VOLUME 11, NUMBER 1 HVAC&R RESEARCH JANUARY 2005

45
Application of CFD Models to Two-Phase Flow
in Refrigerant Distributors
Gang Li Steven Frankel, PhD
James E. Braun, PhD Eckhard A. Groll, PhD
Member ASHRAE Member ASHRAE
Received October 17, 2002; accepted June 10, 2004
The goal of the work described in this paper was to identify an appropriate CFD model for
refrigerant distributors and to apply the model to develop improved designs. There appears to
have been no previous work applying CFD to two-phase flow in refrigerant distributors. As a
first step, results of predictions using different two-phase modeling approaches available within
commercial CFD codes were compared with experimental results from the literature for
two-phase flow distribution with air and water as working fluids. The volume of fluid (VOF) and
algebraic slip mixture (ASM) models failed to predict the experimental data available, whereas
the inter-phase slip algorithm (IPSA) model predictions were in close agreement with the mea-
surements in all cases. However, for a typical refrigerant distributor geometry and set of oper-
ating conditions, predictions of two-phase distribution and separation were quite similar for the
different modeling approaches. As a result, the ASM is applicable for studying refrigerant dis-
tributor designs and was used to evaluate the performance of both existing and improved
designs. In general, it is better to utilize a spherical distributor base as compared with other
shapes and to locate the orifice close to the distributor base. These changes tend to improve the
robustness of the distributor in terms of providing even flow and phase distribution in different
outlet branches when the orifice and/or distributor are not oriented optimally. In addition,
experiments were performed that tend to validate the general trends associated with the CFD
results for refrigerant distributions.
INTRODUCTION
Figure 1 depicts a vapor compression cycle with a refrigerant flow distributor. Following the
expansion device, two-phase refrigerant enters the distribution device and is divided into multi-
ple flow circuits that feed the evaporator. Ideally, the distributor should provide equal mass flow
rates and qualities to each flow circuit of the evaporator. However, it is difficult to design and
manufacture a refrigerant flow distributor that will split a two-phase refrigerant mixture evenly
between different outlet branches. The distribution of liquid and vapor phases is very sensitive
to the conditions and orientation of the flow entering the distributor and to the orientation of the
distributor.
Very little has been published regarding the analysis of refrigerant flow distributors.
Nakayama et al. (2000) did experimental investigations involving a refrigerant distributor and
proposed a new distributor that it has claimed had better performance with respect to flow distri-
bution. The improved design utilized a capillary mixing space rather than an orifice. This design
Gang Li is an engineer at Hayward Pool Products, Clemmons, NC. Steven Frankel and James E. Braun are professors
and Eckhard A. Groll is an associate professor at Purdue University, Ray W. Herrick Laboratories, West Lafayette, IN.
46 HVAC&R RESEARCH
resulted in much more even flow rates for the individual branches within the refrigerant distrib-
utor.
The design of refrigerant flow distributors is typically a trial-and-error process. Modifications
are proposed, a prototype is built, and the device is tested. An alternative approach, described in
this paper, is to use CFD modeling for the problem of flow and phase distribution in refrigerant
distributors. Often, a removable orifice is located within the base of the distributor. The refriger-
ant flows through the orifice, expands from a liquid to a two-phase refrigerant mixture into the
base of the distributor, and then branches into the different (four shown in Figure 1) feeder tubes
leading to the evaporator. A vertically installed distributor with no manufacturing defects and
with symmetrically oriented branches would produce even flow and phase distributions in each
branch. However, it is difficult to orient the orifice perfectly and, as a result, the refrigerant will
exit the orifice in a direction that is not along the centerline of the distributor. In addition, grav-
ity effects will affect the flow and phase distributions when the distributor is oriented horizon-
tally. However, the impact of these imperfections can be reduced through proper design of the
base of the distributor.
The flow through a refrigerant flow distributor involves complicated phenomena of phase
separation and distribution. If the flow were fully mixed and the volume fractions of each phase
were the same everywhere, then the flow would be homogeneous and could be modeled as a sin-
gle-phase problem. However, the volume fractions of each phase are usually not uniform inside
the flow distributor, and multiphase models are necessary to simulate the phase distribution and
separation phenomena. Several CFD codes have been successfully used to predict void fraction
profiles in phase separation and phase distribution applications. However, they have been prima-
rily applied to two-dimensional geometries with air-water mixtures. There appears to be no
application of CFD modeling to refrigerant distributors in the literature.
Figure 1. Vapor compression cycle with refrigerant distributor.
VOLUME 11, NUMBER 1, JANUARY 2005 47
Boisson and Malin (1996) applied a commercial CFD code to predict two-phase flow in bub-
ble columns. An inter-phase slip algorithm (IPSA) was incorporated with the interfacial lift,
drag, and virtual mass forces in this two-fluid model. The predictions were compared with the
measured circulation patterns and void fraction distributions. It was shown that the model can
predict void fraction peaking near the wall for an air-water upflow case and void coring (i.e.,
void fraction concentration near the pipes centerline) for a bubbly column case. The interfacial
lift and virtual mass forces were believed to cause these lateral phase distribution profiles. Class
et al. (1991) and Liu (1991) pointed out that the bubble size effects are also very important. The
importance of bubble size on lateral void phase distribution has also been recognized by other
investigators, including Sekoguchi et al. (1982). Moreover, Monji and Matsui (1991) found that
for bubbly air-water upflows, small bubbles tend to be uniformly distributed across the conduits,
while large bubbles tend to core, and only the intermediate size bubbles concentrate near the
wall of the conduit. Lahey et al. (1993) investigated the phase distribution in complex geometry
conduits, such as triangular and rectangular pipes, and found the void coring and peaking phe-
nomena occurred as well.
Lahey (1990) investigated air-water flow through tee and wye junctions and observed the
phase separation phenomena. It was explained that the vapor phase has far less axial inertia than
the liquid phase; thus, the vapor can be expected to more easily turn the corner into the side
branch.
Two alternatives to the IPSA model, commonly employed in commercial CFD codes, are the
algebraic slip mixture (ASM) and the volume of fluid (VOF) models. These models require sig-
nificantly less computation than the IPSA model, which would be an advantage for application
to design. Furthermore, it is not clear that it is necessary to consider interfacial lift forces for
flow conditions encountered in refrigerant distributors. One of the goals of the current work was
to compare predictions of ASM, VOF, and IPSA models to data available in the literature for air
and water mixtures. This allowed general model validation. Since detailed flow and phase distri-
bution data are not available for refrigerant distributors, comparisons between the ASM and
IPSA models were performed for this application in order to highlight any differences. The goal
was to identify an appropriate model for the refrigerant distributors that could be used to
develop improved designs. FLUENT 5.4 was employed for CFD simulations involving the
ASM and VOF models, whereas PHOENICS 3.3 was used for the IPSA model results.
CFD was then applied to evaluate the performance of both existing and improved designs for
refrigerant distributors. In addition, experiments were performed to validate the general trends
associated with the CFD simulations in terms of flow maldistributions. More details on the mod-
eling and experiments are provided by Li et al. (2001).
CFD MODELING
Due to the complexity involved in two-phase flow modeling, most analytical models avail-
able in the literature have been developed primarily for one-dimensional phenomena (Peixoto
and Bullard 1994; Fiorelli et al. 1998). The most sophisticated of these models are based on a
two-fluid modeling approach. The two-fluid modeling approach has been extended to multidi-
mensional flows (Lahey 1990) and has been adopted by several state-of-the-art multi-dimen-
sional computer codes. Throughout the current study, an adiabatic flow was assumed with no
phase change. A description of the basic modeling approaches considered in this study follows.
Multi-Phase Mass and Momentum Equations
In the absence of phase change within the domain, conservation of mass for each phase leads
to
48 HVAC&R RESEARCH
, (1)
where for phase i,
i
is the volume fraction, is the velocity vector, and
i
is the density.
For each phase, the momentum conservation equation is
, (2)
where P is pressure, which is the same for each phase; is a unit vector in the l-direction; u
il
is
velocity of phase i in direction l; B
il
is the l-direction body force per unit volume of phase i; F
il
is
the friction force exerted on phase i by stresses within that phase; and I
il
is the momentum trans-
fer to phase i resulting from interaction with other phases occupying the same cell. For
steady-state problems, the time-dependent terms disappear in the mass conservation and
momentum conservation equations. For
i
= 1, the equations are applicable to single-phase flow
problems. In this case, the I
il
term vanishes from the momentum equation.
Volume of Fluid Model
In the volume of fluid (VOF) formulation, the different phases assume the same local velocity
and pressure, and properties represent volume-averaged values determined from the local vol-
ume fraction of each of the phases. The densities of the individual phases are assumed to be con-
stant and only a single mass conservation is applied to the vapor phase, such that Equation 1
reduces to
, (3)
where
g
is the volume fraction of the vapor phase. The volume fraction of the liquid phase is
determined from the constraint that the sum of the individual volume fractions is unity.
For volume-averaged properties, Equation 2 reduces to
, (4)
where F
l
is friction force per unit volume exerted by stresses within that location. The average
properties of density, , and viscosity, , are determined from the volume fractions of all phases.
Since only a single momentum equation is solved, there are no interfacial forces between phases
that are considered in the VOF model. More details on the volume of fluid model can be found
in the FLUENT users manual (FLUENT 2003).
Algebraic Slip Mixture Model
The algebraic slip mixture (ASM) model allows for different local velocities and properties of
the phases through the use of an algebraic model that relates phase velocities. An overall mass
conservation equation for the ASM model is
, (5)
where
.

t
----
i

i
( ) div
i

i
Vi ( ) + 0 =
Vi

t
----
i

i
u
il
( ) div
i

i
Viu
il
( ) +
i
l P B
il
+ ( ) F
il
I
il
+ + =
l

t
----
g
( ) div
g
V ( ) + 0 =

t
---- u
l
( ) div Vul ( ) + l P B
l
+ ( ) F
l
+ =

t
----
m
( ) div
m
V ( ) + 0 =

m

i

i
( )
i 1 =
2

=
VOLUME 11, NUMBER 1, JANUARY 2005 49
Again, the local void fractions must sum to unity.
The momentum equation for the mixture is
, (6)
where is the drift velocity for phase i and is given by
, (7)
and where
. (8)
The relative velocity (also referred to as the slip velocity) is defined as the velocity of the sec-
ondary phase (vapor) relative to the primary-phase (liquid) velocity or
. (9)
In the ASM model, the relative velocity is estimated as
, (10)
where is the secondary phase particles acceleration and is the particulate relaxation
time. The particle acceleration is determined from the solution for the velocity field, whereas the
relaxation time is calculated as
, (11)
where d
2
and
1
are secondary phase particle diameter and primary phase dynamic viscosity,
respectively. The user specifies the particle diameter based upon an understanding of the flow
geometry.
Although the ASM model allows velocity slip between phases, only one set of momentum
equations for the mixture is solved and an additional force term related to the slip velocity is
added to the equation. The above version of ASM is also referred to as the drift-flux model. For
more details on this model, please consult the FLUENT manual (FLUENT 2003).
Inter-Phase Slip Algorithm Model
The inter-phase slip algorithm (IPSA) employs individual momentum equations for each
phase with interfacial lift and virtual mass forces. Equations 1 and 2 are applied with the virtual
mass and interfacial lift forces added into the term I
il
, such that
, (12)
where M
vm
and M
lft
are virtual mass and interfacial forces, respectively. As described in the
FLUENT manual (FLUENT 2003), the
lift forces act on a particle mainly due to velocity gradients
in the primary-phase flow field. The lift force will be more
significant for larger particles... Thus, the inclusion of lift
forces is not appropriate for closely packed particles or for
very small particles.

t
----
i

i
u
il
( ) div
i

i
Viu
il
( ) +
i 1 =
2

l P
i
B
il
F
il
div
i

i
VDi ( ) + + [ ]

+ =
VDi
VDi Vi Vm =
Vm

i
Vi

m
----------------



i 1 =
2

=
V1 2 , V2 V1 =
V1 2 ,
2 1 ,
a =
a
2 1 ,

2 1 ,

2
d
2
2
18
1
------------ =
I
il
M
vm
M
lft
+ =
50 HVAC&R RESEARCH
Mathematical details can be found in the manual. Regarding the virtual mass forces, again quot-
ing the FLUENT manual, the
virtual mass effect occurs when a secondary phase acceler-
ates relative to the primary phase. The inertia of the pri-
mary-phase mass encountered by the accelerating particles
(or droplets or bubbles) exerts a virtual mass force on the
particles... The virtual mass effect is significant when the
secondary phase density is much smaller than the primary
phase density (e.g., for air and water in a bubble column).
The interfacial forces are estimated using empirical relations, and for the mathematical details
the reader is referred to the FLUENT manual.
Turbulence Modeling
The k- model for turbulence was employed in this study for all CFD simulations. The friction
force F
il
involves the quantities of turbulent fluctuation velocities with the form of ,
where and are turbulent fluctuation velocity components of u
il
and u
im
, respectively.
According to Boussinesqs assumption (employed in the k- model), it may be expressed as
, (13)
where is the turbulent kinetic energy,
ij
is 1 for i = j and 0 otherwise, and
t
is
the turbulent viscosity (eddy viscosity), which is related to k and to the rate of dissipation by
. (14)
is a model constant with a value of 0.09 in the k- model.
The eddy viscosity, together with molecular viscosity, gives an effective viscosity as
. (15)
The standard k- model is a two-equation turbulent model, which means both k and are
solved from the following transport equations:
(16)
(17)
where C
1
, C
2
,
k
, and

are model constants with values given in Table 1.

u
il
u
im

u
il
u
im

u
il
u
im

t
u
il
x
m
---------
u
im
x
l
----------- +


2
3
---
ij
k =
k
1
2
--- u
il
u
im
( ) =

t
C

k
2

----- =
C

eff

t
+ =

x
il
--------- u
il
k ( )

x
il
---------

k
-----
k
x
il
---------


G
k
G
b
+ + =

x
il
--------- u
il
( )

x
il
---------

k
-----

x
il
---------


C
1

k
--G
k
C
2

2
k
----- + =
VOLUME 11, NUMBER 1, JANUARY 2005 51
G
k
is calculated from
. (18)
Boundary Conditions
All of the geometries simulated in this paper had an inflow boundary, outflow boundaries,
solid walls, and in some cases, a symmetry plane. The inflow boundary had specified inlet
velocity with uniform void fractions. Constant and uniform pressures were specified for all out-
flow boundaries. Although laminar transport is considered in the k- model, it is not applicable
near walls where the turbulence Reynolds number is low. In these regions, equilibrium wall
functions are used in boundary conditions, which are specified at a grid point located in the fully
turbulent region. At this location, the logarithmic law of the wall prevails and the turbulence is
assumed to be in local equilibrium.
MODEL COMPARISONS
As a first step in evaluating the applicability of CFD to two-phase flow in refrigerant distribu-
tors, results of predictions from the ASM, VOF, and IPSA models were compared with experi-
mental results available in the literature. These data are for air and liquid water mixtures having
relatively low inlet velocities (less than 10 m/s). In refrigerant distributors, the velocities exiting
the orifice and entering the distributor are much higher (up to 50 m/s) and the difference in den-
sities between the two phases is much smaller. Modeling of the flow and phase distributions for
the air-water cases is more difficult and therefore the data for these flow geometries allow gen-
eral model validation. Detailed flow and phase distribution data are not available for refrigerant
distributors. However, results from ASM and IPSA were compared with each other for this
application in order to highlight any differences for this case. The goal was to identify an appro-
priate model for the refrigerant distributors that could be used to develop improved designs.
For all of the simulations, a no-slip condition was assumed between the two phases at the inlet
of the computational domain and the standard k- model was employed as the turbulence model.
Air-Water Turbulent Pipe Flow
Experimental measurements for the mean radial profiles of void fraction are available in the
open literature (Seriwaza et al. 1992) for turbulent air-water upflow in a pipe. The Reynolds
number, based on the superficial liquid velocity and the pipe diameter, was 80,000. The inlet
superficial gas and liquid velocities were j
g
= u
g
= 0.077 m/s and j
l
= u
l
(1 ) = 1.36 m/s,
where u
g
, u
l
, and are gas velocity, liquid velocity, and void fraction, respectively.
Two-dimensional, axisymmetric calculations were performed with a computational domain
extending 35 pipe diameters downstream, corresponding to the experimental measuring station
that lies in the region of fully developed flow.
Figure 2 shows comparisons between measurements and CFD predictions of local void frac-
tion as a function of nondimensional radial position from the center within the pipe for one ver-
tical location. The data show void peaking near the wall, whereas the VOF- and ASM-predicted
void fractions are both nearly uniform throughout the pipes cross section. The VOF model does
not consider velocity slip, so a straight pipe with a homogeneous two-phase inlet will have a uni-
Table 1. Constant Values Employed in the k- Turbulence Model
Constant C

C
1
C
2
Value 0.09 1 1.3 1.44 1.92
G
k

t
u
il
x
im
-----------
u
im
x
il
----------- +


u
il
x
im
----------- =
52 HVAC&R RESEARCH
form void fraction distribution in the flow field. The velocity slip scheme in the ASM has the
potential of changing the void fraction profile inside the computational domain. However, there
are no radial forces, such as the interfacial lift force, and the resulting void fraction profile is
also uniform.
Figure 2 also shows that the IPSA predictions of void fraction are in good agreement with the
measurements. The void fraction peaks at the wall, indicating that the gas bubbles have been
driven to the wall by the interfacial lift and virtual mass forces that are incorporated in the IPSA.
The wavy behavior of the void fraction distribution near the pipe wall in the simulations does
not appear to be due to numerical problems. The grid size was reduced with no effect on the
results. In addition, the predicted velocity profile did not exhibit this waviness. The wavy behav-
ior could be real, since there is not enough resolution in the data to capture this effect.
Phase Separation in a Tee Junction
There have been a number of published studies concerned with air-water phase separation in
tee-branching pipes (Collier 1975; Honan and Lahey 1981; Zetzmann 1982). The tee junction
case studied by Honan and Lahey (1981) was chosen to be simulated. The diameters of the inlet
and branch pipes were both equal to 0.038 m. The superficial velocity inlet conditions for the
case considered were j
g
= 10 m/s and j
t
= 1 m/s.
Figure 3 shows the flow configuration in a tee junction and the ratio of branch exit to inlet
quality (x
3
/x
1
) as a function of the ratio of the branch exit to inlet mass flow rate (w
3
/w
1
) for the
experiments and the VOF and ASM model predictions. Over a wide range of mass extraction
ratios (the ratio of the mass flow rate at the exit of the branch pipe to the mass flow rate at the
inlet pipe; i.e., w
3
/w
1
), almost complete phase separation occurred for the experimental data
(i.e., w
3
x
3
= w
1
x
1
). Vapor primarily goes to the side branch and liquid mainly goes to the straight
run. The trends observed in the experiments are attributable to the fact that the vapor phase nor-
mally has a smaller axial momentum than the liquid phase (i.e.,
g
u
g
2
<<
l
u
l
2
). Therefore, the
vapor more easily turns into the side branch. From the data, there is a flow rate ratio that leads to
maximum phase separation.
Both the VOF and ASM models do not predict any phase separation at any flow rate (for
x
3
/x
1
= 1, the flow is homogeneous and the void fraction is uniform throughout). For the VOF
Figure 2. CFD model predictions and measured void fraction radial profiles for Seri-
wazas (1992) bubbly column case.
VOLUME 11, NUMBER 1, JANUARY 2005 53
model, a uniform void fraction distribution results from the lack of a velocity slip and the
homogeneous two-phase inlet. For the ASM, the slip velocity effects were not large enough to
change the void fraction distribution within the branch.
Lahey (1990) used the IPSA with the measured inlet void fraction profile to predict the phase
separation phenomena in two-phase flow through the tee. He found excellent agreement with the
measurements over the entire range of flow rate ratios shown in Figure 3 and was able to predict
the point of maximum phase separation.
Refrigerant Flow Distribution
Interfacial lift and virtual mass forces are important elements in predicting phase separation
and phase distribution phenomena for air-water mixtures at the velocities considered in the liter-
ature. However, refrigerant distributors operate with higher inlet velocities and refrigerant vapor
and liquid properties are much different than those of air and water. As a result, the ASM may
provide accurate predictions of phase separation and distribution for refrigerant distributors.
Since the IPSA has been validated for several flow geometries involving air-water mixtures, it
provides a baseline for evaluating predictions associated with the ASM.
In order to evaluate the application of the ASM to refrigerant distributors, a two-dimen-
sional simulation of the simple distributor geometry shown in Figure 4 was developed. The
geometry and inlet boundary conditions were chosen to be representative of a commercially
Figure 3. Flow configuration and comparison of experimental with ASM and VOF
model predictions for phase distribution in a tee junction (data from Lahey [1990]).
Figure 4. Geometry of the two-dimensional refrigerant distributor.
54 HVAC&R RESEARCH
available distributor with an orifice at the inlet. The properties of the two phases were calcu-
lated from properties of R-22 at 5C. Gravity was not considered. The orifice was assumed to
be tipped slightly in the y direction to simulate manufacturing imperfections and create an
uneven flow distribution. The x and y components of the refrigerant velocities at the inlet were
u
x
= 50.79 m/s and u
y
= 3 m/s.
Table 2 gives comparisons between IPSA and ASM predictions of outlet flow rates and qual-
ities. The predictions of branch liquid and vapor flow rates were very similar for the two codes.
In the addition, the predicted phase distributions and velocity fields were found to be very simi-
lar for the two modeling approaches. It appears that interfacial lift and virtual mass forces are
negligible for this high-momentum two-phase flow.
REFRIGERANT DISTRIBUTOR GEOMETRIES
AND PERFORMANCE INDICES
Table 3 and Figure 5 describe alternative refrigerant distributor geometries that were consid-
ered in this study with R-22 as the working fluid. The figures are drawn to scale and show the
internal volumes where refrigerant flows. Each of the distributors has four branches and an ori-
fice that is located at the centerline of the inlet to the distributor. The Type 1 and Type 3 geome-
tries are commercially available, while the other three incorporate design modifications. In the
Type 1 geometry, the base of the distributor is convex with respect to the flow and comes to a
point. This design was conceived so as to provide a single point of contact for separation of the
flow. The Type 3 design uses a cone-shaped base that is concave with respect to the flow and
was conceived so as to provide a mixing chamber for distribution of refrigerant. Types 2 and 4
are the same as Type 3, except the cone is replaced with flat and spherical surfaces, respectively.
The Type 5 design is a modification of Type 4 where the orifice has been moved closer to the
distributor base and the depth of the chamber has been reduced. For all of the distributors, the
center of the base is on the centerline of the distributor along with the center of the orifice.
Results are presented in terms of uneven flow and quality distribution performance indices.
For a given branch i, the uneven flow and quality indices are
Table 2. Branch Flow Rates and Qualities from IPSA and ASM Models for
Two-Dimensional Refrigerant Distributor
IPSA Model ASM Model
Mass Flow Rate (kg/s) Mass Flow Rate (kg/s)
Vapor Liquid Quality Vapor Liquid Quality
Outlet 1 2.61 12.57 0.161 2.70 14.05 0.161
Outlet 2 1.163 6.025 0.162 1.07 5.55 0.161
Table 3. Description of the Refrigerant Distributors
Type Number Base Characteristics Other Characteristics
Type 1 Sharp-end Commercially available design
Type 2 Flat Same as Type 3 except base surface is flat
Type 3 Cone Commercially available design
Type 4 Spherical Same as Type 3 except base is spherical
Type 5 Spherical Same as Type 4 except orifice position is moved
closer to chamber end; chamber depth is reduced
VOLUME 11, NUMBER 1, JANUARY 2005 55
(19)
(20)
where
m,i
and
x,i
are the indices for uneven flow and quality, and x
i
are individual branch
flow rates and qualities, and and are average mass flow rates and qualities for all four
branches.
For a given distributor, average performance indices for uneven flow and quality are defined
as
(21)
(22)

m i ,
m

i
m

---------------- =
Figure 5. Internal volumes associated with refrigerant distributors described in Table 3.

x i ,
x
i
x
x
------------ =
m

i
m

m
m
i
m

2
i 1 =
4

-------------------------------------- =

x
x
i
x ( )
2
i 1 =
4

x
---------------------------------- =
56 HVAC&R RESEARCH
CFD EVALUATION OF REFRIGERANT DISTRIBUTORS
The inflow boundary had specified inlet velocity with uniform void fractions at conditions
associated with the outlet of the orifice. Constant and uniform pressures were specified for all
outflow boundaries. A three-dimensional incompressible adiabatic flow was assumed with no
phase change. The inlet and outlet pressures, inlet mass flow rate, inlet refrigerant quality, and
refrigerant properties were based upon a 3-ton air-conditioning unit with R-22 as the working
fluid. The refrigerant at the inlet to the distributor was a homogeneous saturated mixture with a
quality of 0.35 and a saturation temperature of 0C. The outlet reference pressure for all
branches was 5 atm, which is close to the evaporating pressure for this specific application.
The goal of the simulations was to evaluate the robustness of each of the distributors with
respect to manufacturing and installation defects. For vertically installed distributors with no
manufacturing defects, all of the distributor designs would produce even flow and phase distri-
butions in each of the branches since the outlet branches are located symmetrically about the
centerline of the distributor. However, it is difficult to orient the orifice perfectly, and as a result
the refrigerant will exit the orifice in a direction that is not along the centerline of the distributor.
To simulate this effect, the direction of refrigerant flow was considered to be off the centerline
by an angle of 3.7. In addition, gravity effects will affect the flow and phase distributions when
the distributor is oriented horizontally. This case was also considered.
As a first step to eliminate the poorer performing devices, single-phase simulations were
performed. Under the assumption of homogeneous equilibrium flow, two-phase flow can be
simplified to single-phase flow with proper averaging methods used for the properties of the
fluids. The density and viscosity for this pseudo-fluid were taken to be 58.34 kg/m
3
and
0.00023 kg/m-s, respectively, corresponding to a saturation temperature of 0C and a quality
of 0.35.
Figure 6 shows flow distribution results for the single-phase simulations applied to distribu-
tor Types 1-4, all oriented upward in the vertical direction with an orifice tilted 3.7 toward
branch 1. The x and y components of the inlet velocity were 50.79 m/s and 3 m/s. For this situ-
ation, the sharp-end (Type 1) distributor performed considerably worse than the other three
designs. The maximum flow rate for the sharp-end distributor occurs in branch 1, which is the
Figure 6. Single-phase simulation results for mass flow distribution in vertically installed
distributors with imperfect orifice orientations.
VOLUME 11, NUMBER 1, JANUARY 2005 57
direction of orientation of the orifice. On the other hand, the blunt-end designs tend to recircu-
late refrigerant prior to distribution so the maximum flow rate occurs in other branches and the
flow rate is more evenly distributed. In general, the sharp-end design is very sensitive to ori-
fice orientation and was not considered in two-phase simulations.
Figures 7 and 8 show flow and quality distribution results for two-phase simulations applied
to distributor Types 2-5, all oriented upward in the vertical direction with an orifice tilted 3.7
toward branch 1. The boundary conditions for two-phase flow were the same as for the sin-
gle-phase simulations, except there was a uniform void fraction specified at the inlet assuming a
no-slip condition. For these blunt-end distributors, the shape of the base plays an important role
in terms of flow recirculation and distribution. Types 2-4 have similar overall performance but
result in different flow and quality distributions for individual branches. The Type 5 distributor
Figure 7. Two-phase simulation results for mass flow distribution in vertically installed
distributors with imperfect orifice orientations.
Figure 8. Two-phase simulation results for quality distribution in vertically installed dis-
tributors with imperfect orifice orientations.
58 HVAC&R RESEARCH
is much less sensitive to orifice orientation and has much more uniform flow and quality among
the different branches.
Figure 9 shows velocity vectors and void fraction contours for a plane that goes through the
center of the distributor, branch 1, and branch 3 for the Type 3 and 5 distributors (oriented
upward in the vertical direction with an orifice tilted 3.7 toward branch 1). For the cone-shaped
base (Type 3), more of the higher momentum liquid refrigerant flows through branch 1 than
branch 3, leading to larger total flow and lower quality (see Figures 7 and 8) through this branch.
However, as shown in Figure 7, the largest flows occur in branches 2 and 4 even though there is
a slight increase in quality of the refrigerant in these branches (Figure 8). The complex recircula-
tion patterns produced by this shape result in relatively asymmetric velocity and phase distribu-
tions in all three dimensions. On the other hand, the Type 5 design (spherical base with a closer
orifice location) results in more symmetric recirculation patterns and velocity and phase distri-
butions.
Figures 10 and 11 show flow and quality distribution results for the distributors mounted in a
horizontal direction with perfect orifice orientations. The flow and quality distributions are
uneven due to the effects of gravity. The spherically shaped bases have more even flow and
phase distributions than the cone and flat bases. In particular, the flat base (Type 2) has much
worse phase distribution than the other designs.
Figure 9. Velocity vectors superimposed on void fraction contours.
Figure 10. Two-phase simulation results for mass flow distribution in horizontally
installed distributors with perfect orifice orientations.
VOLUME 11, NUMBER 1, JANUARY 2005 59
Table 4 summarizes the overall performance of the Type 2-5 distributors in terms of mass
flow and quality distributions at the outlet branches. The Type 5 distributor is much more robust
with respect to imperfections in orifice and distributor installation as compared with the other
designs.
EXPERIMENTAL EVALUATION OF REFRIGERANT DISTRIBUTORS
In an attempt to validate the trends associated with the CFD modeling, experiments were per-
formed for three of the geometries: sharp-end base (Type 1), cone-shaped base (Type 3), and
spherical base with closer orifice location (Type 5). Types 1 and 3 were commercially available,
whereas Type 5 had to be specially built for this project. Different orifices were investigated, but
only the best flow distribution results are presented for each distributor.
Figure 12 shows the experimental test stand that was developed for testing. The inlet pressure
and temperature to the orifice and the pressures at the outlets of the branches were controlled
using adjustable valves. For each test, the pressures at the outlet of each branch were controlled
to the same value. The refrigerant mass flow rate through an individual distributor branch was
estimated from an energy balance on an electrically powered superheater located after the evap-
orator that follows that branch. Measurements of power input and refrigerant inlet and outlet
temperatures and pressure were used along with property data to estimate mass flow rate
through that branch. The quality of the refrigerant leaving each branch was estimated from an
energy balance on the evaporator associated with that branch. The evaporators were
water-cooled and heat transfer rates were determined from water-side flow rates and tempera-
Table 4. CFD Results for Average Two-Phase Unevenness
Distributions for Different Distributors
Horizontal installation Vertical installation with tilted inlet velocity
Type 2 23.48% 13.89% 27.09% 13.48%
Type 3 17.59% 6.77% 48.23% 16.02%
Type 4 2.66% 2.67% 26.87% 11.26%
Type 5 0.91% 2.57% 1.02% 2.25%
m

Figure 11. Two-phase simulation results for quality distribution in horizontally installed
distributors with perfect orifice orientations.
60 HVAC&R RESEARCH
tures. The evaporator inlet enthalpy was determined from a refrigerant-side energy balance
using the water-side heat transfer rate and refrigerant outlet state measurements. The inlet qual-
ity was then determined from refrigerant property data with the estimated inlet enthalpy and
measured pressure.
Based upon an uncertainty analysis, the calculated uncertainty in the mass flow measurements
was in the range of 5% to 10%, whereas the uncertainty in the quality measurements was
between about 10% and 20%. However, the actual uncertainty was undoubtedly much greater
due to additional factors, such as randomness in the orientation of the orifice within the distribu-
tor housing between tests, unmeasured heat gains and losses, unsteady water loop flows, etc. It
was difficult to obtain consistent results, particularly with respect to measurements of outlet
branch qualities. As a result, only overall mass flow rate maldistributions are presented here.
The distributors were tested in a vertical, downward-flowing orientation. Tests were per-
formed at high, medium, and low condenser pressures and repeated twice at each condition. In
addition, each of the tests was repeated with the distributor rotated through all four possible
arrangements of the outlet branches feeding the different evaporator circuits of the test stand.
This was done in order to reduce biases associated with orifice orientation and the test stand.
Average performance indices for uneven flow (Equation 3) was determined for each distributor
by averaging results obtained for all conditions and distributor orientations (24 tests for each
distributor).
Figure 13 shows test results for the different distributors considered. Consistent with the CFD
simulations, the sharp-end distributor had the worst performance of those considered. The per-
formance of the spherical-base distributor was slightly better than the cone-base distributor.
Figure 12. Experimental test stand.
VOLUME 11, NUMBER 1, JANUARY 2005 61
However, the differences were smaller than those determined through simulation. Several fac-
tors could have led to these differences. The actual orientation of the orifice within the distribu-
tor housing was unknown and has a major impact on flow distribution. The experimental results
were averaged for several different tests where the orifice orientation could have changed. The
uncertainty in the flow measurements could be 10% or higher. The simulations assumed a uni-
form void fraction at the outlet of the orifice. The actual void fraction distribution at the orifice
outlet was unknown and a nonuniform distribution would impact the flow distribution at the
branch outlets.
CONCLUSIONS
The VOF and ASM models failed to predict experimental data available in the literature for
two-phase flow distribution and separation with air and liquid water having relatively low inlet
stream momentum. On the other hand, the IPSA predictions were in close agreement with the
measurements in all cases. The IPSA (inter-phase slip algorithm) model solves continuity and
momentum equations for each phase and incorporates interfacial lift and virtual mass forces that
are not present in the VOF and ASM models. However, these forces do not appear to be impor-
tant for the high-momentum inlet flows encountered in refrigerant distributors. For a typical
distributor geometry, predictions of two-phase distribution and separation were quite similar for
the ASM and IPSA. As a result, the ASM is applicable for studying improved refrigerant distrib-
utor designs and was used for this purpose
Experimental results confirmed some general trends arising from CFD simulations of existing
and improved refrigerant distributors. In general, it is better to utilize a spherical base, as com-
pared with other shapes, and to locate the orifice close to distributor base. These changes tend to
improve the robustness of the distributor in terms of providing even flow and phase distribution
in different branches when the orifice and/or distributor are not oriented in an optimal fashion.
Additional detailed experiments should be performed to validate the application of two-phase
CFD models to refrigerant distributions. In particular, it would be useful to obtain data for spa-
tial distributions of liquid and vapor within the distributor and at the outlets and to compare
these data with model predictions.
Figure 13. Comparison of average unevenness flow distributions for different types of
distributors determined from experiments.
62 HVAC&R RESEARCH
ACKNOWLEDGMENTS
The authors appreciate the technical support provided by Jeff Grau and Chris Schadewald
(formerly of Aeroquip Corporation) during the course of this study.
REFERENCES
Boisson, N., and M.R. Malin. 1996. Numerical prediction of two-phase flow in bubble columns. Int J. for
Numerical Methods in Fluids 23:1289-1310.
Class, G., R. Meyder, and W. Sengpiel. 1991. Measurement of spatial gas distribution and turbulence struc-
ture in developing bubbly two-phase in vertical channels. Proceedings of the International Conference
on Multiphase Flows 91Tsububa 1, pp. 473-477.
Collier, J. 1975. Single-phase and two-phase flow behavior in primary circuit components. Symposium on
Two-Phase Flow and Heat Transfer in Water Cooled Nuclear Reactors, Dartmouth University.
Fiorelli, F.A.S., R.A. Peixoto, M.A.S. Paiva, and O.M. Silvaresm. 1998. Numerical study on refrigerant
mixtures flow in capillary tubes. 1998 International Refrigeration Conference at Purdue.
FLUENT Users Manual. 2003.
Honan, T.J., and R.T Lahey, Jr. 1981. The measurement of phase separation in wyes and tees. Nuclear Eng.
& Design 54(1):93-102.
Lahey, R.T., Jr. 1990. The analysis of phase separation phenomena in branch conduits. Dynamics of
Two-Phase Flow, pp.313-346.
Lahey, R.T., Jr., M. Lopez de Bertodano, and O.C. Jones. 1993. Phase distribution in complex geometry
ducts. Nucl. Eng. Design 141, pp. 177-201.
Li, G., J.E. Braun, E.A. Groll, and S. Frankel. 2001. A numerical and experimental investigation of refrig-
erant flow control devices. HL Report #5372-1, Ray W. Herrick Laboratories, Purdue University.
Liu, T.J. 1991. The effect of bubble size on void fraction distribution in a vertical channel. Proceedings of
the International Conference on Multiphase Flows 91-Tsububa 1, pp. 453-457.
Monji, H., and G. Matsui. 1991. Effect of bubble size on the structure of vertical bubble flow. Proceedings
of the International Conference on Multiphase Flow 91-Tsububa, pp. 449-452.
Nakayama, M., Y. Sumida, S. Hirakuni, and A. Mochizuki. 2000. Development of a refrigerant two-phase
flow distributor for a room air conditioner. International Refrigeration Conference at Purdue, pp.
313-319.
Peixoto, R.A., and C.W. Bullard. 1994. A simulation and design model for capillary tube-suction line heat
exchangers. 1994 International Refrigeration Conference at Purdue.
Sekoguchi, K., T. Sato, and T. Honda. 1982. Two-phase flow in vertical noncircular channels. Int. J. Mul-
tiphase Flow 8(6):641-655.
Serizawa, A., I. Kataoka, and I. Michiyoshi. 1992. Phase distribution in bubbly flow: Data set No. 24. Mul-
tiphase Science and Technology 6:257. Washington DC: Hemisphere.
Zetzmann, K. 1982. Phasenseparation und Druckabfall in Zweiphasig Durchstroemten Vertikalen
Rohrabzweigungen. Doctorate thesis, University of Hanover, FRG.