Applied Energy xxx (2017) xxx–xxx

Contents lists available at ScienceDirect

Applied Energy
journal homepage: www.elsevier.com/locate/apenergy

Numerical analysis of deposit effect on nozzle flow and spray

characteristics of GDI injectors
Bo Wang a,c, Yizhou Jiang b,c, Peter Hutchins c, Tawfik Badawy c, Hongming Xu b,c,⇑, Xinyu Zhang a,
Alexander Rack d, Paul Tafforeau d
a
International Doctoral Innovation Centre, University of Nottingham Ningbo China, Ningbo, China
b
State Key Laboratory of Automotive Safety and Energy, Tsinghua University, Beijing, China
c
School of Mechanical Engineering, University of Birmingham, Birmingham, UK
d
European Synchrotron Radiation Facility, 38000 Grenoble, France

h i g h l i g h t s

High resolution X-ray scan was performed for the coked injector tip.

Nozzle deposits can lead to extra cavitation inception points.

Nozzle deposits can restrict the air/fuel interaction inside the counterbore.

Coupled nozzle and spray simulations well captured the effect of deposits.

a r t i c l e i n f o a b s t r a c t

Article history: Injector deposit is a common phenomenon for gasoline direct ignition (GDI) engines that greatly affects
Received 12 January 2017 the spray behavior and consequently the combustion performance and emissions. In this study, the effect
Received in revised form 10 March 2017 of deposit on both the in-nozzle flow dynamics and downstream spray behaviors was numerically inves-
Accepted 21 March 2017
tigated. High-resolution X-ray microtomographic scans were performed first to obtain nozzle and deposit
Available online xxxx
morphologies and topology. In-nozzle simulation was then carried out in the Euler-Euler framework with
cavitation taken into account by a homogeneous relaxation model (HRM). Finally, the effect of deposit on
Keywords:
spray behaviors was evaluated in the Euler-Lagrangian framework, coupled with the in-nozzle simulation
Deposit
Nozzle
results. Results of the nozzle flow simulations highlight that the rough surface of the deposits leads to
Cavitation additional cavitation inception and restricts the flow area, causing mass flow rate losses. Deposits inside
CFD the counterbore act as an extension to the inner orifice and constrain the air recirculation. Turbulent
GDI levels at the exit of the counterbore are lower for the coked injector due to the reduced air/fuel interac-
X-ray tion. Spray simulations have shown that deposits would lead to longer spray penetration, a smaller spray
cone angle and larger droplets diameters. Simulation results agree reasonably well with the available
experimental data.
Ó 2017 Published by Elsevier Ltd.

1. Introduction
Injector coking otherwise known as the injector deposit is a
serious issue for GDI engines and it has been observed to have neg-
ative effects on the engine performance and emissions [1]. Com-
pared to traditional port fuel injection (PFI) engines, GDI engines
are more susceptible to injector coking since their injectors are
directly exposed to harsh in-cylinder conditions. Even in the earli-
⇑ Corresponding author at: State Key Laboratory of Automotive Safety and
Energy, Tsinghua University, Beijing, China.
E-mail address: hmxu@mail.tsinghua.edu.cn (H. Xu).
est publications about GDI systems the risks of deposit formation
on injectors were discussed [2,3].
For the traditional PFI engines, low-level fuel rate loss and spray
pattern distortion caused by injector coking could be compensated
by the engine control system, as they commonly operate under a
homogeneous-charge combustion mode and their air/fuel mixture
process is completed in the intake port [3].
With GDI engines, spray characteristics are more critical to
retain a good and stable engine performance. This is especially true
when GDI engines are running at the stratified-charge combustion
mode, where the formation of the desired air/fuel mixture relies
heavily on the carefully designed spray pattern and its interaction

http://dx.doi.org/10.1016/j.apenergy.2017.03.094
0306-2619/Ó 2017 Published by Elsevier Ltd.

Please cite this article in press as: Wang B et al. Numerical analysis of deposit effect on nozzle flow and spray characteristics of GDI injectors. Appl Energy
(2017), http://dx.doi.org/10.1016/j.apenergy.2017.03.094
2 B. Wang et al. / Applied Energy xxx (2017) xxx–xxx

with the intake charge flow [4]. A relatively small amount of detailed investigations of its effect on the spray behaviors are very
deposit accumulation on the injector can however restrict the fuel limited and simulation work is barely seen. It is not clear how the
flow rate, distort such spray patterns and affect the combustion deposits interact with the inner-nozzle fuel flow and alert the
inside the cylinder. This may lead to lower fuel economy, higher downstream spray. This paper is one of the first studies carrying
emissions or even vehicle drivability problems such as misfire out numerical studies of the injector deposit effect on the in-
[1,5,6]. Thus, the injector coking effect is a significant concern for nozzle flow and spray behaviors. Simulations were carried out with
GDI applications and needs great attention. commercial CFD software, CONVERGE [18]. High resolution syn-
The effect of injector deposit on spray characteristic and engine chrotron coherent X-ray microtomographic imaging was con-
performance has been widely reported [7–12]. Song et al. studied ducted first to obtain the detailed nozzle and deposit
the effect of deposit on multi-hole GDI injectors’ spray behavior morphologies. The 3D reconstruction of the data was completed
with a coked injector having been used in a vehicle for with Avizo 9.0.1 software, which allowed distinction between
58,000 km and they claimed that the deposit would increase the phases and also generation of high-spatial resolution models for
spray cone angle and reduce spray penetration [7]. Lindgren both deposited injectors and the metal surface on which they were
et al. studied the effect of deposit on swirl injectors with spray deposited. Two-phase inner nozzle flow simulations, taking into
visualization technologies and found that the fouled injector pro- consideration cavitation effects, were then carried out to study
duced denser and faster pre-jet compared to that of the clean injec- the deposits impact on the nozzle flow behavior. Finally, coupled
tor [8]. Joedicke et al. conducted an accelerated deposit formation with the information provided by the in-nozzle simulation, spray
test using a fuel with additives that could accelerate the deposit calculations were carried out to study the effect of deposit on the
formation [9]. After 55 h of a dirty-up test, 23.5% fuel rate loss spray characteristics. The results can provide practical guidance
was observed accompanied with 20%, 93% increase of HC, CO emis- for the GDI injector design with better deposit resistance.
sions and 2.45% increase of fuel consumption. In the work of Sand-
quist et al., after running for 60 h with a customized engine cycle,
2. X-ray scan and deposit configuration
an 8.5% fuel rate reduction was observed together with 10% higher
HC emissions under the work load of 5.5–8.5 bar IMEP compared
2.1. X-ray scan setup
with a clean injector [10]. Wang et al. used two coked and one
clean multi-hole injector in a single cylinder spray guided DISI
In this study, tomographic scans with high spatial-resolution
research engine to study the effect of injector coking on engine
were conducted at the imaging beam line ID19 [19] of the Euro-
emissions [11]. The injector was fouled after running continuously
pean Synchrotron Radiation Facility (ESRF, Grenoble, France). In
for 54 h with 3.5–8.5 bar IMEP and 1500 rpm engine speed. It was
order to ensure a sufficiently high photon flux density in the hard
found that the coked injector consistently produced higher emis-
X-ray regime, ID19 was operated in pink beam mode with the
sions and the maximum difference was observed at the highest
beamline’s wiggler insertion device as the source (gap: 26 mm):
engine load of 8.5 bar IMEP, where the fouled injector produced
5.6 mm Al and 0.5 mm W attenuators in combination with a
58% higher PN emissions and 300% higher PM emissions. Wang
1.4 mm diamond window suppressed the softer parts of the spec-
et al. conducted numerical studies of the effect of deposit on the
trum [20–22]. The effective spectrum seen by the detector has a
in-cylinder air/fuel mixing process and found that the deposit
mean energy around 90 keV. Additionally, 64 so-called compound
would lead to severe fuel impingement on the piston and cylinder
refractive lenses made from Be were used to collimate the photons
wall due to longer spray penetration and poorer spray dispersion
on the field-of-view of the detector. A mirror placed between the
[12]. Due to a lack of accurate deposit morphologies inside the
scintillator and the lenses allowed for a right-angle design, which
injector, the spray simulations were conducted on the basis of cal-
kept the dose-sensitive components (lenses, electronics) out of
ibration in their study.
the intense hard X-ray beam. The effective pixel size of the detector
Many studies have applied the Scanning Electron Microscopy
is 0.54 mm. A propagation distance between sample and detector of
(SEM) to investigate the GDI injector deposits [7,13–16]. Most of
110 mm was set in order to improve the contrast by propagation-
these studies focused on the deposit outside the injector tip, which
based inline phase contrast. Then 9216 projection images were
had little effect on the spray behaviors. Among the few who stud-
acquired over a 360 degree rotation, with the axis of rotation being
ied the inner-nozzle deposit, Dearn et al. used the Scanning Elec-
shifted slightly off the center of the field-of-view. This approach
tron Microscopy with Energy Dispersive Spectroscopy (SEM-EDS)
allows stitching of projected images, which are taken at a 180
to study deposit formation inside a multi-hole GDI injector [13].
degree angular distance, in order to form a 180 degree tomo-
They showed that extensive deposits were formed in both internal
graphic scan with a lateral extended field-of-view (effectively,
and external injector holes and the external hole deposits were
the new field of view is 3113 pixels wide). For post-processing, a
radially distributed and collected in the shoulder. Song et al. used
2  2 binning was applied.
the SEM technique to visualize the internal injector deposits and
Data were subject to mild sharpening and denoise processing
tried to link the deposit position with its effect on the spray behav-
and then selective phase segmentation using Avizo 9.0.1. The seg-
iors [7]. However, due to the limitations with the SEM technique,
mented 3D volumes were then converted from label fields to high-
the injectors have to be cut in order to give a view to the inner-
resolution surfaces. These surfaces were converted in .stl format
nozzle structures. Detailed 3-D deposits morphologies were still
for input into CFD simulation. Simplification was performed by
not available.
reducing the full resolution surfaces to 5,000,000 triangles with
In a recent work conducted by Xu et al., a compensative review
appropriate smoothing, due to the complexity of the deposited sur-
of injector deposit was conducted [17]. The authors concluded that
faces; since this mitigated the issue of incomplete surface meshes
while extensive work had been done, the effect of GDI injector
for regions with complex topography and provided files of a data
deposit was still not fully understood. It is also pointed out that
size that were useable for workstation computing.
CFD modeling can provide significant help in understanding the
injector coking effect, but little work has been carried out so far
because information about the 3-D structure and morphology of 2.2. Deposit configuration
the deposits inside the nozzle was generally not available [17].
Based on the literature study, it can be concluded that a com- Clean and coked GDI multi-hole injectors were used in this
plete understanding of the GDI injector deposits is important, but study. The coked injector exhibit deposit formation and morphol-

Please cite this article in press as: Wang B et al. Numerical analysis of deposit effect on nozzle flow and spray characteristics of GDI injectors. Appl Energy
(2017), http://dx.doi.org/10.1016/j.apenergy.2017.03.094
 B. Wang et al. / Applied Energy xxx (2017) xxx–xxx 3

ogy which is typical of fuel injectors of this kind. The injector has This model describes the rate at which the instantaneous mass
six nozzle holes. Reconstructed X-ray scan results for both the fraction of vapor in a two-phase mixture will approach its equilib-
clean and coked injectors are shown in Fig. 1. The configurations rium value. The rate equation is given with a simply linearized
of the nozzle hole and its counterbore for the clean injector are form as:
shown in Fig. 1(a). The coked injector was produced in the authors’
dx
x  x
laboratory using a previously developed fouling cycle [11]. Injector ¼ ð5Þ
fouling tests were carried out on a multi-cylinder (V8) thermal
dt h
engine with engine loads sweeping from idle to 8 bar IMEP at where x is the non-equilibrium vapor quality and

x is the equilib-
2000 rpm over 54 h of an engine run. Fuel injection pressure was rium vapor quality, which is given as:
150 bar. The test fuel was Ron 95 Unleaded Gasoline (ULG95),
h  hl;sat
3.3% Ethanol, 6.1% Ethyl Tertiary Butyl Ether (ETBE) without
x ¼ ð6Þ
hv ;sat  hl;sat
Deposit Control Additives (DCA). By the end of the fouling cycle,
large amounts of deposits were accumulated, mainly in the coun- where h is the actual fluid enthalpy of the liquid and vapor mixture,
terbore and very few inside the orifice, as shown in Fig. 1 and hl,sat and hv,sat stand for saturated liquid and vapor enthalpy
(b) and (c). Fig. 1(d) shows a zoomed-in view of the deposited noz- respectively. The time scale h is given as:
zle hole simulated in this work, where the accumulated deposits  
 p p 
occupied about 43.6% of its counterbore volume. h ¼ 3:84  107 a0:54  sat  ð7Þ
pcrit  psat 
3. Computational setup where a is the gas phase volume fraction and pcrit and psat stand for
the saturation and critical pressures respectively.
3.1. Nozzle flow simulation setup The (Pressure Implicit with Splitting of Operator) PISO algo-
rithm was used for velocity/pressure coupling. Spatial discretiza-
To capture the effect of deposit on the nozzle flow behavior, in- tion was second-order accuracy with a central difference scheme
nozzle simulation was carried out with the CFD software CON- whenever possible. An implicit Euler scheme was used for the time
VERGE in the LES framework, with a dynamic structure model integration. Time steps were controlled by the Courant-Friedrichs-
[23] to account for the sub-grid stress; it treats the sub-grid stress Lewy (CFL) number below 0.2. Previous studies and validations
similarly to the Leonard stresses. The sub-grid stress tensor sij is with similar algorithms used on nozzle flow simulations can be
given as: found in Refs. [25,26].
For computational cost considerations, only one injector hole
Lij
si;j ¼ 2ksgs q
ð1Þ was simulated in this study as shown in Figs. 1(b) and 2. The inlet
Lkk
and outlet boundary conditions were set as constant pressures.
where Lij denotes the Leonard stress tensor and can be calculated Three injection pressures were simulated, namely 100, 150 and
as: 200 bar and the exit pressure was kept at 1 bar. Numerical grids
are shown in Fig. 2. As some deposits were accumulated outside
~d
Lij ¼ u iu
^
i u
~j  u ^
j ð2Þ the counterbore exit, a ‘numerical exit plane’ 20 lm downstream
An additional transport equation for the sub-grid turbulent of the original counterbore exit plane was added to the coked
kinetic energy (TKE) ksgs was also included: injector to account for them. Both fixed embedding and adaptive
  mesh refinement methods were applied with mesh sizes ranging
@q

ksgs @ q ~j

ksgs u @ @ksgs from minimum a 5 lm in the orifice and counterbore region, to a
þ ¼ si;j
Sij  q

e  l ð3Þ
@t @xj @xj @xj maximum 20 lm in the sac, which are justified in Section 4.1.
The total mesh number was around 800,000. Due to the state-of-
where
Sij ¼ ð@ u

i =@
xj þ @ u

j =@
xi Þ=2 and the sub-grid dissipation rate e the-art cutting cell method employed in the CONVERGE, the highly
is given by: irregular shape of the deposit surfaces could be acutely captured,
1:5 as shown in Fig. 2(c). Following Som et al. [27,28], five different
ksgs
e¼ ð4Þ fixed needle-lift positions were simulated, 10, 25, 40, 55 and
D 70 lm (maximum needle lift), to facilitate the so-called ‘quasi-
In this framework, the box filter was employed and the filter dynamic’ coupled simulation. Simulation conditions are summa-
size D was fixed with the computational cell size. rized and listed in Table 1. The simulated fuel was isooctane, as a
The HRM model developed by Schmidt et al. [24] was used to typical surrogate for gasoline. Relevant fuel properties are shown
simulate the mass exchange between liquid and vapor (cavitation). in Table 2. Parameters such as the initial droplet diameter, exit

Fig. 1. X-ray 3D synchrotron microtomography results: (a) clean injector; (b) coked injector; (c) deposit only; and (d) zoomed in figure.

Please cite this article in press as: Wang B et al. Numerical analysis of deposit effect on nozzle flow and spray characteristics of GDI injectors. Appl Energy
(2017), http://dx.doi.org/10.1016/j.apenergy.2017.03.094
4 B. Wang et al. / Applied Energy xxx (2017) xxx–xxx

Fig. 2. 3D reconstructed topology (top) and numerical grid (bottom) for the (a) clean injector and (b) coked injector; and (c) zoomed-in view for the coked injector
counterbore.

Table 1 0.125 mm. A detailed grid convergence study with similar setups
In-nozzle simulation conditions. can be found in Ref. [29].
Parameters Values Spray simulations were conducted in the Eulerian-Lagrangian
framework. The Blob method was used to inject the spray into
Fuel Isooctane
Injection pressures 100, 150, 200 bar
the constant volume vessel. The effective drop diameters were ini-
Back pressure 1 bar tialized with the nozzle flow simulation data. In-nozzle simulation
Needle lift 10, 25, 40, 55, 70 lm results such as exit velocity, TKE, turbulent dissipation rate and
Finest mesh size 5 lm cavitation level were also used to initialize the injected droplets.
Duration 100 ls
Spray breakup was simulated using the KH-ACT-RT model origi-
nally developed by Som et al. [27]. In this model, the primary
breakup was simulated with the KH-ACT model taking account of
the cavitation and turbulence effects except for the aerodynamic
Table 2 effect (Kelvin-Helmholtz instability). The KH-RT model was used
Fuel properties of isooctane at 20 °C. to predict the subsequent secondary droplet breakups, taking into
account both KH and RT (Rayleigh-Taylor) instability. Detailed val-
Parameters Values
idation and discussion of applying this model to simulate GDI
Formula C8H18
sprays can be found in former work by the authors [30]. For all
Density (g/ml) 0.702
Surface tension (mN/m) 24.06 transport equations, an unbalanced central scheme was chosen
Kinetic viscosity (mm/s) 0.716 for accuracy consideration. The RNG k-e model was used for the
Specific heat J/(kg K) 2095 turbulence modeling. Droplet collisions were modeled by a no time
counter method [31]. The effects of turbulence on the droplet were
included using a turbulent dispersion model [32]. The Frossling
correlation was used to model the droplet evaporation and
velocity, TKE and cavitation were calculated at the counterbore the drag force was calculated by the dynamic drag model [33];
exit for each needle-lift position and together provided initializa- based on the assumption that the drag coefficient was dependent
tion for the spray simulation. on the droplet shape which could vary between a sphere and a
disk.
3.2. Spray simulation setup
4. Results and discussion
The full domain of a constant volume vessel was used for the
spray simulation. The vessel was a cylinder of 100 mm diameter The effects of deposit on nozzle flow and spray characteristics
and 150 mm length. The base grid size was fixed at 2 mm were modeled in two stages. The in-nozzle simulation was per-
with an adaptive mesh refinement method applied to the velocity formed first in the Euler framework. Then spray simulation was
field; resulting in the mesh size of the main spray region of conducted in the Euler-Lagrangian framework, with the nozzle
0.25 mm. To resolve the flow in the near nozzle region, a fixed flow data as the initialization. The results and discussion are pre-
embedding method was used with the finest mesh size of sented in the later sections.

Please cite this article in press as: Wang B et al. Numerical analysis of deposit effect on nozzle flow and spray characteristics of GDI injectors. Appl Energy
(2017), http://dx.doi.org/10.1016/j.apenergy.2017.03.094
 B. Wang et al. / Applied Energy xxx (2017) xxx–xxx 5

4.1. Nozzle flow simulation

(1) Model validation and mass flow rate

The grid independency analysis results under the condition of
Pinj = 150 bar and Pback = 1 bar are shown in Fig. 3. Five different
mesh sizes ranging from 15 to 2.5 lm were tested. The simulated
mass flow rate was averaged in a quasi-steady period equal to
around 50 times as the length of time for a fluid particle to travel
from the nozzle inlet to the counterbore exit [34], Fig. 3(a). The
results show that the coarse grid tended to underestimate the mass
flow rate. With the mesh getting finer, the predicted mass flow
rates increased slightly. As the mesh size reached 5 lm, the pre-
dicted mass flow rate was maintained at a stable level, which
demonstrated that the CFD solution was relatively independent
of the grid size. Therefore, the mesh size of 5 lm was used in this
study as a compromise of accuracy and computational cost.
In order to validate the CFD simulation model, the simulated
mean mass flow rate was also compared with the test data, as
shown in Fig. 4 at three different injection pressures. The experi- Fig. 4. Comparison of tested and simulated mass flow rate versus injection
pressure.
mental test was conducted with an in-house built, long tube injec-
tion rate measuring system based on the Bosch method. Details of
the test rig and experiment setup can be found in Ref. [35]. As can (2) Effect of deposit on injector flow characteristic
be observed, at any injection pressure, the CFD results generally
matched well with the test data. Around 9% mass flow rate reduc- Fig. 5 presents the averaged vapor volume fraction (VVF) con-
tion was observed for the coked injector compared to the clean tours for the coked and clean injectors at the mid-plane and four
injector for all the injection pressures. This difference is much cross sections of the nozzle and counterbore with Pinj = 150 bar
higher than the reported deviation between injectors due to man- and Pback = 1 bar with vapor phase shown in red. The mid-plane
ufactory inconsistence [7]; which means that if no adjustment was views indicated that cavitation occurred at the orifice inlet for both
taken, the deposits would restrain the injected fuel amount and the injectors. When flow entered the orifice, the sharp bend
lead to lower power output of the engine. Although the mass flow resulted in large velocity and pressure gradients and caused cavita-
rate loss could be detected and compensated by the ECU by pro- tion inception in this area. Most of the cavitation occurred in the
longing the injection duration, this would lead to longer spray pen- upper surface (right side corner) due to the orifice orientation. Cav-
etration and potentially fuel impingement at the piston and itation restricted the effective flow area and led to lower mass flow
cylinder walls. rate. The cavitation length at the orifice inlet was shorter for the
The deviation between the experimental and computational coked injector. It is believed that the rough surface of deposits
values was less than 4%. For the clean injector, the CFD calculations increased the flow resistance and the local pressure inside the noz-
slightly underestimated the experimental results for all three pres- zle hole which suppressed the cavitation development. In addition
sures; while for the coked injector, the simulation slightly overes- to the orifice inlet, cavitation inception could also be found on the
timated the results. The difference may be because the deposit deposit surfaces inside the counterbore for the coked injector. The
surface roughness was smoothed for the simulation. Due to the surface irregularities and pits formed by the deposit caused flow
complexity of cavitation flow modeling, it is considered that the separation and thus led to cavitation. The residual gas in the
CFD simulation is able to predict the flow behavior with a sufficient deposit pits were also believed to contribute to the cavitation
degree of confidence. inception. These cavitation inception points can be more clearly

Fig. 3. (a) An example of simulation mass flow rate at the counterbore exit and (b) grid independency analysis.

Please cite this article in press as: Wang B et al. Numerical analysis of deposit effect on nozzle flow and spray characteristics of GDI injectors. Appl Energy
(2017), http://dx.doi.org/10.1016/j.apenergy.2017.03.094
6 B. Wang et al. / Applied Energy xxx (2017) xxx–xxx
Fig. 5. Averaged vapor fraction contours for the coked (upper) and clean (bottom) injectors.
observed with the cross-sectional views in Fig. 5. The deposit
caused cavitation developed along the deposit surface through
the counterbore exit.
The velocity contours are shown in Fig. 6. For the clean and
coked injectors, the velocity distributions appeared to be similar
to each other from upstream as far as the orifice inlet. Inside the
counterbore, clear air recirculation zones could be observed for
the clean injector. The air/fuel interaction was beneficial for the
disintegration of the liquid column into droplets and better
atomization of the spray, which also increased the near-field spray
angle [30]. The liquid distribution was uneven due to the nozzle
orientation. Nevertheless, for the coked injector, deposit forma-
tions changed the counterbore geometry, which acted as the
extension of the orifice and constrained the air entrainment. Lower
air entrainment resulted in less air/fuel interaction and conse-
quently weaker atomization and a smaller spray cone angle inside
the counterbore. The weak air/fuel interaction was also believed to
contribute to the downstream smaller spray angle and larger dro-
plet sizes for the coked injector spray, as previously observed by
other researchers [6]. The cavitation vapor initiating from deposits
inside the inner hole produced a low velocity vapor region which
can be observed at the upper surface (right side) of the exiting
jet from the inner hole, Fig. 6. The presence of deposit and the
low velocity vapor region together restricted the effective flow area
and resulted in the flow rate loss shown in Fig. 4.
(3) Effect of deposit under different injection pressures
Fig. 7 shows the VOF contours for the clean and coked injectors
under different injection pressures. The results demonstrated that
cavitation already occurred at 100 bar injection pressure for both
injectors. For the clean injector, the cavitation inception always
occurred at the corners of the orifice inlet. With higher injection
pressure, the cavitation area expanded further downstream, but
the flow always reattached back to the nozzle wall before entering
into the counterbore. Higher injection pressures also contributed
Please cite this article in press as: Wang B et al. Numerical analysis of deposit effect on nozzle flow and spray characteristics of GDI injectors. Appl Energy
(2017), http://dx.doi.org/10.1016/j.apenergy.2017.03.094
to higher flow velocity and stronger fuel/air interaction, as shown
in Fig. 8. More pronounced recirculation zones were found at the
counterbore for the clean injector. For the coked injector, the liquid
phase velocity increased with the injection pressure increase
similar to the clean injector. The deposit caused cavitation inside
the counterbore was found at all three injection pressures for
the coked injector. Larger cavitation caused by the deposit led to
a further blockage at higher injection pressures; this resulted in
the higher fraction of mass flow rate loss shown in Fig. 4
and reduced the discharge coefficient almost linearly by 11.2%,
when the injection pressure was increased from 100 to 200 bar,
Fig. 9(a).
Fig. 9(b) presents a comparison of the mean exit velocity from
the counterbore for the clean and coked injectors at different injec-
tion pressures. As a general trend, higher injection pressure led to
higher exit velocities. Although the rough surface of deposit tended
to increase the flow resistance for the coked injector, the restricted
flow area and smaller cone angle of the flow resulted in higher
counterbore exit velocity. The absence of air/fuel interaction also
restricted the turbulence development, resulting in much lower
TKE for the coked injector as shown in Fig. 9(c). Both the exit veloc-
ity and TKE were calculated for the liquid flow region with
VOF 5 0.5. These differences in the nozzle flow behavior would
have implications on the atomization behavior of the spray, as
the flow velocity, the TKE, as well as the cavitation, play important
roles in the primary breakup.
4.2. Spray simulation
(1) Coupling with in-nozzle simulation
In order to couple the spray simulation with the nozzle flow
simulation results, the nozzle flow characteristics at different nee-
dle lifts were calculated and combined together as inputs for the
spray simulations. The needle was fully open during 80% of the
injection duration. Fig. 10 presents the calculated rate of injection,
 B. Wang et al. / Applied Energy xxx (2017) xxx–xxx 7

Fig. 6. Averaged velocity contours for the coked (upper) and clean (bottom) injectors.

100 bar 150 bar 200 bar

Fig. 7. Averaged vapor fraction contours at different injection pressures.

area coefficient, TKE and turbulent dissipation rate as functions of
time with 150 bar injection pressure and 1 bar as the ambient
pressure. These parameters were all averaged properties at the
counterbore exit obtained by the 3D flow simulation inside the
injector. The five different needle lift positions, where in-nozzle
simulations were performed are indicated by dashed lines. Consis-
tent with the results presented earlier, the deposits lead to lower
ROI and area coefficient due to the enhanced cavitation and
restricted flow area. The TKE and TDR were also lower for the
coked injector due to the restricted air/fuel interaction.

Please cite this article in press as: Wang B et al. Numerical analysis of deposit effect on nozzle flow and spray characteristics of GDI injectors. Appl Energy
(2017), http://dx.doi.org/10.1016/j.apenergy.2017.03.094
8 B. Wang et al. / Applied Energy xxx (2017) xxx–xxx

100 bar 150 bar 200 bar

Fig. 8. Averaged velocity contours at different injection pressures.

Fig. 9. (a) Averaged discharge coefficient; (b) exit velocity and (c) TKE of the liquid flow at the counterbore exit.

Fig. 10. (a) Profile of injected mass and area coefficient and (b) turbulent kinetic energy and dissipation rate.

Please cite this article in press as: Wang B et al. Numerical analysis of deposit effect on nozzle flow and spray characteristics of GDI injectors. Appl Energy
(2017), http://dx.doi.org/10.1016/j.apenergy.2017.03.094
 B. Wang et al. / Applied Energy xxx (2017) xxx–xxx 9

(2) Spray simulation results

Results in the previous sections have shown that the deposit
had a significant influence on the in-nozzle flow behaviors, which
implies that the downstream spray characteristics could also be
affected. Fig. 11 shows the computed spray development of the
clean and coked injectors in terms of droplet diameters at different
times ASOI, with the injection pressure of 150 bar and back pres-
sure of 1 bar. Generally, the overall spray morphologies’ develop-
ment was similar for the both the injectors. Nevertheless, the
coked injector tended to have a longer spray penetration and a
smaller cone angle, compared to the clean injector as a result of
its higher exit velocity and smaller nozzle flow cone angle,
observed in the earlier section. These observations are consistent
with earlier high speed imaging results for the same injectors, as
previously reported by Wang et al. [12]. The smaller spray cone
angle implies weaker spray dispersion and the longer penetration
might lead to fuel impingement at the piston and cylinder walls;
both of which are unfavorable for the consequent combustion per-
formance and emissions.
Fig. 12 presents spray penetration versus time, for the clean and Fig. 12. Measured and calculated penetration length for the clean and coked
coked injectors from both the calculated results and previous injector.
experimental data [12]. Calculations were made for both the KH-
ACT-RT model and the classical KH-RT model. As indicated by
the figure, injector deposit led to larger spray penetration com-
pared to the clean injector, as a result of its smaller spray cone
angle and higher exit velocity. The predicted spray penetration of
the KH-ACT-RT model coupled with in-nozzle simulation matched
well with the experimental data. In contrast, the KH-RT model
tended to over-predict the penetration length. Due to the absence
of nozzle flow information in the KH-RT model, such as cavitation
and turbulence which are the destabilizing forces and promote
atomization for the spray, the results underestimated the spray
atomization and thus this led to longer penetration. On the other
hand, the KH-ACT-RT model captured the effects of cavitation
and turbulence and better predicted the penetration.
The calculated Sauter mean diameters (SMD) from the KH-ACT
model for the clean and coked injectors during the main injection
event are presented in Fig. 13 at different distances from the injec-
tor tip. The simulation results were compared to available PDPA
data with good agreement. The observed increase in the SMD with
increasing distance from the injector tip was due to more rapid
evaporation of small droplets in the spray. The averaged droplet
SMD increase for the coked injector is around 9.5%. A weaker Fig. 13. Measured and calculated SMD for the clean and coked injector at difference
breakup process was noticeable for the coked injector, indicated distance from the injector tip.

Time Clean injector Coked injector

0.1
ms

0.4
ms

0.7
ms

1.0
ms

Fig. 11. Computed spray morphologies of the clean and coked injectors at different times ASOI (Pinj = 150 bar, Pback = 1 bar).

Please cite this article in press as: Wang B et al. Numerical analysis of deposit effect on nozzle flow and spray characteristics of GDI injectors. Appl Energy
(2017), http://dx.doi.org/10.1016/j.apenergy.2017.03.094
10 B. Wang et al. / Applied Energy xxx (2017) xxx–xxx
by its larger SMD. As observed earlier, the deposit acted as an
extension of the orifice and constrained the air/fuel interaction
inside the counterbore; which resulted in a significant lower TKE
value and thus weaker atomization. Although the higher level of
cavitation in the coked injector tended to offset part of this effect,
the coked injector produced larger droplet diameters. This had a
negative effect on the downstream in-cylinder air/fuel mixture
preparation process as the larger droplets will take a longer time
to fully evaporate.
5. Conclusions
This paper has presented a computational study on the effect of
deposit on the nozzle flow and spray characteristics of GDI injec-
tors. Detailed 3-D injector flow simulations were performed for
the nozzle and deposit configurations obtained from high resolu-
tion X-ray microtomography data. The obtained nozzle flow infor-
mation such as the mass flow rate, spray velocity, cavitation and
turbulence levels were then used to initialize the spray simulation.
Results have revealed that the CFD modeling with accurate deposit
morphologies can be effectively used to study the deposit effect.
The main conclusions drawn from the study are as follows:
(1) Spray simulation coupled with the nozzle flow simulation is
able to accurately capture the effect of deposit on the spray
behavior. On the contrary, simulations without nozzle flow
information tend to over-predict the spray penetration.
(2) The deposits inside the counterbore act as an extension of
the orifice and restrict the air recirculation and entrainment.
This leads to the lower exiting turbulent kinetic energy of
the coked injector spray and contributes to the weaker
atomization performance of the coked injector.
(3) The rough surfaces of the deposits create extra cavitation
inceptions inside the counterbore. The deposits and cavita-
tion together restrict the effective flow area, resulting in a
reduction in the mass flow rate. With higher injection pres-
sure, cavitation area growth is observed for both the clean
and coked injectors.
(4) Deposits lead to longer spray penetration, due to the higher
exit velocity and smaller spray cone angle. Although cavita-
tion helps the break-up and atomization process, the effect
is less dominant than the reduction of the TKE, resulting in
larger droplet diameters. Compared to the clean injector,
the SMD of the coked injector is increased by around 9.5%
due to restricted turbulence development.
Acknowledgement
The authors gratefully acknowledge project supported by
National Natural Science Foundation of China with Grant No.
51636003. The finical support by EPSRC program at University of
Birmingham with project code EP/N021746/1 is also gratefully
acknowledged. The authors also acknowledge the financial support
from the International Doctoral Innovation Centre, Ningbo Educa-
tion Bureau, Ningbo Science and Technology Bureau, China’s MOST,
University of Nottingham and the partial support by Ningbo Natu-
ral Science Foundation Program with project code 2013A610107.
References
[1] Zhao F, Lai MC, Harrington DL. Automotive spark-ignited direct-injection
gasoline engines. Prog Energy Combust Sci 1999;25:437–562.
[2] Iwamoto Y, Noma K, Nakayama O, Yamauchi T, Ando H. Development of
gasoline direct injection engine. SAE technical paper; 1997 [1997-05-41].
Please cite this article in press as: Wang B et al. Numerical analysis of deposit effect on nozzle flow and spray characteristics of GDI injectors. Appl Energy
(2017), http://dx.doi.org/10.1016/j.apenergy.2017.03.094
[3] Kinoshita M, Saito A, Matsushita S, Shibata H, Niwa Y. A method for
suppressing formation of deposits on fuel injector for direct injection
gasoline engine. SAE technical paper; 1999 [1999-01-3656].
[4] Stone R. Introduction to internal combustion engines; 2012.
[5] Aradi AA, Evans J, Miller K, Hotchkiss A. Direct injection gasoline (DIG) injector
deposit control with additives. SAE technical paper; 2003 [2003-01-2024].
[6] Arters DC, Macduff MJ. The effect on vehicle performance of injector deposits
in a direct injection gasoline engine. SAE technical paper; 2000 [2000-01-
2021].
[7] Song H, Xiao J, Chen Y, Huang Z. The effects of deposits on spray behaviors of a
gasoline direct injector. Fuel 2016;180:506–13.
[8] Lindgren R, Skogsberg M, Sandquist H, Denbratt I, Sandquist H. The influence of
injector deposits on mixture formation in a DISC SI engine. JSAE 2003.
[9] Joedicke A, Krueger-Venus J, Bohr P, Cracknell R, Doyle D. Understanding the
effect of DISI injector deposits on vehicle performance. SAE technical paper;
2012 [2012-01-0391].
[10] Sandquist H, Denbratt I, Owrang F, Olsson J. Influence of fuel parameters on
deposit formation and emissions in a direct injection stratified charge SI
engine. SAE technical paper; 2001 [2001-01-2028].
[11] Wang C, Xu H, Herreros JM, Wang J, Cracknell R. Impact of fuel and injection
system on particle emissions from a GDI engine. Appl Energy
2014;132:178–91.
[12] Wang B, Badawy T, Jiang Y, Xu H, Ghafourian A, Zhang X. Investigation of
deposit effect on multi-hole injector spray characteristics and air/fuel mixing
process. Fuel 2017;191:10–24.
[13] Dearn K, Xu J, Ding H, Xu H, Weall A, Kirkby P, et al. An investigation into the
characteristics of DISI injector deposits using advanced analytical methods.
SAE Int J Fuels Lubr 2014;7:771–82.
[14] Von Bacho PS, Galante-Fox J, Sant DW. Development of a robust injector design
for superior deposit resistance. SAE technical paper; 2005 [2005-01-3841].
[15] Aradi AA, Imoehl B, Avery NL, Wells PP, Grosser RW. The effect of fuel
composition and engine operating parameters on injector deposits in a high-
pressure direct injection gasoline (DIG) research engine. SAE technical paper;
1999 [1999-01-3690].
[16] Imoehl W, Gestri L, Maragliulo M, Del-Frate L, Klepatsch M, Wildeson R. A DOE
approach to engine deposit testing used to optimize the design of a gasoline
direct injector seat and orifice. SAE Int J Fuels Lubr 2012;5 [2012-01-1642].
[17] Xu H, Wang C, Ma X, Sarangi AK, Weall A, Krueger-Venus J. Fuel injector
deposits in direct-injection spark-ignition engines. Prog Energy Combust Sci
2015;50:63–80.
[18] Richards K, Senecal PK, Pomraning E. CONVERGE (version 2.3.0)
manual. Madison, WI: Convergent Science Inc.; 2015.
[19] Paganin D, Mayo SC, Gureyev TE, Miller PR, Wilkins SW. Simultaneous phase
and amplitude extraction from a single defocused image of a homogeneous
object. J Microsc 2002;206:33–40.
[20] Sanchez S, Ahlberg PE, Trinajstic KM, Mirone A, Tafforeau P. Three-dimensional
synchrotron virtual paleohistology: a new insight into the world of fossil bone
microstructures. Microsc Microanal 2012;18:1095–105.
[21] Weitkamp T, Haas D, Wegrzynek D, Rack A. ANKAphase: software for single-
distance phase retrieval from inline X-ray phase-contrast radiographs. J
Synchrotron Radiat 2011;18:617–29.
[22] Cloetens P, Barrett R, Baruchel J, Guigay J-P, Schlenker M. Phase objects in
synchrotron radiation hard X-ray imaging. J Phys D Appl Phys
1999;29:133–46.
[23] Pomraning E, Rutland CJ. Dynamic one-equation nonviscosity large-eddy
simulation model. AIAA J 2002;40:689–701.
[24] Schmidt DP, Gopalakrishnan S, Jasak H. Multi-dimensional simulation of
thermal non-equilibrium channel flow. Int J Multiph Flow 2010;36:284–92.
[25] Battistoni M, Xue Q, Som S. Large-eddy simulation (LES) of spray transients:
start and end of injection phenomena. Oil Gas Sci Technol – Rev d’IFP Energies
Nouv 2016;71:24.
[26] Zhao H, Quan S, Dai M, Pomraning E, Senecal PK, Xue Q, et al. Validation of a
three-dimensional internal nozzle flow model including automatic mesh
generation and cavitation effects. J Eng Gas Turbines Power 2014;136:92603.
[27] Som S, Aggarwal SK. Effects of primary breakup modeling on spray and
combustion characteristics of compression ignition engines. Combust Flame
2010;157:1179–93.
[28] Som S, Longman DE, Ramírez AI, Aggarwal SK. A comparison of injector flow
and spray characteristics of biodiesel with petrodiesel. Fuel 2010;89:4014–24.
[29] Senecal PK, Pomraning E, Richards KJ, Som S. Grid-convergent spray models for
internal combustion engine computational fluid dynamics simulations. J
Energy Resour Technol 2013;136:12204.
[30] Tu P. Numerical and experimental study of spray characteristics in the
gasoline direct injection engine [PhD thesis]. University of Birmingham; 2016.
[31] Schmidt DP, Rutland CJ. A new droplet collision algorithm. J Comput Phys
2000;164:62–80.
[32] Amsden AA, O’Rourke PJ, Butler TD. KIVA-II: a computer program for
chemically reactive flows with sprays; 1989.
[33] Liu AB, Mather D, Reitz RD. Modeling the effects of drop drag and breakup on
fuel sprays. SAE technical paper 930072; 1993.
[34] Desantes JM, Salvador FJ, Carreres M, Martinez-Lopez J. Large-eddy simulation
analysis of the influence of the needle lift on the cavitation in diesel injector
nozzles. Proc Inst Mech Eng D J Automob Eng 2014;229:407–23.
[35] Tay RS. Fuel injection rate measurement device. Univ Birmingham; 2012 [final
year thesis].