Mesh in Boundary Layer

[luyitang]

Hello,

I have some doubts during my simulation with Code_Saturne.

1) I use k-epsilon model to simulate high reynold number flow in a tube (Re= 7*10^5). The result didn't converge well,the first time step stops at N_iter=N_iter_max =10000(I have already decrease the precision from 10E-08 to 10E-06). And in the listing, it informs me that my mesh is "too refine at the wall,... yplus must be greater than the limit value YPLULI=0.2381E+01". I have an impression that, in the k-eps model the mesh can't be too in the boundary layer, the first mesh should put into the turbulence well developed domain. Is this the reason for the "too refine at the wall" information?

When I enlarge the mesh, the "too refine" information still appear, how should I modify the mesh, I'm confused. I think the LAW OF WALL may help to do the theorical calulation of y+.
Should I do like below:
u_fric=sqrt(tau/rho),
tau is the wall shear stress, tau= mu*(du/dy)|y=0
y+=u_fric*y/nu

Is "y" where I should put my first mesh?

2)When the thermal boundary layer is also involved, how can I define the first mesh?

3)Among the RANS model, which model can allow me see the transient of boudary layer?

Thx a lot!

Luyi

[Yvan Fournier]

Hello,

Did you follow the "best practices" for Code_Saturne (http://code-saturne.org/cms/documentation/BPG) ? If not, you should start reading there.

For a complex geometry, there may be cases where y+ is too refined in some areas, not enough in others, so visualizing the output y+ with ParaView, EnSight, or a similar tool is recommended.

For a canal, this is probably not the case.

You may try the "scalable wall function" advanced option for k-epsilon, but it is always nice to visualize y+ first to see how far from the target you are (the target y+ for the first mesh depends on the turbulence model, with about 1 being recommended for low-Reynolds models, but higher values being required for hight-Reynolds models; heck the best practices here).

I am not sure I understand your last question: RANS models are based on an average, not transient model, although if I remember correctly, the averaging is on the time scale of k/epsilon, so transients with a time scale significantly above this are OK. If you are talking of turbulent transition, the value of y+ is related to whether you are in the laminar, log, or turbulent region, but for most models, Code_Saturne only computes it at the wall (though for a simple geometry, you might rebuild it in cs_user_extra_operations.f90). For temperature, T+ is the equivalent. RANS models this transition rather than captures it. To capture it, DNS would be needed in theory, and LES might work, but I'll let you check with turbulence specialists.

Regards,

Yvan

[luyitang]

Hello Yvan,

Thank you for your respond, it helps a lot. I didn't realize there is a BPG, I was always searching for a documentation like that which can explain how to choose y+, Fourier number,etc.
Thanks again.
Best regards!

Luyi