Downloaded from SAE International by Purdue University, Tuesday, August 21, 2018

SAE TECHNICAL
PAPER SERIES 2007-01-0148

Development of an Ignition Model for

S.I. Engines Simulation
S. Falfari and G. M. Bianchi
DIEM – University of Bologna

Reprinted From: Multi-Dimensional Engine Modeling, 2007

(SP-2125)

2007 World Congress

Detroit, Michigan
April 16-19, 2007

400 Commonwealth Drive, Warrendale, PA 15096-0001 U.S.A. Tel: (724) 776-4841 Fax: (724) 776-0790 Web: www.sae.org
 Downloaded from SAE International by Purdue University, Tuesday, August 21, 2018

By mandate of the Engineering Meetings Board, this paper has been approved for SAE publication upon
completion of a peer review process by a minimum of three (3) industry experts under the supervision of
the session organizer.

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or
transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise,
without the prior written permission of SAE.

For permission and licensing requests contact:

SAE Permissions
400 Commonwealth Drive
Warrendale, PA 15096-0001-USA
Email: [email protected]
Fax: 724-776-3036
Tel: 724-772-4028

For multiple print copies contact:

SAE Customer Service

Tel: 877-606-7323 (inside USA and Canada)
Tel: 724-776-4970 (outside USA)
Fax: 724-776-0790
Email: [email protected]

ISSN 0148-7191
Copyright © 2007 SAE International
Positions and opinions advanced in this paper are those of the author(s) and not necessarily those of SAE.
The author is solely responsible for the content of the paper. A process is available by which discussions
will be printed with the paper if it is published in SAE Transactions.

Persons wishing to submit papers to be considered for presentation or publication by SAE should send the
manuscript or a 300 word abstract of a proposed manuscript to: Secretary, Engineering Meetings Board, SAE.

Printed in USA
 Downloaded from SAE International by Purdue University, Tuesday, August 21, 2018

2007-01-0148

Development of an Ignition Model for S.I. Engines Simulation

S. Falfari and G. M. Bianchi
DIEM – University of Bologna

Copyright © 2007 SAE International

ABSTRACT In the brief review of the models we did not mention all
the most simplest models which prescribe that a given
An ignition model based on Lagrangian approach was mass of fuel burns by a certain time since no physics
set-up. A lump model for the electrical circuit of the is there involved. Focusing attention on the Eulerian
spark plug is used to compute breakdown and glow 1D-modelling, many efforts were spent in order to
energy. At the end of shock wave and very first plasma develop accurate modelling starting from the
expansion, a spherical kernel is deposited inside the considerations arising from the experimental
gas flow at spark plug location. A simple model allows approach. Herweg and Maly [1] developed a detailed,
one to compute initial flame kernel radius and time dependent one-dimensional model for simulating
temperature based on physical mixture properties and flame kernel formation and validated it versus
spark plug characteristics. The sphere surface of the experiments. Boudier et al. [2] proposed the Laminar
kernel is discretized by triangular elements which Ignition (LI) model coupling ignition with CFM
move radially according to a lagrangian approach. combustion model: they mainly applied the model to
Expansion velocity is computed accounting for both cases with low turbulence levels and without
heat conduction effect at the highest temperatures and convection. Duclos et al. [3] derived from the LI model
thermodynamic energy balance at relatively lower a new ignition model, called Small Kernel Ignition
temperatures. Turbulence effects and thermodynamic (SKI), for dealing with convection. They divided the
properties of the air-fuel mixture are accounted for. ignition modelling in two main phases: (i) in the first
Restrikes are possible depending on gas flow velocity phase, computed outside the main computational
and mixture quality at spark location. CFD solver and code, a laminar kernel of 1-2mm is evaluated. In
1D/lagrangian ignition model are closely coupled at particular they calculate “the birth of laminar flame as a
each time step. The model proves to strongly reduce function of the local thermodynamic quantities in the
the grid sensitivity. The physical validation was carried area of the spark”; (ii) in the second phase there is a
out by reproducing the experimental tests by Herweg complete coupling with the combustion model This
and Maly [1]. Comparisons showed a good agreement model “handles a flame deposit, the turbulent stretch,
between experiments and numerical results. the mean curvature and thermal expansion stretch and
a flame holding effect at the spark electrodes”.
INTRODUCTION Andreassi et al. [4] used the model of Herweg and
Maly for quantifying the energy transferred from the
Modelling spark ignition and flame kernel growth plays spark plug to the mixture. Willems et al. [5] developed
a key role in the study of the whole combustion a one-dimensional model to simulate the growth of the
process of spark ignition engines because it influences plasma kernel and the initial flame. Bianchi et al. [6]
the further flame development. Several experimental proposed a model which allowed a lump model of
tests in combustion chambers or in engines have been spark plug electrical characteristics as well as a sub-
carried out in order to find correlations between the model for kernel expansion. The model was
main thermo-physics (i.e. spark current, breakdown successfully tested in conjunction with the ECFM
energy, air flow convection, mixture characteristics) on combustion model by reproducing the Herweg and
plasma and flame kernel characteristics. All these Maly test case where the effects of turbulence, mixture
experimental evidences have been translated into quality and air flow convection were considered. The
physical ignition models for CFD computations. The poor results achieved in testing the grid sensitivity of
flame kernel modelling is quite complex because the model dropped off the applicability to real engine
involves the simultaneous treatment of physics of cases, where cell size is expected to vary from mesh
spark discharge and plasma formation, chemical to mesh.
kinetics, flow and turbulence interactions. Many of
these processes occur at characteristic length scales 1D-Eulerian models likely provide grid-dependent
which are much smaller than usual cell size adopted in results. For this reason, the Lagrangian methods offer
CFD calculations. Finally, one of the most critical issue the potential to reduce the grid sensitivity since it
is the coupling between the ignition model and the decouples the kernel modelling from gas flow
main CFD solver as well as the switching criteria to the modelling allowing one to deal with kernel dynamics at
main combustion model. its own length scales. Reviewing the lagrangian
 Downloaded from SAE International by Purdue University, Tuesday, August 21, 2018

ignition models we could mention those by Fan et al. § 6 ·

[7] and Duclos et al. [8]. In the Discrete Particle ¨ w ¸
w6 wu~i 6 w ¨ §¨ P t P· U ¸
Ignition Model (DPIK) developed by Fan et al. [7] the ¸  P1  P2  P3 6  D (1)
 ¸
wt wx i wx i ¨ ¨ Sc  Sc ¸ wx
flame kernel position is marked by particles which ¨¨ © t ¹ i ¸¸
represent a sole kernel. Duclos et al. [8] developed a © ¹
numerical model called AKTIM for describing the flame
kernel expansion. The model includes a detailed Further details, i.e., expression of production and
description of the electrical inductive system of the destruction terms, are given in [6, 10]. A conditional
spark plug in order to evaluate the voltage and the averaging is used to compute separately burnt and
current among electrodes. The ignition kernels were fresh gas enthalpy. The concentrations of burnt gases
modelled by a set of particles uniformly placed along is computed at burnt gas temperature according to the
the spark path, which was elongated in time by the equilibrium mechanism proposed by Meintjes and
mean flow field. The flame kernels can growth Morgan [11].
according to mixture thermo-physics and turbulence
properties and can be convected by mean flows. An LAGRANGIAN IGNITION MODEL
energy balance of each kernel accounts for the energy
received from the electrical circuit and the energy lost In the present paper the previous ignition model was
to electrodes. In case of strong convection multiple modified by means a 1D-Lagrangian description of the
breakdowns could be handled. The main disadvantage flame kernel growth. As done in the 1D- Eulerian
of this model is the need of detailed information of version [6], the ignition model does not simulate the
spark plug electrical circuit. plasma formation while a 1D model is used to compute
flame kernel expansion before switching to main
In this paper the authors present a revised lagrangian combustion model. A lagrangian tracking of the flame
version of the model presented in [6] in order to surface is carried out.
achieve less grid-dependent numerical results. The
ignition modelling framework is successfully tested by The model is divided into two main sub-models: the
reproducing the Herweg and Maly test cases after a electrical sub-model and the flame kernel expansion
preliminary numerical assessment. The basic test case sub-model.
by Herweg and Maly [1] is used to assess the
capability of the modelling framework in reproducing
1. ELECTRICAL CIRCUIT SUB-MODEL
the physical effects of the major parameters on the
flame kernel development. It must be pointed out that
The aim of the electrical sub-model is to provide the
the focus of the paper is to accomplish to the task to
initial conditions for the flame kernel expansion model.
simulate ignition into a three-dimensional CFD solver.
As done in the previous work [6], it was chosen to not
Several issues arise in such a coupling between a
simulate in detail the phase of the electric arc and the
thermodynamic model and the CFD solver. The
plasma formation because: (i) their duration is
ignition model framework here proposed would like to
negligible (order of nanoseconds) if compared to
address to those issues without any empirical tuning of
plasma/kernel expansion and development (order of
constant, at this stage. Therefore, it is not expected to
microseconds); (ii) a detailed model of the plasma
provide necessary better agreement with experiments
would require the simulation of shock wave and highly
with respect to the ignition models above referenced
unsteady formation and destruction of ionised
and not used in CFD simulations.
chemical species, as done by [12]; (iii) based on
experience on electromagnetic circuit of common rail
CFD CODE DESCRIPTION injection system carried out by present authors [13,
14], a detailed model is potentially more accurate but
The CFD code used for these simulations is a requires a detailed knowledge of all characteristics of
version of KIVA3 modified at University of Bologna: electrical circuit. This is not always possible because
it is used for both SI and Diesel engine combustion many information on the spark plug are not released
simulation. A Quasi Second Order Upwind Scheme by manufacturers. Thus, the electrical model retains
(QSOU) is chosen for the derivative discretization. the basic equations of the AKTIM model [8] and
All the simulations have been carried out with a simplifications have been introduced.
standard k-H turbulence model and turbulent law of
the wall. The code runs an improved rezoning The electrical model computes the breakdown energy
algorithm, in order to accomplish with complex and the electrical power released to gas flow during
geometry of combustion chambers. In order to the glow phase. The breakdown energy Ebd available
improve wall heat predictions the model by Han and at electrodes during the breakdown phase operation is
Reitz [9] is adopted. Premixed combustion is given by an empirical relation [8]:
modelled by using the ECFM combustion model as
proposed by Duclos et al. [3]. In particular, the Vbd2
Ebd (2)
transport equation of the mean flame surface density Cbd2 ˜ d gap
6 is in the form presented by Colin et al. [10]:
where the breakdown voltage Vbd in [kV] can be
evaluated according to [15]:
 Downloaded from SAE International by Purdue University, Tuesday, August 21, 2018

p p conservation the kernel mean expansion velocity is

Vbd 4.3  136  324 d gap (3) evaluated. The mass conservation equation reads:
Tunb Tunb

dm k
where p is the gas pressure in bar, Tunb is the fresh U unb ˜ s lam , k ˜ ( Ak ˜ ; ) (6)
gas temperature in K and dgap is the inter-electrodes dt
distance in mm. The value of the constant Cbd was not
reported by Duclos et al. [8]: the authors of the present where Uunb is the fresh gas density, Ak is the flame
paper adopted a value of 245 kV/(J1/2mm1/2). The kernel surface assumed to be unaffected by
breakdown energy is released to the main CFD solver turbulence, ; is the wrinkling coefficient of the
in the cells where the spark plug is located. flame surface introduced here for accounting for
turbulence influence (i.e., ; t 1). slam,k is the
Similarly, based on the spark plug electrical laminar flame speed which is evaluated according
characteristics, the electrical power released during to Metghalchi and Keck [19] based on local fresh
the glow phase is released to gas flow during the mixture conditions in the spark plug location at the
ignition time. moment of ignition. Differentiation of Eq. (6) gives:

2. FLAME KERNEL EXPANSION SUB-MODEL dV k U unb V dU k

s lam , k ˜ ( Ak ˜ ;)  k (7)
dt Uk U k dt
(A) INITIAL CONDITIONS

At the end of plasma shock wave dynamics, the model and applying the general gas law, one obtains the
starts to simulate the kernel expansion. Thus, variation of the mean flame kernel radius in time:

1. A spherical flame kernel of initial radius ri and drk U unb ªV § 1 dTk 1 dp ·º

s lam , k ˜ ;  « k ¨ ¸
¨ T dt  p dt ¸» (8)
initial temperature Ti is deposited into the gas flow dt Uk «¬ Ak © k ¹»¼
in the cell where spark plug lies.

2. The model by Song and Sunwoo [16] provides the In order to solve Eq. (8), the kernel temperature Tk
initial kernel temperature Ti and radius ri by means and pressure must be evaluated. The pressure is
of equations (4) and (5) according to Kravchik’s assumed to be uniform in burnt and fresh gases. In
model [17]: order to evaluate the flame kernel temperature one
has to consider two phases: in the first one
(PHASE 1) the flame kernel expansion at high
ª 1 § Tb · º temperature is dominated by heat conduction from
Ti « ¨¨  1¸¸  1»Tunb (4)
«¬ k © Tunb ¹ »¼ burned to fresh mixture which prevails over the
contribution given by reaction rate. The contribution
1/ 2
of thermal conduction can be considerate as
ª º negligible only below a flame kernel temperature
« » which is correlated to adiabatic flame temperature,
« k E bd »
ri « k 1 ˜ (5) i.e., Tk=3 Tad as proposed in [5]. Willems et al. [5]
§ T · » highlighted that this transition temperature is not
« p ˜ d gap ¨¨1  unb ¸S »
¸ »
«¬ © Ti ¹ ¼ really critical and an increase of this temperature or
a decrease of some hundreds Kelvin has a
minimum effect on the combustion, as also
The breakdown temperature Tb can be reasonably
remarked by Shen et al. [18]. In particular Shen et
assumed to be 60000 K according to [17].
al. [18] adopted the same choice for switching from
expansion to combustion reactions. In fact below
3. The sphere representing the kernel is discretized this temperature the expansion because of
by a set of triangular elements which move radially chemical reaction starts to prevail and therefore an
depending on burning rate and thermal expansion. energy balance for an open thermodynamic system
The number of elements used is 5000. A can be applied (PHASE 2).
connectivity database is created in order to relate
each element with the nodes belong to as well as
A) PHASE 1: Tkt3 Tad
with adjacent elements.
The temperature variation of the kernel is evaluated
(B) FLAME KERNEL MODELING
by solving the heat conduction equation for the
space-dependent plasma temperature Tpl
At each time step of the main CFD solver the following
according to [1]:
steps are taken:

1. Flame Kernel Expansion Velocity and

Thermodynamic Properties. wTpl § w 2Tpl 2 wTpl · P  Q w
D¨  ¸ (9)
wt ¨ wr 2 r wr ¸¹ Uc pVk
©
The sub-model is based on Herweg and Maly
model [1]. By the application of the mass
which assumes that the heat is dissipated in all
directions [16] and the thermal diffusivity is
 Downloaded from SAE International by Purdue University, Tuesday, August 21, 2018

constant. The power P comes from the power knowledge of the specific enthalpy, the kernel
released by electrical circuit. Q W represents the temperature Tk, i.e. burnt gas temperature, and the
heat power lost to electrodes and according to [5] kernel density Uk are then derived.
its expression is:
2. Turbulence wrinkling
Q W hc ˜ Ac ˜ (Tk  Twall ) (10)
The wrinkling factor ; is introduced to account for
turbulence which is experimentally found to affect
where the temperature at electrodes was assumed the kernel burning rate since its early development,
to be equal to temperature at walls Twall close to the as shown by Herweg and Maly in their experiments
spark plug and Tk is the kernel temperature, i.e. [1]. The wrinkling factor ; is computed as proposed
burnt gas temperature. by Colin et al. [10] and the lagrangian equation for
; reads:
The solution of Eq. (9) provides the kernel
temperature space and time dependent evolution. d;
The numerical solution of Equation (9) is decoupled P1;  Dk (12)
from main CFD solver. Suited sub-cycle time step dt
for time integration and a spherical sub-grid are
adopted. The time step in sub-cycle is 2.0e-8 s and where the destruction term Dk is:
the node spacing is 10 Pm. The temperature initial 1/ 2
conditions are: ª slam ,k º
Dk « ;  13Q t0.5 P11.5 » (13)
¬« 2 ¼»
T pl (0, r ) Ti if 0  r  ri
T pl (0, r ) Tunb if ri  r and P1 is the flame surface production term
calculated as in the ECFM model [8, 10].
while the boundary conditions are:
3. Flame Kernel Expansion: each node of triangular
elements moves at a velocity given by Eq. (8)
T pl Ti if r 0; which provides the kernel radius variation in time.
T pl Tunb if r rk ; The displacement is assumed to be radial and no
convective effect on flame kernel induced by fresh
gas velocity is modeled. The convection effects are
where the kernel radius rk is continuously updated. accounted for in the main combustion model.
Therefore mean CFD solver treats the convection
The kernel temperature Tk is taken as the average transport of the flame at the scale resolved
distribution in space of the plasma temperature Tpl according to grid resolution. The radial
between r=0 and r=rk at a certain instant. displacement of each node is enforced by
computing the local average normal of each
B) PHASE 2: Tk<3 Tad triangular surface the node belongs to.

The heat conduction from burned to fresh gases 4. Coupling with main CFD Solver: For the mass and
becomes negligible once the flame kernel has energy coupling with the CFD solver, three steps
reduced his temperature. Therefore the kernel take place:
temperature is estimated by applying an energy
balance equation for the open system represented (i) Each triangle of the spherical kernel lies in a fluid
by the flame kernel under the following cell which must be identified. Once this task has
assumptions: been accomplished with, the overall contribution in
the i-th cell to flame front surface Si-th can be
x The system is in thermodynamic equilibrium. calculated:
x The general gas law holds.
x The pressure is uniform in burnt and fresh M

gases. S i th ¦ Ak , j th (14)

i 1

Heat losses were not considered in the present

being M the number of triangular element lying in i-
model as done by Willems et al. [5].
th cell having an area Ak,j-th. It follows that the total
flame kernel area is:
Thus, the energy conservation gives the change in
specific enthalpy hk in time: N
Ak ¦ S ith (15)
dhk 1 ª dQch dE ele dQWV dp º hk dmk i 1
   Vk  (11)
dt mk «¬ dt dt dt dt »¼ mk dt
being N the total number of computational cells
where the flame front is located.
where Eele is the energy supplied by the spark plug,
Qch is the energy contribution from chemical
reactions, Qwv the energy lost at walls. From the
 Downloaded from SAE International by Purdue University, Tuesday, August 21, 2018

(ii) In the i-th computational cell containing the COMPARISON WITH DPIK AND AKTIM MODELS
kernel flame front, the flame surface density 6 is
updated according to: The present model differs in many points with respect
to the DPIK and AKTIM lagrangian ignition sub-
S ith ˜ ; models.
6 i th (16)
Vi th
1. The present electrical sub-model is simple since it
provides the spark energy released during
where Vi-th is the volume of i-th cell. In this step breakdown and glow phases based on few global
turbulence wrinkling effect is introduced. parameters. DPIK does not have electrical model
[7]. On the contrary AKTIM is considerably more
(iii) The mass and energy coupling with the main detailed and therefore it is expected to be more
CFD solver occurs by computing the fuel accurate once all data are precisely available [8].
consumption rate in the cell i-th accordingly to what In particular, in the present model re-ignition
done in the ECFM model: strikes are allowed based on the availability of
inflammable fuel mixture. In AKTIM a detailed
Z k ,i  th U unb ˜ s lam , k ˜ 6 i  th (17) positive column voltage model [20] is used and
therefore the voltage fall among electrodes can be
Based on the fuel consumption rate, species and computed depending on flow conditions.
internal energy are updated.
2. The present model uses sub-models for
(C) SWITCHING TO THE MAIN COMBUSTION determining the initial size and temperature of the
MODEL flame kernel after the arc phase. This should take
into account the effect of spark gap, fresh gases
When the kernel radius reaches a value of 2.0 mm, the temperature, breakdown energy on initial kernel
switch to the flamelet ECFM main combustion model conditions. In DPIK a reasonable initial flame
occurs. The ECFM takes as initial conditions for the kernel radius (i.e, 0.5 mm) is assumed. The value
flame surface density 6 the latest value available in is thus imposed without accounting for the effect of
each computational cell occupied by the flame front. the physics or spark plug characteristics. In the
As shown later, the transition from the 1D/Lagrangian AKTIM model, a different concept of the flame
flame kernel model to the main combustion model is kernel is adopted since the particle set does not
smooth thanks to the continuous updating of the flame represent a sole kernel but each particle is the
surface density during the flame kernel expansion. The center of gravity of a possible kernel having a
threshold value of 2.0 mm has been chosen as a statistical weight. The initialization of each kernel
reasonable value of length scale which can be surface comes from the initial mass of each kernel
resolved by RANS CFD thus allowing accounting for particle. At ignition time the latter is assumed to be
flow convective effect. as a fraction of the whole fresh gas mass which
will burn during ignition.
(D) MULTIPLE-IGNITION OPTION
3. In the present model the flame kernel temperature,
When high flow velocities at the gap take place, which affects the expansion velocity, is computed
restrikes could occur. Restrikes are determined by flow based on an accurate evaluation of the heat
induced stretching of the glow discharge till a rising conduction effects. Then, at relatively low
and sustaining voltage reaches the value required for temperatures a thermodynamic energy balance is
breakdown and creation of a new spark among applied.
electrodes [20].
4. In the present paper, the flame expansion velocity
derives from the differentiation of the mass
At that time, if mixture air index O is within inflammable
conservation equation by considering the effect of
limits, a new spark kernel can be formed. The re-
turbulence wrinkling on flame surface. This
ignition frequency is somehow proportional to flow
velocity is used to update the position of each
velocity close to spark plug. A phenomenological
triangular element which discretizes the flame
model for evaluating the total spark path voltage is
kernel during the whole ignition phase. In DPIK
given by the sum of the positive column voltage, the
model each flame marker expands radially, as in
anode fall voltage and the cathode fall voltage. In
the present model, with a velocity whose
particular, the positive column voltage between
expression depends on laminar velocity, flame
electrodes for the glow mode is a function of flow
adiabatic temperature, turbulence kinetic energy.
velocity, discharge current, gap size and pressure. Kim
Its derivation is not given in the paper [7]. In
and Anderson [20] proposed a phenomenological
AKTIM each particle is convected by gas flow
model for evaluating the positive column voltage
depending on its characteristics at spark gap.
during the glow phase. Since many data might be
difficult to be evaluated precisely, the present model
deals with re-ignition in a very simple way. After the 5. In the present model the coupling with CFD solver
end of each ignition phase, which ends with the switch is done in a such way that the flame surface
to the main combustion model, a check of air index at density in the i-th cell occupied by flame front
spark plug is done. A new ignition can occur and a comes from the contribution of the triangular
surface element of the kernel lying in that cell. The
new kernel is formed if the air index O is: 0.3 < O < 1.7.
 Downloaded from SAE International by Purdue University, Tuesday, August 21, 2018

continuous update of the flame surface density one to consider the influence of turbulence,
makes smooth the transition to ECFM combustion mixture quality, air flow convection and lump spark
model. The criterion for switching is given by plug characteristics on flame kernel expansion.
kernel radius.
NUMERICAL VALIDATION
6. A grid dependency analysis is provided for the
present model while not for the others. In order to test numeric behavior, the new ignition
model was applied to predict the ignition of a premixed
EXPERIMENTAL TEST CASE air-fuel mixture at the following conditions:

Herweg and Maly [1] experimentally investigated the x Air index ratio O: 1;
flame kernel formation in a cylindrical transparent side x Turbulent intensity: 1.48 m/s;
engine chamber. The latter was designed to induce a x Temperature: 660 K;
strong swirl in the incoming charge. By varying the x Pressure: 1 MPa;
engine angular speed and spark position, i.e., central x Flow: quiescent.
or peripheral, they studied separately the effect of the
mean flow velocity U and the turbulence intensity u’. The grid sensitivity test was performed by considering
Engine data characteristics can be found in Table 1. an uniform Cartesian grid with three different cell
sizes:
Bore 73 mm
Stroke 67 mm x 'x= 2.0 mm, referred to as “COARSE”;
Length of connecting rod 112 mm x 'x= 1.0 mm, referred to as “MEDIUM”;
Clearance height 1 mm x 'x= 0.5 mm, referred to as “FINE”;
Compression ratio 7.3:1
0.3
Diameter of side chamber 45 mm
COARSE
Clearance of side chamber 19 mm MEDIUM
0.25
TCI -> FINE
Total Energy= 60 mJ in 1.5 ms
Ignition systems 0.2
Burning rate [g/s]

CDI ->
Total Energy= 06 mJ in 0.3 ms
Spark Inter-Electrodes Distance 1 mm 0.15

Engine speeds 300, 500, 750, 1000, 1250 rpm

0.1
Air index 1.0, 1.3, 1.5
Ignition timing (CA deg.) 350 0.05
Mixture temperature
660 K
at ignition timing
0
Pressure at ignition timing 0.5 MPa
0 200 400 600 800 1000 1200 1400 1600
Fuel propane Time [Ps]

Table 1: Test conditions

Figure 2: Effect of grid refinement on burning rate
Two different ignition systems, respectively TCI predictions
(Transistorized Coil Ignition) and CDI (Capacitive
Discharge Ignition), were tested. Flow field properties Figures 2 and 3 plot the evolution in time of the
were determined by LDV measurements. Properties of burning rate and flame surface, respectively, predicted
the ignition system were characterized by their by the three different levels of refinement. From the
current/voltage characteristics. Thin wire electrodes analysis of Figures 2 and 3, it can be concluded that
were chosen in order to minimize additional effects of the approach adopted provides almost grid
the flow field. Kernel formation was measured by independent results. A refinement of 1 mm is
simultaneous high speed Schlieren filming from two considered enough for the convergence. It must be
orthogonal aspects, showing the axial and the radial noted that the ignition model is almost grid-
evolution with respect to the side chamber axis. independent. The effect of grid size reveals after 1000
Further details can be found in Herweg and Maly [1]. Ps when the main combustion model is switched on.
Therefore the grid-sensitivity seems more related to
MODEL VALIDATION the fact that the flame surface is unresolved using cell
whose size is longer than 1 mm.
The model has been validated in order to assess:
Finally, the effect of kernel discretization was
1. Numerics. A grid sensitivity analysis and an assessed by testing the influence of the number of
investigation of the effect of triangular elements on triangles on the kernel sphere surface prediction, as
model predictions have been carried out; one can see in Figure 4. Four different levels of
refinement were considered, from 1000 to 10000
triangles. As one can note, the convergence of the
2. Physics. The model is validated by reproducing
solution is obtained for a number of triangular
the Herweg and Maly test case. The latter allows
elements above 5000.
 Downloaded from SAE International by Purdue University, Tuesday, August 21, 2018

10 Simulations were carried out using the following

COARSE settings:
MEDIUM
8 FINE 1. A time step of 5.e-6 s was adopted;
Flame surface [cm ]
2

6 2. The computational domain was limited to the side

combustion chamber as done in [6]. A structured
grid with maximum cell size of 1.0 mm was used;
4

3. The temperature of the mixture was initialized

2 homogeneously throughout the side chamber at
the value of 660 K;
0
4. The gas mean velocity profile in the combustion
0 200 400 600 800 1000 1200 1400 1600
chamber was initialized according to experimental
Time [Ps]
results taken in [1]. In particular a non dimensional
Figure 3: Effect of grid refinement on flame surface velocity radial profile was imposed as reported in
predictions [1, 6]. In Table 2 the convection flow velocities at
ignition time in the spark plug peripheral location
6 are listed for three engine speeds.
1000
5 2500 5. The turbulence initial values, shown in Table 2,
5000 were set-up depending on the engine angular
10000
speed, as proposed in [1]. In particular Herweg
Flame surface [cm ]
2

4
and Maly introduced different values of the
3
normalized turbulence intensity for mean piston
speed at 350 CA deg. depending on the spark
2
location and the radial distribution.

1 6. The standard k-H turbulence model has been used.

0 Convection Turbulent
0 200 400 600 800 1000 Engine speed velocity [m/s] kinetic energy
Time [Ps]
[rpm] at spark plug at spark onset
2 2
location [m /s ]
1250 10 9.5
Figure 4: Effect of triangular element number on the
flame surface predictions surface 750 5 3.5
300 2 1.0
PHYSICAL VALIDATION
Table 2: Convection velocity and turbulent kinetic
energy values in the peripheral spark plug
In their experiments Herweg and Maly investigated the
location at the ignition onset for 1250, 750
effect of different parameters on the initial flame kernel
and 300 rpm of engine speed
development:

1. Ignition system characteristics; Engine

Ignition angular Air index Spark
2. Mixture quality;
system speed O [-] location
3. Turbulence levels; [rev/min]
CASE 1 TCI 1250 1.0 Peripheral
4. Gas flow convection. CASE 2 TCI 1250 1.3 Peripheral
The validation of the ignition model has been CASE 3 TCI 1250 1.5 Peripheral
CASE 4 TCI 300 1.0 Peripheral
performed by considering the ten different
CASE 5 TCI 750 1.0 Peripheral
combinations proposed in the Herweg and Maly paper CASE 6 CDI 300 1.0 Peripheral
[1] and in the last paper [6] because they are believed CASE 7 CDI 750 1.0 Peripheral
to be representative of the actual engine conditions. CASE 8 CDI 1250 1.0 Peripheral
The operating conditions simulated are summarized in CASE 9 CDI 1250 1.3 Peripheral
Table 3. Comparing Table 3 and the cases considered CASE 10 CDI 1250 1.5 Peripheral
by Herweg and Maly [1] one can note that in the
present validation only the peripheral spark location Table 3: Test conditions analyzed for model validation
was considered. This choice has been made because
the peripheral location was more representative for Figure 5 shows that the wrinkling factor depends on
model assessment than the central one where the the turbulence intensity, i.e. the engine speed: the
effect of flow convection on spark kernel is negligible. greater is the engine speed value, the greater is the
turbulence intensity and of consequence the wrinkling
factor. This particular part of the modeling is important
because the Herweg and Maly [1] experiments show
 Downloaded from SAE International by Purdue University, Tuesday, August 21, 2018

that turbulence affects the flame kernel since the very flow convects the flame kernel away from electrodes
early stage of its development. reducing the amount of energy it can receive. A slower
kernel growth occurs in the case of CDI system for a
Figure 6 provides with qualitative comparison the fixed air index and angular engine speed. All the
capability of the model of reproducing the influence of experimental evidences are qualitatively captured by
convection and turbulence on kernel growth for TCI the model proposed. Moreover as one can see in
system, air index ratio equal to 1 and different engine Figure 8 the predicted flame surface density increases,
speeds. When the engine angular velocity is increased increasing the turbulence intensity as indicated by
the flame kernel experiences a stronger flow motion experiments.
since early stage of its development, as visible in
Table 2 where are reported convection velocity values In Figure 9 is shown the influence of air index O on the
for 300, 750 and 1250 rpm of engine speeds. prediction of the flame surface for TCI system at 1250
Moreover, an increase of the engine angular velocity rpm engine speed. For the same cases Figure 10
results in an increase of mixture turbulence. As one reports the plot of kernel temperature. In Figure 11 the
can note from experimental images, the effect of the kernel radius trend for CDI ignition system at different
turbulence on the flame kernel growth is visible from air index values is depicted: the leaner is the gas
250 Ps after the spark discharge. The flame kernel mixture, the slower is the kernel radius increase.
wrinkling increases with the engine speed and this
induces a faster kernel growth. In Figure 7 the flame At different operating conditions it is interesting to plot
surface prediction for TCI system at O=1.0 for different not only the kernel volume but the time required to
engine speeds is plotted. reach a given volume. Figure 12 compares
experimental and numerical predictions of iso-volume
curves for TCI system at O=1.0 at different engine
speeds. A significant difference between model and
1.5 experimental results is seen only for very lean
1250 mixtures, as noted in [6] and experimentally in [1].
750 Figures 13 and 14 depict respectively the iso-volume
300 curves and lagrangian kernel radius prediction for TCI
system at engine angular speed of 1250 rpm for
Wrinkling factor [-]

1
different air index O. Both pictures show a reasonable
agreement between predictions and measurements.
For evaluating the breakdown frequency reliability of
0.5 Figure 14 authors did not have data available at the
moment. The only reference that could be used for
carrying out a comparison is the prediction of re-
ignition frequency of Duclos et al. [8] obtained applying
0 the AKTIM model. This comparison shows similar
0 50 100 150 200 250 values of the order of 300 Ps for similar convection
Time [Ps] velocity.

Figure 15 shows a sequence of pictures representing

the lagrangian kernel flame development, traced by
Figure 5: Stretch factor trend for different engine
markers, for TCI system at 1250 rpm and O=1.0. One
speeds – TCI ignition system – O=1.0
can note that the kernel re-ignition has occurred at
about 290 Ps as shown in Figure 14. In Figure 16 is
The ignition model and its coupling with flamelet model
traced a comparison between experimental [1] and
is able to capture the effect of the turbulence on the
flame growth. Moreover it is worthnoting that the simulated flame speed trends at air index O=1.0 for
transition from ignition sub-model to the ECFM model TCI ignition system at 1250 rpm. In Figure 17 for CDI
is smooth. Figures 8 provides a comparison between ignition system are reported kernel radius trends for
predicted flame surface density and experimental different engine speeds: as already mentioned in this
images in order to assess the capability of the model case re-ignition does not occur.
to reproduce the effect of different air index O and
ignition systems. This check is important to assess The model has shown to be capable of reproducing
model accuracy because kernel formation and the physics of the ignition process. A slight
dimension depend strongly on the mixture composition underprediction of the flame speed is present in almost
and the ignition system characteristics. It is to note that all conditions. It must be noted that present results
in the case of TCI system a flame holder effect [1] have been achieved without the use of any tuning
occurs despite the fact that the gas flow convects the parameter.
flame kernel away from electrodes. TCI system has a
longer glow phase which lets restrikes to occur. The
CDI system can not sustain the flame kernel
development in the case of a strong convection since it
releases an amount of energy (6 mJ) which is one
order of magnitude less than (60 mJ) the energy
released by TCI system. The effect is that the mean
 Downloaded from SAE International by Purdue University, Tuesday, August 21, 2018

SIMULATION EXPERIMENTS SIMULATION EXPERIMENTS SIMULATION EXPERIMENTS

300 rpm 300 rpm 750 rpm 750 rpm 1250 rpm 1250 rpm

100 Ps

250 Ps

500 Ps

625 Ps

Figure 6: TCI system-peripheral spark location - Ȝ=1.0 –axial view- Effect of convection velocity (Engine angular
velocity: 300 – 750 – 1250 rpm)-Predicted flame surface density iso-concentration (left) – Experimental
images (right) - The arrow shows the ignition point.

4
1250
3.5 750
300
3
Flame surface [cm ]
2

2.5

1.5

0.5

0
0 100 200 300 400 500 600
Time [Ps]

Figure 7: Flame surface trends for different engine speeds at O=1.0, TCI ignition system
 Downloaded from SAE International by Purdue University, Tuesday, August 21, 2018

SIMULATION EXPERIMENTS SIMULATION EXPERIMENTS SIMULATION EXPERIMENTS

O=1.0 O=1.0 O=1.3 O=1.3 O=1.5 O=1.5

250 Ps
TCI

250 Ps
CDI

750 Ps
TCI

750 Ps
CDI

Figure 8: Effect of Ignition system (TCI vs CDI) and air index on flame kernel growth. Engine angular velocity: 1250
rpm. Predicted flame surface density iso-concentration (left) – Experimental images (right)

4
1.0
3.5 1.3
1.5
3
Flame surface [cm ]
2

2.5

1.5

0.5

0
0 100 200 300 400 500 600
Time [Ps]

Figure 9: Flame surface trends for different air index O at 1250 rpm, TCI ignition system
 Downloaded from SAE International by Purdue University, Tuesday, August 21, 2018

4
7 10 2
1.0 1250 Exp
4
6 10 1.3 1250 Sim
1.5 750 Exp

Time after spark onset [ms]

1.5
Kernel temperature [K]
4 750 Sim
5 10
300 Exp
4 300 Sim
4 10
1
4
3 10

4
2 10
0.5
4
1 10

0 0
0 50 100 150 200 250 0 0.2 0.4 0.6 0.8 1 1.2
Time [Ps] Flame kernel volume [cm ]
3

Figure 10: Kernel temperature trends for different air Fig.13: Comparison between predicted and
index O at 1250 rpm, TCI ignition system measured flame kernel volumes for different
engine speeds and O=1.0, TCI System - by
2 modelling and experimental tests [1]
1.0
1.3 2.5
Lagrangian kernel radius [mm]

1.5
1.5 1250
750
2 300

Lagrangian kernel radius [mm]

1
1.5

0.5 1

0.5
0
0 50 100 150 200 250
Time [Ps] 0
50 150 250 350 450
Time [Ps]
Figure 11: Lagrangian kernel radius trends at 1250
rpm for different air index O, CDI ignition
system Figure 14: Lagrangian kernel radius trends for different
engine speeds with air index O=1.0, TCI
2 ignition system
1.0 Exp
1.0 Sim
1.3 Exp
Time after spark onset [ms]

1.5 1.3 Sim

1.5 Exp
1.5 Sim
1 a) b) c) d) e) f)

Figure 15: Flame kernel development for TCI ignition

0.5 system at 1250 rpm, O=1.0 – a) ignition time;
b) 50 Ps after ignition time; c) 200 Ps after
ignition time; d) 250 Ps after ignition time; e)
0 290 Ps after ignition time; f) 350 Ps after
0 0.2 0.4 0.6 0.8 1 1.2 ignition time
3
Flame kernel volume [cm ]

Fig. 12: Comparison between predicted and

measured flame kernel volumes for different
air index O – TCI System – 1250 rpm - by
modelling and experimental tests [1]
 Downloaded from SAE International by Purdue University, Tuesday, August 21, 2018

50 3
Sim TCI
Exp 2.5 CDI

Lagrangian kernel radius [mm]

Flame speed [m/s] 40

2
30
1.5
20
1

10
0.5

0 0
0 50 100 150 200 250 300 0 100 200 300 400 500 600
Time [Ps] Time [Ps]

Figure 16: Comparison between experimental [1] and Figure 18: Comparison between Lagrangian kernel
simulated flame speed trends at air index radius trends with re-ignition at air index
O=1.0, TCI ignition system, 1250 rpm O=1.0 for TCI and CDI ignition systems

2
1250
750 6 10
4
Lagrangian kernel radius [mm]

300
1.5
4
5 10

Kernel temperature [K] 4 10

4
1

4
3 10
0.5
4
2 10

4
0 1 10
0 50 100 150 200 250
Time [Ps] 0
0 50 100 150 200 250 300 350 400
Time [Ps]
Figure 17: Lagrangian kernel radius trends for different
engine speeds with air index O=1.0, CDI
ignition system Figure 19: Kernel temperature trend with re-ignition at
air index O=1.0, TCI ignition system
RE-IGNITION PROCESS
CONCLUSIONS
In the model presented in this paper re-ignition is
modeled by means a simple approach, as previously An ignition model based on Lagrangian approach was
said. presented. A lump model for the electrical circuit of the
spark plug is used to compute breakdown and glow
In Figure 18 it is possible to note the evolution of the energy. At the end of shock wave and very first plasma
kernel radius when multiple breakdowns are occurring. expansion, a spherical kernel is deposited inside the
One can note that at each re-ignition, the lagrangian gas flow at spark plug location. The sphere surface is
kernel radius is reinitialized at the value computed by discretized by triangular elements which move radially
means of Eq. (5), which depends on the local value of according to a lagrangian approach. Turbulence
pressure and temperature. Moreover the new kernel effects and thermodynamic properties of the air-fuel
radius value expands at a rate computed by the mixture are accounted for. Restrikes are possible
Lagrangian model. It is worthnoting that for CDI depending on gas flow velocity and mixture quality at
system the lagrangian kernel radius stops at 1.75 mm spark location.
instead of the reference value of 2 mm because the
spark duration is too short to let the kernel reach the The proposed model provided grid independent
value rk=2mm. Figure 19 shows the trace of the kernel results.
temperature due to multiple ignitions in which it is
notable that after the first ignition there is an increase The model was validated reproducing the experimental
of the value of the initial kernel temperature. tests by Herweg and Maly [1]. The accuracy of the
prediction of the effect of mixture quality, turbulence,
gas velocity and lump electrical spark plug
characteristics on flame kernel evolution was
 Downloaded from SAE International by Purdue University, Tuesday, August 21, 2018

investigated. Comparisons showed a good agreement 4. Andreassi, L., Cordiner, S., Mulone, V., Reynolds,
between experiments and numerical results. C., Evans, R.L., “Numerical-Experimental
Comparison of the Performance of a Partially
NOMENCLATURE Stratified Charged Natural Gas Fuelled Engine”,
ICEF2004-912, 2004 Fall Technical Conference of
A Area [m2] ASME, 2004.
Ac Contact area between burned gas and 5. Willems, H., Sierens, R., “Modeling the Initial
electrodes [m2] Growth of the Plasma and Flame Kernel in SI
cp Specific heat at constant pressure [J/kgK] Engines”, Transaction of ASME 2003, Journal of
cv Specific heat at constant volume [J/kgK] Engineering for Gas Turbines and Power, Vol.
Cbd Constant [kV/(J1/2mm1/2)] 125, pp.479-484.
dgap Inter-electrodes distance [mm] 6. Bianchi, G.M., Falfari, S., “Development and
Eele Electrical energy transferred to the park [J] Validation of an Ignition Model for SI Engines”,
Ebd Breakdown energy [J] ASME paper ICES2006-1432, ASME Spring 2006,
ECFM Extended Coherent Flame Model Aachen, Germany.
h Specific enthalpy [J/kg]
7. L. Fan, G. Li, Z. Han, R.D. Reitz, “Modelling Fuel
hc Convection coefficient [W/(m2K)]
Preparation and Stratified Combustion in a
k cp/cv
Gasoline Direct Injection Engine”, SAE paper
m mass [kg]
p Pressure [Pa] 1999-01-0175, 1999.
P Power [W] 8. Duclos, J-M., Colin, O., “Arc and Kernel Tracking
Ignition Model for 3D Spark-Ignition Engine
Q W Heat power lost to electrodes [W]
Calculations”, COMODIA 2001, IFP, 343-350.
Qch Heat of reaction [J] 9. Han, Z. and Reitz, R.D., “A temperature wall
QWV Heat losses to electrodes and walls [J] function formulation for variable density turbulent
ri Initial radius of the plasma [m] flows with application in engine convective heat
slam,k Flame velocity [m/s] transfer modeling”, Int. J. of Heat and Mass
T Temperature [K] Transfer, 40(3): 613-625, 1997.
Ti Initial temperature of the plasma [K]
10. Colin, O., Benkenida, A., Angelberger, C., „ A 3D
Tb Breakdown temperature [K]
Modelling of Mixing, Ignition and Combustion
Tunb Fresh gas temperature [K]
Phenomena in Highly Stratified Gasoline Engines“,
Tad Adiabatic temperature [K]
Tpl Plasma temperature [K] Oil & Gas Science and Technology – Rev. IFP,
Twall Wall temperature [K] 58,1,47-62.
U Mean flow velocity [m/s] 11. Meintjes, K. and Morgan, A.P., ”Element Variables
u’ Turbulent intensity [m/s] and Solutions of Complex Chemical Equilibrium
Vbd Voltage breakdown [V] Problems”, Los Alamos National Laboratory report
V Volume [m3] LA-10849-MS (October 1986).
12. Thiele, M., Selle, S., Riedel, U., Warnatz, J., ”A
Greeks Detailed Two Dimensional Numerical Study of
Spark Ignition Including Ionization”, SAE paper
D Thermal diffusivity [m2/s] 2002-01-1110, Detroit, 2002.
Qt Turbulent viscosity [m2/s]] 13. Bianchi, G.M., Pelloni, P., Filicori, F., and Vannini,
U Density [kg/m3] G., “Optimization of the Solenoid valve Behavior in
6 Flame front surface density [1/m] Common-Rail Injection Systems”, SAE
; Wrinkling factor [-] Transactions-Section 4. Journal of Fuels and
Lubricants, Vol.109, SAE 2000-01-2042, 2000.
Subscripts 14. Bianchi, G.M., Falfari, S., Pelloni P., Filicori, F.,
Milani, M., “A Numerical and Experimental Study
i initial on the Possible Improvements of C.R. Injectors",
k kernel SAE 2002-01-0500, March 2002.
unb fresh gas 15. Stone, R., Pashley, N., Graham, R., “Ignition
System Measurement Techniques and
REFERENCES Correlations for Breakdown and Arc Voltages and
Currents”, SAE paper 2000-01-0245, 2000.
1. Herweg, R., Maly, R.R., “A Fundamental Model for 16. Jeonghoon Song and Myoungho Sunwoo, “A
Flame Kernel Formation in SI Engines”, SAE Modeling and Experimental Study of Initial Flame
paper 922243, 1992. Kernel Development and Propagation in SI
2. Boudier, P., Henriot, S., Poinsot, T., Baritaud, T, Engines”, SAE paper 2000-01-0960.
24th Symposium (International) on Combution, 17. T. Kravchik, E. Sher, J.B. Heywood, “From spark
The Combustion Institute, 1992, pp. 503-510. ignition to flame initiation”, Comb. Sci. Technol.,
3. J. M. Duclos, G. Bruneaux, T. A. Baritaud, “3D 108: 1-30, 1995.
Modelling of Combustion and Pollutants in a 4- 18. Shen, H., Hinze, P., Heywood, J., “A Model for
Valve SI Engine; Effect of Fuel and Residuals Flame Initiation and Early Development in SI
Distribution and Spark Location”, SAE paper Engine and Its Application to Cycle-to-Cycle
961964, 1996. Variations ”, SAE paper 942049, 1994.
 Downloaded from SAE International by Purdue University, Tuesday, August 21, 2018

19. Metghalchi M., Keck, J.C., 1980, “Laminar burning 20. Kim, J., Anderson, R.W., “Spark Anemometry of
velocity of propane-air mixtures at high Bulk Gas Velocity at the Plug Gap of a Firing
temperature and pressure”. Combust. Flame 38, Engine”, SAE paper 952459, 1995.
143-154.