[thecfduser]

If u are looking for the mean values, u must of course use a turbulence model.
Anyway, it is why we invented turbulence models.

U does not need to refine the mech very much....Only near the solid boundaries if u want to use a low-Re model, or a 2 layer model....

I dont know if a URANS model will be able to capture a non-stationnary flow....but i think no.
We cannot use a DNS for such a flow, unleast u can wait a mont to get a solution

Remember that for such a Re, even in experiences u will not see a vortex schedding (u have it, but u have a lot of other scale eddies...)

[maa]

I need to repeat the simulations because I ´ll work for another Reynolds and I need to know if the results are or not correct.

1- I think that the simulation is converged because I have 20 time step and during the simulation the fluent use only 11 time step to converge.
2- I have done one simulation for RNG steady and the result has the same of the unsteady.
3- In literature they used 2D.
4- I don´t have low-Re modification because I´m using the FLUENT 6.3 and I don´t have that option.

Why shouldn´t I use a turbulence model for Re = 27 000?

[thecfduser]

1-Monitor variables. Residuals are not sufficent to judge convergence. Anyway, in my life, i never used more than 2 iterations per time step.... Refine ure time step instead of using 11 iterations by step. if ure time step is dt, u will only resolve correctly the scales wich lifetime is higher than dt/2.
2-Yes because RNG as k-epsilon model is a bit...lets say robust.
In fact, RANS models are models that take average.
In average, u have 50% probability that the first big eddie comes from the top, and 50% that it comes from the down of the cyclinder. So a RANS model wil not give u a vortex schedding because it is doing averages (unleast the model is not highly diffusive, so it will be the NUMERICAL residuals that makes ure vortex schedding start).
3-OK. They are supposing a very long cylinder and uses RANS models (DNS or LES will need 3D, even if the average mean flow is 2D)
4-U have it, but it is for professional so it can only be activated by typing a command line. Anyway, for RNG u still have it in the interface: differentiel viscosity option. U must year a very fine near wall mesh.
U can use an intermediate near wall resolution, wich is the enhanced treatement in FLUENT.

So, to know wich model using, compare with litterature. Find the best model, that u can use it for the rest of ure simulations.
Verify that solution is independant of the grid.
Use steady RANS if the simulation still converging (monitor variables and not only residuals)
An important thing, do a good grid quality. Gambit does not handle body fitted grid and this is a big limitation....

[pachputesharad]

I modeled 2-D flow over cylinder .ICEMCFD used for grid generation with O-grid.Computational domain like 10d upstream,20d top and bottom,20d downstream.Used BC such that inlet-velocity inlet,oulet-outflow,Top,bottom-symmetry.Firstly for Re=100 with dt=0.01,t=20.compared this Cd=1.31+-0.02 value with research paper data

[rishi_iyengar]

Hi Sanjay

I think the accuracy of capturing vortex shedding is more fundamentally based on the grid that you use.Use a very fine grid in the wake region.Try and control the y+ value also.

I think if you can do this you ca capture vortex shedding with any turbulence model.

Rishikesh

[teguhtf]

Hi all,
I have done the simulation of inviscid flow in circular cylinder (i just follow the step of http://www.et.byu.edu/~maynesrd/clas...20Cylinder.pdf). I think some thing odd there. I find wake behind the cylinder. How it can be?? In my mind, inviscid flow will not give wake. Is any body can explain the reason??
Thank you
Teguh Hady Ariwibowo