[ghorrocks]

I don't think your check or periodic vs symmetry boundaries is valid. I would do a full 3D model with enough elements in the Z direction that you will pick up a flow if it is there.

Does the literature say there should be 3D flow at this Re?

[Josh]

Good question. I'm not sure if there are 3D effects. At this low of a Reynolds number, I wouldn't expect any drastic 3D effects.

Still, we were specifically asked to render the cylinder as infinitely long.

[Tristan]

Josh,

What length scale is your Reynolds number based on? If it's the cylinder diameter then your flow is definitely not laminar nor is it two dimensional. I know I've seen LES and DNS work for cylinder flows with Re_D in 1e3 to 1e4 range.

Tristan

[Josh]

Hi Tristan -

The diameter is my length scale. Since we're interested in drag, we want the frontal area, which is the diameter multiplied by the length. Since the length is theoretically infinite, we assume it to be of unit length and, therefore, the frontal area is just the diameter.

To be consistent, my Reynolds number is also based on the diameter (D = 6 inches = 0.1524 meters). But this should be laminar, should it not?

Re = (density)(velocity)(diameter) / dynamic viscosity
= (1.185 kg/m^3)(4.445 m/s)(0.1524 m) / (1.8E-5 kg/m.s)
= 4.5E4
= laminar region

[Tristan]

Josh,

Your calculations are correct. I know Kravchenko did LES of flow over a cylinder at ReD=3900. His flow was definitely turbulent and three-dimensional . At ReD=40000 your flow will be more complex.

Tristan

[Josh]

Is the publication you're talking about called, "B-spline method and zonal grids for simulations of complex turbulent flows" by A. G. Kravchenko? If not, could I have some more info for that source so I can find it?

So you're suggesting that I run the simulation as 3D and turbulent despite that the problem stipulates an infinite length cylinder and a Reynolds value in the laminar region?

I appreciate the help, I'm just confused about the laminar-turbulent debate. From what I've read in literature, transition does not occur on a cylinder until Re = 3E5. For my Reynolds number of 4E4, I will experience laminar flow with a boundary layer that separates before the 90 degree point (from the front stagnation point) on the cylinder.

[Tristan]

That's the right guy/publication. Infinite length in one direction does not mean that there are no variations in that direction, there are spanwise (i.e. along the cylinder) structures in your flow. Where did you get 3e5 as the transition from laminar to turbulent? I think that's where the drag crisis occurs for a cylinder but the flow over the cylinder becomes turbulent much earlier than that (you just don't get the narrow wake yet).

You can run it 2D but you need a turbulence model and I think this is a challenging flow for turbulence models to accurately compute, as simple as it may seem.

Tristan

[Josh]

Thanks for the help.

I'll try some variations.

[crmorton]

"Any further suggestions/comments as to why my drag values are too low?"

For most transient simulations I have completed on a uniform cylinder, it takes a great deal of simulation time to reach a state of not only periodic vortex shedding, but quasi-steady periodic vortex shedding (meaning the drag fluctuations in time ocillate about a mean drag value over the period of 20 or 30 vortex shedding cycles).
Make sure in your transient simulation that you are saving enough time steps to properly re-produce the drag signal (do not under-sample)

Also, for a nice visualization of the flow, try contours of vorticity, using a log-log scale, with 50 contours or more. It takes a long time to load but looks very nice.

Cheers.

[crmorton]

"So you're suggesting that I run the simulation as 3D and turbulent despite that the problem stipulates an infinite length cylinder and a Reynolds value in the laminar region?
I appreciate the help, I'm just confused about the laminar-turbulent debate. From what I've read in literature, transition does not occur on a cylinder until Re = 3E5. For my Reynolds number of 4E4, I will experience laminar flow with a boundary layer that separates before the 90 degree point (from the front stagnation point) on the cylinder."

If you have library/online access, see "Vortex Shedding in the Cylinder Wake", CHK Williamson.

Transition to turbulence occurs in several stages for the circular cylinder. Transition to turbulent flow in the wake of the cylinder occurs for Re>1000.
As Re increases, you can think of the transition to turbulence moving closer and closer to the front of the cylinder. The shear layer separating off the upper and lower surfaces transitions to turbulence, then eventually the boundary layer on the surface of the cylinder transitions to turbulence (drag crisis)

[Josh]

Thanks for the great suggestions.

After talking to several experts and reading multiple sources of literature, I have come to the conclusion that an LES turbulence model would have worked nicely for my particular simulation. With my 8-processor CPU, each run would take about a week, and I simply don't have that time. The RANS turbulence models have, for the most part, failed me for the drag calculations. Qualitatively, the SST produced nice enough results and proper separation point determination, but produced drag values that were 20% of the theoretical value sizes.

With the LES, I should run the simulation in 3D and follow meshing conditions as specified in the CFX help file (e.g. for a cylinder of length 2 x D, 50 nodes should be placed in the spanwise direction). No symmetry should be used. Using the Strouhal number, I should find the frequency of my vortexes and base the timesteps on that, as well as an average Courant number of 0.5 to 1. I should also initialize the LES simulation with the values from a converged RANS solution (like SST). Also, the Central Difference (rather than High Resolution) advection scheme is preferable.

Chris -
I see you go to Waterloo, not too far down the way from me. I went to Queen's and am now at RMC. I was in Waterloo for Oktoberfest - I think it was a good time, but can't really recall. Thanks for the contour advice - I'll give it a try.

[Josh]

Originally Posted by crmorton

"Any further suggestions/comments as to why my drag values are too low?"

For most transient simulations I have completed on a uniform cylinder, it takes a great deal of simulation time to reach a state of not only periodic vortex shedding, but quasi-steady periodic vortex shedding (meaning the drag fluctuations in time ocillate about a mean drag value over the period of 20 or 30 vortex shedding cycles).
Make sure in your transient simulation that you are saving enough time steps to properly re-produce the drag signal (do not under-sample)

Also, for a nice visualization of the flow, try contours of vorticity, using a log-log scale, with 50 contours or more. It takes a long time to load but looks very nice.

Cheers.

Just to verify...

Do you create a vorticity contour by first creating a vortex core region, then specifying the defining variable as vorticity? Or do you create a regular contour plane and specify the defining variable as velocity.curl? Or am I way off?

I ask because I've tried both with mediocre results.

[Tristan]

Josh,

I use the appropriate component of Velocity.Curl to look at vorticity since that's how vorticity is defined. What do you mean by 'mediocre results'?

Tristan

[Josh]

I shouldn't have said mediocre. Poor choice of words. It's not inaccurate, just not as aesthetically pleasing as I expected, even with 100 contours.

I'd post a picture, but I'm running a simulation!

[crmorton]

I insert an expression for the vorticity in a particular direction: e.g., Velocity v.Gradient X - Velocity u.Gradient Y.
Sorry if the images aren't coming out as nice as you would like. It may have looked nice for me since I was using a very high mesh density in the far wake region.

I am not sure how it would turn out, but you could also try writing expressions for the second invariant of the vorticity tensor, in other words, Q criterion. Q = S^2 - W^2, where S is the strain rate and W is the vorticity tensors I think. If you are using ANSYS V12, they have included this Q criterion as one of the variables for plotting a "vortex core region"

[vmlxb6]

Did you get your problem solver ??????????????

[bbhv]

Hi Josh,

I know it has been a long time since this thread was active, but I hope you do not mind to give some hints.
My results also show much lower, around 50%, Cd and Cl than the reference data.
My object is not a cylinder, but I think you can still provide general hints or tips on how to improve the accuracy of simulation results in CFX.

Thanks,
bbhv

[Josh]

Hi ST,

Have you performed sensitivity analyses of the grid resolution, i.e., refine until the desired results no longer change? How about grid quality improvements? Timestep refinement? Convergence residual refinement? Domain size?

[bbhv]

Hi Josh,

Yes, I have done time and grid independence studies.
Refining the time or space only slightly improve the results, but the error is still large 40-50%.
Can you explain more the influence of thickness that is added to simulate 2D model?
and, did you manage to get an accurate results for your case?

Cheers,
bbhv

[ghorrocks]

Doesn't this suggest that you need to increase your grid and/or time resolution by a lot more than you have tried so far? How much did you change the mesh and/or time step size?