CFD Online Discussion Forums

CFD Online Discussion Forums (http://www.cfd-online.com/Forums/)
-   Main CFD Forum (http://www.cfd-online.com/Forums/main/)
-   -   Lattice Boltzman (http://www.cfd-online.com/Forums/main/6552-lattice-boltzman.html)

Rob Kunz September 10, 2003 18:52

Lattice Boltzman
 
Okay, I am skeptical. I just read the 4-page article in August 2003 "Science" by Chen et al. on Lattice Boltzman Method. The 1st sentence in the abstract reads "Complex fluid physics can be modeled using an extended kinetic (Boltzman)equation in a more efficient way than using continuum Navier-Stokes equations. The paper goes on to describe the method on a very high level, claims better turbulence modeling than using RANS and shows 3 cases that anyone with a RANS code could solve with some experience and a decent turbulence model, including up to Re=3x10^6. The last paragraph states: "the run times of the turbulence simulations may be an order or magnitude faster [than Navier-Stokes]". No quantitative timing comparisons of any sort are made. I have "heard" that LBM approaches require much much more run time for high Reynolds numbers than RANS, but have no direct experience. This coupled with the fact that several of the authors have financial interest in Powerflow makes me skeptical. Whats the real deal?

Free and Simple September 11, 2003 02:46

Re: Lattice Boltzman
 
Using LB Methods now for several years my personal opinion is:

Don't believe everything they claim

LB methods perform best for flows with complex physics (or where the continuum approach of the NS equations is not appropriate). It is somehow more straightforward to include "microscopic" effects (e.g. interaction between several phases) into the LBE. On the other hand most people just adapt "macroscopic" approaches and use it in LB. The best example is turbulence modelling, where everybody in the LB community uses models developed for the NS equations. There is no real progress to use the strength of LB and kinetic theory for alternative turbulence modelling.

The most wide spread turbulence model used in LB is a Smagorinsky type LES (or better VLES). So the turbulence modelling at the present state of LB is as good as the LES models for NS equations. I obtained VERY good results for high turbulent flows using LBM + LES, but ...

... using a large number of cells ... introducing grid refinment ... using advanced wall functions

You see, the same troubles as in every other method for simulating turbulent flows. AFAIK Powerflow is also using conventinal turbulence models, but claims to have very "sophisticated" ones.

You can find some very good papers on the LB method (including the most recent review and an evaluation of the performance of Powerflow) on the webpage of Li-Shi Luo (http://research.nianet.org/~luo/). BTW Li-Shi's papers on this subject are a definite must to read.

My personal conclusion: The LB method is a powerful tool, if you know where to apply it. If your interest is in compressible flows and flows with strong heat transfer, forget it.

Regards, F & S

an ex-powerflow user October 15, 2003 09:31

Re: Lattice Boltzman
 
F& S,

Absolutely agree with you - DON'T BELIEVE WHAT THEY CLAIM. If this 21st century cfd code (this isn't my words, EXA used to advertize powerflow as the 21st century super code back in 1990s) works so well, why do they need to turn back to the traditional (20th cetury) tricks to make it work (e.g. turbulence models, very fine mesh, mesh refinement, near wall treatment etc etc). By the time one used all these tricks, one might just as well go back to a 20th cetrury old fasioned NS code (much cheaper & less memory needed) & get more or less (sometime even better) results......


All times are GMT -4. The time now is 01:46.