You are on page 1of 12

Special Issue Article

High-resolution large eddy simulations


of cavitating gasolineethanol blends

International J of Engine Research


14(6) 578589
IMechE 2013
Reprints and permissions:
sagepub.co.uk/journalsPermissions.nav
DOI: 10.1177/1468087413501824
jer.sagepub.com

Daniel J Duke1,2, David P Schmidt3, Kshitij Neroorkar3,


Alan L Kastengren4 and Christopher F Powell1

Abstract
Cavitation plays an important role in the formation of sprays in fuel injection systems. With the increasing use of
gasolineethanol blends, there is a need to understand how changes in fluid properties due to the use of these fuels can
alter cavitation behavior. Gasolineethanol blends are azeotropic mixtures whose properties are difficult to model. We
have tabulated the thermodynamic properties of gasolineethanol blends using a method developed for flash-boiling
simulations. The properties of neat gasoline and ethanol were obtained from National Institute of Standards and
Technology REFPROP data, and blends from 0% to 85% ethanol by mass have been tabulated. We have undertaken highresolution three-dimensional numerical simulations of cavitating flow in a 500-mm-diameter submerged nozzle using the
in-house HRMFoam homogeneous relaxation model constructed from the OpenFOAM toolkit. The simulations are conducted at 1 MPa inlet pressure and atmospheric outlet pressure, corresponding to a cavitation number range of
1.0661.084 and a Reynolds number range of 15,00040,000. For the pure gasoline case, the numerical simulations are
compared with synchrotron X-ray radiography measurements. Despite significant variation in the fluid properties, the
distribution of cavitation vapor in the nozzle is relatively unaffected by the gasolineethanol ratio. The vapor remains
attached to the nozzle wall, resulting in an unstable annular two-phase jet in the outlet. Including turbulence at the conditions studied does not significantly change mixing behavior, because the thermal nonequilibrium at the vaporliquid interfaces acts to low-pass filter the turbulent fluctuations in both the nozzle boundary layer and jet mixing layer.

Keywords
OpenFOAM, HRMFoam, gasoline, ethanol, X-ray radiography

Date received: 25 February 2013; accepted: 12 July 2013

Introduction
Cavitating channel flow is a complex fluid mechanical
phenomenon which has many applications and has
been studied in detail for some time.1 In particular,
cavitation is a problem in fuel injection for internal
combustion engines, where the fuel travels through a
small nozzle under a large pressure gradient to produce
a spray. Cavitation leads to component wear and can
markedly change the structure of the resulting liquid
spray.2,3 These changes in spray structure can lead to
changes in engine performance, emissions, and efficiency.46
Gasolineethanol blends are becoming an important
factor in injector design as their use increases and the
ethanol ratios also increase. A significant number of
studies have been undertaken on the effects of ethanol
blends on emissions.710 Experimental analysis of the
effects of gasolineethanol blend ratio on the spray
structure from simplified two-dimensional (2D) model

injectors reveals that the morphology of the spray is


not strongly affected by the quantity of ethanol.11
However, little work has been done on the effects of
gasolineethanol blend ratio on the internal flow of
cavitating nozzles. Gasolineethanol blends are azeotropic mixtures whose properties are difficult to model.
1

Energy Systems Division, Argonne National Laboratory, Argonne, IL,


USA
2
Laboratory for Turbulence Research in Aerospace & Combustion,
Department of Mechanical & Aerospace Engineering, Monash University,
Melbourne, VIC, Australia
3
Department of Mechanical & Industrial Engineering, University of
Massachusetts, Amherst, MA, USA
4
X-Ray Science Division, Advanced Photon Source, Argonne National
Laboratory, Argonne, IL, USA
Corresponding author:
Daniel J Duke, Energy Systems Division, Argonne National Laboratory,
9700 South Cass Avenue, Argonne, IL 60439, USA.
Email: dduke@anl.gov

Downloaded from jer.sagepub.com at City University Library on June 22, 2015

Duke et al.

579

Experimental approaches to measuring cavitation


using visible light are often limited to the analysis of
2D nozzles and can only measure the outer structure of
cavitating regions.12,13 Recently, X-ray-based techniques have enabled measurements of the vapor fraction in three-dimensional (3D) cavitating flows.14
Synchrotron X-ray radiography methods which have
been proven useful for measuring the mass distribution
of fuel sprays15 are also being extended to measuring
cavitation in model nozzles. However, X-ray propagation techniques are only capable of measuring mass distribution by attenuation (not velocity, pressure or
temperature). Numerical simulations are still essential
to understanding the fluid dynamics of cavitating nozzles, and how they may be affected by variations in
fluid properties.
A range of approaches have been used to model
cavitating flow. Many cavitation models simplify the
two-fluid problem by assuming that the two-phase mixture can be approximated as a single fluid with varying
sound speed and density, with either compressible or
incompressible mass conservation.1618 In a 3D simulation, interface velocity slip does not need to be modeled
as long as the discretization is fine enough.19 Singlefluid models are often preferred over a full compressible
two-phase calculation, since the time and length scales
become so small in the full compressible approach that
computational cost can become prohibitive, especially
if a transient solution is desired over a reasonable timespan in a large domain. The use of a single-fluid model
often leads to the assumption that the vapor and liquid
phases are in instantaneous local equilibrium. In some
models, a constant empirical time scale for homogenization to local equilibrium has been used, with the continued assumption of thermal equilibrium.20 However,
such assumptions are not valid in cavitating flashboiling flows which experience significant nonequilibrium in both phase change rate and temperature. The
homogeneous relaxation model (HRM) used in this
study allows a variable relaxation time for both phase
change rate and temperature, which depends on the
local pressure and void fraction.21
The modeling of fluid properties is also an important
factor in considering how ethanol blending may affect a
cavitating nozzle flow. Many cavitation models handle
fluid properties via an equation of state. A barotropic
model is often used so that the density field can be calculated from the pressure field using predefined empirical constants. However, when considering the effects of
fluid properties, a more accurate model of the fluid
properties may be desired. An alternative approach is
to precompute a lookup table of fluid properties over
the desired pressure and temperature range.19 Since
lookup tables more accurately capture slight variations
in fluid properties, this approach is ideally suited to an
investigation of the effects of various fuel blends.
Previous experiments on the effect of fuel viscosity and
density on nozzle discharge coefficient suggest a

significant sensitivity to viscosity at megapascal pressures, but little sensitivity to density variation.22
Neroorkar and Schmidt23 have used the HRM to
model the cavitating and flash-boiling of fuel in a 3D
multihole fuel injector nozzle with a range of fluids,
including n-hexane, n-octane, and E60 and E85 ethanol
blends. The fluid properties were obtained from lookup
tables.24 They found that the E60 blend generated more
vapor and a wider spray cone angle. These calculations
assumed a laminar flow, so the question remains as to
how the inclusion of turbulence may affect the mixing
of vapor and liquid. Other questions which need investigation are how the flow may be affected by a wider
range of ethanol blends, and how well the simulation
results might compare with experimental data in a
canonical geometry.
Although previous studies have often neglected turbulence, it is known that it can have a significant effect
on a cavitating flow.25 Where turbulence modeling is
implemented, the most popular approach is k2e, which
performs well for free shear flows at modest pressure
gradients, is computationally efficient, and permits simple analysis of grid convergence.17,18,20,26 Som et al.27
investigated fluid properties by simulating the effects of
diesel and biodiesel fuels, where the effects of turbulence were included using realizable k2e in both a
channel flow and an axisymmetric injector geometry.
They found that the local strain rate due to turbulent
mixing was an important factor in cavitation inception.
The largest turbulent features in nozzle flow are therefore likely to strongly interact with cavitation features.
This behavior can be more accurately captured using
large eddy simulation (LES), which is not as frequently
implemented.
In this article, we have undertaken high-resolution,
3D LESs of cavitating flow in a canonical nozzle geometry at fixed pressure boundary conditions using the
single-fluid HRM.21 The fluid properties have been
modeled using precalculated lookup tables.24,28
Variations in cavitation behavior between a range of
gasolineethanol blends from E0 to E85 have been
investigated. For the pure gasoline case, the numerical
simulations are compared against synchrotron X-ray
radiography measurements of the local void fraction.

Method
The canonical nozzle geometry used in the following
simulations is a submerged nozzle with a throat diameter of D = 0.5 mm, and an expansioncontraction
ratio of 5. The nozzle length is 5D (2.5 mm), and the
contraction and expansion have hemispherical profiles.
The inlet boundary condition is a fixed pressure of
1.034 MPa and a temperature of 300 K, with zero velocity gradient. The outlet boundary condition is a fixed
pressure of 0.1 MPa, with zero velocity gradient. The
geometry and boundary conditions were chosen to
match those used in the X-ray experiments.

Downloaded from jer.sagepub.com at City University Library on June 22, 2015

580

International J of Engine Research 14(6)

The application of appropriate wall boundary conditions in LES remains an open research question. In this
study, we have applied the standard set of wall boundary conditions used in the OpenFOAM solver. Velocity
is assumed zero at the wall, and density gradient is
assumed to be zero. The flow is assumed isenthalpic, so
no boundary conditions for the energy terms (such as
thermal diffusivity) are required. For the LES model
terms, the subgrid scale viscosity uses the Spalding wall
function for mSGS and k has zero gradient.

the boundary conditions are shown in Figure 2. It can


be seen that, consistent with the observations of Kar
et al.,31 the REFPROP code predicts that the gasoline
ethanol blends show azeotropic behavior and the vapor
pressure trend is nonlinear. The sound speed and viscosity are at a minimum around 20% ethanol by volume. The change in properties is relatively large over
the range of blends considered; the saturated liquid
pressure decreases by around 50% and the liquid density by 13% from E0 to E85.

Gasolineethanol blend modeling

HRMFoam solver

In order to obtain the properties of gasoline, a fuel surrogate was employed consisting of four components:
15% isopentane, 20% hexane, 45% isooctane, and
20% decane by mass. This surrogate was used by
Styron et al.29 and was designed to match the distillation curve of a California Phase II gasoline. The fluid
property lookup tables required by the HRMFoam
flow solver were generated by creating mixtures of the
above-mentioned surrogate with ethanol using the
REFPROP fluid property program.28 The REFPROP
program uses equations of states to calculate the thermodynamic and transport properties of fluids and
mixtures.
To confirm that the modeling procedure is accurate,
experimental data from Takeshita et al.30 were used to
compare the distillation curves for the different ethanol
blends in Figure 1. To predict the distillation curves
using the REFPROP code, a subroutine was obtained
from EW Lemmon (April 2012, personal communication). This subroutine solves the vaporliquid equilibrium for the surrogate to calculate the number of
moles lost in the form of vapor during the distillation
process, and consequently the volume of liquid
obtained from the condensation of the vapor. The ratio
of volume condensed to the original liquid volume
gives the evaporated volume fraction.
The saturation pressure, density, viscosity, and
sound speed of the gasolineethanol blend models at

The solver used in the following study is the HRM


(HRMFoam) for cavitating and flash-boiling flows,
which is an in-house code running on the OpenFOAM
toolkit.32 A complete description of the solver is given
by Neroorkar et al.;21 a brief summary of which is given
here. The HRM relates the substantial derivative of the
vapor mass fraction (x) to the nonequilibrium time
scale u
dx
xx
=
dt
u

where x is the equilibrium vapor mass fraction over the


time u. The value of x is calculated based on a flashing
process, as implemented in REFPROP.33
The physical concept behind equation (1) is that the
vapor fraction will return to an equilibrium value over
a time u according to a complicated function which can
be linearized.34 The instantaneous mass fraction is calculated from the vapor volume fraction, density, and
vapor density as x = arv =r, where the vapor volume
fraction is defined as
a=

E10
E20
E40
E60
E10
E20
E40
E60

460
Temperature, K

440
420
400
380

u = u0 a0:54 c1:76
Psat  P
Psat  Pcrit

3
4

where the empirical constant u0 = 3:843107 s.21,23 To


generalize their expression to multicomponent mixtures, we use the bubble point pressure in place of Psat .
In practice, the mass fraction of vapor in any given cell
is almost always small; consequently, the bubble point
is a good representation of the pressure during phase
change. The mass, momentum, and energy conservation equations are expressed in terms of mass flux
f = ru

360
340
320
300
0

rl  r
r l  rv

The time scale u is calculated via an empirical relation proposed by Downar-Zapolski et al.35

c=
480

0.2

0.4

0.6

0.8

Evaporated Volume Fraction

Figure 1. Distillation curves for several gasolineethanol


blends calculated using the REFPROP model (lines), as compared
to the experimental data of Takeshita et al.30 (markers).

r
+r  f=0
t
ru
+ r  fu =  rP + rt
t

Downloaded from jer.sagepub.com at City University Library on June 22, 2015

5
6

581

Sound speed, m/s

Inlet conditions
Outlet conditions

1080
1040
1000
960

Dynamic Visosity, Pa-s

1120

Current Study
Kar et al

40
35
30
25
20
15
10

1x10

-3

8x10

-4

6x10

-4

0.2

0.4 0.6 0.8


Ethanol fraction

780

Inlet conditions
Outlet conditions

4x10-4
2x10

Density, kg/m3

Saturation Pressure (kPa)

Duke et al.

-4

0.2
0.4
0.6
Ethanol fraction

0.8

Inlet conditions
Outlet conditions

760
740
720
700
680
660

0.2

0.4

0.6

0.8

Ethanol fraction

0.2

0.4

0.6

0.8

Ethanol fraction

Figure 2. Calculated properties of gasolineethanol blends at the boundary conditions (inlet pressure 1 MPa, outlet pressure 0.1
MPa). Vapor pressure reference data are from Kar et al.31

rh
P
+ r  fh =
+ u  rP
t
t

where t is the shear stress tensor. For a single-phase


flow, equations (5) and (6) can be closed to form a
pressure equation. However, due to phase change,
equations (5)(7) are not closed, so equations (1)(4)
are implemented to deal with the additional terms. The
off-diagonal terms are represented by H(u) and the
diagonal terms by a, such that aj uj = Hu  rP.34
The derivation is described in full by Neroorkar and
Schmidt23 The implicit solution for the pressure equation is expressed in its final form as
 



1 r 
rP
H(u)
+ rrPfv + r
r Px, h t
aj

8
1
P dx
=0
 rr  rP + 
aj
x P, h dt
The first two terms concern the transmission of pressure waves and are neglected since they result in an
asymmetric linear system for pressure. Such systems
cost more to solve and produce less robust solutions.
Past tests19 found that these compressibility terms had
negligible impact on the predicted flowfield. The incompressible liquid assumption has previously been shown
to yield reasonable results.21,23,25 The pressure equation
is solved using a pressure-implicit split-operator (PISO)
predictor-corrector algorithm with 10 iterations.36
Turbulence is modeled in this study using LES, with a
single-equation subgrid scale model.32,37 This allows us
to investigate the instantaneous interactions between
turbulent mixing and cavitation at large scales. The discretization schemes are second order, using gamma differencing for the r, P, h, and u divergence terms and

upwind differencing for the turbulent effective k and e


divergences. The Laplacian terms are discretized using
second-order central differencing. The pressure field is
solved using the fast simplified diagonal-based incomplete Cholesky preconditioner and a generalized
geometric-algebraic multigrid solver for the final step,
with under-relaxation. The r, h, k, e, and u fields are
solved using a simplified diagonal-based incomplete
LU preconditioner and a preconditioned biconjugate
gradient method for the final step, with under-relaxation. The solution is calculated on a cluster computer at
Argonne National Laboratory, with OpenFOAMs
implementation of Message Passing Interface (MPI)
parallelization over 96 CPUs. The time-stepping
scheme is Courant number limited such that the local
Courant number does not exceed 0.8 at any cell.

Mesh development and resolution study


The mesh used in the following simulations is a 3D
unstructured hex mesh with 8.4 million cells, as shown
in Figure 3. The mesh is successively refined in the
throat region; the mean cell size relative to the throat
diameter is D=d50.
In order to determine the rate of convergence of the
solution with increasing mesh resolution, three successively refined meshes with cell sizes of D=d = 25,
D=d = 34, and D=d = 50 have been solved out to a
simulation time of 2 ms with neat gasoline as the working fluid. Profiles of the steady-state, time-averaged
vapor volume fraction, velocity, turbulent energy, and
pressure are shown in Figure 4, along a streamwise vector which passes near the throat wall where the flow is
strongly cavitating (at a radial position of r=R = 0:96).

Downloaded from jer.sagepub.com at City University Library on June 22, 2015

582

International J of Engine Research 14(6)

Figure 3. Mesh of cavitating nozzle: (a) centerline slice, cropped around the nozzle throat and (b) oblique view.

Figure 4. Convergence of time-averaged vapor volume fraction, velocity magnitude, turbulent energy, and pressure with increasing
grid resolution, plotted along a streamwise vector near the nozzle wall, at a radial position of r=R = 0:96.

The three time-averaged solutions have been used to


perform a Richardson extrapolation to determine how
close the D=d = 50 mesh is to a converged solution.
The sum square error in the centerline pressure is
0.04% with respect to the converged solution since it is
pinned by the boundary conditions. The velocity magnitude is within 0.12%, and the density is within 0.66%.

The turbulent energy and vapor fraction are within


1.2% of convergence. Given that the Richardson extrapolate of three successively refined meshes may not represent the true converged solution, a series expansion
error contribution of order O(dkn + 1 ) is included in the
above error values, where kn is the convergence rate of
the extrapolation.

Downloaded from jer.sagepub.com at City University Library on June 22, 2015

Duke et al.

583

1.4
1.2

0.8

Density, kg/m3

|U|max / UB

1
E0
E10
E20
E40
E60
E85

0.6
0.4
0.2
0
0

0.2

0.4

0.6

0.8

900
800
700
600
500
400
300
200
100
0
-100

E0
E10
E20
E40
E60
E85

0.2

Time, ms

0.4
0.6
Time, ms

0.8

Figure 5. Evolution of density extremes and maximum velocity (normalized by the Bernoulli velocity) in transient solution.

Results and discussion


The solution is run until the net momentum and mass
flux reach a steady state, which occurs after approximately 0.8 ms. The solution is then run out to at least 1
ms. Figure 5 shows the evolution of the minimum and
maximum density and the maximum velocity with
respect to simulation time. The minimum density is a
good marker of cavitation inception; phase change
begins about 0.4 ms after start of injection at the given
pressure gradient and reaches a steady state after 0.5
ms. The maximum velocity magnitude reaches apsteady

state very close to the Bernoulli velocity UB =2 DP=r,


as expected.16 The selected boundary conditions
and fluid properties result in a Reynolds number
range of ReD =UD=nl 2 (1:53104 ,4:03104) and a cavitation number range of K= P1 Pvap =P1 P2 2
(1:066,1:084). Both Re and K vary due to the change in
density, viscosity, and vapor pressure of the varying
blends, while the pressure and geometry remain constant (see Figure 2). The bulk velocity U is the mean
centerline pressure in the nozzle throat, which is close
to the Bernoulli velocity.
Most cavitation vapor is generated at the sharp inlet
to the nozzle. The vapor in the computations sticks to
the wall through the nozzle since nearly all the pressure
drop happens in the inlet region and the pressure gradient along the nozzle throat is flat. Centerline slices of
vapor volume fraction in the nozzle for each of the
blends studied are shown in Figure 6. An annular jet of
vapor is generated at the nozzle outlet. This unstable
configuration rapidly breaks down and the vapor
jet mixes with the surrounding liquid, as shown in
Figure 7. A reasonable fraction of the vapor does not
condense but is convected through the outlet. Very little change in the vapor distribution is observed across
the different blends investigated; this is a surprising
result given the significant change in fluid properties.
In the application of cavitating nozzles to fuel injection systems, the main location of interest is the nozzle
exit plane, where the mass flow rate, velocity profile,
and vapor distribution will have an impact on spray
formation.3 The time-averaged steady-state velocity

Figure 6. Centerline slices of vapor volume fraction at 1 ms


after start of injection. The nozzle outlet is submerged. (a) E0
(neat gasoline), (b) E10 (10% ethanol, 90% gasoline), (c) E20
(20% ethanol, 80% gasoline), (d) E40 (40% ethanol, 60%
gasoline), (e) E60 (60% ethanol, 40% gasoline), and (f) E85 (85%
ethanol, 15% gasoline).

distribution over the outlet plane has been azimuthally


averaged to generate a mean radial profile, as shown in
Figure 8. The velocity magnitude (Figure 8(a)) has a
well-developed profile which does not vary significantly
with ethanol blend. The spanwise velocity components

Downloaded from jer.sagepub.com at City University Library on June 22, 2015

584

International J of Engine Research 14(6)

Figure 7. Iso-contours of cavitation vapor distribution for E20


blend, 1 ms after start of injection.

indicate that the flow becomes slightly more divergent


near the wall as the ethanol ratio increases. The total
mass flux in the outlet plane is shown in Figure 9(a),
normalized as a discharge coefficient
s
m_ 1  b4
CD =
9
A 2rDP

60
50
40

E0
E10
E20
E40
E60
E85

30
20
10
0
-1

-0.5

X-ray radiography measurements


In order to provide experimental comparison of the
simulation results, X-ray radiography measurements of
a cavitating flow in a nozzle of the same nominal

Spanwise velocity (m/s)

Velocity magnitude (m/s)

where b =
ratio
0:2 is the nozzle diameter contraction
1
and m_ = rUdA. For a round nozzle,
Nurick
predicts
p
a discharge coefficient of CD = CC K, where the coefficient CC 0:61 for a sharp-edged nozzle with b ! 0.16
Figure 9(a) indicates that Nuricks relation tracks

reasonably well with the simulation results. Using


CC = 0:61, Nuricks relation underpredicts the mass
flux by about 5%. In addition to the effect of the nonzero b, the hemispherical inlet profile to the nozzle will
cause an increase in CC similar in effect to an obtuse
inlet angle, which would explain the underprediction.
However, the effect
of b is known to be very small; CD
p
varies with b as 1  b4 0:9992 in this geometry.38
The best fit value for the results is CC 0:63, which
agrees reasonably well with previous simulations of the
effects inlet angle.16
The vapor mass flux weighted
by the total mass

flux is calculated as m_ v = xrUdA and is shown in


Figure 9(b). The vapor flux decreases by a factor of 3
from E0 to E85 blends, peaking for the E40 blend. An
increase in vapor flux corresponds to a reduction in
CD . Given that the vapor volume fraction and its spatial extent do not significantly change (Figure 6), this
suggests that the discharge coefficient is not varying
due to area occlusion, but rather due to changes in density caused by variation in fluid properties.

0
r/R

0.5

0.7
0.6
0.5
0.4
0.3
0.2
0.1
0

E0
E10
E20
E40
E60
E85

-1

-0.5

(a)

0
r/R

0.5

(b)

Figure 8. Time-averaged and azimuthally averaged radial velocity profiles in the nozzle outlet plane, for (a) the streamwise
component of velocity and (b) the total spanwise component for the different blends considered.

0.7

3.5

Simulations
Nurick, CC =0.61
Nurick, CC =0.63

0.68

3
2.5

0.64

0.62

1.5

CD

0.66

0.6

Simulations
X-ray Experiment

1
0

0.2

0.4

0.6

Ethanol Fraction

(a)

0.8

0.2

0.4

0.6

0.8

Ethanol Fraction

(b)

Figure 9. Time-averaged fluxes at the nozzle outlet plane, for (a) total mass flux as a discharge coefficient and (b) weighted vapor
mass flux versus blend ratio.

Downloaded from jer.sagepub.com at City University Library on June 22, 2015

Duke et al.

585

(a)
PiN
Diode
Nozzle Wall
Vapor
Liquid

monochromatic
x-ray source

(b)
Figure 10. X-ray radiography experiment. (a) Monochromatic radiography microprobe setup at Sector 7-BM of the Advanced
Photon Source. The diagram is not to scale; the distance from the source to the experiment is 35.5 m. The beam is focused to a 5 3
6 mm spot where it passes through the nozzle. (b) Example cross-section of submerged nozzle, showing the typical interaction of
the X-ray beam with the flow.

geometry as that used in the simulations have been


undertaken using a commercial gasoline surrogate whose
properties are close to the E0 model. Measurements were
performed at the 7-BM beamline of the Advanced
Photon Source (APS) at Argonne National
Laboratory.39 A schematic of the experiment is given in
Figure 10(a). The X-ray beam is monochromatic, with a
mean energy of 8 keV, a bandwidth of 1.4% full width
at half maximum (FWHM), and a flux of approximately
2:531011 ph=s. The beam is focused to a spot size of
536 mm FWHM with a pair of KirkpatrickBaez mirrors. The focused beam acts as a microprobe, allowing
the cavitation vapor fraction over a small area to be
probed along a line of sight. The incoming X-ray intensity is normalized and the transmitted X-rays are collected by a 300-mm-thick PIN diode. The experiment is
placed 36.5 m downstream from the source; the PIN
diode is approximately 5 cm downstream of the experiment. The normalized PIN diode reading is sampled for
dt = 0:5 s at each sample point.
The quantity of cavitation vapor is measured along
a line of sight dz at some point of interest. The experiment is then repeated at a noncavitating condition
where the nozzle is filled with liquid at the same inlet
pressure, but over a very small pressure gradient where
there is no cavitation (denoted by prime). The presence
of vapor results in a relative increase in transmission
due to a reduction in the absorbing mass in the beam
path. The X-ray attenuation through the sample can be

described by the LambertBeer Law. The known density of the liquid phase is used to convert mass per unit
area measurements from radiography to a path length
dzv , which represents the projected depth of vapor in
the path of the beam
!
I1
1
I0
dzv =
loge I91
10
ml rl
I9
0

The quantity of vapor is expressed as the total path


length of vapor dzv in the beam, rather than an average
vapor fraction since the vapor is not evenly distributed
along the line of sight. An example of a typical measurement is given in Figure 10(b); the X-ray beam interacts with both the vapor near the wall and the liquid in
the core of the flow, and the resulting measurement is
integrated along the beam path.
The attenuation coefficient ml is measured by filling
capillary tubes with the working fluid, water, and air,
and measuring the X-ray transmission through the
tubes. It was found that ml rl 0:3124mm1 . We have
neglected the vapor phase attenuation coefficient since
rv  rl . The uncertainty in the time-averaged projected
vapor quantity is determined by error propagation calculations to be edzv 63:5mm of vapor (0.007 D), which
is 2% of the maximum path length of vapor in a typical
measurement.
In earlier X-ray experiments, a significant quantity
of vapor was observed along the nozzle centerline that

Downloaded from jer.sagepub.com at City University Library on June 22, 2015

586

International J of Engine Research 14(6)

(a)

(b)

Figure 11. Projected vapor quantity dzv =R from X-ray measurements at two streamwise locations along the nozzle. The X-ray data
are compared against simulated projections from the numerical data, at 1 ms after start of injection: (a) x=L = 0:1 and (b) x=L = 0:7.

was not captured in the models. Such features were also


observed in the X-ray tomography experiments of
Bauer et al.14 and were proposed to be related to cavitation phenomena; either migration of vapor away
from the wall or perhaps a string type along the nozzle
centerline. A recent numerical modeling study considered the modeling of dissolved gas and cavitation in
pure gasoline in the same submerged geometry40 and
proposed that this centerline vapor may be due to dissolved gas in the fuel coming out of solution, rather
than cavitation. Since the X-ray radiography experiment is unable to distinguish between the two phenomena clearly, the experiment was repeated with the fuel
degassed, and the results of this improved experiment
are those shown in Figure 11. Once dissolved gas is
controlled for, the experiment and model agree reasonably well, although some asymmetry remains. The nozzle wall was smoothed in the experiment to remove
machining defects which may cause asymmetry. The
asymmetry that remains in Figure 11(b) is indicative of
the fact that cavitation inception depends on local
nucleation at the wall, and small machining defects can
lead to large deviations of the vapor distribution further downstream, even in a very smooth nozzle.
In order to directly compare the X-ray data against
the LES, simulated X-ray experiments are performed
on the mesh by projecting several thousand rays
through the domain and computing the projected
vapor distributions. The resulting measurements are a
volume integral of the local vapor fraction normalized
over the X-ray beam cross-section, normalized to units
of micrometers. The results are shown in Figure 11.
Just downstream of the nozzle inlet in Figure 11(a),
both experimental and simulated profiles take the form
of an annular vapor distribution. The differences
between the experiment and the simulation for the pure
gasoline case are significantly larger than the variations
between the different gasolineethanol blends.
The variations between experiment and simulation
are much more significant than the variations between

the different simulation cases, suggesting that ethanol


blend does not strongly affect the distribution of vapor
in the nozzle. Furthermore, the distribution of vapor
can be integrated in a plane to determine the total area
ratio occluded by vapor, and by combining this with
the known mass flow rate of fuel (using a turbine flowmeter), the weighted vapor mass flux in the outlet plane
_ v =A)(rv =r), assuming that
can be estimated as m_ v m(A
the slip velocity is not significant. The resulting vapor
flux is shown in Figure 9(b) and agrees well with the
simulation results.

Influence of turbulence modeling


The effect of using LES for turbulence has also been
investigated. In addition to the HRMFoam cavitating
LES, the pure gasoline case has been simulated with a
laminar assumption (phase change, but no turbulence
model) and a single-phase incompressible LES using a
PISO solver (turbulence modeled, but no phase
change). A comparison of the predicted velocity fields
in centerline slices through the nozzle at 1.2 ms after
start of injection is given in Figure 12. The single-phase
solution (Figure 12(a)) has a wide range of turbulent
length scales and a developing boundary layer in the
throat as expected of a turbulent liquid jet. Here, there
is no phase change nonequilibrium to impede turbulent
mixing in the outlet.
The regions of high vapor fraction are also the
regions of greatest shear (i.e. the nozzle throat boundary layer and jet mixing layer). As such, the nonequilibrium between vapor and liquid acts to eliminate most
of the boundary layer momentum deficit in the nozzle
throat and also acts to low-pass filter the smaller scales
of turbulence in the jet mixing layer. Thus, the singlephase model (Figure 12(a)) shows a higher turbulence
intensity than the cavitating cases (Figure 12(b)
and (c)).
It is interesting to note that turbulence in the jet
wake takes longer to develop in the LES case

Downloaded from jer.sagepub.com at City University Library on June 22, 2015

Duke et al.

587

Figure 12. Centerline slices of velocity magnitude for E0 (neat


gasoline) at 1.2 ms after start of injection. Contour units are m/
s. (a) Single-phase PISO LES (no phase change); (b) HRM
cavitation model with laminar, unsteady flow; and (c) HRM
cavitation model with LES.

(Figure 12(c)) than in the laminar case (Figure 12(b)).


Initially, this seems counterintuitive because the LES
model should show a more unsteady flow, as the effects
of subgrid scale turbulence are now being considered.
However, when the damping effects of cavitation phase
change are considered, it becomes apparent that cavitation will deter turbulent mixing at the subgrid scale
equally effectively as at the cell scale. The result is a
relative decrease in turbulence intensity as compared to
the noncavitating case, which results in an increased
numerical viscosity applied at the cell level when LES is
implemented. The effects of cavitation dominate turbulent effects near the wall where most turbulence is produced, resulting in a slightly steadier flow. Conversely,
a laminar simulation overpredicts the instability of the
flow because the damping effects of phase change are
only applied at a cell level (i.e. at low wavenumbers),
making the flow artificially more unstable. The implementation of LES does not significantly alter the vapor
production prediction, only the turbulent development
length of the jet wake.

Conclusion
A submerged nozzle of 500 mm diameter has been modeled using high-resolution LES and the HRMFoam
cavitation and flash-boiling solver of Neroorkar et
al.,21 using tabulated fluid properties. The purpose of

the simulations was to investigate the effect of varying


gasolineethanol blends on the vapor distribution in a
canonical geometry. The simulations were also compared to an X-ray radiography measurement of fluid
density.
Despite a significant variation in the fluid properties,
the distribution of vapor does not vary much at all from
pure gasoline up to 85% ethanol, with respect to volume fraction and spatial extent. The vapor mass flux in
the nozzle exit plane varies by a factor of 3 from E0 to
E85, peaking at E40. This occurs due to the variation in
fluid density with ethanol ratio since the profile of the
vena contracta does not vary between cases. Given that
K does not change much as the inlet pressure increases
and the nozzle is effectively choked, it is likely that the
effects of fluid properties such as vapor pressure are
saturated at lower cavitation numbers. Furthermore,
laminar and single-phase solutions in the pure gasoline
case indicate that turbulence modeling with LES does
not significantly alter the vapor fraction or velocity
fields with respect to a laminar flow assumption; however, it does act to damp instability in the jet wake. The
Reynolds numbers achieved in the simulations are in
the range of 15,00040,000 depending on the viscosity
and density of the fluids. In a fuel injector, the cavitation number is likely to be much higher due to the
larger pressure gradient, and the Reynolds number will
be similar. Therefore, it may be hypothesized that similar sensitivity to the blend ratio would be observed in a
cavitating axial single-hole injector.
The X-ray cavitation measurement was found to be
highly sensitive to the quantity of dissolved gas in the
fuel, necessitating that the fuel be degassed prior to the
experiment. As such, prediction of the location of
vapor in the nozzle depends on the level of degassing. It
remains to be seen as to how factors not included in the
HRM model such as transmission of pressure waves in
the liquid phase may affect the solution. The separation
of vapor from the wall in the experiment which does
not occur in the simulations is also a matter for further
investigation. The simulations presented here suggest
that the cavitation vapor distribution in a nozzle of sufficiently large cavitation number will not be affected by
gasolineethanol blend ratio.
Acknowledgements
The authors wish to acknowledge Dr Eric Lemmon of
the National Institute of Standards and Technology
(NIST) for his support in the fluid property modeling
process described in section gasolineethanol blend
modeling. They also acknowledge the use of Dr
Lemmons program code to generate the distillation
curves in Figure 1. They 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 also acknowledge General
Motors Research Center, NASA, and the National

Downloaded from jer.sagepub.com at City University Library on June 22, 2015

588

International J of Engine Research 14(6)

Science Foundation for their support of the development of HRMFoam. The X-ray experiments presented
in this research were performed at the 7-BM beamline
of the APS at Argonne National Laboratory. The
authors wish to thank Team Leader Gurpreet Singh
for his support on this work.
Declaration of conflicting interests
The authors declare that there is no conflict of interest.
Funding
Use of the APS is supported by the US Department of
Energy (DOE) under Contract No. DE-AC0206CH11357. The fuel spray research is sponsored by
the DOE Vehicle Technologies Program. D.J.D. was
supported by an ANSTO Fulbright Scholarship in
Nuclear Science & Technology during the research.

References
1. Nurick WH. Orifice cavitation and its effect on spray
mixing. J Fluid Eng: T ASME 1976; 98: 681687.
2. Tamaki N, Shimizu M and Nishida K. Effects of cavitation and internal flow on atomization of a liquid jet. Atomization Spray 1998; 8: 179197.
3. Schmidt DP and Corradini ML. The internal flow of diesel fuel injector nozzles: a review. Int J Engine Res 2001;
2(1): 122.
4. Pierpont DA and Reitz RD. Effects of injection pressure
and nozzle geometry on DI diesel emissions and performance. SAE paper 950604, 1995.
5. Shoji T. Effect of cycle-to-cycle variations in spray characteristics on hydrocarbon emission in DI diesel engines.
JSME Int J B: Fluid T 1997; 40(2): 312319.
6. Blessing M, Konig G, Kruger C, Michels U and Schwarz
V. Analysis of flow and cavitation phenomena in diesel
injection nozzles and its effects on spray and mixture formation. SAE technical paper 2003-01-1358, 2003.
7. Assad MS, Kucharchuk IK, Penyazkov OG, Rusetskii
AM and Chornyi AD. Influence of ethanol on the operating parameters of an internal-combustion engine.
J Eng Phys Thermophys 2011; 84(6): 13111317.
8. Jia L-W, Shen M-Q, Wang J and Lin M-Q. Influence of
ethanolgasoline blended fuel on emission characteristics
from a four-stroke motorcycle engine. J Hazard Mater
2005; 123(13): 2934.
9. He B-Q, Wang J-X, Hao J-M, Yan X-G and Xiao J-H.
A study on emission characteristics of an EFI engine
with ethanol blended gasoline fuels. Atmos Environ 2003;
37(7): 949957.
10. Hsieh WD, Chen RH, Wu TL and Lin TH. Engine performance and pollutant emission of an SI engine using
ethanolgasoline blended fuels. Atmos Environ 2001;
36(3): 403410.
11. Gao J, Jiang D and Huang Z. Spray properties of alternative fuels: a comparative analysis of ethanolgasoline
blends and gasoline. Fuel 2007; 86(1011): 16451650.
12. Suh HK and Lee CS. Effect of cavitation in nozzle orifice
on the diesel fuel atomization characteristics. Int J Heat
Fluid Fl 2008; 29: 10011009.

13. Payri R, Salvador FJ, Gimeno J and de la Morena J.


Study of cavitation phenomena based on a technique for
visualizing bubbles in a liquid pressurized chamber. Int J
Heat Fluid Fl 2009; 30: 768777.
14. Bauer D, Chaves H and Arcoumanis C. Measurements of
void fraction distribution in cavitating pipe flow using xray CT. Meas Sci Technol 2012; 23(5): 055302.
15. Kastengren A, Powell CF, Liu Z and Wang J. Time
resolved, three dimensional mass distribution of diesel
sprays measured with X-ray radiography. SAE paper
2009-01-0840, 2009.
16. Schmidt D, Rutland C and Corradini M. A numerical
study of cavitating flow through various nozzle shapes.
SAE paper 971597, 1997.
17. Jia M, Xie M, Liu H, Lam WH and Wang T. Numerical
simulation of cavitation in the conical-spray nozzle for
diesel premixed charge compression ignition engines. Fuel
2011; 90(8): 26522661.
18. Singhal AK, Athavale MM, Li H and Jiang Y. Mathematical basis and validation of the full cavitation model.
J Fluids Eng 2002; 124: 617623.
19. Schmidt DP, Gopalakrishnan S and Jasak H. Multidimensional simulation of thermal non-equilibrium channel flow. Int J Multiphas Flow 2010; 36(4): 284292.
20. Weische S. Numerical simulation of cavitation effects
behind obstacles and in an automotive fuel jet pump.
Heat Mass Transfer 2005; 41: 615624.
21. Neroorkar K, Shields B, Grover R Jr, Plazas Torres A
and Schmidt DP. Application of the homogeneous
relaxation model to simulating cavitating flow of a diesel
fuel. SAE Paper 2012-01-1269, 2011.
22. Dernotte J, Hespel C, Foucher F, Houille S and Mounaim-Rousselle C. Influence of physical fuel properties
on the injection rate in a Diesel injector. Fuel 2012; 96:
153160.
23. Neroorkar K and Schmidt D. A computational investigation of flash-boiling multi-hole injectors with gasolineethanol blends. SAE paper 2011-01-0384, 2011.
24. Neroorkar K and Schmidt D. Modeling of vapor-liquid
equilibrium of gasolineethanol blended fuels for flash
boiling simulations. Fuel 2010; 90(2): 665673.
25. Roth H, Gavaises M and Arcoumanis C. Cavitation
initiation, its development and link with flow turbulence
in diesel injector nozzles. SAE paper 2002-01-0124, 2002.
26. Andriotis A, Gavaises M and Arcoumanis C. Vortex flow
and cavitation in diesel injector nozzles. J Fluid Mech
2008; 610: 195215.
27. Som S, Longman DE, Ramirez AI and Aggarwal S.
Influence of nozzle orifice geometry and fuel properties
on flow and cavitation characteristics of a diesel injector.
In: Lejda K (ed.) Fuel injection in automotive engineering.
New York: InTech, 2012, pp.111126.
28. Lemmon EW, Huber ML and McLinden MO. NIST
standard reference database 23: reference fluid thermodynamic and transport propertiesREFPROP. Version 8.0,
April 2007. Boulder, CO: National Institute of Standards
and Technology.
29. Styron JP, Kelly-Zion PL, Lee CF, Peters JE and White
RA. Multicomponent liquid and vapor fuel distribution
measurements in the cylinder of a port-injected, sparkignition engine. SAE paper 2000-01-0243, 2000.
30. Takeshita E, Rezende R, Guelli U, de Souza S and Ulson
de Souza A. Influence of solvent addition on the

Downloaded from jer.sagepub.com at City University Library on June 22, 2015

Duke et al.

31.

32.

33.

34.

35.

589

physicochemical properties of Brazilian gasoline. Fuel


2008; 87(1011): 21682177.
Kar K, Last T, Haywood C and Raine RR. Measurement
of vapor pressures and enthalpies of vaporization of gasoline and ethanol blends and their effects on mixture preparation in an SI engine. SAE paper 2008-01-0317, 2008.
Weller HG, Tabor G, Jasak H and Fureby C. A tensorial
approach to computational continuum mechanics using
object-oriented techniques. Comput Phys 1998; 12(6):
620631.
Lemmon EW, Huber ML and McLinden MO. NIST
standard reference database 23: reference fluid thermodynamic and transport propertiesREFPROP. Version
9.0.1. Boulder, CO: National Institute of Standards and
Technology, June 2012.
Gopalakrishnan S and Schmidt DP. A computational
study of flashing flow in fuel injector nozzles. SAE paper
2008-01-0141, 2008.
Downar-Zapolski P, Bilicki Z, Bolle L and Franco J. The
non-equilibrium relaxation model for one-dimensional

36.

37.

38.
39.

40.

flashing liquid flow. Int J Multiphas Flow 1996; 22(3):


473483.
Issa RI. Solution of the implicitly discretised fluid flow
equations by operator-splitting. J Comput Phys 1986;
62(1): 4065.
Dejoan A and Schiestel R. LES of unsteady turbulence
via a one-equation subgrid-scale transport model. Int J
Heat Fluid Fl 2002; 23: 398412.
Hewitt GF. Measurements of two phase flow parameters.
London and New York: Academic Press, 1978.
Kastengren AL, Powell CF, Arms D, Dufresne EM, Gibson H and Wang J. The 7BM beamline at the APS: a
facility for time-resolved fluid dynamics measurements. J
Synchrotron Radiat 2012; 19: 654657.
Battistoni M, Som S and Longman DE. Comparison of
mixture and multi-fluid models for in-nozzle cavitation
prediction (ICEF2013-19093). In: Proceedings of the
ASME 2013 internal combustion engine division fall technical conference (ICEF2013), Dearborn, MI, USA, 13
October 2013. American Society of Mechanical
Engineers.

Downloaded from jer.sagepub.com at City University Library on June 22, 2015

You might also like