code_saturne User's Forum

Hello,

I'm using k-omega SST turbulence model to simulate flow around a circular cylinder (code_saturne 5.0.4) in order to study the vortex shedding phenomenon. I use a refined mesh to ensure that y+ is less than 1, so that no wall function is necessary. The simulation results in a much larger force coefficients that experimental data and the results in the literature (Reynolds number is 10000, instead of cl (lift coefficient, fluctuating part) = 0.5-0.7 and cd (drag coefficient, average part) = 1.1-1.2 in the literature, I got cl = 1.6 and cd = 1.55). I use turbulent intensity and turbulent viscosity to initialize k and omega for the internal field and also to specify boundary condition for k and omega at the inlet.

I notice that in code_saturne, the boundary condition at a wall for omega is given as a Neumann type, however in the literature, people use mostly the Dirichlet type, and the latter one seems to me more logical since what we want to impose is a large omega value at a wall in order to have a tiny turbulent viscosity. I tested these two types of boundary condition, only small difference between these two was observed. The explanation could be that omega drops abruptly as we get away from the wall, the results of these two types of boundary condition can be very similar.

However, when I modified the imposed value of omega at a wall, the numerical results changed greatly. This is not normal, since as long as we impose a great value for omega, the result should be very close (this statement is validated by using the Openfoam code). In addition, the numerical results are very dependent on the inlet conditions of k and omega that I give, which seems to me very abnormal as well. Lastly, the results are very mesh dependent. For two meshes differing only in the axial direction of the cylinder, and both satisfying y+ < 1, the numerical results could be very different.

Something is going wrong in my simulation, maybe a bad setting in the parameters (numerical solver, relaxation factor, etc). You can find in attached files some information files (the .zip file includes all the necessary elements to reproduce the simulation) related to my simulation, could you please help me to find the problem?

Thanks in advance!

Best regards,
Lei

Hello,

I'm not a turbulence expert, so I'll check with colleagues, but your results do seem a bit strange, since in most cases, k-omega is less sensitive to near-wall mesh refinement than most other RANS models.

Did you try the advanced turbulence parameters, such as scalable wall functions ? I seem to recall the basic k-omega as implemented in Code_Saturne works well mainly above y+ = 1. I'll try to get info on why we use a Neumann condition, or suggest someone more expert in that area to check this thread.

Best regards,

Yvan

Hello,

Thanks for checking with your colleagues!

I have tried k-omega SST model with wall function using a coarse mesh (y+ around 30), which worked well. However, as I want to resolve explicitly the viscous boundary layer in order to obtain the flow separation point exactly, the better solution should be using a refined mesh without wall function.

I wonder if the implementation of k-omega SST support only the case with wall function. I'll be glad to see if you can share a test case involving k-omega SST model using a refined mesh without wall function.

Thanks a lot!

Best regards,
Lei

Hello Lei,

Thanks a lot for sharing your test case. You are right concerning the Neumann type BC for Omega in the code. We are currently working on an similar issue. Could you try to set the Neumann constant of Omega BC to 6.75 instead of 120 and see if this improve your results ?
Could you also please attach the modified clptur.f90 source in which you implemented the Dirichlet BC ?

Regarding the k-omega model itself, its boundary condition that goes to infinity at the wall makes it very sensitive to mesh refinement when y+<1. It can be hard to observe a real convergence when progressively refining the mesh. It is usually advise to keep y+ around 1.

I will come back to you as soon as I have a look on your test case. Keep me informed of the change in the Neumann BC.

Regards,

Jean-François

Hello,

Thanks a lot for your response!

As you suggested, I changed the omega wall bc from 120 to 6.75, however the numerical results show no improvements. You can find clptur.f90 in the attached file, where instead of using "call set_neumann_scalar", I use "call set_dirichlet_scalar". It seems that no modifications are needed in turbkw.f90, where the BC conditions are recalculated taking into account of the variation of sigma in the k-omega model.

I tried different values of omega to impose at the wall, which leads to very different lift force coefficient. In addition, when I doubled the number of cells in the axial direction of the mesh around the cylinder, the lift force coefficient could change from 1.5 to 0.4. (the maximum of y+ drops from 0.4 to 0.2). All these are quite strange, and I believe there are somethings wrong in my simulation.

I am looking forward to your feedback of the test case!

Best regards,
Lei