Journal of the Energy Institute 93 (2020) 2444e2455

Contents lists available at ScienceDirect

Journal of the Energy Institute

journal homepage: www.journals.elsevier.com/journal-of-the-energy-institute

Development of a turbulent burning velocity model based on flame

stretch concept for SI engines
Behdad Afkhami*, Yanyu Wang, Scott A. Miers, Jeffrey D. Naber
Department of Mechanical Engineering-Engineering Mechanics, Michigan Technological University, Houghton, MI, USA

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

Article history: According to the US Energy Information Administration, fossil fuels will remain the main source of
Received 27 February 2020 energy for transportation over the next decades and thus the combustion of these fuels remains an
Received in revised form important concern.
28 July 2020
This research studied the flame propagation under engine in-cylinder conditions and developed a
Accepted 4 August 2020
correlation for turbulent burning velocity based on the global flame stretch concept. To study the impact
Available online 19 August 2020
of engine operation on flame stretch, two speeds, two loads, and three fuel-air mixtures were investi-
gated. The flame front was determined by processing images of the flame natural luminosity.
Keywords:
Burning velocity model
A turbulent burning velocity model was developed using dimensional analysis. The model showed that
Flame stretch the turbulent burning velocity decreased due to flame stretching. Higher engine speeds increased the
Markstein number turbulent burning velocity by increasing the turbulent intensity, yet a tradeoff between the flame stretch
Spark-ignition engine and the turbulent burning velocity due to higher engine speed was observed. In cases where the flame
distortion was very high, the flame stretch may cancel out any benefits of a large enflamed area.
Incorporating the flame stretch into the burning velocity model and coupling the developed model
with GT-Power simulation software revealed that the stretch may result in a 35% reduction in turbulent
burning velocity.
© 2020 Energy Institute. Published by Elsevier Ltd. All rights reserved.

1. Introduction to evaluate the burning velocity is to understand the laminar

burning velocity, which corresponds to the speed of a one-
The energy consumption projection based on a US Energy In- dimensional planar adiabatic flame in laminar conditions without
formation Administration report shows that fossil fuels will remain instability [10]. However, this fundamental laminar burning ve-
the main source of energy for power generation and transportation locity cannot be considered as the speed of an expanding flame
over the next several decades [1]. Improving combustion of fossil since flames are affected by flame stretch. A curved flame, in a non-
fuels remains an important concern not only for today, but also for uniform flow, is subject to a continual frontal area change, known
the foreseeable future [2e4]. This becomes increasingly important as flame stretch. Since Karlovitz et al. [11] introduced the first
in the transportation sector because internal combustion engines attempt to calculate the flame stretch as a possible mechanism for
must meet ever-increasing emission regulations. Understanding flame quenching, there have been several studies to quantify the
the behavior of the flame in an internal combustion engine leads to effect of flame stretch on the flame speed. The flame has a finite
improvements in fuel economy and exhaust emissions [5e8]. thickness and the flame surface has no precise meaning. Based on
Flame velocity that propagates normal to itself and relative to the Damko € hler’s theory [12], flame-turbulence interaction can be
flow into the unburned mixture is one of the most important classified as either a small scale or large-scale turbulence. In the
quantities in combustion [9]. It provides the capability of predicting small-scale turbulence regime, small eddies interact with transport
the fuel consumption rate in a combustor (from internal combus- mechanisms within the flame and they modify the transport pro-
tion engines to gas turbines and industrial furnaces). The first step cesses in the preheat zone [13]. In the large-scale turbulence, the
turbulence length scale is assumed to be much larger than the
flame thickness and hence the flame (preheat and reaction zones)
can be considered as a surface [14]. Implicit in the large-scale tur-
* Corresponding author.
E-mail address: [email protected] (B. Afkhami). bulence approach it is assumed that the small turbulent eddies do

https://doi.org/10.1016/j.joei.2020.08.002
1743-9671/© 2020 Energy Institute. Published by Elsevier Ltd. All rights reserved.
 B. Afkhami et al. / Journal of the Energy Institute 93 (2020) 2444e2455 2445

not modify chemistry and transport inside the flame, which retains increase in flame surface area as a result of turbulence (Damko€ hler
its laminar structure with transport, and chemistry accounted for theory), but also the relative drop in local burning velocity as a
through the Markstein length. By utilizing the large-scale turbu- result of flame stretching. Fogla et al. [13] examined the propaga-
lence approach, Bauwens et al. [15] experimentally examined tion of premixed turbulent flames using a hydrodynamic model
spherical-flame propagation of propaneeair flames and considered and applying a hybrid Navier-Stokes/front tracking methodology.
the outer surface of the spherical flame they imaged as the flame They showed that for moderate-to-high turbulence intensities
front. Gaponov [16] analytically studied flame propagation by a (u’=S a1; the ratio of turbulence intensity to laminar flame speed),
L
premixed gas mixture at high turbulence and by applying the large- the effect of the flame stretch on the turbulent flame speed can be
scale turbulence approach, he concluded that the turbulent flame
expressed as shown in Eq. (3), where K and AT are respectively the
speed is defined by properties of its forward front. Consider a point
mean flame stretch experienced by the turbulent flame and the
on the flame surface that moves along the surface as a result of its
mean turbulent flame surface area, in time and cross sectional area.
underlying fluid velocity field but remains on it at later times. A
They observed a 10e15% reduction in turbulent flame speed due to
surface element of area A would be formed with a set of such points,
flame stretching. However, unlike engine in-cylinder conditions,
which is continually deformed with respect to time. The stretch of a
they assumed that the system pressure was nearly constant.
general flame surface can be obtained by calculating the fractional
area change of a Lagrangian surface element and is formulated as

shown in Eq. (1). ST L K AT
¼ 1 (3)
SL SL AL
1 dA
K¼ (1) Brequigny et al. [10] studied different airefuel mixtures
A dt (different fuels or equivalence ratios) with similar un-stretched
In general, flame stretch results from the non-uniform flow field laminar burning velocities and thermodynamic properties but
across the curved front (hydrodynamic strain) or from the motion different responses to the flame stretch. The mixtures were studied
of the moving curved front (effect of curvature). By utilizing kine- in a turbulent spherical vessel and the engine-like flow and com-
matic arguments to calculate the rate of area change, dA/dt, for a bustion conditions were produced using a fan at 5000, 6000, and
flame of arbitrary shape, Matalon [17] showed that the deviations 7000 rpm. They concluded that flame stretch sensitivity properties
of the flame temperature from the adiabatic flame temperature are such as Markstein length and Lewis number are relevant parame-
locally proportional to the flame stretch. In a research to experi- ters that need to be considered to predict the global performance of
mentally explore the effect of density ratio on turbulent flame fuels. They also extended their research to an optical engine and by
speed, Lipatnikov et al. [18] observed that the flame stretch resulted defining an “arbitrary flame stretch sensitivity”, they showed that
in decreasing mean local consumption and hence the flame increasing engine speed decreased the flame sensitivity to the
displacement speed. flame stretch.
Giannakopoulos et al. [19] conducted an analytical research on Although there are studies on flame stretch, only a few of them
the effect of flame stretch on flame speed in a spherically-shaped studied the effect of flame stretch under direct fuel injection and in-
flame. They showed that the flame stretch affects the (local) cylinder engine conditions. This becomes more important when it
burning velocity and the proportionality is measured by a coeffi- is realized that burning velocity and flame extinction during the
cient known as Markstein length. The Markstein length shows the flame kernel growth period can be significantly affected by the
sensitivity of the flame to the stretch and depends on the fuel type, flame stretch and have a direct effect on engine performance and
its reactivity, and the mixture composition. In their study, they exhaust emissions. The present research focuses on the flame
specifically assumed that the flame thickness characterized by the stretch inside an engine cylinder and the way it affects the turbu-
diffusion length lf ≡ Dth/SL, where Dth is the thermal diffusivity of the lent burning velocity. For this reason, an optically-accessible engine
combustible mixture and SL is the laminar flame speed, is much was utilized and the flame was imaged with a high-speed camera.
smaller than the length scale L that characterizes the flow field. In The flame images were processed and a burning velocity model
the limit lf/L/ 0, the entire flame reduces to a surface e the flame was developed at different engine speeds, loads and fuel-air mix-
front, which is uniquely determined. Therefore, an expression for tures stoichiometry. A summary of the studies on the effect of flame
the (local) burning velocity, Sf, can be written as shown in Eq. (2). stretch on burning velocity is shown in Fig. 1.
The equation shows the linear dependence of burning velocity on
the flame stretch, where L is the Markstein length.

Sf ¼ SL  L K (2) 2. Markstein length and Markstein number

Deshaies and Cambray [20] conducted research on a stationary
axisymmetric flame-flow configuration for three lean propane- The Markstein length L , has units of length and is proportional
oxygen-nitrogen mixtures to study the effect of flame stretch on to flame thickness, lf. It represents the effects of reaction and
flame velocity. They concluded that the proportionality between diffusion and therefore depends on the mixture composition and
burning velocity and stretch results in up to a 50% decrease in its reactivity. The mixture composition is represented by the Lewis
burning velocity in situations where the flame stretch decreased effective number, Leeff and the reactivity is represented by the
the flame speed. For a larger decrease in flame burning velocity, Zel’dovich or activation energy parameter. It is convenient to
they observed extinction of the flame. introduce a non-dimensionalized form of the Markstein length as
Theoretical studies have adopted Damko € hler’s theory [12] that the ratio of Markstein length to flame thickness. This is known as
the ratio of turbulent flame speed, ST, to the laminar flame speed, SL, the Markstein number and is expressed as shown in Eq. (4). Note
is equal to the increase in surface area of the wrinkled flame, such that in asymptotic hydrodynamic theory the objective is the
that ST/SL ¼ AT/AL where AT and AL are the surface areas of turbulent determination of the large-scale flow field (represented by the
l
and laminar flames, respectively. But recently the controlling fac- length scale L) induced by the propagating flame. In the limit Lf /0,
tors on turbulent flame speed were fully recognized and it was the flame of small thickness reduces to a surface that propagates
shown that the ratio of ST/SL depends not only on the relative relative to the fresh mixture.
2446 B. Afkhami et al. / Journal of the Energy Institute 93 (2020) 2444e2455

Fig. 1. Summary of the studies on the effect of flame stretch on burning velocity.

of methane-air mixtures by Seshadri and Go €ttgens [23].

8   9
 < s ðs lðtÞ b Leeff  1 ðs lðtÞ
s  1 = The relationship between the burning velocity and the flame
M ¼L lf ¼ dt þ * ln dt stretch is through the Markstein length. With the definition of M ¼
:s  1 t 2ðs  1Þ t t1 ;
1 1 L =lf for the Markstein number, the Markstein length can be
(4) calculated if the flame thickness, lf, is known. According to Deme-
soukas et al. [25] the laminar flame thickness can be calculated as
where s ≡ ru/rb¼ Tb/Tu is the unburned to burned density ratio shown in Eq. (5), where Dth is the thermal diffusivity of the
(thermal expansion parameter), b is the Zel’dovich number, and l is combustible mixture and SL is the laminar flame speed. Thus, the
the thermal conductivity of the mixture. The dependency of ther- relationship between the burning velocity and the flame stretch
mal conductivity on temperature for most combustible mixtures is (Markstein length) can be computed.
of the form lðtÞ  t0:7 [19]. In this research, the state of fresh un-
burned gas is chosen as the reference so t ¼ T/Tu. Dth
lf ¼ (5)
The Lewis number is defined as the ratio of thermal diffusivity to SL
mass diffusivity, Dth/D. The effective Lewis number of the mixture,
Leeff, is a weighted average of individual Lewis numbers of fuel and
oxidizer, LeF and LeO, and it is weighted more heavily with respect to
the deficient component in the mixture. Detailed calculations of the 3. Experimental set up
effective Lewis number are discussed in Refs. [19].
The Zel’dovich number in Eq. (4) shows the sensitivity of 3.1. Optical engine test cell setup
chemical reactions to the variation of the maximum flame tem-
perature and the inverse of it physically denotes an “effective The core pieces that make up the optical engine test cell are the
dimensionless width of the reaction zone” [21]. Go € ttgens et al. [22] engine, dynamometer, fuel system, air intake/exhaust system, en-
showed that b is a function of equivalence ratio (4), however, the gine control system, and flame imaging system. Further details of
dependency is weak and may be ignored over a small range of 4 the instrumentation and experimental set up are in Ref. [26].
(~0.8e1.2); an assumption supported by chemical-kinetic models
[23]. Bradley et al. [24] evaluated b for an equivalence ratio between 3.1.1. Engine and dynamometer
0.8 and 1.2 and they observed a slight increase with pressure and The engine used in this research was a 0.601-Liter, single-
interpreted the overall increasing trend of b with pressure as fol- cylinder, direct-injection, spark-ignition, naturally-aspirated,
lows. As the pressure increases, the temperature of fuel consump- optically-accessible engine. The engine specifications and oper-
tion, or inner layer, increases. The inner layer temperature ating conditions are shown in Table 1. The primary reason for using
characterizes the balance between the chain-branching reactions an optical engine was to capture flame natural luminosity images
and the chain-breaking effect of the fuel and the recombination during actual engine operation. The spark advance shown in Table 1
reactions. The inner layer temperature is interpreted as the critical for different test condition are selected based on the maximum
temperature “at and above which chemical reactions take place” brake torque (MBT) spark timing of the metal engine identical to
[24]. The temperature is determined by the relative rates of the optical engine utilized in the current research.
important kinetic steps. Hence the reaction is driven to higher An electric heater was utilized to warm up the engine oil to a
temperatures and this has the effect of increasing the Zel’dovich typical working temperature (50  C) and hold it constant during
number. This effect was also observed in the computational studies experiments. Flame imaging experiments were conducted based on
 B. Afkhami et al. / Journal of the Energy Institute 93 (2020) 2444e2455 2447

Table 1
Engine specifications and operating conditions.

NO. OF CYLINDERS 1

BORE X STROKE 87.5 mm  100 mm

DISPLACEMENT 0.601 L
COMPRESSION RATIO 10.01:1
ASPIRATION NATURALLY ASPIRATED
FUEL DELIVERY SYSTEM DIRECT INJECTION
FUEL INJECTION PRESSURE 5.3 MPa
MAXIMUM INDICATED MEAN EFFECTIVE PRESSURE (IMEP) 1.3 MPa
MAXIMUM SPEED 3000 RPM; 2500 RPM PREFERRED
START OF INJECTION 300 CA BTDC
◦
SPARK ADVANCE 1000 RPM 250 kPa IMEP 21 CA
◦
300 kPa IMEP 20 CA
◦
1500RPM 250 kPa IMEP 19 CA
◦
300 kPa IMEP 18 CA

a skip-firing strategy to avoid extra heat in the engine, with a ratio The start of imaging was triggered at 2 CA before spark timing and
of 10 fired to 15 skip-fired cycles. In the current study, E10 was used 130 images were taken from the flame in every fired cycle. A total of
as the fuel with a research octane number (RON) of 91.7 and motor 50 fired cycles were imaged and ensemble averaged for each test
octane number (MON) of 82.5, consisted of 9.31%v/v ethanol. 9.3% condition.
ethanol and 90.7% indolene (a gasoline reference fuel [27,28]) were
used to compute other properties of E10. Experiments were con-
ducted under two different engine loads; 250 and 300 kPa net 4. Image processing and flame properties calculation
Indicated Mean Effective Pressure (IMEP). Three equivalence ratios
were tested; stoichiometric, lean with 4 ¼ 0.84, and rich with Flame image processing begins by loading the natural lumi-
4 ¼ 1.18. Spark timing was kept constant at each engine speed and nosity images into an image processing software. In this study, post
load, shown in Table 1, in an attempt to minimize changes to the processing was performed in MATLAB. A detailed discussion on
fluid dynamical state of flow in the cylinder. The fuel was injected image processing was presented in our previous study [26].
directly into the cylinder early in the intake stroke (see Table 1) to A graphical representation of how the flame front length pro-
promote a homogeneous mixture preparation. gresses is shown in Fig. 3 at an engine speed of 1000 rpm and
250 kPa net IMEP. After MFB90, there was minimal fuel left to burn
and processing images does not give any meaningful concept. For
3.1.2. Engine instrumentation and measurements
this reason, Fig. 3 was limited to the duration between ignition and
The engine was instrumented with various sensors, in order to
MFB90. As the flame front progresses, it eventually touches the
measure the optical engine operating parameters. In Table 2, pa-
walls of the combustion chamber. The locations where the flame
rameters that were measured with different sensors and their
touches the combustion chamber are not utilized in computing the
specifications are listed.
flame front properties.
To record in-cylinder pressure, a high-speed Advance Combus-
It was important to determine when the flame front impinged
tion Analysis Program (ACAP) data acquisition system was used. To
on the piston surface as it was discussed by Ratzke et al. [29]. They
record steady state variables, a National Instruments (NI) PXIe-1078
derived a turbulent flame speed model from the mean in-cylinder
chassis was utilized with two NI TB-4353 thermocouple modules to
pressure using a zonal cylinder model and showed that the in-
collect temperature data while oil pressure was transferred to the
cylinder pressure curve of the model was in good agreement
chassis through an NI TB-2706 terminal block.
with measurements until the flame front touched the piston sur-
face. After that, higher deviation was observed between the
3.2. Flame imaging measured and the modeled in-cylinder pressure. Hence, they chose
the instant of time before the flame impinged on the piston for
High-speed images of the flame were captured using a Photron further studies. In this research, a novel approach was followed to
APX RS camera at a rate of 10,000 frames per second (fps). At this calculate the flame radius. First, a binary mask was applied to each
rate, the maximum size of the image was 768  336 pixels. The image to ensure that the part of the flame front which is in contact
spatial resolution of 0.0769 mm/pixel was assigned to the images with the combustion chamber is excluded from the images. Then
using the known dimensions of the spark plug, which was clearly the distance between the instantaneous centroid of the flame and
visible in many images. The schematic view of the test set up is each pixel on the flame front was calculated. Finally, the instanta-
shown in Fig. 2. neous average radius of the flame was estimated as the average
Images were taken at the tumble plane with the intake valve on distance between the centroid and the flame front. To find the
the right side and the exhaust valve on the left side of the image. average radius of the flame, 50 fired cycles were averaged. The

Table 2
Engine instrumentation and measurement.

PARAMETER PRODUCT USED SPECIFICATIONS

IN-CYLINDER PRESSURE PCB 115A04 0e241.4 BAR ABSOLUTE

ENGINE POSITION (ENCODER) BEI XH20DB-37-SS-720-ABZC-28V/V-SM18 RESOLUTION: 0.5 CAD
COOLANT, OIL, INTAKE AND EXHAUST TEMPERATURES OMEGA K TYPE
RANGE: 200 TO 1250  C
OIL PRESSURE GP50 211 0e34.5 BAR
2448 B. Afkhami et al. / Journal of the Energy Institute 93 (2020) 2444e2455

Fig. 2. Schematic view of the optical SIDI engine. a) Side view; b) Bottom view.

instantaneous average flame radius, R, can then be expressed as

!,
shown in Eq. (6), where px is the number of pixels on the flame X
px
front and ri is the distance between the instantaneous centroid and R¼ ri px (6)
the ith pixel on the flame front. 1

An alternative approach for calculation of flame radius was

introduced by Salazar and Kaiser [30]. In their method, a projected
circular flame with the area equal to the real flame was assumed.
From the already-known flame area, they calculated the respective
equivalent radius R. Detailed comparison of the two different ap-
proaches is discussed in Refs. [26]. In order to represent the dif-
ference between the two approaches, flame radius versus  CA is
shown in Fig. 4. There is good agreement between the two ap-
proaches for the first section of the flame propagation with sepa-
ration occurring when the flame touches the piston surface. This
behavior can be interpreted according to previous studies [31,32].
They showed that the average flame is fairly spherical (circular in 2-
D), either imaged in the swirl plane or the tumble plane. The sep-
aration point is an advantage of the current approach that gives
better understanding of when the flame is no longer spherical. By
calculating the flame radius, the crank angle where the flame front
touches the piston is determined. In addition, in Fig. 4 the

Fig. 3. Variation of flame front length during one cycle for stoichiometric fuel-air
mixture at 1000 rpm and 250 kPa net IMEP; A: at MFB10; B: at TDC; C: at MFB50; Fig. 4. Comparison between the approaches used to calculate flame radius; current
D: at MFB90. approach and Salazar [30] approach.
 B. Afkhami et al. / Journal of the Energy Institute 93 (2020) 2444e2455 2449

maximum and minimum calculated equivalent flame radii using 5. Results and discussion
the current approach for the 50 fired cycles are shown with dashed
lines. 5.1. Turbulent burning velocity model development
Assuming a spherically-propagating flame with radius R, Eq. (1)
was re-written as Eq. (7) where dt was the time interval between Qualitative assessment of the turbulent flame structure in the
images from the high-speed camera. In this study, there was no studied optical engine can be accomplished by utilizing the Borghi/
spatial resolution in the third dimension because a single camera Peters diagram, which provides an understanding about the tur-
was used to collect images. Extracting quantitative measurements bulent combustion regimes [9]. The diagram evaluates the struc-
from 2-D flame images provides valuable information pertaining to ture of the flames based on their interaction with the turbulent flow
flame stretch, which was discussed in a previous study [33]. Esti- field and the different regimes of turbulent combustion are iden-
mation of 3D flame surface from planar images in turbulent pre- tified by comparing turbulence and chemical characteristic time
mixed combustion [34,35] showed that a 2D tomography image scales. In order to acquire the information about the in-cylinder
underestimates the flame surface by around 15% compared to true flow field of the optical engine, the engine was modeled in GT-
3D values. However, the underestimation in flame surface area does Power, a 1-D engine simulation program. The results for turbu-
not affect the computation of flame stretch, because area appears in lent intensity, u’, and integral length scale, LI, were utilized to po-
both the numerator and denominator in Eq. (1). Using Eq. (7) for the sition the turbulent combustion regime in the optical engine on the
flame stretch results in global flame stretch compared to the local Borghi/Peters diagram as shown in Fig. 6 for a stoichiometric fuel-
flame stretch, which requires detailed information of the fluid flow air mixture at two different engine speeds.
field [10,36]. This information was not available for this study. It is evident that increasing the engine speed increased the
Global flame stretch is more favorable than local flame stretch for a turbulence intensity. The duration of combustion shown on Fig. 6 is
spherically expanding flame due to difficulties in fluid flow field from ignition until the flame impinges on the piston surface. Based
measurements after ignition in an engine [10,33,37,40]. on Fig. 6, it is reasonable to assume that the flame thickness is very
small compared to the in-cylinder flow field turbulent length scale
as discussed in MARKSTEIN NUMBER section. In the large-scale
2 dR
K¼ (7) turbulence regime, it is assumed that the small turbulent eddies
R dt
do not modify chemistry and transport inside the flame. The flame
To investigate Damko €hler’s theory under engine in-cylinder retains its laminar structure and therefore, the large-scale turbu-
conditions, calculation of the ratio of wrinkled (turbulent) flame lence regime is valid for this study.
surface area to laminar flame surface area was necessary. In general, Detailed discussion on the flame stretch, burning velocity and
the flame front moves through the combustion chamber with the Markstein number under the test conditions shown in Table 1 was
laminar flame speed. However, the surface of the flame is wrinkled presented in our previous study [26]. Dimensional analysis was
due to in-cylinder turbulence, resulting in an enlargement of the utilized to modify Damko €hler’s theory [12] and develop a turbulent
surface and thus an acceleration of the combustion process. To burning velocity model that incorporated the effect of flame
quantify the flame distortion due to turbulence, Aghdam [31], Weib stretch. The variables used for the dimensional analysis to predict
et al. [38], and Ritzinger [39] suggested dividing the wrinkled flame turbulent burning velocity (ST) included a tuning constant (C), in-
front length (LT) by the length of unwrinkled flame front (Ll), which cylinder pressure (Pcyl), barometric pressure at standard condition
was also applied in this study. Hence, for spherical flames the (P0), Markstein length (L ), flame stretch (K), laminar flame speed
surface area ratio can be estimated as Eq. (8). (SL), and laminar and turbulent flame surface area (AL, AT, respec-
tively). The dependency was expressed in the form of ST ¼ f ðC;Pcyl ;

2 P0 ; L ; K; SL ; AL ; AT Þ. Using dimensional analysis, the non-
AT L
z T (8) dimensional groups relating ST and the variables were deter-
AL Ll
mined as Eq. (9). The dimensionless groups were selected to cover
Fig. 5 represents the defined turbulent and laminar flame sur-
face areas and burning velocities in an internal combustion engine.
The method for the flame wrinkling quantification assumes that
the wrinkling of the surface is independent of the orientation of the
measurement plane and similar in all directions. As a consequence,
the measured flame front sections are statistically representative
for the wrinkling of the flame in every direction.

Fig. 6. Borghi/Peters diagram showing the combustion regimes for stoichiometric fuel-
Fig. 5. Turbulent flame propagation in an internal combustion engine. air mixture at 250 kPa net IMEP and two engine speeds; 1000 rpm and 1500 rpm.
2450 B. Afkhami et al. / Journal of the Energy Institute 93 (2020) 2444e2455

the effects of load (through Pcyl), stoichiometry (through L ), and

engine speed (through AT and its proportionality to turbulent in-
tensity [12]).



ST Pcyl m L K n AT
¼C 1 (9)
SL Po SL AL
An iterative solution was used to maximize R-squared and
minimize residuals to determine m and n. Note that, comparing the
Eq. (9) with reference [13] reveals that a tuning constant (C) and the
effect of changing system pressure (Pcyl/P0) are considered in the
developed model. The beauty of the model presented by Refs. [13]
was that they proved that C was equal to 1, which eliminates the
need for changing based on different applications. Similar approach
was also followed in this research and the C was set to 1 in the
attempt to develop a simple model that did not over-constrain our
experimental data. Only the portion of combustion that was after
spark and before the flame touched the piston was selected to
determine the correlation exponents. To provide better under-
Pcyl 0:47  1:3
standing of the dependency of ST/SL on Pcyl/P0 and LSLK, Fig. 7 is Fig. 8. Po 1  LSLK versus the ratio of (ST/SL)/(AT/AL) calculated from experi-
shown for three different fuel-air mixtures at 1000 rpm engine mental results.
speed and 250 kPa net IMEP. Fig. 7 shows that ST/SL increases with
increasing cylinder pressure and decreasing flame stretch.
Increasing cylinder pressure results in a destabilizing effect and 1 Increasing in-cylinder pressure increases the turbulent burning
increases burning velocity [41,42] and increasing flame stretch velocity which is in agreement with [41,42] who experimentally
decreases burning velocity for fuel-air mixtures with positive studied propagation characteristics of turbulent premixed
Markstein number [13]. flames at elevated pressures. Based on their finding, the effects
To avoid overfitting to the data sets and proceed with one value of pressure on ST/SL are significant. While in Ref. [41] the pres-
for m and another for n, all the data sets for the test cases shown in sure exponent, m, for methane and propane Bunsen-type tur-
Table 1 were combined. Eq. (10) suggests a correlation with bulent premixed flames in a high-pressure chamber were found
empirical coefficients determined by a fit to the data. to be 0.38 and 0.4, respectively, in the current study the value of
m for E10 was 0.47.
2 The values for LSLK was less than unity and implies that the



ST Pcyl 0:47 L K 1:3 AT turbulent burning velocity decreases as flame stretch increases.
¼ 1 (10)
SL Po SL AL The exponent of (1  LSLK), was found to be 1.3 which was in good
An investigation of data scatter is shown in Fig. 8 for agreement with [13], who analytically examined propagation of

0:47
1:3 premixed turbulent flames by applying a hybrid Navier-Stokes/
Pcyl
Po 1  LSLK versus the ratio of (ST/SL)/(AT/AL) calculated front tracking methodology. They determined n to be 1.
3 It has been shown that increasing engine speed increases flame
from experimental results with the solid line showing the 1:1 ratio stretch, especially during the very early stage of flame kernel
and the dashed lines showing one standard deviation around the growth [26,43]. However, the turbulent intensity (u’) also in-
developed model containing 68% of the data points. The data points creases with engine speed [44]. According to Ref. [13], AT/AL
shown in Fig. 8 are from all test cases discussed in section 1.1. follows a direct scaling law depending on turbulent intensity
Scatter could be reduced if more dimensionless groups were added and therefore AT/AL increases with increasing engine speed
to the model. However, an attempt was made to develop a simple (AT =AL fu’ =SL Þ. There is a tradeoff between AT/AL and the flame
model that did not over-constrain the data. stretch, as shown in Eq. (10). In cases where the flame distortion
From the developed turbulent burning velocity model (Eq. (10))
the following observations can be made:

Fig. 7. Dependency of ST/SL on left: Pcyl/P0, and right: LK for different fuel-air mixtures at 1000 rpm and 250 kPa net IMEP.
SL
 B. Afkhami et al. / Journal of the Energy Institute 93 (2020) 2444e2455 2451

is very high (high stretch), the flame stretch may cancel out any presented the burning velocity for a SI engine with compression
benefits of a large enflamed area [43]. ratio of 10.2 operating with isooctane stoichiometric air/fuel
mixture at 1500 rpm and 20  CA bTDC ignition timing, which is
very comparable to the test cases shown in Table 1. In Fig. 11 the
turbulent burning velocity for the 250 and 300 kPa IMEP at
5.2. Validation of turbulent burning velocity model 1500 rpm using the developed model is compared to the similar
test cases from Ref. [32] using the “Zimont” and “Leeds KLe” tur-
5.2.1. Validation results from optical engine bulent burning velocity models. A good agreement between the
The developed turbulent burning velocity model was coupled developed model and the “Zimont” and “Leeds KLe” turbulent
with GT-Power engine simulation software to validate the model. burning velocity models from Ref. [32] can be observed from Fig. 11.
The intake, cylinders, cranktrain, and exhaust sections of the optical
engine were modeled in GT-Power. In order to couple the devel-
oped turbulent burning velocity model with GT-Power, a Dynamic 5.2.2. Extrapolation of developed turbulent burning velocity model
Link Library (.dll) file was utilized. FORTRAN programing language test setup and procedure
was required to build the user interface code. The default turbulent The proposed turbulent burning velocity model (Eq. (10)) was
combustion model, which was utilized in GT-Power, was developed developed using only data from the optical engine, which operated
by Morel et al. [45]. The laminar flame speed model that was pre- at a relatively slow speed and low load. To test the robustness of the
programmed in GT-Power was utilized for this validation test. All model, a liquid-cooled, four-stroke, two-cylinder, carbureted en-
coefficients related to the turbulent burning velocity and heat gine was instrumented to collect engine performance and com-
transfer in GT-Power were set to their default value. In GT-Power, bustion data. The engine specifications are shown in Table 3 and the
the default values for the turbulent flame speed and the standard details of the experimental setup were discussed in our previous
Woschni heat transfer multipliers are equal to 1. Setting the mul- study [46]. The brake torque was held constant at 40 Nm (7.5 bar
tipliers to their default values eliminated the need to tune any BMEP) and the engine speed varied from 1800 to 3600 RPM.
combustion-related parameters. Extrapolation of the developed turbulent burning velocity
To investigate the validity of the developed turbulent burning model was also conducted by coupling the developed model with
velocity model, in-cylinder pressure was compared between the GT-Power engine simulation software. The laminar flame speed
experiments and the GT-Power. In Fig. 9 the in-cylinder pressure for model that was pre-programmed in GT-Power was utilized for this
two of the test cases (stoichiometric fuel-air mixture, 250 kPa IMEP extrapolation test. The parameters which were used to build the
at 1000 and 1500 rpm) are shown. In addition, to study the GT-Power model are shown in Table 4.
robustness of the developed burning velocity model in predicting
the in-cylinder pressure for off-stoichiometric mixtures, Fig. 10 is 5.3. Extrapolation results
shown for two test cases at 1000 rpm, 300 kPa net IMEP one for the
lean and another for the rich mixtures. In each figure, the results for In-cylinder pressure results from GT-Power are compared with
the default turbulent burning velocity model are also shown, to the experimental results from 1800 to 3600 RPM in Figs. 12- 15. In
show the improvement with the developed turbulent burning ve- each figure, the results for the default turbulent burning velocity
locity model. From Figs. 9 and 10 it can be seen that incorporating model are also shown, to show the improvement with the devel-
the effect of flame stretch into the burning velocity improved the oped turbulent burning velocity model. From Fig. 12 through Fig. 15
prediction of the in-cylinder pressure. it is evident that incorporating the effect of flame stretch into the
In addition, to gain better insight into the developed model, it is turbulent burning velocity model significantly improved the pre-
advantageous to compare the results of the turbulent burning ve- diction of the in-cylinder pressure trace, the maximum in-cylinder
locity with other literature. Note that the engine design parameters pressure and the phasing of the combustion. The Markstein num-
as well as the operating parameters can affect the turbulent ber is positive for the hydrocarbon-air mixtures. This means that
burning velocity. Therefore, the selected literature should have the effective (stretched) burning velocity will be lower than the
utilized the engine with the design and operating parameters as unstretched burning velocity. When flame stretch is taken into
close as possible to the tested engine in this study. Ref. [32] account, the result is slower flame propagation and hence slower

Fig. 9. In-cylinder pressure for stoichiometric fuel-air mixture at 250 kPa net IMEP and left: 1000 rpm, right: 1500 rpm from experiment and GT-Power results using the default and
the developed turbulent burning velocity models.
2452 B. Afkhami et al. / Journal of the Energy Institute 93 (2020) 2444e2455

Fig. 10. In-cylinder pressure at 1000 rpm, 300 kPa net IMEP and left: lean, right: rich fuel-air mixtures from experiment and GT-Power results using the developed turbulent
burning velocity models.

both engines; especially in the early phase of combustion.

To further understand the impact of flame stretch on the overall
model results, the dimensionless group containing flame stretch
was plotted for different engine speeds, as shown in Fig. 16. During
the early flame development period, a 35% reduction in turbulent
burning velocity was realized through the implementation of flame
stretch. The effect of the flame stretch dimensionless group on
reducing the burning velocity increased with increasing engine
speed.
In-cylinder pressure showed that incorporating the effect of
flame stretch into the burning velocity model improved prediction
of the pressure trace and the phasing of the combustion. To
quantitatively evaluate the improvement, it was useful to compare
typical engine performance parameters between the experimental
results and GT-Power. Therefore, IMEP, BSFC, CA50 and D10-90
were selected to respectively study load, fuel conversion effi-
ciency, combustion phasing and combustion duration. In Fig. 17
engine operating parameters (IMEP, BSFC, CA50, and D10-90) are
shown. Significant improvements in predicting the engine perfor-
mance parameters were observed when the developed turbulent
Fig. 11. Comparison of the developed turbulent burning velocity with the models used burning velocity model was coupled with GT-Power compared to
in Ref. [32] at 1500 rpm and 20  CA bTDC ignition timing. the default turbulent burning velocity model in GT-Power. The
developed model was able to predict the IMEP and BSFC with
maximum error of 6.4% and 4.4%, respectively, among the four
Table 3
validation test cases. From the CA50, it can be seen that as the
Engine specifications for turbulent burning velocity model extrapolation.
engine speed increased, the model showed a delay in the com-
Engine Parameter Value bustion phasing; however, the combustion duration (D10-90) was
No. of cylinders 2 underpredicted by about 5 CA for all the test cases.
Bore  Stroke 80 mm  67 mm
Displacement volume 674 cm3
Compression ratio 8.5:1 5.4. Limitations of the developed burning velocity model
Aspiration Naturally Aspirated
Maximum speed 3750 rpm Every developed model is expected to have kind of limitations
Fuel delivery system Carburetor
and to study the limitations of the developed turbulent burning
Cooling system Liquid
Spark timing 40  CA BTDC velocity model in this study, it was beneficial to turn back to the
assumption that was made earlier to develop the model. It was
specifically assumed and shown on the Borghi/Peters diagram that
the flame thickness was much smaller than the turbulent length
heat release compared to a model that does not account for flame scale, LI, that characterizes the flow field. In the limit lf/LI/ 0, the
stretch. However, the capability of predicting the in-cylinder entire flame shrinks to a surface e the flame front, which is
pressure trace seemed to diminish as the engine speed increased. uniquely determined. However, as the turbulence intensity in-
In addition, note that while the spark advance for the optical engine creases inside an engine cylinder with increasing engine speed
was nearly 20  CA bTDC (Table 1), this timing for the metal engine [44], the effect of eddies of the size of flame thickness become more
was 40  CA bTDC (Table 3). Therefore, despite the significant dif- important. Therefore, Borghi/Peters diagram is shown in Fig. 18 for
ferences in the spark advance between the two engines, the model the Kohler LH690 and the different test conditions are located on
showed a reasonable capability of predicting the pressure trace of the diagram to better represent how the regime of combustion may
 B. Afkhami et al. / Journal of the Energy Institute 93 (2020) 2444e2455 2453

Table 4
Key parameters used in GT-Power model (controlled parameters).

Parameter Description

Engine speed 1800, 2400, 3000, 3600 rpm to cover a wide range of operating condition
Equivalence ratio 1.25, 1.23, 1.26, 1.36 according to experimental results for the corresponding engine speed
Heat transfer multiplier Default (1); for standard Woschni heat transfer model
Turbulent burning velocity multiplier Default (1)
Spark timing Anchored at 40 CA BTDC
Cam timing and lift Measured on the tested engine; Rocker arm ratio 1.5:1.

Fig. 15. In-cylinder pressure at 3600 rpm and 7.5 bar BMEP from experiment and GT-
Power results using the default and the developed turbulent burning velocity models.
Fig. 12. In-cylinder pressure at 1800 rpm and 7.5 bar BMEP from experiment and GT-
Power results using the default and the developed turbulent burning velocity models.

Fig. 13. In-cylinder pressure at 2400 rpm and 7.5 bar BMEP from experiment and GT-
1:3
Power results using the default and the developed turbulent burning velocity models. Fig. 16. The value of 1  LSLK versus crank angle at four different engine speeds.

change as the engine speed increases. At higher engine speeds the

developed turbulent burning velocity may be used with care. In
other words, the application of the developed burning velocity
model may be limited at very higher engine speed. Due to the
limitation of the Kohler LH690 it was not possible to conduct ex-
periments at very high engine speed. In addition, it has been proven
[44] that the turbulence intensity in the internal combustion en-
gines has its maximum value at the start of intake stroke, and it
diminishes as the piston moves toward the top dead center (TDC).
Therefore, the later spark advance in the optical engine (see
Table 1) resulted in narrower flame regime compared to the Kohler
LH690 engine (see Table 3).

6. Conclusions and future studies

Fig. 14. In-cylinder pressure at 3000 rpm and 7.5 bar BMEP from experiment and GT- Recent studies have shown that flame stretch can have a sig-
Power results using the default and the developed turbulent burning velocity models. nificant impact on the turbulent burning velocity. An optically-
2454 B. Afkhami et al. / Journal of the Energy Institute 93 (2020) 2444e2455

Fig. 17. Comparison of performance parameters between experiments and GT-Power.

the fluid flow field.

In order to incorporate the effect of pressure and flame stretch
into a turbulent burning velocity model, dimensional analysis was
conducted and a model was developed based on the flame stretch
concept. The dependency of the turbulent burning velocity, ST, on
the in-cylinder pressure, Pcyl, and flame stretch, K, appeared to be

0:47
1:3
Pcyl
in the form of Po and 1  LSLK , respectively, showing

that the turbulent burning velocity increases due to pressure and

decreases due to stretching. In addition, although it is shown in
previous studies that increasing engine speed increases both the
flame stretch and the turbulent burning velocity, a tradeoff be-
tween the flame stretch and the burning velocity due to higher
engine speed was observed in the developed model. In cases where
the flame distortion is very high, the flame stretch may cancel out
any benefits of a large enflamed area.
While the turbulent burning velocity model was developed for
an optically-accessible DISI engine at low engine speed and load, it
Fig. 18. Borghi/Peters diagram of combustion regimes for the Kohler LH690 engine. was coupled with GT-Power engine simulation software and the
results were compare to experimental data for Kohler LH690 at
higher engine speed and load. Comparison of the engine in-
accessible engine was utilized to experimentally investigate the cylinder pressure and performance parameters from simulation
effect of flame stretch in an internal combustion engine. The flame and experiments for the Kohler LH690 revealed that when the
propagation process was captured by imaging the flame natural developed turbulent burning velocity model was coupled with GT-
luminosity using a high-speed camera on the tumble plane. The Power, it improved the predictive combustion capability of GT-
flame images were processed to find the flame front and a novel Power compared to the default turbulent burning velocity model
approach was introduced to determine the propagation of equiv- in GT-Power.
alent spherical flame. Computation of the flame stretch on CA basis In addition, simulation results showed that the flame stretch
was conducted using the definition of global flame stretch; lead to 35% decrease in turbulent burning velocity at very early
compared to local stretch, which requires detailed information of stage of combustion compared to the default turbulent burning
 B. Afkhami et al. / Journal of the Energy Institute 93 (2020) 2444e2455 2455

velocity model in GT-Power that did not account for the effect of [20] B. Deshaies, P. Cambray, The velocity of a premixed flame as a function of the
flame stretch: an experimental study, Combust. Flame 82 (3-4) (1990)
flame stretch. Therefore, the slower burning velocity due to flame
361e375.
stretch was then responsible for a slower increase in the heat [21] Xiao Jun Gu, Md Zahurul Haq, Malcolm Lawes, R. Woolley, Laminar burning
release. Also, simulation results at different engine speeds showed velocity and Markstein lengths of methaneeair mixtures, Combust. Flame 121
that as the engine speed increases, the effect of the flame stretch on (1) (2000) 41e58.
€ttgens, F. Mauss, N. Peters, Analytic approximations of burning velocities
[22] J. Go
reducing the burning velocity increases, as well. and flame thicknesses of lean hydrogen, methane, ethylene, ethane, acetylene,
Because the effect of equivalence ratio inhomogeneity is of and propane flames, in: Symposium (International) on Combustion, vol. 24,
important matter for the flame stretch, studying local burning ve- Elsevier, 1992, pp. 129e135, 1.
€ttgens, Structure of the oxidation layer for stoichiometric and
[23] K. Seshadri, J. Go
locity and the flame response to stretch may be an interesting lean methane-air flames, in: Reduced Kinetic Mechanisms and Asymptotic
research topic, especially in direct injection engines with late in- Approximations for Methane-Air Flames, 1991, pp. 111e136.
jection timing. [24] D. Bradley, R.A. Hicks, M. Lawes, C.G.W. Sheppard, R. Woolley, The measure-
ment of laminar burning velocities and Markstein numbers for iso-octaneeair
and iso-octaneen-heptaneeair mixtures at elevated temperatures and pres-
Declaration of Competing Interest sures in an explosion bomb, Combust. Flame 115 (1) (1998) 126e144.
[25] Sokratis Demesoukas, Pierre Brequigny, Christian Caillol, Fabien Halter,
Christine Mounaïm-Rousselle, 0D modeling aspects of flame stretch in spark
The authors declare that they have no known competing ignition engines and comparison with experimental results, Appl. Energy
financial interests or personal relationships that could have (2016) 401e412.
appeared to influence the work reported in this paper. [26] Behdad Afkhami, Yanyu Wang, Scott A. Miers, Jeffrey D. Naber, Experimental
determination of flame speed and flame stretch using an optically accessible,
spark-ignition engine, Int. J. Engine Res. (2019), 1468087419877718.
References [27] Yu Cheng Liu, Anthony J. Savas, C. Thomas Avedisian, Comparison of the
burning characteristics of indolene and commercial grade gasoline droplets
[1] International Energy Outlook 2016 with Projections to 2040, U.S. Energy In- without convection, Energy Fuel. 26 (9) (2012) 5740e5749.
formation Administration, May 2016. [28] Mohamad Metghalchi, James C. Keck, Burning velocities of mixtures of air
[2] M. Song, L. Zeng, Z. Chen, Z. Li, Q. Zhu, M. Kuang, Industrial application of an with methanol, isooctane, and indolene at high pressure and temperature,
improved multiple injection and multiple staging combustion technology in a Combust. Flame 48 (1982) 191e210.
600 MWe supercritical down-fired boiler, Environ. Sci. Technol. 50 (3) (2016) [29] Ansgar Ratzke, et al., Validation of turbulent flame speed models for meth-
1604e1610. aneeair-mixtures at high pressure gas engine conditions, Combust. Flame 162
[3] I. Bonefaci
c, I. Wolf, P. Blecich, Improvement of fuel oil spray combustion (7) (2015) 2778e2787.
inside a 7MW industrial furnace: a numerical study, Appl. Therm. Eng. 110 [30] Victor Salazar, Sebastian Kaiser, Influence of the flow field on flame propa-
(2017) 795e804. gation in a hydrogen-fueled internal combustion engine, SAE International
[4] Martia Shahsavan, Mohammadrasool Morovatiyan, John Hunter Mack, The Journal of Engines 4 (2011), 2011-24-0098 (2011): 2376-2394.
Influence of Mixedness on Ignition for Hydrogen Direct Injection in a Constant [31] Martin Goschütz, Christof Schulz, A. Sebastian, Kaiser, Endoscopic imaging of
Volume Combustion Chamber, 2018. early flame propagation in a near-production engine, SAE International
[5] B. Afkhami, A.H. Kakaee, K. Pouyan, Studying engine cold start characteristics Journal of Engines 7 (2014) 351e365, 2014-01-1178.
at low temperatures for CNG and HCNG by investigating low-temperature [32] Aghdam Abdi, Ebrahim, Improvement and Validation of a Thermodynamic SI
oxidation, Energy Convers. Manag. 64 (2012) 122e128. Engine Simulation Code, PhD diss, University of Leeds, 2003.
[6] Ehsan Ansari, Menucci Tyler, Mahdi Shahbakhti, Jeffrey Naber, Experimental [33] P.G. Aleiferis, J. Serras-Pereira, D. Richardson, Characterisation of flame
investigation into effects of high reactive fuel on combustion and emission development with ethanol, butanol, iso-octane, gasoline and methane in a
characteristics of the Diesel-Natural gas Reactivity Controlled Compression direct-injection spark-ignition engine, Fuel 109 (2013) 256e278.
Ignition engine, Appl. Energy 239 (2019) 948e956. [34] Denis Veynante, Lodato Guido, Domingo Pascale, Luc Vervisch, R. Evatt,
[7] Hamidreza Mahabadipour, Kalyan K. Srinivasan, Sundar R. Krishnan, A second Hawkes, Estimation of three-dimensional flame surface densities from planar
law-based framework to identify high efficiency pathways in dual fuel low images in turbulent premixed combustion, Exp. Fluid 49 (1) (2010) 267e278.
temperature combustion, Appl. Energy 202 (2017) 199e212. [35] Meng Zhang, Jinhua Wang, Wu Jin, Zuohua Huang, Hideaki Kobayashi, Ma Lin,
[8] Derek Johnson, Mahdi Darzi, Chris Ulishney, Mehar Bade, Nima Zamani, Estimation of 3D flame surface density and global fuel consumption rate from
Methods to improve combustion stability, efficiency, and power density of a 2D PLIF images of turbulent premixed flame, Combust. Flame 162 (5) (2015)
small, port-injected, spark-ignited, two-stroke natural gas engine, in: ASME 2087e2097.
2017 Internal Combustion Engine Division Fall Technical Conference, Amer- [36] Swetaprovo Chaudhuri, Abhishek Saha, K. Chung, Law. "On flameeturbulence
ican Society of Mechanical Engineers, 2017. V002T07A008-V002T07A008. interaction in constant-pressure expanding flames, Proc. Combust. Inst. 35 (2)
[9] N. Peters, Turbulent Combustion, Cambridge university press, 2000. (2015) 1331e1339.
[10] Pierre Brequigny, et al., Fuel performances in Spark-Ignition (SI) engines: [37] Philipp Schiffmann, David L. Reuss, Sick Volker, Empirical investigation of
impact of flame stretch, Combust. Flame 166 (2016) 98e112. spark-ignited flame-initiation cycle-to-cycle variability in a homogeneous
[11] B. Karlovitz, D.W. Denniston, D.H. Knapschaefer, F.E. Wells, Fourth Symposium charge reciprocating engine, Int. J. Engine Res. (2017), 1468087417720558.
(International) on Combustion, Williams and Wilkins, Baltimore, MD, 1953, [38] M. Weiß, N. Zarzalis, R. Suntz, Experimental study of Markstein number ef-
pp. 613e619. fects on laminar flamelet velocity in turbulent premixed flames, Combust.
[12] G. Damko €hler, Der einfluß der turbulenz auf die flammengeschwindigkeit in Flame 154 (4) (2008) 671e691.
gasgemischen, ZeitschriftfürElektrochemie und angewandtephysikali- [39] J. Ritzinger, Einfluss der Kraftstoffe RON95, Methan und Ethanol auf Flam-
scheChemie 46 (11) (1940) 601e626. menausbreitung und Klopfverhalten in Ottomotoren mit Abgasrückführung,
[13] N. Fogla, F. Creta, M. Matalon, Effect of folds and pockets on the topology and Ph.D. thesis, ETH Zürich, 2013.
propagation of premixed turbulent flames, Combust. Flame 162 (7) (2015) [40] Sokratis Demesoukas, Pierre Brequigny, Christian Caillol, Fabien Halter,
2758e2777, https://doi.org/10.1016/j.combustflame.2015.04.012. Christine Mounaïm-Rousselle, 0D modeling aspects of flame stretch in spark
[14] R. Borghi, On the structure and morphology of turbulent premixed flames. ignition engines and comparison with experimental results, Appl. Energy
Recent Advances in the Aerospace Sciences, Springer US, 1985, pp. 117e138. (2016) 401e412.
[15] C. Regis Bauwens, Jeffrey M. Bergthorson, Sergey B. Dorofeev, Experimental [41] Hideaki Kobayashi, Experimental study of high-pressure turbulent premixed
study of spherical-flame acceleration mechanisms in large-scale propaneeair flames, Exp. Therm. Fluid Sci. 26 (2e4) (2002) 375e387.
flames, Proc. Combust. Inst. 35 (2) (2015) 2059e2066. [42] Grunde Jomaas, C.K. Law, J.K. Bechtold, On transition to cellularity in
[16] Sergey A. Gaponov, About a flame propagation by a premixed gas mixture at expanding spherical flames, J. Fluid Mech. 583 (2007) 1e26.
high turbulence, 1, in: AIP Conference Proceedings, vol. 1939AIP Publishing [43] M. Moslemin Koupaie, A. Cairns, H. vafamehr, T. Lanzanova, Cyclically
LLC, 2018, 020036. resolved flame and flow imaging in an SI engine operating with future ethanol
[17] M. Matalon, On flame stretch, Combust. Sci. Technol. 31 (3-4) (1983) fuels, in: SAE Technical Paper 2017-01-0655, 2017, https://doi.org/10.4271/
169e181. 2017-01-0655.
[18] A.N. Lipatnikov, W.Y. Li, L.J. Jiang, S.S. Shy, Does density ratio significantly affect [44] John B. Heywood, Internal Combustion Engine Fundamentals, 1988.
turbulent flame speed? Flow, Turbul. Combust. 98 (4) (2017) 1153e1172. [45] Thomas Morel, Rifat Keribar, A Model for Predicting Spatially and Time
[19] G.K. Giannakopoulos, A. Gatzoulis, C.E. Frouzakis, M. Matalon, Resolved Convective Heat Transfer in Bowl-In-Piston Combustion Chambers,
A.G. Tomboulides, Consistent definitions of “flame displacement speed” and SAE Technical Paper, 1985. No. 850204.
“Markstein length” for premixed flame propagation, Combust. Flame 162 (4) [46] Behdad Afkhami, Yu Zhao, Miers Scott, Carbureted SI Engine Air Flow Mea-
(2015) 1249e1264. surements, SAE Technical Paper, 2016. No. 2016-01-1082.