comparison of mixture and multifluid models for in-nozzle cavitation prediction
TRANSCRIPT
1 Copyright © 2013 by ASME
Proceedings of the ASME 2013 Internal Combustion Engine Division Fall Technical Conference ICEF2013
October 13-16, 2013, Dearborn, Michigan, USA
ICEF2013-19093
COMPARISON OF MIXTURE AND MULTI-FLUID MODELS FOR IN-NOZZLE CAVITATION PREDICTION
Michele Battistoni1,2, Sibendu Som1 and Douglas E. Longman1 1Energy Systems Division, Argonne National Laboratory, Argonne, IL, 60439, USA
2Department of Industrial Engineering, University of Perugia, 06125, Italy
ABSTRACT Fuel injectors often feature cavitation because of large
pressure gradients which in some regions lead to extremely low
pressure levels. The objective of this paper is to compare the
prediction capabilities of two computational fluid dynamics
(CFD) codes for modeling cavitation in small channel flows, like
those used in diesel or gasoline fuel injectors. Numerical results
are assessed against quantitative high resolution experimental
data collected at Argonne National Laboratory using synchrotron
x-ray radiography of a model nozzle. The first numerical
approach is based on the homogeneous mixture model, phase
change is modeled via the Homogeneous Relaxation Model
(HRM), and it is implemented in CONVERGE code. The second
approach is based on the multi-fluid non-homogeneous model as
implemented in AVL-FIRE code, and it uses the Rayleigh bubble
dynamics model to account for cavitation. One key feature of the
work is that both models take into account the presence of
dissolved gases in the multi-phase flow. This effect has been
accounted for by running compressible three-phase flow
simulations.
Results indicate that both modeling approaches are capable
of capturing the local and global effects of cavitation and void
formation due to either phase change or expansion of the
dissolved air. From a quantitative standpoint, the amount of void
predicted by the multi-fluid model is in good agreement with
measurements, while the mixture model overpredicts the values.
Qualitatively, void regions look similar and compare well with
the experimental measurements, especially in the nozzle
entrance region where most of the vapor is produced. Some
difference has been noted in the centerline of the channel towards
the exit, where void due to dissolved gas expansion has been
observed.
The study highlights the importance of accounting for
dissolved gases in the liquid, especially if the outlet pressure is
low. If dissolved gases are neglected, results underestimate the
centerline void formation. Grid converged results have been
achieved for the prediction of mass flow rate, while grid-
convergence for void fraction is still an open issue. The paper
also includes a discussion about the effect of turbulent pressure
fluctuations on cavitation inception.
INTRODUCTION Prediction and control of sprays emerging from fuel
injectors are considered a key aspect for improving the
performance of modern direct injection engines. The flow
characteristics developed inside an injector determine many
aspects of the resulting jet and of the following combustion
process. The ultimate purpose of the spray is to increase the
liquid surface area exposed to the mass, momentum, and heat
transfer phenomena. Cavitation of the fuel is commonly
encountered even at relatively low injection pressure, but can be
enhanced by either increasing the injection pressure or
decreasing the outlet pressure. Nozzles can experience fully
developed cavitation or even string-type cavitation depending on
the sac flow field [1]. Several studies carried out on Diesel or
gasoline direct injection (GDI) injectors [2,3,4,5,6,7,8,9]
reported that cavitation inside the nozzle orifice generates
increased turbulence that contributes greatly to the disintegration
of the liquid jet, improving the primary breakup and the
subsequent atomization process. Another positive outcome of
cavitation, which is usually observed, is a larger spray cone
angle. However, lower injected mass is generally achieved
because of the reduction of the effective outlet area. In addition,
material erosion and damages to the internal surfaces are
observed. These advantages and disadvantages both provide
significant motivation to study the phenomena.
The size of typical fuel nozzles have made the observation
of the flow field extremely difficult for quite long time, and
researchers have used scaled-up and transparent nozzles to study
NOT PUBLISHED
2 Copyright © 2013 by ASME
the internal fluid dynamics. However, over the past decade, the
advances in instrumentation technology have allowed more
information to be obtained in real-size injector nozzles
[10,11,12,13]. These experiments have pointed out the presence
of various types of cavitation structures inside the injectors like
attached cavitation, bubble clouds, and string-type cavitation.
More recently, x-ray radiography has been used to provide
quantitative measurements of local void fractions in cavitating
flows, as reported by Duke et al. [14]. Medical x‐ray tomography
has also been used by Bauer et al. [15] to measure void fraction
in a scaled‐up cavitating pipe flow. Lastly, x-ray phase contrast
imaging experiments of internal injector geometry and needle lift
have also been reported by Kastengren et al. [16,17], very
recently.
At the same time, numerous modeling studies of nozzle flow
cavitation can be found in literature. An extensive coverage of
the topic is given by Schmidt and Corradini [18] or by
Giannadakis [19]. Different approaches can be adopted to treat
the two-, or multi-, phase flow. They can be classified either
according to the model of multi-phase fluid adopted or according
to the mass transfer mechanism assumed for cavitation.
Concerning the multi-phase model criteria, and with specific
reference to cavitation inside Diesel fuel nozzles, at least three
approaches can be identified: the homogeneous Eulerian models
[20,21,22,23], the multi-fluid Eulerian models [24, 25], and the
Lagrangian models [27,28].
The common feature of the homogeneous method is the
assumption that all the phases share the same velocity. In this
case the fluid is a continuous mixture of liquid and vapor. They
can assume different forms depending on how density and
pressure are formulated. The mixture is usually compressible. In
some implementation the effects of turbulence have been
neglected [32]. From the numerical perspective the Volume-of-
fluid (VOF) model, which allows the resolution of sharp
interfaces, is very similar to the homogeneous model. A single
momentum equation is calculated for all phases that interact
using the VOF model. An implementation of this has been
proposed by Marcer et al. [23]. The basic drawback of this
approach is the assumption that cavitation area is delimited by a
large-scale interface, but this rules out the possibility of
dispersed bubbly flows as observed in experiments.
The second aforementioned approach is the Eulerian multi-
fluid method. This method is characterized by different sets of
conservation equations, one for each phase, thus the main feature
is that each phase has its own velocity. One of the first
application of this method to fuel injector flows is reported by
Alajbegovic et al. [24,25].
The third approach is the Lagrangian method. Here, only the
liquid is a continuum, while the vapor is the dispersed phase and
it is represented by parcels of bubbles. Vapor bubble trajectories
are tracked integrating the Newton equation of motion for each
parcel. Examples of this method applied to nozzle flows are
reported by Giannadakis et al. [27, 28].
Concerning the classification criteria based on the type of
mass transfer model, two macro approaches can be identified.
The first one uses thermodynamic considerations in which the
pressure is treated only as a thermodynamic variable, and
thermal equilibrium (Homogeneous Equilibrium Model - HEM)
between liquid and vapor is assumed [29,30,31]. Also a thermal
non-equilibrium version has been proposed for flash boiling
phenomena (Homogeneous Relaxation Model - HRM) [32,33].
This type of approach is generally implemented in the
homogeneous mixture fluid model.
The second group of cavitation models are based on the
assumption that the pressure difference between the inner bubble
and the surrounding liquid acts as a mechanical force which
determines whether a vapor bubble is expanding or collapsing.
The Rayleigh-Plesset (R-P) equation is generally used for
tracking the bubble dynamics and from its time integration the
mass transfer rate can be derived. Applications of this approach
to nozzle flow simulations are given by Giannadakis et al. [27,
28] and Alajbegovic at al. [24,25,34].
In the context of such a wide panorama it is difficult to find
studies that have addressed the quantitative comparison of
cavitation models with comprehensive experimental data. In the
present paper an attempt to validate numerical models has been
made by comparing the results with void fraction measurements
performed at Argonne National Laboratory using synchrotron x-
rays [14]. The paper focuses on the analysis of cavitation, using
a gasoline type fuel in a 500 m diameter nozzle.
The primary objective of the paper is to present a
quantitative comparison of two CFD codes against the
experimental data. One code is CONVERGE and it is based on
the homogeneous flow assumption. The cavitation model is a
non-equilibrium thermal model (HRM) and it uses an empirical
time-scale correlation to allow for finite-rate evaporation
process. The second code is FIRE and it uses an Eulerian multi-
fluid description. The cavitation model is based on the R-P
equation, and bubble number and size are predicted by a poly-
dispersed model.
The second objective of the paper is to analyze numerically
the effect of the amount of non-condensable gases in the multi-
phase flow, by means of three-phase flow simulations. In
addition, some insight on the effect of turbulent pressure
fluctuations on the cavitation inception is presented.
The paper is organized as follows. First, a description of the
available experimental data is given. Second, the CFD models
concerning the mixture and the multi-fluid approaches are
described. Lastly, numerical results are assessed against
experimental data, and in addition some sensitivity and
parametric studies are discussed.
NOMENCLATURE '''A interfacial area density [m-1]
c speed of sound
CN cavitation number satoutoutin pppp
d orifice diameter
D bubble diameter
f
body forces per unit mass
h enthalpy
k turbulence kinetic energy
3 Copyright © 2013 by ASME
L orifice length
m mass
ijM
momentum interfacial exchange between phases i, j
n number of phases involved in the multi-fluid model '''N bubble number density [m-3]
p pressure
R bubble radius
Re Reynolds number
Rj source/sink of bubble number due coalescence/breakup
Rph source/sink of bubble number due to phase change
T temperature
v velocity
Y mass fraction
Greek symbols
volume fraction
ij mass interfacial exchange term between phases i, j
turbulence dissipation rate
time scale
kinematic viscosity
density
stress tensor
j source/sink of interfacial area due to coales./breakup
ph source/sink of interfacial area due to phase change
Subscripts
1 or l liquid phase
2 or v vapor phase
3 or a air phase
b bubble
crit critical (used for p or T of the fluid at the critical point)
D drag
eff effective
g gaseous (air + vapor)
in inlet
out outlet
ph phase change
sat saturation
TD turbulent dispersion
Abbreviations
AMR Adaptive Mesh Refinement
CFD Computational Fluid Dynamics
GDI Gasoline Direct Injection
HEM Homogeneous Equilibrium Model
HRM Homogeneous Relaxation Model
LES Large Eddy Simulation
RANS Reynolds Averaged Navier Stokes
R-P Rayleigh-Plesset
VOF Volume of Fluid
PROBLEM DESCRIPTION AND EXPERIMENTAL DATA All the numerical tests reported in this paper refer to nozzle
flow simulations with a gasoline surrogate. The purpose of this
section is to describe the available data and the conditions in
which they have been collected. Experiments have been carried
out recently at Argonne National Laboratory by Duke et al. [14]
using synchrotron x-ray radiography of a polycarbonate nozzle.
The model nozzle has sharp inlet edges and it is cylindrical with
a diameter (d) = 500 m and a length (L) = 2.5 mm. Geometry is
shown in Figure 1; test conditions and fluid properties are shown
in Table 1. As a result of the low outlet pressure of 0.87 bar (abs),
the cavitation number (CN) = 11.2 is very high, even with a
modest pressure drop of about 10 bar and a relatively low Re
number of 1.5·104. The gasoline surrogate used is called Viscor
which has been extensively used in the past in x-ray
measurements [14,17].
Figure 1. Nozzle geometry tested by Duke et al. [14].
Table 1. Experimental test conditions and fluid
properties [14].
Inlet p, T Outlet p,
T
Mass flow rate Re CN
10.6 bar
25 °C
0.87 bar
25 °C
5.82 g/s 1.58x104 11.2
Fluid type Density
@ 20 °C
Dynamic
Viscosity
@ 20 °C
Saturation
Pressure
@ 25 °C
Viscor 16br
(gasoline
surrogate)
781.8 kg/m3 9.35x10-4 Pa.s 640 Pa
Typical experimental images and data are reproduced in
Figure 2. Figure 2.a provides a high resolution (1.67 m)
qualitative image of cavitation at the entrance to the nozzle.
Figure 2.b provides quantitative void fraction measurements
over a line-of-sight collected by means of an x-ray microprobe
radiography technique. Probe resolution was 5x6 m and it was
moved over a raster grid with 100 transverse points and 19 axial
locations. Images have been then reconstructed by streamwise
interpolation. Both these measurements represent time-averaged
4 Copyright © 2013 by ASME
results. It is also important to mention that the experiments are
performed under steady flow conditions (there is no opening or
closing of needle), hence comparing time averaged experimental
data against steady state CFD results is justified.
These data show that cavitation is generated near the sharp
inlet edge. At halfway of the channel length the vapor layer
attached to the wall collapses, while a significant cloud of void
starts being visible in the core of the channel and then it extends
up to the channel outlet. Duke et al. [14] observe that the void
accumulation along the centerline is different from the canonical
distribution found in channels of similar size, such as those
investigated by Winklhofer et al. [35] using diesel fuel. The
presence of a void in the center, according to [14] is more likely
due to vapor detached from the walls and transported in the core
rather than due to a new nucleation process, since the pressure in
that region is not supposed to be low enough for cavitation
inception/triggering. At the same time it is also recognized that
dissolved gases in the fuel could also cause the centerline void.
The small amount of air eventually present is estimated to be of
the order of 10-3 mole fraction in the experiments but actual
measurements are not available.
Bauer et al. [15], in a cavitating pipe flow using water,
observed presence of void in the centerline in a similar manner.
They claim that isolated nucleation events occur due to the low
pressure in the core region. It should be noted from all the
experimental studies cited here that the techniques cannot
capture the differences between dissolved gases and fuel vapor.
The experiments can detect only a void cloud in different regions
of the channels.
Figure 2. X‐ray measurements of cavitation in a 500 m
polycarbonate nozzle, obtained via (a) radiographic imaging and
(b) raster‐scan microprobe radiography using a monochromatic
synchrotron source at 8‐10keV. Images reproduced with
permission from Duke et al. [14].
MIXTURE MODEL In this section, the Navier-Stokes equations for a
homogeneous multi-phase mixture will be introduced along with
a description of the HRM non-equilibrium cavitation model.
These models are implemented in the CFD code CONVERGE
[36,37,38].
Basic Equations
In a single-fluid approach, the homogeneous multi-phase
mixture model is governed by one set of conservation equations
for mass, momentum, and energy, with the addition of a
turbulence closure model for Reynolds Averaged Navier-Stokes
(RANS) equations. The homogeneous mixture model is based on
the assumption that the velocity, temperature, and pressure
between the phases are equal. Mass and momentum equations
are given below
0
v
t
(1)
fpvvt
v
(2)
where and v are the mixture density and velocity; p is the
pressure; is the mixture stress-strain tensor due to molecular
and turbulent viscosity. Analogous formalisms apply for energy
conservation and turbulence closure models [36,37,38].
Equations are omitted here for brevity.
In the present study, the multi-phase system is comprised of
a liquid phase (1), a vapor phase (2) and non-condensable gases
(3). The sum of vapor and non-condensable gases will be
referred to with the subscript g. The concept of pseudo-density
is used and the mixture density is computed with the following
equation:
lggg 1332211 (3)
The volume and mass fractions are related through
iii Y (4)
In the present implementation, the void fraction is not
transported directly, but the species are first solved using the
species transport equation and then the void fraction g is
calculated:
iiiii SYDvY
t
Y
(5)
ii
gg
gY
Y
(6)
The liquid phase is treated as incompressible and the gas phases
are treated as compressible.
Cavitation Model
In this code the mass exchange between the liquid and vapor
is based on a Homogenous Relaxation Model (HRM) as
developed by Bilicki and Kestin [32], and recently implemented
by Schmidt and co-workers [20,21,22,32], proposed for
describing non-equilibrium flash boiling processes. Both
cavitation and flash-boiling are evaporation phenomena driven
by a pressure drop. Cavitation is pressure-driven vaporization
(a) (b)
5 Copyright © 2013 by ASME
occurring at low temperatures in which the vapor density is so
small that the latent heat flow does not affect the phenomenon.
As a result, the time scale of heat transfer is much faster than the
time scale of bulk motion, therefore the latter is basically
controlled by the inertia of the liquid. On the contrary, a flash-
boiling process occurs at elevated temperatures and in this case
vapor density is much higher, therefore the liquid must provide
more energy per unit volume of vapor. The process takes a non-
negligible time scale and the dynamics is controlled by finite-
rate heat transfer rather than by inertia [39]. In this framework
the HRM approach assumes a first order rate equation for the
evolution of the instantaneous mass fraction of vapor Yv towards
its equilibrium value vY over a given time-scale . The model
as proposed by Bilicki and Kestin [32] is given below
vvv YY
dt
dY (7)
The equilibrium vapor quality vY is a function of the
thermodynamic properties at the local pressure, i.e.
lvlv hhhhY . The time-scale is evaluated using an
empirical fit proposed by Downar- Zapolski [40] 76.154.071084.3 [s] (8)
where
satcrit
sat
pp
pp
(9)
and it was determined in experiments of flashing flows of water
in pipes with upstream pressures of in excess of 10 bar.
EULERIAN MULTI-FLUID MODEL In this section, the governing equations for the multi-phase
flow modeled using a non-homogeneous multi-fluid approach
will be introduced along with a description of the Rayleigh-
Plesset cavitation model. These models are implemented in the
CFD code FIRE.
Basic Equations
The Eulerian multi-fluid approach is based on the
assumption of co-existence of n different phases, treated as
interpenetrating continua. In the present study, for the purpose of
cavitation modeling, the multi-phase system is comprised of a
liquid phase (1), a vapor phase (2) and incondensable gases (3).
Each phase is considered as a continuous medium and an
ensemble averaging procedure is applied to remove the
microscopic interfaces, at sub-grid scale level [44,45]. For each
phase a set of conservation equations is applied which is
analogous to the RANS equations, with the addition of a scalar
quantity, i.e. the volume fraction i of the phase i. Furthermore,
relevant source/sink terms for inter-phase exchange of the
conserved quantities are introduced [46,47]. For n = 1, equations
actually reduce to the RANS equations for a single phase flow.
Mass and momentum averaged equations for the multi-fluid
model are given below, according to [45, 49]
n
ijj ijiiiii v
t ,1
(10)
n
ijj iji
n
ijj ijiiiiii
iiiiiii
vMfp
vvt
v
,1,1
(11)
where , and v are the averaged volume fraction, density, and
velocity for each phase; p is the pressure and it is shared by all
phases, i.e. p = pi; i is the phase stress-strain tensor due to
molecular and turbulent viscosity; subscripts i and j are the phase
indicators; Γij is the mass change rate due to cavitation and ijM
is the inter-phase momentum transfer term. Analogous
formalisms apply for energy conservation and turbulence closure
models [45,49]. Equations are here omitted for brevity.
Each fluid can be treated as incompressible or compressible.
In the latter case, for a gas phase the density is determined based
on the equation of state for an ideal gas:
iii TRp (12)
For the liquid phase the density is determined using the
barotropic relation:
2
, irefrefii cpp (13)
where ci is the liquid speed of sound.
Cavitation Model: Mass Transfer between Liquid and Vapor
In order to describe the mass transfer due to cavitation, a
vaporization/condensation rate model is required. In this code
the Rayleigh equation for the dynamics of a spherical bubble
growth constitutes the starting point [41]
RR
ppR
l
effsat 3
2 (14)
Plesset [39,42] also added two terms accounting for surface
tension and liquid viscosity effects. Although these terms could
be easily incorporated into the model their contribution is
marginal, therefore they have been neglected in this study.
The concept of effective pressure peff has been introduced to
reflect the fact that the dynamics of a single bubble can be
affected by other flow parameters as well. In particular, due to
the local liquid turbulence the pressure around a bubble is
subjected to fluctuations, which leads to the instantaneous
pressure experienced by the bubble becoming lower than the
mean pressure, at least for certain time intervals. In view of this
the following expression is used:
llEeff kCpp 3
2 (15)
where CE is the Egler coefficient with a value of 1.2 as suggested
by Hinze [43] and Giannadakis et al. [27], and kl is the local
turbulent kinetic energy in the surrounding liquid. In this model
therefore the effective surrounding pressure peff is assumed lower
than the RANS averaged pressure p. This is a way for
representing the likelihood of cavitation inception at average
pressure levels higher than the saturation pressure because of
turbulent pressure fluctuations. At the same time it is implicitly
assumed that bubble dynamics and turbulent eddies dynamics
are different, so that once bubble growth is started, a rapid re-
6 Copyright © 2013 by ASME
increase of the pressure will not necessarily cause a
corresponding quick collapse.
The mass change rate of a single vapor bubble due to
isothermal evaporation or condensation is
RRtm vb24 (16)
In general, assuming a poly-dispersed flow, a probability
distribution function for bubble radius Rf can be defined.
Therefore, the integral mass exchange of the population can be
calculated. Statistically, two average quantities are required to
close the problem, e.g. the bubble number density '''N and the
interfacial area density '''A . The related transport equations are
given below
phj RRvN
t
N )0('''''' (17)
phjvA
t
A )2('''''' (18)
where )0(v
and )2(v
are the 0th and 2nd moment average veloci-
ties, respectively, of the bubble size distribution, defined as:
max
min
max
min
)(R
R
kR
R
kk dRRfRdRRfRvv . (19)
The terms on the right hand side of the equations (17) and (18),
i.e. jR and
j , represents the net rate of change in the
corresponding transported quantities due to fluid particle
interactions such as coalescence and disintegration, while phR
and ph are the fluid particle source/sink rates due to phase
change. Closure models for these terms along with further
detailed descriptions of the model are provided in [47,48]. By
means of '''N and '''A the mass exchange term 12 in eq. (10) and
(11) can be calculated.
Momentum Exchanges
The momentum exchange term 12M
takes into account at the
grid scale level the microscopic effects exerted at the interface
between liquid and vapor.
2112121
'''
21128
1 kCvvvvACM TDD
(20)
Drag and turbulent dispersion effects are considered, whereas
inertia and lift effects are neglected. The drag coefficient CD is
based on the drag law of a single sphere. The turbulent dispersion
force, which accounts for the bubble dispersion due to the
turbulent mixing process, has been modeled setting CTD = 0.1. In
addition, since a bubbly flow induces turbulence at a
microscopic level and basically this is caused by momentum
interactions at the interface, a bubble induced turbulent viscosity
term is added to the liquid phase turbulent viscosity term,
according to Sato [50].
221
1
2
1,1 2
vvRC
kC Satot
(21)
The value here used is SatoC = 0.6.
The momentum exchange between liquid and air 13M
was
accounted for using formally a model identical to that already
described in eqns. (20) and (21) for the vapor. Air was considered
the dispersed phase and liquid the continuous phase. The only
difference is the assumption of a constant diameter for air
bubbles, set to 1 m. Therefore in eq. (20) the interfacial area
density simplifies to
33
3/2
3
'''3/1'''2
3
'''
3 636 DNNDA (22)
The coefficients for turbulence dispersion CTD and for bubble-
induced turbulent viscosity CSato associated with air have been
set to the same values adopted for vapor. If locally 3 becomes
greater than 0.5, the continuous and dispersed phases are
switched. In addition, no momentum exchange between air and
vapor was assumed, i.e. we set 023 M
.
NUMERICAL SETUP Acknowledging that the two codes are based on different
models, it is understandable that the implementations are also
different. In this study, extensive effort has been made in order
to present a fair comparison, hence, same settings have been
applied wherever possible. Most of the numerical parameters are
summarized in the following Table 2 and are discussed in the
following paragraphs. Some intrinsic differences exist, which in
essence characterize each code and each approach, and they will
be discussed below as well.
Table 2. Baseline numerical set-up.
CONVERGE FIRE
Fluid model Mixture model Multi-fluid model
Cavitation model HRM Rayleigh
Effective pressure
which triggers
cavitation
ppeff llEeff kCpp 3
2
setting CE = 0
Species / phases liquid, vapor, air liquid, vapor, air
Dissolved gases at
the inlet boundary
2·10-5 by mass 2·10-5 by mass
Compressibility compressible compressible
Turbulence model k- k-
Grid type hexahedral, with
fixed embedding
mainly hexahedral
Base grid size [m] 150 150
Min grid size [m] 9 9
Geometry 90 deg sector with
periodic boundar.
90 deg sector with
periodic boundar.
Pressure-Velocity
coupling
PISO SIMPLEC
Time integration Euler 1st order Euler 1st order
Spatial discretiz. 2nd order 2nd order
Time step [s] lower than 5·10-9 lower than 5·10-8
The main difference as outlined in the previous section lies
in the homogeneous vs. non-homogeneous approach used for
representing the multi-phase system. CONVERGE uses a
mixture model which is intrinsically homogeneous, for example
velocity is common to all the phases, therefore no slip exists.
7 Copyright © 2013 by ASME
This is reasonably true in a cell where the volume fraction of one
phase largely dominates over all the others; in this case the latter
can be thought as transported without velocity slip.
Unfortunately, the approach is questionable when comparable
amounts of different fluids are present in the same cell. FIRE
instead solves different sets of equations for each fluid, therefore
slip phenomena can be taken into account.
The numerical solution is based on a finite volume
discretization of the governing equations, as available in both the
commercial codes. Simulations are transient, so all calculations
were run until both the inflow and outflow had stabilized. This
usually required at least 1.0 ms of physical time, but in some
cases simulations were run beyond 1.5 ms. The first-order
implicit Euler scheme has been employed for the time
integration. For the baseline case comparison, shown in Table 2,
calculations were run taking into account gas-phase
compressibility, therefore time step size was extremely small for
stability reasons. FIRE required a maximum time step of the
order of 5·10-8 s and CONVERGE an even smaller time-step of
approximately 5·10-9 s.
Even though large gradients exist in nozzle flows and
stability has been a main concern, in the present study convection
terms have been discretized using second-order spatial schemes
for all quantities. To this end, calculations have been started with
first order upwind schemes when needed and after reaching a
steady state solution they have been extended with second order
schemes. The overall iterative solution procedure has been based
on the PISO algorithm for CONVERGE and on the SIMPLEC
algorithm for FIRE. Turbulence has been accounted for using the
standard k- model with standard wall treatment in both codes.
Grids are slightly different in terms of cell type and hence in
terms of total number of cells. This is a consequence of different
meshing approaches used by the two codes. CONVERGE uses a
modified cut-cell Cartesian method for grid generation and the
grid is generated internally at runtime. Both fixed cell
embedding and AMR cell embedding can be employed.
Nevertheless, in this study only fixed embedding has been used
for fairer comparison of the results. FIRE can handle structured
and unstructured grids and it uses polyhedral cells in the
transition zone between two neighboring refinement levels.
Examples of the meshes are given in Figure 3. In both cases the
base grid size is 150 m and 4 levels of refinements are used to
reach the minimum value of approximately 9 m. The whole
channel region has been uniformly discretized using the smallest
cells, while coarsening has been applied only outside, in the inlet
and outlet chambers. Lastly, only one quarter of the channel has
been modeled and periodic boundaries have been applied on the
side surfaces.
A noteworthy aspect of this work is that the effects of
dissolved gases in the multi-phase calculations were accounted
for in the simulations. As outlined in a previous section, the main
reason is that the experimental tests present low outlet pressures
therefore expansion of small fractions of air inside the channel
could be emphasized at these low pressures. In both codes the
amount of air introduced as a third species in the fluid was set to
2.10-5 mass fraction. This value corresponds approximately to the
standard value for gasoline (or water) exposed to the ambient
pressure [51]. It is worth noting that in CONVERGE this input
is straightforward, since it solves for the species mass transport,
while in FIRE it has been converted to the corresponding amount
of volume fraction at the inlet pressure (approximately 1.31.10-3
by vol. at 10.6 bar).
Figure 3. Examples of grids generated in CONVERGE (a)
and FIRE (b) with 150 m base grid size and 9 m min. cell size.
The cavitation models are also different, as described in the
previous section. CONVERGE uses the HRM model, while
FIRE uses the Rayleigh equation based model. Both codes
assume that cavitation occurs if the fluid pressure drops below
the saturation pressure of the fluid at the local temperature. Since
the flow is turbulent and pressure fluctuations can exists, the
problem of what would be the effective triggering pressure
arises, especially in a RANS approach. In this context, FIRE
allows to include this contribution by means of the Egler
coefficient CE shown in eq. (15). Nevertheless, the value chosen
is arbitrary since the only theoretical basis [43] is valid for
isotropic turbulent flows. Hence the value can be changed to
enhance or reduce cavitation by influencing the local pressure
distribution. Furthermore, the present implementation of
CONVERGE neglects this aspect; therefore in this study for the
baseline set-up, in order to present a fair comparison, the value
of CE in FIRE has been set to 0, as shown in Table 2.
RESULTS AND DISCUSSION In this section CONVERGE and FIRE results concerning
the cavitating nozzle under the conditions described in Table 1
will be compared against x-ray data [14].
Model assessment
Figure 4 shows the results of the simulations for the
cavitating nozzle compared to the experimental data from Duke
et al. [14]. Simulation settings and boundary conditions for both
codes are reported in Table 2 and Table 1, respectively. Results
shown in this sub-section refers to the minimum grid size of 9
m for both codes. CFD data have been post-processed to
evaluate the projected amount of void, computed as the line
integral of g along the transverse direction, throughout all the
channel width. The spatial resolution of this evaluation
corresponds to the cell size and shape existing on the mid cut-
(a)
(b)
8 Copyright © 2013 by ASME
plane, i.e., the integration is performed starting from each cell
belonging to this plane and moving normally to the plane itself
in both directions. This type of result will also be referred to as
radiographic image.
In terms of global mass flow rate the experimental value is
5.82 g/s, and the computed values are 5.25 g/s (-10%) and 5.36
g/s (-8%) with CONVERGE and FIRE respectively.
Figure 4. Predicted void fraction integrated along the
transverse direction using CONVERGE and FIRE codes,
compared to experimental data [14]. Dimensions are in mm.
Flow is from bottom to top.
Observing radiographic void contours in Figure 4, a
reasonable qualitative agreement can be seen for both codes.
Each model is capable of reproducing the presence of cavitation
just after the inlet corners. In this region the predicted integrated
void along the line-of-sight has a peak which occurs slightly off
from the wall, and 0.25 mm downstream of the inlet edge. The
CONVERGE mixture model predicts a peak value around
220 m. The FIRE multi-fluid model, instead, predicts a local
peak value of 150 m. In this region, measurements show two
different local maxima (they are not symmetrical due to
machining imperfections of the nozzle), whose average can be
estimated close to 150 m. In view of that, the mixture model
overestimates the local void peak, while the multi-fluid model is
quite close to the experimental data.
As far as the void cloud in the middle of the channel is
concerned, both models capture this effect and this is
noteworthy, since in the literature it is fairly common to find
simulation results with cavitation clouds developing just along
the wall, without penetrating significantly into the core
[20,26,31,33]. Anyway, for both codes the predicted void cloud
is a bit shorter, and it does not extend up to the outlet section as
in the experiment. FIRE predicts a shorter central cloud
compared to the experimental data. CONVERGE predicts
slightly better the axial location of this central void, but the local
maximum still occurs early. Another interesting point is that in
the experiments the centerline void is rather disconnected from
the inlet edge void. This aspect is better captured by the multi-
fluid model.
Lastly, FIRE is able to capture the pressure recovery and
hence the reattachment of the liquid phase to the wall which
occurs approximately at one third of the channel length.
CONVERGE is not able to capture this effect and it predicts
presence of void along the wall all the way up to the exit.
Figure 5 shows a second quantitative comparison in terms
of total void fraction along the channel axis in ten specific axial
locations as available from the experiments [14]. Numerical
values are calculated as the volume integral of the void fraction,
normalized to the volume itself, over small channel slices with a
thickness of one cell layer. Experimental data are in fact
evaluated on 6 m width cross cuts. Right after the inlet edge (at
x/L = 0.1-0.2) we can observe the maximum value of void
formation. Thereafter, both codes predict a fairly constant
amount of total void up to x/L = 0.6. Then a certain decrease as
a function of the axial position is predicted, while the
experimental value is still constant around 21%. Results obtained
with CONVERGE over-predict by 30% the experimental data in
the first half of the channel, while downstream results tend to
approach the measured values. Void fractions computed by FIRE
agree very well for the most part of the channel, and only at the
exit the value is underestimated. It should be noted that the
experimental uncertainties associated with such measurements
are not available.
Figure 5. Computed total volume fraction g (vapor + air)
along the channel axis evaluated in one cell layer thick slices,
compared to the experimental data [14].
It is worth noting that results shown so far (Figure 4 and
Figure 5) refer to the total void g as the sum of vapor and air,
without addressing the separate contribution of cavitation and
expansion of dissolved gases. While it is understood and
accepted (and it will be shown in the next sub-section) that the
void after the inlet edge is generated by cavitation due to the
extremely low pressures, the most difficult region to be predicted
is the second part of the channel, along the axis. In this region
either new cavitation or vapor transport or air expansion or a
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
tota
l vo
id f
ract
ion
g
x/L
Experimental dataCONVERGEFIRE
9 Copyright © 2013 by ASME
combination of all these effects can play a significant role in
contributing to the void. A specific sub-section, following a grid
sensitivity study, will address this issue in more detail.
Grid sensitivity
Since both codes have similar mesh embedding capabilities,
the cell sizes shown are the minimum grid sizes. For both codes
two sizes have been tested, i.e., 18 m and 9 m. Results in terms
of mass flow rates are given in Table 3. As mentioned previously,
predicted values are underestimated within a range of 7% to
11%, but the effect of changing the grid size is limited to 1.5%
for each code, therefore the calculations can be considered grid-
converged with respect to this parameter. Future studies will
consider finer cell sizes to comprehensively address the issue of
grid-convergence.
Figure 6 shows the overall void fraction predicted along the
channel axis. This quantity is affected by higher dispersion, but
taking into account inherent modeling difficulties, the results are
still encouraging. A good trend is shown by both codes, since
refining the mesh numerical results tend to get closer to the
experiments. In fact, it can be noted that mesh refinement has
opposite effects on the two codes, i.e., with CONVERGE mesh
refinement results in increase of the total void fraction while in
FIRE the vice versa happens.
Table 3. Cell size effect on mass flow rate.
Min. cell size[m] Mass flow rate [g/s]
CONVERGE FIRE
18 5.17 5.46
9 5.25 5.36
Experimental value 5.82
Figure 6. Cell size effects on total void fraction g along the
channel. Experimental data [14] are also plotted for comparison
purposes.
Further Discussion
In order to better understand the underlying physics and to
further assess the modeling capabilities and limitations, the
computational results can be used to analyze other features of
interest, such as velocity, pressure, and split of vapor fraction and
air fraction. Figure 7 plots all the aforementioned quantities
along a cut-plane through the nozzle axis; the two columns show
the comparison between the two codes.
The total void g on the cut-plane is shown in Figure 7.c.
Despite the fact that these patterns are quantitatively different
from the radiographic images described in Figure 4, they
substantially present common features, therefore this allow us to
analyze the void formation process also on this basis.
The pressure field is mainly responsible for the void
production and can be seen in Figure 7.b (please note that the
scale is narrowed to 0-1 bar) and pressure profiles along the
channel wall and axis can be seen in Figure 8 (scale is zoomed
as well).
The flow separates off of the inlet edge and an extremely
low pressure region is generated just behind it. In this region both
codes predict local values well below the saturation pressure
(Figure 8.b) where local values close to the wall are of the order
of a few tens of Pa. As a result vapor is produced and expands
along the wall (Figure 7.c,d,f). It is worth to point out that
because of the compressible model, there is no unphysical
undershooting of the pressure below 0 bar, as would occur using
an incompressible approach. In the region downstream of the
inlet edge, there is significant amount of cavitation predicted by
both codes. Here, vapor mass fraction is distributed along the
channel wall and sticks to it without penetrating significantly
towards the axis (Figure 7.d,f). The total void g reaches 100%
locally and the cavitation contribution to the total void is very
important. However, it is worth noting that the predictions about
the amount of vapor mass generated after the inlet corner (cf.
Figure 7.f,g) are very different between the two codes. The Y2
peak value predicted by CONVERGE is around one tenth of the
air mass fraction Y3, while the corresponding value predicted by
FIRE is close to ten times the air mass fraction. Such relatively
big differences are due to the fact that FIRE using a multi-fluid
model, predicts separation of air from vapor in this region as a
result of momentum exchange (cf. Figure 7.g at the inlet edge),
while CONVERGE using a homogeneous model, keeps the air
mass fraction constant everywhere.
Along the wall at one third of the nozzle length, FIRE
predicts a pressure recovery (marked by the arrow in Figure 8.a
at x/L = 0.3) which causes reattachment of the liquid (Figure
7.a,c). This feature has been already pointed out in Figure 4 and
matches with experiments. CONVERGE does not capture this
effect.
Along the nozzle axis pressure rapidly decreases and
reaches a new local minimum at x/L = 0.8-0.9 (see second arrow
in Figure 8.b). In this region the predicted pressure is not low
enough to generate additional vapor, but it is low enough to allow
a significant air expansion (Figure 7.c,d,e). Here the local void
fraction g reaches 50% and is entirely due to air expansion.
Further downstream, a steep pressure rise occurs (the flow goes
through local shocks) and thereafter pressure oscillates before
finally recovering the outlet chamber level. Both codes predict a
similar behavior, the only difference lies in the location of the
pressure rise.
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
tota
l vo
id f
ract
ion
g
x/L
Experimental dataCONVERGE - 18 micronsCONVERGE - 9 micronsFIRE - 18 micronsFIRE - 9 microns
10 Copyright © 2013 by ASME
Figure 7. Predicted velocity (a), pressure (b), volume fractions (c,d,e) and mass fractions (f,g). Left column shows CONVERGE
results, right column shows FIRE results. Color scale range is shown above each row.
0.0 0.2 0.4 0.6 0.8 1.0-20
0
20
40
60
80
100
120
0.0 0.2 0.4 0.6 0.8 1.0-20
0
20
40
60
80
100
120
Pre
ssure
[kP
a]
x/L
nozzle axis - FIRE
nozzle axis - CONVERGE
Pre
ssure
[kP
a]
x/L
nozzle wall - FIRE
nozzle wall - CONVERGE
Figure 8. Pressure profiles along the nozzle wall (a) and
along the axis (b).
As far as the comparison between the two codes is
concerned, velocity and total void fractions are similarly
predicted. Two main differences can be observed. The first one
lies on the prediction of the amount of vapor generated at the
inlet edge, since FIRE predicts more vapor production than
CONVERGE. The second one lies on the reattachment of the
liquid predicted by FIRE at x/L = 0.3, which CONVERGE is not
able to reproduce.
From this discussion it is evident that although the total
amount of void matches the experimental data quite well (Figure
4 and Figure 5), the relative amount of vapor and air is still not
fully assessed. This is mainly due to the fact that the experiments
cannot distinguish between vapor and dissolved gases. A
stronger validation would be possible if pressure data was
available, since the vapor formation and the air expansion in
simulations are extremely sensitive to these low pressure levels.
In addition, the exact knowledge of the dissolved air fraction
would be essential. With respect to the latter, the following
section will give more details.
(a)
(b)
(c)
(d)
(e)
(f)
(g)
(a)
(b)
Mixture Model
(CONVERGE)
Multi-Fluid Model
(FIRE)
11 Copyright © 2013 by ASME
PARAMETRIC STUDY Following the validation of the two codes, some parametric
studies have been carried out using parameters of interest as
outlined below in Table 4. First, a study about changing the inlet
and outlet pressure, at equal cavitation number CN is shown.
Afterwards, the influence of the amount of dissolved gases on
cavitation is presented, since some uncertainties on the
knowledge of this quantity exists. Both these assessments are
performed with CONVERGE code. Lastly cavitation-turbulence
interaction is discussed with FIRE code. All the other parameters
not mentioned in Table 4 have been kept unchanged, as already
shown in Table 2.
Table 4. Parametric studies with CONVERGE and FIRE
codes.
CONVERGE FIRE
Inlet – outlet pressure [bar]
106.0 – 8.7,
10.6 – 0.87
(same CN=11.2)
Amount of dissolved air,
mass fractions [-]
0, 2E-7, 2E-6,
2E-5, 2E-4
Turbulent pressure
fluctuations on effective
pressure llEeff kCpp
3
2
CE = 0,
CE = 1.2
Influence of outlet pressure on cavitation
In order to quantify the importance of using a three-phase
model which includes dissolved air in the fuel, instead of a
simpler two-phase approach, a simple test is presented. The
baseline case (pin = 10.6 bar and pout = 0.87 bar) already
discussed in the previous sections, is compared to a higher
injection and back-pressure case, i.e., pin = 106 bar and pout = 8.7
bar. This allows a comparison at the same CN number.
The comparison is shown in Figure 9 and Figure 10.
Concerning the void production at the inlet corner no significant
differences can be noticed, since the local pressure is always
extremely low and sufficient to trigger cavitation (Figure 9.a).
On the contrary, along the channel axis the pressure minimum,
occurring at approximately x/L = 0.8, presents very different
values (Figure 9.b). Pressure is above 5 kPa for the high outlet
pressure case, while it is as low as 1 kPa for the low outlet
pressure case. This means that in this region, the appearance of
void can be substantially offset by high back-pressure
conditions, regardless of the amount of dissolved gases or the
value of the saturation pressure of the fuel (at least for the typical
values encountered).
The difference visible in Figure 9.a up to x/L = 0.6 occurs
because the 8.7 bar case has higher velocities that bring the
pressure down to very low levels for a greater length.
0.0 0.2 0.4 0.6 0.8 1.00
10
20
30
40
50
0.0 0.2 0.4 0.6 0.8 1.00
10
20
30
40
50
Pre
ssure
[kP
a]
x/L
Axis profiles:
outlet pressure = 8.7 bar
outlet pressure = 0.87 bar
Pre
ssure
[kP
a]
x/L
Wall profiles:
outlet pressure = 8.7 bar
outlet pressure = 0.87 bar
Figure 9. Outlet pressure effects under equal CN value.
Pressure profiles along the nozzle wall (a) and along the axis (b)
are shown using CONVERGE code.
Figure 10. Outlet pressure effects under equal CN value.
Predicted total volume fractions g using CONVERGE code.
Dissolved air set to the baseline value of 2.10-5 mass fraction
(Table 2).
Dissolved air effect
The influence of the amount of non-condensable gases on
void formation is discussed here. The experimental uncertainty
concerning the exact knowledge of this data motivated us to
quantify the effects of different concentrations. All the
parameters have been kept unchanged as shown in Table 4, apart
from the air fraction Y3 which has been varied from 0 to 2·10-4
by mass.
Results are shown in Figure 11 and Figure 12. The former
plots the value of g integrated and normalized over channel
slices at different axial locations, the latter shows the contours of
vapor mass fraction Y2 and of total void fraction g. From Figure
11 it can be inferred that the volume fraction of vapor and non-
condensable gases remains fairly constant as the amount of air is
increased, except for the highest level of Y3 = 2·10-4. This
indicates that adding air to the liquid inhibits the inception of
cavitation at very low pressures. This occurs because while the
pressure decreases, the expansion of non-condensable gases
opposes a further pressure drop and in turn hinders cavitation by
keeping the local pressure at high values.
(b)
(a)
12 Copyright © 2013 by ASME
Figure 11. Effect of dissolved air mass fraction Y3 on total
void fraction g predicted along the channel using CONVERGE
code.
At extreme conditions, for Y3 = 0 the vapor and void
formation is confined along the wall; while for Y3 = 2·10-4 there
is a practical absence of cavitation but a huge production of void
at the wall and in the core due to the air expansion. It is also
important to point out that only introducing the highest level of
non-condensable gases (Y3 = 2·10-4) the total void produced
starts to get increased, with respect to all the other cases (Figure
11), as it would be intuitively expected.
The fact that the total void evaluated as integral at different
axial location (cf. Figure 11) is quite independent of the amount
of non-condensable gases suggests that further data are needed
to fully validate the models and to accurately predict the physics.
Both a direct measure of the dissolved air and the measurement
of the pressure at some locations could help in complete
understanding.
Figure 12. Effect of the dissolved air mass fraction Y3 on the vapor production and on the total void fraction patterns using
CONVERGE code. Left column shows vapor mass fraction Y2, right column shows total void fraction g. Color scale range is shown
above each column.
Turbulence effect
In this last sub-section the influence of the turbulent pressure
fluctuations accounted for by means of the Egler factor CE, are
presented using FIRE code. In all the previous results, this
parameter has been set to 0, therefore possible turbulent effects
on the cavitation inception have been ignored.
Locally, the effective liquid pressure peff that is responsible
for cavitation inception (cf. eqns. [12] and [13]) could be lower
than the average value p, because of turbulent fluctuations. To
the best of our knowledge many Authors [32,25,27] recommend
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
tota
l vo
id f
ract
ion
g
x/L
Y3 = 0
Y3 = 2E-07
Y3 = 2E-06
Y3 = 2E-05
Y3 = 2E-04
13 Copyright © 2013 by ASME
the use of CE = 1.2, but no evident proof of that assumption has
been given.
A comparison between assuming CE = 0 and CE = 1.2 is
shown in Figure 13 and Figure 14. The former plots the value of
g integrated and normalized over channel slices at different
axial locations, the latter shows the contours of various fluid
dynamic variables. Clearly, the value of g is increased when
turbulence is included in the cavitation model, as visible in
Figure 14, but the overall value is still comparable to the
experimental data especially near the exit of the channel.
Observing Figure 14.d,f, when a value of CE = 1.2 is used, a
larger cavitation cloud is generated downstream of the inlet edge
and in addition cavitation is predicted in the middle of the
channel too. The corresponding pressure levels (Figure 14.b) are
slightly increased as a countereffect of more momentum
exchange between the vapor and the liquid phase. These results
suggest that the phenomena are extremely complex and further
validation is required to shed light on the physics. It would also
be advisable to model the effect of turbulence in a more
straightforward manner, i.e., by means of large eddy simulations
(LES) to remove great part of the uncertainties associated to this
aspect.
Figure 13. Effect of turbulent pressure fluctuations on total
void fraction g predicted along the channel using FIRE code.
Figure 14. Effect of turbulent pressure fluctuations on cavitation development using FIRE code.
0
0.05
0.1
0.15
0.2
0.25
0.3
0.35
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
tota
l vo
id f
ract
ion
g
x/L
Experimental data
CE=0
CE=1.2
(a)
(b)
(c)
(d)
(e)
(f)
(g)
14 Copyright © 2013 by ASME
CONCLUSION In this paper two cavitation models have been studied. We
compared a mixture model in conjunction with the HRM phase
change model, as implemented in CONVERGE, with a
multi-fluid model with Rayleigh bubble dynamics for phase
change, as implemented in FIRE. The numerical results have
been assessed against experimental data collected at Argonne
National Laboratory using x-ray techniques. The study focused
on a cavitating nozzle tested with gasoline fuel at atmospheric
outlet pressure. Numerical results were compared to
measurements of mass flow rate and high-resolution void
fraction pattern, evaluated as radiographic projections.
From an engineering point of view, the two models showed
good predictive capabilities. In more detail, on equal grid size,
mass flow rates are underestimated by 8% using the multi-fluid
model, and by 10% using the mixture model. The multi-fluid
model is able to match reasonably well the total void fraction as
a function of the axial position. The mixture model over-predicts
this quantity by up to 30% for large part of the channel. The
multi-fluid model is also able to predict the liquid stream
reattachment to the wall occurring approximately at one third of
the channel length, while the mixture model does not capture this
effect. The axial location of the centerline void is slightly better
predicted by the mixture model instead, but the multi-fluid
model captures the disconnection between the void region at the
inlet edge and the void in the centerline, as it is observed in the
experiments.
The simulations have been carried out using a compressible
three-phase flow approach, in both codes, and this aspect was
shown to be essential for capturing the formation of the void
cloud in central region of the channel. Grid convergence has
been checked, and with respect to the total void fraction some
work is still needed. Future studies will consider finer cell sizes
to comprehensively address this issue. Parametric study on the
amount of non-condensable gases revealed that this quantity has
a strong effect on the amount of cavitation produced. The higher
the amount of air, the lower the vapor generation. However,
surprisingly, the total amount of the two gaseous phases is quite
independent of the amount of non-condensable gases. Therefore,
further data is needed to fully validate the models and split the
two contributions. If the operating pressures are increased, the
void cloud in the channel core is reduced, while the cavitation at
the inlet edge remains largely unaffected. Future work will
address further validations including the comparison with
experimental pressure measurements inside the channel. Also
LES studies will be performed to better address the effect of
turbulence on cavitation inception.
ACKNOWLEDGMENTS The submitted manuscript has been created by UChicago
Argonne, LLC, Operator of Argonne National Laboratory
(“Argonne”). Argonne, a U.S. Department of Energy Office of
Science laboratory, is operated under Contract No. DE-AC02-
06CH11357. The U.S. Government retains for itself, and others
acting on its behalf, a paid-up nonexclusive, irrevocable
worldwide license in said article to reproduce, prepare derivative
works, distribute copies to the public, and perform publicly and
display publicly, by or on behalf of the Government. This
research was funded by DOE’s Office of Vehicle Technologies,
Office of Energy Efficiency and Renewable Energy under
Contract No. DE-AC02-06CH11357. The authors wish to thank
Gurpreet Singh, program manager at DOE, for his support.
We gratefully acknowledge the computing resources
provided on "Fusion," a 320-node computing cluster operated by
the Laboratory Computing Resource Center at Argonne National
Laboratory.
The authors would also like to acknowledge Dr. Eric
Pomraning and Dr. Hongwu Zhao at Convergent Science Inc. for
helping with the set-up in CONVERGE software.
We also acknowledge AVL-Graz for providing support with
the FIRE code during the research activity.
Lastly, the authors would also like to acknowledge Dr.
Daniel Duke and Dr. Chris Powell at Argonne National
Laboratory for sharing the experimental data and many helpful
discussions.
REFERENCES 1. Andriotis, A., Gavaises, M. and Arcoumanis, C., J. Fluid
Mech. (2008), vol. 610, pp. 195–215.
2. Payri, F., Bermudez, V., Payri, R. and Salvador, F.J., Fuel 83
(2004) 419–431.
3. Som S., Aggarwal S. K., El-Hannouny, E. M., Longman D.
E., 2010, J. Eng. Gas Turbines Power 132(4), 042802
(2010).
4. Som S., Ramirez A. I., Longman D. E., Aggarwal S. K.,
(2011), Fuel, 90, 1267–1276.
5. Battistoni, M., and Grimaldi, C. N., Applied Energy, (2012)
Vol. 97, 656-666.
6. Postrioti, L., Mariani, F., Battistoni, M., Fuel, (2012) Vol.
98, 149-163.
7. Befrui, B., Corbinelli, G., Hoffmann, G., Andrews, R. J. and
Sankhalpara, S. R., SAE Technical Paper 2009-01-1483,
(2009).
8. Battistoni, M., and Grimaldi, C. N., SAE Technical Paper
No. 2010-01-2245 (2010), Int. Journal of Fuels and
Lubricants, vol. 3, issue 2, pp. 879-900.
9. Battistoni, M., Grimaldi, C. N., and Mariani, F., SAE
Technical Paper No. 2012-01-1267 (2012).
10. Arcoumanis, C., Gavaises, M., Flora, H., Roth, H., Mec.
Ind., 2, pp.375-381, (2001).
11. Arcoumanis C, Badami M, Flora H and Gavaises M,
Transactions Journal of Engines, SAE Paper 2000-01-1249,
Vol. 109-3, (2000).
12. Hayashi, T., Suzuki, M. and Ikemoto, M., ICLASS 2012,
Contribution No. 1375.
13. Payri, R., Salvador, F.J., Gimeno, J., Venegas, O.,
Experimental Thermal and Fluid Science 44 (2013) 235–
244.
14. Duke, D., Kastengren, A., Tilocco, F. Z., Powell, C., 25th
Annual Conference on Liquid Atomization and Spray
Systems, ILASS-Americas, Pittsburgh, Paper No. 8, (2013).
15 Copyright © 2013 by ASME
15. Bauer, D., Chaves, H. and Arcoumanis, C., Meas. Sci.
Technol. 23 (2012).
16. Kastengren, A.L., Tilocco, F.Z., Powell, C.F., Manin, J.,
Pickett, L.M., Payri, R., and Bazyn, T., Accepted in
Atomization and Sprays, (2013).
17. Kastengren, A., Powell, C.F., Liu, Z., Fezzaa, K., and Wang,
J., (2009), Proceedings of the ASME Internal Combustion
Engine Division Spring Technical Conference, Paper
ICES2009-76032.
18. Schmidt, D. P. and Corradini, M. L. (2001), Intl J. Engine
Res. 2, 1–22.
19. Giannadakis, E., (2005), Ph.D. thesis, Imperial College,
London.
20. Rakshit, S., and Schmidt, D.P., ILASS Americas, 24th
Annual Conf. on Liquid Atomization and Spray Systems,
San Antonio, TX, 2012, paper no. 62.
21. Neroorkar K.D., Mitcham, C.E., Plazas, A.H., Grover, R.O.,
and Schmidt D.P., ILASS Americas, 24th Annual Conf. on
Liquid Atomization and Spray Systems, San Antonio, TX,
2012, paper no. 88.
22. Shields B., Neroorkar K., and Schmidt D.P, ILASS
Americas, 23rd Annual Conf. on Liquid Atomization and
Spray Systems, Ventura, CA, 2011, paper no. 110.
23. Marcer, R., Le Cottier, P., Chaves, H., Argueyrolles, B.,
Habchi, C., Barbeau, H., SAE Technical Paper 2000-01-
2932, (2000).
24. Alajbegovic, A. (1999). In Proc. Second Annual Meeting
Inst. Multifluid Sci. Technol., Santa Barbara, CA pp. III. 97–
III.103.
25. Von Berg, E., Alajbegovic, A., Tatschl, R., Krüger, C. and
Michels, U., ILASS-Europe (2001).
26. Greif, D., De Ming, W., Proceedings of the ASME
FEDSM’06, Paper FEDSM2006-98501 (2006).
27. Giannadakis, E., Gavaises, M. and Arcoumanis, C., J. Fluid
Mech. (2008), vol. 616, pp. 153–193.
28. Giannadakis, E., Papoulias, D., Gavaises, M., Arcoumanis,
C., Soteriou, C.and Tang, W., SAE Technical Paper 2007-
01-0245, (2007).
29. Avva, R. K., Singhal, A. and Gibson, D. H. (1995), ASME
FED, pp. 63–70.
30. Wallis, G. B. (1969) One-Dimensional Two-phase Flow, p.
143. McGraw-Hill.
31. Ning, W., Reitz, R.D., Diwakar, R., Lippert, A. M., SAE
Technical Paper No. 2008-01-0936 (2008).
32. Bilicki, Z., Kestin, J., (1990). Proc. Roy. Soc. Lond. A. 428,
379–397.
33. Schmidt, D.P., Gopalakrishnan, S. and Jasak, H., Int. J. of
Multiphase Flow 36 (2010) 284–292.
34. Alajbegovic A., G. Meister, D. Greif, B. Basara,
Experimental Thermal and Fluid Science 26(6-7), pp. 677–
681, (2002).
35. Winklhofer, E., Kull, E., Kelz, E., and Morozov, A., ILASS-
Europe, Zurich (2001).
36. Richards, K. J., Senecal, P. K., and Pomraning, E.,
CONVERGE (Version 1.4.1) Manual, Convergent Science,
Inc., Middleton, WI, (2012).
37. Senecal, P. K., Richards, K. J., Pomraning, E., Yang, T., Dai,
M. Z., McDavid, R. M., Patter-son, M. A., Hou, S., and
Shethaji, T., SAE World Congress Paper No. 2007-01-0159
(2007).
38. Zhao, H., Quan, S., Dai, M., Pomraning, E., Senecal, E.,
Xue, Q., Battistoni, M., and Som, S., Proceedings of the
ASME 2013 Internal Combustion Engine Division Fall
Technical Conference ICEF2013-19167, Dearborn, MI
(2013).
39. Plesset, M. and Prosperetti, A., Annu. Rev. Fluid Mech. 9,
145 (1977).
40. Downar-Zapolski, P., Bilicki, Z., Bolle, L., Franco, J., 1996.
The non-equilibrium relaxation model for one-dimensional
flashing liquid flow. IJMF 22, 473–483.
41. Lord Rayleigh, (1917). Philosophical Magazine 34, 94–98.
42. Plesset, M. S. 1949. J. Appl. Mech. 16:277-82.
43. Hinze J. O., (1975), Turbulence, p. 309. McGraw-Hill.
44. Drew, D.A., Ann. Rev. Fluid Mech., V. 15, (1983).
45. Drew, D.A. and Passman, S.L., Theory of Multicomponent
Fluids, Springer-Verlag, New York, (1999).
46. Ishii, M., Thermo-Fluid Dynamic Theory of Two-Phase
Flow, Eyrolles, Paris, (1975).
47. Kocamustafaogullari G., Ishii M., (1995), Int. J. Heat Mass
Transfer. Vol. 38, No. 3, pp. 481-493.
48. Ishii M., Sun, X., Kim, S., (2003), Annals of Nuclear Energy
30 1309–1331.
49. Avl List GmbH, AVL Fire v.2011 – Eulerian Multiphase
(2011).
50. Sato, Y. and Sekoguchi, K., (1975), Int. J. Multiphase Flow,
2 (79).
51. Perry, R.H., Green, D.W., Perry’s Chemicals Engineers’
Handbook, McGraw-Hill, 7th Ed., (1997).