Downloaded from SAE International by University of Birmingham, Sunday, August 19, 2018

SAE TECHNICAL
PAPER SERIES 2007-01-0939

Study of Cyclic Variation in an SI Engine Using

Quasi-Dimensional Combustion Model
E. Abdi Aghdam, A. A. Burluka, T. Hattrell,
K. Liu and C. G. W. Sheppard
School of Mechanical Engineering, The University of Leeds

J. Neumeister and N. Crundwell

MAHLE Powertrain Ltd.

Reprinted From: Modeling of SI and Diesel Engines, 2007

(SP-2079)

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 University of Birmingham, Sunday, August 19, 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: permissions@sae.org
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: CustomerService@sae.org

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 University of Birmingham, Sunday, August 19, 2018

2007-01-0939

Study of Cyclic Variation in an SI Engine Using

Quasi-Dimensional Combustion Model
E. Abdi Aghdam, A. A. Burluka, T. Hattrell, K. Liu and C. G. W. Sheppard
School of Mechanical Engineering, The University of Leeds

J. Neumeister and N. Crundwell

MAHLE Powertrain Ltd.

Copyright © 2007 SAE International

ABSTRACT Ozdor et al. [3] suggests that total elimination of cycle-

to-cycle variation would result in a 10% increase in brake
The paper is concerned with the effects of cyclic varia- power output for the same fuel consumption. The cur-
tion in turbulence (expressed in terms of rms turbulent rent trend in legislation towards reducing NOx emissions
velocity) on the burn rate and subsequent cyclic varia- is driving engine designers to increase the amount of ex-
tion in in-cylinder pressure derived parameters. The task haust gas recirculation and encouraging development of
has been addressed by applying a thermodynamic en- lean-burn concepts. Both of these strategies tend, how-
gine modelling approach for simulations of two very dif- ever, to result in increased cyclic variability. In particu-
ferent engines; a single cylinder research engine in which lar, cycle-to-cycle variation in early flame development re-
sources of cyclic variation other than turbulence had been stricts lean operation for any particular fuel [4]. Reduc-
minimised and a multi-cylinder production engine. The ing cycle-to-cycle variations in combustion would allow
cyclic variability in the two engines had a number of sim- the engine to operate “under average conditions that are
ilar features; the effects of turbulence variation cycle-to- closer to the limiting ones” [3], thus permitting engine de-
cycle proved dominant in the production engine, mixture signers to develop power plants with better performance in
strength secondary and prior-cycle residual concentration both driveability and environmental impact. Pre-requisite
feedback marginal. to reducing ccv at the engine design stage are modelling
tools for assessing the impact of cyclic variation effects;
INTRODUCTION this is addressed in the current work within the framework
of a so called quasi-dimensional engine model.
Cycle-to-cycle variation (ccv) in internal combustion en-
gine performance, defined as non-repeatability of instan-
taneous combustion rate between different cycles at nom- MEASURES OF CYCLIC VARIATIONS A number of
inally identical operating conditions, has long been identi- different parameters can be used as measures of the
fied [1] as a limiting factor in determining the performance amount of cyclic variability in an engine. The most com-
of an engine. It is now generally recognised that under monly used parameters are [1, 5, 6]:
operating conditions where an engine is liable to knock,
the octane requirement, maximum compression ratio and
1. indicated mean effective pressure (IMEP);
spark timing are limited by the propensity for autoignition
to occur in the fastest burning cycles [2]. Where autoigni- 2. in-cylinder pressure, either maximum, or taken at a
tion is not a problem, the spark timing for a given running specific crank angle; alternatively, the crank angle at
condition is typically optimised for the heat release profile which maximum or some given pressure occurs;
of the most frequently occurring cycle. Penalties in terms
of lost power and efficiency are inevitably incurred for cy- 3. the rate of in-cylinder pressure change;
cles which deviate from the mode.
4. the time at which the flame arrives at a given point in
It has been suggested [1] that a coefficient of variation in the cylinder;
IMEP of greater than 10% is noticeable to the driver as
a deterioration in the vehicle driveability. The review of 5. variably defined combustion duration or “ignition de-
lay”;

1
 Downloaded from SAE International by University of Birmingham, Sunday, August 19, 2018

6. the mass fraction burnt (MFB) at a prescribed crank is because, although they affect ccv magnitude, they do
angle; not vary in themselves between cycles and therefore can-
not be a primary cause of ccv.

An appropriate standard deviation or coefficient of vari-

ation (COV) can be defined for any of these parame- RELATIVE IMPORTANCE OF CCV ORIGINS As early
ters, with a larger value indicating greater cyclic variabil- as 1953 Mickelsen and Ernstein [9] observed that the
ity. However, because of the very different nature of these burning rate of a free turbulent flame in a wind tunnel was
variables, their absolute values cannot be compared di- a random quantity, the distribution of which was approx-
rectly. For example, a standard deviation of 5% in peak imately Gaussian. For stronger turbulence and for non-
pressure denotes a much smaller variability than that indi- stoichiometric mixtures the variance of burning rate was
cated by a standard deviation of 5% in the duration of the found to increase. The variations in burning rates ob-
period between 10 and 90% mass fraction burnt. The sit- served by Mickelsen and Ernstein [9] must be attributed
uation is somewhat complicated by the lack of correlation entirely to the first of the above-listed factors, because the
between these parameters, e.g. the cycles with the high- effect of variation in the spark discharge properties and
est IMEP are not those cycles with the highest peak pres- other parameters were ruled out. It was found that there
sure [7]. Arguably, of the above-listed criteria, the most was very little correlation between either the spark cur-
commonly used in the automotive industry (the variation in rent, spark energy or spark kernel displacement and the
IMEP) is the most confusing as it compounds several fac- observed turbulent flame speed. Large variations in burn-
tors [6]. For research purposes, instantaneous pressure- ing rate have been observed for turbulent deflagrations in
related parameters are more useful than the IMEP as a fan-stirred bombs, where the charge composition is per-
measure of cyclic variability but their use is complicated by fectly uniform and well controlled [10, 11]. Shen et al.
the fact that the cylinder pressure is affected not only by [12] performed simulations of an SI engine using different
combustion but also by the changing combustion chamber sub-models to simulate variability in flame kernel convec-
volume and heat transfer between the gas and walls. The tion, level of turbulence experienced by the flame kernel
most interesting measures of ccv in combustion for a real during the early stages of combustion and the level of tur-
engine are those directly related to the actual combustion bulence experienced during the main flame propagation.
process, characterised either in terms of flame propaga- They suggested that ccv in turbulence, particularly during
tion speeds or in terms of integral mass burning rate. Be- the early stages of combustion, had the largest impact.
cause of the disparate nature of these different parame- The predominance of the fluid flow factor (more specifi-
ters and the complexity of their inter-relationship, the only cally, variations in the turbulence conditions experienced
means of assessing the combined effects of the above by the developing flame kernel) on ccv in freely propagat-
variables is to have a model which predicts from funda- ing explosions has been demonstrated by Lipatnikov and
mental principles the cyclic variability in the combustion Chomiak [13]. This work established the link between the
event. internal intermittency of turbulence, which is manifested
in the log-normal distribution for the instantaneous values
of the dissipation, and the amplitude of the spread in the
ORIGINS OF CYCLIC VARIATIONS The source of ccv turbulent diffusivity defining the flame growth.
has been associated with a number of interacting fac-
tors [1, 5, 8]: In an IC engine, in addition to turbulence effects, varia-
tion in the second and third factors in the above list can
be appreciable. This is especially true under part-load,
1. charge motion and “turbulence” (usually expressed in
low volumetric efficiency, and/or stratified charge and/or
terms of rms turbulent velocity) in the cylinder during
lean mixture conditions [4]. However, Hinze and Miles
combustion;
[14] found that the spatial variation in residual fraction in
2. the amounts of fuel, air, and residual and/or recircu- a measurement volume near the spark plug decreased
lated exhaust gas in the cylinder; throughout the compression stroke; to an rms value of
less than 1% at 15° bTDC, an amount unlikely to cause
3. uniformity of the mixture composition within the cylin- any significant ccv. For port fuel injection engines or
der, especially near the spark plug, associated with gasoline direct injection engines where the injection oc-
imperfect mixing between the air, fuel and resid- curs early in the intake stroke, although the charge may
ual/recirculated exhaust; be strongly non-homogeneous during the intake stroke,
turbulent diffusion during the compression stroke is often
4. spark discharge characteristics, such as breakdown strong enough to reduce inhomogeneities in the charge
energy and initial flame kernel random displacement. composition to a negligible level. This is the assumption
made in the present work.
Some of the above-listed factors are of much greater im-
portance than others. A number of other factors, such as There is general agreement based on earlier observations
the type of spark plug and spark gap orientation, which in disc-chamber engines [15–17, and references therein]
are sometimes cited as causes of ccv are not listed. This that characteristics of the spark have little influence upon

2
 Downloaded from SAE International by University of Birmingham, Sunday, August 19, 2018

subsequent flame propagation (provided the spark energy lence in a fan-stirred vessel. In order to determine a value
exceeds the minimum ignition energy). This conclusion is of u
, velocity measurements must be averaged over a
supported by the recent measurements of Aleiferis et al. certain time. They showed that, if many measurements of
[18]; however, their measurements clearly demonstrated u
were made, the standard deviation of a set of measure-
that, in a pent-roof engine with a repeatable mean veloc- ments, σu
, decreased from approximately 0.2u
to 0.03u

ity pattern at the ignition instant, the spark location and as the averaging period used for a particular set increased
orientation may have some impact on early stage com- from one to one hundred integral time scales. Moreover,
bustion. Cycle-to-cycle variations in combustion rate are their work suggested that variations in turbulent flame
manifest from the very moment of ignition [16]. Despite speed were caused by the limited duration of a deflagra-
this, their observed magnitude cannot be attributed en- tion event during which time the flame is subject to a fluc-
tirely to the initial flame formation and development period tuating u
. Qualitatively, in an engine, the total duration of
[19]. Indeed, quoting from this latter work: “even when the combustion is just a few integral time scales and the mea-
variations in mean velocity were coupled to the effects of surements of Al-Khishali et al. [10] show that the variation
both heat losses to the electrodes and spark-energy re- σu
in the rms velocity averaged over this period is about
lease patterns, the combined effect was found to account 10 to 20% of its average value. For engine conditions a
for no more than 40-50 per cent of the observed cyclic similar value of 26% has been reported by Shen et al. [12],
variability in flame-kernel radius at the 5% MFB timing of who obtained this value assuming that the turbulence ki-
40° CA AIT”. Finally, it is worth noting that variations at netic energy was a constant fraction of the total charge ki-
the instant of 5% MFB usually account for no more than netic energy. Choi and Guezennec [24] conducted three-
40-50% of the total ccv magnitude [20]. dimensional PIV measurements, with N = 100 samples,
in a water analog engine. They reported an even higher
value of σu
amounting to 0.45u
at certain locations within
CYCLIC VARIABILITY IN TURBULENCE The above the cylinder. In the present work, the standard deviation
studies and a large number of other [16] observations of the rms velocity σu
is introduced into the expression
suggest that the first factor in the above list, that is cyclic for the burning rate used in a quasi-dimensional model.
variability in the flow field affecting the burning rate, is pre- Due to the rather large uncertainty in the available mea-
dominant. The flow motion in any piston engine is un- surements, in the current work the exact magnitude of
steady and has an average pattern and a fluctuating ve- σu
is implied by comparison between the calculated and
locity component. Discerning an “appropriate” turbulent measured ccv amplitude at certain engine operating con-
parameter is not trivial. A clear separation between mean ditions.
and fluctuating motions exists in neither kinetic energy
(important for the proper orthogonal modes decomposi-
tion [21]), nor characteristic frequencies, (important for CYCLIC VARIABILITY MODEL PHILOSOPHY A num-
Fourier-modes decomposition). Nevertheless, in a num- ber of former studies [8, 12, 25] have attempted to model
ber of experimental studies the turbulent rms velocity has cyclic variability in engines; using separate sub-models
been deduced from the sum of the high-frequency and the for ignition, combustion, flame geometry and turbulence.
deviation from the mean low-frequency parts of the mea- Depending on the particular type of model, input parame-
sured velocity spectrum [22]. Taking into account that the ters for AFR, turbulence, residual gas fraction etc. have
temporal resolution is not the same in these studies and been provided either as random variables, computed from
that under engine conditions it is difficult to resolve the a knowledge of previous cycles, or some combination of
smallest scales of turbulence, a value so obtained should both methods. By varying the simulation input parame-
not be much different from ters from one cycle to another, such models then attempt
  2 1/2 to replicate the changing conditions found in a real engine
1  N
1 N
combustion chamber. If the engine simulation code is re-
u
(Θ, N ) = u(Θ, i) − u(Θ, i)  (1)
N N alistic and detailed enough, these changing conditions will
i i
be reflected by variations in combustion and hence cylin-
which is the definition used in PIV (Particle Image Ve- der pressure and temperature. Such models aim to repro-
locimetry) studies [23]. In Eq. 1 u
(Θ, N ) is the rms veloc- duce the level of variations observed for a given engine
ity at crank angle Θ averaged over N cycles and u(Θ, i) is condition. Because full-scale CFD simulation of the air-
the instantaneous velocity value in the i−th cycle. When exchange process coupled with combustion simulation is
the number of sampled cycles N tends to infinity, the limit computationally demanding, so-called thermodynamic or
of u
(Θ) = lim u
(Θ, N ) is the sum of both the proper tur- zonal models seem to offer a sensible compromise be-
N →∞
bulent rms velocity and the rms contribution coming from tween computational complexity and accuracy of predic-
cycle-to-cycle variation in the pattern of the mean veloc- tions for the large number of simulated cycles required.
ity. When N is finite, u
(Θ, N ) is a fluctuating quantity;
of which both the average value u
(Θ) and the variance Study of ccv within the framework of a thermodynamic
σu
are functions of N . This is similar to the averaging zonal model has been previously attempted in a number
of a continuous signal over a finite time considered by Al- of works [12, 26, 27]. The present study has adopted a
Khishali et al. [10] who presented LDV measurements of similar framework with a more refined combustion model
velocity in stationary, homogeneous and isotropic turbu- and account of crevice and blow-by flow. The simulation

3
 Downloaded from SAE International by University of Birmingham, Sunday, August 19, 2018

“predictions” have first been tested against experimental ical on the mean during the early part of the cycle only.
results obtained in an engine where residual gas and prior Nevertheless, the spherical flame assumption has been
cycle effects have been eliminated through skip-fired op- retained for assessment of the effects of cyclic variation in
eration. These experiments provided well quantified and turbulence in the simulation of the production engine; for
well controlled initial conditions, such as charge pressure, its simplicity, transparency and the lack of a viable proven
temperature and composition at the moment of inlet port alternative.
closure. This enabled omission of details of the engine
breathing in simulations of this engine, placing empha- The flame propagation model employed in the current
sis on the importance of burning rate variation on ob- study is a three zone entrainment and burn-up model
served pattern of ccv on measured parameters. The sim- based on the ideas of Blizard and Keck [34]. Fresh mix-
ulations have been performed with the quasi-dimensional ture is entrained into the leading edge of the flame at a
thermodynamic computer code “LUSIE”, an acronym for rate proportional to a turbulent burning velocity ute and
Leeds University Spark Ignition Engine, which describes the area of the (entrainment) flame front Af :
the closed part of the engine cycle [28, 29].
dme
= ρu Af ute (2)
Simulations of the closed part of the cycle in a skip-fired dt
engine do not cover all the possible factors which can af- where me is the mass of fresh gas entrained into the flame
fect ccv in a typical production engine. The second stage brush and ρu is the fresh gas density. The flame sur-
of this study therefore concerns application of the model face area Af is found from a look-up table generated for
to a multi-cylinder modern engine. Crucial for this applica- a given chamber geometry and spark position assuming
tion is reliable modelling of the engine breathing, this fa- that the flame shape is a truncated sphere, as discussed
cility is provided by “GT-Power”; a commercially available above. The expression for the entrainment turbulent burn-
engine modelling suite developed by Gamma Technolo- ing velocity ute includes a flame development term derived
gies Inc. The LUSIE and GT-Power computer codes have by Lipatnikov and Chomiak [35, 36] by analogy with the
been interfaced into a hybrid GT-LU code in which LUSIE Taylor theory of diffusion from a point source:
combustion routines are called from within GT-power. De- 
 1/2
scription of this hybrid code and results of simulations ob- τ
t
ute = ul + ut0 1 + exp −
− 1 f (3)
tained with it can be found in the work of Hattrell et al. t τ
[30].
where ul is the laminar flame speed, t is time elapsed from
ignition, τ
= 0.55τt is a turbulent time scale, τt = L/u
is
DESCRIPTION OF THE MODEL
the turbulence integral timescale and f is a function which
accounts for the decrease in the flame propagation veloc-
Within LUSIE, the cylinder charge is divided into two
ity as the flame approaches the cylinder wall. From the
zones, burnt and unburnt, each with different tempera-
great number of published models for the fully developed
tures and chemical compositions but of equal pressure.
flame speed ut0 , the one selected for the current study is
Mixture composition is considered to be fixed and homo-
the expression originally proposed by Zimont [37]:
geneous in the unburned zone, converted by combustion
1/2 −1/4
into a mixture of species in chemical and thermodynam- ut0 = Au
· Da−1/4 = Au
3/4 ul κ L1/4 (4)
ical equilibrium in the burnt zone. The temperatures of
both burnt and unburnt zones are considered spatially uni- which has been shown [38] to have the particular merit of
form. Flame geometry is assumed to be spherical, trun- correctly predicting the effect of pressure on the turbulent
cated at contact with the chamber walls and centred on burning velocity. In Eq. 4, A is a model constant, Da is the
the spark gap. Flame centroid movement caused by bulk Damköhler number, κ is the molecular thermal diffusivity
gas motion is assumed to be small and is neglected. For of the fresh gas calculated using the method described
the single cylinder engine, described later, these assump- by Bird et al. [39] and L is the turbulence integral length-
tions have been shown to be valid for the whole duration of scale.
the combustion event [28, 31]. For the production engine,
these assumptions are more questionable. The impact of The entrained mixture is converted to burnt gas at a rate:
the spherical flame assumption on the computed ccv am- dmb me − mb
plitude has been studied by Shen et al. [12], who found = (5)
dt τb
that the effects of an oblate flame can be subsumed by
variations in the burning rate. Experimentally, flame shape L
τb = Cτ b (6)
deviation from a sphere has been studied by Witze [32]; ul
he concluded that, for an engine with a strong swirl flow where mb is the mass of burnt gas and Cτ b is a second
and off-centre ignition, a sphere approximates the flame model constant. The laminar burning velocity ul used in
shape well for flame radii less than one-third of the equiv- the above expressions is calculated using the correlation
alent radius based on the flame volume. This conclusion of Metghalachi and Keck [40].
is supported by the results of Murad [33]. For a simpli-
fied pent-roof combustion chamber engine having a bulk The entrainment rate is very sensitive to the rms turbulent
flow, he showed that the flame remained close to spher- velocity u
. The turbulent rms velocity u
for a given i-th

4
 Downloaded from SAE International by University of Birmingham, Sunday, August 19, 2018

cycle is taken here as the sum of a random variation and where t
= t/τt is the dimensionless time elapsed since
an average crank-angle dependent value u
m (Θ) defined ignition.
from the usual steady-state analysis. This random varia-
tion is introduced using K(i), a random variable which is Heat transfer to the cylinder walls and piston crown was
updated between cycles. The distribution of K(i) is as- calculated using the standard Woschni model [44]. The
sumed to be Gaussian with zero mean and standard de- same look-up table which was used for the calculation
viation σu
proportional to u
m when the piston is at the top of the flame surface area in Eq. 2 was used for the cal-
dead centre position: culation of the wetted contact areas. Though often ne-
glected in other works, the amount of charge mass escap-
u
(Θ, i) = u
m (Θ) + K(i)u
m (Θ = 0) (7)
ing the main combustion event through flow into the piston
It is well established [41] that turbulence decays near to top-land crevice and the inter-ring spaces (blow-by) have
TDC and the above approach allows this to be taken into been found to be non-negligible. These flows were calcu-
account, however, the rate of decay is not varied cycle- lated assuming isentropic flow [1] through the ring gaps,
to-cycle. Choice of the standard deviation of K in Eq. 7 the areas of which were measured in the cold engine with
is described later. Within the framework of the thermody- an allowance subsequently made for the thermal expan-
namic approach, the turbulent velocity is characterised in sion of the materials.
terms of a single rms value which is averaged over the en-
tire combustion chamber volume. No allowance is made EXPERIMENTAL FACILITIES
for any spatial variation of the rms velocity and this is one
of the limitations of the present approach. Two engines were used to provide experimental data,
a single cylinder ported research engine and a modern
Both Eqs. 4 and 6 require as an input the integral length poppet-valve production engine. The single cylinder en-
scale of turbulence within the combustion chamber. For gine, although lacking some of the features of a typical
the disc-shaped engine this was assumed to be spatially car power plant, brings greater control of other variables
uniform and was calculated following the work of Fraser such as charge composition, head and barrel tempera-
et al. [42] who found that in a disc geometry combustion tures, as well as affording exceptional optical access. This
chamber the integral scale was well approximated by: allowed experiments under conditions where some of the
L = 0.2hc (Θ) (8) previously mentioned causes of ccv were substantially re-
duced or eliminated. This is contrasted with a produc-
where hc is the clearance height at a particular crank an- tion engine in which all the causes of ccv are active and
gle Θ. against which the trends predicted by the model could be
comprehensively evaluated.
The later stages of the combustion period are charac-
terised by a rate of combustion which decreases as the
flame approaches the walls. To describe this decrease SINGLE CYLINDER ENGINE A single-cylinder ported
in burning rate, it has been assumed that the approach- research engine known as LUPOE1-D (acronym for
ing wall cuts the self-similar flame brush so that the mass Leeds University Ported Optical Engine—Disc chamber),
burning rate is decreased compared with a planar uncon- having complete overhead optical access, was developed
strained flame. It has been observed [38] that the spatial at Leeds University from a commercial JLO L372 two-
profile of the mean flame progress variable c̄ for a flame stroke engine specifically for the study of combustion in
with a brush thickness δt located at x0 collapses onto a SI engines [28, 45]. The engine featured a simple disc-
universal curve under a wide range of conditions and is shaped combustion chamber with breathing through two
well described by: diametrically opposed intake ports and a single exhaust


1 x − x0 port inclined at a right angle to the two intakes. The ports
c̄(x) = 1 − erf (9) were located on the cylinder barrel sidewall and were cov-
2 δt (t)
ered and uncovered at set points in the cycle by the piston
For a planar flame with a wall located at xw , the rate of allowing breathing to take place.
combustion is therefore decreased compared with an un-
constrained flame by the following factor: Intake air was supplied under pressure and exhaust gases
 xw   
 ∂c̄(x
) 
xw − x
were expelled at just below atmospheric pressure in-
f=  
 ∂x
 dx = erf δt (t)
= erf(Ra) (10) stead of using the conventional two stroke pressurised
−∞
crankcase induction system. The engine was skip fired to
where Ra = (Rb − rf l )/δt is the dimensionless flame ra- ensure that all residual gases were well scavenged from
dius. the cylinder between firing cycles. This operation mode
ensured a negligible contribution to ccv from any inho-
For the purpose of calculating f it has been assumed that mogeneity arising from imperfect mixing of fresh charge
the growth of the turbulent flame brush thickness is mainly with residual gas. The geometry of the ports was de-
governed by the turbulent diffusion law [43]: signed to impart minimum tumble and swirl to the cylinder

1/2


1 −t

charge. Absence of swirl and tumble was confirmed with
δt (t) = u τt 2t 1 −
1 − e (11) LDA measurements [46]. For the selected operating con-
t
5
 Downloaded from SAE International by University of Birmingham, Sunday, August 19, 2018

Table 1: LUPOE1-D engine geometry

Bore [mm] 80
Stroke [mm] 74
Effective stroke [mm] 53
Connecting-rod length [mm] 148
Effective compression ratio 7.6
Inlet port open/close [CA° b/aTDC] 115.7
Exhaust port open/close [CA° b/aTDC] 108.5

PRODUCTION ENGINE A high power output multi-

cylinder four-valve gasoline direct injection engine, re-
ferred to as the “production” engine henceforth, with con-
tinuously variable valve timing on both intake and exhaust,
was used to provide experimental data from a radically
different type of engine in order to assess the generality
of the studied model. The combustion chamber was of a
Figure 1: Pressure traces from 151 cycles recorded for pent-roof design incorporating wall guided direct injection.
LUPOE1-D. Under full load conditions the specific power output of the
engine was approximately 75 kW/l. The engine was in-
strumented with cylinder, runner and manifold pressure
ditions the turbulence proved very nearly homogeneous
transducers. A pressure pegging procedure, similar to
and decayed linearly with crank angle according to:
that used on the single cylinder LUPOE1-D, was applied
to the signals from the cylinder and runner pressure trans-
u
m (Θ) = C1 + C2 S p + C3 Θ (12) ducers. Thermocouples were used to measure the intake
air and water jacket temperatures. The rate of air con-
where C1 , C2 and C3 are constants with values 0.47 m/s, sumption was derived from the rate of fuel consumption
0.6, and 0.024 m/s° respectively, S p is the mean piston and knowledge of the equivalence ratio, monitored us-
speed and Θ is the crank angle in degrees aTDC. The ing a gas analyser, to allow comparison of predicted and
above equation gives the mean values of the rms velocity measured volumetric efficiencies. No flowfield/turbulence
used in LUPOE1-D modelling. measurements were available for this engine.

Cylinder pressure was recorded using two pressure trans- The engine was operated at Mahle at a low speed
ducers; a piezoelectric dynamic pressure transducer, (2000 rpm) low load (2.5 bar IMEP) condition with fuel
mounted in the cylinder head which remained exposed injection occurring early during the intake stroke; such
for the entire cycle, and a piezoresistive absolute pres- that the charge could be considered effectively homo-
sure transducer mounted part way down the barrel (and geneous at the moment of ignition. Inlet valve open-
therefore isolated from contact with the hot combustion ing/closing occurred at 324/592° aTDC and exhaust valve
gases close to TDC). The absolute pressure transducer
was used to provide a reference pressure against which
the signal from the dynamic transducer, prone to drift over
long periods of time and errors caused by thermal shock,
could be pegged.

The cylinder head and intake ports were heated and ther-
mostatically maintained at a constant temperature such
that the head and intake temperatures remained at 70° C.
The engine was run on a stoichiometric mixture of iso-
octane and air for the purpose of the current study. In
order to ensure a homogeneous charge, fuel was added
to the supplied air well before the intake port. The in-
take ducts were heated along their length to ensure full
vaporisation and mixing. This study uses 151 cycles of
LUPOE1-D data recorded at 1500 rpm with an ignition
timing of 20° bTDC. Cyclic variability in these data can be
clearly seen in the pressure traces shown in Fig. 1. Other
engine particulars are given in Table 1.
Figure 2: Pressure traces from 300 consecutive cycles
recorded for the production engine.

6
 Downloaded from SAE International by University of Birmingham, Sunday, August 19, 2018

opening/closing occurred at 106/370° aTDC. Ignition tim- lowed the ideas of Poulos and Heywood [47], and Morel
ing was set at 29° bTDC and the engine was run on a stoi- and Keribar [48], in that the turbulent kinetic energy and
chiometric mixture of gasoline and air. At these nominally its dissipation were calculated taking into account the ki-
fixed conditions, the magnitude of cyclic variability, see netic energy of the intake and exhaust flows as well as
Fig. 2, proved quite comparable to that of the LUPOE1-D turbulence generation by the motion of the piston. The
engine. laminar burning velocity values for gasoline required by
the model were taken as those of indolene, using the ex-
MEAN CYCLE MODELLING pression suggested in Reference [40].

The combustion model presented above contains two ad- An obvious prerequisite for accurate combustion simula-
justable constants, see Eqs. 4 and 6. The first step in
modelling was the determination of their values so that
the predicted pressure—crank angle curve matched well
an “average”, or “middle”, measured cycle at one refer-
ence condition. Data from the single cylinder engine was
used to provide this middle cycle, as this dataset included
simultaneous measurements of cylinder pressure and en-
trainment flame radius; thus providing two measurements
against which the two unknown constants could be cali-
brated.

Defining a suitable middle pressure trace proved prob-

lematic, as taking an ensemble average of the dataset
shown in Fig. 1 gave a pressure record which did not
closely match any real cycle from the dataset. Ensem-
ble averaging in particular resulted in a flattening of the
pressure curve in the region of peak pressure, near the
end of the combustion period. Obviously, if the distribu-
tion of pressures at any given crank angle were Gaussian
then an ensemble average curve would be close to a cy-
cle which could be observed. Analysis of the experimental
dataset showed that this was not the case and, therefore,
the middle pressure trace had to be taken from the cy-
cle which had a peak pressure closest to the mean peak Figure 3: A comparison between the experiment middle
pressure of the entire data set. The model constant val- cycle and the model for LUPOE1-D.
ues were then adjusted to produce the minimum deviation
between predictions and measurements for both pressure
and flame radius curves for this middle cycle. The values
so found remained unchanged for all subsequent calcu-
lations, including those performed on the multi-cylinder
production engine. The level of agreement obtained for
this middle cycle is illustrated in Fig. 3. A similar level of
agreement, without parameter adjustment, pertained over
a wide range of engine speeds, mixture strengths and ig-
nition timings [28]. The burnt mass fraction shown was
found using a reverse thermodynamic analysis employ-
ing the same sub-models as employed in LUSIE. Differ-
ences in the later stages of the cycle seen in Fig. 3 can
be attributed to thermal drift in the pressure transducer
mounted in the cylinder head. The pressure recorded by
the transducer mounted in the barrel wall should be less
susceptible to thermal drift. The difference between the
value recorded by this pressure transducer and the pre-
dicted pressure later in the cycle proved very small.

Modelling of the mean cycle of the production engine was

performed using the same values of the model constants.
Because data were not available for the rms turbulent ve-
locity or for the scales of turbulence, use was made of Figure 4: A comparison between the experiment middle
the turbulence model built into GT-Power. This model fol- cycle and the model for the production engine.

7
 Downloaded from SAE International by University of Birmingham, Sunday, August 19, 2018

tion in a multi-cylinder environment is adequate represen-

tation of the engine breathing characteristics. This has
been verified by a good agreement between the time-
resolved model predictions and measured data for both
volumetric efficiency and pressures in the manifold run-
ners. A comparison between the predicted and measured
cylinder pressure traces and burnt mass fraction curves
for an experimental middle cycle from the production en-
gine and the GT-LU model is shown in Fig. 4. The burnt
mass fraction for the production engine was calculated us-
ing the reverse analysis available with GT-Power. It is dif-
ficult to determine with any certainty the exact cause of
the small error seen later in the cycle, any error in the
measured pressure signal should be negligible as this en-
gine was equipped with a water cooled pressure trans-
ducer. It is possible that an error in the model prediction
was caused by inaccurate prediction of the final amount
of mass to be burnt or the magnitude of heat transfer. Figure 5: Experimental and simulated ensemble fast mid-
dle slow cycles for LUPOE1-D.
CYCLIC VARIABILITY MODELLING AND DISCUSSION

SINGLE CYLINDER ENGINE As shown by the pres-

sure traces in Fig. 1, despite the measures taken to elim-
inate cyclic variation in parameters other than turbulence,
there was a large amount of cyclic variability. The pres-
sure traces can be seen to begin to deviate from the mean
at approximately 13° CA after ignition; before this point
the amount of burnt mass is too small to cause a notice-
able pressure rise. Combustion was completely finished
in most cycles by 45° CA after ignition. Between these
two points, considerable spread can be seen between the
fastest and slowest cycles.

For cyclic variability modelling, the only change to the

LUSIE code was to the rms turbulent velocity; which was
calculated according to Eq. 7 rather than using a deter-
ministic mean value. For the implementation of this equa-
tion, a new value for K was randomly selected from a pre- Figure 6: Coefficient of variation in IMEP for simulations of
generated list of random Gaussian numbers at the start of the production engine and LUPOE1-D with various values
each cycle. Three different series of 500 cycles were sim- of u
standard deviation.
ulated; with σu
values of 5, 12.5 and 20%. These three
values were chosen somewhat arbitrarily to give a good
spread in the variation in simulated cycles, but were not unexpectedly, increasing/decreasing the standard devia-
too different from the findings of Al-Khishali et al. [10]. tion of u
increases/decreased the spread of the pressure
curves.
Calculated and experimental representative fast, middle
and slow cycles are shown in Fig. 5. Fast and slow cycles The effect of change in standard deviation in u
on the
were chosen, experimental and calculated alike, as the magnitude of cycle-to-cycle variations in IMEP, as quan-
cycles with peak pressure nearest to the mean peak pres- tified by the coefficient of variation (COV) is shown in
sure plus or minus two standard deviations, respectively. Fig. 6. Dotted lines in this figure show the measured COV
The middle cycle was selected in a similar way, as having of IMEP and the corresponding best fit in terms of σu
.
a peak pressure equal to the mean peak pressure for the In common with the observations for pressure, a stan-
entire data set. The section describing mean cycle mod- dard deviation of u
of 12.5% gives a reasonable corre-
elling presents a discussion on why this methodology was spondence between the measured and predicted COV of
adopted. As expected, increasing the spread in u
leads IMEP. It should be noted that as the LUSIE code simulates
to a larger variation in pressure between the fast and slow only the closed part of the cycle, values of IMEP derived
predicted cycles (whilst having very little effect on the mid- from simulated data used idealised intake and exhaust
dle cycles). A spread of u
values of 12.5% of the mean pressures. Given the skip-fired nature of the LUPOE1-D
value at TDC resulted in a magnitude of cyclic variability engine, it is unlikely that any significant error is introduced
in pressure development quite close to that observed. Not

8
 Downloaded from SAE International by University of Birmingham, Sunday, August 19, 2018

by this assumption.

Displayed in Figs. 7, 8 and 9 are measured pressure prob-

ability densities at different crank angles for the data il-
lustrated in Fig. 1, compared with predictions obtained
with different standard deviations of u
. Variation of u
by
12.5%, as has already been noted, produced a spread
in pressures comparable to that observed. The calcu-
lated distribution of pressures at TDC and 10° aTDC can
be seen to be quite symmetrical, without any noticeable
skewness; the same can be said of the experimental
points, although the number of sampled cycles was insuf-
ficient for precise definition of the distribution. Later in the
cycle, at 20° aTDC, the distribution of pressures shows a
definite skewness with a small number of cycles exhibit-
ing rather low pressures. This behaviour is attributed to
the fact that, by this point (in a large number of cycles),
combustion is close to completion. In the small number Figure 7: Experimental and simulated probability density
of cycles with slow combustion however, pressure is still of pressure at TDC for LUPOE1-D.
rising and this explains the skewness of the distributions
at 20° aTDC, Fig. 9.

Probability densities of peak cylinder pressure, and its an-

gle of occurrence, for the LUPOE1-D data set shown in
Fig. 1 are given in Figs. 10 and 11. Displayed in the same
figures are model predicted probability densities for differ-
ent standard deviations of u
. As before, a 12.5% stan-
dard deviation of u
yielded a good agreement between
the calculated and observed data. The calculated distri-
butions show a noticeable skewness not exhibited by the
measured distributions; as previously noted, the number
of measured cycles may not have been sufficient to reveal
the finer details of the experimental distribution profile.

The cycles which cause the skewness in Figs. 10 and 11

are those cycles where the peak pressure is low and oc-
curs late in the cycle, in other words those cycles where
Figure 8: Experimental and simulated probability density
the rate of combustion is much lower than the mean. The
of pressure at 10° aTDC for LUPOE1-D.
origin of these slow cycles is in their low K(i), and hence
low u
. In order to understand the significance of this low
u
and why the distributions shown in Figs. 10 and 11 are
skewed even though the distribution of u
is symmetrical
it is necessary first to understand how the magnitude of
the pressure and temperature rise early in the cycle can
feedback to give a rapid rate of combustion later in the
cycle.

In a typical cycle, the rate of conversion of fresh gas to

burnt products is characterised by three main phases:
An initial phase immediately after the spark during which
the cylinder pressure remains close to the motoring pres-
sure and the rate of production of burnt gas is low; a
main phase where the burning rate increases as the cylin-
der pressure begins to rise above the motoring pressure;
and a termination phase where flame-wall interactions be-
come important and combustion eventually ceases. The
increase in the rate of combustion between the initial and
Figure 9: Experimental and simulated probability density
main phases is driven by the relationship between the
of pressure at 20° aTDC for LUPOE1-D.
laminar burning velocity and the unburnt gas temperature.
Immediately after ignition, the pressure and temperature

9
 Downloaded from SAE International by University of Birmingham, Sunday, August 19, 2018

of the unburnt gas begin to rise due to combustion. Ini-

tially the rate of this rise is slow but as more gas is burnt
the pressure and temperature rise becomes large enough
to increase the laminar burning velocity. This, in turn,
reduces the residence time of the fresh mixture behind
the flame front, see Eq. 6, in other words entrained gas
spends less time behind the flame front before it is con-
verted to burnt gas. As more mass is burnt and the flame
area becomes larger, this effect becomes stronger, with
the elevated laminar burning velocity causing an increase
in the pressure and temperature of the unburnt mixture
causing an increase in the laminar burning velocity and
so on and so on. In this way, the rate of combustion later
in the cycle is elevated significantly above than its initial
level with the time taken for this increase to occur being
strongly dependent on the initial entrainment and burn up
rates.
Figure 10: Experimental and simulated peak pressure
probability density for LUPOE1-D. As the entrainment rate is a strong function of u
, cycles
with a low u
also exhibit a low rate of entrainment of fresh
gas compared with the mean. Ultimately these cycles
burn less mass early in the combustion period resulting
in a correspondingly lower temperature, and hence lami-
nar burning velocity, later in the cycle. This effect is com-
pounded by the fact that by the time these slow cycles do
burn enough mass to cause a significant pressure rise the
piston has moved further past TDC than for typical cycles
and the chamber volume is increasing. This on its own
though does not explain the skewness in the distributions
in Figs. 10 and 11.

Cycles in which K(i) is much higher than the mean do

not exhibit a correspondingly higher rate of combustion
as there comes a point where the laminar burning veloc-
ity is large enough that entrained fresh gas is converted
to burnt gas almost immediately. Any further increase in
laminar burning velocity serves only to increase the en-
Figure 11: Experimental and simulated probability density trainment velocity, which is less sensitive to changes in
of crank angle at peak pressure for LUPOE1-D. ul than changes in u
, see Eq. 4. In this way the feed-
back effect is non-linear and it is this non-linearity which
causes the observed skewness in the data in Figs. 10
and 11. This implies that the cycles with the poorest com-
bustion performance in a real engine are those in which
the early combustion performance, for whatever reason is
poor. This effect has been noted previously by other au-
thors, e.g. Peters and Borman [7], Aleiferis et al. [18].

Set out in Fig. 12 are data for experimental in-cylinder

peak pressure and the crank angle of its occurrence, to-
gether with corresponding simulation results. The exper-
imental data show a degree of scatter in the peak pres-
sure corresponding to any given crank angle (∼11% at the
middle cycle peak pressure crank angle of ∼18° CA). This
scatter (not seen in the simulations) is to a certain extent
inevitable in any experiment and may reflect slight cyclic
variations in fuelling, air supply, ignition (in spite of the
Figure 12: Measured and calculated peak pressure plot- measures to eliminate these for the idealised engine con-
ted against crank angle at peak pressure for LUPOE1-D. ditions) as well as experimental measurement errors. The
Standard deviation in model u
of 12.5%. observed difference in the gradient of the experimental
and modelled peak pressure/crank angle curves is more

10
 Downloaded from SAE International by University of Birmingham, Sunday, August 19, 2018

interesting. This may indicate that the 12.5% value for

σu
is perhaps not quite optimal (although the modelled
gradient did not change significantly with varying σu
) or
may reflect physical differences. In the model, turbulence
is assumed spatially homogenous – hence initially fast
burn cycles will be stay fast and vice-versa for slow cy-
cles. This may not be true in the experiment, where for a
given nominal mean u
spatial variation in local turbulence
intensity may result in relatively fast early burn and slow
later burn (and vice-versa) with a consequent flatter gra-
dient in Fig. 12. Alternatively, the differing gradients could
be a result of a systematic error in the model where for cy-
cles with slow combustion the final amount of mass burnt
is too small. This area merits further investigation.

PRODUCTION ENGINE The cyclic variability present in

the production engine was simulated following the same Figure 13: Experimental and simulated fast middle slow
approach as used for the single cylinder LUPOE1-D, ex- cycles for the production engine.
cept that here the LUSIE code was substituted for GT-LU
to allow simulation of the gas exchange processes. The
simulation was first run for 50 cycles without any pertur-
bation to the value of u
predicted by GT-Power, allowing
the solution for the flow in the runners and manifolds to
converge. Subsequently five hundred cycles were simu-
lated where the value of u
, predicted by the model used
in GT-Power, was varied according to Eq. 7. Three series
of simulations were run with values of σu
of 6, 10 and
14%, these values having been chosen for similar rea-
sons as the ones used when simulating the single cylin-
der engine. As GT-Power includes full simulation of the
engine breathing processes, this approach captured vari-
ations in residual/recirculated exhaust gas and any prior
cycle “feedback” effects induced by the variations in the
burning rate.

Presented in Fig. 13 are representative fast, middle and

slow measured cycles, compared with calculated cycles Figure 14: Experimental and simulated peak pressure
for different standard deviations of u
. A standard de- probability density for the production engine.
viation of u
of 10% yielded a spread in pressure de-
velopment comparable to that observed in this engine.
This value is close to that inferred for the much sim-
pler LUPOE1-D research engine, however, this agree-
ment may be fortuitous as the absolute values of u
at
TDC were quite different for the two engines. The rela-
tionship between COV of IMEP and standard deviation of
u
, although of similar order, appears to be more sensi-
tive for the production engine than for the simple disc en-
gine, Fig. 6. This perhaps reflects the difference in flame
surface area versus mean flame radius for the two en-
gines [1], with greater flame propagation rate feedback (in
increased temperature and pressure) for the production
engine than with the more truncated flame area disc en-
gine.

Probability density distributions of peak pressure and the

angle at peak pressure are set out in Figs. 14 and 15 for
observed and simulated cycles. In common with the re- Figure 15: Experimental and simulated probability density
sults for the single cylinder engine shown in Figs. 7, 8 of crank angle at peak pressure for the production engine.
and 9, the predicted distributions show a clear asymmetry.

11
 Downloaded from SAE International by University of Birmingham, Sunday, August 19, 2018

A model standard deviation of u
of 10% proved sufficient

to replicate the observed spread in the data.

One of the effects of ccv in combustion in a non skip-fired

engine should be to introduce variations in the charge
composition cycle-to-cycle. The probability densities of
the proportion of trapped burnt gas at IVC, shown in
Fig. 16 demonstrate clearly that variability in combustion
does cause dispersion in the charge composition cycle-
to-cycle. This variability is caused by variation in exhaust
gas temperature (and hence density) as well as the ratio
of burnt to unburnt gas remaining in the cylinder cycle-to-
cycle, i.e. the completeness of combustion predicted by
the model. Previous studies have shown that variations
in the completeness of combustion can be significant in
a real engine [49, 50] but the consequent variability in
the amount of trapped burnt gas has very little effect on
combustion [12]. Despite the fact that the ratio of fresh Figure 17: Measured and calculated peak pressure plot-
to residual gas varied noticeably cycle-to-cycle, predicted ted against crank angle at peak pressure for the produc-
variations in other quantities were much smaller. For ex- tion engine. Standard deviation in model u
of 10%.
ample, for the most extreme case simulated where the
standard deviation of u
was 14% and the variability in
combustion was very large, see Fig. 13, the temperature
at ignition ranged only from 773 to 781 K and the laminar Given the thermodynamic nature of the model, fluid dy-
burning velocity at ignition ranged from 1.02 to 1.08 m/s. namic effects such as these will never be predicted. In
The predicted variation in other parameters such as vol- Fig. 17, the gradient of the peak pressure vs angle for its
umetric efficiency or air-fuel ratio were so small as to be occurrence can be seen to match the experimental trend
negligible. rather better than in the case of the single cylinder en-
gine. This might be a result of more consistent in-cylinder
flow and turbulence generation in the production pent-
roof geometry or the greater feedback effect associated
with the greater flame surface area vs mean flame radius
in faster initial burn cycles remaining faster through the
whole combustion event.

Even though the standard deviation of u
was selected

(from the three analysed in each case) to give approx-
imately the same COV of IMEP for the two engines,
greater scatter in experimental values at a given crank
angle of peak pressure (∼11% c.f. ∼20%) are evident for
the production engine, Fig. 17, vis a vis the single cylinder
engine, Fig. 12. This may in part reflect the greater sen-
sitivity to σu
noted for the production engine, Fig. 6, as
well as additional variation in mixture strength and resid-
ual concentration largely eliminated in the case of the sin-
gle cylinder engine.

Among the possible causes of variation in the peak pres-

Figure 16: Simulated probability density of burnt gas frac- sure occurring at a point in the cycle is variation induced
tion at inlet valve closure for the production engine. by the fuel induction system. To assess the relative im-
portance of this factor, another series of calculations were
Despite the inclusion of prior cycle effects in the produc- performed where u
was not perturbed but the mass of in-
tion engine simulation, points of predicted peak pressure jected fuel was varied from cycle to cycle around its mean
against crank angle at peak pressure, shown in Fig. 17, value. Again, for this variation, a Gaussian distribution
still essentially fell on a single line, with no discernible was assumed with zero mean and a standard deviation of
spread in peak pressures occurring at the same crank an- 5%. This value is quite arbitrary and was assumed only
gle. It is possible that the large valve overlap and high for a preliminary estimation however, it is of a comparable
levels of residual gas present for the production engine, magnitude to a measured amount of more than 8% [51].
coupled with poorer in-cylinder mixing or mixing variations It can be seen from Fig. 18 that the effect of this plausi-
from cycle-to-cycle, could contribute to the “scatter” exhib- ble variation in mixture strength is rather less than that of
ited by the experimental data for the production engine.

12
 Downloaded from SAE International by University of Birmingham, Sunday, August 19, 2018

ensemble averaged u
estimated on the basis of the

turbulence model within GT-Power, a similar level of
agreement pertained for representative mean cycles
for a production engine of rather different combustion
chamber geometry.
• The range of cyclic variation in in-cylinder pressure
derived parameters for the production engine could
similarly be reproduced assuming a variation in rms
turbulent velocity remarkably similar to that for the
single cylinder engine. Since the results may be a
function of the operating conditions, this agreement
might, to a certain extent, be fortuitous; requiring ad-
ditional tests at other conditions.
• The production engine proved more sensitive to vari-
ation in rms turbulent velocity, probably related to the
Figure 18: Measured and calculated peak pressure plot- steeper flame surface area to mean flame radius re-
ted against crank angle at peak pressure for the produc- lationship vis a vis the simple disc engine.
tion engine. Standard deviation in model equivalence ratio
• At a given crank angle for peak cylinder pressure,
of 5%.
experimental data scatter was rather greater for the
production engine, associated with the above effects
variation in u
even though it might contribute to the overall with superimposed additional cyclic variation in mix-
variation noted in cylinder pressure. ture strength and residual/recirculated exhaust gas
concentration.
CONCLUSIONS
• At the production engine condition tested, the ef-
fects of variation in rms velocity were dominant, vari-
Variations in the rms turbulent velocity experienced by a
ations in mixture strength (at typical measured levels)
growing flame are unavoidable because the entire dura-
proved secondary and effects of variations in prior cy-
tion of a deflagration in an SI engine does not exceed
cle generated residuals proved marginal.
a more than a few integral time-scales. An attempt to
simulate the effect of these variations has been made by
perturbing, according to a Gaussian distribution, the level ACKNOWLEDGEMENTS
of turbulence experienced by a flame within a thermody-
namic SI engine model. This procedure has been applied The financial support of MAHLE Powertrain for this work
to two quite different engines; one a single cylinder disc- is gratefully acknowledged. Parts of the LUSIE engine
shaped chamber engine in which variation in all parame- simulation software were developed and the LUPOE1-D
ters other than turbulence had largely been eliminated, validation data generated under the EU funded GET-CO2
the other a representative production engine. The follow- Project, Project number: GRDA-2000-25618. Discus-
ing conclusions may be drawn: sions with Prof. D. Bradley and Dr. M. Lawes have been
invaluable.
• Adopting measured ensemble-averaged values of
rms turbulent velocity, a satisfactory fit to pressure REFERENCES
and simultaneous flame radius data for the single [1] J. Heywood. Internal Combustion Engine Funda-
cylinder engine could be obtained over a wide range mentals. McGraw Hill, N.Y., 2nd edition, 1988.
of engine operating conditions. [2] J. Pan, C.G.W. Sheppard, A. Tindall, M. Berzins, S.V.
• The coefficient of variation in indicated mean effec- Pennington, and J.M. Ware. End gas inhomogeneity,
tive pressure has been shown sensitive to the stan- autoignition and knock. In SAE Technical Paper Se-
dard deviation in rms turbulent velocity and the ex- ries, number 982616. 1998.
perimentally noted variation in a wide range of engine [3] N. Ozdor, M. Dulger, and E. Sher. Cyclic variability
pressure related parameters could be well produced in spark ignition engines: A literature survey. In SAE
with an appropriate variation in rms turbulent velocity. Technical Paper Series, number 940987. 1994.
• Linear variation in rms turbulent velocity induces non- [4] P.G. Aleiferis, Y. Hardalupas, A.M.P.K. Taylor, K. Ishii,
linear changes in pressure development associated and Y. Urata. Flame chemiluminescence studies of
with temperature/pressure feedback effect on flame cyclic combustion variations and air-to-fuel ratio of
propagation. the reacting mixture in a lean-burn stratified-charge
• With all model parameters held constant and a mean spark-ignition engine. Comb. Flame, 136:72–90,
2004.

13
 Downloaded from SAE International by University of Birmingham, Sunday, August 19, 2018

[5] W. Dai, N. Trigui, and Y. Lu. Modeling of cyclic vari- [19] P.G. Aleiferis, A.M.P.K. Taylor, K. Ishii, and Y. Urata.
ations in spark-ignition engine. In SAE Technical Pa- The relative effects of fuel concentration, residual-
per Series, number 2000-01-2036. 2000. gas fraction, gas motion, spark energy and heat
losses to the electrodes on flame-kernel develop-
[6] S.O. Bade Shrestha and G.A. Karim. Considering the
ment in a lean-burn spark ignition engine. Proc. In-
effects of cyclic variations when modeling the perfor-
stn. Mech. Engrs. D: J. Automobile Eng., 218:411–
mance of a spark ignition engine. In SAE Technical
425, 2004.
Paper Series, number 2001-01-3600. 2001.
[20] E. Zervas. Correlations between cycle-to-cycle vari-
[7] B.D. Peters and G.L. Borman. Cyclic variations and
ations and combustion parameters of a spark ignition
average burning rates in a spark-ignition engine. In
engine. Appl. Thermal Eng., 24:2073–2081, 2004.
SAE Technical Paper Series, number 700064. 1970.
[8] C.R. Stone, A.G. Brown, and P. Beckwith. Cycle-by- [21] M. Fogleman, J. Lumley, D. Rempfer, and D. Ha-
cycle variations in spark ignition engine combustion - worth. Application of the proper orthogonal decom-
part II: Modelling of flame kernel displacements as a position to datasets of internal combustion engine
cause of cycle-by-cycle variations. In SAE Technical flows. J. Turbulence, 5(23):1–18, 2004.
Paper Series, number 960613. 1996. [22] T. Urushihara, T. Murayama, Y. Takagi, and K.-H.
[9] W.R. Mickelsen and N.E. Ernstein. Growth rates of Lee. Turbulence and cycle-by-cycle variation of
turbulent free flames. In IVth Symp. (Int.) on Comb., mean velocity generated by swirl and tumble flow and
pages 325–333, Baltimore, 1953. Williams & Wilkins. their effects on combustion. In SAE Technical Paper
Series, number 950813. 1995.
[10] K.J. Al-Khishali, P.M. Boston, D. Bradley, M. Lawes,
and M.J. Pegg. The influence of fluctuations in turbu- [23] D.L. Reuss. Cyclic variability of large-scale turbu-
lenceupon fluctuations in turbulent burning velocity. lent structures in directed and undirected IC engine
Proc. Instn. Mech. Engrs., (C49/83):175–180, 1983. flows. In SAE Technical Paper Series, number 2000-
01-0246. 2000.
[11] K. Liu, A.A. Burluka, R. Woolley, C.G.W. Sheppard,
and A.N. Lipatnikov. On the development of the tur- [24] W.-C. Choi and Y.G. Guezennec. Measurement of
bulent mass burning rate. In 3rd Asia-Pacific Confer- cycle-to-cycle variations and cycle-resolved turbu-
ence on Combustion, pages 303–306, Seoul, South lence in an IC engine using a 3-D particle tracking
Korea, 2001. Seoul Nat. Univ. velocimetry. JSME Int. J., 41:991–1003, 1998.
[12] H. Shen, P.C. Hinze, and J.B. Heywood. A study of [25] F. Ma, H. Shen, C. Liu, D. Wu, G. Li, and D. Jiang.
cycle-to-cycle variations in SI engines using a modi- The importance of turbulence and initial flame kernel
fied quasi-dimensional model. In SAE Technical Pa- centre position on the cyclic combustion variations
per Series, number 961187, 1996. for spark-ignition engine. In SAE Technical Paper Se-
ries, number 961969. 1996.
[13] A.N. Lipatnikov and J. Chomiak. Randomness of
flame kernel development in turbulent gas mixture. In [26] D.D. Brehob and C.E. Newman. Monté-carlo simu-
SAE Technical Paper Series, number 982617. 1998. lation of cycle by cycle variability. In SAE Technical
[14] P.C. Hinze and P.C. Miles. Quantitative measure- Paper Series, number 922165. 1992.
ments of residual and fresh charge mixing in a mod- [27] A.G. Brown, C.R. Stone, and P. Beckwith. Cycle-by-
ern SI engine using spontaneous raman scaterring. cycle variations in spark ignition engine combustion -
In SAE Technical Paper Series, number 1999-01- part I: Flame speed and combustion measurements
1106. 1999. and a simplified turbulent combustion model. In SAE
[15] P.G. Hill. Cyclic variation and turbulence structure Technical Paper Series, number 960612. 1996.
in spark-ignition engines. Comb. Flame, 72:73–89, [28] E. Abdi Aghdam. Improvement and Validation of a
1988. Thermodynamic S.I. Engine Simulation Code. PhD
[16] P.G. Hill and D. Zhang. The effects of swirl and tum- thesis, School of Mechanical Engineering, The Uni-
ble on combustion in spark-ignition engines. Progr. versity of Leeds, 2003.
Energy Comb. Sci., 20:373–429, 1994.
[29] C.G.W. Sheppard and S. Merdjani. Gasoline engine
[17] J. Heywood. Combustion and its modelling in spark- cycle simulation using the Leeds turbulent burning
ignition engines. In COMODIA Int. Symp., pages 1– velocity correlations. In SAE Technical Paper Series,
15, 1994. number 932640, 1993.

[18] P.G. Aleiferis, A.M.P.K. Taylor, J.H. Whitelaw, K. Ishii, [30] T. Hattrell, C.G.W. Sheppard, A.A. Burluka,
and Y. Urata. Cyclic variations of initial flame kernel J. Neumeister, and A. Cairns. Burn rate impli-
growth in a Honda VTEC-E lean-burn spark-ignition cations of alternative knock reduction strategies
engine. In SAE Technical Paper Series, number turbocharged SI engines. In SAE Technical Paper
2000-01-1207. 2000. Series, number 2006-01-1110. 2006.

14
 Downloaded from SAE International by University of Birmingham, Sunday, August 19, 2018

[31] A. Cairns. Turbulent flame development in a spark [45] A. Cairns and C.G.W. Sheppard. Cyclically resolved
ignition engine. PhD thesis, School of Mechanical simultaneous flame and flow imaging in a SI engine.
Engineering, The University of Leeds, 2001. In SAE Technical Paper Series, number 2000-01-
2832. 2000.
[32] P.O. Witze. Interpretation of head-gasket ionization-
probe measurements using a two-zone spherical [46] K. Atashkari. Experimental Study of Flow and Tur-
flame model. In Int. Symp. COMODIA ’94, pages bulence in a V-flame Burner and a SI Engine. PhD
453–458. 1994. thesis, School of Mechanical Engineering, The Uni-
versity of Leeds, 1997.
[33] A.E.M.A. Murad. Flow and combustion in disc and
pent roof SI engines. PhD thesis, School of Mechan- [47] S.G. Poulos and J.B. Heywood. The effect of cham-
ical Engineering, The University of Leeds, 2006. ber geometry on spark-ignition engine combustion.
In SAE Technical Paper Series, number 830334.
[34] N.C. Blizard and J.C. Keck. Experimental and theo- 1983.
retical investigation of turbulent burning model for in-
ternal combustion engines. In SAE Technical Paper [48] T. Morel and R. Keribar. A model for predicting spa-
Series, number 740191. 1974. tially and time resolved convective heat transfer in
bowl-in-piston combustion chambers. In SAE Tech-
[35] A.N. Lipatnikov and J. Chomiak. A simple model of
nical Paper Series, number 850204. 1985.
unsteady turbulent flame propagation. In SAE Tech-
nical Paper Series, number 972993. 1997. [49] J.K. Ball, R.R. Raine, and Stone C.R. Combustion
analysis and cycle-by-cycle variations in spark ig-
[36] A.N. Lipatnikov and J. Chomiak. Transient and geo-
nition engine combustion part 1: an evaluation of
metrical effects in expanding turbulent flames. Comb.
combustion analysis routines by reference to model
Sci. and Tech., 154:75–117, 2000.
data. Proc. Instn. Mech. Engrs. Part D, 212:381–399,
[37] V.L Zimont. The theory of the turbulent combustion 1998.
at high reynolds numbers. Fizika Gorenia i Vzryva,
[50] J.K. Ball, R.R. Raine, and Stone C.R. Combustion
15:23–28, 1979. (English translation: Combustion,
analysis and cycle-by-cycle variations in spark igni-
Explosion and Shock Waves, vol. 15, pp. 305-311,
tion engine combustion part 2: a new parameter for
1979).
completeness of combustion and its use in modelling
[38] A.N. Lipatnikov and J. Chomiak. Turbulent flame cycle-by-cycle variations in combustion. Proc. Instn.
speed and thickness: phenomenology, evaluation Mech. Engrs. Part D, 212:507–523, 1998.
and application in multi-dimensional simulations.
[51] J.H. Wright and J.A. Drallmeier. Cyclic variability of
Progr. Energy Comb. Sci,, 28:1–74, 2002.
pulsed spray vapor fields. Exp. Fluids, 25:329–336,
[39] R.B. Bird, W.E. Stewart, and E.N. Lightfoot. Trans- 1998.
port phenomena. J. Wiley & Sons, Inc., N.Y., 1960.
[40] M. Metghalachi and J.C. Keck. Burning velocities
of mixtures of air with methanol, iso-octane and in-
dolene at high pressure and temperature. Comb.
Flame, 48:191–210, 1982.
[41] E.S. Semenov. Studies of turbulent gas flow in piston
engines. In L.N. Khitrin, editor, Combustion in turbu-
lent flow, pages 122–147. Israel program for scien-
tific translations, Jerusalem, 1963.
[42] R.A. Fraser, P.G. Felton, F.V. Bracco, and D.A. San-
tavicca. Preliminary turbulence length scale mea-
surements in a motored IC engine. In SAE Technical
Paper Series, number 860021. 1986.
[43] A.N. Lipatnikov and J. Chomiak. Modeling of pres-
sure and non-stationary effects in spark-ignition en-
gine combustion: A comparison of different ap-
proaches. In SAE Technical Paper Series, number
2000-01-2034. 2000.
[44] G. Woschni. A universally applicable equation for the
instantaneous heat transfer coefficient in the internal
combustion engine. In SAE Technical Paper Series,
number 670931. 1967.

15