Documente Academic
Documente Profesional
Documente Cultură
1682-RP
STUDY TO IDENTIFY CFD MODELS FOR
USE INDETERMINING HVAC DUCT FITTING
LOSS COEFFICIENTS: CFD PROGRAM
Approval: December 2016
Shaping Tomorrow’s
Built Environment Today
©2012 ASHRAE www.ashrae.org. This material may not be copied nor distributed in either paper or digital form without
ASHRAE’s permission. Requests for this report should be directed to the ASHRAE Manager of Research and Technical
Services.
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
STUDY TO IDENTIFY CFD MODELS FOR USE IN DETERMINING
HVAC DUCT FITTING LOSS COEFFICIENTS:
CFD PROGRAM
ASHRAE RP-1682
Submitted by
to
March 2016
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
ABSTRACT
This report presents results from a systematic study to establish whether CFD
techniques are capable of predicting pressure drop in duct fittings including single elbow,
close-coupled five-gore elbows having nominal diameters of 203 mm (8 in.) and 304.8
mm (12 in.) and turning radii r/D = 1.5. The close-coupled elbow combinations
comprised either a Z-shape or a U-shape. In every instance the duct length separating the
center points of the elbows was systematically varied. An experimental program was
likewise conducted to verify the CFD predictions, and data from the measurements are
included. Zero-length pressure loss coefficients were predicted using the k- and k-
models, as well as the Reynolds Stress Model (RSM), and compared to experimental
data. Two-equation turbulence models except the k-model predicted incorrect trends
when applied to flow in U- and Z-configuration ducts. However, the k- and the
Reynolds Stress Models (RSM) with enhanced wall treatment were generally able to
correctly predict elbow loss coefficients with an error of less than 15%. None of the two-
equation and RSM models could predict the experimental velocity profiles of Z- and U-
configurations accurately.
ii
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
TABLE OF CONTENTS
ABSTRACT ........................................................................................................................ ii
LIST OF TABLES ............................................................................................................. iv
LIST OF FIGURES ............................................................................................................ v
NOMENCLATURE ........................................................................................................ viii
INTRODUCTION .............................................................................................................. 1
TEST PROGRAM .............................................................................................................. 3
CFD MODELING .............................................................................................................. 6
EXPERIMENTAL DATA REDUCTION ........................................................................ 14
EXPERIMENTAL RESULTS.......................................................................................... 18
CFD RESULTS ................................................................................................................ 26
DISCUSSION ................................................................................................................... 47
SUMMARY OF CAPABILITIES AND LIMITATIONS OF TURBULENCE MODELS
........................................................................................................................................... 50
SUMMARY OF COMPUTATIONAL EFFORT AND SOLUTION BEHAVIOR ........ 53
SUMMARY OF GRID CONSIDERATIONS FOR TURBULENT FLOW
SIMULATION .................................................................................................................. 54
CONCLUSIONS............................................................................................................... 56
SUMMARY OF GENERAL RECOMMENDATIONS .................................................. 58
SUMMARY OF RECOMMENDATIONS FOR DUCT FITTINGS ............................... 59
REFERENCES ................................................................................................................. 60
APPENDIX A ................................................................................................................... 64
APPENDIX B ................................................................................................................... 69
iii
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
LIST OF TABLES
Table 1. 203 mm (8.0 in.) Diameter Straight Duct Tare Pressure Loss Testing
Table 2. 203 mm (8.0 in.) Diameter Single Elbow Pressure Loss Testing
Table 3. 203 mm (8.0 in.) Diameter Double Elbow Pressure Loss Testing: Z-
Configuration
Table 4. 203 mm (8.0 in.) Diameter Double Elbow Pressure Loss Testing: U-
Configuration
Table 6. Error for 203 mm (8.0 in.) Diameter Double Elbow Loss Coefficient: U-
Table 7. Error for 308.4 mm (12.0 in.) Diameter Double Elbow Loss Coefficient: U-
Table 8. Error for 203 mm (8.0 in.) Diameter Double Elbow Loss Coefficient: U-
Table 9. Error of loss coefficient for 304.8 mm (12.0 in.) Diameter Double Elbow Z-
Configuration (LoD=10)
Table 10. Maximum % Error - Experimental Data vs. Different CFD Models
iv
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
LIST OF FIGURES
Figure 1. Secondary Flows at the Center (Left) and Exit (Right) of Square Cross Section
Figure 5. 203 mm (8.0 in.) Diameter Straight Duct Moody Diagram - Comparison of
Figure 6. 203 mm (8.0 in.) Diameter Straight Duct Moody Diagram - Comparison of
CFD k-realizable k- and RNG k- turbulence models with wall roughness to
experimental results.
Figure 7. 203 mm (8.0 in.) Diameter Straight Duct Moody Diagram - Comparison of
CFD standard k-and SSTk- turbulence models with wall roughness to experimental
results.
Figure 8. 203 mm (8.0 in.) Diameter Straight Duct Moody Diagram - Comparison of
Figure 9. 203 mm (8.0 in.) Diameter Single Elbow Loss Coefficient - Comparison of
Figure 10. 203 mm (8.0 in.) Diameter Double Elbow Loss Coefficient: Z-Configuration
Lint = 2.52 m (8.28 ft) - Comparison of CFD RSM, k- and k- turbulence models to
experimental results
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
Figure 11. U-configuration duct fitting considered in this study is shown by Figure 11
Figure 12. 203 mm (8.0 in.) Diameter Double Elbow Loss Coefficient: U-Configuration
Figure 13. Numerical grid for 304.8 mm (12.0 in.) Diameter Double Elbow U-
Figure 14. Scaled residuals for 304.8 mm (12.0 in.) Diameter Double Elbow U-
Figure 15. Contours of static pressure (in Pa) for 304.8 mm (12.0 in.) Diameter Double
Figure 16. Contours of dynamic pressure (in Pa) for 304.8 mm (12.0 in.) Diameter
Figure 17. Comparison of total pressure loss vs velocity pressure for 304.8 mm (12.0 in.)
Figure 18. CFD velocity vectors for 304.8 mm (12.0 in.) Diameter Double Elbow U-
Figure 19. CFD velocity profile for 304.8 mm (12.0 in.) Diameter Double Elbow U-
Figure 20. Comparison of experimental and CFD velocity profile at x/d = 1 for 304.8 mm
(12.0 in.) Diameter Double Elbow U-Configuration Lint = 2.52 m (8.28 ft)
vi
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
Figure 21. Comparison of experimental and CFD velocity profile at x/d = 7 for 304.8 mm
(12.0 in.) Diameter Double Elbow U-Configuration Lint = 2.52 m (8.28 ft)
Figure 22. Comparison of experimental and CFD velocity profile at x/d = 9 for 304.8 mm
(12.0 in.) Diameter Double Elbow U-Configuration Lint = 2.52 m (8.28 ft)
Figure 23. Comparison of total pressure loss vs velocity pressure for 203 mm (8.0 in.)
Figure 24. Comparison of loss coefficient for 304.8 mm (12.0 in.) Diameter Double
Figure 25. Comparison of total pressure loss vs velocity pressure for 304.8 mm (12.0 in.)
Figure 26. CFD velocity vectors for 304.8 mm (12.0 in.) Diameter Double Elbow Z-
Figure 27. CFD velocity profile for 304.8 mm (12.0 in.) Diameter Double Elbow Z-
vii
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
NOMENCLATURE
k+ = k s u , dimensionless
L
Q = volumetric flow rate, s (cfm)
viii
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
u = friction velocity, m/s (ft/min)
u = dimensionless velocity
m ft
V = velocity, (s)
s
GREEK SYMBOLS
kg lbm
ρ = density, m3 ( ft3 )
Ns lbm
μ = dynamic viscosity, m2 ( s ft )
SUBSCRIPTS
e = exit plane
x = plane 1, 2, - - -, n, as applicable
z = upstream plane
ix
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
INTRODUCTION
This report presents a systematic study to establish whether CFD techniques are
capable of obtaining pressure drop predictions in elbows that are accurate to within 15%
of experimental loss coefficients. A test program was initiated to measure the friction
factor in a straight duct, and the loss coefficient of both a single five-gore elbow and two
CFD turbulence models in predicting pressure drop for each of the configurations was
models, including k-ε, k-ω and the Reynolds Stress Model (RSM) to establish their
Since 1988 there have been several experimental research projects sponsored by
ASHRAE, e.g., RP-551 (1991), RP-690 (1994), RP-1319 (2008), RP-1488 (2011), and
RP-1606 (2014), designed to populate the Duct Fitting Database (DFDB, 2014). The
results from these studies have been published in the open literature. The pressure loss for
flat oval straight ducts has been reported in Townsend et al. (1994). For flat oval tee and
lateral fittings, it has been found that branch loss coefficients were power law functions
of branch-to-common flow rate ratio, and the main loss coefficients were power law
functions of the main-to-common flow rate ratio, and generally less than 0.02 at high
flow rate ratios; refer to Townsend et al. (1996a), Idem and Khodabakhsh (1999), Idem
(2003), Kulkarni et al. (2011), and Gibbs and Idem (2012). Loss coefficient data for flat
oval elbows have been expressed as power law functions of aspect ratio and hydraulic
diameter; refer to Townsend et al. (1996b) and Kulkarni et al. (2009). The goal of RP-
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
1606 (2014) was to evaluate loss coefficients of flat oval to flat oval transitions where the
major and minor axes were either concentric or eccentric. Recently RP-1493 (2011) was
completed. The primary objective of that project was to ascertain whether Computational
Fluid Dynamics (CFD) could be used to determine loss coefficients for divided flow duct
fittings, in the absence of empirical input. The ultimate goal of that effort was to reduce
the number of expensive and time-consuming laboratory tests required to enhance the
DFDB. The main findings of the project were that the trends of the pressure loss
coefficients were predicted correctly, while a CFD prediction accuracy within 15% of
experimental data was not achieved. The results of that study are summarized in Sleiti et
al. (2013), Liu et al. (2012), Manning et al. (2013), and Gutovic et al. (2013). An
pressure loss in single or close-coupled fittings; these references include Gan and Riffat
(1995), Mahank and Mumma (1997), Mumma et al. (1997, 1998), Rahmeyer (2002),
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
TEST PROGRAM
ASHRAE under the same RP-1682. Here, a brief description of the experimental work is
provided for the convenience of the reader. An experimental apparatus was constructed to
measure the friction factor in a straight duct, and the loss coefficient of both a single five-
gore elbow and two close-coupled five-gore elbows. The measurements of pressure drop
and volumetric flow rate through the ductwork and fittings were performed in accordance
with ASHRAE Standard 120-2008. The elbow pressure loss experiments were preceded
by a series of tests designed to evaluate the friction factor of straight ducts. A bellmouth
was mounted at the entrance of the ductwork to ensure uniform inlet flow. Pressure taps
soldered to the ducts were employed to measure the pressure drop at specific distances
prescribed in Standard 120-2008. In order to establish the baseline loss coefficient for a
single elbow, the straight duct setup was modified by inserting an elbow and measuring
the pressure drop across the single fitting. The setup shown in Figure 16 of Standard 120
was used to measure the friction factor and relative roughness of the duct connected to
the elbows. Similarly Figure 17 of Standard 120 depicts the setup that was employed to
determine loss coefficient for a single elbow. Two close-coupling configurations were
studied in the present experiments. Per the terminology employed by Sami and Cui
(2004), in one instance the close-coupled test apparatus constituted a Z-shape, whereas in
other instances the experiments conformed to a U-shape. In each case the straight duct
length inserted between the upstream elbow and the downstream elbow was
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
In every instance the turning radii of the elbows was fixed at r/D = 1.5. The
elbows and ductwork were constructed from 26-gauge galvanized steel. The ducts
upstream and downstream of the test section were each 1.20 m (4 ft) in length, and were
connected by slip couplings. For close coupled elbows the intermediate length x,
measured from the exit plane of the upstream elbow to the entrance plane of the
downstream elbow, was varied from 0 m (0 ft) to 3.05 m (10 ft) in increments of 0.60 m
(2 ft). For an intermediate length of 0 m (0 ft), the two elbows were butted together, and
the joint wrapped tightly by a rubber sheet for mechanical integrity. There were no joints
in the intermediate duct mounted between the close-coupled elbows. All joints and fitting
A 30 hp centrifugal fan provided air flow through the test apparatus in the forced
120-2008 was used to measure the volume flow rate through the test setup. Pressure drop
measurements over the test section were performed using a liquid-filled micromanometer
having a measurement accuracy of 0.025 mm (0.001 in). Likewise, the static gage
pressure upstream and downstream of the test section was measured with electronic
manometers having a readability of 0.25 mm (0.01 in). The nozzle pressure loss was
Static pressure was measured in the chamber (to assess air density) using an electronic
manometer having the scale readability of 0.25 mm (0.01 in.). A VFD was used to
control the flow rate through the test section. Various combinations of flow nozzles were
employed, depending on the desired flow rate, and unused nozzles were plugged with
smooth vinyl balls. The air temperature in the nozzle chamber was measured using a
4
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
mercury thermometer having a scale readability of 0.6C (1F). The dry-bulb
temperature and wet-bulb temperature of the ambient air were measured using an
mercury. At least eight test points, evenly spaced over the range of test velocities, were
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
CFD MODELING
straight duct, and in both single and close-coupled five gore elbows (Z-configuration and
U-configuration). The finite volume method was the numerical solution approach used to
the governing equations to algebraic equations that were solved numerically. This control
volume technique consisted of integrating the governing equations about each control
basis.
cannot be solved analytically, except for a few special cases. Consequently for most
Numerical Simulations (DNS) and Large Eddy Simulations (LES) are the most accurate
but they require large computational resources. Although Reynolds Averaged Navier
resources, and are widely used for industrial applications. Thus, in this report RANS
In RANS, the quantities in the NS-equation could be divided into mean and
of a parameter over time. Applying the Reynolds averaging to the NS-equations results in
the Reynolds Averaged Navier-Stokes equations. This procedure produces the Reynolds
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
stresses, i.e., ij u i u j , which are unknown and need to be evaluated using turbulence
modeling. This is referred to as the closure problem with Reynolds averaging. To model
the Reynolds stresses there are two approaches, the eddy-viscosity models (EVM), and
the Reynolds stress models (RSM). Details on EVM and RSM turbulence modeling are
provided in Appendix A. In RSM the actual stresses are solved, while in the EVM the
Boussinesq hypothesis is employed to estimate τij. In EVM the directional properties are
not considered, and the turbulence is assumed to be isotropic. While this is true for
smaller eddies at high Reynolds numbers, the large eddies are anisotropic due to the
strain rate of the mean flow. The local equilibrium assumption in EVM is another
problem where the production is assumed to be equal locally to the dissipation term. For
uni-directional flow and for flow where turbulence is evolving at a sufficiently rapid rate,
such that the effects of past events do not dominate the dynamics, the estimates based on
local scales can give relatively accurate results. Typical flows where two-equation
models have been shown to fail are flows with sudden changes in mean strain rate,
curved surfaces, secondary motions, rotating and stratified fluids, flows with separation,
EVMs were used to perform the CFD simulations: (i) the standard k-ε model, (ii) the
renormalization group k-ε model, (iii) the realizable k-ε model, (iv) the standard k-ω
model, and (v) the Shear-Stress Transport (SST) k-ω model. The Reynolds Stress Model
(RSM) solves for the Reynolds Stresses using individual transport equations. RSM is also
used in this report and its predictions are compared to EVM predictions.
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
The complex geometry, imposed centrifugal forces, and the boundary conditions
within the ducts complicate the flow fields. In this section these factors and how they
influence the flow are discussed. Duct walls have several effects on flow and turbulence.
These effects include: (i) the no-slip condition, because of viscous effects, (ii) the
blocking effect that makes the turbulence anisotropic by suppressing the fluctuations in
the wall normal direction, (iii) the shearing mechanism in the flow increases the turbulent
production, (iv) the wall reflection effect through reduction of stress components
The surface roughness effects are of considerable importance. When the walls are
smooth, the shear stress at the surface is transmitted to the flow via a viscous sublayer,
where the velocity in this sublayer varies linearly, as in laminar Couette flow, such that
lam (const)u 2 . The constant is the value of the velocity where it ceases to behave
accounts for conditions at the boundary, and the dimensionless distance from the wall
y u y / . It adds a uniform velocity to the entire flow with no change in its internal
structure. For a rough surface with roughness height larger than lam , the stress is
transmitted by pressure forces in the wakes of the roughness elements, rather than by
1 y
viscosity. The form of the profile given in the log layer is then U ln 8.5 ,
k s
8
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
where ks is Nikuradse’s equivalent roughness height, and is related to the roughness
geometry at the boundary. For boundary layers that are considered fully rough,
2.5
x
pressure boundary layers is C f 2.87 1.58 log . Numerically, the wall node
k s
must be placed at some distance above the equivalent roughness. This follows from the
physical argument that the flow cannot exist inside the wall, whose edges effectively
In this study both wall functions and enhanced near wall treatment approaches
were used and compared for near wall treatment. Details of both wall functions and
enhanced near wall treatment approaches are provided in APPENDIX B. The standard
wall functions are semi-empirical formulas used to bridge the viscous sublayer and buffer
layer near the wall, while in the enhanced wall treatment approach the viscosity-affected
region is resolved all the way to the wall, including the viscous sublayer.
There are two types of secondary flows: (i) secondary flows generated by inviscid
effects (first kind secondary flows) and (ii) secondary flows generated by Reynolds
stresses (second kind secondary flows). Secondary flows of the first kind are generated
when span wise pressure gradients occur in the mean flow. Secondary flows of the
second type are generated by Reynolds stresses (turbulence induced) which develop in
9
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
the corners of the duct. There the cross-stream gradients of the Reynolds stresses generate
weak streamwise vorticity. Because these secondary flows are driven by gradients of
Effects of curvature
Duct systems usually have curved ducts in the form of one or more elbows. As the
with a large turning radius, the pressure gradient varies almost linearly from the inner to
the outer radii. The high momentum fluid in the duct center tends to drift outwards.
Continuity requires that the outward motion in the center of the duct be balanced by a
reverse flow along the walls, where the centrifugal force is less because the streamwise
velocity is lower. This flow behavior generates a circular motion in the cross-stream
plane. At the exit of the bend two opposing rotating cells with vorticity in the streamwise
direction will appear, see Figure 1. The Dean number, De, is the parameter that measures
the curvature effect relative to viscous effect, where De ReD h R 2 . In this instance
1
Re is the Reynolds number and R is the radius of the curvature. The Dean number gives a
measure of the degree of stability. The flow is considered unstable when exceeding the
critical Dean number according to Rayleigh’s criteria. For the flow in a bend the convex
side (inner) is stable, while the concave side (outer) is unstable. The fluid viscosity
affects the stability of the flow through bend and hence it is important for turbulence
modeling.
10
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
Figure 1. Secondary Flows at the Center (Left) and Exit (Right) of Square Cross
Section U-bend, Re=25,000, Density Ratio = 0.13, RSM Predictions.
Modeling details
In this report the CFD code FLUENT was employed to perform the simulations.
The numerical solution method used to solve N-S equations was the finite volume
to algebraic equations that could be solved numerically. This control volume technique
consisted of integrating the governing equations about each control volume, yielding
each model to experimental measurements in a straight duct, a single elbow, and close-
coupled Z-configuration and U-configuration elbows. The numerical grid for the studied
11
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
Figure 3. Numerical grid for single elbow.
12
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
Figure 4. Numerical grid for Z-configuration duct geometry.
For enhanced wall treatment, y for the first cell next to a wall must be of order
unity. To resolve the near wall viscous region 10 grid points were placed in the boundary
layer near all walls. The minimum convergence criterion for all velocity components and
turbulence quantities error was 10-6. A grid refinement study was conducted to determine
the optimum grid distribution for all studied geometries. The results of the grid
13
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
EXPERIMENTAL DATA REDUCTION
In order to accurately calculate the total pressure loss across the test elbow(s) it
was first necessary to determine the Darcy friction factor, which is defined as follows
p12 / L12
f (1 SI)
1
2 1V12 /( D / 1000)
p12 / L12
f (1 I-P)
1 V1 / 1097 / D / 12
2
The plane locations are provided in Figure 17 of Standard 120-2008. The volumetric flow
2 p5-6
Q 1000 Yn
ρ5
(C n A n ) (2 SI)
p5-6
Q 1098 Yn
ρ5
(C n A n ) (2 I-P)
where 5 denotes the section upstream of the nozzle, and 6 indicates the nozzle throat.
Additional equations necessary to support the flow calculation per Equation 2 can be
found in ASHRAE Standard 120. Determination of the flow rate required the
measurement of the pressure drop across the nozzle board, the static pressure of the
plenum chamber, and the temperature inside the plenum chamber. The density of air
in the test section was calculated by means of the correlations presented in ASHRAE
Standard 120 based on measurements of the ambient dry and wet bulb temperature
and barometric pressure, and the test section temperature and average static pressure.
14
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
1 V1 D1 / 1000
Re1 (3 SI)
1
where V is average air velocity and D is the measured duct diameter. The average duct
velocity is the ratio of the measured volumetric flow rate to the cross-sectional area of the
test section
Q / 1000
V1 1 (4 SI)
A
Q
V1 1 (4 I-P)
A
The Moody diagram was plotted using the Colebrook formula. The plots are curves of
constant relative roughness /D as function of friction factor f and the Reynolds number
1 ε/D 2.51
2 log 1 (5 SI)
f 3.7 Re1 f
The appropriate value of relative roughness of the duct was determined by fitting the
The elbow pressure loss coefficient is defined as the ratio of the total pressure loss
15
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
Δpt,12
C (6)
pv7
The velocity pressure depends on the measured average velocity in the duct
1
pv7 ρ V2 (7)
2
Referring to Standard 120, the total pressure loss across an equal area elbow can be
determined as
p
p t ,12 ps,78 L 71 L 28 f (8)
L
The subscripts in the above equation refer to the planes indicated in Figure 17 of
Standard 120. The terms L7-1 and L2-8 represent the separation distance between the
upstream taps and the center point of the elbow, and center point of the elbow to the
downstream pressure taps, respectively. The pressure friction loss per unit length p f is
L
the duct tare pressure loss per unit length, as calculated by Equation 9
p f p v1 f
(9)
L D1
In the case of experiments that were performed on two close-coupled elbows, the
loss coefficient was again calculated using Equation 7. However, the total pressure loss
p
p t ,12 ps,78 L 71 L 28 L int f . (10)
L
In this instance the quantity Lint refers to the measured distance from center point to
center point of the two close-coupled elbows. The friction pressure loss p f was
L
16
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
p f
calculated from Equation 9. The factor Lint accounts for the additional loss due to
L
The total pressure losses were measured experimentally at each flow rate for all
setups, and the least squares method was employed in order to obtain an overall loss
p t ,12 C p v 7 . (11)
In the above equation the tare friction pressure loss has been subtracted from the overall
pressure drop across the fitting thus yielding a zero-length loss coefficient; refer to
Equation 10 for example. The slope of the curve Δp t,12 plotted against p v 7 is interpreted
17
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
EXPERIMENTAL RESULTS
Pressure drop measurements were performed on the 203 mm. (8 in.) diameter
straight duct in order to evaluate the tare pressure loss The Darcy friction factor was
calculated for each flow rate by Equation 1 and plotted against the Reynolds number on a
Figure 5. 203 mm (8.0 in.) Diameter Straight Duct Moody Diagram - Comparison of
CFD k- turbulence model to experimental results.
18
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
Figure 6. 203 mm (8.0 in.) Diameter Straight Duct Moody Diagram - Comparison of
CFD k-realizable k- and RNG k- turbulence models with wall roughness to
experimental results.
19
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
Figure 7. 203 mm (8.0 in.) Diameter Straight Duct Moody Diagram - Comparison of
CFD standard k-and SSTk- turbulence models with wall roughness to
experimental results.
20
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
Figure 8. 203 mm (8.0 in.) Diameter Straight Duct Moody Diagram - Comparison of
CFD RSM turbulence model with wall roughness to experimental results.
The Colebrook equation was used to determine the relative roughness values by the
least squares method. To a close approximation, the friction factor data followed a single
relative roughness curve /D = 0.0009. The tare pressure loss was calculated using
Equation 5, in conjunction with Equation 9. Therein the total pressure loss across the
elbow(s) was calculated using Equation 8 for a single elbow, or Equation 10 for two
close-coupled elbows, and plotted as a function of the velocity pressure through the
elbow(s), thereby obtaining the zero-length loss coefficient for each elbow combination.
Figure 9 illustrates the pressure loss coefficient data obtained for a single 203 mm. (8 in.)
diameter 5-gore 90º elbow with a dimensionless turning radius R/D = 1.5.
21
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
70
CFD RSM
60
CFD k-
50
Total Pressure Loss (Pa)
30
20
10
0
0 50 100 150 200 250 300 350
Velocity Pressure (Pa)
Figure 9. 203 mm (8.0 in.) Diameter Single Elbow Loss Coefficient - Comparison of
CFD RSM, k- and k- turbulence models to experimental results.
Figure 10 depicts the measured pressure loss coefficient data for two close-coupled
Lint = 2.52 m (8.28 ft). In every instance the experimental loss coefficient data are
22
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
180
140
120
Total Pressure Loss (Pa)
Experimental
100 Data
CFD k-
80 CFD k-
60
40
20
0
0 50 100 150 200 250 300 350 400 450 500
Velocity Pressure (Pa)
Figure 10. 203 mm (8.0 in.) Diameter Double Elbow Loss Coefficient: Z-
Configuration Lint = 2.52 m (8.28 ft) - Comparison of CFD RSM, k- and k-
turbulence models to experimental results
The experimental results are summarized in Tables 1 through 4, which also provide
test setup dimensions, per Standard 120. The single elbow loss coefficient for a 203 mm
(8 in.) diameter five-gore elbow with a dimensionless turning radius R/D = 1.5 was
determined to equal C = 0.19. By comparison, for that single elbow geometry the
ASHRAE Duct Fitting Database (2015) indicates a value C = 0.23. Uncertainty estimates
elbow loss coefficients, i.e., either C or 2C, are likewise tabulated. The case where Lint
= 0.48 m (1.57 ft) corresponds to cases where the two elbows were butted together to
form a close-coupled pair with no straight duct present between the elbows.
23
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
Table 1. 203 mm (8.0 in.) Diameter Straight Duct Tare Pressure Loss Testing
Table 2. 203 mm (8.0 in.) Diameter Single Elbow Pressure Loss Testing
Table 3. 203 mm (8.0 in.) Diameter Double Elbow Pressure Loss Testing: Z-
Configuration
24
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
Table 4. 203 mm (8.0 in.) Diameter Double Elbow Pressure Loss Testing: U-
Configuration
25
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
CFD RESULTS
Figure 5 shows a comparison between friction factors for a smooth straight duct
calculated using the Colebrook equation, and CFD k- turbulence model predictions. A
standard k- turbulence model was used with a 60 200 grid and enhanced near wall
treatment. The results are summarized as follows. The CFD predictions were accurate to
within 5% for Re less than 250,000. For higher Re the error increased to more than 10%.
The reason for the increased error is that y increases for high Re numbers, which makes
the solution less accurate. For the present study let the error be defined as: (CFD Friction
refinement study was conducted to determine the optimum grid distribution for cases
with Re less than 250,000. The results of the grid refinement study are summarized in
Table 5.
Grid Maximum
Remarks
Distribution Error
30 200 8% Needs refinement in radial direction
60 200 5% Acceptable
90 200 4.7% Refinement in radial direction is not justified
90 100 7% Needs refinement in axial direction
90 400 4.8% Refinement in axial direction is not justified
In addition, Figure 5 also compares friction factors for a smooth straight duct
26
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
CFD k- model. In this case, the numerical grid points were increased near the wall to
maintain y less than 25 for cases where Re is more than 250,000. The error in this case
became less than 3% compared to values obtained from the Colebrook equation for a
smooth duct. Likewise a standard k- turbulence model was used with a 60 200 size
and standard wall functions to predict the friction factor of a straight rough duct with /D
= 0.0009, and to compare to experimental data from this study, as well as friction factors
determined using the Colebrook equation. These results show that the CFD predictions
are accurate within 7% error for high Re (more than 250,000) and within 12% for low Re
and three CFD k- turbulence models, assuming a duct with a roughness /D = 0.0009. In
every instance a 60 200 grid and standard wall functions were employed by the models.
Results determined using the k- turbulence model indicate that the error in friction factor
predictions ranged from 2% (for high Re) to more than 13% (for low Re). Predictions
obtained using the RNG k- turbulence model revealed that errors in the friction factor
predictions ranged from 6% to more than 17%. Likewise calculations of the friction
factor achieved using the realizable k- turbulence model yielded errors in friction factor
and two CFD k- turbulence models, with a prescribed duct wall roughness /D =
0.0009. In each case a 60 200 grid and standard wall functions were employed by the
CFD models. The friction factor predictions obtained using the standard k- model show
27
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
that the errors ranges from 2% (for high Re) to more than 13% (for low Re). Similarly
friction factor predictions achieved by means of the SST k-model ranged from 6% for
high Re to more than 18% for low Re. Figure 8 illustrates a comparison between
experimental friction factors and RSM turbulence model predictions with a wall
roughness /D = 0.0009. The results indicate that errors in friction factor predictions
using the RSM CFD model ranged from 13% to more than 22%.
turbulence model predictions of total pressure loss versus velocity pressure for a single
elbow. The turbulence models were used with a grid size of 60 200 in the entrance
region, 60 22 in the elbow curve region and 60 160 in the exit region. Standard wall
functions were employed, and a wall roughness /D = 0.0009 was assumed. The results
show that errors in CFD predictions of the elbow loss coefficient based on the Standard
k-model ranged from 5% to more than 14%. Likewise prediction errors obtained by
means of the standard k- model ranged from 9% to more than 19%. Moreover CFD
predictions resulting from use of the RSM turbulence model ranged from 4% to 11%.
This approach was more accurate than either the k- or k- turbulence models.
turbulence model predictions for the Z-configuration with Lint = 2.52 m (8.28 ft). For
28
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
every case considered in this study the turbulence models employed a grid size of 60
200 in the entrance region, 60 22 in the curved regions, and 60 160 in the exit region.
Likewise standard wall functions were used, and a duct roughness of /D = 0.0009 was
employed. The error in CFD predictions resulting from the standard k- model ranged
from 0.1% to more than 14%. In addition the error in CFD predictions derived from the
standard k- model ranged from 7% to more than 24%. By comparison the error from
CFD predictions obtained using RSM ranged from 0.4% to more than 12%. Once again
RSM loss coefficient predictions were more accurate than k- and k- turbulence models.
experimental and CFD (using k- model) loss coefficient is provided by Figure 12 for U-
Configuration for 8 in diameter with separation distance LoD = 10. The error is defined
29
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
Figure 11. U-configuration duct fitting considered in this study is shown by Figure
11 based on ANSI/ASHRAE Standard 120-2008: Figure 17.
30
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
180
160 Experimental
CFD
140 Linear (Experimental) y = 0.352x
R² = 0.994
120
Total Pressure Loss (Pa)
100
80
60
40
20
0
0 100 200 300 400 500
Velocity Pressure (Pa)
Figure 12. 203 mm (8.0 in.) Diameter Double Elbow Loss Coefficient: U-
Configuration Lint = 2.52 m (8.28 ft)
Table 6. Error for 203 mm (8.0 in.) Diameter Double Elbow Loss Coefficient:
U-Configuration Lint = 2.52 m (8.28 ft)
pv8 Re8 pf,7-8 pt,1-2 Error
C
Pa Pa/m Pa %
159.5 222237 15.9 58.0 0.363 2.39
219.7 259702 21.5 77.7 0.354 3.88
289.5 299337 28.4 99.2 0.343 1.88
365.8 335362 35.6 122.5 0.335 5.20
454.4 373857 44.1 149.1 0.328 5.87
129.5 199503 13.0 48.1 0.371 -1.67
233.9 267881 23.1 81.8 0.350 1.53
368.8 336590 36.0 123.4 0.335 2.52
35.1 104739 3.8 16.0 0.455 6.66
64.4 141000 6.7 27.6 0.428 6.73
31
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
CFD - U-Configuration for 12 in diameter duct
In this section, a detailed CFD modeling and results are provided for the 12 in
diameter duct as the experimental results showed different trend of the loss coefficient
The numerical grid for the 12 in duct U-Configuration with separation distance.
LoD = 10 is shown in Figure 13. A structured grid is used considering the finer mesh
near walls and with aspect ratio of not more than 10.
Figure 13. Numerical grid for 304.8 mm (12.0 in.) Diameter Double Elbow U-
Configuration Lint = 2.52 m (8.28 ft)
32
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
The numerical convergence of the solution is shown by Figure 14 in terms of
scaled residuals. The average number of iterations for such cases is about 1300 iterations.
Figure 14. Scaled residuals for 304.8 mm (12.0 in.) Diameter Double Elbow U-
Configuration Lint = 2.52 m (8.28 ft)
Contours of static pressure (in Pa) are shown by Figure 15. The effect of the two
elbows on the static pressure is clearly shown, where the static pressure is increasing at
the outer radii and decreasing at the inner radii. The opposite is predicted for the dynamic
pressure (Figure 16), where the dynamic pressure is decreasing at the outer radii and
33
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
Figure 15. Contours of static pressure (in Pa) for 304.8 mm (12.0 in.) Diameter
Double Elbow U-Configuration Lint = 2.52 m (8.28 ft)
Figure 16. Contours of dynamic pressure (in Pa) for 304.8 mm (12.0 in.) Diameter
Double Elbow U-Configuration Lint = 2.52 m (8.28 ft)
34
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
Figure 17. Comparison of total pressure loss vs velocity pressure for 304.8 mm (12.0
in.) Diameter Double Elbow U-Configuration Lint = 2.52 m (8.28 ft)
The error in the loss coefficient (c) is defined as follows: Error = (Experimental –
CFD) / Experimental. This error is provided in Table 7 for this case where it was found
35
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
Table 7. Error for 308.4 mm (12.0 in.) Diameter Double Elbow Loss Coefficient: U-
Configuration Lint = 2.52 m (8.28 ft)
pv8 Re8 pf,7-8 pt,1-2 Error
C
Pa Pa/m Pa %
146.9 304346 8.6 38.0 0.259 -8.88
265.5 411320 15.4 64.0 0.241 -9.42
428.0 516918 24.1 96.0 0.224 -9.76
197.0 347458 11.2 51.0 0.259 -8.21
376.0 481710 21.1 90.0 0.239 -9.89
64.3 197017 3.8 18.8 0.292 0.73
115.0 264535 6.6 31.7 0.276 0.04
180.6 332431 10.3 47.7 0.264 -0.30
217.7 365916 12.4 56.4 0.259 -0.72
36.2 149418 2.1 11.4 0.315 3.06
99.9 250579 6.0 28.5 0.285 -3.44
The k- CFD velocity vectors near the two elbows of the U-Configuration 12 in
duct fitting are shown by Figure 18. Flow separation can be seen clearly in the vicinity of
the elbows. However, whether this prediction of the flow separation is sufficient to be
compared to the experimental velocity profiles is another question. The velocity profile is
shown by Figure 19 as a function of the separation distance location. It is noted that close
to the elbows, there is a y velocity component, which means there is a flow separation.
The experimental velocity profiles were compared to CFD velocity profiles for
x/d values of 1, 7 and 9 as shown in Figures 20, 21 and 22, respectively. It is found that
CFD k-model is not capable of predicting the velocity profiles accurately, especially in
the vicinity of the elbows. This is because of the assumptions involved in the turbulence
model, which don’t account for curvature and flow separation accurately.
36
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
Figure 18. CFD velocity vectors for 304.8 mm (12.0 in.) Diameter Double Elbow U-
Configuration Lint = 2.52 m (8.28 ft)
37
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
Figure 19. CFD velocity profile for 304.8 mm (12.0 in.) Diameter Double Elbow U-
Configuration Lint = 2.52 m (8.28 ft)
East West
0.80
North South
0.60 CFD-East West
0.40
0.20
r/R
0.00
0.00 0.20 0.40 0.60 0.80 1.00 1.20 1.40 1.60
-0.20
-0.40
-0.60
-0.80
-1.00
V/Vr=0
Figure 20. Comparison of experimental and CFD velocity profile at x/d = 1 for 304.8
mm (12.0 in.) Diameter Double Elbow U-Configuration Lint = 2.52 m (8.28 ft)
38
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
LoD=10 Position 7 Velocity Profile
1.00
East West
0.80
North South
0.60
CFD-East West
0.40
0.20
0.00
r/R
-0.40
-0.60
-0.80
-1.00
V/Vr=0
Figure 21. Comparison of experimental and CFD velocity profile at x/d = 7 for 304.8
mm (12.0 in.) Diameter Double Elbow U-Configuration Lint = 2.52 m (8.28 ft)
0.40
0.20
r/R
0.00
0.00 0.20 0.40 0.60 0.80 1.00 1.20 1.40 1.60
-0.20
-0.40
-0.60
-0.80
-1.00
V/Vr=0
Figure 22. Comparison of experimental and CFD velocity profile at x/d = 9 for 304.8
mm (12.0 in.) Diameter Double Elbow U-Configuration Lint = 2.52 m (8.28 ft)
39
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
CFD predictions for U-Configuration for 8 in diameter with short separation
distance of LoD = 4
Figure 23 shows comparison between experimental and k- CFD total pressure loss vs
velocity pressure. The maximum error for this case is about 9% as shown in Table 8.
180
160
Experimental
140 CFD
Linear (Experimental)
120 Linear (CFD)
Total Pressure Loss (Pa)
100
80
60
40
20
0
0 50 100 150 200 250 300 350 400 450 500
Velocity Pressure (Pa)
Figure 23. Comparison of total pressure loss vs velocity pressure for 203 mm (8.0
in.) Diameter Double Elbow U-Configuration Lint = 1.304 m (4.28 ft)
40
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
Table 8. Error for 203 mm (8.0 in.) Diameter Double Elbow Loss Coefficient: U-
Configuration Lint = 1.304 m (4.28 ft)
Re8 Pv8 Pt,1-2 Error
C
Pa Pa %
225883 165.89 53.14 0.32 -6.65
263625 227.60 72.01 0.32 -7.04
301890 296.10 91.70 0.31 -7.42
338994 374.54 116.16 0.31 -8.38
376285 479.00 145.60 0.30 -8.75
200805 135.40 43.78 0.32 -6.43
269187 240.60 75.70 0.31 -7.14
337239 379.00 115.40 0.30 -7.53
104309 36.09 14.68 0.41 -5.56
140529 65.50 25.30 0.39 -6.97
function of Re. The maximum error for this case is about 11%.
Re8 C Error
%
301004 0.21 6.81
404439 0.20 3.27
507307 0.19 -4.76
341181 0.21 1.46
471027 0.19 0.42
193871 0.24 2.95
259835 0.22 7.30
324139 0.21 10.99
356805 0.20 9.18
145584 0.26 11.70
243354 0.23 10.12
Figure 24. Comparison of loss coefficient for 304.8 mm (12.0 in.) Diameter Double
Elbow U-Configuration Lint = 1.304 m (4.28 ft)
41
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
CFD predictions for Z-Configuration for 12 in diameter with long separation
distance of LoD = 10
Figure 25 shows comparison between experimental and k- CFD total pressure
loss vs velocity pressure. The maximum error for this case is 7% as shown in Table 9.
Figure 25. Comparison of total pressure loss vs velocity pressure for 304.8 mm (12.0
in.) Diameter Double Elbow Z-Configuration (LoD=10)
42
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
Table 9. Error of loss coefficient for 304.8 mm (12.0 in.) Diameter Double Elbow Z-
Configuration (LoD=10)
V8 pv8 Re8 pf,7-8 pt,1-2 Error
C
m/s Pa Pa/m Pa %
15.9 151.7 330599 9.0 37.0 0.244 -3.12
21.5 274.9 442178 16.0 63.0 0.229 -5.91
27.0 433.4 553526 25.0 94.0 0.217 -6.54
18.2 197.7 374064 11.6 47.0 0.238 -5.89
25.3 378.8 515830 21.9 84.0 0.222 -3.71
10.3 63.5 212150 3.9 17.5 0.275 1.89
13.9 114.8 284791 6.9 29.0 0.253 6.76
17.5 181.0 356390 10.7 43.5 0.240 6.57
19.3 219.0 390952 12.9 52.0 0.237 7.65
7.9 36.6 160217 2.1 10.8 0.294 8.22
13.2 103.1 270303 6.2 26.5 0.257 4.31
The k- CFD velocity vectors near the two elbows of the Z-Configuration 12 in
duct fitting are shown by Figure 26. Flow separation can be seen more clearly than in the
U-configuration in the vicinity of the elbows. The velocity profile is shown by Figure 27
that close to the elbows, there is a y velocity component, which means there is a flow
separation.
43
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
Figure 26. CFD velocity vectors for 304.8 mm (12.0 in.) Diameter Double Elbow Z-
Configuration Lint = 2.52 m (8.28 ft)
44
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
Figure 27. CFD velocity profile for 304.8 mm (12.0 in.) Diameter Double Elbow Z-
Configuration Lint = 2.52 m (8.28 ft)
Comparison between CFD and experimental results: Maximum Error for all cases
of Z- and U-Configurations
Comparison of CFD RSM, k- and k- turbulence model predictions to
experimental results for several Lint lengths are provided in Table 10 in terms of the
maximum percent error. These results demonstrate that the RSM turbulence model
yielded more accurate predictions than k- or k- turbulence models, and that the k-
model outperformed the k- model in most cases. Therefore it is recommended to use the
the k- model can be used for Z- and U- configurations in cases where slightly higher
error is acceptable.
45
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
Table 10. Maximum % Error - Experimental Data vs. Different CFD Models
46
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
DISCUSSION
The results of the present study showed the advantages of RSM relative to the
two-equation models. This is mainly due to the assumptions involved in closing the two-
equation models. While the two-equation models are complete in that no new information
is needed, they are to some degree limited to flows that do not depart significantly from
turbulent production and dissipation given by the k-ε equation are equal, which implies
that the scales of turbulence are locally proportional to the scales of the mean flow.
Therefore the two-equation models predicted incorrect trends when applied to non-
equilibrium flow studied in this report. The local equilibrium assumption follows from
the fact that the Reynolds stresses must be estimated at every point in the flow field.
Hence, the eddy viscosity is defined to be the proportional constant between the Reynolds
stresses and the mean strain rate. This is the essence of the Boussinesq hypothesis. Since
the turbulence and mean scales are proportional, the eddy viscosity can be estimated
based on dimensional reasoning by using either the turbulent or mean scales. Thus, for
the k-ε model t k 2 and for the k-ω model t k . When production does not
balance dissipation, as in the cases studied in this report, then the ratio of Reynolds
stresses to the mean strain rate is not a local constant and should be a function of both
turbulent and mean scales. Based on local equilibrium assumption, while the transport
effects are included for turbulent scales they are neglected for turbulent Reynolds
stresses. In fact Reynolds stresses depend on the local conditions with some history
effects. RSM accounts for these history effects, while two-equation models do not. The
flow inside the elbow(s) cannot be assumed in local equilibrium because the flow is three
47
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
dimensional, with sudden changes in mean strain rate, as well as the presence of
secondary flows, separation, and fluid rotation. Hence the two-equation models failed in
this study.
fluctuations, u, v, w are locally isotropic or equal. This is true of the smallest eddies at
high Re, but the large eddies are in a state of steady anisotropy due to the strain rate of
the mean flow. This assumption results in equal normal stresses at a point in the flow
field. For the cases studied in this report, RSM predictions showed that this is not true,
The k-ε equation is based on the ε equation, which actually represents the
mechanism of the smallest eddies that physically accomplishes the dissipation. What is
actually needed in the model is a length or time scale relevant to the large, energy-
containing eddies that are responsible for most of the turbulent stresses and fluxes. This
leads to questions about how relevant the exact dissipation equation is, when the desired
quantity is a length scale, characteristic of the large eddies. The k-ω model solves for
only the rate at which the dissipation occurs. The equation governing ω has traditionally
been formulated based on physical reasoning in light of the processes normally governing
the transport of a scalar in a fluid. Usually the coefficients required for CFD model
closure are determined by setting their values such that the model obtains reasonable
for one application are not necessarily suitable for dissimilar cases.
RSM with enhanced wall treatment that resolves the sublayer region attempts to
account for history effects, as the Reynolds stress equation includes convection and
48
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
diffusion terms for the stresses and eliminates the need for equilibrium assumption and
local isotropy. Also, the Reynolds stress equations include production and body force
terms that can respond to the effects of the streamline curvature, rotation, and buoyancy.
The two-equation models need corrections for streamline curvature, buoyancy and
predictions.
49
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
SUMMARY OF CAPABILITIES AND LIMITATIONS OF TURBULENCE
MODELS
in duct fittings is limited by the assumptions and the class of flow configuration, besides
other factors that these models were developed for. In this section, suggestions and a
50
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
High Re model
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
4 additional transport equations are required in 2D flows and 7 in 3D.
Accounts for the effects of streamline curvature, swirl, rotation, and rapid
changes in strain rate
limited by the closure assumptions employed to model various terms in
the exact transport equations for the Reynolds stresses.
52
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
SUMMARY OF COMPUTATIONAL EFFORT AND SOLUTION BEHAVIOR
o The realizable k- model requires slightly more computational effort than the
standard k- model.
o RNG k- model takes 10 to 15% more CPU time than with the standard k-
o The k- models require about the same computational effort.
o RSM requires about 50 to 60% more CPU time per iteration compared to the k-
and k- models. 15 to 20% more memory is needed.
o RSM takes more iterations to converge
53
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
SUMMARY OF GRID CONSIDERATIONS FOR TURBULENT FLOW
SIMULATION
In this section, some useful grid considerations for CFD flow simulations in duct
It is recommended to use fine meshes for the regions where the mean flow
changes rapidly and there are shear layers with a large mean rate of strain.
We can check the near-wall mesh by plotting the values of y+, and Rey.
y+ and Rey are not fixed geometrical quantities. They are solution-dependent. For
example, when doubling the mesh, the new y+ does not necessarily become half
of the y+ for the original mesh.
considering the range over which the log-law is valid. The distance is usually measured
54
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
At least 10 cells needed within the viscosity-affected near wall region (Rey <
200) to be able to resolve the mean velocity and turbulent quantities in that
region. (Rey = y k1/2 / )
55
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
CONCLUSIONS
Based on the CFD results obtained using different turbulence models provided in
For smooth duct walls, use the k- turbulence model for high Re (Re above
100,000) flow in straight ducts with enhanced wall treatment grid requirements
(y+ = 3 to 4).
For rough duct walls, use the k- turbulence model for high Re (Re above
100,000) flow in straight ducts with wall functions grid requirements (y+ = 25 to
30)
For rough duct walls of all geometries, always use wall functions grid
requirements (y+ = 25 to 30). Enhanced near wall treatment can’t be used for
rough walls.
For single elbow and double elbow ducts, use the RSM turbulence model
is concluded:
models need a length scale correction and corrections to account for streamline curvature.
The standard wall functions do not yield accurate results when applied under conditions
different from those under which the law of the wall is derived. The velocity profile near
the wall is altered by the pressure gradients and non-equilibrium flow. RSM (with
56
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
enhanced wall treatment that resolves the sublayer region) accounts for history effects, as
the Reynolds stress equation includes convection and diffusion terms for the stresses and
eliminate the need for equilibrium assumption and local isotropy. Also, the Reynolds
stress equations include production and body force terms that can respond to the effects
of the streamline curvature. Hence, RSM predictions were superior to the two-equation
57
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
SUMMARY OF GENERAL RECOMMENDATIONS
It is not recommended to use realizable k- model for duct flow simulation
because it produces non-physical turbulent viscosities due to the fact that the
model includes the effects of mean rotation in the definition of the turbulent
viscosity.
It is not recommended to use k- turbulence models for duct flow simulations
effects, compressibility, and shear flow spreading, which makes them more
appropriate for far wakes, mixing layers, and plane, round, and radial jets.
The RSM might not always yield results that are clearly superior to the simpler
However, use of the RSM is a must when the flow features of interest are the
flows in ducts.
58
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
SUMMARY OF RECOMMENDATIONS FOR DUCT FITTINGS
Use k- model with wall function wall treatment to predict pressure loss within
15% error
Numerical grid: structured with at least 31 grid points in radial direction and
Aspect Ratio of not more than 6 in longitudinal direction. Finer mesh is needed
for high Re
In case velocity profiles need to be predicted, use RSM with enhanced near-wall
treatment grid
accurately.
59
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
REFERENCES
Engineers, Inc.
Flow Resistance of HVAC Ducts and Fittings. Atlanta: American Society of Heating,
3. ASHRAE. 2015. Duct Fitting Database Version 6.00.04. Atlanta: American Society
4. Gan, G. and S.B. Riffat. 1995. K-Factors for HVAC Ducts: Numerical and
Technology, 16(3):133-139.
5. Gibbs, D.C. and S. Idem. 2012. Measurements of Flat Oval Diverging-Flow Fitting
6. Gutovic, M., H. Lulic, and E. Sirbubalo. 2013. CFD Analysis of Pressure Losses in
7. Idem, S. and F. Khodabakhsh. 1999. Influence of Area Ratio on Flat Oval Divided
8. Idem, S. 2003. Main Loss Coefficients for Flat Oval Tees and Laterals. ASHRAE
Transactions 109(1):456-461.
60
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
9. Kulkarni, D., J. Cui and S. Idem. 2011. Laboratory Testing of Converging Flow Flat
17(5):710-725.
10. Kulkarni, D., S. Khaire and S. Idem. 2009. Measurements of Flat Oval Elbow Loss
11. Liu, W., Z. Long and Q. Chen. 2012. A Procedure for Predicting Pressure Loss
Research, 18(6):1168-1181.
12. Mahank, T.A. and S.A. Mumma. 1997. Flow Modeling of Flat Oval Ductwork
177.
13. Manning, A., J. Wilson, N. Hanlon, and T. Mikjaniec. 2013. Prediction of Duct
411.
14. Mumma, S.A., T.A. Mahank, and Y. P. Ke. 1997. Close Coupled Ductwork Fitting
15. Mumma, S.A., T.A. Mahank, and Y. P. Ke. 1998. Analytical Determination of Duct
16. Mylaram, N.K. and S. Idem. 2005. Pressure Loss Coefficient Measurements of Two
17. Rahmeyer, J. 2002. Pressure Loss Coefficients for Close-Coupled Pipe Ells.
61
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
18. RP-551. 1991. Laboratory Study to Determine Flow Resistance of HVAC Duct
19. RP-690. 1994. Laboratory Study to Determine Flow Resistance of Oval Ducts and
20. RP-1319. 2008. Laboratory Testing of Duct Fittings (Flat Oval Elbows) to Determine
Loss Coefficients.
21. RP-1488. 2011. Laboratory Testing of Flat Oval Fittings to Determine Loss
22. RP-1493. 2011. CFD Shootout Contest – Prediction of Duct Fitting Losses. Final
Report.
23. RP-1606. 2014. Laboratory Testing of Flat Oval Transitions to Determine Loss
24. RP-1682. 2015. Study to Identify CFD Models for Use in Determining HVAC Duct
25. Sami, S. and J. Cui. 2003. Numerical Study of Pressure Losses in Close-Coupled
26. Sleiti, A., J. Zhai, and S. Idem.2013. Computational Fluid Dynamics to Predict Duct
62
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
29. Townsend, B., F. Khodabakhsh, and S. Idem.1996b. Loss Coefficient Measurements
63
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
APPENDIX A
Turbulence Models
Governing Equations
Computational approach
64
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
k- models
65
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
How different terms are modeled for k- models
Details on how different terms are modeled for k- models are provided in Table A2
66
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
Reynolds Stress Model (RSM)
67
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
68
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
APPENDIX B
The k- models were designed to be applied throughout the boundary layer,
provided that the near-wall mesh resolution is sufficient, see Figure B.1.
69
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.
Wall Functions vs. Near-Wall Model
Comparison between wall functions and enhanced near wall treatment approaches is
shown by Figure B.2
70
This file is licensd to Tsz Hung Law (simoncarte31@gl.) This downla was made avilbe with suport from Rital. Copyright ASHRE 20.