Hello,

I'm doing a simulation of the 2D flow of a circular cylinder at high Re-Numbers with k-eps with wall functions. For Re=1e6 the mean drag coefficient is 0.4 which fits to data from literature. For Re=1e7 the drag coefficient is the same, but should be around 0.8. If I mesh the boundary layer I'm limited by the y+ value. In the results it seems that the boundary layer is only one cell thick but further mesh refinement is not possible due to y+ value. Perhaps the false drag coefficient is because of the low resolution of the boundary layer ?? Any comments will be appreciated !

Tim

you're right it should be obvious that you need more b.l. grid points. why can't you refine the grid more?

Because the y+ gets to small for using wall functions. It should work with a two-laayer-modell or a low-Re turbulence modell.

I think you can cluster the grid at the wall vicinity only, good result is expected.

Clifford,

my problem is that in my actual grid level the y+=20-40. As far as I know further refinement would violate the assumption of the wall function. If I look at my results in the first grid cell next to the cylinder I have already half of the free stream velocity. So it seems that the boundary layer thickness is only up to y+=100. Is this o.k. ?

Tim

the above post by Joern Beilke has some good points you can check it out. i had thought you were using a wall function but weren't sure. can you also model the turbulence right up to the wall, ie not use the wall law, you will need much more grid points but setting up the problem might be easier

HI! Franke,

With wall functions, you are correct. You should stay in the log layer region. But what is more important with respect to drag coefficient prediction is, how are you doing the wall boundary condition for the shear stress. From wall functions, you have to impose the shear stress on the wall based on the U_tau calculated (stress = rho*u_tau**2). THis shear stress has to be imposed on the wall boundary face while computing your viscous fluxes on that face. The velocity profile that then comes out will show a velocity difference that can be supported by this shear. If you have a y+ of 20-30 you will see a smaller velocity difference between the wall point and the first point off the wall than if you had a y+ of 150.

Now, for drag coefficient calculation, you have to use this shear stress imposed to compute the Cd. You cannot use the standard delta_u/delta_n type of computation, this will give you the wrong Cd. you have to store the shear stress imposed on the wall face and compute teh Cd from this. if you use the delta_u/delta_n type of approach, you will typically over predict the Cd.

(At lower reynolds numbers, the resolution might be good enough that the shear can support a velocity difference that is close to the finite difference shear, at larger reynolds numbers, your cell reynolds number drops and this is no longer the case.)

Hope this helps Srinivasan

Hi Tim,

Sorry to bother you with a question instead of an answer:

Predicting drag on cylinders should be a question of predicting the separation properly. I just wonder what your resoltion along the cylinder wall is, and if an increase of surface resolution has an impact on the drag?

Lennard