I am trying to carry out a solution for flow in a standard pipe. Due to several incovenients, I must use a code usually used for naval cases. I'm wondering if this code is usefull for me, becouse the results I get are not realistic. Should I obtain another code? The boundary conditions I am imposing are the following: celerity at the inlet flow cross section given; a roughness wall law (unused commercial steel, k=0.05E-6 m); Smagorinsky formulation of turbulence (the flow I'm considering is always turbullent, Re>10E5). Is that correct? I think the mesh is OK, so the problem is not in that way.

This is a first simulation to obtain the loss of pressure/velocity because of the roughness, then I want to get the velocity and pressure fields over a valve situated in the pipe. This is where I began, but the results were not correct, so I thought I could estimate the loss of charge to see what wall law or turbullence modelling to use, and that's the reason for the pipe without the valve.

Thank you for your time and please forgive my poor english. Samuel

The Smagorinsky formulation implies the use of LES to represent the turbulence rather than the use of a statistical turbulence model plus wall functions. Are you doing LES or turbulence modelling?

As far as I know, I can model the turbulence with the Smagorinsky formulation, wich implies the use of LES, but I thougt that besides there was also needed a wall law to take in account the loss of pressure due to the roughness of the pipe.

Do you mean that with the use of Smagorinsky (or any other model of turbullence) there is no need to apply a wall law over the pipe?

Hello Samuel,

For using LES for the Reynolds number you are looking at would need a very huge mesh. In LES, the large scales (which are anisotropic) are computed and the small scales (which are isotropic) are modeled. The model you are referring to is meant to model the small scales. Getting a mesh which would resolves the scales down to the isotropic scales would be too expensive. To make things worse, the simulation is going to be an unsteady one. Then, you need to worry about proper inlet and outlet boundary conditions to get the accurate turbulent statistics in the region of interest. It is important to note that the numerical techniques generate most of the errors at the small scales.

You are better of trying a 2-equation model like the RNG version of the k-e model with wall function (this is a standard feature in most of the commercial codes today). In this case, you would be solving for the mean flow, which is good enough for engineering applications

Good Luck,

Thomas

My question amounts to asking whether you performing a unsteady 3d calculation using the Smagorinsky model to model the sub-grid scales (LES), or are you performing a steady 2d calculation using the Smagorinsky's model for closing the unknown correlations representing the turbulent fluxes of momentum? If it is latter, the approach is wrong and as Thomas suggests, you should use a statistical turbulence model, say for example, the standard high-Re form of the k-e model plus a logarithmic wall law using Jayatilleke-Nikaradse "sand-grain" roughness.

Then I was a bit confused with turbullence. I thought that the Smagorinsky was more or less the same than the k or the k-epsilon formulations, but both the lasts use differential equations to model and the Smagorinsky uses empirical results to do the same.

Thanks for the answers, Thomas and Michael. I will try with the 2 equations plus wall law as you say.

Samuel

Hi Michael,

Why a 2D axisymetric unsteady simulation of a pipe flow using Smagorinsky model shwon wrong results ?

Please forgive my poor english

Thank you, Norma

Far before being used to modelize the viscosity effect of unresolved scales for LES, the Smagorinski model was dedicated to compute a turbulent viscosity used to close the RANS equations throught a Boussinesq assumption. Some CFD codes, like pamflow, still used it that way.

So Smagorinski model doesn't necessary mean LES.

Regards,

Sylvain

Strictly, Smagorinsky's model should be used in LES, and Prandtl's model in statistical turbulence modelling. The two models look very similar except that Smagorinsky's model employs the local mesh size, whereas Prandtl's model uses the mixing length.

In fact, I used the k-e model at first with Reindhardt wall law, but the results where not reallistic (in some of them even the model didn't converge). Then I tried the Smagorynsky without wall law, and the results where OK.

So I think my code is the kind of that you where saying.

One important result of this little study is that the velocities obtained near the valve (a floodgate one) for a little grade of opening are on the range of 70-90 m/s, with a loss of pressure of near 1 KPa. This mean that maybe there is cavitation in this zone. For more opened, there is not such problem.

Thank you again to everyone and is a pity I can't put here a jpg with the results so that you could see them.