You are on page 1of 19

Computers & Fluids 103 (2014) 215233

Contents lists available at ScienceDirect

Computers & Fluids


j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / c o m p fl u i d

Fluid-dynamic and numerical aspects in the simulation of direct CNG


injection in spark-ignition engines
Mirko Baratta , Nicola Rapetto
Dipartimento Energia, Politecnico di Torino, Italy

a r t i c l e

i n f o

Article history:
Received 25 September 2013
Received in revised form 20 June 2014
Accepted 29 July 2014
Available online 7 August 2014
Keywords:
Gas engines
Numerical simulation
Underexpanded jet
Direct injection

a b s t r a c t
This paper presents a detailed discussion on the numerical simulation of the underexpanded gas efux
from an outward-opening poppet-valve injector into an engine combustion chamber. The aim of the
paper is to optimize the numerical simulation strategy for direct gas injection, in view of its application
to internal combustion (IC) engines. In the rst part of the paper, the widely studied case of a two-dimensional compressible ow is examined, and the main guidelines for the development of an effective
numerical model for compressed natural gas (CNG) direct injection simulation are given, with specic
reference to IC engines. The second part of the paper is devoted to the description of the numerical model
developed and validated by the authors within the Star-CD environment, which is characterized by the
presence of two distinct meshes. The rst is built manually and covers the region surrounding the injector
exit, whereas the second one covers most of the engine chamber and is built using the Es-ICE tool. A careful grid-independence study has been carried out in both the rst and second part of the paper, and the
inuence of the spatial discretization of the convective uxes has been discussed as well.
The analyses have shown that a resolution of 40 cells in the nozzle height should be adopted to describe
the typical phenomena that characterize an underexpanded free jet, unless a second order scheme can be
implemented. However, as far as the simulation of the jet penetration time-history and its mixing with
the surrounding air is concerned, sufciently accurate results can also be obtained by using 20 cells per
nozzle diameter and the rst-order upwind scheme. As for the direct injection engine model, 16 cells
across the nozzle lift represent a good compromise between accuracy and reliability of the results and
the required computational time. The model has been validated with the support of experimental PLIF
images in an optical-access engine, and has shown overall good accuracy and reliability, thus suggesting
it is suitable for mixture formation analysis.
2014 Elsevier Ltd. All rights reserved.

1. Introduction and present work


In recent years, air quality has become a particularly severe
problem and the growing concern about exhaust emissions from
IC engines has resulted in the implementation of strict emission
regulations in many countries, particularly in the United States
and Europe. Natural gas (NG), which is primarily composed of
methane, is regarded as one of the most promising alternative
fuels, because of its high H/C ratio and high research octane number (about 130). NG allows CO2, HC and PM emissions to be
reduced, while the gap in performance, with respect to gasoline
engines, can be recovered by means of turbocharging [13]. NG
Corresponding author. Address: Dipartimento Energia, Politecnico di Torino,
Corso Duca degli Abruzzi, 24, 10129 Torino, Italy. Tel.: +39 011 090 4484; fax: +39
011 090 4599.
E-mail address: mirko.baratta@polito.it (M. Baratta).
http://dx.doi.org/10.1016/j.compuid.2014.07.028
0045-7930/ 2014 Elsevier Ltd. All rights reserved.

engine performance can be further improved by adopting a direct


injection (DI) concept, in which the gaseous fuel is injected directly
into the combustion chamber. The absence of fuel in the port during the intake phase in fact leads to appreciable benets in terms of
volumetric efciency and, in turn, of engine torque, particularly at
low engine speeds, through a more effective scavenging and the
optimization of the compressor working point. Moreover, DI
engines have been shown to enhance the combustion characteristics and to extend the fuel-lean operating limit of normal engine
operation, compared to port fuel injection [2,4].
In recent years, considerable interest has been generated in the
application of three-dimensional computational uid-dynamic
(CFD) models to simulate direct injection of gas and subsequent
mixture formation within the combustion chamber. The application of these models can cut down the cost and time required for
the development of engine hardware and testing. However, the
simulation of gas DI into an engine combustion chamber is a

216

M. Baratta, N. Rapetto / Computers & Fluids 103 (2014) 215233

Nomenclature
ASoI
BTDC
C
cp
cv
CA
CCD
CFD
CNG
CH4
CO
CO2
DI
EC
env
EOI
ff
H
h
hneedle
HC
IC
ICCD
imep
k
LIF
Ls
M
_
m
~ in
m
MARS
MCE
~
n
ne
NG

after start of injection


before top dead center
carbon
specic heat at constant pressure
specic heat at constant volume
crank angle
charge-coupled device
computational uid dynamics
compressed natural gas
methane
carbon monoxide
carbon dioxide
direct injection
European Commission
environment
end of injection
ammable fuel mass fraction
hydrogen
height
needle lift
hydro-carbon
internal combustion
intensied CCD
indicated mean effective pressure
turbulent kinetic energy
laser induced uorescence intensity
shock cell spacing
Mach number
mass ow rate
mass ow rate per unit volume in source cells
monotone advection and reconstruction scheme
multi-cylinder engine
cell face normal
nozzle exit
natural gas

challenging task, due to the high pressure ratios that are used and
to the substantial difference in dimensions (roughly between two
and three orders of magnitude) between the injector nozzle and
the engine cylinder length scales [5,6]. The study becomes even
more critical when the injection is carried out by means of poppet
valves instead of axial-symmetric orices, as a complex timedependent area of the nozzle cross-section is generated.
From a uid-dynamic point of view, the problem is closely
related to the widely studied case of an underexpanded compressible ow issuing from an orice. In fact, the rail pressure under
normal engine operation is 20 bar or higher, whereas the pressure
in the combustion chamber during the injection is usually between
approximately 0.5 and 3 bar, thus causing the ow to become
chocked. From a numerical point of view, the simulation of the
gas ow from a nozzle to the combustion chamber implies that a
number of additional issues must be faced, mainly connected to
the time-dependent problem geometry, which requires the adoption of a moving mesh strategy in which the cell number, size
and connectivity may vary at the same time. During this process,
since the combustion chamber generally has a complex geometry,
some cells with high skewness or a high aspect ratio may appear.
In order to keep the number of such cells to a minimum, the moving mesh strategy needs careful optimization.
Other investigations have been made, with the aim of obtaining
a better understanding of the physics of gaseous injection in

NPR
p
p0
p
PLIF
PM
R
RAFR
rpm
SU
SU,add
SCE
SI
SOI
ST
T
T
t
TDC
TMA
TVD
~
U
U
u, v, w
u00
UV
Vc

nozzle pressure ratio


pressure, Reynolds-averaged pressure
pressure turbulent uctuation
total pressure
planar laser induced uorescence
particulate matter
gas constant
relative airfuel ration
revolution per minute
source term of U
additional source term in source cells
single-cylinder engine
spark ignition
start of injection
spark timing
temperature
total temperature
time
top dead center
trimethylamine
total variation diminishing
velocity vector
velocity magnitude
velocity components
density-weighted (Favre) velocity uctuation
ultra violet
volume of the computational cell
e
turbulent kinetic energy dissipation rate
q
density
r
standard deviation
/ = ast/a equivalence ratio
U
generic ux property
c
cp/cv
C
generic diffusion coefcient
Xc
cell boundary

engines, as well as of the critical issues connected to its numerical


simulation [521].
As far as the modeling approach is concerned, the most accurate
way of simulating the expansion dynamics that the NG undergoes
inside the injector and directly downstream from it, is to use the
entire domain, including the ow within the injector. Such an
approach is rather expensive, as a result of the ne grid and the
small time steps required. This is, for example, the procedure that
has been followed in [5,8,12]. However, in [8], the injector needle
was kept at its maximum lift position during the whole injection
period, and the rather simple geometry of the combustion chamber
allowed a variable-size axial-symmetric mesh to be used to discretize the computational domain. In [17] the nozzle throat region
was included in the simulation domain, and two different kinds of
inlet boundaries were compared, namely, the mass-ow inlet
(which imposes the mass ux and the total temperature of the rail)
and the pressure inlet (which imposes the total pressure and temperature of the rail). The needle position was xed, which resulted
in less penetration during the transients. With reference to the
injection of hydrogen from a single-hole injector, in [15,18,19]
two separate grids were adopted: a computational grid was used
for the full engine geometry (including valves and ducts) to compute the in-cylinder ow-eld during the gas-exchange phase,
prior to fuel injection, whereas a grid without the intake and
exhaust ports, but including the nozzle geometry, was used to

M. Baratta, N. Rapetto / Computers & Fluids 103 (2014) 215233

simulate the direct injection and mixture formation processes. The


ow eld, state variables, and turbulence-model parameters
calculated from the gas-exchange phase were set as the initial
conditions for the subsequent simulation of direct injection and
mixture formation. This approach allowed the authors to separate
the gas exchange from the injection process, which require different grid resolutions. In such a case, the model cannot be applied to
the simulation of injection during the induction phase (advanced
injection). An alternative procedure has been applied in [9,10],
among others. In these cases, a detailed simulation of the ow issuing from the nozzle was not conducted and an empirical model
was instead used to describe the supersonic ow outside the injector. This approach does not seem to be applicable to poppet valve
injector simulation under variable lift conditions. Another
approach consists of using 3-D CFD models developed for liquid
injection and adapting them to gaseous fuel injection by means
of the introduction of ctitious gas droplets [10,11]. Although this
procedure allows very reduced grid densities to be adopted, it does
not seem to be suitable to simulate poppet-valve injectors operating under variable lift conditions, as a great deal of information
needs to be specied concerning the initial position, size and
density of the droplet and so on.
Another key factor that has to be dened is the mesh size. Many
researchers have discussed the problem of optimizing the number
of cells to be used in the simulation of gas injection, in order to
obtain accurate results and guarantee reasonable calculation
efforts at the same time [5,7,9,18]. A grid size of half the nozzle
diameter has been proposed for practical engine computations
[9,22], in which the main goal is to achieve a reasonably good accuracy in capturing the essential features of the mixture formation
mechanism, such as the jet penetration rate. However, in [7] it
was pointed out that at least 4 cells should be used within the nozzle diameter in order to obtain jet-velocity results that are independent of the ambient values of the turbulence scales. With
reference to gas injectors with poppet valves, in [12] 6 cells were
used to discretize the nozzle-throat section at the maximum needle lift, and the model proved to be suitable for mixture formation
analysis with different combustion chamber, bowl-in-piston, and
injector geometry congurations. However, a large gradient along
the ow direction of all the gas properties was found in the nozzle
outow region, and it was therefore necessary to increase the grid
resolution in the same direction to obtain a high accuracy in the
evaluation of the ow properties in the nozzle throat [5]. Moreover, it is reported that a high grid density is required normally
to the ow direction, in order to allow the compressible ow
details (i.e., the expansion fan departing from the nozzle exit) to
be described accurately [7,18,23,24].
The above discussion has highlighted the opportuness of optimizing the strategy for the numerical simulation of direct gas
injection, in view of its application to IC engines. Therefore, the
present paper offers a detailed discussion on the numerical simulation of the underexpanded gas efux from injector nozzles.
The present paper is complementary to Ref. [24]. More specifically, in [24] a thorough analysis of the mixture formation
process in a prototype DI-CNG engine is provided, from both a
numerical and experimental point of view, and the mixture
characteristics were correlated to the available pollutant
emission data. In the present paper, the uid-dynamic problem
of the underxpanded compressible ow from a poppet-valve
injector is analyzed in detail. The objectives of the paper are
listed hereafter:
Numerical and phenomenological characterization of twodimensional compressible jets in open environments. This
study aims at characterizing such ows independently from
any specic application.

217

Denition of the main guidelines for the development of a


suitable numerical model for the simulation of gas injection
and mixing phenomena in an engine combustion chamber.
Detailed description of the model developed over the last
few years within the InGAS research project of the European
Community (www.ingas-eu.org).
Discussion of the inuence of the grid size and topology,
discretization schemes and turbulence modeling on the
model results.
A preliminary version of the model was developed within the
framework of a previous project and the results obtained with it
were presented by Baratta et al. [12]. Since then, the model has
been generalized by including the possibility of simulating the
whole engine cycle, and its accuracy has been enhanced to a great
extent.
2. Two-dimensional compressible ow from a orice
In the present section, the supersonic underexpanded free-jet
issuing from a two-dimensional nozzle is studied by means of
numerical simulation. The structure of supersonic underexpanded
gas jets has been well documented in the literature [2527]. In
such ows, the pressure at the exit plane of the nozzle is higher
than the pressure in the downstream environment, so that the
gas expansion is completed in the downstream environment. The
expansion waves departing from the nozzle lip reect on the
constant pressure jet boundary as compression waves and subsequently coalesce, thus forming a shock structure which is commonly known as barrel shock. Depending on the ow condition
(nozzle pressure ratio, NPR), the shock may reect regularly at
the centerline (moderately underexpanded jet, Fig. 1a), generating
a periodic structure of shock cells, or it may terminate in a triplepoint Mach disk conguration (highly underexpanded jet, Fig. 1b).
The region behind the Mach disk is a subsonic ow and is bounded
by a slip line emanating from the triple-point [25].
As stated in Section 1, the rst part of the present paper is
devoted to the analysis of two-dimensional underexpanded ows
in Cartesian coordinates. The general purpose CFD package StarCD (version 4.12) has been used to perform the calculations. The
aim of the work was to assess how accurately a commercial CFD
solver can describe highly compressible ows, like those relevant
for CNG direct injection. As a matter of fact, such ows require a
ne grid resolution at the nozzle exit, so as to accurately describe
the PrandtlMeyer expansion fan that occurs at the nozzle exit as
well as the subsequent recompression with shock formation and
reection. Moreover, the overall grid topology needs to be optimized in order to account for both the need of high accuracy in
the supersonic region and the opportunity of limiting the computational time required. The resulting mesh structure was then used
as a reference to setup the engine chamber mesh. In fact, the poppet-valve injector used in the engine under consideration usually
gives rise to a toroidal jet in the cylinder (see Section 3), in which
the distribution of velocity and thermodynamic properties, in each
section through the axis, is closely similar to that usually observed
in two-dimensional jets.
2.1. Computational domain, numerical grid and boundary conditions
The geometry, the dimensions and the main features of the
considered computational domain are given in Fig. 2. All the
domain dimensions have been specied as functions of the width
of the nozzle exit, wne, which was set equal to 1 mm. The uid
(air) enters into the calculation domain with given stagnation conditions p, T and with Mach number M = 1.1. It is worth observing

218

M. Baratta, N. Rapetto / Computers & Fluids 103 (2014) 215233

Jet
boundary

Barrel
shock

Nozzle exit

Expansion
fan

Reflected
shock

Secondary
expansion

(a) regular reflection


Jet
boundary

Reflected shock

Mach
disk

Expansion
fan

Nozzle exit

Incident
shock

Subsonic
core

Triple point

Secondary
expansion

(b) Mach reflection


Fig. 1. Features of moderately (a) and highly (b) underexpanded gas jets.

Domain exit

Inlet

Wall

wne

512 wne

Wall

Domain exit

960 wne
Fig. 2. Computational domain and boundary conditions.

that the present section is mainly meant to be a preliminary analysis to the main focus on the paper, that is, the simulation of gas
injection into the engine chamber. Since the considered injector
is known to have a slightly diverging portion downstream of the
throat section, a higher-than-unity Mach number could be
expected at the injector exit. For that reason, the value M = 1.1
has been selected in the preliminary investigations presented
hereafter.
Before being issued into the constant pressure downstream
environment, the uid ows in a constant section duct whose
length is three times the nozzle width (3wne). The duct has been
introduced to avoid any interference between the outowing jet
and boundary to the left of the domain. As the ow at the domain

inlet is supersonic, the velocity magnitude U, the temperature T


and the density q of the uid are imposed as boundary conditions.
In the cases discussed in this paper, such conditions were already
established in the inlet section at the beginning of the transient
simulation. From an operative point of view, the values of M, T,
p, are imposed indirectly through the following relations:

U
M p
cRT

p qRT

U2
T T
2cp
c
  c1
T

p p
T

3
4

By combining Eqs. (1), (3) and (4), and by recalling that cp = c/


(c  1)  R, the injection pressure (that is, the stagnation pressure
upstream of the nozzle) can be expressed as a function of the static
pressure at the domain inlet (nozzle outlet) and the Mach number:


c
c  1 2 c1
p p  1
M
2

The static pressure was xed at 10 bar in most of the cases


considered in this paper (see Table 1 and Section 2.3), and
consequently Eq. (5) gives p = 21.35 bar. The computational
domain extends 960wne in the jet axial direction and 512wne in
the direction normal to the jet centerline. The mesh is composed

219

M. Baratta, N. Rapetto / Computers & Fluids 103 (2014) 215233


Table 1
Summary of the considered test cases.
Test-case
no.

NPR = p/penv
()

T
(K)

Mne
()

penv
(MPa)

Min. grid size


(mm)

1
2
3
4
5
6
7

7.12
7.12
7.12
21.35
21.35
21.35
4.27

393
393
393
393
393
393
393

1.1
1.1
1.1
1.1
1.1
1.1
1.1

0.3
0.3
0.3
0.1
0.1
0.1
0.1

wne/10
wne/20
wne/40
wne/10
wne/20
wne/40
wne/20

of hexahedral elements with an aspect ratio equal to 1. In order to


obtain insight into the grid requirements of underexpanded compressible ows, the calculations were performed using three different grid resolutions, featuring minimum grid sizes, near the nozzle
exit, of 0.025wne, 0.05wne and 0.1wne, respectively. The region
around the nozzle exit is in fact critical from a computational point
of view, given the very high pressure and velocity gradients that
occur within it. As can be inferred from Fig. 3, for the mesh featuring a minimum cell size of 0.05wne, the grid becomes progressively
coarser as the distance from the nozzle increases, until a maximum
size of 3.2wne is reached in the outer domain. The latter mesh size
was established in order to obtain the same ratio between the
maximum and minimum cell dimensions as the one in the engine
model. This ratio is virtually xed by accuracy and practical constraints, as will be discussed later on, and it was kept the same
for the three considered meshes. The boundary on the left of the
downstream environment is a wall in which adiabatic and no-slip
conditions are applied. The far-eld boundaries, normal and parallel to the jet, are modelled as Riemann boundaries and are meant
to enable weak pressure waves to leave the solution domain without reection. Stronger pressure waves might actually be partially
reected and might moderately inuence the numerical solution
achieved. It should be noted, however, that this inuence is also
mitigated to a great extent by the considerable distance between
the jet and the boundaries, as well as by the numerical dissipation
effect due to the coarseness of the mesh close to the boundaries,
which was purposely designed to minimize the boundary-induced

3.2 wne

1.6 wne
0.8 wne

0.8 wne
0.4 wne
0.2 wne
0.1 wne
0.05 wne

Fig. 3. Grid-size distribution for the mesh with 20 cells across the nozzle section
(minimum size = 0.05 wne).

solution perturbation. Riemann boundaries require the value of the


pressure (penv) and temperature (Tenv) in the environment outside
the boundary to be specied.
An alternative solution for the gas injection was tested by
substituting the inlet boundary condition with source cells. This
facility provides an alternative and convenient means of introducing uid mass into the calculation domain, and allows the virtual
injector model to be adopted (see Section 3). The latter was purposely developed and validated in [5]. From a topological point
of view, in the source cell approach, the inlet boundary is
replaced by a wall one, whereas the injection process is modeled
as an additional source term (SU,add) in the nite volume equation
of the cells in which the injection occurs. The term is of the form:

~ in Uin
SU;add m

~ in is the mass ow rate of the injected stream per unit


where m
volume, and Uin is the value of a generic ow property per unit
mass (momentum, enthalpy, etc.) attributed to the stream [28]. A
term for each solved nite volume equation needs to be specied
at the selected cell(s). On the basis of Eq. (6), the generic conservation law for each extensive property is modied in the source cell
locations as follows:

@
@t

Z
VC

qUdV C

~U  CrUdXC
~
n  qU
XC

~ in Uin dV C
SU m
VC

7
In which VC denotes each nite volume of the computation
domain, XC is the boundary that delimits the volume and SU
represents the source term of U for standard cells, i.e., cells
without gas injection.
2.2. Numerical schemes and turbulence modeling
The mass, momentum and total energy conservation equations
are solved using the nite volume method with a cell-centered
colocated arrangement.
The diffusive terms are discretized using a centered scheme
while the convective uxes are treated using a rst-order upwind
scheme or the MARS scheme, depending on the test case, as specied in Section 2.3. MARS is a second-order accurate TVD scheme
that operates in two separate steps: reconstruction and advection
[29]. The scheme incorporates a variable compression level that
controls the amount of second order upwinding and which can
usually be set to 1 in the momentum equation, but cannot exceed
0.6 in the energy equation, in order to reach a stable solution. Calculations were also performed with lower compression levels.
The PISO algorithm was used to solve the pressurevelocity
coupling while an implicit scheme, based on the implicit Euler
scheme and explicit deferred correctors, is used for time marching.
The resulting method has a formal accuracy between rst and second order. As far as the allowed time-step is concerned, when the
rst order upwind scheme is selected, the time step can be
adjusted to have a maximum Courant number of between 2 and
5, depending on the grid resolution, whereas the MARS scheme
needs the Courant number to be decreased to 0.5 in order to obtain
a stable solution when high compression levels are used. If the
compression level is set to very low values, the Courant number
can be increased to a value suitable for the rst order upwind
scheme. Typical time-step values that have been used in the simulation of 2-D under-expanded jets are between 20 and 200 ns. Once
the stability had been guaranteed, the results accuracy showed to
be not affected by the adopted time-step. As a matter of fact, a time
scale for the ow under study was evaluated according to Vuorinen
et al. [30], and a value of about 1.3 ls was obtained, which was
more than six times higher than the longest time step.

220

M. Baratta, N. Rapetto / Computers & Fluids 103 (2014) 215233

As far as the turbulence modeling approach is concerned, it


should be pointed out that, to the best of the authors knowledge,
very few studies have analyzed in detail the effect of the constitutive relation for the turbulent stress tensor. Moreover, such works
have dealt with axial-symmetric jets rather than two-dimensional
ones. Predictions and measurements were compared in [31] for a
fully expanded axial-symmetric Mach 2 jet by implementing Wilcoxs two-equation kx model [32]. The turbulence model was
shown to allow the mean jet geometry close to the jet outlet to
be reproduced with acceptable accuracy. A broader agreement,
within 10%, was achieved further downstream for the axial velocity. In [33] numerical tools adapted for the simulation of the near
eld of highly underexpanded jets were developed and combined,
and corrections of the ke model adopted to take into account
compressibility effects on turbulence were assessed. More in
detail, the density-weighted averaging procedure suggested by
Favre [34] was introduced to derive the averaged mean conservation equations, and the additional terms with respect to the corresponding incompressible equations (namely, the pressure work
@p
G u00i  @x
and the pressure dilatation Pd p0 
i

@u00i
)
@xi

were modeled

by means of proper correlations. Qualitatively good results were


obtained by implementing linear eddy-viscosity closures. However, adapted non-linear closures of the Reynolds stress tensor
components were tested in order to improve the accuracy of the
potential core length prediction. Unfortunately, the model
appeared to make the numerical method unstable, even for very
low time steps. In the simulation of underexpanded jets presented
in [13], the observed underestimation of the jet penetration for values above 6 cm was attributed to the effect of the Standard ke
model, which overestimates jet diffusion. This result was in agreement with the ndings reported in [11,35]. Scarcelli et al. [20,21]
numerically analyzed gas injection into a cylindrical chamber
and validated their results through a comparison with high-denition, time-resolved measurements of gas jet mass distribution
using X-ray radiography. Three ke versions (standard, RNG, realizable) were evaluated, along with three solutions for the discretization of the convective uxes (1st, 2nd and 3rd order). The
standard ke model provided the best accuracy, in terms of jet
width. Moreover, the 3rd order scheme provided a higher level of
detail of the Mach disk. Nevertheless, it resulted to be less robust
than the linear model, and provided less accuracy in terms of mixing of the jet with the ambient gas far from the nozzle.
An example of the inuence of the ke model variant on the 2D
simulation results is provided in the following section. However,
on the basis of the above considerations, the compressible version
of the Standard ke model was used to perform most of the 2-D jet
simulations, mainly due to its well proven stability. The stability of
the model has in fact resulted to be the main issue in the engine
simulation, as discussed later on.

cases were run using the rst-order upwind scheme for the spatial
discretization of convective uxes. Cases from 4 to 6 were also carried out with the MARS scheme.
Fig. 4 reports the jet Mach number distribution along the jet
symmetry axis computed with 3 different grids, using the rstorder upwind scheme for convective uxes for test cases 13, in
which the nozzle pressure ratio was 7.12. It can be inferred that
a reduced grid size leads to higher peaks in the Mach number
upstream of the shock, as well as to stronger shocks. This behavior
can be ascribed to a decrease in the grid numerical viscosity. However, for the considered pressure ratios the shock cell wavelength
is virtually unaffected by the grid size and in particular the length
of the potential-core-like region was much the same in tests 2 and
3. Similar results were obtained in [36] on an axial-symmetric jet.
Furthermore, the difference in the Mach number distributions of
tests 2 and 3 is lower than that of tests 1 and 2. The change in
the maximum Mach number is of about 6% from test 1 to test 2,
and of about 3.5% from test 2 to test 3. This is an indicator that,
in principle, completely grid independent results can be achieved
with a grid ner with respect to test 3. However, due to the very
low grid size of test 3 as well as to the associated small time step,
a further grid renement might be not feasible and the results
obtained in test 3 can be considered satisfactory. It is also worth
pointing out that the assessment of grid-independence actually
depends on what results are considered. Complete grid independence cannot be claimed for the Mach number peak values of Mesh
3, as already said. On the other hand, concerning the shock position
as well as the velocity prole after the shock-cells region, increasing further the number of cells does not produce appreciably different results with respect to those found in the paper. The jet
axial penetration (related to three different mass concentrations
of the injected gas) versus time is presented in Fig. 5 for the same
NPR. Amongst the values in Fig. 5, the mass concentration of 0.05
can be considered as a measure of the jet-tip location. The considered time interval (2.5 ms) is characteristic of an injection duration
at a part load operating condition of an IC engine. It is evident that,
as far as the mixing process and the jet penetration are concerned
(which are the main outcomes of the simulation of gas injection in
internal combustion engines), the grid size has no impact and an
almost grid independent solution can already be reached with a
resolution of 10 cells in the nozzle diameter for the analyzed
NPR. It is worth pointing out that, for high penetration values,
the result may be affected by the fact that the grid size at the jet
tip, for the considered tests, is the same. However, as already
mentioned, the 2-D jet study should be considered as a preliminary
one for the main application, which is the fuel direct injection in

test 1
2.3. Results and discussion

test 3
Mach [-]

The simulations were performed under different conditions,


which were chosen to be representative of likely NG engine
injector operating points. Table 1 summarizes the tests that were
performed and the respective boundary conditions.
In tests 16, the static pressure was pne = 1 MPa and the total
temperature was T = 393 K. In tests 13, the downstream
environment pressure was set to penv = 0.3 MPa, whereas it was
penv = 0.1 MPa in tests 46. In a DISI engine, these values are
representative of an injection during the compression stroke (late
injection) and during the intake stroke (early injection), respectively. An additional test was performed in which the static pressure at the nozzle exit was lowered to pne = 0.2 MPa and the
environment pressure was set to penv = 0.1 MPa (test 7). All the

test 2

2.5

1.5

1
0

10

15

20

25

30

35

40

x / wne [-]
Fig. 4. Mach number distribution along the jet axis at time = 2.5 ms; test cases 13,
rst order upwind scheme.

221

M. Baratta, N. Rapetto / Computers & Fluids 103 (2014) 215233

Mass conc. = 0.5

300

300

x / wne [-]

75

test 1

250

test 2
test 3

50
25

200
150
100

0.5

1.5

2.5

test 2
test 3

200
150
100
50

50

test 1

250

test 2
test 3

x / wne [-]

test 1

100

x / wne [-]

Mass conc. = 0.05

Mass conc. = 0.3

125

0.5

time [ms]

1.5

2.5

0.5

time [ms]

1.5

2.5

time [ms]

Fig. 5. Jet axial penetration versus time for test-cases 13.

natural gas engines. In view of this, in this paper the opportunity of


rening the mesh around the jet-tip location was not considered,
because it would not have been possible in the engine application,
due to the limitations of the overall combustion-chamber mesh.
However, given the small differences in the maximum Mach number of the shock-cell, shown in Fig. 4, in general a limited inuence
of the grid size on penetration can be expected.
A grid independence study has also been carried out for a test
with a higher NPR (NPR = 21.35, tests 46). The calculations were
performed with both the rst-order upwind and the second-order
MARS schemes (with different compression levels). In this case, the
shock reection at the centerline was no longer regular. Instead,
Mach reection occurred, with the formation of a subsonic ow
area behind the Mach disk. The Mach number distribution along
the jet symmetry axis is reported in Fig. 6. When combined with
the rst-order upwind scheme, the coarsest grid resulted to be
unable to reproduce the transition from regular to Mach reection
and the ow behind the rst shock cell was still supersonic. An
increase in grid resolution to 20 cells per nozzle diameter allowed
the normal shock that occurs on the jet axis to be captured. A further decrease in the grid spacing produced a higher peak Mach
number, the difference being about 3%, without affecting the shock
position to any great extent. Table 2 reports the Mach disk height
for the considered meshes. It can be seen from the table that a
decrease in the grid size by a factor of 2 produces an increase in
the Mach-disk height by 2030%. Fig. 6 also reports the calculation
results obtained with the second-order MARS scheme, for test
cases 4 and 5. Though the predicted shock position did not differ

test 4 UW

test 5 UW
test 6 UW
4

test 4 MA

Mach [-]

test 5 MA

Table 2
Mach disk height for test cases 46, NPR = 21.35.
Mesh size

Upwind scheme

MARS scheme

wne/10
wne/20
wne/40

NA
1.95 wne
2.6 wne

2.30 wne
2.70 wne

from the previous cases, the Mach reection was correctly


described, even with the grid featuring the minimum resolution.
The use of higher compression levels increases the order of the
scheme and thus allows the sharpness of the shock to be increased
(results not shown in the paper for the sake of conciseness), but the
computational cost also increases due to the lower time step
allowed. It is also worth pointing out that the accuracy could be
further improved by combining the mesh with a size of wne/40 to
the second order scheme, although, in this case, the required computational effort would be not acceptable.
The jet axial penetration versus time is presented in Fig. 7 for
the previously considered NPRs. As far as the periphery of the jet
is concerned (related to a mass concentration of the injected gas
of 0.05), it is shown that a lower grid resolution produces higher
jet penetration in the rst half of the injection period, while all
the grids give similar results after t = 1.5 ms. In addition, the
employment of the second-order MARS scheme results in a slightly
lower jet penetration prediction. The picture on the left in Fig. 7
refers to a higher mass concentration (0.5), which is more representative of the core of the jet plume; in this case, the effect of
the grid resolution on the jet penetration is higher in the second
part of the injection period. In short, with a higher NPR, the grid
with a cell size of 0.1wne is not sufcient to get a fully grid-independent evolution of the fuel concentration patterns versus time,
unless a high-resolution spatial scheme (MARS) is considered.
The length of the rst shock cell (Ls) is plotted in Fig. 8 as a function of the fully expanded, isentropic, jet Mach number (Mj), given
by:

v
!
u
  c1=c
u 2
p
t
Mj
1
c  1 penv

The circles represent the present computations (tests 2, 5 and 7,


grid resolution of wne/20) while the solid line refers to Tams
theoretical relation [36]:

0
0

10

15

20

25

30

35

40

x / wne [-]
Fig. 6. Mach number distribution along the jet axis at t = 6 ms. Test cases 46. Solid
lines: rst-order upwind scheme. Dotted lines: MARS scheme.

2
1
1=2 1 2 c  1M j
Ls
Md
2
M 2j  1
hne
Mj
1 12 c  1M 2d

!1c=2c1
9

Md being the nozzle design Mach number (1.1 in the considered


tests). Eq. (9) holds for Mj > Md. The agreement is relatively good,

222

M. Baratta, N. Rapetto / Computers & Fluids 103 (2014) 215233

Mass conc. = 0.5

Mass conc. = 0.3

500

Mass conc. = 0.05

500

500

400

400

test 5 UW
test 6 UW

[-]

300

ne

test 5 MA

x/w

x / w ne [-]

test 4 MA

300

x / w ne [-]

test 4 UW

400

200
100

test 4 UW

200

test 5 UW
test 6 UW

100

300
test 4 UW

200

test 5 UW
test 6 UW

100

test 4 MA

test 4 MA
test 5 MA

test 5 MA

0.5

1.5

2.5

0.5

time [ms]

1.5

2.5

0.5

time [ms]

1.5

2.5

time [ms]

Fig. 7. Jet axial penetration versus time for test-cases 46. Solid lines: rst order upwind scheme. Dotted lines: MARS scheme.

3. CNG injection in an engine chamber

20

Tam corr.
Simulated

Ls / wne [-]

15

10

1.5

2.5

M j [-]
Fig. 8. Shock-cell spacing Ls/wne as a function of the jet Mach number Mj. Solid line:
Tams analytical solution [36]. Circles: present computations (resolution: wne/20).

thus denoting a rather good accuracy of the overall approach for the
simulation of compressible ows.
In conclusion, the preliminary analysis discussed in this section
shows that, in order to accurately describe the typical phenomena
that characterize an underexpanded free jet (PrandtlMeyer
expansion, barrel and reected shocks), a high cell resolution is
required (up to 40 cells across the nozzle diameter), unless a second-order scheme can be implemented. However, good results
can be achieved, in terms of penetration and mixing process prediction, with a cell size of 0.05wne and when the rst-order
upwind scheme is used, particularly at relatively low NPR.
The results presented in Figs. 48 were obtained by adopting
the Standard ke model for the turbulence closure. However, the
inuence of the specic ke model on results was investigated as
well. The Standard and the Realizable variants were considered
and the results are shown in Figs. 9 and 10. Fig. 9 shows that the
Realizable ke model predicts a slightly higher shock intensity
and, correspondingly, jet penetrations. However, such differences
are lower than those related to the mesh size and discretization
scheme. Moreover, from Fig. 10 one can infer that the jet shape
and the Mach distribution are almost identical for the two models.
Tests 2 and 5 were also repeated using source cells to model the
injection, in order to validate this approach. Fig. 11 shows that the
results are completely equivalent to those obtained with the standard inlet boundary condition. This conrms that the source cell
approach allows consistent results to be obtained when underexpanded compressible ows are simulated and confers further reliability to the application of the virtual injector model [5] in the
engine-ow simulations.

This section is focused on the description and validation of the


numerical model developed for the simulation of the CNG direct
injection in the InGAS spark ignition engine. Two engines were
actually built during the InGAS CP for the experimental investigations; a multi-cylinder engine (MCE) for an analysis of the performance and emissions and an optical-access single-cylinder (SCE)
engine. The present model refers to the SCE, which was designed
for the experimental investigation of mixture formation processes
by means of the Planar Laser Induced Fluorescence technique. The
experimental results obtained on the SCE were also used for the
validation of the numerical model. The main characteristics of
the engine are reported in Table 3, whereas Figs. 12 and 13 show
a schematic of the cylinder head and the piston of both engines,
as well as a representation of the poppet-valve-shaped needle
and of its cartridge.
The computational domain considered for the model development takes into consideration the complete engine geometry,
including the intake valves and ports, for the whole engine cycle
simulation, covering the induction and compression strokes. Given
the symmetry of the computational volume, with respect to a
plane through the cylinder axis, only half has been considered to
build the mesh.
The main problems related to mesh generation in SI internal
combustion engines are due to the complex geometry of the cylinder head and to the presence of canted valves, whose motion needs
to be taken into account during the simulation along with the piston movement. The simulation of CNG direct injection into an
engine combustion chamber involves additional challenges, mainly
due to the high operating pressure ratio of the injector, which
requires a high grid resolution to capture the main features of
the supersonic underexpanded gas jet issuing from its exit. As
mentioned in Section 1, it is also worth pointing out that a substantial difference in dimensions exists between the injector nozzle
and the engine.
3.1. Numerical model description
The model developed to simulate a gaseous injection in an
engine chamber is discussed in this section. The model was
developed within the Star-CD environment (http://www.cd-adapco.com/products/star_cd/) and the Es-ICE tool has been adopted to
build the mesh and to set-up the grid motion.
The parametric grid generation process of Es-ICE is based on a
two-dimensional template and allows a hexahedral mesh of the
engine to be easily built, whereas the automatic generation of a
time-dependent grid allows the valves and the piston movement
to be simulated. However, this tool does not allow a mesh of the

223

M. Baratta, N. Rapetto / Computers & Fluids 103 (2014) 215233

Test 2; t = 2.5 ms

Test 2; t = 2.5 ms
1

Standard
2.5

Realizable

Mach [-]

Inj. Gas conc. [-]

Standard
0.75

0.5

Realizable

1.5

0.25

0
0

100

200

300

400

500

x / wne [-]

10

20

30

40

x / wne [-]

Fig. 9. Standard versus Realizable ke model: injected-gas concentration (left) and Mach number (right) along the jet axis.

Fig. 10. Standard versus Realizable ke model: injected-gas concentration (left) and Mach number (right) contours.

region inside the injector to be generated. This would require a


maximum cell size near the nozzle of less than 1/10 of the maximum needle lift, which is much lower than that recommended
for common engine-ow simulation.
For this reason, two different meshes have been built and
assembled in the current work. The rst one, built in the Prostar
environment (the standard pre/post processor of Star-CD) accounts
for the computational domain in the injector vicinity (from now on
referred to as injector seat), while the second one, built in Es-ICE,
involves the remaining part of the domain (i.e., the portion of the
chamber volume not included in the injector seat, the intake and
the exhaust ports). Fig. 14 shows the computational domain of
the injector seat, whereas Fig. 15 shows the 2D template used to
generate the Es-ICE mesh, which can be seen in Fig. 16. The latter
gure also shows the different renement levels that have been
used inside the Prostar mesh.
The grid architecture needs to be optimized in order to obtain
the best compromise between accuracy of the results and computational time required. The key features of the numerical grid are
not only the minimum grid size (which affects the computational
time step) and the total number of cells, but also the transition
in mesh size between ner and coarser grids within the injector

seat mesh, as well as the interface between the injector seat and
the Es-ICE meshes and nally the arrangement of the different
regions within the Es-ICE grid.
The injector seat region can be considered as the most critical
domain from the computational point of view, owing to the very
high velocity and density gradients that occur within very short
distances. The overall grid resolution has a signicant inuence
on the numerical results in a direct injection process. However,
the main inuence stems from the cell size within the nozzle exit
[19]. The 2D test cases described in Section 2 suggest employing a
higher spatial resolution than 1/10 of the needle lift (whose maximum value ranged from 0.15 to 0.25 mm in the considered tests)
and this grid size constraint has been respected in the mesh for the
engine model, in order to maintain an high degree of accuracy in
the supersonic region. The manual denition of the injector seat
mesh (Fig. 14) allows a very high cell quality hexahedral structure
to be obtained. Moreover, a cylindrical-based structure has dened
in the region around the injector axis, in order to t the axially
symmetric geometry of the injector throat section. As far as the
mesh size transition is concerned, 6 renement levels are needed
to obtain a smooth transition from the high cell density zones near
the nozzle area to the relatively coarser grids adopted for the

224

M. Baratta, N. Rapetto / Computers & Fluids 103 (2014) 215233

Test 5; t = 2.5 ms

Test 2; t = 2.5 ms
5

inlet

inlet
4

source

Mach [-]

Mach [-]

2.5

1.5

source

3
2
1
0

1
0

10

20

30

40

10

20

30

Test 2; t = 2.5 ms

Test 5; t = 2.5 ms

inlet
0.75

inlet

source

Inj. Gas conc. [-]

Inj. Gas conc. [-]

40

x / wne [-]

x / wne [-]

0.5

0.25

0.75

source

0.5

0.25

0
0

100

200

300

400

500

x / wne [-]

100

200

300

400

500

x / wne [-]

Fig. 11. Comparison of the results obtained with the standard inlet boundary condition (solid lines) and the source cell approach (dotted lines). From left to right, NPR = 7.12
and NPR = 21.35. First row: Mach number distribution along the jet axis. Second row: injected gas mass concentration along the jet axis.

Table 3
Characteristics of the InGAS SCE engine.
Number of cylinders
Bore
Stroke
Conrod length
Cylinder displacement
Compression ratio

1
82 mm
85 mm
136.5 mm
449 cm3
8.7

discretization of the combustion chamber. This value turned out


to be the best compromise between the transition gradualness
and the distance between two consecutive levels, which should

correspond to at least 45 cell layers. In view of this, the dimensions of the injector seat mesh should be increased as much as possible, also to cover most of the domain characterized by the
supersonic ow. However, its dimension is limited in the radial
direction by the presence of the grid blocks that are necessary to
simulate the valve motion and in the axial direction by the piston
position at TDC. The triangular shape adopted for the injector seat
domain allows the structure of the Es-ICE grid to be tted. In fact,
the seat mesh dimensions perfectly correspond to those of the centrally located triangular-shaped region of the Es-ICE mesh (see the
2D template in Fig. 15). Another important aspect is the quality of
the interface (transition) between the Prostar and the Es-ICE

Fig. 12. Combustion chamber of the InGAS engine: cylinder head (left) and piston (right).

M. Baratta, N. Rapetto / Computers & Fluids 103 (2014) 215233

225

Fig. 13. InGAS injector: poppet-valve needle (left) and needle-cartridge assembly (right).

The optimization of the template for the Es-ICE mesh has been a
critical task (Figs. 15 and 16), since the peripheral position of the
intake valve could have determined the appearance of a large number of skewed cells. Specic mesh-parameter settings were thus
selected carefully to optimize the process. The average size of the
cells in the Es-ICE mesh was set to 0.5 mm and it could reach values of up to 1.2 mm in the deforming layer areas that are necessary
to simulate the piston motion. The nal model was obtained by
connecting the two meshes (see Fig. 16, in which the Prostar mesh
for the injector seat is highlighted). The intake and exhaust
domains are deactivated at valve closure timings, in order to save
computational time.
3.2. Numerical schemes and boundary conditions

Fig. 14. Computational mesh of the injector seat. Colors denote the source cells for
the virtual injector model implementation. (For interpretation of the references to
color in this gure legend, the reader is referred to the web version of this article.)

Fig. 15. 2D template within Es-ICE.

meshes. As a matter of fact, the presence of the interface is strictly


related to that of the Prostar mesh and cannot be avoided. Should
the cells of the Prostar mesh be exposed to cells of the Es-ICE mesh
with very different size, a low-quality interface would arise, with
the appearance of further numerical diffusion. In order to avoid
this, the meshes were built in such a way that cells of comparable
size and shape were obtained at the interface.

The Star-CD code has been used to solve the mass, momentum
and total energy conservation equations with the nite volume
method. The diffusive terms were discretized using a centered
scheme, while the convective uxes were treated using the
rst-order upwind scheme. This choice was mainly due to the possibility of adopting larger time steps and saving computational
time. The second-order MARS scheme, which was successfully
employed in the two-dimensional test case, here required a large
decrease in the time step size in order to obtain a stable numerical
solution, and this resulted in excessive computational costs. The
PISO algorithm was used to couple pressure and velocity.
Given the adoption of the Upwind differencing scheme, as well
as the hexahedral cell structure that was purposely designed
within the supersonic region, the model stability limits imposed
nearly the same limitations as the 2-D jet model on the Courant
number, whose maximum value ranged between 2 and 5 during
the injection. The time step was hence adjusted case by case in
order to meet these requirements. The values of the time step
adopted in the different tests are reported in Table 4. Three different meshes are reported in the table, as will be discussed in Section
3.3. As a result, the injection simulation on an eight Intel Xeon CPU
X5472 3.00 GHz processor workstation required 420 days,
depending on the injection duration and the in-cylinder pressure
at SOI.
As far as the boundary conditions are concerned, the experimental pressure proles and temperature mean values were
imposed at the inlet and outlet sections of the domain (intake
and exhaust manifold), on the basis of the SCE measurements.
The simulation of the gas exchange started at the end of the
exhaust phase, 30CA before inlet valve opening. The initial

226

M. Baratta, N. Rapetto / Computers & Fluids 103 (2014) 215233

Prostar mesh

Fig. 16. Final computational grid. Cell size distribution refers to Mesh 2 (see Section 3.3).

Table 4
Computational time steps adopted in the different test-cases.
First half of induction stroke
Late induction stroke till IVC
Compression stroke
Injection

Table 5
Main characteristics of the considered computational grids.

0.02 CA deg
0.05 CA deg
0.10 CA deg
2000x3 EOI 50 (Mesh 1)
2000x3 EOI 50 (Mesh 2)
2000x3 EOI 50 (Mesh 3)
2000x3 EOI 220 (Mesh 2)
1500 WOT (Mesh 2)

300 ns
150 ns
30 ns
60 ns
200 ns

pressure and temperature in the cylinder, exhaust and intake manifold were also set on the basis of experimental measurements. As
an example, Fig. 17 shows the intake-valve mass-ow rate prole
related to a case at 2000 rpm, partial load and stratied mixture.
The virtual injector model [5] was used to simulate the injection of methane into the combustion chamber. On the basis of this
approach, source cells were set at the critical section (denoted in
color in Fig. 14), and the computational domain upstream from this
section was eliminated. For a complete description of the virtual
injector model, the reader can refer to Baratta et al. [5].
3.3. Inuence of grid resolution
The main features of the numerical grids adopted for the
calculation are reported in Table 5. The three grids basically differ
0.025

Mass Flow Rate [kg/s]

0.02
0.015
0.01
0.005
0
-0.005
-0.01
300

360

420

480

540

600

CA [deg]
Fig. 17. Intake mass ow rate (2000 rpm, imep = 3 bar, RAFR = 1.8, stratied
mixture).

Total number of cells (SOI)


Cells across the needle lift

Mesh 1

Mesh 2

Mesh 3

1,180,000
8

1,350,000
16

2,095,000
32

as far as the resolution within the injector seat area is concerned,


while the grid architecture and the overall grid resolution in the
Es-ICE grid remain unchanged. This decision was based on the evidence that the effect of spatial discretization in the near nozzle
region is preponderant and the dimensions of the rened area
are not negligible compared to the whole combustion chamber.
Moreover, the overall grid resolution adopted in the combustion
chamber (reference size 0.5 mm) is already higher than the typical
mesh size of internal combustion engine applications (12 mm),
and consequently it can be hardly increased.
The inuence of the grid resolution is here analyzed for a partload operation point at 2000 rpm, under stratied conditions.
Figs. 18 and 19 show the effect of the grid size on the simulation
of the supersonic ow-eld near the injector exit. A time instant
of 8CA ASoI has been considered, when a quasi-steady conguration of the jet near the injector exit is reached. The Mach number
contour plots obtained with the three different grids are shown
in Fig. 18, while Fig. 19 reports Mach number distribution along
the mean jet path, starting at the needle exit. It is worth noting that
M > 1 at the nozzle exit. This is due to the presence of a diverging
nozzle portion between the throat and the exit sections, in which
further ow expansion occurs. The coarsest grid gives a rather poor
description of the typical shock-cell structure that occurs in underexpanded gas jets. The maximum Mach number, as well as the
length of the potential-core-like region, are underestimated considerably, while the jet diffusion is higher. The jet computed with
the second mesh clearly shows two shock cells, and a higher level
of detail of the underexpanded region is achieved. Obviously, a better description of the periodic structure of the shock cells is
obtained with the third mesh, which also predicts a higher Mach
number and is affected less by numerical diffusion. However, the
jet Mach number predicted with the second and third grids
becomes more and more similar as the distance from the nozzle
increases. The difference in the maximum Mach number is of about
12% between Mesh 1 and Mesh 2, and of about 8% between Mesh 2
and Mesh 3. Moreover, Mesh 2 gives nearly the same Mach number
downstream of the rst shock as Mesh 3. Finally, the percentage

227

M. Baratta, N. Rapetto / Computers & Fluids 103 (2014) 215233

Mesh 1

Mesh 3

Mesh 2

Mach [-]

Fig. 18. Inuence of the grid resolution on the computed jet Mach number in the supersonic under-expanded region close to the injector exit. Section through the combustion
chamber symmetry plane. From left to right, Mesh 1, Mesh 2 and Mesh 3.

during the second part of the injection. In addition, the increase in


resolution from grid 2 to 3 seems to have a negligible effect on the
fuel dispersion and air entrainment within the jet. Fig. 21 shows
the equivalence ratio distributions for the three meshes at 700
and 720CA (ring TDC). Moreover, Table 6 shows the values of
ru, i.e. the standard deviation of / over the control volumes as well
as the ammable fuel mass fraction ff:

mesh 1

Mach [-]

2.5

mesh 2
mesh 3

1.5

ff
1

0.5
0

10

15

20

x / hneedle [-]
Fig. 19. Inuence of the grid resolution on the computed Mach number. Data
collected along the mean jet path starting from the injector exit.

difference in the Mach number predicted by Mesh 2 and Mesh 3


along the mean jet path decreases to 5% for x/hneedle = 15 (see
Fig. 19).
The inuence of the grid resolution on the jet shape and
penetration can be observed in Fig. 20, which shows the fuelair
equivalence ratio (/) contours during an injection event, for the
same test case with partial load, stratied lean mixture. The overall
RAFR was 1.8. The results obtained with the three grids are
presented from left to right, while each row refers to a different
instant of the injection (from top to bottom 5CA, 8CA and
13CA ASoI). The poppet valve behaves as a bluff body over which
the injected gas ows, producing an almost axial-symmetric annular jet into the cylinder [5,8].
The results show the limited ability of the coarsest grid to
provide an accurate prediction of the jet penetration and evolution.
The penetration (measured with reference to the mass isoconcentration line with value of 0.05) is overpredicted, since the
rst part of the injection, which results in a narrower jet, and the
fuel concentration inside the plume is considerably overestimated
with respect to Meshes 2 and 3. This behavior is due to the higher
numerical diffusion associated to mesh 1, and to the consequently
lower jet velocity (see Fig. 18). In fact, the lower jet velocity causes
the ow patterns to collapse towards the axis at shorter distance
from the injector. The results obtained by using the second and
third grid are quite similar in terms of jet shape, width and penetration. However, the reduced resolution of mesh 2 leads to a
slightly higher equivalence ratio in the core of the jet, particularly

mass of fuel included in the flammable mixture


total mass of the fuel

10

It is apparent from the gure that Mesh 2 is able to give a full


grid-independent / distribution around the ignition timing for
the considered test case, being the variation of the fuel mass
fraction and of ru with respect to Mesh 3 always within 0.01. Thus,
Mesh 2 offers sufcient accuracy to assess different engine design
and calibration solutions.
3.4. Concluding remarks on grid resolution
It is evident that the grid size in the near-nozzle region has a
great inuence on the fuel jet evolution and fuelair mixing. As
already pointed out, the assessment of grid-independence actually
depends on what results are considered.
As far the mixture stratication is concerned, the grid independence of results from Mesh 2 can be concluded from the almost
identical equivalence ratio contours that can be seen in Fig. 21,
as well as from the very similar jet shapes during the injection that
can be observed in Fig. 20 (second and third column). Given the
target of the simulation of CNG direct injection in a IC engine,
i.e., the denition of the best injection strategy among different
possibilities, the mixture stratication before combustion onset is
the main result that is expected from the simulation model. Hence,
its grid independence is essential. However, the mixture stratication cannot be the only parameter to be examined for assessing the
grid independence. In fact, also the essential features of the supersonic region need to be captured with the best possible accuracy,
as they determine the jet shape evolution and, in turn, the fuel
air mixing process (as thoroughly discussed in [5]). In this case,
the grid independence of Mesh 2 results is reached to a lower
extent, but it should be considered that the cell count of Mesh 3
is more than 1.5 times higher than Mesh 2 (see Table 3), so a compromise with the computational time needed to be found. Thus,
the authors considered Mesh 2 as the grid which provided the best
compromise between accuracy and computational requirement. It
is worth recalling that Mesh 2 features a resolution of 16 cells
across the nozzle lift.

228

M. Baratta, N. Rapetto / Computers & Fluids 103 (2014) 215233

Mesh1

Mesh2
5 CA ASoI

Mesh3

8 CA ASoI
7
6
5
4

13 CA ASoI

3
2
1
0

Fig. 20. Fuelair equivalence ratio contours: inuence of the grid resolution (section through the combustion chamber symmetry plane).

1.7

1.4

Mesh1

Mesh3

Mesh2
700deg CA

1.0

720deg CA
0.6

0.3
Fig. 21. Fuelair equivalence ratio contours: inuence of the grid resolution (section through the combustion chamber symmetry plane).

Table 6
Quantitative indicators of mixing degree for different meshes.

Flammable mixture mass fraction

r/

CA (deg)

Mesh 1

Mesh 2

Mesh 3

710
720
710
720

0.600
0.608
0.480
0.487

0.619
0.660
0.475
0.490

0.616
0.670
0.474
0.491

Whenever a reacting ow is to be simulated, it should be taken


into account that the model CPU requirement increases to a great
extent at combustion onset, due to the increase of the conservation
equations to be solved. In this case, usually the mesh adopted for

the simulation of gas injection cannot be used for the combustion


phase. The most suitable strategy could be to switch from Mesh 2
to a coarser mesh shortly before the combustion onset, so as to
meet the very different grid requirements which pertain to the
mixing and the combustion simulation.
3.5. Turbulence modeling
The RANS (Reynolds-Averaged NavierStokes) approach has
been adopted in the present work to model turbulence by means
of the two-equation ke model. This is the most widely used
turbulence model for internal combustion engine simulations
and different variants are available in literature, although many
researchers have successfully employed the kx model. Based on

229

M. Baratta, N. Rapetto / Computers & Fluids 103 (2014) 215233

the available literature, the choice of the turbulence model and its
tuning do not inuence the jet penetration or the overall evolution
throughout the cylinder to a great extent [18]. The effect of the turbulent closure is reported to become evident at the end of the
injection process [37]. According to [37], the standard ke model
seems to provide the best agreement with experimental data. This
is conrmed in [21] in which the simulation results obtained with
three variant of the ke model have been validated against experimental data. The standard model provided the most accurate
results in terms of jet width and mixing with the surrounding
air. The RNG version signicantly under-predicts the jet width
while Realizable ke model represented a good compromise
between penetration and mixing, in agreement with the ndings
from a previous study of the same authors [19]. The results
obtained with the Standard and the Realizable ke formulations
have been compared in Figs. 22 and 23, in terms of fuel concentration evolution versus time, for the same test case as Figs. 1821.
Fig. 22 compares the jet shape at three instants during the injection
process, whereas Fig. 23 shows the mixture stratication around
spark timing (left column) and at ring TDC (right column). Both
gures conrm the limited inuence of the turbulence model on
the overall jet characteristics, although the obtained jet details
appear slightly different from one another, particularly during
the jet impingement on the piston surface and the subsequent
recirculation. A slightly higher homogeneity degree is obtained
with the Realizable ke model (Fig. 23).
The standard formulation of the two equation ke model has
been adopted in the present work. This is also in agreement with
the choice made by Kim et al. [6,8].

3.6. Numerical model validation and result analysis


The developed numerical model has been validated by comparing experimental PLIF images [37,38], which were acquired by AVL
List GmbH as part of the EC InGAS Collaborative Project, with the
computed contours of the fuelair ratio, as suggested by Reboux
et al. [39,40]. Based on the above discussion, the Mesh 2 was
chosen.

A detailed description of the experimental apparatus, along


with an extensive numerical and experimental analysis of the mixture formation and performance of a Direct Injection CNG engine,
can be found in [24]. Fig. 24 shows the optical access engine on
the test bed for PLIF measurements in the AVL laboratories,
whereas Fig. 25 reports a picture of the transparent engine liner
and of the laser sheet position with respect to the engine chamber.
The transparent engine was fuelled through bombs, with an
already formed mixture of methane containing 0.4% in volume of
trimethylamine (TMA, C3H9N) as the tracer, in order to enable PLIF
visualization. The double sheet laser images were taken at dened
crank angle positions, using an intensied CCD camera with an
image-intensier quantum efciency of around 20% at 300 nm.
The ICCD camera was tted with a UV-compatible Micro Nikkor
lens.
As suggested in [38,41], an image with no tracer in the fuel
under motored conditions was used to remove the background
level, and a picture of forty cycles of a uniform mixture was used
to establish the gain between the luminosity level and equivalence
ratio. More specically, the calculation of A/F in each point of the
section is carried out according to the following equation:

Scorr

SB
UB

11

In which S, B, and U indicate the ensemble average of the


intensity in each point for the sample picture (gas jet under investigation), for the background and for the uniform mixture images,
respectively. Scorr indicates the corrected and scaled sample image.
It is worth noting that the correction is applied after having subtracted the background image to both the sample and the uniform
images. The uncertainty in the quantication process of the experimental equivalence ratio has been estimated to be about 30% on
the basis of the ndings of other studies in the literature [38,42].
The result of this quantitative model evaluation is reported in
Fig. 26, for a part-load operation point at 2000 rpm under stratied
conditions, and in Fig. 27, for a part-load operation point at
2000 rpm under homogeneous conditions. A triplet of images is
reported in Fig. 26 at each crank angle (indicated above each line):
the rst one was obtained from the CFD simulation, and the second

Standard k-
5 CA ASoI

8 CA ASoI

13 CA ASoI
7
6
5
4

Realizable k-

3
2
1
0

Fig. 22. Fuelair equivalence ratio contours during the injection event: comparison between Standard and Realizable ke models (section through the combustion chamber
symmetry plane).

230

M. Baratta, N. Rapetto / Computers & Fluids 103 (2014) 215233

Standard k-
700CA

720 CA

Realizable k-

Fig. 23. Fuelair equivalence ratio contours at the end of compression stroke: comparison between Standard and Realizable ke models (section through the combustion
chamber symmetry plane).

Fig. 24. Optical arrangement on the transparent engine test bed.

one is the ensemble-averaged / eld extracted from the experimental pictures, while the third one is the raw black and white
ensemble picture. The latter is not reported in Fig. 27. Both the
numerical and the experimental contours have been plotted on
the laser sheet plane represented in Fig. 25. Fig. 28 compares some
experimental and numerical trends of the ammable fuel mass
fraction (ff, rst row) and of the jet penetration (L, second row),
for the tests in Figs. 26 and 27 as well as for a test at 1500 rpm
and full load [24].
Experimental values of ff were determined considering the
measured / values on the PLIF plane. The axial penetration was con-

sidered for the jet penetration of jets like those in Fig. 25, whereas
the radial penetration (jet width) was plotted for the case of Fig. 27.
As previously mentioned, the accuracy target for the numerical
simulations was established so that the differences between the
simulated and experimental contours are within 30%. The accuracy
of the model is on average rather good, as can be appreciated from
the jet and the mixture pictures in Figs. 26 and 27. The evolution of
the jet shape versus time is sufciently well captured by the model,
over a wide range of in-cylinder thermo-dynamic conditions and
engine operation modes. However, the model tends to overpredict
the fuel concentration inside the plume, and in particular in second
part of the injection period (see Fig. 26, h = 50CA BTDC), and to
slightly underpredict the homogenization process after the end
of injection (EOI). This effect is due to the residual numerical viscosity, which had been minimized, but not fully eliminated, by
the mesh-optimization process. The consequently lower velocity
makes the ow patterns to collapse towards the axis after a shorter
path with respect to the experiments. As discussed above, a compromise has to be made between model accuracy and the required
CPU time, thus the minimum cell size has been limited and, in turn,
the total number of volumes in the model. A satisfactory agreement is also obtained for homogeneous operation at partial load
(Fig. 27). The simulated jet is slightly narrower than the experimental one at h = 225CA BTDC, due to the damping effect of the
residual numerical viscosity on the maximum jet velocity. In the
subsequent instants, the differences tend to vanish, thus indicating
that the accuracy of the model increases after the end of injection.

Fig. 25. Optical access to the SCE combustion chamber.

M. Baratta, N. Rapetto / Computers & Fluids 103 (2014) 215233

231

2000 rpm RAFR=1.8 Stratified operation imep=3 bar


= 58 CA BTDC (5 CA ASoI)
7
6
5
4

= 50 CA BTDC (13 CA ASoI)

3
2
1
0

= 30 CA BTDC

Fig. 26. Numerical model validation under stratied operation at partial load simulated (left) versus elaborated (center) equivalence-ratio contours, and average raw
luminosity distributions, at the indicated crank angles (laser-sheet plane, nominal EOI = 50, actual EOI = 37 CA BTDC).

2000 rpm RAFR=1.0 Homogeneous operation imep=3 bar


= 225 CA BTDC (8 CA ASoI)

= 205 CA BTDC

= 215 CA BTDC (18 CA ASoI)

= 190 CA BTDC

Fig. 27. Numerical model validation under homogeneous operation at partial load Simulated (rst and third columns) versus elaborated (second and forth columns)
equivalence-ratio contours, at the indicated crank angles (Laser-sheet plane, nominal EOI = 220, actual EOI = 207 CA BTDC).

This observation also holds for the test case of Fig. 26. It is worth
underlining that the apparently lower penetration, in the axial
direction, that appears in Fig. 27 when the piston is far from
TDC, can actually be attributed to a lack of luminosity in the lower
portion of the combustion chamber. In all the considered test
cases, the pictures of the most delayed crank angle show a comparable degree of stratication in the simulations and experiments,
as the rich mixture portions are located in almost the same position. Thus, the accuracy of the model can be expected to increase
even more as the spark-timing (ST) angle is approached.

The model accuracy is also testied by the graphs in Fig. 28, in


which the most signicant indicators of the jet evolution and of the
mixture formation are considered. In this gure, the experimental
values are represented along with their own error band, whose
width is related to the estimated uncertainty of the F/A ratio quantication process. The rst row in the gure shows a generally
good agreement, in terms of ammable fuel mass fraction, which
means that the model is able to reproduce the main features of
the air and fuel mixing process. The second row in Fig. 28 represents the time history of the jet penetration, which has been

232

M. Baratta, N. Rapetto / Computers & Fluids 103 (2014) 215233

experimental

1500 rpm Full load


[24]

simulated

2000 rpm 3 bar strat. lean


(Fig. 25)

2000 rpm 3 bar homog. stoich.


(Fig. 26)

ff [-]

0.8
0.6
0.4
0.2
0

-180

-150

-120

-90

-60

-30

-90

[deg CA]

-60

-30

-240

-180

[deg CA]

-120

-60

[deg CA]

100

L [mm]

75

50

Axial penetration
Radial penetration

Axial penetration

25

0
-180

-165

-150

-135

-120

-75

-60

[deg CA]

-45

[deg CA]

-30

-240

-225

-210

-195

-180

[deg CA]

Fig. 28. Comparison between the experimental and computed values of the fuel mass fraction (rst row) and jet penetration (second row).

dened as the distance between the injector tip and the point at
which the fuel mass concentration is equal to 0.05. Two distinct
penetrations are considered: the axial penetration is selected for
the rst two columns, where the injector produces an annular jet
in the cylinder (cloud-like jet [14,24]), whereas, in the right column, the ow issuing from the injector is attracted to the upper
wall (umbrella-like jet), thus the radial penetration is more significant than the axial one. Fig. 28 shows that the jet penetration
time-history is captured by the model within the accuracy limits
of the available experimental data. The discrepancy between the
axial penetration values that are detected in the rst case after
the end of injection is due to the scarce intensity of the laser light
in the lower combustion chamber region, as already mentioned,
whereas the last experimental penetration value in the second column does not have any physical meaning, since the jet has entered
the piston bowl and cannot be detected correctly by the camera.
The above discussion highlights that the model is accurate
enough to allow different mixture formation strategies to be
assessed in a direct injection engine. In fact, it is able to reproduce
the fuel mass fraction time-history with an accuracy comparable to
that of the experimental PLIF data, and also to capture the spatial
distribution of fuel, at least in a relative sense, when two different
combustion chamber layouts or injection strategies are compared.
The developed numerical models have provided useful information for the investigation and characterization of the mixture formation process in the InGAS direct injection CNG engine. A survey of
the results and a combined numericalexperimental investigation
have been presented in [24], whereas the results of a more detailed
numerical investigation of the process will be fully documented in
two companion papers, which are currently being prepared.
4. Conclusion
A numerical model for the simulation of direct gas injection in a
combustion chamber of an internal combustion engine has been

developed and validated in this paper. In the rst part of the paper,
the case of two-dimensional compressible ow, which is well
known from the literature, has been examined, and the main
guidelines for the development of an effective numerical model
for direct CNG injection simulation have been established, with
specic reference to IC engines. The second part of the paper has
been devoted to the description of the numerical model of the
engine that has been developed and validated by the authors.
The main conclusions are summarized as follows.
A resolution of 40 cells in the nozzle height should be adopted
in order to describe the complex gas-dynamic structure that
characterizes an underexpanded free jet, which originates from
the expansion fan departing from the nozzle edge and features
several shock-cell elements in which the gas is alternately
expanded and compressed. Should the resolution be of 20 cells
per nozzle height, the difference in the maximum Mach number
would be of about 3%, but the overall jet structure would be correctly reproduced.
A resolution of 10 cells can give acceptable results in terms of
jet penetration but is not able do describe the Mach reection
that occurs for highly underexpanded jets when coupled to
the rst-order upwind scheme.
The second-order MARS scheme allows more accurate results to
be obtained for a given mesh size, but the required computational effort could be not acceptable for ne grids.
As far as the simulation of the jet penetration time-history and
its mixing with the surrounding air are concerned, sufciently
accurate results can also be obtained can be simulated with
acceptable accuracy also by using a 20 cell per nozzle diameter
and a rst-order upwind scheme.
Slightly higher shock intensity and penetration are usually
obtained with the Realizable ke model with respect to the
Standard one, but the differences are lower than those related
to the mesh size and discretization scheme.

M. Baratta, N. Rapetto / Computers & Fluids 103 (2014) 215233

The outcomes in terms of grid size, discretization scheme and


time step from the 2-D jet analysis were used as reference for
the development of the engine numerical model, for which
additional challenges arise due to the complex time-dependent
geometry.
The mesh with 16 cells across the nozzle lift has shown to represent a good compromise between accuracy of the results and
the required computational time. On one hand, a mesh with 8
cells across the lift cannot accurately predict the jet penetration,
evolution and mixing with the surrounding air. On the other
hand, with a resolution of 32 cells some appreciable difference
in the supersonic region properties can be obtained, but the
equivalence ratio spatial distribution keeps the same as in the
case with 16 cells.
The MARS scheme, which was successfully employed in the
two-dimensional test case, required a large decrease in the time
step size in order to obtain a stable numerical solution when
applied in the engine model, and this resulted in excessive computational costs.
A slightly higher homogenization degree can be obtained with
the Realizable ke model, though the overall inuence of the
specic turbulence model variant is limited.
The model has been validated by examining experimental PLIF
images in an optical-access engine and has demonstrated to
be is accurate enough to allow different mixture formation
strategies to be assessed in a direct injection engine. In fact, it
is able to capture the spatial distribution of fuel, at least in a relative sense, when two different combustion chamber layout
solutions or injection strategies are compared.

Acknowledgements
The present research activity has been carried out as part of the
InGAS Collaborative Project of the European Community, VII FP.
The authors would like to thank Dr. Alois Fuerhapter and Harald
Philipp, of AVL List GmbH, for providing the experimental database
used in this paper, Dr. K. Roessler, of Daimler AG, for providing the
SCE geometrical data for the numerical model set up, Dr. Matthias
Gerlich and Wolfgang Zoels, of Siemens AG Corporate Technology,
for providing the geometrical and performance data of the injector,
as well as Dr. H. Schuele, of Continental Automotive GmbH, for
supplying the ECU calibration data.
References
[1] Kato K et al. Development of engine for natural gas vehicle. SAE tech. paper
1999-01-0574; 1999.
[2] Cho HM, He BQ. Spark ignition natural gas engines a review. Energy Convers
Manage 2007;48:60818.
[3] Korakianitis T, Namasivayam AM, Crookes RJ. Natural-gas fueled sparkignition (SI) and compression-ignition (CI) engine performance and
emissions. Prog Energy Combust Sci 2011;37:89112.
[4] Zeng K et al. Combustion characteristics of a direct-injection natural gas engine
under various fuel injection timings. Appl Therm Eng 2006;26:80613.
[5] Baratta M, Catania AE, Pesce FC. Multidimensional modeling of natural gas jet
and mixture formation in direct injection spark ignition engines development
and validation of a virtual injector model. J Fluids Eng 2011;133:0413041041304-14.
[6] Kim GH, Kirkpatrick A, Mitchell C. Supersonic virtual valve design for
numerical simulation of a large-bore natural gas engine. J Eng Gas Turbine
Power 2007;129:106571.
[7] Abraham J. What is Adequate resolution in the numerical computations of
transient jets? SAE tech paper 970051; 1997.
[8] Kim GH, Kirkpatrick A, Mitchell C. Computational modeling of natural gas
injection in a large bore engine. J Eng Gas Turbine Power 2004;126:65664.
[9] Ra Y et al. Multidimensional modeling of transient gas jet injection using
coarse computational grids. SAE tech paper 2005-01-0208; 2005.
[10] Chiodi M, Berner HJ, Bargende M. Investigation on different injection strategies in
a direct-injected turbocharged CNG engine. SAE tech paper 2006-01-3000; 2006.
[11] Hessel RP et al. gaseous fuel injection modeling using a gaseous sphere
injection methodology. SAE tech paper 2006-01-3265; 2006.

233

[12] Baratta M et al. Multi-dimensional modeling of direct natural gas injection and
mixture formation in a stratied-charge SI engine with centrally mounted
injector. SAE Int J Engines 2009;1:60726.
[13] Andreassi L, Facci AL, Ubertini S. Numerical simulation of gaseous fuel
injection: a new methodology for multi-dimensional modeling. Int J Numer
Method Fluids 2010;64:60926.
[14] Baratta M, Catania AE, Pesce FC. Computational and experimental analysis of
direct CNG injection and mixture formation in a spark ignition research
engine. ASME paper ICEF2010-35103, 2010 fall technical conference of the
ASME internal combustion engine division, September 1215, 2010, San
Antonio, TX, USA; 2010.
[15] Scarcelli R et al. CFD and optical investigation of uid dynamics and mixture
formation in a DI-H2 ICE. ASME paper ICEF2010-35084, 2010 fall technical
conference of the ASME Internal combustion engine division, September 12
15, 2010, San Antonio, TX, USA; 2010.
[16] Sukumaran S, Kong SC. Numerical study on mixture formation characteristics in
a direct-injection hydrogen engine. Int J Hydrogen Energy 2010;35:79918007.
[17] Douailler B et al. Direct injection of CNG on high compression ratio spark
ignition engine: numerical and experimental investigation. SAE tech paper
2011-01-0923; 2011.
[18] Scarcelli R et al. Numerical and optical evolution of gaseous jets in direct
injection hydrogen engines. SAE tech paper 2011-01-0675; 2011.
[19] Scarcelli R et al. Mixture formation in direct injection hydrogen engines: CFD
and optical analysis of single- and multi-hole nozzles. SAE Int J Engines
2011;4:236175.
[20] Scarcelli R et al. CFD and X-ray investigation of the characteristics of underexpanded gaseous jets. In: COMODIA 2012 the eight international
conference on modeling and diagnostics for advanced engine systems, July
2326, 2012, Fukuoka, Japan; 2012. p. 36873.
[21] Scarcelli R et al. High-pressure gaseous injection: a comprehensive analysis of
gas dynamics and mixing effects. In: ASME paper ICEF 201292137, fall
technical conference of the ASME internal combustion engine division,
September 2326, 2012, Vancouver, Canada; 2012.
[22] Li G et al. Optimization study of pilot-ignited natural gas direct-injection in
diesel engines. SAE tech paper 1999-01-3556; 1999.
[23] Li Y et al. Characteristic and computational uid dynamics modeling of highpressure gas injection. J Eng Gas Turbine Power 2004;126:1927.
[24] Baratta M et al. Numerical and experimental analysis of mixture formation
and performance in a direct injection CNG engine. SAE tech paper 2012-010401; 2012.
[25] Adamson Jr TC, Nicholls JA. On the structure of jets from highly
underexpanded nozzles into still air. J Aerosp Sci 1959;26:1624.
[26] Crist S, Sherman PM, Glass DR. Study of highly underexpanded sonic jet. AIAA J
1966;4:6871.
[27] Chuech SG, Lai MC, Faeth GM. Structure of turbulent underexpanded free jets.
AIAA J 1989;27:54959.
[28] Ferziger JH, Peric M. Computational methods for uid dynamics. 3rd ed. Berlin,
Heidelberg, New York: Springer-Verlag; 2002.
[29] LeVeque RJ. Finite volume methods for hyperbolic problems. Cambridge:
Cambridge University Press; 2002.
[30] Vuorinen V et al. Large-eddy simulation of highly underexpanded transient gas
jets. Phys Fluids 2013;25:016101.
[31] Rona A, Zhang X. Time accurate numerical study of turbulent supersonic jets. J
Sound Vib 2004;260:297321.
[32] Wilcox DC. Turbulence modeling for CFD. 1st ed. La Canada: DCW Industries
Inc.; 1993.
[33] Lehnasch G, Bruel P. A robust methodology for RANS simulations of highly
underexpanded jets. Int J Numer Methods Fluids 2008;56:2179205.
[34] Favre A. Equations des Gaz Turbulents Compressibles. J Mecanique
1965;4(3):36190.
[35] Birkby P, Page GJ. Numerical prediction of turbulent underexpanded sonic jets
using a pressure-based methodology. Proc Inst Mech Eng Part G
2001;215:16573.
[36] Tam CKW. The shock-cell structures and screech tone frequencies of rectangular
and non-axisymmetric supersonic jets. J Sound Vib 1988;121:13547.
[37] Eckbreth AC. Laser diagnostics for combustion temperature and species. 2nd
ed. Amsterdam, The Netherlands: Gordon and Breach Publishers; 1996.
[38] Zhao H, Laddomatos N. Optical diagnostics for in-cylinder mixture formation
measurements in IC engines. Prog Energy Combust Sci 1998;24:297336.
[39] Reboux J, Puechberty D, Dionnet F. A new approach of planar laser induced
uorescence applied to fuel/air ratio measurement in the compression stroke
of an optical S.I. engine. SAE paper no. 941988; 1994.
[40] Reboux J, Puechberty D, Dionnet F. Study of mixture in-homogeneities and
combustion development in a S.I. engine using a new approach of laser
induced uorescence (FARLIF). SAE paper no. 961205; 1996.
[41] Hishinuma H et al. Development of a technique for quantifying in-cylinder A/F
ratio distribution using LIF image processing. JSAE Rev 1996;17:2559.
[42] McGee J et al. Evaluation of a direct-injected stratied charge combustion
system using tracer PLIF. SAE tech paper 2004-01-0548; 2004.

Web references
[43] www.ingas-eu.org InGas Project website (accessed 03.08.12).
[44] www.cd-adapco.com/products/star_cd CD-adapco STAR-CD
(accessed 04.07.13).

website

You might also like