You are on page 1of 17

Simulation of high pressure, direct injection

processes of gaseous fuels by a density-


based OpenFOAM solver
Cite as: Phys. Fluids 33, 066104 (2021); https://doi.org/10.1063/5.0054098
Submitted: 14 April 2021 • Accepted: 27 May 2021 • Published Online: 15 June 2021

Francesco Duronio, Stefano Ranieri, Andrea Di Mascio, et al.

ARTICLES YOU MAY BE INTERESTED IN

Large-eddy simulation of highly underexpanded transient gas jets


Physics of Fluids 25, 016101 (2013); https://doi.org/10.1063/1.4772192

Numerical investigation of the flow characteristics of underexpanded methane jets


Physics of Fluids 31, 056105 (2019); https://doi.org/10.1063/1.5092776

LES of H2-air jet combustion in high enthalpy supersonic crossflow


Physics of Fluids 33, 035133 (2021); https://doi.org/10.1063/5.0040398

Phys. Fluids 33, 066104 (2021); https://doi.org/10.1063/5.0054098 33, 066104

© 2021 Author(s).
Physics of Fluids ARTICLE scitation.org/journal/phf

Simulation of high pressure, direct injection


processes of gaseous fuels by a density-based
OpenFOAM solver
Cite as: Phys. Fluids 33, 066104 (2021); doi: 10.1063/5.0054098
Submitted: 14 April 2021 . Accepted: 27 May 2021 .
Published Online: 15 June 2021

Francesco Duronio,a) Stefano Ranieri, Andrea Di Mascio, and Angelo De Vita

AFFILIATIONS
 degli studi dell’Aquila - Piazzale Ernesto Pontieri, Monteluco di Roio, 67100 L’Aquila (AQ), Italy
Universita

a)
Author to whom correspondence should be addressed: francesco.duronio@univaq.it

ABSTRACT
The direct injection of a gaseous fuel in internal combustion engines involves under-expanded supersonic jets and complex air/fuel fluid-
dynamics. Furthermore, with the high pressure ratios between the injector and the cylinder, the gaseous flow usually becomes choked even inside
the injector. Knowledge of all these phenomena is essential to achieve a deeper understanding of the air–fuel mixing process that follows, influenc-
ing combustion and pollutant formation. In this framework, this study deals with the development and validation of a fully explicit, density-based
solver for supersonic compressible flows, using the OpenFOAM library and featuring Runge–kutta fourth order time discretization and the
Kurganov central flux splitting scheme. This methodology was applied to analyze the inner and the external flow of an innovative, multi-hole,
high pressure injector for heavy-duty vehicle applications. The adoption of multi-hole patterned injectors in gaseous fuel combustion systems is
believed to be an efficient way of achieving a better air/fuel mixture and, therefore, improving the combustion reaction. The present work aims to
evaluate the reliability of the aforementioned mathematical approach for such kinds of complex flows and, especially, provide a comprehensive
characterization of the multi-jet spray. It was found that shock waves in the internal-nozzle deeply modify the flow development and the external
Mach disk as shock cells move the mixing activity on a lateral shear layer. It was also observed that a methane cloud grows downstream and,
although flammable conditions are present, it later inhibits air recirculation toward the near nozzle zone.
Published under an exclusive license by AIP Publishing. https://doi.org/10.1063/5.0054098

I. INTRODUCTION the vehicle market. In 2019, electric cars accounted for only 2.6% of
The harmful impact of internal combustion engine (ICE) emis- global car sales,2 and so it is essential to continue reducing tailpipe
sions on both humans and the environment has led to the introduc- emissions and improving fuel savings for conventional ICEs, by
tion of increasingly stringent regulations in transportation. Regarding improving existing technologies and developing new ones. Among the
exhaust emissions of greenhouse gases (GHG), in Europe the most different solutions, improvements in fuel economy have been made
stringent standards have been set with the aim of reducing CO2 emis- possible by using the most recent direct injection engines,3–5 either on
sions by 2030. Furthermore, several post Euro-6 regulation changes their own or in combination with other existing technologies such as
were considered, including tightening limits on CO, NOx, unburnt the Miller cycle, advanced turbocharging, improved thermal manage-
hydrocarbons (HC) and particulate matter (PM) emissions, as well as ment,6 water injection, and lean homogeneous combustion (pre-
adding the emission measurements of several new gas species and the chamber jet ignition). The use of cleaner fuels in these advanced
introduction of a new urban real drive emission (RDE) cycle.1 engines can further reduce tailpipe emissions.1
Vehicle industries must undergo significant and rapid transfor- Compressed natural gas (CNG) is commonly recognized as an
mations to meet the normative requirements of reducing GHG and attractive fuel for both spark-ignition (SI) and compression-ignition
pollutant emissions. However, if hybrid and electric vehicles are con- (CI) engines. CNG has a higher-octane number and hydrogen/carbon
sidered to be the most promising long-term solution, due to their zero ratio, can be produced in renewable ways, and is more widespread and
tailpipe emissions and the possibility of producing electricity from cheaper than liquid fossil fuels.7,8
renewable sources, issues regarding storage batteries, driving range, There are at least three options for building a natural gas ICE:9
and lack of refueling stations and costs may hinder their spread across bi-fuel gasoline/NG engines, which involves converting a conventional

Phys. Fluids 33, 066104 (2021); doi: 10.1063/5.0054098 33, 066104-1


Published under an exclusive license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

SI engine to burn a pre-mixed mixture and adding a fuel feeding sys- the mass flow rate, using Dalton’s law, showing that this is indepen-
tem for CNG, with a limit to the maximum compression ratio that can dent of chamber pressure because the natural gas flow is choked at the
be achieved; dual-fuel diesel/NG engines, which involves converting a nozzle exit. Furthermore, mass flow rate increases at higher injection
CI engine by replacing the fuel system and adding a pilot quantity of pressures, allowing the required amount of fuel to be injected.
diesel fuel injected near the top dead center (TDC) to ignite the air-gas Investigations of a jet issuing from a multi-hole injector can be
mixture, with higher compression ratio and better fuel economy; and challenging due to the complex physical phenomena that occur in the
dedicated NG vehicles, involving the design of a new engine for the near-nozzle zone for under-expanded jets and the interaction of the
specific purpose of burning natural gas only, with high design individual plumes.18–22
flexibility. In an early study, Salazar et al.18 investigated the evolution of the
Typically, for all of the above configurations, natural gas is deliv- mixing process of a nitrogen/hydrogen mixture for several multi-hole
ered in the air intake port. Due to its low density, the gaseous fuel injectors by means of Planar Laser-Induced Fluorescence (PLIF). For a
occupies 4–15% of the intake passage volume, resulting in a significant 13-hole nozzle, they observed that all jets merged into a single col-
reduction in volumetric efficiency when compared to liquid fuels, and lapsed jet, immediately downstream of the under-expanded zone. This
this causes a reduction in engine power.10 A plausible solution could phenomenon is called the Coanda Effect, and it stems from a lack of
lie in delivering NG directly to the combustion chamber, namely, air entrainment due to the close proximity of interacting individual
CNG direct injection (CNG-DI). Moreover, CNG-DI technologies plumes. The data from such experimental studies were used by
permit both homogeneous and stratified combustion, leading to an Scarcelli et al.19 to validate numerical results obtained with a CFD-
increase in combustion efficiency for every engine load condition.11 RANS simulation. While CFD results accurately predict the jet pene-
However, no CNG-DI vehicle is currently commercial, since the tration and the fuel distribution for all the multi-hole nozzles where
complex mixture formation and combustion processes have not yet no jet-to-jet interaction was observed, a worse agreement with experi-
been well-understood or adequately developed. mental data was recorded for the 13-hole nozzle. This was due to the
Mixture formation in a CNG-DI engine depends, among other strong influence that the presence of under-expanded structures
parameters, on the ratio between the total pressure upstream of the exerted on the mixing process and on the plume-to-plume interaction.
injector nozzle and the in-cylinder pressure [often referred to as net They concluded that increasing the grid resolution and performing a
pressure ratio (NPR)], the nozzle geometry, and the injector type.12 Large Eddie Simulation (LES) could improve the accuracy of numeri-
Choosing these parameters correctly is crucial for the proper design of cal prediction. Friedrich et al.20 studied the jet-jet-interaction of the
the CNG injector. Due to the low density and compressibility, natural individual plumes issuing from a multi-hole injector. Nitrogen was
gas presents a slower penetration rate, which could be increased by injected into an optically accessible single-cylinder engine, and Particle
maintaining the injection momentum of the jets as high as possible. Image Velocimetry (PIV) and Laser-Induced Fluorescence (LIF) were
This requires a suitably high injection pressure. When a compressible simultaneously used to quantify air entrainment velocity and fuel con-
fluid is injected at high pressure into an ambient lower pressure, centration. Among other results, they highlighted the strong impact
depending on the NPR, complex shock structures may appear, which that under-expanded structure formations (Mach disk) and free jet
affect the mixing process.13 In order to inject enough gas for every interactions have on both flow velocity and mixture quality.
operating engine condition and to ensure the possibility of both homo- Furthermore, Twellmeyer et al.21,22 have shown, by means of RANS
geneous and stratified combustion strategies, large flow areas of the numerical simulations, that for high pressure ratios between the injec-
injector nozzles are required.12 In this regard, it is not completely clear tor and the cylinder, a gaseous flow can become chocked, even inside
which type of injector (single-hole, multi-hole, or hollow cone) is most the injector. For a multi-hole injector, this leads to a complex and
suitable for the direct injection of natural gas. Anyhow, multi-hole asymmetric flow-structure inside each nozzle that affects the jet forma-
injectors allow part of the injected gas to be directed to specific areas tion immediately downstream from the nozzle exit. Further investiga-
of the combustion chamber, making them particularly suitable for tions on the effects of the in-nozzle under-expanded structure
optimizing the formation of the mixture and, therefore, suitable for formation and the subsequent air/fuel interaction at the nozzle exit
CNG-DI engines.14–17 can provide valuable information for improving the mixing of a gas-
Chiodi et al.15 have investigated the characteristics of a CNG jet eous fuel issuing from a multi-hole injector. In this study, a fully
issuing from both single and multi-hole GDI injectors, by means of explicit, density-based solver of supersonic compressible flows (featur-
experimental measurements into a single-cylinder engine and numeri- ing Runge–kutta fourth order time discretization and the central flux
cal simulations. They have shown that a multi-hole injector is more splitting scheme of Kurganov) and an appropriately fine computa-
suitable for mixture homogenization or stratification than single-hole tional grid were used. An LES turbulence framework, multi-species
injectors. A numerical investigation into the effects of combustion transport model, and variable thermophysical properties were adopted
chamber geometry, injection parameters, cylinder head shape, and in order to properly investigate the inner and near nozzle flow of a
injector type on the mixing of air-methane in a CNG-DI engine was multi-hole prototype injector. The same numerical model had been
also conducted by Yadollahi et al.16 They found that a multi-hole previously applied by the authors for investigating the far-field struc-
injector delivers a slightly more flammable, stratified mixture than a ture of a methane spray injected by the same injector.23 However, as
single-hole injector. The larger flammable mass fraction in multi-hole far as the authors know, this is the first time that such a numerical
injection can be attributed to a wider jet spread and thus higher levels approach has been used for investigating the internal flow of a multi-
of mixing. Erfan et al.17 used the schlieren photography technique to hole gaseous injector. The results show the jet evolution inside the
investigate the effect of varying injection parameters on the structure injector nozzle, the formation of shock waves, and the evolution of the
of a CNG jet issuing from a multi-hole injector. They also calculated under-expanded, down-flow structures in the near nozzle zone,

Phys. Fluids 33, 066104 (2021); doi: 10.1063/5.0054098 33, 066104-2


Published under an exclusive license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

achieving a deeper understanding of the air–fuel mixing process that This function is valid between a lower and upper limit at temper-
strongly influences the combustion and pollutant emissions. ature Tl and Th, respectively. Two sets of coefficients a0 … a4 are
specified, the first set for temperatures above a common tempera-
II. NUMERICAL METHODOLOGY
ture Tc (and below Th), and the second for temperatures below Tc
A. Governing equations of compressible flows (and above Tl).
In order to properly describe the physics of the injection process,
a new solver was developed in the OpenFoam library. Considering In order to study the mixing of multiple species after the injection
compressible flows in the presence of shock waves and strong disconti- process, the set of governing Eqs. (1)–(4) is completed by a transport
nuities, the governing equations (mass, momentum and total energy equation for each species
conservation) are solved using the conservative variables q, qu, and qe
@ ðqYi Þ
as reported below:24,25 þ r  quYi ¼ r  ðqDi rYi Þ; (6)
@t
@q
þ r  ðuqÞ ¼ 0; (1) where, for the i-species, Yi is concentration, and Di is the diffusivity,
@t
equal to
@ðquÞ
þ r  ½uðquÞ þ p ¼ r  r; (2) l
@t Di ¼ ;
@ðqeÞ qSc
þ r  ½uðqeÞ þ p ¼ r  ðr  uÞ þ r  q; (3)
@t with Schmdit number Sc equal to 0.7, after Vuorinen et al.27
where:
• p is pressure, which is assumed to be coupled to density and tem- B. Discretization and solution
perature via the perfect gas state equation The simulation of high-speed, under-expanded jets requires
p numerical discretization that can capture flow discontinuities such as
¼ RT; (4) Mach disks and shock waves and, at the same time, avoids unwanted
q
oscillations. Methodologies based on Riemann solvers, such as the
• e is the total energy given by the sum of internal Ui and kinetic Piecewise Parabolic Method (PPM) and Weighted Essentially Non-
energy Oscillatory (WENO), give an efficient reproduction of compressible
1 flow but have important drawbacks. Such approaches involve charac-
e ¼ Ui þ jju2 jj; teristic decomposition and Jacobian evaluation, which make them
2
complex and difficult to be implemented in co-located, unstructured
with Ui being expressed as grids that, instead, allow us to discretize complex geometries and are
Ui ¼ cv T; much more flexible than structured grids.28,29 Valid alternative
approaches were developed; one of them used the so-called central
cv represents constant volume specific heat and T is temperature; schemes formulations of Kurganov (KNP) and Kurganov and Tadmor
• r is the stress tensor that, for a Newtonian fluid is equal to (KT).30,31 These methods are non-staggered second order central
h i   schemes that evaluate the fluxes on the cell faces using the values at
2
r ¼ l ru þ ðruÞT  lr  u I ¼ 2ldevðDÞ ; the cell center. The cell to face flow interpolation is subdivided into an
3
inward and outward direction with respect to the face owner cell. A
with l dynamic viscosity. full and detailed description of KNP and KT central schemes, used in
D strain tensor: the developed solver, can be found in Ref. 28.
h i In order to ensure the simulation stability and to avoid solution
1
D¼ ru þ ðruÞT ; oscillations, OpenFOAM allows the adoption of flux limiters and so a
2 VanLeer flux limiter was adopted in the performed simulations.28,32
• q is the heat flux vector computed as Previous studies demonstrated that, in order to properly simulate
under-expanded jets, higher order time discretization methods, such
q ¼ krT; as Runge–Kutta fourth (RK4), are required.25,27 Similar to the previ-
ously cited works, the classic RK4 explicit method was implemented in
where T is the fluid temperature and k is the heat conduction
the newly developed OpenFOAM solver for integrating the solution
coefficient computed as
over time. The time-stepping coefficients of the RK4 method that were
lcp
k¼ ; chosen were33
Pr
1 1
with cp being the constant pressure specific heat and Pr is the a1 ¼ a4 ¼ ; a2 ¼ a3 ¼ : (7)
6 3
Prandtl number.
cp is computed as a function of temperature from JANAF tables26 A complete description of the method can be found in Refs. 25
with the following relation: and 33. The developed solver was summarized in accordance with the
flow chart shown in Fig. 1. It illustrates all of the operations performed
cp ¼ Rðððða4 T þ a3 ÞT þ a2 ÞT þ a1 ÞT þ a0 Þ: (5) during a generic time step and for each RK4 step: the q; u; e fields are

Phys. Fluids 33, 066104 (2021); doi: 10.1063/5.0054098 33, 066104-3


Published under an exclusive license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

updated, the relative fluxes are recomputed, and the turbulence model
is solved.
Finally, the temporal integration of species and turbulence trans-
port equations was performed using a second-order backward implicit
method.

C. Turbulence modeling
The solution of the governing Eqs. (1)–(3) is valid when the com-
putational grid is fine enough to resolve all of the flow scales.27 This
would be a Direct Numerical Simulation (DNS) of the flow which,
today, is not affordable because it is too complex and time demanding.
So, other modeling techniques, such as Large Eddy Simulation (LES),
are preferred for under-expanded jet simulations.24,25,27
The concept behind the LES approach is to model the smaller
scales, which are universal and not affected by the flow geometry,
while explicitly solving the larger ones. This is performed by mathe-
matically filtering the governing equations and introducing the so-
called Sub-Grid Stress (SGS) tensor (ssgs).34
The SGS term modeling involves an eddy viscosity approxima-
tion. A one-equation eddy viscosity model for compressible flows was
adopted.35,36 This kind of approach, which is different from zero equa-
tion models, uses a transport equation to compute the local SGS
kinetic energy ksgs. Then, the sub-grid scale eddy viscosity  sgs is calcu-
lated using the ksgs field and the filter dimension D (usually evaluated
from the grid size) according to the following relation:
qffiffiffiffiffiffiffi
 sgs ¼ Ck D ksgs ; (8)

where Ck is a model constant whose default value is 0.094.


Finally, using this parameter, an effective viscosity (leff) is
obtained and used for the solution of the governing Eqs. (1)–(3).25
The molecular viscosity is computed by Sutherland’s law as a function
of temperature
pffiffiffiffi
As T
lm ¼ ; (9)
1 þ Ts =T
As and Ts are specific constants of the single species (Table I).37
The developed code was validated using the shock tube case.38
Full details and validation results are reported in Appendix.
III. CASE STUDY SETUP: MULTI-HOLE INJECTOR
The simulations that were performed dealt with an innovative,
prototype, high-pressure multi-hole, inwardly opening injector, specif-
ically designed and realized for the direct injection of compressed nat-
ural gas (maximum injection pressure of 50 bar) in heavy-duty
applications (Fig. 2).
The device has interchangeable heads, each characterized by
different nozzle geometries, which drive the flow into a particular

TABLE I. Sutherland’s law As and Ts constants, adopted values.

Species As ðkg  m1  s1  K0:5 Þ Ts ðKÞ

N2 1:4067  106 111.0


FIG. 1. Flowchart representing the developed solver structure and summarizing the CH4 1:2529  106 197.8
operations performed in each calculation step.

Phys. Fluids 33, 066104 (2021); doi: 10.1063/5.0054098 33, 066104-4


Published under an exclusive license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

FIG. 2. Silhouette of the high-pressure injector.


FIG. 4. Current command signal and nozzle inlet pressure time behavior.
multi-hole spray configuration. The particular injector head investi-
gated in this study is characterized by a central hole and six radially, temperature was fixed at 293.15 K, and the intensity of the turbu-
equally spaced, slots. Such a component was analyzed by means of x- lent kinetic energy was fixed at 0.05. The mass fraction of meth-
ray tomography, in order to define the exact internal geometry shown ane entering the system was fixed at unity (CH4 ¼ 1) and,
in Fig. 3. This information was used to create a CAD model of the noz- consequently, N2 ¼ 0.
zle geometry. • Walls: walls of the injection environment.
The injection process was controlled using a programable • NozzleWalls: nozzle internal walls. A slip condition was set for
Electronic Control Unit (PECU) with the energizing current profile velocity. Considering the injection pressure values and, as a con-
shown in Fig. 4 (blue curve). Following magnetic and mechanical sequence, the high-velocity flow generated, accurate modeling of
opening delays, the injector needle begins to lift, and the pressure at the in-nozzle viscous boundary layer requires significant spatial
the nozzle inlet increases. The time behavior of such a parameter was resolution locally. Therefore, common practice in the literature is
measured and provided by the injector manufacturer. It is represented
by the red curve in Fig. 4.
The computational domain is represented in Fig. 5.
Boundary conditions are:
• Inflow: this is the section at the top of the divergent duct where
the injected gas entered the domain. The manufacturer provided
the time-behavior of the pressure at this section, as represented
in Fig. 4. Consequently, using the OpenFOAM boundary condi-
tions, a uniformFixedValue was imposed for the pressure while
a zeroGradient was used for the velocity. The value of the

FIG. 3. X-ray tomography of the injector showing its internal geometry. FIG. 5. The boundary conditions.

Phys. Fluids 33, 066104 (2021); doi: 10.1063/5.0054098 33, 066104-5


Published under an exclusive license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

TABLE II. Case study initial conditions. TABLE III. Refinement regions details.

Ambient pressure 1 bar Refinement region Refinement level


Ambient temperature 293.15 K
Duct 5
Disk 5
to consider a slip condition on the solid wall in order to avoid Bulk 1 4
the formation of any artificial boundary layer27,39 Bulk 2 3
Bulk 3 2
The integration domain was initialized with the parameters Bulk 4 1
reported in Table II.
The geometry was discretized with two unstructured grids, gener-
ated using the snappyHexMesh tool embedded in OpenFOAM. The
IV. RESULTS
two meshes were topologically equal, and, in particular, the fine grid
had base dimensions of 1.6  1.6  1.6 mm, while the coarse grid’s The simulation allowed us to describe many aspects of the flow,
base dimensions were 3.2  3.2  3.2 mm. Figure 6 and Table III both in the injector’s internal ducts and the mixing region, which
show details of the refinement regions. would otherwise be very difficult to obtain.
The meshing strategy was based on the adoption of refinement The first set of result is regarding model validation. Figure 7
regions with different refinement levels. The cell dimensions are given shows the comparison of experimental and numerical axial penetra-
by the following relation: tion of the jet over time (for further information see Montanaro
et al.23). The uncertainty of the numerical results was assessed in
dxb
dxi ¼ ; (10) accordance with the classical paper by Roache,40 in which the uncer-
2RL tainty assessment procedures now adopted by AIAA, ITTC, and IEEE
where dxi is the dimension of the grid in the generic refinement region, are described and discussed in detail. Following these recommenda-
dxb is the base grid dimension, and RL is the refinement level. tions, the solution was computed on two grid levels, the finer obtained
In this way, for the fine grid, the near-nozzle zone was discretized with a refinement ratio equal to 2. A three-grid level assessment was
with elements of 50 lm then, as shown in Table III the size gradually impossible because another finer grid with a refinement ratio of 2
increased up to base dimensions. The total cell count was of 40 M. would consist of about 320 M points, which was beyond the computa-
The time step adopted was variable and depended on the Courant tional resources available to us. On the contrary, grid coarsening with
number, which was set to be lower than 0.8. the same ratio would yield too coarse a mesh and be unable to prop-
Numerical results were then post-processed using Paraview erly capture the main flow features. The uncertainty was then esti-
software. mated by Richardson extrapolation of the first two computed

FIG. 6. The computational domain grid.

Phys. Fluids 33, 066104 (2021); doi: 10.1063/5.0054098 33, 066104-6


Published under an exclusive license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

simulations. Verification and validation, obtained by comparing the


simulations with an analogous physical experiment, will help to quan-
tify the reliability of the reported numerical results. The morphology
and evolution of the spray are well represented, the interaction
between the jets being properly reproduced, as well as the methane
cloud that surrounds the far-field. The volumetric evolution of the jet
resembles the one recorded by Montanaro et al.,23 demonstrating the
validity of the numerical approach adopted and, in particular, the tur-
bulence model chosen.
A very complex flow field also appears within the injector, as
shown by the temporal sequence in Fig. 10. The Mach number con-
tours are reported on the left hand side, together with the logarithm of
density gradient on the right.
As a matter of fact, the apparatus consists of an initial duct, which
branches out into seven smaller ducts: a central circular duct with a
constant diameter of 0.66 mm and six larger inclined ducts, whose sec-
tion is a circular crown with a hydraulic diameter varying in a
stream–wise direction, with a value Oð1  3Þ mm, that drives the mix-
ture into the main chamber.
FIG. 7. Experimental and numerical axial jet tip penetration vs time. It should be noted that the jet evolution is not steady. Indeed, as
shown in Fig. 4, the inlet section pressure varies in time and, therefore,
solutions; then, given the two-grid assessment procedure a safety fac- the actual flow is characterized by a complex transient evolution.
tor (as suggested in Ref. 40) was used. Nevertheless, the transient flow can be described as a succession of
In Fig. 7, both numerical and experimental jet tip penetration is quasi-steady state problems, which is useful in understanding some
illustrated with the estimated uncertainties. As can be observed, the relevant features. Initially, a normal shock travels within the nozzle’s
agreement between the penetrations measured from the experimental ducts, propagating sonic conditions across the domain. Then, after
images and those computed from the simulations is very good and approximately 0.5 ms a steady-state condition is reached, and the flow
well within the combined numerical uncertainty for the whole of the structure should resemble the one sketched in Fig. 11; the supersonic
simulated period. flow expands and is accelerated into the diverging nozzle.
Figure 8 presents a comparison of the experimental picture in Corresponding to the section where flow branches out into the smaller
Ref. 23, recorded by means of schlieren imaging, against the numerical ducts, due to the change of hydraulic diameter, a normal shock is rec-
results of the logarithm of density gradient. ognizable across the central duct. Such features initially reestablish the
The temporal evolution of the jet is reported in Fig. 9; the numer- subsonic condition but, later, the flow accelerates back to the sonic
ically reproduced spray is represented by a methane concentration condition.
(CH4) volumetric rendering. Figure 12 presents quantitative insights about this evolution by
The flow in the main chamber is characterized by a very complex means of a Cartesian plot of the Mach number along the axis of the
mixing region. There are no simple theories capable of describing the nozzle’s central duct, starting from the inlet section.
jet development, and so the flow description must rely on the Conversely, a more complex flow structure, consisting of a series
of oblique shocks, appears in the side ducts; these shocks are required
to divert the supersonic flow along the lateral duct and, given the
Mach number of the approaching flow and the deflection angle (very
small), they are expected to be attached to the corner. Therefore, the
flow reduces its speed across the oblique shocks but remains
supersonic.
Figure 13 also reports the Mach number along the axis of the
nozzle for an oblique duct; this time the plot begins from the section
where the duct branches off. After that, the brief transient extinguishes,
and the oblique shocks are clearly recognizable. The first shock takes
place about 1 mm downstream and, then, their intensity diminishes
downstream of the axis of the duct. The oblique shocks are weaker
than the normal one presented on the central ducts and, for this rea-
son, the Mach number is approximately 3.5 on the circumferential
ducts. On the central duct, after 0.7 ms, the Mach number reaches a
FIG. 8. Comparison of experimental Schlieren image recorded by Montanaro
maximum value of 1.3.
et al.23 with the numerical results plotted as methane mass fraction volumetric ren-
dering at 1.0 ms after the SOI. The experimental Schlieren image is reproduced As expected from the upstream-to-downstream pressure ratio,
with permission from SAE Technical paper 2020–01–2124, 2020, copyright SAE the external flow is also characterized by the appearance of under-
International. expanded structures developing during the whole of the injection

Phys. Fluids 33, 066104 (2021); doi: 10.1063/5.0054098 33, 066104-7


Published under an exclusive license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

FIG. 9. Temporal evolution of the spray development region represented by methane concentration (CH4) volume rendering.

period. Such features are important because they inhibit the air–fuel were not reached. Indeed, upstream within the nozzle, the pressure is
entrainment, as will be shown in the following. Figure 14 reports the lower than the central channel, due to the presence of the oblique
evolution of the logarithm of the density gradient obtained by slicing shocks that dissipate pressure energy and accelerate the fluid, creating,
the domain with an axial plane. 0.2 ms, after the SOI, the jet is outside, shock diamonds with converging oblique shocks. Besides this,
completely subsonic. In the second snapshot at t ¼ 0.5 ms, the circumferential holes have a circular crown section, which creates a
upstream pressure increases and shock diamonds appear downstream particular geometry of the shock wave.
from the central nozzle. The oblique shocks become a Mach disk after For both configurations, a potential core (with 100% methane
0.7 ms, the latter lasting until the end of the simulated period. concentration) extends for various diameters downstream, as shown
Different behavior is recognizable in the outflow from the circumfer- in the sequence in Fig. 9. The potential core inhibits the air–fuel mix-
ential ducts, where the conditions for the appearance of Mach disks ing, which only begins downstream.

Phys. Fluids 33, 066104 (2021); doi: 10.1063/5.0054098 33, 066104-8


Published under an exclusive license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

FIG. 11. Internal flow structure.

shock, in terms of both amplitude and length. This is to be ascribed to


the fact that the jet structure changes as the injection pressure
increases (see Fig. 14). The Mach number, upstream from the shock, is
almost equal to 4 and, after, it reaches at a maximum of 1.5, this dem-
onstrates the presence of a completely developed under-expanded jet.
Downstream from the Mach disk, the presence of further shock waves
creates fluctuations in the Mach number.
Figure 16 shows the temporal evolution of the vortical structures
represented by Q-criterion iso-surfaces, colored by total turbulent
kinetic energy (TKE). As can be observed, at the beginning of the
injection, the spray’s TKE is quite small, no great vortex structures
appear and the methane is confined within the potential core. As the
injection continues, a larger vortex grows downstream, owning to

FIG. 10. Temporal evolution of the spray within the injector’s ducts with a multi-plot
of Mach number on the left hand side and logarithm of density gradient [logðrqÞ]
on the right side of each picture.

Figure 15 is a plot of the temporal evolution of Mach number for


the central jet along the vertical axis. The curve at t ¼ 0.5 ms shows the
oscillations of the thermodynamic proprieties across the shock dia-
monds. The amplitude of the oscillation of all the thermodynamic
fields is reduced to finally reach the ambient values at a distance from
the hole exit of about 20 mm. The appearance of the Mach disk FIG. 12. Mach number spatial and temporal evolution in the central duct of the
reduces the flow propriety fluctuations downstream from the principal injector.

Phys. Fluids 33, 066104 (2021); doi: 10.1063/5.0054098 33, 066104-9


Published under an exclusive license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

FIG. 13. Mach number spatial and temporal evolution in the oblique lateral ducts of FIG. 15. Numerical evaluation of Mach number plotted along the vertical axis for
the injector. the central jet.

greater TKE. This behavior enhances the mixing process, which is turbulent mixing processes. High values of v denote high fuel concen-
now relevant and allows the disruption of the potential core, creating tration gradients while, conversely, low values indicate zones where
the conditions for an effective air–fuel mixing. the fuel concentration is approximately constant.
This is also confirmed by plotting the scalar dissipation rate Figure 17 reports the iso-lines representing flammable limits and
(SDR). Such a parameter, computed as v ¼ ljrCH4 j, is relevant in the stoichiometric conditions, in addition to the SDR plot.

FIG. 14. Transient evolution of under-expanded structures.

Phys. Fluids 33, 066104 (2021); doi: 10.1063/5.0054098 33, 066104-10


Published under an exclusive license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

FIG. 16. Q-criterion contour colored by total kinetic energy (TKE).

As expected, the near nozzle zone is characterized by high values activity changes. Indeed, in the far field a large volume, characterized
of log ðvÞ, denoting the presence of potential cores rich in methane by flammable conditions, grows, revealing that the interaction between
while, further downstream, the methane concentration gradient is the two species is stronger.
lower and the SDR tends to zero. The mixing activity is strongly influenced by the nozzle geometry,
Furthermore, at t ¼ 0.6 ms the flammable region is a thin layer, which is a relevant design parameter to be considered in direct gaseous
which surrounds each spray jet but, as injection continues, the mixing injections. Figure 18 represents the velocity vectors superimposed onto

Phys. Fluids 33, 066104 (2021); doi: 10.1063/5.0054098 33, 066104-11


Published under an exclusive license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

FIG. 17. Scalar dissipation rate plot for


two temporal instants. High values of such
scalar identify zone where there is an
intense mixing activity.

the methane mass fraction: the single jets develop allowing, in a first process. Therefore, the mass weighted probability density function
phase, air recirculation within the jet from the bottom of the injection [PDFk ðCH4Þ] was calculated as shown in Eq. (11),
environment toward the top but, later, it is not the same. The growing 8
of the already-mentioned methane cloud in the far field inhibits such >
> m ¼ qi dVi CH4;i
> CH4;i
N >
< for CH4;k  CH4;i  CH4;kþ1
processes by not allowing air to come back toward the injector tip. 1 X (
This is an undesirable effect, because it obstructs air/fuel mixing, espe- PDFk ðCH4 Þ ¼ (11)
Mtot i¼1 >
> CH4;i < CH4;k
cially in the near nozzle zone and should be avoided. An optimized >0 for
>
: CH4;i > CH4;kþ1 ;
hole pattern could surely help to solve this issue by enhancing air/fuel
mixing and making it faster.
where
Achieving a quantitative estimation of the global mixture quality
has a relevant importance, especially with regard to the combustion • N is the total cell count;

FIG. 18. Representation of the velocity vector superimposed on the methane mass fraction for two temporal instants.

Phys. Fluids 33, 066104 (2021); doi: 10.1063/5.0054098 33, 066104-12


Published under an exclusive license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

• k is an integer in the interval ½0; …; 100;


• CH4;k ¼ k=100 for k ¼ 1; …; 100 and CH4;0 ¼ 0:001, is the dis-
crete random variable;
• PDFk ðCH4 Þ is the probability density function relative to CH4;k ;
• dVi is the ith cell volume;
• CH4;i is the ith cell fuel mass fraction;
• Mtot is the total injected mass at a certain temporal instant;
Such a parameter was integrated over different ranges of CH4;k in
order to estimate the percentage of a lean, flammable, and rich mix-
ture. A mixture with CH4 volume fraction lower than 0.044 (hereafter
for brevity named lean), i.e., with a mass fraction lower than 0.025, is
too lean to be ignited. A mixture with CH4 volume fraction higher
than 0.17 (hereafter named rich), i.e., with a mass fraction higher than
0.102, is otherwise too rich to be ignited. Therefore, a flammable mix-
ture has the fuel mass fraction in the range 0:025  CH4  0:102.41
Figure 19 shows the temporal evolution of such mixture catego-
ries. The rich mixture initially grows very quickly, reaching a value of
the unity after 0.1 ms; then, it begins to decrease because the mixing
activity becomes more and more intense.
Advancing toward t  1 ms flammable mixture is approximately
the 40% of the total mass of the fuel injected while the rich mixture is
the 60%. The mass of the lean mixture is small compared to the others
(approximately 5%).
Figure 20 shows the temporal evolution of the average value of
the fuel mass fraction and its relative variance. Both parameters have
the same behavior: after an initial growth they invert the trend,
decreasing and finally assuming a horizontal plateau. At the end of the
simulated period, the mean value is approximately 0.15, which is
almost included in the flammable range. This, together with decreasing
variance, demonstrates that the distribution has ever-increasing flam-
mability features.
V. CONCLUSIONS
The current work deals with the development of a high-order,
density-based solver and simulation of the spray issued from of an

FIG. 20. Average fuel mass fraction value in the injection environment and relative
variance.

innovative, high pressure, multi-hole device for the direct injection of


gaseous fuels.
Based on the numerical results, the following conclusions can be
pointed out:
• The developed approach is capable of reproducing a spray’s mor-
phology observed by means of the schlieren technique and the
jet-tip penetration experimentally measured.
• The flow discontinuities, as shock waves and Mach disks, are prop-
erly captured by the high order discretization schemes chosen.
• The spray development in the injection environment is well
FIG. 19. Percentage of injected fuel mass in lean, flammable, and rich conditions reproduced, for the whole injection duration, by the LES turbu-
during the injection period. lence model adopted.

Phys. Fluids 33, 066104 (2021); doi: 10.1063/5.0054098 33, 066104-13


Published under an exclusive license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

• The flow within injector ducts, after 0.5 ms, presents a quasi steady- TABLE IV. Validation case initial conditions.
state sequence of oblique and reflecting shocks which dissipate pres-
sure energy. Conversely, outside of the injection environment, spray Region p (bar) T (K) l (ms1)
has a transient nature, developing under-expanded structures, espe-
0  l  0:5 m 1.0 348.4 0
cially on the central hole where barrel shocks, shock diamonds and
Mach disks appear in sequence. Different behavior is recognizable for 0:5 < l  1 m 0.1 278.7 0
the circumferential holes where the geometry brings the formation of
particular, under-expanded structures and the net pressure ratio
• The fuel interacts with the surrounding air, creating a large meth-
inhibits the eventual presence of Mach disks.
ane cloud, which strongly influences the air-circulation within
It seems that the more relevant findings are: the spray’s plumes. On the one hand, mixing activity is relevant
• The under-expanded structures strongly affect the mixing activity but, on the other hand, the nozzle pattern should be improved in
in the near nozzle zone where potential cores rich in methane are order to achieve a stronger air recirculation capable of transport-
present. Air/fuel mixing begins further downstream, where the ing air from the bottom toward the injector tip, promoting mix-
largest eddies have the higher turbulent kinetic energy content. ing activity.

FIG. 21. Shock tube validation case: com-


parison of exact solution with the numeri-
cal computed pressure.

FIG. 22. Shock tube validation case: com-


parison of exact solution with numerical
computed density.

Phys. Fluids 33, 066104 (2021); doi: 10.1063/5.0054098 33, 066104-14


Published under an exclusive license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

• The air–fuel mixture quality was quantified through a statistical APPENDIX: CODE VALIDATION
approach. It was observed that after 1.0 ms 40% of the injected
mass is in a flammable condition while only a small part (5%) is Shock-tube is a common test case that is used in the literature to
in lean conditions. The overall average value of the mass fraction evaluate the accuracy of algorithms for compressible flow simulations
computed over the injection domain is almost embedded within and their capability for capturing shock waves and other typical fea-
the flammable limits. tures. The shock tube considered in this work was 1.0 m long, contain-
ing nitrogen and with the initial conditions listed in Table IV.
The domain was subdivided in two regions along the length
ACKNOWLEDGMENTS and initialized with different pressures and temperatures. The com-
putational domain was then discretized using a grid with 1000 ele-
All the simulations were performed with the developed solver ments uniformly spaced in the direction of the shocktube length;
on the Galileo cluster of HPC CINECA facilities within the Class C the time stepping value was 10–8 s. The simulation results after
Project named HPGAS. 0.6 ms were plotted (see Figs. 21–24) to show a complete agreement

FIG. 23. Shock tube validation case: com-


parison of exact solution with numerical
computed temperature.

FIG. 24. Shock tube validation case: com-


parison of exact solution with numerical
computed velocity magnitude.

Phys. Fluids 33, 066104 (2021); doi: 10.1063/5.0054098 33, 066104-15


Published under an exclusive license by AIP Publishing
Physics of Fluids ARTICLE scitation.org/journal/phf

19
with the exact solution for each computed field. Moreover, the code R. Scarcelli, T. Wallner, N. Matthias, V. Salazar, and S. Kaiser, “Mixture forma-
capability to capture shock-waves and the absence of solution oscil- tion in direct injection hydrogen engines: CFD and optical analysis of single-
and multi-hole nozzles,” SAE Int. J. Engines 4, 2361–2375 (2011).
lations are evident. 20
W. Friedrich, R. Grzeszik, M. Helmich, T. Bossmeyer, and M. Wensing,
DATA AVAILABILITY “Characterization of high pressure natural gas injections for stratified operation of
a spark ignited engine by simultaneous fuel concentration and air entrainment
The data that support the findings of this study are available measurements,” in 18th International Symposium on the Application of Laser and
within the article. Imaging Techniques to Fluid Mechanics (FAU, 2016), pp. 4–7.
21
A. Twellmeyer, F. K€ opple, and B. Weigand, “A novel CFD approach for model-
ling the high-pressure direct injection and mixture formation in a spark-ignited
REFERENCES CNG engine,” Int. J. Comput. Methods Exp. Meas. 4, 424–433 (2016).
1 22
A. Joshi, “Review of vehicle engine efficiency and emissions,” SAE Technical A. Twellmeyer, F. Kopple, and B. Weigand, “Evaluating different measures to
Paper No. 2020-01-0352 (2020). improve the numerical simulation of the mixture formation in a spark-ignition
2
“Global EV Outlook 2020–Analysis—IEA,” Technology Report, 2020. CNG-DI-engine,” SAE Technical Paper No. 2017-01-0567.(2017).
3 23
F. Duronio, A. De Vita, L. Allocca, and M. Anatone, “Gasoline direct injection A. Montanaro, L. Allocca, A. De Vita, S. Ranieri, F. Duronio, and G.
engines—A review of latest technologies and trends. Part 1: Spray breakup Meccariello, “Experimental and numerical characterization of high-pressure
process,” Fuel 265, 116948 (2020). methane jets for direct injection in internal combustion engines,” SAE
4
F. Duronio, A. De Vita, A. Montanaro, and C. Villante, “Gasoline direct injec- Technical Paper No. 2020-01-2124 (2020).
24
tion engines—A review of latest technologies and trends. Part 2,” Fuel 265, V. Vuorinen, A. Wehrfritz, J. Yu, O. Kaario, M. Larmi, and B. J. Boersma, “Large-
116947 (2020). eddy simulation of subsonic jets,” J. Phys.: Conf. Ser. 318, 032052 (2011).
5 25
F. Duronio, A. De Vita, L. Allocca, A. Montanaro, S. Ranieri, and C. Villante, A. Hamzehloo and P. G. Aleiferis, “LES and RANS modelling of under-
“CFD numerical reconstruction of the flash boiling gasoline spray expanded jets with application to gaseous fuel direct injection for advanced
morphology,” in Conference on Sustainable Mobility (SAE International, 2020). propulsion systems,” Int. J. Heat Fluid Flow 76, 309–334 (2019).
6 26
F. Fatigati, D. D. Battista, and R. Cipollone, “Design improvement of volumet- M. Chase, NIST-JANAF Thermochemical Tables, 4th ed. (NIST, 1998).
27
ric pump for engine cooling in the transportation sector,” Energy 231, 120936 V. Vuorinen, J. Yu, S. Tirunagari, O. Kaario, M. Larmi, C. Duwig, and B. J.
(2021). Boersma, “Large-Eddy simulation of highly underexpanded transient gas jets,”
7
G. T. Chala, A. R. A. Aziz, and F. Y. Hagos, “Natural gas engine technologies: Phys. Fluids 25, 016101 (2013).
28
challenges and energy sustainability issue,” Energies 11, 2934 (2018). C. J. Greenshields, H. G. Weller, L. Gasparini, and J. M. Reese,
8
M. I. Khan, T. Yasmin, and A. Shakoor, “Technical overview of compressed “Implementation of semi-discrete, non-staggered central schemes in a colo-
natural gas (CNG) as a transportation fuel,” Renewable Sustainable Energy cated, polyhedral, finite volume framework, for high-speed viscous flows,” Int.
Rev. 51, 785–797 (2015). J. Numer. Methods Fluids 63, 21 (2009).
29
9
M. Muralidharan, A. Srivastava, and M. Subramanian, “A technical review on H. Versteg and W. Malalasekera, in An Introduction to Computational Fluid
performance and emissions of compressed natural gas—diesel dual fuel Dynamics (Pearson Education Limited, Harlow, 2007).
30
engine,” SAE Technical Paper No. 2019-28-2390 (2019). A. Kurganov and E. Tadmor, “New high-resolution central schemes for non-
10
A. Boretti, P. Lappas, B. Zhang, and S. K. Mazlan, “CNG fueling strategies for linear conservation laws and convection-diffusion equations,” J. Comput. Phys.
commercial vehicles engines-a literature review,” SAE Technical Paper No. 160, 241–282 (2000).
31
A. Kurganov, S. Noelle, and G. Petrova, “Semidiscrete central-upwind schemes
2013-01-2812 (2013).
11 for hyperbolic conservation laws and Hamilton–Jacobi equations,” SIAM J. Sci.
D. Sankesh and P. Lappas, “Natural-gas direct-injection for spark-ignition
Comput. 23, 707–740 (2001).
engines—A review on late-injection studies,” SAE Technical Paper No. 2017- 32
B. van Leer, “Towards the ultimate conservative difference scheme. II.
26-0067 (2018).
12 Monotonicity and conservation combined in a second-order scheme,”
A. Hajialimohammadi, A. Abdullah, M. AghaMirsalim, and I. Chitsaz,
J. Comput. Phys. 14, 361–370 (1974).
“Experimental and numerical investigation on the macroscopic characteristics 33
V. Vuorinen, J. P. Keskinen, C. Duwig, and B. J. Boersma, “On the implemen-
of the jet discharging from gaseous direct injector,” J. Mech. Sci. Technol. 28,
tation of low-dissipative Runge-Kutta projection methods for time dependent
773–781 (2014). flows using OpenFOAMV R ,” Comput. Fluids 93, 153–163 (2014).
13
L. Allocca, A. Montanaro, G. Meccariello, F. Duronio, S. Ranieri, and A. De 34
S. B. Pope, in Turbulent Flows (Cambridge University Press, 2000).
Vita, “Under-expanded gaseous jets characterization for application in direct 35
S. Huang and Q. S. Li, “A new dynamic one-equation subgrid-scale model for
injection engines: Experimental and numerical approach,” SAE Technical large Eddy simulations,” Int. J. Numer. Methods Eng. 81, 865 (2009).
Paper No. 2020-01-0325 (2020). 36
A. Yoshizawa and K. Horiuti, “A statistically-derived subgrid-scale kinetic
14
A. Hajialimohammadi, D. Honnery, A. Abdullah, and M. A. Mirsalim, “Time energy model for the large-Eddy simulation of turbulent flows,” J. Phys. Soc.
resolved characteristics of gaseous jet injected by a group-hole nozzle,” Fuel Jpn. 54, 2834 (1985).
113, 497–505 (2013). 37
F. M. White, Viscous Fluid Flow, McGraw-Hill Series in Mechanical
15
M. Chiodi, H. J. Berner, and M. Bargende, “Investigation on different injection Engineering (McGraw-Hill, 1991).
strategies in a direct-injected turbocharged CNG-engine,” SAE Technical Paper 38
G. A. Sod, “A survey of several finite difference methods for systems of nonlin-
No. 2017-01-2275 (2006). ear hyperbolic conservation laws,” J. Comput. Phys. 27, 1–31 (1978).
16 39
B. Yadollahi and M. Boroomand, “The effect of combustion chamber geometry A. Hamzehloo and P. Aleiferis, “Large Eddy simulation of highly turbulent
on injection and mixture preparation in a CNG direct injection SI engine,” Fuel under-expanded hydrogen and methane jets for gaseous-fuelled internal com-
107, 52–62 (2013). bustion engines,” Int. J. Hydrogen Energy 39, 21275–21296 (2014).
17 40
I. Erfan, A. Hajialimohammadi, I. Chitsaz, M. Ziabasharhagh, and R. J. P. J. Roache, “Quantification of uncertainty in computational fluid dynamics,”
Martinuzzi, “Influence of chamber pressure on CNG jet characteristics of a Annu. Rev. Fluid Mech. 29, 123 (1997).
41
multi-hole high pressure injector,” Fuel 197, 186–193 (2017). European Committee for Standardization, “Explosive atmospheres—Part 20-1:
18
V. M. Salazar and S. A. Kaiser, “An optical study of mixture preparation in a Material characteristics for as and vapour classification—Test methods and
hydrogen-fueled engine with direct injection using different nozzle designs,” data,” Standard No. EN 60079–20-1:2010 (European Committee for
SAE Int. J. Engines 2, 119–131 (2009). Standardization, 2010).

Phys. Fluids 33, 066104 (2021); doi: 10.1063/5.0054098 33, 066104-16


Published under an exclusive license by AIP Publishing

You might also like