CFD Online Discussion Forums

CFD Online Discussion Forums (
-   CFX (
-   -   Swirling flow in a diffuser: K-E over SST ? (

Felix February 27, 2007 10:48

Swirling flow in a diffuser: K-E over SST ?
Hello to you all CFDers,

I am modeling an axisymmetric diffuser with an inlet swirling flow. Experimentaly, the flow did not separate from the wall nor did it present a recirculation zone in the core. So far, the K-E model results I get are much closer to the experimental data than the SST ones (although not perfect) and that gets me a litle confused. In fact, the SST predicts a separation early in the diffuser causing a jet-type flow field and thus cannot predict anything correctly afterwards.

The case I am investigating was studied experimentaly by Clausen and numericaly by Armfield (late 80s/early 90s) and has a total divergence angle of 20 degrees. All the data is available on the Ercoftac database.

I haven't proved mesh independence yet but I took care of using Y+ < 2 for the SST and 25 < Y+ < 40 for the K-E. Has anyone ever simulated this case in SST ? What were the results ? How did you get them ? I'm getting more and more curious about the real predictive ability of the SST turbulence model.

Thanks to all, best regards,


matej February 27, 2007 12:34

Re: Swirling flow in a diffuser: K-E over SST ?

I did not modelled this case, but I've just spend all day putting together the results comparing k-eps and SST with experiments on swirling combustor and it seems to me, that non of the models is better than the other one. While tangential vel. profiles are closer to the measurements for SST, the size of recirculation zone is to my surprise better predicted by k-eps. not easy to decide.

maybe SAS has the answer....?


HekLer February 27, 2007 18:44

Re: Swirling flow in a diffuser: K-E over SST ?
Generally the two equation models on their own are not particularily suited to strongly swirling flows.

SST is a blend between k-epsilon and k-omega. If you throw the 25<y+<40 grid at SST it should behave similarly to k-epsilon.

That being said the early separation must be a result of the eddy viscosity prediction being less (flow is not as thick) which is tied to prediction of the local pressure gradient and mean velocity gradients.

One other thing to do is contact the CFX help desk and ask if this case has been run by them. It may have been.

Glenn Horrocks February 27, 2007 22:03

Re: Swirling flow in a diffuser: K-E over SST ?

k-epsilon has well-known deficiencies in modelling swirling flow. SST has some terms in it to account for swirling flows but if you want to model swirling flow with a RANS approach you should really consider the Reynolds stress models.

Glenn Horrocks

All times are GMT -4. The time now is 13:48.