[huangxianbei]

Hi,all:
I have just done a LES calculation of channel flow. The size of the channel is 2pi*h, 2h, 2pi*h, Re_tau=397 and Re=7000, the mass flow rate is fixed so pressure gradient is refreshed each timestep. As known, the Reynolds stress in LES should be UPrime2Mean+Rmean(or BMean),So the SGS part is added to obtain the total Reynolds Stress

[huangxianbei]

Oh,dear, I find that the RMean should do a sqrt(RMean) when using rms values, so here is the correct curves:



Now I have a question about the much overpredicted urms, is there any way to improve this?(The vrms and wrms is much better in dynamicSmagorinsky)

[huangxianbei]

Here is Eugene's results in his Phd's thesis

He use oneEq model and obtain much better result than mine, I don't know why this happens because I only change the calculation domain's size and mesh,64x128x128, the results are averaged over about 40 flow-through time(t*=Lx/Ub as defined in Eugene's thesis)

[huangxianbei]

No one? Ok, I think I should run the same case with Eugene and figure out what's the problem.

[Villo]

Hi Huang, check if your y+ is the same as Eugene: this could be the first reason of a mismatching of fluctuating values near by the wall where velocity gradients are high.
Second thing, check your fvScheme, normally for LES the best numerical scheme (for a linear geometry as a channel flow) is Gaus linear

[Villo]

Oh, when you will check the mesh size check also the time step! If you are applying a higher time step you could be not able to catch all the spectrum of the transient turbulent phenomenon.
Last thing about settings... check your LES filtering settings if is the same as Eugene.
I saw that you`re applying a fixed mass flow rate... the other way is to apply a fixed velocity as inlet condition with periodic boundaries for inlet-outlet including system/fvOptions dictionary in order to keep a pressure gradient

[huangxianbei]

Originally Posted by Villo

Oh, when you will check the mesh size check also the time step! If you are applying a higher time step you could be not able to catch all the spectrum of the transient turbulent phenomenon.
Last thing about settings... check your LES filtering settings if is the same as Eugene.
I saw that you`re applying a fixed mass flow rate... the other way is to apply a fixed velocity as inlet condition with periodic boundaries for inlet-outlet including system/fvOptions dictionary in order to keep a pressure gradient

Hi,Villo:
Thank you for your suggestions. Well, my domain is in fact different from Eugene's, 6.28x2x6.28, his is 4x2x2, and, considering my mesh ,y+ is about 0.8. So I think it's enough because the finest mesh in Eugene's thesis is y+=1. Yes, there are two ways--fixed gradP or fixed U_bar, while in order to compare with the DNS data with is obtained under a fixed U_bar , I chose to use the latter way.Here is the fvscheme, as you can see , is of no difference with Eugene's in the tutorial.

ddtSchemes
{
    default         backward;
}

gradSchemes
{
    default         Gauss linear;
    grad(p)         Gauss linear;
    grad(U)         Gauss linear;
}

divSchemes
{
    default         none;
    div(phi,U)      Gauss linear;
    div(phi,k)      Gauss limitedLinear 1;
    div(phi,B)      Gauss limitedLinear 1;
    div(B)          Gauss linear;
    div(phi,nuTilda) Gauss limitedLinear 1;
    div((nuEff*dev(T(grad(U))))) 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(DBEff,B) Gauss linear corrected;
    laplacian(DnuTildaEff,nuTilda) Gauss linear corrected;
}

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

snGradSchemes
{
    default         corrected;
}

fluxRequired
{
    default         no;
    p               ;
}

I have run a case which is completely the same as Eugene's , I'll do some post-process to see if it's really agree well with DNS data and what's wrong with my case.

[Villo]

Ok, you`re in line with him about fvScheme and y+. Your setting about momentum source correction is right (fixed Ubar).
The last things that you should check are time step, mesh stretching ratio and LES filter.
The difference in length of your domain can effect results. The risk for a short domain, as the one of Eugene, is that fluctuating fields are still correlated between inlet and outlet: periodic boundary conditions will introduce in each flow to time a certain amount of error (you can apply two-points correlations in order to verify it). On the other side, your domain is longer... you need an higher flow to time to reach the convergence! let it run for a longer time.
If you want to make tuning on your LES computation to find your best settings... yes, use a DNS computation: Run your LES on a domain with the same dimensions used from a DNS computation (Moser maybe).

[huangxianbei]

Originally Posted by Villo

Ok, you`re in line with him about fvScheme and y+. Your setting about momentum source correction is right (fixed Ubar).
The last things that you should check are time step, mesh stretching ratio and LES filter.
The difference in length of your domain can effect results. The risk for a short domain, as the one of Eugene, is that fluctuating fields are still correlated between inlet and outlet: periodic boundary conditions will introduce in each flow to time a certain amount of error (you can apply two-points correlations in order to verify it). On the other side, your domain is longer... you need an higher flow to time to reach the convergence! let it run for a longer time.
If you want to make tuning on your LES computation to find your best settings... yes, use a DNS computation: Run your LES on a domain with the same dimensions used from a DNS computation (Moser maybe).

Yes,you are right, the shorter domain can be more sensitive. I use the following criteria to judge if it's converged:
-uv+Re_tau^(-1)*dU/dy be linear to y
so I can make sure it's converged.
The domain is same to the DNS, now the only this I'm not sure is whether a different U_bar used in the calculation can affect. Since if a smaller U_bar is used, the timestep can be bigger and make the calculation more quickly to be fully developed. So I use a different U_bar=0.07, the DNS is about 0.1.

By the way, have you notice that perturbU generates much similar fields no matter which Re_tau you specified and also the U_bar?

[Villo]

I never used perturbU for similar simulations if you have available a previous LES simulation (also if at different Re_tau) map the solution with mapFields: your simulation will reach convergence faster.
About your convergence criteria, do you have any reference for it? I use to check residuals

[huangxianbei]

Originally Posted by Villo

I never used perturbU for similar simulations if you have available a previous LES simulation (also if at different Re_tau) map the solution with mapFields: your simulation will reach convergence faster.
About your convergence criteria, do you have any reference for it? I use to check residuals

oh,I see. The criteria is from papers such as Kristofferson's "Direct simulations of low-Reynolds-number turbulent flow in a rotating channel,1993"

[Villo]

Thank you Huang for the reference!
Good luck with your LES

[huangxianbei]

Originally Posted by Villo

Thank you Huang for the reference!
Good luck with your LES

Here again. Well, a disappointing fact appears that the solution ,which is the same as Eugene still not converged when running 35000s... the friction velocity is only half the right one. I'm thinking to use the mapFields instead of perturbU.