Is the turbulence model really accurate,or it's just by chance?

keryfluid · June 6, 2011, 09:39

Hello everyone,
The turbulence model may get pretty good results for one case,compared to the experiment. But if we switch to a problem with similar flow mechanism, the same turbulence model may get quite bad results. I think some times the error just cancel with each other and we may get a 'good' simulation result by chance. The problem is, how to define if the model really works well or it just happens to work well?
As for simulation of complexed problem, the case available may be quite limited. Sometimes we may even just simulate a single case. In this case I find it really hard to judge which model behaves better with such little imformation. Is there any advice?

Kevin De Smet · June 6, 2011, 13:45

While I don't have the answer to your question, it's very true that wrong answers do look like right answers, and that's more often than we'd like to think. Just with the mesh already you've got two primary mechanisms at work: mesh divergence and mesh dilution.

So you might have a concentration which makes a certain value jump really high but at the same time have the mesh to coarse so it spreads across a wider area. Both yank at the other, and you can have a seemingly valid answer. Dangerous stuff.

Martin Hegedus · June 6, 2011, 14:16

Before one trusts a code or the implementation of a turbulence model, the code and model should be verified and validated. Something along the lines of http://turbmodels.larc.nasa.gov/index.html.

One should understand the flow physics being modeled, the turbulence model itself, and the experiment.

One should also run many simple cases and compare to what is out there to gain experience. Try to break the model, then back off to see where it works.

Yes, RANS codes can give bad results. It is just a model. But many bad results are also due to bad usage or bad practices.

June 6, 2011, 09:39	Is the turbulence model really accurate,or it's just by chance?	#1
keryfluid New Member Kan Rui Join Date: May 2011 Posts: 7 Rep Power: 15	Hello everyone, The turbulence model may get pretty good results for one case,compared to the experiment. But if we switch to a problem with similar flow mechanism, the same turbulence model may get quite bad results. I think some times the error just cancel with each other and we may get a 'good' simulation result by chance. The problem is, how to define if the model really works well or it just happens to work well? As for simulation of complexed problem, the case available may be quite limited. Sometimes we may even just simulate a single case. In this case I find it really hard to judge which model behaves better with such little imformation. Is there any advice?

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
SimpleFoam case with SpalartAllmaras turbulence model implemented	nedved	OpenFOAM Running, Solving & CFD	2	November 30, 2014 23:43
Wrong calculation of nut in the kOmegaSST turbulence model	FelixL	OpenFOAM Bugs	27	March 27, 2012 10:02
Low Reynolds k-epsilon model	YJZ	ANSYS	1	August 20, 2010 14:57
KOmega Turbulence model from wwwopenFOAMWikinet	philippose	OpenFOAM Running, Solving & CFD	30	August 4, 2010 11:26
SSG Reynolds Turbulence Model	Georges	CFX	1	February 28, 2007 17:15

June 6, 2011, 13:45		#2
Kevin De Smet New Member Vertex Wrangler Join Date: Apr 2011 Location: Belgium Posts: 25 Rep Power: 15	While I don't have the answer to your question, it's very true that wrong answers do look like right answers, and that's more often than we'd like to think. Just with the mesh already you've got two primary mechanisms at work: mesh divergence and mesh dilution. So you might have a concentration which makes a certain value jump really high but at the same time have the mesh to coarse so it spreads across a wider area. Both yank at the other, and you can have a seemingly valid answer. Dangerous stuff.

June 6, 2011, 14:16		#3
Martin Hegedus Senior Member Martin Hegedus Join Date: Feb 2011 Posts: 500 Rep Power: 19	Before one trusts a code or the implementation of a turbulence model, the code and model should be verified and validated. Something along the lines of http://turbmodels.larc.nasa.gov/index.html. One should understand the flow physics being modeled, the turbulence model itself, and the experiment. One should also run many simple cases and compare to what is out there to gain experience. Try to break the model, then back off to see where it works. Yes, RANS codes can give bad results. It is just a model. But many bad results are also due to bad usage or bad practices.