You are on page 1of 21

International Journal of Heat and Mass Transfer 59 (2013) 451471

Contents lists available at SciVerse ScienceDirect

International Journal of Heat and Mass Transfer


journal homepage: www.elsevier.com/locate/ijhmt

Numerical investigation of hydrodynamics and heat transfer of elongated


bubbles during ow boiling in a microchannel
M. Magnini a,, B. Pulvirenti a, J.R. Thome b
a
b

Dipartimento di Ingegneria Energetica, Nucleare e del Controllo Ambientale, Universit di Bologna, Bologna, Italy
Laboratory of Heat and Mass Transfer (LTCM), Ecole Polytechnique Fdrale de Lausanne (EPFL), Lausanne CH-1015, Switzerland

a r t i c l e

i n f o

Article history:
Received 29 June 2012
Received in revised form 21 November 2012
Accepted 2 December 2012
Available online 23 January 2013
Keywords:
Flow boiling
Microchannel
Volume Of Fluid
Evaporation
Heat transfer

a b s t r a c t
Flow boiling within microchannels has been explored intensively in the last decade due to their capability to remove high heat uxes from microelectronic devices. However, the contribution of experiments to
the understanding of the local features of the ow is still severely limited by the small scales involved.
Instead, multiphase CFD simulations with appropriate modeling of interfacial effects overcome the current limitations in experimental techniques. Presently, numerical simulations of single elongated bubbles
in ow boiling conditions within circular microchannels were performed. The numerical framework is
the commercial CFD code ANSYS Fluent 12 with a Volume Of Fluid interface capturing method, which
was improved here by implementing, as external functions, a Height Function method to better estimate
the local capillary effects and an evaporation model to compute the local rates of mass and energy
exchange at the interface. A detailed insight on bubble dynamics and local patterns enhancing the wall
heat transfer is achievable utilizing this improved solver. The numerical results show that, under operating conditions typical for ow boiling experiments in microchannels, the bubble accelerates downstream
following an exponential time-law, in good agreement with theoretical models. Thin-lm evaporation is
proved to be the dominant heat transfer mechanism in the liquid lm region between the wall and the
elongated bubble, while transient heat convection is found to strongly enhance the heat transfer performance in the bubble wake in the liquid slug between two bubbles. A transient-heat-conduction-based
boiling heat transfer model for the liquid lm region, which is an extension of a widely quoted mechanistic model, is proposed here. It provides estimations of the local heat transfer coefcient that are in
excellent agreement with simulations and it might be included in next-generation predictive methods.
2012 Elsevier Ltd. All rights reserved.

1. Introduction
Microscale ow boiling is one of the most promising cooling
technologies to dissipate high heat uxes from microprocessors.
The two-phase cooling, applied directly on the chip through microchannels evaporators, is nowadays succeeding in removing more
than 300 W/cm2 from the electronic chip itself. Besides the capability of removing high heat power densities, Agostini et al. [1] argued that the main advantages of two-phase ow boiling heat
transfer compared to other high heat ux cooling methods are:
lower mass ow rate of the coolant due to the high energy absorption by the latent heat of vaporization, lower pressure drop due to
this lower mass ow rate, lower temperature gradients due to saturated ow conditions and the heat transfer coefcient increases
with heat ux. Within microchannels, once nucleation begins at
one location, the vapor bubble grows rapidly and lls the entire
Corresponding author. Address: via Terracini, 34, 40128 Bologna, Italy. Tel.: +39
051 2090541.
E-mail address: mirco.magnini@unibo.it (M. Magnini).
0017-9310/$ - see front matter 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijheatmasstransfer.2012.12.010

cross-section of the channel. Hence, bubbly ows are suppressed


already at low values of vapor quality, while slug (elongated bubble) and annular ow regimes occupy a large area on the ow map
[2]. In particular, slug ows lead to very efcient heat transfer
mechanisms due to the following local ow structures: the recirculating ows within the liquid slugs enhance heat and mass transfer
from the liquid to the wall; the large interfacial area promotes liquidvapor mass transfer; the presence of the bubbles separating
the liquid slugs prevents the ow to become thermally fully developed and shorter liquid slugs lead to higher local Nusselt number
[3]; the evaporation of the thin liquid lm surrounding the bubble
increases strongly the local heat transfer coefcient [4].
Due to the importance of the slug ow regime in microchannels,
the availability of reliable predicting methods for boiling heat
transfer and pressure drop is fundamental for industrial manufacturing of cooling systems. Models and correlations developed for
the macroscale do not apply well when extrapolated to channel
sizes below 1 or 2 mm, thus highlighting the presence of a macroto microscale transition. Therefore, in the last decade, researchers
have focused on the development of new physically-based models

452

M. Magnini et al. / International Journal of Heat and Mass Transfer 59 (2013) 451471

Nomenclature
Roman Letters
A
area
a
acceleration
B
generic interfacial effect
 2
Boa
bubble acceleration Bond number qaD
r
b
generic uid property
 
Ca
capillary number lrU 
h
i1=2 
r
Co
connement number
2
g DqD
!
PNDf t

Cr

Courant number

cp
D
Fr
f
G
g
H
h
hlv
I
L
La
Lh
Ls
L2, L1
M
_
m
_g
m
_i
m
_o
m
N
n
p
p
q
qe
R
Re

constant pressure specic heat


diameter
surface tension force vector
interface line
mass ux
gravity acceleration vector
height function
heat transfer coefcient
latent heat
VOF indicator function
length
adiabatic length
heated length
slug length
error norm
molecular weight
interphase mass transfer
global rate of vapor creation
mass ow rate across the inlet section
mass ow rate across the outlet section
number of computational cells
interface unit norm vector
pressure
dimensionless pressure
heat ux
evaporation equivalent heat ux
radius
 
Reynolds number qUD
l

Rg
r, z
T
t
U

universal gas constant


cylindrical reference frame
temperature
time
velocity

V=

u
u
V
We

velocity vector
dimensionless velocity vector
volume
 2 
Weber number qUr D

x, y
x
xS
Y
zG
zh

Cartesian reference frame


position vector
interface position vector
eigenfunction
axial position of the center of gravity
axial distance from the entrance in the heated region

uf nf Af

accounting for the actual two-phase ow structure. However, the


contribution of experiments to the understanding of the local features of the two-phase ow is still limited, due to the small scales
involved, while multiphase CFD techniques are emerging as powerful tools to provide detailed and interesting insights into the local
hydrodynamics and wall heat transfer.
2. Literature review
2.1. Experiments on ow boiling in microchannels
Excellent reviews on microchannel ow boiling are available in
Garimella and Sobhan [5], Bertsch et al. [6] and in the Wolverine
Engineering Data Book III [7] written by Thome.
Agostini and Thome [8] analyzed 13 published experimental
studies on ow boiling heat transfer in microchannels. These studies reported a broad agreement on the increasing of the heat trans-

Greek Letters
a
volume fraction
at
thermal diffusivity
b
eigenvalue
c
accommodation coefcient
D
mesh element size
d
liquid lm thickness
dS
delta-function
dT
thermal boundary layer thickness
j
interface curvature
k
thermal conductivity
l
dynamic viscosity
n
Scriven model growth constant
q
density
r
surface tension coefcient
/
interface kinetic mobility
Subscripts
0
initial conditions
1, 2
primary, secondary phase
b
bubble
c
computational cell centroid value
ex
exact value
f
computational cell face centroid value
if
interfacial
l
liquid
N
bubble nose
sat
saturation
sp
single phase
tp
two-phase
v
vapor
w
wall
z, zz
rst, second order derivatives with respect to z
1
far system conditions

fer coefcient with heat ux, weak effect of the mass ux, but
conicting trends with the vapor quality. This suggested to the
authors that additional phenomena, negligible in the macroscale,
must come into play in microchannels.
Due to the substantial heat ux dependency of the heat transfer
coefcient which is typical of a nucleate boiling controlled regime
in the macroscale, many authors concluded that nucleate boiling is
the governing heat transfer mechanism in the microscale as well.
However, Thome observed in [7] that there is not any experimental
proof to conclude that nucleate boiling is the prevalent regime in
microchannels, hence he advised against the application of macroscale ideas to derive microscale ow boiling methods.
Bertsch et al. [6] compared predictions of 25 published correlations for ow boiling heat transfer against 10 independent data
sets from the published literature. They reported that models
developed specically for the microscale gave no better results
than those for conventional channels, and that Coopers pool

M. Magnini et al. / International Journal of Heat and Mass Transfer 59 (2013) 451471

boiling correlation [9] provided the overall best prediction. However, Coopers correlation was able to predict only the 48% of the
data set with a deviation within 30%. Thus, Bertsch and coauthors remarked the clear need for additional research into the
mechanism of ow boiling in small channels.
Thome et al. [4] developed a three-zone heat transfer model for
the evaporation of elongated bubbles in microchannels. The model
assumes that the evaporation of the thin liquid lm trapped
between the bubble and the channel wall is the dominant heat
transfer mechanism rather than nucleate boiling and estimates a
time-averaged heat transfer coefcient in the liquid lm zone by
one-dimensional steady-state heat conduction across the lm.
The bubble is modeled as a trapezoid and empirical correlations
are used to evaluate its size, which in turn lead to ve adjustable
empirical parameters. Recent papers [1012] have eliminated
one of these, nding that the lm dryout thickness can be set to
the measured wall roughness. The three-zone model is very sensitive to the set of parameters chosen, this has led to very good predictions [1315] as well as poor ones [6,12] when employed to
predict independent ow boiling heat transfer databases.
Recently, Harirchian and Garimella [12] employed the Thome
et al. three-zone model [4] to estimate boiling heat transfer in
the slug ow regime, as part of a comprehensive ow regimebased heat transfer model. They modied the three-zone model
original correlations for initial and minimum lm thickness. With
a new set of empirical parameters, the percentage of data predicted
within 30% increased from 36% obtained with the original
three-zone model to 82% of the new one with the proposed
modications when t to their data set for FC-72. Han et al. [16]
performed liquid lm thickness and wall temperature measurements under ow boiling conditions for water and ethanol and
they found a good agreement between the heat transfer coefcient calculated from measured liquid lm thickness and that
obtained directly from wall temperature measurements. This
conrmed that thin-lm evaporation played a dominant role on
the heat transfer within microchannels.
The three-zone model approach, based on thin lm evaporation
as the prevalent heat transfer mechanism and attempting to reconstruct the actual ow conguration, laid the foundation for a more
reliable physics-based modeling of microscale ow boiling. However, new generation methods cannot prescind from accurate models for the macro-to-micro transition, ow patterns and geometry
of the liquid-vapor interface. Kew and Cornwell [17] for example
recognized the unimportance of gravitational forces as a peculiar
effect of microscale conditions. They observed that heat transfer
and ow characteristics deviate considerably from macroscale

1=2
trends when the Connement number Co gDrqD2
had values
above 0.5. Recently, Ong and Thome [18] measured the lm thickness above and below elongated bubbles in ow boiling conditions
and observed that gravity forces are fully suppressed, such that the
ow is symmetric, when Co > 1. Revellin and Thome [19] and
Harirchian and Garimella [20] proposed diabatic vapor quality versus mass ux ow pattern maps, with transition lines captured
through mechanistic models in order to quantitatively distinguish
the different ow regimes. The accurate estimation of the liquid
lm thickness d surrounding the bubble in the slug ow regime
is fundamental for boiling heat transfer models based on thin lm
evaporation [4,12], as the local heat transfer coefcient in the lm
region is computed as h = k/d. Han and Shikazono performed a
large experimental study of liquid lm thickness measurements
for bubbles in steady motion [21] and under acceleration [22].
By applying a scaling analysis to the forces acting on the bubble
to t their experimental data, they proposed the following
relationship to estimate the lm thickness in laminar ow
conditions:

" 
 
  #
d
d
d
min
;
D
D steady
D accel

453

where d is the liquid lm thickness at the beginning of the at lm


region which follows the bubble nose region. Values of the thickness under steady and accelerated conditions are obtained by the
following empirical expressions:

 
d
0:67Ca2=3

2
2=3
D steady 1 3:13Ca 0:504Ca0:672 Re0:589  0:352We0:629
 
d
0:968Ca2=3 Bo0:414
a

3
D accel 1 4:838Ca2=3 Bo0:414
a
where the Capillary number Ca lrU, the Reynolds number Re qUD
l ,
the Weber number We = Ca  Re and the bubble acceleration Bond
2
number Boa qaD
have to be evaluated at the actual bubble
r
velocity.
2.2. Numerical simulations of ow boiling within microchannels
Recent advances on multiphase computational uid dynamics
allow numerical solution of boiling ows within microchannels
to be performed, thus providing essential information on the local
structure of the ow. Interface capturing techniques for xed computational grids, such as Level Set (LS) [23] or Volume Of Fluid
(VOF) [24] methods, are emerging as one of the best mathematical
and numerical treatments of multiphase ow physics due to their
easiness of implementation, accuracy and robustness of the
algorithms.
Talimi et al. [25] provided a comprehensive review of numerical
studies concerning adiabatic and diabatic slug ow in microchannels without phase change. The simulations of slug ows with heat
transfer reported impressive enhancement of the wall heat transfer
performance along the liquid lm region, as well as remarkable increase of heat transfer coefcients in the wake behind the bubble,
due to local recirculation patterns forced by the bubble motion.
Mukherjee and Kandlikar [26] simulated the ow boiling of a
water vapor bubble within a square microchannel, by use of a LS
method to track the interface. They studied the bubble growth rate
for different liquid superheats and ow velocities and observed
that the vapor bubble grew spherically with a linear timelaw
for the growth rate until it approached the channels walls. Subsequently, the bubble stretched and generated a thin liquid lm,
eventually forming some dry patches, while the growth rate
time-law became exponential.
Suh et al. [27] studied, by means of a LS method, the bubble
dynamics and the associated ow and heat transfer in parallel
microchannels, in order to investigate the conditions leading to
ow reversal. They showed that backows may occur in parallel
microchannels when the bubble formation is not simultaneous in
adjacent channels. This leads to a drop in the heat transfer performance at the wall of the channel where reversed ow occurs and
such an instability is boosted by higher wall superheats and smaller contact angles.
Mukherjee [28], simulating the ow boiling of a bubble in contact with the heated surface of a microchannel, investigated the
role of advancing and receding contact angles between the bubble
interface and channel wall. He reported that the wall heat transfer
is improved by a smaller contact angle, as it promotes the formation of a thin liquid layer trapped between the bubble interface
and the channel wall, thus indicating that thin lm evaporation
is the primary wall heat transfer mechanism in microscale ow
boiling as proposed earlier in [29].
Mukherjee et al. [30] performed a parametric study to assess
the inuence of wall superheat, Reynolds number, surface tension
and contact angles on bubble growth rate and wall heat transfer.

454

M. Magnini et al. / International Journal of Heat and Mass Transfer 59 (2013) 451471

They observed that an increase of the wall superheat favored the


bubble nose to move downstream to the channel, allowing a longer
liquid lm to exist thus increasing the wall heat transfer. The Reynolds number was observed to have little effect on the ow, because the velocities associated to the evaporation phenomena
were much higher than that of the liquid inow.
Zu et al. [31] carried out numerical simulations of bubble nucleation, detachment and then ow boiling within a rectangular
microchannel by employing the VOF method included in the commercial CFD software ANSYS Fluent. The phase change at the liquidvapor interface due to evaporation was modeled by
implementing an approximate method based on the concept of
pseudo-boiling. Bubble deformation while growing at the wall
and its trajectory during partially conned growth were successfully compared with experimental results and 1-D theoretical
models.
Zhuan and Wang [32] studied ow patterns and related transitions within circular microchannels in ow boiling conditions by
means of a VOF algorithm. They explored the effect of bubble
lift-off size, heat ux, mass velocity, frequency of bubble generation and uid properties on the location of transition lines from
bubbly ow to semi-annular ow and obtained results in good
agreement with the Revellin et al. [2] experimental ndings.
The direct numerical simulation of the interface has intrinsic
limitations related to the specic interface capturing scheme
adopted. Level Set methods have issues with mass conservation
while the Volume Of Fluid approach tends to suffer from poor estimation of the interface topology, which is involved in the surface
tension force calculation.
In the microscale, the interface temperature condition has to account for the interfacial resistance to mass transfer and the Laplacian jump in pressure across the interface [33]. Microscale effects
such as disjoining pressure and microlayer evaporation, which
act at length scales several orders of magnitude smaller than typical mesh element sizes, should be modeled and coupled with macroscopic calculations when dealing with very thin liquid lms or
wall adhesion [34,35] and dynamic contact angles at the solid
liquidvapor three-phase contact line should be assigned.
Boiling ows require typically very ne computational grids,
such that computations become extremely time-consuming, especially when performing three-dimensional simulations.

3. Objectives and methodology


The objective of this study is to analyze the hydrodynamics and
heat transfer of slug ow in microchannels in ow boiling conditions. This is accomplished by performing numerical simulations
of a single vapor bubble owing within a heated channel and
growing as a consequence of evaporation of liquid at the interface.
An elongated bubble at saturation conditions is patched at the upstream of a horizontal circular microchannel. The dynamics of the
bubble during evaporation, the thermal and ow elds within the
channel and the local variations of the heat transfer coefcient at
the heated wall are investigated, for different refrigerant uids
and operating conditions. The effect of the uid properties on the
bubble growth rate and the governing heat transfer mechanisms
along bubble and bulk liquid regions are explored. The results of
the computations suggest modications to the above mentioned
three-zone model for boiling heat transfer in microchannels.
Simulations are performed by means of the nite-volume commercial CFD solver ANSYS Fluent version 12 where the solvers default VOF algorithm is adopted to capture the interface. In order to
overcome VOF issues on poor interface reconstruction, a Height
Function algorithm [3639] is implemented by self-developed subroutines to replace ANSYS Fluent default estimation of the surface

tension force. An evaporation model is introduced within the solver through additional subroutines to estimate the rates of mass
and energy exchange at the interface due to evaporation. The evaporation model allows the interface temperature to deviate from the
saturation condition, according to a physical model developed by
Schrage [40] for interphase mass transfer. Dryout and microlayer
effects are avoided in our simulations by appropriate choice of
the operating conditions for each case. Gravitational effects are
made negligible by choosing operating conditions which lead to
Co > 1 [18]. This allows a two-dimensional axisymmetrical formulation of the ow problem, such that the entire computational effort is aimed to very ne mesh grids and long channels, up to 72
diameters. Abundant use of parallel computations were implemented to decrease the computational time, using up to 128 processors for the simulation run with the longest channel.
4. Numerical framework
4.1. The VOF method
The two-phase ow problem is formulated through a singleuid approach, such that a unique velocity, pressure and temperature eld is shared among the phases. A single set of ow equations
is written and solved throughout the domain and the phases are
treated as a single uid, whose properties change abruptly across
the interface. The ow problem, along with boundary conditions,
is similar to that of a single phase ow; however additional
arrangements are necessary: denition of a marker function to
identify each uid, a method to update the marker function as
the interface evolves, mathematical modeling of interfacial effects
and discretization on the computational grid.
The Volume Of Fluid method denes a marker function I(x,t) as
a multidimensional Heavyside step function with the value 1 in the
primary phase and 0 in the secondary phase. The discrete version
of the indicator function is the volume fraction a, obtained by integration of I(x,t) over the computational cell of volume V:

1
V

Ix; t dV

The so-dened volume fraction represents the ratio of the cell volume occupied by the primary phase. It is 1 if the cell is lled with
the primary phase, 0 if lled with the secondary phase and
0 < a < 1 for an interfacial cell with both phases inside. The generic
uid property b for every domain cell can be expressed in terms of a
as follows:

b b2 b1  b2 a

where b1 and b2 are primary and secondary phases specic properties. Since the volume fraction is transported as a passive scalar by
the ow eld, its values can be updated by solving a transport equation. The interfacial effects are modeled as delta functions concentrated at the phases interface. By referring to dS = d(x  xS) as a
multidimensional delta function which is non-zero only on interface points xS, the generic interfacial effect B(x) is introduced within
the ow equation as the source term B(x)dS. According to the VOF
approach, the delta function is represented in the computational
grid as dS = jraj. Since ra 0 on the few layers of cells laying
across the interface, the interface is meant as a transition region
with nite thickness, where interfacial effects are concentrated
and the uid properties vary according to Eq. (5).
4.2. Governing equations
In this work both phases are always assumed incompressible. It
is worth to note that the density of the vapor phase may decrease

M. Magnini et al. / International Journal of Heat and Mass Transfer 59 (2013) 451471

considerably in the streamwise direction due to the pressure drop


which occurs within the microchannel, but the operating conditions simulated in the present work made such effect negligible.
Hence, the mass conservation equation for an incompressible ow
is expressed as:

ru

qv

ql




1
1
_ S
_ raj

md
mj

qv

ql

where the r.h.s. of Eq. (6) accounts for the uid expansion due to
_ The forphase change by introducing the interphase mass ux m.
_ is the task of the evaporation modmulation and computation of m
el, which is presented in Section 4.3. Note that, within the bulk
phases, the Eq. (6) reduces to the well-known divergence-free condition for the velocity eld r  u = 0.
To evolve the interface location, the following volume fraction
conservation equation for ow with phase change is solved:

@a
1
1
_
_ raj
md
mj
r  au
@t
qv S qv

where a represents the vapor volume fraction. Only the primary


phase volume fraction Eq. (7) is solved, while the secondary volume
fraction eld (the liquid phase in this case) is obtained as 1  a at
the end of the calculation.
The single-uid momentum conservation equation for Newtonian uids in laminar ow takes the following form:

 

@qu
r  qu  u rp r  l ru ruT qg F r
@t

where Fr is the surface tension force. By means of the Continuum


Surface Force (CSF) method proposed by Brackbill et al. [41], the
capillary force is converted into a volume force:

F r rjndS rjnjraj

which is concentrated only at interfacial cells, where ra 0. n


identies the interface unit norm vector. The surface tension coefcient r is considered constant. The local interface curvature j is not
available explicitly in the VOF method, but it is implied in the volume fraction eld. Various approaches exist to estimate the curvatures from the volume fractions, see Cummins et al. [36] for
reference. In this work ANSYS Fluents default scheme is replaced
by a self-implementation of the Height Function interface reconstruction algorithm, which is briey introduced in Section 4.4.
The set of ow equations is completed by the energy conservation equation:

@qcp T
_ lv  cp;v  cp;l Tjraj
r  qcp uT  r  krT mh
@t
10
which at the r.h.s. shows the energy source terms given by the
_ lv , with hlv being the latent heat of vaporization,
evaporation mh
_ p;v T and that of the liquid rethe enthalpy of the vapor created mc
_ p;l T, with cp being the constant pressure specic heat.
moved mc
The energy equation (10) does not include the viscous heating
term. Following the dimensional analysis proposed by Morini
[42], the viscous heating contribution was estimated here on the
uid bulk temperature for a single phase ow under operating conditions representative of the cases simulated. It was found to be of
the order of 104 compared with the rise in temperature generated
by the wall heat ux, and therefore the viscous heating effect is
negligible in our simulations. The variation of the uid temperature in the simulations performed is sufciently small such that
the uid specic properties are considered constant throughout
the ow domain.

455

4.3. Evaporation model


The task of the evaporation model is to provide an estimation of
_ due to the evaporation of the
the local interphase mass transfer m
liquid at the interface, depending on the local temperature eld.
The foundation of the evaporation model is the interfacial condition assigned to the temperature. Schrage [40] assumed that, at
the interface, vapor and liquid temperatures are at their thermodynamic equilibrium saturation values, but he supposed an interfacial jump in the temperature to exist, such that at the interface
Tsat(pl) = Tl Tv = Tsat(pv). Schrage applied the kinetic theory of
gases to express the net ux of molecules crossing the interface
due to the phase change, as a function of the temperature and pressure jumps. When phase change occurs, a fraction c of the molecules from the bulk phase strike and cross the interface, that is
evaporate or condense, while the fraction 1  c is reected. The
evaporation and condensation fractions ce and cc are often considered equal and are referred to as the accommodation coefcient.
_ is given by the difference
The net mass ux across the interface m
in the liquid-to-vapor and vapor-to-liquid mass uxes and according to [43] is given by the following expression:

_
m



1=2 
2c
M
pv
pl
p
 p
2  c 2pRg
Tv
Tl

11

where M is the molecular weight and Rg = 8.314 J/mol  K is the universal gas constant, pv and Tv are the vapor pressure and temperature at the interface, pl and Tl are the liquid pressure and
temperature at the interface.
The accommodation coefcient is difcult to be measured
experimentally and it is known only for a few liquids, with a large
degree of uncertainty. Marek and Straub [44] analyzed the published data for water and reported values in the range from 103
to 1. Rose [45] performed a review of experimental results on dropwise condensation and concluded that the most reliable values for
c were close to unity. Wang et al. [46] showed that non polarliquids have an experimentally determined accommodation
coefcient of unity. As it will be discussed in the Section 5.3, we
found the best agreement with analytical solutions by setting c = 1.
Tanasawa [47] assumed that for small interface temperature
jumps, such that (Tv  Tif)  Tv, the interphase mass ux depends
linearly on the temperature jump between the interface and the
vapor phase:

_
m


1=2
qv hlv T if  T v
2c
M
2  c 2pRg
T 3=2
v

12

The operating conditions simulated in our work involve small


Laplacian pressure jumps across the interface, such that at the interface Tv Tsat(p1) with p1 being the system pressure. Therefore, the
evaporation model implemented in the numerical framework computes the interphase mass transfer through the following modied
version of the Tanasawa expression (12):

_ /T  T sat p1
m

13

where / is the so-called kinetic mobility:


1=2
2c
M
qv hlv
2  c 2pRg
T 3=2
sat p1

14

and 1// can be meant as the interfacial resistance to mass transfer.


In Eq. (13), the kinetic mobility and the saturation temperature are
constant throughout the domain, and therefore the evaporation
model computes the rate of mass transfer at the interface proportional to the local interface superheating. For each superheated
_ raj of vapor is created and the same
interface cell, an amount mj
mass of liquid disappears. The latent heat of the evaporating liquid

456

M. Magnini et al. / International Journal of Heat and Mass Transfer 59 (2013) 451471

is subtracted from the energy stored within the cell, such that the
temperature drops locally to a value close to the saturation condition. Therefore, the interface temperature is always equal or little
above the saturation value.
The mass and energy source terms at the r.h.s. of Eqs. (7) and
(10) are concentrated at the 23 cells laying across the interface.
Such source terms localized in a narrow region may give rise to
numerical instabilities when the rate of mass and energy production is high. Hence, a mathematical procedure was implemented
to smear the source terms over a few cells across the interface,
as proposed by Hardt and Wondra [48]. The solution of a diffusion
equation for the original evaporation rate provides a smooth evaporation rate eld. Then, the use of volume fractions and normalization factors allows the newly created vapor to concentrate on the
vapor side of the interface and liquid to disappear on the liquid
side, always ensuring global mass conservation. The reader is referred to [48] for a detailed description and validation of the procedure. In order to test the entire evaporation model, a vapor bubble
growing in superheated liquid was simulated and the results are
reported in Section 5.3.
4.4. Height Function algorithm
Generally speaking, once an approximation of the interface unit
normal vector n is built, the local curvature can be derived as
j = r  n [41]. ANSYS Fluent (version 12 and earlier) computes
the interface unit norm vector as n = ra/jraj according to early
Youngs PLIC (Piecewise Linear Interface Calculation) formulation
[49], and it estimates the curvature by differencing volume fractions. However, such an approach is known to have poor accuracy
as the volume fraction changes abruptly across the interface and
standard derivation schemes do not converge when applied to
strongly discontinuous functions. The consequence is the creation
and growth of unphysical velocities, known as spurious velocities
or parasitic currents [50], which may lead to unreal deformation
of the interface, up to its breaking-off. Furthermore, these numerical artifacts articially increase heat convection at interfaces,
speeding up evaporation or condensation phenomena through a
purely numerical process.
To overcome this limitation, a Height Function algorithm was
implemented to replace the ANSYS Fluent default method to estimate curvatures. Let y = f(x) be the mathematical function identifying the interface line in a Cartesian (x,y) reference frame, as shown
in Fig. 1. The height function H(x;D) represents the height of the
interface line f(x), averaged within a local stencil of width D and
centered on x:

Hx; D

1
D

xD=2

f tdt

xD=2

Fig. 1. Illustration of the height function on the continuous domain.

15

For axisymmetrical domains with revolution around the z  axis,


the interface unit normal vector and curvature can be represented
by geometrical considerations as:

1
1 Hz 2 1=2

j r  n 

Hz ; 1

16


Hzz
2 3=2

1 Hz 


Hzz
1
jHzz j f z1 Hz 2 1=2

17

where Hz and Hzz denote the rst and second order derivatives with
respect to z and f(z) is the local elevation of the interface over the
revolution axis.
The HF algorithm implemented here is a combination of the
Malik et al. [51] and Hernandez et al. [52] versions, with the addition of a self-developed routine to estimate the local interface elevation f(z). The algorithm is written for two-dimensional and
axisymmetrical geometries, with constant grid spacings. In Section 5, the performance of our implementation of the HF algorithm
is assessed by several validation benchmarks.
4.5. The ow solver
ANSYS Fluent discretizes and solves the ow equations by
means of a nite-volume scheme. The volume fraction Eq. (7) is
discretized in time with a rst order explicit scheme and the convective term is computed through a geometrical PLIC [49] reconstruction of the uxes across the faces of each interfacial cell.
The numerical stability of the explicit PLIC scheme poses a limitation on the maximum time step allowed to solve the volume fraction equation, since the interface must travel less than one grid cell
at each time interval. The time step for the volume fraction equation is calculated by the solver according to the maximum Courant
number (Cr) allowed for interface and near-interface cells. The
Courant number is a dimensionless number that compares the
simulation time step Dt and the time it would take for the uid
to empty out of the cell:

Cr

Dt
PN f
V= f uf  nf Af

18

where V is the volume of the cell and the sum loops on the Nf
boundary faces of the cell. The Fluent default value of Cr = 0.25
was used here.
The momentum and energy equations are discretized in time
with a rst order implicit formulation, which allows a coarser time
step than the volume fraction one. A variable time step technique
was adopted with a maximum Courant number of 0.5 and therefore the volume fraction eld is updated more frequently than
the velocity and temperature elds. The convective terms within
momentum and energy equations are discretized with a third order MUSCL (Monotonic Upstream-centered Scheme for Conservation Laws) [53] scheme, while the diffusive terms are discretized
with a central nite-difference scheme. The cell-centered gradients
of each scalar eld are computed from the scalar values at the cell
face centroids, by means of the Green-Gauss theorem. The Fluent
GreenGauss node-based formulation [54] was proved to be the
best option to enforce the balance among pressure and surface tension within the momentum equation when the HF algorithm is
employed, leading to spurious velocities of two orders of magnitude smaller than other options available in the solver. Furthermore, Gupta et al. [55] reported smaller unphysical pressure
oscillations at the interface of a bubble when the node-based formulation was employed rather than a cell-based one.
The mass conservation equation (6) is turned into a pressure
correction equation which is coupled to the momentum equation
(8). The pressurevelocity coupling is handled by a PISO (Pressure
Implicit Splitting of Operators) [56] algorithm, which was proved

M. Magnini et al. / International Journal of Heat and Mass Transfer 59 (2013) 451471

to converge more quickly than the other options available in the


solver. The control-volume integration of the pressure gradient
within the momentum equation turns the pressure gradient into
differences among face-centered pressures. Fluent uses a collocated technique, in which the ow equations are solved for cellcentered variables. Fluent offers a body-force-weighted scheme
to interpolate cell-centered pressures at the cell faces and a PRESTO (PRessure STaggering Option) algorithm that solves the pressure correction equation for a staggered control volume, thus
leading to face-centered pressures without the need of interpolations. An evaluation here showed that the PRESTO algorithm generated spurious velocities of three orders of magnitude lower than
the body-force-weighted scheme.
The Height Function and evaporation models are implemented
in the ANSYS Fluent software by means of user-dened subroutines. Fluent default surface tension computation is disabled by
setting a zero value of the surface tension in the solver. For each
interfacial cell, the HF-computed local curvature is used to estimate the local surface tension force by means of Eq. (9), where
the actual value of the surface tension coefcient is set. Finally,
the estimated local surface tension force is introduced within the
momentum equation as a source term. ANSYS Fluent does not handle any default evaporation model. Therefore, the evaporation
model discussed in Section 4.3 computes the r.h.s. terms of Eqs.
(7) and (10), and hence they are introduced in the related equations as source terms.
5. Validation benchmarks
In Sections 5.1 and 5.2 the HF performance in the reconstruction
of a circular interface and simulation of an inviscid static droplet
are compared with the Fluent-default PLIC-based algorithm, in
the following referred to as Youngs method. In Section 5.3, the
HF and evaporation models are tested by the simulation of a vapor
bubble growing in superheated liquid and compared with analytical solutions. Finally, in Section 5.4 the adiabatic ow of an elongated bubble within a circular horizontal microchannel is
simulated and results are validated through comparison with correlations available in the literature. Further benchmarks for our
implementation were discussed in [57].
5.1. Reconstruction of a circular interface
This test case involves the curvature calculation algorithm
alone, without solving the ow equations. A two-dimensional circular droplet of radius R = 5 mm is placed within a L = 4R side
square domain and the HF and Youngs algorithm performances
on local interface curvature estimation under mesh renement
are compared. The coarsest computational mesh has 10  10 elements, with R/D = 2.5. The most rened mesh has 160  160 elements, with R/D = 40. The circle center is placed randomly
around the domain center (0, 0) in the interval ([0, D], [0, D]). For
each test case, the analytical surface representing the droplet is
intersected with the domain mesh, then the volume fraction eld
is mapped through the numerical computation of the areas. For
each mesh element size, 50 runs are performed to span the range
of possible positions for the circle interface, then the results are
averaged. The comparison is performed by computing the following curvature error norm for each test run:

L1 j

max jji  jex j

jex

for i 1; . . . ; Ni

19

where Ni is the number of interfacial cells, ji is the ith cell computed curvature and jex = 1/R is the exact curvature of the droplet.
Fig. 2 reports the values of L1(j) under mesh renement for HF and

457

Fig. 2. L1(j) error norm convergence rate. White circles are HF errors and black
diamonds are those of Youngs. The solid line is the second order convergence curve.

Youngs algorithms. Youngs curvature estimation worsens as the


mesh is rened, because the standard derivation schemes do not
converge when differencing volume fractions. On the other hand,
the HF approach gives estimations that converge with the second
order of the mesh size, which is consistent with the second order
accurate nite-difference schemes used to differentiate the local
heights. A second order convergence rate was detected for the
Height Function algorithm by various authors as well [36,38,39],
thus proving the accuracy of our implementation. Note that at the
highest mesh resolution tested, the HF evaluates curvatures with
a maximum error of four orders of magnitude lower than Youngs
one, which is a considerable improvement.
5.2. Inviscid static droplet
Height Function and default Youngs schemes are separately employed to compute curvatures in the numerical simulation of an
inviscid static droplet in equilibrium without gravity. The absence
of gravitational and viscosity effects tests exclusively the accuracy
of the implementation of the surface tension term within the
momentum equation and the solution algorithm. A circular droplet
is centered on a square domain and the mesh size range R/D = [5,40]
is investigated. The geometrical conguration is the same as described in Section 5.1. Surface tension and phase densities are set
to unity. Viscous and gravity effects are neglected. A constant value
of the pressure is set on all domain boundaries. The time step for the
solution of the momentum equation is xed to Dt = 5  107 s. The
initial velocity eld is null throughout the domain. With such operating conditions, the exact solution of the momentum Eq. (8) is a
null velocity eld, a constant pressure within and outside the droplet and a pressure jump at the interface given by the Laplace law
Dpex = r/R. Errors in surface tension estimation generates unphysical
ows and an erratic pressure eld, hence the magnitude of the
velocities and the errors in the pressure are optimal parameters to
compare the efciency of HF and Youngs schemes within the
numerical solver. The following dimensionless velocity and interface
pressure jump error norms are observed:

L1 ju j


1=2
2qR

 maxjui j for i 1; . . . ; N

20

where N is the number of domain cells and (r/2qR)1/2 is a chosen


velocity scale,

v
uN
int
uX
1
Dpi  Dpex 2
t
L2 Dp
Dpex i1
Nint

21

458

M. Magnini et al. / International Journal of Heat and Mass Transfer 59 (2013) 451471

where Nint is the number of interior droplet cells. Fig. 3(a) reports
the velocity error norms after one simulation time step. The
Youngs-based algorithm generates spurious vortices whose magnitudes increase with the mesh renement, as the curvature estimation gets worse by rening the numerical grid. The Height Function
method exhibits velocities which scale with a convergence rate between the rst and second order with respect to the mesh element
size, in good agreement with Francois at al. [38] observations. When
the nest R/D = 40 computational grid is employed, the parasitic
currents induced by the HF algorithm are three orders of magnitude
lower than Youngs generated ones. Such an improvement of the
standard commercial code is favorable to simulate elongated bubbles within microchannels, because the domain mesh is typically ner than R/D = 40. When the Fluent default estimation of curvatures
is employed, the magnitude of the spurious ows may easily approach that of the physical phenomenon being simulated, thus rendering the results unusable. Fig. 3(b) plots the errors in the
computed pressure eld after one simulation time step. The Height
Function computation exhibits a convergence rate between the rst
and second order as the mesh is rened. The Youngs algorithm gives
a converging pressure eld only when R/D 6 10; however at higher
mesh resolutions the average pressure within the droplet deviates
only by 1% from the exact value, which is a reasonable error.
5.3. Vapor bubble growing in superheated liquid
The heat-transfer-controlled growth of a spherical vapor bubble
in an innitely extended superheated liquid was simulated.
According to the analysis of Plesset and Zwick [58] on the bubble
growth process, the heat-transfer-controlled growth reaches an
asymptotic stage in which the growth of the bubble is limited by
heat transport to the interface. The pressure within the bubble is
equal to the liquid pressure increased by the pressure jump at
the interface and the temperature of the vapor is equal to the saturation temperature for that pressure. Scriven [59] has derived an
analytical solution for this stage neglecting viscous and surface
tension effects and considering the interface to be at the saturation
temperature. He obtained the following time-law for the bubble
radius:

p
Rt 2n at t

22

where n is a growth constant whose details can be found in [59] and


at is liquid thermal diffusivity. This solution is used to validate the
simulations here.
A spherical vapor bubble of radius R0 = 0.1 mm is initialized at
the center of the axis of an axisymmetrical domain of radius 4R0

and length 8R0. A uniform mesh size is chosen, with 1 lm element


size. Such a ne grid is necessary in order to solve the thin thermal
boundary layer surrounding the bubble interface. A constant pressure is set at all boundaries except for the axis. Gravity effects are
neglected. The initial bubble size is large enough to neglect vapor
saturation temperature rise due to pressure jump across the interface, so that the saturation temperature is equal in both phases. As
initial conditions, the velocity eld is zero, the vapor phase is at the
saturation temperature while the liquid is superheated at T1 =
Tsat + 5 K. Since the simulation starts at t = t0, when R(t0) = R0, a
thermal boundary layer has already been developing on the liquid
side around the bubble since the beginning of the heat-transfercontrolled growth stage. The temperature eld within the layer
at t = t0, and thus its thickness dT, can be extrapolated from the Scriven solution [59]. The temperature prole within the layer serves
as the initial condition for the temperature eld. In order to avoid
that the thermal layer overlaps the vaporliquid interface on the
computational grid, it is initialized with a bit of misplacement,
about 12 cells, outside the bubble interface. Three different uids
were tested. Water at atmospheric pressure and HFE-7100 at
0.52 bar, both with n = 15.1 and dT = 7 lm, and R134a at 0.84 bar,
with n = 9.34 and dT = 11 lm. The choice of each system pressure
was done in order to have similar growth constants for the uids.
All vapor and liquid properties for the uids are considered constant at the saturation temperature. The accommodation coefcient within Eq. (14) for the evaporation model is set as unity.
Fig. 4 shows the bubble radius evolution compared to the analytical solutions for all three uids. Numerical data show very good
agreement with the analytical results.
For each uid, the bubble
p
numerical growth rate follows a t proportional law, as it should
be from Eq. (22). This does not happen during the initial growth
phase, when the initially misplaced thermal boundary layer rearranges itself to t the interface position. This settlement phase is
reected on the numerical growth rates being lower than analytical ones at the beginning of the simulations. As detected by Kunkelmann and Stephan [35], the liquid thermal conductivity is the
parameter that rules the length of this thermal layer settlement
phase. The higher the liquid thermal conductivity is, the faster is
thermal layer arrangement. HFE-7100 has the lowest thermal conductivity among the uids employed, for this reason its numerical
bubble growth rate deviates slightly from the analytical curve in
Fig. 4. The deviation for water at its higher time steps is due to
the presence of weak parasitic vortices across the interface, which
neither the use of the Height Function algorithm can overcome at
all. Their effect is more evident for water because it has the highest
surface tension coefcient (around four times higher than the

Fig. 3. (a) L1(juj) and (b) L2(Dp) error norms after one simulation time step. White circles are HF errors and black diamonds are Youngs ones. The dashed line is the rst
order convergence curve and the solid line is the second order curve.

M. Magnini et al. / International Journal of Heat and Mass Transfer 59 (2013) 451471

PN
zc ac V c
zG Pc1
N
c1 ac V c

459

23

where N is the number of computational cells, zc the axial location


of the cth cell centroid and V the volume of the cell.
The width of the liquid lm surrounding the bubble, the bubble
terminal velocity and the pressure drop across the channel at the
steady state of the ow are compared with correlations available
in the literature. The Han and Shikazono correlation for the liquid
lm thickness at steady conditions reported in Eq. (2) is considered. It involves dimensionless groups computed by referring to
the bubble velocity. A prediction for the steady velocity of the bubble Ub can be obtained by applying a mass balance to part of the
channel which includes the owing bubble, which for axisymmetrical ows leads to the following relationship:

Ub
Fig. 4. Vapor bubble radius over time for analytical (lines) and numerical (symbols)
solutions.

refrigerant uids considered here) and, as observed by various


authors [50,60,61], the spurious velocities magnitude is proportional to r. However, the maximum deviation between numerical
and analytical bubble radius for water stays under the 10%
throughout the simulation.
5.4. Simulation of the adiabatic ow of an elongated bubble inside a
horizontal microchannel
The isothermal ow of an elongated bubble within a horizontal
microchannel, pushed by a liquid ow rate, is simulated under several operating conditions. Such a ow conguration corresponds to
the adiabatic version of the evaporating ow which this work deals
with, and therefore this test is an optimal benchmark to evaluate
the accuracy of the numerical framework by comparison of simulation results with published correlations. An elongated gas bubble
is initialized at the upstream of a horizontal circular channel and a
constant liquid ow rate feeds the channel. The bubble, pushed by
the liquid ow, accelerates and deforms until a steady motion is attained. The ow domain is modeled as a two-dimensional axisymmetrical channel with diameter D = 1 mm and length L = 8D. The
bubble is initialized as a cylinder with spherical rounded ends.
The initial length of the bubble is 3D and the lm thickness is d/
D = 0.045. However preliminary tests showed that the steady ow
achieved is independent of the bubble initial shape and size. The
liquid ow into the channel is modeled as a fully developed laminar velocity prole set as the boundary condition at the channel inlet and outlet. The average velocity of the liquid at the inlet and
outlet boundaries is xed at Ul = 0.25 m/s for all the simulations.
The specic properties of gas and liquid phases are different for
each simulation run. The values chosen give the Capillary numbers
of Ca = 0.025 and 0.0125 and Reynolds numbers within the range
of 15.625625, with both the groups computed by referring to
the liquid average velocity. The liquid to gas density ratio is set
to 1000 and the viscosity ratio to 50. The computational grid is a
uniform D/D = 100 mesh for the simulations with Ca = 0.025 and
D/D = 200 for those with Ca = 0.0125, in order to always have at
least ve cells discretizing the predicted liquid lm thickness in
accordance with the Gupta et al. [55] recommendation. Each simulation is run until a steady state condition for the bubble velocity
is reached. The bubble velocity is estimated at every time step by
differencing in time the position of the center of gravity zG computed as:

Ul


1  4 Dd 1  Dd

24

where d is the thickness of the liquid lm. Predictions of terminal


liquid lm thickness and bubble velocity are obtained by solving
iteratively Eqs. (2) and (24), introducing the average liquid velocity
Ul as the initial guess.
Kreutzer et al. [62] suggested the following correlation for the
pressure drop in a channel with ow of an elongated bubble:



 r
64 1
L
Dp 1:08
qU 2 s
3Ca2=3 Re1=3
Re 2
D
D

25

where Ls is the length of the liquid slug, U is the supercial velocity


of the ow, Ca and Re have to be evaluated by referring to U. Eq.
(25) is used as the benchmark for pressure drop in the simulations.
Fig. 5 shows the bubble terminal shapes obtained by the numerical simulations. At a xed Ca, an increase of the Reynolds number
tends to sharpen the bubble nose, to atten the rear and to thicken
the liquid lm due to the effect of inertia, as already observed by
Aussillous and Qur [63] and Kreutzer et al. [62]. At low Reynolds
numbers the liquid lm is at but, as Re is increased, the bubble
rear enlarges thus squeezing locally the lm. At the highest Reynolds number simulated, the bubble rear in Fig. 5(a) and (b) shows
some capillary waves whose wavelength seems to be a function of
the surface tension. Liberzon et al. [64] observed similar waves on
the surface of short Taylor bubbles rising in vertical pipes and concluded that their length is a function of liquid density, surface tension and distance from the bubble nose. Due to these waves, the
liquid lm does not reach a constant thickness because from the
bubble nose toward the rear the lm becomes thinner and then
it also becomes wavy. By comparing the shapes obtained with
Ca = 0.0125 and Ca = 0.025, the thinning of the liquid lm is evident, as the inuence of the higher surface tension tends to round
off and thus shorten the bubble.
Table 1 summarizes the numerical results on pressure drop, liquid lm thickness and bubble terminal velocity compared with
the considered correlations. The simulations reproduce quite well
the rise in the pressure drops with the increase of the Capillary
and Reynolds numbers, where the maximum deviation from Eq.
(25) is 17.6% and the average error is around 10%. The thickness
of the liquid lm in each simulation is not constant along the bubble. At low Re the value reported refers to the thickness at the central region of the lm, where it is almost constant. At high Re, a
zone with constant thickness is not observed and the value reported refers to the location where the smooth prole of the bubble matches the wavy prole at the rear. The thickening effect of
the inertia on the liquid lm is well captured by the simulations
as well as the thinning effect of the capillary forces, such that the
lm thickness increases as Ca and Re grow. The agreement with
Han and Shikazono correlation is very good and the errors are

460

M. Magnini et al. / International Journal of Heat and Mass Transfer 59 (2013) 451471

Fig. 5. Bubble terminal shapes for different test conditions.

Table 1
Comparison of numerical results and correlations. The errors between parenthesis are computed as
Ca

d/D
Ub [m/s]

 100.

0.025

Re

Dp [Pa]

jcorrnumj
corr

Eq. (25)
num (err)
Eq. (2)
num (err)
Eq. (24)
num (err)

15.625

62.5

312.5

625

60
52 (13.3)
0.0494
0.05 (1.2)
0.308
0.308 (0)

65
54 (16.9)
0.0495
0.047 (5.1)
0.308
0.305 (1)

69
71 (2.9)
0.0526
0.05 (4.9)
0.312
0.305 (2.2)

73
80 (9.6)
0.0578
0.06 (3.8)
0.32
0.315 (1.6)

Ca

0.0125

Re

15.625

62.5

312.5

625

62
57 (8.1)
0.0331
0.0279 (15.7)
0.287
0.279 (2.8)

67
58 (13.4)
0.0333
0.028 (15.9)
0.287
0.278 (3.1)

74
61 (17.6)
0.0352
0.035 (0.6)
0.289
0.285 (1.4)

80
82 (2.5)
0.0379
0.038 (0.3)
0.293
0.291 (0.7)

Dp [Pa]
d/D
Ub [m/s]

Eq. (25)
num (err)
Eq. (2)
num (err)
Eq. (24)
num (err)

within 5% with the exception of the cases with Ca = 0.0125 and


Re = 15.625 and 62.5, whose errors are about 15%. However, the local lm thicknesses in the simulations range within 0.0220.034
for Re = 15.625 and from 0.02 to 0.038 for Re = 62.5, thus including
the predicted values. Eq. (24) estimates the terminal velocity of the
bubble in excellent agreement with the numerical simulations for
all the simulation runs, with errors within 3%.
6. Flow boiling of an elongated bubble within a horizontal
microchannel: simulation set-up
6.1. Flow conditions
Five different simulation runs are performed and their operating conditions are summarized in Table 2. The circular microchannel is modeled as a two-dimensional axisymmetrical channel with
a diameter D = 0.5 mm and length L that varies depending on the
simulation run. The channel is always split into an adiabatic region

of length La followed by a heated region of length Lh. An elongated


vapor bubble of length 3D at saturation conditions is initialized as
a cylinder with spherical rounded ends, placed at the upstream of
the channel as depicted in Fig. 6, which reports an illustration of
the initial conguration for the Case 1. The bubble is pushed by a
saturated liquid inow of mass ux G, introduced within the channel through the inlet section upstream to the channel. A constant
heat ux q is applied at the wall of the heated region of the
Table 2
Operating conditions for ow boiling simulation runs. La stands for adiabatic length of
the channel, Lh for the heated length.
Case

Fluid

G [kg/m2 s]

Tsat [C]

q [kW/m2]

La

Lh

1
2
3
4
5

R113
R113
R245fa
R134a
R245fa

600
600
600
500
550

50
50
50
31
31

9
20
20
20
5

8D
8D
8D
8D
16D

12D
22D
22D
22D
56D

M. Magnini et al. / International Journal of Heat and Mass Transfer 59 (2013) 451471

461

UP

Fig. 6. Initial temperature eld within the channel, wall temperature (dashed line) and heat transfer coefcient (solid line) for simulation Case 1. The bubble interface is
represented by the black line prole at the upstream of the channel. The channel image is stretched vertically to enlarge the thermal boundary layer at the heated wall.

channel. The gravitational force is neglected. A constant velocity of


the liquid Ul = G/ql is imposed at the inlet section of the channel. At
the channels outlet, the Patankars outow boundary treatment
[65] is set for the velocity and temperature elds. This mimics a
zero gradient boundary condition for velocity and temperature,
which is a valid assumption provided that the Peclet number of
the ow exceeds 1 and that no ow reversal occurs at the outlet
section. The initial velocity and temperature elds are obtained
as results of a preliminary liquid-only steady state simulation
run under the same ow conditions. For reference, Fig. 6 reports
for Case 1 the initial temperature eld within the channel, the wall
temperature prole and the heat transfer coefcient computed as:

q
T w  T sat

26

where Tw is the local wall temperature.


Due to the implemented version of the Height Function algorithm, only uniform grids with square elements can be adopted.
The choice of the computational grid for the simulations is addressed in the next section.
6.2. Grid convergence analysis
The minimum number of mesh elements to be tted within the
liquid lm to obtain converged results is the object of this grid convergence analysis. Fig. 6 shows that the heat load applied to the
diabatic wall of the channel generates a superheated thermal
boundary layer. The bubble ows downstream and when the vaporliquid interface reaches the superheated thermal layer the
evaporation process starts. Due to the elongated prole and the
sharp nose of the bubble, such contact likely occurs in the liquid
lm region, where the thickness of the thermal layer approaches
the height of the lm. The evaporation model cools down locally
the liquid temperature to maintain the bubble interface close to
the saturation temperature, and hence the liquid temperature
within the lm drops sharply from the highest value at the channel
wall to the lowest value at the bubble interface. For this reason, in
the lm region high temperature gradients take place and the computational grid needs to be ne enough in order to solve the gradients properly. The accurate discretization becomes fundamental in
conjunction with the evaporation model, which computes the rate
of evaporation proportional to the local interface superheating. The
Gupta et al. [55] criterion is valid for adiabatic or diabatic ows
without phase change, but it cannot be assumed a priori in the
present conguration.
Four computational grids are employed to perform the Case 1
run, whose operating conditions are listed in Table 2. For such

operating conditions, Eq. (2) predicts a liquid lm thickness of


d/D = 0.04. The four grids used have channel diameter to mesh element size ratio of D/D = 100, 200, 300, 400, such that the predicted
lm thickness to mesh element ratio ranges from 4 to 16. For each
computational grid the simulation is run until the nose of the bubble reaches the end of the channel. The grid convergence analysis is
performed by examining the differences in the bubble dynamics
obtained with the computational grids employed. The bubble
dynamics are reconstructed by computing at each time instant
the bubble nose velocity as dzN/dt and the bubble growth rate as
dVb/dt, where zN is the bubble nose position and Vb is the volume
of the vapor bubble.
Fig. 7(a) reports the bubble nose velocity as a function of the
time. There is a short stage that lasts less than one millisecond in
which the bubble shape modies from the initialized shape to attain a steady motion. After this settlement period the bubble ows
steadily up to the heated region with a constant velocity of
Ub = 0.47 m/s and liquid lm thickness d/D = 0.045 which are the
same for all the computational grids employed. Hence, even the
coarsest mesh is sufcient to accurately capture the dynamics of
the adiabatic stage of the ow. Fig. 7(a) suggests evaporation to
start at around t = 6 ms, when the bubble nose begins to accelerate.
As well, the beginning of the evaporation phenomenon is clearly
illustrated by the plot of the bubble volume growth rate in
Fig. 7(b). Henceforth the different performances of the computational grids become evident. The computed evaporation rate decreases as the mesh is rened, converging to similar proles for
the grids with D/D = 300 and 400. Instead, an inadequate mesh element size within the liquid lm leads to a faster growth of the bubble, as a consequence of the inaccuracy on the discretization of the
local temperature gradients. The computational grid with D/
D = 300 is a good compromise between accuracy and computational cost of the simulations. The minimum lm thickness, measured at the highest crest of the interfacial wave occurring at the
bubble rear, to mesh element size ratio is dmin/D  7, thus suggesting that 7 cells is the minimum resolution of the liquid lm able to
solve the local temperature eld in ow boiling conditions. All the
test cases reported in Table 2 are characterized by a similar liquid
lm thickness, and hence the mesh grid employed is always the
one with D/D = 300.
6.3. Error evaluation
The error on the mass balance throughout the simulation is
observed for Case 1 run with the chosen D/D = 300 mesh grid.
The mass balance between the terminal sections of the channel
yields:

462

M. Magnini et al. / International Journal of Heat and Mass Transfer 59 (2013) 451471

Fig. 7. (a) Velocity of the bubble nose, (b) bubble growth rate.

Fig. 8. Errors in mass conservation throughout the simulation for Case 1 run with
the D/D = 300 computational grid. The absolute error is reported in kg/s units.



q  qv
_im
_ g t l
_ o t  m
m

qv

27

_ i and m
_ o are the mass ow rates across the inlet and outlet
where m
_ g is the global rate of vapor creation, obtained by
sections and m
_ raj throughout the domain. Fig. 8 depicts the erintegration of mj
rors on the mass balance expressed in Eq. (27). The relative error,
intended as absolute error normalized by the l.h.s. of Eq. (27), always remains under 106 as time elapses and vapor is being created. This means that mass conservation is ensured during bubble
growth. In addition, according to tests, at each time step the unbalance between vapor created and liquid disappeared is of the order
of machine accuracy, and hence all the liquid removed due to evaporation is actually converted into vapor.
7. Flow boiling of an elongated bubble within a horizontal
microchannel: results and discussion
7.1. Bubble dynamics and comparison with a theoretical model
The bubble dynamics during ow boiling for Case 1 is analyzed.
Fig. 9 reports some snapshots of the bubble evolution while owing within the microchannel. The bubble nose enters the heated region of the channel a little after 4.5 ms, while the plot of the bubble
growth rate shown in Fig. 7(b) (blue color) suggests evaporation to
start at 6 ms, when the bubble interface gets in contact with the

superheated thermal boundary layer developing at the heated


wall. Before evaporation starts, we observed that the bubble rear
oscillates with a constant time-period. Liberzon et al. suggested
in [64] that these oscillations generate capillary waves on the bubble surface, which are clearly visible here in Fig. 5(g) and (h). While
evaporation occurs, the rear of the bubble stabilizes and proceeds
downstream with a constant velocity, equal to the adiabatic velocity of the bubble, while the bubble nose accelerates making the
bubble longer. At t = 12.5 ms the bubble volume is twice the value
before evaporation began. As an effect of the acceleration, the
thickness of the liquid lm is increased from the adiabatic value
of d/D = 0.045 to d/D = 0.05, in good agreement with both Han
and Shikazono correlations for adiabatic [21] and accelerated
[22] ows, which predict respectively d/D = 0.056 and d/D = 0.054
if the actual bubble velocity and acceleration at 12.5 ms are used.
In Fig. 7(a) the velocity of the bubble nose appears to increase
linearly with time. Actually, it has an exponential behavior which
is little perceptible as the simulation ends after few milliseconds.
Such hypothesis is proved below by comparing the simulation results with an exponential law for the bubble nose position derived
theoretically by Consolini and Thome [66]. Under the assumptions
of thermodynamic equilibrium between the phases, thus no liquid
superheating, axisymmetrical ow conguration and applying a
constant heat ux only to evaporate the liquid, they obtained the
following time-law governing the position of the bubble nose:

zN t z0





G qv hlv D
4q
exp
t  t 0  1
ql 4q
qv hlv D

28

where z0 and t0 are the axial location and time instant at which the
bubble nucleates. In the present simulations, the bubble is generated before entering within the heated region, hence z0 locates
the entrance in the heated region and t0 the time instant at which
the nose of the bubble crosses z0. In the theoretical model the velocity of the bubble when t = t0 equals the velocity of the liquid inow,
while in the simulations the bubble velocity before evaporation begins exceeds that of the liquid as expressed by Eq. (24). Thus, in order to have the same initial velocity of the bubble for the simulation
and the model, the second term on the r.h.s. of Eq. (28) is multiplied
by the reciprocal of the denominator of the r.h.s. of Eq. (24).
Fig. 10(a) shows the comparison of simulation and theoretical
bubble nose positions for Case 1. The exponential time-law is well
captured by the simulation. The curves overlap at the initial stage
of the bubble evaporation, but afterwords the model underestimates the bubble velocity and, as time elapses, this deviation
slowly grows. The origin of the gap is within the mentioned
assumptions of the theoretical model. In the model the bubble

M. Magnini et al. / International Journal of Heat and Mass Transfer 59 (2013) 451471

463

Fig. 9. Bubble evolution during evaporation for Case 1. The black dashed line indicates the entrance in the heated region.

Fig. 10. (a) Time evolution of bubble nose position for Case 1 obtained with simulation compared with Consolini and Thome theoretical model [66]. (b) Equivalent heat ux
absorbed by evaporation.

grows only due to the wall heat ux, while in the simulation the
bubble grows also absorbing the sensible heat of the superheated
liquid, which is able to store energy before the bubble transits.
Thus, the heat used to evaporate the liquid may be smaller than
the wall heat ux or exceed it. In order to obtain an estimation
of the time-varying equivalent heat ux qe absorbed by the evaporation, the heat absorbed to generate vapor is divided by the
heated surface actually traveled by the bubble at the time instant t:

qe

dV b =dtqv hlv
pDzN t  z0

29

Fig. 10(b) shows that qe > q from t = 8 ms on, where 8 ms is the time
instant at which the bubble has entered completely within the

heated region. As time elapses and the bubble nose ows downstream to the heated channel, it comes across regions more and
more superheated, and therefore qe rises monotonically as a consequence of the increasing evaporation rate. Within the hypothesis of
wall heat ux used only to evaporate liquid, qe represents the wall
heat ux actually felt by the bubble, and therefore when qe > q Eq.
(28) leads to the underestimation observed in Fig. 10(a).
Numerical and theoretical results are compared in Fig. 11 for
simulation cases 2, 3, 4 and 5, and despite the systematical underprediction of the model at the later stage of the growth as noted
above, the agreement is good. Note that among the simulations
run under the same heat ux (cases 2, 3, 4), the operating uid
R113 gives the fastest growing bubble, while the vapor bubble of
R134a is the last to reach the end of the channel. Due to similar

464

M. Magnini et al. / International Journal of Heat and Mass Transfer 59 (2013) 451471

Fig. 11. Time evolution of bubble nose position obtained with simulations compared with Consolini and Thome theoretical model [66].

heat uxes and mass ow rates, the bubble growth rate depends
mainly on liquid-to-vapor density ratio, kinetic mobility and liquid
specic heat of the operating uids. The specic heat represents
the capacity of the uid to store thermal energy with low temperature gradients, hence it is responsible for the rate of liquid superheating at the wall which determines the local interphase mass
ux, see Eq. (13). Furthermore, from Eq. (13) it follows that the
mass of vapor created is proportional to the kinetic mobility. The
density ratio has a strong effect on bubble growth in volume, as
the lighter is the gas phase than the liquid which it replaces, the
more the liquid evaporated expands when it is converted into vapor. The R113 has the lowest kinetic mobility, 24 kg/m2s K against
58 kg/m2s K of R245fa and 111 kg/m2s K of R134a at the given
operating conditions. Thus its rate of vapor creation would be the
lowest one if all the uids share the same temperature eld. However, R113 has also the lowest liquid specic heat (943 J/kg K
against about 1400 J/kg K of the others), and thus it leads to the
highest liquid superheating and this partially balances the lower
kinetic mobility. What makes the R113 the fastest growing vapor
bubble is its higher liquid-to-vapor density ratio, which is three
times that of R245fa and six times R134a. Therefore, even though
at the end of the simulations R134a has the highest rate of mass
of vapor created, the evaporation of R113 generates the largest
bubble and it is the rst to reach the end of the channel.
7.2. Analysis of ow and temperature eld around the bubble
The temperature and ow elds obtained by the simulation of
Case 1 (uid R113, see Table 2) are analyzed, in order to investigate
the local patterns of the ow and the dominating heat transfer
mechanisms. The ow is captured at the time instant t = 12.5 ms,

before the bubble exits the ow domain. At this time instant the
velocity of the nose of the bubble is 1.07 m/s, the center of gravity
moves at 0.76 m/s and the bubble rear at 0.47 m/s. The axial location of the bubble nose at the center of the channel is z/D = 18.91
and the location of the rear is z/D = 12.43.
Fig. 12 depicts the contours and isolines of the velocity eld
within the heated region of the channel, together with the proles
of the axial and radial velocities averaged within the crosssectional area occupied by the liquid. Average values are evaluated
as:

umean z

2
2

R  R  dz

ur; zrdr

30

Rdz

where d(z) = R in the absence of the bubble. Isotherms and temperature contours are reported in Fig. 13 together with the wall heat
transfer coefcient estimated as expressed in Eq. (26). The single
phase heat transfer coefcient refers to the preliminary steady
state simulation run with only liquid. A black dashed line represents the thickness of the thermal boundary layer dT for the single
phase case computed as dT(z) = R  r(T = Tsat + 0.01(Tw(z)  Tsat)),
where the value 0.01 was arbitrarily chosen to identify the thermal
boundary layer as that region whose temperature exceeds the
saturation value by at least 1% of the local wall superheating
(Tw(z)  Tsat). Fig. 14 reports the two-phase boiling heat transfer
coefcient htp along the heated wall, relative to the local single
phase value hsp obtained as result of the preliminary single phase
simulation.
Four separate regions can be identied by observation of the
ow and temperature elds across the bubble and each region is
ruled by specic wall heat transfer mechanisms:

M. Magnini et al. / International Journal of Heat and Mass Transfer 59 (2013) 451471

465

Fig. 12. Average liquid axial and radial velocity (above) and contours of the velocity eld (below). The thick black solid line identies the bubble interface.

Fig. 13. Heat transfer coefcient (above) and contours of temperature eld (below). The thick black solid line identies the bubble interface. The black dashed line represents
the width of the thermal boundary layer for the single phase case.

wake region far away from the bubble, located approximately at


z/D = [8,10]. The isolines of velocity reported in Fig. 12 are horizontal within this region and the mean radial velocity of the
liquid is null, thus suggesting that the bubble disturbance on
the liquid ow is negligible. This hypothesis is conrmed by
the proles of axial velocity plotted in Fig. 15(a), which suggest
that deviations from a single phase ow become noticeable only
for z/D > 11, thus from one and a half diameters behind the bubble. Differently, the bubble inuence on the thermal eld is
noticeable up to the entrance in the heated region (z/D = 8).
The temperature contours reported in Fig. 13 show that the wall
thermal boundary layer is still restoring to the steady situation
that was holding before the bubble transited, thus during this

transient stage the thickness of the layer is smaller and the


liquid is cooler. Fig. 16 proves that at z/D = 10 the wall temperature is decreased by 1 K compared with the single phase case.
In this region the wall heat transfer is enhanced by the transient
heat convection mechanism, such that the improvement with
respect to the single phase case increases monotonically in
the streamwise direction, as shown in Fig. 14.
wake region nearby the bubble, located within z/D = [10,12.5].
The velocity isolines located at r/D < 0.35 converge on the channel axis, thus indicating that the liquid is slowing down along
the centerline of the channel. The undisturbed liquid axial
velocity around the channels axis is higher than that of the
bubble rear, and hence the bubbles presence slows down

466

M. Magnini et al. / International Journal of Heat and Mass Transfer 59 (2013) 451471

Fig. 14. Enhancement of the heat transfer coefcient given by the boiling twophase ow (subscript tp) relative to the single phase case (sp). The black dashed
lines identify the bubble rear and nose positions.

locally the liquid. As a consequence, the axial velocity nearby


the wall has to increase in order for the ow to maintain a constant liquid volumetric ow rate behind the bubble, and a at
velocity prole is generated. This is demonstrated by the proles of axial velocity in the wake reported in Fig. 15(a). At sections closer to the bubble rear, the liquid velocity prole
deviates from an expected parabolic to at prole and at
z/D = 12.4 the liquid velocity at the centerline of the channel
matches the bubble rear velocity (0.47 m/s). Note that the inter-

sections between at and undisturbed velocity proles behind


the bubble occur at r/D  0.35 at all the axial locations reported
in Fig. 15(a), in agreement with the observations of Gupta et al.
[67] for ow without phase change. The higher liquid velocity at
the wall delays the time-development of the thermal layer,
which appears thinner than upstream locations in Fig. 13. The
temperature proles depicted in Fig. 16 show that the wall temperature drops by more than 1 K. In this region the combined
effect of transient heat convection and at velocity prole augments the heat transfer coefcient up to 30% with respect to the
liquid-only case.
liquid lm region, z/D = [12.5,19]. In this region the bubble interface is in contact with superheated liquid and evaporation
occurs. The interfacial resistance to mass transfer is very low,
such that the interface always stays at the saturation temperature. The bubble interface squeezes the thermal boundary layer
against the channel wall and lm evaporation cools down
locally the superheated liquid to the saturation condition, and
thus the isotherms in Fig. 13 are more dense. The undulations
of the interface prole near the rear of the bubble create local
recirculation patterns, which make the liquid average velocities
oscillate around zero, but with values low in magnitude. Clockwise vortices are detected upon each valley while anticlockwise
vortices are evident at each crest of the interfacial wave.
Fig. 15(b) reports the proles of axial velocity within the liquid
lm and at z/D = 13, where an anticlockwise vortex occurs, the
axial velocity is negative. Moving downstream along the bubble,
the amplitude of the interfacial wave decreases and the shape of

Fig. 15. Proles of liquid axial velocity in (a) the bubble wake and (b) liquid lm.

Fig. 16. Proles of liquid temperature in the bubble wake. Solid line: two-phase ow, dashed line: single phase ow.

M. Magnini et al. / International Journal of Heat and Mass Transfer 59 (2013) 451471

467

Fig. 17. Proles of liquid temperature in the lm region. Solid line: two-phase ow, dashed line: single phase ow. Thick solid line: location of the bubble interface.

Fig. 18. Temperature (left) and axial velocity (right) proles at z/D = 19. Solid line: two-phase ow; dashed line (left): single phase ow; dashed line (right): laminar fully
developed ow with the same liquid ow rate.

the axial velocity proles within the thickening liquid lm is


determined by the negligible shear stress at the vapor-liquid
interface. The wall temperature is weakly affected by the
dynamics of the vortices existing at the bubble rear and the
enhancement of the heat transfer coefcient depicted in
Fig. 14 decreases smoothly toward the bubble nose. The temperature proles within the lm illustrated in Fig. 17 report that
the wall temperature decreases by less than 1 K compared with
the single phase case, thus conrming that the heat transfer
performance is worse than that measured in the wake region
nearby the bubble. The liquid lm evaporation is indeed the
heat transfer mechanism governing this region. The enhancement on the heat transfer coefcient is maximum at the bubble
rear (25%) where the lm thickness is minimum, and it drops as
the liquid lm thickens toward the nose. In the Thome et al. [4]
three-zone model, the heat transfer in the lm region is estimated by assuming one-dimensional heat conduction, the
validity of such an approximation for the simulations under
analysis is discussed in Section 7.3.
liquid region ahead of the bubble, z/D > 19. The liquid is strongly
accelerated by the evaporation such that the axial velocity prole plotted in Fig. 18 is attened with respect to a laminar fully
developed ow. The liquid ow rate is twice the value of the
preliminary single phase simulation, and therefore in Fig. 18 a
parabolic velocity prole with the same ow rate is reported
as reference. Convective heat transfer is only slightly inuenced
by the higher liquid ow rate and by the at velocity prole.

The heat transfer coefcient is only 3% higher than the


single phase case value and hence the liquid temperature
prole within the wall thermal boundary layer shifts only by
some tenths of degree, as reported in Fig. 18.
7.3. Heat transfer model for the lm region
In the Thome et al. three-zone model [4], the heat transfer coefcient in the lm region is obtained by assuming one-dimensional
steady-state heat conduction in the stagnant liquid lm deposited
between the bubble and the wall. The solution of the Fourier heat
conduction equation, by imposing saturation temperature at the
bubble interface and constant heat ux at the channel wall, leads
to the following linear temperature eld at a given axial location:

q
Ty; t T sat y
k

31

where y is a vertical coordinate, which ranges from 0 at the bubble


interface to d(t) at the channel wall. Fig. 19 shows the comparison
between temperature proles within the liquid lm given by Eq.
(31) and simulation results for Case 2, as the bubble is traveling
at zh/D = 5, with zh being the axial distance from the entrance in
the heated region. The elapsed time t is intended as the time interval from the transit of the bubble nose at the axial location observed. The theoretical model gives unsatisfactory results as it
generally underestimates the wall temperature by several degrees
Kelvin. By introducing the wall temperature expressed through

468

M. Magnini et al. / International Journal of Heat and Mass Transfer 59 (2013) 451471

Fig. 19. Proles of the liquid temperature within the lm region for the simulation Case 2, evaluated at zh/D = 5. Solid line with asterisks: simulation data; dash-dotted line
with diamonds: steady heat conduction model Eq. (31); dashed line with circles: transient heat conduction model Eq. (37); horizontal thick line: location of the bubble
interface.

Eq. (31) within the relationship (26), the following relationship for
the local heat transfer coefcient yields:

ht

k
dt

32

The consequence of the model errors in the wall temperature prediction is the overestimation of the heat transfer coefcient.
Fig. 20(a)(d) prove that the model leads to coefcients substantially higher than the present simulations, with time-averaged values more than 60% above for all the runs.
The three-zone model has been widely validated through comparison of the predicted time-averaged boiling heat transfer coefcient with several experimental databases. However, the original
three-zone model supposes a lm thickness of the order of 1/100
the channel size, while in the present simulations d/D  1/20.
Within such a thick liquid lm, the thermal inertia of the liquid
might be not negligible, because the time scale of the thermal phenomenon approaches that of the residence time of the liquid lm.

In this situation, a heat transfer model based on transient heat conduction might capture the local physics of the lm better than a
steady-state model. In order to test this hypothesis for the operating conditions simulated in cases 2 to 5, an extended version of the
heat transfer model which leads to the Eq. (31) is obtained by adding the transient term to the one-dimensional Fourier equation:

@T
@2T
at 2
@t
@y

33

solved with the boundary conditions:

Ty 0; t T sat
@T
 k y dt; t q
@y

34
35

and with the initial condition:

Ty; t 0 Fy

36

M. Magnini et al. / International Journal of Heat and Mass Transfer 59 (2013) 451471

469

Fig. 20. Time-evolution of the heat transfer coefcient for simulation runs 2 to 5. Solid line: simulation data; dash-dotted line: steady heat conduction model with actual
time-varying liquid lm thickness; dashed line: transient heat conduction model with constant lm thickness.

By considering a constant lm thickness d the thermal problem allows the following analytical solution [68]:
m
q
2X
2
Ty; t T sat y
eat bi t Y i y
k
d i1

Y i y0 Fy0 dy

37

where bi and Yi(y) are eigenvalues and eigenfunctions of the spatial


problem. The formulation of the mathematical problem in Cartesian
coordinates would be an approximation for a domain which actually is axisymmetrical. However, since the width of the liquid lm
is sufciently smaller than the channel radius for the cases under
analysis (d/D  1/20), the deviation between Eq. (37) and a solution
based on cylindrical coordinates is negligible. The axial location
zh/D = 5 is again considered to compare improved model and
numerical results for simulation Case 2. The initial temperature
F(y) is the steady temperature prole at zh/D = 5 that holds before
the bubble ows and it is obtained as a result of the preliminary
liquid-only simulation. The liquid lm thickness is not constant
while the bubble is crossing the axial location analyzed. Since the
analytical solution (37) requires a constant value of d, the average
width of the lm while the rear of the bubble is passing at
zh/D = 5 is considered as a representative value. The comparison is
displayed in Fig. 19. At t = 0.55 ms, the theoretical prole does not
match the simulation temperature near the interface due to the
higher width of the actual lm in the nose region and to the initial
temperature prole F(y) set. However, the wall temperature is
already well predicted, as the deviation is around 0.3 K. As time
goes by and the actual width of the lm thickness approaches the

value set in the model, the theoretical and numerical proles get
closer and then overlap partially after 2.81 ms. Afterwards, the
wavy prole of the bubble rear transits and the temperature prole
in the simulation follows the dynamics of the interface. The wall
temperature is little affected by local vortices and lm thickness
oscillations, and hence the wall temperature estimation remains
reliable in this region as well. The bubble rear crosses the axial location analyzed after about 4 ms. The transient heat conduction model suggests that quasi-steady-state conditions for the temperature
within the liquid lm would be reached only after 20 ms. This conrms that, for the operating conditions simulated here, the transient heat conduction is the dominant heat transfer mechanism.
However, Eq. (37) estimates that 0.3 ms would be sufcient for
the temperature eld to attain a steady condition in a lm ten times
thinner, and hence a steady heat conduction modeling might be satisfactory in that case.
Fig. 20(a)(d) show that the prediction of the heat transfer coefcient given by the extended model agrees very well with the results of the numerical simulations for Cases 25. Case 1 was not
studied as the channel is too short. At initial time-steps the proles
may differ due to the initialization of the temperature within the
model, but as the time elapses the enhancement of the heat transfer performance is well captured. The transient heat conduction
model gives time-averaged predictions of the heat transfer coefcient within the 5% of numerical results for Cases 2, 3, 4 and 9%
higher for Case 5. The reason of the systematic model overprediction is twofold. In the simulations, the value of the lm thickness

470

M. Magnini et al. / International Journal of Heat and Mass Transfer 59 (2013) 451471

set in the model is reached only at the latest stage of the bubble
passage, while before it is actually thicker such that the heat removal is less efcient. In addition, the transit of the bubble nose
(which identies t = 0) may not coincide with the beginning of
the heat-conduction-controlled stage for the ow and the energy
transport by heat convection might be dominant until the lm
thickness drops below a threshold value.

8. Conclusions
We have modied the commercial CFD solver ANSYS Fluent by
implementing a Height Function interface reconstruction algorithm and an evaporation model as external self-developed functions. This improved solver enabled a very detailed insight in the
bubble dynamics and local phenomena affecting the wall heat
transfer during ow boiling in a microchannel. The working conditions investigated in this work include three different refrigerant
uids, namely R113, R134a and R245fa, circular channels of
0.5 mm diameter, mass uxes ranging from 500 kg/m2s to
600 kg/m2s, heat uxes from 5 kW/m2 to 20 kW/m2 and saturation
temperatures of 31 C and 50 C. As the bubble enters the heated
region of the channel and comes in contact with the superheated
thermal boundary layer developing at the wall, the bubble nose
accelerates downstream following an exponential time-law, in
good agreement with a theoretical model available in the literature. Under similar operating conditions, the high liquid-to-vapor
density ratio of R113 leads to the most rapidly expanding bubble.
The bubble transit at a given axial location pushes liquid against
the microchannel wall and locally cools down the thermal boundary layer as an effect of the evaporation occurring along the liquid
lm region. The heat transfer coefcient increases monotonically
as the lm thickness falls, up to 25% of the liquid-only case near
the bubble rear, thus conrming that lm evaporation is the primary wall heat transfer mechanism. However, the enhancement
of the heat transfer performance remains considerable as well in
the wake region next to the bubble, as a consequence of the superposition of liquid-wall transient heat convection and attened liquid velocity prole. The latter prevents the wall thermal layer
from restoring to the steady situation that was holding before
the bubble transited. A heat transfer method for the liquid lm region based on steady one-dimensional heat conduction overpredicts the local values of the heat transfer coefcient measured in
the present simulations, as the thermal inertia of the liquid is not
negligible (D/d  20 in the present simulations). A new method
including thermal transient conduction within the mathematical
model is able to predict local heat transfer in remarkable agreement with numerical results for all the simulated runs. This model
might be used to extend the existing microchannel boiling heat
transfer methods to cover a wider range of diameter-to-lm-thickness ratios.

References
[1] B. Agostini, M. Fabbri, J.E. Park, L. Wojtan, J.R. Thome, B. Michel, State of the art
of high heat ux cooling technologies, Heat Transfer Engineering 28 (2007)
258281.
[2] R. Revellin, V. Dupont, T. Ursenbacher, J.R. Thome, I. Zun, Characterization of
diabatic two-phase ows in microchannels: Flow parameter results for R-134a
in a 0.5 mm channel, Int. J. of Multiphase Flow 32 (2006) 755774.
[3] P.A. Walsh, E.J. Walsh, Y.S. Muzychka, Heat transfer model for gas-liquid slug
ows under constant heat ux, Int. J. of Heat and Mass Transfer 53 (2010)
31933201.
[4] J.R. Thome, V. Dupont, A.M. Jabobi, Heat transfer model for evaporation in
microchannels. Part I: presentation of the model, Int. J. of Heat and Mass
Transfer 47 (2004) 33753385.
[5] S.V. Garimella, C.B. Sobhan, Transport in microchannels - a critical review,
Annu. Rev. Heat Transfer 13 (2003) 150.

[6] S.S. Bertsch, E.A. Groll, S.V. Garimella, Review and comparative analysis of
studies on saturated ow boiling in small channels, Nanoscale and Microscale
Thermophysical Engineering 12 (2008) 187227.
[7] J.R. Thome, Wolverine Engineering Data Book III, Available free at http://
www.wlv.com/products/databook/db3/DataBookIII.pdf, 2004.
[8] B. Agostini, J.R. Thome, Comparison of an extended database for boiling heat
transfer coefcients in multi-microchannels elements with the three-zone
model, in: ECI Heat Transfer and Fluid Flow in Microscale, Castelvecchio
Pascoli, Italy.
[9] M.G. Cooper, Heat ow rates in saturated nucleate pool boiling - A wideranging examination using reduced properties, Advances in Heat Transfer 16
(1984) 157239.
[10] B. Agostini, J.R. Thome, M. Fabbri, B. Michel, D. Calmi, U. Kloter, High heat ux
ow boiling in silicon multi-microchannels - Part II: Heat transfer
characteristics of refrigerant R245fa, Int. J. of Heat and Mass Transfer 51
(2008) 54155425.
[11] C.L. Ong, J.R. Thome, Macro-to-microchannel transition in two-phase ow:
Part 2 -Flow boiling heat transfer and critical heat ux, Experimental Thermal
and Fluid Science 35 (2011) 873886.
[12] T. Harirchian, S.V. Garimella, Flow regime-based modeling of heat transfer and
pressure drop in microchannel ow boiling, Int. J. of Heat and Mass Transfer 55
(2012) 12461260.
[13] V. Dupont, J.R. Thome, A.M. Jabobi, Heat transfer model for evaporation in
microchannels. Part II: comparison with the database, Int. J. of Heat and Mass
Transfer 47 (2004) 33873401.
[14] G. Ribatski, W. Zhang, L. Consolini, J. Xu, J.R. Thome, On the prediction of heat
transfer in micro-scale ow boiling, Heat Transfer Engineering 28 (2007) 842
851.
[15] L. Consolini, J.R. Thome, Micro-channel ow boiling heat transfer of R-134a, R236fa, and R-245fa, Microuid Nanouid 6 (2009) 731746.
[16] Y. Han, N. Shikazono, N. Kasagi, The effect of liquid lm evaporation on ow
boiling heat transfer in a microtube, Int. J. of Heat and Mass Transfer 55 (2012)
547555.
[17] P.A. Kew, K. Cornwell, Correlations for prediction of ow boiling heat transfer
in small-diameters channels, Applied Thermal Engineering 17 (1997) 705
715.
[18] C.L. Ong, J.R. Thome, Macro-to-microchannel transition in two-phase ow:
Part 1 -Two-phase ow patterns and lm thickness measurements,
Experimental Thermal and Fluid Science 35 (2011) 3747.
[19] R. Revellin, J.R. Thome, A new type of diabatic ow pattern map for boiling
heat transfer in microchannels, J. of Micromechanics and Microengineering 17
(2007) 788796.
[20] T. Harirchian, S.V. Garimella, A comprehensive ow regime map for
microchannel ow boiling with quantitative transition criteria, Int. J. of Heat
and Mass Transfer 53 (2010) 26942702.
[21] Y. Han, N. Shikazono, Measurement of the liquid lm thickness in microtube
slug ow, Int. J. of Heat and Mass Transfer 30 (2009) 842853.
[22] Y. Han, N. Shikazono, The effect of bubble acceleration on the liquid lm
thickness in micro tubes, Int. J. of Heat and Fluid Flow 31 (2010) 630639.
[23] M. Sussman, P. Smereka, S. Osher, A level set approach for computing solutions
to incompressible two-phase ow, J. of Computational Physics 114 (1994)
146159.
[24] C.W. Hirt, B.D. Nichols, Volume of uid (VOF) method for the dynamics of free
boundaries, J. of Computational Physics 39 (1981) 201225.
[25] V. Talimi, Y.S. Muzychka, S. Kocabiyik, A review on numerical studies of slug
ow hydrodynamics and heat transfer in microtubes and microchannels, Int. J.
of Multiphase Flow 39 (2012) 88104.
[26] A. Mukherjee, S.G. Kandlikar, Numerical simulation of growth of a vapor
bubble during ow boiling of water in microchannel, Microuid Nanouid 1
(2005) 137145.
[27] Y. Suh, W. Lee, G. Son, Bubble dynamics, ow, and heat transfer during ow
boiling in parallel microchannels, Numerical Heat Transfer, Part A 54 (2008)
390405.
[28] A. Mukherjee, Contribution of thin-lm evaporation during ow boiling inside
microchannels, Int. J. of Thermal Sciences 48 (2009) 20252035.
[29] A.M. Jacobi, J.R. Thome, Heat transfer model for evaporation of elongated
bubble ows in microchannels, J. of Heat Transfer 124 (2002) 11311136.
[30] A. Mukherjee, S.G. Kandlikar, Z.J. Edel, Numerical study of bubble growth and
wall heat transfer during ow boiling in a microchannel, Int. J. of Heat and
Mass Transfer 54 (2011) 37023718.
[31] Y.Q. Zu, Y.Y. Yan, S. Gedupudi, T.G. Karayiannis, D.B.R. Kenning, Conned
bubble growth during ow boiling in a mini-/micro-channel of rectangular
cross-section part II: Approximate 3-D numerical simulation, Int. J. of Thermal
Sciences 50 (2011) 267273.
[32] R. Zhuan, W. Wang, Flow pattern of boiling in micro-channel by numerical
simulation, Int. J. of Heat and Mass Transfer 55 (2012) 17411753.
[33] D. Juric, G. Tryggvason, Computations of boiling ows, Int. J. of Multiphase
Flow 24 (1998) 387410.
[34] G. Son, V.K. Dhir, Numerical simulation of nucleate boiling on a horizontal
surface at high heat uxes, Int. J. of Heat and Mass Transfer 51 (2008) 2566
2582.
[35] C. Kunkelmann, P. Stephan, CFD simulation of boiling ows using the Volumeof-Fluid method within OpenFOAM, Numerical Heat Transfer, Part A 56 (2009)
631646.
[36] S.J. Cummins, M.M. Francois, D.B. Kothe, Estimating curvature from volume
fractions, Computers and Structures 83 (2005) 425434.

M. Magnini et al. / International Journal of Heat and Mass Transfer 59 (2013) 451471
[37] G. Bornia, A. Cervone, S. Manservisi, R. Scardovelli, S. Zaleski, On the properties
and limitations of the height function method in two-dimensional Cartesian
geometry, J. of Computational Physics 230 (2011) 851862.
[38] M.M. Francois, S.J. Cummins, E.D. Dendy, D.B. Kothe, J.M. Sicilian, M.J. Williams,
A balanced-force algorithm for continuous and sharp interfacial surface
tension models within a volume tracking framework, J. of Computational
Physics 213 (2006) 141173.
[39] S. Popinet, An accurate adaptive solver for surface-tension-driven interfacial
ows, J. of Computational Physics 228 (2009) 58385866.
[40] R.W. Schrage, A theoretical study of interphase mass transfer, Columbia
University Press, New York, 1953.
[41] J.U. Brackbill, D.B. Kothe, C. Zemach, A continuum method for modeling
surface tension, J. of Computational Physics 100 (1992) 335354.
[42] G.L. Morini, Viscous heating in liquid ows in micro-channels, Int. J. of Heat
and Mass Transfer 48 (2005) 36373647.
[43] V.P. Carey, Liquid-vapor phase change phenomena, Taylor and Francis, 1992.
[44] R. Marek, J. Straub, Analysis of the evaporation coefcient and the
condensation coefcient of water, Int. J. of Heat and Mass Transfer 44 (2001)
3953.
[45] J.W. Rose, On interphase matter transfer, the condensation coefcient and
dropwise condensation, Proceedings of the Royal Society of London A 411
(1987) 305311.
[46] H. Wang, S.V. Garimella, J.Y. Murthy, Characteristics of evaporating thin lm in
a microchannel, Int. J. of Heat and Mass Transfer 50 (2007) 39333942.
[47] I. Tanasawa, Advances in condensation heat transfer, in: J.P. Hartnett, T.F.
Irvine (Eds.), Advances in Heat Transfer, Academic Press, San Diego, 1991.
[48] S. Hardt, F. Wondra, Evaporation model for interfacial ows based on a
continuum-eld representation of the source terms, J. of Computational
Physics 227 (2008) 58715895.
[49] D.L. Youngs, Time-dependent multi-material ow with large uid distortion,
in: K.W. Morton, M.J. Baines (Eds.), Numerical Methods for Fluid Dynamics,
Academic Press, 1982, pp. 273285.
[50] B. Lafaurie, C. Nardone, R. Scardovelli, S. Zaleski, G. Zanetti, Modelling merging
and fragmentation in multiphase ows with SURFER, J. of Computational
Physics 113 (1994) 134147.
[51] M. Malik, E.S.-C. Fan, M. Bussmann, Adaptive VOF with curvature-based
renement, Int. J. for Numerical Methods in Fluids 55 (2007) 693712.
[52] J. Hernandez, J. Lopez, P. Gomez, C. Zanzi, F. Faura, A new volume of uid
method in three dimensions-Part I: Multidimensional advection method with

[53]

[54]
[55]
[56]
[57]

[58]
[59]
[60]
[61]

[62]

[63]
[64]
[65]
[66]

[67]
[68]

471

face-matched ux polyhedra, Int. J. for Numerical Methods in Fluids 58 (2008)


897921.
B. van Leer, Towards the ultimate conservative difference scheme. V.A secondorder sequel to Godunovs method, J. of Computational Physics 32 (1979) 101
136.
Fluent 6.3 Users Guide, Fluent Inc., Lebanon, NH, 2006.
R. Gupta, D.F. Fletcher, B.S. Haynes, On the CFD modelling of Taylor ow in
microchannels, Chemical Engineering Science 64 (2009) 29412950.
R.I. Issa, Solution of the implicitly discretized uid ow equations by operatorsplitting, J. of Computational Physics 62 (1985) 4065.
M. Magnini, B. Pulvirenti, Height function interface reconstruction algorithm
for the simulation of boiling ows, in: A.A. Mammoli, C.A. Brebbia (Eds.),
Computational Methods in Multiphase Flows VI, Wessex Institute of
Technology, UK, 2011.
M.S. Plesset, S.A. Zwick, The growth of vapor bubbles in superheated liquids, J.
of Applied Physics 25 (1954) 493500.
L.E. Scriven, On the dynamics of phase growth, Chemical Engineering Science
10 (1959) 113.
R. Scardovelli, S. Zaleski, Direct numerical simulation of free-surface and
interfacial ow, Annual Review of Fluid Mechanics 31 (1999) 567603.
D.J.E. Harvie, M.R. Davidson, M. Rudman, An analysis of parasitic current
generation in Volume Of Fluid simulations, Applied Mathematical Modelling
30 (2006) 10561066.
M.T. Kreutzer, F. Kapteijn, J.A. Moulijn, C.R. Kleijn, J.J. Heiszwolf, Inertial and
interfacial effects on pressure drop of taylor ow in capillaries, A.I.Ch.E. Journal
51 (2005) 24282440.
P. Aussillous, D. Qur, Quick deposition of a uid on the wall of a tube, Physics
of Fluids 12 (2000) 23672371.
D. Liberzon, L. Shemer, D. Barnea, Upward-propagating capillary waves on the
surface of short taylor bubbles, Physics of Fluids 18 (2006).
S.V. Patankar, Numerical Heat Transfer and Fluid Flow, Hemisphere Publishing,
New York, 1980.
L. Consolini, J.R. Thome, A heat transfer model for evaporation of coalescing
bubbles in microchannel ow, Int. J. of Heat and Fluid Flow 31 (2010) 115
125.
R. Gupta, D.F. Fletcher, B.S. Haynes, CFD modelling of ow and heat transfer in
the Taylor ow regime, Chemical Engineering Science 65 (2010) 20942107.
M.N. Ozisik, Heat Conduction, John Wiley and Sons Inc., New York, 1993.

You might also like