Académique Documents
Professionnel Documents
Culture Documents
0306-2619(95)
ABSTRACT
The available data from A S H R A E and CIBSE on pressure losses in
H V A C duct fittings are of unsatisfactory accuracy and insufficient
quantity, which may lead to inaccurate sizing and poor energy efficiencies
in air supply systems. This situation is receiving increasing international
attention and much effort is being spent in its rectification, especially
using computation fluid dynamics (CFD). This paper presents an
investigation of the accuracy of the CFD approach. It was found that
the combination of the k - e model and the higher order QUICK scheme
produces the highest accuracy (with a relative error of 10%). Grid
dependency tests showed that a relatively low grid density in the straight
sections upstream~downstream of the duct fitting is sufficient but a higher
density in the duct axis direction is required in the section containing
the duct fitting. In addition, a computational domain set-up with short
upstream~downstream sections is adequate. The examination of CFD
accuracy and its maximisation should be extended to other fittings. This
paper illustrates the important factors involved and the procedures to
follow.
INTRODUCTION
Installation of heating, ventilation and air conditioning (HVAC) ducts in
modern buildings is widespread and an accurate knowledge of pressure
loss in the ducts is crucial for correct plant sizing and energy efficiency.
The pressure loss in ductwork supplying air to various zones can be
calculated using computer models which incorporate pressure loss k
factors for duct fittings, based on data given in the CIBSE Guide
233
234
L. Shao, S. B. Riffat
factor is given by
k = AP/(O.5p Vz)
(1)
where
AP -- pressure loss across a duct fitting
p = density of air
V -- average air velocity in the duct
The existing guides do not include many of the duct fittings used in
HVAC systems. This compels designers to make intelligent guesses for
some of the k values used in their calculations. Despite continuous efforts,
notably by the American Society of Heating, Refrigeration and AirConditioning Engineers and other workers 3 to expand and to update the
k factor databases, there are still many duct fittings for which performance
data have not been established. The situation is not likely to change if
experimental testing remains the sole source of data, as these tests are
lengthy and incur high costs.
Computational fluid dynamics offers an alternative for improving
k factor databases. 4 This method has the advantage of not requiring
measurement equipment or assembly of experimental duct systems and is
more efficient for the determination of k factors under a large spectrum
of flow conditions. The duct flows in HVAC systems are relatively simple
in terms of turbulence structures and chemical/physical processes and can
be accurately modelled by the k - e and Reynolds stress models commonly
employed in current CFD software. Furthermore, the numerical simplicity
associated with modelling these components, which have relatively
regular shapes and simple boundary conditions, also assists the accurate
application of CFD.
In addition to providing a fast and cost effective method for obtaining
new k factors, CFD can also predict the performance of a duct during
design stages (i.e. prior to its manufacture) to assist design optimisation.
Furthermore, CFD can be used in conjunction with experiments to
improve the measurement accuracy of k factors. Significant inaccuracies
exist in experimentally obtained data and there are differences of up to
several hundred per cent between k factors quoted in the CIBSE,
ASHRAE and BSRIA guides. Previous research has shown that this is
due to the lack of knowledge of flow patterns within the duct fittings. As a
result, pressure sensors may be located within a disturbance section, causing
large measurement errors. The detailed flow field information provided
by CFD would be valuable for guiding the correct set-up of tests. 5
Before CFD can be considered viable for predicting k factors, its
accuracy and consistency must be determined. This paper examines the
235
Elbow1
Ou~rw']l-- I
Innerwall~ ]
[
]
Elbow2~ J
Fig. I.
D E S C R I P T I O N OF C O M P U T A T I O N
The computations were performed using the commercial flow simulation
software F L U E N T . The Reynolds stress terms in the averaged Navier
Stokes equations were computed using the standard two-equation k - e
model (or RSM); in the region of low Reynolds number close to the
walls, wall functions were used instead. Although the k e model is
robust, efficient and very widely used, it is known that in highly swirling
flows or in flows where significant stream curvature exists, this model
becomes inadequate. In such cases, which could include flow in a number
of duct fittings, the RSM generally offers greater accuracy by modelling
the Reynolds stresses directly. The relative advantages of the two models
in simulating the double elbows will be discussed in the following section.
The assumption was made that during the computation, the air velocity
distribution at the entrance of a duct is uniform and the flow direction is
normal to the inlet cross-section. The value of flow speed is unimportant as
it does not affect the k factor; it is assumed to be 6 m/s, i.e. representative
236
L. Shao, S. B. Riffat
237
The double elbow is symmetrical with respect to the plane defined by its
curved axis. Figure 2 shows an example of a grid set-up in this central
symmetrical plane, Grid lines are either parallel or perpendicular to the
duct axis. The pressure distribution along each of the grid lines parallel to
the axis for a double elbow with zero separation (between the two elbows)
is shown in Fig. 3. The horizontal axis of the figure is the cell number in
the I direction (i.e. the duct axis direction), which is used in preference to
'distance from duct entrance' because the number of cells in the elbow
section is identical for all the parallel grid lines and this enables direct
comparison between the pressures on these grid lines. The 'distance from
entrance' for them are not identical. In addition the length of the elbow
is around 0.3 m compared to more than 7 m for the complete duct,
which makes it difficult to display the pressure in that region with clarity
if the horizontal axis of the figure is based on distance. The horizontal
axis based on cell number will be used repeatedly in the following. As
shown in Fig. 3, the pressure distributions in the straight section (i,e. of
59 cells) upstream of the elbow are identical for all the parallel grid lines
and they coincide to form a single line. The pressure distributions diverge
dramatically within the double elbow (20 cells), with relative differences
between them of up to several hundred per cent. The top and bottom
curves correspond to the outer and inner curved walls of the double
238
L. Shao, S. B. Riffat
l0
10
20
30
40
50-
,~-~/~/~.
'~'
80
90
I00
P
-10
~_
"-...~'~
-20
-30
-40
-50
-60
Fig. 3.
elbow, respectively, and the other curves correspond to the grid lines in
between. As the flow leaves the elbow section, the pressure distributions
converge, again into a single line. An implication o f the result is that in
pressure loss measurements, the pressure sensors can be located anywhere
on the duct wall to obtain the same accurate reading, so long as they are
away from the elbow section. This is confirmed by results from another
computation in which two elbows separated by a straight section of 1.2 m
were examined. Figure 4 shows the pressure distribution in the double
10 ~
20
40
r.ff~80
100
120
-10
~- -20
,,~ -30
ft.
-40
-50
-60
Cell number in 1 direction
Fig. 4.
Pressure distribution along grid lines parallel to the duct axis, for a double elbow
with 1-2 m separation.
20
40
+
60
80
I
:.)39
100
120
J
-5
-10
-15
-20
-25
u~
.~ -30
o.,
-35
-40
---q:r-- RSM
-50
-60
C e l l n u m b e r in l d i r e c t i o n
Fig.
5.
elbow, with diverging regions for the two elbow sections and converged
regions for the straight sections. The top, middle and bottom curves
corresponds to the outer wall, duct axis and inner wall, respectively.
Effect of turbulence models
240
L. Shao, S. B. Riffat
-5
-10
-15
v
-20
-25
-30
k--epsilon
[] RSM
-35
-40
0 RSM h i g h e r order
k - e p s i l o n h i g h e r order
-45
-50
I
20
I
40
I
60
I
80
J
100
C e l l n u m b e r in 1 d i r e c t i o n
-10
'~
-15
~"
20
10
30
40
50
60
70
80
90
~--i--
-25
20
241
Single j & k
Double j & k
3O
two grids are identical, the original grid density is sufficient. Differences
between the solutions indicate that the original grid density is insufficient.
Grid dependency tests are not always carried out, however, due to the
computing resources required; each test would require a new grid eight times
larger with a dramatic increase in CPU time and memory requirement.
In this study, an alternative approach was adopted: the doubling of grid
density was not implemented simultaneously in all three dimensions and
not implemented simultaneously in all sections of the computational
domain. This arrangement enables the grid dependency tests to be carried
out without dramatically increasing demands on computing resources.
Figure 7 shows the effect of doubling the J and K grid densities. The
definition of I, J and K is given in Fig. 8, where X, Y and Z indicate the
ZOO
Fig. 8.
Definitions of I, J and K.
L. Shao, S. B. Riffat
242
0.1
0E+00
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
-5.00E+00 I
-I.00E+01
-I.50E+01
J
-2.00E+01
Double "i"
wmmeWmuim
-2.50E+01
3.00E+01
N o r m a l i s e d cell n u m b e r
Fig. 9.
direction of the duct axis, height and width./, J and K grid densities are the
densities of grid (grid planes, to be exact) in the X, Y and Z directions,
respectively. The dark curve shows the pressure distribution along the
axis of the duct and the light curve shows the distribution along the same
axis when the J and K grid densities are doubled. The double elbows
in this case have zero separation between them and correspond to cell
numbers 61-80. Cell numbers 1-60 and 81-90 correspond to the straight
sections before and after the elbows. As can be seen, the two curves
virtually coincide, so indicating that the original grid densities in the J
and K directions are adequate.
Grid dependency tests in the I direction yielded very different results.
Figure 9 shows the effect of doubling the I grid density uniformly along
the full length of the duct. The dark curve shows the pressure distribution along the axis of the duct and the light curve shows the distribution
along the same axis when the I grid density is doubled. The horizontal
axis is the normalised I cell number. This is necessary to enable direct
comparison of results from two cases with a different number of cells in
the I direction. The normalised cell number is defined as the ratio of the
I cell number to the total number of cells in the I direction. Similar arrangements of the horizontal axis are used repeatedly in the following,
and in each case the use of the normalised I cell number ensures that a
particular position in the duct corresponds to a unique value of the
normalised cell number, regardless of the grid distributions of individual
cases, so that direct comparisons can be made of results based on different
0.1
0.2
0.3
0.4
0.5
0.6
0.7
243
0.8
0.9
1.0
-10
am
r~
~,
15
e~
20
A 2u2e2d
luleld
[] 2u2e4d
-25
nnlmuwlmw!
-30
Normalised cell n u m b e r
Fig. 10. P r e s s u r e d i s t r i b u t i o n s , c o r r e s p o n d i n g to t w o d i f f e r e n t 1 grid d e n s i t i e s in t h e
s t r a i g h t s e c t i o n d o w n s t r e a m o f t h e b e n d , a l o n g t h e d u c t axis for a d o u b l e e l b o w with zero
separation.
I grid densities to determine their effects. The double elbows in this case
are separated by a straight section of 1-2 m length and are signified by
the two humps in both curves. The two curves virtually coincide in the
straight section upstream of the elbows, so indicating that the grid density
there is adequate. However, differences between the two curves are quite
significant in the elbow sections, thereby indicating insufficient grid density
in that region.
The adequacy of the grid density in the straight section downstream of
the elbows was examined and the results are shown in Fig. 10. The double
elbows in this case have zero separation between them. The dark curve
shows the pressure distribution along the axis of the duct based on the
original grid, and the curve marked by the triangles correspond to a
grid with doubled density. The former is annotated in the graph where
u, e and d denote upstream straight section, elbow section (including two
elbows and the section separating them) and downstream section, respectively. The number before u, e and d indicates the ratio of grid density in
the respective section to that of the original set-up. Therefore, the second
curve with double grid density in all sections is marked by 2u2e2d and the
third curve corresponds to a case that is identical to the second (2u2e2d)
except that the grid density in the downstream section is doubled again.
This method of describing the grid density will be adopted in the following
discussion. As can be seen from Fig. 10, the effect of the increase in density
is negligible.
244
L. Shao, S. B. Riffat
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
-5
-10
v
.o
.o
-15
m 2u4e4d
2u8e4d
-25
Normalised cell n u m b e r
Fig. 11.
P r e s s u r e d i s t r i b u t i o n s , c o r r e s p o n d i n g to t h r e e d i f f e r e n t I g r i d d e n s i t i e s in t h e
elbow section, along the duct axis for a double elbow with zero separation.
Figure 11 shows the effect of increasing the I grid density in the elbow
section. The double elbow in this case has zero separation and the three
curves correspond to cases (2u2e2d, 2u4e4d, 2u8e4d) with three levels of
grid density in the elbow section. As the density increases, the pressure
drop across the elbow decreases but the rate of this decrease slows down.
The difference between the pressure losses predicted by 2u4e4d abd 2u8e4d
is less than 1 Pa and a further increase in density would not lead to more
than a negligible difference. The high density in the elbow section, eight
times the original level, can therefore be regarded as adequate.
Pressure loss in duct fittings, based on a large number of grid set-ups,
has been computed. The largest number of grid lines in the J and K
direction is 21 and in the I direction 202. Simultaneous increases of grid
density in all dimensions and all sections of the computational domain
would have led to a grid size many times over the limit that can be handled
with the computing resources available to the authors. The alternative
approach presented above enabled the grid dependency test to be carried
out without resorting to larger and faster computers.
0.2
0.4
0.6
0.8
245
1.0
1.2
-5
-10
~. -15
2u2e2d
- 2u2e2d+ex tended-d-section
-20
-25
N o r m a l i s e d cell number
Fig. 12.
P r e s s u r e d i s t r i b u t i o n s , c o r r e s p o n d i n g to t w o d i f f e r e n t l e n g t h s o f t he d o w n -
stream section, along the duct axis of a double elbow with zero separation.
section 2, should be sufficiently long to avoid artificially restricting flow
development, altering the flow field and affecting the prediction accuracy.
The adequacy of the length of the downstream straight section was
examined by computing the pressure loss based on two domain set-ups,
which are identical except that the downstream section of one is twice as
long as that of the other (i.e. the original length). The double elbows
have zero separation between them. As can be seen from Fig. 12, the
predictions based on the two set-ups are virtually identical, indicating
that the original length is sufficient.
All the computation results presented above show that the pressure
loss curves corresponding to the upstream sections are virtually straight
lines. In other words, there is little variation in the pattern of pressure
loss for virtually the full length of the upstream section, which can be
shortened without affecting the prediction for the elbow section. The
length of the upstream section is sufficient, but the apparent absence of a
'development section', which exhibits different pressure loss characteristics
as described in section 2, is somewhat unexpected. This is probably due
to the relatively thin boundary layer, on the duct wall, associated with
the low viscosity of air and relatively large dimensions of the duct. This
boundary layer remains thin especially under the turbulent conditions of
HVAC flow and it never merges to form the 'fully developed' velocity
profile. The C F D results indicate that the viscosity in the duct flow
manifests itself in the form of thin boundary layers rather than in 'fully
developed' flow. This is supported by both pressure and velocity data.
L. Shao, S. B. Riffat
246
0.2
0.4
0.6
1.0
1.2
-5
-10
-15
9
e~
-20
-25
-30
Normalised cell number
Fig. 13.
Accuracy achieved
The computation of the pressure loss was carried out using a large
number of cases with various turbulence models, discretisation schemes,
grids and computation domains. Results from a selection of these cases
(all based on a double elbow with zero separation) are shown in Fig. 13
to demonstrate the differences between the predicted pressure losses,
which can be as high as several hundred per cent.
Despite the large differences between the k factor values given by
CIBSE and ASHRAE for many duct fittings, their values for double elbows
are consistently close to each other and so can be regarded as accurate.
CIBSE and A S H R A E values for double elbows with zero separation are
0.25 and 0.24 respectively. The most accurate prediction achieved in this
study is 0.22, which is based on the k-e model and the higher order QUICK
scheme and represents a relative error of about 10%. This value of k
factor is also the smallest obtained; any other combination of turbulence
model and discretisation scheme would lead to relative errors in the region
of + 100% or higher. It is also based on a grid and computational domain
that has been tested in the way presented above. For the example of double
elbows, relatively low grid densities in the J and K directions and the I
direction in the straight sections upstream/downstream of the elbow sections
are sufficient but a higher density is required in the elbow section in the I
direction. The exact grid densities for particular fittings should be deter-
247
mined by grid dependency tests. It has also been shown that a computational domain set-up with short upstream/downstream sections is adequate.
CONCLUSIONS
An investigation has been made into the accuracy of using CFD for the
prediction of pressure loss in HVAC duct fittings, and the factors affecting
its accuracy. The well-known CFD package FLUENT was used to predict
pressure loss in a common type of duct fitting, viz,, double elbows. It was
found that the pressure loss predicted by the Reynolds stress model is
more than double that predicted by the k - e model, and the prediction
based on power law discretisation/interpolation scheme is much higher
than that based on the higher order Q U I C K scheme. The most accurate
prediction achieved in this study (relative error of about 10%) was based
on the k - e model and the higher order QUICK scheme; any other
combination of turbulence model and discretisation scheme would lead
to relative errors in the region of +100% or higher.
Grid dependency testing by doubling grid density in three dimensions
simultaneously is costly, even impractical, in terms of computing resources.
This study adopted an alternative approach in which grid doubling is not
implemented simultaneously in all dimensions and sections of the computational domain. The arrangement enables the grid dependency tests to
be carried out without dramatically increasing demands on computing
resources. It was found that relatively low grid densities in the J and K
directions and the I direction in the straight sections upstream/downstream
of the elbow sections are sufficient but a higher density is required in the
elbow section in the I direction. It should be pointed out that exact grid
densities for particular fittings should be determined by grid dependency
tests. It has also been shown that computational domain set-up with
short upstream/downstream sections is adequate.
The above conclusions refer specifically to double elbows and may not
necessarily apply to other fittings, as one specific set-up which yields high
accuracy in simulation of one fitting may lead to large errors in that of
another. The accuracy of CFD and its maximisation should be examined
on a fitting-by-fitting basis. This paper illustrates the important factors
involved and the procedures to follow.
ACKNOWLEDGEMENT
This work is funded by the UK Engineering and Physical Sciences Research
Council.
248
L. Shao, S. B. Riffat
REFERENCES
1. CIBSE, Reference Data. The Chartered Institution of Building Services
Engineers, London, 1986.
2. ASHRAE, Fundamentals. American Society of Heating, Refrigeration and
Air Conditioning Engineers, Atlanta, 1989.
3. Cheong, K. W. & Riffat, S. B., A new method for determination of velocity
pressure loss-factors for HVAC system components. Proceedings 13th AIVC
Conference, France, 1992, pp. 546-61.
4. Riffat, S. B., Shao, L. & Woods, A., Determination of k-factors by measurement
and CFD modelling. 15th AIVC Conference, France, 1994.
5. Shao, L. & Riffat, S. B., CFD for improvement of k-factor accuracy in
HVAC systems. Ventilation '94, Sweden, 1994.