Advances in Engineering Software 49 (2012) 43–53

Contents lists available at SciVerse ScienceDirect

Advances in Engineering Software

journal homepage: www.elsevier.com/locate/advengsoft

CFD optimization for GDI spray model tuning and enhancement

of engine performance
M. Costa ⇑, U. Sorge, L. Allocca
CNR – Istituto Motori, Viale Marconi 8, 80125 Naples, Italy

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

Article history: Coupling a 3D Computational Fluid Dynamics (CFD) tool with a rigorous method of decision making is
Received 14 October 2011 becoming indispensable in the design process of complex systems, as internal combustion engines.
Received in revised form 3 February 2012 CFD based optimization (CFD-O) is here carried out on a single cylinder, four-valve, four-stroke gasoline
Accepted 25 March 2012
direct injection (GDI) engine, to enhance mixture formation under stratified charge operation, hence to
choose between the single or double injection strategy maximizing the engine power output. A 3D engine
model is coupled with the Simplex algorithm to find the optimal synchronization of both injection and
Keywords:
spark timing within the working cycle.
Gasoline direct injection
Spark ignition engines
CFD-O is also addressed to perform the validation of the gasoline spray model, that otherwise reveals
Multidimensional modelling tedious and time-consuming. The Simplex algorithm is used to tune the constants entering a model
Split injection developed by authors, as applied to three different high pressure GDI injectors, preliminary experimen-
CFD based optimization tally characterized.
Charge stratification Fully automatic procedures are assessed to be exploited in the phase of engine design, whose contribu-
tion may be of great importance to reduce development costs and time-to-market of new technologies.
Ó 2012 Elsevier Ltd. All rights reserved.

1. Introduction Tools and techniques used for the design of yesterday engines,
indeed, are insufficient for the challenges of today new engines,
The worldwide concern about the environmental impact of en- due to their increased complexity, or even to the fact that the fuels
ergy conversion systems has, as a relapse in the automotive field, landscape is continuously evolving and becoming more complex.
the imposition of strict government regulations relevant to pollu- Major issues of concern in engine simulations are chemical kinetics
tant emissions and fuel-efficiency standards of vehicles. Nowadays, and spray modelling. Most of the spray sub-models needs calibra-
the preferred route towards the reduction of both engine exhaust tion, namely tuning of a number of involved constants, whose va-
noxious emissions and fuel consumption remains the control of lue is to be changed, possibly, as the injector is changed, or even if
the mixture formation and combustion processes taking place the injection pressure is raised. This requires that valuable innova-
within the engine combustion chamber, a complicated task, af- tion time must be spent in adjusting or adding complexity to the
fected by many variables. The high cost and time needed to achieve model, or in finding the optimal combination of spray model
optimization through bench testing alone has drawn interest of en- parameters and grid size (the grid dependence of the spray calcu-
gine developers towards the use of Computational Fluid Dynamics lations is another detrimental aspect on model reliability). Analo-
(CFD) analyses. The critical link between the need to reduce green- gous problems are encountered in the assessment of combustion
house gas emissions and the use of advanced engine simulation is sub-models. Increasing attention, therefore, is today being devoted
highlighted in a document recently published by one of the scien- towards coupling traditional 3D engine simulation tools with algo-
tific laboratories of the US Department of Energy (DOE), where it is rithms able to explore the model constants space in an automatic
said that engine manufacturers need to ‘‘change from a test-first way, as genetic or robust search methods. During each iteration,
culture to an Analysis-Led Design Process’’ and that ‘‘a predictive the decision variables are manipulated using various operators
simulation toolkit would accelerate the market transformation to (selection, combination, crossover or mutation) to create new de-
high-efficiency, clean power sources for transportation’’ [1]. sign populations, i.e. new sets of decision variables. Optimization
algorithms, on the other hand, may be used to drive the choice of
a design solution, or configuration, between various alternatives,
hence in a role that is more congenial and traditional [2]. Examples
⇑ Corresponding author. of application of genetic algorithms to the design of diesel engines
E-mail address: m.costa@im.cnr.it (M. Costa). are found in the paper by Wickman et al. [3] and De Risi et al. [4]. A

0965-9978/$ - see front matter Ó 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.advengsoft.2012.03.004
44 M. Costa et al. / Advances in Engineering Software 49 (2012) 43–53

more recent application by Dempsey and Reitz [5] allows exploring

the possibility of reducing the pollutant emissions in a reactivity
controlled engine fed with diesel and gasoline fuels. Design fitness
determination through coupling between a 3D engine model and a
1D gas dynamics code for the simulation of the gas exchange pro-
cess (intake and exhaust phases) is addressed in the work of Shriv-
astava et al. [6].
Present work is aimed at discussing two activities performed by
authors within the framework of the numerical study of high per-
formance gasoline direct injection (GDI) spark ignition (SI) en-
gines: the development of a 3D numerical model for the spray
dynamics, whose prediction capability is improved through a Sim-
plex optimization algorithm [7], and the assessment of a procedure
for the engine fuel consumption reduction, based on the optimal
synchronization of the injection event within the working cycle.
Fig. 1. Homogeneous (left) and stratified (right) charge mode in a GDI engine [9].
The work represents a revised and updated version of the paper
authored by Sorge et al. [8], presented at the 7th International Con-
ference on Engineering Computational Technology – 2010. sion ignition ones. Split injections have already been adopted un-
In order to assess a 3D spray model characterized by a high por- der special operating regimes, as to increase the temperature for
tability, three commercial multi-hole injectors are preliminary the converter light-off, to achieve a smooth idle and to reduce
experimentally characterized both at the mass flow rate test bench the engine tendency to knock [11–13]. The effects of a single or
and in an optically accessible vessel. At each injection pressure, the double injection strategy on the performance of a single cylinder
error between the numerically computed and the experimentally four-valve, four-stroke GDI engine are here numerically analysed.
measured penetration length is minimized by varying the value The runs of a 3D engine model are driven by the Simplex algorithm
of properly chosen constants in a preliminary defined design of to search for the injection timing and spark advance that reveal
experiment (DOE) space. This approach represents a novelty with optimal for the combustion development. In this sense, present
respect to the work performed in Ref. [8], where just one of the work represents an important step beyond the mere parametric
three injectors was considered, and the tuning of the relevant analysis performed in Ref. [8], since the engine performance
spray model constants was made through a trial and error improvement is here guaranteed by the use of an optimization
procedure. algorithm that automatically manages the 3D model towards
The optimization problem intended to the reduction of the fuel exploiting the actual advantages of the injection strategy.
consumption of a GDI engine is aimed at the choice of the injection
strategy most proper to realize an effective energy conversion pro-
cess under a lean-mixture, moderate-load, moderate-speed case. 2. Numerical simulation of GDI sprays
GDI is nowadays one of the most pursued solutions to improve
the performance of SI engines from both an energetic and an envi- The problem of spray modelling in internal combustion engines
ronmental point of view. This is mainly due to the possibility to applications has been discussed in numerous monographs and re-
precisely control and adapt the fuel amount and injection timing view papers (e.g. [14–16]). Sprays belong to a specific type of two-
to the specific load and speed operating condition. The greatest phase flows, characterized by a dominating direction of motion
fuel efficiency advantages of GDI engines are achieved by resorting and involving a liquid phase in the discrete form of droplets and
to different combustion concepts, namely homogeneous stoichi- ligaments, and a gas phase as a continuum. Sprays are the results
ometric and stratified overall lean. At the higher loads the engine of high pressure-driven liquid jets injected through one or several
has to work with a homogeneous charge, realized with an early injector nozzle orifices into a gaseous atmosphere. The jet atomizes
injection during the intake stroke, so that the air-to-fuel ratio is into different liquid fragments, and, finally, into droplets, which
maintained stoichiometric or slightly rich. At the lower loads the may suffer evaporation.
possibility to resort to stratified charges, rich around the spark plug Currently, the most common spray description is based on the
and with an increasing air-to-fuel ratio towards the cylinder walls, Lagrangian discrete droplet method (DDM) [17]. While the contin-
is to be exploited, that allows working under overall lean condi- uous gaseous phase is described by the standard Eulerian conser-
tions for low fuel consumption, without combustion instabilities. vation equations, the transport of the dispersed phase is
The lowest temperature of the mixture close to the cylinder walls calculated by tracking the trajectories of a certain number of rep-
reduces the wall heat losses, the HC and CO formation, and resentative parcels (particles). A parcel consists of a number of
strongly increases the engine volumetric efficiency. droplets and it is assumed that all the droplets within one parcel
Fig. 1, from Ref. [9], shows the schematic view of the combus- have the same physical properties and behave equally when they
tion chamber of two GDI engines, both exhibiting an intake and move, break up, hit a wall or evaporate. The coupling between
an exhaust valve, the spark plug in central position and the injector the liquid and the gaseous phases is achieved through source terms
on the right. The left and the right sides of the figure represent the representing the exchange of mass, momentum and energy. Vari-
homogeneous charge and the stratified concepts. The time of injec- ous sub-models account for the effects of turbulent dispersion
tion is chosen during the intake stroke (piston moving from the top [18], coalescence [19], evaporation [20], wall interaction [21] and
to the bottom) for homogeneous charge, during the compression droplet break up [22].
stroke (piston moving from the bottom to the top) for stratified In the present section, the assessment of a simulation tool
charge. reproducing the spatio-temporal dynamics of sprays issuing from
Achievement of optimal charge conditions in direct injection new generation high pressure injectors under various operating
stratified charge (DISC) lean combustion operation is affected by conditions is presented. The model, developed within the AVL
many parameters, whose effects are complex and overlapped FireTM code environment [23], is conceived to exploit a properly
[10]. This is the reason why injection modulation and splitting made experimental campaign. Collected data serve in part as input
are being considered also in SI engines, in analogy with compres- parameters, in part as terms of comparison for the numerical
 M. Costa et al. / Advances in Engineering Software 49 (2012) 43–53 45

Table 1 the spray propagation. Alignments of the jet directions with re-
Geometrical and flow rate characteristics of the three tested injectors. spect to the camera axis are actuated by a wet seal spherical holder
Injector type # Holes Hole diameter Static flow
enabling to tilt the injector in the angular range ±15°. The tip pen-
(mm) at 10 MPa (g/s) etration of the considered jet, as well as the cone angle, is collected
Injector #1 BOSCH HDEV 5.1 6 0.193 13.7
as a function of time. Processing of images is based on background
Injector #2 BOSCH HDEV 5.1 7 0.179 13.7 subtraction, filtering and edges determination. All the measure-
Injector #3 CONTINENTAL 6 0.190 13.1 ments are made on five-image averaged pictures for a statistical
analysis of the cycle-to-cycle dispersion. A plateau value of the
cone angle is achieved when the spray is completely developed
(t  500 ls).
results. A brief description of the experimental activity is given in The injection strategies in the experimental campaign cover the
the following. entire injection pressure range for the three injectors. The pulse
durations are calibrated to deliver 10, 20 and 50 mg of gasoline
2.1. Experimental characterization of the GDI spray dynamics from at different injection pressures. Some single injection tests are re-
multi-hole injectors ported in Table 2. Fig. 3 reports a typical energizing current signal
to the solenoid for injecting 20 mg of fuel at the pressure of
Three commercial multi-hole injectors suitable to be mounted 10 MPa, and the corresponding fuel injection rate signals, as col-
on high-performances SI engines are tested. Table 1 reports the lected for the three injectors. The signals are averaged over one
number and diameter of holes, as well as the exact flow rates of hundred shots. A shift of 0.35 ms is registered between the start
the considered injectors. The axes of the single jets coming from of the energizing current and the exiting of the fuel from the noz-
the nozzles are configured to depict different spray footprint struc- zle, indicating a postponed answer of the mechanical parts. This
tures. Two injectors are manufactured by Bosch, type HDEV 5.1, delay remains practically unchanged for all the three devices. Dif-
differing for the number of holes, six for Injector #1, seven for ferently, the fuel injection rate signals show different rise times:
Injector #2, distributed regularly on a circumference to form an for both the Bosch injectors it is of about 70 ls, while for the
ellipsoidal-like hollow-cone geometry. The third injector is a six-
hole Continental device, with five holes distributed over a circum- Table 2
ference and the sixth one in central position. Fig. 2 represents three Time durations of the pulses for the desired fuel amounts at the indicated injection
sketches, each drawing the position of the holes on the relevant pressures.
injector and the footprint of the spray axes on a plane placed at a Pinj (MPa) 3 6 10 10 15 20 23
distance of 30 mm from the holes themselves. tinj (ls) 1000 1900 1450 3600 2900 2600 2500
The maximum operating pressure for all the three injectors is Qinj (mg/str) 10 20 20 50 50 50 50
forced up to 25 MPa for Injector #1 and #2, up to 20 MPa for Injec-
tor #3. Commercial gasoline is used (q = 740 kg/m3), delivered by a
hydro-pneumatic injection system without rotating organs. The
12
system is managed by a programmable electronic control unit Pinj 10 MPa
solenoid current [A]

10 tinj = 1,45 ms
(PECU) enabling the definition of the strategy typology in terms
of number of injection events, durations and dwell times. 8 Qinj = 20.00 mg/str
Two types of analysis are conducted: (a) instantaneous mass 6
flow rates of issuing gasoline are measured by means of an AVL 4
meter operating on the Bosch principle [24,25], under both single 2
and double injection strategies; (b) image processing techniques 0 0.35 ms
are applied to derive the single jet penetration length and cone an- Inj. 1 - 6 holes
gle over time in the single injection case. 20 Inj. 2 - 7 holes
fuel inj. rate [mg/ms]

The measured instantaneous mass flow rate profile is integrated 15 Inj. 3 - 6 holes
over the injection interval of time to gain the total injected mass,
10
and to verify that the value of this last quantity is in accordance
5
with that measured by means of a precision balance. The study
of the fuel dispersion, instead, is realized in an optically-accessible 0

high-pressure quiescent vessel containing air at atmospheric back- -5

pressure and ambient temperature. The jets are enlightened by
0,0 0,5 1,0 1,5 2,0 2,5 3,0
powerful flashes at different instants from SOI. Images are cap- time [ms]
tured through a high resolution CCD camera, 0.5 ls shutter time,
12 bit, at different times from SOI. The optical axis of the CCD is ori- Fig. 3. Energizing solenoid current (top) and fuel injection rates (bottom) for the
ented either in a parallel or in an orthogonal way with respect to three considered injectors at Pinj = 10 MPa, Qinj = 20 mg/str.

0.016 0.016 0.024

0.016
0.008 0.008
y (m)

y (m)

y (m)

0.008
0 0
z=0 z=0 0 z=0
z = 30 mm z = 30 mm z = 30 mm
-0.008 -0.008 -0.008
-0.02 -0.01 0 0.01 0.02 -0.02 -0.01 0 0.01 0.02 -0.024 -0.016 -0.008 0 0.008 0.016 0.024
x (m) x (m) x (m)

Fig. 2. Holes distribution and spray footprint on a plane placed at 30 mm from the injector tip. Injector #1 (left), Injector #2 (centre) and Injector #3 (right).
46 M. Costa et al. / Advances in Engineering Software 49 (2012) 43–53

Continental one it is of about 170 ls. Furthermore, the closure time distribution, whose expected value is given by the following theo-
for Injector #3 is longer than for the others, namely it is of about retical diameter:
70 ls. The precise overlapping between signals relevant to Injec- !
tors #1 and #2 is indicative of an analogous behaviour of the mov- 2psf
ing equipment, while the Injector #3 has a slight larger inertia, Dth ¼ C d k ð1Þ
qg u2rel
hence greater opening and closing delays. This implies a different
promptness availability of the fuel with the same command
being sf the gasoline surface tension, qg the surrounding gas den-
signals.
sity, urel the relative velocity between the fuel and the gas, Cd a con-
An idea of the behaviour of Injectors #1 and #2 under double
stant of the order of the unity (indeed taken equal to the unity), and
strategies is given in Fig. 4. This reports the fuel injection rate sig-
the parameter k deriving from the hydrodynamic stability analysis
nal collected for a double-pulse strategy at the injection pressure of
and indicating the dimensionless wavelength of the more unstable
6 MPa for Injector #1, together with the timing of the solenoid
perturbation to the liquid–gas interface at the injector exit section.
driving current. Each pulse is equal to 0.9 ms in duration, hence
The variance of the distribution, r, is another parameter of the mod-
the gasoline injected mass is split in percentages equal to 50% plus
el to be properly tuned.
50% of the total amount. Stability and repetitiveness of the injec-
The definition of a probabilistic distribution of initial size of
tion events is studied by varying the value of the dwell time, dw,
droplets at the nozzle exit section corresponds to specify the occur-
from the minimum value up to 1.5 ms. The minimum value of this
rence probability for each particle diameter entry in the particle
variable, below which the opening of the second injection event
size distribution. The sum of all elements is used to normalize
interferes with the closing of the previous one, due to the elec-
the distribution. The number of particles per parcel is determined
tro-hydraulic inertia of the internal mobile equipment, is equal
by the particle probability distribution, the number of introduced
to 320 ls.
parcels per time step and the assigned mass flow rate. The number
of introduced parcels per time step is fixed a priori, and the injec-
2.2. Numerical model formulation and validation tion velocity is evaluated in such a way to fulfil the continuity
equation. The single jet cone angle is set as an input parameter,
In order to numerically simulate the tests made in the labora- according to the effected measurements.
tory, the spray is hypothesized to enter the top surface of a prop- In order to assess the numerical results dependency on the grid
erly dimensioned computational domain of cylindrical shape, cell size, preliminary tests are made. As shown in Fig. 5, the pene-
where the injector is supposed to be placed in central position. tration length in a certain test case, as averaged on the six jets issu-
According to the DDM, the spray is considered as a train of droplets ing from the Injector #1, is practically unchanged as computed
of given size, suffering various concurring effects as they travel over a grid made of 35,000 cells and over a grid made of 96,000
within the computational domain. The spray model is described cells. The former, therefore, is found being sufficient to perform
into detail in Refs. [8,26], as developed for Injector #1. Here it is the computations with a reasonable accuracy and low computa-
worth pointing out that break-up is simulated according to the tional effort. In the following, however, results relevant to a grid
sub-model of Huh and Gosman [27], whose constant C1 (regulating made of about 60,000 cells are presented, since, in this case, the
the break-up time) is properly adjusted in the tuning procedure. typical cell size is comparable with that relevant to the zone inter-
Initial size of droplets at the nozzle exit section, is considered as ested by the spray within the GDI engine 3D model described in
not constant, but variable according to a probabilistic log-normal paragraph 3.
The tuning of the Huh–Gosman model constant, C1, and of the
distribution variance, r, is here performed, for all the three injec-
30
tors, by means of an automatic procedure instead than through a
Fuel injection rate (mg/ms)

trial and error procedure, as was made in Refs. [8,26] for Injector
20 #1. A sketch of the tuning methodology, as developed within the
ModeFrontier software, is represented in Fig. 6. At each injection
pressure, the log-normal distribution of the initial droplet size at
10
the injector exit section is built (within a Microsoft Excel sheet)
starting from the value of r chosen in the DOE space and from
0 the expected value computed according to Eq. (1). The distribu-
0.4 ms
tion profile is transferred to the FireTM spray model, that also
-10
receives the value of C1 from the DOE space. The model realizes
the spray computation in the interval of time needed to inject a
25

20
Solenoid current (A)

0.1
0.9 ms
15 0.9 ms
Penetration length (m)

0.08
10 0.32 ms
0.06
5
0.04
0
Grid A - 19440 number cells

-5 0.02 Grid B - 35000 number cells

Grid C - 96000 number cells
-0.5 0 0.5 1 1.5 2 2.5 3 3.5
0
Time (ms)
0 0.0004 0.0008 0.0012 0.0016 0.002
Fig. 4. Fuel injection rate for a double injection strategy at the minimum dw (top) Time (s)
with the corresponding exciting solenoid currents (bottom) for Injector #1.
Pinj = 6 MPa. Fig. 5. Results dependency on the grid size.
 M. Costa et al. / Advances in Engineering Software 49 (2012) 43–53 47

Fig. 6. Sketch of the tuning procedure for the numerical spray sub-model constants.

given mass of gasoline (according the experimental measure- 16

inj 1 - 6 holes
ments), and furnishes, as an output, the penetration length of
inj 2 - 7 holes
the jets compounding the spray. The error between the numeri- inj 3 - 6 holes
cally computed penetration length, as averaged over the six (or 12
seven) jets, and the experimentally measured one is minimized
C1

by the Simplex algorithm. The objective function, therefore, is

defined as: 8
X
n
Objjr;C 1 ¼ ½lex ðt i Þ  lnum ðti Þ2 ð2Þ
i¼1
4
where n represents the number of discrete instants of time in 0.8

which the injection interval of time is subdivided, ti is the ith in- inj 1 - 6 holes
0.7 inj 2 - 7 holes
stant of time, and lex(ti) and lnum(ti) the values, respectively, of the
inj 3 - 6 holes
experimentally measured and the numerically computed penetra-
0.6
tion length at ti. The experimentally measured penetration length,
σ

indeed, is evaluated by means of a smoothing spline passing

0.5
through the actual measurements points in the time-length
plane.
0.4
The results of the automatic model tuning are reported in Fig. 7.
The top of the figure represents the value of the C1 constant mini- 0.3
mizing the error between numerical and experimental data, as a
4 6 8 10 12 14 16 18 20 22
function of injection pressure for the three injectors. The bottom Injection pressure (MPa)
of the figure reports the value of the variance of the corresponding
distribution of initial size of droplets. A similitude of behaviour be- Fig. 7. Results of the tuning procedure of the spray sub-model constants.
tween Injector #1 and #2 is evident, since the values of C1 are
found comparable. The values relevant to Injector #3 are slightly
higher, probably due to the differences in the injector geometrical
characteristics, hence in the way internal perturbations affect the relative velocity between the liquid and the air, that moves
issuing flow. A slight increase of the variance with injection pres- towards lower values as injection pressure is raised. The distribu-
sure may be assumed. A similar conjecture may apply to the trend tion variance may be maintained almost constant or slightly
of the Huh–Gosman constant, which slightly increases to account increasing with injection pressure, as confirmed by the here pre-
for the greater injection velocity. sented optimization procedure.
The low dispersion of the optimal values of the model con- Fig. 8, as an example, reports the distributions used for the four
stants confirms the good prediction capability of the model. considered injection pressures for Injector #1, as resulting from the
The combined use of the Huh–Gosman break-up model and of constants tuning automatic procedure. Analogous shapes are rele-
a properly defined log-normal distribution for the initial size of vant to the other two injectors. All the distributions are cut at the
droplets, that is an innovative idea proposed by authors, allows value corresponding to the diameter of the nozzle holes and exhi-
overwhelming the problems generally encountered in the simu- bit, according to physical considerations, expected values decreas-
lation of GDI sprays from new generation injectors. The actual ing with injection pressure.
physics of the dependence of the droplet break-up process upon The prediction capability of the model is demonstrated in Fig. 9,
injection pressure is well reproduced by the proposed approach, where a comparison between the computed and the measured
since the expected value of the initial distribution of the diame- penetration lengths relevant to the three injectors are reported.
ter of droplets, at the nozzle exit section, is a theoretical diame- The agreement is satisfactory under all the considered injection
ter linked to the injection pressure through the value of the pressures.
48 M. Costa et al. / Advances in Engineering Software 49 (2012) 43–53

0.8 that makes for the fuel spray to be injected directly in proximity
of the ignition location. A 3D numerical model of the thermo-flui-
pinj = 6 MPa - Dth = 4.99e-5 m - σ = 0.56
pinj = 10 MPa - D th = 2.54e-5 m - σ = 0.55
dynamic processes occurring within the cylinder and intake and
pinj = 15 MPa - D th = 1.67e-5 m - σ = 0.58 exhaust ducts is developed by authors within the AVL FireTM envi-
0.6 pinj = 20 MPa - D th = 1.27e-5 m - σ = 0.6 ronment [26]. Gasoline injection is simulated according to the pre-
viously discussed spray sub-model, under both single and double
strategies, as issuing form Injector #1.
Probability density

The discretisation of the computational domain is made

through the same pre-processing software included in the FireTM
0.4
Graphical User Interface (GUI), namely the Fame Engine Plus
(FEP) module [23,26]. This allows to roughly control the grid cell
size by locally thickening nodes where particular geometric con-
formations of the outer surfaces are present, or where intense gra-
0.2
dients of the thermo-fluidynamic variables are expected.
The considered engine is at a design stage, hence no test bench
data are available that may serve to the multidimensional model-
ling validation. Nevertheless, the lack of experimental data is ful-
0 filled by exploiting results of a 1D model of the entire propulsion
0E+000 4E-005 8E-005 1E-004 2E-004 2E-004
system, developed at the University of Naples ‘‘Federico II’’ – DIME
Diameter (m)
[29], and validated with reference to engines of similar size and
specific power. The 1D model uses information derived by a preli-
Fig. 8. Log-normal distribution of the initial size of droplets for Injector #1 as a minary experimental evaluation of the considered engine intake
function of injection pressure. and exhaust valves discharge coefficients. Inlet total pressure and
temperature and outlet static pressure, as a function of time, de-
rived from the 1D analysis, are set on the 3D domain boundaries.
3. The performance optimization of a GDI engine Inlet and outlet sections coincide with the entrance of the intake
ducts and the exit of the exhaust ducts, respectively.
The optimization of the energy conversion process of a GDI The turbulent in-cylinder flow in the gaseous phase within the
engine is here pursued through the assessment of a proper optimi- 3D engine model is described according to the k–f–f model [23],
zation problem that allows the choice of the most proper injection combustion according to the extended coherent flame model
strategy and spark timing within the working cycle. A single cylin- (ECFM) model, using standard values of the entering constants
der engine, four-valve, four-stroke, 638 cc displacement, suitable [30]. Finally, it is to be pointed out that in the application of the
for motorbike applications is analysed. The mixture formation pro- spray model described in paragraph 2 to the GDI engine, spray-
cess is assumed to occur under the spray guided mode [28], hence wall interaction is also taken into account. The impingement of
with a reduced spacing between the injector and the spark plug, the spray on the engine walls gives rise to the deposition of a liquid
Penetration length (m)

0.12 0.12 0.12 0.12

0.08 0.08 0.08 0.08

0.04 0.04 0.04 0.04

pinj = 6 MPa pinj = 10 MPa pinj = 15 MPa pinj = 20 MPa
0 0 0 0
0 1.5 3 0 1.5 3 0 1.5 3 0 1.5 3
Time (ms) Time (ms) Time (ms) Time (ms)
Penetration length (m)

0.12 0.12 0.12 0.12

0.08 0.08 0.08 0.08

0.04 0.04 0.04 0.04

pinj = 6 MPa pinj = 10 MPa pinj = 15 MPa pinj = 20 MPa
0 0 0 0
0 1.5 3 0 1.5 3 0 1.5 3 0 1.5 3
Time (ms) Time (ms) Time (ms) Time (ms)
Penetration length (m)

0.12 0.12 0.12 0.12

0.08 0.08 0.08 0.08

0.04 0.04 0.04 0.04

pinj = 6 MPa p inj = 10 MPa pinj = 15 MPa pinj = 20 MPa
0 0 0 0
0 1.5 3 0 1.5 3 0 1.5 3 0 1.5 3
Time (ms) Time (ms) Time (ms) Time (ms)

Fig. 9. Numerical (continuous line) and experimental (dashed line with dots) penetration lengths for Injector #1 (top), #2 (centre), #3 (bottom).
 M. Costa et al. / Advances in Engineering Software 49 (2012) 43–53 49

film on the walls and to a secondary atomization of the impacting split in two parts. The choice of the range of variation of the sam-
droplets. The process is strongly affected by the conditions of pres- ples, as well as the step between successive samples, are a subjec-
sure, velocity and temperature of the surrounding gas and by the tive matter, strongly affecting the efficiency and speed of the
value of the wall temperature. These aspects are treated within optimization procedure. Here physical considerations are made,
the ‘‘wallfilm’’ module of the FireTM code that involves the Kuhnke as avoiding injection in the valves overlap period, or considering
model [23]. the existence of a maximum brake torque (MBT) value correspond-
The entire four stroke engine cycle is simulated with the in-cyl- ing to a given SI, that helps in limiting the interval of variation of
inder initialization also made according to the 1D results. A valida- this last quantity, as described in the following.
tion of the entire four strokes motored cycle is here not reported Fig. 11, as an example, shows the flow-chart of the optimization
for the sake of brevity. A check about the repeatability of the problem in the case of split injection.
achieved results is made by simulating more than one working cy- The design variables are indicated as SOI1, Dwell and SI. The
cle. It is found that it is convenient to start the computation at the first two allows defining the injection law, starting from a mea-
crank angle of exhaust valves opening (EVO). Fig. 10 schematically sured mass flow rate profile, as the one shown in Fig. 4. The injec-
represents the working cycle of the considered engine. The EVO is tion law is transferred to the FireTM model, together with the value
at the crank angle of 130°, in an angular reference frame with ori- chosen for SI in the DOE space. The model computes the intake,
gin at the piston top dead centre (TDC) and a clockwise direction compression and expansion phases of the engine cycle and trans-
for increasing crank angles. The gases exhaust phase ends at 390° fers the values of the in-cylinder mean pressure to a routine for
(exhaust valves closing, EVC). The intake stroke starts at 330° (in- the evaluation of the IMEP (objective function):
take valves opening, IVO) and ends at 608° (intake valves closing, Z EVO
IVC). The valves overlap lasts 60° around TDC. The closed valves 1
ObjjSOI1 ;Dwell;SI ¼ IMEPclosedv alv es ¼ pcyl dV; ð3Þ
period, starting at IVC and ending at EVO, is indicated in the figure Vd IVC
with a thicker line. During the closed valve period, there is the
occurrence of the spark ignition (some degrees before the top dead with Vd as the cylinder displacement, pcyl as the mean in-cylinder
centre (BTDC)) and the consequent combustion process (around pressure and V as the instantaneous cylinder volume. Maximizing
720°). The gasoline injection is represented in Fig. 10 as a single the engine IMEP corresponds to maximize the useful work, hence
pulse during intake, in the interval of crank angles ranging be- to reduce the gasoline consumption (if the injected mass is main-
tween 475° and 537°. tained constant).
The considered engine performance optimization is realized
within a single objective optimization problem, where the devel- 3.1. Discussion of the results
oped 3D Fire model is automatically run by the ModeFrontier soft-
ware. Two successive analyses are effected, one for single injection, A moderate-load, moderate-speed condition, characterized by
one for injection split at the 50 + 50%. In both the situations the an air-to-fuel ratio equal to 17, is discussed with the aim of high-
Simplex algorithm is used to search for the inputs maximizing lighting how the mixture formation process can be optimized to
the indicated mean effective pressure (IMEP) in the closed valve assure a lean combustion the most complete and stable. Speed is
period. The underlying design variables of the optimization prob- assumed equal to 5000 rpm.
lem are identified in the time of spark ignition (SI) and in the start The case where injection is realized in a single event is first ana-
of the single injection event. More into detail, the variable is just lyzed, as characterized by an injection pressure equal to 6 MPa,
the hereafter called SOI (start of injection) in the case of single hence by a duration, at the considered engine speed, of about
injection, or the start of the first injection event, SOI1, and the dwell 60°. The in-cylinder pressure is preliminary evaluated within a
time between two successive pulses, Dwell, in the case injection is parametric analysis where the start of injection (SOI) is fixed at
450° and the time of spark ignition (SI) is varied between 670°
and 710°, step 5°. The value of SOI is assumed on the ground of
the physical consideration that the injection has to fully exploit
the motion of the air entering the cylinder, hence by accounting
for the fact that the maximum intake valve lift occurs at 470°.
The indicated mean pressure, relevant to the closed valve period
and normalized with respect to its maximum value, is reported
in Fig. 12 as a function of SI. The maximum value of the curve,
hence the MBT is attained in correspondence of SI at 680°, namely
40° BTDC. This situation is hereafter considered as a reference case
to be used as a term of comparison for the following numerical re-
sults (starting point). The individuation of the crank angle of MBT
allows a more proper definition of the interval of variation of the
samples of the SI variable.
Results of the optimization analysis relevant to the single injec-
tion case are reported in Fig. 13, where the IMEP of each computed
cycle in the closed valve period, made dimensionless with respect
to the value relevant to the starting point cycle (IMEPref), is repre-
sented in a bubble plot as a function of the input variables SOI–SI
plane. The starting and the ending point of the optimization pro-
cess are indicated, as well as the sequence of iterations performed.
It is evident that the couple of values of SOI and SI maximizing the
engine performance is SOI placed at 475° and SI at 680°. The gain in
the IMEP with respect to the starting point is of about the 5.3%. The
dramatic effect of the change of SOI on the in-cylinder pressure at
Fig. 10. The GDI engine working cycle. the considered moderate-load case is visualized in Fig. 14, where
50 M. Costa et al. / Advances in Engineering Software 49 (2012) 43–53

Fig. 11. Flow chart of the optimization problem in the case mixture formation is realized through split injection.

0.95
(IMEP/IMEP max )closed valves

0.9

0.85

0.8

0.75
670 680 690 700 710
SI (°)

Fig. 12. Normalized mean indicated pressure in the closed valve period for single
injection as a function of SI.

five different in-cylinder pressure cycles are plotted, chosen be-

tween those computed during the optimization procedure. All of
them have the SI at the same crank angle, 680°. It is evident that
injection has to be not so much delayed, and has to really exploit Fig. 13. Bubble plot of the optimization procedure in the single injection case in the
the motion of the air entering the cylinder, before the valve lift de- SOI–SI plane.
creases too much.
Effects of splitting the injection in two successive events
characterized by a same injected mass is discussed by varying a plane passing through the maximum value of this variable, at
SI and the starts of both the first and the second injection event. four different crank angles. The sequence on the top refers to the
Fig. 15 represents the optimization problem results, with the optimal single injection case, the sequence on the bottom to the
optimal values being SOI1 at 450° and a Dwell equal to 80°. The optimal double injection case. The spark position corresponds to
optimal spark advance remains at 680°, although the relevant the centre of the burnt zone visible in the images placed more on
graph is here not reported. Splitting injection in two events al- the left. The double injection event, with the second pulse entirely
lows an increase of the work greater than the 8.6% with respect realized during the compression stroke, gives rise to a faster prop-
to the case assumed as starting point, with injection realized in agation of the flame front. This is to be considered as the major
one shot. consequence of the more effective charge stratification around
The comparison between the in-cylinder pressure cycles found the spark plug achieved by splitting injection, as also highlighted
as optimal in the single injection case and in the double injection by Li et al. [31].
case is made in Fig. 16. As already said, the total injected mass is The optimization procedure, as conceived, also allows deter-
the same, namely 20 mg/str. The increase in the cycle area is well mining the effects of the injection strategy on the main pollutants
visible. emitted at the engine exhaust. As an example, Fig. 18 represents,
The explanation of the better performance relevant to the dou- for each of the solution computed in the case injection is realized
ble injection case can be drawn by looking at Fig. 17, where the in one shot, the value of the unburned equivalence ratio on the
burnt fuel mass fraction is represented, in the engine cylinder, on abscissa and the related value of the nitrogen oxide (NO) mass
 M. Costa et al. / Advances in Engineering Software 49 (2012) 43–53 51

3.00E+006 3.00E+006

SOI = 445°
SOI = 475°
SOI = 485°
SOI = 495°
SOI = 535°
2.00E+006 2.00E+006
Pressure (Pa)

Pressure (Pa)
1.00E+006 1.00E+006

optimal single SOI=475°

optimal double SOI 1=450° dwell=80°

0.00E+000 0.00E+000
640 680 720 760 800 640 680 720 760 800
Crank angle (°) Crank angle (°)

Fig. 14. Pressure cycles relevant to five different values of SOI. Fig. 16. Optimal pressure cycles relevant to single injection and double injection.

fraction, at EVO. The unburned equivalence ratio gives an idea of 4. Conclusions

the unburned hydrocarbons (HC) emitted at the exhaust, whereas
the NO amount is computed through the classic model of Zel’do- Two fully automatic procedures are developed, suitable to be
vich [23]. The bubble plot represents also the optimized variable used in the phase of design of innovative internal combustion
IMEP/ IMEPref. The starting and the optimal points are drawn in engines.
grey and black, respectively. The optimal synchronization of the The first one is finalised to the validation of a 3D GDI spray
single injection, beside increasing the useful work, also allows a model. In particular, it is applied to three different high pressure
slight reduction of both the considered pollutants. Fig. 19, finally, multi-hole injectors, preliminary experimentally characterized at
represents the effects of a double injection on the emitted HC both the mass flow rate test bench and in an optically accessible
and NO. The optimal single injection and optimal double injection vessel. The automatic tuning of the entering constants serves to
cases are drawn in grey and black, respectively. Double injection prove the predictive capability of a model proposed by authors.
tends to reduce the amount of unburned hydrocarbons. When cor- The second automatic procedure consists in a CFD optimization
rectly synchronized, it substantially reduces these pollutants, with of the mixture formation process in a GDI engine, finalised to the
the penalty of a slight increase in the produced NO (consequent the minimization of the fuel consumption.
increase in the in-cylinder temperature). Several points character- Achievement of optimal charge conditions at all the engine
ized by lower values of the NO mass fraction, but also lower IMEP, loads and speeds in modern gasoline GDI SI engines is undoubtedly
are visible. a challenging task, especially if the so-called mixed mode boosting

(a) 1.1
(b) 1.1

1.08 1.08
(IMEP / IMEPref)closed valves

(IMEP / IMEPref)closed valves

1.06 1.06

1.04 1.04

1.02 1.02

1 1

440 445 450 455 460 465 20 40 60 80 100 120

SOI (°) Dwell (°)

Fig. 15. Optimal synchronization of two injection events w.r.t. SOI (a) and dwell time (b).
52 M. Costa et al. / Advances in Engineering Software 49 (2012) 43–53

Fig. 17. Burnt fuel mass fraction on a plane passing through its maximum value in the optimal single injection case (top) and double injection case (bottom) at four crank
angles.

Fig. 18. Bubble plot of the optimization procedure in the single injection case in the Fig. 19. Bubble plot of the optimization procedure in the double injection case in
unburned equivalence ratio – NO mass fraction plane. the unburned equivalence ratio – NO mass fraction plane.

is to be realized, with homogeneous rich mixtures at the higher

loads, and stratified lean mixtures at the lower ones. This is the in the numerical study of mixture formation and combustion con-
reason why fully automatic procedures to be used in the phase sequent a double injection. The start of the first pulse and the dwell
of definition of the engine governing parameters are strongly time between two successive pulses, both characterized by the
demanded. same gasoline mass, are searched, together with the best time of
The considered engine is equipped with a six-hole new genera- SI. The optimal double injection is shown to reduce the fuel con-
tion high pressure injector. A properly developed 3D model of the sumption with respect to the case injection is realized in one shot,
in-cylinder processes is coupled with a multi-objective optimiza- thus confirming the possibility to resort to split injections to im-
tion tool to optimize the engine performance under a moderate- prove the quality of the charge stratification under lean operation.
speed, moderate-load lean mixture condition. Both single and dou- Effects of the injection strategy on the main pollutants are also
ble injection events are taken under examination. Optimal choice discussed.
of both the start of the single injection strategy and the time of The assessed procedures, that may have a broad field of applica-
spark advance is realized by means of the Simplex algorithm to tion, since easily expandable to port fuel injection (PFI) or com-
maximize the engine power output. The same criterion is adopted pression ignition engines, are valuable tools in the phase of
 M. Costa et al. / Advances in Engineering Software 49 (2012) 43–53 53

engine design to reduce development cost and time-to-market of [14] Sirignano WA. Fluid dynamics and transport of droplets and sprays. Cambridge
University Press; 1999.
new technologies.
[15] Fritsching U. Spray simulation. Cambridge University Press; 2004.
[16] Loth E. Numerical approaches for motion of dispersed particles, droplets and
References bubbles. Progr Energy Combust Sci 2000;26:161–223.
[17] Dukowicz JK. A particle-fluid numerical model for liquid sprays. J Comput Phys
[1] Carling RW. Predictive simulation of combustion engine performance in an 1980;35:229–53.
evolving fuel environment. Sandia National Laboratories; 2010. [18] Gosman AD, Ioannides E. Aspects of computer simulation of liquid-fueled
[2] Thévenin D, Janiga G, editors. Optimization and computational fluid combustors. J Energy 1983;7:482–90.
dynamics. Berlin Heidelberg: Springer-Verlag; 2008. [19] O’Rourke PJ. Modeling of drop interaction in thick sprays and a comparison
[3] Wickman DD, Senecal PK, Reitz RD. Diesel engine combustion chamber with experiments. IMechE – Stratified Charge Automotive Engines Conference;
geometry optimization using genetic algorithms and multi-dimensional spray 1980.
and combustion modeling. SAE Paper 2001–01-0547; 2001. [20] Dukowicz JK. Quasi-steady droplet phase change in the presence of
[4] De Risi A, Donateo T, Laforgia D. Optimization of the combustion chambre of convection. Los Alamos, Report LA-7997-MS; 1979.
direct injection diesel engines. SAE Paper 2003–01-1064; 2003. [21] Naber JD, Reitz RD. Modeling Engine Spray/Wall Impingement. SAE Paper
[5] Dempsey AB, Reitz RD. Computational optimization of reactivity controlled 880107; 1988.
compression ignition in a heavy-duty engine with ultra low compression ratio. [22] Liu AB, Reitz RD. Modeling the Effects of Drop Drag and Breakup on Fuel
SAE Paper 2011–24-0015; 2011. Sprays. SAE 930072; 1993.
[6] Shrivastava R, Hessel R, Reitz RD. CFD optimization of DI diesel engine [23] AVL Fire v2008 Users Guide – ICE Physics & Chemistry.
performance and emissions using variable intake valve actuation with boost [24] Bosch W. The fuel rate indicator: a new measuring instrument for display of
pressure, EGR and multiple injections. SAE Paper 2002–01-0959; 2002. the characteristics of individual injection. SAE Paper 6607496; 1966.
[7] Antoniou A, Lu W. Practical optimization algorithms and engineering [25] Wallace I. Injection rate gauge: pass off information and user
applications. Springer; 2007. instructions. Graz: Fuel & Engine Management Systems; 2002.
[8] Sorge U, Costa M, Allocca L. Optimization of the mixture formation process by [26] Costa M, Sorge U, Allocca L. Numerical study of the mixture formation process
split injection in a gasoline direct injection engine for two-wheel applications. in a four strokes GDI engine for two-wheels applications. Simul Modell Pract
In: Topping BHV, Adam JM, Pallarés FJ, Bru R, Romero ML, editors. Proceedings Theory J 2011;19(4):1212–26.
of the seventh international conference on engineering computational [27] Huh KY, Gosman AD. A phenomenological model of diesel spray atomisation.
technology, Stirlingshire, UK: Civil-Comp Press; Paper 53, 2010. doi:10.4203/ International conference on multiphase flows 1991. Tsukuba, Japan; 1991.
ccp.94.53, ISBN: 978-1-905088-41-6. [28] Stan C. Direct injection systems for spark-ignition and compression-ignition
[9] Çelik MB, Özdalyan B. Gasoline direct injection. In: Siano D, editor. Fuel engines. SAE Publication; 2000.
Injection, Sciyo; 2010. <http://www.intechopen.com>. [29] Bozza F, Torella E. The employment of a 1D simulation model for the A/F ratio
[10] Alkidas AC. Combustion advancements in gasoline engines. Energy Convers control in a VVT Engine. SAE Trans J Eng 2004;3.
Manage 2007;48:2751–61. [30] Colin O, Benkenida A, Angelberger C. 3D modeling of mixing, ignition and
[11] Patent N. 7415348 B1 – multiple injection blend for direct injected engines combustion phenomena in highly stratified gasoline engines, Oil & Gas Science
assignee. Detroit, MI (US): GM Global Technology Operation, Inc.; 2008. and Technology – Rev. IFP Energies Nouv 2003;58(1):47–62.
[12] Kuwahara K, Ueda K, Ando H. Mixing Control Strategy for Engine Performance [31] Li T, Nishida K, Zhang Y, Hiroyasu H. Effect of split injection on stratified charge
Improvement in a Gasoline Direct Injection Engine. SAE Paper 980158; 1998. formation of direct injection spark ignition engines. Int J Eng Res
[13] Arcoumanis C, Kamimoto T, editors. Flow and combustion in reciprocating 2007;8:205–18.
engines. Berlin, Heidelberg: Springer-Verlag; 2008.