Grid Sensitivity Analysis
1,214 views
Skip to first unread message
Christopher Wood
Jan 11, 2008, 4:23:16 PM1/11/08
to fds...@googlegroups.com, Bryan Klein, Kevin McGrattan
At the suggestion of Dr. McGrattan I am posting this preliminary, unpublished work to foster discussion and input from the group.
I did a number of runs re: grid sensitivity of an axi-symmetric plume. Please see the write-up for some scant details and a few comments. I am looking for feedback, comments, suggestions, concerns, etc. as I hope to generate a paper from this work. Also any suggestions on improving the presentation of results regarding the graphics would also be much appreciated. However, this is admittedly not a paper and is not presented as such – it contains no references.
I look forward to your thoughtful comments.
dr_jfloyd
Jan 11, 2008, 5:04:36 PM1/11/08
to FDS and Smokeview Discussions
One rule of thumb has been to try and get 8 to 10 cells in D*. For
your fires that would suggest 7-9, 12-15, and 16-20 cm. Another rule
of thumb is to try to get at least 6 to 8 cells across the burner.
That would suggest 12 - 16 cm. Your observations are not odds with
these; however, your fire sizes and pool diameter are such that these
rules of thumb wind up being similar. It would be interesting to also
have comparisons with a fixed fire size and varied pool diameter.
In terms of the graphs I would recommend the use of bold distinct
colors rather than a pastel pallet. You may wish to drop a couple of
curves (25 cm and 11 cm perhaps) to make the plot more readable.
For the surface plots you really aren't plotting an error. We don't
have a real known value. You are plotting a relative difference.
On Jan 11, 4:23 pm, "Christopher Wood"
> 863KDownload
your fires that would suggest 7-9, 12-15, and 16-20 cm. Another rule
of thumb is to try to get at least 6 to 8 cells across the burner.
That would suggest 12 - 16 cm. Your observations are not odds with
these; however, your fire sizes and pool diameter are such that these
rules of thumb wind up being similar. It would be interesting to also
have comparisons with a fixed fire size and varied pool diameter.
In terms of the graphs I would recommend the use of bold distinct
colors rather than a pastel pallet. You may wish to drop a couple of
curves (25 cm and 11 cm perhaps) to make the plot more readable.
For the surface plots you really aren't plotting an error. We don't
have a real known value. You are plotting a relative difference.
On Jan 11, 4:23 pm, "Christopher Wood"
<christopher.w...@firelinkllc.com> wrote:
> At the suggestion of Dr. McGrattan I am posting this preliminary,
> unpublished work to foster discussion and input from the group.
>
> I did a number of runs re: grid sensitivity of an axi-symmetric plume.
> Please see the write-up for some scant details and a few comments. I am
> looking for feedback, comments, suggestions, concerns, etc. as I hope to
> generate a paper from this work. Also any suggestions on improving the
> presentation of results regarding the graphics would also be much
> appreciated. However, this is admittedly not a paper and is not
> presented as such - it contains no references.
> At the suggestion of Dr. McGrattan I am posting this preliminary,
> unpublished work to foster discussion and input from the group.
>
> I did a number of runs re: grid sensitivity of an axi-symmetric plume.
> Please see the write-up for some scant details and a few comments. I am
> looking for feedback, comments, suggestions, concerns, etc. as I hope to
> generate a paper from this work. Also any suggestions on improving the
> presentation of results regarding the graphics would also be much
> appreciated. However, this is admittedly not a paper and is not
>
> I look forward to your thoughtful comments.
>
> prelim_release_FDS_group_20080111.pdf
> I look forward to your thoughtful comments.
>
> 863KDownload
dr_jfloyd
Jan 12, 2008, 10:20:00 AM1/12/08
to FDS and Smokeview Discussions
To continue with the objective of fostering group discussion, I am
replying to your email (attached below) here:
My comments on D* were meant to say that the grid size suggested by D*
turned out to be similar to what you observed as an adequate grid. In
your writeup you identified that < 12 cm and there was not as much
change in the results and that dimension was consistent with both
rules of thumb : number of cells in D* and number of cells across the
burner
In terms of the pool fire size yes I meant pick a fire size (2 MW
would be good) and then make change the area of the vent for the
fire. But keep the grid constant over the entire domain just as you
did with the prior runs, in fact use some of the same grids as
before.
Jason,
Thanks for your response. Please see my question below.
On Jan 11, 5:04 pm, dr_jfloyd <drjfl...@gmail.com> wrote:
> One rule of thumb has been to try and get 8 to 10 cells in D*. For
> your fires that would suggest 7-9, 12-15, and 16-20 cm. Another rule
> of thumb is to try to get at least 6 to 8 cells across the burner.
> That would suggest 12 - 16 cm. Your observations are not odds with
> these; however, your fire sizes and pool diameter are such that these
> rules of thumb wind up being similar. It would be interesting to also
> have comparisons with a fixed fire size and varied pool diameter.
>
I am wondering about the specifics of your suggestion. For example, I
was outside D* with the calculations that I made so that addresses a
little bit of your comment. As far as the different pool diameters, I
am trying to figure out if you are saying something like a 2MW fire
maybe in a 2 m x 2 m vent and then a .5 m x .5 m vent? If so, how are
you suggesting that I break up the grid spacing across that space?
Without fulling explaining it in my note, I was thinking in terms of
cells per meter and the HRRPUA for which I used a constant size fire
(1 m^2).
> For the surface plots you really aren't plotting an error. We don't
> have a real known value. You are plotting a relative difference.
>
Yes, as I said in my note it is not really an error calculation. Your
terminology is better.
replying to your email (attached below) here:
My comments on D* were meant to say that the grid size suggested by D*
turned out to be similar to what you observed as an adequate grid. In
your writeup you identified that < 12 cm and there was not as much
change in the results and that dimension was consistent with both
rules of thumb : number of cells in D* and number of cells across the
burner
In terms of the pool fire size yes I meant pick a fire size (2 MW
would be good) and then make change the area of the vent for the
fire. But keep the grid constant over the entire domain just as you
did with the prior runs, in fact use some of the same grids as
before.
Jason,
Thanks for your response. Please see my question below.
On Jan 11, 5:04 pm, dr_jfloyd <drjfl...@gmail.com> wrote:
> One rule of thumb has been to try and get 8 to 10 cells in D*. For
> your fires that would suggest 7-9, 12-15, and 16-20 cm. Another rule
> of thumb is to try to get at least 6 to 8 cells across the burner.
> That would suggest 12 - 16 cm. Your observations are not odds with
> these; however, your fire sizes and pool diameter are such that these
> rules of thumb wind up being similar. It would be interesting to also
> have comparisons with a fixed fire size and varied pool diameter.
>
was outside D* with the calculations that I made so that addresses a
little bit of your comment. As far as the different pool diameters, I
am trying to figure out if you are saying something like a 2MW fire
maybe in a 2 m x 2 m vent and then a .5 m x .5 m vent? If so, how are
you suggesting that I break up the grid spacing across that space?
Without fulling explaining it in my note, I was thinking in terms of
cells per meter and the HRRPUA for which I used a constant size fire
(1 m^2).
> For the surface plots you really aren't plotting an error. We don't
> have a real known value. You are plotting a relative difference.
>
terminology is better.
Kevin
Jan 12, 2008, 7:45:13 PM1/12/08
to FDS and Smokeview Discussions
This is precisely the kind of analysis that we ought to permanently
archive in our Validation Guide. We have been very busy with the
release of FDS 5 and the various bug fixes and question answering that
we have not had time yet to polish the new Validation Guide, which is
going to supplement the FDS Tech Guide. If you Google search on
NUREG-1824, you will find the FDS V&V study conducted by the US
Nuclear Regulatory Commission. It was done with FDS 4, and we have re-
run all the cases in FDS 5 and plan to create our own version of it
that we will update and add to as appropriate. Over the years, we have
used these various plume correlations as you have done to test new
algorithms, but we have yet to create a permanent home for this kind
of thing. We used to publish in journals, but it has become clear to
us that this may not be the proper venue for this kind of routine
sensitivity analysis. In fact, I have reviewed or read numerous papers
testing the various versions of FDS in ways similar to what you have
done. The work is valuable, but it should not go into an archival
journal because the results might change (not dramatically, but still
a bit) even before it is published. Archival journals ought to be
reserved for fundamental developments in the physical model, things
that will last a substantial amount of time. Conference proceedings
are OK for more time-sensitive info. Our V&V Guide ought to be a
permanent repository that people can refer to when they want to know
how well the model works for a particular problem.
Of course, this is all my opinion. Some will argue that everything
ought to be published in archival journals, but I fear that the
journals might get choked with too much routine V&V work.
Off the soapbox -- there is something interesting in your calculations
that I have noticed myself when doing this sort of thing. Certainly,
FDS converges towards the correlation in the far field (high up in the
plume) as the grid is resolved. That's exactly what LES ought to do.
But look down low. Notice the temps at the base getting hotter and
hotter as the grid is refined. This is the real problem, and it has
little to do with LES or any particular turbulence model. It has to do
with assumptions we must make about the combustion on coarse grids --
and all these grids you are using are "coarse" in regard to
combustion. There is still too much grid dependence at the base of the
fire, and this effects heat feedback to the fuel bed and ultimately
flame spread. In these cases you have fixed the fuel flow, but suppose
you didn't. Then the problem is exascerbated by the two-way coupling
between the gas and solid phase. A goal for us this year is to lessen
the grid dependence in the near field. We currently have to use an
upper bound on the local volumetric heat release rate due to excess
diffusion that inevitably results from the transport of fuel and
oxygen on coarse grids. I believe that there are relatively simple
things we can do to improve are current assumptions, so stay tuned.
That's enough for now -- the Patriots are on.
K
> 863KDownload
archive in our Validation Guide. We have been very busy with the
release of FDS 5 and the various bug fixes and question answering that
we have not had time yet to polish the new Validation Guide, which is
going to supplement the FDS Tech Guide. If you Google search on
NUREG-1824, you will find the FDS V&V study conducted by the US
Nuclear Regulatory Commission. It was done with FDS 4, and we have re-
run all the cases in FDS 5 and plan to create our own version of it
that we will update and add to as appropriate. Over the years, we have
used these various plume correlations as you have done to test new
algorithms, but we have yet to create a permanent home for this kind
of thing. We used to publish in journals, but it has become clear to
us that this may not be the proper venue for this kind of routine
sensitivity analysis. In fact, I have reviewed or read numerous papers
testing the various versions of FDS in ways similar to what you have
done. The work is valuable, but it should not go into an archival
journal because the results might change (not dramatically, but still
a bit) even before it is published. Archival journals ought to be
reserved for fundamental developments in the physical model, things
that will last a substantial amount of time. Conference proceedings
are OK for more time-sensitive info. Our V&V Guide ought to be a
permanent repository that people can refer to when they want to know
how well the model works for a particular problem.
Of course, this is all my opinion. Some will argue that everything
ought to be published in archival journals, but I fear that the
journals might get choked with too much routine V&V work.
Off the soapbox -- there is something interesting in your calculations
that I have noticed myself when doing this sort of thing. Certainly,
FDS converges towards the correlation in the far field (high up in the
plume) as the grid is resolved. That's exactly what LES ought to do.
But look down low. Notice the temps at the base getting hotter and
hotter as the grid is refined. This is the real problem, and it has
little to do with LES or any particular turbulence model. It has to do
with assumptions we must make about the combustion on coarse grids --
and all these grids you are using are "coarse" in regard to
combustion. There is still too much grid dependence at the base of the
fire, and this effects heat feedback to the fuel bed and ultimately
flame spread. In these cases you have fixed the fuel flow, but suppose
you didn't. Then the problem is exascerbated by the two-way coupling
between the gas and solid phase. A goal for us this year is to lessen
the grid dependence in the near field. We currently have to use an
upper bound on the local volumetric heat release rate due to excess
diffusion that inevitably results from the transport of fuel and
oxygen on coarse grids. I believe that there are relatively simple
things we can do to improve are current assumptions, so stay tuned.
That's enough for now -- the Patriots are on.
K
> At the suggestion of Dr. McGrattan I am posting this preliminary,
> unpublished work to foster discussion and input from the group.
>
> I did a number of runs re: grid sensitivity of an axi-symmetric plume.
> Please see the write-up for some scant details and a few comments. I am
> looking for feedback, comments, suggestions, concerns, etc. as I hope to
> generate a paper from this work. Also any suggestions on improving the
> presentation of results regarding the graphics would also be much
> appreciated. However, this is admittedly not a paper and is not
> presented as such - it contains no references.
> unpublished work to foster discussion and input from the group.
>
> I did a number of runs re: grid sensitivity of an axi-symmetric plume.
> Please see the write-up for some scant details and a few comments. I am
> looking for feedback, comments, suggestions, concerns, etc. as I hope to
> generate a paper from this work. Also any suggestions on improving the
> presentation of results regarding the graphics would also be much
> appreciated. However, this is admittedly not a paper and is not
>
> I look forward to your thoughtful comments.
>
> prelim_release_FDS_group_20080111.pdf
> I look forward to your thoughtful comments.
>
> 863KDownload
Chris
Jan 14, 2008, 2:36:14 PM1/14/08
to FDS and Smokeview Discussions
Thank you Kevin for that very interesting study NUREG-1824. I'm still
reading it. This is my conclusion from the report: FDS 4 is predicting
much too high Smoke Concentrations, 50% higher than the measured data.
I'm very interested to know whether FDS 5 is better! Thank you for
linking this useful report.
Chris
> > 863KDownload- Zitierten Text ausblenden -
>
> - Zitierten Text anzeigen -
reading it. This is my conclusion from the report: FDS 4 is predicting
much too high Smoke Concentrations, 50% higher than the measured data.
I'm very interested to know whether FDS 5 is better! Thank you for
linking this useful report.
Chris
>
> - Zitierten Text anzeigen -
dr_jfloyd
Jan 14, 2008, 3:38:26 PM1/14/08
to FDS and Smokeview Discussions
Pretty much everyone was over predicting the smoke concentration. I
think this results from the general assumption that the commonly used
models make that smoke acts just like the other gas phase combustion
products when in fact some of it will deposit on surfaces. For
example the BE 3 tests were ~ 1 MW for 20 minutes. For illustration
assume 40 kJ/kg and a 2 % soot yield. 30 kg of fuel (0.025 kg for
1200 s) would make ~600 g of soot. 50 % too high a prediction would
require that only 1/3 of the soot deposit on surfaces or 200 g.
Assume this plates out evenly over the upper half of the test
compartment and you would only need a soot layer ~1/3 micron thick.
That would be probably not even be discernable on an orignally white
wall unless you streak a finger across it. It seems entirely likely
that this is the cause of the over prediction. FDS 5 does not do
anything different in terms of soot than FDS 4 did.
> > - Zitierten Text anzeigen -- Hide quoted text -
>
> - Show quoted text -
think this results from the general assumption that the commonly used
models make that smoke acts just like the other gas phase combustion
products when in fact some of it will deposit on surfaces. For
example the BE 3 tests were ~ 1 MW for 20 minutes. For illustration
assume 40 kJ/kg and a 2 % soot yield. 30 kg of fuel (0.025 kg for
1200 s) would make ~600 g of soot. 50 % too high a prediction would
require that only 1/3 of the soot deposit on surfaces or 200 g.
Assume this plates out evenly over the upper half of the test
compartment and you would only need a soot layer ~1/3 micron thick.
That would be probably not even be discernable on an orignally white
wall unless you streak a finger across it. It seems entirely likely
that this is the cause of the over prediction. FDS 5 does not do
anything different in terms of soot than FDS 4 did.
>
> - Show quoted text -
Khalid Moinuddin
Jan 15, 2008, 1:00:55 AM1/15/08
to fds...@googlegroups.com
I did some work on grid sensitivity analysis for pool fire within an ISO
9705 room using FDS4 where the fire size was not prescribed. It appears
that to have proper convergence D* value needs to be in the order of
100. The paper is attached here.
9705 room using FDS4 where the fire size was not prescribed. It appears
that to have proper convergence D* value needs to be in the order of
100. The paper is attached here.
You will find that due to the use of enhnacement routine you can get the
same fire as the converged fire using D* in the order of 10 (see the
powerpoint to see the effect of the routine).
Unfortunately the enhancement routine is excluded from FDS5.
With regards,
Khalid
Kevin
Jan 15, 2008, 8:24:19 AM1/15/08
to FDS and Smokeview Discussions
Let's not leap to conclusions. In FDS versions 2-4 we used a single
mixture fraction variable Z. To get the HRR, we would integrate the
gradient of Z across the flame sheet Z=Z_F, the stoichiometric value.
The "enhancement" was really just a slection of a different value of
Z_F to get the flame in the right place on coarse grids. While it
worked to some extent, it became too difficult to apply in flashover
simulations, plus we have adopted a new multiple mixture fraction
variable formulation in FDS 5 that no longer requires computing
gradients of Z or the "enhancement."
As I said in a previous post, it is difficult to get the right spatial
distribution of heat release rate on coarse meshes. The "enhancement"
was one of several ways to account for coarse grids. We believe that
the methodology in 5 is ultimately more robust, but we cannot expect
the same results in 5 as we had in 4. This is why I want to organize
the grid sensitivity calculations in the Validation Guide.
On Jan 15, 1:00 am, "Khalid Moinuddin" <Khalid.Moinud...@vu.edu.au>
wrote:
> 880KDownload
>
> Enhancement routine.ppt
> 348KDownload
> > > 863KDownload- Hide quoted text -
mixture fraction variable Z. To get the HRR, we would integrate the
gradient of Z across the flame sheet Z=Z_F, the stoichiometric value.
The "enhancement" was really just a slection of a different value of
Z_F to get the flame in the right place on coarse grids. While it
worked to some extent, it became too difficult to apply in flashover
simulations, plus we have adopted a new multiple mixture fraction
variable formulation in FDS 5 that no longer requires computing
gradients of Z or the "enhancement."
As I said in a previous post, it is difficult to get the right spatial
distribution of heat release rate on coarse meshes. The "enhancement"
was one of several ways to account for coarse grids. We believe that
the methodology in 5 is ultimately more robust, but we cannot expect
the same results in 5 as we had in 4. This is why I want to organize
the grid sensitivity calculations in the Validation Guide.
On Jan 15, 1:00 am, "Khalid Moinuddin" <Khalid.Moinud...@vu.edu.au>
wrote:
> I did some work on grid sensitivity analysis for pool fire within an ISO
> 9705 room using FDS4 where the fire size was not prescribed. It appears
> that to have proper convergence D* value needs to be in the order of
> 100. The paper is attached here.
>
> afmc_VPAC_grid-print.pdf
> 9705 room using FDS4 where the fire size was not prescribed. It appears
> that to have proper convergence D* value needs to be in the order of
> 100. The paper is attached here.
>
> 880KDownload
>
> Enhancement routine.ppt
> 348KDownload
Khalid Moinuddin
Jan 24, 2008, 10:16:10 PM1/24/08
to FDS and Smokeview Discussions
Hi Kevin,
I found only one option to reply in that thread, therefore I am
replying through another relevant thread.
Thanks for your reply.
Now looking at the example of section 15.3.1 of the FDS5 User guide, I
feel that if we change the number of radiation angle from 100 to 300 -
the grid sensitivity related to the radiation loss can be improved. I
understand that it will computationally very expensive.
Regards,
Khalid
This information is very valuable, but unfortunately it is only I who
is getting it. Why not post this to the Group?
FYI, I am still trying to improve the grid sensitivity of FDS. The
problem is that all of these percentages are tied to the grid.
The real issue, and that which few have really talked about, is the
upper bound we use for the HRRPUV, which is 200/dx kW/m3. This is
needed because if we did not use a limit, then all combustion would
take place in the nearest cell to the boundary, because the rate of
oxygen and fuel diffusion is orders of magnitude greater on a coarse
mesh of say 10 cm, than it would be for an actual flame. All
practical CFD fire models have some sort of combustion limiter,
whether it be what I have said, or the Eddy Breakup, or whatever.
However, most non-modelers cannot understand these because of the
complicated mix of numerics and combustion physics.
Until we overcome this source of grid sensitivity, talking about
absorption coefficients is moot. It confuses the issue.
Thanks
Kevin
Kevin McGrattan
National Institute of Standards and Technology 100 Bureau Drive, Mail
Stop 8663 Gaithersburg MD 20899
Phone: 301 975 2712
Fax: 301 975 4052
----- Original Message -----
From: "Khalid Moinuddin" <Khalid.M...@vu.edu.au>
To: "Kevin" <mcgr...@gmail.com>
Sent: Thursday, January 24, 2008 1:07 AM
Subject: Re: Absorption coefficient
Hi Kevin,
I ran two ethanol_pan cases with FDS 5.1.0 (one with 5cm cell and the
other with 2.5 cm cell). I removed absorption coefficent and used
Radiation_Fraction=0.35
Radiation loss for 5 cm case is 38.5% and for 2.5 cm is 41% - roughly
the same as you got with your cases.
Integrated HEAT FLUX to the surface is larger for 5 cm case, hence the
higher HRR is obtained.
I have now run two more ethanol_pan cases with FDS 5.1.0 (one with 5cm
cell and the other with 2.5 cm cell). I removed absorption coefficent,
but used Radiation_Fraction=0.0
This time radiation loss for 5 cm case is 25% and for 2.5 cm is
30.5% . This time I have found that the integrated HEAT FLUX to the
surface is lesser for 5 cm case, hence the lower HRR (22%) is
obtained.
So in FDS5, higher radiation loss resulted in lower HRR when
Radiation_Fraction=0.35 and the opposite happened when
Radiation_Fraction=0.0
Now I looked back the simulations with FDS4 with
Radiation_Fraction=0.0 (though FDS4 does not have many features like
FDS5 and has a different combustion model). I find that Radiation loss
for 5 cm case was 21% and for 2.5 cm is 23.5% - which resulted in 5%
higher HRR for 2.5 cm case. So in FDS4, higher radiation loss resulted
in higher HRR when Radiation_Fraction=0.0 which is the same for the
case of FDS5.
I hope this information has some value to you.
Regards,
Khalid
On Nov 13 2007, 1:05 am, Kevin <mcgra...@gmail.com> wrote:
> Khalid
>
> I have done some work on this problem last week. The ethanol_pan case
> is difficult to work with because there are so many parameters. So I
> decided to just look at a simple "burner" in FDS. Nothing more than a
> fixed HRRPUA, with no specified material properties. I had to add a
> new output to FDS in order to integrate the HEAT_FLUX to the burner
> surface. I noticed that for cases with grids of 10 cm, 5 cm and 2.5
> cm, the radiative loss (which is roughly proportional to the
> integrated heat flux to the burner) varied from 35% to 42%. The 35% is
> understandable in the 10 cm case, because FDS is just using the given
> RADIATIVE_FRACTION. In the 2.5 cm case, the sigma*T^4 term is also
> playing a role. In the ethanol_pan case, RADIATIVE_FRACTION=0, and the
> radiative loss is calculated purely from temperature and gas/soot
> composition. It appears that this is the leading cause of grid
> dependence in your case and in mine.
>
> Our goal is to eliminate as much as possible grid dependence in the
> gas phase calculation. We have had good success for total HRR and
> flame height. Now we need to turn our attention to the radiative loss.
> If we succeed in this, then we can be confident that the overall
> spatial distribution of the fire's energy is independent of the grid,
> and we can then turn our attention to the details of the pyrolysis. We
> are trying to decouple gas and solid phase phenomena. Results such as
> yours are a combination of the two. By looking at the burner only, we
> focus on the gas. On the other hand, we have simple examples of just
> the solid (or liquid) phase. Grid dependence is mainly in the gas
> phase. For the solid, we have as fine a grid as we need.
>
> I might bump this thread out into a separate one, to keep open the
> lines of communication, and to let others on the team know what we're
> up to. K
>
> On Nov 1, 11:31 pm, "KhalidMoinuddin" <Khalid.Moinud...@vu.edu.au>
> wrote:
>
>
>
> > Hi Kevin and Simo,...
>
> > read more »
>
> > ethanol_pan_grid.doc
> > 78KDownload
>
> > I have run ethanol_pan case with three different cell sizes i.e. 50 mm,
> > 25 mm and 12.5 mm. The result is attached to this mail. You can see that
> > the results vary greatly.
>
> > I apologise that I did not do this during the beta testing. At that
> > time, I thought that the simulation result of fire in open space would
> > be grid independent at 50 mm cell as found with FDS4 which is shown in
> > reference [36] of FDS5 User's guide.
>
> > Regards,
>
> >Khalid
>
I found only one option to reply in that thread, therefore I am
replying through another relevant thread.
Thanks for your reply.
Now looking at the example of section 15.3.1 of the FDS5 User guide, I
feel that if we change the number of radiation angle from 100 to 300 -
the grid sensitivity related to the radiation loss can be improved. I
understand that it will computationally very expensive.
Regards,
Khalid
This information is very valuable, but unfortunately it is only I who
is getting it. Why not post this to the Group?
FYI, I am still trying to improve the grid sensitivity of FDS. The
problem is that all of these percentages are tied to the grid.
The real issue, and that which few have really talked about, is the
upper bound we use for the HRRPUV, which is 200/dx kW/m3. This is
needed because if we did not use a limit, then all combustion would
take place in the nearest cell to the boundary, because the rate of
oxygen and fuel diffusion is orders of magnitude greater on a coarse
mesh of say 10 cm, than it would be for an actual flame. All
practical CFD fire models have some sort of combustion limiter,
whether it be what I have said, or the Eddy Breakup, or whatever.
However, most non-modelers cannot understand these because of the
complicated mix of numerics and combustion physics.
Until we overcome this source of grid sensitivity, talking about
absorption coefficients is moot. It confuses the issue.
Thanks
Kevin
Kevin McGrattan
National Institute of Standards and Technology 100 Bureau Drive, Mail
Stop 8663 Gaithersburg MD 20899
Phone: 301 975 2712
Fax: 301 975 4052
----- Original Message -----
From: "Khalid Moinuddin" <Khalid.M...@vu.edu.au>
To: "Kevin" <mcgr...@gmail.com>
Sent: Thursday, January 24, 2008 1:07 AM
Subject: Re: Absorption coefficient
Hi Kevin,
I ran two ethanol_pan cases with FDS 5.1.0 (one with 5cm cell and the
other with 2.5 cm cell). I removed absorption coefficent and used
Radiation_Fraction=0.35
Radiation loss for 5 cm case is 38.5% and for 2.5 cm is 41% - roughly
the same as you got with your cases.
Integrated HEAT FLUX to the surface is larger for 5 cm case, hence the
higher HRR is obtained.
I have now run two more ethanol_pan cases with FDS 5.1.0 (one with 5cm
cell and the other with 2.5 cm cell). I removed absorption coefficent,
but used Radiation_Fraction=0.0
This time radiation loss for 5 cm case is 25% and for 2.5 cm is
30.5% . This time I have found that the integrated HEAT FLUX to the
surface is lesser for 5 cm case, hence the lower HRR (22%) is
obtained.
So in FDS5, higher radiation loss resulted in lower HRR when
Radiation_Fraction=0.35 and the opposite happened when
Radiation_Fraction=0.0
Now I looked back the simulations with FDS4 with
Radiation_Fraction=0.0 (though FDS4 does not have many features like
FDS5 and has a different combustion model). I find that Radiation loss
for 5 cm case was 21% and for 2.5 cm is 23.5% - which resulted in 5%
higher HRR for 2.5 cm case. So in FDS4, higher radiation loss resulted
in higher HRR when Radiation_Fraction=0.0 which is the same for the
case of FDS5.
I hope this information has some value to you.
Regards,
Khalid
On Nov 13 2007, 1:05 am, Kevin <mcgra...@gmail.com> wrote:
> Khalid
>
> I have done some work on this problem last week. The ethanol_pan case
> is difficult to work with because there are so many parameters. So I
> decided to just look at a simple "burner" in FDS. Nothing more than a
> fixed HRRPUA, with no specified material properties. I had to add a
> new output to FDS in order to integrate the HEAT_FLUX to the burner
> surface. I noticed that for cases with grids of 10 cm, 5 cm and 2.5
> cm, the radiative loss (which is roughly proportional to the
> integrated heat flux to the burner) varied from 35% to 42%. The 35% is
> understandable in the 10 cm case, because FDS is just using the given
> RADIATIVE_FRACTION. In the 2.5 cm case, the sigma*T^4 term is also
> playing a role. In the ethanol_pan case, RADIATIVE_FRACTION=0, and the
> radiative loss is calculated purely from temperature and gas/soot
> composition. It appears that this is the leading cause of grid
> dependence in your case and in mine.
>
> Our goal is to eliminate as much as possible grid dependence in the
> gas phase calculation. We have had good success for total HRR and
> flame height. Now we need to turn our attention to the radiative loss.
> If we succeed in this, then we can be confident that the overall
> spatial distribution of the fire's energy is independent of the grid,
> and we can then turn our attention to the details of the pyrolysis. We
> are trying to decouple gas and solid phase phenomena. Results such as
> yours are a combination of the two. By looking at the burner only, we
> focus on the gas. On the other hand, we have simple examples of just
> the solid (or liquid) phase. Grid dependence is mainly in the gas
> phase. For the solid, we have as fine a grid as we need.
>
> I might bump this thread out into a separate one, to keep open the
> lines of communication, and to let others on the team know what we're
> up to. K
>
> On Nov 1, 11:31 pm, "KhalidMoinuddin" <Khalid.Moinud...@vu.edu.au>
> wrote:
>
>
>
> > Hi Kevin and Simo,...
>
> > read more »
>
> > ethanol_pan_grid.doc
> > 78KDownload
>
> > I have run ethanol_pan case with three different cell sizes i.e. 50 mm,
> > 25 mm and 12.5 mm. The result is attached to this mail. You can see that
> > the results vary greatly.
>
> > I apologise that I did not do this during the beta testing. At that
> > time, I thought that the simulation result of fire in open space would
> > be grid independent at 50 mm cell as found with FDS4 which is shown in
> > reference [36] of FDS5 User's guide.
>
> > Regards,
>
> >Khalid
>
Kevin
Jan 25, 2008, 8:25:06 AM1/25/08
to FDS and Smokeview Discussions
Khalid
I always use the Discussion Group interface to read the threads, and I
make entries by hitting "Reply." If you hit "Reply to Author" it is
like emailing the author at his or her gmail account.
Increasing the number of radiation angles can help in some cases. Best
way to determine this is to look at the pattern of RADIATIVE_FLUX on
the solid surfaces. If you see a star-like pattern in the area of
interest, then you should increase the number of angles until you get
a smoother pattern. The star pattern is a direct indication of a
limited number of solid angles.
As for cost, one way to play it is to increase the number of angles,
but decrease the frequency of updating the radiation solver. Simo
Hostikka cleverly designed into his radiation solver the ability to
control both the spatial and temporal resolution of the RTE. By
default, a full update of the RTE is done every 15 time steps (as I
recall, every fifth angle is updated once every 3 time step). In a
typical FDS calc, the time step is on the order of a hundredth of a
second, thus the radiation is solved every 0.15 s. Unless you actually
want to resolve the pulsations of heat on a near-field target, you may
be able to relax this number considerably.
K
On Jan 24, 10:16 pm, Khalid Moinuddin <Khalid.Moinud...@vu.edu.au>
wrote:
> > >Khalid- Hide quoted text -
I always use the Discussion Group interface to read the threads, and I
make entries by hitting "Reply." If you hit "Reply to Author" it is
like emailing the author at his or her gmail account.
Increasing the number of radiation angles can help in some cases. Best
way to determine this is to look at the pattern of RADIATIVE_FLUX on
the solid surfaces. If you see a star-like pattern in the area of
interest, then you should increase the number of angles until you get
a smoother pattern. The star pattern is a direct indication of a
limited number of solid angles.
As for cost, one way to play it is to increase the number of angles,
but decrease the frequency of updating the radiation solver. Simo
Hostikka cleverly designed into his radiation solver the ability to
control both the spatial and temporal resolution of the RTE. By
default, a full update of the RTE is done every 15 time steps (as I
recall, every fifth angle is updated once every 3 time step). In a
typical FDS calc, the time step is on the order of a hundredth of a
second, thus the radiation is solved every 0.15 s. Unless you actually
want to resolve the pulsations of heat on a near-field target, you may
be able to relax this number considerably.
K
On Jan 24, 10:16 pm, Khalid Moinuddin <Khalid.Moinud...@vu.edu.au>
wrote:
Khalid Moinuddin
Jan 28, 2008, 6:45:30 PM1/28/08
to fds...@googlegroups.com
Thank you, Kevin.
I will run some simulations by increasing the number of angles and decreasing the frequency of updating the radiation solver.
Regards,
Khalid
-----Original Message-----
From: fds...@googlegroups.com [mailto:fds...@googlegroups.com] On Behalf Of Kevin
Sent: Saturday, 26 January 2008 12:25 AM
To: FDS and Smokeview Discussions
Kevin
Jan 29, 2008, 8:23:49 AM1/29/08
to FDS and Smokeview Discussions
Great, thanks. While you're at it, look at the following outputs:
&SLCF PBY=..., QUANTITY='HRRPUV' /
&SLCF XB=0.0,0.0,0.0,0.0,0.0,3.6, QUANTITY='HRRPUL' / Directly on
fire plume centerline, from bottom to top
&DEVC XB=-0.4,0.4,-0.4,0.4,0.10,0.10,
QUANTITY='HEAT_FLUX',IOR=3,ID='int flux', STATISTICS='SURFACE
INTEGRAL', SURF_ID='BURNER' /
&DEVC XB=-0.4,0.4,-0.4,0.4,0.10,0.10,
QUANTITY='CONVECTIVE_FLUX',IOR=3,ID='int flux', STATISTICS='SURFACE
INTEGRAL',SURF_ID='BURNER' /
&DEVC XB=-0.4,0.4,-0.4,0.4,0.10,0.10,
QUANTITY='RADIATIVE_FLUX',IOR=3,ID='int flux', STATISTICS='SURFACE
INTEGRAL', SURF_ID='BURNER' /
Some of these are not yet listed in the Guide. HRRPUL is the HRR Per
Unit Length. It is aligned with the plume centerline, and it records
the integral of HRRPUV in the x and y directions. It tells us the
amount of heat released as a function of height over the burner. The
DEVC's do integrals of the heat fluxes across the burner. Ideally, we
want these quantities all to be relatively insensitive to grid size.
If HRRPUL is not, then the heat flux to the surface is not going to be
either.
We're currently exploring ways to make the distribution of HRRPUV
insensitive to grid. That's what drives everything, and it is more
than just a matter of flame height, which is what we have focussed on
in the past.
On Jan 28, 6:45 pm, "Khalid Moinuddin" <Khalid.Moinud...@vu.edu.au>
> > - Show quoted text -- Hide quoted text -
&SLCF PBY=..., QUANTITY='HRRPUV' /
&SLCF XB=0.0,0.0,0.0,0.0,0.0,3.6, QUANTITY='HRRPUL' / Directly on
fire plume centerline, from bottom to top
&DEVC XB=-0.4,0.4,-0.4,0.4,0.10,0.10,
QUANTITY='HEAT_FLUX',IOR=3,ID='int flux', STATISTICS='SURFACE
INTEGRAL', SURF_ID='BURNER' /
&DEVC XB=-0.4,0.4,-0.4,0.4,0.10,0.10,
QUANTITY='CONVECTIVE_FLUX',IOR=3,ID='int flux', STATISTICS='SURFACE
INTEGRAL',SURF_ID='BURNER' /
&DEVC XB=-0.4,0.4,-0.4,0.4,0.10,0.10,
QUANTITY='RADIATIVE_FLUX',IOR=3,ID='int flux', STATISTICS='SURFACE
INTEGRAL', SURF_ID='BURNER' /
Some of these are not yet listed in the Guide. HRRPUL is the HRR Per
Unit Length. It is aligned with the plume centerline, and it records
the integral of HRRPUV in the x and y directions. It tells us the
amount of heat released as a function of height over the burner. The
DEVC's do integrals of the heat fluxes across the burner. Ideally, we
want these quantities all to be relatively insensitive to grid size.
If HRRPUL is not, then the heat flux to the surface is not going to be
either.
We're currently exploring ways to make the distribution of HRRPUV
insensitive to grid. That's what drives everything, and it is more
than just a matter of flame height, which is what we have focussed on
in the past.
On Jan 28, 6:45 pm, "Khalid Moinuddin" <Khalid.Moinud...@vu.edu.au>
Dave McGill
Feb 11, 2008, 11:36:06 AM2/11/08
to FDS and Smokeview Discussions
Hi Chris,
I have a few minor suggestions for your graphs.
1. Figures 1-5
The lines on these graphs curve back on themselves. I would plot the
independent variable on the x-axis and the dependent variable on the y-
axis. (No-one else has mentioned this so I may harbouring an old-
school convention.)
2. Use a consistent value for the maximum time.
3. Delete the (default) horizontal grid lines.
4. I assume you are submitting this to a traditional print
publication where colour is not available. I looked at a couple of
journals, the maximum graph size is in the range of 3 1/2" x 2 1/2".
Your only way to distinguish the different lines is by thickness,
dashes and gray-scale. I think you will have to drastically reduce
the number of lines per graph. (There is a technique I've seen in one
of Edward Tufte's books that comprises a series of small graphs. I can
find the reference if you are interested.)
I hope this helps.
Regards
Dave
I have a few minor suggestions for your graphs.
1. Figures 1-5
The lines on these graphs curve back on themselves. I would plot the
independent variable on the x-axis and the dependent variable on the y-
axis. (No-one else has mentioned this so I may harbouring an old-
school convention.)
2. Use a consistent value for the maximum time.
3. Delete the (default) horizontal grid lines.
4. I assume you are submitting this to a traditional print
publication where colour is not available. I looked at a couple of
journals, the maximum graph size is in the range of 3 1/2" x 2 1/2".
Your only way to distinguish the different lines is by thickness,
dashes and gray-scale. I think you will have to drastically reduce
the number of lines per graph. (There is a technique I've seen in one
of Edward Tufte's books that comprises a series of small graphs. I can
find the reference if you are interested.)
I hope this helps.
Regards
Dave
Kevin
Feb 11, 2008, 11:47:10 AM2/11/08
to FDS and Smokeview Discussions
Dave -- I suggested to Chris above that this work will be valuable
within the Validation Guide, which we're currently putting together.
We're currently experimenting with various techniques to reduce the
grid dependence, especially in the near field region (in the fire,
essentially). That is what I noticed right away in Chris' plots -- the
temperatures in the fire are very grid dependent. If Chris were to
publish this now, we will probably change FDS before it even goes to
press. What we really need is a standard "plume" case, along with the
grid sensitivity and comparison to the correlation/experiment. Then,
as we improve FDS, we can rerun the cases to make sure that we are
getting better, not worse.
within the Validation Guide, which we're currently putting together.
We're currently experimenting with various techniques to reduce the
grid dependence, especially in the near field region (in the fire,
essentially). That is what I noticed right away in Chris' plots -- the
temperatures in the fire are very grid dependent. If Chris were to
publish this now, we will probably change FDS before it even goes to
press. What we really need is a standard "plume" case, along with the
grid sensitivity and comparison to the correlation/experiment. Then,
as we improve FDS, we can rerun the cases to make sure that we are
getting better, not worse.
john
Feb 12, 2008, 4:06:08 AM2/12/08
to FDS and Smokeview Discussions
Hi All,
This is excellent type of work for beginners and students, because it
makes you read SFPE etc. to obtain the comparative equation such as
Alpert or Heskestad. This is how I started out with FDS in Version
4, and I then compared the plume centreline temperatures with
Heskestad. I did not obtain good agreement until I had at least 11-18
cells across the burner. I then extended this for a larger compartment
and used all sort of meshing schemes. I found that an ultra-fast fire
growing to 20MW in 25m high space (6850sqm) with 11cells across (250mm
grid) the burner increasing to 500mm and then 1000mm in the furthest
regions did not give me as good smoke layer definition in comparison
to the 250mm grid throughout in fact up to 5m height differences were
observed.
I know the following does not apply to Christopher's results but I
have noticed that some FDS plume centreline numerical experiments do
not employ a free-burning plume, or the fire has developed beyond the
ceiling-jet phase. I think I am correct here in assuming for such
that Alpert and Heskestad comparative equations are being used beyond
their experimental limits. It would be good to see grid sensitivity
studies using FDS that are being presented to NIST or Conference
highlighting briefly the comparative correlation limitations.
Best Regards, John
This is excellent type of work for beginners and students, because it
makes you read SFPE etc. to obtain the comparative equation such as
Alpert or Heskestad. This is how I started out with FDS in Version
4, and I then compared the plume centreline temperatures with
Heskestad. I did not obtain good agreement until I had at least 11-18
cells across the burner. I then extended this for a larger compartment
and used all sort of meshing schemes. I found that an ultra-fast fire
growing to 20MW in 25m high space (6850sqm) with 11cells across (250mm
grid) the burner increasing to 500mm and then 1000mm in the furthest
regions did not give me as good smoke layer definition in comparison
to the 250mm grid throughout in fact up to 5m height differences were
observed.
I know the following does not apply to Christopher's results but I
have noticed that some FDS plume centreline numerical experiments do
not employ a free-burning plume, or the fire has developed beyond the
ceiling-jet phase. I think I am correct here in assuming for such
that Alpert and Heskestad comparative equations are being used beyond
their experimental limits. It would be good to see grid sensitivity
studies using FDS that are being presented to NIST or Conference
highlighting briefly the comparative correlation limitations.
Best Regards, John
Kevin
Feb 12, 2008, 8:27:56 AM2/12/08
to FDS and Smokeview Discussions
I am currently taking another look at McCaffrey's original fire plume
measurements (a copy of the report is in the "Library" linked to the
FDS homepage). I've nothing against Alpert or Heskestad, but McCaffrey
is my mother's brother, so family loyalty dictates that I use his
data. Plus, he did his work at NBS (National Bureau of Standards), the
former name of NIST. So there are people here who remember the
project. McCaffrey's correlation resulted from 5 methane fires of
sizes 14 kW, 22 kW, 35 kW, 48 kW and 57 kW (more or less), all in a 30
cm by 30 cm square sand burner. Perfect for FDS simulation. I'm
currently tracking down the individual measurements of each fire. The
correlation is a pretty good fit to the data, but there are subtle
trends tied to the fire size that I'd like to see if the model
captures. Of particular interest to me is the near field, the so-
called flame and intermittent regions. I've focussed mainly on the
plume region in the past, because our primary interest in smoke
movement apps was to get the right entrainment and dilution of the
plume. Near-field effects were not as important. In most large scale
smoke movement calcs, people do not use enough grid cells over the
fire to capture anything but the plume. That's OK so long as the
objective of the calc is to assess smoke filling/movement. But if
you're interested in near-field effects, like radiative flux to
targets or fire spread, you've got to resolve the fire better. As
always, our goal is to be as accurate as we can using grids of modest
resolution. DNS calcs are great fun, but rarely of practical value.
measurements (a copy of the report is in the "Library" linked to the
FDS homepage). I've nothing against Alpert or Heskestad, but McCaffrey
is my mother's brother, so family loyalty dictates that I use his
data. Plus, he did his work at NBS (National Bureau of Standards), the
former name of NIST. So there are people here who remember the
project. McCaffrey's correlation resulted from 5 methane fires of
sizes 14 kW, 22 kW, 35 kW, 48 kW and 57 kW (more or less), all in a 30
cm by 30 cm square sand burner. Perfect for FDS simulation. I'm
currently tracking down the individual measurements of each fire. The
correlation is a pretty good fit to the data, but there are subtle
trends tied to the fire size that I'd like to see if the model
captures. Of particular interest to me is the near field, the so-
called flame and intermittent regions. I've focussed mainly on the
plume region in the past, because our primary interest in smoke
movement apps was to get the right entrainment and dilution of the
plume. Near-field effects were not as important. In most large scale
smoke movement calcs, people do not use enough grid cells over the
fire to capture anything but the plume. That's OK so long as the
objective of the calc is to assess smoke filling/movement. But if
you're interested in near-field effects, like radiative flux to
targets or fire spread, you've got to resolve the fire better. As
always, our goal is to be as accurate as we can using grids of modest
resolution. DNS calcs are great fun, but rarely of practical value.
cwood
Feb 12, 2008, 10:12:04 AM2/12/08
to FDS and Smokeview Discussions
Kevin,
I was going to move to the Heskestad equations. Notwithstanding
family loyalty, the McCaffrey fires are all quite small in terms of
"typical" fire sizes in design and forensics. I believe that
Heskestad has a larger range of fire sizes. Do you think that has any
impact in the applicability of the McCaffrey equations regarding
general applicability of FDS? The information seems quite good for
the use you discussed.
In one of your comments on near-field effects you mention, "radiative
flux to targets." I just want to make sure that I understand the
context in which you make this statement. In the Tech Ref (SVN 1132)
it refers to the inability to accurately predict the temperature and
therefore the calculation method is modified from using the cell
temperature as the source energy. However, I am not clear to the
extent to which the "flame zone" extends or at what physical limit the
equation applies. If you could explain this further I would
appreciate it.
Christopher
I was going to move to the Heskestad equations. Notwithstanding
family loyalty, the McCaffrey fires are all quite small in terms of
"typical" fire sizes in design and forensics. I believe that
Heskestad has a larger range of fire sizes. Do you think that has any
impact in the applicability of the McCaffrey equations regarding
general applicability of FDS? The information seems quite good for
the use you discussed.
In one of your comments on near-field effects you mention, "radiative
flux to targets." I just want to make sure that I understand the
context in which you make this statement. In the Tech Ref (SVN 1132)
it refers to the inability to accurately predict the temperature and
therefore the calculation method is modified from using the cell
temperature as the source energy. However, I am not clear to the
extent to which the "flame zone" extends or at what physical limit the
equation applies. If you could explain this further I would
appreciate it.
Christopher
Kevin
Feb 12, 2008, 11:16:52 AM2/12/08
to FDS and Smokeview Discussions
I prefer to compare FDS calcs directly to experiments. Correlations
are great for collapsing lots of data into easy to use formulae, but
these correlations are not exactly the experimental results. Over the
years, people have complained when FDS does "converge" to a particular
correlation. It shouldn't, as a correlation is an average of lots of
experiments.
We can include all the correlations in the V&V Guide, in fact, it is
good to show that there is a variation. But I also want to show that
FDS can reproduce, to whatever extent, the behavior of a very specific
fire.
As for radiation -- in practical calcs, FDS takes a fixed fraction of
the HRRPUV (HRR Per Unit Volume) as the source in the radiation
transport equation. Temperature and soot concentration cannot be
relied upon in a coarse calc. So long as we're matching the total HRR
and its spatial distribution, our raditive fluxes to nearby targets/
walls shouldn't be too bad. What your sensitivity analysis will show
is that FDS gets the flame height right, but as you refine the grid,
you see that the distribution of HRRPUV (check the FDS output
quantity) varies more than we'd like. If HRRPUV is relatively grid
insensitive, and we apply the fixed "chi_R" rule, then we should see
less sensitivity in flame spread calcs.
are great for collapsing lots of data into easy to use formulae, but
these correlations are not exactly the experimental results. Over the
years, people have complained when FDS does "converge" to a particular
correlation. It shouldn't, as a correlation is an average of lots of
experiments.
We can include all the correlations in the V&V Guide, in fact, it is
good to show that there is a variation. But I also want to show that
FDS can reproduce, to whatever extent, the behavior of a very specific
fire.
As for radiation -- in practical calcs, FDS takes a fixed fraction of
the HRRPUV (HRR Per Unit Volume) as the source in the radiation
transport equation. Temperature and soot concentration cannot be
relied upon in a coarse calc. So long as we're matching the total HRR
and its spatial distribution, our raditive fluxes to nearby targets/
walls shouldn't be too bad. What your sensitivity analysis will show
is that FDS gets the flame height right, but as you refine the grid,
you see that the distribution of HRRPUV (check the FDS output
quantity) varies more than we'd like. If HRRPUV is relatively grid
insensitive, and we apply the fixed "chi_R" rule, then we should see
less sensitivity in flame spread calcs.
Khalid Moinuddin
Feb 12, 2008, 8:07:50 PM2/12/08
to FDS and Smokeview Discussions
Because good agreement is obtained when there are at least 11-18 cells
across the burner - does not mean that it is grid-independent result.
Please refer to section 6.3.5 of FDS5 user guide which states "In
general, you should build an FDS input file using a relatively coarse
mesh, and then gradually refine the mesh until you do not see
appreciable differences in your results." Any CFD result has little
meaning unless it is grid-independent irrespective of having good
agreement with the experiemental result or correlation.
Regards,
Khalid
. I then extended this for a larger compartment
> and used all sort of meshing schemes. I found that an ultra-fast fire
> growing to 20MW in 25m high space (6850sqm) with 11cells across (250mm
> grid) the burner increasing to 500mm and then 1000mm in the furthest
> regions did not give me as good smoke layer definition in comparison
> to the 250mm grid throughout in fact up to 5m height differences were
> observed.
Kevin
Feb 13, 2008, 8:47:34 AM2/13/08
to FDS and Smokeview Discussions
I agree. I often see various rules of thumb about grid resolution,
either in terms of an "optimal" grid cell size or number of cells
spanning the fire. A better indicator of resolution is D*/dx, where D*
is a characteristic fire diameter based on the total HRR, and dx is
the cell size. For fires with Q*=1, D* is the actual fire diameter, in
which case D*/dx is literally the number of cells spanning the burner.
But be careful. Sometimes Q* is very small -- a big burner with very
low fuel flow rate. This fire is not as well "resolved" with 10 cells
spanning the burner as would be a fire of Q*=1.
All this being said, relatively low values of Q*/dx may capture far-
field plume behavior reasonably well, but not near field. And as I've
said in the above posts, we're still working on getting convergence in
near field behavior for relatively high values of D*/dx.
I'd like to see another quantity like D*/dx that we can apply to flame
spread problems. D*/dx falls out of the non-dimensional form of the
Navier-Stokes equations. Thus, it is really just a fluid dynamics
scaling parameter. Flame spread involves additional heat transfer
parameters that are not captured by D*/dx. Scale modelers -- have at
it!
On Feb 12, 8:07 pm, Khalid Moinuddin <Khalid.Moinud...@vu.edu.au>
wrote:
> > Best Regards, John- Hide quoted text -
either in terms of an "optimal" grid cell size or number of cells
spanning the fire. A better indicator of resolution is D*/dx, where D*
is a characteristic fire diameter based on the total HRR, and dx is
the cell size. For fires with Q*=1, D* is the actual fire diameter, in
which case D*/dx is literally the number of cells spanning the burner.
But be careful. Sometimes Q* is very small -- a big burner with very
low fuel flow rate. This fire is not as well "resolved" with 10 cells
spanning the burner as would be a fire of Q*=1.
All this being said, relatively low values of Q*/dx may capture far-
field plume behavior reasonably well, but not near field. And as I've
said in the above posts, we're still working on getting convergence in
near field behavior for relatively high values of D*/dx.
I'd like to see another quantity like D*/dx that we can apply to flame
spread problems. D*/dx falls out of the non-dimensional form of the
Navier-Stokes equations. Thus, it is really just a fluid dynamics
scaling parameter. Flame spread involves additional heat transfer
parameters that are not captured by D*/dx. Scale modelers -- have at
it!
On Feb 12, 8:07 pm, Khalid Moinuddin <Khalid.Moinud...@vu.edu.au>
wrote:
Reply all
Reply to author
Forward
0 new messages