
[Sponsors] 
January 6, 2012, 07:00 
Kepsilon vs. komega turbulence modelling

#1 
New Member
Join Date: Jan 2012
Posts: 3
Rep Power: 13 
Hello,
I am currently working on a heat transfer model involving strong vortical cooling effects in the pressure based solver. To reduce computational expense, the turbulence models of choice are the realisable kepsilon and SST komega two equation models, with the aim to fully resolve the boundary layer with a suitably defined fine mesh density. These two models have returned sizeable numerical differences. I am trying to understand why this could be. Could anyone give any information regarding these turbulence models for this particular case? Cheers Tom 

January 14, 2012, 01:45 

#2 
Senior Member
Lucky
Join Date: Apr 2011
Location: Orlando, FL USA
Posts: 5,112
Rep Power: 60 
The SST komega model reduces to a standard kepsilon model away from walls.
Near walls, the SST komega uses the the omega formulation whereas the realizable kepsilon uses the epsilon formulation. The two models differ in their approach both near walls and away from walls. The SST komega model is generally thought to be better near walls, but because it reduces to the standard kepsilon model away from walls and not the realizable kepsilon model, it also has weaknesses. The two models are very different and it should be no surprise that significant differences in results can be obtained from each model (this should be considered the norm rather than unusual). 

January 15, 2012, 14:38 

#3 
New Member
Join Date: Jan 2012
Posts: 3
Rep Power: 13 
Thank you for your input.
You have alerted me to one point regarding the turbulence models I have used; that being the SST komega model which has a blending function to reduce the model to the standard kepsilon model. I overlooked this feature, and was expecting that the solutions provided would support the application of the SST komega model, given its suggested superiority in the near wall flow regions. It would be good to gauge what turbulence models other users of Fluent have adopted to capture steady state heat transfer, particularly when flow characteristics such as separation, reattachment, and recirculation have a strong presence. 

January 24, 2012, 08:13 

#4 
New Member
Avinash
Join Date: Jan 2011
Posts: 18
Rep Power: 14 
Hi,
I have used SST komega model, Realizable and RNG kepsilon models for a similar problem involving heat transfer in a duct (internal flow) with a lot separation, reattachment and recirculations at nearwall regions. Significant differences in wall temperatures are seen between the turbulence models. Unfortunately very high temperatures are obtained with the usage of SST komega model compared to the kepsilon models, which has resulted in badcomparison with the results from the literature. This finding is contrary to other numerical simulation papers. The recirculation size in case of SST komega model is also very different from the kepsilon models. y+ less than 1 was maintained for all the cases. The reason for high temperatures with the usage of SST komega model evades me. Please let me know of your suggestions, as to what can be wrong. Regards Avinash 

February 7, 2013, 00:42 

#5 
New Member
Join Date: Oct 2012
Posts: 16
Rep Power: 12 
Hi,
did you find the reason for the wrong behavior of the SST? I recognized the same behavior when applying the SST in separated flows (although in OpenFoam) getting to high wall temperatures and a significant underestimation of the Nusselt number respectively. Regards, Pascal 

May 20, 2016, 04:56 

#6  
Senior Member

Quote:
1. Nu is poor (kw SST )  good ( kepsilon (real) ) 2. Friction excellent (kw SST )  poor ( kepsilon (real) ) 

May 21, 2016, 13:27 

#7 
Senior Member
Lucky
Join Date: Apr 2011
Location: Orlando, FL USA
Posts: 5,112
Rep Power: 60 
I understand why everyone would like to compare Nu and f for two models, because it feels like a low hanging fruit.
For separated flows, kepsilon and kw SST models can and generally do give wildly different results. Don't just compare Nu and f. To appreciate the difference, just compare the separation patterns of the two models. Generally, kepsilon predicts a much weaker separation than kw model. In some cases, kepsilon predicts no separation and kw model predicts a VERY large separation bubble. Flow separation is considered by many (to be a primary flow pattern, as opposed to secondary flow). When your simulation is missing something as important as a separation (or contains a separation bubble when it shouldn't), then it should be no surprised to anyone that it also predicts the wrong Nu and f. 

May 22, 2016, 04:35 

#8 
Senior Member

Dear Lucky,
Thank you for your kind remarks. You are right that separation is well captured in case of kw SST but my geometry is a bit complicated. Its a pipe with grooves in its thickness. Those grooves are helically extruded into the pipe thickness. I have modelled 1/10 th of geometry with periodic conditions side wise. There are total 10 grooves. Since I was getting a lot of difference from experiment in case of k omega at high Re. I checked the model constants, I modified the Energy Prandtl number and Wall Prandtl number from 0.85 to 0.35 and got my results very close to experiment. It seemed logical to me because its the wall Pr number that controls the thermal boundary layer. So decreasing this value will not increase the wall temperature much which should be logical because at high Re the flow conduction time is not much as compared to the flow velocity and so the fluid is not well heated and the delta T , the conduction temp and the bulk temp diff should be less. Therefore, h will be more and hence the Nu will be higher. I hope you will agree with me. 

May 22, 2016, 04:36 

#9 
Senior Member

Dear Lucky,
I can check the separation but the issue is that the geometry is not consistent in axial direction (z drn) all the way long. cutting a plane at a particular location may/may not give true separation region. 

May 22, 2016, 11:00 

#10 
Senior Member
Lucky
Join Date: Apr 2011
Location: Orlando, FL USA
Posts: 5,112
Rep Power: 60 
I agree that if you change some model constants that shouldn't be changed, that it can can an effect on the results. Yes.
Yet again there is no surprise that you can change the turbulent Prandtl number and force Nu to make your experimental result. Remember von Neumann's infamous statement. With four parameters I can fit an elephant, and with five I can make him wiggle his trunk. Interestingly, you set the turbulent Prandtl number to ~0.3 when many experiments suggest that typical turbulent Prandtl number in ribbed/grooved pipe flow is closer to 0.9. You can tune any number of parameters to get a desired surface result but they are all meaningless from a CFD perspective unless you consider, verify, and validate the entire flowfield result. I can also tune a random number generator to give me any desired number. A lot of people also tend to miss a very important problem. The fact that there is a turbulent Prandtl number. There is also a closure problem on the energy equation (just like the momentum equation) that needs additional models. If you dig into it, you'll realize how little development has been done on the closure model for the energy equation. I.e. what models can you use to predict the correct local turbulent Prandtl number, the same way we try to predict the local eddy viscosity. For ribbed flows, the work at ITLR Stuttgart comes to mind with rather impressive results. At the end of the day, why is anyone surprised that you predict the wrong heat transfer when there is this magical missing model to close the energy equations in the first place? 

May 22, 2016, 11:17 

#11 
Senior Member

thanks Tran
Could you please give me the reference for turbulent Pr of 0.9 for ribbed flows? I am now worried a lot because my supervisor has asked me to publish a paper and now I do not have results matching with experiment (with 0.85 Pr) , in some cases kepsilon is good. But how do I justify because globally kepsilon is not recommended for such flows. [] 

May 24, 2016, 20:46 

#12 
Senior Member
Lucky
Join Date: Apr 2011
Location: Orlando, FL USA
Posts: 5,112
Rep Power: 60 
Hi, I am assuming you are working with air.
The laminar or molecular Prandtl number for air is 0.7. If turbulence enhances heat transport (via turbulent fluctuations) then how can the turbulent Prandtl number be less than 0.7? Although this argument is not perfect (it assumes Boussinesq hypothesis roughly holds), any turbulent Prandtl number less than 0.7 is nonphysical. Even where the Boussinesq hypothesis does not hold exactly, everyone can appreciate that advection does not decrease the total rate of heat transfer to below that of pure conduction. The exception is the extremely rare case when the velocity fluctuations are perfectly out of phase with the turbulent fluctuations (which generally occurs only if there is a forcing function). 

May 25, 2016, 06:19 

#13 
Senior Member

Hi there,
1. No, I am working with water and solid material is Stainless Steel. It seems a bit harder to absorb your point of advection that it does not decrease the total rate of heat transfer to below pure conduction. Do you mean that the ratio of Pr is related to adv/cond, and you may be referring to that? 2. Could you please provide reference for your statement, that "many experiments suggest that typical turbulent Prandtl number in ribbed/grooved pipe flow is closer to 0.9." 

Tags 
heat transfer, kepsilon, komega, turbulence 
Thread Tools  Search this Thread 
Display Modes  


Similar Threads  
Thread  Thread Starter  Forum  Replies  Last Post 
SST K omega turbulence model  mb.pejvak  Main CFD Forum  8  September 16, 2011 09:52 
Homogeneous Turbulence Modelling  Zaktatir  CFX  0  December 15, 2010 10:37 
SimpleFoam k and epsilon bounded  nedved  OpenFOAM Running, Solving & CFD  1  November 25, 2008 21:21 
a book for turbulence modelling  Prapanj  Main CFD Forum  1  August 21, 2007 06:13 
Multigrid Method in Turbulence Modelling  Bipin Lokhande  Main CFD Forum  1  November 11, 1998 20:20 