Is there any free 2D LES Code for analysing turbulent flows over bluff bodies on the net or if someone can lend me one (even for a small price)? I have access to a few 2D k-epsilon codes for that purpose, but I am desperately looking for something using LES, using either FEM or FVOL or FDIF. Else please let me know how difficult it is to convert a k-epsilon code to a LES code? Would really appreciate a detailed explanation of the steps taken to do so.. Also some good references or books to do so would be great.

Thanks in advance to everyone.... Anindya

There is no such thing as a 2D LES code if LES has it's traditional meaning. 2D turbulent motion is an interesting phenomenum but pretty rare. Nearly all turbulence models for engineering use assume the normal energy cascade under the action of (3D) vortex stretching.

An LES code is easy to write if you are comfortable coding and have a previous code to draw upon. It should certainly take less than a day for an experienced developer to put together a simple LES code (virtually all the code is a subset of what is required for a RANS code).

However, to create a fast, accurate and reasonably general LES code requires a lot more effort. Most of the numerical decisions made for RANS codes need revisiting and reworking for an LES code. Speed of execution is much more important for an LES code than a RANS code. The required numerical properties of the discretization scheme for the convection terms are very different. The pros and cons of structured vs. unstructured grids is different. Boundary conditions are a much bigger concern for LES than RANS. Etc...

Although I have never used the codes (and if I recall correctly since my links are old and dead) both the OVERTURE and FEAT codes have LES models. I would expect several other open source codes to also have LES models (possibly KIVA).

Like Andy said, 2D turbulence are completely different than 3D turbulence and so it is of course more realistic to simulate the full 3D problem otherwise you will get wrong results.

3D turublence are characterized by a cascade of energy (direct cascade) from the large scale to the small scale, and the energy is dissipated in the small scale due to the viscosity. Incompressible flow can be subject to an processes that are purely three-dimensional (for example streamwise vortices can be associated with the transition to turbulence).

2D turbulence are characterized by an INVERSE cascade of energy from the small scale to the large scale. As a result the energy accumulate in the largest scales: small vortices merge to form larger vortices. Eventually a few big vortices might be present and dominate the flow (like the Great Red Spot of Jupiter). The instability of these flow has nothing to do with 3D processes but only with 2D processes.

Only in a few cases some general results might be related in 2D and 3D simulations (like for Jets for example), but the interpretation of the results has to be done very carefully.

PG.

Hi,

For (1) Compressible turbulence

(2) Turbulent combustion (with variable density)

the difference between the 2D and the 3D cases seems to be less than that for incompressible turbulence.

This is probably due to the fact that the 2D compressible flows have mechanisms to generate vorticity (even though they have no vorticity generation by vortex stretching) wheras the 2D incompressible flows have no vorticity generation mechanism whatsoever.

Olus Boratav

When talks about turbulence modelling, one should never forget the physics of turbulence - the main characteristics of turbulence is its unsteadiness and three dimensionality. In statistically averaged approach (such as RANS), you can assume 2D approach not because the fluctuation in third dimension is zero (which is definitely not), but because the gradient in third dimension is zero.

When LES is used, the larger scales of turbulence are explicitly solved using filtered unsteady N-S equations. That is we are solving large part of turbulence without modelling, therefore one has to solve it 3D to catch that part of turbulence (if you have chance to look at the LES results in a simple channel, you would realise what it means - only the averaged data has 2D meaning, the instantaneous quantities are very RANDOM).

SO there is no 2D LES code - sorry about this.

Hi,

Many of the comments made on this subject implicitly assume that the flow is incompressible. For this case, 2D and 3D cases are very different.

But there are turbulent flows in which this is not the case. Consider turbulent, buoyant, nonpremixed flames, as an example. For such flows, the dominant physics is by baroclinic production, which can mostly be captured in a 2D analysis/ simulations/arguments (so all the interesting stuff happens in a plane and the dominant vorticity is out-of-plane). One does not gain much (extra) by assuming that the flow is 3D. If you do not believe it, simulate a flame by 2D and then by 3D equations and compare the evolution of the integral of vorticity-squared (integral over the whole domain).

The situation is even more interesting for non-buoyant, turbulent, nonpremixed flames. For these flows, the vorticity field near the flame gets totally depleted (by the so called 'heat release effects), and the flow near the flame is close to being 2D (and almost irrotational !) even though one has a 3D turbulence background. It is difficult to fit such flows in any classical category, but one will not be totally off if one starts modelling such a flow near the flame as being 2D.

So, here is the point: Don't get carried away by buzz words on turbulence. There is turbulence and then there is turbulence.

Olus Boratav

(1). I am not sure I understand the difference between 2-D and 3-D turbulent flow simulations. (2). In 2-D, one can only get u' and v', the w' will be missing. So, the TKE will have only two components. (3). In 3-D, you will have all three components, u', v' and w'. (4). What you are saying is in some cases, the mean flow will be the same regardless of whether a 2-D simulation or a 3-D simulation is used. But still a two-component TKE is different from a three-component TKE. Am I right? (when you say almost irrotational, does that mean the flow is almost non-turbulent.)

John,

Yes to all the questions. In one case we will have 2 components of the velocity field and in the other 3 components. But the general trends (dominant term contributing to the vorticity production) will be similar in the flows I mentioned. Such a similarity will not exist when you have incompressible turbulent flows.

And yes, the flow I mentioned will be locally non-turbulent since it will have little vorticity i.e almost irrotational locally but overall i.e. away from the flame zone, the field will be a turbulent field with lots of vorticity. This funny situation exists if you have a nonpremixed flame in a very low gravity environment somehow superposed on a turbulent field.

Olus Boratav

I understand what you people said. Thanks a lot for the valuable information and lively discussions. I think I should have said, that I am looking for a 3D LES code for turbulent flows over bluff bodies. The term 2D did create a lot of confusion. If anyone has any information on such a 3D LES code, or if anyone has one which one can hand over to me, then please let me know about it. It would be very helpful on my research work. I am trying to visualize the effects of fluctuating and turbulent input velocity profiles on the pressures over bluff bodies, over a period of time. Thanks .... Anindya

Am I right that all direct numerical simulations should be three dimentionsal ?

Thx.

-1) The similarities can in certain cases be extended further, not only between compressible 2D and compressible 3D, but between incompressible 3D and compressible 2D. The compressibility in the 2D case adds an additional degree of freedom which compensates for the third dimension. For example the 3D incompressible atmosphere is easily represented by 2D polytropic compressible models. When the polytropic index is 1 (gamma=2), and the polytropic constant is chosen appropriately (to fit the speed of the waves), then the polytropic model is equivalent to the shallow water approximation: this is where the 2D compressible meets with the 3D incompressible for (e.g.) an atmosphere with surface gravitational waves.

-2) The similarity between 2D and 3D compressible in Jets give good results on the largest scale, but the results are very poor on the small scale where the flow is more three dimensional.

-3) Also two-dimensionality implies symmetry, and is only a particular case of the real three dimensional picture. Again in Jets, 2D implies axisymmetric modes, while only 3D simulations can give some hints on the non-axisymmetric modes of the instability (body modes - sonic modes; and surface modes - Kelvin Helmholts instability).

Nice examples, Patrick.

More on 2D/3D stuff: Here is an example of an important 2D DNS:

In some instabilities such as the Richtmeyer-Meshkov instability, it is very common to simulate and analyze the flow using 2D equations. These are not exactly turbulent flows, but very important flows for which compressibility/density variation is important and a complete 3D analysis is not expected to contribute significantly to our understanding of the underlying physics.

Olus Boratav

(1). It is nice to know that there are 2-D special cases one can study and still give useful results for 3-D flows. (2). But I guess, to be on the safe side, a 3-D simulation would be the ultimate goal to simulate the real flow without the add on limitations. (similar to the situations that useful results can be obtained using a constant viscosity model in some jet, wake problems)

Actually, often, it is difficult to guess how (un)important a certain term is in a governing equation/model without making the full 3D simulation/analysis. Also, quite often, common sense turns out to be misleading and one encounters nice surprises.

2D vs 3D arguments, in a sense, are similar to the LES vs DNS arguments: The quality of a LES model can only be understood by comparing its predictions to results from DNS.

Olus Boratav

Some phenomena can be regarded as 2D, e.g. thin film. My former PhD supervisor, Dr. J. Chasnov, have carried out DNS in 2D.

(1). The Reynolds number for the thin film flow must be very low in this case. (2). A drop of gasoline on the surface of water can be very thin indeed. And the flow would be laminar. I mean, 2-D transient laminar flow.

I have not read the papers on the experiment so I better withhold any comment on this. (I am not working in 2D)

Regarding to the simulations by Chasnov, he is using low Reynold numbers (Re defined by using Taylor microscale), O(10). The Re is however changing due to inverse cascade. If interested, you can grap his paper in his homepage (www.math.ust.hk/~machas), particularly the one in viscous convective subrange.

Regards.

Frank

(1). Thank you for the web site information. I have just visited the site, there are nice color pictures of flow field along with a picture taken in a Chinese restaurant. (2). When I have time, I will download the paper and take a look at it.