LES and RANS
 

Hi dear colleagues,
i have simulated an hydroclone using steady k-epsilon RNG model, then using LES with dynamic smagorinsky model for SGS.
The mesh for LES is finer than the RNG mesh.
I compared with experimental data and......i see that RANS gave better results!!!!!
Is that normal????( i would like to know if this happened be4 with any of you.)

if yes, what are the limits of LES????? is it capable of predicting all flow types (at least with a gross manner)?
I mean is there any instability that cant be captured using LES?

Thanks in advance

The only decent RANS model to simulate cyclonic flow is the RSM model as the k-e RNG model is unable to accurately calculate the axial velocity component and is usually found to give very different results to experimental data. Also a hydrocyclone is not a steady problem and the unsteady solver must be used to accurately model the problem to account for the precessing vortex core. Another problem when using the steady solver (seggregated) is that it will not give you the correct pressure distribution in the chamber and is unable to calculate the reversed flow region along the chamber axis. It may be worth using your k-e RNG solution as a starting point for a RSM simulation and compare the results. Pressure discretisation needs to be PRESTO and everything else QUICK to give the best results.

If you are using LES have you managed to construct a grid which allows for y+=1 at the walls? Its pretty tricky to construct the boundary layer mesh for the whole chamber especially around the inlets and it is more appropriate to use DES instead as wall effects are not of major concern with this type of problem.

Hi
thanks for ure help.
About RSM, in fact i tried it in different simulations: sometimes it did not converge (oscillatory solution, because i am using steady solver maybe) and sometimes it gace very bad solution......
The hydrocone is cylindrical and not cylinder-on-cone: maybe that is why i have different results than literature because RNG gace me a better solution than RSM, and predicts a reverse flow region. However the reverse flow region vanishes a bit faster than in experiments. That is why i tried LES.
1-I used wall functions with LES. That is the big difference with DNS (since my LES uses an implicit filter). I mean i used a very fine mesh, so if u want my LES is a sort of DNS with wall functions.
2- For discretisation, as u said, i used QUICK for all variables.
3-For the precessing vortex core, can u please explain to me this phenomenon in details???? I would like to know also what is a vortex breakdown.
4-Returning to the basic question, is it normal that a LES cannot predict accurately such a flow???? What are the limites of LES???

Hi,
I faced the same problem while modeling a cylindrical cyclone.
I have performed several simulations using RANS models, DES and LES.
With LES, i used the dynamic germano model for SGS modeling. I have used a coarse mesh for near-wall-region (typically, the first cell falls in the logarithmic layer). Results from LES and DES were not clearly better than RANS ones.
I am convinced that the problem is coming from the near wall treatement: wall functions are underestimating the turbulence production.
So now, i am performing another LES simulation, but this time, as you are advicing, i am using a boundary layer meshing near the walls.
My question is :
In the boundary layer region, small scale eddies do not obey to Kolmogorov similarity law, so the dynamic germano model must not be the convenable choice for such simulations.
Can u advice me which SGS model to use between those proposed by Fluent?

Thank you.

Hi,
for vortex breakdown, it has several definitions. The most used, is to consider that it is a stagnation point, with a reversed turbulent flow in its downstream. It is similar to a wake flow behind of a cylindre (roughly speaking).
Precessing vortex core is the rotation of the vortex around an immaginary axis (the radius of the rotation mouvement will be typically about 1 cm). The phenomenon has been well studied in the case of combustion chambers. The precessing core generally ends on the confining walls surface, according to Helmholtz third theorem (of course, it is not always the case as Helmholtz theorem concerns inviscid flows). I can send to you some papers about this topic.

I am working on a problem that is very much similar to yours. It is about gas-liquid cyclones, so i hope that we can share our experience in the domain.

I also performed LES simulations (with standard wall functions for calculating the wall shear stress), and its results were not much better than RANS ones. I think that the problem is coming from an inadequate turbulence production rate in the wall region. In fact, the velocity profile is not stable there (according to the Rayleigh criterion) and wall functions cannot take this factor into consideration. Now i am doing another LES but with a boundary layer mesh for near wall regions. I will tell you about my results once finished ( i am not sure that the SGS proposed by Fluent can perform good in the boundary layer region because of its inverse cascade, while those models rely on the Kolmogrov forward cascade theory).

I also have some questions and i hope you can give me answers You said that your hydroclone is cylindrical. As i know, typical hydroclones have a conical geometry. So can you please explain to me what is the big difference between a cylindrical and conical hydroclone (from an hydrodynamical point of view)? are you using a vortex finder at the top outlet?

About multiphase modeling, are you using eulerian-eulerian models? If yes, how are you choosing your dispersed phase size? And what is the rate of success in your simulations (compared to experimental data)?

Thank you, and i hope to hear from you soon.

Hi
thank u for the help. do u know if RANS models can find the true locstion of the vortex brekdonw?
of course we can share oure experience. I have a very little experience with gas-liquid cyclone also: for simulation u dont need eulerian model.
Dont put a liquid outlet. At inlet, put only gas. Siulate ure gas liquid interface with VOF, than do lagrangian to the droplets. Ure problem is to chose a crtierion to know if ure droplets are captured by the wires...

About my hydroclone, in fact it is a 3 phase hydroclone: oil water sand. That is why the geometry is different and includes a core finder. Now i am doing simulations using eulerian model (RNG as turbulence model) and choosing a dispersed phase size of 1mm. There is no other manner to do...
Results seems good but i dont have experimental data to compare with. I will simulate sand trajectory by lagrangian.