International Journal of Heat and Mass Transfer 150 (2020) 119291

Contents lists available at ScienceDirect

International Journal of Heat and Mass Transfer

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

3-D computational investigation and experimental validation of effect

of shear-lift on two-phase flow and heat transfer characteristics of
highly subcooled flow boiling in vertical upflow
Jeongmin Lee, Lucas E. O’Neill, Issam Mudawar∗
Purdue University Boiling and Two-Phase Flow Laboratory (PU-BTPFL), School of Mechanical Engineering, Purdue University, 585 Purdue Mall, West
Lafayette, IN 47907, USA

a r t i c l e i n f o a b s t r a c t

Article history: The present study is focused on development of a 3-D computational approach to predict highly sub-
Received 23 August 2019 cooled nucleate flow boiling in vertical upflow. Investigation of existing computational methodology
Revised 11 December 2019
based on Volume of Fluid (VOF) approach revealed fundamental weaknesses in modeling multiphase
Accepted 29 December 2019
flows, which stems mostly from inadequate representation of shear-lift force on bubbles. A user defined
Available online 18 January 2020
function is adopted which provides detailed information relating to this important effect, and computed
Keywords: results are validated through comparison with experimental results and analytic predictions of single
Two-phase flow bubble trajectory. The computational method is then used to model the entire flow field for subcooled
Subcooled flow boiling flow boiling in a rectangular channel heated on two opposite walls, and predictions are validated against
Shear-lift force FC-72 experimental data for four different mass velocities. Overall, computationally predicted interfa-
CFD cial behavior, flow pattern, and heat transfer parameters (wall temperature and heat transfer coefficient)
show good agreement with experimental data. The model is also shown highly effective at predicting
local phenomena (velocity and temperature profiles) not easily determined through experiments.
© 2020 Elsevier Ltd. All rights reserved.

1. Introduction from devices with high-flux surfaces by phase change from liq-
uid to vapor, and the latter involves heat rejection to the ambient
1.1. Application of two-phase flow boiling for thermal management by returning the vapor to liquid state. The appeal of phase change
methods is derived mostly from the coolant’s natural ability to uti-
While developments in modern cutting-edge technologies have lize its both sensible and latent heat.
led to many performance benefits, these benefits were achieved at Compared to condensation, boiling often represents the greater
the expense of greatly increased amount of dissipative heat. This challenge for design engineers. This is due to greater difficulty in
is especially the case with electronic and power devices, which are controlling pressure drop and heat transfer coefficient to obtain
found in such applications as supercomputers, electronic data cen- stable operation, often related to presence of instabilities in the
ters, medical equipment, advanced military radar, directed energy system [3–5]. Furthermore, over-intense boiling can lead to criti-
lasers, aircraft and spacecraft avionics, hybrid vehicle power elec- cal heat flux (CHF), where vapor occlusion of heated surfaces leads
tronics, and energy storage system (ESS) batteries [1,2]. High power to rapid temperature rise capable of destroying whole systems.
densities and dissipative heat fluxes encountered in these applica- CHF is inarguably the most important design factor and safety
tions mean single-phase liquid cooling methods are no longer ef- parameter for all high heat flux controlled boiling applications.
fective, as they cannot fulfill cooling requirements across operating Overheating beyond CHF greatly reduces cooling performance,
conditions. causing the device being cooled to physically burn out or melt
As thermal management has become an enabling technology down. In applications avoiding this risk, however, boiling is a very
for many modern applications, diverse two-phase methods have effective approach to cooling modern devices due to its ability
been proposed, utilizing boiling and its reciprocal process of con- to greatly increase heat transfer coefficient, thereby decreasing
densation. Of these two, the former deals with heat acquisition surface temperature for a given heat flux, helping minimize sys-
tem weight and volume, and requiring relatively small amounts of
coolant compared to single-phase liquid cooling. Also, depending
∗
Corresponding author. on boiling regime and operating conditions, devices can be main-
E-mail address: [email protected] (I. Mudawar). tained at a fairly uniform temperature near the coolant’s saturation

https://doi.org/10.1016/j.ijheatmasstransfer.2019.119291
0017-9310/© 2020 Elsevier Ltd. All rights reserved.
2 J. Lee, L.E. O’Neill and I. Mudawar / International Journal of Heat and Mass Transfer 150 (2020) 119291

Nomenclature g vapor phase

i initial bubble condition; index for phase
a half of bubble diameter sat saturation
CL shear-lift coefficient sub subcooling
cp specific heat at constant pressure
c mesh (cell) size
d diameter for circular tubes temperature. Because of these practical merits, boiling schemes
E energy per unit mass have been target of extensive investigations at the Purdue Uni-
F force versity Boiling and Two-Phase Flow Laboratory (PU-BTPFL) and
Fb buoyancy force other organizations worldwide. Configurations considered include
Fcp contact pressure force capillary-pumped [6], pool boiling [7–9], macro-channel flow boil-
Fdu unsteady drag force ing [10–12], mini/micro-channel flow boiling [13,14], spray [15–17],
Fh hydrodynamic pressure force jet impingement [18,19], and hybrid cooling schemes [20].
Fqs quasi-steady drag The present study is focused entirely on channel flow boiling.
Fs surface tension force Presently, predicting two-phase flow and heat transfer in chan-
Fsl shear-lift force nels relies mostly on empirical correlations (e.g., [21–25]) and, to a
G mass velocity lesser extent, theoretical models [26,27]. For thermal design engi-
Gs dimensionless shear rate neers in industry, correlations are highly desirable because of their
g gravitational acceleration ease of use, however, they are valid only for the specific fluid(s)
H height of flow channel’s cross-section and operating ranges of databases upon which they are based.
h heat transfer coefficient
hfg latent heat of vaporization 1.2. Utilization of computational methods for prediction of two-phase
k thermal conductivity flow and heat transfer
keff effective thermal conductivity
L total length of computational domainin individual The present study focuses on two-phase boiling flow and heat
references (Table 3) transfer characteristics in vertical upflow. Key points are to pre-
Ld development length of flow channel in experiment dict boiling regime and heat transfer coefficient under given oper-
Le exit length of flow channel in experiment ating conditions. To overcome the limitations of empirical correla-
Lentrance entrance length of computational domain tions, computational approaches have gained increased popularity
Lexit exit length of computational domain for thermal system design. Computational approaches have evolved
Lh heated length of flow channel over the past few decades along with improvements in comput-
m˙ mass transfer rate ing performance of computer hardware. Computational methods
P pressure have shown credible success in predicting single-phase fluid flow
q heat flux and heat transfer in a variety of flow configurations and operat-
Re Reynolds number ing environments, verified by agreement with experimental data.
ri mass transfer intensity factor Over time, significant improvements in numerical schemes for en-
Sh volumetric energy source hanced accuracy, stability and computation speed have been made
T temperature for complicated single-phase flows.
Tin temperature at channel inlet Computational prediction of two-phase flow involving phase
t time change is still challenging, but continuing research is closing the
ts thickness of solid wall in computational domain gap with single-phase flow. One key facet of predicting two-phase
Tsub inlet subcooling flow is to track the interface between liquid and gas phases as be-
Ur relative velocity havior of each phase and interaction along interface define flow
u velocity regime that in turn affects heat transfer. In the case of boiling,
uτ friction velocity the behavior of vapor bubbles (i.e., bubble departure, deformation,
V volume coalescence, and breakup) can determine the stage in the boiling
W width of flow channel’s cross-section process, and each stage has distinct heat transfer characteristics.
y coordinate normal to wall A recent review article by Kharangate and Mudawar [28] intro-
y+ dimensionless distance perpendicular to the wall duced several interface tracking methods that have shown success
z axial direction coordinate in numerical modeling of multi-phase flow. Lagrangian and Eule-
rian methods are two such schemes. The former provides accu-
Greek symbols rate and smooth interfaces between phases, but it is complex and
α volume fraction; void fraction computationally expensive because of the need to remesh every
θa advancing angle iteration. This method is suitable for modeling a single bubble or
θi inclination angle film-wise condensation with a simple flow configuration. On the
θr receding angle other hand, Eularian methods are relatively less complicated and
μ dynamic viscosity expensive, but show difficulty in clearly capturing interface (al-
ν kinetic viscosity though some interface reconstruction schemes compensate for this
ρ density weakness). Even Eulerian methods, however, require high perfor-
φ fluid property mance computing resources if full scale, realistic boiling problems
(e.g., full micro-channel heat sinks) are to be simulated.
Subscripts Another key factor in proper computational prediction of two-
b bubble phase flows is modeling the phase change mechanism. In general,
f liquid phase phase change models are combined two-phase schemes that track
or capture interface position and evaluate mass transfer across the
 J. Lee, L.E. O’Neill and I. Mudawar / International Journal of Heat and Mass Transfer 150 (2020) 119291 3

interface. Selection of appropriate mass transfer model is critical stability in mass conservation when tackling multi-phase flow with
to accurate computational predictions and will be discussed in phase change.
greater detail in following sections. In spite of these merits, VOF model often provides poor pre-
The ability of the selected computational approach to fully cap- diction of flows with a discrete fluid in a continuous fluid (e.g.,
ture crucial interfacial dynamics and heat/mass transfer at in- bubbly flow or mist/droplet flow) due to the absence of momen-
terfaces largely determines its ability to predict bulk flow phe- tum equations for each phase. As a single set of momentum equa-
nomenon including flow regime and heat transfer. These small- tions is shared by phases, forces acting at/across interfaces cannot
scale, local calculations of mass, momentum, and energy trans- be precisely calculated due to less accurate estimation of relative
fer across interfaces are also common causes of numerical diver- velocities between phases. Inaccurate force prediction for interac-
gence if not configured properly for given boundary conditions. tion between different phases leads to errors in predicting bubble
Many studies have focused on computational/numerical prediction detachment or lift-off (particularly in high-velocity convective boil-
of nucleate boiling (ONB) [29–31], slug flow [32], and annular flow ing where these effects are critical), causing severe errors regarding
[33] using computational modeling, with differing degrees of suc- bubble motion along the flow direction. These errors become even
cess predicting important design parameters such as wall tempera- worse in the case of a few dispersed bubbles surrounded by con-
ture, heat transfer coefficient, and void fraction along flow channel. tinuous fluid and large difference in velocities between phases. In
Despite the proliferation of recent works regarding computa- a previous study by the current authors relating to 2-D flow boil-
tional phase change modeling, many difficulties remain in select- ing simulations [34], it was clearly shown that streamlines pen-
ing and utilizing proper sub-models (selecting and implementing a etrate across interfaces and there is no stagnation point around
proper mass transfer model is particularly difficult depending on bubbles, as seen in Fig. 1. Hence, forces acting on bubbles can be
operating conditions under investigation). The following sections under-/over-predicted, which causes pre-CHF at high heat flux con-
will discuss pros and cons of different approaches and outline the ditions as a result of wrong prediction of bubble lift-off. One of the
methodology adopted in the present study. missing forces relating to the bubble lift-off is a transverse force
called shear-lift force which a particle experiences when moving
in a shear flow [35]. The influence of shear-lift force on lateral mi-
1.3. Choice and limitations of volume of fluid (VOF) method, and gration in shear flow for rigid spherical particles was first demon-
need for 3-D modeling strated experimentally by Segre and Silberberg [36]. Because of
this effect, these particles have a certain equilibrium position in
Of the two popular numerical methods for multi-phase flow, a tube, irrespective of the injected position. Similarly, treatment
Eulerian has been more commonly used to track liquid-vapor in- of the shear-lift force governing a bubble’s lateral migration may
terface due to the practicality of implementation. The most popu- have a profound effect on improving results for continuum flow
lar of Eulerian methods is the volume-of-fluid (VOF) model, which model computations used to predict large-scale dispersed liquid-
solves a continuity equation for the volume fraction of each phase gas flows.
and a single momentum equation throughout computational do- In addition to insufficient treatment of interfacial forces by cur-
main. Theoretically, volume fraction is a continuous function in rent computational approaches, there is the possibility of large dis-
time and space and only applicable in the case where each phase crepancies between 3-D and 2-D boiling simulations not addressed
is not immiscible. VOF model is comparatively simpler, less time- in most prior works. As 3-D approaches fully consider all bound-
consuming, and more stable in tackling mass conservation than an- ing walls of a flow channel, predicted shear stress is more accu-
other common approach called the level-set (LS) model, though it rate, affecting bubble behavior near the walls. Further differences
is less accurate in capturing interface topology. VOF model can also between 2-D and 3-D simulations are attributable to differences in
be easily combined with phase change models computing mass bubble shape, with bubbles close to sphere/semi-sphere in 3-D as
transfer between phases incorporated in the continuity equation, opposed to cylinder or truncated cylinder in 2-D. This means area
while LS model suffers numerical divergence associated with in- in contact with liquid phase will be different, affecting the amount

Fig. 1. Streamlines around vapor bubble in vertical upflow for G = 836.64–833.14 kg/m2s, q = 191–246 kW/m2 and ATsub = 30.81–31.24 °C.
4 J. Lee, L.E. O’Neill and I. Mudawar / International Journal of Heat and Mass Transfer 150 (2020) 119291

of phase change, which is a key factor of two-phase cooling. More- 3) Illustration of the inability of current VOF formulation to
over, unpredicted shear stress from sidewalls of a rectangular flow capture interfacial behavior. A new approach to better ac-
channel in 2-D tends to stimulate pre-CHF as shear-lift force from count for interfacial forces (primarily the shear-lift force)
sidewalls (captured in 3-D simulations) pushes bubbles towards will be outlined, and its benefits over the unmodified ap-
the channel center and aids liquid replenishment along the heated proach detailed.
walls. In 2-D computations, absence of this extra shear effect leads 4) Comparison of 3-D predictions (accurately accounting for in-
to premature predictions of CHF. Thus, 3-D boiling simulation and terfacial forces) and prior 2-D results with experimental re-
accurate modeling of forces acting on bubble are crucial in the case sults to establish the new 3-D approach as state-of-the-art
of relatively large-scale computational domain with high heat flux for prediction of flow boiling in macro-channels.
conditions.

2. Experimental methods
1.4. Objectives of study
Experimental flow boiling results are used throughout the
The present study is a continuing part of NASA’s Flow Boiling present study to offer validation of computational predictions.
and Condensation Experiment (FBCE), an ongoing collaboration be- Thus, it is relevant to provide a brief summary of experimental
tween NASA Glenn Research Center and the Purdue University Boil- methods employed in gathering results.
ing and Two-Phase Flow Laboratory (PU-BTPFL). The primary goal The test section used in the current work is the Flow Boil-
of this initiative is to develop an experimental facility for the In- ing Module (FBM), part of NASA’s Flow Boiling and Condensa-
ternational Space Station (ISS) capable of gathering long term flow tion Experiment (FBCE). Fig. 2(a) provides an exploded view of
boiling and flow condensation data in microgravity that can be its construction, highlighting how it is comprised of three polycar-
contrasted with data amassed from terrestrial experiments in or- bonate plates compressed between two aluminum support plates.
der to develop predictive tools (empirical, analytical and computa- The flow channel is milled out of the central polycarbonate piece,
tional) capable of accurately accounting for effects of body force on which is polished to optical-grade clarity, allowing detailed high-
two-phase flow and heat transfer in channels. A summary of sci- speed flow visualization measurements to be made.
entific developments realized as a part of the project can be found Fig. 2(b) provides a schematic of the flow channel. The chan-
in a recent review article by Mudawar [37], including details of nel is rectangular with 2.5-mm width and 5-mm height, providing
extensive prior flow boiling testing performed on ground and in a 3.33-mm hydraulic diameter (meaning the test section is best
parabolic flight [38,39]. described as a macro-channel). A 32.79-cm developing length is
The present study uses both experimental and computational located just downstream of a flow straightener near the channel
results to explore fluid flow and heat transfer characteristics for inlet, provided to ensure hydrodynamically fully developed flow
highly subcooled vertical upflow of FC-72. The present computa- at the inlet of the heated length. The 11.46-cm heated length is
tional approach offers significant improvements over prior work comprised of two copper slabs recessed into the right and left
[34] by incorporating (1) modeling of shear-lift force acting on walls of the channel, each with six resistive heaters soldered to
bubbles in forced convective boiling, and (2) simulating the full 3- the back. Direct fluid temperature measurements are made using
D computational domain (including both fluid and solid sections type-E thermocouples just prior to the flow straightener (module
for conjugate heat transfer). Conditions for four different mass ve- inlet) and at the end of the 6.09-cm exit length downstream of
locities and heat fluxes are examined. Shear-lift force acting on va- the boiling region (module exit).
por phase as well as the method of identifying effective bubble size Fig. 2(c) showcases the construction and instrumentation of the
influenced by this force are provided using a user-defined function heated length, with type-E thermocouples mounted in the copper
(UDF). Validation of the effect of shear-lift force on bubble behav- slab just upstream and downstream of consecutive resistors (the
ior in computational modeling is performed by comparing predic- positions of these thermocouples are important as their measure-
tions for single bubble trajectory in a flow channel with experi- ments will be compared directly to computational predictions of
ments by Nahra and Motil [40]. Two other shear-lift models devel- wall temperature in later sections). As mentioned previously, opti-
oped by Mei and Klausner [41] and Legendre-Magnaudet [35] are cally polished polycarbonate sidewalls are left unobstructed to al-
also numerically investigated. Vapor void fraction is used to vali- low for simultaneous heat transfer and flow visualization measure-
date set-up of phase change model relating to the mass transfer in- ments.
tensity factor incorporated in computational analysis as dominant Finally, Fig. 2(d) provides a schematic of the flow loop used
interfacial diffusion that can arise from advection in 3-D. Tempo- for experiments. Working fluid FC-72 (a dielectric manufactured by
ral and spatial averaging method is employed to predict the wall 3 M Company) is circulated using an Ismatech MCP-z magnetically
temperature and heat transfer coefficient. Overall, this study in- coupled gear pump and first passes through a filter to remove any
vestigates computationally important characteristics of flow boil- particulates. A turbine flow meter is used to measure flowrate, af-
ing which cannot be obtained from experiment, including void ter which the fluid enters a Cast-X circulation heater (pre-heater)
fraction, velocity and temperature distributions for both along and used to set inlet conditions at the test section.
across flow channel. The test section is mounted on a rotating platform to allow
Key goals for the present work are: for testing in different orientations, although all results presented
in the current work correspond to vertical upflow orientation.
1) Investigate benefits of scaling-up from a 2-D to a 3-D com- A variac is used to control power supplied to the heated walls,
putational domain, evaluated in terms of accurate predic- and simultaneous flow visualization images are captured at 20 0 0
tion of flow boiling hydrodynamics and corresponding heat frames per second (fps) with a pixel resolution of 2040 × 174
transfer. spanning the entire 11.46-cm heated length.
2) Further outline proper use of Lee mass transfer model Exiting the test section, flow passes through an air-cooled
[42] for subcooled flow boiling simulations, focusing on the Lytron heat exchanger, used to return the working fluid to single-
use/non-use of identical values of mass transfer intensity phase liquid, prior to passing an accumulator (used to compensate
factor (ri ) for interfacial evaporation and condensation cal- for volume changes associated with vapor production) and return-
culations. ing to the pump.
 J. Lee, L.E. O’Neill and I. Mudawar / International Journal of Heat and Mass Transfer 150 (2020) 119291 5

Fig. 2. (a) Exploded view of flow boiling module (FBM). Schematics of (b) FBM fluid path and (c) heated wall temperature measurement locations. (d) Schematic of flow
loop.
6 J. Lee, L.E. O’Neill and I. Mudawar / International Journal of Heat and Mass Transfer 150 (2020) 119291

All data collection is handled by a National Instruments NI addition, to consider turbulence effects in flow boiling, the two-
SCXI-10 0 0 data acquisition system controlled by LabView. Type-E equation Shear-Stress Transport (SST) k-ω turbulence model with
thermocouples with an accuracy of ±0.5 °C are used to measure turbulence dampening is used. This model is a combination of k-
fluid and heated wall temperatures throughout the system. Pres- ω and k-ε models, applying the former in the near-wall region
sure measurements are made using transducers with an accuracy and the latter for free-stream flow (with a blending function ap-
of ±0.1% and pressure drop accuracy of ±0.2%. The turbine flow plied between them). It has been observed the SST k-ω turbulence
meter has an accuracy of ±0.1%, and wall heat input is measured model shows improved prediction of temperature gradients near
with an accuracy of ±0.5 W. Overall uncertainty in determining the liquid-vapor interfaces, likely due to its treatment of turbulence
heat transfer coefficient is ±8%. dampening in this region. Use of the k-ω model in the near-wall
For additional details on fluid components used, measurement region allows consideration of low Reynolds effects and shear flow
techniques employed, and operating conditions tested, the reader spreading, capturing most physical phenomena of present in two-
is advised to consult a prior experimental study [43]. phase flows [44,45]. The use of another two-equation turbulence
model, such as the k-ε model, was avoided as this offers less accu-
3. Computational methods racy in the near-wall region. High-order turbulence models, while
potentially more accurate, were also avoided as they incur high
3.1. Governing equations computational expense. Overall numerical details and discretiza-
tion methods used in the present study are similar to ones ap-
A 3-D, transient, VOF approach is adopted in ANSYS FLUENT for plied in recent computational modeling of 2-D flow boiling by the
tracking interfacial behavior during flow boiling with double-sided present authors [34], with key differences being the 3-D domain
heated walls. Within this method, volume fractions of each phase and inclusion of shear-lift force in conservation of momentum in
are calculated for every computational cell and the sum of volume the present work.
fractions of phases must equal unity. In order to obtain a sharp
interface between liquid and vapor phases, spatial discretization 3.2. Shear-lift force
scheme Geo-Reconstruct is explicitly used with interfacial anti-
diffusion treatment (only applied to interfacial cells to suppress Klausner et al. [46] and Zeng et al. [47] provided detailed anal-
the excess numerical diffusion that can arise from 3-D advection ysis of forces acting on a bubble in directions parallel and normal
effects and high aspect-ratio cells). Body force is implicitly consid- to the heated surface to understand the influence of shear flow
ered to take into account vapor detachment from and liquid re- conditions on bubble detachment in flow boiling. Investigation of
plenishment to the heated surface. It is assumed the flow is incom- shear-lift force and wake effect acting on single and multiple bub-
pressible and properties of each phase are constant with temper- bles (with sizes much larger than bubbles in the current study)
ature. Continuity of the volume fraction (including mass transfer in shear liquid flow was conducted computationally by Rabha and
between phases) is expressed as Buwa [48], showing good prediction of bubble trajectory, change in
bubble shape, and vorticity distribution for rising dispersed bub-
∂α f   1  
+ ∇  αf u
f = m˙ gf − m˙ fg for liquid phase (1) bles. As bubble growth and detachment are important to the boil-
∂t ρf ing process, force balance on bubbles should be carefully evalu-
and ated. As discussed in the introduction, however, forces generated
by differences in phase velocities cannot be accurately evaluated
∂ αg 1  
+ ∇  (αg u
g ) = m˙ f g − m˙ g f for vapor phase, (2) because of limitations characteristic of the VOF method. As illus-
∂t ρg trated in Fig. 3, force balance for single bubble on vertical wall is
where α , u , m˙ , and ρ are, respectively, volume fraction, velocity given as
vector, mass transfer rate, and density. Subscripts f and g refer to  duy
liquid and vapor, respectively. Fy = Fs,y + Fdu,y + Fsl + Fh + Fcp = ρgVb
dt
Momentum and energy equations are solved throughout the
for forces normal to the surface, (6)
computational domain, with properties density, ρ , dynamic viscos-
ity, μ, and thermal conductivity, keff , depending on local volume and
fraction. Momentum and energy equations are given by  duz
Fz = Fs,z + Fdu,z + Fqs + Fb = ρgVb
∂    dt
(ρ u ) + ∇  (ρ u u ) = −∇ P + ∇  μ ∇ u + ∇ u T + F , (3) for forces parallel to the surface.
∂t (7)
and In the above equations, Fs is surface tension force, Fdu unsteady
∂   drag force (due to transient and asymmetrical bubble growth), Fsl
(ρ E ) + ∇  (u (ρ E + P ) ) = ∇  ke f f ∇ T + Sh , (4) shear-lift force produced by the velocity gradient, Fh hydrodynamic
∂t
pressure force, Fcp contact pressure force (due to pressure differ-
where P and E refer to pressure and energy per mass. All proper- ence between inside and outside of the bubble at the reference
ties used in above equations are determined according to point), Fqs quasi-steady drag in the flow direction, Fb buoyancy

φ= αi φi , (5) force, and Vb volume of bubble. Also in Fig. 3, θ a , θ r , and θ i are,
respectively, advancing, receding, and inclination angle. In this for-
where φ is the fluid property evaluated using a phase-weighted mulation, dynamic effects of turbulence and interfacial waviness
average of liquid and vapor values. The Continuum Surface Force are ignored.
model (CSF) [44] is used for surface tension force as a part of Among these forces, shear-lift force induced by shear rate of
the source term F in the momentum equation. Similarly, the ex- oncoming flow is carefully examined in the present study as it
tra energy transfer from boiling and condensation is incorporated is poorly predicted by the VOF method and found to affect bub-
as a source term, Sh , in the energy equation. It is defined as the ble lift-off significantly. In addition, for a typical bubble of around
product of mass transfer rate and latent heat, expressed as m˙ hfg . 0.3 mm, shear-lift force is of the same order of magnitude as sur-
A shear-lift force acting on only the vapor phase is also included face tension and unsteady drag forces (occurring due to asymmet-
in the present formulation, which is discussed in next section. In rical bubble growth which acts opposite to surface normal), and
 J. Lee, L.E. O’Neill and I. Mudawar / International Journal of Heat and Mass Transfer 150 (2020) 119291 7

Fig. 3. Schematic of forces acting on single bubble on the heated surface in shear
flow.

Fig. 4. Flow chart for computing and applying shear-lift force.

shear-lift has significant role in bubble detachment while the other
forces play lesser roles [46]. For this study, models developed for
a solid sphere in slow shear flow [49], a bubble in inviscid flow average vapor bubble diameter (which changes in time) through-
with weak shear [50], or in viscous linear shear flow [35] were out the computational domain, updated every iteration according
excluded from analysis. To improve prediction of bubble behavior to the ratio of total vapor volume within the domain to interfacial
and thus heat transfer, the shear-lift force formulation derived by area.
Mei and Klausner [41], considering local shear strain rate, is em- Another challenge is to identify the relative velocity between
ployed as a source term in the momentum equation using a user- oncoming flow and bubble motion from the computational simu-
defined function (UDF) in ANSYS FLUENT. It should be noted that lation. As fluid flow develops and nucleate boiling occurs, velocity
they modified Saffman’s model [49] to tackle vapor bubbles and, vectors of oncoming liquid flow and the vapor motion within the
combined with Auton’s study [50], developed an interpolation for bubble are not in the same direction, and they continue to change
the shear-lift force over a wide range of Reynolds number. Shear- every iteration. Intense computational resources are required to
lift coefficient in their formulation is expressed as perform the necessary integration and vector calculation for every
Fsl
  m 1/m single cell, so similar to that done in determining bubble diameter,
CL = = 3.877G1s /2 × Re−m/ 2
+ 0.344G1s /2 , the difference between constant liquid velocity and vapor velocity
1
2
ρ 2
f Ur π a2 b
varying with space and time is used for relative velocity.
with m = 4, (8) The detailed computing procedure for the UDF is described in
where the flow chart illustrated in Fig. 4. As subcooled liquid initially
dUr a passes through the flow channel, shear lift force plays no role at
Gs = , (9) the beginning. Once phase change occurs, this algorithm identifies
dy Ur
cells occupied by vapor phase whose volume fraction is greater
Gs is the dimensionless shear rate based on relative velocity, than or equal to a specific reference value of 0.5 and obtains to-
Ur = Uf – Ug , between the vapor bubble and liquid, Reb is bub- tal vapor volume within the entire domain through summation
ble Reynolds number based on the characteristic length of bubble of cells with vapor. It also sums all face areas where vapor con-
diameter, and a is half the bubble diameter. tacts the thermally conjugated surface between fluid and solid
In order to apply these formulations to CFD, determining bubble cells. These two values are saved into allocated memory, recalled
diameter is most crucial for the current flow boiling application. It by the UDF, and used to calculate the mean diameter of vapor
is both computationally complex and expensive to determine the every numerical iteration by dividing the total volume of vapor
diameter of every bubble along the flow channel as bubble size cells by the total surface area of vapor. Shear-lift force is then ob-
varies with time and space. In nucleate flow boiling, tiny bubbles tained according to Eq. (8). It should be noted that shear-lift force
are generated, and they get bigger downstream because of indi- is only applied to the vapor cells within the velocity boundary
vidual bubble growth and coalescence between bubbles. Further, layer (based on the velocity profile) because of the large shear
as each bubble deforms in anisotropic fashion during flow boiling, rate in the boundary layer. This is necessary to prevent overes-
it is hard to identify exact bubble diameter for every single bub- timation of shear-lift force, as overestimated positive and nega-
ble. Because of this complicating factor, the present study uses an tive shear-lift forces would tend to squeeze the bubble and change
8 J. Lee, L.E. O’Neill and I. Mudawar / International Journal of Heat and Mass Transfer 150 (2020) 119291

Fig. 5. Schematics of 3-D computational domain and mesh of cross section area.

its shape computationally to thin and long strings, which is not

physically correct. The shear-lift force UDF is not applied if vapor
cells are located beyond the velocity boundary layer (close to the Fig. 6. (a) Grid independence test based on spatially and time-averaged wall tem-
core region). Validation of this methodology will be discussed in perature, (b) Axial variation of heated wall y +.
Section 4, performed through comparison with single-bubble ex-
periments by Nahra and Motil [40].
Grid independence is verified by averaging wall temperature
3.3. Computational domain and initial/boundary conditions
over the entire copper heating walls for four different grid sizes,
comparing the spatial and time averaged value with experimental
Fig. 5 shows the computation domain used in the present study.
data after reaching computational steady-state. For grid indepen-
This domain is three-dimensional and full sized, simulating ver-
dence testing, intermediate mass velocity of G = 453.16 kg/m2 s
tical upflow boiling in a rectangular channel with cross sectional
and heat flux of q = 143.24 kW/m2 are applied as bound-
area of 2.5 mm × 5 mm (W × H) and uses actual dimensions
ary conditions. As shown in the authors’ prior study involv-
of copper heating walls used in the experimental flow boiling
ing 2-D simulation [34], near-wall cell size below 14-μm ex-
module. The length of domain is extended slightly to provide an
hibited grid-independent results, so cell sizes of c = 5.5, 6.5,
adiabatic entrance length upstream (Lentrance = 5 mm) and exit
7.5, 9.5 μm near wall are tested for the present 3-D simulation.
length downstream (Lexit = 10 mm), done to eliminate any poten-
Fig. 6(a) shows asymptotic convergence of average wall tempera-
tial numerically induced entrance/exit effects. Two solid regions of
ture achieved for near-wall cell size below around 7.5 μm. A cell
2.5 mm × 1.04 mm × 114.6 mm (W × ts × Lh ), representing cop-
size of c = 6.5 μm near wall is selected in the present study
per heating walls, are applied on the right and left of the fluid re-
to provide peak accuracy with minimum computing time. As the
gion to simulate conjugate heat transfer from solid to fluid. Mesh-
present case corresponds to turbulent flow, it is also important to
ing is done by ANSYS ICEM CFD, with quadrilateral mesh adopted
evaluate non-dimensional distance, y+ , from the wall based on the
for the entire domain. Non-uniform mesh sizing is used with re-
selected cell size. It is defined as
finement near walls to capture phase change starting from vapor
embryos, bubble formation, and accurately predict shear stress and yuτ
fluid-heat transfer interaction in the viscous sublayer. y+ = , (10)
ν
 J. Lee, L.E. O’Neill and I. Mudawar / International Journal of Heat and Mass Transfer 150 (2020) 119291 9

Table 1
Mass velocities and thermophysical properties used in computational model.

G Tsat hfg ρf Cp f kf μf ρg C pg kg μg σ
(kg/m2 .s ) (◦ C ) (J/kg mol ) (kg/m3 ) (J/kg × K) ( W/m · K ) (kg/m · s ) (kg/m3 ) (J/kg · K ) ( W/m · K ) (kg/m · s ) ( N/m )
176.52 61.97 2.775 × 107 1608.8 1117.1 0.0536 3.858 × 10−4 15.874 942.06 0.0142 1.209 × 10−5 0.0080
445.75 60.16 2.761 × 107 1605.2 1120.1 0.0534 3.786 × 10−4 16.591 946.91 0.0143 1.215 × 10−5 0.0079
836.64 62.20 2.974 × 107 1608.2 1117.6 0.0536 3.846 × 10−4 15.992 942.87 0.0142 1.210 × 10−5 0.0080
2453.51 69.37 2.701 × 107 1589.6 1133.1 0.0527 3.497 × 10−4 19.953 967.97 0.0149 1.243 × 10−5 0.0073

Table 2
Numerical details and discretization methods.

Pressure-velocity coupling Pressure-implicit with splitting of operators (PISO)

Gradient Least square cell based

Pressure PRESTO!
Momentum Third-order monotonic upstream-centered scheme for conservation laws (MUSCL)
Volume fraction Geo-reconstruct
Turbulent kinetic energy First-order upwind
Specific dissipation rate First-order upwind
Energy Second-order upwind
Transient formulation First-order implicit

where y, uτ , and v are, respectively, the absolute distance from oration in bulk flow at any location where fluid temperature ex-
the wall, friction velocity, and kinetic viscosity of mixture. Non- ceeds local saturation temperature, and effectiveness of this model
dimensional distance represents a standard measure of mesh re- was verified in the authors’ previous study [34]. The mass trans-
finement for heat and mass transfer phenomena of interest. This fer rate per unit volume computed according to the Lee model is
parameter fluctuates along the channel due to property variation given by
and local phase change. As shown in Fig. 6(b), the selected cell size  
T f − Tsat
yields y + < 5 along the heated wall for the entire channel length, m˙ fg = ri α f ρ f for evaporation, (11)
which is small enough to capture relevant physical phenomena in Tsat
the viscous sublayer. and
Simulations for four different combinations of mass velocity
(Tsat − Tg )
and wall heat flux are performed in the present study. Fully m˙ gf = ri αg ρg for condensation, (12)
Tsat
developed velocity profiles corresponding to each mass velocity
(G = 176. 96, 453.16, 837.32, 2438.28 kg/m2 s) and accompany- where ri is mass transfer intensity factor, different values of which
ing turbulent properties are applied at the inlet boundary. Specific can affect evaporation and condensation rates. In regards to boil-
pressure values equal to experimental data are used in the out- ing, it is noted that ri values can have an impact on growth of va-
let boundary. No-slip boundary condition is applied to all walls. por bubble, interfacial behavior, flow regime, and total amount of
Contact angle estimated from the vapor-solid to vapor-liquid inter- heat rejection predicted along the flow channel. Generally, high ri
face for wall adhesion effect is 175°. Contact angle influences wall values can cause unstable numerical convergence while low ri val-
adhesion effects in conjunction with the surface tension model ues can result in a discrepancy between interface and saturation
[44], adjusting the curvature of the surface normal in cells near temperatures.
the heated surface and affecting the body force term during the Table 3 provides a summary of literature relating to computa-
surface tension calculation. Wall heat fluxes correspond to ~ 42– tional flow boiling and flow condensation work performed using
45% of CHF as measured during experiments [11] (corresponding the VOF method with the Lee mass transfer model. Combination
to q = 104.34, 143.24, 191.56, 194.87 W/m2 ). The range of inlet of VOF method and Lee model is the most easily implemented ap-
subcooling is between 30 and 36 °C. A coupled heat flux condition proach among those available for computational two-phase flow
is applied to the interface between solid and liquid to address con- modeling and is capable of predicting a wide range of flow regimes
jugate heat transfer. Indicated in Table 1 are properties of FC-72 for including bubbly, slug, and stratified/wavy flow, rendering it a pop-
each case, which are determined based on the corresponding mea- ular choice for design predictions of two-phase cooling applica-
sured saturation pressure, spanning P = 113.82–151.83 kPa. Liquid tions. However, lack of understanding on how best to determine
initially occupies the entire flow channel with constant velocity ri value can limit its correct application in industry. Of special im-
corresponding to each mass velocity case. For numerical stability, portance to the current study is appropriate determination of ri in
a global Courant number (ut/c) of unity and variable time-step 3-D simulations that accounts for prevailing operating conditions,
size ranging from 10−5 to 10−7 s are used. Full numerical details fluid properties, and flow geometry. Notice that all studies listed
and discretization methods are summarized in Table 2. in Table 3 employ identical ri values for both evaporation and con-
densation in Eqs. (11) and (12), respectively.
3.4. Phase change model and mass transfer intensity factor in Firstly, use of identical ri values for evaporation and conden-
subcooled flow boiling sation is dependent on how much fluid is subcooled or super-
heated. Most studies in Table 3 deal with low subcooling or sat-
It is paramount to use an appropriate phase change model urated flow conditions. In this case, numerical error arising from ri
to achieve accurate computational prediction of heat and mass value (which can yield discrepancies between interfacial and satu-
transfer. As the present study deals with highly subcooled liq- ration temperatures) is relatively minor. However, as high subcool-
uid flow and begins from liquid-only state, mass transfer models ing of around 34 °C is considered in present study, interfacial heat
that require pre-existing interfaces to calculate mass transfer (e.g., diffusion is significant, making interface-tracking difficult and ac-
Schrage model [51]) cannot be used. The Lee model [42] is adopted celerating vapor bubble condensation in non-physical fashion. Sec-
in the current approach due to its effectiveness at predicting evap- ondly, in regards to geometry (especially cross-section area), the
10 J. Lee, L.E. O’Neill and I. Mudawar / International Journal of Heat and Mass Transfer 150 (2020) 119291

Table 3
Summary of computational works on channel flow condensation and flow boiling employing volume-of-fluid method and Lee mass transfer model.

Mass transfer intensity

Author(s) System Dimensions Refrigerant Operating conditions factor (ri )

Tsub [°C] G [kg/m2 s] q [kW/m2 ]

Chen et al. [52] Rectangular micro-channel (3D) L = 300 mm FC-72 0 100–150 10–30 100
Flow condensation H = 1 mm
W = 1 mm
Wu et al. [53] Serpentine tube (3D) L = 560 mm R141B 3 40–200 7–36 0.1
Flow boiling d = 6 mm
Yang et al. [54] Coiled tube (3D) L = 842.6 mm R141B 8–10 120–180 7–25 100
Flow boiling d = 6 mm
Bahreini et al. [55] Vertical mini-channel (2D) L = 200 mm HFE-7100 5 20–210 20–160 1
Flow boiling H = 6 mm
Lorenzini & Joshi [56] Micro-channel (3D) L = 5, 6 mm Water 0–10 500 0 −2700 0.1
(Non-uniform)
Flow boiling H = 150,
100 μm
W = 300,
200 μm
De Schepper et al. [57] Single circular channel L = 11.3 m Gasoil 30 336.759 30.6–44.2 0.1
Flow boiling d = 52.5 mm
Fang et al. [58] Micro-channel L = 5 mm Water 3.5 269.73 60–250 100
H = 150 μm
Flow boiling W = 300 μm
Alizadehdakhel et al. Thermosyphon Lh = 40 mm Water – – 14.7–29.3 0.1
[59]
Pool boiling and condensation d = 1.75 cm

present setup is quite small. While entire micro-channels can be

engulfed by the thermal boundary layers, in macro-channels (such
as the present setup), there are core regions not covered by the
thermal boundary layers. These regions interrupt bubble produc-
tion, growth and existence along the flow channel by allowing for
entrainment and collapse (depending on local subcooling) of bub-
bles produced along the heated walls. Lastly, thermal conductiv-
ity and velocity of working fluid are also important factors. Dur-
ing transient computations of subcooled flow boiling, for a given
time step, heat is quickly transferred to the subcooled core region
while fluid with low thermal conductivity (vapor phase) cannot
transfer heat, influencing development of the thermal boundary
layer. Furthermore, subcooled flow with high velocity can flush out
the boundary layer or cause generation of a thin thermal bound-
ary layer, which is not a favorable environment for computational
predictions of phase change. Thus, use of identical ri values for
both evaporation and condensation in the Lee model is a good
assumption only in cases with low subcooling or saturated inlet,
small sized cross-section area, fluid with high thermal conductiv-
ity, and low inlet velocity. It must be noted that the value of ri
should be theoretically different for evaporation and condensation
[60], but it is not trivial to tune these values to match experimental
data because of unknown, undefined, or inconsistent variables such
as bubble diameter, interface shape, and accommodation coeffi-
cient; these are important factors required to obtain evaporation-
condensation flux developed from the Hertz-Knudsen formula and
kinetic theory.
The operating conditions in this investigation (highly sub-
cooled FC-72, low thermal conductivity, high flow velocity, macro-
channel) mean use of identical ri values is no longer effective.
Fig. 7 compares, for G = 453.16 kg/m2 s, q = 143.24 kW/m2 , and Fig. 7. Comparison of the influence of using identical versus different values for
evaporation and condensation in the Lee model on vapor bubble formation along
Tsub = 32.57 °C, bubble formation predicted by the Lee model with the channel.
(i) identical mass transfer intensity factor value, ri = 100, used
for both evaporation and condensation, and (ii) different values,
ri = 100 for evaporation and 0.1 for condensation. Predicted vapor
condensation of vapor bubbles by surrounding subcooled liquid,
bubble formations within the channel show appreciable discrep-
and consequently unrealistic depiction of bubble formation. On the
ancy from experimental results in the case with identical ri values
other hand, use of different ri values for evaporation and condensa-
for evaporation and condensation. It is obvious that void fraction
tion predicts bubble formation and behavior (including growth, co-
predicted using identical ri is much smaller due to much faster
alescence and departure) that are closer to those observed exper-
 J. Lee, L.E. O’Neill and I. Mudawar / International Journal of Heat and Mass Transfer 150 (2020) 119291 11

imentally. Unfortunately, quantitative comparison cannot be con- ble under microgravity conditions is considered, thereby eliminat-
ducted due to inherent difficulties present in calculating experi- ing any influence of buoyancy force. Analytic trajectory calculations
mental void fraction using only 2-D images. In 2-D images, a large are performed using the models of Mei and Klausner [41] and Leg-
number of bubbles are piled up and randomly deformed, so it is endre and Magnaudet [35] for shear-lift and analyzing force bal-
difficult to track exact outlines of every single bubble and calcu- ance of a bubble moving in Poiseuille flow as a function of time.
late volume of vapor present at any given moment. Comparison For the force balance, drag force, shear lift force, and momentum
of average heated wall temperature was performed, but the tem- terms including virtual mass force are considered. Computational
perature difference was small even though distinct differences in predictions with and without the UDF for shear-lift force are per-
bubble formation were observed. formed for comparison. The 2-D computational domain for this
To finally determine most appropriate ri value for evaporation simple case is illustrated in Fig. 8(a) along with actual dimensions
in 3-D flow boiling, five different values (10, 10 0, 30 0, 50 0, and of the experiment [40] used for validation. The adiabatic channel
10 0 0) are examined and ri = 100 (along with ri = 0.1 for conden- is 40-cm long and 2-cm high and contains air-water flow. Fully de-
sation) is selected based on thorough assessment of ability to pre- veloped velocity profile is used for inlet boundary, and mean liquid
dict experimental bubble formation, interfacial behavior, and wall velocity is 7 cm/s. A 4.5-mm diameter bubble with initial velocity
temperature. components ui = 2.8 cm/s (parallel to the wall) and vi = 2.0 cm/s
(perpendicular to the wall) is generated with its center 0.43 cm
away from the wall.
4. Results and discussion
Fig. 8(b) compares experimental single bubble trajectory with
those calculated using analytic (solved numerically) and compu-
4.1. Impact of shear-lift force
tational methods. It shows the computational method using the
UDF for shear-lift force provides best agreement with experiment.
Shear-lift force plays a critical role in the bubble detachment
It also illustrates how the bubble rapidly moves upwards be-
process for cases where high flow inertia liquid occupies most of
cause of strong shear-lift force near the wall. As the bubble mi-
the flow channel and acts to suppresses formation and growth of
grates, relative velocity between the two phases gets smaller, re-
small vapor bubbles produced on the heated surface. In order to
ducing shear-lift force, and converges towards the channel cen-
validate the effectiveness of the present computational methodol-
terline (y = 1.0 cm) due to equilibrium of positive and negative
ogy at capturing shear-lift force, a sample case simulating a single
shear-lift forces at this point. Trajectory predicted computation-
bubble’s lift-off and trajectory after detachment is examined and
ally without inclusion of the UDF for shear-lift force also shows
compared to experimental data as well as analytic models. To bet-
the bubble approaching the channel’s centerline, albeit at a much
ter capture the effect of the shear-lift force, a single vapor bub-

Fig. 8. (a) Example of single air bubble released in adiabatic Poiseuille water flow, which is used for evaluating effectiveness of present computational methodology at
capturing force balance on a bubble, (b) Comparison of single bubble trajectory using experimental, computational, and analytic results, and investigation into the effect of
user-define function (UDF) on the bubble’s (c) normal velocity and (d) axial velocity.
12 J. Lee, L.E. O’Neill and I. Mudawar / International Journal of Heat and Mass Transfer 150 (2020) 119291

slower rate than observed experimentally and predicted when us- an accelerating or decelerating bubble must push its way through
ing the UDF. This shows how, without the UDF, there is no ad- surrounding liquid) cannot be carefully calculated every iteration.
ditional force acting on the bubble in the transverse direction, and It still ends up converging into the core region of the flow chan-
initial y-velocity is dominant in determining bubble motion. This is nel, however, due to shear-lift working in the opposite direction
also proven in Fig. 8(c), which shows, without the UDF, y-velocity after it initially overshoots the centerline. On the other hand, the
exhibits a monotonic decrease with time, while with the UDF ini- Legendre and Magnaudet model predicts the bubble linearly lifts
tially increases above its initial velocity, then decreases because of off away from the initial location. This model considers only rela-
negative force in the y direction, before fluctuating some as the tive velocity between a bubble and liquid flow velocity to calculate
bubble approaches the channel centerline. These differences in y- the shear-lift force, but not any other local fluid flow character-
velocity also manifest in x-velocity, Fig. 8(d), as the case with the istics. Overall, both analytic solutions neglect to consider interfa-
UDF shows the bubble penetrates the high-velocity liquid core ear- cial behavior, bubble deformation, or other local interactions be-
lier and is significantly accelerated, while that without the UDF, tween the bubble and surrounding liquid, meaning projected area
shows the bubble remains in the near-wall region for far longer. for drag and shear-lift forces is assumed constant with time. This
It is important to point out that, with the UDF, Figs. 8(c) is a key area where the computational approach has a clear ad-
and 8(d) also capture the instantaneous fluctuations of x- and y- vantage, evidenced by the computational approach with modified
velocities, which are the outcome of ability to account for spatial shear-lift modeling providing best predictions of bubble lift-off and
and temporal variations of local liquid velocities surrounding the trajectory.
vapor bubble as the bubble rises and deforms, which also affect
the bubble trajectory. On the other hand, without the UDF, changes 4.2. Fluid flow and heat transfer characteristics of highly subcooled
in local velocities of liquid caused by bubble rise are not predicted, boiling predicted by 3-D computational model with shear-lift force
which results in non-physical smooth bubble drift along stream- UDF
lines determined mainly by the inlet liquid profile.
Fig. 8(b) also shows how, when it comes to analytic calculations Fig. 9 shows computational visualization results obtained us-
based on force balance for single bubble moving in time and space, ing the aforementioned computational method with the shear-lift
the Mei and Klausner model over predicts bubble trajectory as vir- UDF. Captured are images of flow boiling along the entire heated
tual mass force (which is the inertia added to a bubble because length as well as enlarged discrete regions (upstream, middle and

Fig. 9. Computationally predicted flow visualization images of entire heated channel for different mass velocities and different heat fluxes corresponding to fairly similar
percentage of CHF.
 J. Lee, L.E. O’Neill and I. Mudawar / International Journal of Heat and Mass Transfer 150 (2020) 119291 13

Fig. 10. Comparison of computed axial variations of vapor void fraction for four
different mass velocities.

downstream) for vertical upflow boiling with highly subcooled in-

let conditions and mass velocities of G = 176.52, 445.75, 836.64,
and 2432.51 kg/m2 s. Heat fluxes in each case are different, but
are all about 43% of CHF. Individual images illustrate instantaneous
and temporal tracking of interfacial behavior and wall temperature
contours corresponding to each set of operating conditions.
Vapor growth, bubble departure, entrainment, coalescence, and
break-up are key facets of subcooled flow boiling. In Fig. 9, pure
Fig. 11. Vector field of cross-sectional area volume-of-fraction contours at
liquid with high subcooling of approximately 33 °C enters the z = 79.2 mm for G = 176.52 kg/m2 s and q = 104 kW/m2 .
heated portion of the flow channel, and tiny bubbles begin to form
in entrance region. These bubbles grow in the axial direction as
they coalescence and the rate of condensation diminishes due to 445.75, 836.64, and 2432.51 kg/m2 s computed at seven axial lo-
increasing bulk fluid temperature along the channel. Vapor bubbles cations (z = 5.4, 22.7, 40.0, 57.3, 74.6, 91.9, and 109.2 mm) then
in the downstream region are seen to partially detach from the spatially and time-averaged after reaching steady state. As distor-
surface and migrate towards the liquid core. All heated wall phe- tion of bubble shapes becomes severe as mass velocity increases,
nomena observed are almost symmetrical in the computational re- improper guesses of bubble shape from experimental video images
sults, but not exactly identical because of differences in small scale can cause serious errors when attempting to estimate local void
wake effects and rate of local vapor collision and bubble coales- fraction. This is why, unlike the authors’ prior 2-D work [34], ex-
cence. perimental results are not compared with the present 3-D compu-
Fig. 9 shows obvious differences between cases with high mass tational void fraction results. Overall, predicted vapor void fraction
velocities and those with lower mass velocities. As mass velocity in the present study shows a trend similar to that seen in the pre-
increases, deformation of vapor bubbles is appreciable and pinch- vious 2-D study. However, because the fraction of cross-sectional
ing of vapor bubbles becomes noticeable, leading large bubbles to area occupied by liquid is much larger than by vapor in the 3-D
break into small discrete bubbles in the downstream region. This simulations, void fraction values here are lower than those from
is largely due to strong shear stresses generated from both heated the 2-D simulations.
and adiabatic surfaces for higher mass velocities. Bubble shape is Fig. 10 shows void fraction starts from very low values (less
also seen to be less spherical than that under low mass velocity than 1%) in the upstream region and increases gradually towards
conditions. the exit for all cases. The results (excepting the highest mass
For the highest mass velocity case of G = 2432.51 kg/m2 s, velocity case) show large increments of void fraction with dis-
single-phase convection plays an important heat transfer role, tance. The lowest velocity case in particular shows void fraction
small bubbles generated along the surface are unable to grow into increases rapidly after the fifth point (z = 74.6 mm) where large
large bubbles because of the ability of bulk liquid with high iner- bubbles are observed after coalescence begins to occur. It is inter-
tia to maintain its subcooling along the entire channel, and a thin esting that values of void fraction for all cases except the highest
thermal boundary layer is maintained along the entire channel. mass velocity are somewhat similar despite predicted bubble sizes
High flow velocity in this case also prevents any gradual evapora- and interactions between bubbles differing, as shown in Fig. 9.
tion due to minimal residence time along the heated walls. These Results for the highest mass velocity case, meanwhile, highlight
effects make only a few computational cells near heated surfaces how increased velocity suppresses bubble growth, stimulates bub-
satisfy the saturation requirement for boiling (as outlined for the ble break-up, and invigorates condensation, resulting in very small
Lee mass transfer model), which results in decreased vapor void void fractions in the range of ~ 0–2%.
fraction. Fig. 9 also shows the variation of surface temperature be- Fig. 11 depicts instantaneously captured vector field in cross-
tween inlet and exit is around 5.5 °C for all mass velocity cases. sectional area at z = 79.2 mm along with a volume-of-fraction
This is evidence flow boiling is highly effective at maintaining rel- contour plot for G = 176.52 kg/m2 s and q = 104 kW/m2 . This
atively uniform temperature compared to purely single-phase heat figure captures the moment two large bubbles attached each to an
transfer. opposite surface begin to depart. It is clear only one vector field
Axial variation of vapor void fraction along the channel is quan- for both liquid and vapor exists due to formulation of the VOF
titatively examined to understand the influence of mass velocity on method as described previously. However, the UDF added shear-
vapor generation. Fig. 10 compares vapor fraction for G = 176.52, lift force Fsl acts on the cell occupied by vapor as an extra source
14 J. Lee, L.E. O’Neill and I. Mudawar / International Journal of Heat and Mass Transfer 150 (2020) 119291

term in the momentum equation, yielding the detailed 3-D vector

field with components vectors u (x-direction), v (y-direction nor-
mal to the surface), and w (z-direction), which are projected onto
the x-y plane in Fig. 11. Inside vapor cells (red color) Fsl points in
a direction normal to the surface.
This 3-D simulation also shows a large recirculation zone
next to the lower bubble, which may contribute additional phase
change by increasing liquid replenishment into near-wall region
and encouraging further nucleation; an effect that cannot be cap-
tured in 2-D simulations. This is further validation that shear-lift
both affects bubble detachment and enhances heat transfer by gen-
erating recirculation zones.

4.3. Validation of computed results

To more explicitly highlight advantages of moving from 2-D to

3-D simulations, Fig. 12(a) provides a comparison of flow visu-
alization predicted by 2-D and 3-D computational methods (in-
cluding the shear-lift modeling) with that captured experimen-
tally. The lowest mass velocity case of G = 176.52 kg/m2 s with
q = 104 kW/m2 is excluded since it was not computed in the pre-
vious 2-D study. Predicted flow regimes from 3-D simulations are
similar to both 2-D and experimental results. Even though more
bubbles are generated in 3-D simulations compared to 2-D, this
may be attributed to the increase in heated wall area in contact
with liquid. Interfacial behavior from 3-D simulations is also in
good agreement with experiments as well as 2-D simulations, and
any over- or under-prediction of shear-lift force acting on vapor
phase using the formulation described in Section 3 are not notice-
able. But one interesting difference is the location where nucleate
boiling initiates in 3-D simulation is further upstream compared to
2-D.
Fig. 12(b) shows a comparison of channel-averaged steady-state
wall temperature versus mass velocity from 2-D and 3-D simula-
tions alongside experimental results. This plot clearly shows the
superiority of the present 3-D computational work with modified
shear-lift modeling over the prior 2-D results as the 3-D curve is
noticeably closer to experimental values for each mass velocity.
Additionally, 3-D results match the slope of the experimental curve
exactly, while 2-D results show a departure as mass velocity is in-
creased. For the lower mass velocity case (G = 445.75 kg/m2 s), av-
erage wall temperature predicted by 3-D simulation is only ~2 °C
greater than the experimental value, while the 2-D simulation pre-
dicts a value ~7 °C higher. These differences increase with increas-
ing mass velocity, with 3-D results offset from experiments by
~5 °C for both G = 836.64 and 2432.51 kg/m2 s cases, while 2-D re-
sults over-predict temperature by ~15 °C and ~25 °C, respectively.
The superiority of 3-D simulations results from the increased area
for phase change, additional turbulent and wake effects in the span
Fig. 12. (a) Comparison of flow visualization from experiment with predictions of
direction (produced by shear stress from the adiabatic sidewalls), present 3-D computational method and authors’ previous 2-D method [34], (b)
and increased bubble departure rate due to the shear-lift force. Comparison of corresponding average wall temperature with mass velocity.
Bubble detachment is suppressed by liquid with high inertia in 2-
D simulations, but modified shear-lift modeling in the present 3-D
work makes bubbles behave more physically. to the surface from z = 0 mm to ~20 mm. Temperature differ-
Fig. 13(a) and 13(b) show wall temperature and heat transfer ence between experimental results and 3-D simulations is mini-
coefficient variations along the heated portion of the channel. Here, mal for the lowest mass velocity and increases slightly with in-
cases with four different mass velocities and heat fluxes are com- creasing mass velocity. Once nucleate boiling begins, 3-D predicted
pared with one another. All values presented in this figure are spa- wall temperature is largely constant along the channel, with values
tially and time-averaged over a few seconds of steady-state simula- near 67 °C for the cases of G = 176.52 and 445.75 kg/m2 s, 75 °C
tion. Results from the authors’ prior 2-D computational simulations for G = 836.64, and 81 °C for G = 2432.51 kg/m2 s. This is seen as
[34] are averages of left and right wall values, while those from a sizable improvement compared to 2-D simulation results, which
present 3-D simulations are indicated for each wall separately. show values that are both higher than measured and continuously
Fig. 13(a) shows wall temperature increases in the entrance re- increasing in the axial direction due to artificial vapor adhesion to
gion for all mass velocities tested due to dominance of single- the heated walls, which is rectified in the 3-D simulations through
phase heat transfer in this region (despite some minimal vapor modified shear-lift modeling. The improvement with 3-D simula-
generation as shown in Fig. 9). This entrance region corresponds tions is most noticeable for the highest mass velocity case, with
 J. Lee, L.E. O’Neill and I. Mudawar / International Journal of Heat and Mass Transfer 150 (2020) 119291 15

Fig. 13. Comparison of measured axial variations of (a) wall temperature and (b) heat transfer coefficient with predictions using present 3-D computational method and
authors’ previous 2-D method [34].

cleation, bubble growth, and departure to repeat more frequently.

Despite this appreciable improvement, it remains difficult to cap-
ture the experimental temperature decrease in the exit region, al-
though it is believed further improvements in capturing interfacial
phenomena will resolve this discrepancy.
Fig. 13(b) shows axial variations of local heat transfer coef-
ficient, h, with distance for the four mass velocities and heat
fluxes ranging around 43% of CHF. Similar to wall temperature,
predicted local heat transfer coefficient is also spatially and time-
averaged over a few seconds of steady-state simulation. Predicted
heat transfer coefficient is computed by dividing heat flux by tem-
perature difference (between solid and fluid mixture temperatures,
both determined through area-weighted averaging). Experimental
local heat transfer coefficient is determined using a simple control
volume energy balance centered on each thermocouple illustrated
in Fig. 2(c). Fig. 13(b) shows 3-D computed heat transfer coeffi-
cients are well predicted over the different operating conditions.
Fig. 14. Comparison of measured and 3-D computed average, wall temperature and
Experimental heat transfer coefficient values are seen to decrease
average heat transfer coefficient for four different mass velocities.
in the entrance region, maintain a constant value in the middle re-
predicted wall temperature in the present study 20–25 °C lower gion, and increase in the exit region. Heat transfer coefficient for
than 2-D results in the middle and downstream regions. This too is the lowest mass velocity of G = 176.52 kg/m2 s predicted using
attributable to modified shear-lift modeling and shear-stress effects the present 3-D simulations gradually increases with distance, as
provided by sidewalls in the 3-D domain leading the cycle of nu- continuous thermal boundary layer development allows continu-
16 J. Lee, L.E. O’Neill and I. Mudawar / International Journal of Heat and Mass Transfer 150 (2020) 119291

Fig. 15. Instantaneous variation of computed (a) axial velocity and (b) fluid temperature across the channel at three axial locations (entrance, middle, and exit regions) for
different operating conditions.
 J. Lee, L.E. O’Neill and I. Mudawar / International Journal of Heat and Mass Transfer 150 (2020) 119291 17

ous phase change and bulk flow temperature increase. For higher ture in the direction normal to the surface grows, so prediction of
velocities, a slight reduction in h is seen in the entrance region, shear-stress transport within the boundary layer is important.
followed by a slight increase along the remaining channel length Moving along the channel across all cases appreciable increases
(likely due to flow acceleration brought on by increasing void frac- in core temperature are observed. For the highest mass velocity,
tion). All 3-D predicted heat transfer coefficients are in reasonable however, velocity and temperature profiles at all three locations
agreement with experimental values, compared to less accurate are similar to their inlet values because bubble migration toward
predictions using the 2-D simulations. Due to the presence of adi- the center of the flow channel is suppressed by high flow inertia.
abatic walls in the 3-D domain, shear stress becomes increasingly Boiling still takes place in this case with small bubbles forming and
influential as mass velocity increases, resulting in increasingly bet- lifting off from heated surfaces, but highly subcooled liquid leads
ter prediction of h over 2-D results. These facts further demon- to rapid condensation as they attempt to enter bulk flow. This lim-
strate superiority of the current 3-D approach over prior 2-D re- its their influence on velocity and temperature profiles.
sults.
Overall, deviation between measurements and 3-D predicted h 5. Conclusions
is less than 400 W/m2 K throughout the middle region of the chan-
nel, with a minimum difference of only 0.3% for the second lowest The present study focused on use of a 3-D computational ap-
mass velocity case. Differences between predicted and measured proach with advanced modeling for shear-lift force in ANSYS-
values are comparatively large in the upstream region, however, FLUENT to predict highly subcooled nucleate flow boiling of FC-
with maximum deviation of around 25% for the highest mass ve- 72. Four different mass velocity cases were evaluated, with heat
locity case. Deviation in the entrance region is largely attributable fluxes of ~ 41–46% of CHF in each case applied to the two oppo-
to over-prediction of wall temperature in this portion of the chan- site heated surfaces of a rectangular channel in vertical upflow. De-
nel. tailed information relating to the effect of shear-lift force and the
Fig. 14 shows 3-D predicted average values of wall temperature ability of ANSYS-FLUENT to accurately capture its effects was pro-
and heat transfer coefficient (with averaging achieved over entire vided through comparison with experimental results and analytic
heat length) match experimental trends well, further demonstrat- predictions. Comparison of 3-D with prior 2-D predictions and ex-
ing the effectiveness of the current 3-D computational method at perimental results showed the ability of the current computational
predicting the boiling process more physically than prior 2-D work. approach to surpass prior 2-D results by closely matching exper-
In addition to inclusion of modified shear-lift force and shear stress imental measurements. Detailed local computational results were
from the sidewalls, the 3-D simulations allows for more realistic also utilized to highlight local phenomena (velocity and tempera-
bubble shapes compared to 2-D. This leads to more physical con- ture profiles) not easily determined through experiments. Key find-
tact area between phases as well as 3-D turbulent effects which ings from present study are as follows.
Fig. 11 showed impacts heat transfer as well.
Despite these significant improvements, it is still difficult to (1) The 3-D computational approach outlined here captures
model the initial boiling process of vapor growth in cavities on complicated subcooled flow boiling behavior along the
the heated surface as vastly differing length scales mean surface heated portion of the channel, including vapor generation,
roughness effects are neglected and only bulk phenomena consid- bubble sliding, bubble departure, coalescence, and bubble
ered. Simulating bubble growth from physical cavities can improve break-up. Prediction of wall temperature and heat transfer
prediction accuracy of wall temperature and heat transfer coeffi- coefficient is appreciably improved from prior 2-D work due
cient in the entrance region, although current computing resources to the increase of area in contact with liquid phase as well
are insufficient to tackle this type of multi-scale problem. as shear-lift force and sidewall shear stress leading to more
aggressive phase change.
(2) Modeling shear lift force is required to overcome innate lim-
4.4. Predicted velocity and temperature profiles itations of the VOF model which employs only a single mo-
mentum equation for two phases and thus struggles to accu-
Fig. 15(a) and 15(b) show, respectively, cross-sectional profiles rately represent interfacial forces. Inclusion of shear-lift force
of computed axial velocity and fluid mixture temperature at three model prevents premature prediction of CHF through vapor
axial locations for all four operating conditions. These profiles are blanketing by providing more realistic capture of bubble de-
instantaneously exported after steady state along a diagonal line parture from heated walls. By enhancing physicality of bub-
on the cross section at each location (entrance, middle, and exit) to ble lift-off predictions, addition of modified shear-lift force
understand the influence of shear stress (produced by both heated provides more opportunity for liquid to replenish the near
and adiabatic surfaces) on local velocity and temperature. wall region, improving heat transfer. The 3-D nature of the
For all cases, the velocity profiles at z = 5.4 mm are fairly sim- present simulations also allowed for prediction of recircula-
ilar to that used as inlet condition as little vapor formation has tion zones, further enhancing physicality of the heat transfer
occurred. As nucleate boiling becomes appreciable in the middle results.
region, vapor motion begins to affect profiles. Distortion from the (3) Mass transfer intensity factor ri must be carefully tuned sep-
initial profile becomes considerable as vapor bubble penetrates to- arately for evaporation and condensation in 3-D computa-
ward the core region of the channel following bubble departure tional models for highly subcooled flow boiling of low con-
and coalescence. ductivity fluids in macro-scale channels at high flow rates.
Similar to velocity profiles, fluid mixture temperature grad- This is due to significant interfacial heat diffusion leading to
ually develops with distance, leading to the development of difficulty tracking interfacial behavior. Use of non-identical
a thermal boundary layer. Distortion of temperature profiles is mass transfer intensity factors for evaporation and conden-
clearly observed in the downstream region due to thermal dif- sation prevents over-prediction of vapor condensation while
fusion, advection from heated surfaces, as well as vapor bub- within subcooled bulk liquid.
ble motion. Interestingly, for temperature profiles for the case of (4) The 3-D computational methodology outlined here exhibits
G = 836.64 kg/m2 s, subcooled liquid replenishment is captured capability to predict realistic flow boiling results across var-
at z = 109.2 mm as there are two troughs around wall regions. ious operating conditions. This computational approach also
As mass velocity increases, the gradients of velocity and tempera- shows substantial improvement over prior 2-D results. Pre-
18 J. Lee, L.E. O’Neill and I. Mudawar / International Journal of Heat and Mass Transfer 150 (2020) 119291

diction of nucleate boiling in the entrance region is still a [20] M.K. Sung, I. Mudawar, Single-phase hybrid micro-channel/micro-jet impinge-
challenge, however, requiring additional modeling of boiling ment cooling, Int. J. Heat Mass Transf. 51 (2008) 4342–4352.
[21] R.W. Lockhart, R.C. Martinelli, Proposed correlation of data for isothermal
in small cavities to alleviate current deviations occurring in two-phase, two-component flow in pipes, Chem. Eng. Prog. 45 (1949) 39–48.
the upstream region. This is highly demanding computation- [22] S.W. Churchill, R. Usagi, A general expression for the correlation of rates of
ally and requires further thought on how to best implement transfer and other phenomena, AIChE J. 18 (1972) 1121–1128.
[23] H.J. Lee, S.Y. Lee, Pressure drop correlations for two-phase flow within hori-
to achieve more accurate results. zontal rectangular channels with small heights, Int. J. Multiph. Flow 27 (2001)
783–796.
Declaration of Competing Interest [24] S.M. Kim, I. Mudawar, Universal approach to predicting saturated flow boiling
heat transfer in mini/micro-channels part I. Dryout incipience quality, Int. J.
Heat Mass Transf. 64 (2013) 1226–1238.
None. [25] S.M. Kim, I. Mudawar, Review of databases and predictive methods for pres-
sure drop in adiabatic, condensing and boiling mini/micro-channel flows, Int.
Acknowledgments J. Heat Mass Transf. 77 (2014) 74–97.
[26] S.M. Kim, I. Mudawar, Theoretical model for local heat transfer coefficient for
annular flow boiling in circular mini/micro-channels, Int. J. Heat Mass Transf.
The authors are grateful for financial support provided by the 73 (2014) 731–742.
National Aeronautics and Space Administration (NASA) under grant [27] S.M. Kim, I. Mudawar, Review of two-phase critical flow models and investi-
gation of the relationship between choking, premature CHF, and CHF in mi-
no. NNX17AK98G, and technical support of the NASA Glenn Re-
cro-channel heat sinks, Int. J. Heat Mass Transf. 87 (2015) 497–511.
search Center, Cleveland, Ohio. This work was also supported by [28] C.R. Kharangate, I. Mudawar, Review of computational studies on boiling and
NASA Space Technology Research Fellowship NNX15AP29H. condensation, Int. J. Heat Mass Transf. 108 (2017) 1164–1196.
[29] E.J. Davis, G.H. Anderson, The incipience of nucleate boiling in forced convec-
tion flow, AIChE J. 12 (1966) 774–780.
Supplementary materials [30] J.-H. Wei, L.-M. Pan, D.-Q. Chen, H. Zhang, J.-J. Xu, Y.-P. Huang, Numerical simu-
lation of bubble behaviors in subcooled flow boiling under swing motion, Nucl.
Supplementary material associated with this article can be Eng. Des. 241 (2011) 2898–2908.
[31] R. Zhuan, W. Wang, Flow pattern of boiling in micro-channel by numerical
found, in the online version, at doi:10.1016/j.ijheatmasstransfer. simulation, Int. J. Heat Mass Transf. 55 (2012) 1741–1753.
2019.119291. [32] M. Magnini, B. Pulvirenti, J.R. Thome, Numerical investigation of the influence
of leading and sequential bubbles on slug flow boiling within a microchannel,
References Int. J. Therm. Sci. 71 (2013) 36–52.
[33] S. Chen, Z. Yang, Y. Duan, Y. Chen, Di Wu, Simulation of condensation flow in
[1] I. Mudawar, Two-phase microchannel heat sinks: theory, applications, and lim- a rectangular microchannel, Chem. Eng. Process. 76 (2014) 60–69.
itations, J. Electron. Packag. 133 (2011) 041002–041031. [34] J. Lee, L.E. O’Neill, S. Lee, I. Mudawar, Experimental and computational inves-
[2] I. Mudawar, Recent advances in high-flux, two-phase thermal management, J. tigation on two-phase flow and heat transfer of highly subcooled flow boiling
Therm. Sci. Eng. Appl. 5 (2013) 021012–021015. in vertical upflow, Int. J. Heat Mass Transf. 136 (2019) 1199–1216.
[3] W. Qu, I. Mudawar, Measurement and prediction of pressure drop in [35] D. Legendre, J. Magnaudet, The lift force on a spherical bubble in a viscous
two-phase micro-channel heat sinks, Int. J. Heat Mass Transf. 46 (2003) linear shear flow, J. Fluid Mech. 368 (1998) 81–126.
2737–2753. [36] G. Segre, A. Silberberg, Behaviour of macroscopic rigid spheres in Poiseuille
[4] S. Lee, V.S. Devahdhanush, I. Mudawar, Investigation of subcooled and satu- flow Part 2. Experimental results and interpretation, J Fluid Mech. 14 (1962)
rated boiling heat transfer mechanisms, instabilities, and transient flow regime 136–157.
maps for large length-to-diameter ratio micro-channel heat sinks, Int. J. Heat [37] I. Mudawar, Flow boiling and flow condensation in reduced gravity, Adv. Heat
Mass Transf. 123 (2018) 172–191. Transf. 49 (2017) 225–306.
[5] H.J. Lee, D.Y. Liu, S.-C. Yao, Flow instability of evaporative micro-channels, Int. [38] H. Zhang, I. Mudawar, M.M. Hasan, Experimental and theoretical study of
J. Heat Mass Transf. 53 (2010) 1740–1749. orientation effects on flow boiling CHF, Int. J. Heat Mass Transf. 45 (2002)
[6] T.J. LaClair, I. Mudawar, Thermal transients in a capillary evaporator prior to 4463–4478.
the initiation of boiling, Int. J. Heat Mass Transf. 43 (20 0 0) 3937–3952. [39] C. Konishi, I. Mudawar, Review of flow boiling and critical heat flux in micro-
[7] I. Mudawar, T.M. Anderson, Parametric investigation into the effects of pres- gravity, Int. J. Heat Mass Transf. 80 (2015) 469–493.
sure, subcooIing, surface augmentation and choice of coolant on pool boiling [40] H.K. Nahra, B.J. Motil, Measurements of shear lift force on a bubble in channel
in the design of cooling system for high-power-density electronic chips, J. Elec- flow in microgravity, NASA/TM-2003-212113 (2013) 1–16.
tron. Packag. 112 (1990) 375–382. [41] R. Mei, J.F. Klausner, Shear lift force on spherical bubbles, Int. J. Heat Fluid
[8] I.C. Bang, S. Heung Chang, Boiling heat transfer performance and phenomena Flow 15 (1994) 62–65.
of Al2O3–water nano-fluids from a plain surface in a pool, Int. J. Heat Mass [42] W.H. Lee, A pressure iteration scheme for two-phase flow modeling, in:
Transf. 48 (2005) 2407–2419. T.N. Veziroglu (Ed.), Multi-Phase Transport: Fundamentals, Reactor Safety, Ap-
[9] H.-Y. Kim, Y.G. Kim, B.H. Kang, Enhancement of natural convection and pool plications, 1, Hemisphere Publishing, Washington, DC, 1980.
boiling heat transfer via ultrasonic vibration, Int. J. Heat Mass Transf. 47 (2004) [43] L.E. O’Neill, I. Mudawar, M.M. Hasan, H.K. Nahra, R. Balasubramaniam,
2831–2840. N.R. Hall, A. Lokey, J.R. Mackey, Experimental investigation into the impact of
[10] C.O. Gersey, I. Mudawar, Effects of heater length and orientation on the trigger density wave oscillations on flow boiling system dynamic behavior and stabil-
mechanism for near-saturated flow boiling CHF - I. Photographic and statistical ity, Int. J. Heat Mass Transf. 120 (2018) 144–166.
characterization of the near-wall interfacial features, Int. J. Heat Mass Transf. [44] J.U. Brackbill, D.B. Kothe, C. Zemach, A continuum method for modeling surface
38 (1995) 629–642. tension, J. Comput. Phys. 100 (1992) 335–354.
[11] C.R. Kharangate, L.E. O’Neill, I. Mudawar, Effects of two-phase inlet quality, [45] C.R. Kharangate, H. Lee, I. Mudawar, Computational modeling of turbulent
mass velocity, flow orientation, and heating perimeter on flow boiling in a evaporating falling films, Int. J. Heat Mass Transf. 81 (2015) 52–62.
rectangular channel: part 1 - Two-phase flow and heat transfer results, Int. [46] J.F. Klausner, R. Mei, D.M. Bernhard, L.Z. Zeng, Vapor bubble departure in
J. Heat Mass Transf. 103 (2016) 1261–1279. forced convection boiling, Int. J. Heat Mass Transf. 3 (1993) 651–662.
[12] C.R. Kharangate, L.E. O’Neill, I. Mudawar, Effects of two-phase inlet quality, [47] L.Z. Zeng, J.F. Klausner, D.M. Bernhard, R. Mei, A unified model for the predic-
mass velocity, flow orientation, and heating perimeter on flow boiling in a tion of bubble detachment diameters in boiling systems-II. Flow boiling, Int. J.
rectangular channel: part 2 - CHF experimental results and model, Int. J. Heat Heat Mass Transf. 36 (1993) 2271–2279.
Mass Transf. 103 (2016) 1280–1296. [48] S.S. Rabha, V.V. Buwa, Volume-of-fluid (VOF) simulations of rise of sin-
[13] S. Mukherjee, I. Mudawar, Pumpless loop for narrow channel and micro-chan- gle/multiple bubbles in sheared liquids, Chem. Eng. Sci. 65 (2010) 527–537.
nel boiling from vertical surfaces, J. Electron. Packag. 125 (2003) 431–441. [49] P.G. Saffman, The lift on a small sphere in a slow shear flow, J. Fluid Mech. 22
[14] J. Lee, I. Mudawar, Critical heat flux for subcooled flow boiling in micro-chan- (1965) 385–400.
nel heat sinks, Int. J. Heat Mass Transf. 52 (2009) 3341–3352. [50] T.R. Auton, The lift force on a spherical body in a rotational flow, J. Fluid Mech.
[15] W.P. Klinzing, J.C. Rozzi, I. Mudawar, Film and transition boiling correlations for 183 (1987) 199–218.
quenching of hot surfaces with water sprays, J. Heat Treat. 9 (1992) 91–103. [51] R.W. Schrage, A Theoretical Study of Interphase Mass Transfer, Columbia Uni-
[16] M. Visaria, I. Mudawar, Application of two-phase spray cooling for thermal versity Press, New York, 1953.
management of electronic devices, IEEE Trans.-CPMT 32 (2009) 784–793. [52] S. Chen, Z. Yang, Y. Duan, Y. Chen, D. Wu, Simulation of condensation flow in
[17] W.-L. Cheng, F.-Y. Han, Q.-N. Liu, R. Zhao, H.-L. Fan, Experimental and theo- a rectangular microchannel, Chem. Eng. Process. 76 (2014) 60–69.
retical investigation of surface temperature non-uniformity of spray cooling, [53] H.L. Wu, X.F. Peng, P. Ye, Y. Eric Gong, Simulation of refrigerant flow boiling in
Energy 36 (2011) 249–257. serpentine tube, Int. J. Heat Mass Transf. 50 (2007) 1186–1195.
[18] M. Monde, T. Inoue, Critical heat flux in saturated forced convective boiling on [54] Z. Yang, X.F. Peng, P. Ye, Numerical and experimental investigation of two
a heated disk with multiple impinging jets, J. Heat Transf. 113 (1991) 722–727. phase flow during boiling in a coiled tube, Int. J. Heat Mass Transf. 51 (2008)
[19] M.E. Johns, I. Mudawar, An ultra-high power two-phase jet-impingement 1003–1016.
avionic clamshell module, J. Electron. Packag. 118 (1996) 264–270.
 J. Lee, L.E. O’Neill and I. Mudawar / International Journal of Heat and Mass Transfer 150 (2020) 119291 19

[55] M. Bahreini, A. Ramiar, A.A. Ranjbar, Numerical simulation of subcooled flow [58] C. Fang, M. David, A. Rogacs, K. Goodson, Volume of fluid simulation of boiling
boiling under conjugate heat transfer and microgravity condition in a vertical two-phase flow in a vapor-venting microchannel, Front. Heat Mass Transf. 1
mini channel, Appl. Therm. Eng. 113 (2017) 170–185. (2010) 0132002.
[56] D. Lorenzini, Y.K. Joshi, Computational fluid dynamics modeling of flow boil- [59] A. Alizadehdakhel, M. Rahimi, A.A. Alsairafi, CFD modeling of flow and heat
ing in microchannels with nonuniform heat flux, J. Heat Transf. 140 (2018) transfer in a thermosyphon, Int. Commun. Heat Mass Transf. 37 (2010)
011501–1. 312–318.
[57] S.C.K. De Schepper, G.J. Heynderickx, G.B. Marin, Modeling the evaporation of [60] ANSYS FLUENT Theory Guide, ANSYS Inc., Canonburg, PA, 2009.
a hydrocarbon feedstock in the convection section of a steam cracker, Comput.
Chem. Eng. 33 (2009) 122–132.