[robo]

URANS is based on the idea that you can do a time average of the flow such that the turbulence is completely averaged out, but some unsteady forcing component that is unrelated to turbulence is not. For this to be valid, the timescale of the unsteadiness must be clearly uncorrelated with the timescale of the turbulence.

When the Reynold's decomposition (U + u') is subbed into the Navier Stokes equations, three terms arise due to the non-linearity: the mean convection term UU, the Reynold's stress term u'u', and a cross term Uu'. URANS assumes that the Uu' term is negligible due to the timescales being very different and neglects this term. In this way the unsteady component of the mean velocity U is assumed to not interact with the turbulent unsteadiness u'.

This will in general not be true for unsteady problems. If your problem has an unsteady external force at a relatively low frequency, then you might be able to get away with URANS, I would typically give a slowly oscillating airfoil as an example. However, for problems like an unsteady wake, I would say that it is likely a poor assumption. LES could be used for any URANS problem, but the reverse is not true. URANS is only attractive in comparison to LES because it is cheaper.

[hwsv07]

Originally Posted by robo

URANS is based on the idea that you can do a time average of the flow such that the turbulence is completely averaged out, but some unsteady forcing component that is unrelated to turbulence is not. For this to be valid, the timescale of the unsteadiness must be clearly uncorrelated with the timescale of the turbulence.

When the Reynold's decomposition (U + u') is subbed into the Navier Stokes equations, three terms arise due to the non-linearity: the mean convection term UU, the Reynold's stress term u'u', and a cross term Uu'. URANS assumes that the Uu' term is negligible due to the timescales being very different and neglects this term. In this way the unsteady component of the mean velocity U is assumed to not interact with the turbulent unsteadiness u'.

This will in general not be true for unsteady problems. If your problem has an unsteady external force at a relatively low frequency, then you might be able to get away with URANS, I would typically give a slowly oscillating airfoil as an example. However, for problems like an unsteady wake, I would say that it is likely a poor assumption. LES could be used for any URANS problem, but the reverse is not true. URANS is only attractive in comparison to LES because it is cheaper.

this explanation makes sense to me. do u have any source that explains that it like that?

[omar.rahul]

Hi,

I've tried to model a Trickle Bed reactor by both options 1) Three phase eulerian 2) 2 phase eulerian with porous media. Both simulations had convergence problem.

Problem for Gas & liquid phase is how to define the bubble dia considering gas is continuous phase.

For the 3 phase case, I patched solid vol fraction i.e. (1 - bed voidage ) in the solution domain, but to my surprise, solid volume fraction keeps on changing and eventually get replaced by the liquid & gas which is not correct.

Anybody has any clue?

Response will be appreciated.

email- omar.rahul51@gmail.com

[Eslam Reda]

To my little knowledge, the models used in RANS allow the NSE NOT to consider the turbulent motions. So when you make the grid-dependence analysis of the problem, you reach grid-independence at a large cell size (the grid size depends solely on the geometry).

On the other hand, in LES the SGS models can solve the small scale turbulent motions and the NSE will have to solve the large scale turbulent motions which means the grid-independence will not be reached till the grid size is very small (the grid size depends on the geometry and the turbulent scales).

If you read a comparative study between RANS and LES (e.g. flow over cylinder) you will find the grid size in case of LES is much much larger than that for RANS.

I suppose that in DNS where there is no supporting model at all the NSE will have to solve the whole turbulent scale range i.e. the grid size will have to be tremendously small.

[FMDenaro]

Originally Posted by Eslam Reda

To my little knowledge, the models used in RANS allow the NSE NOT to consider the turbulent motions. So when you make the grid-dependence analysis of the problem, you reach grid-independence at a large cell size (the grid size depends solely on the geometry).

On the other hand, in LES the SGS models can solve the small scale turbulent motions and the NSE will have to solve the large scale turbulent motions which means the grid-independence will not be reached till the grid size is very small (the grid size depends on the geometry and the turbulent scales).

If you read a comparative study between RANS and LES (e.g. flow over cylinder) you will find the grid size in case of LES is much much larger than that for RANS.

I suppose that in DNS where there is no supporting model at all the NSE will have to solve the whole turbulent scale range i.e. the grid size will have to be tremendously small.

Hello,
what I wrote need some correction...

1) RANS could have a coarse grid convergence but that is for the fact that the order of magnitude of the RANS-based turbulence modelling is greater than the local truncation error. Therefore grid convergence can be quite fast after some refinement. However, some models can require fine grids to reach convergence.

2) LES when performed with implicit filtering has no grid convergence in term of filtered variable. The grid convergence will be the DNS solution as the filtered velocity tends to the pointwise field and the SGS model vanishes. Conversely, convergence study for LES can be correctly performed using explicit filterin.

[kwardle]

Hello - Sorry to zombify this thread, but there are some really interesting points here. Can anyone point me to references where LES and URANS are compared for an unsteady problem on the same coarse mesh (one made for URANS and 'not suited' for LES) against experimental data and it is definitively shown that URANS is better on a coarse mesh? I understand what is being said re mesh requirements of LES, wall interactions and the like and do not dispute that. My question is whether one can stretch to VLES (or even VeryVeryLES) and still get a better solution than URANS on the same mesh. So perhaps I am saying, sure, maybe LES is bad on a mesh that is too coarse, but can't URANS still be worse? One can say that LES on a coarse mesh is worse than LES on a fine one, but it does not necessarily follow that URANS is better than coarse LES. I would love to see a comparison with real data.
Thanks.

[FMDenaro]

Have a look here and the references
https://www.researchgate.net/publica...LIKE_BEHAVIOUR

[kwardle]

Thanks for the response and the interesting reference. I have not digested it completely, but the paper seems to indicated that URANS in some cases can capture unsteadiness but in general is inherently constrained in terms of temporal instabilities. Also as a point of clarification, the paper says that URANS = VLES which is not what I have understood and certainly not the way I am using the term. What I mean is LES on a coarse mesh but still with a SGS model like Smagorinsky. I don't see that the paper addresses this question of whether poorly resolved LES is worse than URANS on the same mesh.

[FMDenaro]

Originally Posted by kwardle

Thanks for the response and the interesting reference. I have not digested it completely, but the paper seems to indicated that URANS in some cases can capture unsteadiness but in general is inherently constrained in terms of temporal instabilities. Also as a point of clarification, the paper says that URANS = VLES which is not what I have understood and certainly not the way I am using the term. What I mean is LES on a coarse mesh but still with a SGS model like Smagorinsky. I don't see that the paper addresses this question of whether poorly resolved LES is worse than URANS on the same mesh.

Well, I don't think that using a coarse grid and comparing URANS and LES solutions on the same grid makes sense... Formally, you are comparing two different things...

[kwardle]

I get your hesitation, but I am not talking about comparing them to each other but rather each to experimental data.

[FMDenaro]

Originally Posted by kwardle

I get your hesitation, but I am not talking about comparing them to each other but rather each to experimental data.

And that is not as same as comparing each other???

[kwardle]

If you can't compare two different models to experimental data (or analytical solution, for that matter) and determine which is more accurate, what exactly is the point of discussing which is 'better'?

[FMDenaro]

Originally Posted by kwardle

If you can't compare two different models to experimental data (or analytical solution, for that matter) and determine which is more accurate, what exactly is the point of discussing which is 'better'?

An analytical solution is something totally different from an experimental measurement... When you have an experiment you are doing an "experimental simulation" with some boundary and initial conditions that are rarely exactly introduced in a numerical simulation.
Then, URANS and LES produce variables that have a different meaning and you have to know how to match at the best with the experimental measures.
Often, comparisons with experiments cannot give enough infos to address which formulation is better....

[kwardle]

Hmm. Not sure I am getting across what I am meaning. It is not my intent to discuss whether an experiment is accurate or not and I am certainly assuming that one is comparing the same values between URANS and LES.

How about this instead: is it possible to determine whether a URANS simulation or a coarse LES simulation on the same mesh is a 'better' fit to the appropriately time averaged results obtained from a DNS simulation? "LES requires a fine mesh" My question is that if you ignore that and use a coarse mesh do you get results that are better or worse than a URANS simulation of the same phenomena on the same mesh. I don't think it is a crazy question. Probably it is problem dependent, but I would be happy to see any example which makes a case one way or the other.

[FMDenaro]

Originally Posted by kwardle

Hmm. Not sure I am getting across what I am meaning. It is not my intent to discuss whether an experiment is accurate or not and I am certainly assuming that one is comparing the same values between URANS and LES.

How about this instead: is it possible to determine whether a URANS simulation or a coarse LES simulation on the same mesh is a 'better' fit to the appropriately time averaged results obtained from a DNS simulation? "LES requires a fine mesh" My question is that if you ignore that and use a coarse mesh do you get results that are better or worse than a URANS simulation of the same phenomena on the same mesh. I don't think it is a crazy question. Probably it is problem dependent, but I would be happy to see any example which makes a case one way or the other.

What you want is something like this
http://www.iaeng.org/publication/WCE..._pp673-678.pdf

However, For performing an LES, you need to specify time-dependent inflow conditions that match the experiment and this is a very difficult goal. URANS is quite different.
Concerning the match with a DNS, the LES grid can be coarsened (but you have to fulfill the requirement that the grid filter lies in the inertial range) which means you have some issue about the role of the filtering size.
Furthermore, Using URANS you cannot obtain high order statistics therefore the comparison with LES and DNS is just in terms of averaged profiles. That is a limit in the answer of the better formulation.
You could find some examples of coarse LES solutions compared to DNS here
https://www.researchgate.net/publica...mulation_codes

[kwardle]

First, thanks for sticking with me on my follow-up questions! Hopefully you are up for a few more...
The first reference is looking at comparing LES to LES, all on the same mesh. The second one does consider URANS and LES and appears to do so on the same mesh (though I don't see that it is explicitly stated). However, they used a very different time step, 4e-2s for URANS, 2.5e-3s for LES which makes a 16x difference. What is the driver for the much smaller time step if the mesh and presumably the flow velocities are equivalent (i.e. same Courant)?

[FMDenaro]

The time step in LES is taken small in order to resolve all temporal characteristic scales. Practically one uses a DNS time step to eliminate filtering effects in time.

[sanazm]

Hi, I have problem in downloadingthem, would you please send it to me?


Thanks in advance,
Sanaz

Originally Posted by mb.pejvak

This is a very useful document in simple language provided by Prof. Lars Davidson. You can find several fundamental points related to turbulence and fluid dynamics in it, and ansewer of most of these questions were asked in the threat.

"Fluid mechanics, turbulent flow and turbulence modeling"; you can download it from:
http://www.tfd.chalmers.se/˜lada/MoF/lecture notes.html
and
http://www.tfd.chalmers.se/˜lada/comp turb model/lecture notes.html

if you have problem in downloading it, send me an email. I 'll willingly send it to you.