Zero equation turbulence model gives better result??
 

Hello All,

I am doing flow simulation of 90 Deg. duct bend. Fluid is air at 500 Deg C. The area of duct upstream is low while that of downstream is quite more. Hence the flow separation is taking place. When I tried with k-e, SST & RSM model, flow is packing towards the outer wall of the bend, resulting in high velocity in that part. Though this is practical, the separation region is over-predicted (Separation region is more in area). Hence the velocity values are also over-predicted (as more quantum of fluid is passing through less area).

When I tried with zero equation turbulence model, though separation region is under-predicted; the velocity values are near to ground data (as flow is more distributed). This obviously confused me because as per my knowledge, 2 equation models are more accurate than zero equation model.:confused:

I can not use zero equation model throughout the domain, as there are some turbulent dominant areas in the domain.

Is there any intermediately accurate model between zero equation & SST?

I would appreciate all the help I could get.

Thanks in advance.

Regards,

--Sanyo

I like the logic - zero equation is underpredicted, two equation is over predicted so a one equation model should be about right. Unfortunately it does not work that way. Predicting separation lengths is very tricky for turbulence models. I think your best bet would be to use SST and activate some of the advanced features, I think some of these include models to improve separation lengths. I cannot remember the names off hand, hopefully people can recall the options I have in mind.

If this does not work then you should consider DES, LES or SAS approaches.

Hi Glenn,

I would like to admit that your replies always help me a lot & would like to thank you for the efforts you take for CFD community.

Probably I missed to convey my message clearly. I didnt want to say that zero equation model under predicts & hence its more accurate. I just found that two equation model or even RSM somehow was not able to capture the flow distribution at bend (where flow changes its direction at 90 Deg.) & giving the much different velocity values against ground data. On the other hand zero equation model gave the results much closer to ground data. I just wanted to know, if I was missing something in turbulence modeling.

I also came across to the low Re K-e model of Fluent. I didnt find such model in CFX. I also read that K-w could be such model, but I didnt find the results promising. So I was wondering if such model could be implemented in CFX?

Do you suggest to use the advance feature of SST such as "streamline curvature correction model"? In that case, how to determine the values of constants?

Thank you for your help.

Regards,

-Sanyo

I would not use any of the unusual turbulence models like low Re k-e or k-w unless you have a good reason to do so. The SST model is the best all-purpose turbulence model in CFX by far so use that one. Additionally the SST model has the extra options of things like curvature correction which the others do not.

No need to adjust the constants in the model. Leave them at the defaults.

What Re is the duct flowing at?

Hi Glenn,

Thanks for the reply.

The Re at duct is 9.5e+5.


Regards,

-Sanyo

OK, mid level turbulence. SST should be as good as anything for this flow regime.

Hi Glenn,

Sorry for late reply.

Can you suggest what model to be used if Re is around or less than 35000?
In my case, for another set of operating conditions, Re is 37000.

Thanks in advance.

Regards,

-Sanyo

SST is the best general purpose turbulence model.

Thanks a lot.

I'm not sure if this will be helpful to your situation or not, but I experienced what sounds like a similar problem with turbulent flow over a backward facing step. The reattachment point didn't correspond to experiment and, upon further inspection, the cross-section flow profiles were "off" as well. The problem turned out to be convergence-related. The default RMS value of 1E-4 was not sufficient for achieving a fully converged solution.

Hi jtipton2,

Thank you very much for your reply.

I will definitely try to achieve more stringent convergence norm.

Regards,

-Sanyo