I am using the Stochastic tracking model of Fluent code to simulate the turbulent dispersion of sub-micron particles.
In the Stochastic tracking approach, Fluent used the Duscrete Random Walk (DRW) model, in which the Time cale Constant must be specified. I have also turned-on the Random Eddy Lifetime module.
Could any one let me know why I get different particle tracking result in every display?
Also, any one can give me some literature regarding the Duscrete Random Walk (DRW) model.
Is the Stochastic tracking model of Fluent capable to predict particle behavior with resonable accuracy?
Re: patricle tracking
Dear C. H. Huang
Following statement is my opinion, not a generally accepted theory.
1) If you track 'a particle' several times in the real turbulent flow system, you will get different particle trajectory at every trial. Trajectory(or streakline) of your cigarette smoke is a good example. CFD calculate one of stochastically possible trajectories. That's why you get different particle trajectory in every different trial.
2) Particle tracking equation is a kind of auxiliary equation which is coupled with N-S equation. Some aux. eq. is universally accepted by everyone, for example, equation of state under low pressure condition, but there is no universally accepted equation for particle behavior in the turbulent flow system. So, for me, as far as particle phase is concerned, I always expect qulitatively, not quantitatively, consistent result. CFD can do much, but I do not think CFD can do everything. However, I think CFD can give you qualitatively consistent result. Decesion whether calculated particle behavior is reasonable or not, is entirely upto your insight about your flow system.
Hoping it will help you.
Sincerely, Jin Wook LEE
Re: particle tracking
I completely agree. Here's an example... experiment has shown spiralling (swirling flow) entering a large opening isokinetic sampler whose opening was oriented at 120 degrees to external flow in a particle settling chamber. Particles, having diameter of 3 micrometers, behaved in the spiral fashion consistently from result to result during the experiment. However, for CFD models , only a minimal number of iterations is required to get results approximating (using this word loosely...) experiment.
If more iterations are applied, the spirals almost always go away completely and if not, they are certainly much reduced and thus not useful for comparison against experiment. Stochastic modeling (random walk model) of particle behavior in simulated turbulent flows is useful, but only to a certain degree since the "lifetime" of the eddy determines the length of time that the randomness occurs. As the book says, "...The DRW model may give nonphysical results in strongly inhomogeneous diffusion-dominated flows [submicrometer would fit this of course]...the DRW will show a tendency for such particles to concentrate in low-turbulence regions of the flow..."
So what I've done is make really fine meshes and solve les or full NS without turbulence models so that the Langrangian reference is applied and solved are the forces including drag, Saffman, and gravity. I've gotten better consistency this way for the particle sizes I'm interested in (1-30 micrometers).
Re: patricle tracking
I am not a Fluent user - I can only diagnose the problem based on the methodology you are using:
The random walk method would have to use a random number generator (and the appropriate probability density function) to mimic stochasticity, and to emulate the process that DRW is expected to emulate. For example, if the process is diffusion, then the random walk method picks numbers "randomly" from a Gaussian probability distribution.
"All" random number generators need an initial "seed" value to begin the process. If you assign a fixed value for the seed then the code will become reproducible (that is, you will get the same solution every time). However, if you change the seed for each run, you _will_ get different results for different runs (more correctly, you will get a different realization of the same flow, to the extent that the random walk method is accurate). To get variable seeds one can link their value to machine's internal clock, as an example. So I assume this is probably the mechanism/reason behind your observation.
As for the accuracy of the random walk ... convergence of the solution is proportional to the square root of timestep, 1/Re (Re= Reynolds number), 1/N (N= number of particles per unit volume). So, the random walk method works quite well for HIGH Reynolds number cases, but does quite poorly for lower Reynolds numbers - in agreement with the earlier observation that DRW doesn't work well in highly diffusive flows.
As for the validity of this approach (and its correspondence to reality) I agree mostly with the previous posters
|All times are GMT -4. The time now is 14:08.|