Hi I am trying to model a swirling flow with a expansion and a contraction (in a circular geometry at Re=10000) with the standard k-e model and it's different versions (rng, non-linear etc). The agreement with measurements is quite poor. But I have read about something that is called "swirl correction" terms, and I wonder what that is. Does someone know if there is anything published with a swirling flow and "swirl corrections" terms? Thanks in Advance / Ulf

Any turbulence model that uses the eddy viscosity assumption as a model for the Reynolds stresses will rapidly drive a strongly swirling flow towards solid body rotation. There are a variety of kludges available but they rarely introduce more physics and want viewing with a reasonable degree of caution. They may well work better for the cases they were setup to address and very much worse for cases they were not (more physics does not have this property).

The next model up that does introduce more physics is a Reynolds stress transport model. This does not suffer the solid body rotation problems and can predict streamline curvature effects on turbulence reasonably well. The curvature effects on the convection/production terms of the Reynolds stresses are now exact and not modelled as they would be with an eddy viscosity model. This was one of the major driving forces behind the developemnt of the model during the 70s and 80s (things died down a bit after that). Unfortunately, you may have to use a "swirling flow" set of coefficients rather than a "boundary layer" set of coefficients to get good agreement. Several of the commercial packages have implementations and I would strongly recommend you give it a go. The improvement will almost certainly be significant.

The next model up again is LES. This is computationally expensive and it is often difficult to generate reasonable boundary conditions. It has given excellent results for swirling flows. Last time I checked there were no sensible implementations in commercial codes but this may have changed in recent releases.

Try: Analysis cntr > other > misc. > Blending factor . I have flow with swirl number >1.6, the best agreement with measurement was for "swirl,0.05".

Sorry, wrong forum.

Hi for swirling flows free or forced there are two non-dimensional groups that govern the status of the flow namely the Reynolds and circulation number. For Re between 0 and 400,000 vortex bursting can occur depending upon the circulation number defined as (circulation/bulk velocity*pipe diameter)therby giving rise to severe flow gradients so check out the circulation number and don't bother with the RNG-k-e model as it is formulated for weakly swirling flows.

Regards

k-e models are not suitable for hihgly swirling flows at all. You should use Reynolds stress model instead which gives excellent results. RSM predicts very well streamline curvatures and turbulence anisotropy which is the case of the swirling flow.

I have additional questions on swirling flows.

In rotating equipment at turbulent conditions one can distinquish random turbulent fluctuations and periodic turbulent fluctuations.

Are the periodic fluctuations resolved by either the Reynolds-Stress or k-epsilon-Models in a time dependent calculation or do we require LES/DNS for this?

If they are resolved is RSM still better (I think so)?

If they are not resolved, to what extent can we rely on the calcluations?

Regs, Astrid

The separation of coherent motion from turbulent motion in a RANS simulation depends largely on the respective time scales. If the coherent motion is significantly slower than the energy containing turbulent motion it is reasonable to use a RANS model. If not one must use an approach such as LES/DNS which can simulate both types of motion. There is no difference between k-e and RST in this respect since they both use Reynolds averaging.

If the coherent motion is amenable to removal by transforming the governing equations one might make some progress taking the route. However, I suspect the additional terms in the turbluence transport equations would prove difficult to model.

Andy,

Thank you for answering. What do you exactly mean by 'taking the route'? LES/DNS?

Astrid

Not really. I was refering to transforming the governing equations in order to change the residue of "time averaging" from the Reynolds stresses into something else. For example, in variable density flows it is common to density weight before time averaging (Favre averaging) in order to get terms more amenable to closure. In this case one could consider, for example, inclining time in the blade to blade direction in order to replace the strong spatial variation of the wake with an effectively constant inlet "velocity" which rises and falls through time. This would largely remove the coherence from the "Reynolds stress" like terms (would still exist in the streamwise direction) making an eddy-viscosity type closure more reasonable. However, we do now have a solution which varies over the time between wakes (like the flow in a piston engine), the terms to be simulated by the turbulence closure may be odd and the governing equations and boundary conditions may be odd.

I strongly suspect the cons are going to outweigh the pros and now regret having made the comment in the first place!

Hi,

I just finish some simulations of swirl flow cases with FLUENT5 software. It can say that both k-e and RSM are unsuccessful to simulate strong swirl flow. There are large errors compared with experimental data.

The errors are from two sources: 1.numerical error; 2. turbulent model. In fact the former provide a large contribution for errors. However, they are always negelected in the past years. A higher order scheme should be employed, especially for 3D problems to reduce the numerical diffusion. The latter also is important to simulate correctly cases. RSM is not a general validated model. General speaking, it is better for swirl flow because it consider the anisotropy. However, if the geometry or flow field is complex, it is possible that RSM can not correctly simulate the flow. Particularlly, for 3D problems when you employ a higher order scheme, the computing become very expensive. LES has same problem.

k-e is not suitable for swirl flow in its standard model. In fact, its model constants are from the plate flow experimental data. Hence, if it is used for swirl flow, it should be curve modification. Generally, a Richard number is introduced for correction of curve. You can refer Launder's reference:ASME J of fluid ENg. 1977, 99:231-239. If you use the commercial code, you can modify the model constants.

Good luck.