Hi.

We are considering the use of CFD. We have a case (an airdistributor) with an outlet Re=10000 at a velocity of 8m/s. The air is guided through the distributor where it changes direction 2 times (2x90° bends).

Can anyone give an estimate of the expected accuracy of a simulation? We have measured the axial velocity out of the airdist. and would like to compare this with CFD calc. for further evaluation. Can CFD provide results like this within, say 2% accuracy? Or is CFD a waste of time?

Thank you in advance.

Christian

From the description of your problem, I'd say: yes, CFD can do this. CFD can do this also within your desired accuracy. Except that the Code can probably not tell you if or when you reached this accuracy. CFD is no black box. If you don't really know what your algorithm is doing, CFD may be a lack of time for you.

Hi Christian,

The Reynolds number in your case is quite high so it is turbulent.I have never worked with a commerial CFD code myself but what I have heard is that they are not very accurate for turbulent flows in general. Even for quite simple geometries like turbulent pipe flow they give significant deviations. I would guess that the accuracy of CFD is closer to 20% than to 2%, 2% is rather optimistic. Even with DNS and LES this accuracy is very difficult to achieve. But perhaps there are some people that find this view too pestimistic.

Tom

A particular part of your question I would be concerned about is whether the 90 degree bends form a "U" or an "S" and are close together or far apart. Big difference in pressure drop/flow pattern in one situation vs. the other.

Christian,

CFD can certainly provide you with the information and accuracy that you need. Let me give you an example.

About a year and a half ago I performed a project for a friend of mine at Target Rock Corporation. They needed help designing a vacuum relief valve with a moderately complex flow path and were not getting the required performance.

I performed several analyses and suggested some design changes before we got to a final design that they built and tested. The flow rate through the valve that I predicted was within 1.5% of the flow rate that they measured.

If you would like more information, feel free to contact me (just take the REMOVETHIS out of the e-mail address above), or get a copy of the paper that we wrote: Reich, Athavale, and DiMeo, "Coupled Fluid Structure Interaction Simulation of the Opening of the Target Rock Vacuum Relief Valve," ASME Pressure Vessels and Piping Conference, July 2001.

Regards, Alton

Two articles that might be interesting for you are: Luo & Lakshminarayana (1997) "Predictions of strongly curved turbulent duct flows with Reynolds stress model" AIAA journal, Vol. 35, pp. 91-98.

Bradshaw, Launder & Lumley (1996) "Collaborative testing of turbulence models" J. Fluids Eng. Trans. ASME 118, pp. 243-247.

That you can predict every kind of flow with an accuracy of 2% is complete nonsense. It is not without reason that people are still working on the improvement of turbulence models or on the development of LES. For example, there is nowadays a strong interest in the modelling of turbulent flows with streamline curvature, the case that you want to study. Or would there still be an interest in improving models even if you can predict it with an accuarcy of 2% with a modern CFD code?

Tom

Well, there are a lot of parameters that influence the solution. In this particular case it might be the right turbulence model and grid density. To reach 2% of accuracy with CFD is a very hard job. It can be done for a couple of problems and it can't be done for more than a couple of problems! My experience is, that for a cold channel flow without any reactions (chemical, combustion or what ever) you can reach easy an accuracy of about 5-10%. To be better needs much more efforts.

Thomas

The Reynolds number in this case suggests us that we are dealing with a turbulent model. I work on the commercial code. I would say that the CFD commercial code might not predict turbulent flow as accurately as we expect CFD to do! There are a couple of turbulent flow modules in the code. You might want to test which one of them might be closer to your experimental/theoretical predictions .