Direct Numerical Simulation

fadoobaba · June 1, 2018, 13:40

Does laminar flow option in commercial CFD packages count as direct numerical simulation, even if the flow is turbulent? Like, if I disable the turbulence model in Fluent or Flow Simulation Premium. Or any other package or in my own code.

flotus1 · June 1, 2018, 13:47

"Laminar" options should disable all kinds of explicit turbulence models. But you should still read the documentation of your specific CFD package to find out what it actually does.
Yet it does not necessarily turn your simulation into a DNS. IMO in order to be called DNS, the simulation also needs to resolve all scales both in time and space. Setting up the simulation correctly in this regard is still up to you. Otherwise you end up with something called "under-resolved DNS".

fadoobaba · June 1, 2018, 14:10

I'll run CFD on a 2D cylinder, Re=500, with time-step of 1e-10s and 1000x1000 grid. If my computer doesn't explodes, I post the results.

Is the time-step suitable? Can I use 1e-5 s?

FMDenaro · June 1, 2018, 14:24

Quote:

Originally Posted by fadoobaba

I'll run CFD on a 2D cylinder, Re=500, with time-step of 1e-10s and 1000x1000 grid. If my computer doesn't explodes, I post the results.

Is the time-step suitable? Can I use 1e-5 s?

No. If you want to perform a DNS the flow problem MUST be 3D.

LuckyTran · June 1, 2018, 16:26

In principle running an unsteady laminar flow that resolves all important scales is DNS (it's quite subjective what this means). But the question is whether a particular (commercial) CFD code actually does this. One thing you have control of is the mesh. One thing you don't have much control over is the discretization scheme.

For your time-step size, you should target a courant number just under 1 (e.g. 0.5).

And what is the purpose of doing 2D DNS? Just as a joke? Because real turbulence is 3D and requires the domain to be 3D.

flotus1 · June 1, 2018, 16:55

You never know, maybe he wants to study 2D "turbulence"?

selig5576 · June 1, 2018, 19:55

It is possible to do 2D "turbulence." However, I have only seen this in the geophysical fluids community. The argument is that geophysical flows (quasi-geostrphic flows) are statistically 2D. A very well known paper is one by McWilliams (http://www.o3d.org/abracco/jfm84.pdf.) With that said, this is done in a periodic domain, so maybe this does not satisfy the domain you had in mind.

In terms of DNS, I agree with the previous posters in that it is fundamentally 3D. A note on using CFD packages: I would be skeptical of solvers like Fluent due to the stabilization techniques. For DNS, or even LES, you want to minimize artificial dissipation in your solution.

FMDenaro · June 2, 2018, 04:10

That's right, 2D turbulence is well characterized for geophysical flows where the third dimension contains scales of the flow much more small than those in the plane. There is also a paragraph in the book of Lesieur.
The flow over a cylinder requires 3D.

Commercial CFD code can be used for DNS, the key is only in the grid resolution

fadoobaba · June 3, 2018, 16:48

Thank you very much everybody!

June 1, 2018, 13:40	Direct Numerical Simulation	#1
fadoobaba New Member Join Date: Oct 2017 Posts: 18 Rep Power: 9	Does laminar flow option in commercial CFD packages count as direct numerical simulation, even if the flow is turbulent? Like, if I disable the turbulence model in Fluent or Flow Simulation Premium. Or any other package or in my own code.

June 1, 2018, 13:47		#2
flotus1 Super Moderator Alex Join Date: Jun 2012 Location: Germany Posts: 3,427 Rep Power: 49	"Laminar" options should disable all kinds of explicit turbulence models. But you should still read the documentation of your specific CFD package to find out what it actually does. Yet it does not necessarily turn your simulation into a DNS. IMO in order to be called DNS, the simulation also needs to resolve all scales both in time and space. Setting up the simulation correctly in this regard is still up to you. Otherwise you end up with something called "under-resolved DNS". FMDenaro, lcarasik and fadoobaba like this.

June 1, 2018, 19:55	2D turbulence	#7
selig5576 Senior Member Selig Join Date: Jul 2016 Posts: 213 Rep Power: 11	It is possible to do 2D "turbulence." However, I have only seen this in the geophysical fluids community. The argument is that geophysical flows (quasi-geostrphic flows) are statistically 2D. A very well known paper is one by McWilliams (http://www.o3d.org/abracco/jfm84.pdf.) With that said, this is done in a periodic domain, so maybe this does not satisfy the domain you had in mind. In terms of DNS, I agree with the previous posters in that it is fundamentally 3D. A note on using CFD packages: I would be skeptical of solvers like Fluent due to the stabilization techniques. For DNS, or even LES, you want to minimize artificial dissipation in your solution.

June 2, 2018, 04:10		#8
FMDenaro Senior Member Filippo Maria Denaro Join Date: Jul 2010 Posts: 6,882 Rep Power: 73	That's right, 2D turbulence is well characterized for geophysical flows where the third dimension contains scales of the flow much more small than those in the plane. There is also a paragraph in the book of Lesieur. The flow over a cylinder requires 3D. Commercial CFD code can be used for DNS, the key is only in the grid resolution selig5576 likes this.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
TimeVaryingMappedFixedValue for Direct Numerical Simulation inlet	johndeas	OpenFOAM	5	May 21, 2014 08:11
numerical simulation of gas bubbles	lg88	OpenFOAM	4	October 25, 2011 00:14
numerical simulation of the electrostatic to DPM model	jeremyding	Main CFD Forum	0	September 8, 2011 04:27
About numerical filtering in direct simulation?	leaf	Main CFD Forum	0	June 20, 2006 02:57
numerical simulation of a hydrocyclone	caowei	Main CFD Forum	0	January 22, 2002 05:09

June 1, 2018, 14:10		#3
fadoobaba New Member Join Date: Oct 2017 Posts: 18 Rep Power: 9	I'll run CFD on a 2D cylinder, Re=500, with time-step of 1e-10s and 1000x1000 grid. If my computer doesn't explodes, I post the results. Is the time-step suitable? Can I use 1e-5 s?

June 1, 2018, 16:26		#5
LuckyTran Senior Member Lucky Join Date: Apr 2011 Location: Orlando, FL USA Posts: 5,754 Rep Power: 66	In principle running an unsteady laminar flow that resolves all important scales is DNS (it's quite subjective what this means). But the question is whether a particular (commercial) CFD code actually does this. One thing you have control of is the mesh. One thing you don't have much control over is the discretization scheme. For your time-step size, you should target a courant number just under 1 (e.g. 0.5). And what is the purpose of doing 2D DNS? Just as a joke? Because real turbulence is 3D and requires the domain to be 3D.

June 1, 2018, 16:55		#6
flotus1 Super Moderator Alex Join Date: Jun 2012 Location: Germany Posts: 3,427 Rep Power: 49	You never know, maybe he wants to study 2D "turbulence"?

June 3, 2018, 16:48		#9
fadoobaba New Member Join Date: Oct 2017 Posts: 18 Rep Power: 9	Thank you very much everybody!