I am having trouble with a calculation. The fluid is air. The velocity is max 10m/s and much lower in the plenum. It is a design with a tangential inlet into a plenum chamber. The air is then guided from the plenum to the outlet tube via 3 x 90° turns. It is not at cyclone, but the flow does rotate in the plenum.

A physical model has been made and measurements have been done. And the calculations are not the least comparable to the measurements (axial outlet velocity in segments). The measurements show that the segment axial velocity deviation is approx 3% and the calculation say approx 50%. This is a huge difference.

I use RSM.

Can anyone comment on possible errors in setup? I know that it impossible for you to debug my calculation, but if you could come with some general suggestions/thoughts related to rotating flows.

If you need more information regarding the design please let me know, and I will provide them.

Thank you in advance.

There are a wide range of possibilities here:

* the measurements are in error. A problem I have now seen withseveral experimental groups is to use an inappropriate reference static pressure for the probes in the presence of swirl. Is this the case here?

* the predictions are in error. Is the solution predicting solid body rotation when the profile should be relatively flat? The culprit here may well be numerical diffusion if you are using a Reynolds stress transport model. Are you using tetrahedrals?

Is your RST model converged? The pressure gradient in swirling flows makes the momentum equations stiff and the axial profile can take many thousands of iterations to converge. A typical convergence profile will show a steady drop to a lowish plateau, followed by a very long flat period during which the axial profile develops slowly followed, eventually, by the residual dropping again and heading towards roundoff.

andy gave you a bunch of good advice, let me add one thing - your case sounds quite complex. Did you try to validate your code/model on a more simple swirling flow case first. You can find links to cases in the Resources/References/Test Cases and Databases section of CFD Online. For example, the following looks promising:

ERCOFTAC Nexus Classical Database, Confined Flows, Swirling Flows in Ducts & Pipes

As I'm sure you know swirling flows are notriously difficult to predict so don't be supriced if you will not get good results in the end. This is one of the few applications where I would use an RSM model.

Hi Andy. Thank you for your detailed discussion.

* I don't thin that the measurements are inaccurate. I have done several follow-up measurements, and they all show the same. I use a (thick = approx. 0.5mm) hotwire anemometer. I only measure the axial velocity, as the hotwire is semi-encapsulated and therefore only good for bulk flow (the outlet is small, and I think that capturing components other that the axial may be hard with the equipment).

* I will look into the solid body vs. flat profil. Thank you. I use mainly hex'es, only in the tangential inlet area is there tets, and I have made this area as small as possible. I tried using tets for the whole design, but it could not converge. A problem I have seen before in swirling flows. If RSM/RST can be a problem, what would you suggest? I would think that k-e would be unable to capture the swirl correctly.

* Well. I think it is converged. It has run for several thousand itt, and the residuals has flattened out. It is however not a straight line. It is horizontal (or vertical when I turn the monitor with some small-amplitude "white noise" on the line. It has been flat for several hundred itt. I also monitor the outlet axial velocity in two places, and they are both flat AND straight.

Jonas: Thank you for your input. I will look closer on the database you referred to.

Do you have a significant swirl component at your exit or is the axial velocity the dominant component? (concerned here about the semi-encapsulation).

The business about solid body rotation (swirl velocity is proportional to radius making the flow spin like a solid) is that it is an indicator of the presence of diffusion (modelled or numerical). A turbulence model which uses an eddy viscosity to model the Reynolds stresses (e.g. k-e) is always going to drive swirling flow towards solid body rotation. A Reynolds Stress Transport model generally will not. However, the presence of numerical diffusion will also drive the flow towards solid body rotation. Tetrahedral elements are usually low-accuracy and a source of numerical diffusion hence the question. Avoiding them is generally a good idea.

Another serious source of problems for swirling flows in commercial codes is pressure smoothing. The "low accuracy" version of pressue smoothing will generate significant mass flow to oppose any deviation from a linear pressure gradient. One can live with this for boundary layers and the like where the pressure gradients are relatively gentle. Unfortunately, for swirling flows the pressure gradients are strong and not linear. A quick to check to see if you have a problem is to look at the cells next to the wall which have a significant swirl component: if the axial flow of the cell next to the wall is not following the wall but turned strongly inward/outward but the axial flow of the cell two away from the wall is sensible then you almost certainly have serious pressure smoothing problems. To be fair, at least one commercial supplier advises one not to use their simple base model for swirling flows because of this problem and advise using one of their more sophisticated/expensive solution procedures instead.

If the residuals have flattened out at a level higher than machine roundoff this is an indication of problems and not that you have reached convergence. It is telling you that something is happening which is preventing convergence. After blowing out the initial transient an RST model with signficant swirl will have a long flat period while the flow field gradually changes via the stiff terms.

Hi Andy.

I am sorry for the delay. I needed time to think about your answer, and to check the calc. I do appreciate your time.

There not much swirl in the physical model. I have not measured it, only a visually evaluation. The calc however, the tangential component are as much as 10%. I don't think that is correct.

I cannot see why a certain amount of solid body rotation is wrong. I thought that it was "normal" close to the axis to have a "solid core". However, it does not seem to be the case in my calc. If the calc shows too mush solid core what will be your recommendation?

The velocity vectors are all pointing in the rotational direction, so I guess that is ok. No funny directions of the flow.

I am curious. What do you mean by more expensive procedures? The k-e cannot capture a rotational flow very well, and is inexpensive. RSTM is more expensive (and more unstable) but should be able to capture the rotational characteristics (well that's what I thought until last week Do you mean LES or are you thinking of modifications to the RSTM?

Do you have any recommendations to steps making the calculation work.

Ps. Funny thing is that I have made an calc for about 1 year ago that is very similar to this one, and it worked fine. I have looked it over to find any differences in setup, but cannot see any. I guess that I was lucky the last time, and that my knowledge of rotating flow needs an upgrade.