I am conducting simulation on the vortex street behind cylinders near a boundary. I'm currently conducting 2D sims with the standard k-esp model with the standard wall function.

The Re= 50000 (5e4) I have successful realistic shedding according to the Cl values I obtain.

the question is I have a fine mesh near the cylinder and I have y+ values ranging from 2~35. The manual says to avoid the region y+ = 30~300. I'm smack bang in th damn middle. This would mean I need to increase my first cell spacing by a magnitude of 10 and this does not look right. Quite rightly this looks wrong.

I have used the Enhanced wall treatment on this mesh and it seems to give no different solution. However refering to the manual again, it says to resolve viscous sublayer with 10 or more cells in the inner layer (i.e. y+<300) and put first cell centroid in the viscous sublayer y+=1~5.

Then comes the final blow, k-esp preforms poorly for seperation, which I have from vortex shedding and high presure gradients, which again I believe I have from the vortecies.

All these revilations have made me a bit lost. Some guidance or examples of other peoples parameters would be appreciated.

So thanking you all in advance

Allan

i can not say much about the yplus, for enhanced wall function you will need yplus around 1 and for standard wall functions you will need it greater than 30. and yes, standard k-e model does not perform well with separating flows. But for cylinders, where the most important thing to predict is drag, k-e realisable model gives good results with enhanced wall functions. however if you want very good results for separating flows then you have to move to DES or LES. See if that is a possiblity. With LES, you can try werner wengle wall functions if your yplus is in the range you mentioned.

A good method for dealing with the problem of variable y+ is to use grid adaption. For this you need a grid which has y+ values just in the range you experienced (this is perfectly normal, the cylinder is a typical application where y+ takes very different values along the wall, because wall shear stress is big up to separation point and then drops behind the cylinder).

So, for steady case use once y+ adaption with 0.9-1.1 as minimum and maximum values (be carefull to include geometry adaption as well) and for unsteady case use dynamic adaption for every time step, keeping both "refine" and "coarsen" options ON.

BUT, this approach is "valid" only for 2d flows, in 3d the number of aditional cells generated by this method is too great and a grid which is good from the beggining, is prefferable (good means y+ no more than 2 near the separation points)

As for turbulence model, the most important feature you must be able to obtain accurately is the separation point. On this depends mostly the accuracy of the whole calculation. So, you must use very good turbulence model. Realisable k-e with EWT will do a sufficiently good job, but if you want better you must use RSM. But of course, the best remains LES (the only problem is you need very good quality grids, high number of elements and lots of patience).

Best whishes, Razvan

I have worked with LES for separated flow cases. The problem is of course that a very fine near-wall grid is required (typically around y+=1). However, we were able to reduce the grid size tremendously by using the trimmed cell refinement near the walls. A structured grid that would give the same near wall resolution would have been around 13 million and we were able to get the same results from a grid of 1.15 million cells.

For the cylinder case, you would require to refine the wake region as well to get good separation characteristics and wake development. Of course you understand that LES is reliable only in 3D. I have worked with 2D LES for separated flows over airfoils but the vorticity tends to amplify in this case and corrupts the solution.