# URANS 2D Square Cylinder Problems

 Register Blogs Members List Search Today's Posts Mark Forums Read

 May 23, 2017, 13:18 URANS 2D Square Cylinder Problems #1 New Member   Join Date: May 2017 Posts: 3 Rep Power: 8 Dear Community, I found a strange behavior on some URANS bluff body aerodynamics simulations that I am carrying out. I am studying the square section geometry in 2D. The Reynolds number based on the cross-section is 75600. The Cartesian geometry is created with blockMesh (OF3.0). The use of such mesh generator prevent from the mesh skewness and non-orto problems. The domain is very wide, so I got no problems of reflection or blockage. The mesh resolution is very accurate (from 60k to 250k cells), because the aim is the use of SA model and kw-SST model WITHOUT wall functions (low-Re). With my mesh I reach a y+ max (locally, at the edges) between 4 and 2 depending on the mesh accuracy. n_w/D ranges between 6e-4 for the coarsest mesh and 3.5e-4 for the finest one. The stretching factor is at least 1.3. I use OF 2.3.1 to solve the equations. To interpolate the discretized RANS equation terms I use second-order schemes such as: linearUpwind for div(phi,U) term linearUpwind for the turbulence model transport term(s) linear for the diffusive term I solve the equations with the PIMPLE pressure-velocity algorithm. The iterations advances in time with the backward second-order scheme. The time step varies according to the maxCo imposed at any iteration. I carried out a lot of simulations, but I would like to ask your opinion. It seems that the numerical solution of the problem could converge to two distant solutions: one in good agreement with the experimental results (CD=2.1, St=0.125, CL'=1.5), one far away from it (CD=2-2.4, St=0.09-0.1(!!!), CL'=1.3-1.8). In particular I noticed: - With the SA model: with coarse mesh and large time step the solution converges to the expected. Increasing the mesh resolution or reducing the time step (down to maxCo=1) the model seems to miss some additional numerical viscosity and converges to a solution (with Strouhal = 0.09) far away from the experimental one expected. - With the kw-SST model: I found more problems than with the SA. Also for coarser grids or larger time step the solution converges to the same values above mentioned far away from the expected one. It is a long time I am trying to vary all the possible parameters: mesh generator, numerical schemes, OF version, etc. I am struggling with the interpretation of these results. I would like to ask your opinion, especially from OF expert. I apologize for disturbing you, but I am not so good reading the code and maybe there is something there that I am missing. Sincerely, Andrea

 May 23, 2017, 14:04 #2 Senior Member     CFD Join Date: Nov 2010 Location: United States Posts: 243 Rep Power: 16 Hi, What values did you use for k and omega when you used komegaSST? I guess generally the problem similar to that if you replace the square by a cylinder in terms of physics. Best,

 May 24, 2017, 01:09 #3 Senior Member     Uwe Pilz Join Date: Feb 2017 Location: Leipzig, Germany Posts: 744 Rep Power: 15 With the coarse mesh you have the wall function working. If you finer the mesh an lower the tim step the wall function works to a much lesser degree: That is the difference I see. Nevertheless, the close to DNS simulation should get reasonable results too. But I think that is the direction you should look for what is different. __________________ Uwe Pilz -- Die der Hauptbewegung überlagerte Schwankungsbewegung ist in ihren Einzelheiten so hoffnungslos kompliziert, daß ihre theoretische Berechnung aussichtslos erscheint. (Hermann Schlichting, 1950)

 May 24, 2017, 09:46 #4 New Member   Join Date: May 2017 Posts: 3 Rep Power: 8 Dear tareqkh, if you mean the k and omega values at the surface, I employed the values suggested in Menter's paper for omega, while I set k very small (1e-20). Please, can you explain better your statement "generally the problem similar to that if you replace the square by a cylinder in terms of physics"? Thank you

 May 24, 2017, 09:49 #5 New Member   Join Date: May 2017 Posts: 3 Rep Power: 8 Dear piu58, as I wrote in the main message I do not use wall function. I mean, I do not set any wall function in the dict files. Are there some WF in the code working irrespectively to my numerical set up? You wrote "the close to DNS simulation should get reasonable results too", but how can I perform DNS in a two-dimensional domain? Thank you

May 24, 2017, 21:01
#6
Senior Member

CFD
Join Date: Nov 2010
Location: United States
Posts: 243
Rep Power: 16
Quote:
 Originally Posted by and_user Dear tareqkh, if you mean the k and omega values at the surface, I employed the values suggested in Menter's paper for omega, while I set k very small (1e-20). Please, can you explain better your statement "generally the problem similar to that if you replace the square by a cylinder in terms of physics"? Thank you
Dear and_user,

Well, you should expect von karman vortex according to your Reynolds number. Please have a look at the following link http://www.mediafire.com/file/wyf8w1...dyCylinder.pdf. I created this document a long time ago for the laminar flow over a cylinder. You might find it helpful at the same time you might find some typo here and there etc. I still have the case files as well. By the way, what is your wall distance? How fvSchemes looks like in your case?

Regards,

 Tags aerodynamics, bluff body wake, square, urans