4.1/lib/linux64GccDPOpt/libOpenFOAM.so"
#2 __restore_rt in "/lib64/tls/libc.so.6"
#3 Foam::DICPreconditioner::calcReciprocalD(Foam::Fie ld<double>&, Foam::lduMatrix const&) in "/home/brblo/OpenFOAM/OpenFOAM-1.4.1/lib/linux64GccDPOpt/libOpenFOAM.so"
#4 Foam::DICPreconditioner::DICPreconditioner(Foam::l duMatrix::solver const&, Foam::Istream&) in "/home/brblo/OpenFOAM/OpenFOAM-1.4.1/lib/linux64GccDPOpt/libOpenFOAM.so"
#5 Foam::lduMatrix::preconditioner::addsymMatrixConst ructorToTable<foam::dicprecond itioner>::New(Foam::lduMatrix::solver const&, Foam::Istream&) in "/home/brblo/OpenFOAM/OpenFOAM-1.4.1/lib/linux64GccDPOpt/libOpenFOAM.so"
#6 Foam::lduMatrix::preconditioner::New(Foam::lduMatr ix::solver const&, Foam::Istream&) in "/home/brblo/OpenFOAM/OpenFOAM-1.4.1/lib/linux64GccDPOpt/libOpenFOAM.so"
#7 Foam::PCG::solve(Foam::Field<double>&, Foam::Field<double> const&, unsigned char) const in "/home/brblo/OpenFOAM/OpenFOAM-1.4.1/lib/linux64GccDPOpt/libOpenFOAM.so"
#8 Foam::fvMatrix<double>::solve(Foam::Istream&) in "/home/brblo/OpenFOAM/OpenFOAM-1.4.1/lib/linux64GccDPOpt/libfiniteVolume.so"
#9 main in "/home/brblo/OpenFOAM/OpenFOAM-1.4.1/applications/bin/linux64GccDPOpt/simpleFoam "
#10 __libc_start_main in "/lib64/tls/libc.so.6"
#11 Foam::regIOobject::readIfModified() in "/home/brblo/OpenFOAM/OpenFOAM-1.4.1/applications/bin/linux64GccDPOpt/simpleFoam "
Floating point exception


I tried to play around with relaxation factors and nonorthogonal correcors without success. I would be very happy if a FOAM expert could give me some hints. As mentioned before checkMesh reports some errors but other solvers run on this mesh without problems.

Thanks

[vinz]

Hi Bastil,

As I allready said , I got the same problems.

What I didn't mentioned earlier is that my mesh was also running fine with other solvers but not with OpenFOAM (1.4.1 at least). So I guess it's really a problem of stability with respect to cell shape.

By the way, is your mesh refined a lot at the wall? Because I also noticed that simpleFoam doesn't really like very refined meshes at walls.

I would also be interested in openfoam expert solution toughts on the subject.

Vincent

[fra76]

At first, you can try to start from the potential solution.
Then, you can try switching off any turbulence model, at the beginning.
After a few iteration, switch it on again, starting from the laminar solution.

Good luck!
Francesco

[bastil]

Vincent: I have two prims layers near the walls. I do not know if this is "very refined". But in every case it is a mesh suitable for turbulence model + wall function.
Francesco: Thanks for these hints. I will try to turn off turbulence first, maybe this will help.

Bastil

[vega]

BastiL,

I had better success with GAMG solver for pressure. See it in the setup files I posted in this thread.
Also, see if your initial k and eps values are reasonable
--
R

[bastil]

Rajneesh,

I agree. I get better results with the settings from Hrv:

http://www.cfd-online.com/OpenFOAM_D...es/1/4245.html

Nevertheles in 1.4.1 use GAMG instead of AMG. I have 68 Timesteps so far and it looks much better. I will keep you up.

[vinz]

Hi everybody,

did someone tried to change the liftDrag utility to account for turbulence. I mean replacing dragCoefficient(...) by turbDragCoefficient(...).
I face a stupid problem. For the turbDragCoefficient computation, one need to pass the turbulence model since the first function argument is:
const autoPtr<turbulencemodel>& turbulence
my question is what should I add to liftDrag.C to pass this turbulence model?

Thanks in advance.

Vincent

[ariorus]

I had written time ago this application which computed at runtime the forces on a body. It is not too efficient but it should work.
It runs by this command:

ssimpleFoam <root> <case> "(patch1 patch2 patch3)"

where patch1-3 are the patches of the body you want to calculate the force on.

I run it in version 1.3 but it I verified it works with 1.4 as well (I don't have installed 1.4.1).

Hope this is useful.

ssimpleFoam.tgz

to Rajneesh: I had a quick look at the setting of the Amhmed body you posted. How did you choose turbulent parameters? I mean k and epsilon at inlet.
Why did you start by zero velocity as initial condition?

[vega]

Rosario,

Initially I had picked values corresponding to the turbulence intensity of 0.6 and length scale of 0.02. It gives k and eps of order of 0.1. After that I played with several other values. Simulation does blow up if these values are too much out of reasonable limits.

There is no particular reason for starting from 0. Just thats the way I have been doing at my work for Fluent simulations. This choice seems safest even though it may take little longer to converge for some cases.

[ariorus]

Thank you Rajneesh for your answer.

Actually I was thinking to use an higher value for turbulence intensity, but I don't know the details of the ahmed body experiment.

Are your result affected much by the turbulence set at inlet? (I'm just curious).

[vinz]

Hi Rajneesh,

I come back to you after I run your case.
I didn't change anything and the computation went fine to the end.
Now I would like to compute the lift and drag and I get some strange results.
Here is what I get from my liftDrag utility:

Create mesh for time = 1500

Time = 1500
Reading U

Reading p

Reading/calculating face flux field phi

Selecting incompressible transport model Newtonian
Selecting turbulence model RNGkEpsilon

Inlet velocity: (60 0 0)
pressure Coefficient: 0.00540718 viscous Coefficient: 0.000145663 turb Coefficient: -0.00139343
Wall patch 3 named a.forebody :
Reference area: 1 Reference length: 0.182 Total Turbulent drag coefficient: 0.00415941 Lift coefficient: (0 -0.0314594 0.00310949)

pressure Coefficient: -1.42641e-12 viscous Coefficient: 0.000213444 turb Coefficient: -0.000637851
Wall patch 4 named a.side :
Reference area: 1 Reference length: 0.862199 Total Turbulent drag coefficient: -0.000424407 Lift coefficient: (0 -0.0411145 8.50987e-07)

pressure Coefficient: -2.85131e-07 viscous Coefficient: 0.000111487 turb Coefficient: -0.000347901
Wall patch 5 named a.top :
Reference area: 1 Reference length: 0.645392 Total Turbulent drag coefficient: -0.000236699 Lift coefficient: (0 3.45158e-06 0.0261528)

pressure Coefficient: 0 viscous Coefficient: 0.000150484 turb Coefficient: -0.00048029
Wall patch 6 named a.bot :
Reference area: 1 Reference length: 0.862 Total Turbulent drag coefficient: -0.000329805 Lift coefficient: (0 -1.36859e-07 -0.0361365)

pressure Coefficient: 0.00288352 viscous Coefficient: 3.65491e-05 turb Coefficient: -4.66086e-05
Wall patch 7 named a.baklite :
Reference area: 1 Reference length: 0.216612 Total Turbulent drag coefficient: 0.00287346 Lift coefficient: (0 1.51287e-06 0.0129836)

pressure Coefficient: 0.0133659 viscous Coefficient: 3.91805e-07 turb Coefficient: 0.000242045
Wall patch 8 named a.rear :
Reference area: 1 Reference length: 0 Total Turbulent drag coefficient: 0.0136083 Lift coefficient: (0 -1.37909e-05 5.13891e-06)


One thing to notice is that I prescribe a reference area of 1m². So the numbers have to be divided by the real frontal area of the ahmed body which in your case is 0.5x0.288x0.389 (because it's a half body,am I right?).
So, to obtain the pressure coefficient for instance, I sum all pressure coefficients of each body part and divide by the number explained above. Then I get a rounded pressure coefficient of 0.387 which is exactly twice as yours. The same can be done with the viscous coefficient and I obtain 0.012 which is also twice as yours.

My questions are then, are your drag coefficients computed using the half body frontal area or the whole frontal area? are my computations wrong? Where is my mistake?

Thans a lot for your help.

Regards,

Vincent

[vega]

Vincent,

Its quite possible that I am using area for the
full body. We have to use the half model frontal area, exactly as you mentioned. I will check it from home tonight and let you know

I had modifed the computeForces.H program by hardcoding the area
value (full body area = 0.1120). I have been lately using full model as well to see if the symmetry assumption makes any difference. The Fluent setup that predicted Cd exactly for 12.5 deg body is way off for this 30 deg model.

BTW if you are interested here is full mesh model for 30 Deg Ahmad body.It should be available on this link for next two-three days.

http://download.yousendit.com/09F309E86E2932AB

Rosario: I was playing with various turbulence conditions to see if the solution converges or blows up. I never really ran it long enough to get final values.

All: I am still trying to figure out best schemes for the second order accuracy for external aero simulations. Please post your observations.
--
Rajneesh

[vinz]

Ok, so I guess the drag coefficient you gave before in this thread should be multiplied by 2.

This actually closer to the results I get by using my own mesh.

Unfortunately this is not a good news since it is twice as large as ahmed results. And this factor 2 is very disturbing.

I'm going to download you 30° mesh and run it to see if we obtain the same results (i.e. drag coefficient twice as large as ahmed for a 30° slant angle).

Anyway thanks a lot for sharing your work.

Vincent

[vega]

Yes, confirmed I was using area of the full model.
So Sorry for the confusion.

By changing to second order schemes, Pressure Cd does come down to 0.320 (with correct area) but it is still quite far from the experimental values.

--
R

[bastil]

Hi all,

I started working on the half model from Rajneesh yesterday.First calculations went fine. Concerning the differences we should consider the following (I do not have th evalues of th ehmed experiment...)

1.) Make sure drag/lift is computed correct. Which tools are you using? I can not find any of the draglift tools in OF 1.4.1, are they? Where can I get them?

2.) Get rid of mesh-dependencies concerning near wall resolution and farfield. I guess mesh from Rajneesh is nice but hexcore mesh would be nicer. Also wake and stagnation point refinements are the key to validation.

3.) Turbulence modelling is another key.

BastiL

[vinz]

Hi everybody,

So, unfortunatly I was right, the drag coefficients were wrong. But as you said, Rajneesh, by changing fvSchemes we get closer to the right values. Actually I changed the schemes to these ones:

ddtSchemes
{
default steadyState;
}

gradSchemes
{
default fourth;
grad(p) fourth;
grad(U) fourth;
}

divSchemes
{
default none;
div(phi,U) Gauss SFCD;
div(phi,k) Gauss SFCD;
div(phi,epsilon) Gauss SFCD;
div(phi,R) Gauss SFCD;
div(R) Gauss SFCD;
div(phi,nuTilda) Gauss SFCD;
div((nuEff*dev(grad(U).T()))) Gauss linear;
}

laplacianSchemes
{
default none;
laplacian(nuEff,U) Gauss linear corrected;
laplacian((1|A(U)),p) Gauss linear corrected;
laplacian(DkEff,k) Gauss linear corrected;
laplacian(DepsilonEff,epsilon) Gauss linear corrected;
laplacian(DREff,R) Gauss linear corrected;
laplacian(DnuTildaEff,nuTilda) Gauss linear corrected;
}

interpolationSchemes
{
default linear;
interpolate(U) linear;
}

snGradSchemes
{
default corrected;
}


and I'm getting a value of 0.27 (using th right area) for the drag coefficient which is not so bad. I'm still playing with the schemes to find the best ones to use.

Then, we'll have to find the best turbulence model to see if we get better results as Bastil has mentioned.

For the liftDrag utility, I use one I found on this forum. Use the search command to find the thread, there is an explanation on how to install the liftDrag tool on previous version of OpenFOAM (and this also works with 1.4.1).

Vincent

[vega]

A good news I guess
Fluent simulation for the same mesh (0.63 Mill Cells) as used for OpenFoam predicts Cd of 0.309.
Now OF results do not look that bad. Fluent model
of ~3 Mill cells gives same result as test. I will
try that on my machine sometime soon.

From now on, I will be uploading interesting results, flow vis comparisons to the googlepage site
linked in my profile.

[vinz]

Actually, yes, it looks like a good news.
I didn't understand how OpenFOAM was so far from Fluent, but the mesh difference is certainly one of the reasons.

I'm waiting forward to see what you get. I'll keep you updated if I find better results with other shemes.