search

firemodels / fds

Fire Dynamics Simulator
https://pages.nist.gov/fds-smv/
Other
631 stars 612 forks source link

Radiative heat flux distribution trend different for different diameter #883

Closed gforney closed 9 years ago

gforney commented 9 years ago 
Please complete the following lines...

Application Version: V5.4
SVN Revision Number:
Compile Date:15/09/2009
Operating System:XP sp2

The 30 cm and 50cm dimeter pool fire is simulated. Fuel is gasoline. For 30
cm the radiative heat flux trend matches with the experimental measurements.

For 50 cm pool radiative heat flux distribution trend doent match with
experimental measurements.(graph attached for your reference)

This problem was observed with diesel too in past.

kindly let me know how to justify this variability? how to fix this problem .
Is it appropriate to conclude that FDS doesnt work well for the blended
fuels like gasoline,diesel. etc.?

Regards....
Saumil

Original issue reported on code.google.com by saumil2525 on 2009-10-22 06:59:03

gforney commented 9 years ago 
This result probably has nothing to do with the nature of the fuels. It has more to

do with the relatively simple combustion model in FDS. Radiative emission is a 
complicated function of soot growth and oxidation, turbulent combustion, fluid 
dynamics, and gas species spectral properties. Currently, the source term in the 
radiative transport equation consists is a function of the heat release rate per 
unit volume and the specified radiative fraction. This is a simplification of the 
real physics. To better understand this, look at vertical contour slices 
of 'INTEGRATED INTENSITY' and 'HRRPUV'. Then, try adding FDS6=.TRUE. to the MISC 
line to see if the experimental routines under development change the result. Let me

know how it goes.

Original issue reported on code.google.com by mcgratta on 2009-10-22 12:03:11

gforney commented 9 years ago 
Please be careful about jumping to conclusions like, "FDS doesnt work well for the
blended fuels like gasoline,diesel. etc."?

Have you run a series of grid resolutions? Have you studied the effect of the
RADIATIVE_FRACTION?  The fuels you are looking are EXTREMELY complicated. The
distillation curves are not simple, the chemistry is not simple, and the
turbulence-radiation interactions (TRI) certainly are not simple.  Yet you are likely
just specifying a HRRPUA with a fixed composition and FDS treats the TRI by a simple
model with RADIATIVE_FRACTION. After looking at your plots, I would also bet that you
do not have the turbulence intensity correct at the base of the pool (more on this
below).

When you set FDS6=T, also add CHECK_KINETIC_ENERGY=T and output slices of 'TURBULENCE
RESOLUTION'.  You should have the time average <MTR> <= 0.2 (roughly) for a good LES.

Another trick that I discovered is essential to getting good results with helium
plumes is to set the inlet into a "pan", which most likely mimics the real experiment
better.  This helps the issue of the turbulence intensity at the inlet.  Put a layer
of obstacles 1 or 2 cells thick at the base of the simulation and cut a hole out for
the pool.  This creates a lip that generates turbulence with the entrained air stream
and gives a getter representation of the shear layer between the fuel and air at the
base of the pool.  Getting this shear layer correct is vital to the overall quality
of the simulation because the turbulent mixing between the fuel and air depends on
it.

Original issue reported on code.google.com by randy.mcdermott on 2009-10-22 12:51:36

gforney commented 9 years ago 
Dear sirs
I apologize for the conclusion made in previous post but simulation ran for the fuel
like heptane and diesel in last few months has forced us to conclude.

As per suggestion I have run FDS with FDS6 = T and check KE = T.,results are attached
in the document.you can observe from the results attached MTR is almost 0.2 for all
the cases hence satisfying the one of conditions proposed.further observations are
listed in last page of the attached doc.More over, in simulation, we have put steel
plated surrounding the pool which may take care of the shear layer probably.For your
reference FDS file is attached herewith.

Point I would like to highlight is HRR predicted by the simulation is matching with
given input HRR. The grid independent study and solid angle independent study is done
before running actual case.In sensitivity study influence of grid size and number of
solid angle on centerline temperature and radiative flux is studied.

looking forward for your valuable inputs. 
regards..
Saumil

Original issue reported on code.google.com by saumil2525 on 2009-10-27 07:04:34

gforney commented 9 years ago 
Saumil,

Thanks for sharing this study.  This is very interesting work.

Your conclusions are not surprising.  FDS 5 will say that you have better resolution
relative to solution of the LES equations because the solution is smoother overall
due to the constant coefficient Smagorinsky model.  This does not necessarily mean
that you have a better solution.  So, it is worthwhile to compare <MTR> (i.e. time
average) for different grid resolutions with the same turbulence model, but it is
hard to relate <MTR> between two different models. We need to amass experience with
MTR for specific models.

Based on your plot of radiant heat flux (very nice, by the way!) it seems that the
radiant fraction is (not surprisingly) very important, but I would also be that the
turbulence characteristics of your inlet condition are still a major factor in
getting this problem right.  I have attached my set for my helium plume case.  Notice
that the pool is depressed and this is a key in the local mixing.  I think also that
the inlet plane for your pool fire should be depressed into the pan!

With regard to MTR, we do not yet know if 0.2 is sufficient or if it is overkill for
any specific problem.  It may be that we need 0.1 in order to get good results with
FDS 6 for gasoline fires.  I doubt this, but it is possible.  So, we still need to
run at least 3 different grid resolutions to start to get an idea of the right value
for MTR for the gasoline pool fire.

So, you have a very good start here.  What I would do at this stage is:

1. Change the inflow pan to have a depression.  It does not have to have a flat floor
like my He plume case, but you should be able to have thin obstructions that make up
the pan.

2. With FDS6=T and radiant fraction (RF)=0.35, run the case at half the grid
resolution and at twice the grid resolution.  Look at MTR and the desired metric,
like location of peak radiant flux.

3. The last parameter that will be critical in your results is C_EDC on the REAC
line.  This the constant in the Eddy Dissipation Model time scale.  The default is
0.1.  The rate of reaction is min(Y_F,s*Y_O)/tau, where s is a stoichiometric
coefficient and tau is the turbulent mixing time tau = C_EDC*dx^2/D_t and D_t is the
turbulent diffusivity.  So, increasing C_EDC increases the mixing time (slower
mixing, lower heat release).  For a given RF, run a few different values of C_EDC,
say 0.05, 0.1, 0.2, for 3 different grid resolutions.

So, in effect, I am arguing that the first order parameters in your model are (a) the
grid resolution, (b) C_EDC, and (c) RF.  You need to get a good handle on how these
affect your results.  Ideally, we would be able to settle on values of C_EDC and RF
that are grid independent.  This is hard to do unless you have data that allows you
to vary only one parameter while the others are unimportant or known.  I think the
best hope at the moment is the settle on a value of RF from the literature (I hope
something exists, if not use 0.35) and use C_EDC as the parameter to vary at the
highest grid resolution you can afford computationally.  From there, varying C_EDC,
match the data as best you can.  What MTR do you have at this point?  Is it possible
to back off the resolution without harming the results?  If so, at what MTR are your
results unsatisfactory (try to use a quantifiable value, e.g., location of peak
radiant flux and value of peak radiant flux).

Well, I think this is plenty to work on right now.  I will be interested in your results.
Best,
Randy

Original issue reported on code.google.com by randy.mcdermott on 2009-10-27 14:39:10

gforney commented 9 years ago 
I suggest that you also try setting RADIATIVE_FRACTION=0 in the RADI line. This will

force FDS to use the gas temperature and absorption coefficient (essentially 
4*kappa*sigma*T^4) instead of the RAD_FRAC*HRRPUV as the source term in the 
radiation transport equation. Your plots of 'INTEGRATED INTENSITY' suggest that the

emitting region of the simulated fire extends to a height that is greater than that

of the real fire. The measured heat flux profile has higher values lower in the fire

and lower values higher in the fire. This suggests the region of emission should be

concentrated more towards the base of the fire. Randy's suggestion about a pan lip

might impact this.

In the FDS Validation Guide, there is a set of experiments called 'Hamins Methane 
Burner Heat Flux Measurements'. The FDS predictions of heat flux from a small 
methane burner match better the overall shape of the heat flux distribution. These

calculations were done at relatively high resolution and they do not use 
RADIATIVE_FRACTION.

Original issue reported on code.google.com by mcgratta on 2009-10-27 17:31:44

gforney commented 9 years ago 
In case of improved grid resolution simulation time may increase many fold but that's
not of the interest at the moment. 

I will get back to you soon.

Regards..
Saumil

Original issue reported on code.google.com by saumil2525 on 2009-10-28 09:55:58

gforney commented 9 years ago 
Hello sir/s

Answers to your Questions:

1)As presented in previous study MTR is close to 0.2.
2)In present study some optimum for grid size is worked out.pl comment whether it is
convincing or not. 
3)At MTR = 0.2 maxima of the RHF is at Y/D = 2 compared to experimental measurements
at Y/D = 1. but present study shows lipheight has affected the location of the pick
RHF.

I have partially completed first part of study. All results are presented for the
here are for C_EDC = 0.1 and RF = 0.35. The file is attached here with. You will find
conclusions at the end of each study.

Also I have not changed C_EDC in present study as I am not in position to justify the
different mixing time to my colleagues but I will present in the next part.Even many
of the my colleague have taken objection against fixing the radiative fraction in
spite of having enough evidences of RF 0.3-0.45 (for most of the combustible fuels)
in literature.

Finally,In next part of first study I have fixed the grid size to 3cm,lip height to
20 mm and the C_EDC is varied. Results are awaited.

Part 2 will have RF= 0 and max possible grid resolution(1.5 cm),lip height 20 mm.

Pl comment on the results attached. 

Regards...
Saumil

Original issue reported on code.google.com by saumil2525 on 2009-10-31 07:50:44

gforney commented 9 years ago 
Dear sir 

I have checked the effect of C_EDC on one of the cases . the detailed results are
attached here with.Kindly refer page 9 and 10.It seems for the case considered,the
effect of E_EDC on the FDS output is negligible.also it seems best result is within
25% of experimental measurement. Now only parameter left in our hand is RF.We are
planning experiments for measurement of RF too. Awaiting your valuable inputs. 

I have started with the part 2 which was suggested by Dr Mc Grattan.By tomorrow
result will be out.
Thanks for the help

Regards..
Saumil

Original issue reported on code.google.com by saumil2525 on 2009-11-02 05:45:56

gforney commented 9 years ago 
Saumil,

From what I can see:

a) You did not modify the lip in the way I suggested.  You need a DEPRESSION, not an
extension. Or at least send me a photo of the experimental set up so that I can
understand what this is supposed to look like.  I still think this boundary condition
is the major problem in the qualitative trend.
b) You cannot make conclusions about the effect of C_EDC when RF=0 (I guess this
pertains to your plans for 'Part 2' of your study).
c) I pray that we can avoid the conclusion that an "optimal" grid size exists, except
in sense of efficiency: fastest solution for a given error tolerance.  The fact that
you got "better" results with 3cm res than 1.5cm res is an indication that something
is wrong.

In short, I do not agree with your conclusions.  Further, if your colleagues have a
suggestion for how to vary the RF without increasing the cost of the simulation, by
all means, we would like to hear their ideas.

Randy

Original issue reported on code.google.com by randy.mcdermott on 2009-11-02 13:57:30

gforney commented 9 years ago 
I am out of station.will reply tomorrow.
Regards..
Saumil

Original issue reported on code.google.com by saumil2525 on 2009-11-05 11:50:48

gforney commented 9 years ago 
Dear sir

1. The schematic of the experiments is attached here with.
2. There is some mistake from my end,I had uploaded the old doc. pls see the
attchement on page 9 and 10. Rf is not taken as zero it 0.35 only.C_EDC is changed
for RF=0.35,Grid size = 3 cm, Lipheigth = 20 mm.

Also find part 2 of the study,because of the limitation of hardware I could not
decrease size bellow 1.5 cm.pls ref attachment for observations and conclusion.

Regards..
Saumil

Original issue reported on code.google.com by saumil2525 on 2009-11-06 06:00:43

gforney commented 9 years ago 
Saumil,

Thank you for your notes.  However, I still have questions about your conclusions.
I
would like to see:

a) a photo or video of the pool fire
b) a photo of the experimental set up
c) an ascii file with the heat flux data
d) a detailed description of the experimental conditions
e) you latest FDS input file with the "lip"
f) some visualization results for instantaneous flow fields near the pool, with 3D
perspective so that we can see what is going on; it is very hard for me to tell what
is happening with these time-averaged 2D slices; in fact, if you could zip up the smv
and slice files so that I could visualize the case myself, that would be good.

Thanks.
Randy

Original issue reported on code.google.com by randy.mcdermott on 2009-11-06 13:19:51

gforney commented 9 years ago 
Dear sir

Sorry for the delay.Asked details are as follows.

a & b) Img 33& 36 is the photo of 30 cm heptane pool.
Since i don't have the 50 cm gasoline pool images/video 30 cm heplane pool is
attached.In photo radiometer (mounted on lead screw) is circled red.

c) ascii file heat flux data. (Of 3cm grid resolution)
d) exp procedure/procedure for computing the radiative flux from thermal images 
e) fds file with grid size 3 cm and lip height 20 mm
f) there are 9 slice files 28 mb each hence not in position to upload.I am trying to
upload on some file sharing site.till I upload pls bear with me.

Regards.
Saumil

Original issue reported on code.google.com by saumil2525 on 2009-11-11 07:25:58

gforney commented 9 years ago 
Saumil,

Thanks for this!  There is no rush.  I am swamped and will not be able to look
closely at this stuff for a couple of weeks.  But then I will look closely since I
am
also putting together a report on flame height and radiation distribution.

Cheers,
Randy

Original issue reported on code.google.com by randy.mcdermott on 2009-11-11 15:08:28

gforney commented 9 years ago 
I am transferring this case to Randy McDermott.

Original issue reported on code.google.com by mcgratta on 2009-11-17 14:21:34

gforney commented 9 years ago 
Dear Mr McDermott
I am not in position of uploading these large files. one has to rerun at your end.

Regards..
Saumil

Original issue reported on code.google.com by saumil2525 on 2009-12-26 15:14:59

gforney commented 9 years ago 
Samuil,

Sorry, I have been off working on other things, plus out for the holiday.  Where do
we stand on this problem?  If you need to send me large files, try
http://www.yousendit.com/

Hopefully soon I will get a chance to start looking at flame height validation and
my
plan is to incorporate your test data and case into this study.

Best,
Randy

Original issue reported on code.google.com by randy.mcdermott on 2010-01-07 15:52:04

gforney commented 9 years ago 
Dear Mr McDermott
Overview of the problem is given bellow.For further clarification you may refer the
documents attached in previous posts.

Also request you to download .smv and slice files from the following file sharing portal.

http://www.4shared.com/dir/27651195/19b079b/50_cm_Gasoline_.html

Problem overview:
Radiative heat flux distribution trend for 50 cm gasoline pool

1)As suggested, MTR was checked which was fund close to 0.2
2)FDS6 was invoked.
Observation:
It had very little effect on the trend of RHF distribution but it underestimated the
RHF compared to FDS6=F.
3)lip height effect was studied.
Observation:
Maxima of RHF distribution shifted toward base with addition of lipheight

4)Effect of gridsize was studied
Observation:
RHF found almost same for grid size 2 cm and 1.5 cm grid. 

5)Effect of turbulent mixing (C_EDC) was studied
Observation:
Very little effect output was observed

Regards..
Saumil

Original issue reported on code.google.com by saumil2525 on 2010-01-19 07:05:05

gforney commented 9 years ago 
Saumil,

Thanks for this.  When I try to access the file sharing web site, it asks for a
password.  Can you please either post the password or remove this restriction?

Thanks.
Randy

Original issue reported on code.google.com by randy.mcdermott on 2010-01-19 13:25:14

gforney commented 9 years ago 

Pswd:me1234

Regards..
Saumil

Original issue reported on code.google.com by saumil2525 on 2010-01-28 05:19:32

gforney commented 9 years ago 
Saumil,

Thanks for this.  I have downloaded the files.  To me, this "looks" pretty good.

But, I see there might be a problem with your input file.  You are only supposed to
use one MISC line.  We should probably throw an error on this.  I will work on that.
 But in the mean time.  I think FDS6=T is the only line that is being considered in
your calculation.  At one point I had CHECK_KINETIC_ENERGY=T as a default for FDS6
which is why you were able to output MTR.  But it looks like your ambient temperature
is indeed 23F, so this is odd.  It makes me wonder what is going on.  You may not be
using FDS6 at all.

Regarding the claim that C_EDC does not affect the results, this is quite surprising.
 It means that that mixing time is always smaller than the time step.  As you
increase C_EDC you must reach a point at which it has an effect, for if C_EDC is
infinite the heat release rate is zero.  But this would tend to stretch the flame.

What this tells me is that you are still somewhat under-resolved (in spite of what
MTR may say) and you are not invoking the physics of the EDC model.  Ideally, we want
to see MIX_TIME > DT on average, which likely means better resolution and a larger
value for C_EDC.

There have been some significant changes to FDS6.  Please have a try with the latest
code when you get time, and also make sure to only use one MISC line.

Best,
R

Original issue reported on code.google.com by randy.mcdermott on 2010-01-28 14:23:00

gforney commented 9 years ago 
After thinking about this a bit more, it may also be the case that you are
continuously running into the HRRPUA_SHEET limitation of 200 kW/m2.  It would
interesting to remove this bound (set it equal to something large on the REAC line)
to see if that is the case.

Original issue reported on code.google.com by randy.mcdermott on 2010-01-28 15:12:01

gforney commented 9 years ago 
Dear sir

I reran the simulation increasing he upper bound on HRR_SHEET.The radiative heat flux
increased it means it was running at the upper limit , but still i get high RHF at
higher heights.I have attached some results for your ref.I fill results are fairly
ok.

Now I should be able to produce the results for all other diameter (0.7m and 1m)on
similar lines. I request you not to close this issue till i get the results for the
other diameters.

Apart from this i will try to understand the sensitivity of C_EDC.

Thanks for the help.

Regards..
Saumil

Original issue reported on code.google.com by saumil2525 on 2010-02-01 05:20:58

gforney commented 9 years ago 
Saumil,

Thanks for this.  The HRRPUA_SHEET limit is clearly in play here.  Decreasing C_EDC
(i.e. faster mixing) should bring the tail of the RHF curve more in line with the
data.  But so would decreasing the turbulent Schmidt number (SC on MISC).  It becomes
very difficult to settle on the correct parameters when you start model validation
with the most difficult case imaginable.  Just be aware that there are many knobs to
turn.  The default for SC is 0.5.  I would not go below 0.4 on SC (the literature
would not support it).  The interplay between HRRPUA_SHEET, C_EDC, SC, and the grid
resolution needs to be well understood.  I do not know if there is support for
HRRPUA_SHEET above 200 in the literature with gasoline.  Maybe someone else can chime
in here. It will be important that you examine the new results at a few different
grid resolutions.

Further, we have made some recent changes to the way we handle the baroclinic
correction in FDS6.  So please use the latest version of the code for your new runs.

Thanks for your continued work on this problem!
Best,
Randy

Original issue reported on code.google.com by randy.mcdermott on 2010-02-01 13:27:15

gforney commented 9 years ago 
HRRPUA_SHEET has been estimated from small fires that are just in the transition 
region between laminar and turbulent -- like what you see in the cone calorimeter 
(10 cm by 10 cm base). We take the total HRR and divide by an estimate of the 
surface area of the flame region. Values tend to be between 100 kW/m2 and 150 kW/m2.

200 was put into FDS as an upper bound to prevent excessive "hot spots" near the 
base of the fire where we do not resolve well the amount of mixing of fuel and 
oxygen. It was intended to stop very coarse calculations from becoming unstable. It

is a limit we'd like to get rid of, as I do not think it is physically correct and

should only be used as an upper bound to prevent instability. I think it more 
prudent to look at mixing times and parameters like Sc and Pr, as they have more 
physical validity.

Original issue reported on code.google.com by mcgratta on 2010-02-01 13:35:45

gforney commented 9 years ago 
Thanks Kevin.

So, Saumil, this suggests to me that we should take HRRPUA_SHEET out of the equation
by setting it equal to 1e10, or something very large.  Then we will see just the
effect of the physics in MIXING_TIME, which scales linearly with C_EDC and SC.  If
you lower SC, also lower PR accordingly.

Best,
Randy

Original issue reported on code.google.com by randy.mcdermott on 2010-02-01 16:24:02

gforney commented 9 years ago 
Dear sirs 

We have observed the effect of HRR_SHEET on the result is nil We will share the 
report soon.

Now we will run following cases.

1. case with SC=0.5,PR=0.5 and different C_EDC (0.05,0.08,0.15)
1. case with different grid sizes 
2. case with SC=0.4,PR=0.4 different C_EDC.
3. case with SC=0.4,PR=0.4,HRR_SHEET= 1e10 and best C_EDC
4. case with the latest version.

your inputs are welcomed.will share the data soon.

Regards..
Saumil

Original issue reported on code.google.com by saumil2525 on 2010-02-02 12:27:06

gforney commented 9 years ago 
Saumil,

I suppose you mean that HRRPUA_SHEET has no effect if set > 400 kW/m^2?  Because
there was a dramatic effect going from 200 to 400 per your post a couple of days ago.
 To make certain it has no effect, it is important to set it a large value as in your
case 3.

I don't understand why you put case 4 as last.  You should be running ALL your cases
with the latest version.

Further, now that we know that the HRRPUA_SHEET limit was being applied in earlier
runs, it may be again important to examine the effect of XR, the radiant fraction.

My suggestion is that you run a few coarse cases that run relatively fast to test
qualitative effects of some of these combinations.  Then do the grid sensitivity
study last once you have settled on the important parameters.  So, my case matrix
would look more like:

ALL: latest code,HRRPUA_SHEET=1e10,FDS6=T
1. SC=0.4, PR=0.4 (default parameters otherwise)
2. XR=[.1,.35,.5] (use better of SC,PR values)
3. C_EDC=[.2,.1,.05] (use better of SC,PR,XR values)
4. refine the grid (use best of all values)

Best,
Randy

Original issue reported on code.google.com by randy.mcdermott on 2010-02-02 13:31:13

gforney commented 9 years ago 
Dear sir
I will run with SVN5220 as of now our whole work was on SVN4697.

Regards..
Saumil

Original issue reported on code.google.com by saumil2525 on 2010-02-03 06:43:35

gforney commented 9 years ago 
I tried to run 70 cm and 50 cm gasoline pool with latest version 5.4.3 svn5414. i
resulted in numerical instability. the same fds file is working ok in 5.4.1 svn4697.
I am attaching the same along with the .out file.

From .out file I have notices that time step is decreasing continuously to zero.that
means CFL and VN are not limiting the time step.

Regards..
Saumil

Original issue reported on code.google.com by saumil2525 on 2010-02-05 11:50:43

gforney commented 9 years ago 
Saumil,

I have modified your input file (see attached).  Please make sure you use one MISC
and REAC line.  Also, monitor HRRPUA.

The problem seems to stem for the choice of model parameters, not a bug.  The time
step restriction based on the flow divergence was responsible for the instability
stop.  Let's see why this happened.

If unlimited, the heat release rate is determined by the mass consumption rate of the
fuel:

mdot_F = rho*min(Y_F,Y_O/s)/max(dt,tau_mix)

where

tau_mix = C_EDC*DX^2/TURBULENT_DIFFUSIVITY

With HRRPUA_SHEET=large and C_EDC=small, then we were consuming pretty much all the
fuel very near the burner exit in one time step.  This generates an enormous heat
release, which causes a flow expansion, and the then the time step has to be reduced
in order to prevent mass from evacuating the cell.

When we want to get rid of HRRPUA_SHEET as a parameter we need to set it something
really large, not a value like 600 that may or may not still play a role.  So, I have
set it to 1E10 in the attached input file.  Now we can really see the effect of
C_EDC.  The place to start is simply C_EDC=1.

I have also attached an image after 6 seconds of run time.  I ran this with the
latest code (SVN 5540) at half the grid resolution you were using.  Try
C_EDC=[0.5,1,2] at this lower resolution.  Observe the qualitative results and then
go on to study the effects of RADIATIVE_FRACTION and SOOT_YIELD before refining the
grid.

I am interested in your findings. Thanks!
Randy

Original issue reported on code.google.com by randy.mcdermott on 2010-02-05 14:03:14

gforney commented 9 years ago 
Dear sir
Pls tell me your system configuration because I have to upgrade system. I can not go
beyond 2.4 million cells.
existing config are:

Q9650 Quad 2 core 3GHz
DDR3 4 Gb RAM FBS1333MHz

also latest available release is 5414 only 

Regards...
Saumil

Original issue reported on code.google.com by saumil2525 on 2010-02-06 05:29:42

gforney commented 9 years ago 
You need to use a 64 bit operating system with more than 4 GB RAM, or you need to 
use the MPI version of FDS. 2.4 million is close to the limit for one 32 bit machine.

Original issue reported on code.google.com by mcgratta on 2010-02-06 16:02:44

gforney commented 9 years ago 
Dear sir 
I have upgraded my system to 64 bit windows 7 pro with 8 GB ram. but simulation is
running too slow. so pls bear with me. producing this many results will take quite
a
long time.

Regards..
Saumil

Original issue reported on code.google.com by saumil2525 on 2010-02-11 05:33:41

gforney commented 9 years ago 
Saumil,

This is why you need to reduce your resolution until we are confident that the model
parameters are producing 
the expected qualitative trends.  Cut your resolution in half in each direction.

Randy

Original issue reported on code.google.com by randy.mcdermott on 2010-02-11 13:04:41

gforney commented 9 years ago 
Dear sir
there were some issues with system. I hve done as suggested and will get back to you
soon within 2 days.

Regards..
Saumil

Original issue reported on code.google.com by saumil2525 on 2010-02-15 06:01:01

gforney commented 9 years ago 

I changed the C_EDC and observe the effect on radiative heat flux measured at L/D =
1.the parameter for this study were SC= 0.4, Pr = 0.4, Grid size = 4cm, and
HRRPUA_SHEET = 1e10.

It seems (probably) smaller C_EDC, better represent the trends. so started  radiative
fraction study with RF 0.5 and E_EDC 0.5 but that resulted into num instability (Due
to same reason explained in previous post by Dr Mc Randy dermott). so I have changed
the SC and Pr back to 0.5.

pls see the result of partial study .

Regards...
Saumil

Original issue reported on code.google.com by saumil2525 on 2010-02-16 11:44:54

gforney commented 9 years ago 
Saumil,

Thanks for this.  Okay, now we are getting somewhere.

First, you cannot look at this plot and conclude that C_EDC=0.5 "better represents
the trends".  The encouraging result is simply that C_EDC has a first order effect
on
the heat distribution, as we expect it should.  Remember that before you had
concluded that this parameter had no effect, and you can see that this conclusion
came because of the effect of HRRPUA_SHEET.

Second, the effect of SC and PR will be proportional.  So, leave these as SC=PR=0.4
(good support for this in the literature) and vary RADIATIVE_FRACTION with C_EDC=1
and this coarse mesh resolution.  Again, we are making sure that the heat
distribution is affected correctly.  Then, we will refine the grid and fine tune the
simulation.

Best,
Randy

Original issue reported on code.google.com by randy.mcdermott on 2010-02-16 13:08:30

gforney commented 9 years ago 
Dear sir

Change in Radaitive fraction has some effect on radiative heat flux computed at L/D
=
1 although little (may be because of very course grid). pls see the image attached.
Now I have started the grid sensitivity. will update you soon.

parameters taken in the study are as follows.
C_EDC = 1
SC=PR=0.4
Grid size = 4 cm
HRRPUA_SHEET = 1e10

Regards..
Saumil

Original issue reported on code.google.com by saumil2525 on 2010-02-17 10:56:28

gforney commented 9 years ago 
It looks to me like there is no real reason to change XR from its default of 0.35.

In your grid sensitivity study, run cases with C_EDC=1 and 2 at the finest grid you
can afford.

Original issue reported on code.google.com by randy.mcdermott on 2010-02-17 13:04:15

gforney commented 9 years ago 
Dear sir
I ran simulation for C_EDC = 1 ,RF = 0.4 and grid size 4 cm & 2.4 cm. some results
are attached.I think the refined grid wil only shift maxima of the RHF distribution
near the surface of the pool.hence we may have to reduce the E_EDC but for you info
C_EDC = 0.5 with HRR_SHEET =1e10 and 2 cm grid size resulting in num instability. 

Though I am currently running simulation for 1.6 cm (5 million cells) grid and C_EDC
= 1 Rf = 0.4. this will take long time. 

Regards..
Saumil

Original issue reported on code.google.com by saumil2525 on 2010-02-18 04:52:57

gforney commented 9 years ago 
Dear sir

I ran case with Rf=0.4 HRRPUA_SHEET = 1e10,C_EDC = 1 and grid size 1.6 cm . but
resulted into numerical instability.Currently Case 2 with C_EDC = 2 and grid size 1.6
cm is running. 
will get back soon.

Regards...
Saumil

Original issue reported on code.google.com by saumil2525 on 2010-03-02 04:22:24

gforney commented 9 years ago 
Saumil,

Please attach you latest input file, so that I know exactly what is crashing.

Thanks.
R

Original issue reported on code.google.com by randy.mcdermott on 2010-03-02 13:29:53

gforney commented 9 years ago 
Dear sir 

The problem is same as reported earlier. Timestep gradually reducing to zero.

Regards..
Saumil

Original issue reported on code.google.com by saumil2525 on 2010-03-03 05:19:51

gforney commented 9 years ago 
Dear sir
I have completed simulation for C_EDC = 2 for various grid size as we had discussed.

Attaching results for your coments. since predicted heatflux distribution was very

different from the exp one, I reduced C_EDC to 0.2 simulated for different grid 
sizes. Partial results are attached here with in second graph.

Observation is, larger value of C_EDC flattens the radiative heat flux distribution

curve so smaller the C_EDC better are the result.

FYIP:Currently simulation with E_EDC = 0.2 and grid size 1.6 cm is running.

Regards..
Saumil

Original issue reported on code.google.com by saumil2525 on 2010-03-08 05:14:14

gforney commented 9 years ago 
Saumil,

Thanks for this.  This is good news because last week we did some work on a 1 m
methane pool fire that showed C_EDC*SC = 0.2 gave good results for the puffing
frequency for both 3 cm and 1.5 cm resolution.

It is important we think of things in terms of C_EDC*SC because this is the
coefficient that changes in the reaction model.

Randy

Original issue reported on code.google.com by randy.mcdermott on 2010-03-08 13:12:11

gforney commented 9 years ago 
Dear sir 
In this case C_EDC*Sc=0.2*0.5 = 0.1, now with change in grid resolution to 1.6 cm
from 2 cm resulted in numerical instability. I feel it is better to consider grid
resolution also i.e. C_EDC*Sc*(dx*dy*dz)^(2/3) 

Is rightly to conclude, this is the best achievalbe results (refer previous post) for
70 cm gasoline pool because we have tried lot of combinations and it seem the results
are not changing much.For your study I am attaching fds file for C_EDC = 0.2 sc = 0.5
and grid = 1.6 cm which resulted into num instability( timestep gradually reduced to
zero )

For you info find puffing frequency analysis on 0.3 m and 0.5 m pool fires.

Regards..
Saumil

Original issue reported on code.google.com by saumil2525 on 2010-03-09 04:37:05

gforney commented 9 years ago 
Saumil,

Thanks for this.  Of course, I agree we must consider grid resolution.  But my point
was that we should eventually find that our results are insensitive to dx, whereas
we
will always see a first order influence of C_EDC*SC.  So, your conclusion about
including the square of the mesh spacing as a key parameter is not correct.

I will try to work on the instability problem now.  What SVN have you been running?

Randy

Original issue reported on code.google.com by randy.mcdermott on 2010-03-09 12:47:58

gforney commented 9 years ago 
By the way, how are you computing your "puffing frequency"?  I doubt you should get
such drastic differences depending on which parameter you use, as you show in your
.xls file.

The best way to look at this is to compute the power spectrum from a time series of
the data.  Do not limit the DT_DEVC.  In fact, set it to something small.  (Change
the total time of the calculation to 20 sec to decrease the file size.)  Add a DEVC
for W-VELOCITY.

As an example, I have attached the plot for the 1 m methane pool fire and a Matlab
script to compute the power spectrum.

Compute the power spectrum of W between t = [10,20] sec.  Let u = fft(W). p(k) =
u*conj(u).  You will see a peak in p(k) at the puffing frequency and a -5/3 fall off.

Original issue reported on code.google.com by randy.mcdermott on 2010-03-09 13:03:50

gforney commented 9 years ago 
Saumil,

I am starting to set up a validation case based on your experiments.  I need a write
up of the experiment.  Is there a paper available?

Thanks,
Randy

Original issue reported on code.google.com by randy.mcdermott on 2010-03-09 13:19:45