Closed gforney closed 9 years ago
I do not see this behavior with the current release (6.2.0).
Original issue reported on code.google.com by drjfloyd
on 2015-04-20 12:12:26
Now with the current release (6.2.0) it does work. But at parallel running I still see
this behavior (fds_mpi)
Original issue reported on code.google.com by he.tauber
on 2015-04-21 05:50:41
Your domain is 30 m high. By default, FDS assumes that the atmosphere is stratified.
That is, the pressure and density decrease with altitude. This causes small temperature
fluctuations, as would be expected in an outdoor flow. If you don't want this effect,
set STRATIFICATION=.FALSE. on the MISC line.
Original issue reported on code.google.com by mcgratta
on 2015-04-21 18:23:33
I set STRATIFICATION=.FALSE. on the MISC line, but the behavior still remains. This
effect starts at a wind velocity of about 0.7 m/s. When I set the temperature range
from 19.99 to 20.01 the effect starts at about 127 s and builds up. As seen in the
attached file you get a strange pattern.
This effect I noticed in some other simulations when the mesh cells were on the order
of a millimeter. By setting DNS=.TRUE. the behaviour was not seen anymore.
Original issue reported on code.google.com by he.tauber
on 2015-04-22 08:13:19
In any numerical solution of partial differential equations there will be some numerical
error. The error in FDS is second order in space. An error on the order of 0.01 K given
a 0.5 m grid size does not seem unreasonable.
Original issue reported on code.google.com by drjfloyd
on 2015-04-22 12:29:49
The file Eddies_0637.jpg shows only the beginning of the strange behavior but as you
can see in figure Eddies2.jpg it grows to an error on the order of about 15°C (17 to
32°C).
Original issue reported on code.google.com by he.tauber
on 2015-04-23 05:56:27
Running with the release version of 6.2.0 on Windows I cannot reproduce this. For the
case in comment #4 I show temperatures = 20 C for the length of the simulation for
both serial (1 process 4 threads) and mpi (2 processes 4 threads/process).
Original issue reported on code.google.com by drjfloyd
on 2015-04-23 13:30:15
I get this problem only at parallel running (fds_mpi) in Windows as well in Linux. In
the attached file you see the recorded min and max temperatures from the total area.
Original issue reported on code.google.com by he.tauber
on 2015-04-24 06:01:51
After studying User's Guide about "Numerical Stability Parameters" I tested a few parameters.
By changing the following parameters the problem was solved:
CFL_VELOCITY_NORM=1 or
FLUX_LIMITER=0 or
FLUX_LIMITER=5
Could you please confirm (correct) that I have to consider anyone of these parameters
for the parallel running.
Original issue reported on code.google.com by he.tauber
on 2015-04-24 08:47:18
There is no fds_mpi for FDS 6.2.0, there is only one executable for FDS now. If you
are running fds_mpi, then you are using 6.1.2 or earlier. I cannot replicate your
issue with FDS 6.2.0.
Original issue reported on code.google.com by drjfloyd
on 2015-04-24 11:16:59
If 6.2.0 was properly installed, fds_mpi.exe will be the same as fds.exe. It is just
a pointer. We did this to prevent old versions of fds_mpi from running. In any case,
type
fds
or
fds_mpi
to see what version you have.
Original issue reported on code.google.com by mcgratta
on 2015-04-24 12:32:08
I think one of the things you are experience is that you haven't changed the default
lapse rate from 0 C/m. In typical usage, how we base our defaults, where one is dealing
with flows in buildings, there isn't a temperature lapse as heating and cooling systems
prevent it. In atmospheric flows there is a drop in temperature with height. The
attached images are your eddies2 case with the default LAPSE_RATE=0 (eddies2) and setting
an adiabatic lapse (eddies3) [LAPSE_RATE=-0.0098]. Note with the adiabatic lapse that
you don't see odd temperature behavior. With no lapse, when higher density air at
the bottom is forced up into a lower pressure region it expands and cools (and vice
versa). This causes what you saw in comment 2. With an adiabatic lapse, the cooling
by expansion is match by the drop in temperature with height and the temperature profile
remains constant.
Original issue reported on code.google.com by drjfloyd
on 2015-04-24 14:13:11
Now I see the problem. Last week I updated to the latest release. When I type fds I
get the version 6.2.0 and when I type fds_mpi I get the version 6.1.2. That means that
I always did the parallel run with the back version (fds_mpi). Now, when I do a parallel
run by typing: "mpiexec -n 2 fds Eddies.fds" I get the right result according to your
figures. If I got you right I do not have to use the fds_mpi file anymore. Am I right?
Original issue reported on code.google.com by he.tauber
on 2015-04-27 13:59:58
Yes. I suggest that you uninstall FDS, and then reinstall 6.2.0. This will remove old
executables. Or just type the name of the executable before running a job to ensure
you have the right version.
Original issue reported on code.google.com by mcgratta
on 2015-04-27 14:02:51
I uninstalled FDS and reinstalled 6.2.0 as you suggested. As far as I can see the wrong
version was the only reason for the behaviour. Finally I thank you for your help to
find out the answer.
Original issue reported on code.google.com by he.tauber
on 2015-05-04 04:35:59
Thanks for letting us know.
Original issue reported on code.google.com by mcgratta
on 2015-05-04 12:04:06
Original issue reported on code.google.com by
he.tauber
on 2015-04-20 10:32:38