LES of turbulent flow in annular pipe
 

Hi everybody,

I am trying to set up a LES simulation using OpenFOAM 2.4.0 for a turbulent flow in an annular pipe. The solver I am using is pisoFoam and the turbulent model is dynamicSmagorinsky. The (structured) mesh is the one used by Lee et al. 2003, which was suitable for the LES they set up at that time.

My first goal is to reach the statistically-steady state, before doing any statistics. To do so I am monitoring two variables: volumeAverageUx2 and volumeAverageUy2 which means the volume averaged of the square of the velocity components normal to the flow direction (z-direction in my case). They should give an indication of the turbulent kinetic energy, as explained in this tutorial: http://www.ara.bme.hu/~lohasz/LEScou...nel_tut_v1.pdf . I should expect a peak of these variables and then an oscillation around a constant value when the statistically-steady state is reached.

The problem is the following: using the central scheme for the convective term (i.e. Gauss linear) the solution diverges after some time, depending on the initial condition. I also tried to use the MUSCL scheme and the same occurs. Even by fixing the maxCo to a small value such as 0.2, the schemes diverge (because the velocities become too large and the timestep very very small) The only scheme for the convective term which is stable is the upwind scheme. I tried to use the upwind scheme for the transition and then the CDS in the statistically-steady part, but the scheme diverges anyway.

The initial conditions used so far are: laminar parabolic profile unperturbed, laminar parabolic profile with 2 type of perturbations on top of it. I tried to adopt the initial condition used by Eugene De Villiers to my geometry but nothing improves. I also tried to use "perturbU.tgz" for "any ducted flow" that can be downloaded online, but I cannot graphically see any perturbation, despite my average Re_tau=147 is quite a common value.

Any ideas about how to tackle the problem? Do you think it depends on the initial condition, numerical scheme or maybe mesh? In the tutorial mentioned above, the CDS is used and no problems are met.

Thank you very much,
Francesco

I found the problem. The crash was due to the OpenFOAM utility "funkySetFields" which changed my velocity BC at the wall from "uniform ( 0 0 0 )" to "zeroGradient".
It is enough to execute funkySetFields and then to change the BC in the 0/U file before running the simulation.