See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/215864320

Spray Atomization Models in Engine

Applications, from Correlations to Direct
Numerical Simulations
Article in Oil & Gas Science and Technology September 2011
Impact Factor: 0.75 DOI: 10.2516/ogst/2011116

CITATIONS

READS

81

2 authors:
Fabien Dos Santos

Luis Le Moyne

University of Burgundy

University of Burgundy

4 PUBLICATIONS 4 CITATIONS

89 PUBLICATIONS 295 CITATIONS

SEE PROFILE

All in-text references underlined in blue are linked to publications on ResearchGate,

letting you access and read them immediately.

SEE PROFILE

Available from: Luis Le Moyne

Retrieved on: 23 April 2016

ogst100079_DosSantos

17/11/11

11:16

Page 801

Oil & Gas Science and Technology Rev. IFP Energies nouvelles, Vol. 66 (2011), No. 5, pp. 801-822
Copyright 2011, IFP Energies nouvelles
DOI: 10.2516/ogst/2011116

Dossier

R&D for Cleaner and Fuel Efficient Engines and Vehicles

R&D pour des vhicules et moteurs plus propres et conomes

Spray Atomization Models in Engine Applications,

from Correlations to Direct Numerical Simulations
F. Dos Santos and L. Le Moyne
ISAT Universit de Bourgogne, 49 rue Mlle Bourgeois, 58000 Nevers - France
e-mail: [email protected] - [email protected]

Rsum Modles de spray dans les applications moteur, des corrlations aux simulations
numriques directes Les sprays sont parmi les principaux facteurs de qualit, dans la formation du
mlange et la combustion, dans un grand nombre de moteurs ( combustion interne). Ils sont de toute
premire importance dans la formation de polluants et lefficacit nergtique, bien quune modlisation
adquate soit encore en dveloppement. Pour un grand nombre dapplications, la validation et la
calibration de ces modles demeurent une question ouverte. Aussi, prsentons-nous un aperu des
modles existants et proposons quelques voies damlioration. Les modles sont classs en nondimensionnels et dimensionnels allant de formules simples ddies des applications proches du temps
rel des descriptions dtailles des premiers stades de latomisation.
Abstract Spray Atomization Models in Engine Applications, from Correlations to Direct Numerical
Simulations Sprays are among the very main factors of mixture formation and combustion quality in
almost every (IC) engine. They are of great importance in pollutant formation and energy efficiency
although adequate modeling is still on development. For many applications, validation and calibration of
models are still an open question. Therefore, we present an overview of existing models and propose
some trends of improvement. Models are classified in zero dimensional and dimensional classes ranging
from simple formulations aimed at close-to-real-time applications to complete detailed description of
early atomization stages.

ogst100079_DosSantos

17/11/11

802

11:16

Page 802

Oil & Gas Science and Technology Rev. IFP Energies nouvelles, Vol. 66 (2011), No. 5

INTRODUCTION
Spray applications range from large scale dispersion of insecticides to nanometer thin film deposits with ion sources.
What they have in common is the transformation of a continuous phase (mostly liquids) into a number of separated
droplets created by a specific device, or atomizer. For characterization of sprays, the most commonly used qualities are
size/number distribution of droplets, penetration and spray
angle. Atomizers technologies are based on a number of principles that achieve break-up of continuous phase with surface
or volume forces. The main forces used and their corresponding atomizers used in injection applications for engines are
listed non-exhaustively in Table 1.

depend on space coordinates, although in order to obtain

those results dimensional equations must be solved (Le
Moyne, 2010).

1 0D MODEL
Most of the investigations on sprays conclude with empirical
or semi-empirical laws, which predict the characteristics of
the spray as a function of several parameters.

SMD

TABLE 1
Main types of forces and corresponding atomizers
Force

Atomizer type

Inertial

Rotating, vibration, impinging, swirl type, etc.

Aerodynamic drag

Pressure hole, swirl, two-fluid, air-assisted, etc.

Gas bubbles
growth/collapse

Pressure hole, cavitation, flash,

effervescent, void, etc.

The cost of calibration of the many parameters to be

optimized in modern engines renders real engine tests prohibitively high. It is therefore a need to develop effective
models able to give a description of mechanisms with a less
cost. Simple 0D models are currently integrated into complete phenomenological engine models able to find, through
optimization algorithms, ensembles of parameter values that
comply with pollutant regulations and fuel efficiency for a
number of functioning conditions. 1D to 3D models are used
in design tasks at every stage of engine development.
In the last years, intensive use of CFD (Computational
Fluid Dynamics) codes in engines has still increased and generalized. Yet, among the features needed for more effective
models, there is the atomization and spray sub-models of
many engine models. From pioneering work (Wakuri et al.,
1960; Dent, 1971; Hiroyasu et al., 1978) on penetration correlations to recent efforts in LES and DNS modeling (Apte et
al., 2003; De Villiers et al., 2004; Menard et al., 2005; Fuster
et al., 2009; Lebas et al., 2009), spray models expand
through a very large range of concepts. It is our aim to have
an overview of existing tools for modeling and prospective
insight of future techniques. We focus our interest in the liquid and droplet treatment on a surrounding gas, putting aside
evaporation and mixing models which we consider as out of
scope of this paper.
First we list non-dimensional models and correlations for
spray penetration, spray angle, liquid length and characteristic drop size. In this category, we include atomization models
for primary and secondary break-up as their results do not

Spray
angle

Liquid
length
Spray tip penetration
Figure 1
Different spray parameters.

A spray can be roughly described by four parameters

(Fig. 1): the spray tip penetration S, the spray angle , the
break-up length Lb (or liquid length) and the global Sauter
mean diameter x32. All together are related in the process of
disintegration of the spray and can be modeled by 0D models. Other parameters, like air entrainment which results from
the spray, wont be analyzed in this study. Also, according
with the data, the models are for hole injectors.
1.1 Spray Angle Models
Spray angle (defined in Fig. 1) is influenced by the injector
characteristics, fuel properties and ambient conditions.
Basically during injection, the angle increases rapidly, reaches
a maximum and then decreases to reach a constant value
(Lefebvre, 1989). Many authors try to assess and calculate to
give the constant value for the spray angle. The collected 0D
spray angle models are shown in Table 2.
1.1.1 Reitz and Bracco Model

Reitz and Bracco (1979) proposed a model for the spray

angle by employing the aerodynamic break-up model of
Ranz (1958). It includes the ratio of Reynolds and Weber
numbers of the liquid flow in the function f ():
4 g

A l

0.5

tan () =

f ( )

(1)

ogst100079_DosSantos

17/11/11

11:16

Page 803

F Dos Santos and L Le Moyne / Spray Atomization Models in Engine Applications, from Correlations to Direct Numerical Simulations 803

TABLE 2
Different correlations for spray angle
Model

Correlation

References

4 g

A l

0.5

Reitz and Bracco

Reitz and Bracco

Simplified

tan () =

Reitz and
Bracco, 1979

3
6

Heywood,
1988

4 g

A l

0.5

tan () =

4 g

A l

0.5

Ruiz and Chigier

f ( )

tan () =

0.25

Re
f ( ) l
Wel

Pd 2
= 0.025 g 2 o
g

0.25

Arai

Hiroyasu
and Arai

0.22

l
2 = 83.5 o
do

do

dsac

0.15

Ruiz and
Chigier, 1991
Arai et al.,
1984

g

l

0.26

Hiroyasu and
Arai, 1990

Arrgle

tan () = do0.508 Pinj0.009430.335

g

Arrgle et al.,
1999

Siebers

0.5
0.19

tan () = C g 0.0043 l
l
g

Siebers, 1999

where g and l are the gas and liquid density, respectively. A

is a constant depending on the nozzle design and may be
extracted from experiments or approximated by Equation (2):
l
A = 3.0 + 0.28 o
do

(2)

where do is the nozzle diameter and lo the length of the

nozzle hole. f () is a weak function of the physical properties
of the liquid and the injection velocity:
f ( ) =

3
10
1 e( )
6

Re
= l l
Wel g

Wel =

l
lVinj2 do

Vinj = Cv

l
2P
l

Heywood (1988) simplified Equation (3) into Equation (1)

for cases with high-pressure sprays because the liquid warms
and his viscosity decreases. In these cases, f() becomes
asymptotically equal to 31/2/6 because of the rise of . These
operating conditions can be found in most of modern Diesel
injectors:
4 g

A l

0.5

tan () =

3
6

(8)

1.1.3 Ruiz and Chigier Model

Ruiz and Chigier (1991) examined the previous aerodynamic

break-up model and suggested a modification based on injection parameters atypical of modern Diesel engines:
4 g
tan () =

A l

0.5

0.25

Re
f ( ) l
Wel

(9)

1.1.4 Arai Model

Arai et al. (1984) gave the following equation in order to

determine the spray angle:
Pd 2
= 0.025 g 2 o
g

0.25

(10)

where g is the gas density, P the difference between fuel

injector pressure and ambient gas pressure, do the nozzle
diameter, and g the gas dynamic viscosity.
1.1.5 Hiroyasu and Arai Model

(4)

where Rel and Wel are the Reynolds number and the Weber
number respectively, based on liquid properties and nozzle
diameter:
lVinj do

1.1.2 Reitz and Bracco Simplified Model

(3)

Rel =

where Vinj is the injection velocity, Cv the velocity coefficient,

l the liquid dynamic viscosity, l the surface tension of
liquid and P the difference between injection pressure and
ambient pressure.

(5)

(6)

(7)

Hiroyasu and Arai (1990) have proposed an empirical equation for the spray angle which includes some characteristics
of the nozzle:
0.22

l
2 = 83.5 o
do

do

dsac

0.15

g

l

0.26

(11)

where lo is the length of the nozzle, do the diameter of the

nozzle, and dsac the diameter of the sack chamber. All these
parameters are intrinsic characteristics of injector used.
1.1.6 Arrgle Model

Arrgle et al. (1999) used a simple equation with different

coefficients which have been fitted with experimental results
on a Diesel spray. This equation depends only of the nozzle

ogst100079_DosSantos

17/11/11

804

11:16

Page 804

Oil & Gas Science and Technology Rev. IFP Energies nouvelles, Vol. 66 (2011), No. 5

diameter, the injection pressure and the gas density. Equation

(12) is the result:
tan () = d

0.508 0.00943 0.335

o
inj
g

Spray angle vs injection pressure

do = 0.241 mm, g = 30 kg/m3, Tg = 450 K

20

(12)
18

where Pinj is the injection pressure. We can see that the spray
angle does not depend on injection pressure because the
exponent is very small. This is in agreement with some other
models.

Siebers (1999) has suggested a model in which the spray

angle depends on the liquid and gas densities, and a constant:
0.5
0.19

g
l

tan () = C 0.0043
(13)
l
g

where C is a coefficient depending on the injector characteristics.

1.1.8 Analysis and Validation

In order to compare the previous models, experimental data

with high-pressure injectors has been used (from Sandia
(2010)s online database). The 0D results are compared to
the measured experimental data. The fuel is the Phillips
research grade DF2 fuel (l = 699 kg/m3 at Tl = 450 K).
Experimental and simulated values are considered for three
different nozzle diameters, whose characteristics are shown
in Table 3.

14
Spray angle ()

1.1.7 Siebers Model

16

12
10
8
6
4
2
60

80

100
120
140
Liquid injection pressure (Pa)

160

Experiment

Arai

Reitz and Bracco

Hiroyasu and Arai

Reitz and Bracco simplified

Arrgle

Ruitz and Chigier

Siebers

Figure 2
Evolution of spray angle with the injection pressure.

TABLE 3
Characteristics of nozzles
Orifice
diameter

Discharge
coefficient

Area-contraction Velocity
coefficient
coefficient

Length-todiameter

do (mm)

Cd

Ca

Cv

lo/do

0.185

0.64

0.93

0.68

5.4

0.241

0.71

0.92

0.77

4.2

0.33

0.66

0.89

0.74

3.0

The ambient temperature Tg is set to 300 K for diameter

0.330 mm and 450 K for diameters 0.241 and 0.185 mm
(non-vaporizing experiments). The percentage of oxygen is
set to 0%, in order to disable combustion.
Experimental data show that the injection pressure doesnt
seem to influence the spray angle value (Fig. 2). This has
been observed as well by Arrgle et al. (1999), resulting in a
tiny injection pressure exponent in equation 12 (0.00943,
while exponent for nozzle diameter is 0.508 and exponent for
ambient density is 0.335, proving that the injection pressure
does not really affect spray angle). Considering that, many
spray models dont take into account the injection pressure in
their equations.

Some other models show an evolution in Figure 2 because

the injection pressure is indirectly included in their equation.
For example in Arai model with the term P and in Ruiz and
Chigier and Reitz and Bracco models with Reynolds and
Weber numbers.
As discussed in (Siebers, 1999), Appendix A, the only
parameter that seems to have a clear influence on spray
angle, apart from the injector orifice characteristics, is the
ambient gas to fuel density ratio. Thus, for a given fuel, the
parameter of interest is the ambient density. Experimental
data (Fig. 3) show a rise of the spray angle when the ambient
density increases.
It can be noted that Reitz and Bracco, and Reitz and
Bracco Simplified models are similar in the cases with low
ambient densities but not after. In fact when the ambient density increases, the first term (Rel /Wel)2 in the gamma function
rises a bit because of the change of the injection velocity, but
gamma value is much more influenced (in our case) by the
second term l /g, which reduces faster. Consequently,

ogst100079_DosSantos

17/11/11

11:16

Page 805

F Dos Santos and L Le Moyne / Spray Atomization Models in Engine Applications, from Correlations to Direct Numerical Simulations 805

Measured and predicted spray angle

Spray angle vs gas density

do = 0.33 mm, Pinj = 140 Mpa, Tg = 300 K

45

40

40

Reitz and Bracco

simplified
R2 = 90.47%

Predicted spray angle ()

30

30
Spray angle ()

Reitz and Bracco

R2 = 92.53%

35

35

25
20
15

Arai R2 = 94.17%

15

100
Gas density (kg/m3)

150

200

Experiment

Arai

Reitz and Bracco

Hiroyasu and Arai

Reitz and Bracco simplified

Arrgle

Ruitz and Chigier

Siebers

Arrgle
R2 = 90.74%
Siebers
R2 = 56.47%

50

Hiroyasu and Arai

R2 = 87.41%

20

10

Ruitz and Chigier

R2 = 92.12%

25

10

Experiment

10
Measured spray angle ()

Figure 4
Comparison between measured and predicted spray angle.

Figure 3
Evolution of the spray angle with the ambient density.

gamma decreases and f() is no longer equal to 31/2/6.

Another variable which can change considerably the value
and the evolution of f() is the fuel temperature. When it
increases, the fuel viscosity l decreases a lot, and changes
the Reynolds number. The term (Rel /Wel)2 increases and
becomes more significant than l /g. In this case (high
fuel temperature), both models should be equal (because
f() 31/2/6).
In Figure 4 are represented 27 points with at least one
different characteristic between each other (ambient density,
injection pressure, nozzle diameter...). The coefficient of
determination R2 shows a good correlation between models
and experimental data. Concerning Siebers model, our values
chosen for the constant C dont seem to be adapted to the
injectors used for this study, but the evolution on spray angle
for a selected injector is well represented (Fig. 3).
The coefficient of determination and the mean absolute
percentage error for each model are given in Table 4 (the best
values are in bold). In our case, Hiroyasu and Arai gives a

20

TABLE 4
Comparison results between models and data
Spray angle
model

Coefficient of
determination

Mean absolute
percentage error

Reitz and Bracco

92.53%

23.91%

Reitz and Bracco simplified

90.47%

38.21%

Ruiz and Chigier

92.12%

87.85%

Arai

94.17%

106.57%

Hiroyasu and Arai

87.41%

7.15%

Arrgle

90.74%

67.69%

Siebers

56.47%

24.27%

spray angle value closed to the measured one but its prediction on the evolution of the spray angle is not really good. On
the contrary, Arai model has a good correlation with the
experimental data, but must be calibrated in order to give
values close to the measured spray angle.
1.2 Spray Tip Penetration Models
The knowledge of spray tip penetration is important in design
of Diesel and Gasoline Direct Injection (GDI) engines. The

ogst100079_DosSantos

17/11/11

806

11:16

Page 806

Oil & Gas Science and Technology Rev. IFP Energies nouvelles, Vol. 66 (2011), No. 5

This model is different from other models for considering the

ambient temperature Tg effects by the term (294/Tg):

TABLE 5
Different correlations for spray tip penetration
Model

Correlation

References

P
S = 1.189Ca0.25
g

0.25

Wakuri

P
S = 3.07
g

do t

tan()

Wakuri
et al., 1960

294

Tg

Dent, 1971

0.5

0.25

Dent

2P
S = 0.39
t
l

0.5

0.5

Hiroyasu

P
S = 2.95
g

0 < t < tb
Hiroyasu and
Arai, 1990

0.25

( do t )

P
S = 1.414Cv0.5
g

0.25

Schihl

0.5

t > tb

do t

tan()

0.5

0.25

Schihl
et al., 1996

Naber and
Siebers

1
1 n 1 n n
S = + 0.5
t t

Naber and
Siebers, 1996

Arrgle

S = do0.307 Pinj0.262 g0.406 t 0.568

Arregle
et al., 1999

( do t )

spray penetration is a function of several engine operating

parameters such as nozzle geometry, spray angle, injection
conditions including the injection pressure, ambient density
and temperature. The collected 0D spray tip penetration models are shown in Table 5.
1.2.1 Wakuri Model

Wakuri et al. (1960) used momentum theory to develop the

fuel spray model by assuming that the relative velocity
between fuel droplets and entrained air can be neglected and
the injected liquid droplet momentum is transferred to the
homogeneous fuel droplet-entrained air mixture for the density ratio range of 40-60. Their model is expressed by
Equation (14):
P

g

0.25

S = 1.189C

do t

tan()

(15)

Hiroyasu and Arai (1990) proposed Equations (16) and (17),

which are derived from data obtained during the investigation (Hiroyasu et al., 1978) and from applying the jet disintegration theory by Levich (1962). They used two different
equations, one from the beginning of the injection to the jet
break-up time, where the penetration is proportional to time.
The other occurs at times exceeding the jet break-up time,
where the penetration is proportional to the square root of
time:
0.5
2P
(16)
S = 0.39
t 0 < t < tb
l
P
S = 2.95
g

0.25

0.5

(14)

where Ca is the coefficient of contraction (Cd = 1 in this

model), P the difference between fuel injector pressure and
ambient gas pressure, g the gas density, do the nozzle diameter and the spray angle.
1.2.2 Dent Model

Equation (15) of spray penetration is the jet mixing model,

based on gas jet mixing theory, proposed by Dent (1971).

( do t )

0.5

t > tb

(17)

where tb is the jet break-up time, which can be evaluated by

Equation (18):
tb = 28.65

0.25
a

294

Tg

0.25

0.5

1.2.3 Hiroyasu Model

0.25

( do t )

P
S = 3.07
g

l d o

( P )

(18)

0.5

1.2.4 Schihl Model

Schihl et al. (1996) analyzed the existing spray penetration

models and proposed the following phenomenological cone
penetration model:
P

g

0.25

S = 1.414C

0.5
v

do t

tan()

0.5

(19)

1.2.5 Naber and Siebers Model

Equation (20) is a correlation given in Appendix C of Naber

and Siebers (1996):
1
1 n 1 n n
(20)
S = + 0.5
t t
~

where S is the dimensionless penetration, t~ the dimensionless

time and n a model constant (n = 2.2):
S
S = +
x

(21)

t
t = +
t

(22)

ogst100079_DosSantos

17/11/11

11:16

Page 807

F Dos Santos and L Le Moyne / Spray Atomization Models in Engine Applications, from Correlations to Direct Numerical Simulations 807

where t+ and x+ are the time scale and length scale respectively, they are defined by Equations (23) and (24):

x+ =

(23)

tan ()
Ca0.5 do

t+ =

0.12

l
g

0.10

l
g

(24)

tan () Vinj

1.2.6 Arrgle Model

Arrgle et al. (1999) used a simple equation with different

coefficients which have been fitted with experimental results
on a Diesel spray. This equation depends only on the nozzle
diameter, the injection pressure, the gas density and the time.
This is the result:
S = do0.307 Pinj0.262 g0.406 t 0.568

0.08

0.06

Experiment

0.04

Wakuri
R2 = 95.79%
Dent
R2 = 92.6%
Hiroyasu
R2 = 96.96%
Schihl
R2 = 96.75%

0.02

Naber and
Siebers
R2 = 96.6%
Arrgle
R2 = 94.45%

(25)
0

where Pinj is the injection pressure. We can observe that the

spray penetration is proportional to a value close to the
square root of the time, as in the other models.

0.05
0.10
Measured spray penetration (m)

Figure 6
Comparison between measured and predicted spray tip
penetration.

1.2.7 Analysis and Validation

The data come from the same experiment that previously for
the spray angle. The start of injection is defined as the first
mass leaving the nozzle.

Comparaison between experiment and models

do = 0.241 mm, Pinj = 140 Mpa, Tg = 450 K, g = 14.8 kg/m3
0.10
Spray tip penetration (m)

Predicted spray penetration (m)

Ca0.5 do

Measured and predicted spray tip penetration

0.08

Figure 5 shows predicted and measured spray tip penetrations for one example. The shape of penetration is similar for
all the models because they are all proportional to the square
root of time. Concerning the break-up time tb used by
Hiroyasu, it has an influence on the first part of the penetration,
until a value comprised between 4.105 and 4.104 seconds in
our case.
In Figure 6 are represented 1 040 points (from 27
penetrations) for each model. All the models seem to be
predictive.
The coefficient of determination and the mean absolute
percentage error for each model are given in Table 6 (the best
TABLE 6

0.06

Experiment

Comparison results between models and data

Wakuri
Dent

0.04

Hiroyasu
Schihl
Naber and
Siebers

0.02

Arrgle

0.2

0.4

0.6
0.8
Time (s)

1.0

Figure 5
Example of 0D results for spray tip penetration.

1.2

x 10-3

Spray angle
model

Coefficient of
determination

Mean absolute
percentage error

Wakuri

95.79%

7.57%

Dent

92.60%

11.14%

Hiroyasu

96.96%

12.61%

Schihl

96.75%

8.77%

Naber and Siebers

96.60%

15.26%

Arrgle

94.45%

17.86%

ogst100079_DosSantos

17/11/11

808

11:16

Page 808

Oil & Gas Science and Technology Rev. IFP Energies nouvelles, Vol. 66 (2011), No. 5

values are in bold). In our case, Wakuri model gives a spray

tip penetration value closed to the measured one. But the best
prediction is given by Hiroyasu model. Also, the additional
term with the ambient temperature, in Dent model equation,
seems to reduce its validity.
1.3 Liquid Length Models
The purpose of this study is to investigate the applicability of
the existing 0D liquid length models for spray. The collected
spray angle models are show in Table 7.
TABLE 7
Different correlations for liquid length
Type of model

Correlation

References

Chehroudi

Lb = Cdo l
g

Beale and
Reitz

Lb = 0.5B1do l
g

Chehroudi
et al., 1985
0.5

Beale and
Reitz, 1999

r P
Lb = 7.0do 1 + 0.4 o g 2
do lVinj

0.05

l
o
do

0.13

Beale and Reitz (1999) proposed a modified equation of the

previous model:

Lb = 0.5B1do l
g

0.5

l
g

0.5

Hiroyasu and
Arai, 1990

(27)

where B1 is the well known breakup constant of Kelvin

Helmholtz model. According with Su et al. (1996), this
constant can be set to 60.
1.3.3 Hiroyasu and Arai Model

The following equation was given by Hiroyasu and Arai

(1990) for the liquid length from experimental data which
covered a wide range of conditions:

r P
Lb = 7.0do 1 + 0.4 o g 2
do lVinj

0.05

0.5

Hiroyasu
and Arai

1.3.2 Beale and Reitz Model

lo

do

0.13

l
g

0.5

(28)

where Lb is the liquid length, ro is the radius of inlet edge of

hole, Pg the ambient gas pressure, Vinj the injection velocity,
lo the length of the nozzle hole and g the gas density.
This equation includes the effect of the inlet edge rounding which shifts the inception of cavitation and turbulence to
higher injection pressures and increases the liquid length. The
influence of cavitation and turbulence is also included via the
2 ).
term (Pg/lVinj
1.3.4 Siebers Model

b l

a g

0.5

Siebers

Lb =

do Ca0.5 2
+1 1
tan() B
2

294

Lb = 93
Tg

1.43

Enhanced
model

do l
g

Siebers, 1999

0.5

Siebers (1999) has developed a scaling law to predict the

liquid length. The assumptions made include quasi-steady
flow with a uniform rate, uniform velocity, uniform fuel concentration and uniform temperature profiles (perfect mixing
inside the spray boundaries), and finally no velocity slip
between the injected fuel and the entrained gas. The model
equation is the following:
b
Lb = l
a g

0.5

1.3.1 Chehroudi Model

The more famous and simple equation for the liquid length
expresses the fact that the core length is dependent on the
ratio of liquid and gas density and that it is directly proportional to the nozzle hole diameter:

Lb = Cdo l
g

0.5

(26)

where do is the nozzle diameter, l the liquid density and g

the gas density. C is an empirical constant which expresses
the influence of the nozzle flow conditions and other effects
that cannot be described in detail. Chehroudi et al. (1985)
performed some experiments and proposed that the constant
C should be in the range 7-16. However, for our experiments,
a value of 16.3 gives the best result.

do Ca0.5 2
+1 1
tan() B
2

(29)

where a and b are two constants (0.66 and 0.41 respectively)

and the spray angle. B is the ratio of the fuel and ambient
gas mass flow rates resulting in complete vaporization of the
fuel:
B=

Z g (Ts ,Pg Ps ) Ps M l
hg (Tg ,Pg ) hg (Ts ,Pg Ps )
=
(30)
hl (Ts ) hl (Tl ,Pg )
Zl (Ts ,Ps ) Pg Ps M g

where Z is the compressibility factor, M the molecular

weights, P the pressure, T the temperature and h the enthalpy.
The subscript g, l and s represents the ambient gas, the vaporized liquid and the saturated liquid vapor condition at the
liquid length respectively.

ogst100079_DosSantos

17/11/11

11:16

Page 809

F Dos Santos and L Le Moyne / Spray Atomization Models in Engine Applications, from Correlations to Direct Numerical Simulations 809

1.3.5 Enhanced Model

Experimental data shows a correlation between liquid length

and ambient temperature (Fig. 8). Previous models dont take
into account this phenomenon, except Sieberss model.
By using Chehroudis model, an additional term can be
added. The resulting enhanced model is expressed in
Equation (31):

294 2
do l
Lb = C1
g
Tg
C

Liquid length vs injection pressure

do = 0.246 mm, g = 14.8 kg/m3, Tg = 700 K
0.05

0.04
Liquid length (m)

The unknown, Ts, can be solved iteratively, given the fuel

and ambient gas properties and initial fuel and ambient gas
conditions. Once determined, Ts defines B, as well as the
pressures, temperatures, and enthalpies of the fuel and ambient gas at the liquid length location.
The term B given by Equation (30) is analogous to the
mass and thermal transfer numbers used in droplet vaporization studies (Lefebvre, 1989). Furthermore, the method of
solving for Ts is analogous to that used in determining the
surface temperature of a vaporizing liquid droplet (Lefebvre,
1989).

Experiment

Hiroyasu and Arai

Chehroudi

Enhanced model

Beale and Reitz

0.03

0.02

0.01

0
60

80

100
120
140
Liquid injection pressure (Pa)

160

Figure 7

0.5

(31)

where Tg is the ambient gas temperature. C1 and C2 are two

constants to determine by a linear regression method. In our
case a value of 93 and 1.43 is chosen, respectively.
1.3.6 Analysis and Validation

The experiments are different to the previous ones; the ambient temperature Tg is increased (700 K to 1 300 K instead of
450 K). The nozzles characteristics are listed in Table 8.
TABLE 8
Characteristics of nozzles
Length-todiameter

Evolution of liquid length with the injection pressure.

All the models include a density ratio in their equations, so

their prediction for a variation of density is roughly the same.
But concerning the injection pressure, only one model
includes it (Hiroyasu and Arai model, indirectly with the
injection velocity Vinj in the cavitation term (Pg/lV2inj)).
Figure 7 shows that the liquid length decreases when the
injection pressure rises, and Hiroyasu and Arai model predicts very well this phenomenon.
Concerning the ambient temperature, its effect is to
decrease the liquid length when it rises. The phenomenon,
shown in Figure 8, is well predicted by the enhanced model.
The coefficient of determination and the mean absolute
percentage error for each model are given in Table 9 (the best
values are in bold). The coefficient of determination is very
low for the first models because the ambient temperature is
not taken into account. The enhanced model is clearly more
predictive with its additional term.

Orifice
diameter

Discharge
coefficient

do (mm)

Cd

Ca for 72 Mpa

Ca for 138 Mpa

lo/do

0.100

0.80

0.91

0.86

4.0

0.180

0.77

0.85

0.82

4.2

0.251

0.79

0.88

0.79

2.2

0.246

0.78

0.89

0.81

4.2

TABLE 9

0.267

0.77

0.89

0.82

8.0

Comparison results between models and data

0.363

0.81

0.85

4.1

0.498

0.84

0.94

0.88

4.3

Area-contraction coefficient

180

The fuel is 2,2,4,4,6,8,8 heptamethylnonane (C16H34),

with a density l = 689 kg/m3 for a temperature Tl = 436 K.
The percentage of oxygen is set to 0%, in order to disable
combustion.

Spray angle
model

Coefficient of
determination

Mean absolute
percentage error

Chehroudi

54.78%

23.30%

Beale and Reitz

54.78%

70.41%

Hiroyasu and Arai

51.33%

61.51%

Enhanced model

98.06%

6.07%

ogst100079_DosSantos

17/11/11

810

11:16

Page 810

Oil & Gas Science and Technology Rev. IFP Energies nouvelles, Vol. 66 (2011), No. 5

0.09

TABLE 10

Liquid length vs ambient temperature

do = 0.246 mm, g = 7.3 kg/m3, Pinj = 136.27 Mpa

Correlation of SMD for different models

0.08

Type
of model

0.07

Hiroyasu

Correlation

References
Hiroyasu
et al., 1989

x32 = 2.33 10 3 P 0.135 0.121

ml0.131
g

Liquid length (m)

Experiment

0.06

Chehroudi

x32 = 8.7do ( Rel Wel )

Varde

Varde et al.,
1984

0.28

Beale and Reitz

0.05

x32 = max x

Hiroyasu and Arai

Enhanced model

32 LS

,x

32 HS

= 4.12do Rel0.12Wel0.75 l
g

l
g

= 0.38do Rel0.25 Wel0.32 l
g

l
g

0.54

0.04

Hiroyasu
and Arai

0.03

0.37

x
0.02
0.01
400

32 LS

600

800
1 000
1 200
Ambient temperature (K)

1 400

Merrington
and
Richardson
Elkotb

Figure 8

32 HS

x32 =

0.18

0.47

500do1.2 l0.2
Vinj

Hiroyasu and
Arai 1990

Merrington
and Richardson,
1947

0.54
x32 = 3.085l0.385 l0.737 l0.737 0.06
g Pl

Elkotb, 1982

Evolution of liquid length with the ambient temperature.

1.4.1 Hiroyasu Model

Experiments used ambient conditions similar to the ones in

an engine combustion chamber, and the injector is also comparable to the ones used in car engines. So our values of the
constant C1 and C2 should be a good start in order to model
the liquid length in an internal combustion engine.
1.4 Global Sauter Mean Diameter Models
The global Sauter Mean Diameter (SMD or x32) characterizes a single droplet with the same volume to surface area
ratio as the ratio of the respective quantities integrated over
the whole droplet size distribution present in a real spray:

x32 =

3
i

i=1
N drops

(32)
d

x32 = 2.33 10 3 P 0.1350.121

ml0.131
g

2
i

(33)

where P is the difference between fuel injector pressure and

ambient gas, g the gas density and ml is the amount of fuel
delivery.
1.4.2 Varde Model

Varde et al. (1984) related the SMD to the diameter of the

injection nozzle in the following equation:
x32 = 8.7do ( RelWel )

N drops

The following equation is an empirical expression for the

SMD for typical Diesel fuel properties and for hole type
nozzle given by Hiroyasu et al. (1989):

0.28

(34)

where do is the nozzle diameter, Rel the Reynolds number

and Wel the Weber number.

i=1

1.4.3 Hiroyasu and Arai Model

The SMD is a quantity characterizing the average droplet

size of a spray. Many studies and correlations describe the
SMD depending on the distance to the nozzle or on the
breaking type or phase (Wu and Faeth, 1995; Faeth, 1996;
Lee et al., 2007).
Some global SMD models are listed in Table 10, but the
available data, mostly obtained by time-resolved point
measurements is not pertinent for comparison.

Hiroyasu and Arai (1990) studied more in detail the effects of

various parameters on the SMD like ambient pressure, injection pressure or the nozzle characteristics. Dimensionless
analysis leaded to the following experimental equations for a
complete spray:
x32 = max x

32 LS

,x

32 HS

(35)

ogst100079_DosSantos

17/11/11

11:16

Page 811

F Dos Santos and L Le Moyne / Spray Atomization Models in Engine Applications, from Correlations to Direct Numerical Simulations 811

l
g

l
g

= 0.38do Rel0.25Wel0.32 l
g

l
g

0.18

0.54

32 LS

0.12
l

= 4.12do Re

0.75
l

We

0.37

32 HS

(36)

0.47

(37)

where l and g are the dynamic viscosity of the liquid and

the gas respectively.
1.4.4 Merrington and Richardson Model

500do1.2 l0.2
Vinj

(38)

where l is the kinematic viscosity of the liquid and Ving the

injection velocity.
1.4.5 Elkotb Model

Elkotb (1982) has developed a model that takes into account

numerous properties of both liquid and gas. In the end, experimental formula provides:
0.54
x32 = 3.085l0.385 l0.737l0.7370.06
g Pl

2 MULTIDIMENSIONAL MODEL: RANS AND LES

In this part are listed the actual multidimensional models and
methods used to model sprays, some results are shown for
the most popular methods.
2.1 RANS; Liquid is Pure Lagrangian

The disintegration of liquid jets injected into stagnant air was

studied empirically (Merrington and Richardson, 1947).
Drop-size distribution curves were obtained by a stationary
nozzle directed vertically downward in an enclosed spray
tower. The pressure supply was held constant during each
test run by using a pressure-regulated gas cylinder.
(Lefebvre, 1989) gives a revision of the Merrington and
Richardson SMD correlation including the effect of orifice
diameter on the resulting SMD. The correlation for SMD is
provided in Equation (38):
x32 =

in a combustion chamber, multidimensional models should

are needed.

2.1.1 Primary Break-Up

The task of a primary break-up model is to determine the

starting conditions of drops which penetrate into the chamber
(initial drop size and its velocity components). These conditions are principally influenced by the flow conditions inside
the nozzle.
Blob Method
The blob method, developed by Reitz and Diwakar (1987), is
the most popular of the models because of its simplicity. It is
assumed that a detailed description of the atomization and
breakup processes within the primary breakup zone of the
spray is not required. This method creates big spherical
droplets (Fig. 9) with the same diameter (usually equal to the
nozzle hole diameter), which are then subject to secondary
break-up. The number of drops injected per unit time is
determined from the mass flow rate and the conservation of
mass gives the injection velocity of the blobs.

(39)

where P is measured in bar. SMD value is given in m.

Uinj

1.5 Conclusion on 0D Model

We have seen in this part some spray characteristics of sprays
like tip penetration or spray angle. Several 0D models have
been considered and analyzed, and the results of the comparisons with experimental data show that these characteristics
can be accurately predicted, and thus the spray correctly
described.
Despite not being as exact as multidimensional models,
0D models offer important advantages like their reduced simulation time as well as their low computational requirements.
For that reason, these models can be useful when a large
number of simulations is needed.
But like there is not a spatial discretization in these models,
the interaction between spray and the flow or the walls can
not be modeled. Also the ambient conditions are fixed
(volume, temperature...). In order to describe a spray evolving

Nozzle hole

Blob
Secondary break-up

Figure 9
Blob method.

Kuensberg Sarre et al. (1999) suggested an enhanced

version of the blob method. This method allows calculating
an effective injection velocity and an effective injection particle diameter taking into account the reduction of the nozzle
cross section due to cavitation (by decreasing the initial blob
size and estimating a more realistic initial velocity).

ogst100079_DosSantos

812

17/11/11

11:16

Page 812

Oil & Gas Science and Technology Rev. IFP Energies nouvelles, Vol. 66 (2011), No. 5

Altogether, the blob methods are a good way to define the

initial starting conditions for the liquid entering the chamber.
But these methods do not represent a detailed physical and
satisfying modeling of the relevant processes during primary
break-up.
Distribution Functions
This method assumes that the fuel is already fully atomized
at the nozzle exit and that the distribution of drop sizes can be
described by mathematical functions. In this case, a distribution of droplet sizes is injected.
According to Levy et al. (1998), the droplets diameter D
at the nozzle exit should be sampled from a X2-law in order
to get a good agreement between measured and simulated
downstream drop size distributions:
1
P ( D) =
D 3 e DD
6D 4

(40)

where D = SMD/6. Various authors have used distribution

functions in Diesel spray modeling like Long et al. (1994)
and Lefebvre (1989).
It is also possible to predict the distribution function by
using the Maximum Entropy Formalism (MEF) instead of a
mathematical distribution function. Due to the use of a physically based criterion, the entropy of the distribution given by
Shannon (1948), it is possible to estimate the most probable
distribution function. Cousin and Desjonqures (2003) have
used this method in order to predict drop size distributions in
sprays from pressure-swirl atomizers.
Kelvin-Helmholtz Break-Up Model (KH Model)
The Kelvin-Helmholtz model (KH model or WAVE, Fig. 10)
has been proposed by Reitz (1987). The model is based on a
first order linear analysis of a Kelvin-Helmholtz instability
growing on the surface of a cylindrical liquid jet that is penetrating into a stationary, inviscid and incompressible gas with
a relative velocity.

2r

Figure 10
Illustration of the KH model.

2.B0.

Urel

From the solution of a general dispersion equation, maximum

growth rate KH and corresponding wave length KH are
given by the following equations:
r 3
KH l 0

0.5

0.34 + 0.38We1.5
g

(1 + Z ) (1 +1.4T 0.6 )

(41)

(1 + 0.45Z 0.5 )(1 + 0.4T 0.7 )

KH
= 9.02
0.6
r0
(1 + 0.87We1g.67 )

(42)

rnew = 0.61 KH

(43)

where r0 is the initial radius of the droplet, Z = Wel /Rel and

T = Z Weg. The new radius of droplet rnew is given by:

The new droplet continuously looses mass while penetration into the gas, its radius r is expressed in Equation (44):
r =

r rnew
bu

bu = 3.788B1

r
KH KH

(44)
(45)

where bu is the characteristic time span.

Turbulence-Induced Break-Up
Huh and Gosman (1991) have published a model of turbulence-induced atomization for full-cone Diesel sprays. They
assume that the turbulent forces within the liquid emerging
from the nozzle are the producers of initial surface perturbations, which grow exponentially due to aerodynamic forces
and form new droplets.
The droplets break up with a characteristic atomization
length scale LA and time scale A. The characteristic atomization length is proportional to the turbulent length scale Lt:
LA = C1Lt = C2Lw

(46)

where L w is the wavelength of surface instability, determined

by turbulence. According with the authors, C1= 2 and
C2= 0.5. The characteristic atomization time scale A can be
calculated under the assumption that the time scale of atomization is a linear combination of the turbulence time scale t
and the wave growth time scale w:
A = C3 t +C4 w = spn + exp

(47)

where C3 = 1.2 and C4 = 0.5. spn and exp indicate spontaneous wave growth time and exponential growth time.
This model shows good agreement with available
experimental data for the spray cone angle of steady-flow
single-hole experiments, but the effects of cavitation are not
included.

ogst100079_DosSantos

17/11/11

11:16

Page 813

F Dos Santos and L Le Moyne / Spray Atomization Models in Engine Applications, from Correlations to Direct Numerical Simulations 813

Cavitation-Induced Break-Up
Arcoumanis and Gavaises (1997) have presented a primary
break-up model that takes into account cavitation, turbulence,
and aerodynamic effects. The initial droplet diameter is set
equal to the effective hole diameter (blob method), and the
first break-up of these blobs is modeled using the KelvinHelmholtz mechanism in the case of aerodynamic-induced
break-up, the previous model of Huh and Gosman (1991) for
turbulence-induced break-up, and a new phenomenological
model in the case of cavitation-induced break-up.
The cavitation bubbles are transported to the blob surface
by the turbulent velocity inside the liquid and either burst on
the surface or collapse before reaching it, depending of the
characteristic time scale. The radius of the bubbles is given
by the following equations:
rcav = r02 reff2
reff =

Aeff

(48)

where rcav is the radius of bubble, reff the effective radius and
Aeff is the effective area. The characteristic time scales for
collapse coll and bursting of bubbles burs are:

burst =

l
g,bubble

r0 rcav
uturb

(50)

(51)

where uturb is the turbulent velocity (uturb = 2k /3, k is the

turbulent kinetic energy). The smaller characteristic time
scale causes the break-up.

Cylindrical shape
parent parcel
(contain one droplet)

Fturb = surface dynamic pressure = D

Pressure gradient force to side

surface of parent parcel

Nozzle
hole

Faero =

2do2
2
0.5g urel
3

Fsurf = D

Surface vibrates by
turbulent energy

Figure 11
Primary break-up model of Nishimura and Assanis (2000).

(52)

(53)
(54)

The diameter of the original cylinder is reduced until

Fturb + Faero = Fsurf again.
2.1.2 Secondary Break-Up

Secondary break-up is the disintegration of already existing

droplets into smaller ones due to the aerodynamic forces.
These forces are induced by the relative velocity between
droplet and surrounding gas. The result is an unstable growth
of waves on the droplet surface which finally leads to the disintegration of the droplet into new ones.

Child
parcel

No cavitation and no
turbulence in child parcel

Child droplet size determined by wave length,

and all turbulence energy in child parcel
is transformed to tangential velocity

Parent parcel continues

to generate child parcel until
bubbles collapse totally
Cavitation
bubble

2D l 2
uturb
3 2

It breaks up if the sum of Fturb and the aerodynamic drag

force Faero is no longer compensated by the surface tension
Fsurf :

(49)

coll = 0.9145rcav

Cavitation and Turbulence-Induced Break-Up

Nishimura and Assanis (2000) have proposed a cavitation
and turbulence-induced primary break-up model for full-cone
Diesel sprays. During the injection period, discrete fuel
parcels enter in the chamber with an initial diameter equal to
the nozzle hole diameter. Each parcel contains bubbles (see
Fig. 11), according to the volume fraction and size distribution
at the hole exit, computed from a phenomenological cavitation
model inside the injector.
The authors assume that the velocity fluctuations inside
the cylinder induce a deformation force on its surface:

When breakup force becomes bigger

than surface tension, parent parcel emits
excess volume as one child parcel

ogst100079_DosSantos

17/11/11

814

11:16

Page 814

Oil & Gas Science and Technology Rev. IFP Energies nouvelles, Vol. 66 (2011), No. 5

Taylor-Analogy Break-Up Model (TAB Model)

The Taylor Analogy Break-up model was proposed by
ORourke and Amsden (1987). It is based on an analogy
between a forced oscillating spring-mass system and an
oscillating and distorting droplet (see Tab. 11).

where K is the liquid-gas density ratio and N is the liquid-gas

viscosity ratio. The droplets start with zero deformation
(y = 4/3). The break-up happens when the deformation of
droplet reaches the value given by:
a Weg
=
r
6

TABLE 11
Comparison of a spring-mass system to a distorting droplet
Spring-mass system

Distorting and oscillating droplet

Restoring force of spring

Surface tension forces

External force

Droplet drag force

Damping force

Droplet viscosity forces

The deformation of the droplet is given by:


y=

CF g urel
Ck
y Cd
y
2
3
l r
l r 2
Cb l r

(55)

where Cf = 1/3, Cb = 0.5, Ck = 8 and Cd = 5. urel is the relative

velocity between droplet and gas and is the surface tension.
Tanner (1997) showed that the TAB model under predicts
the droplets sizes of full cone sprays and has developed an
enhanced version of this model (ETAB). The difference is
that an initial oscillation is chosen for the spherical drops
emerging from the nozzle (against no oscillation for TAB
model).
Droplet Deformation and Break-Up Model (DDB Model)
Ibrahim et al. (1993) presented a droplet deformation and
break-up model. It assumes that the liquid droplet is
deformed from the initial spherical shape into an ellipsoidal
one (see Fig. 12).
The distortion of droplet that causes break-up is governed
by the following equation:
6
3 3
4N 1
27 2
(56)

Ky +
y
+
y

y
1
2
=

4 8
Reg y 2
16Weg

where a is the major semi-axis of the ellipsoidal cross section

of the oblate spheroid.
The DDB and TAB model have a very similar breakup
mechanism, so the DDB model can be regarded as an alternative to the TAB model.
Rayleigh-Taylor Break-Up Model (RT Model)
The Rayleigh-Taylor model (RT model, Fig. 13) is based on
the theoretical work of Taylor (1950), who investigated the
instability of the interface between two fluids of different
densities in the case of an acceleration (or deceleration) normal to this interface.
From the assumption of linearized disturbance growth
rates and negligible viscosity, the frequency of the fast-growing RT wave is given by:
3/ 2

2 a (l g )
=

l + g
3 3

0.5

RT

RT =

2C RT
K RT

(59)

a (l g )

K RT =
3

0.5

(60)

where a is the acceleration of the droplet and CRT is a

constant (CRT = 0.3).

Front
a

Urel

Urel

(58)

The corresponding wavelength RT and wave number KRT

are given by:

Back

(57)

b
Figure 12

Figure 13

Schematic diagram of the deforming half drop.

Illustration of the RT model.

ogst100079_DosSantos

17/11/11

11:16

Page 815

F Dos Santos and L Le Moyne / Spray Atomization Models in Engine Applications, from Correlations to Direct Numerical Simulations 815

Figure 14
Lagrangian particles (color scale for diameter in meters) and Eulerian flow (color scale for velocities in m/s).

Comparaison between experiment and models

do = 0.241 mm, Pinj = 140 Mpa, Tg = 450 K, g = 58.5 kg/m3
0.09

Spray tip penetration (m)

0.08
0.07
0.06
0.05
0.04
0.03
Experiment

0.02

Wakuri 0D

0.01
0
0

is necessary to use combined models (usually a primary and

a secondary break-up) in order to improve the accuracy of
prediction.
Patterson and Reitz (1998) suggested the KH-RT hybrid
breakup model with the concept that the competition between
KH instability and RT instability causes the droplet breakup.
Park et al. (2003) investigated the prediction accuracy of
various hybrid models for high-speed Diesel fuel sprays and
reported that the results of KH-RT model, KH-DDB model,
and Turbulence-DDB model agree well with the experimental
results.
2.1.3 Calculation

KH-RT 3D

0.5

1.0

1.5
Time (s)

2.0

2.5

3.0
x 10-3

Figure 15
Penetration results for KH-RT Lagrangian particles in
Eulerian flow.

If the corresponding wavelength is smaller than the

droplet diameter, the instability of the RT wave increases.
While the wave is growing, the wave growth time is tracked,
and then compared with the break-up time. In RT break-up,
the break-up time RT and the radius after break-up rnew are
defined by the following equations:
CKH
rnew =
(61)
K RT
RT =

C
RT

(62)

Combined Models
A single break-up model is usually not able to describe all
break-up processes and break-up regimes of engine sprays. It

Figure 14 shows an example of utilization of the KH-RT

model in order to model a Diesel high pressure spray. The
software used is OpenFOAM1.5-dev which includes natively
the KH-RT model. The Lagrangian particles are injected in
an Eulerian RANS flow using the k- turbulence model.
Figure 15 is the penetration curve from 3D model compared to correlation and experimental data, showing a good
agreement.
2.2 RANS; Liquid is Pure Eulerian
Wan and Peters (1999) developed an Eulerian approach from
droplet equations, integrating over the radial direction
to model Diesel sprays with a 1D model. Their ICAS
(Interactive Cross sectional Averaged Spray) model is effective in simulating the vicinity of the nozzle up to the zone
where the gas velocity becomes the predominant mixing
mechanism. Vallet et al. (2001) and Beheshti and Burluka
(2004) proposed a fully Eulerian spray atomization model by
generalizing Kolmogorov hypothesis on turbulence to characteristic scales of spray. The principles of such modeling are
currently used for the development of Eulerian-Lagrange
coupled codes for spray simulation. The compressible Favreaveraged flow equations are applied to a single fluid with
variable properties. The mass and momentum conservation

ogst100079_DosSantos

816

17/11/11

11:16

Page 816

Oil & Gas Science and Technology Rev. IFP Energies nouvelles, Vol. 66 (2011), No. 5

equations consider the contributions from one phase on

another. The model is completed by k- equations for the
two-phase turbulent kinetic energy (which includes kinetic
energy in gas, liquid and the mean velocity gradient of the
two phases) and dissipation rate, and then closed by an
Algebraic Stress Model (ASM) of the Reynolds stress tensor.
Specific treatment of atomizing liquid consists of conservation equation of liquid mean mass fraction:
Y uiY
ui ' Y '
+
=
t
xi
xi
with:

1 Y 1 Y
= +
l
g

and a state law:

1 Y rgTg
p=
Y
1
l

(63)

way, a critical scale of atomization where inertial forces

balance surface tension:
Wec =

c rc uc2
1

(67)

On the condition that balance between different processes of

droplet formation of radius rc (coalescence, break-up) and
turbulent diffusion is obtained. Such a model includes some
constants that can be tuned for solution. A correct calibration
of the model can therefore lead to good simulation efficiency
but the physical mechanisms of atomization are not explicitly
represented resulting in the potential need for calibration of
constants for each different case and hence limited generality.

(64)
2.3 RANS; Liquid is Eulerian-Lagrangian Represented

(65)

along with a transport equation of mean interfacial area per

unit volume:

Ds

ui
xi
(66)
+
=
+ ( A + a ) Va 2
t xi
xi

(= diffusion + production destruction)

Diffusion term is written classically with a diffusion
coefficient Ds that depends on turbulence and liquid fraction gradient. The production consists of two mechanisms,
first is the action of velocity gradients that, according to a
Kelvin-Helmholtz type instability, will stretch liquid-gas
interface and make the surface larger. The authors choice
is to give it a characteristic time A1 equivalent to the one of
turbulent kinetic energy production. The second is the
action of turbulence which can also stretch and augment
liquid-gas interface, with a characteristic time a1 equivalent to the integral scale of turbulence. Last, the destruction
term allows compensating production in such a way that at
small scales, inertial forces be of the same order as surface
tension: We = 1. The destruction term exponent being
greater than 1, value of 2 is given to include the possibility
of having two interfaces interacting inside a given volume.
In this approach, oriented towards large Reynolds and
Weber number flows with a main velocity direction (hole
injectors of Diesel type), surface tension, and viscosity are
only effective at small scales, and where curvature and
velocity gradients are important. At large scales, the characteristics of the flow are dependent only on fluid density. It
is the generalization to atomization of Kolmogorovs results
for turbulence. Also, it is possible to define in the same

The Eulerian multiphase method can also be used for modeling the liquid in the near nozzle dense spray region combined
to the Lagrangian method for zones where the spray is sufficiently diluted (far away from the nozzle). This method is
called ELSA (Eulerian-Lagrangian Spray Atomization) and
has been used by Demoulin et al. (2007) and Lebas et al.
(2009).
2.4 LES; Liquid Fuel is Pure Lagrangian
Bharadwaj et al. (2009) have modeled an non-evaporative
Diesel spray using the Lagrangian method with large eddy
simulation. Their results show a good agreement. It was
shown that a high speed Diesel spray can create significant
energy at the sub-grid scale in the near nozzle region. This
sub-grid kinetic energy is important in the models of sub-grid
shear stress and droplet turbulent dispersion.
2.5 LES; Liquid Fuel is Pure Eulerian
Some studies have been to model a spray in full Eulerian in
LES, like De Villiers et al. (2004). Their approach combines
multiphase Volume-Of-Fluid (VOF) and large eddy simulation methodologies. It is used to perform quasi-direct transient fully three dimensional calculations of the atomization
of a high-pressure Diesel jet, providing detailed information
on the processes and structures in the near nozzle region,
which is difficult to obtain by experimentation.
This methodology allows separate examination of diverse
influences on the breakup process and is expected in due
course to provide a detailed picture of the mechanisms that
govern the spray formation. It is a powerful tool for assisting
in the development of accurate atomization models for practical applications.
This approach has been used by Bianchi et al. (2007),
who are also interested by the flow inside the nozzle which
can influence the processes of atomization. In the case of

ogst100079_DosSantos

17/11/11

11:16

Page 817

F Dos Santos and L Le Moyne / Spray Atomization Models in Engine Applications, from Correlations to Direct Numerical Simulations 817

high-pressure injectors, cavitation can occur in the nozzle and

changes the rest of the spray, Peng Karrholm et al. (2007)
has worked on this phenomenon and has performed some
calculations in order to model cavitation, with LES.

3 MULTIDIMENSIONAL MODEL: DNS

As classic simulations carry problematic issues concerning
grid dependency, in order to correctly represent the spray,
alternative solutions must be considered. The simplest is to
carry direct numerical solutions. This of course can involve
costly computer features if trying to solve the integral space
domain. We propose here an alternative consisting on simulating only the very close vicinity of the spray and considering only the main processes in that region. For example,
given the length and time scales, we do not consider evaporation (although some marginal and not precisely estimated
amount of numerical evaporation is hard to avoid with the
interface capturing techniques employed), Tomar et al.
(2010). The idea is to carry independent simulations with
varying parameters that can furnish adequate initial conditions (velocity, position and size of droplets) to spray models.
Thus, DNS is carried up to the point where droplets reach a
size small enough to be represented by Lagrangian particles.
The principle has been studied and implemented by Tomar et
al. (2010) based on GERRIS code by Popinet (2003), which
has been applied to spray simulations (Fuster et al., 2009).
When droplets reach a size small enough they are simply
taken out from the fluid domain and implemented as
Lagrangian particles. Balance of exchanges between droplets
and gas is made for each cell, thus modifying gas flow.
Probably the main interest of DNS applied to spray consists
on the possibility of modeling unsteady complex processes
for which no validated models exist, for example the beginning
of jet injection.
Simulations presented here are intended to explore the
actual possibilities of the methodology. They are carried with
an adaptive 3D mesh of a maximum of 29(512) cells per
dimension when detailed aspects of liquid interface are
required, or 27(128) when global quantities like penetration
are estimated. With 29 cells the equivalent size of the smallest
cells is 2.3 m for an initial jet diameter of 0.2 mm, with 27
cells it is of 9.4 m. The Reynolds and Weber number of
experiments and simulations are theoretically 30 103 and 75
103 respectively although the effects of numerical viscosity
and surface tension are still to be measured. Simulation results
are therefore still to be validated and we only show here a
comparison with spray penetration and qualitative aspects.
Density ratio (gas/liquid) of simulation and experiments is .
Results are presented in non-dimensional variables S* and t*:
S
S* =
(68)
do

t* =

t
d
o
Vinj

(69)

Figure 16 shows an example of the results obtained for the

simulation of a pressure-hole type of spray at time t* = 20
with the minimum mesh size, along with details of the adaptive mesh used by GERRIS code. The traced value is the isosurface 0.5 for VOF tracer (0 is gas only, 1 is liquid only).
Note that surface instabilities grow from inlet in a symmetric
way due to meshing of the boundary conditions. They later
develop into ligaments that ultimately form the droplets that
detach and sub-divide. Although this can be considered as
non-physical because dependent on meshing characteristics,
surface instabilities growth are proven to be responsible for
atomization. The real origin of such perturbations can be
attributed to turbulence, roughness and other causes.
Enhancement of such perturbations differs in a less iso-tropical distribution but growth depends on flow characteristics
and relatively low on meshing. Globally, droplet and ligament formation is realistic. For computing time reasons
droplets below a certain size have been removed. They can
be treated in a Lagrangian way for full-spray calculations.

Figure 16
View of a pressure-hole type spray simulated by DNS with
GERRIS code (top: VOF tracer 0.5 iso-surface), details of
mesh in the cross section (bottom left) and of the squared
zone (bottom right). Minimum cell size is 1/29.

ogst100079_DosSantos

818

17/11/11

11:16

Page 818

Oil & Gas Science and Technology Rev. IFP Energies nouvelles, Vol. 66 (2011), No. 5

Figure 17
Spray aspect for non-dimensional time t* = 80 (top: VOF tracer 0.5 iso-surface), corresponding mesh (middle) and detail of mesh (below).
Minimum cell size is 1/27.

In Figure 17, the same spray is represented at time t* = 80

for a coarser mesh. Note that details on ligaments have not
been lost although they are less refined. For values like penetration, the mesh change from 29 to 27 induces variations of
less than 5%, for the time span we have tested. For later
development of spray, as the liquid core vanishes, this observation may not remain valid.
In Figures (18-20), we show a comparison of simulation
with images issued from experiments in similar conditions.
Imaging is obtained through LIF of spray droplets with a
four-hole injector (do = 0.2 mm) in a pressure vessel using
iso-octane doped with fluoranthene as described in work by
Le Moyne et al. (1997). Only one of the 4 sprays is correctly
illuminated by the laser sheet intersecting it in its axis. A PIV
set-up with a micro-telescope equipped camera is used.
Figures 18 to 20 concern times t* at 20, 40, and 80. In the
experimental reference they correspond to 200, 400 and 800
microseconds after start of injection. The start of injection
t = 0, is detected experimentally by the appearance of liquid
at the nozzle hole.
Figure 21 shows a comparison of penetration for experimental and simulation data issued from a single simulation,
showing good agreement. Low dispersion of this parameter
has been verified for a limited set of simulations. Note that
even for very early steps of spray development, penetration
follows a power law. In this case:
S * (t * )

0.6772

(70)

This is contrary to some assumptions made previously

in literature, as no experimental data was available for so
small times.
Simulation is carried removing all small droplets, so only
the liquid core is represented here. Globally the general
aspects of the liquid core are well captured by the simulation.
Qualitatively the length scales of ligaments and droplets correspond to those of experiment images. As no effective
method to measure droplet sizes in this very dense region of
spray is available, a quantitative comparison of droplets sizes
between simulation and experiments is not possible. Two
representations of spray are shown with a cross section view
of VOF tracer value and a 3D view of iso-surface at value 0.5
of VOF tracer. One can see that differences with experiments
exist although no model for atomization or turbulence is
implemented. One reason is that numerical aspects may not
allow capturing the very complex interface. Another reason
may be that boundary conditions for the simulation are not
realistic.
Indeed, a parabolic velocity profile is injected into the
simulation domain and break-up of droplets is not effective
in the simulation until surface instabilities have grown
enough. One can notice from experiment images that some
atomization is visible from the very nozzle exit and that
the smooth cylinder of simulations is not apparent in
experiments.
The precocity of this atomization may be due to fastest
growing instabilities or intra-hole phenomena. As the code is

ogst100079_DosSantos

17/11/11

11:16

Page 819

F Dos Santos and L Le Moyne / Spray Atomization Models in Engine Applications, from Correlations to Direct Numerical Simulations 819

Figure 20
Spray image experiment and simulation at t* = 80.

40
Data

35

Simu

30

S*

25

Figure 18

20
15

Spray image experiment and simulation at t* = 20.

10
5
0

20

40

t*

60

80

100

Figure 21
Non-dimensional penetration versus non-dimensional time
for experiments and simulation.

Figure 19
Spray image experiment and simulation at t* = 40.

incompressible, and with no phase change for the time being,

it is not possible to model cavitation-like mechanisms.
However, as shown in Figure 22, by solving the 3D flow
inside the injector it is possible to induce early atomization,
namely from t* ~ 3. This suggests that the velocity profile
induced by complex flow inside the injector with vorticity
sources, induced by rugosity for example, may play a role at
least on early stages of atomization. Figure 23 shows a view

of the spray at a very early stage of injection. The injection

pressure has probably not reached its steady value as injector
needle may still be moving. Expansion in the lateral directions and the corrugated aspect of the interface suggests that
high levels of turbulence prevail from spray inception.
Further work is needed to validate the results and assumptions made, but as the DNS allows to capture the finest
details of atomization it is, with the growing power of
processors, a possible way to explore the spray formation
mechanisms.

4 DISCUSSION
All the models cited in the previous sections show all the
diversity of the results concerning spray formation. As the

ogst100079_DosSantos

820

17/11/11

11:16

Page 820

Oil & Gas Science and Technology Rev. IFP Energies nouvelles, Vol. 66 (2011), No. 5

Figure 22
Example of DNS simulation for inter-injector flow (inside
injector view left t* = 3 and t* = 20; and out-coming liquid
right for t* = 20).

process of atomization is of singular complexity the efforts to

resume to few equations the effects of some parameters on
strongly non-linear mechanisms give limited agreement with
experiments. In the sense that when applied to different conditions to those where model constants and hypothesis were
validated one often obtains non physical results. Every
known spray model has a limited range of validity as quite
different atomization regimes exist with a single device
(Rayleigh regime, wind induced, atomization, flash boiling,
effervescent atomization, etc.). Moreover, the experimental
techniques available do not always allow access to necessary
data like drop size and velocity in dense regions of spray. As
a whole, when confronted to the need of simulating combustion processes in engines one often finds difficulties in choosing the correct spray sub-model and setting its several constants. Very often the final objective is completely reversed
as a specific set of experiments is needed to calibrate the
spray model. The semi-empirical correlations and 0D models
have been extensively used and have proven their utility and
limits. The widely spread use of Eulerian-Lagrangian models
with RANS equations for gas phase has given relatively low
valid results as the constants and parameters in spray models
lack of a proven methodology for correct adjustment.
Moreover, the issue of grid dependency, when addressed,
asks difficult numeric questions at a certain computer cost.
Finally, great hope is put on the LES and DNS simulations as
the possibilities now at range allow starting to address the
difficult points of spray modeling.

ACKNOWLEDGMENTS
Authors would like to thank Profs. Guibert and Zaleski,
Danielson Engineering and Conseil Rgional de Bourgogne
for continuous support.

REFERENCES
Apte S.V., Gorokhovski M., Moin P. (2003) LES of Atomizing
Spray with Stochastic Modeling of Secondary Breakup, Int. J.
Multiphase Flow 29, 1503-1522.
Arai M., Tabata M., Hiroyasu H., Shimizu M. (1984) Disintegrating
Process and Spray Characterization of Fuel Jet Injected by a Diesel
Nozzle, SAE International, SAE paper 840275.
Arcoumanis C., Gavaises M. (1997) Effect of Fuel Injection
Processes on the Structure of Diesel Sprays, SAE International, SAE
paper 970799.
Arrgle J.M., Pastor J.V., Ruiz S. (1999) The Influence of Injection
Parameters on Diesel Spray Characteristics, SAE International, SAE
paper 1999-01-0200.
Figure 23
Early view of spray at t* = 3.

Beale J.C., Reitz R.D. (1999) Modeling Spray Atomization with the
Kelvin-Helmholtz/Rayleigh-Taylor Hybrid Model, Atomization
Sprays 9, 623-650.

ogst100079_DosSantos

17/11/11

11:16

Page 821

F Dos Santos and L Le Moyne / Spray Atomization Models in Engine Applications, from Correlations to Direct Numerical Simulations 821

Beheshti N., Burluka A. (2004) Eulerian Modelling of Atomisation

in Turbulent Flows, 19th Annual Meeting of the Institute for Liquid
Atomization and Spray Systems (Europe), Nottingham, 6-8
September, 207-212.
Bharadwaj N., Rutland C., Chang S. (2009) Large Eddy Simulation
Modelling of Spray-Induced Turbulence Effects, Int. J. Engine Res.
10, 97-119.
Bianchi G. Minelli F., Scardovelli R., Zaleski S. (2007) 3D Large
Scale Simulation of the High-Speed Liquid Jet Atomization, SAE
International, SAE paper 2007-01-0244.
Chehroudi B., Chen S.-H., Bracco F.V., Onuma Y. (1985) On the
Intact Core of Full-Cone Sprays, SAE International, SAE paper
850126.
Cousin J., Desjonqures P. (2003) A New Approach for the
Application of the Maximum Entropy Formalism on Sprays,
ICLASS 2003, Sorrento, Italy, 13-17 July.
De Villiers E., Gosman A., Weller H. (2004) Large Eddy
Simulation of Primary Diesel Spray Atomization, SAE
International, SAE paper 2004-01-0100.
Demoulin F.X., Beau P.A., Blokkeel G., Mura A., Borghi R. (2007)
A New Model for Turbulent Flows with Large Density
Fluctuations: Application to Liquid Atomization, Atomization
Sprays 17, 315-345.

Lee K., Aalburg C., Diez F.J., Faeth G.M., Sallam K.A. (2007)
Primary breakup of turbulent round liquid jets in uniform crossflows, AIAA J. 45, 1907-1916.
Lefebvre H.A. (1989) Atomization and Sprays. Combustion: An
International Series, Hemisphere, New-York, 434 p.
Levich V. (1962) Physicochemical Hydrodynamics, Prentice-Hall
Inc., pp. 639-650.
Levy N., Amara, S., Champoussin J.C. (1998) Simulation of a
Diesel Jet Assumed Fully Atomized at the Nozzle Exit, SAE
International, SAE paper 981067.
Long W., Hosoya H., Mashimo T., Kobayashi K., Obokata T.,
Durst F., Xu T. (1994) Analytical Functions to Match Size
Distributions in Diesel-Sprays, International Symposium COMODIA,
Yokohama, Japan, 11-14 July.
Menard T., Beau P., Tanguy S., Demoulin F., Berlemont A. (2005)
Primary break-up: DNS of liquid jet to improve atomization modelling, WIT Transactions on Engineering Sciences 55, 343-352.
Merrington A.C., Richardson E.G. (1947) The break-up of liquid
jets, Proc. Phys. Soc. 59, 1.
Naber J.D., Siebers D.L. (1996) Effects of Gas Density and
Vaporization on Penetration and Dispersion of Diesel Sprays, SAE
International, SAE paper 960034.

Dent J.C. (1971) Basis for the Comparison of Various Experimental

Methods for Studying Spray Penetration, SAE International, SAE
paper 710571.

Nishimura A., Assanis D.N. (2000) A Model for Primary Diesel

Fuel Atomization Based on Cavitation Bubble Collapse Energy,
ICLASS 2000, Pasadena, CA, 16-20 July, pp. 1249-1256.

Elkotb M. (1982) Fuel Atomization for Spray Modelling, Progr.

Energ. Combust. Sci. 8, 61-90.

ORourke P.J., Amsden A.A. (1987) The Tab Method for

Numerical Calculation of Spray Droplet Breakup, SAE
International, SAE paper 872089.

Faeth G. (1996) Spray Combustion Phenomena, Symp. Int.

Combust. 26, 1593-1612.
Fuster D., Bagu A., Boeck T., Moyne L.L., Leboissetier A.,
Popinet S., Ray P., Scardovelli R., Zaleski S. (2009) Simulation of
Primary Atomization With an Octree Adaptive Mesh Refinement
and VOF Method, Int. J. Multiphase Flow 35, 550-565.
Heywood J. (1988) Internal Combustion Engine Fundamentals,
McGraw-Hill.
Hiroyasu H., Arai M. (1990) Structures of Fuel Sprays in Diesel
Engines, SAE International, SAE paper 900475.
Hiroyasu H., Arai M., Tabata M. (1989) Empirical Equations for the
Sauter Mean Diameter of a Diesel Spray, SAE International, SAE
paper 890464.
Hiroyasu H., Kadota T., Tasaka S. (1978) Study of the Penetration
of Diesel, JSME International Journal 44, 3208-3219.
Huh K.Y., Gosman A.D. (1991) A Phenomenological Model of
Diesel Spray Atomization, Proceedings of International Conference
on Multiphase Flows, Tsukuba, Japan, 24-27 September.
Ibrahim E.A., Yang H.Q., Przekwas A.J. (1993) Modeling of Spray
Droplets Deformation and Breakup, J. Propuls. Power 9, 651-654.
Kuensberg Sarre C., Kong S.C., Reitz R.D. (1999) Modeling the
Effects of Injector Nozzle Geometry on Diesel Sprays, SAE
International, SAE paper 1999-01-0912.
Le Moyne L. (2010) Trends in atomization theory, Int. J. Spray
Combustion Dynamics 2, 49-84.
Le Moyne L., Maroteaux F., Guibert P., Murat M. (1997) Model
and Measure of Flows at the Intake of Engines, J. Phys. III France
7, 1927-1940.
Lebas R., Menard T., Beau P., Berlemont A., Demoulin F. (2009)
Numerical simulation of primary break-up and atomization: DNS
and modelling study, Int. J. Multiphase Flow 35, 247-260.

Park S.W., Lee C.S., Kim H.J. (2003) Investigation of Atomization

Characteristics and Prediction Accuracy of Hybrid Models for
High-Speed Diesel Fuel Sprays, SAE International, SAE paper
2003-01-1045.
Patterson M.A., Reitz R.D. (1998) Modeling the Effects of Fuel
Spray Characteristics on Diesel Engine Combustion and Emissions,
SAE International, SAE paper 980131.
Peng Karrholm F., Weller H., Nordin N. (2007) Modelling injector
flow including cavitation effects for Diesel applications,
Proceedings of FEDSM2007 5th Joint ASME/JSME Fluids
Engineering Conference, San Diego, CA, USA, 30 July - 2 August.
Popinet S. (2003) Gerris: a tree-based adaptive solver for the incompressible Euler equations in complex geometries, J. Computat.
Phys. 190, 572-600.
Ranz W.E. (1958) Some Experiments on Orifice Sprays, Can. J.
Chem. Eng. 36, 175-181.
Reitz R. (1987) Modeling atomization processes in high pressure
vaporizing sprays, Atomization Sprays 3, 309-337.
Reitz R.D., Bracco F.B. (1979) On the Dependence of Spray Angle
and Other Spray Parameters on Nozzle Design and Operating
Conditions, SAE International, SAE paper 790494.
Reitz R.D., Diwakar R. (1987) Structure of High-Pressure Fuel
Sprays, SAE International, SAE paper 870598.
Ruiz F., Chigier N. (1991) Parametric Experiments on Liquid Jet
Atomization Spray Angle, Atomization Sprays 1, 23-45.
Sandia National Laboratories (2010) online database,
http://www.sandia.gov/ecn/cvdata/dsearch.php.
Schihl P., Bryzik W., Altreya A. (1996) Analysis of Current Spray
Penetration Models and Proposal of a Phenomenological Cone
Penetration Model, SAE International, SAE paper 960773.

ogst100079_DosSantos

822

17/11/11

11:16

Page 822

Oil & Gas Science and Technology Rev. IFP Energies nouvelles, Vol. 66 (2011), No. 5

Shannon C.E. (1948) A Mathematical Theory of Communication,

Bell Syst. Tech. J. 27, 379-423, 623-656.
Siebers D.L. (1999) Scaling Liquid-Phase Fuel Penetration in Diesel
Sprays Based on Mixing-Limited Vaporization, SAE International,
SAE paper 1999-01-0528.
Su T.F., Patterson M.A., Reitz R.D., Farrell F.V. (1996)
Experimental and Numerical Studies of High Pressure Multiple
Injection Sprays, SAE International, SAE paper 960861.
Tanner F.X. (1997) Liquid Jet Atomization and Droplet Breakup
Modeling of Non-Evaporating Diesel Fuel Sprays, SAE
International, SAE paper 970050.

Vallet A., Burluka A.A., Borghi R. (2001) Development of a

Eulerian Model for the Atomization of a Liquid Jet, Atomization
and Sprays 11, 24.
Varde K.S., Popa D.M., Varde L.K. (1984) Spray Angle and
Atomization in Diesel Sprays, SAE International, SAE paper
841055.
Wakuri Y., Fujii M., Amitani T., Tsuneya R. (1960) Studies of the
Penetration of a Fuel Spray in a Diesel Engine, JSME International
Journal 3, 123-130.
Wan Y., Peters N. (1999) Scaling of Spray Penetration with
Evaporation, Atomization and Sprays 9, 111-132.

Taylor G.I. (1950) The Instability of Liquid Surfaces When

Accelerated in a Direction Perpendicular to their Planes, Proc. R.
Soc. London, 192-196.

Wu P.-K., Faeth G. (1995) Onset and End of Drop Formation Along

the Surface of Turbulent Liquid Jets in Still Gases, Phys. Fluids 7,
2915-2917.

Tomar G., Fuster D., Zaleski S., Popinet S. (2010) Multiscale

Simulations of Primary Atomization, Comput. Fluids 39, 1864-1874.

Final manuscript received in March 2011

Published online in September 2011

Copyright 2011 IFP Energies nouvelles

Permission to make digital or hard copies of part or all of this work for personal or classroom use is granted without fee provided that copies are not made
or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for components of this
work owned by others than IFP Energies nouvelles must be honored. Abstracting with credit is permitted. To copy otherwise, to republish, to post on
servers, or to redistribute to lists, requires prior specific permission and/or a fee: Request permission from Information Mission, IFP Energies nouvelles,
fax. +33 1 47 52 70 96, or [email protected].