search

Yinan-Scott-Shi / fds-smv

Automatically exported from code.google.com/p/fds-smv
0 stars 0 forks source link

FDS 5.2.4 vs FDS 5.3.0 #684

Closed GoogleCodeExporter closed 9 years ago

GoogleCodeExporter commented 9 years ago 
Application Version: FDS 5.2.4, FDS 5.3.0
SVN Revision Number:
Compile Date:
Operating System: Windows XP

Describe details of the issue below:
I have a question regarding the differences between FDS 5.2 and 5.3.
I did a few simulations, the first simulation with FDS 5.2 and the last 
ones with 5.3. The geometry is complex and I think Randys ongoing work on 
the pressure solver might have some influences but both simulations (FDS 
5.2 and 5.3) were done in serial Mode, where the pressure solver should 
have no significant influence.
As you can see the results differ very much (which is o.k. because FDS is 
in a continuous state of change).
Long story short:
The FDS release notes for version 5.3.0 do not show any changes that might 
explain these differences between 5.3 and 5.2 in my point of view.

You find 2 comparable screenshots attached.

Boris

Original issue reported on code.google.com by I...@F-Sim.de on 25 Mar 2009 at 2:21

Attachments:

GoogleCodeExporter commented 9 years ago 
We'll take a look at it.

Original comment by mcgra...@gmail.com on 25 Mar 2009 at 2:40

GoogleCodeExporter commented 9 years ago 
I ran your two input files exactly as you posted them. I ran with the parallel 
versions of 5.2.4 and 5.3.0 on our linux cluster. Attached are two snapshots of 
the 
calculations at z=1.8 at 60.8 s. I do not see a dramatic difference in the flow 
behavior. Because of changes made to various parts of the algorithm, you cannot 
expect the same exact flow. In fact, the temperature slices that you posted are 
vertically oriented through the door. The plume does not align with this 
vertical 
plane, and therefore you are seeing the result of the "butterfly effect." The 
hot 
gases from the smallest compartment are swirling about in the large 
compartment. I 
would not expect any one snapshot through any one plane to be exactly the same 
from 
version to version. It would be more useful to look at time histories (DEVC) of 
TEMPERATURE at more than one location and compare 5.2 and 5.3. Results that are 
suitably time-averaged should not change significantly.

Note that the pressure solver is important in both serial and parallel mode. It 
is 
the use of multiple meshes, not the use of MPI, that makes the pressure solving 
more 
challenging. In fact, there should be no significant difference between a 
multi-mesh 
calculation run in serial or parellel. 

Also, you might want to try running without PRESSURE_CORRECTION and 
CHECK_VOLUME_FLOW. We have found a few instances where this feature has caused 
unphysical flow behavior. It does not appear to be the case in your example, 
but 
just for your information, the model seems to be working well with or without 
the 
correction.

Let me know if you see significantly different results in the time history of 
the 
temperature at different points in the domain.

Original comment by mcgra...@gmail.com on 28 Mar 2009 at 2:03

Attachments:

GoogleCodeExporter commented 9 years ago 
Kevin,

thanks for your answer. Your slices seem to show much better outcomes.
I will do some new simulations at monday and post my findings.

Boris

Original comment by I...@F-Sim.de on 28 Mar 2009 at 5:58

GoogleCodeExporter commented 9 years ago 
Kevin,

in advance (I haven't done new simulations yet) I attached some screenshots 
showing 
PLOT3d ISO surfaces.
Both after 120 s, both showing 29°C.

FDS 5.3 shows a layer height between 0 an max. 2 m nearly everywhere. The 
simulation 
done with FDS 5.2 shows wide areas with a height >> 2 m.

I will work out these results with DEVCs to make the differences clearer... 

Boris

Original comment by I...@F-Sim.de on 31 Mar 2009 at 9:56

Attachments:

GoogleCodeExporter commented 9 years ago 
These are not good methods to assess whether or not there is a noticeable 
change in 
FDS. Let's look at time histories of temperature for the entire duration, at 
various 
points. 29 C is just above ambient temperature. I would expect to see 
differences in 
this contour.

Original comment by mcgra...@gmail.com on 31 Mar 2009 at 12:53

GoogleCodeExporter commented 9 years ago 
Kevin,

I understand your retention and will take a look at some time histories.

Boris

Original comment by I...@F-Sim.de on 31 Mar 2009 at 1:15

GoogleCodeExporter commented 9 years ago 
Kevin,

I set some temperature DEVCs in the middle of MESH 1.
The interpretation of the plots time vs. temperature in all 
(test_FDS_52_devc_all.pdf and test_FDS_53_devc_all.pdf) is not very useful (in 
my 
opinion) because there is no trend for all heights visible. 

For a better visualisation of what might in my opinion be a problem, I focused 
on 
the DEVCs between 0 and 2 m height and just showed 200 s < time < 300 s 
(test_FDS_52_devc_200-300.pdf and test_FDS_52_devc_200-300.pdf).

The two last plots show that while FDS’ 5.2 outcomes show something like a 
lower 
layer with no significant rise of temperature, FDS 5.3 has no more lower layer 
without a rise of temperature.
The differences may be small, but the VENT standing for a fire is also a very 
small 
one. Relating to a simulation of fire, these could for example mean that the 
escape 
routes are free of smoke (FDS 5.2) or that there is no (cool and smokefree) 
lower 
layer for escape routes.

Boris

Original comment by I...@F-Sim.de on 6 Apr 2009 at 9:33

Attachments:

GoogleCodeExporter commented 9 years ago 
These charts are useful. I need some time to look them over. Thanks.

Original comment by mcgra...@gmail.com on 6 Apr 2009 at 12:16

GoogleCodeExporter commented 9 years ago 
Boris,

Sorry I did not dig into this at the start.  It seemed like it was not clear 
there
was a problem.

These are long runs and it may be hard to diagnose quickly.  But now that I have
taken a quick look at your input files, the first thing I would do is to get 
rid of
PRESSURE_CORRECTION=T and CHECK_VOLUME_FLOW=T.  In my opinion this "feature" 
has been
oversold as the fix for the multimesh version of FDS.  We have found it may 
actually
cause more problems than it fixes.  The new-found stability of the multimesh 
runs is
due to other new (default) features of the algorithm like continuous definition 
of
the Dirichlet pressure boundary condition, the averaging of the normal 
components of
velocity at a mesh interface at the beginning of a time substep, and 
adjustments to
the projection algorithm.  Perfect volume conservation is not guaranteed, but 
neither
is perfect velocity alignment guaranteed using PRESSURE_CORRECTION (thus mass
conservation is not guaranteed to machine precision) -- this does not mean we
encounter gross mass errors.  In fact, Susan Kilian has been shown me several
multimesh runs using her global Poisson solver and the solutions were
indistinguishable from default FDS multimesh runs.

So, please run your cases with the default (why NOISE=F?) scheme and let's see 
how
things look. (Also, no need for ISOTHERMAL=F and SYNCH=T).  In the end this may 
not
be the only problem, but it could help us figure out where to look.

Thanks,
Randy

Original comment by randy.mc...@gmail.com on 6 Apr 2009 at 2:40

GoogleCodeExporter commented 9 years ago 
Randy,

I redid the simulations with

&MISC  PRESSURE_CORRECTION=.FALSE., NOISE=.FALSE., RADIATION=.FALSE./
&TIME TWFIN=300.  /Simulationszeit
&DUMP CHECK_VOLUME_FLOW=.FALSE., NFRAMES=1800, DT_DEVC=5., DT_PL3D=30./

The plots changed a little bit, but there is no general advancement.

I am attaching the new plots.
Boris

Original comment by I...@F-Sim.de on 7 Apr 2009 at 7:18

Attachments:

GoogleCodeExporter commented 9 years ago 
I have narrowed my search to two changes made between the release of FDS 5.2.4 
and 
5.3.0. The first is SVN 2912, where we changed the velocity "slip" factor  
(SLIP_FACTOR) from its default value of 0.5 to a grid dependent value of 
1-0.01/dx, 
which in your case (dx=0.2) is 0.95, close to a free slip condition. The second 
change (SVN 3108) was a small modification of the way the velocity boundary 
condition is handled at corners, which in your case would be the door way 
between 
the small compartment and the large one.

I am currently running a case with 5.3.0 using the old SLIP_FACTOR of 0.5. 
Hopefully 
this will narrow down the possibilities even further.

Two notes -- I looked at the temperature slices of your two cases at z=6 m. The 
5.3.0 version exhibits more "swirling" in the clockwise direction, I suspect 
because 
there is less drag exerted on the air by the walls. More simply, the air more 
easily 
slips past the walls. It is difficult to say without experimental validation 
data 
that the 5.3.0 case is more accurate than the 5.2.4, but in my experience, more 
flow 
dynamics is better than less. In your case, I would not expect the walls to 
impede 
the flow as much as they appear to do in the 5.2.4 case. However, this is just 
speculation.

Next, Randy McDermott has recently implemented in FDS a better treatment of the 
velocity boundary condition at the wall (Werner and Wengle model) and he has 
tested 
it for tunnel flows. This will replace the old "slip factor" approach and it 
will 
eliminate, hopefully, these kinds of discrepancies. This model must be run 
against 
the full validation test suite, and should be the default in 5.4.0. As with the 
change you have noticed, we do not change functionality without a change in 
minor 
release number.

Original comment by mcgra...@gmail.com on 7 Apr 2009 at 12:37

GoogleCodeExporter commented 9 years ago 
Attached are three snapshots from three different cases. test_FDS_52 and 
test_FDS_53 
are the original cases run with 5.2.4 and 5.3.0 (parallel). test_FDS_53_SF=p5 
is the 
same as test_FDS_53, except it uses the old default SLIP_FACTOR=0.5. Each image 
is 
time-averaged over 30 s. The results indicate to me that the SLIP_FACTOR 
explains 
only part of the difference between 5.2 and 5.3. I believe that the other 
change 
that I mentioned above explains the rest.

Original comment by mcgra...@gmail.com on 7 Apr 2009 at 2:33

Attachments:

GoogleCodeExporter commented 9 years ago 
I will do your simulations over night and have a look tomorrow morning.
Thanks for the help.
Boris

Original comment by I...@F-Sim.de on 7 Apr 2009 at 2:42

GoogleCodeExporter commented 9 years ago 
Here is the input file I used.

Original comment by mcgra...@gmail.com on 7 Apr 2009 at 2:50

Attachments:

GoogleCodeExporter commented 9 years ago 
Looks better. Somewhere between 5.2 and 5.3.
I am curios what the Werner and Wengle model will bring.

Thanks for the explanations on this issue.
Boris

Original comment by I...@F-Sim.de on 8 Apr 2009 at 3:14

Attachments:

GoogleCodeExporter commented 9 years ago 
Boris,

Set WERNER_WENGLE_WALL_MODEL=T on MISC with the latest code.

Cheers,
Randy

Original comment by randy.mc...@gmail.com on 8 Apr 2009 at 3:24

GoogleCodeExporter commented 9 years ago 
Boris

Randy is right. The W-W model is already in FDS. We will release 5.3.1 today or 
tomorrow, but the W-W model will still be an option that you must specify. 
After we 
run the entire validation suite, we will release 5.4.0 with the W-W model as 
the 
default. However, it would be good if you run your case with 5.3.1 and let us 
know 
if there is something dramatically different. The benefit of the W-W model is 
two-
fold. First, it is physically valid, documented, and widely accepted in the CFD 
community. Second, it no longer requires the user to specify an ad hoc "slip 
factor." The W-W model will have an optional roughness length, but this is a 
well-
recognized parameter for which values can be found in the literature for 
various 
materials.

I'll mark the case as "Verified," but let us know if there are new developments.

Original comment by mcgra...@gmail.com on 8 Apr 2009 at 3:49

GoogleCodeExporter commented 9 years ago 
I started a test case with W_W=.TRUE. I will post the outcomes next week 
(Easter 
holidays).

Boris

Original comment by I...@F-Sim.de on 9 Apr 2009 at 4:22

GoogleCodeExporter commented 9 years ago 
I am attaching the actual plot with w_w=.True. (FDS 5.3.1).
The values are somewhere between FDS 5.2.4 and FDS 5.3.0.

Boris

Original comment by I...@F-Sim.de on 14 Apr 2009 at 6:57

Attachments:

GoogleCodeExporter commented 9 years ago 
Thanks for checking this for us. One issue to keep in mind when talking about 
velocity boundary conditions is that in real buildings, the walls are not 
perfectly 
smooth like steel pipes. There is "roughness". We are going to try to capture 
this 
effect with the W-W model.

Original comment by mcgra...@gmail.com on 14 Apr 2009 at 12:11