International Journal of Multiphase Flow 174 (2024) 104772

Contents lists available at ScienceDirect

International Journal of Multiphase Flow

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

Subcooled forced convection boiling flow measured using high-resolution

techniques at the COSMOS-L facility and accompanying CFD simulation
employing an interface-tracking scheme
Florian Kaiser a, Yohei Sato b, Stephan Gabriel a, *
a
Karlsruhe Institute of Technology, Germany
b
Paul Scherrer Institute, Switzerland

A R T I C L E I N F O A B S T R A C T

Keywords: This paper describes state-of-the-art experimental measurement techniques and computational fluid dynamics
Subcooled boiling methods applied to subcooled forced convection boiling flow at 2-3 bar, both featuring high spatial- and time-
Critical heat flux resolution measurements. The objectives are: (i) to provide the measured boiling flow data, e.g. average and
Boiling crisis
RMS profiles of axial and radial velocities and bubble size distributions, and (ii) to evaluate the capabilities of
Two phase flow
Interface tracking
state-of-the-art numerical simulation techniques to model the observed phenomena, by comparison between
measured data and numerical prediction. In the experiment, the bubble velocities in the axial and radial di­
rections were measured using laser Doppler anemometry, and the bubble size was estimated by a shadowgraphy
method. The simulation results, incorporating an interface capturing technique, showed good agreement with
measurement in terms of the wall temperature and axial velocity, but underestimation of the radial velocity, and
overestimation of bubble sizes. The paper shows that the discrepancies between measurement and simulation
may be traced to: (i) the uncertainties in the nucleation site density model; (ii) too coarse a grid being adopted,
which is not able to resolve the thermal boundary layer around the bubbles; and (iii) spurious numerical bubble
coalescence, a feature not seen in the experiment. The mechanism of bubble lift-off from the hot surface, without
any sliding motion, as observed in the experiment, is discussed in detail, based on the simulation results, and the
condensation on the bubble cap as the bubble is ‘sucked’ into the bulk flow is also examined.

1. Introduction CFD, simulations have been specifically developed to model boiling

flows, in order to understand the phenomena taking place, and to
Boiling heat transfer is used in a variety of engineering applications: simulate boiling heat transfer under various imposed conditions. As a
e.g. nuclear power plants, refrigerators, heat pipes and chemical pro­ consequence of these techniques, with high resolution, local phenomena
cessing devices. Important parameters are the overall heat transfer co­ relating to boiling flow and boiling heat transfer have been measured,
efficient and the maximum heat flux − the so-called Critical Heat Flux, and simulated, in order to gain insights into the physics controlling CHF.
CHF − and have been intensively studied in this field, for the purpose of In this paper, we present state-of-the-art measurements, and associated
effective, efficient and safe operation of the different systems. Boiling CFD simulations, for subcooled, forced-convection boiling flow, and
heat transfer and CHF have been measured mainly using thermocouples, discuss the capabilities and the limitations of the numerical modelling
and associated empirical correlations and look-up tables have been potential.
assembled for the purpose of design and operation of the various ap­
plications: e.g. Levy (1959) and Rohsenow (1971). In addition, to better
1.1. Measurements of boiling flow
understand the mechanisms of boiling flow and boiling heat transfer,
measurement techniques have been developed which can resolve local
Numerous measurement methods and instruments have been
temperatures, velocities and void-fractions (Dhir, 1998). In parallel to
developed to quantify boiling flows. To produce CFD-grade validation
the measurements being carried out, Computational Fluid Dynamics,
data, accurate measurements are required for the dispersed gas phase

* Corresponding author.
E-mail address: stephan.gabriel@kit.edu (S. Gabriel).

https://doi.org/10.1016/j.ijmultiphaseflow.2024.104772
Received 24 October 2023; Received in revised form 23 December 2023; Accepted 14 February 2024
Available online 24 February 2024
0301-9322/© 2024 Published by Elsevier Ltd.
F. Kaiser et al. International Journal of Multiphase Flow 174 (2024) 104772

and the continuous liquid phase. In the case of the gaseous phase, the 1.2. Numerical simulation methods for boiling
focus is principally on the bubble size and number density, as well as on
their trajectories and velocities. The continuous liquid phase is charac­ In boiling flow simulations, it is essential to resolve liquid-vapor
terized by its velocity and temperature distributions, and on its turbulent interface morphology and area, because vaporization and condensa­
properties. The choice of measurement technique, though as always tion rates are strongly influenced by the heat flux at the liquid/vapor
based on the concept of the ideal, is limited by numerous technical and interface (Sato et al., 2018). Thus, an interface tracking approach, which
physical restrictions. In general, non-intrusive measurement techniques is able to resolve the liquid-vapor interface, is considered to be the most
are to be preferred, to avoid disturbing the morphology of the flow, and appropriate for simulations with phase-change phenomena (Yadigar­
optical accessibility of the flows under investigation is limited, espe­ oglu, 2005), although fine computational meshes are required to spe­
cially at high vapor fractions, where refraction and diffraction signifi­ cifically resolve the interface itself. There are several types of interface
cantly reduce accessibility in the visible light range. Methods such as tracking approaches: e.g. the Arbitrary Lagrangian-Eulerian, ALE,
high-speed videometry should also be mentioned here, where, for approach (Fuchs et al., 2006; Lee and Nydahl, 1989; Welch, 1998), the
example, images can be generated using the shadow method (Gabriel Level Set, LS, method (Huber et al., 2017; Son et al., 1999), the Volume
et al., 2018), which can be processed into discrete measured values Of Fluid, VOF, method (Kunkelmann and Stephan, 2009; Welch and
using the methods of digital image processing. For a system with mul­ Wilson, 2000), the Color Function Method, CFM, (Giustini et al., 2017;
tiple cameras, this offers good access to the flow even at high bubble Murallidharan et al., 2016; Sato and Niceno, 2013), the Front Tracking
densities (Büttner et al., 2018). Method, FTM, (Esmaeeli and Tryggvason, 2004), and the Phase Field
In particular, laser-based methods, such as Laser Doppler Anemom­ Method, PFM, (Badillo, 2012; Jamet et al., 2001; Takada and Tomiyama,
etry, LDA, and Particle Image Velocimetry, PIV, e.g. Gabriel (2014), can 2007). The differences between these approaches stem from the pro­
be used to obtain velocity data and information on the turbulence cedure adopted to represent the gas-liquid interface, though all have
characteristics of continuous flow situations. Both methods face chal­ been used successfully to simulate nucleate boiling and film boiling, as
lenges in the investigation of boiling flows, however, since in the con­ reviewed for example by Kharangate and Mudawar (2017). However,
ventional set-up they require the presence of tracer particles in the flow boiling simulations at high heat flux, i.e. near CHF, are rare, because (i)
to mark the fluid motion. In addition to the question of whether these the flow field becomes violent, due to the high mass-transfer rate, which
particles also act as heterogeneous boiling nuclei in a boiling flow sit­ introduces numerical instability, and (ii) an appropriate nucleation site
uation, the question also arises of whether a tracer can be found that model at high heat flux is required (Sato and Niceno, 2018), a situation
represents the flow well enough in terms of its density and tracing ability that requires further study.
without physically interfering with it. Due to temperature gradients in A nucleate pool boiling simulation near CHF has been performed by
the flow, as well as the phenomena of evaporation and condensation, Son and Dhir (2008), in which multiple nucleation sites were incorpo­
this idealization can probably only be achieved in an approximate way. rated, with the liquid-vapor interface captured using the LS method. Due
With respect to the measurement of the phase distribution, other to the restriction imposed by the authors of constant temperature over
non-intrusive possibilities exist, such as ultrafast X-ray tomography the heat-transfer surface, the bubble waiting time, i.e. the time interval
(Fischer and Hampel, 2010). Use of this measurement technique allows between bubble departure and subsequent bubble nucleation, could not
a temporally and spatially high-resolution view into the depth of the be directly computed, and had to be prescribed a priori. Nonetheless, the
flow to be obtained. In the field of intrusive measurement, different authors were able to obtain acceptable results for nucleate pool boiling
techniques have been proposed in the context of the quantification of of water for applied heat fluxes up to 800 kW/m2 at atmospheric pres­
dispersive/boiling flows. The sensors or probes commonly adopted sure. Li et al. (2015) and (Gong and Cheng, 2015), (Gong and Cheng,
mostly utilize optical or electrical/capacitive methods to detect bubbles 2016) have simulated the transition from nucleate boiling to film boiling
and their properties, such as bubble size and velocity. Examples are fiber regimes through CHF using a Lattice-Boltzmann approach. In these
optic probes that reflect an injected laser beam, based on the different works, the influence of the wettability of the heat-transfer surface on the
refractive indices of the vapor and liquid phases, and direct it to a heat transfer coefficient was specifically taken into account. However,
photodiode, or not, depending on the fluid present (Bruder et al., 2019). the computations are limited to two dimensions, and no validation
The phase fraction at the discrete location of the probe tip can then be experiment has been reported. Utilizing the sharp-interface, pha­
determined by evaluating the time curve of the luminosity signal se-change model of Sato and Niceno (2013), and its associated deplet­
received by the photo diode. able micro-layer model (Sato and Niceno, 2015), the authors of these
Apart from the optical sensors, there are further possible measuring works were able to simulate the pool boiling experiment of Gaertner
techniques, of which at least the two electrical ones should be (1965) from the discrete-bubble regime to the film-boiling regime
mentioned. Two electrodes are placed in the flow at a small distance through to Departure from Nucleate Boiling, DNB; Sato and Niceno
apart; the actual arrangement can vary depending on the experimental (2017), (2018). Unlike the simulation undertaken by Son and Dhir
set-up. If the liquid phase is electrically conductive and the gas phase is (2008), in these studies, the wall temperature was not prescribed a
not, the electrical resistance between the two electrodes can be priori, and the computed heat transfer coefficient was in good agreement
measured continuously. Also, in this case, the phase distribution can be with experimental data. In the present paper, we apply the same simu­
calculated from temporal analysis of the measurement signal. For a non- lation method originally developed for pool boiling flows by Sato and
conductive fluid, it is also possible to use its electrical capacity for Niceno (2018) to a subcooled, forced-convection boiling flow, and
measurement. A probe of this type, with a single measuring site, has present comparisons with the experimental measurements taken at the
already been used by Stäbler (2007). The significantly more complex COSMOS-L facility.
wire mesh sensors (Prasser, 1999) have a much higher number of
measuring points, and can record the phase distribution in an entire flow 1.3. Objectives of the current study
channel. For these sensors, wires are stretched in two planes orthogonal
to each other. The measuring sites are located at the intersection points In this paper, we present state-of-the-art measurement and simula­
of two cross-wires, and are scanned by the measuring electronics in tion results for subcooled, forced-convection boiling flow, to demon­
sequence. A disadvantage of the method is that the probe body itself strate recent progress in this field. The measurements were performed at
interacts with the flow, and small bubbles cannot always be detected in the COSMOS-L loop at the Karlsruhe Institute of Technology, KIT; Haas
some cases (Tompkins et al., 2018). (2012), Haas et al. (2018). Bubble shapes/motions were measured using
high-speed video cameras, and local velocities by Laser Doppler
Anemometry, LDA. The measurement of local quantities with high

2
F. Kaiser et al. International Journal of Multiphase Flow 174 (2024) 104772

temporal and spatial resolution enables us to directly compare with the heated section and the pressure hull. The glass tube is sealed by an O-
numerical simulations. The CFD calculations were performed at the Paul ring, and centered by the PEEK ring in the inlet block. The heated section
Scherer Institute, PSI, using the in-house, open-source CFD code PSI-­ begins where the fluid enters the annular channel formed by the central
BOIL1, which employs an interface tracking technique. The supercom­ Zircaloy-4 tube and the glass tube surround. The heated inner tube has a
puter at the Swiss National Supercomputing Centre, CSCS, was length of 326 mm. The test section is connected to the outlet block, and
employed for the computations. The measured and computed results, e. the working fluid then flows into a large horizontal channel of diameter
g. profiles of void fraction, velocity and bubble size distribution, are 100 mm. The vapor and liquid phases separate from each other in the
compared, in order to clarify any limitations and open issues in the both condenser downstream of the outlet block, Fig. 1. The glass tube and the
the experimental data and simulation results. upper copper conductor, are both centered, and sealed, in the outlet
The limitation of the current study is different topology of the test section. Outside of the test section, flexible cables for the power supply
section, which is a double tube in the experiment while a rectangular in are connected to the bottom and top ends of the inner tube assembly.
the CFD simulations. This is due to the restriction of the CFD code, which The inlet block is fixed and positioned relative to the outlet assembly as a
can only employ an orthogonal Cartesian grid. Although the dimensions notionally “stiff” structure.
of the test section, i.e. the axial length, circumferential/lateral length,
and the gap between the heated surface and the wall on the other side, 2.2. Measurement apparatus
are same between the experiment and CFD, the ratio of the volume of the
channel to the area of the heat transfer surface by 30%. In addition, we 2.2.1. Thermocouples and pressure sensors
infer that the velocity profiles in the wall normal direction differ even for The loop is equipped with several temperature and pressure sensors,
single-phase liquid flow, i.e., the peak appears in the center in the as indicated in Fig. 1, the test section itself being equipped with sensors
rectangular domain, while it is shifted to the outward in the double-tube for such measurements. A lance, consisting of twelve thermocouples, is
domain. inserted into the zircaloy tube, as shown in Fig. 3. In addition to tem­
The structure of this paper is as follows: the test facility and mea­ perature measurement, an important function of the thermocouples is
surement methods are described in Section 2, and the CFD simulation the detection of CHF, which is explained in section 2.3 below. The
method is presented in Section 3, which includes a parametric and grid- thermocouples are of type K, with a diameter of 0.36 mm. The pressures
dependence study. The measurement and simulation results are at the inlet and outlet block were continuously monitored, the outlet
compared and discussed in Section 4, and overall conclusions presented pressure being considered the system pressure.
in Section 5. The positions of the thermocouples are illustrated in Fig. 2 (right). At
each height, two thermocouples are placed on opposite sides of the in­
2. Test facility and measurement techniques side of the zircaloy tube. The height distribution has been chosen ac­
cording to the experience gained from the determination of CHF in
2.1. Test facility earlier experiments. For the chosen test section gap (8.5 mm), CHF was
observed mostly at the very top of the test section in the experiments
The experiments were carried out at the COSMOS-L (Critical-heat- reported here.
flux On Smooth and MOdified Surfaces – Low pressure) test facility
located at the Institute of Thermal Energy Technology and Safety, ITES, 2.2.2. High speed video camera
of KIT. COSMOS-L is a low pressure water loop for thermo-hydraulic Using video cameras, it is possible to observe the liquid-vapor
experiments of two-phase flow with heat and mass transfer, operating interface during the test; i.e. bubble growth, bubble motion and vapor
at low pressure, ranging from 1 to 3 bar. The loop was designed to agglomerations. A standard action camera with a frame rate of 30 fps,
measure the flow under DNB-type CHF conditions, the fluid to the test and a resolution of 3840 × 2160 pixels, is used for the overall obser­
section inlet being liquid subcooled water; the heated section is most vation of the test section, in particular continuous observation of the
probably not long enough for dryout-type CHF. The working fluid in the upper part. For the purposes of observing transient phenomena at high
present case was deionized water. Electrical conductivity and the oxy­ resolution, three high-speed video cameras were installed at different
gen content were continuously monitored during the experiments. axial height positions, as illustrated in Fig. 4. The framerate was 3000
The layout of the facility, i.e. the Piping and Instrumentation Dia­ fps, with a resolution of 1280 × 1024 pixels.
gram (P&ID), is presented in Fig. 1. First traversing a flow meter (F), the
water flows through a preheater, where the fluid temperature is adjusted 2.2.3. LDA for bubble size and bubble velocity measurement
to a defined level before entering the test section. The mass flow at the Bubble sizes and bubble velocities are essential for mechanistic
test section is set by a control valve, situated between the preheater and model descriptions. The measurement of these parameters allows a
the test section. direct comparison with the predicted values of the different model
The test section itself consists of inner and outer tubes, the working theories to be made, and represent important quantities for the valida­
fluid flowing vertically upwards through the annular gap between them. tion of such models. In addition, the local vapor content plays a key role
The inner tube is made of Zircaloy-4, and it is electrically heated. The in all models, as it hinders the supply of fresh cooling medium to the
outer diameter of the inner tube is 9.5 mm, and the thickness of the tube heated surface, and can thus promote CHF conditions.
itself is 0.57 mm. The outer tube is made of glass, for the purposes of The outer glass tube enabled the application of further optical
optical measurement. The inner diameter of the glass tube is 18.0 mm, measurement technologies to be utilized. The choice of measurement
and the thickness of glass is 2 mm. Fig. 2 shows the cross-section of the method needs to be made carefully to avoid the influence of the physical
test section (right) and a photograph of the assembled test section (left). presence of in-situ measurement devices, e.g. probes or wire meshes, on
Referring to the bottom of Fig. 2 (left), the water enters the inlet block the boiling process. Thus, only non-invasive methods have been used in
from four sides, each connected to a high pressure hose, and is then the tests reported here. The vapor bubble velocities were measured
directed upwards into the unheated annular section. The lower copper using a LDA system. A Diode-Pumped Solid-State Laser, DPSSL, with the
part of the inner test tube is centered in the inlet block by a ring with four maximum power of 300 mW and wavelength of 532 and 561 nm, having
spacers. The ring is made of the thermoplastic material PEEK (Polyether been was employed. A focal lens of 300 mm was used, providing a
ether ketone), which also insulates against electrical contact between measuring volume of 0.7 mm3. In order to avoid disturbance of the flow
by the use of tracer particles, only small bubbles (~100 µm) of steam, or
small amounts of non-condensable gases, were used as tracers. The
1
https://github.com/Niceno/PSI-BOIL. measurement times are defined in such a way as to ensure a sufficient

3
F. Kaiser et al. International Journal of Multiphase Flow 174 (2024) 104772

Fig. 1. Piping and instrumentation diagram of the COSMOS-L facility.

number of measured values are recorded at each measurement point, in 2.3. Measurement procedure
order to make a reliable estimate of the mean value, and to determine
the measurement uncertainty. The optical refraction at the transition First, the CHF condition was determined a priori by performing a
between the water and the outer glass tube was taken into account by series of experiments under different boundary conditions. Knowing the
calibration of the optical system, and a correction function derived from value of the critical heat flux, and the frequency distribution for the
it. Velocity profiles ranging from 0.06 mm ≤ r ≤ 4.3 mm, with an interval critical power from a large number of pre-test experiments, the boiling
of 0.07 mm, were measured at all three positions: Positions 1, 2 and 3 in characteristics of the test section could be estimated at reduced heating
Fig. 4. A comparison of the estimated mass flow rate based on LDA, and power in relation to the critical CHF power. This was achieved using the
the superficial velocity measured by the flow meter upstream of the test same boundary conditions (pressure, inlet fluid temperature/subcooling
section, was undertaken continuously during the experiment (Fig. 5). and mass flow rate). A steady-state measurement under CHF conditions
In addition to the bubble velocities, the bubble size distribution has is impossible in water, since CHF induces a high temperature increase on
also been measured in the tests. To this purpose, a shadowgraphy the heater surface with a gradient of about 200 K/s. The heater would
method was employed. However, due to frequent bubble coalescence then be damaged in a short time. The criterion for CHF estimation is
processes, and the rapid evaporation and condensation of the bubbles, based here on the prerequisite that CHF occurs spontaneously at a spe­
simultaneous detection of the bubble sizes and the velocities proved to cific applied power level. In addition to a temperature criterion, a time
be unachievable. It was observed that the bubbles that formed the criterion is defined within which CHF can occur. Each level of heating
heated surface deformed so rapidly that an automated assignment of the power was applied for two minutes before the next increment was
bubbles on double images was not possible. The time differences be­ initiated. Within these two minutes, the thermal-hydraulic parameters
tween the double images had to be chosen large enough to ensure that had been observed to stabilize to the extent that a quasi-stationary flow
the bubbles move by a measurable distance, but the deformation of the situation could be assumed.
bubbles within this period proved to be too large, which prevented the When CHF conditions are reached, a rapid shutdown of the heating
alignment to be achieved. Therefore, the velocity measurements were power must be triggered, in order to protect the test section from
carried out using LDA instead, focusing on the small bubbles already damage. However, within the short time span between the heating
present in the flow at the designated measurement locations. The set-up power increase and the occurrence of CHF, no reproducible measure­
of the shadowgraphy measurements is shown in Fig. 6. An Nd:YAG laser ment of thermal-hydraulic parameters could be realized. Therefore,
provides the illumination required, prepared by a diffusor and several specifically defined plateaus of the heating power had to be selected, for
matt screens, resulting in a homogeneous background illumination. On which a more detailed measurement of the parameters could be carried
the opposite site of the test section, a Charge-Coupled Device, CCD, out. Consequently, a measuring point was selected corresponding to an
camera had been placed, which registers an image of the flow in which inlet temperature of Tin = 80◦ C, mass flux of G = 400 kg/(m2 s), and a
bubbles appear darker than the continuous liquid, due to light re­ system pressure at the outlet of the test section of 2 bar. This is equiv­
fractions. Such images were taken at a frequency of 2 Hz. For any given alent to a thermal subcooling of 40.2 K. CHF for this case was found to be
bubble distribution, a minimum of 50 images needed to be taken into 1.23 MW/m2. Thermal-hydraulic parameters were adjusted at 50% of
consideration. The observation time for the images corresponds in total this heating power, corresponding to CHF at the top of the test section.
to 250 s, within which the images are taken at equidistant time intervals,
to avoid over-interpretation of exceptional flow events.

4
F. Kaiser et al. International Journal of Multiphase Flow 174 (2024) 104772

Fig. 2. Photo (left) and schematic (right) of the annular test section of COSMOS-L.

Fig. 3. Lance of thermocouples inserted into the inner tube.

3. CFD simulation method explained in Sato and Niceno (2015, 2018). The parametric study con­
ducted for the nucleation site density modelling, and grid dependency
The numerical method used for the CFD simulations is briefly study, are presented in this paper, together with the simulation results
explained here. The open-source CFD code, PSI-BOIL, is employed for all specifically in the context the COSMOS-L facility.
the simulations. The vapor and liquid phases are assumed to be
incompressible fluids, and the solid phase is incorporated according to 3.1. Governing equations
the immersed boundary method (Mittal and Iaccarino, 2005). The
method is essentially the same as that developed for the pool boiling The governing equations may be written (Sato and Niceno, 2013):
flow simulations of Sato and Niceno (2017, 2018), the difference being ( )
1 1
that the presence of inlet and outlet boundaries for forced convective ∇⋅→
u = − ṁ (1)
ρv ρl
flow in the present context. Detailed explanations of the fluid flow
equations solver with mass transfer are given in Sato and Niceno (2013),
and the modelling for nucleation sites and micro-layer development are

5
F. Kaiser et al. International Journal of Multiphase Flow 174 (2024) 104772

∂→u
ρ + ρ{∇⋅(→
u ⊗→
u )− →
u (∇⋅→
u )} = − ∇p
∂t
{ ( )} →
+ ∇⋅ (μ + μt ) ∇→
u + (∇→
u )T + f
(2)

∂ϕ 1
+ ∇⋅(ϕ→
u ) = − ṁ (3)
∂t ρl
( )
∂T →
Cp + u ⋅∇T = ∇⋅((λ + λt )∇T) + Q. (4)
∂t
Equations (1)-(2) are the mass and momentum conservation equa­
u (m/s) is the velocity vector, ρ (kg/m3) the
tions, respectively, in which →
density, and the subscripts l and v denote the liquid and vapor phases,
respectively; ṁ (kg/m3s) is the phase change rate (a positive value for
vaporization, and a negative value for condensation); t (s) is the time, p
(Pa) the pressure, μ (Pa⋅s) the dynamic molecular viscosity, μt (Pa⋅s) the
Fig. 4. Axial height positions for the measurements of the thermal hydrau­ →
lic parameters. turbulent eddy viscosity, and f (N/m3) the body-force vector. Equation
(3) is the governing equation for the color function f, which signifies the
volume fraction of liquid inside a given control volume. The average
density and viscosity within such a control volume (in this formulation a
computational mesh/cell) are respectively defined as:
ρ = ϕρl + (1 − ϕ)ρv and μ = ϕμl + (1 − ϕ)μv (5)
The advection term in Eq. (3) is discretized according to the rational

Fig. 5. LDA measurements applied at the annual gap test section (left) and a magnified view of the test section during a measurement.

Fig. 6. Shadowgraphy measurement system for bubble size distribution.

6
F. Kaiser et al. International Journal of Multiphase Flow 174 (2024) 104772

CIP-CSL2 scheme (Nakamura et al., 2001; Xiao et al., 1996). To prevent available experimental data to model the active nucleation site density
smearing of the color function, an interface sharpening algorithm (Sato as a function of the measured wall temperature. In cases in which one
and Niceno, 2012) is employed. Equation (4) is the energy balance needs to simulate nucleate boiling without recourse to experimental
equation, in which Cp (J/m3K) is the specific heat at constant pressure, T data, a correlation for the active Nucleation Site Density, NSD, for
(K) the temperature, λ (W/mK) the thermal conductivity, λt (W/mK) the example those of Hibiki and Ishii (2003) or Kocamustafaogullari and
turbulent thermal conductivity, and Q (W/m3) the volumetric heat Ishii (1983), may be used instead. In this study, the model proposed by
source. The turbulent thermal conductivity is calculated according to Prt the latter authors is used, because it specifically takes into account the
= (μt/ρ)/(λt/Cp), where Prt (= 0.9) is the turbulent Prandtl number. flow rate and the degree of subcooling, thereby reducing the number of
Detailed derivations of Eqs. (1)-(4) are given in Appendices A and B of assumption.
Sato and Niceno (2013). The depletable micro-layer model used here includes the wall-adhe­
Turbulence is modeled here using Large Eddy Simulation (LES), the sion model developed earlier by Sato and Niceno (2015). It has been
Smagorinsky subgrid-scale model (Smagorinsky, 1963) being adopted to shown that the model is a necessary addition to a cell-wise CFD simu­
estimate the turbulent eddy viscosity μt. A Smagorinsky constant of Cs = lation, and is required to resolve the phase-change phenomena taking
0.17, which is the theoretical value derived by Lilly (1967), has been place in the thin liquid film (micro-layer) beneath each growing bubble.
employed in this study. Note that any interaction between the liquid If one were to incorporate the physics of the micro-layer directly into the
turbulence and the liquid-vapor interface is not taken into account in the CFD simulation, the disparity in cell size between the micro-layer and
present formulation. On the right side of Eq. (2), the body-force vector f
→ the bulk of the flow domain would render the calculation unfeasible.
includes both gravity and surface tension forces. To take into account Hence, a micro-layer model of some description is a necessity. In the
the effect of buoyancy, the Boussinesq approximation (Tritton, 1977) is context of the depletable micro-layer model, the micro-layer thickness δ
introduced to the body force via the density difference between liquid (m) is treated as a variable stored at the center of wall-adjacent cells (i.e.
and vapor. The surface tension coefficient is set to a constant value at the the fluid cells next to the wall), with the tacit assumption that the wall is
saturation temperature, meaning that the Marangoni effect (Burdon, coincident with a cell boundary. The thickness of the micro-layer de­
2014) is neglected in the simulations. Brackbill’s Continuum Surface creases as a result of vaporization. In the micro-layer model of Sato and
Force (CSF) model (Brackbill et al., 1992) is employed to represent the Niceno (2015), the initial micro-layer thickness δ0 is defined as δ0 =
effects of surface-tension. In order to avoid strong parasitic currents Cslope rL, where Cslope is constant, obtained either directly from mea­
being generated, we use the density-scaled form of the CSF model here surement or via numerical experiments, and rL (m) is the horizontal
(Brackbill et al., 1992). distance from the nucleation site to the cell center of the wall-adjacent
The phase-change rate at the liquid-vapor interface is computed cell containing the triple line, i.e. the line in which the liquid, vapor
directly from the heat fluxes on the two sides. The temperature at the and solid phases jointly come into contact (on the scale of the CFD
liquid-vapor interface is assumed to be the saturation temperature at the mesh). The linear modeling assumption, i.e. δ0 = Cslope rL, is based on the
prevailing pressure, though away from the interface the conditions of observations of Utaka et al. (2013). For example, Cslope = 4.46 × 10− 3
superheated liquid and subcooled vapor are properly taken into account. was deduced for water, and Cslope = 1.02 × 10− 2 for ethanol, in each case
The phase change rate m˙ at the liquid-vapor interface is modeled as ṁ = relating to pool boiling from a heated quartz glass surface at atmospheric
ql +qv Sint 2 pressure; in addition, Cslope appears to be almost independent of the
L V , where ql and qv are the heat fluxes (W/m ) derived from the
applied heat flux. In our model, we consider that Cslope depends on both
liquid and the vapor sides of the interface, respectively; L (J/kg) is the
the material and surface-roughness properties of the heat-transfer sur­
latent heat of vaporization, Sint (m2) the area of the liquid-vapor inter­
face, as well as on the ambient pressure. Thus, when we simulate
face in the computational cell under consideration, and V (m3) is the cell
nucleate boiling with an unknown combination of working fluid, ma­
total volume. The area Sint is calculated by means of the marching cube
terial properties, surface-roughness and pressure, we first perform a
algorithm (Lorensen and Cline, 1987), which has been proven successful
parameter study to determine Cslope in such a way that the bubble growth
for this type of flow (Sato and Niceno, 2013). The heat fluxes at the
rate agrees with the corresponding measurement. In this study, we adopt
interface are defined as ql = (λl +λt )(∇Tl )⋅→ n and qv = − (λv +
the value Cslope = 2.45 × 10− 2, which had been obtained for the simu­
λt )(∇Tv )⋅→ n , where Tl and Tv are the temperatures in the liquid and lations of Gaertner’s pool boiling experiment (Gaertner, 1965), as
vapor phases, respectively, → n the unit normal vector to the interface, described by Sato and Niceno (2018).
pointing from the vapor to the liquid phase. Note that the temperature The effect of pressure on micro-layer formation has recently been
defined in cells filled with liquid phase and the temperature at measured by using interferometry for pool boiling (Wang et al., 2023)
liquid-vapor interface are both used for the computation of ∇Tl , and the and convective boiling (Kossolapov, 2021) of water, and both the
same is true for the vapor phase. More details of the sharp-interface measurements showed that micro-layers are apparently formed up to
phase-change model are reported in Sato and Niceno (2013). system pressures of 3 bar. Note that Indium Tin Oxide (ITO) was used as
the heat transfer surface in both the experiments, which is differ from
3.2. Nucleation-site model and micro-layer model Zircaloy-4 used in the COSMOS-L experiment, and thus the micro-layer
formation, i.e. thickness and length, may also different especially
The nucleation-site model originally proposed by Sato and Niceno because of the contact angle. Although the micro-layer formation on
(2017, 2018) is used in this work. The locations of the nucleation sites Zircaloy-4 is unknown, we used the micro-layer model in the CFD
are prescribed a priori on the heat-transfer surface, together with a simulation because the COSMOS-L experiment presented in this paper
nucleation activation temperature Tact. Each nucleation site location is was performed at the system pressure of 2 bar, which is lower than the
selected using a non-biased, random number generator in order to threshold of 3 bar.
distribute the nucleation sites on the heat-transfer surface without
examining the micro-scale imperfections of the surface. When the tem­ 3.3. Discretization
perature at any nucleation site reaches Tact, a small vapor bubble (seed
bubble) is placed at the site. The seed bubble is assumed to be initially The governing equations are discretized using a finite-volume
hemi-spherical in shape, the radius being one cell width of the under­ approach. A staggered-variable arrangement (Harlow and Welch,
lying grid. This of course means that the model is dependent on the grid 1965) is adopted, with the vector velocity and body-force components
spacing adopted, and a grid refinement study is then required to eval­ defined on cell faces, and the scalar variables − pressure, temperature,
uate the influence of this assumption on the model predictions. color function and phase-change rate − defined at cell centers. The
The nucleation activation temperature Tact is obtained from the projection method (Chorin, 1968) is used for the coupling between the

7
F. Kaiser et al. International Journal of Multiphase Flow 174 (2024) 104772

pressure and velocity fields. The spatial discretization is based on a surface roughness, which has impact on nucleation site density. Thus,
second-order-accurate, upwind scheme for the advection terms, and a we used the contact angle θ as a tuning parameter to take into account the
second-order, central-difference scheme for the diffusion terms. The unmodelled influence (e.g. surface roughness) in such a way to fit the
time discretization is first-order implicit for the diffusion terms, and calculated temperature to the measurement. Three cases of contact
first-order explicit for the advection terms. A first-order scheme is angle at 27.5˚, 55˚and 110˚have been evaluated. The computed NSD is
employed for the time discretization, instead of a second-order scheme, shown in Fig. 8, and compared with those obtained from the
such as that of Adams-Bashforth, (Ferziger and Perić, 2002). The time Lemmert-Chawla model (Lemmert and Chawla, 1977).
increment Δt is then limited both by the CFL condition (Courant et al., Three simulation cases were computed, each with an applied wall
1928) associated with the explicit time discretization of the advection heat flux of 736 kW/m2. The computational cells are uniform cubes of
term in Eq. (2), and by the surface tension treatment, also in Eq. (2). The 116 μm for the fluid domain and uniform rectangular of thickness 71 μm
details related to the time increment are described in Sato and Niceno for the solid domain in each case, which corresponds to the Medium grid
(2018). It was found that no instability arises in the present calculations used for the grid dependence study in the following section. The
if a safety factor of 0.25 is applied to the CFL limit of Δt. computed wall temperature profiles are compared with measured data
from the COSMOS-L facility in Fig. 9. Here, the computed wall tem­
perature is the averaged value in the lateral direction at the given height.
3.4. Computational domain and boundary conditions
As the contact angle decreases, the wall temperature increases, which is
considered to be reasonable, since NSD is smaller for lower contact
Because the CFD code PSI-BOIL can only deal with an orthogonal
angle. The case with a contact angle of 27.5˚is closest to the measure­
Cartesian grid, the test section was modelled as a rectangular domain, as
ment. Consequently, we adopt θ = 27.5˚for the numerical simulations
illustrated in Fig. 7 (right). This modification artificially decreases the
described in the following sections.
ratio of the volume of the channel to the area of the heat-transfer surface
by 30%, which may result in discrepancies between experimental
measurements and CFD predictions. The computational domain is 356 ×
3.6. Grid dependency study
29.7 × 4.82 mm3 in the axial, lateral and wall-normal directions,
respectively, with 4608 × 384 × 72 cells in each coordinate directions
A grid dependency study has been performed for the three cases with
(≈127 million cells in total). Twelve cells out of the 72 in the wall-
different cell sizes, as listed in Table 1. The directions of Z, Y and r are,
normal direction are used to represent the solid heater. The cell size in
respectively, the axial, lateral and normal-to-wall directions depicted in
the fluid domain is 77 μm in each direction.
Fig. 7. The applied heat flux was set at 596 kW/m2, and the contact angle
The inlet/outlet boundaries are labeled in Fig. 7 (right). At the inlet,
used for the NSD model was set to 27.5˚, for the reasons given in Section
turbulent flow conditions are introduced by means of an anisotropic
3.5. The computations were continued until the wall temperature
Gaussian random process (Li et al., 1994). The boundary conditions at
reached a pseudo-steady condition. The evolution of the temperature
the side of the computational domain are periodic. The outlet boundary
field for the Fine Grid simulation is shown in Fig. 10. The temperatures
conditions are of the Neumann type (i.e. zero gradient) for the velocity,
displayed in the figure correspond to the locations of the thermocouples
pressure and temperature fields. As mentioned earlier, the locations of
in the test section (Fig. 4). In each case, the value is the average tem­
the nucleation sites are chosen using a non-biased, random number
perature in the Y-direction at the given height. The Fine Grid simulation
generator in the same way as for the pool boiling simulations performed
ran for around two months using 512 cores on the supercomputer at
earlier (Sato and Niceno, 2017, 2018).
CSCS, which uses Intel® Xeon® E5-2695 v4 @ 2.10GHz as CPUs.
The computed profiles for the wall superheat are compared in Fig. 11
3.5. Parametric study for nucleation site density, NSD, modelling with the measured data. Note that two measured data points exist for
each height because two thermocouples were employed, which in­
The nucleation site density model proposed by Kocamustafaogullari dicates the uncertainty of temperature with respect to the circumfer­
and Ishii (1983) is employed in this study because it can take into ac­ ential position, and general lack of axial symmetry of the flow. As the
count the influence of flow rate and degree of subcooling on nucleation grid is made finer, higher wall temperatures are predicted. The calcu­
site density. However, the model does not have parameter for e.g., lated temperatures display large grid dependency at the lower location,

Fig. 7. Schematics of (a) the experimental set-up and (b) the computational domain and boundary conditions.

8
F. Kaiser et al. International Journal of Multiphase Flow 174 (2024) 104772

Fig. 8. Comparison of Nucleation Site Density, NSD, as the function of wall superheat between the models of Lemmert-Chawla and the Kocamustafaogullari-Ishii at 2
bar pressure and a mass flow rate 400 kg/m2s with contact angles at 27.5˚, 55˚and 110˚, respectively.

Fig. 9. Comparison of the wall superheat profiles between the measurement at

COSMOS-L and the CFD simulations with different contact angles in the Fig. 10. Computed time-history of wall temperature at the locations corre­
Kocamustafaogullari-Ishi model. sponding to the thermocouples for the Fine grid simulation.

Table 1
Computational grid for the grid dependency study.
Grid Cell size (µm) Number of cells Total number of
name fluid solid X Y r cells
domain domain

Coaese Cube of 155 × 155 2304 192 36 15′925′248

155 × 95
Medium Cube of 116 × 116 3072 256 48 37′748′736
116 × 71
Fine Cube of 78 × 78 × 4608 384 72 127′401′984
78 48

that between the Coarse and Fine simulations being 6 K at the lowest
point. This result indicates that the solution still depends on the grid
size, especially at the lower position, even if the Fine Grid option is Fig. 11. Comparisons of vertical temperature profiles between measurement
employed. Ideally, we should use still finer grids to evaluate further the and simulation for different sizes of grid.
grid dependency effect. However, the computational resources we can
currently access are too limited to allow such a study to be undertaken at criterion, the local exceeding of which is considered as a DNB event.
this time. Thus, we use the results from the Fine Grid simulations in the Unfortunately, this criterion is in opposition to requirements to conduct
following analysis. measurements of bubble velocity, flow and bubble distributions. While
the detection of the spontaneously occurring, CHF requires the shortest
4. Results of experiment and simulation possible reaction time, in combination with a fast power shut-off, the
measurements of the averaged bubble velocity, flow observations and
In the experiment, DNB-type CHF was measured for an applied heat the bubble size distribution require a certain minimum time in excess of
flux of 1.23 MW/m2 in the case of Tin = 80˚C, G = 400 kg/m2s at 2 bar. this. On reaching DNB, the electrical heating must be shut off immedi­
The criterion for DNB was defined to appear spontaneously at a specific ately to avoid damage to the heater. In contrast, estimation of the
applied heat during the pseudo-steady-state measuring procedure. Then averaged bubble velocity and bubble size distributions involves a
a strong local temperature increase of about 200 K/s could be observed. continuous observation of these parameters for some duration in order
Therefore, a maximum heater temperature was selected as a switch-off to avoid overemphasis of singular events.

9
F. Kaiser et al. International Journal of Multiphase Flow 174 (2024) 104772

Therefore, reduced heat flux levels needed to be chosen to investi­ increases rapidly because of the inefficiency of single-phase heat trans­
gate the wall temperatures, flow velocities and bubble size distributions fer. Then, in the downstream of single-phase flow, nucleation sites are
at steady-state conditions. In the studies relating to this paper, the focus activated due to the high wall temperature. The highest temperature
lies on comparison of the experimental and numerical data at a heat flux (≈160˚C) is observed in the zone where the nucleation sites are not yet
corresponding to 50% of the CHF level. At this reduced level, coales­ activated. Downstream of the highest temperature region, the wall
cence phenomena associated with DNB, and hence CHF, are not strongly temperature decreases due to the bubble nucleation and sliding bubbles
pronounced. Thus, the focus here can be set on bubble generation and on the wall.
recondensation without additionally facing the challenging task of Fig. 15 gives a magnified view of the region around Pos 2 at t = 0.80 s
modelling bubble coalescence phenomena. Hence, we concentrate on and t = 0.81 s, in which streaks of high wall temperature are observed
the two vertical measurement positions 1 and 2 (Fig. 4), located at the downstream of a bubble. By comparing Tw at different times, one can
beginning and the middle of the test section, at which coalescence is notice that the high wall temperature streak disappears, as the bubble
almost negligible. slides over it, as indicated by the black dashed-line circle in the figure.

4.1. Flow observations 4.2. Wall superheat

During the experiments, numerous images of the flow were taken Comparisons of the wall temperatures between the measurement and
using standard and high-speed cameras. The videos are provided as simulation are given in Fig. 16. To detect trends in the temperature, for
supplementary information: illustration purposes, we have also included the cases with applied heat
fluxes of 30% and 80% of CHF. In general, the simulation results agree
• Video 12 for the standard camera, and well with measurement. In the case of an applied heat flux 80% of CHF,
• Video 23 for the high-speed cameras. the temperature decreases with height in the region above 250 mm ac­
cording to the simulation, which agrees well with the measurement
Visual observation of the flow gives an insight into bubble formation, data. The decrease in wall temperature downstream can be explained by
bubble lift-off and interaction, as well as of condensation. Selected im­ the increase of the mean flow velocity resulting from the total volume
ages from Video 1 and Video 2 are shown in Fig. 12 and Fig. 13, expansion associated with boiling.
respectively. The first video (snapshots in Fig. 12) shows the upper part The good agreement of the wall superheat between measurement
of the test section for various applied heat fluxes and permits a global and simulation implies that the local heat transfer at the wall, which
view of the flow and evolution of the relevant flow structures to be comprises multiple heat transfer mechanisms resulting from single-
attained. Beginning at a heat flux at 40% of the CHF-value, the images phase liquid, sliding of bubbles along the wall, evaporation and
from the video clearly show the increasingly dynamic nature of the flow. quenching, have all been computed appropriately in the CFD
As can be seen, bubble growth and bubble condensation phenomena simulations.
occurring very rapidly.
Fig. 13 shows the boiling flow at a heat flux corresponding to 50% of 4.3. Void fraction
CHF at three discrete positions, Pos 1-3, which represents flow is the
focus of the present work. The selected images provide representative The computed void fraction at Pos 1-3 are shown in Fig. 17 (left),
insights into the growth and condensation of the steam fraction. How­ representing time- and space-averaged values, α(r). The time-averaging
ever, an important observation from the video obtained from of the has been performed from 0.6 s to 0.9 s, and the space averaging per­
high-speed cameras (Video 2) is that the rates of bubble growth and formed in the Y-direction at the specific elevation and radial position.
condensation are not constant/steady in time, and a periodic pulsation Specifically,
phenomenon is clearly evident. Unfortunately, for positions 1 and 2,
individual images cannot be evaluated from the high-speed video for 1 1
∫t1 ∫Y1
quality reasons, so the time spans of the pulsation could not be consid­ α(r) = α(Y, r, t)dYdt, (6)
t1 − t0 Y1 − Y0
ered exactly. Nevertheless, the images give a good overall impression of t0 Y0

the flow conditions, and show that the relevance of coalescence phe­
nomena at this lower heat flux are almost negligible compared to those where (t0, t1) = (0.6, 0.9) s, and (Y0, Y1) = (0, 29.8) mm; see Fig. 7 (b) for
after critical power. This is a consequence of the low void fraction, the definition of Y.
meaning that the bubbles rarely approach each other and coalesce. The RMS of the void fraction is shown in Fig. 17 (right). As can be
Overall, the videos thus indicate the dynamic character of the flow, with seen, the RMS value is larger than the average value, which indicates
the processes of bubble growth and bubble condensation occurring very that there are large fluctuations of void fraction both in space and time.
rapidly. The time-averaged void fraction in the sliced plane Y2 = 19.6 mm,
A comparison of the bubble shapes observed in the between exper­ α(Y2 , r), at the elevations corresponding to Pos 1-3 are shown in Fig. 18
iment and predicted from the simulation is given in Fig. 14. From the (left), as representative Y-positions. The time-averaged value is defined
experimental findings, only the bubbles on a sliced plane through the as:
annulus have been visualized. The computed bubble shapes and tem­ ∫t1
perature distribution refer to a snapshot taken at t = 0.8 s, when pseudo- α(Y2 , r) =
1
α(Y2 , r, t)dt. (7)
steady-state conditions have been achieved for the wall temperature, see t1 − t0
t0
Fig. 10. As an overall trend, the experimental and the simulation data
display optically good agreement; e.g. the number of bubbles and bubble A sharp peak is observed for Pos 3 at r = 0.4 mm, which is considered
sizes increase in the downstream region. More quantitative comparisons to be caused by the existence of a nucleation site upstream of this
of the bubble shapes and sizes are given in the following sections. The location. The RMS value in the same sliced plane is given in Fig. 18
computed temperature on the heat-transfer surface Twall is also visual­ (right).
ized in Fig. 14. The wall temperature Twall is low at the inlet (80˚C), but
4.4. Bubble velocity

2
https://publikationen.bibliothek.kit.edu/1000126774. Measurements of bubble velocity were made at each of the three
3
https://publikationen.bibliothek.kit.edu/1000126559. vertical measurement positions (Pos 1-3 in Fig. 4) in the axial and radial

10
F. Kaiser et al. International Journal of Multiphase Flow 174 (2024) 104772

Fig. 12. Snapshots of the boiling flow recorded with the standard camera (Video 1) for different applied heat-flux.

Fig. 13. Detailed image of the flow at the three measuring positions with short exposure time recorded with the high-speed camera (Video 2).

directions over the entire annular gap. These measurements were

ρl − ρv dH g Gr
repeated at least three times for every parameter set, so that enough Ri = = 2, (8)
ρl w2 Re
detected bubbles were available for a statistical evaluation. Fig. 19
shows the measured, time-averaged bubble velocities of the axial and
where dH is the hydraulic diameter, Gr the Grashof number, and Re the
radial components in the experiment. The error bars in this figure
Reynolds number. The Richardson number provides a measure of the
represent the deviations of the individual measurements.
strength of the buoyancy force to that of the shear flow force. For values
During the vertical ascent through the annular gap, the bubbles
Ri << 1, the buoyancy force can be neglected. For the present appli­
theoretically experience an acceleration due to (i) their buoyancy and
cation, Ri = 0.14, so that buoyancy terms are considered non-influential,
(ii) acceleration of the liquid phase. Here, the buoyancy force contrib­
resulting in a constant difference between the bubble and upward liquid
utes to the difference of upward velocities between the bubble and liquid
velocities throughout the vertical annulus. Note that this assumption is
phase, while the acceleration of the liquid phase is caused by the in­
only valid if bubble coalescence is negligible, and the bubbles remain
crease of the vapor fraction. However, the measured upward bubble
small. Consequently, the similar upward velocity profiles at Pos 1-3 are
velocity did not show an acceleration with the ascent, as shown in
considered to be caused by (i) non-influential buoyancy force owing to
Fig. 19 (left). We discuss the reason for this here.
Ri << 1, and (ii) the non-significant acceleration of the liquid phase
The liquid-phase velocity field was not specifically measured in the
along the ascent; i.e. no increase of vapor fraction because of subcooling
experiment, but the upward current velocity of the liquid phase has
of liquid in the bulk region.
estimated to be slightly lower than that of the bubbles due to buoyancy
It must be pointed out that the measurement uncertainty of bubble
of bubbles. By considering the Richardson number, the influence of the
velocity in the outer region of r > 3.5 mm is higher than that in the inner
buoyancy force can be quantified:
region, due to the reflection of the laser at the glass tube. Especially, w of

11
F. Kaiser et al. International Journal of Multiphase Flow 174 (2024) 104772

Fig. 14. Comparison of the bubble shapes between experiment and simulation. The experimental images were obtained using high-speed video cameras located at
Pos 1-3 defined in Fig. 4.

Fig. 15. Computed flow field around Pos 2 at t = 0.80 s (left) and t = 0.81 s (right).

Pos 3 in the range r > 3.5 mm is lower than that of Pos 1 and 2; Fig. 19 vertical position; the time-averaging is performed from 0.6 s to 0.9 s. The
(left), which was classified as valid by the measuring system, but velocity profiles are not smooth in the far field, away from the heat-
considered to be affected by the measurement uncertainty. Additionally, transfer surface, because there are insufficient bubbles in this region.
the probe size of the LDA measurement was chosen to be 0.7 mm, so that The peak in the axial vapor velocity profile appears around r = 0.8 ~ 1.1
only bubbles smaller than this were taken into account. mm, where the void fraction reaches its maximum, as seen in Fig. 17
Considering the velocity boundary layer on the heated surface, one (left). The peak observed in Fig. 20 (left) increases in the upward/
can notice that the boundary layer thickness increases downstream, see downstream direction; i.e. the peak at Pos 3 is greater than that at Pos 1.
Fig. 19 (left). The boundary layer is thickened by the radial velocity near Since the axial vapor velocity is accelerated by the buoyancy force
heated wall, as shown in Fig. 19 (right), which results from the bubble acting on the bubble, this result is considered to be reasonable. How­
growth, and lift off process. ever, these features are not seen in the experiment. The peak in the
The numerical simulation also indicates an influence of bubble for­ radial velocity profile, Fig. 20 (right), is around 0.2 m/s, and appears in
mation at the heated rod on the velocity distribution. Fig. 20 shows time the vicinity of the heat-transfer surface. Compared to the measurement
and space-averaged results of the axial and radial velocities. The aver­ result − Fig. 19 (right) − the peak near the wall is similar, but the radial
aging in space is taken over the entire circumferential direction for each velocity in the region away from the wall is markedly different.

12
F. Kaiser et al. International Journal of Multiphase Flow 174 (2024) 104772

Fig. 16. Comparison of wall temperatures between measurement and calculation for an applied heat flux at 30% (left), 50% (middle) and 80% (right) of CHF.

Fig. 17. Computed time- and space-averaged void fraction α(r) at Pos 1-3 (left) and its RMS value (right).

Fig. 18. Computed time-averaged void fraction in the plane Y2 = 19.6 mm α(Y2 , r) at the elevation of Pos 1-3 (left) and its RMS value (right).

Fig. 19. Measured, time-averaged bubble velocity profiles in the annular gap for the axial and radial components at different vertical positions for boundary
condition for 50% applied CHF power.

13
F. Kaiser et al. International Journal of Multiphase Flow 174 (2024) 104772

Fig. 20. Computed velocity profiles of vapor, averaged in space and time.

According to the measurement, the radial velocity asymptotes to zero for experimental data lies in the definition of the geometry. Whereas the
r > 0.5 mm (Fig. 19 (right)). The discrepancy between the measurement experiments were conducted in a cylindrical annular gap, the numerical
(Fig. 19) and numerical prediction (Fig. 20) is considered to be princi­ study was undertaken in a rectangular geometry (Fig. 7). This limitation
pally caused from underestimation of the condensation rate in the CFD of the CFD model has an impact on the behavior of the rising bubbles. In
simulation, as a result of using too coarse a mesh. The bubbles generated Fig. 24, the theoretically available space for growing bubbles in the two
at the heat-transfer surface did not condense sufficiently, which led an geometries is visualized. In this figure, the shaded region indicates the
increased in the radial velocity, which subsequently remained positive available space for the bubble cavities when they have reached their
in the far field (0.5 mm < r < 2 mm); see Fig. 20 (right). maximum size, and start touching each other. For the detachment pro­
The computed velocity profiles of the liquid phase are shown in cess, the available space for the annular gap in the experiments is
Fig. 21; these were not measured in the experiment. The axial velocity expanding, whereas the space available in the numerical set-up stays
profile at Pos 1 is symmetric, because of the rectangular computational constant, as visualized in Fig. 24 (left). This mismatch also results in a
domain, but the higher velocity region moves toward the heat-transfer difference in the pressure distribution across the channel. The differ­
surface at upper elevations, due to the influence of the increased axial ences in pressure could have an additional accelerating effect on the
velocity of the bubbles in this region, and their induced drag on the bubbles towards the outer glass tube in the experiment, though this has
liquid phase. The radial velocity in the liquid phase is much smaller, and not been specifically investigated.
almost zero for r > 3 mm at all elevations.
What is remarkable regarding the experimental results is the 4.5. Bubble size distribution
continuously increasing RMS value of the vertical bubble velocity in the
downstream direction, as shown in Fig. 22 (left). The axial component of The bubble size has an influence on the velocity and temperature
the RMS velocity, wRMS, at the highest position is identical to the value of fields, and is one of the key factors in accurately predicting the total heat
the averaged vertical velocity, as seen in Fig. 19. This is a consequence of and mass transfer rates. Fig. 25 (left) gives the distribution of the bubble
the observed strong pulsations in the flow. The radial RMS velocities, diameters measured at Pos 1 and Pos 2. Almost 90% of the bubbles are in
uRMS, for the lower two positions are much larger than that at the highest the range 0-200 µm, and bubbles of diameter larger than 400 µm have a
position, and indeed much larger than wRMS values. However, the width relative frequency below 6%. Bubbles larger than 700 µm in diameter
of influence is largest for the highest position, Pos 3, and this extends can be found only at the Pos 2. The bubble size distribution is similar for
over almost the entire annular gap. the two positions, since the bubbles collapse quickly after detachment
The calculated RMS values of the bubble velocities are shown in from the heated surface, and do not flow downstream, a feature which
Fig. 23. The values in the far field (r > 2.5 mm) fluctuate because there could also be observed in Video 1.
are insufficient bubbles representing the vapor phase for a statistical Fig. 25 (right) shows the computed distributions. Again, a bubble
average to be meaningful. The RMS predictions for the axial velocity are diameter of 0-200 µm represents about 90% of the total. However, the
similar between Pos 1-3, and the value at Pos 3 resembles the experi­ bubble size distribution above 300 µm diameter displays a different
ment (Fig. 22 left): the peak wRMS is 0.4 m/s, located near the heat- tendency. From the measurements, the relative frequency of the
transfer surface, and decreases to half that value at r = 0.5 mm. From appearance of such bubbles gradually decreases as the bubble diameter
the simulation, the RMS for the radial velocity is generally smaller than increases. In contrast, according to the simulation, bubbles of diameter
that measured, which is considered to be also due to the underestimation 300-700 µm are absent entirely, and those larger than 1000 µm are
of condensation by too coarse a mesh being employed. observed instead. The aberration is considered to be caused by numer­
One of the major differences between the numerical and ical error, the so-called spurious numerical bubble coalescence

Fig. 21. Computed velocity profiles of the liquid phase averaged in space and time.

14
F. Kaiser et al. International Journal of Multiphase Flow 174 (2024) 104772

Fig. 22. RMS values corresponding to the averaged velocity profiles of Fig. 19.

Fig. 23. Computed RMS of the vapor velocity, averaged over space and time.

because of the reduced interfacial area density. As a result, sliding of

large bubbles along the heated surface is predicted in the simulation, as
indicated in Fig. 15, but not observed in the experiment.
It is of interest to investigate the bubble size distribution in the radial
direction at the different elevations from the viewpoint of the bubble
dynamics as influenced by the overall liquid-vapor mass transfer. Fig. 26
compares experiment (left) and simulation (right) in this regard. Both
figures reveal that the larger bubbles appear mainly in the vicinity of the
Fig. 24. Differences in geometry between the numerical (left) and experi­ heat-transfer surface, whereas the smaller bubbles, i.e. those of di­
mental (right) studies. ameters less than 200 µm, are present across the entire width of the
annulus. The differences in bubble size distribution between Pos 1 and
phenomenon (Coyajee and Boersma, 2009). When the distance of the Pos 2 are marginal for both the experiment and simulation. The
interfaces between two separate bubbles becomes narrower than one measured bubble size distribution shows the presence of larger bubbles
computational grid size, the bubbles are automatically merged in the (dia. > 500 µm) near the heated surface, i.e. at r ≈ 0.5 mm. With
simulation. To avoid this artificial numerical bubble coalescence, a more increasing distance from the heated surface, the bubble diameters
sophisticated model, e.g. a multiple marker formulation (Cifani et al., continuously decrease, and from r = 1.0 mm outwards only small bub­
2018; Kwakkel et al., 2013) must be used. This extension of the model is bles, i.e. with diameters below 200 µm, are present in the flow.
here postponed to a future work. Once a larger bubble is generated According to the computation, Fig. 26 (right), most of the larger
artificially, then it condenses more slowly than the smaller bubbles, bubbles lie along the line dia. = 2r, and the smaller ones slightly above

Fig. 25. Bubble size distribution measured (left) and computed (right) for an applied heat-flux of 50% CHF.

15
F. Kaiser et al. International Journal of Multiphase Flow 174 (2024) 104772

Fig. 26. Bubble diameter distribution as a function of radial position, measured (left) and computed (right), for an applied heat-flux of 50% CHF.

the line dia. = grid spacing (= 76 µm). A bubble with dia. = 2r represents temperature at the site attained the activation temperature of 135.2˚C.
a spherical bubble attached to, or sliding along, the heat transfer surface. Note that the locations of the sites are prescribed in the simulation,
The bubbles clustered around the line dia. = 2r are considered to be an randomly distributed on the heat-transfer surface, with each site having
artifact of the spurious bubble coalescence alluded to earlier. The a specific activation temperature. The bubble grows, and the tempera­
computed bubbles slightly above the line dia. = grid spacing may indicate ture of the solid wall beneath the bubble decreases, as seen in Fig. 28 (1)-
the capability of the interface tracking method to predict the bubble (3). The temperature drop in the solid around the nucleation site reflect
dynamics in the low-void regions, since the measurements reflect similar the energy expended through mass transfer (latent heat of evaporation),
behavior. as observed in Fig. 29 (1)-(3). The bubble shape at time (1) is squat, as a
consequence of the micro-layer modelling, the bubble diameter on the
wall reaching a maximum at image (2). Then, the bubble continues to
4.6. Bubble liftoff without sliding motion
grow in the wall normal direction, as shown in (3)-(4). During this
period, vaporization takes place at the bottom of bubble and concur­
In this section, we consider the mechanisms of bubble liftoff without
sliding motion, as was observed in the experiment. Fig. 27 illustrates a rently condensation occurs at the top, see Fig. 29 (3)-(4). In the situation
typical evolution of a seed bubble, as observed in the experiment with a seen in Fig. 29 (5), the rate of vaporization at the base of the bubble
temporal resolution of 3000 fps. The bubble then is first nucleated be­ decreases, and the vapor condenses at the top, which “pulls” the bubble
tween the timeframe (1) and (2) in Fig. 27, and grows on the wall during away from the wall. The computed bubble shape in Fig. 28 (5), i.e.
timeframes (2)-(3). The bubble lifts off at timeframe (4), any sliding of elongated in the wall normal direction, is very similar to that observed in
the bubble along the wall not being observed. The bubble moves to the the experiment, Fig. 27 (6). From these observations, one can conclude
bulk region and condensates, as shown in the timeframes (5)-(6). The that the bubble does not slide along the wall, but lifts off immediately
bubble shape is elongated in the wall normal direction, as seen in following bubble growth, due to the condensation at the upper surface of
timeframe (6). the bubble in the subcooled flow stream.
Qualitatively, similar bubble motion is predicted from the simula­ The bubble dynamics during subcooled, forced-convection nucleate
tion, as evidenced in Fig. 28, in which the liquid-vapor interface and the boiling, i.e. bubble growth, sliding, lift-off and collapse, is significantly
temperature distribution are visualized at a frequency of 6000 fps. It is to influenced by the degree of subcooling in the bulk flow. In the case that
be noted that not all the bubbles lift off without sliding, as shown in this the degree of subcooling is large, i.e. Tsub > 55 K for water at atmo­
figure, but many slide along the wall without lift off, as displayed in spheric pressure (Gunther, 1950), bubbles grow and collapse while they
Fig. 15, and can also be seen in Video 1 of the experiment. Since the are still attached to the heated surface, without sliding or lifting off. In
distribution in a sliced plane is drawn in this figure, only a few bubbles contrast, in the case of saturated boiling, bubbles grow, slide and then
are to be seen. In contrast, multiple bubbles are observed in the exper­ lift off; observations of this phenomenon, using high-resolution mea­
iment (Fig. 27), as a result of the perspective view. The evolutions of the surement techniques, were reported, for example by Klausner et al.
mass-transfer rate distribution and bubble shape are displayed in (1993) and Maity (2000). The experiment of forced convection nucleate
Fig. 29, the location being the same as in Fig. 28. boiling flow in a vertical channel performed by Maity (2000) at satu­
In Fig. 28 (1), the nucleation site has been activated because the ration conditions has been simulated by Li and Dhir (2007) and Sato

Fig. 27. Evolution of bubble growth and lift-off recorded by the high-speed video camera at Pos 3, visualized at 3000 fps.

16
F. Kaiser et al. International Journal of Multiphase Flow 174 (2024) 104772

Fig. 28. Evolution of bubble and temperature distributions in a Y-constant plane, visualized at 6000 fps.

Fig. 29. Evolution of bubble (black line) and mass transfer distributions (color) in a Y-constant plane, visualized at 6000 fps.

et al. (2013), and sliding and lift off were well reproduced in the CFD simulation. In the COSMOS-L experiment, a high-speed camera was used
simulations. to capture the bubble dynamics, LDA was employed to measure the
The feature of the bubble motion measured and simulated in the bubble velocities, and shadowgraphy used to estimate the bubble-size
present study, i.e. lift off without any apparent sliding motion, has also distribution. The water quality, flowrate, inlet temperature and heater
been reported by Van Helden et al. (1995), though the degree of sub­ power were strictly controlled during the experiment, and the temper­
cooling is not mentioned in the paper. The authors considered that the ature distribution of the heater rod monitored using 12 thermocouples.
temperature difference in the liquid phase between the near-wall and In the CFD simulation, an interface-tracking scheme was employed
bulk flow regions results in a change in the surface tension coefficient, to capture the vapor/liquid boundaries, together with a sharp-interface
which produces a Marangoni flow (Burdon, 2014), which was assumed phase-change model and a micro-layer model, as described in earlier
to induce lift off. They named the force induced by the Marangoni flow works. As a boundary condition at the heat-transfer surface, an empir­
the temperature drop force. ical correlation for nucleation site density, based on the Kocamusta­
In contrast, the Marangoni force is not modelled in the present faogullari and Ishii model, was employed, since the nucleation site
simulation, since the Continuum Surface Force, CSF, model is employed density was not specifically measured in the experiment. This intro­
for the surface tension force, which cannot take into account this effect duced uncertainties into the simulation. Ideally, the phase-change
(Brackbill et al., 1992). However, the bubble dynamics, i.e. lift-off model requires at least 4~5 computational cells to capture the ther­
without sliding, shows good agreement with the measurement, which mal boundary layer around the liquid-vapor interface. But we could not
appears to imply that the Marangoni effect is not the main driving force achieve this requirement, since the thermal boundary layer is too thin in
in the lift-off phenomenon. What is considered important in the present this application, as a result of the subcooled boiling flow, even though
work is the condensation that takes place at the top of the bubble, i.e. we have used a supercomputer for the numerical simulation. Nonethe­
that close to the central, subcooled flow, as visualized in Fig. 29 (3)-(5). less, the CFD results have been compared with measurement, in order to
The condensation and associated shrinkage of volume at the “top” of the clarify the limitations of the current CFD interface-tracking algorithm
bubble, “sucks” the bubble into the bulk, triggering the lift-off event. for applications of this type. Good agreement has been obtained for the
wall temperatures at all the available measurement points, i.e. at six
5. Conclusions different elevations. This implies that the total heat transfer at the wall,
which includes the heat transfer due to single-phase liquid, evaporation,
In this paper, we have investigated subcooled forced convection sliding of bubbles along the wall and quenching, were all computed
boiling flow around a heated rod at 2 bar, as measured in the COSMOS-L appropriately.
facility, and computed using a CFD simulation tool. The ultimate goal of Consequently, we expect to learn the following from this paper:
the study is to increase understanding of the mechanism of DNB-type
CHF, and the direct prediction of it via numerical simulation. As a • The kind of data that can be obtained using a state-of-the-art mea­
first step, a flow with an applied heat flux of 50% CHF has been inves­ surement system consists of a test section incorporating large optical
tigated, to evaluate the capabilities and limitations of CFD simulation windows, LDA, shadowgraphy, thermocouples, and a high-speed
tools currently being applied to the phenomenon. Since DNB is essen­ camera. This last produces highly resolved photos of subcooled
tially a transient, local phenomenon, methods are needed that can convective nucleate boiling at the high frequency of 6000 fps,
resolve the boiling flow situation with sufficient space and time reso­ together with transient data concerning bubble velocities, bubble
lution; this is true both for the measurement techniques and for the size distributions and temperature profiles.

17
F. Kaiser et al. International Journal of Multiphase Flow 174 (2024) 104772

• The CFD simulation, with interface tracking, can provide a three- Ferziger, J.H., Perić, M., 2002. Methods for Unsteady Problems. In: Ferziger, J.H.,
Perić, M. (Eds.), Computational Methods for Fluid Dynamics. Springer Berlin
dimensional distribution of the flow field, including velocity, pres­
Heidelberg, Berlin, Heidelberg, pp. 135–156.
sure, temperature, void and mass transfer rate distributions. Fischer, F., Hampel, U., 2010. Ultra fast electron beam X-ray computed tomography for
• Based on the measurement and simulation data collected, the bubble two-phase flow measurement. Nucl. Eng. Des. 240, 2254–2259.
dynamics at bubble nucleation and lift off can be uniquely clarified. Fuchs, T., Kern, J., Stephan, P., 2006. A transient nucleate boiling model including
microscale effects and wall heat transfer. J. Heat. Transfer. 128, 1257–1265.
• Direct comparison between measurement and simulation reveals the Gabriel, S., Schulenberg, T., Albrecht, G., Heiler, W., Miassoedov, A., Kaiser, F.,
limitations of the CFD simulation: i.e. overprediction of bubble Wetzel, T., 2018. Optical void measurement method for stratified wavy two phase
merging and underprediction of the condensation process in the bulk flows. Exp. Therm. Fluid. Sci. 97, 341–350.
Gabriel, S.G., 2014. Experimental investigation of droplet separation in a horizontal
flow, due to too coarse a computational grid being employed, this as countercurrentair/water stratified flow. KIT Scientific Publishing, Karlsruhe.
a result of the complexity of the physics, and restricted computa­ Gaertner, R.F., 1965. Photographic study of nucleate pool boiling on a horizontal surface.
tional capability. J. Heat. Transfer. 87, 17–27.
Giustini, G., Walker, S.P., Sato, Y., Niceno, B., 2017. Computational fluid dynamics
analysis of the transient cooling of the boiling surface at bubble departure. J. Heat.
CRediT authorship contribution statement Transfer. 139, 091501.
Gong, S., Cheng, P., 2015. Lattice Boltzmann simulations for surface wettability effects in
saturated pool boiling heat transfer. Int. J. Heat. Mass Transf. 85, 635–646.
Florian Kaiser: Investigation, Methodology, Validation, Writing – Gong, S., Cheng, P., 2016. Two-dimensional mesoscale simulations of saturated pool
original draft. Yohei Sato: Methodology, Software, Validation, Writing boiling from rough surfaces. Part II: Bubble interactions above multi-cavities. Int. J.
– original draft, Writing – review & editing. Stephan Gabriel: Heat. Mass Transf. 100, 938–948.
Gunther, F.C., 1950. Photographic study of surface-boiling heat transfer to water with
Conceptualization, Funding acquisition, Methodology, Resources, Su­ forced convection.
pervision, Writing – original draft, Writing – review & editing. Haas, C., 2012. Critical heat flux for flow boiling of water at low pressure on smooth and
micro-structured zircaloy tube surfaces. KIT.
Haas, C., Kaiser, F., Schulenberg, T., Wetzel, T., 2018. Critical heat flux for flow boiling
Declaration of competing interest of water on micro-structured Zircaloy tube surfaces. Int. J. Heat. Mass Transf. 120,
793–806.
Harlow, F.H., Welch, J.E., 1965. Numerical calculation of time-dependent viscous
The authors declare the following financial interests/personal re­ incompressible flow of fluid with free surface. Physics of Fluids 8, 2182–2189.
lationships which may be considered as potential competing interests: Hibiki, T., Ishii, M., 2003. Active nucleation site density in boiling systems. Int. J. Heat.
Mass Transf. 46, 2587–2601.
Stephan Gabriel reports financial support was provided by the Huber, G., Tanguy, S., Sagan, M., Colin, C., 2017. Direct numerical simulation of nucleate
German Society for Plant and Reactor Safety (GRS) Projekt No. pool boiling at large microscopic contact angle and moderate Jakob number. Int. J.
1501473B. If there are other authors, they declare that they have no Heat. Mass Transf. 113, 662–682.
Jamet, D., Lebaigue, O., Coutris, N., Delhaye, J.M., 2001. The second gradient method
known competing financial interests or personal relationships that could for the direct numerical simulation of liquid-vapor flows with phase change.
have appeared to influence the work reported in this paper. J. Comput. Phys. 169, 624–651.
Kharangate, C.R., Mudawar, I., 2017. Review of computational studies on boiling and
condensation. Int. J. Heat. Mass Transf. 108, 1164–1196.
Data availability
Klausner, J.F., Mei, R., Bernhard, D.M., Zeng, L.Z., 1993. Vapor bubble departure in
forced convection boiling. Int. J. Heat. Mass Transf. 36, 651–662.
Data will be made available on request. Kocamustafaogullari, G., Ishii, M., 1983. Interfacial area and nucleation site density in
boiling systems. Int. J. Heat. Mass Transf. 26, 1377–1387.
Kossolapov, A., 2021. Experimental investigation of subcooled flow boiling and CHF at
prototypical pressures of light water reactors. Nuclear Science and Engineering.
Acknowledgements Massachusetts institute of technology.
Kunkelmann, C., Stephan, P., 2009. CFD simulation of boiling flows using the volume-of-
fluid method within OpenFOAM. Numerical Heat Transfer, Part A: Applications 56,
The authors would like to thank the former colleague Dr B. Smith for 631–646.
many valuable discussions, English corrections and for his careful Kwakkel, M., Breugem, W.-P., Boersma, B.J., 2013. Extension of a CLSVOF method for
droplet-laden flows with a coalescence/breakup model. J. Comput. Phys. 253,
reading of the final manuscript. This work was partially supported by a
166–188.
grant from the Swiss National Supercomputing Centre (CSCS) under Lee, R.C., Nydahl, J.E., 1989. Numerical calculation of bubble growth in nucleate boiling
project ID psi05. Furthermore, part of the experimental work was car­ from inception through departure. J. Heat. Transfer. 111, 474–479.
Lemmert, M., Chawla, J.M., 1977. Influence of flow velocity on surface boiling heat
ried out as part of the NUBEKS BMWi 1501473B project.
transfer coefficient. In: Hahne, E., Grigull, U. (Eds.), Heat transfer in boiling.
Academic Press.
References Levy, S., 1959. Generalized correlation of boiling heat transfer. J. Heat. Transfer. 81,
37–42.
Li, A., Ahmadi, G., Bayer, R.G., Gaynes, M.A., 1994. Aerosol particle deposition in an
Badillo, A., 2012. Quantitative phase-field modeling for boiling phenomena. Phys. Rev. E
obstructed turbulent duct flow. J. Aerosol. Sci. 25, 91–112.
86, 041603.
Li, D., Dhir, V.K., 2007. Numerical study of single bubble dynamics during flow boiling.
Brackbill, J.U., Kothe, D.B., Zemach, C., 1992. A continuum method for modeling surface
J. Heat. Transfer. 129, 864–876.
tension. J. Comput. Phys. 100, 335–354.
Li, Q., Kang, Q.J., Francois, M.M., He, Y.L., Luo, K.H., 2015. Lattice Boltzmann modeling
Bruder, M., Sembach, L., Lampl, D., Hirsch, C., Sattelmayer, T., 2019. Local
of boiling heat transfer: The boiling curve and the effects of wettability. Int. J. Heat.
measurements on vertical subcooled flow boiling of refrigerant Novec 649. Int. J.
Mass Transf. 85, 787–796.
Multiphase Flow 119, 108–122.
Lilly, D.K., 1967. The representation of small-scale turbulence in numerical simulation
Burdon, R.S., 2014. Surface Tension and the Spreading of Liquids, paperback ed.
experiments. In: IBM sci. comp. symp. on environmental sciences. New York.
Cambridge University Press, Cambridge.
Lorensen, W., Cline, H., 1987. Marching cubes: A high resolution 3D surface construction
Büttner, F., Heiler, W., Gabriel, S., Kuhn, D., 2018. Experimental investigation of bubble
algorithm. In: SIGGRAPH ’87: Proceedings of the 14th Annual Conference on
entrainment by a vertical jet plunging into a liquid water pool, Flüssigbad
Computer Graphics and Interactive Techniques. ACM, pp. 163–169.
eintauchenden Freistrahl. Vortrag gehalten auf 26. Fachtagung Experimentelle
Maity, S., 2000. Effect of velocity and gravity on bubble dynamics. University of
Strömungsmechanik (2018), Rostock, Germany.
California, Los Angeles.
Chorin, A.J., 1968. Numerical solution of the Navier-Stokes equations. Math. Comput.
Mittal, R., Iaccarino, G., 2005. Immersed boundary methods. Annu Rev. Fluid. Mech. 37,
22, 745–762.
239–261.
Cifani, P., Kuerten, J.G.M., Geurts, B.J., 2018. Highly scalable DNS solver for turbulent
Murallidharan, J., Giustini, G., Sato, Y., Ničeno, B., Badalassi, V., Walker, S.P., 2016.
bubble-laden channel flow. Comput. Fluids. 172, 67–83.
Computational fluid dynamic simulation of single bubble growth under high-
Courant, R., Friedrichs, K., Lewy, H., 1928. Über die partiellen Differenzengleichungen
pressure pool boiling conditions. Nucl. Eng. Technol. 48, 859–869.
der mathematischen Physik. Math. Ann. 100, 32–74.
Nakamura, T., Tanaka, R., Yabe, T., Takizawa, K., 2001. Exactly conservative semi-
Coyajee, E., Boersma, B.J., 2009. Numerical simulation of drop impact on a liquid–liquid
Lagrangian scheme for multi-dimensional hyperbolic equations with directional
interface with a multiple marker front-capturing method. J. Comput. Phys. 228,
splitting technique. J. Comput. Phys. 174, 171–207.
4444–4467.
Prasser, H.M., 1999. Wire-mesh sensors for two-phase flow investigations.
Dhir, V.K., 1998. Boiling heat transfer. Annu Rev. Fluid. Mech. 30, 365–401.
Forschungszentrum Rossendorf, Germany, pp. 23–28.
Esmaeeli, A., Tryggvason, G., 2004. Computations of film boiling. Part I: numerical
Rohsenow, W.M., 1971. Boiling. Annu Rev. Fluid. Mech. 3, 211–236.
method. Int. J. Heat. Mass Transf. 47, 5451–5461.

18
F. Kaiser et al. International Journal of Multiphase Flow 174 (2024) 104772

Sato, Y., Lal, S., Niceno, B., 2013. Computational fluid dynamics simulation of single Stäbler, T., 2007. Experimentelle Untersuchung und physikalische Beschreibung der
bubble dynamics in convective boiling flows. Multiphase Sci. Technol. 25, 287–309. Schichtenströmung in horizontalen Kanälen. Karlsruhe Institute of Technology.
Sato, Y., Niceno, B., 2012. A conservative local interface sharpening scheme for the Takada, N., Tomiyama, A., 2007. Numerical simulation of isothermal and thermal two-
constrained interpolation profile method. Int. J. Numer. Methods Fluids. 70, phase flows using phase-field modeling. Int. J. Mod. Phys. C 18, 536–545.
441–467. Tompkins, C., Prasser, H.-M., Corradini, M., 2018. Wire-mesh sensors: a review of
Sato, Y., Niceno, B., 2013. A sharp-interface phase change model for a mass-conservative methods and uncertainty in multiphase flows relative to other measurement
interface tracking method. J. Comput. Phys. 249, 127–161. techniques. Nucl. Eng. Des. 337, 205–220.
Sato, Y., Niceno, B., 2015. A depletable micro-layer model for nucleate pool boiling. Tritton, D.J., 1977. Thermal Flows: Basic Equations and Concepts. In: Tritton, D.J. (Ed.),
J. Comput. Phys. 300, 20–52. Physical Fluid Dynamics. Springer, Netherlands, Dordrecht, pp. 127–134.
Sato, Y., Niceno, B., 2017. Nucleate pool boiling simulations using the interface tracking Utaka, Y., Kashiwabara, Y., Ozaki, M., 2013. Microlayer structure in nucleate boiling of
method: Boiling regime from discrete bubble to vapor mushroom region. Int. J. Heat. water and ethanol at atmospheric pressure. Int. J. Heat. Mass Transf. 57, 222–230.
Mass Transf. 105, 505–524. Van Helden, W.G.J., Van Der Geld, C.W.M., Boot, P.G.M., 1995. Forces on bubbles
Sato, Y., Niceno, B., 2018. Pool boiling simulation using an interface tracking method: growing and detaching in flow along a vertical wall. Int. J. Heat. Mass Transf. 38,
From nucleate boiling to film boiling regime through critical heat flux. Int. J. Heat. 2075–2088.
Mass Transf. 125, 876–890. Wang, J., Wang, H., Xiong, J., 2023. Experimental investigation on microlayer behavior
Sato, Y., Smith, B.L., Niceno, B., 2018. Examples of Pool-Boiling Simulations Using an and bubble growth based on laser interferometric method. Front. Energy Res. 11.
Interface Tracking Method Applied to Nucleate Boiling, Departure from Nucleate Welch, S.W.J., 1998. Direct simulation of vapor bubble growth. Int. J. Heat. Mass Transf.
Boiling and Film Boiling, Encyclopedia of Two-Phase Heat Transfer and Flow III. 41, 1655–1666.
WORLD SCIENTIFIC, pp. 225–263. Welch, S.W.J., Wilson, J., 2000. A volume of fluid based method for fluid flows with
Smagorinsky, J., 1963. General circulation experiments with the primitive equations. phase change. J. Comput. Phys. 160, 662–682.
Mon. Weather. Rev. 91, 99–164. Xiao, F., Yabe, T., Ito, T., 1996. Constructing oscillation preventing scheme for advection
Son, G., Dhir, V.K., 2008. Numerical simulation of nucleate boiling on a horizontal equation by rational function. Comput. Phys. Commun. 93, 1–12.
surface at high heat fluxes. Int. J. Heat. Mass Transf. 51, 2566–2582. Yadigaroglu, G., 2005. Computational Fluid Dynamics for nuclear applications: from
Son, G., Dhir, V.K., Ramanujapu, N., 1999. Dynamics and heat transfer associated with a CFD to multi-scale CMFD. Nucl. Eng. Des. 235, 153–164.
single bubble during nucleate boiling on a horizontal surface. J. Heat. Transfer. 121,
623–631.

19