Meshes for fires in atriums - all the way up?
355 views
Skip to first unread message
Henrik Nordenstedt
Oct 6, 2010, 10:23:33 AM10/6/10
to FDS and Smokeview Discussions
Hi all,
I'm about to model an atrium in a hospital 33 metres high. I want to
model both 5 and 10 MW fires to check sensitivity for both smoke
exhaust designs but also temperatures to surfaces surrounding the
atria.
The geometry is pretty simple except for some walk ways and
"balconies" to center and sides that obstruct free smoke movement in
some parts.
How do you go about setting up meshes for this high-type buildings. If
I understand the manuals correctly I don't want to divide the flow
into several coarser meshes vertically, because a lot of reactions and
velocity gradients take place there. On the other hand, if I keep a
vertical relatively small mesh all the way to the top (33 metres),
nodecount and runtime skyrocket.
A) Is there a good way of maintaining credible results without keeping
the mesh in the vertical dimension?
B) For some of the cases the geometries are simple enough to allow
handcalcs and some are not, depending of the placement of the fire.
Also for atria as high as this one there are risks of stratification
of the smoke when bouyancy is lost due to temperature loss to the
surrounding surfaces. (of course in which case heat transfer to the
side no longer are a problem.
Does anyone have experiences with modelling high atrias?
Cheers,
Henrik
I'm about to model an atrium in a hospital 33 metres high. I want to
model both 5 and 10 MW fires to check sensitivity for both smoke
exhaust designs but also temperatures to surfaces surrounding the
atria.
The geometry is pretty simple except for some walk ways and
"balconies" to center and sides that obstruct free smoke movement in
some parts.
How do you go about setting up meshes for this high-type buildings. If
I understand the manuals correctly I don't want to divide the flow
into several coarser meshes vertically, because a lot of reactions and
velocity gradients take place there. On the other hand, if I keep a
vertical relatively small mesh all the way to the top (33 metres),
nodecount and runtime skyrocket.
A) Is there a good way of maintaining credible results without keeping
the mesh in the vertical dimension?
B) For some of the cases the geometries are simple enough to allow
handcalcs and some are not, depending of the placement of the fire.
Also for atria as high as this one there are risks of stratification
of the smoke when bouyancy is lost due to temperature loss to the
surrounding surfaces. (of course in which case heat transfer to the
side no longer are a problem.
Does anyone have experiences with modelling high atrias?
Cheers,
Henrik
brad c
Oct 6, 2010, 1:23:00 PM10/6/10
to FDS and Smokeview Discussions
Henrik, except for worrying much about temperatures to surfaces and
the 10 MW fire, I have a little experience. I did hand calcs from the
building code, then modeled the same thing with 1 meter cells-
everything appeared to be ok. From there I input the exits, since the
make-up air was open doors-- I input the max over-pressure of the
fans, and modeled some prevailing wind speed and direction- still with
1 meter cells- i increased the exhaust rate to above Bldg Code
minimum. Your fire in the middle will likely not be the most
demanding, and not affect the temp of the surroundings much- this is
the one you can do by hand?
You are worried about the 'Spill Plume'? Why 10MW? Will the atrium be
used to store 20 big dry Christmas trees? :)
the 10 MW fire, I have a little experience. I did hand calcs from the
building code, then modeled the same thing with 1 meter cells-
everything appeared to be ok. From there I input the exits, since the
make-up air was open doors-- I input the max over-pressure of the
fans, and modeled some prevailing wind speed and direction- still with
1 meter cells- i increased the exhaust rate to above Bldg Code
minimum. Your fire in the middle will likely not be the most
demanding, and not affect the temp of the surroundings much- this is
the one you can do by hand?
You are worried about the 'Spill Plume'? Why 10MW? Will the atrium be
used to store 20 big dry Christmas trees? :)
dr_jfloyd
Oct 6, 2010, 2:06:56 PM10/6/10
to FDS and Smokeview Discussions
If you are going to be using a CFD model to design life safety
systems, you need to use that model in a manner that is likely to
obtain accurate results. In some cases, the size of the facility
being modeled and the size of the design fire are such that large
numbers of grid cells are required and run times will be lengthy. It
is not unusual for atrium calcs I have done to have run times of 1 to
2 weeks or more. Run times can be reduced by using multiple computers
or by reducing grid resolution in "uninteresting" regions of the
domain (for example areas that serve no purpose in the calculation
other than providing the appropriate free air volume but are not
anticipated to have any significant flow develop in them).
One of the most important modeling decisions to make is the grid size
as one needs to have adequate grid resolution to be able to resolve
the important flow structures. If you are doing a smoke control
system, those flow structures are tied to the length scale of the fire
plume and if not resolved, you will not be resolving the entrainment
into the plume. A 1 m grid would be incapable of resolving any flow
structures smaller than about 3 m in size (or about one typical floor
height) as it takes a few cells to generate a vortex.
systems, you need to use that model in a manner that is likely to
obtain accurate results. In some cases, the size of the facility
being modeled and the size of the design fire are such that large
numbers of grid cells are required and run times will be lengthy. It
is not unusual for atrium calcs I have done to have run times of 1 to
2 weeks or more. Run times can be reduced by using multiple computers
or by reducing grid resolution in "uninteresting" regions of the
domain (for example areas that serve no purpose in the calculation
other than providing the appropriate free air volume but are not
anticipated to have any significant flow develop in them).
One of the most important modeling decisions to make is the grid size
as one needs to have adequate grid resolution to be able to resolve
the important flow structures. If you are doing a smoke control
system, those flow structures are tied to the length scale of the fire
plume and if not resolved, you will not be resolving the entrainment
into the plume. A 1 m grid would be incapable of resolving any flow
structures smaller than about 3 m in size (or about one typical floor
height) as it takes a few cells to generate a vortex.
Henrik Nordenstedt
Oct 6, 2010, 5:30:48 PM10/6/10
to FDS and Smokeview Discussions
Areas no directly under the atria are sprinkled and lesser of a
problem.
Ref to Kristoffer Overholt a moderate grid size would be about 0,113 m
x 0,113 m x 0,113 m or about 5,600,000 cells. Which I can do using
multiple cores/computers. However there is no way I could half that to
check for grid sensitivity.
When approaching the upper limit of nodes I can handle, how do I check
this?
Could I verify the approach with hand calcs to check if they show
similar results? In this case, is it not just faster to just go with
hand calcs?
Since it's relatively simple geometries, could I ignore balconies etc
for effects and just go atria hand calcs?
Thanks for your help guys!
Regards,
Henrik
problem.
Ref to Kristoffer Overholt a moderate grid size would be about 0,113 m
x 0,113 m x 0,113 m or about 5,600,000 cells. Which I can do using
multiple cores/computers. However there is no way I could half that to
check for grid sensitivity.
When approaching the upper limit of nodes I can handle, how do I check
this?
Could I verify the approach with hand calcs to check if they show
similar results? In this case, is it not just faster to just go with
hand calcs?
Since it's relatively simple geometries, could I ignore balconies etc
for effects and just go atria hand calcs?
Thanks for your help guys!
Regards,
Henrik
Kevin
Oct 6, 2010, 5:50:23 PM10/6/10
to FDS and Smokeview Discussions
I suggest you compare the vertical profile of plume temperature and
velocity with correlations of either Heskestad or McCaffrey. Then
decide if the accuracy of the calculation is sufficient for your
analysis. You can do these calculations in the open, without including
the building. Just model an open plume with the same fire size, grid
spacing, domain height as your building.
> > > > Henrik- Hide quoted text -
>
> - Show quoted text -
velocity with correlations of either Heskestad or McCaffrey. Then
decide if the accuracy of the calculation is sufficient for your
analysis. You can do these calculations in the open, without including
the building. Just model an open plume with the same fire size, grid
spacing, domain height as your building.
>
> - Show quoted text -
Rein
Oct 7, 2010, 3:24:03 AM10/7/10
to FDS and Smokeview Discussions, Candido G Montes
You might find of interest the work of the group at University of
Jaen. They have a recent series of papers (see list and links bellow)
in the journal 'Building and Environment' simulating fires inside a 20
m cubic atrium and comparing to experimental data. The data set is
most impressive. The fires range from 1.2 to 2.3 MW, and the models
are FLUENT and FDSv4. The data collected is quite extensive (includes
gas-phase and wall temperatures, and heat fluxes) and the effect of
the mechanical/natural ventilation, make-up air velocities and vents
are also investigated.
In particular, the effect of the grid size for FDSv4 vs. plume theory
and vs. experimental data is investigated in Figs 9 and 10 in [2].
Once the best grid is chosen, comparison with detail experimental data
is provided. The conclusion is "It has been observed that FDS4
simulations significantly over-predict by 40–80% the plume temperature
near the flame (below 9 m) but only slightly over-predict by 10–25%
the plume temperature above 9 m. [...] At the exhaust fans and the
upper parts near the walls (above 10 m) the agreement is good too, and
at the lower parts of the walls the predictions are poor, as with the
plume."
The list of papers in the series is:
[1] Numerical model and validation experiments of atrium enclosure
fire in a new fire test
facility. Building and Environment 43 (11), pp 1912–28, 2008.
http://dx.doi.org/10.1016/j.buildenv.2007.11.010
[2] Experimental Data and Numerical Modelling of 1.3 and 2.3 MW Fires
in a 20 m Cubic Atrium, Building and Environment 44, pp. 1827–1839,
2009. http://dx.doi.org/10.1016/j.buildenv.2008.12.010. Open access
version at http://www.era.lib.ed.ac.uk/handle/1842/2761
[3] Influence of different make-up air configurations on the fire-
induced conditions in an atrium, Building and Environment 45 (11), pp
2458-2472, 2010. http://dx.doi.org/10.1016/j.buildenv.2010.05.006.
G.
brad c
Oct 8, 2010, 1:08:35 PM10/8/10
to FDS and Smokeview Discussions
Thanks Dr. Floyd, great points, but do not worry, especially when it
comes to modeling real projects, I am the x-ray technician (i know how
to run the machine) not the Doctor! I had made up some idealized
atriums with all the numbers straight from the Bldg code, forced
supply air- everything smooth and nice. The mixture fraction iso rose
and hoovered a little way above the elevation I had used in the hand
calc for the last several hundred sec of the 1200. I guess i was
lulled into thinking a 1 m model was at least as good as the hand
calcs. Then I got an atrium that backed up to the side of hill, so the
front exit was on floor 1 and the back exit on floor 3- and those
doors were the make-up air. (how do you limit the speed of the make-up
air in the direction of the fire using doors?). I actually thought "uh
oh, for these fans to work, the back exit doors would have to remain
closed during an emergency"! I knew any outside wind (western Kansas)
would only aggravate things. The runs were taking about 18 hours with
the 1m cells, so I did a few, and when I turned it over to the pros
for diagnosis, I mentioned that I did not think the minimum from the
Code was going to be adequate in this case. I got back to work on
sprinklers and forgot to ask about the final outcome.
* * * * * *
I am still very curious, Henrik, why 10MW?
comes to modeling real projects, I am the x-ray technician (i know how
to run the machine) not the Doctor! I had made up some idealized
atriums with all the numbers straight from the Bldg code, forced
supply air- everything smooth and nice. The mixture fraction iso rose
and hoovered a little way above the elevation I had used in the hand
calc for the last several hundred sec of the 1200. I guess i was
lulled into thinking a 1 m model was at least as good as the hand
calcs. Then I got an atrium that backed up to the side of hill, so the
front exit was on floor 1 and the back exit on floor 3- and those
doors were the make-up air. (how do you limit the speed of the make-up
air in the direction of the fire using doors?). I actually thought "uh
oh, for these fans to work, the back exit doors would have to remain
closed during an emergency"! I knew any outside wind (western Kansas)
would only aggravate things. The runs were taking about 18 hours with
the 1m cells, so I did a few, and when I turned it over to the pros
for diagnosis, I mentioned that I did not think the minimum from the
Code was going to be adequate in this case. I got back to work on
sprinklers and forgot to ask about the final outcome.
* * * * * *
I am still very curious, Henrik, why 10MW?
Henrik Nordenstedt
Oct 10, 2010, 11:18:06 AM10/10/10
to FDS and Smokeview Discussions
Kevin: Can I model an open fire plume like a very high shaft open to
the surrounding or do I have to include a larger part of the plume?
When checking for temperatures and velocity both I'm really only
interested in comparing the maximum (centerline) to empirical data? Do
I miss important turbulence parameters by modeling this way?
If using more than one mesh, how can I make a statement about that the
coarser mesh is providing me with accurate enough results? With a
accurate enough fire/higher velocities, will the coarser mesh have a
marginal impact further out because of the larger turbulent eddies?
Rein: Good and very relevant articles, especially since we are
concerned with keeping the smoke layer to the higher parts of the
building. Thank you :)
Brad: Not sure about the 10MW, but 5 MW is a design case for a
sprinkler-controlled fire. About the grid sizes, I think dr_floyd is
right. It seems you need to have about 10-20 cells across the fire
diameter. The only way to "know" is to control the fire to empirical
data and see if it behave like I expect it to.
Have a nice remainder of the weekend!
Henrik
the surrounding or do I have to include a larger part of the plume?
When checking for temperatures and velocity both I'm really only
interested in comparing the maximum (centerline) to empirical data? Do
I miss important turbulence parameters by modeling this way?
If using more than one mesh, how can I make a statement about that the
coarser mesh is providing me with accurate enough results? With a
accurate enough fire/higher velocities, will the coarser mesh have a
marginal impact further out because of the larger turbulent eddies?
Rein: Good and very relevant articles, especially since we are
concerned with keeping the smoke layer to the higher parts of the
building. Thank you :)
Brad: Not sure about the 10MW, but 5 MW is a design case for a
sprinkler-controlled fire. About the grid sizes, I think dr_floyd is
right. It seems you need to have about 10-20 cells across the fire
diameter. The only way to "know" is to control the fire to empirical
data and see if it behave like I expect it to.
Have a nice remainder of the weekend!
Henrik
Kevin
Oct 11, 2010, 8:50:54 AM10/11/10
to FDS and Smokeview Discussions
One of our validation cases (See Validation Guide, VTT Test cases)
involves the VTT test building which is 19 m tall. I model this with a
relatively fine mesh for the plume (all the way to the ceiling) and
less fine meshes filling out the rest of the building. To test, I just
run the fine mesh with open boundaries. In my experience, if you match
centerline correlation data to your desired degree of accuracy, the
other measurements will be simulated as accurately. Matching
centerline decreases in temperature and velocity implies you have
resolved well-enough the entrainment of air into the plume.
involves the VTT test building which is 19 m tall. I model this with a
relatively fine mesh for the plume (all the way to the ceiling) and
less fine meshes filling out the rest of the building. To test, I just
run the fine mesh with open boundaries. In my experience, if you match
centerline correlation data to your desired degree of accuracy, the
other measurements will be simulated as accurately. Matching
centerline decreases in temperature and velocity implies you have
resolved well-enough the entrainment of air into the plume.
Henrik Nordenstedt
Oct 26, 2010, 3:18:21 AM10/26/10
to FDS and Smokeview Discussions
Thanks Kevin.
Some additional questions:
1. I ran into some trouble modeling 4 grids (3 x 0,125m and 1 x 0,25m)
of about 1,7M nodes each on 4 separate cores with 4GB ram each and new
processors. Is there a "limit" to how many cells are allowed within a
grid? Suggestions how to overcomce this problem? I have trouble
dividing the grids further because of the high building and I don't
want to 2 different grids, even when they are the same node size over
the fire. Correct?
2. I have used K.Overholts tool somewhat in trying to define
appropriate grid size which I use as a basis for a grid sensitivity
analysis. When building models with bounding walls etc it's convenient
to use, say 0,125/0,25/0,5 etc grids because it's easy to fit to the
walls. My question is this: What happens if I use say 0,183m grid in
one of my models?
3. How far away from the fire is it OK to place coarser grids?
Regards,
Henrik
Some additional questions:
1. I ran into some trouble modeling 4 grids (3 x 0,125m and 1 x 0,25m)
of about 1,7M nodes each on 4 separate cores with 4GB ram each and new
processors. Is there a "limit" to how many cells are allowed within a
grid? Suggestions how to overcomce this problem? I have trouble
dividing the grids further because of the high building and I don't
want to 2 different grids, even when they are the same node size over
the fire. Correct?
2. I have used K.Overholts tool somewhat in trying to define
appropriate grid size which I use as a basis for a grid sensitivity
analysis. When building models with bounding walls etc it's convenient
to use, say 0,125/0,25/0,5 etc grids because it's easy to fit to the
walls. My question is this: What happens if I use say 0,183m grid in
one of my models?
3. How far away from the fire is it OK to place coarser grids?
Regards,
Henrik
Henrik Nordenstedt
Oct 26, 2010, 5:58:58 AM10/26/10
to FDS and Smokeview Discussions
4. Even the coarser grids takes forever to run, the eq now running on
8 cores/4GB each. Each timestep is 0,03 which seems to be slowing down
the processes.
This is only the result dumps, right? Where can I change them? Which
are the main parameters to affects calculation speed (omitting grid
resolution), that I could try and change?
5. If I shut off the radiation solver does the model still deduct the
radiative fraction from the HRR? If not, how do I compensate?
6. Is it possible to run several meshes on one processor AND use MPI?
Thanks!
Henrik
8 cores/4GB each. Each timestep is 0,03 which seems to be slowing down
the processes.
This is only the result dumps, right? Where can I change them? Which
are the main parameters to affects calculation speed (omitting grid
resolution), that I could try and change?
5. If I shut off the radiation solver does the model still deduct the
radiative fraction from the HRR? If not, how do I compensate?
6. Is it possible to run several meshes on one processor AND use MPI?
Thanks!
Henrik
dr_jfloyd
Oct 26, 2010, 8:43:23 AM10/26/10
to FDS and Smokeview Discussions
1) There is a limit in the number of cells. It is function of
available memory and the details of your input file. If you are
getting allocation or other memory errors and your run doesn't
successfully start timestepping, then you probably have too many
cells.
2) FDS doesn't care whether or not you use nice numbers for grid cell
sizes
3) This is something you will need to determine for your specific
scenario
4) 0.03 s is not an unusual time step size. It is actually fairly
large. You are running a CFD model, not a zone model. Expect that
simulations of large buildings are likely to take days to weeks.
5) Read the User's Guide section 11.3 (Why would you shut off the
radiation solver? What unintended effects might this have on your
results and what impact will it have on life safety decisions being
made for the hospital?)
6) Read the User's Guide section 6.3.3
available memory and the details of your input file. If you are
getting allocation or other memory errors and your run doesn't
successfully start timestepping, then you probably have too many
cells.
2) FDS doesn't care whether or not you use nice numbers for grid cell
sizes
3) This is something you will need to determine for your specific
scenario
4) 0.03 s is not an unusual time step size. It is actually fairly
large. You are running a CFD model, not a zone model. Expect that
simulations of large buildings are likely to take days to weeks.
5) Read the User's Guide section 11.3 (Why would you shut off the
radiation solver? What unintended effects might this have on your
results and what impact will it have on life safety decisions being
made for the hospital?)
6) Read the User's Guide section 6.3.3
Kevin
Oct 26, 2010, 8:44:58 AM10/26/10
to FDS and Smokeview Discussions
> 1. I ran into some trouble modeling 4 grids (3 x 0,125m and 1 x 0,25m)
> of about 1,7M nodes each on 4 separate cores with 4GB ram each and new
> processors. Is there a "limit" to how many cells are allowed within a
> grid? Suggestions how to overcomce this problem? I have trouble
> dividing the grids further because of the high building and I don't
> want to 2 different grids, even when they are the same node size over
> the fire. Correct?
Depends on your machine. Do you have a 32 bit or 64 bit OS. If 32 bit,
> of about 1,7M nodes each on 4 separate cores with 4GB ram each and new
> processors. Is there a "limit" to how many cells are allowed within a
> grid? Suggestions how to overcomce this problem? I have trouble
> dividing the grids further because of the high building and I don't
> want to 2 different grids, even when they are the same node size over
> the fire. Correct?
the limit per grid is about 1.5 million cells. Are you sure you have 4
GB RAM for each core? Just run a small case and gradually increase the
number of cells, when the case no longer runs, you've hit the limit.
>
> 2. I have used K.Overholts tool somewhat in trying to define
> appropriate grid size which I use as a basis for a grid sensitivity
> analysis. When building models with bounding walls etc it's convenient
> to use, say 0,125/0,25/0,5 etc grids because it's easy to fit to the
> walls. My question is this: What happens if I use say 0,183m grid in
> one of my models?
>
> 3. How far away from the fire is it OK to place coarser grids?
>
define a metric of accuracy and then test your grid selection against
that.
Kevin
Oct 26, 2010, 8:47:07 AM10/26/10
to FDS and Smokeview Discussions
> 4. Even the coarser grids takes forever to run, the eq now running on
> 8 cores/4GB each. Each timestep is 0,03 which seems to be slowing down
> the processes.
>
> This is only the result dumps, right? Where can I change them? Which
> are the main parameters to affects calculation speed (omitting grid
> resolution), that I could try and change?
Nothing really. It's all the grid size. Run on more meshes.
> 8 cores/4GB each. Each timestep is 0,03 which seems to be slowing down
> the processes.
>
> This is only the result dumps, right? Where can I change them? Which
> are the main parameters to affects calculation speed (omitting grid
> resolution), that I could try and change?
>
> 5. If I shut off the radiation solver does the model still deduct the
> radiative fraction from the HRR? If not, how do I compensate?
>
> 6. Is it possible to run several meshes on one processor AND use MPI?
>
Henrik Nordenstedt
Oct 26, 2010, 11:34:39 AM10/26/10
to FDS and Smokeview Discussions
On Oct 26, 2:43 pm, dr_jfloyd <drjfl...@gmail.com> wrote:
> 2) FDS doesn't care whether or not you use nice numbers for grid cell
> sizes
What happens when a wall for example is partly included in a cell,
what is lost?
>
> 3) This is something you will need to determine for your specific
> scenario
error?
For grid sensitivity there is plume eq and other empirical data. What
can I compare coarser grids to?
>
> 5) Read the User's Guide section 11.3 (Why would you shut off the
> radiation solver? What unintended effects might this have on your
> results and what impact will it have on life safety decisions being
> made for the hospital?)
particularly interested in radiation it's better to refine the size of
the grid further to improve the quality of the results. I will read
the section.
>
> 6) Read the User's Guide section 6.3.3
dr_jfloyd
Oct 26, 2010, 11:43:43 AM10/26/10
to FDS and Smokeview Discussions
On Oct 26, 11:34 am, Henrik Nordenstedt <henkeb...@gmail.com> wrote:
> On Oct 26, 2:43 pm, dr_jfloyd <drjfl...@gmail.com> wrote:
>
> > 2) FDS doesn't care whether or not you use nice numbers for grid cell
> > sizes
>
> Does that mean I could model a 5 x 5 metre room using a 0,183m grid?
> What happens when a wall for example is partly included in a cell,
> what is lost?
>
cells. As to what happens this is discussed in the User's Guide. (you
could also just create some simple test cases to test this yourself
which is the best way to learn these things)
>
>
> > 3) This is something you will need to determine for your specific
> > scenario
>
> Yes. Is there a convenient way to do this or simply by trial and
> error?
> For grid sensitivity there is plume eq and other empirical data. What
> can I compare coarser grids to?
>
objective is, and many other things play a role in to what is or is
not an acceptable error for your specific intended use. If your
coarse grids are located in regions with low flows or flows that are
mostly unidirectional, then they probably won't have much impact on
the plume.
Henrik Nordenstedt
Oct 27, 2010, 4:20:26 AM10/27/10
to FDS and Smokeview Discussions
drfloyd and kevin, thanks for taking the time!
Greatly appreciated.
Have a nice remainder of the week!
Regards,
Henrik
Greatly appreciated.
Have a nice remainder of the week!
Regards,
Henrik
Henrik Nordenstedt
Nov 5, 2010, 10:28:59 AM11/5/10
to FDS and Smokeview Discussions
I have a validation question with regards to the reports from Rein and
validating a FDS model with empirical data.
As Rein describes (above), in the atria FDS seem to overpredict the
temperatures closer to the flame and slightly underpredict the
temperatures higher up. I get similar results when conducting a grid
sensitivity analysis and comparing to Heskestad plume temperatures
(thanks kevin). I used 0.0625, 0.125, and 0,25 m grids to for a
uniform mesh for a 2MW fire with 1m x 1m base and shaft extending 45m
up. The drew the conclusion that 0.0625 and 0.125 produce similar
results for this setup, hence I want to use the 0.125 grid for
modeling for time reasons.
I have a few questions about this sensitivity analysis:
1) I'm interested in good results "far" away from the fire where the
modeled results seem to converge with the empirical ones. What happens
if I model the same fire in a room only 4 metres high? Is the grid
sensitivity analysis no longer valid for this case? Would the model
give me good comparison with Mccaffrey plume model if I did it
correctly? I find things get complicated once I'm in the flame :)
2) I'm using exhaust vents with 10 m3/s capacity. Some of the vents
are located further away from the fire where the grids are coarser (up
to 0,5 metres). Since the exhaust vents is "action" territory, do I
need a finer grid here as well? If yes, do I need the finer grid from
the fire, all the way up to that vent?
3) The design fire is a 2MW fire. Because of the complexity of subject
furniture such as chairs of unknown quality etc I have used heptane
with a HRR of 2MW. Instead of a heptane fire with an area of
0,4347..m2 I used a 1m2 burner with HHR 2000kw/m2.
The "Reaction" line only used for species etc, correct? Do I give up
any accuracy of calculation with this approach with regards to
temperatures, smoke fill etc? (since there wont be a heptane pool
there anyway, but stacked chairs etc)
Warm regards,
Henrik
validating a FDS model with empirical data.
As Rein describes (above), in the atria FDS seem to overpredict the
temperatures closer to the flame and slightly underpredict the
temperatures higher up. I get similar results when conducting a grid
sensitivity analysis and comparing to Heskestad plume temperatures
(thanks kevin). I used 0.0625, 0.125, and 0,25 m grids to for a
uniform mesh for a 2MW fire with 1m x 1m base and shaft extending 45m
up. The drew the conclusion that 0.0625 and 0.125 produce similar
results for this setup, hence I want to use the 0.125 grid for
modeling for time reasons.
I have a few questions about this sensitivity analysis:
1) I'm interested in good results "far" away from the fire where the
modeled results seem to converge with the empirical ones. What happens
if I model the same fire in a room only 4 metres high? Is the grid
sensitivity analysis no longer valid for this case? Would the model
give me good comparison with Mccaffrey plume model if I did it
correctly? I find things get complicated once I'm in the flame :)
2) I'm using exhaust vents with 10 m3/s capacity. Some of the vents
are located further away from the fire where the grids are coarser (up
to 0,5 metres). Since the exhaust vents is "action" territory, do I
need a finer grid here as well? If yes, do I need the finer grid from
the fire, all the way up to that vent?
3) The design fire is a 2MW fire. Because of the complexity of subject
furniture such as chairs of unknown quality etc I have used heptane
with a HRR of 2MW. Instead of a heptane fire with an area of
0,4347..m2 I used a 1m2 burner with HHR 2000kw/m2.
The "Reaction" line only used for species etc, correct? Do I give up
any accuracy of calculation with this approach with regards to
temperatures, smoke fill etc? (since there wont be a heptane pool
there anyway, but stacked chairs etc)
Warm regards,
Henrik
Kevin
Nov 5, 2010, 4:08:07 PM11/5/10
to FDS and Smokeview Discussions
1. Plume correlations no longer apply when a smoke layer forms in the
compartment.
2. Going from a fine to coarse to fine mesh is not useful. Information
is lost on the coarse that cannot be recovered.
3. Do the calculations both ways and find out if there is a big
difference.
A grid resolution study is the answer to most of your questions.
> > > > Thank you!- Hide quoted text -
compartment.
2. Going from a fine to coarse to fine mesh is not useful. Information
is lost on the coarse that cannot be recovered.
3. Do the calculations both ways and find out if there is a big
difference.
A grid resolution study is the answer to most of your questions.
Henrik Nordenstedt
Nov 8, 2010, 5:19:04 AM11/8/10
to FDS and Smokeview Discussions
Yes, and how do I perform a grid sensitivity analysis for a fire in a
department store or similar setting with room heights of maybe 3-4
metres?
When placing a fire in a room I have previously gotten some funny
results of conductivity losses. I have just placed a burner vent on
the floor with surrounding surfaces inert. If the surrounding surfaces
are not inert does this lead to a large portion of heat being "led"
away from the flame by conduction? Do I have to build a box and place
a fire on top?
Regards,
Henrik
department store or similar setting with room heights of maybe 3-4
metres?
When placing a fire in a room I have previously gotten some funny
results of conductivity losses. I have just placed a burner vent on
the floor with surrounding surfaces inert. If the surrounding surfaces
are not inert does this lead to a large portion of heat being "led"
away from the flame by conduction? Do I have to build a box and place
a fire on top?
Regards,
Henrik
Kevin
Nov 8, 2010, 8:29:50 AM11/8/10
to FDS and Smokeview Discussions
I would not expect the surrounding surfaces to have a significant
impact, but I do not know exactly what you are doing. A grid
sensitivity analysis is the same for any simulation -- run the case
with smaller and smaller grid cells to determine if your results
change significantly. Monitor the temperature and heat flux of various
regions of the gas and solids.
> > > - Show quoted text -- Hide quoted text -
impact, but I do not know exactly what you are doing. A grid
sensitivity analysis is the same for any simulation -- run the case
with smaller and smaller grid cells to determine if your results
change significantly. Monitor the temperature and heat flux of various
regions of the gas and solids.
Henrik Nordenstedt
Nov 9, 2010, 11:28:25 AM11/9/10
to FDS and Smokeview Discussions
Node size, h 35 m 30 m 25 m 20 m 15 m 10 m
GRID 0,0625 28,97 31,70 34,73 41,36 50,15 75,61
GRID 0,125 28,86 31,87 36,10 43,39 55,93 102,36
GRID 0,25 29,68 35,12 43,30 60,85 95,32 152,16
Heskestad 28,31 30,75 34,61 41,34 54,96 90,94
ok, here is my grid sensitivity analysis. high up and far away from
the fire both 0,125 and 0,0625 seems to predict, in this case, flame
temperatures well enough. GRID 0,25 seems to overpredict the
temperatures.
Now, if you would look at the 10m column the temperature varies much
from grid to grid. I suspect as we get closer to the flame region that
the results would wary further.
Now I want to model a room 4 metres high. As rein pointed out above,
FDS seem to overpredict temperatures/velocities close to the fire and
only slightly underpredict them further away. Do I keep decreasing the
grid till I hit similar temperatures/velocities or how do I go about
setting up a grid sensitivity analysis?
Regards,
Henrik
GRID 0,0625 28,97 31,70 34,73 41,36 50,15 75,61
GRID 0,125 28,86 31,87 36,10 43,39 55,93 102,36
GRID 0,25 29,68 35,12 43,30 60,85 95,32 152,16
Heskestad 28,31 30,75 34,61 41,34 54,96 90,94
ok, here is my grid sensitivity analysis. high up and far away from
the fire both 0,125 and 0,0625 seems to predict, in this case, flame
temperatures well enough. GRID 0,25 seems to overpredict the
temperatures.
Now, if you would look at the 10m column the temperature varies much
from grid to grid. I suspect as we get closer to the flame region that
the results would wary further.
Now I want to model a room 4 metres high. As rein pointed out above,
FDS seem to overpredict temperatures/velocities close to the fire and
only slightly underpredict them further away. Do I keep decreasing the
grid till I hit similar temperatures/velocities or how do I go about
setting up a grid sensitivity analysis?
Regards,
Henrik
Kevin
Nov 9, 2010, 5:22:35 PM11/9/10
to FDS and Smokeview Discussions
First -- do not expect to "converge" exactly to the Heskestad profile.
An empirical correlation is an average of many experiments. It appears
that you are quite close in the upper plume.
Second, near the fire, I would expect that details of the combustion
model may affect the convergence. The objective of grid sensitivity is
just to determine when resolving the grid no longer matters. For your
room simulation, you just want to determine when the results no longer
change significantly.
An empirical correlation is an average of many experiments. It appears
that you are quite close in the upper plume.
Second, near the fire, I would expect that details of the combustion
model may affect the convergence. The objective of grid sensitivity is
just to determine when resolving the grid no longer matters. For your
room simulation, you just want to determine when the results no longer
change significantly.
Henrik Nordenstedt
Nov 16, 2010, 8:17:44 AM11/16/10
to FDS and Smokeview Discussions
Ah, Kevin
Missed your reply.
Yes. But does independent grid resolution mean that the model
acceptably predict reality?
To check for validity I want to compare the model with experiments.
How can I make a statement whether the model fire predicts what I want
it to, when I have smaller compartment heights and in some cases flame
contact with ceiling?
Cheers,
Henrik
Missed your reply.
Yes. But does independent grid resolution mean that the model
acceptably predict reality?
To check for validity I want to compare the model with experiments.
How can I make a statement whether the model fire predicts what I want
it to, when I have smaller compartment heights and in some cases flame
contact with ceiling?
Cheers,
Henrik
Kevin
Nov 16, 2010, 9:01:03 AM11/16/10
to FDS and Smokeview Discussions
No, CFD models typically "converge" to a unique solution as the grid
is refined (cells made smaller and smaller). But this solution may not
reflect reality. Grid sensitivity studies are a form of Verification
-- they test if the equations are being properly solved. Validation is
where you compare your converged solution to experiments. The appeal
of LES is that the solution "converges" towards reality because your
simulation changes from LES to DNS (Direct Numerical Solution). This
is generally true for isothermal flow calculations, like low speed air
flow over blunt bodies. In such cases, the flow is just a function of
the Reynolds number and your finer and finer calculations "capture"
the turbulent motion on smaller and smaller scales until you get to a
point where you are solving the actual flow equations without
resorting to a "model" of sub-grid scale turbulence.
For FDS, there are still issues related to the combustion routine
(like mixing times, etc) that prevent the fire from converging towards
reality. Much of the work on FDS 6 involves ensuring proper
convergence towards reality in the near field. However, in the plume,
away from the fire, the solution tends towards reality. At least that
has been our experience looking at hundreds of experimental datasets
of plumes and compartment fire experiments. These results are
documented in the FDS Validation Guide.
is refined (cells made smaller and smaller). But this solution may not
reflect reality. Grid sensitivity studies are a form of Verification
-- they test if the equations are being properly solved. Validation is
where you compare your converged solution to experiments. The appeal
of LES is that the solution "converges" towards reality because your
simulation changes from LES to DNS (Direct Numerical Solution). This
is generally true for isothermal flow calculations, like low speed air
flow over blunt bodies. In such cases, the flow is just a function of
the Reynolds number and your finer and finer calculations "capture"
the turbulent motion on smaller and smaller scales until you get to a
point where you are solving the actual flow equations without
resorting to a "model" of sub-grid scale turbulence.
For FDS, there are still issues related to the combustion routine
(like mixing times, etc) that prevent the fire from converging towards
reality. Much of the work on FDS 6 involves ensuring proper
convergence towards reality in the near field. However, in the plume,
away from the fire, the solution tends towards reality. At least that
has been our experience looking at hundreds of experimental datasets
of plumes and compartment fire experiments. These results are
documented in the FDS Validation Guide.
Rein
Nov 17, 2010, 3:08:41 PM11/17/10
to FDS and Smokeview Discussions
The computational domain in fire modelling can be divided into the
near-field and the far-field. The near-field is close to the flames,
whereas the far-field is far from them. In these general terms, your
question can be recasted as "How can I make a statement whether the
model fire predicts what I want it to in the near field?" I have asked
the same questions many times, and I still ask myself and my
colleagues.
The state of the art of CFD is known to do reasonable well in the far-
field where smoke movement dominates the thermofluid problem if the
HRR and the geometry are known. For the near-field the situation is
different; it is more complicated to model it accurately. Note that
flame combustion experts (conducting fundamental research more towards
physics than engineering) still struggle to model accurately 10 cm
laboratory ideal flames of simple gas fuel (and they use up to
millions of reactions). Thus, I think that an engineer cannot claim
accuracy in the near field, and a worst case scenario approach should
be adopted.
You might find of help the work that we have conducted comparing FDS
v4 (sorry, we run out of funds to repeat it with v5) with experimental
measurements in the near- and the far-fields from a realistic and non-
trivial fire load (IKEA residential scenario). The work has just been
accepted for publication in the journal 'Building and Environment',
but you can read most of the work in the Chapter 2 of this PhD thesis:
http://www.era.lib.ed.ac.uk/handle/1842/3418.
If anyone is interested in the full paper, send me an email and I will
be please to send you a copy when it is published in the next few
weeks.
Cheers
G.
near-field and the far-field. The near-field is close to the flames,
whereas the far-field is far from them. In these general terms, your
question can be recasted as "How can I make a statement whether the
model fire predicts what I want it to in the near field?" I have asked
the same questions many times, and I still ask myself and my
colleagues.
The state of the art of CFD is known to do reasonable well in the far-
field where smoke movement dominates the thermofluid problem if the
HRR and the geometry are known. For the near-field the situation is
different; it is more complicated to model it accurately. Note that
flame combustion experts (conducting fundamental research more towards
physics than engineering) still struggle to model accurately 10 cm
laboratory ideal flames of simple gas fuel (and they use up to
millions of reactions). Thus, I think that an engineer cannot claim
accuracy in the near field, and a worst case scenario approach should
be adopted.
You might find of help the work that we have conducted comparing FDS
v4 (sorry, we run out of funds to repeat it with v5) with experimental
measurements in the near- and the far-fields from a realistic and non-
trivial fire load (IKEA residential scenario). The work has just been
accepted for publication in the journal 'Building and Environment',
but you can read most of the work in the Chapter 2 of this PhD thesis:
http://www.era.lib.ed.ac.uk/handle/1842/3418.
If anyone is interested in the full paper, send me an email and I will
be please to send you a copy when it is published in the next few
weeks.
Cheers
G.
Reply all
Reply to author
Forward
0 new messages