Download as pdf or txt
Download as pdf or txt
You are on page 1of 14

Propulsion and Power Research 2020;9(1):37e50

http://ppr.buaa.edu.cn/

Propulsion and Power Research

w w w. s c i e n c e d i r e c t . c om

ORGINAL ARTICLE

Interaction between a diesel-fuel spray and


entrained air with single- and double-injection
strategies using large eddy simulations
Jonathan Brulatout*, François Garnier, Patrice Seers

Mechanical Engineering Department, École de Technologie Supérieure, 1100 Notre-Dame Street West, Montreal,
H3C 1K3, Canada

Received 14 January 2019; accepted 5 December 2019


Available online 27 February 2020

KEYWORDS Abstract Interaction between fuel and air in a combustion chamber is one of the main drivers
of the mixing process. Experimentally, flow visualizations are limited by high droplet density in
Diesel fuel spray;
Large eddy simulation; the spray. Numerically, the ability of large eddy simulations (LES) to resolve large scales of
Multiple injection; flow offers good perspectives on capturing flow structures issued from the interaction between
Coherent structures; the Lagrangian (fuel droplets) and Eulerian (ambient gas) phases. This study examined these
Lagrangian-Eulerian interactions first during a single injection using 3D and 2D criteria for both phases. As for
interaction; the 3D criteria, the spray shape was analyzed in parallel to the Q-criteria applied to the Eulerian
Air-spray interactions phase, making it possible to relate the spray deformations to some specific Eulerian structures.
Secondly, 2D criteria were the fuel mass-fraction field and Eulerian streamlines, both taken in
the mid-plane of the spray. This last analysis allows for identifying certain mechanisms
involved in the Eulerian phase's structure generation and relates it to high fuel-concentration
areas in the fuel mass-fraction visualizations.
ª 2020 Beihang University. Production and hosting by Elsevier B.V. on behalf of KeAi. This is an open
access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/).

*Corresponding author. 1. Introduction


E-mail address: jonathan.brulatout.1@ens.etsmtl.ca (Jonathan Brulatout).
Peer review under responsibility of Beihang University. The mixture formation in internal combustion engines is
a key feature in better controlling the combustion process
and emissions produced. This process results from the
Production and Hosting by Elsevier on behalf of KeAi interaction between the fuel spray and the surrounding gas,

https://doi.org/10.1016/j.jppr.2019.12.001
2212-540X/ª 2020 Beihang University. Production and hosting by Elsevier B.V. on behalf of KeAi. This is an open access article under the CC BY-NC-ND
license (http://creativecommons.org/licenses/by-nc-nd/4.0/).
38 Jonathan Brulatout et al.

which has been studied experimentally and, more recently, 2. Computational method
numerically. This has been possible thanks to the increased
access to computational power allowing the use of the 2.1. Numerical model
large-eddy-simulation (LES), method providing finer de-
tails of the flow compared to the conventional URANS The following equations were resolved with STAR-
method. CCMþ commercial code, from CD-adapco, and are fol-
Using optical techniques, experimental studies of these lowed by a short description of the dynamic Smagorinsky
interactions concluded that coherent structures are generated model used for SGS modeling. Sagaut [10], Garnier et al.
in the surrounding gas. Indeed, they are indirectly detected [11], and STAR-CCMþ documentation [12] should be
at the periphery of the spray thanks to the rolling motion of consulted if more details about the mathematical formula-
the droplets. As an example, Sasaki et al. [1] presented fuel- tions are needed.
spray visualizations showing structures with a diameter The LES approach was implemented to resolve the
between 3 and 5 mm about 30 mm downstream from the largest Eulerian phase eddies, whereas the smallest ones
injector's nozzle. Even if Cao et al. [2] and Bruneaux et al. were modeled with an SGS model. Thus, a variable F can
[3] carried out the same type of visualization, the high be decomposed as a filtered component F and subgrid-scale
density and velocity of droplets in the rest of the spray fluctuations F0 , as presented in Eq. (1).
prevents current experimental techniques from detecting the
formation of such structures [4]. Therefore, numerical F Z F þ F0 ð1Þ
simulation is also a tool of choice in gaining a better un-
The resolved component F is extracted from the field of
derstanding of the phenomena at play during the injection
F by separating the scales. This operation is done through a
and mixture-formation processes, as it offers direct insight
spatial filtering function GD , defining F as in Eq. (2)
into the interaction between the Eulerian and Lagrangian
Z
phases.
Fðx; tÞ Z Fðx; tÞGD ðx  x; tÞd 3 x ð2Þ
Banerjee et al. [5,6] pursued this approach applying the
Q-criterion, making it possible to detect coherent structures U

into the Eulerian phase. Zhou et al. [7,8] used the same
where U is the control volume and x are the coordinates.
post-treatment to the fuel spray near the end of injection,
Herein, the filtering function GD is a box-type function
highlighting coherent structures of helical shape close to
whose cutting length D is closely tied to the mesh base size,
the nozzle, while also observing more fragmented struc-
as per Eq. (3)
tures downstream near the fuel-spray tip. Recently, Gha-
semi et al. [9] used the swirling strength to characterize the D Z V 1=3 ð3Þ
impact of injection pressure on the entrained-air and
mixing process obtained by URANS. They observed where V is the cell volume. Since the flow is modeled in a
smaller Eulerian structures at the spray tip with increasing compressible framework, the Favre-filtering formulation,
injection pressure as well as higher fuel concentration presented in Eq. (4), is used.
along the spray axis, which was devoid of swirling ~
structures. rF Z rF ð4Þ
As the above short survey illustrated, there is an interest
Thus, the filtered compressible Navier-Stokes equations
in characterizing the Eulerian phase of diesel-fuel spray to
are resolved: Eq. (5) satisfies the mass conservation,
help in identifying the mechanism at play during the
whereas Eqs. (6) and (7) take care of the momentum and
mixture-formation process. So far, studies have looked at
energy conservation, respectively.
single injection, while most diesel injection strategies
involve short multiple injections. Thus, this paper focuses vr v
þ ðr~ui Þ Z 0 ð5Þ
on studying the interaction between the Lagrangian and vt vxi
Eulerian phase of a diesel-fuel spray by comparing single-
v   vp vtij vtij
and double-injection strategies to identify the mechanism at sgs
v
play during the formation process from a non-evaporative ðr~ui Þ þ r~uj ~ui Z  þ þ þ si ð6Þ
vt vxj vxi vxj vxj
fuel-spray perspective. To do so, the paper is constructed as
follows. First, the mathematical formulation of the large
v v   vqj v    
eddy simulation used herein is presented and validated. ðr~eÞ þ ðr~e þ pÞ~uj Z  þ tij ~ui  v Cp qsgs
Then, the study of a simple-injection strategy will be done vt vxj vxj vxj vxj j

using tri- and bi-dimensional criteria. In the third section, ð7Þ


the same analysis strategy will be applied to split injection
and the impact of the first injection on the fuel dispersion of In Eqs. (5)e(7), r is the ambient gas density, ui the ve-
the second injection will be emphasized to identify the flow locity components, xi the coordinates, p the pressure, si the
differences responsible for a higher mixing associated with source term for the Lagrangian-Eulerian coupling, e the total
the split-injection strategy. energy as described in Eq. (8), qj the heat transfer, and tij
Interaction between a diesel-fuel spray and entrained air 39

the shear stress for a Newtonian flow as in Eq. (9). tsgs


ij is the
subgrid-scale stress tensor, which reduces to the expression
of Eq. (10) [10] as the box-type filter used herein is a
Reynolds operator [13].
p 1 tsgs
r~e Z uj ~
þ r~ ui þ ii ð8Þ
g1 2 2
 
 v~
ui v~
uj 2 v~
uk
tij Z m þ  dij ð9Þ
vxj vxi 3 vxk

g00 00
ij Z  rui uj
tsgs ð10Þ

where m the dynamic viscosity of the fluid and u00i represents


the Favre fluctuations of the velocity. The SGS model is
based on a hypothesis originally proposed by Boussinesq
[14] that leads to using a subgrid-scale viscosity msgs, as
presented in Eq. (11).
 
1 sgs  1
tsgs
ij  t d ij Z  2msgs S ij  S d
ij ij ð11Þ
3 kk 3
Figure 1 Mesh.
where Sij the rate-of-strain tensor of the resolved velocity a droplet to the gaseous phase while crossing a cell is
field, which is computed with Eq. (12). computed as per Eq. (15).
 
 1 v~ui v~uj dud;i 1
S ij Z þ ð12Þ md Z Cd rAfr;d jur;i jur;i ð15Þ
2 vxj vxi dt 2

The subgrid-scale viscosity msgs is modeled using the where md is the droplet mass, ud;i the i component of the
dynamic Smagorinsky model, which has been shown to droplet velocity, Cd the droplet drag coefficient, Afr;d the
successfully reproduce DNS results d especially the Rey- frontal area of the droplet, and ur;i the i component of the
nolds stresses d of thermal turbulent air jets [15]. This relative velocity computed as per Eq. (16).
dynamic model was chosen as it also appears to provide
satisfactory results when used in fuel-spray simulations, as ur;i Z ud;i  ~ui ð16Þ
proper spray shape and Lagrangian phase characteristics
(liquid and vapor penetration, mean droplet diameter) were
reported by Zhou et al. [7,16,17]. 2.2. Discretization and boundary conditions
The SGS model herein is based on the Smagorinsky
model [18], which computes SGS viscosity as in Eq. (13). The domain simulated was a parallelepiped 60 mm wide
2 and 110 mm long, as shown in Figure 1. On the top face, a
msgs Z rðCs DÞ jSj ð13Þ
wall boundary condition was implemented, while the rest of
the faces had pressure outlet conditions. The domain was
where Cs is the Smagorinsky constant, which is computed
discretized with a cubic mesh, with a side length of 0.25 mm
using the Germano-Lilly dynamic procedure [19,20] in the
in the center and 0.5 mm in the rest of the domain, as shown
case of the dynamic Smagorinsky model. Even if this
in Figure 1. The spatial discretization scheme was a second-
method can be conducive to numerical instabilities due to
order version of the upwind scheme, and the temporal
the possible computation of negative values of Cs , no such
scheme was a second-order Euler-type scheme.
instabilities were detected and, therefore, no limitations
were applied to this constant. Lastly, the norm of the rate-of-
strain tensor is computed as per Eq. (14), and the strain rate 3. Eulerian phase validation
tensor itself as per Eq. (12).
  The validation of the computational setup (i.e., mesh,
 Z 2Sij Sij 1=2
jSj ð14Þ discretization schemes, SGS model) for the Eulerian phase
of LES diesel-fuel-spray simulations is not straightforward
Regarding the Lagrangian-Eulerian coupling, the non- [21]. The complexity in validating is due, among other
evaporative application studied herein limits exchanges with things, to the lack of experimental data for the Eulerian
the Eulerian phase to a momentum transfer due to droplet phase because current optical methods provide limited ac-
drag, as seen in Eqs. (5)e(7). The momentum transferred by cess to the flow inside the fuel spray, especially for non-
40 Jonathan Brulatout et al.

evaporative fuel spray [4,22,23]. Other aspects include the


high cost for ensemble averaging of numerical data
requiring a significant number of LES simulations [21,24]
and the nature of dense fuel-spray simulations that prevents
a grid-independent solution [25e28] even when fine grids
are used. Due to these limitations, the Eulerian phase is
validated by using the initial boundary conditions of the
inlet flow based on the entrained-air characteristics of a
URANS diesel-fuel spray from Brulatout et al. [29] to
generate a statistically stationary turbulent air jet having an
initial diameter (D) of 4 mm and a fully developed-velocity
shape.
The domain used for the turbulent-jet simulation is the
same as the geometry of the fuel-spray cases, i.e., 30D (i.e.,
120 mm) in the axial direction and 12.5D wide (i.e., 50 mm).
It should be noted that this computational domain is limited by
the diesel-fuel spray application targeted in this paper, which Figure 2 Nondimensional axial-velocity profile.
does not allow for reaching the autosimilar state of the tur-
bulent round jet. The cell base size is the same as for the fuel the jet axis, which is within the range of the experimental
spray to be studied and is equal to 0.0625D (i.e., 0.25 mm), as data reported by Weisgraber et al. [34].
has also been successfully used in the turbulent round-jet [15] To complete the validation of the Eulerian phase, second-
and fuel-spray literature [8,30]. Spatial and temporal dis- order velocity moments were also compared with experi-
cretization schemes along with the time step were kept as in mental correlations established by Weisgraber et al. [34] in
the spray case in order to allow proper validation. the transition area, and by Hussein et al. [33]. Figure 4
The computed volume was considered as quiescent at the presents the results. It should be noted that the moments
beginning of the simulation, and an initial perturbation was u0 u0 , u0 v0 and w0 u0 are not presented, as their profiles are
introduced using the synthetic-eddy method (SEM) [31], similar to those for v0 v0, for the first two moments, and w0 v0 ,
allowing for faster turbulence development. This mesh- for the latter.
independent method generated eddies whose spins and po- The profiles of the second-order moments presented
sitions are determined according to a normal distribution. herein are in fairly good agreement with the experimental
The two main characteristics are the intensity (herein 10%) correlation of Weisgraber et al. [34] and Hussein et al. [33]
and length scale. The latter was greater than twice the mesh with an equivalent order of magnitude and trend, which was
base size for a proper discretization of the eddies, as rec- confirmed with other papers in the literatures
ommended by the STAR-CCMþ user manual [12]. [15,32,35e37]. These successful comparisons confirm the
First, the temporal-averaged axial velocity along the resolution quality provided by the model herein, especially
main axis of the flow Wc was nondimensionalized by the in the outer area of the turbulent jet, where most of the
exit velocity of the jet Wj , as presented in Figure 2. From the mixing process takes place.
nozzle exit to 10D, the profile presents only small variations
due to the low momentum exchange between the center of
the jet and the ambient gas [32]. After 10D, when the mo-
mentum exchange is established, this velocity ratio can be
represented mathematically with a linear function, as rep-
resented by Eq. (17) [32,33].

Wj 1  z z0 
Z  ð17Þ
Wc Bu D D

where Bu is a constant, z the position along the central axis,


and z0 the axial position of the nozzle.
Figure 3 presents the profile of the axial velocity as a
function of the nondimensional radial distance rnd , along
with the experimental correlation of Weisgraber et al. [34].
The axial velocity W is nondimensionalized by temporal-
averaged axial velocity Wc at a distance of 25D from the
nozzle as done in the experimental results used as reference
[34]. Numerical results showed good agreement with the Figure 3 Nondimensional axial velocity versus radial distance at
experimental correlation with a slight discrepancy far from 25D compared with the Weisgraber et al. [34] experimental profile.
Interaction between a diesel-fuel spray and entrained air 41

Table 1 KHRT-model constants [29].

Constant B0 B1 C3 Ct

Value 0.6 40 0.05 1.55

strategies by providing the mass flow rate measured


experimentally by Plamondon et al. [38], as presented in
Figure 5. Based on these profiles, parcels were introduced
into the domain using the blob method [39,40]. Their
introduction in a quiescent domain caused fragmentation,
which is accounted for with the KHRT model with the
Brulatout et al. [29] calibration presented in Table 1.
In addition, the interaction between droplets was handled
using the NTC model [41] as it has been extensively used
and validated in the literature as the computation is based on
the O'Rourke model [42]. Also, a turbulence dispersion
model was used to take into account the interaction of the
droplet and subgrid scale turbulence [26,27].
Also, the Lagrangian-Eulerian approach used herein re-
lies on the point-particle approximation which implies a
void fraction close to 1, meaning a negligible presence of
the Lagrangian phase in the cell volume. For diesel fuel
spray such as herein, this condition is not respected,
particularly so, near the injector nozzle where fuel concen-
tration is higher. However, this approach is widely used in
the fuel spray literature as seen in the introduction and is
said to offer better statistical representation of the fuel spray
than Eulerian-Eulerian approaches [28].

4.2. Numerical setup validation for spray simulation

The validation of the fuel-spray simulation was first


quantitatively pursued based on the liquid penetration
(defined as the position containing 95% of the fuel mass
[7,17,43e45]) for which the current model yielded a fairly
good prediction, as presented in Figure 6. The figure shows
that the model closely matches the experimental results from
Tétrault et al. [46], between 250 ms and 1600 ms, with varia-
tions within experimental uncertainty. Early liquid penetra-
tion before 250 ms, however, shows the model's difficulty in
representing the early transient state of the fuel-injection
process, because the model does not take into account the
transient dynamics of needle displacement inside the injector
[47,48]. Moreover, the discrepancy might also be due to the
inaccuracies associated with the blob injection method which
will impact the primary atomization.
Figure 4 Radial turbulent profiles for the gas round jet at 25D In addition, the SMD temporal evolution is presented in
compared with the Weisgraber et al. [34] experimental profile. Figure 7. The values decrease with time as the fragmen-
tation process occurs and the spray penetrates farther in the
ambient gas. This evolution is consistent with the behavior
4. Lagrangian phase validation reported by Park et al. [49], with a final overall SMD of
44 mm. Around 1150 ms, a short increase of the SMD can
4.1. Atomization model be seen and is linked to the blob method injecting constant
diameter droplets toward the end of injection. Moreover, it
Based on the Eulerian phase results, the computational should be noted that the liquid penetration and the mean
setup was used to simulate single- and double-injection droplet diameter were not influenced by the ensemble
42 Jonathan Brulatout et al.

Figure 5 Single-(o) and double-(*) injection profiles. Figure 7 Sauter mean diameter versus time.

average (6 runs herein), as also shown by Senecal et al.


[50].
The main interest of using an LES approach for a diesel-
fuel spray is to better predict the mixing rate, which depends
on air entrainment. Figure 8 presents the cumulated mixing
rate tc defined by Eq. (18) for the fuel-spray simulation
obtained with the LES and the experimental data from
Sepret et al. [51].

tc Z m_ e m_ ð18Þ
f

where m_ e is the entrained air mass flow rate and m_ f is the


instantaneous fuel mass flow rate. It should be noted that
the air entrainment for the LES diesel-fuel spray was
computed using 1% velocity contour on 70% of the
penetration [29]. Both curves show similar trends but with
higher predicted air entrainment for the simulation when
compared to the experimental steady-state fuel-spray in-
jection. The difference might be due to how entrained air is Figure 8 Numerical cumulated mixing rate compared to experi-
mental results from Sepret [52].
defined. Indeed, in Sepret et al. [51], a constant fictive cone

having characteristics equal to the experimental steady-


state fuel-spray injection was used. Herein, however, the
boundary of the fictive cone is defined by the velocity flow
field of the Eulerian phase, for which the characteristics
may change over time as it converges to a fully initiated
state [51].

5. Results and analysis

5.1. Simple injection strategy

First, the interaction between the fuel-spray droplet and


surrounding air were considered for a single-injection
strategy (see Figure 9 to Figure 14). In order to better
visualize the three-dimensional nature of the flow, 3D
Lagrangian phase visualization is complemented by the Q-
Figure 6 Spray penetration versus time (numerical data is from an criteria for the Eulerian phase. The Q-criteria was based on
ensemble averaged of 6 simulations, using the 95% criteria). the antisymmetric Uij and symmetric Sij parts of the velocity
Interaction between a diesel-fuel spray and entrained air 43

used to give insight into the fuel-spray middle plane.


Because of the non-evaporative character of the fuel spray
studied herein, the mass fraction Yfuel was defined from the
fuel mass in a cell (mfuel, cell) and the maximum fuel mass in
a cell on all the computational domain U, as defined by Eq.
(22).

mfuel; cell
Yfuel Z   ð22Þ
maxU mfuel; cell

It should be noted that, during all the following analysis,


the area of interest is located in the fuel-spray surroundings
and streamline patterns far from the latter can be ignored
because of the very low velocity in the constant-volume
vessel.
Figure 9 (top, left) presents the Lagrangian phase at 200
ms after the start of injection (SOI). At the nozzle exit, the
droplets reached velocities between 250 m/s and almost
400 m/s, rapidly decreasing as the droplets fragmented and
decelerated due to drag to reach low velocities (~50 m/s)
within a few millimeters from the injector nozzle (visible at
the fuel-spray periphery). At the same time, Q-criteria
visualization in the same time-step in Figure 9 (bottom, left)
shows the creation of an annular structure around the fuel-
spray axis. This coherent structure was created by the
Lagrangian phase penetrating the quiescent ambient gas, as
was previously observed by Yeom et al. [54] with 2D fuel-
spray simulation, and also in the case of impulsive gaseous
jets [55]. In the latter application, the generation of this
structure is attributed to the creation of a shear layer that
Figure 9 2D and 3D representations at 200 ms after SOI. Droplet quickly rolls to create an annular recirculation zone. Herein,
velocity (top left), fuel mass fraction (top right), Eulerian structures the momentum impulsion is provided by the drag of the
(bottom left), and streamlines at mid-plane (bottom right).
droplets injected into the ambient gas. Simultaneously, at a
distance between 8 mm and 10 mm from the injector nozzle,
a nascent helical structure can be observed, which becomes
gradient tensor, and were calculated using Eqs. (19) and (20) developed at 300 ms (not presented herein for the sake of
[53]. brevity). Figure 9 (top, right) shows two circular zones
  containing a high fuel concentration at 17 mm, that is to say,
1 vui vuj
Uij Z  ð19Þ 5 mm downstream from the Eulerian phase structures visible
2 vxj vxi in Figure 9 (bottom, right). Indeed, the Eulerian phase
  motion was initially induced by the momentum transferred
1 vui vuj from the Lagrangian phase, which penetrates with a much
Sij Z þ ð20Þ
2 vxj vxi higher velocity (reaching 400 m/s) than the Eulerian phase.
This high velocity acts as an impulsion, as discussed in the
From those quantities, the Q-criteria was computed with previous 3D analysis.
Eq. (21) and allows for quantifying the excess of rotation Globally, the fuel-spray shape presented at 400 ms in
with respect to the shear of a fluid particle. Figure 10 (top left) does not show any differences in
1  behavior with respect to Figure 9 (top left). The helical
QZ Uij Uij þ Sij Sij ð21Þ structure creation, however, detected by Q-criteria
2
(Figure 10, bottom left) continued developing over time
In addition, the main characteristic investigated with LES with the transformation of the donut shape at the spray front
fuel-spray simulations was the ability of the model to predict into multiple smaller coherent structures. Comparing
the mixing of the fuel and ambient air. Thus, two- streamlines in Figure 9 (bottom right) to those in Figure 10
dimensional visualization, combining fuel mass fraction of (bottom right), allows for visualizing the deformation of the
the Lagrangian phase and Eulerian phase streamlines, was annular structure detected by the Q-criteria and illustrated as
44 Jonathan Brulatout et al.

injector. This phenomenon is observable in the fuel mass


fraction (Figure 11, top right) as two zones of low fuel mass
concentration can be seen at 25 mm and 35 mm above and
below high fuel concentration which is located between
recirculation zones, as can be observed with streamlines
(Figure 11, bottom right). This observation agrees with the
experimental results of Cao et al. [2], who reported high fuel
concentration downstream of structures and fuel entrain-
ment upstream by recirculation. This dynamic creates a
higher fuel concentration in the periphery of the structures,
whereas their centers are almost free of fuel, increasing the
contact area between the fuel and ambient air and favoring
mixing [2].
The continued deformations of the Eulerian structure
and its impact on the spray deformations and fuel distri-
bution is more visible at 800 ms after SOI, as presented in
Figure 12 (top). Indeed, Q-criteria (Figure 12, bottom left)
shows a complex flow structure with the detection of
intertwined ligament-like structures that emanated from
the initial annular structure. This type of ligament structure
was also highlighted with respect to fuel-spray de-
velopments by Banerjee et al. [5,6], Irannejad et al. [56]

Figure 10 2D and 3D representations at 400 ms after SOI. Droplet


velocity (top left), fuel mass fraction (top right), Eulerian structures
(bottom left), and streamlines at mid-plane (bottom right).

a recirculation zone in streamlines. This stretch in the spray-


axis direction can be linked to the fuel-spray tip in Figure 10
(top right) with the consequence that the more visible in the
region of high fuel concentration (Ycarb close to 1) moves
away from the spray central axis and continues its lateral
dispersion. Moreover, it can be observed that those high-
fuel-concentration areas were influenced by the annular
recirculation structure due to the droplets’ low velocity
below 20 m/s and identified in Figure 10 (top left).
At 600 ms after the SOI, the Lagrangian phase (Figure 11,
top left) presents some minor distortion as a loss of sym-
metry began appearing and can be linked to Eulerian
structures (Figure 11, bottom left). Indeed, the helical
structure there extends 10 mme25 mm from the injector
nozzle, while the annular structure underwent deformation,
leaving small fragmented structures as it moved forward
with the fuel spray (~35 mm). This deformed annular
structure at the forefront interacted with the low-velocity
droplets located at the spray periphery, dispersing droplets
Figure 11 2D and 3D representations at 600 ms after SOI. Droplet
from the spray, while other droplets were entrained by the
velocity (top left), fuel mass fraction (top right), Eulerian structures
recirculation structures with velocities of 20 m/s towards the
(bottom left), and streamlines at mid-plane (bottom right).
Interaction between a diesel-fuel spray and entrained air 45

toward small low-energy structures. This fragmentation of


the structure increases the Lagrangian phase fluctuations, all
the more that droplets with very low velocity (few meters
per second) are more receptive to the Eulerian-velocity field
fluctuations. Comparing streamlines at 1200 ms with results
obtained at 800 ms reveals the dissipation of the upstream
right-side annular structure, while the left-side structure
persisted (located at ~37 mm); a new recirculation zone can
be observed in Figure 13 (bottom right) at ~47 mm (right
side). This new recirculation zone is related to the dynamic
of the fragmented structures identified using the Q-criteria.
Moreover, the presence of these structures is responsible for
the fuel dispersion illustrated in Figure 13 (top right),
resulting in an increase in the contact area between the
ambient air and fuel, leading to a decrease in the fuel mass
fraction close to the spray axis.
At the end of the simulation (1600 ms), the fuel drop-
lets at the spray periphery (Figure 14, top left) are char-
acterized as having low velocities (~10 m/s) and, as such,
interact more easily with the dissipating structures in the
Eulerian phases (Figure 14, bottom left). The dynamics
shown at time 1200 ms was thus emphasized over time
because the Lagrangian phase quickly decelerated and

Figure 12 2D and 3D representations at 800 ms after SOI. Droplet


velocity (top left), fuel mass fraction (top right), Eulerian structures
(bottom left), and streamlines at mid-plane (bottom right).

and Zhou et al. [7,8]. The formation of these ligaments


indicates the presence of a more turbulent flow in this
region responsible for changes in fuel-spray geometry by
interacting with low-velocity fuel droplets. The interaction
between the structures and fuel droplets can be observed in
the fuel mass fraction (Figure 11, top right) along the
central plane, where the lack of symmetry is apparent. The
streamlines presented in Figure 12 (bottom right) confirm
the asymmetry in the flow structure and also reveal the
disappearance of one structure previously detected at
600 ms.
At the end of injection (EOI), which occurred at 1200 ms,
the fuel-spray deformation at the bottom half is evident in
Figure 13 (top left) as fuel droplets interacted with the
complex structure of the Eulerian phase (Figure 13, bottom
left). During the injector closure, the momentum of the fuel
droplet injected decreased, resulting in structures that
are more easily dissipated in the upper half of the spray,
while downstream structures will start dissipating as the
momentum transfer from the upstream decreases and ceases.
The downstream structures now fragment into smaller lig- Figure 13 2D and 3D representations at 1200 ms after SOI. Droplet
aments, implying that the energy transfer is then directed velocity (top left), fuel mass fraction (top right), Eulerian structures
(bottom left), and streamlines at mid-plane (bottom right).
46 Jonathan Brulatout et al.

5.2. Double injection strategy

5.2.1. Injection conditions


The double injection strategy simulated in this sec-
tion consists of a split energization time, initially of 525
ms in the case of the simple injection, in two periods of
265 ms, one at the beginning of the calculation and a
second one at time 1000 ms [38]. Those experimental
conditions provided the fuel mass flow presented in
Figure 5.

5.2.2. Flow analysis


While the first injection is shorter than the single injec-
tion studied in the previous section, the trend observed are
identical but differed in intensity. Therefore, an annular
structure due to the initial impulsion of the fuel spray is
observed which progress with time. Because a lower mo-
mentum is injected due to a shorter injection, the fragmen-
tation of the annular structure is faster and is simply linked
to the injection duration. Following the end of the first in-
jection, dissipation of the structures and their fragmentation
progress as reported previously. For that reason, the analysis
begins at the SOI of the second injection at time 1000 ms as

Figure 14 2D and 3D representations at 1600 ms after SOI. Droplet


velocity (top left), fuel mass fraction (top right), Eulerian structures
(bottom left), and streamlines at mid-plane (bottom right).

interacted with the Eulerian phase structures. While the


Q-criteria has shown that the ligament structures are
fragmented from the EOI and are mainly located in the
downstream region of the fuel spray, the consequence on
the flow is illustrated with Figure 14 (bottom right) at a
time of 1600 ms. Vortex-like structures regularly spaced at
the interface between the Lagrangian and Eulerian phase
can be observed, as reported experimentally by Cao et al.
[2]. These structures dispersed the fuel droplet at
the spray periphery, while decreasing the fuel mass
fraction at its center (Figure 13, top right). It therefore
illustrates the mixing-formation process after the end of
injection.
From the above analysis of the time evolution of the fuel
spray a partial conclusion can be reached. Overall, fluctuations
of the Lagrangian phase are related to coherent structures of
the Eulerian phase which are responsible to dispersing fuel
droplet and illustrates the mixing process. It is noted that the
annular structure created by the impulsion of the spray at the Figure 15 2D and 3D representations at 1000 ms after SOI of the
start of injection appears to have an important influence on the first injection. Droplet velocity (top left), fuel mass fraction (top right),
ensuing fuel mixing as well as the spray deformations. Eulerian structures (bottom left), and streamlines at mid-plane (bottom
right).
Interaction between a diesel-fuel spray and entrained air 47

generated an annular structure similarly to the first injection


(not presented herein) and to the single injection case
(Figure 9). Figure 15 (bottom right) presents the Eulerian
phase streamlines where a symmetric pair of recirculation
structures provoked by the second injection can be
observed, as expected from past results. This pair of struc-
tures can still be associated with two high fuel-concentration
areas in Figure 15 (top right), as in the single-injection case
(Figure 9). This process happens while structures from the
first injection keep transporting the fuel radially at the bot-
tom of the first spray.
The annular structures identified in the upper part of
Figure 15 (bottom right) at time 1200 ms were rapidly
transformed into ligaments by the EOI (at time 1400 ms
from the beginning of the simulation), as can be observed in
the upper part of the domain in Figure 17 (bottom left).
Simultaneously, the structure generated by the first injection
dissipated at the bottom of the spray (Figure 16, bottom left)
with streamlines (Figure 17, bottom right), showing a more
chaotic flow at the bottom center as the second spray
collided with the first. This collision increased the fuel
mass-fraction distribution in the area of impact, as can be
seen in Figure 16 (top right).
Lastly, at the end of the simulation corresponding to a
time of 1600 ms, the Lagrangian phase (Figure 18, top left)

Figure 16 2D and 3D representations at 1200 ms after SOI of the


first injection (second injection visible at top). Droplet velocity (top
left), fuel mass fraction (top right), Eulerian structures (bottom left),
and streamlines at mid-plane (bottom right).

illustrate with Figure 15 that presents the flow characteris-


tics where intertwined ligament structures are present in the
bottom half of the first spray (Figure 15, bottom-left) with
low velocity (10 m/s range) droplets covering the spray
volume (Figure 15, top-left). The ligaments detected with
the Q-criteria matched the recirculation zone highlighted by
the streamlines (Figure 15, bottom). These high perturba-
tions on the Eulerian phase, even inside the spray, explains
the more homogeneously and radially spread fuel mass
fraction as presented on Figure 15 (top, right) which favors
the contact surface between the fuel and the ambient air,
showing an enhanced mixing process.
At 200 ms after the SOI of the second injection (time
1200 ms since the beginning of the simulation), high ve-
locity droplets emanating from the second injection are
visible in the first 22 mm in Figure 16 (top left), which also
shows the droplet interaction between the first and second
injections. It is noted that the second injection had a faster
penetration than the first injection, which reached approxi-
mately 18 mm during the same time interval (Figure 8, top
Figure 17 2D and 3D representations at 1400 ms after SOI of first
left) and thus confirming the impact of the entrained air of injection. Droplet velocity (top left), fuel mass fraction (top right),
the first injection on the second one. The Q-criteria of Eulerian structures (bottom left), and streamlines at mid-plane (bottom
Figure 16 (bottom left) shows that this second injection right).
48 Jonathan Brulatout et al.

Figure 19 Evolution of the fuel mass distribution along the spray


central axis for the single and double injection strategies at 1000 ms and
1600 ms after SOI.

offered a fuel mass that is distributed from the injector to the


spray tip as the injection is still on-going. On the contrary, the
double injection strategy showed a lower fuel mass distribu-
tion due to a lower fuel mass injected. That distribution is
mainly located between 25 mm and 50 mm from the injector
nozzle. At 1600 ms after the start of the injection, the single
Figure 18 2D and 3D representations at 1600 ms after SOI of first and double injection strategies offered similar shape and
injection. Droplet velocity (top left), fuel mass fraction (top right), amplitude for the mass distribution but a higher penetration is
Eulerian structures (bottom left), and streamlines at mid-plane (bottom observed for the single injection strategy. Another observation
right). for the double injection strategy is that the fuel mass distri-
bution is fairly similar for both times considered when one
considers the shape and amplitude of the distribution.
and coherent structures (Figure 18, bottom left) from the
second injection completely fused with those from the first 6. Conclusions
injection. In particular, the coherent structures obtained at
the end of the simulation by the double injection did not This paper focused on studying the differences between
exhibit major qualitative differences with the result obtained the mixing process induced by a single- and double-
at the end of the simulation for the single-injection case, injection strategy using a LES fuel-spray model. Thus, the
even if the momentum quantity of the double injection was computational setup was first validated by comparing the
less than one of the single injection. Contrary to the first Lagrangian characteristics to experimental data from the
injection, however, the helical structure around the spray literature. As the possibilities for validating the computa-
axis is not clearly visible. This could be the consequence of tional setup is under certain numerical and experimental
the flow initialization produced by the first injection, which limitations, a validation was done by assuming a similarity
reduces shear between the two phases during the second between the Eulerian phase of the fuel spray and a gaseous
injection and disadvantages the structure creation in the air jet. The latter was computed using the same computa-
Eulerian phase. Figure 18 (top right) also shows that the tional setup with initial conditions extracted from fuel-
second spray penetrated deeply into the first spray due to spray air entrainment. Results led to a good match in term
entrainment with a higher fuel mass fraction along the spray of average and turbulent profile for the gaseous round jet
axis (Figure 18, top right) transported radially by the and thus, by similarity, for the Eulerian phase of the fuel
recirculation structures, as was observed at the end of the spray. Then, the mixing process for the single- and double-
simulation for the single-injection case (Figure 13). injection strategies was analyzed using 2D and 3D visu-
In addition to the 3D visualization, Figure 19 compares the alization tools.
evolution of the fuel mass distribution along the central axis of In the single-injection case, the creation of an annular
the spray for the simple and double injection strategy at two structure was highlighted. A similarity with impulsive
different times namely at 1000 ms and 1600 ms. At 1000 ms, gaseous jets allowed the conclusion that this structure
the simple injection strategy, and longer duration injection, resulted from the momentum impulsion induced by the fuel
Interaction between a diesel-fuel spray and entrained air 49

spray in the quiescent environment. Further in flow devel- [9] A. Ghasemi, R.M. Barron, R. Balachandar, Spray-induced air motion
opment, a helical structure was detected using the Q-crite- in single and twin ultra-high injection diesel sprays, Fuel 121 (2014)
281e297.
rion but was not visible in the streamlines. Later, the annular [10] P. Sagaut, Large Eddy Simulation for Incompressible Flows: an
structure close to the fuel-spray front fragmented, while the Introduction, Springer-Verlag Berlin Heidelberg, 2006.
recirculation zone was observed above and conveyed the [11] E. Garnier, N. Adams, P. Sagaut, Large Eddy Simulation for
fuel radially, thus increasing the radial diffusion of the fuel. Compressible Flows, 2009.
In the last section, the same analysis was applied to the [12] CD-Adapco, Star-CCMþ Version 8.04, 2013.
[13] L.C. Berselli, T. Iliescu, W.J. Layton, Mathematics of Large Eddy
double-injection strategy. It has been observed that the fuel- Simulation of Turbulent Flows, 2006.
spray tip penetration of the second injection was faster than [14] J. Boussinesq, Essai sur la théorie des eaux courantes, 1877.
the first due to air entrainment. Collision between the first [15] C. Ferreira Gago, F. Garnier, F. Utheza, Direct testing of subgrid scale
and second spray created a large zone of high fuel con- models in large-eddy simulation of a non-isothermal turbulent jet, Int.
centration that were dispersed by the recirculating zone. J. Numer. Methods Fluid. 42 (2003) 999e1026.
[16] L. Zhou, M. Xie, M. Jia, Q. Zhou, C. Xu, Influences of subgrid tur-
Overall, the general mixing dynamics was similar to the bulent kinetic energy and turbulent dispersion on the characteristics of
single-injection strategy, but two annular structures were fuel spray, SAE Tech. Pap. Ser. 1, 2011.
observed during the second injection that appeared to [17] L. Zhou, M.Z. Xie, K.H. Luo, M. Jia, Q. Zhou, H. Liu, Mixing effects
enhance the mixing process as the dwell time between in- of early injection in diesel spray using LES model with different
jections, allowing structures to transport fuel farther radially, subgrid scale models, SAE Tech. Pap. 2013-01-1111, 2013.
[18] J. Smagorinsky, General circulation experiments with the primitive
favoring the mixing process of the double-injection strategy. equations, Mon. Weather Rev. 91 (1963) 99e164.
[19] M. Germano, U. Piomelli, P. Moin, W.H. Cabot, A dynamic subgrid-
scale eddy viscosity model, Phys. Fluids A Fluid Dyn. 3 (1991)
Acknowledgements 1760e1765.
[20] D.K. Lilly, A proposed modification of the germano-subgrid-scale
closure method, Phys. Fluids A Fluid Dyn. 4 (1992) 633e635.
Computations were made on the Guillimin supercom- [21] C.J. Rutland, Large-eddy simulations for internal combustion engines
puter at McGill University, managed by Calcul Québec and - a review, Int. J. Engine Res. 12 (2011) 421e451.
Compute Canada. The operation of this supercomputer is [22] M. Meijer, L.M. Malbec, G. Bruneaux, L.M.T. Somers, Engine
combustion network: ‘spray A’ basic measurements and advanced
funded by the Canada Foundation for Innovation (CFI),
diagnostics, in: 12th Trienn. Int. Conf. Liq. At. Spray Syst, (ICLASS
NanoQuébec, RMGA and the Fonds de recherche du 2012), Heidelberg, Germany, 2012, pp. 2e6.
Québec - Nature et technologies (FRQ-NT). [23] R. Payri, J.P. Viera, H. Wang, L.-M. Malbec, Velocity field analysis of
The authors also acknowledge AUTO21 Network of the high density, high pressure diesel spray, Int. J. Multiphas. Flow 80
Centers of Excellence and Natural Sciences and Engineering (2016) 69e78.
[24] D.C. Haworth, Large-eddy simulation of in-cylinder flows, Oil Gas
Research Council of Canada (NSERC) for financially sup-
Sci. Technol. 54 (1999) 175e185.
porting this work. [25] J. Abraham, What is adequate resolution in the numerical computa-
Special thanks go to C. Mounaïm-Rousselle and C. tions of transient jets? SAE Tech. Pap. 970051, 1997.
Hespel from the Université d'Orléans (France) for their [26] C. Baumgarten, Mixture Formation in Internal Combustion Engines,
support with the diesel-fuel spray experiments. 2006.
[27] G. Stiesch, Modeling Engine Spray and Combustion Processes,
Springer, Berlin Heidelberg, 2003.
[28] S. Subramaniam, Lagrangian-Eulerian methods for multiphase flows,
References Prog. Energy Combust. Sci. 39 (2013) 215e245.
[29] J. Brulatout, F. Garnier, C. Mounaïm-Rousselle, P. Seers, Calibration
[1] S. Sasaki, H. Akagawa, K. Tsujimura, A study on surrounding air flow strategy of diesel-fuel spray atomization models using a design of
induced by diesel sprays, SAE Tech. Pap. 980805, 1998. experiment method, Int. J. Engine Res. (2015) 1e19.
[2] Z.M. Cao, K. Nishino, S. Mizuno, K. Torii, PIV measurement of in- [30] C.-W. Tsang, C. Rutland, Effects of numerical schemes on large eddy
ternal structure of diesel fuel spray, Exp. Fluid 29 (2000) S211eS219. simulation of turbulent planar gas jet and diesel spray, SAE Int. J.
[3] G. Bruneaux, M. Causse, A. Omrane, Air entrainment in diesel-like Fuels Lubr. 9 (2016) 149e164.
gas jet by simultaneous flow velocity and fuel concentration mea- [31] N. Jarrin, S. Benhamadouche, D. Laurence, R. Prosser, A synthetic-
surements, Comparison of Free and Wall Impinging Jet Configura- eddy-method for generating inflow conditions for large-eddy simula-
tions, SAE Int. J. Engines 5 (2011) 76e93. tions, Int. J. Heat Fluid Flow 27 (2006) 585e593.
[4] A. Coghe, G.E. Cossali, Quantitative optical techniques for dense [32] C.G. Ball, H. Fellouah, A. Pollard, The flow field in turbulent round
sprays investigation: a survey, Optic Laser. Eng. 50 (2012) 46e56. free jets, Prog. Aero. Sci. 50 (2012) 1e26.
[5] S. Banerjee, C. Rutland, On LES grid criteria for spray induced tur- [33] H.J. Hussein, S.P. Capp, W.K. George, Velocity measurements in a
bulence, SAE Tech. Pap. 2012-01-0141, 2012. high-Reynolds-number, momentum-conserving, axisymmetric, turbu-
[6] S. Banerjee, C.J. Rutland, Study on spray induced turbulence using lent jet, J. Fluid Mech. 258 (1994) 31e75.
large eddy simulations, At. Sprays. 25 (2015) 285e316. [34] H.T. Weisgraber, D. Liepmann, Turbulent structure during transition
[7] L. Zhou, K.H. Luo, S.J. Shuai, M.Z. Xie, Large-eddy simulations of to self-similarity in a round jet, Exp. Fluid 24 (1998) 210e224.
diesel spray with a fine grid in a constant-volume vessel, Proc. Inst. [35] C. Bogey, C. Bailly, Large eddy simulations of transitional round jets:
Mech. Eng. - Part D J. Automob. Eng. 229 (2015) 247e260. influence of the Reynolds number on flow development and energy
[8] L. Zhou, K.H. Luo, W. Qin, M. Jia, S.J. Shuai, Large eddy simulation dissipation, Phys. Fluids 18 (2006) 1e14.
of spray and combustion characteristics with realistic chemistry and [36] H. Foysi, J.P. Mellado, S. Sarkar, Large-eddy simulation of variable-
high-order numerical scheme under diesel engine-like conditions, density round and plane jets, Int. J. Heat Fluid Flow 31 (2010)
Energy Convers. Manag. 93 (2015) 377e387. 307e314.
50 Jonathan Brulatout et al.

[37] S. McIlwain, A. Pollard, Large eddy simulation of the effects of mild [47] M.R. Turner, S.S. Sazhin, J.J. Healey, C. Crua, S.B. Martynov, A
swirl on the near field of a round free jet, Phys. Fluids 14 (2) (2002), breakup model for transient diesel fuel sprays, Fuel 97 (2012) 288e305.
https://doi.org/10.1063/1.1430734. [48] M. Battistoni, Q.L. Xue, S. Som, Large-eddy simulation (LES) of
[38] É. Plamondon, P. Seers, Development of a simplified dynamic model spray transients: start and end of injection phenomena, Oil Gas Sci.
for a piezoelectric injector using multiple injection strategies with Technol. - Rev. d’IFP Énergies Nouv. 71 (2016), https:
biodiesel/diesel-fuel blends, Appl. Energy 131 (2014) 411e424. //doi.org/10.2516/ogst/2015024.
[39] R.D. Reitz, R. Diwakar, Effect of drop breakup on fuel sprays, SAE [49] S.H. Park, H.J. Kim, H.K. Suh, C.S. Lee, A study on the fuel injection
Tech. Pap. 860469, 1986. and atomization characteristics of soybean oil methyl ester (SME), Int.
[40] R.D. Reitz, R. Diwakar, Structure of high-pressure fuel sprays, SAE J. Heat Fluid Flow 30 (2009) 108e116.
Tech. Pap. 870598, 1987. [50] P.K. Senecal, E. Pomraning, K.J. Richards, S. Som, An investigation
[41] D.P. Schmidt, C.J. Rutland, A new droplet collision algorithm, J. of grid convergence for spray simulations using an LES turbulence
Comput. Phys. 164 (2000) 62e80. model, SAE Tech. Pap. 2013-01-1083, 2013.
[42] P.J. O'Rourke, Collective Drop Effects on Vaporizing Liquid Sprays, [51] V. Sepret, R. Bazile, M. Marchal, G. Couteau, Effect of ambient
Los Alamos National Lab., NM, USA, 1981. density and orifice diameter on gas entrainment by a single-hole diesel
[43] K. Jagus, X. Jiang, G. Dober, G. Greeves, N. Milanovic, H. Zhao, spray, Exp. Fluid 49 (2010) 1293e1305.
Assessment of large-eddy simulation feasibility in modelling the un- [52] V. Sepret, Application de la PIV sur traceurs fluorescents à l’étude de
steady diesel fuel injection and mixing in a highspeed direct-injection l’entraînement d’air par un spray diesel, Influence de la densité ambiante
engine, Proc. Inst. Mech. Eng. - Part D J. Automob. Eng. 223 (2009) et du diamètre de trou d’injecteur, University of Toulouse, 2009.
1033e1048. [53] P. Chakraborty, S. Balachandar, R.J. Adrian, On the relationships
[44] G.M. Magnotti, C.L. Genzale, Influence of liquid penetration metrics between local vortex identification schemes, J. Fluid Mech. 535
on diesel spray model validation, SAE Int. 2013-01-1102, 2013. (2005) 189e214.
[45] S. Tonini, M. Gavaises, A. Theodorakakos, G.E. Cossali, Numerical [54] J.-K. Yeom, K. Ashida, J. Senda, H. Fujimoto, T. Dan, Analysis of
investigation of a multiple injection strategy on the development of diesel spray structure by using a hybrid model of tab breakup model
high-pressure diesel sprays, Proc. Inst. Mech. Eng. - Part D J. Auto- and vortex method, SAE Tech. Pap. 2001-01-1240, 2001.
mob. Eng. 224 (2010) 125e141. [55] D.T.H. New, S.C.M. Yu, Vortex Rings and Jets: Recent Developments
[46] P. Tetrault, E. Plamondon, M. Breuze, C. Hespel, C. Mounaïm-rous- in Near-Field Dynamics, Springer, Singapore, 2015.
selle, Fuel spray tip penetration model for double injection strategy, [56] A. Irannejad, F. Jaberi, Numerical study of high speed evaporating
SAE Tech. Pap. 2015-01-0934, 2015. sprays, Int. J. Multiphas. Flow 70 (2015) 58e76.

You might also like