FDS has overpredicted a heat loss of 2 times for fluid travelling along a long duct or a tunnel

[GoogleCodeExporter]

Please complete the following lines...

Application Version: 5.2.5
SVN Revision Number: 2828
Compile Date:Dec 2008
Operating System: Win-xp

Describe details of the issue below:

I find FDS has overpredicted a temperature drop by 2 times, compared with 
the results obtained with empirical model. As I am reading the users 
technical guide (Section 4.4.2 Convective Heat Transfer to Solid Objects), 
I find formula (4.115) that FDS used has two issues:

(1) Hydraulic diameter assumption of 1.0 is not right. Why not use the 
hydraulic diameter based on the cross section dimension?
(2) A too large coefficient of 0.037 in front of Re is used, and no 
reference is given. According to my reference (Incropera & Dewitt, 
Introduction to Heat Transfer, 3rd Edition, Eq 8-60), this coefficient 
should be 0.023.

If these two issued are solved, then FDS can predict very good heat losses 
for fluid travelling along a long duct or a tunnel.

Original issue reported on code.google.com by yunlo...@gmail.com on 20 Feb 2009 at 12:45

[GoogleCodeExporter]

Kevin,

I just did a hand calculation and confirmed it is because formula 4.115 that 
cause 
twice of the heat loss into the wall.

My duct have a dimension of 1800m long x 4m wide x 3m high, which give a 
hydraulic 
diameter of 3.4m. If I take into account of the difference of the coefficient 
in 
front of the Re, which should be 0.023, not 0.037, then I can easily reduce 
heat 
transfer coefficient to the wall by 50% approximately. 

0.023/0.037 x (3.4**0.8/3.4)=1/2.05

This is why I recorded twice of the temperature drop using FDS when compared to 
the 
empirical model (http://en.wikipedia.org/wiki/Heat_transfer_coefficient, as 
well as 
Incropera & Dewitt, Introduction to Heat Transfer, 3rd Edition, Eq 8-60).

I am sure the heat transfer into the solid wall will be very close to the 
results 
from the empirical model. Of course this is for the longitudinal flow along a 
duct.

If you are reluctant to change this for FDS which is designed not only for the 
longitudinal tunnel/long duct, then I proposed release another version of FDS 
specially for this? this is my opinion.

Jason

Original comment by yunlo...@gmail.com on 20 Feb 2009 at 3:11

[GoogleCodeExporter]

(1) How is FDS going to know what the characteristic length of a flow geometry 
is? 
It can't possibly know that.  The correlation is of the form Re_L^0.8 / L, so 
it is
not strongly dependent on L.  
(2) correlation in FDS is for heat transfer to a flat plate (and for the 
typical user
that is the appropriate correlation) not heat transfer to a duct.  

Original comment by drjfloyd on 20 Feb 2009 at 3:12

• Changed state: WontFix
• Added labels: Application-FDS

[GoogleCodeExporter]

drjfloyd,

(1) the characteristic length of a flow geometry is based on the hydraulic 
diameter, 
which is: 4x cross-Section-Area/Perimeter. This is no technical difficulties to 
do 
this.
(2) Most fo the heat transfer text book recommended this:
Nu=0.023xRe^0.8xPr^0.3
Why FDS use: Nu=0.037xRe^0.8xPr^0.3?

I can not understand, and no reference is given in the users guide.
At least another special version of FDS should be released specially for the 
longitudinal flow and heat transfer cases.

Jason

Original comment by yunlo...@gmail.com on 20 Feb 2009 at 5:12

[GoogleCodeExporter]

drjfloyd & Kevin,

I also deteced this over-predict of heat loss through the wall when I was 
comparing 
with a experiment. FDS modelling get a lower temperature when compared to the 
experiment in a smoke door leakage test at Warrington. I believe this is 
extactly 
the reason.
Would you please let me know in which subroutine this fluid-solid heat transfer 
is 
implemented? as I am very keen to valid this if I can compile the code myself.
Do I need both the FORTRAN and C compiler to compile to a Win-Xp executable 
file? 
Thanks.

Jason

Original comment by yunlo...@gmail.com on 20 Feb 2009 at 5:22

[GoogleCodeExporter]

No technical difficulties?  OK write an algorithm that can examine every 
possible
geometry that a user can input and in the context of the flow field and 
temperature
gradient pick the appropriate length scale and form of the heat transfer 
correlation.
 FDS doesn't know that you have input a duct.  It only knows that there is a set of
blockages.  It is very easy for us as people to look at a geometry and say "OH, 
thats
a duct with a diameter of x, I need to use a duct heat transfer correlation" but
getting a computer to do that is not a trivial matter.

We will add a reference to the guide.  Wall heat transfer is done in wall.f90. 
Compilation instructions can be found on the Wiki page (tab at the top of the 
issue
tracker).

Not familiar with that dataset, but what is the total uncertainty in the wall
temperature?  This would include uncertainties in HRR, wall properties, leakage,
compartment dimensions, the correlation for h used in FDS, etc.  Unless you
prediction is in error by more than the total uncertainty you can't conclude the
model is in error.

Original comment by drjfloyd on 20 Feb 2009 at 12:49

[GoogleCodeExporter]

All -- seems like this discussion has spread across multiple threads. In any 
case, I 
think everybody has a point, and this discussion is worth having. Jason (not 
Floyd) -
- you're working on one particular application. We have to work on everybody's 
application and it's hard to configure FDS for every possible geometry. The 
trouble 
with most mass and heat transfer correlations is that they are geometry 
specific, 
especially in regards to a characteristic length scale. Usually, we just set 
that 
length to 1 m for lack of anything better. I think the solution to this problem 
is 
to make it clearer in the User's Guide how one might change the coefficients of 
the 
correlations if that is indeed the only issue. Changing the mathematical form 
may be 
another issue entirely and we'll have to approach on a case by case.

Keep in mind that we do not have strong opinions about the exact form of any 
particular correlation. Our selection of one over another usually comes down to 
how 
much information the correlation requires from the resolvable flow field. 

Original comment by mcgra...@gmail.com on 20 Feb 2009 at 1:32

[GoogleCodeExporter]

drjfloyd:

I can adjust the C_forced to account for the coefficient of 0.023 and the 
length 
scale. Thanks 

Kevin,

There is no need to change the form of the formula, only make the length scale 
possible to input is OK, as C_forced input is aready there. Thanks for the 
solution.

Jason

Original comment by yunlo...@gmail.com on 20 Feb 2009 at 7:46