LES: dynSmagorinsky does not match analytical y+/u+ profile

aylalisa · September 30, 2014, 11:46

Dear Foamers,

I've simulated an LES canal flow and used dynamicSmagorinsky (by Alberto Pasquale) with vanDriest damping function.

The dimensionless profiles y+/u+ don't match the analytical log-law. How can I improve these results?

The finer mesh gives a worse result. Why?
Mesh resolution:
Case510: $\Delta$ x+ ~ 18, $\Delta$ y+ ~ 1,6 - 18, $\Delta$ z+ ~ 10
Case511: $\Delta$ x+ ~ 13, $\Delta$ y+ ~ 1,4 - 15, $\Delta$ z+ ~ 7

I expected this resolution to be o.k. for LES.

Does dynamicSmagorinsky underestimate $\tau_{wall}$ ?

Is it correct to use vanDriest damping in combination with a dynamic SGS model?

Best regards,
Aylalisa

itchy · October 1, 2014, 08:24

Hi,

first:
In dynamic Smagorinsky you use Germano procedure to estimate the Smagorinsky constant. With this you get the y³-law of the wall for nuSgs (that is what you want). If you use van Driest with dynamic procedure you get unphyisical behaviour for nuSgs at the wall. So do never use van Driest with dynSmagorinsky

second:
post your case, then we can help you. Without your case file it is nearly impossible to find the error.

kind regards
Florian

itchy · October 1, 2014, 09:31

Hi again,

how do you calculate tau_w???

Do not calculate tau_w with the velocity gradient near the wall, because you use a inner
variable that depends on your variable! If you do this, then OF (and any other solver) will underpredict tau_w.

A better way is calculating tau_w or u_tau using the pressure gradient (fvOptions).

What kind of LES do you solve??? Near wall resolution LES (NWR) or near wall modelling LES (NWM). If you do NWR, then your y+ is to high.

kind regards
Florian

aylalisa · October 7, 2014, 11:02

Hi Florian,

thanks for your answer, sorry for late reply!

Quote:

how do you calculate tau_w???

Do not calculate tau_w with the velocity gradient near the wall, because you use a inner
variable that depends on your variable! If you do this, then OF (and any other solver) will underpredict tau_w.

A better way is calculating tau_w or u_tau using the pressure gradient (fvOptions).

uTau:

uTau.boundaryField()[patchi] =
sqrt
(
nuEff.boundaryField()[patchi]
*mag(U.boundaryField()[patchi].snGrad())
);

Is the inner variable "nuEff"?

The other way I'm aware of is to compute tau_w with the Stress Tensor Beff:
tau_w = unit_vector & Beff
I've expected that both methods should give the same result.
How can I use fvOptions file for computation of tau_w? I've assumed that this file allows the definition of source terms (if necessary), to avoid additional coding.
I've only added a momentumSource of type pressureGradientExplicitSource in fvOptions to keep up the flow.

Quote:

What kind of LES do you solve??? Near wall resolution LES (NWR) or near wall modelling LES (NWM). If you do NWR, then your y+ is to high.

LES with NWR. $\Delta$ y+ ~ 1,5 at the wall and y+ $_{average}$ =0,4, y+ $_{max}$ =0,6, y+ $_{min}$ =0,2 at wall 'bottom'. How small does y+ have to be for a LES with NWR on a single structured grid? y+ < 1 not small enough? My goal was to locate the first cell center in the inner layer of the boundary layer.
Additionally I've chosen nutUSpaldingWallFunction as wall function. For NWR it should not be necessary but could it cause a problem?

Quote:

first:
In dynamic Smagorinsky you use Germano procedure to estimate the Smagorinsky constant. With this you get the y³-law of the wall for nuSgs (that is what you want). If you use van Driest with dynamic procedure you get unphyisical behaviour for nuSgs at the wall. So do never use van Driest with dynSmagorinsky

o.k.!

Quote:

second:
post your case, then we can help you. Without your case file it is nearly impossible to find the error.

all files?

Code:

/*--------------------------------*- C++ -*----------------------------------*\
| =========                 |                                                 |
| \\      /  F ield         | OpenFOAM: The Open Source CFD Toolbox           |
|  \\    /   O peration     | Version:  2.2.1                                 |
|   \\  /    A nd           | Web:      www.OpenFOAM.org                      |
|    \\/     M anipulation  |                                                 |
\*---------------------------------------------------------------------------*/
FoamFile
{
    version     2.0;
    format      ascii;
    class       dictionary;
    location    "system";
    object      fvSchemes;
}
// * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * //

ddtSchemes
{
    //default         steadyState;
    //default         Euler;
  default    backward;
}

d2dt2Schemes
{
}

gradSchemes
{
    default         Gauss linear;
    
    //neu:
    grad(nuTilda)   cellLimited Gauss linear 1;
    grad(U)           cellLimited Gauss linear 1;
}

divSchemes
{
    default                    none;
    //div(phi,U)               bounded Gauss upwind;
    //div(phi,U)               bounded Gauss upwind;
    div(phi,U)                 Gauss LUST unlimitedGrad(U);
    //div(phi,k)                bounded Gauss upwind;
    div(phi,k)                  Gauss limitedLinear 1;
    div(phi,epsilon)          bounded Gauss upwind;
    div(phi,R)                  bounded Gauss upwind;
    div(R)                       Gauss linear;
    div((nuEff*dev(T(grad(U)))))     Gauss linear;
    //neu
    div(phi,nuTilda)             Gauss limitedLinear 1;
}

laplacianSchemes
{
    default         Gauss linear limited corrected 0.33;
    
    //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;
}

interpolationSchemes
{
    default         linear;
}

snGradSchemes
{
    //default         corrected;
    default         limited corrected 0.33;
}

fluxRequired
{
    default         no;
    p              ;
}

I've set upwind schemes as discretization schemes for:
div(phi,epsilon) bounded Gauss upwind;
div(phi,R) bounded Gauss upwind;
Could that cause additional numerical dissipation in my simulation?

Code:

/*--------------------------------*- C++ -*----------------------------------*\
| =========                 |                                                 |
| \\      /  F ield         | OpenFOAM: The Open Source CFD Toolbox           |
|  \\    /   O peration     | Version:  2.2.2                                 |
|   \\  /    A nd           | Web:      www.OpenFOAM.org                      |
|    \\/     M anipulation  |                                                 |
\*---------------------------------------------------------------------------*/
FoamFile
{
    version     2.0;
    format      ascii;
    class       dictionary;
    object      fvSolution;
}
// * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * //

solvers
{
    p
    {
        solver           GAMG;
        tolerance        1e-6;
        relTol           0.1;

        smoother         GaussSeidel;
        nPreSweeps       0;
        nPostSweeps      2;

        cacheAgglomeration true;

        nCellsInCoarsestLevel 50;//10;
        agglomerator     faceAreaPair;
        mergeLevels      1;
    };

    pFinal
    {
        solver           GAMG;
        tolerance        1e-6;
        relTol           0;

        smoother         GaussSeidel;
    //smoother         DICGaussSeidel;
        nPreSweeps       0;
        nPostSweeps      2;

        cacheAgglomeration true;

        nCellsInCoarsestLevel 50;//10;
        agglomerator     faceAreaPair;
        mergeLevels      1;
    };

    "(U|k)"
    {    
        solver           PBiCG;
        preconditioner   DILU;
          //solver     smoothSolver;
        //smoother     symGaussSeidel;    
    tolerance        1e-08;
        relTol           0;
    };

    "(U|k)Final"
    {
    $U;
        tolerance        1e-08;
        relTol           0;
    };

    B
    {
        solver           PBiCG;
        preconditioner   DILU;
        tolerance        1e-07;
        relTol           0;
    };
    
    /*
    nuTilda
    {
        solver           PBiCG;
        preconditioner   DILU;
        tolerance        1e-07;
        relTol           0;
    };
    */
}

PISO
{
    nCorrectors             2;
    nNonOrthogonalCorrectors     1;
}

PIMPLE
{
    //nOuterCorrectors        1;  
    nCorrectors             2;
    nNonOrthogonalCorrectors     1;
    
    pRefCell                0;
    pRefValue               0;
}

relaxationFactors
{
    "U.*"               1;
    "nuTilda.*"         1;
}


// ************************************************************************* //

Best regards,
Lisa

itchy · October 8, 2014, 03:19

Hi Lisa,

uTau:
The inner variable is the velocity (mean velocity). In your first approach you use U to calculate utau. And then you use utau to scale U. With this you get uncertainties in utau. Not a good way.
Another possibility is to calculate utau with the gradientP. You can extract gradP of your logfile "pressure gradient = ..." and take the mean of it. Then for channel-flow you get:
utau = sqrt(0.5*h*gradP),
where h is the height of your channel.

Grid-resolution:
your grid should be fine enough. Sorry, I misunderstood your specifications.

fvSchemes:
I would use Gauss linear, Gauss linear corrected. But this should not change so much.

fvSolution:
Do you solve B and k???? I don't know the dynSmagorinsky-code of Alberto Pasquale.
You should insert a residual control, smth. like this:
PIMPLE
{
nOuterCorrectors 10;
nCorrectors 1;
nNonOrthogonalCorrectors 3;
pRefCell 1001;
pRefValue 0;
residualControl
{
U
{
relTol 0;
tolerance 1e-6;
}
p
{
relTol 0;
tolerance 1e-6;
}
}
}

Is your case statistical stationary when you begin your averaging??

Is your avaraging long enough??

Do you have proper reference data to check accuracy??

I have never done a LES for channel, only for pipe (but with my own dynSmagorinsky-code). If you don't get it managed with the dynSmagorinsky-code of Alberto Pasquale I can send you my code.

Best regards
Florian

aylalisa · October 8, 2014, 07:09

Hi Florian,

uTau:
I will check and see how curves change!

Grid-resolution:
Especially the fact, that the coarser grid of both simulations gives seemingly the better result, is astonishing. Could this behaviour point to the problem?

fvSchemes/fvSolution:
I'll change and try out.

statistical stationary:
I start averaging after 150 flows through time. Averaging is done for 200 flow throughs.

Quote:

Do you have proper reference data to check accuracy??

Do u refer to one-point/ two-point correlations etc.?

I'll try out all these changes and start simulations again. It will take some time, but I'll post results! If it does not work it would be great if I can try your code.
Thank's!

Best regards,
Lisa

itchy · October 11, 2014, 07:47

Hi Lisa,

averaging time that the flow passes 200 times the domain should be enough for a stable mean value. For the variances pherhaps not.

I mean some experimental data of channelflow?? Is there some abnormal behaviour concerning log-law for channelflow as a function of Re?? In pipeflow it is.

With proper reference data I mean some experimental data of exactly your problem?? e.g experimental data of u+ vs. y+ .

Do you have some result??

kind regards
Florian

aylalisa · October 12, 2014, 14:01

Hi Florian,

I was busy at work and completely lag my ToDo list. Tomorrow I can start simulation. It will take some time because I still work on a stand-alone work station. If u

is the problem I must recompute that value and correct curves!

proper reference data: I've been searching the net and found data sources
like http://torroja.dmt.upm.es/turbdata/agard/ ()and http://turbulence.ices.utexas.edu/da.../chan590/pdfs/ . I am not yet clear about how to use this data. These sources contain DNS canal data comparable to my case (Re $_\tau$ ~ 400).

I've read papers and literature like the work from Jochen Fröhlich who shows y+/u+ for canal flow (Re $_\tau$ ~590). Till now I did not find any hint that the analytical profile shows a different behavior in a canal flow. Though this assumption could be wrong!

To see if the simulation converges a statistical steady state I've plotted the resolved kinetic energy and the turbulent kinetic energy (case511, other is very similar --> attached pic) for start phase and averaging phase.
Additionally you see the u $^+_{rms}$ , v $^+_{rms}$ , w $^+_{rms}$ .

u $_{tau}$ :
wallShearStress.C for RANS computes wSS like

Code:

wallShearStress.boundaryField()[patchI] =
        (
           -mesh.Sf().boundaryField()[patchI]
           /mesh.magSf().boundaryField()[patchI]
        ) & Reff.boundaryField()[patchI];

where Reff is the deviatoric part of shear stress tensor resulting from chosen RAS model. In the forum it was discussed that Reff could be replaced by Beff and then this code should compute the wSS for an LES case (and finally u $_{tau}$ ). This seems to be a further method to compute u

.
Does B contain the sgs stresses computed by the chosen LES model? Where is the difference between tensor fields B and UPrime2Mean?

answer to last question: B -> unresolved stresses (sgs model), UPrime2Mean (var(U)) -> resolved normal stresses ???

Best regards,
Lisa

huangxianbei · October 19, 2014, 08:29

Hi, if a turbulence model is used, the profile may goes upon the analytical profile. I did a similar case with LES before and got similar results as yours. If no turbulence model is used ,then they are commute.

I have a question about the way to calculate the wallShearStress
tau_wall=rho*nu*gradU;
while in the wallShearStressLES, it's calculated as unit_wallnormal &Beff
In my case of a 90 bend duct, the two results deviates about 30-100 times, what's your opinion about that?

aylalisa · October 21, 2014, 12:21

Dear huangxianbei,

thank's for your hint according to the turbulence model!

My function for computing $u_\tau$ , using $\tau_{wall}$ according to unit_wallnormal &Beff ,does not work either.

Problems:
1) how to compute $\tau_{wall}$ ?
$\tau_{wall}$ = (mag((-Sfp/magSfp) & Beffp)*magSfp)

(-Sfp/magSfp) & Beffp should give an vector.
I thought, multiplied by the cell face area should give the desired shear stress

...

2) how to compute $u_\tau$ with $\tau_{wall}$ ?
Dividing it by $\rho$ ?

I haven't solved it yet.

Best regards,
Aylalisa

huangxianbei · October 22, 2014, 08:32

Quote:

Originally Posted by aylalisa

Dear huangxianbei,

thank's for your hint according to the turbulence model!

My function for computing $u_\tau$ , using $\tau_{wall}$ according to unit_wallnormal &Beff ,does not work either.

Problems:
1) how to compute $\tau_{wall}$ ?
$\tau_{wall}$ = (mag((-Sfp/magSfp) & Beffp)*magSfp)

(-Sfp/magSfp) & Beffp should give an vector.
I thought, multiplied by the cell face area should give the desired shear stress

...

2) how to compute $u_\tau$ with $\tau_{wall}$ ?
Dividing it by $\rho$ ?

I haven't solved it yet.

Best regards,
Aylalisa

Hi,Aylalisa:
If the fluid is incompressible, tau_wall should divide rho to get the right value. However, in my case, I compared the results with experiment's, the way that tau_wall=rho*nu*gradU is in the same order as the experiment's, while the wallShearStress didn't. I don't know why

So I'd prefer the way I metioned here.

Xianbei

aylalisa · October 24, 2014, 04:56

Hi Xianbei,

sorry that I ask again but I'd like to be sure to understand you correctly.

You compute $\tau_{wall} = \rho * \nu$ * gradU and this result matches the experimentally received $\tau_{wall}$ .
u $_{\tau computed}$ is equal to u $_{\tau measured}$ .

The problem is that $\tau_{wall}$ = unit_wallnormal &Beff is not equal to $\tau_{wall} = \rho * \nu$ * gradU (and experimental result)?
How do you compute $\tau_{wall}$ = unit_wallnormal &Beff exactly?

Aylalisa

huangxianbei · October 24, 2014, 05:22

Quote:

Originally Posted by aylalisa

Hi Xianbei,

sorry that I ask again but I'd like to be sure to understand you correctly.

You compute $\tau_{wall} = \rho * \nu$ * gradU and this result matches the experimentally received $\tau_{wall}$ .
u $_{\tau computed}$ is equal to u $_{\tau measured}$ .

The problem is that $\tau_{wall}$ = unit_wallnormal &Beff is not equal to $\tau_{wall} = \rho * \nu$ * gradU (and experimental result)?
How do you compute $\tau_{wall}$ = unit_wallnormal &Beff exactly?

Aylalisa

Hi,
I use the wallShearStressLES http://www.cfd-online.com/Forums/ope...ulation-2.html

You can see from the code that it's computed using unit_wallnormal &Beff similar to the case in RANS with wallShearStress utility
Xianbei

aylalisa · October 24, 2014, 07:44

I started from this thread too but I am still looking for the reason why the result does not agree with $\tau = \rho * \nu * gradU$ :-(

aylalisa · October 27, 2014, 13:55

Hi huangxianbei,

I've computed $\tau_{wall}$ and $u_\tau$ by the different ways.

$u_\tau$ = sqrt( $\nu_{eff}$ * snGrad(U))
$u_\tau$ = sqrt(mag((-Sfp/magSfp) & $B_{eff}$ ))
$\rho$ is not needed.

I get different results if I compute $\tau_{wall}$ with the different methods, the difference is not as big as in your case:

Code:

wSS_gradU.boundaryField()[patchi] = 
          nuEff.boundaryField()[patchi] 
        * mag(U.boundaryField()[patchi].snGrad());

Code:

ssp = (-Sfp/magSfp) & Reffp

I abstain from multiplication with $\rho$ .

I don't know how the values develop but these are the values in the beginning of calculation:

wSS (snGrad()) = 0.00103336
wSS (devReff()) = 0.000889885

Though I'd like to know why there is a difference, because $u_\tau$ results are very similar.

Code:

uTauLES uTauLES output:
    patch bottom u_tau : min = 0.0135431, max = 0.0430144, average = 0.0315161
utau_Beff_LES utau_Beff_LES output:
    patch bottom u_tau : min = 0.0135416, max = 0.0430147, average = 0.0315165

Aylalisa

huangxianbei · October 27, 2014, 20:04

Quote:

Originally Posted by aylalisa

Hi huangxianbei,

I've computed $\tau_{wall}$ and $u_\tau$ by the different ways.

$u_\tau$ = sqrt( $\nu_{eff}$ * snGrad(U))
$u_\tau$ = sqrt(mag((-Sfp/magSfp) & $B_{eff}$ ))
$\rho$ is not needed.

I get different results if I compute $\tau_{wall}$ with the different methods, the difference is not as big as in your case:

Code:

wSS_gradU.boundaryField()[patchi] = 
          nuEff.boundaryField()[patchi] 
        * mag(U.boundaryField()[patchi].snGrad());

Code:

ssp = (-Sfp/magSfp) & Reffp

I abstain from multiplication with $\rho$ .

I don't know how the values develop but these are the values in the beginning of calculation:

wSS (snGrad()) = 0.00103336
wSS (devReff()) = 0.000889885

Though I'd like to know why there is a difference, because $u_\tau$ results are very similar.

Code:

uTauLES uTauLES output:
    patch bottom u_tau : min = 0.0135431, max = 0.0430144, average = 0.0315161
utau_Beff_LES utau_Beff_LES output:
    patch bottom u_tau : min = 0.0135416, max = 0.0430147, average = 0.0315165

Aylalisa

Hi,Aylalisa:
Why it seems that Reff can be used to calculate the LES case? As you wrote in the thread, you use Reff. Isn't it a RANS variant instead of LES? The latter 2 results are close enough, is it the same case as the former?

Xianbei

aylalisa · October 28, 2014, 03:17

Hi Xianbei,

to figure out $R_{eff}$ for LES:
I've choosen one of the turbulence models, for example oneEqEddy.C.
Check the source code, line 65 --> Function call GenEddyVisc()
Then check the file GenEddyVisc.C (in my case for incompressible fluids), line 92:

Code:

return -nuEff()*dev(twoSymm(fvc::grad(U())));

This line seems to be $\tau_{ij}^r = - 2 * \nu_{eff} * S_{ij}$ .
$S_{ij}$ can be computed from grad(U()), it is the symmetric part of the deviatoric part of tensor grad(U()). The entity of that expression is

, not yet a shear stress (I think for incompressible fluids it is better to pass the multiplication with $\rho$ ). Apart from multiplication with $\rho$ you have the stress tensor coming from the model. Have a look at LESModel.H, line 218, there you find nuEff().

Aylalisa

huangxianbei · October 28, 2014, 04:52

Quote:

Originally Posted by aylalisa

Hi Xianbei,

to figure out $R_{eff}$ for LES:
I've choosen one of the turbulence models, for example oneEqEddy.C.
Check the source code, line 65 --> Function call GenEddyVisc()
Then check the file GenEddyVisc.C (in my case for incompressible fluids), line 92:

Code:

return -nuEff()*dev(twoSymm(fvc::grad(U())));

This line seems to be $\tau_{ij}^r = - 2 * \nu_{eff} * S_{ij}$ .
$S_{ij}$ can be computed from grad(U()), it is the symmetric part of the deviatoric part of tensor grad(U()). The entity of that expression is

, not yet a shear stress (I think for incompressible fluids it is better to pass the multiplication with $\rho$ ). Apart from multiplication with $\rho$ you have the stress tensor coming from the model. Have a look at LESModel.H, line 218, there you find nuEff().

Aylalisa

Hi,Aylalisa:
Thank you for your explaination. I'm a little confused about the Reff and Beff, is there any difference between them?

Xianbei

aylalisa · October 28, 2014, 14:47

Hi Xianbei,

Quote:

Hi, if a turbulence model is used, the profile may goes upon the analytical profile. I did a similar case with LES before and got similar results as yours. If no turbulence model is used ,then they are commute.

I don't understand. How can I run a LES without an LES turbulence model

Quote:

I'm a little confused about the Reff and Beff, is there any difference between them?

I hope anybody could support us here about the reason why defReff() is used for computation of the wallShearStress!

I think:
defReff()

Code:

return -nuEff()*dev(twoSymm(fvc::grad(U())))

extracts a shear stress tensor field computed with nuEff, without the isotropic part (I think you could also call it hydrostatic part). Since we want to compute the wallShearStress we neglect the isotropic part of the tensor but need to multiply with nuEff.

B()

Code:

return ((2.0/3.0)*I)*k() - nuSgs_*twoSymm(fvc::grad(U()));

seems to be:
$\tau_{ij} - \frac{\delta_{ij}}{3}\tau_{kk}$ = -2 $\nu_{T}\bar{S_{ij}}$
$\tau_{ij} = \frac{\delta_{ij}}{3}\tau_{kk} -2\nu_{T}\bar{S_{ij}}$
The isotropic part is computed from the kinetic energy ( $\frac{1}{2}u^2$ )that comes, depending on the model, from a transport equation. The result B(), including the isotropic part, computed with $\nu_T$ (eddy viscosity of residual motions) should be used for solving that equation.

Aylalisa

huangxianbei · October 28, 2014, 19:48

Quote:

Originally Posted by aylalisa

Hi Xianbei,

I don't understand. How can I run a LES without an LES turbulence model

Aylalisa

Hi,Aylaisa:
Sorry,that's actually not a LES case, but a case without turbulence model, say "under-resolved DNS" or something else.
According to your discription, it may be right to compute wallshearstress using B() instead of defReff. While for incompressible flow, dev_Sij=Sij as dui/dxi=0. So in that case, defReff can be used. Am I right?

Xianbei

October 27, 2014, 13:55		#15
aylalisa Senior Member Join Date: Nov 2012 Location: Bavaria Posts: 145 Rep Power: 13	Hi huangxianbei, I've computed $\tau_{wall}$ and $u_\tau$ by the different ways. $u_\tau$ = sqrt( $\nu_{eff}$ * snGrad(U)) $u_\tau$ = sqrt(mag((-Sfp/magSfp) & $B_{eff}$ )) $\rho$ is not needed. I get different results if I compute $\tau_{wall}$ with the different methods, the difference is not as big as in your case: Code: wSS_gradU.boundaryField()[patchi] = nuEff.boundaryField()[patchi] * mag(U.boundaryField()[patchi].snGrad()); Code: ssp = (-Sfp/magSfp) & Reffp I abstain from multiplication with $\rho$ . I don't know how the values develop but these are the values in the beginning of calculation: wSS (snGrad()) = 0.00103336 wSS (devReff()) = 0.000889885 Though I'd like to know why there is a difference, because $u_\tau$ results are very similar. Code: uTauLES uTauLES output: patch bottom u_tau : min = 0.0135431, max = 0.0430144, average = 0.0315161 utau_Beff_LES utau_Beff_LES output: patch bottom u_tau : min = 0.0135416, max = 0.0430147, average = 0.0315165 Aylalisa

October 28, 2014, 03:17		#17
aylalisa Senior Member Join Date: Nov 2012 Location: Bavaria Posts: 145 Rep Power: 13	Hi Xianbei, to figure out $R_{eff}$ for LES: I've choosen one of the turbulence models, for example oneEqEddy.C. Check the source code, line 65 --> Function call GenEddyVisc() Then check the file GenEddyVisc.C (in my case for incompressible fluids), line 92: Code: return -nuEff()dev(twoSymm(fvc::grad(U()))); This line seems to be $\tau_{ij}^r = - 2 \nu_{eff} * S_{ij}$ . $S_{ij}$ can be computed from grad(U()), it is the symmetric part of the deviatoric part of tensor grad(U()). The entity of that expression is , not yet a shear stress (I think for incompressible fluids it is better to pass the multiplication with $\rho$ ). Apart from multiplication with $\rho$ you have the stress tensor coming from the model. Have a look at LESModel.H, line 218, there you find nuEff(). Aylalisa

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Unsteady Boundary Profile with data file	Arianna	FLUENT	34	July 29, 2019 16:35
createPatch Segmentation Fault (CORE DUMPED)	sam.ho	OpenFOAM Pre-Processing	2	April 21, 2014 02:01
gmsh2ToFoam	sarajags_89	OpenFOAM	0	November 24, 2009 22:50
LES Near-Wall Velocity Profile	Alex May	FLUENT	2	June 12, 2008 12:36
What's wrong with my RMS profile in LES	Jiang Ning	FLUENT	0	March 31, 2005 05:30

October 1, 2014, 08:24		#2
itchy Member Florian Ries Join Date: Feb 2014 Location: Darmstadt, Germany Posts: 88 Rep Power: 12	Hi, first: In dynamic Smagorinsky you use Germano procedure to estimate the Smagorinsky constant. With this you get the y³-law of the wall for nuSgs (that is what you want). If you use van Driest with dynamic procedure you get unphyisical behaviour for nuSgs at the wall. So do never use van Driest with dynSmagorinsky second: post your case, then we can help you. Without your case file it is nearly impossible to find the error. kind regards Florian

October 1, 2014, 09:31		#3
itchy Member Florian Ries Join Date: Feb 2014 Location: Darmstadt, Germany Posts: 88 Rep Power: 12	Hi again, how do you calculate tau_w??? Do not calculate tau_w with the velocity gradient near the wall, because you use a inner variable that depends on your variable! If you do this, then OF (and any other solver) will underpredict tau_w. A better way is calculating tau_w or u_tau using the pressure gradient (fvOptions). What kind of LES do you solve??? Near wall resolution LES (NWR) or near wall modelling LES (NWM). If you do NWR, then your y+ is to high. kind regards Florian

October 8, 2014, 03:19		#5
itchy Member Florian Ries Join Date: Feb 2014 Location: Darmstadt, Germany Posts: 88 Rep Power: 12	Hi Lisa, uTau: The inner variable is the velocity (mean velocity). In your first approach you use U to calculate utau. And then you use utau to scale U. With this you get uncertainties in utau. Not a good way. Another possibility is to calculate utau with the gradientP. You can extract gradP of your logfile "pressure gradient = ..." and take the mean of it. Then for channel-flow you get: utau = sqrt(0.5hgradP), where h is the height of your channel. Grid-resolution: your grid should be fine enough. Sorry, I misunderstood your specifications. fvSchemes: I would use Gauss linear, Gauss linear corrected. But this should not change so much. fvSolution: Do you solve B and k???? I don't know the dynSmagorinsky-code of Alberto Pasquale. You should insert a residual control, smth. like this: PIMPLE { nOuterCorrectors 10; nCorrectors 1; nNonOrthogonalCorrectors 3; pRefCell 1001; pRefValue 0; residualControl { U { relTol 0; tolerance 1e-6; } p { relTol 0; tolerance 1e-6; } } } Is your case statistical stationary when you begin your averaging?? Is your avaraging long enough?? Do you have proper reference data to check accuracy?? I have never done a LES for channel, only for pipe (but with my own dynSmagorinsky-code). If you don't get it managed with the dynSmagorinsky-code of Alberto Pasquale I can send you my code. Best regards Florian

October 11, 2014, 07:47		#7
itchy Member Florian Ries Join Date: Feb 2014 Location: Darmstadt, Germany Posts: 88 Rep Power: 12	Hi Lisa, averaging time that the flow passes 200 times the domain should be enough for a stable mean value. For the variances pherhaps not. I mean some experimental data of channelflow?? Is there some abnormal behaviour concerning log-law for channelflow as a function of Re?? In pipeflow it is. With proper reference data I mean some experimental data of exactly your problem?? e.g experimental data of u+ vs. y+ . Do you have some result?? kind regards Florian

October 19, 2014, 08:29		#9
huangxianbei Senior Member Huang Xianbei Join Date: Sep 2013 Location: Yangzhou,China Posts: 302 Rep Power: 13	Hi, if a turbulence model is used, the profile may goes upon the analytical profile. I did a similar case with LES before and got similar results as yours. If no turbulence model is used ,then they are commute. I have a question about the way to calculate the wallShearStress tau_wall=rhonugradU; while in the wallShearStressLES, it's calculated as unit_wallnormal &Beff In my case of a 90 bend duct, the two results deviates about 30-100 times, what's your opinion about that?

October 21, 2014, 12:21		#10
aylalisa Senior Member Join Date: Nov 2012 Location: Bavaria Posts: 145 Rep Power: 13	Dear huangxianbei, thank's for your hint according to the turbulence model! My function for computing $u_\tau$ , using $\tau_{wall}$ according to unit_wallnormal &Beff ,does not work either. Problems: 1) how to compute $\tau_{wall}$ ? $\tau_{wall}$ = (mag((-Sfp/magSfp) & Beffp)*magSfp) (-Sfp/magSfp) & Beffp should give an vector. I thought, multiplied by the cell face area should give the desired shear stress ... 2) how to compute $u_\tau$ with $\tau_{wall}$ ? Dividing it by $\rho$ ? I haven't solved it yet. Best regards, Aylalisa

October 24, 2014, 04:56		#12
aylalisa Senior Member Join Date: Nov 2012 Location: Bavaria Posts: 145 Rep Power: 13	Hi Xianbei, sorry that I ask again but I'd like to be sure to understand you correctly. You compute $\tau_{wall} = \rho * \nu$ * gradU and this result matches the experimentally received $\tau_{wall}$ . u $_{\tau computed}$ is equal to u $_{\tau measured}$ . The problem is that $\tau_{wall}$ = unit_wallnormal &Beff is not equal to $\tau_{wall} = \rho * \nu$ * gradU (and experimental result)? How do you compute $\tau_{wall}$ = unit_wallnormal &Beff exactly? Aylalisa

October 24, 2014, 07:44		#14
aylalisa Senior Member Join Date: Nov 2012 Location: Bavaria Posts: 145 Rep Power: 13	I started from this thread too but I am still looking for the reason why the result does not agree with $\tau = \rho * \nu * gradU$ :-(