[sasanghomi]

Hello guys,

I am trying to simulate the fluid flow over a cylinder and calculate convective heat transfer over a cylinder with constant temperature.
The problem is that my result is not in harmony with some empirical correlations (churchill & Bernstein)
Incidentally, my turbulence model is K-W SST and the grid is refined perfectly.

Does anybody know how I can improve the accuracy of my results? Is there any turbulence model advanced options that open doors to this discrepancy?

Your help is appreciated deeply in advance.

Best Regards,
Sasan

[Gert-Jan]

What is your Y-plus?

[sasanghomi]

Originally Posted by Gert-Jan

What is your Y-plus?

The value of Y-plus is about 5-10.
Should it be below 1?

Any suggestions? What about discretization methods? They are put on high resolution. (advection & turbulence)
Do you think upwind method could come in handy in this case?

Best Regards

[Gert-Jan]

It depends on your Prandtl number. If you use water at 20°, then you Prandtl number will be around 5, meaning that your thermal bundary layer is 5 times as thin as your hydrodynamic boundary layer. So, to capture your thermal boundary, you need a fine mesh. Finer than required from hydrodynamic point of view.

To be on the safe side, make Y-plus everywhere below 1 and keep on using SST.

Never ever use the word 'Upwind' again.

[sasanghomi]

Thank you so much for your explanation.
Air (at 25 C) is regarded as a fluid passing through a cylinder with 80 C wall temperature.
What is your idea about the advanced options for SST model? I mean using "Reattachment Modification". It adds a source term to K-equation. What is your idea about using this option for such simulation? It seems that this option corrects the stress comes from turbulence after separation.

[Gert-Jan]

I am not a real turbulence expert. So, I don't know. I suggest to raise a new question on this topic. Higher chance of getting a valuable answer.

[ghorrocks]

Note that all the options in CFX (such as the reattachment modification) have problems and are not universally applicable. If they were they would be the default option. You need to understand these options to determine whether they apply to you.

[sasanghomi]

Thank you so much for your response. There is still one more question in my mind. Is it right to simulate the fluid flow around a cylinder as a laminar flow when Reynolds number is around 1000 (even though there are vortexes around the cylinder)?

[Gert-Jan]

Originally Posted by sasanghomi

Thank you so much for your response. There is still one more question in my mind. Is it right to simulate the fluid flow around a cylinder as a laminar flow when Reynolds number is around 1000 (even though there are vortexes around the cylinder)?

In my opinion: no

[ghorrocks]

Sorry Gert-Jan, you are wrong here.

At Re=1000 flow over a smooth cylinder is laminar, so a laminar model is the correct approach here. Note that it will be a transient flow with a vortex street. But - and this is a vital point - a transient flow with vorticies is not necessarily turbulent! Turbulence is a chaotic flow with a wide range of length and time scales. At Re=1000 a cylinder will produce a vortex street with a fixed frequency, and therefore is just a limited range of time and length scales.

Flow over a smooth cylinder at Re=1000 is not turbulent! Turbulence transition occurs at around Re = 5.0E5 (ref: https://www.princeton.edu/~asmits/Bi...web/blunt.html)

[sasanghomi]

Dear Glenn Horrocks,

Thank you, I have always benefited from your comments. Could you please give me hints for converging the flow around a cylinder?
In fact, my problem is that the imbalance of H Energy oscillates unexpectedly. I am simulating the fluid flow around a cylinder whose temperature is 80 C and inlet flow stands at 25 C.
I have refined the grid perfectly, however I have serious convergence problems and I need to capture the convective heat transfer coefficient around the cylinder.
Do you have any suggestions? Is it a really problematic simulation or I am making any egregious mistakes?
By the way, Reynolds number is about 5e3 and the flow is considered laminar in my simulation.

I appreciate your attention.

Best Regards

[ghorrocks]

Previous comments have been assuming an isothermal flow. If heat transfer is occurring that may change things. Is natural convection significant in your case?

[Gert-Jan]

Here are my 50 cents to take into consideration, again.

- Take a step back and first make sure your Strouhal number is correct. Is this the case? If so, then look at heat transfer. First things first.
- You refer to convergence. I presume you mean in a transient calculation, and convergence within a time step? What is your time step? How many coefficient loops do your take? What is your convergence criterium? Look at the FAQ.
- If your Re is 5E3, how can you model this as laminar. In the boundary layer, ok, but in the wake............? In my opinion you should use a low-Reynolds Turbulence model, like SST or k-w (Wilcox).
- What do you define on your inlet? No turbulence at all? Is that valid? Does it fit your measurements/application? Or is it just a hypoothetical case?
- Do your fluid properties depend on temperatures? if you demand (very) accurate results (especially for a hypothetical case), this might be relevant since your delta T is quite large.

(The title of this thread is Convective Heat Transfer Coefficient (Flow over cylinder). So heat transfer is relevant from day 1.)

[ghorrocks]

- If your Re is 5E3, how can you model this as laminar. In the boundary layer, ok, but in the wake............?

The wake will not be turbulent either. There is not enough energy at this Reynolds number to overcome viscous dissipation at any location in the flow so nowhere is turbulent. A laminar flow model is the correct model for this simulation.

Good point on the title. I missed that Too many threads running at once to keep track of things.

[Gert-Jan]

My advice is worth only 50 cents......

[sasanghomi]

Originally Posted by ghorrocks

Is natural convection significant in your case?

Thank you for your response. However, I am comparing my results with some empirical correlations in which the effect of buoyancy force is omitted mostly. (This is my guess)

[sasanghomi]

Let me clarify my experience in this case which may be helpful for others. I used laminar model and did not use turbulence models (Re=5.32 E3). Moreover, I increased dynamic viscosity tangibly and kept the Reynolds number constant by justifying inlet velocity. Also, Prandtl number was kept constant by changing conductivity. The point is that the process of convergence got much better and the vortex street took shape easier.
The results shows that Nusselt number is around an acceptable range (comparing with churchil correlation), however the drag coefficient is a bit high.(it is oscillating around 1.5 but it is supposed to be close to 1)

Best Regards

[Gert-Jan]

Did you determine the various dimensionless numbers. Like Grashof and Rayleigh number. Based on these numbers you should be able to determine whether if buoyancy is important.

[sasanghomi]

Thank you,
I have another question about this modeling,
Is it correct to simulate Turbulent fluid flow around a cylinder with a two dimensional geometry?

P.S. I know that in CFX there is no 2D solver but we can handle 2D simulations by setting 1 cell in the third direction.

Best Regards

[ghorrocks]

A turbulence model will function properly in a 2D model.