RE: ODT model on flame D,E and F

[Kerstein, Alan]

Ricky,

 

The only point that I will add to John's answers is that the dissipation is proportional to the turbulent production of scalar fluctuations, which increases in proportion to A.  So you can increase A to match the scalar dissipation, but when doing so, adjust beta to match the spreading rate.

 

I can find a citation to the formula that John mentioned if you're interested. 

 

For general guidance on resolution, time stepping, etc. you might find it useful to download a code and documentation that are available at http://groups.google.com/group/odt-research .  These are for channel flow so they don't cover jet-specific details like the use of beta.  One feature of the channel-flow code is that the time between eddy samplings is itself randomly sampled in a way that improves accuracy and gives converged results without needing to force the acceptance probability as small as would be needed for uniform time between eddy samplings.  The code also has a control algorithm that adjusts the sampling process based on acceptance probability statistics.  This is especially useful for the jet, in which there are large variations of the time scales as the jet evolves.  Another feature of the code is that it uses a different expression than Echekki et al. Eq. 6 that has basically the same physics, but instead of the hard-wired coefficient ( = 1/16) of the viscous term, it has an adjustable parameter that controls the small-eddy cutoff, and therefore ultimately controls the ranges of scales of property fluctuations.  I can send a paper that gives the version of the model that is coded.

 

I copied this to the google group.  If you would like to join and get emails sent to the group, please let me know.

 

Alan

---

From: Hewson, John C
Sent: Tuesday, April 14, 2009 4:05 PM
To: Ricky Chien
Cc: jyc...@me.berkeley.edu; Kerstein, Alan
Subject: RE: ODT model on flame D,E and F

Ricky,

I am happy to help to the extent that I can.  Unfortunately, I am getting ready to head out the door for some travel, so my response time may not be very good. 

To quickly answer what I can:
1) The scalar dissipation rate is generally chi = 2*D*(grad Z)^2.  I am not sure if there is any great trick to computing this.  One thing that does arise is that the scalar dissipation rate is lognormally distributed so that there is a preponderance of very small values.  The actual average value that you compute will be fairly sensitive to rare instances of large dissipation rates--the intermittence of turbulence.  It is possible that running for longer time periods will help, although you generally have to run for a very long time (1000's of samples perhaps) to get a really good average.  I don't recall what the measured values are for the D/E/F flames, so I cannot say how far off 50 [1/s] is.    I would typically also average the logarithm of the dissipation rate since that should be, roughly, normally distributed.  From the mean and variance of the logarithm it is possible to get the mean and variance of the regular scalar dissipation rate, though I don't have the formula handy. 

Another consideration is where you are averaging.  Are you averaging at a specific spatial location?  Or are you averaging over a large volume?  If you are averaging over the specific location, are you checking to make sure your ODT domain is not moving around too much.  Depending on how you do things, your flame can move around in the ODT domain, so you should take a look at a series of instantaneous shots. 

2) I don't think that there is such a thing as a "too small" time step for stirring sampling as long as you keep the probability of accepting an eddy small.  The stirring rate should be basically proportional to A. 

3) To start, I would suggest estimating a Kolmogorov scale, and trying to have at least ten grid points across that.  In fact, I now recall that if you are under resolved, you will definitely get a dissipation rate that is too low because your high dissipation rates are dominated by numerical dissipation (including regridding).  Do not skimp on grid resolution in ODT in the early stages or you will forever be repeating your work.  Once you get to a point of knowing that you are well resolved, you can look at reducing resolution, but I usually find it easier to be over resolved and not have to worry about it.  ODT creates so much data that you will have plenty to look at while your next simulation is running. 

Good luck with this.  I may not be able to respond again before next week.  I will CC Alan Kerstein who is also very helpful with ODT.

John

-----Original Message-----
From: Ricky Chien [mailto:lcc...@berkeley.edu]
Sent: Tuesday, April 14, 2009 3:43 PM
To: Hewson, John C
Cc: jyc...@me.berkeley.edu
Subject: ODT model on flame D,E and F

Dear Dr. Hewson

My name is Ricky and I am a PhD student in Berkeley currently working with Prof. J.Y. Chen on the finite rate chemistry effects using ODT. I learned from your paper that you are an expert running ODT. I am wondering if you could help me with a few questions or point me to the right direction.

1) We are currently running ODT on the Sandia flame D, E and F.
But the mean scalar dissipation rate at stoichiometry we got is too low(~50 1/s). We doubled checked the postprocessing code and couldn't find any thing wrong. We are wondering if you have similar experience before with the scalar dissipation rate? What should be an appropriate way to calculate the scalar dissipation rate?

2) We found that the stirring time step is at the order of 1.e-9. We are wondering if this is too small for flame D-F(Re=22400-44800)? Could you explain a little on how other parameters( target probability, A,
beta..) could impact this stirring time step?

3) In terms of the resolution, what would be a better way to estimate the grid points needed to resolve the flames under the ODT framework?

I understand you must be busy with your work. Any of your suggestions is greatly appreciated.

Sincerely yours,
Ricky


--
Li-Chun Chien (Ricky)
Combustion Modelling Lab
University of California, Berkeley
246 Hesse