Computational Methods in Multiphase Flow III 251

Modeling multi-phase flow using CFD with

related applications
D. Lin, P. Diwakar, V. Mehrotra, B. Rosendall & J. Berkoe
Bechtel Corporation, USA

Abstract
Concurrent liquid-gas flow manifests itself into different flow regimes depending
on fluid properties, flow rates, heat inputs and geometry. The Baker chart has
identified at least seven flow regimes, viz. stratified, wavy, plug, slug, annular,
dispersed and bubbly flow. Each flow regime exhibits a unique flow pattern and
may transition from one to another depending on the flow conditions in the pipe.
Computational Fluid Dynamics (CFD) with multiphase flow equations add a
capability by which these flow regimes can be visualized and understood. In this
paper stratified and slug flow regimes are studied by using CFD modeling. The
modeling results are validated by comparison with published experimental data.
The benchmark comparison shows that CFD is successful in predicting the
following aspects in two-phase flow:
• Quantitative prediction of stream-wise velocities in both phases.
• Qualitative prediction of flow regime, phase distribution and pressure
distribution.
• Transition between certain, but not all, flow regimes.
On the other hand, more research and development is needed to accurately
predict pressure drop and vapor/liquid interface location when the velocity
difference between phases is relatively large.
The application of the validated CFD model is extended to the same flow
regime in a tee joint distributor system. The results demonstrate that the
mal-distribution of flow can occur in complex piping systems with sudden
transition.
Keywords: validation, transition, stratified flow, slug flow, CFD.

1 Introduction
Two-phase flow is a challenging problem because the interaction of liquid and
vapor phase can generate different flow patterns depending on properties of the

WIT Transactions on Engineering Sciences, Vol 50, © 2005 WIT Press

www.witpress.com, ISSN 1743-3533 (on-line)
252 Computational Methods in Multiphase Flow III

fluids, flow rates, heat inputs and geometry. Baker [1] has identified seven flow
regimes for concurrent flows in a horizontal pipe: stratified, wavy, plug, slug,
annular, dispersed, and bubbly flow. Some of the flow regimes are transient and
migrate from one regime to another. Slug flow is intrinsically unsteady and is
characterized by a succession of slugs of liquid separated by large gas bubbles
whose diameter approaches that of a pipe. It is considered one of the most
challenging flow types to model because of its intermittent nature and lack of
understanding in the mechanism. Stratified flow is characterized by liquid
running at the bottom and vapor at the top of a pipe due to gravity effects. It is
considered the simplest flow regime. This study focuses on modeling these two
flow regimes.
CFD employs different mechanistic models to predict different flow regimes.
It is very important to choose the correct modeling approach that captures the
physics of the application. The best mechanistic multi-phase CFD model for
predicting the behavior in stratified and slug flows is the VOF (Volume of Fluid)
approach (Fluent Manual). The Eulerian-Granular model can successfully
predict the level of solid-liquid mixing when compared to experimental tests
(Rosendall and Berkoe [5]).
Due to the complex nature of two-phase flows, it is necessary to validate the
CFD model with experimental data before applying the analysis to more
complex systems. Validation with experimental results for stratified and slug
flow regime is covered in Section 3. More complex geometry and some results
(flow pattern, velocity and pressure) are presented in Section 4. Finally, some
concluding remarks about using CFD to two-phase flow problems are presented
in Section 5.

2 CFD multiphase model description

2.1 Governing equations

The Volume of Fluid (VOF) model is a surface-tracking technique applied to a

fixed Eulerian mesh. It is designed for two or more immiscible fluids where the
position of the interface between the fluids is of interest. In addition to solving
the continuity and momentum equations, VOF model solves one additional set of
equation for the volume fractions. These equations are listed below (Fluent
Manual):

∂ρ G
+ ∇ • ( ρv ) = 0
∂t
∂ G GG G G G
( ρv ) + ∇ • ( ρv v ) = −∇p + ∇ • [ µ (∇v + ∇v T )] + ρg
∂t
∂α q G
+ v • ∇α q = 0
∂t

WIT Transactions on Engineering Sciences, Vol 50, © 2005 WIT Press

www.witpress.com, ISSN 1743-3533 (on-line)
 Computational Methods in Multiphase Flow III 253
G
where ρ = density for each phase, v = velocity vector that is shared for all the
phases, µ = viscosity of each phase, g = gravitation = 9.81 m/s2, and αq = the
volume fraction for the qth phase of fluid.

2.2 Assumptions

The following assumptions were made for all calculations in this study:
• No phase changes were considered in liquid and vapor.
• Both phases were considered incompressible. The compressibility of
the gas phase is not expected to have a significant effect on the results.
• The air and water were assumed of constant density and viscosity.
• The surface tension for the air-water interaction was assumed constant.

3 Validation of CFD Model for various flow regimes

3.1 Slug flow validation

The CFD results presented in this section are compared with experimental results
on slug flow in a horizontal 2” pipe (Lewis et al. [2]). This paper provides
average velocity profile measurements for comparison.
In the initial CFD testing, it was observed that the generation of small bubbles
and the latter regrouping phenomenon are very important in predicting slug flow.
These require:
1. High velocity differential and mixing (between vapor and liquid) at the inlet.
This can be achieved by using very small injection or impingement (on fences
that provide sufficient restriction), which enhances liquid and vapor mixing
and ensures the generation of small bubbles.
2. A fine grid throughout so that shear stress is not smeared or smoothed out.
This will ensure proper break up and coalesce of bubbles downstream of the
inlet.

3.1.1 Geometry, mesh and boundary conditions:

To create a 3D pipe flow model of 200 diameters long with a vapor injection
zone of 100mm pores will require millions of calculation cells and take a long
time to converge. Based on this consideration, a 2D CFD model was used for
this study.
Nine air and water inlets were created along with slots downstream of the
inlets to enhance mixing. The fence also creates vapor bubbles in the pipe. It is
expected that the simplified CFD model may have a different transition length
for a fully developed slug flow and the different pressure drop through the pipe
length. These two aspects were not investigated in the experiment, however.
The VOF time-dependent segregated solver was used for tracking the break up
and coalesces of the vapor bubbles. For a slug flow, the larger bubbles are of
interest.
The air inlet velocity is 3.3 m/s, and the water inlet velocity was 2.5 m/s.
This translates to air superficial velocity of 1.1 m/s and water superficial velocity

WIT Transactions on Engineering Sciences, Vol 50, © 2005 WIT Press

www.witpress.com, ISSN 1743-3533 (on-line)
254 Computational Methods in Multiphase Flow III

of 1.65 m/s, which corresponds to case (b) in Lewis et al. [2]. The average air
volume fraction was calculated to be 0.4. The outlet was specified as outflow
boundary condition. The walls were specified as stationary walls. The
maximum aspect ratio of the computational cells was about 2 and the total grid
was about 210,000 in this calculation.

3.1.2 Analysis and discussion

3.1.2.1 Generation of air and water bubbles The formation of air bubbles is
represented by contours of volume fraction as illustrated in Fig. 1. Red color
indicates air and blue represents water. As can be seen, air bubbles of different
sizes were generated immediately downstream of the slots. Some air
accumulated and broke off behind the slots. As the air bubbles traveled
downstream, they coalesced and broke again. Influenced by gravity, air bubbles
rose up to the top portion of the pipe and water settled towards the bottom of the
pipe. Some eventually formed bigger bubbles that resembled the flow pattern
shown in Fig. 2.

Air Inlet
slots
Flow
2 in
direction

Water Inlet walls

200 in.

Figure 1: Slug flow creation using slotted inlets and fences.

Figure 2: Tracking large bubbles in slug flow.

3.1.2.2 Slug flow downstream of developing zone Fig. 3 shows a snap shot of
air and water distribution in the whole length of pipe. Fig. 3(a) shows the first
50 diameters of pipe and 3(b) from 50 to 100 diameters. Starting from about 40
diameters downstream, the air accumulates at the top of the pipe. The water
exhibited wavy pattern with small air bubbles trapped inside.

WIT Transactions on Engineering Sciences, Vol 50, © 2005 WIT Press

www.witpress.com, ISSN 1743-3533 (on-line)
 Computational Methods in Multiphase Flow III 255

(a)

(b)

Figure 3: (a) Flow pattern in the first 50 diameters; (b) Flow patterns from 50
diameters to 100 diameters.

3.1.2.3 Velocity profile The velocity profile at x/D=75 is shown in Fig. 4

(right), where x is measured along the length of the pipe and D=2”. Fig. 4 shows
that the calculated air speed is higher than the water speed by about 10%. A
similar trend was also observed in the experiment (Lewis et al. [2]).

4.00 4 2
vel@x/D=75
3.50 3.5 experiment
3.00 3 vof@x/D=75 1.5

air volume of fraction

2.50
Uave(m/s)

velocity (m/s)

2.5
2.00
2 1
1.50
1.5
1.00
CFD 1 0.5
0.50
Experiment 0.5
0.00
-1 -0.5 0 0.5 1 0 0

r/R -1 -0.5 0 0.5 1

r/R

Figure 4: Comparison of calculated time-averaged velocity profile with

measurements at x/D=100 and x/D=75.

Vapor
Vapor

Liquid
Liquid

(a) 2D channel flow (b) 3D pipe flow

Figure 5: Illustration of slug distribution in cross-sectional view.
At x/D=100 (Fig.4 left) the air had gained more speed than water by the time
it reaches the outlet. For the region of r/R < -0.4, the velocity profile agreed very
well with the measurements because this region was occupied by water in both
cases. The velocity difference increases for the region r/R > -0.4 due to the
difference in locations of vapor/liquid interface. With the same vapor volume
fraction, a large slug is expected to displace more water along the vertical
centerline in a 3D pipe, whereas a 2D channel is more likely to have a flat
interface from a cross-sectional view (Fig. 5). This explains the major
discrepancies in velocity profiles between CFD and experiment. The accuracy of
the predicted velocity is about ±20%.

WIT Transactions on Engineering Sciences, Vol 50, © 2005 WIT Press

www.witpress.com, ISSN 1743-3533 (on-line)
256 Computational Methods in Multiphase Flow III

3.2 Stratified flow validation

Two-phase stratified flow is one of the simplest flow regimes since the liquid
and vapor are just separated by gravity. However, there are some challenges
with modeling due the wavy interface between the liquid and vapor layer. Fabre
et al. [3] have published experimental results for stratified flows in a closed
slightly inclined channel.
In the CFD model, the computational domain for the stratified flow is in a
channel that is 12 meters long, 0.2 m wide and 0.1 high. The dimensions are
identical to the experimental setup described in Fabre et al. [3].
Velocity inlet boundary conditions are specified at the air and water inlet.
The water inlet velocity was 0.476 m/s and air velocity at 5.5 m/s with mean
water depth at 0.0315m. This condition corresponded to run #400 in the
experiment. Downstream outlet is specified as an outflow. The four sidewalls
were specified as stationary walls. A total number of 310,000 calculation cells
were generated for this model.
The results of the stratified flow modeling in the form of phase distribution
and velocity fields are shown in Fig. 6. The vapor and liquid interface with
small wave motion at the interface is clearly seen.

3.2.1 Velocity profile

Time averaged data was recorded at various locations of the flow. A comparison
with experimental measurement is shown in Table 1. At Z = 0, the flow is
expected of zero velocity due to the non-slip wall boundary condition.
Therefore, a location of Z = 0.0125m was chosen in CFD for comparison with
experiments. It is suspected that the experimentalist had chosen the measuring
points close to the wall but inaccurately presented the location as Z = 0, instead
of Z ≈ 0.

X
Air
0.1m 12m

0.2m Water
Air and water distribution Z

Figure 6: Air and water distribution of the stratified flow. Black represents
water and gray represents air.
The y = 40 mm location is close to the liquid and vapor interface. At y > 40,
the space is occupied by vapor and below is liquid. Comparing the velocity
along the channel, U, the prediction and experiment agrees very well away from
the interface. At the interface, large discrepancies occurred. It is possible that

WIT Transactions on Engineering Sciences, Vol 50, © 2005 WIT Press

www.witpress.com, ISSN 1743-3533 (on-line)
 Computational Methods in Multiphase Flow III 257

grid might be too coarse at the interface to be able to capture the wavy interface
correctly. This might have also affected the velocity component along the y and
z direction in the cross sectional area of the channel. This is shown by the large
differences in the V velocity in Table 1. It was reported (Fabre, et al. [3]) that
there was a secondary swirling motion in the cross sectional area of the channel,
caused by the wavy liquid and vapor interface. The model was not able to predict
the secondary motions in the pipe. .
Table 1: Comparison of time-average velocity prediction with experiment
(X = 9.1 m).

Y(mm) U (m/s) U (m/s) V (m/s) V (m/s)

Experiment CFD Experiment CFD
@Z=0 @ Z = 0.0125m @Z=0 @ Z =0.0125m
1.2 0.365 0.37 N/A
13 0.54 0.47 -0.0245 -0.00028
28 0.53 0.51 -0.0226 -0.0003
40 5.96 3.96 0.366 -0.0069
75 5.93 6.02 0.366 -0.0102
96 5.07 5.29 0.067 -0.0058

3.2.2 Pressure loss

The pressure gradient in the gas phase was also compared with experimental
data. As shown in Table 2, CFD over predicts the pressure drop by a factor of
1.7. It is expected that the pressure loss is a result of the wall and liquid/vapor
interface friction. However, in the Volume of Fluid (VOF) model, only one set
of momentum equations are solved and the velocity field shared between phases.
It is possible that the velocities were averaged out in the cells that were occupied
by both liquid and vapor. In reality, vortices occur in these locations and
produce drag force in both phases. The effect of the approximation could have
adversely affected the liquid/vapor interface prediction when the velocity
difference between the phases is large.
Based on experimental researches (Paras et al. [4] and Levy, [2]), the pressure
loss at the liquid/vapor interface was found to be a function of velocity
differential between phases, wave height and the properties of both phases. It is
not clear how these parameters were taken into account for the pressure
calculation in the VOF model.
Table 2: Comparison of pressure gradient in gas phase.

Pressure gradient (Pa/m), Pressure gradient (Pa/m),

Experiment CFD
6.7 11.6

4 Applications of stratified flow and slug flow in piping

This section uses the knowledge base developed during the validation study and
applies it to a slightly more complex piping geometry. One example of such

WIT Transactions on Engineering Sciences, Vol 50, © 2005 WIT Press

www.witpress.com, ISSN 1743-3533 (on-line)
258 Computational Methods in Multiphase Flow III

applications is flow through a vertical tee. Section 4.1 presents a slug flow
through a tee junction and section 4.2 describes a stratified flow through a tee.

4.1 Slug flow in a tee

Using a similar geometry and boundary conditions described in Section 3.1, the
model was extended with a vertical tee at the outlet of the slug flow described.
Fig. 7 illustrates the geometry for this model. Due to similar runtime constraints,
the model was built in 2-D. The air inlet velocity was 3.3m/s, and the water inlet
velocity was 2.5m/s. The top and bottom outlets of the tee were specified as
outflow boundary conditions.

Air Inlets Water Inlets slots

Low air
Flow direction
velocity

walls

Water trapped
in air
eventually
flows down

High velocity
region

Gravity
Recirculation
zone

Figure 7: Velocity vectors superimposed on phase distribution at the tee

joint. Black arrows indicate velocity vectors. Black color
represents water and gray color represents air.

Fig. 7 shows an instantaneous velocity vector plot superimposed with phase

distribution in a tee. A high velocity region was predicted at the tee due to the
gravity force acting on the water as it flowed into the vertical branch. As water
accelerates towards the bottom, a significant amount of air was entrained into the
bottom branch. Some air was trapped and re-circulated at the low velocity
region at the lower corner. The air in the top branch moves very slowly due to
the small pressure gradient in this section.
Fig. 8 shows velocity magnitude and static pressure distribution in the tee
joint at the same instance. It is clear that the pressure distribution is not the same

WIT Transactions on Engineering Sciences, Vol 50, © 2005 WIT Press

www.witpress.com, ISSN 1743-3533 (on-line)
 Computational Methods in Multiphase Flow III 259

between the top and bottom branch due to gravity and flow distribution. This
causes the slow air velocity on the top.

Velocity distribution Static pressure distribution

(Scale: 0 – 10 m/s) (Scale: -54000 to –63000 Pascal,
gauge).

Figure 8: Instantaneous phase, velocity and pressure distributions in a tee

joint.
This CFD model can also show the distribution of phase with time. Due to
the unsteady nature of a slug flow, as the mixture impinges at the tee joint, most
liquid flow through the bottom branch while some vapor was entrained with
liquid. The remaining vapor occupies the top branch. Some liquid was trapped
in the vapor flow on the top branch, but influenced by gravity, it eventually
attached to the wall and flows downward. A small amount of vapor was re-
circulating at the lower left corner of the tee due to the low velocity in this
region.
4.2 Stratified flow in a tee joint
The application of a stratified flow running through a tee connection is described
in this section. The upstream geometry and boundary conditions are similar to
the one used for validation with a vertical tee added at the end. Fig. 9 illustrates
the geometry for this model. The air inlet velocity was 5.5m/s, and the water
inlet velocity was 0.476m/s. The top and bottom outlets of the tee were specified
as outflow boundary conditions. The entire channel and tee joint was also tilted
at a slope of 0.001.
The results of the stratified flow through a tee are also shown in Fig. 9 in the
form of phase distribution and velocity magnitude distribution in the vertical
symmetry plane. As expected, water flows down to the bottom branch at the tee
joint due to gravity. Due to the acceleration of the water stream in the vertical
branch, water layer was stretched thin and was broken up as it flowed
downwards. The breaking of the water layer allowed air to penetrate to the left
side of the bottom branch.
Because the pressure at the bottom branch is lower than the top, a significant
amount of air is diverted to the lower branch. The highest velocity region was
found in the right side of the branch immediately downstream of the tee joint.

WIT Transactions on Engineering Sciences, Vol 50, © 2005 WIT Press

www.witpress.com, ISSN 1743-3533 (on-line)
260 Computational Methods in Multiphase Flow III

Velocity distribution

Phase distribution
Top
outflow
boundary

Air Inlet

Water Inlet

Bottom
outflow
boundary

Figure 9: Geometry and velocity magnitude (m/s).

5 Conclusions and recommendations

The complexity of two-phase flow can cause flow mal-distribution, vibration and
excess thermal and mechanical stress in equipment. Understanding two-phase
flows is important for equipment design with potential multi-phase flow. This
study shows that slug and stratified flow patterns can be analyzed with CFD
analysis and the results can be used to determine flow mal-distributions in multi-
phase applications.
The analysis also shed lights on some of the advantages and limitations of
CFD:
Advantages:
• CFD can predict mean stream-wise flow velocities and flow patterns
with reasonable accuracy.
• CFD helps in visualizing flow patterns in complex systems and can
identify problem zones in design, in terms of flow and phase
distribution, re-circulations.
• For unsteady slug flows, CFD helps understand the development and
physics of such flows.
• Unsteady reaction forces due to a slug flow may be deduced with
reasonable accuracy for equipment support and stress calculations, since
the reaction forces are a direct function of stream-wise velocity.
• Qualitatively, CFD can predict the pressure gradient in stratified and
slug flow.
Limitations:
• The inaccurate prediction of liquid and vapor wavy interface when the
velocity differences between phases are large. Increasing the number of

WIT Transactions on Engineering Sciences, Vol 50, © 2005 WIT Press

www.witpress.com, ISSN 1743-3533 (on-line)
 Computational Methods in Multiphase Flow III 261

cells at the interface may minimize the inaccuracy. However, the VOF
model may also need to be improved for better prediction.
• More research is needed in CFD for better prediction of pressure
gradient in a two-phase flow.

Resolving the limitations in CFD rely on our understanding of two-phase

flow behavior, incorporating these inputs into correct boundary conditions in
CFD and a good-quality mesh. There are several application areas that need
predictions of two-phase flow for design and trouble-shooting purposes.
Providing a valid two-phase CFD model will be crucial during the design stage.

References
[1] Baker, O. Multiphase flow in pipes, Oil and Gas Journal, Nov. 1958.
[2] Lewis, S., Fu, W. L., Kojasoy, G. Internal flow structure description of
slug flow-pattern in a horizontal pipe, International Journal of Heat and
Mass Transfer, 2000.
[3] Fabre, J., Masbernat, L., Suzanne, C. Stratified flow, part I: local
structure, Multiphase Science and Technology, volume 3, 1999.
[4] Paras, S.V., Vlachos, N. A., Karabelas, A. J. Liquid layer characteristics
in stratified-atomization flow. Int. J. Multiphase Flow, Vol. 20, 1994.
[5] Rosendall, B. and Berkoe, J. Success of computational fluid dynamics
(CFD) modeling of fluidics designs using pulsed jet mixers in the Hanford
Waste Treatment Plant (WTP). Bechtel WTP project memorandum,
CCN: 066517.
[6] Levy, "Two-phase flow in complex systems", Wiley-Interscience, 1999.

WIT Transactions on Engineering Sciences, Vol 50, © 2005 WIT Press

www.witpress.com, ISSN 1743-3533 (on-line)