LES-Model Constant in a quasi-DNS

JonN · February 13, 2019, 11:26

Hi everybody,

currently I am facing the following problem and I would appreciate your advice:

I'm conducting a 3D finite volume simulation of a several very small scale gauzes in a flow that contains upstream turbulence, generated on larger scales. As the problem also contains heat transfer and a heterogenous reactions on the gauze surface, I need to have a very fine spatial discretization of the boundary layer. A fine discretization of the bulk is comparable cheap.

The upstream turbulence entering the domain is generated using a spectral synthesizer. As the viscous forces are dominating in the spaces between the gauzes, turbulence is damped and the flow is assumed to become laminar after the second gauze.
Flow is fully laminar (with transient boundary layer detachment) for the same average inlet velocity without turbulent fluctuations.

As my grid is fine, the cells are smaller than the Kolmogorov length scale of the upstream turbulence, this is why I call it a "quasi-DNS".

The results for sub-grid turbulent kinetic energy (k) / subgrid turbulence viscosity however don't become 0 when using a dynamic LES procedure (Germano + Smagorinsky), because the shear is high close to the boundary layer and therfore the production of k is not 0 (unphysical behaviour of LES model, because it is not turbulence but laminar shear).

Which is the best way to handle this type of flow?
A) Split domain in inlet zone for turbulence generation (LES) and a zone for the gauzes (transient laminar) and force subgrid turbulence for the interface to be 0

B) Use a standard Smagorinsky LES model and force the model constant to be 0 to obtain DNS behaviour using LES equations

C) ?

It is hard to find literature regarding this topic and any hint would be great!

Thanks in advance

sbaffini · February 13, 2019, 11:40

You say nothing about your solver and/or computational approach, but I guess you are using Fluent.

Few facts:

1) The dynamic procedure is known to return the constant at the level of the test filter, not at the resolved level. Thus, the flow has to be well resolved at both levels to have a null (i.e., negligible) constant out of it.

2) There is no reason, in principle, to not trust the dynamically computed constant in certain zones.

3) You can achieve DNS behavior by simply not using an sgs model. This can be done in Fluent as well without resorting to a constant Smagorinsky model with null constant. You go to solve/set/expert and enable all the available schemes, so that you can use the central scheme with the laminar viscous model (i.e., no model at all). The other ways are doable as well, I guess.

LuckyTran · February 13, 2019, 11:42

It is unlikely the sgs kinetic energy or turbulent visosity will ever become 0 regardless of how fine your mesh is. But why do you need it to be 0? Is it not laminar enough?

You can switch to a laminar model, but given that you are running a super-fine mesh you will still end up with resolved k. Similarly, you can turn off the sgs in your LES, but you still have resolved k. Or you can do true DNS and still end up with resolved k and resolved production of k. Why specifically do you want the sgs k to be 0?

JonN · February 13, 2019, 12:16

Thanks for the quick and helpful answers!
You are right, I'm using fluent as I don't have experience with specialized DNS solvers.

I'm aware that resolved k will always be >0, as it is a transient Problem. The reason why I am concerned about a sub grid turbulence viscosity in LES is that the turbulence damping might be higher than for dns even if the grid is sufficiently fine.

The main reason to use LES at all is the availability of the spectral synthesiser inlet condition, so that the inlet velocity profile can be computer on the fly and I don't have to care about r/w speed of my filesystem. I'll check if I can activate it in the menu sbaffini suggested.

FMDenaro · February 13, 2019, 14:08

Just to suggest to inspect the ni_sgs/ni_molecular ratio in your solution to see if it tends correctly to zero in the zones where your resolution is up to the Kolmogorov scale.

sbaffini · February 13, 2019, 14:22

No, in this case you won't be able to use the spectral synthesizer. Use the static model with a 0 constant.

Note however that up to version 15, It was seriously flawed. Use the vortex method instead.

February 13, 2019, 11:40		#2
sbaffini Senior Member Paolo Lampitella Join Date: Mar 2009 Location: Italy Posts: 2,152 Blog Entries: 29 Rep Power: 39	You say nothing about your solver and/or computational approach, but I guess you are using Fluent. Few facts: 1) The dynamic procedure is known to return the constant at the level of the test filter, not at the resolved level. Thus, the flow has to be well resolved at both levels to have a null (i.e., negligible) constant out of it. 2) There is no reason, in principle, to not trust the dynamically computed constant in certain zones. 3) You can achieve DNS behavior by simply not using an sgs model. This can be done in Fluent as well without resorting to a constant Smagorinsky model with null constant. You go to solve/set/expert and enable all the available schemes, so that you can use the central scheme with the laminar viscous model (i.e., no model at all). The other ways are doable as well, I guess. FMDenaro and JonN like this.

February 13, 2019, 11:42		#3
LuckyTran Senior Member Lucky Join Date: Apr 2011 Location: Orlando, FL USA Posts: 5,668 Rep Power: 65	It is unlikely the sgs kinetic energy or turbulent visosity will ever become 0 regardless of how fine your mesh is. But why do you need it to be 0? Is it not laminar enough? You can switch to a laminar model, but given that you are running a super-fine mesh you will still end up with resolved k. Similarly, you can turn off the sgs in your LES, but you still have resolved k. Or you can do true DNS and still end up with resolved k and resolved production of k. Why specifically do you want the sgs k to be 0? JonN likes this.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
VOF model + mixture model + RAS or LES model	ebtedaei	OpenFOAM Running, Solving & CFD	23	May 12, 2018 03:36
LES constant velocity boundary conditions	balrog_f	CFX	4	April 28, 2017 04:33
LES model for buoyant flows other than air	manuc	OpenFOAM Running, Solving & CFD	0	May 5, 2016 05:22
Implement an les turbulence model	pante	OpenFOAM Programming & Development	19	December 5, 2014 16:16
Implementation of wall model to LES	saeedi	Main CFD Forum	3	August 24, 2012 14:51

February 13, 2019, 12:16		#4
JonN New Member Jonathan Nees Join Date: Nov 2018 Posts: 3 Rep Power: 7	Thanks for the quick and helpful answers! You are right, I'm using fluent as I don't have experience with specialized DNS solvers. I'm aware that resolved k will always be >0, as it is a transient Problem. The reason why I am concerned about a sub grid turbulence viscosity in LES is that the turbulence damping might be higher than for dns even if the grid is sufficiently fine. The main reason to use LES at all is the availability of the spectral synthesiser inlet condition, so that the inlet velocity profile can be computer on the fly and I don't have to care about r/w speed of my filesystem. I'll check if I can activate it in the menu sbaffini suggested.

February 13, 2019, 14:08		#5
FMDenaro Senior Member Filippo Maria Denaro Join Date: Jul 2010 Posts: 6,764 Rep Power: 71	Just to suggest to inspect the ni_sgs/ni_molecular ratio in your solution to see if it tends correctly to zero in the zones where your resolution is up to the Kolmogorov scale.

February 13, 2019, 14:22		#6
sbaffini Senior Member Paolo Lampitella Join Date: Mar 2009 Location: Italy Posts: 2,152 Blog Entries: 29 Rep Power: 39	No, in this case you won't be able to use the spectral synthesizer. Use the static model with a 0 constant. Note however that up to version 15, It was seriously flawed. Use the vortex method instead.