[Anna Tian]

Hi,

In CFX, we can't use different turbulent model to different fluid domain. We have to choose only one to all of them. But for my case, some of my fluid domain have laminar flow and the others are turbulent. So I have to use turbulent model to not only the turbulent part but also the laminar part. May I ask will that bring any inaccuracy?

Thank you very much.

[ghorrocks]

Yes, it will cause an inaccuracy. It can be minimised by using a turbulence model which degenerates nicely to zero turbulence. k-e get a divide by zero error at zero turbulence (unless it has a low Re correction), but SST and k-w work fine with zero turbulence.

So I recommend you use the SST or k-w turbulence models as they handle turbulence levels of zero (or very low levels of turbulence) gracefully, and will only add a small amount of dissipation to the model.

[evcelica]

I believe you can apply different turbulence models to different domains.
EDIT > OPTIONS > GENERAL > ENABLE BETA FEATURES > then uncheck CONSTANT DOMAIN PHYSICS

But maybe you don't need to if you take Glenn's advice.

[ghorrocks]

I remember years ago I compared a flow which was quite laminar with a simulation using laminar and SST turbulence models. The SST model ran fine and the results were close enough to be useful. But I recommend you check this for your configuration.

I do not think you can use non-constant domain physics if the domains are connected. The interface will not be well specified and I am sure the simulation will crash. But it might work if the domains are not connected.

[Anna Tian]

Originally Posted by ghorrocks

Yes, it will cause an inaccuracy. It can be minimised by using a turbulence model which degenerates nicely to zero turbulence. k-e get a divide by zero error at zero turbulence (unless it has a low Re correction), but SST and k-w work fine with zero turbulence.

So I recommend you use the SST or k-w turbulence models as they handle turbulence levels of zero (or very low levels of turbulence) gracefully, and will only add a small amount of dissipation to the model.

Could Reynolds stress model also be okay for laminar and transitional flow?

[Anna Tian]

Originally Posted by ghorrocks

I remember years ago I compared a flow which was quite laminar with a simulation using laminar and SST turbulence models. The SST model ran fine and the results were close enough to be useful. But I recommend you check this for your configuration.

I do not think you can use non-constant domain physics if the domains are connected. The interface will not be well specified and I am sure the simulation will crash. But it might work if the domains are not connected.

What parameter did you use to compare these two models' simulation results? Could SST gives a good estimation of the wall friction loss? Could it resolve laminar boundary layer well?

[ghorrocks]

Could Reynolds stress model also be okay for laminar and transitional flow?

No. The issue is what happens when the turbulence level goes to zero. The important thing is the behaviour of epsilon/omega. The turbulence energy is simple, it just goes to zero. But epsilon is badly behaved as turbulence goes to zero but omega is.

What parameter did you use to compare these two models' simulation results? Could SST gives a good estimation of the wall friction loss? Could it resolve laminar boundary layer well?

I cannot remember the model, but it simply compared a laminar and SST simulation for a flow which was actually laminar. SST should be able to resolve most laminar flow features, but it will have more dissipation than a true laminar model so you will need to validate it carefully. Or even better use a laminar flow model.

[ghorrocks]

Could Reynolds stress model also be okay for laminar and transitional flow?

No. The issue is what happens when the turbulence level goes to zero. The important thing is the behaviour of epsilon/omega. The turbulence energy is simple, it just goes to zero. But epsilon is badly behaved as turbulence goes to zero but omega is.

What parameter did you use to compare these two models' simulation results? Could SST gives a good estimation of the wall friction loss? Could it resolve laminar boundary layer well?

I cannot remember the model, but it simply compared a laminar and SST simulation for a flow which was actually laminar. SST should be able to resolve most laminar flow features, but it will have more dissipation than a true laminar model so you will need to validate it carefully. Or even better use a laminar flow model.

[QCFD]

Originally Posted by ghorrocks

Yes, it will cause an inaccuracy. It can be minimised by using a turbulence model which degenerates nicely to zero turbulence. k-e get a divide by zero error at zero turbulence (unless it has a low Re correction), but SST and k-w work fine with zero turbulence.

So I recommend you use the SST or k-w turbulence models as they handle turbulence levels of zero (or very low levels of turbulence) gracefully, and will only add a small amount of dissipation to the model.

I realize this is an old thread, but I have recently run a simulation back to back with no turbulence model (laminar) and SST for a case that is transitional (laminar in most of domain but turbulent in other places.

I have observed a few things:

-the laminar solution is significantly more grid-sensitive in terms of pressure drop

- the distribution of flow between several potential flowpaths is different (by up to 10%) using SST than laminar

I think both of these come down to near wall treatment. In the bulk flow I understand SST is just going to add turbulence viscosity, which if tke is low (which i checked and it is essentially 0 in these regions) will be small. However, I am not clear on the near wall behavior of laminar simulation in CFX. It seems to me the wall shear will change continuously with mesh refinement at the wall, whereas the SST model gets around that with wall functions.

That is fine, but then what are the impacts I should expect from applying turbulent wall functions on a laminar problem? Does anyone have a good feel for this?

[ghorrocks]

Interesting comparison, thanks for posting it.

I had never thought about the relative grid sensitivity of laminar versus turbulent flow, but I guess laminar flows are more grid sensitive. The reduced dissipation in the laminar model means errors in the mesh are more apparent.

And it is not surprising that this leads to small flow differences (10%).

Laminar flow integrates to the wall by definition. There is no such thing as a wall function for laminar flow. And at low Re SST will be integrating to the wall as well. But as you correctly say, SST will be increasing the viscosity with the turbulent component. Other more subtle things are happening as well - I can't remember the exact option but I think the default pressure interpolation scheme is different for laminar versus turbulent flows. The laminar one is more accurate but less stable. So there is more changing than simply the addition of a turbulence model.

No, in a well meshed laminar flow the wall shear does not change with mesh refinement. The tangential speed of the first node will change (as the distance from the wall changes), but the wall shear remains the same.

So adding a turbulence model to a laminar flow will cause the flow to be damped a little more than it should. The wall function difference is not important (if you are using a scheme which integrates to the wall).