Energy Conversion and Management 165 (2018) 263–275

Contents lists available at ScienceDirect

Energy Conversion and Management

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

A new quasi-dimensional flame tracking combustion model for spark T

ignition engines
⁎
Momir Sjerića, , Darko Kozaraca, Henrik Schuemieb, Reinhard Tatschlb
a
Department of IC Engines and Mechanical Handling Equipment, Faculty of Mechanical Engineering and Naval Architecture, University of Zagreb, Croatia
b
AVL List GmbH, Graz, Austria

A R T I C LE I N FO A B S T R A C T

Keywords: A new quasi-dimensional combustion model based on the flame tracking approach is described and presented in
Spark ignition the paper. The new quasi-dimensional flame tracking model is able to simulate the turbulent combustion process
Flame tracking in premixed fuel/air/residual gas mixtures. A new method for the description of the geometry of the combustion
Quasi-dimensional chamber and flame front was developed enabling the visualisation of flame front movement across the com-
Combustion
bustion chamber. The control of local turbulence quantities in the flame front near the wall enables that the
developed flame tracking model can predict the entire turbulent combustion process after the flame kernel
development in spark-ignition engines without case-dependent calibration requirement. The developed quasi-
dimensional combustion model was validated with the experimental and results of multidimensional model of a
single cylinder spark ignition engine on averaged cycles. The model was integrated with the previously devel-
oped ignition, mixture stratification and cyclic variability sub-model that enable the simulation of cyclic com-
bustion variability triggered by the stochastic variations of flow angle at the spark plug, mixture stratification
and in-cylinder turbulence level. Due to the novelty of the model which includes the control of local integral
length scale and turbulent kinetic energy in the flame segments the predictive capability of quasi-dimensional
model is achieved with the application of single set of parameters related to average cycle and cyclic combustion
variability. Flame tracking model with the low computational time represents a promising tool to calculate the
turbulent combustion process including cyclic combustion variability in modern spark ignition engines.

1. Introduction Modelling of the combustion process in these models can be performed

The majority of transport worldwide will depend on fossil fuels for a
number of years to come. The transport is recognized as one of the
major sources of pollution and therefore governments around the world
have imposed a series of regulations that aim at reducing the harmful
emissions from internal combustion (IC) engines. As a result of these
regulations and general industry development, the research of IC en-
gines has been directed towards the optimization of the combustion
process so that more effective and cleaner processes can be achieved
[1].
As the computer power is getting more and more increased over the
last three decades, numerical simulations became very powerful en-
gineering tools used in variety of applications during the entire engine
development process from the conceptual phase to the calibration of the
engines. In the field of modelling of the IC engine working process, the
engine cycle-simulations, usually called 1-D/0-D models, offer a good
trade-off between the simulation model accuracy and calculation time.
by the application of empirically based combustion models (e.g. Vibe
function [2]) or by the application of predictive quasi-dimensional
combustion models. If the predictive quasi-dimensional combustion
models are properly calibrated, in the combination with 1-D flow model
across the intake and exhaust pipes, it can be used as a predictive si-
mulation tool for estimation of engine performance and emissions. In
the study presented in [3] the calibrated quasi-dimensional combustion
model was applied to predict the maximum engine torque and power
over the different engine speeds when the engine was fuelled by dif-
ferent fuels and the maximum achieved relative difference of measured
and simulated engine torque was ± 3%. The predictive capabilities of
quasi-dimensional combustion model were demonstrated in [4] where
the simulated emission formation of CO and NO from SI engine fuelled
by gasoline and natural gas matched well the experimental data.
The most commonly used quasi-dimensional combustion models for
spark-ignition (SI) engines are the fractal combustion model and the
turbulent entrainment model. The fractal combustion model is based on

⁎
Corresponding author at: Department of IC Engines and Mechanical Handling Equipment, Faculty of Mechanical Engineering and Naval Architecture, University of Zagreb, Ivana
Lučića 5, 10 002 Zagreb, Croatia.
E-mail address: momir.sjeric@fsb.hr (M. Sjerić).

https://doi.org/10.1016/j.enconman.2018.03.008
Received 16 November 2017; Received in revised form 1 February 2018; Accepted 3 March 2018
Available online 27 March 2018
0196-8904/ © 2018 Elsevier Ltd. All rights reserved.
M. Sjerić et al. Energy Conversion and Management 165 (2018) 263–275

Nomenclature red reduction

s secondary
A surface area (m2) SP spark plug
B bore (m) t turbulent
F reduction factor (–) tria triangle
Hs lower heating value of mixture (J/kg)
k turbulent kinetic energy per unit mass (m2/s2) Abbreviations
Li integral length scale of i-th flame point (m)
m mass (kg) AKI anti-knock index
RBZ individual gas constant of the burned zone (J/kg K) AKTIM arc and kernel tracking ignition model
Ri radius of i-th flame point (m) BDC bottom dead center
Ti,flame turbulence intensity in the flame (–) BTDC before top dead center
t time (s) CA crank angle (°)
u flame speed (m/s) CAD computer-aided design
u′ turbulent velocity pulsation (m/s) CCV cycle-to-cycle variations
V volume (m3) CFD computational fluid dynamics
CFR cooperative fuel research
Greek symbols CoV coefficient of variation (%)
CO carbon monoxide
λ excess air ratio (–) CR compression ratio (–)
υ kinematic viscosity (m2/s) DNS direct numerical simulation
ρ density (kg/m3) ECFM extended coherent flame model
Ω solid angle (sterad) EVO exhaust valve open
FTM flame tracking model
Subscripts FTPM flame tracking particle model
HC hydrocarbon
act active IC internal combustion
af anode fall IVC intake valve closure
bd breakdown IMEP indicated mean effective pressure (bar)
BZ burned zone LES large eddy simulation
cf cathode fall NOx nitric oxides
cg column of gas Q-D quasi-dimensional
f flame QDIM quasi-dimensional ignition model
max maximum ROHR rate of heat release (J/°CA)
N normal SI spark ignition
opp opposite TDC top dead center
p piston 0-D, 3-D zero, three dimensional
q quenching

the fractal theory considering the highly wrinkled thin flame front that
propagates across the combustion chamber by the laminar flame speed.
In the fractal combustion model integral length scales of turbulent flow
field are assumed to be larger than the thickness of reactive flame front
which means that the flame is considered in the wrinkled flamelet re-
gime [5]. The concept of the fractal combustion model is applicable to a
fully developed and freely expanding turbulent flame that is not the
case during the entire combustion process in SI engines. The application
of such combustion model and its predictive feature was shown in [6]
where SI engine was fuelled by hydrogen enriched compressed natural
gas. Although the equation of fractal dimension was improved ac-
counting for the hydrogen mole fraction in the mixture, the simulation
results demonstrated that the early flame stage and the late combustion
stage cannot be captured in a good agreement with the experimental
data without additional modifications of fractal dimension during these
two stages of combustion process. The transition from laminar to fully
developed turbulent flame that occurs during the early flame stage can
be described so that the progressive increase function is imposed on the
definition of maximum fractal dimension. Due to decrease of flame
wrinkling near the walls the late combustion stage is characterized as
the laminar one and another modification of the burning rate has to be
defined. Hence, the required additional modifications of the fractal
combustion model increase the number of calibration constants that
have to be calibrated. The extended fractal approach applied together
with the previously mentioned additional modifications that account
for the early and late combustion stage has shown slightly improved
prediction capability of the fractal combustion model over the different
engine speed compared to the standard fractal model, as presented in
[7]. Although the total number of calibration constants remains un-
changed, the potential solution to eliminate the necessity of artificial
wall combustion mode and to better predict the late combustion stage is
the application of two zone turbulence model where the unburned
turbulence quantities are used for the definition of flame wrinkling [8].
The turbulent entrainment model, firstly developed by Blizard and
Keck [9], assume the combustion process to take place in two steps. The
unburned mixture is firstly entrained by the smooth spherical flame
front that propagates by the entrainment velocity. After that, the en-
trained turbulent eddies in the fresh mixture burn in a specific time that
is proportional to the characteristic eddy size and laminar flame speed.
There are many versions of the turbulent flame entrainment model
based on the previously mentioned two steps of the combustion process,
with differences primarily arising from the choice of the characteristic
turbulent eddy size (integral length scale) and the entrainment velocity
which governs the entrainment of fresh mixture by the flame front. The
validation of entrainment turbulent combustion model using the ex-
perimental data of in-cylinder pressure and flame imaging was pre-
sented in [10]. It was shown that two combustion coefficients have to
be calibrated for each operating condition of the engine, especially the
characteristic integral length coefficient which indicates the significant
influence of operating parameters on the scale at which burn-up occurs

264
M. Sjerić et al.
behind the flame front. It was also concluded that the main area of
improvement of such quasi-dimensional combustion model could be in
more detailed definition of the sizes of reacting gas pockets. In other
words, the better prediction of local turbulence quantities in the reac-
tion zone can improve the predictive capabilities of quasi-dimensional
combustion model with reduced calibration requirements. It is evident
from the literature review that there is tendency to continuously im-
prove the predictive capabilities of quasi-dimensional combustion
models by the introduction of more and more details and additional
modifications that increase the model complexity and the number of
calibration constants.
Within this study the new quasi-dimensional combustion model was
developed with the aim to improve the predictive feature of the model
while satisfying the model simplicity and low number of calibration
constants. The model is based on the flame tracking approach where the
infinitely thin and smooth flame segment entrains the fresh mixture by
the turbulent flame speed. The main novelty of the flame tracking
model (FTM) is that the flame surface is described by a set of discretised
triangular elements whose corners shift along the predefined rays di-
rected toward the combustion chamber walls. The number of triangular
elements and rays is constant over the time and it is defined by the
icosahedron subdivision of the unit sphere. The second main novelty of
the model is the tracking and the control of turbulent quantities in the
flame discretization points which captures more 3-D details and de-
scribes the early flame stage as well as the effect of flame-wall inter-
action that is dominant during the late combustion stage. Therefore, the
unique equation for the combustion burning rate can be applied over
the entire combustion period without any additional modelling of ar-
tificial early flame development and flame-wall interaction, as it was
required in the fractal combustion model [7]. On the other hand, the
overall combustion process is governed by the definition of local tur-
bulence quantities in the flame which is detailed enough that it elim-
inates the requirement for the calibration of constants when engine
operating conditions are changed.
The newly developed quasi-dimensional combustion model was
Fig. 1. Projection of flame particle rays of the unit icosahedron sphere from spark plug position on chamber walls: example of the flat-flat combustion chamber
design.
265
Energy Conversion and Management 165 (2018) 263–275
validated by the comparison of simulation results with experimental
and 3-D CFD data of several operating points of a single cylinder ex-
perimental engine. The FTM predictive capability was investigated on
the cycle-average and cycle-resolved basis with the single set of cali-
bration constants. In order to simulate the cyclic combustion variability
in the combustion of SI engines, as one of the research topics that at-
tracts a lot of attention, the FTM was integrated with the previously
developed approach for the modelling of cyclic variability [11]. It was
shown that the new FTM offers insights into the detailed results in the
flame front as well as the flame front positions in the combustion
chamber on the basis of individual cycles. The visualization of the flame
front confirmed that the main part of cyclic combustion variability is
generated during the early stages of the combustion process in SI en-
gines.
2. Flame tracking model
The new quasi-dimensional combustion model is based on the flame
tracking approach and it was developed and integrated into a com-
mercial cycle-simulation software AVL BOOST™ [12]. Since there is no
spatial discretization of the combustion chamber using finite control
volumes when a 0-D approach is applied, several assumptions have to
be adopted for the new quasi-dimensional combustion model:
• Uniform pressure across the combustion chamber is assumed.
• The in-cylinder mixture is divided into unburned and burned zone
that are uniform in composition and separated by an infinitely thin
flame front represented by a set of triangular elements.
• The number of flame face triangles is determined by the applied
icosahedron subdivision of the unit sphere. The total number of
flame particles and triangles is constant over the combustion pro-
cess.
• The flame particle movement is defined only in the radial direction
(from the flame center towards the chamber walls) and its move-
ment is caused by the turbulent combustion and expansion of
M. Sjerić et al.
burned gases. The effect of local in-cylinder convective movement
on the flame front propagation is neglected.
• The flame faces contribute to the combustion process as long as they
reach the combustion chamber walls. The maximum path of each
flame face in the radial direction is defined as distance between the
flame origin and intersection of flame ray with chamber walls (cy-
linder head, liner and piston top) for each piston position.
• The calculation of the combustion burning rate is based on the as-
sumption that smooth flame front entrains the fresh mixture by the
turbulent flame speed.
Since the in-cylinder mass is divided into the burned and unburned
zone, the energy and continuity equation are applied to each zone [13].
In comparison with the Flame Tracking and Particle Method (FTPM)
that is used in the multidimensional discretized simulation model [14],
the newly developed quasi-dimensional combustion model considers
the flame front particles only, while the pre- and post-flame Lagrangian
particles are not modelled. Instead, the effect of pre-flame reactions and
knock conditions in the quasi-dimensional combustion model can be
calculated by application of the previously developed 0-D knock model
that uses detailed characterization of fuel and temperature stratification
in the unburned zone [15].
2.1. Geometry of the combustion chamber
The newly developed quasi-dimensional combustion model based
on the flame tracking approach describes the geometry of the com-
bustion chamber by calculating the maximum path of each flame face in
radial direction for the given piston position. The procedure of the
geometry description consists of two main steps. In the first step, the
data of an icosahedron sphere was defined where the surface of unit
sphere is discretised by uniformly distributed points that are connected
to 20 equilateral triangles. Each equilateral triangle can be additionally
subdivided into 4 equilateral triangles and the subdivision procedure
can be repeated several times. Therefore, the unit sphere can be dis-
cretized into Ntria = 20∗4n triangular faces, as shown at the top of Fig. 1
for subdivision levels n = 3, 4 and 5.
For each discretization equilateral triangle of the sphere, the center
of gravity can be determined and the total number of discretization
points is equal to Ntot = Ntria + (Ntria/2) + 2. For every discretization
point the unit vector in x-, y- and z- direction is defined, while the
corresponding solid angle Ωj is calculated for each discretization equi-
lateral triangle. In the second step of the geometry description, the
Fig. 2. Illustration of multiple intersections of flame rays with the walls of more complex combustion chamber geometry: (a) example of 3 intersections in the case of
a disc-chamber cylinder head, (b) decomposition of total tetrahedron volume on the particular volumes.
266
Energy Conversion and Management 165 (2018) 263–275
origin of the unit icosahedron sphere is translated to the location of the
spark plug position in the cylinder domain and the discretization points
are projected on the combustion chamber walls where the maximum
path in the radial direction is calculated. An example of the dis-
cretization point projection on the chamber walls (head, liner and
piston surface) for the different number of discretization triangles
(Ntria = 1280, 5120 and 20480) is given in Fig. 1 describing the overall
combustion chamber geometry at the specific piston position. De-
pending on the considered spark plug location and the particular ray
direction there are two groups of the calculated maximum paths of each
discretization point: rays whose maximum path does not change with
the piston movement from top dead center (TDC) to the bottom dead
center (BDC) and rays whose maximum path depends on the in-
stantaneous piston position:
Rmax, i = f (sp), (1)
where sp (–) represents the relative piston position sp = Sp/ S defined as
ratio between an arbitrary piston position Sp (m) and engine stroke S
(m). The maximum ray radii for both groups are stored in the following
matrix shape:
ps =0 ps =1
⎡ R max, 1 … R max, 1 ⎤
Rmax = ⎢ : : ⎥,
⎢ sp= 0 sp = 1 ⎥
⎢ R … R max, Ntot ⎥
⎣ max, Ntot ⎦ (2)
where Ntot (–) represents the total number of discretization points.
It is evident from the example of rays’ projection on the walls of flat-
flat combustion chamber design shown in Fig. 1 that each ray has only
one intersection with the chamber wall. For more complex geometries
of the cylinder head and piston top, multiple intersections can occur, as
shown in Fig. 2a in the case of a disc-chamber cylinder head.
When multiple intersections are detected the total tetrahedron vo-
lume is decomposed into the particular volumes as shown in the ex-
ample of 3 intersections in Fig. 2b. The flame origin and the first in-
tersection point of the triangle-center define tetrahedron volume V1, the
second intersection determines volume V2 and third intersection defines
the largest tetrahedron volume V3. The total tetrahedron volume is
equal to the sum of particular decomposed volumes where the second
volume is defined as the negative one. Such summation will give the
volume that is inside the combustion chamber volume. On the other
hand, the summation of resulting volume of particular tetrahedrons will
approximate the overall combustion chamber volume for the given
piston position. The developed procedure for the evaluation of the
M. Sjerić et al.
combustion chamber volume can handle up to 5 intersections (1, 3 or 5
intersections are possible solutions). In the case of 5 intersections, the
same analogy with the decomposed volume sign is applied - the second
and fourth decomposed volumes are negative so that the resulting tet-
rahedron volume corresponds to the one that is inside the combustion
chamber. In the case of multiple intersections the matrix presented by
Eq. (2) is generated for each intersection number and then it becomes a
3-D matrix shape.
The developed procedure for the combustion chamber geometry
description can be applied within an internal and external evaluation.
When the internal evaluation of combustion chamber geometry is ap-
plied only the simplified cylinder head and piston top designs [13] can
be calculated restricted to an assembly of plane, cylinder and sphere
segments. If the cylinder head and piston top have more complex
shapes which cannot be adequately captured by the simplified geo-
metries, an external evaluation of the geometry is needed. Although the
developed quasi-dimensional combustion model can only be applied for
the combustion modelling in SI engines, the module for the combustion
chamber geometry description can be applied to an arbitrary design of
combustion chamber. For the illustration purposes typical diesel-engine
piston bowl using external evaluation of combustion chamber geometry
is shown in Fig. 3.
In the case of external geometry evaluation, first it is necessary to
create the CAD model that represents the in-cylinder domain, as shown
in Fig. 3a. After that, the in-cylinder surfaces are discretized using the
appropriate surface mesh tool. The surface mesh of the combustion
chamber requires the specification of three surface selections that re-
present cylinder head, liner and piston top surface, as shown in Fig. 3b.
Beside surface selections, it is necessary to define the spark plug loca-
tion in the coordinate system of the mesh and engine stroke that are
together with the surface mesh used for the calculation of icosahedral
sphere projection on the chamber walls. The output results as maximum
paths of each ray are stored as function of piston position in the output
file that is afterwards read in during the pre-processing and initializa-
tion of the quasi-dimensional combustion model. The connection of
projected discretization points in the discretization triangles enables the
visualization of combustion chamber volume, as it can be seen in Fig. 3c
where the chamber volume at TDC is shown using 20,480 triangles.
The previously described internal and external procedure calculates
the maximum path of each discretization point for the arbitrary piston
position which enables the evaluation of instantaneous combustion
chamber volume during the FTM pre-processing. On the other hand,
when the combustion process is calculated and flame propagation is
considered, the burned zone volume can be determined using the same
approach where the instantaneous path of triangle centers and sum-
ming of all tetrahedron volumes is applied. In the following sections the
details of the burning rate calculation of the new combustion model are
given.
2.2. The flame surface area calculation
The newly developed quasi-dimensional flame tracking combustion
Fig. 3. Illustration of external evaluation of the combustion chamber geometry with ω shaped piston bowl; (a) CAD model, (b) discretized surface of combustion
chamber by 8192 triangular elements; c) projection of icosahedral sphere with 20,480 triangles; *all previews are captured at TDC.
267
Energy Conversion and Management 165 (2018) 263–275
model considers the flame front propagation across the combustion
chamber. The free flame front propagates and entrains the fresh mix-
ture by the turbulent flame speed. Since the flame front surface area is
described by the set of discretization triangles it is necessary to calcu-
late the free flame surface area of each discretization triangle so that the
combustion burning rate can be defined. Due to the model assumption
that all discretization points move only in the radial direction from the
flame center, the corresponding solid angle Ωj shown in Fig. 4 is con-
stant over the time. The corresponding solid angles are defined during
the model pre-processing using the unit vectors of icosahedron rays.
During the combustion process, each triangle corner and triangle
center of gravity (point G in Fig. 4a) moves along the path whose di-
rection is defined by the unit vector components. Due to different ve-
locities at the triangle corners the shifting of triangle corners within the
considered time step is different and the triangle can be rotated in any
direction around its center of gravity G, as illustrated with the unshaded
non-equilateral triangle in Fig. 4a. For calculation of the combustion
burning rate, the projection of such unshaded non-equilateral triangle
on the plane where an equilateral triangle exists (shaded triangle in
Fig. 4a) is needed. If the radius Rj, tria (m) of a triangle center is known
the total triangle surface AN,j (m2) can be approximated:
2
AN,j = R j,tria ·Ωj, (3)
where Ωi (sterad) is the corresponding solid angle of the tetrahedron
[16]. Eq. (3) is valid as long as the entire triangle is inside the com-
bustion chamber volume. The combustion model is developed in such a
way that the flame discretization points are shifted across the com-
bustion chamber and they are shifted outwards of the cylinder by the
mean flame front radius once they touch the chamber wall. If one or
more triangle corner points are shifted out of the cylinder domain the
free flame surface area is not equal to total triangle surface calculated
by Eq. (3). Therefore, a reduction function Fred, j for each triangle surface
area is defined and calculated according to the following expression:
Fred, j
⎛ ⎛ 1
= min ⎜1, max ⎜0, ·
2
⎝ ⎝
⎛1 + (Rj,1,max −Rj,1) + (Rj,2,max −Rj,2) + (Rj,3,max −Rj,3) ⎞ ⎞ ⎞,
(|Rj,1,max −Rj,1 | + |Rj,2,max −Rj,2 | + |Rj,3,max −Rj,3 |) ⎠ ⎟ ⎟
⎜ ⎟
⎝ ⎠⎠ (4)
where radii with indexes ‘max’ represent the maximum radii defined
within the pre-processing, while the others radii present the actual ra-
dius of each triangle corner. Due to imposed limitations in Eq. (4) the
value of reduction function Fred, j linearly decreases from 1 to 0 after one
of triangle corner reaches the chamber wall. The free flame surface area
inside the combustion chamber that affects the combustion burning rate
can be defined by the multiplication of the total triangle surface area
(Eq. (3)) by the value of reduction function (Eq. (4)):
∗
AN, j = Fred, j· AN, j . (5)
The overall free flame surface area inside the combustion chamber
M. Sjerić et al.
Fig. 4. Illustrations of free surface area (a) and burned tetrahedron volume (b).
∗
is equal to the sum of all particular triangle surface areas AN, j. In the
case of multiple intersections of the flame rays, as illustrated in Fig. 2,
after the first intersection with wall the particular flame point become
inactive in the combustion process. Once the mean flame front radius
(average value of active flame points) is higher than the second ray
intersection, the flame point becomes active again. This control enables
the reconstruction of flame faces behind the wall obstacles.
2.3. Turbulence modelling
The combustion process in SI engines is governed by the fuel oxi-
dation chemistry and effects of the in-cylinder flow field. The fuel
oxidation chemistry is described by the empirical correlation for the
laminar flame speed of considered fuel, while the turbulence effects can
be mainly attributed to the flame wrinkling increasing the flame speed
compared to the laminar one. For the prediction of turbulence effects on
the flame front propagation, the modelling of specific turbulence
quantities i.e. turbulent velocity fluctuations and integral length scale is
of crucial importance.
The available turbulence models that can be applied for the cycle-
simulation modelling are 1-equation (algebraic model) or 2-equation
models (k-ɛ model) [17]. Since the cylinder volume is not discretized on
finite control volumes when 0-D approach is applied, an average value
Fig. 5. The reduction of characteristic length scale of a turbulent eddy at the flame point close to the combustion chamber wall (a) and freely propagating turbulent
flame (b).
268
Energy Conversion and Management 165 (2018) 263–275
of turbulent kinetic energy over the time is calculated. At the defined
ignition time, the average value of in-cylinder turbulent kinetic energy
defines the initial turbulent kinetic energy for the flame propagation. In
the presented study, the previously developed k-ɛ turbulence model
[18] was used. The average integral length scale LI (m) of turbulent
eddies is assumed to be proportional to the instantaneous height of the
simplified cylinder [19]:
LI = CL ·(VCact /((B /2)2 ·π )), (6)
where VCact (m3)
is the actual cylinder volume, B (m) is the engine bore
and CL (−) is a scaling parameter. After the definition of average values
of turbulent kinetic energy and integral length scale the uniform dis-
tribution of both turbulence quantities is applied to the triangle corners.
Since the turbulence properties change over time the actual turbulent
kinetic energy and the actual integral length scale at triangle corners
are calculated as:
k iact = k iold·(1 + Δk∗), (7)
Liact = Liold ·(1 + ΔL∗), (8)
where the non-dimensional (relative) change of turbulent kinetic en-
ergy Δ k∗ (−) and non-dimensional change of integral length scale
Δ L∗ (−) are calculated as:
M. Sjerić et al. Energy Conversion and Management 165 (2018) 263–275

k act−k old combustion burning rate.

Δ k∗ = ,
k old (9)
2.5. Turbulent burning rate calculation
Lact −Lold
Δ L∗ = .
Lold (10)
The flame front is described by the set of triangles and the turbu-
The superscripts marked with ‘act’ are related to the turbulence lence quantities are controlled over the triangle corners. The turbulence
property at the given time, while ‘old’ superscripts present the values of quantities in flame points outside the wall combustion conditions are
turbulence property at the end of the previous time step. equal to the average value of these properties calculated by the applied
The value of non-dimensional change of turbulent kinetic energy turbulence sub-model. For each flame discretization triangle the tur-
Δ k∗ depends on the applied turbulence model. If the single zone k-ɛ bulent combustion heat release rate can be calculated and the sum-
turbulence model is applied, the non-dimensional change of turbulent mation of particular combustion rates over the triangles gives the
kinetic energy depends on the profile of the overall in-cylinder turbu- overall frontal combustion heat release rate:
lent kinetic energy [18], while in case of the two zone k-ɛ turbulence N tria 3
model, the non-dimensional change of turbulent kinetic energy depends Q̇f = Hs·
dm f ⎡ ⎛ 1 ·A ∗ ⎞⎤
dt
= Hs· ∑ ⎢ρUZ · ∑ ⎜ n norm,j · ut , i ⎟ ⎥,
on the profile of the unburned zone turbulent kinetic energy [8]. j=1 ⎣ i = 1 ⎝ j, act ⎠⎦ (12)

2.4. Wall combustion conditions where nj, act (−) represents the number of triangle corners inside the
combustion chamber for j-th flame triangle. The turbulent flame speed
The main feature of the combustion in SI engines is the propagation u t, i is calculated according to the engine correlation proposed in [21]:
of the flame across the combustion chamber. The existence of boundary 1/2
⎡ ⎛ ⎛ u′i ⎞ uN ·L i 1/4 ⎞ ⎤
layers and specific geometry segments (e.g. crevice zone) lead to flame u t, i = uN ·max ⎢1.0, ⎜−4.37 + 1.13· u ·⎛ ⎞ ,
υ ⎠ ⎟⎥
⎜ ⎟

deceleration and occurrence of flame quenching due to the high heat ⎣ ⎝ ⎝ N⎠ ⎝ ⎠⎦ (13)
losses to the wall. The main novelty of the new quasi-dimensional
where uN (m/s) is the laminar flame speed, and υ (m2/s) is the kinematic
combustion model is the control of turbulence quantities for the flame
viscosity. In the proposed quasi-dimensional combustion model the
points that are close to the combustion chamber walls, thus the flame
laminar flame speed is calculated using the empirical correlation of
deceleration effect can be captured. As the flame approaches the wall,
Metghalchi and Keck [22], while the kinematic viscosity is approxi-
the existence of large scale turbulent eddies is disrupted due the high
mated using the unburned zone temperature [13].
viscous forces (high dissipation) and limited mean flow velocity (low
Due to turbulent flame speed which is calculated at triangle corners,
production of turbulence). Since the maximum path of each ray Rmax, i is
each corner is shifted along the predefined direction from the previous
calculated during the pre-processing, the reduction of particular in-
position, as illustrated in Fig. 5b. At the end of the calculation time step
tegral length scale can be correlated to the difference between the
Δt , the shift of triangle corner ΔRi caused by the frontal combustion is
maximum path Rmax, i and the actual flame point radius R i . The reduc-
equal to:
tion of integral length scale in the flame point will be performed only if
the actual value of the integral length scale is higher than the geometry ΔRi = ut, i ·Δt (14)
limit L i,wall for that flame point decreasing the local turbulent flame where ut, i (m / s ) is the turbulent flame speed calculated by Eq. (13). As
speed. The segment of flame front and specific radii for the selected the flame point moves to the combustion chamber wall, the reduction
flame point that is close to the chamber wall are shown in Fig. 5a. of turbulent quantities becomes significant reducing the local turbulent
The application of a homogeneous and symmetric icosahedron flame speed and combustion burning rate.
subdivision of the unit sphere (see Fig. 1), defines that particular flame
point has its pair that has an opposite direction from the flame origin, as
2.6. Expansion of burned gas
shown in Fig. 5a with the blue and red radius vector. Under the as-
sumption of the integral length scale being the diameter of a turbulent
The movement of triangle corners affects the movement of the tri-
eddy, the reduced value of the integral length scale represents the
angle center of gravity and the tetrahedron volume defined by the tri-
minimum value defined from the flame point forward and backward
angle position continually increases. During the combustion the in-
direction criteria:
ternal energy of burned gas increases and the burned zone expands. At
L i,wall = min (Ct ·2·(Ri,max,opp + Ri ), Ct ·2·(Ri,max −Ri )), (11) the end of each time step it is possible to calculate the ‘geometrical’
burned zone volume and the burned zone volume from the gas equa-
where Ct (−) is a user defined calibration parameter. If the reduction of
tion. ‘Geometrical’ volume means that the real position of flame tri-
the integral length scale is performed for a certain flame point, the
angles is taken into account for the definition of tetrahedrons. If ‘geo-
reduction of its turbulent kinetic energy will also be performed as-
metrical’ burned volume and burned volume defined from the gas
suming the linearity between turbulent kinetic energy and integral
equation are not the same, an expansion of burned gases exists and an
length scale. In other words, if the integral length scale is reduced by
additional flame point shifting is required. The ‘geometrical’ burned
30%, the turbulent kinetic energy is reduced by the same ratio. Since
zone volume can be calculated by summing the particular tetrahedron
the combustion in SI engines is initiated by the electric spark discharge
volumes. The separated tetrahedron volume of the burned zone defined
that occurs close to the cylinder head, the certain flame points at the
by flame center (FC ) and actual positions of flame points (R1, R2 and R3 )
beginning of turbulent combustion are under the wall combustion
is illustrated in Fig. 4b.
conditions. The proposed control of wall combustion conditions leads to
The overall ‘geometrical’ burned zone volume can be evaluated by
a significant decrease of turbulence quantities in the flame front during
summing of particular tetrahedron volumes inside the cylinder with the
the late combustion phase as the entire flame front approaches the
base of the equilateral flame triangle:
combustion chamber walls. The mentioned effect of the local flame-
Ntria Ntria
wall interaction on the turbulent burning rate cannot be taken into 1 3
account with the application of two zone turbulence and quasi-di-
VBZ = ∑ Vbz,j =
3
· ∑ R j,tria ·Ωj
j=1 j=1 (15)
mensional combustion models, such as the fractal combustion model
[19] or the entrainment combustion model [20], where the average If the burned zone volume calculated from the gas equation is
values of turbulence quantities (for entire cylinder mass or for the un- higher than the one calculated by Eq. (15), the shifting of triangle
burned zone) are calculated and applied for the calculation of corners is made until the volume equilibrium is achieved. This means

269
M. Sjerić et al.
that the overall movement of a triangle corner within the calculation
time step consists of two parts: movement due to turbulent flame speed
and shifting due to expansion of burned gas. After that, the new time
step is initiated and the calculation of turbulent burning starts with the
new triangle position. The entire procedure is repeated until the entire
mixture is burned.
3. Validation results
The validation of the new quasi-dimensional combustion model was
performed by comparisons of the cycle-simulation results with the
available 3-D CFD and experimental results. The analyzed operating
points of experimental engine include the variation in engine speed and
spark timings.
3.1. Experimental setup and operating points
The experimental single cylinder Waukesha CFR F4 engine was used
for the measurement of reference data. The experimental results include
the in-cylinder pressure data measured with the KISTLER 7061B water-
cooled pressure transducer, crank angle resolved intake pressure, intake
and exhaust temperatures measured with K-type thermocouples and air
equivalence ratio measured with the wide band lambda sensor. The
port fuel injection and spark timings were managed by the MOTEC M4
control unit. More details about the entire experimental setup and
particular components can be found in [11,23]. For each operating
point of the engine 300 in-cylinder pressure traces have been recorded
with the resolution of 0.1 °CA. The offline application with the Savitzky-
Golay filter was used for the pressure data filtering and after that the
recalculation of combustion rate for each individual cycle was per-
formed based on the energy equation. The averaging of recalculated
rate of heat release (ROHR) over the crank angle (for 300 cycles) de-
fines the cycle-averaged reference data used for the calibration of the
new combustion model when the average cycle is simulated.
The specification of the measured operating points of the CFR en-
gine is given in Table 1. The engine was fuelled by gasoline with stoi-
chiometric mixture and with the compression ratio set to 8. Two engine
speeds were considered and a certain spark sweep at each engine speed
was made representing 7 knock-free operating points.
In the real operating conditions of SI engines, even with the constant
imposed intake boundary conditions the instantaneous in-cylinder
pressure profiles show non-repeatable and stochastic behavior known
as cycle-to-cycle variations (CCV) in combustion. There are several
factors that have been identified as the main sources of CCV in SI en-
gines: the variation of in-cylinder flow and turbulence characteristics,
the variation of global air equivalence ratio, the variation of mixture
composition (imperfect mixing of fuel, air and combustion products
[24]), the variation of ignition energy and spark discharge character-
istics. All these factors have certain impact on the combustion varia-
bility from the early flame kernel growth phase to the combustion
termination. The experimental in-depth studies of CCV in combustion
can be made with the optically accessible combustion chamber of SI
engine. The influence of two intake air swirl ratios on the early flame
kernel growth and combustion stability has been quantified experi-
mentally in [25] with the application of proper orthogonal decom-
position. It was shown that 10 times increase of intake air swirl ratio
increases early flame speed 2 times, while the cyclic variability in
combustion was reduced by 50%. The experimental study using the
laser induced fluorescence performed in [26] showed that strong in-
cylinder flow and increased swirl ratio decreases the variation in the
fuel distribution and therefore the lower combustion variability occurs.
Due to presence of the cyclic variability in extreme operating conditions
of SI engine, the fast burning cycles can reach knock combustion, while
the slow burning cycles can reach misfire increasing the exhaust gas
emission and fuel consumption. The experimental analysis of misfires in
a direct injection engine was performed in [27] using the proper
270
Energy Conversion and Management 165 (2018) 263–275
orthogonal decomposition. It was shown that partially burned and
misfire cycles has leaner mixtures and lower convection speeds around
the spark plug that lead to insufficient flame kernel energy for sus-
taining the flame propagation.
When the CCV in IC engines are desired to be numerically analyzed
and predicted, Large Eddy Simulation (LES) or Direct Numerical
Simulation (DNS) have to be employed so that the stochastic behavior
of fluid motion can be captured correctly. The main drawback of such
simulation models is requirement for the fine mesh resolution of the
computational domain that dramatically increases computational time.
Therefore, LES are usually applied to define the specific variations of
particular physical properties over the number of simulated operating
cycles (e.g. variation of flow characteristic) and then such perturbations
are imposed into 1-D/0-D simulation models where large number of
operating cycles is simulated due to low computational time. The same
approach was applied in the study presented in [28] where LES was
applied for the simulation of motored engine condition, and calculated
distributions of the tumble number, integral length scale and convec-
tion lengths (at spark plug location) were imposed as Gaussian dis-
tributions into 1-D combustion model. Although the variation in mix-
ture stratification and the early flame kernel growth were not modelled
in 1-D model, the simulation results were in a good agreement with
measured data with slightly under-predicted combustion rates for slow
burning cycles.
In this study, the newly developed FTM is integrated with the pre-
viously developed ignition, mixture stratification and cyclic variability
sub-model where the independent variations of in-cylinder turbulence,
stratification of air equivalence ratio [29] and flow angle at the spark
plug [11] from cycle-to-cycle can be defined. For the considered op-
erating points of the CFR engine, a single set of model constants and
standard deviations that define Gaussian distributions of particular
property, specified in Table 2, were applied.
The cycle-simulation model constants specified in Table 2 are di-
vided into 3 main groups: constants related to the ignition sub-model
(QDIM), turbulence model constants and combustion model constants
(FTM). The variability of the in-cylinder turbulence level from cycle to
cycle was initiated by the variation of intake production constant S10
defined by its relative standard deviation σrel, S10 . The variability of
mixture stratification from cycle to cycle is defined by the relative
standard deviation of air equivalence ratio σϕ that defines the dis-
tribution of air equivalence ratio at the start of combustion. Both
standard deviations need to be tuned so that the correct prediction of
cyclic combustion variability, usually expressed via the coefficient of
variation (CoV) of the indicated mean effective pressure (IMEP), is
achieved. All other constants specified in Table 2 are related to the
calculation of an average cycle.
Since the experimental studies show importance of early flame
kernel growth and its influence on the cyclic combustion variability
[25,27], the initial flame kernel growth in the simulation should in-
clude the main sources that affect these variations. The variation in the
early flame kernel growth in the applied simulation model is defined by
the variations of flow angle at the spark plug, convective flow velocity
and air equivalence ratio. The variation of flow angle at the spark plug
changes the heat losses from the hot kernel to the spark plug electrodes
[30] and changes the final flame center location for the evaluation of
flame rays that define free flame surfaces. The variations of convective
Table 1
Specification of measured operating points of CFR engine.
Fuel Gasoline (AKI 91)
Tintake (°C) 32
pintake(bar) 1.013
λ (–) 1.00
CR (–) 8
n (rpm) 900 1200
ST (°CA BTDC) 5 10 15 20 10 15 20
M. Sjerić et al. Energy Conversion and Management 165 (2018) 263–275

Table 2 boundary temperatures of cylinder head and piston were set to constant
Overview of the single set of cycle-simulation model constants. values of 130 °C and 180 °C, respectively. The temperature of cylinder
Ignition model (QDIM) r0 (mm) 5.00 liner changes linearly as the piston moves from IVC to TDC when the
kW (W/m2·K) ; Es (0) (J) 1500; 90 maximum temperature of 130 °C is defined. The same boundary tem-
TSP (K) ; Tq (K) 500; 1600 peratures were defined in the cycle-simulation model of AVL BOOST™.
RS (Ω); LS (H) 20000; 20 The initial pressure and temperature in the cylinder at IVC are set to
Vaf (V) ; Vcf (V) 18.75; 252 1.16 bar and 97 °C according to the experimental data, while the re-
Ccg (–); C bd (V/ J·mm ) 40.46; 1.34·105 sidual gas content is defined from the cycle-simulation model and set to
k-ε turbulence model S00 (–); S10 (s/m) 0; 1.2·10−6 5.8%. The initialization of the flow was performed assuming the tumble
C00 (–); C10 (bar−1) 3.50; –1.80 flow with the initial tumble ratio of 0.65 at IVC, while the initial tur-
Cε ; C2 (–) 2.30; 1.92 bulent kinetic energy was set to 2.6 m2/s2 so that the turbulent velocity
Variation of turbulence σrel, S10 (–) 0.09
pulsations at TDC are not higher than the mean piston speed. The
New FTM CL (–) 0.15
Ti,flame (–) 0.022 modelling of turbulence was performed with the default k-zeta-f
Ct (–) 0.50 [31,32] turbulence model in AVL FIRE™, while the simulation of
Variation of mixture stratification σϕ (–) 0.04 combustion process was made by the application of Extended Coherent
Flame Model (ECFM-3Z) [33] that is based on the flame surface density
approach. The simulation of spark ignition process was captured by the
flow velocity at the spark plug are directly correlated to the turbulent Arc and Kernel Tracking Ignition Model – AKTIM [34]. The calculation
velocities calculated by the applied turbulence sub-model. This means time step of 0.5 °CA was set in both simulation models. The 3-D CFD
the operating cycles that have higher in-cylinder turbulence level have results of the combustion and turbulence were used for the additional
larger convective flow velocity at the spark plug and vice versa. The validation of newly developed quasi-dimensional combustion model. In
convective flow velocity defines the shifting of the flame kernel, thus the next section the experimental results are compared with the 3-D
affects the kernel growth speed due to the effect of different heat losses. CFD and new Q-D simulation results.
The variation of air equivalence ratio is defined by the previously ap-
plied stratification sub-model which defines the instantaneous air
equivalence ratio in the flame and its change as the combustion ap- 3.3. Cycle-averaged results
proaches to the end [11]. By means of air equivalence ratio variation
the laminar flame speed changes resulting in the variation of early The calculated average cycle results of in-cylinder pressure profiles
flame kernel growth and the remaining part of the combustion process. for the considered operating points of the CFR engine are plotted and
In addition, the variation of in-cylinder turbulence level from cycle-to- compared with the measured and 3-D CFD data in Fig. 6. The simula-
cycle defines the variation in the turbulent flame speed and therefore tion results of averaged cycle calculated by the application of the new
the variation in the combustion burning rate occurs. Although the 1-D/ quasi-dimensional combustion model matched well the measured and
0-D approach is applied in the presented study, the imposed variations 3-D CFD data although the single set of constants applied. The faster in-
in flow characteristics (angle, velocity magnitude and turbulence ve- cylinder pressure development in the case of new quasi-dimensional
locity pulsation) and mixture composition include most of the factors combustion model occurs at all operating points, while the over-pre-
that contribute to the dynamic behavior and cyclic combustion varia- diction (≈10%) of in-cylinder peak pressure can be observed for op-
bility in SI engines. erating point ST5 at an engine speed of 900 rpm.
The comparisons of ROHR profiles, shown in Fig. 7, give more in-
3.2. 3-D CFD model sights regarding the prediction of particular combustion phases that
cannot be directly seen from the pressure traces. The 3-D CFD simula-
Four operating points of experimental CFR engine that have engine tion results at engine speed of 900 rpm correctly predict the entire
speed fixed to 900 rpm were simulated using the 3-D CFD simulation combustion progress, while at the pressure traces the small over-pre-
model made in AVL FIRE™. The high pressure cycle (from intake valve dictions for ST5 and ST10 operating points can be observed during the
closure – IVC, to exhaust valve open – EVO) was simulated with the expansion phase. The potential reason for such difference is the under-
primarily attention to the modelling of combustion process. The moving prediction of heat losses and blow by effects that are not modelled in
computational mesh of hexahedron control volumes was defined. In the applied 3-D CFD simulation model. The new Q-D combustion model
order to avoid the excessive deformation of control volumes during the correctly predicts the combustion duration for all operating points,
piston movement, the re-layering option was applied which changes the even in ST5 operating point at 900 rpm where the peak value in ROHR
number of control volume from 35,000 at TDC to 170,000 at EVO. The is about 20% higher. The over prediction of peak in-cylinder pressure

Fig. 6. Comparison of averaged in-cylinder pressure profiles of CFR engine.

271
M. Sjerić et al. Energy Conversion and Management 165 (2018) 263–275

Fig. 7. Comparisons of ROHR over the analyzed operating points of CFR engine.

and ROHR in the case of ST5 operating point can be caused by the
several factors. The mixture stratification over the combustion chamber
can exist on the average cycle, while in the new Q-D fully homogeneous
mixtures were considered when the average cycles were simulated. In
the real operating conditions the average temperature in the fresh
mixture before the reaction flame zone that defines the laminar flame
speed can be different from the average unburned zone temperature
that is not the case in the new Q-D model. In addition, the existence of
convective flow movements (e.g. tumble motion) during the combus-
tion can distort the flame shape, while in the new Q-D combustion
model this phenomenon is not modelled. Therefore, all mentioned local
effects can slightly change the combustion profile compared to the one
observed from the experimental data.
It can also be observed that new Q-D combustion model slightly
over-predicts the early flame stage when transition from laminar to
fully developed turbulent flame occurs over the analyzed operating
points. The reason for that is the over-prediction of in-cylinder turbu-
lence level, as it can be seen in Fig. 8 where the 3-D CFD and 0-D results
of turbulent kinetic energy during the combustion for ST5 operating
point are compared. The slightly higher turbulence during the first
20 °CA after ignition potentially increases the turbulent flame speed
(see Eq. (13)) and therefore the combustion burning rate slightly in-
creases. The simulation results of ROHR calculated by the new Q-D
model shown in Fig. 7 demonstrate that the predictive capability of
developed combustion model is achieved, while the number of cali-
bration parameters is even lower compared to the existing fractal
combustion models.
The values of calculated IMEP for different operating conditions of
the CFR engine are compared with the measured results and shown in
Fig. 9. Although the simulated peak in-cylinder pressure is larger by
10% compared to measured data for ST5 operating point, the IMEP
value shown in Fig. 9 is only larger by 3%. The over prediction of in-
cylinder pressure profile during the combustion partially compensates
the pressure differences during gas exchange, compression and expan-
sion. The pressure differences during gas exchange, compression and
expansion may be caused by the differences in the wall heat transfer
and blow by effects that are simulated by the single set of model
parameters. Bearing in mind that a single set of cycle-simulation model
constants is applied for 7 considered operating points, the simulations
results are in a very good agreement with the measured data.
3.4. Cycle-resolved results
The cyclic combustion variability is usually expressed by means of
CoV of IMEP that is calculated by dividing the standard deviation of
IMEP by the average value of IMEP. The simulated results and results of
CoV of IMEP determined from the measurement are compared and
shown in Fig. 10. As already mentioned before, the cyclic combustion
variability in the cycle-simulation model was initiated by the in-
dependent variations of the in-cylinder turbulence level, mixture stra-
tification and flow angle at the spark plug that oscillate from cycle to
cycle. The flow angle variation follows a uniform distribution [13],
while the variation of intake turbulence constant and initial difference
in the air equivalence ratio from the average value are defined by user
defined relative standard deviations σrel, S10 and σϕ , as specified in
Table 2. Both standard deviations were tuned manually to achieve the
cyclic combustion variability close to the measured data and the single
values of standard deviations were found for the considered operating
points of the CFR engine. The predicted CoV of IMEP is in a very good
agreement with the measured results for the considered operating
points that include a change of engine speed and spark sweep.
Fig. 8. Specific in-cylinder turbulent kinetic energy (TKE) for ST5 operating
point.

272
M. Sjerić et al.
Fig. 9. Comparison of average cycle results of the indicated mean effective pressure (IMEP): measurement vs. simulation.
Therefore, the proposed FTM that is coupled with the previously de-
veloped approach for the modelling of cyclic combustion variability
represents a well promising, predictive and fast simulation tool for the
further analysis of cyclic combustion variability in SI engines that be-
comes very important in predicting the knock occurrence [35].
For the selected operating point ST20 at engine speed of 900 rpm,
300 consecutive pressure traces are plotted and compared in Fig. 11. In
both, measured and simulated cases the slow and fast burning cycles are
shown with the green solid lines whereas the averaged cycles are shown
with the dashed lines. The slow burning cycles have lower in-cylinder
turbulence level and lean mixture at the start of combustion, while the
opposite case occurs for the fast burning cycles. It can be seen from
Fig. 11 that the new FTM is able to correctly predict the fast burning
cycles that become very important for the modelling of knock occur-
rence and NOx emissions, while the appropriate prediction of slow
burning cycles creates better conditions for the modelling of emissions
of CO and unburned HC in the cycle-simulation models.
The advantage of the new FTM that uses the flame discretization
points is the possibility of flame front visualization inside the com-
bustion chamber at any given moment. The instantaneous combustion
chamber geometry is described by the projection of the icosahedron on
the chamber walls with the instantaneous piston position, while the
flame front is described by the actual radius of the triangle center that is
inside the combustion chamber. A comparison of flame front positions
at the selected time sequences for the slow, average and fast burning
Fig. 10. Comparison of coefficient of variation (CoV) of IMEP: measurement vs. simulation (sample size is 300 cycles).
273
Energy Conversion and Management 165 (2018) 263–275
cycle of operating point ST20 is given in Fig. 12.
The mean flame front radius Rf for individual cycles and at the
elapsed crank angle (CA) of 6.5, 10.8, 27 and 54° after spark onset is
also shown in Fig. 12. The comparison of slow and fast burning cycles
shows that the ratio between the mean radius of the fast burning cycle
and of the slow burning cycle at 6.5 °CA after spark onset is around 2.
As the combustion process goes to the end, this ratio continuously de-
creases and at the final combustion phase it is close to 1. This confirms
that the main part of cyclic combustion variability is generated during
the early stages of the combustion process in SI engines.
4. Conclusions
A new quasi-dimensional combustion model based on the flame
tracking approach was developed and integrated into the environment
of a commercial 1-D/0-D simulation model. For the description of the
combustion chamber geometry and flame front a new approach was
applied which uses an icosahedron subdivision of the unit sphere whose
surface can be discretized by the different number of equilateral tri-
angular elements. The projection of the icosahedron sphere from the
spark plug position toward the chamber walls describes the entire
combustion chamber volume for the given piston position. The pro-
posed FTM controls the turbulence quantities in the flame points as the
flame front approaches the combustion chamber wall. The reduction of
integral length scale and turbulent kinetic energy in the flame points
M. Sjerić et al.
Fig. 11. Comparison of measured and simulated in-cylinder pressures for 300 consecutive cycles of operating point ST20 at 900 rpm.
Fig. 12. Comparison of flame front positions at the selected time sequences for slow, average and fast burning cycle of operating point ST20 at 900 rpm.
enables that the FTM can be applied for the calculation of the entire
turbulent combustion process in SI engines. The overall flame front
movement in the developed FTM is caused by the turbulent flame speed
and expansion of burned gases. The developed FTM is validated with
the measured and 3-D CFD results of a single cylinder SI engine. The
validation with the measured data required the integration of FTM with
the previously developed approach for the modelling of cyclic com-
bustion variability. After the validation of the FTM model with ex-
perimental and 3-D CFD results and flame front visualization for one
operating point, the following conclusions can be drawn:
• The proposed control of turbulence quantities (turbulent kinetic
energy and integral length scale) in the flame discretization points
close to the chamber walls represents the control of wall combustion
conditions. Such model feature enables that the FTM can be applied
for the simulation of the entire combustion process without any
additional sub-models for the laminar-turbulent flame transition and
artificial wall combustion sub-model. The combustion process is
274
Energy Conversion and Management 165 (2018) 263–275
governed by the flame front movement and the evolution of tur-
bulence quantities in the flame discretization points over the com-
bustion chamber.
• The new quasi-dimensional FTM correctly responds on the imposed
cyclic variability in the cycle-simulation environment. Hence, the
model is able to simulate cyclic combustion variability and to pre-
dict the fast burning cycles that become very important for the
modelling of knock occurrence and NOx emissions, and slow
burning cycles that contribute to better conditions for the modelling
of emissions of CO and unburned HC.
• The cyclic combustion variability in SI engines is mainly generated
during the early stages of the combustion process, as can be seen
from the flame front visualization shown in Fig. 12.
The newly proposed quasi-dimensional combustion model (FTM) is
on one hand characterized by the fast computational time (≈2 s per
operating cycle with the application of Ntria = 5120 ) while on the other
is detailed enough to simulate the entire turbulent combustion process
M. Sjerić et al.
in SI engines including the cyclic variability without case-dependent
calibration of model constants. Therefore, the FTM represents a well
promising simulation tool for predicting the engine performance of
modern SI engines of different configurations.
Acknowledgments
The authors are very grateful to the Combustion analysis Laboratory
of University of California, Berkeley and to Andrew Van Blarigan, PhD,
for the help in the realization of the experiment. The numerical study
was done within the DUFCOROC (HrZZ-1089) project “Experimental
research, optimization and characterization of a piston engine opera-
tion with dual-fuel combustion“ funded by the Croatian Science
Foundation.
References
[1] Mikulčić H, Vujanović M, Duić N. Reducing the CO2 emissions in Croatian cement
industry. Appl Energy 2013;101:41–8. http://dx.doi.org/10.1016/j.apenergy.2012.
02.083.
[2] Vibe II. Brennverlauf und Kreisprozess von Verbrennungsmotoren. Berlin: VEB
Verlag Technik Berlin; 1970.
[3] da Silva Trindade WR, dos Santos RG. 1D modeling of SI engine using n-butanol as
fuel: Adjust of fuel properties and comparison between measurements and simula-
tion. Energy Convers Manage 2018;157:224–38. http://dx.doi.org/10.1016/j.
enconman.2017.12.003.
[4] D’Errico G. Prediction of the combustion process and emission formation of a bi-fuel
s.i. engine. Energy Convers Manage 2008;49:3116–28. http://dx.doi.org/10.1016/j.
enconman.2008.06.012.
[5] De Bellis V, Severi E, Fontanesi S, Bozza F. Hierarchical 1D/3D approach for the
development of a turbulent combustion model applied to a VVA turbocharged en-
gine. Part II: combustion model. Energy Procedia 2014;45:1027–36. http://dx.doi.
org/10.1016/j.egypro.2014.01.108.
[6] Ma F, Li S, Zhao J, Qi Z, Deng J, Naeve N, et al. A fractal-based quasi-dimensional
combustion model for SI engines fuelled by hydrogen enriched compressed natural
gas. Int J Hydrogen Energy 2012;37:9892–901. http://dx.doi.org/10.1016/j.
ijhydene.2012.03.045.
[7] Daniela T. A quasi-dimensional SI combustion model: A bi-fractal approach. Energy
Procedia 2017;126:931–8. http://dx.doi.org/10.1016/j.egypro.2017.08.166.
[8] Sjerić M, Kozarac D, Tomić R. Development of a two zone turbulence model and its
application to the cycle-simulation. Therm Sci 2014;18:1–16. http://dx.doi.org/10.
2298/TSCI130103030S.
[9] Blizard NC, Keck JC. Experimental and theoretical investigation of turbulent
burning model for internal combustion engines. In: SAE Tech. Pap. 740191, 1974.
doi: http://dx.doi.org/10.4271/740191.
[10] Irimescu A, Merola SS, Valentino G. Application of an entrainment turbulent
combustion model with validation based on the distribution of chemical species in
an optical spark ignition engine. Appl Energy 2016;162:908–23. http://dx.doi.org/
10.1016/j.apenergy.2015.10.136.
[11] Sjerić M, Kozarac D, Tatschl R. Modelling of early flame kernel growth towards a
better understanding of cyclic combustion variability in SI engines. Energy Convers
Manage 2015;103:895–909. http://dx.doi.org/10.1016/j.enconman.2015.07.031.
[12] AVL BOOSTTM v2017 Manual; 2017.
[13] Sjerić M. New physically based sub-models for the cycle-simulation of spark-igni-
tion engine, University of Zagreb, Faculty of Mechanical Engineering and Naval
Architecture, Ph.D. Thesis; 2014.
[14] AVL FIRETM v2017 Manual; 2017.
275
Energy Conversion and Management 165 (2018) 263–275
[15] Kozarac D, Tomic R, Taritas I, Chen J-Y, Dibble RW. A model for prediction of
knock in the cycle simulation by detail characterization of fuel and temperature
stratification. SAE Int J Engines 2015;8:1520–34. http://dx.doi.org/10.4271/2015-
01-1245.
[16] Wikipedia - Solid angle, n.d. https://en.wikipedia.org/wiki/Solid_angle#
Tetrahedron.
[17] Ramos JI. Internal combustion engine modeling. 1st ed. Washington, USA:
Hemisphere Publishing Corporation; 1989.
[18] Sjerić M, Kozarac D, Ormuž K. Cycle-simulation turbulence modelling of IC engines.
Int J Automot Technol 2016;17:51–61. http://dx.doi.org/10.1007/s12239-016-
0004-2.
[19] Bozza F, Gimelli A, Merola SS, Vaglieco BM. Validation of a fractal combustion
model through flame imaging. SAE Tech. Pap. SP-1969; 2005. p. 1–17. doi: http://
dx.doi.org/10.4271/2005-01-1120.
[20] Verhelst S, Sheppard CGW. Multi-zone thermodynamic modelling of spark-ignition
engine combustion - An overview. Energy Convers Manage 2009;50:1326–35.
http://dx.doi.org/10.1016/j.enconman.2009.01.002.
[21] Frolov SM, Ivanov VS, Ya BV, Belyaev AA, Basara B, Priesching P, Suffa M. Flame
tracking – particle method for 3d simulation of normal and abnormal (knocking)
operation of spark-ignition automotive engines. In: Int. Automot. Conf. NUMV,
Belgrade, Serbia; 2015. p. 83–91.
[22] Metghalchi M, Keck JC. Laminar burning velocity of propane-air mixtures at high
temperature and pressure. Combust Flame 1980;38:143–54. http://dx.doi.org/10.
1016/0010-2180(80)90046-2.
[23] Van Blarigan A, Kozarac D, Seiser R, Chen JY, Cattolica R, Dibble R. Spark-ignited
engine NOx emissions in a low-nitrogen oxycombustion environment. Appl Energy
2014;118:22–31. http://dx.doi.org/10.1016/j.apenergy.2013.12.007.
[24] Heywood JB. Internal combustion engine fundamentals. USA: McGraw-Hill Inc;
1988.
[25] Chen H, Xu M, Hung DLS, Zhuang H. Cycle-to-cycle variation analysis of early flame
propagation in engine cylinder using proper orthogonal decomposition. Exp Therm
Fluid Sci 2014;58:48–55. http://dx.doi.org/10.1016/j.expthermflusci.2014.06.017.
[26] Yang J, Xu M, Hung DLS, Wu Q, Dong X. Influence of swirl ratio on fuel distribution
and cyclic variation under flash boiling conditions in a spark ignition direct injec-
tion gasoline engine. Energy Convers Manage 2017;138:565–76. http://dx.doi.org/
10.1016/j.enconman.2017.02.024.
[27] Chen H, Reuss DL, Sick V. Analysis of misfires in a direct injection engine using
proper orthogonal decomposition. Exp Fluids 2011;51:1139–51. http://dx.doi.org/
10.1007/s00348-011-1133-z.
[28] Pera C, Richard S, Angelberger C. Exploitation of multi-cycle engine LES to in-
troduce physical perturbations in 1D engine models for reproducing CCV. In: SAE
Tech. Pap.; 2012. doi: http://dx.doi.org/10.4271/2012-01-0127.
[29] Sjerić M, Kozarac D, Šagi G. Development of a stratified fractal combustion model.
In: 23rd JUMV Int. Automot. Conf. with Exhib., 19th - 21st April 2011, Belgrade,
Serbia; 2011.
[30] Sjerić M, Taritaš I, Kozarac D. Effect of spark plug geometry on the cyclic com-
bustion variability and fuel consumption of gasoline engines. J Energy Eng
2017;143. http://dx.doi.org/10.1061/(ASCE)EY.1943-7897.0000492.
[31] Basara B. Eddy viscosity transport model based on elliptic relaxation approach.
AIAA J Tech Notes 2006;44:1686–90. http://dx.doi.org/10.2514/1.20739.
[32] Hanjalić K, Popovac M, Hadžiabdić M. A robust near-wall elliptic-relaxation eddy-
viscosity turbulence model for CFD. Int J Heat Fluid Flow 2004;25:1047–51. http://
dx.doi.org/10.1016/j.ijheatfluidflow.2004.07.005.
[33] Colin O, Benkenida A. The 3-zones extended coherent flame model (ECFM3Z) for
computing premixed/diffusion combustion. Oil Gas Sci Technol – Rev IFP
2004;59:593–609.
[34] Duclos JM, Colin O. Arc and kernel tracking ignition model for 3D spark-ignition
engine calculations. In: Int. Symp. COMODIA 2001, Nagoya, Japan; 2001.
[35] Sjerić M, Taritaš I, Tomić R, Blažić M, Kozarac D, Lulić Z. Efficiency improvement of
a spark-ignition engine at full load conditions using exhaust gas recirculation and
variable geometry turbocharger - numerical study. Energy Convers Manage
2016;125:26–39. http://dx.doi.org/10.1016/j.enconman.2016.02.047.