Accuracy Problem with Flow over 2D cylinder at Transcritical Reynolds Number
I am trying to perform what I thought would be a fairly straightforward validation case in which I want to predict the Strouhal Number and mean drag coefficient for a 2D circular cylinder at a Reynolds number of 5,000,000 but am getting nowhere near the results I should be, can anyone help?!
I am comparing against data given in a paper by Roshko (1961) found here: http://authors.library.caltech.edu/10105/1/ROSjfm61.pdf
From this paper I should be achieving a Cd of approx 0.7 and St of approx 0.27 (I know other authors cite anywhere between 0.25 and 0.27 so I'd be happy with anything like this!). My results are way off the mark so any advice would be great!
My setup is:
Cylinder diameter: 5m
Reynolds number: 5,000,000.
Domain extents: upstream 10D, downstream 20D, Upper and Lower 10D.
SST turb model (runs both with and without trans turbs gamma theta model)
Timestep of approx 1/50 expected shedding frequency, my monitor points look really smooth so I don't think there is any issue here although I will do a time-step sensitivity study when I have checked everything in the setup.
Mesh: Fully hexahedral with an average yplus oscillating between 0.6 and 0.9. There are approx 125 streamwise nodes along the cylinder surface.
Convergence criteria at 1E-5, each iteration is solving within 4 coefficient loops and the RMS Courant Number is around 4.4.
Inlet: Normal velocity inlet 15.45m/s (Based on air at 25C properties, isothermal flow and required Reynolds Number of 5,000,000).
Outlet: Average static pressure 0Pa (across whole outlet)
Upper and Lower: Free slip walls.
Cylinder end planes: Symmetry.
Cylinder: No-slip smooth wall.
Default turbulence intensity of 5%.
Without transitional turbulence: Cd=approx 0.4, St=approx 0.32
With trans turbulence (gamma theta): Cd=approx 0.23, St= approx 0.4
Results are independent of mesh but no time-step dependency study has been done, I want to have my setup checked first.
Any help would be gratefully received, I know there are other posts of this nature but I cant see anything that helps me past where I already am.
I can post the CCL for the most recent run if neccessary,
I would do a time step size and convergence tolerance sensitivity check, but if you say you are converging to 4 coeff loops per iteration then your current time step looks pretty good.
There is an FAQ on accuracy but you appear to have looked at most of the issues discussed on the page. Still it is worth a look, it might remind you of something you missed: http://www.cfd-online.com/Wiki/Ansys..._inaccurate.3F
Please post your CCL and images of you mesh.
Thanks for looking at this. I've put a link below to the CCL for the run with no transitional turbulence, can put the other up if necessary. Also in the same place are a few mesh images
Any help greatly appreciated :confused:
* Use adaptive time stepping, or at least do a time step sensitivity check on the time step you are using.
* Check that the inlet turbulence you have specified is not diffusing the vorticies. Try with a very low inlet turbulence.
* Your y+ is averaged over the whole cylinder. This includes the front section which has a boundary layer, and the rear section which will have a large separation. Getting the average y+ over both these regions does not appear useful. I would think calculating y+ up until separation (of maybe just the front half) would be useful.
* Check whether your separation point is about right. It might give you a clue.
* You have first order turbulence numerics. You might need second for this.
Thanks for the advice, been a bit distracted by another job but will look. At these points on Monday...............
|All times are GMT -4. The time now is 15:28.|