Fuel 231 (2018) 396–403

Contents lists available at ScienceDirect

Fuel
journal homepage: www.elsevier.com/locate/fuel

Multi-zone thermodynamic modeling of combustion and emission formation T

in CNG engines using detailed chemical kinetics
⁎
Mirko Baratta, Alessandro Ferrari , Qing Zhang
Energy Department, Polytechnic University of Turin, Turin, Italy

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

Keywords: This study proposes an efficient and accurate methodology that combines detailed chemical kinetics with a
Natural gas engine validated multi-zone thermodynamic model to calculate the emissions of a spark-ignition, natural gas fueled
Chemical kinetics engine. The relative air–fuel ratio has been changed from 0.8 to 1.53, and the nominal brake mean effective
Thermodynamic model pressure has been varied from 0.2 MPa to 1.29 MPa.
Emission prediction
The multi-zone model in-cylinder pressure and temperature traces, as well as other experimental engine
quantities, have been considered as input data for the chemical submodel. The objective has been to reproduce
and interpret the measured engine-out data in order to obtain insight into the in-chamber combustion and
pollutant formation processes. The concentrations of nitric oxide, carbon monoxide, hydrogen, carbon dioxide
and hydrocarbon, as well as the oxygen concentration after ignition, have been compared with experimental
data and some of them have been compared with the results of conventional models.
The model results, based on detailed chemistry simulation, have been found to be in good agreement with
experimental data for all the species at engine exhaust, and a higher prediction capability than that obtained
through simplified reaction and chemical equilibrium methods has been shown. The influences of the un-
certainties in RAFRs and unburned mass fractions on the calculated results are discussed. The unburned gas
fraction, derived from the calculated and measured oxygen concentration at the engine exhaust, has been shown
to be a way of correcting the hydrocarbon emissions.

1. Introduction formation simulation in NG engines to assist experimental activities and

The emission performance of the internal combustion engine (ICE)
is driven by more and more challenging regulations. Owing to the
benefits of the “shale gas revolution” [1], natural gas (NG) has met an
increasing diffusion in ICEs. It is utilized either in the form of com-
pressed natural gas (CNG) or as liquefied natural gas (LNG). Compared
to conventional fuels, NG combustion causes fewer hydrocarbons (HC)
and CO2 emissions, due to the chemical structure of the fuel and the
absence of any evaporation phase [2]. The low carbon characteristics of
the fuel are known to restrain the formation of benzene, and thus ex-
tremely low particulate matter (PM) levels are emitted. In spark igni-
tion engines, NG can sustain higher compression ratios, owing to its
high knocking resistance; as a consequence, higher thermal NOx engine
out emissions are generated, but a lower carbon monoxide (CO) frac-
tion occurs at the engine exhaust [3–6]. However, the slower flame
propagation speed requires an advance of the spark timing, and this
reduces the port fuel injection engine performance [7].
A reliable diagnostic tool is essential for combustion and emission
support critical phenomena interpretation as well as new combustion
concept analyses.
Models that simulate the combustion process in IC engines can be
subdivided into single-zone models, multi-zone models and multi-di-
mensional models. The single-zone model is the simplest, since the in-
cylinder temperature and composition are considered homogeneous
(pure zero dimensional modeling). Thus, the performance of this model
is not accurate enough to quantify peak temperature values and engine
emissions. On the other hand, multi-dimensional models [8–10] usually
couple a turbulence model (this can be either a RANS or a LES model)
with a combustion model in order to simulate the flow and temperature
fields in the cylinder during an engine cycle. However, these methods
are cumbersome, even when standard RANS turbulence models are
applied; furthermore, the chemical processes are calculated by means of
a reduced reaction mechanism or by means of empirical laws.
Multi-zone based simulation codes are suitable for the diagnostics of
existing engines, for performing parameter studies and for predicting
optimum settings without resorting to complex multidimensional

⁎
Corresponding author.
E-mail address: alessandro.ferrari@polito.it (A. Ferrari).

https://doi.org/10.1016/j.fuel.2018.05.088
Received 5 December 2017; Received in revised form 14 May 2018; Accepted 15 May 2018
Available online 29 May 2018
0016-2361/ © 2018 Elsevier Ltd. All rights reserved.
M. Baratta et al.
models [11]. In particular, multi-zone models are believed to be sui-
table for homogeneous charge compression ignition (HCCI) engines,
since a subdivision of the chamber content into homogeneous zones is
more realistic.
A two-zone combustion model, which couples an adiabatic core
zone with a heat transfer model for the thermal boundary layer zone,
was presented in [12] for HCCI engines. This model was later extended
by adding a ring-pack zone (third zone) and was validated under turbo-
charged conditions, although no heat transfer between the core and
other zones was considered [13]. A multi-zone combustion model,
which included mass transfer between zones and implemented a re-
duced set of chemical reactions, was developed in [14] to assess the
effects of mass transfer on the formation of HC and CO emissions. In
[15], an “accelerated multi-zone model” that did not need to consider
the heat transfer was developed. The methodology calculated the
thermal stratification inside the cylinder on the basis of a computa-
tional fluid dynamics calculation for motored conditions. This model
was benchmarked against coupled thermo-fluid dynamics and chemical
kinetics calculations for a turbocharged (TC) gasoline HCCI engine and
showed satisfactory agreement over a wide range of operating condi-
tions, in terms of in-cylinder pressure and heat release rate predictions.
Neshat [16–18] proposed a new multi-zone combustion tool that im-
plemented an innovative heat transfer submodel and used not so de-
tailed chemical kinetics mechanism to simulate the combustion process
of n-heptane in an HCCI engine. This model contains a crevice zone, a
boundary layer zone, some outer zones and a core zone. Heat and mass
transfer between the zones are considered. The results are in line with
experimental data pertaining to the prediction of in-cylinder pressure,
NOx, CO and UHC emissions.
Bissoli has recently developed a predictive multi-zone model for
HCCI combustion [19], in which many factors that have a significant
impact on combustion and emissions are taken into account: turbu-
lence, heat and mass exchanges, crevices, residual burned gases,
thermal and fuel stratification. The model improves the description of
the mixture stratification phenomena by coupling a wall heat transfer
model with a suitable turbulence model. A general overestimation of
intermediate species is observed by comparing simulated with experi-
mental data.
Unlike HCCI engine models, in which the mixture simultaneously
ignites at multiple sites, the multi-zone model for conventional spark
ignition (SI) engines needs to consider both ignition and flame propa-
gation. Rakopoulos [20,21] and Asgari [22] developed a multi-zone
model for the prediction of nitric oxide emissions in SI engines fueled
with syngas and CNG, respectively. The super-extended Zeldovich
mechanism [23,24], which considers six reactions, has been employed
to calculate NO emissions [20–22], and the reaction rate model for CO
prediction, which is described in [25] has been used. However, these
NO and CO reaction mechanisms are simple and are isolated, and the
predictions are very sensitive to some calibration constants that should
be fitted to each engine working point.
In this paper, the combustion and emission formation processes of a
spark-ignition (SI) CNG engine are analyzed with a multi-zone model
that is coupled with a detailed chemical submodel for methane oxida-
tion. The matching of an accurate multi-zone tool with a detailed
chemical kinetics model represents a novelty in the literature on CNG
engines. The calculated results have been compared with the experi-
mental data as well as with the simulation results from previously de-
veloped thermodynamic models: the validation of the new model has in
particular been carried out on different species at the engine exhaust
(NO, CO, HC, O2, CO2, H2 and CH4). The objective has been to un-
derstand the role played by both thermodynamic modeling and che-
mical modeling in the simulation of combustion and emission formation
in CNG engines.
397
Fuel 231 (2018) 396–403
Table 1
Main specifications of the TC engine.
Bore 70.4 mm
Stroke 78.8 mm
Compression ratio 10.10
Cylinder number 4
Displacement 1242 cm3
2. Methodology
2.1. Experimental setup
The engine is a downsized turbocharged CNG engine for automotive
applications [26]. The main geometrical specifications are given in
Table 1. The engine has specifically been developed and optimized for
CNG fueling, even though it also features bi-fuel capability. In fact,
gasoline fueling, combined with a conservative spark timing strategy, is
only limited to back-up operations in the case of a lack of CNG. The
engine head features a disc-shape combustion chamber, two valves per
cylinder and one side-located spark plug [27]. Furthermore, a waste-
gate valve, which allows selectable levels of turbocharging to be settled,
controls the high-performance engine turbocharger. A Magneti Marelli,
multipoint, sequential-injection module that features optimized cali-
bration maps has been used.
The steady-state test cell is equipped with an eddy-current dynam-
ometer that can adsorb a maximum power of 150 kW and a maximum
torque of 300 Nm, and an encoder has been installed on the engine
crankshaft to measure the engine speed. The engine has been in-
strumented with a hot-film air-mass sensor, which has been installed
inside the intake system, two Coriolis mass-flow meters, installed in the
fuel distribution pipes to measure both gasoline and CNG consumption,
two air-fuel ratio universal exhaust gas oxygen (UEGO) sensors (one for
rich mixtures and the other for lean mixtures), installed in the exhaust
system, a pressure sensor, installed in the injection system rail, a hy-
grometer, installed in the intake system and thermocouples for mea-
suring the temperatures of the intake flow, of the fuel and of the ex-
haust gases, and a water-cooled piezoelectric transducer, for taking
accurate pressure time-histories within the combustion chamber. In
addition, the bench is equipped with a multipurpose exhaust gas ana-
lyzer that can measure the THC (total hydrocarbons), MHC (methane
hydrocarbons), NO, CO, CO2, and O2 levels in the exhaust gases of
engines running on gasoline, diesel or alternative fuels (such as lique-
fied petroleum gas or CNG). All of the measured data are collected by a
data acquisition system.
The accuracy of the measured flow-rates is 1.5% for CNG and 3% for
air. The raw exhaust gas analyzers have a relative accuracy of 1% of the
full-scale range, whereas span gases are provided with a 2% error of
their nominal concentration.
2.2. Multi-zone model
The multi-zone model is based on the application of the first law of
thermodynamics as well as of the perfect gas law to the combustion
chamber content between inlet valve closing and exhaust valve
opening. The diagnostic model requires the pressure time-history,
measured in the SI engine cylinder, as a boundary condition. Fig. 1
reports a schematic of the engine where the combustion chamber vo-
lume is divided into an unburned gas zone (Vu) and a burned gas region
(Vb), which here has been split into six ‘zones’ as an example. The
numbers in the picture show each of the burned zones. In real simu-
lations, a burned zone is generated each 5° crank angle (CA), and thus
15–20 burned zones (the exact number depends upon the combustion
duration) are generally generated at the end of combustion (EOC). The
following equation is obtained by constraining the sum of the volumes
of the unburned region (Vu) and of the burned zones (Vb,i, where
M. Baratta et al. Fuel 231 (2018) 396–403

R3
N+ OH → NO + H (8)
R4
H+ N2 O ⎯→
⎯ N2 + OH (9)
R5
O+ N2 O → N2 + O2 (10)
R6
O+ N2 O ⎯→
⎯ NO + NO (11)

The R1–R6 reaction rates, used to solve this equation system, were
taken from [24]. The thermal mechanism was coupled with the NO
prompt formation model proposed by Fenimore in [28]. The influence
of the prompt NOx formation mechanism was studied in a previous
work [29] for both CNG and gasoline. The results for CNG show that the
additional implementation of the prompt mechanism in the NO for-
mation model leads to a prediction improvement, with respect to the
experimental NO levels, in both the rich and the lean mixture fields.
The CO formation submodel is based on the following simplified
mechanism by Bowman:
R7
CO + OH → CO2 + H (12)
R8
CO2 + O⎯→
⎯ CO + O2 (13)

Reaction rates R7 and R8 are taken from [25].

Specific information about the procedures for the start-of-combus-
tion detection, EOC detection, heat release model, heat loss submodel
and emission formation submodels can be found in [26,29,30].

2.3. Detailed chemical submodel and matching with the multi-zone model

An understanding of the natural gas combustion process and of the

Fig. 1. Burned (Vb) and unburned (Vu) region volumes in the engine combus-
formation of pollutant emissions requires an appropriate detailed me-
tion chamber.
chanism of the methane and of the higher-order alkanes, such as ethane
and propane. The GRI-Mech 3.0 [31] used in this paper consists of 325
subscript i ranges from 1 to n) in order to be equal to the instantaneous element reactions and includes 53 species. The detailed mechanism
cylinder volume (V): includes a reaction scheme to model natural gas combustion, including
n NOx and CO formation, as well as the reburn chemistry. It is the most
dVu + ∑ dVb,i = dV widely used mechanism for natural gas, and has been validated by
i=1 (1)
means of a comparison with several experimental data [32–34].
The mass conservation law, applied to the cylinder content, yields: The composition of the CNG used in a set of the tests performed in
this research investigation is listed in Table 2 as an example. It is nei-
d (mf + ma + mr ) = dmu + dmb,n = 0 (2) ther possible nor necessary to use the exact composition in the detailed
where the subscripts f, a, r, u and b refer to the fuel, air, residual gas, chemistry submodel. The alkanes with four or more carbons have been
and the unburned and burned zones, respectively. Furthermore, the replaced by methane, ethane and propane with an appropriate per-
energy conservation equation for the unburned zone and for each centage to reproduce the effective average stoichiometric ratio and H/C
burned zone i can be written as ratio.
Each zone is considered as a 0-D homogenous batch reactor, and
−q u Au dt + Vu dp = mu diu (3) thus, the detailed chemical kinetics model is based on the assumption
−q b,i Ab,i dt + Vb,i dp = mb,i dib,i that the mixture is homogenous. The generation process in each burned
(4)
zone is described, starting from the inflow, the composition of which is
The terms qb,iAb,i and quAu express the moduli of the global heat
transfer from the considered zone (u or i) to the adjacent zones and to Table 2
the combustion chamber walls, while iu and ib,i are the specific enthalpy Composition and average properties of CNG.
of the corresponding zones. Moreover, cylinder pressure p is given by Species Actual volume fraction Adjusted volume fraction
relation [%] [%]
n
Methane 88.33 88.33
pV = mu Ru Tu + ∑ mb,i Rb,i Tb,i
Ethane 6.30 5.81
i=1 (5)
Propane 1.38 2.15
where Ru and Rb,i are the constants of the unburned gas and of the Iso-butane 0.19 0
Iso-pentane 0.05 0
burned gas belonging to zone i, respectively. Hexane and higher 0.04 0
The super-extended Zeldovich mechanism, which is used to calcu- Nitrogen 2.35 2.35
late NOx emissions, is given by the following chemical equations: Carbon dioxide 1.32 1.32
Helium 0.04 0.04
R1
N2 + O→ NO + N (6) Average properties
Stoichiometric ratio 15.86 15.86
R2 H/C ratio 3.81 3.81
N+ O2 → NO + O (7)

398
M. Baratta et al. Fuel 231 (2018) 396–403

identical to the composition of the unburned zone. Then, the reactant 12

composition at the beginning of each time step is calculated as the Zeld. model
mass-weighted average between the products at the end of the previous 10 Chem. model

NO mole fraction( ×10-3)

time step and the inflow. This method is realized by combining the
multi zone model, written in C language, with the CHEMKIN com- zone1
8
mercial software, which implements the reaction mechanism and solves
it during the 330–450 CA deg phase.
6 zone3
In the CHEMKIN tool, the in-cylinder pressure p, the temperature Ti
of each zone i, the burned mass fraction fb,i that is variable with time
and the composition of the unburned mixture are required as input data 4 zone5
for the calculus of the chemical combustion process and of the emis- zone9 zone11
zone7
sions. The first three parameters can be obtained from the multi-zone 2
thermodynamic model code, adopting a temporal resolution of 0.1 °CA.
Instead, the composition of the unburned mixture is calculated on the
0
basis of the local experimental composition of air (including humidity), 320 340 360 380 400 420 440
of the CNG composition at the considered relative air-fuel ratio (RAFR)
Crank angle (deg)
and of the residual gas composition and fraction.
The RAFR values used in this calculation have been obtained from Fig. 2. NO versus crank angle for the different zones.
the water-gas reaction method rather than from the universal exhaust
gas oxygen (UEGO) sensor or from the directly measured air and fuel 0.20
mass flow rates [35]. The selected method is in fact expected to provide fb
higher accuracy in the evaluation of the RAFR value. (Ti)max 2900
0.16

Maximum temperature (K)

3. Results and discussion
The chemical model has been developed to predict the emission
formation of the tested engine for a wide range of brake mean effective
pressure values (0.2 MPa ≤ BMEP ≤ 1.29 MPa) and relative air-fuel
ratios (0.8 ≤ RAFR ≤ 1.53). The model results on emissions are com-
pared with data obtained from simplified reaction schemes that have
been implemented in the original multi-zone tool and with measured
values obtained from the emission analyzer. The plotted results refer to
two sets of tests: a load sweep at engine speed N = 2000 rpm and
RAFR ≈ 1 (Case A) and a RAFR sweep at N = 3000 rpm and
BMEP = 0.585 MPa (Case B). Table 3 shows the operating conditions
and the tested fuel composition.
3.1. NO formation
In the framework of Case A, Fig. 2 shows the NO formation process,
calculated by means of both the detailed chemistry model (Chem.
model) and the conventional super-extended Zeldovich model (Zeld.
model) [24], under the condition RAFR = 1.016, BMEP = 0.385 MPa.
Furthermore, Fig. 3 reports the maximum temperatures and the final
mass fractions of the burning zones according to the multi-zone model.
The super-extended Zeldovich model is isolated, which means the
NO level only depends on several related species, and some of these
species are assumed to be in an equilibrium state. Instead, almost all the
species are connected in the chemical model, although some species are
not closely related to NO production. Moreover, some of the reaction
rate parameters are different, because the super-extended Zeldovich
model is specially designed for NO prediction. This means it is possible
that the reaction rate constant in the simplified model may not be the
true one for the specified reaction, but it is considered as the best value
to achieve a good result.
A peak concentration of NO can be observed in Fig. 2, and the NO
Table 3
Operating conditions and fuels.
Case A Case B
Fuel composition 99.75% CH4, 0.25% CO2 (volume %) CNG in Table 2
Engine speed N (rpm) 3000 2000
RAFR close to 1 0.8–1.53
BMEP (MPa) 0.2–1.29 close to 0.6
Mass fraction
0.12 2800
0.08
2700
0.04
2600
0.00
1 2 3 4 5 6 7 8 9 10 11 12
Zone mumber
Fig. 3. Mass fraction and maximum temperature of the zones.
concentration reaches an asymptotic value. The pattern of the NO
versus crankshaft angle distribution for each zone has been verified to
be similar to that of the corresponding temperature versus crankangle
curve. The final NO concentration (detected at 450 CA) for each zone
correlates well with the maximum temperature in Fig. 3: the higher the
NO value at 450 CA, the higher the maximum temperature. As a con-
sequence, the NO level at 450 CA reduces as the zone number increases.
By combining the NO concentrations in Fig. 2 with the mass burned
fractions in Fig. 3, it results that most of the NO is produced from zone 4
to zone 90.
In general, the NO predictions of the chemical and the Zeldovich
models are similar. One difference between the two models is that the
NO value predicted for the first zones by the detailed chemistry model
is lower than that predicted by the conventional model and the opposite
situation exists for the last zones. Nevertheless, the difference between
the two models is less than 300 ppm for each zone and is therefore not
so significant.
Fig. 4 compares the global NO production calculated by the detailed
chemical model, by the Zeldovich method and by a chemical model that
uses the equilibrium data of zone 7 (Equil. model), which is the most
representative zone because features the maximum burned mass frac-
tion. The temperature profile, evaluated with the multi-zone tool and
used in the Chem. model case as input datum for the chemical kinetics,
is also reported.
The Equil. model is the most sensitive to temperature variations. A
NO concentration peak value is in fact reached at 370 deg, even though
with a certain delay compared to the peak temperature crankangle
value, and the NO concentration then drops quickly as the in-cylinder
temperature decreases. On the other hand, the NO concentration,

399
M. Baratta et al. Fuel 231 (2018) 396–403

6 3000 4.0 1.04 2900

Chem. GRI 3.0 Exp. Zeld.
Zeld. model
3.5 Tmax Chem.
1.03 2850
5 RAFR
2500
NO mole fraction(×10-3)

Equil. model 3.0

NO mole fraction(×10-3)
T 1.02 2800

Temperature (K)
4
2000 2.5
1.01 2750

RAFR

Tmax
3 2.0
1500 1.00 2700
1.5
2
1.0 0.99
2650
1000
1 0.5 0.98
11 2600
0 500 0.0
360 380 400 420 440 0.0 0.4 0.8 1.2 1.6
BMEP (MPa)
Crank angle (deg)
Fig. 6. NO versus BMEP distributions (RAFR ≈ 1).
Fig. 4. NO concentration and temperature distributions (BMEP = 0.385 MPa).

Mech. 3.0 model results are instead satisfactory for BMEP ≤ 0.93 MPa).
calculated by means of the detailed chemical model, becomes flat as the
Theoretically, the NO emissions are related to both Tmax and RAFR.
temperature decreases for higher crank angle values than 380 deg,
However, in Fig. 6, it seems that the experimental NO value is related
owing to frozen decomposition. The same occurs for the Zeldovich
more to the in-cylinder max temperature (NO rises with Tmax), while the
model.
calculated results tend to be very sensitive to the RAFR trend under
The addition of the Equil. model results to the graph shows that the
near-stoichiometric conditions. This means that the actual NO emission
simplified Zeldovich mechanism provides the same results as the GRI-
largely depends on the in-cylinder temperature and that the model
Mech 3.0 model, even though the kinetics of reactions is the pre-
overestimates the influence of RAFR on the NO emissions in the
dominant aspect (this is highlighted by the differences between either
neighborhood to the left of the-stoichiometric ratio condition (RAFR →
the Zeldovich or GRI Mech. 3.0 model and the Equil. model).
1−). Therefore, the chemical model is also very sensitive to the in-
Fig. 5 shows a comparison between the NO emissions obtained from
accuracy in the RAFR measurement within this slightly rich zone (the
experimental measurements, with the Zeldovich model and with the
uncertainties of the RAFR measurement method of the air-to-fuel ratio
detailed chemical model, for Case B. A satisfactory agreement can
could therefore also be responsible for the misleading prediction of the
generally be observed between the experimental and predicted NO le-
NO emissions).
vels for the indicated operating conditions. In general, both the fitted
The Zeldovich model results at BMEP = 1.16 MPa and 1.29vMPa
Zeldovich model and the detailed chemical model can be considered as
show the same RAFR trend as those of the detailed chemical model.
good tools for engine emission diagnosis and prediction, as they both
However, the Zeldovich model seems to be related less to RAFR than
denote a high capability of correctly capturing the pollutant trend with
the chemical model (cf. also point BMEP = 0.93 MPa) in the neigh-
respect to RAFR (NO emissions are very sensitive to the relative air-to-
borhood to the left of RAFR = 1.
fuel ratio). However, it can be noted that the detailed chemistry model
gives slightly better results than the fitted Zeldovich model for fuel rich
conditions. 3.2. CO formation
Fig. 6 shows the NO emissions and the RAFR values, which are close
to 1, and maximum gas temperature during the engine cycle as func- Fig. 7 reports the level of CO at engine exhaust as a function of
tions of BMEP. The experimental NO emissions are compared with RAFR for the indicated engine operating conditions: these conditions
those predicted by means of either the Zeldovich model or the detailed are related to Case B (a) and to working point BMEP = 4.4 bar and
chemical models for Case A. The calculation gives an accurate predic- N = 4600 rpm (b), which is quoted as Case VI in [30]. The same multi-
tion of the NO level for some RAFR values, but the NO value predicted zone thermodynamic model is used for the simulation tests in Fig. 7a
by means of the GRI-Mech. 3.0 model deviates from the experimental and b. However, a detailed CO chemical model is used in Fig. 7a,
measurements at BMEP = 1.16 MPa and at BMEP = 1.29 MPa (the GRI- whereas a simplified reaction mechanism, taken from Bowman [25,36],
and in which several species (i.e. H, OH, O, O2 and CO2) are assumed to
4 achieve chemical equilibrium, is used in Fig. 7b. As far as the CO ex-
haust levels in Fig. 7b are concerned, the Bowman submodel generally
Exp. tends to underestimate the measured CO level. The discrepancy can be
Zeld. model
NOx mole fraction(×10-3)

ascribed to the CO that derives from the HC oxidation path, which exists
3 Chem. model
in the real engine combustion process, but is not simulated in the
Bowman mechanism. The detailed methane reaction mechanism ac-
counts for the reaction path from HC to CO, and this makes it more
2 consistent with the CO experimental data.
The most representative zone of Case B at RAFR = 0.9, i.e. zone 7
with fb ≈ 19%, has been selected in order to analyze the CO formation
1 process and prove the contribution of HC to the CO path. Fig. 8 shows
the rate of production (ROP) of CO as soon as zone 7 becomes active.
Reactions #99, #284 and #166-168 of the chemical model (these re-
0 actions are reported in the graph) are the most important reactions that
0.8 1.0 1.2 1.4 1.6 contribute to CO consumption and formation, respectively (the sum of
RAFR the CO contribution from these reactions accounts for more than 60% of
the total CO ROP). According to [36–38], the formation and decom-
Fig. 5. NO versus RAFR distributions (BMEP = 0.585 MPa). position of CO is mainly controlled by dissociation reaction #31 (cf.

400
M. Baratta et al. Fuel 231 (2018) 396–403

6 3.3. H2, O2, CO2, CH4 and HC emissions

Case B Exp.
5 One of the advantages of adding the detailed chemistry model to the
Chem. model
multi-zone thermodynamic model is that information about all the
CO mole fraction (×10-3)

4 chemical species can be obtained with the simulation. This not only
provides a more comprehensive simulation of the combustion process,
3 but can also be used to guide a refinement of the multi-zone thermo-
dynamic model through a comparison of the numerical outcomes of the
2 engine exhaust species with the corresponding experimental data.
Fig. 9 plots the exhaust level of H2 as a function of RAFR for Case B.
Hydrogen is a very important element, since it increases the flame
1
speed in the gas mixture. The H2 concentration, calculated by means of
the detailed chemical model, shows a satisfactory agreement with that
0
0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6 derived from experimental data, and the CO that is sensitive to RAFR is
in fact reproduced accurately.
RAFR Fig. 10 shows the variations in the exhaust level of O2 with RAFR for
(a) Case B. The numerical results match well with the experimental ones
6 for RAFR > 1. Both the experimental data and the calculations show a
Case VI in Ref. [30] non-negligible O2 concentration in the fuel-rich mixtures. By calcu-
Exp. lating the final O2 concentration of each burned zone, it has been found
5
Simp. model that the O2 molar fractions in all the burned zones are below 10 ppm
CO mole fraction ( ×10-3)

when RAFR ≤ 0.9 for the considered in-cylinder temperature and

4
pressure values. The remaining amount of O2 is due to the unburned
zone of the mixture, and most of the O2 therefore comes from this zone.
3
In real engines, the unburned region in the engine exhaust process is
often caused by the gas close to the combustion chamber walls or is due
2 to the early opening of the exhaust valve.
The difference between the experimental and numerical O2 results
1 in Fig. 10 (RAFR < 1) can be ascribed to an overestimation of the un-
burned mass fraction in the multi-zone model. The unburned mass
0
0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6
40
RAFR
Exp.
(b) 35
Chem. model
Fig. 7. CO versus RAFR distributions. The simulation methods are the detailed 30
H2 mole fraction(×10-3)

chemical kinetics (a) and conventional method (b). 25

20
Fig. 8) and #99 in both rich and lean conditions; but it has been stated
[38] that reaction #31 is only remarkable for high temperatures. In the 15
fuel rich working conditions of Fig. 8, dissociation reaction #31 is
10
negligible, even though the zone temperature reaches 2500 K.
In short, the simplified Bowman mechanism for the CO prediction of 5
methane engines needs to be optimized: the effect of reactions #166-
0
168 (HCO to CO) and #284 should be included and tuned. 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6
RAFR
Fig. 9. H2 versus RAFR distributions.

70
Exp.
60 Chem. model

50
O2 mole fraction(×10-3)

40

30

20

10

Fig. 8. The rate of production of CO (RAFR = 0.9). 0

0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6
RAFR
Fig. 10. O2 versus RAFR distributions.

401
M. Baratta et al. Fuel 231 (2018) 396–403

through Eq. (14) is applied to the calculation of HC and CH4 for

Exp.
80 RAFR ≤ 1, the numerical results (case Chem. mod. fu) show a sa-
Cal.
tisfactory agreement for CH4 and total HC at fuel-rich working condi-
CO2 mole fraction(×10-3)

tions, and reduce the gap with experimental data at RAFR = 1.

60

40 4. Conclusions

A chemical model, based on a detailed reaction mechanism of me-

20
0 better explanation of the generation of the species which are detected at
0.7 0.8 0.9 1.0 1.1 1.2 1.3
RAFR
Fig. 11. CO2 versus RAFR distributions.
fraction is obtained through an EOC detection procedure that is in-
cluded in the multi-zone model [26]. This procedure detects the end of
combustion as the crank angle at which the heat release reaches a value
that corresponds to an appropriately calibrated percentage of the
overall heat release and thus to a suitable value of the combustion ef-
ficiency. This method is accurate enough for the prediction of the heat
release and emissions of NO gases, but does not give the right input for
those gases that are sensitive to the unburned mass fraction.
A point is therefore to try to optimize the unburned gas submodel in
the multi-zone model when fuel-rich working conditions are con-
sidered. The optimal values of the unburned gas fraction are here
provided for the case in Fig. 10:
[O2]Exp
fu′ = ·fu RAFR ⩽ 1
[O2]Cal
Fig. 11 shows the exhaust level of CO2 as a function of RAFR for
Case B. The CO2 concentration, calculated by means of the chemical
model, generally shows a good agreement with the experimental data.
The differences increase slightly when RAFR ≥ 1.1, and a maximum
percentage error of 5% is reached at RAFR = 1.414.
Fig. 12 shows the CH4 and HC molar fractions at the engine exhaust.
It is generally recognized that the hydrocarbon concentration at the
EOC, especially for RAFR ≤ 1, is dominated by the unburned mass
fraction of a mixture in a CNG engine. Discrepancies can be observed
between the experimental data and the simulation ones that are based
on the original unburned mass fractions which are worked out by the
multi-zone model and which are used in the chemical model (Chem. ori.
fu in Fig. 12). If the adjusted unburned mass fraction worked out
thane, has been developed as a supplementary tool in order to improve
a multi-zone thermodynamic model for CNG engines and to provide a
1.4 1.5 1.6 the engine exhaust.
The Zeldovich mechanism has been confirmed to be very accurate,
and this proves that the other elementary reactions in the detailed
methane mechanism are not fundamental for a precise prediction of the
NO at the engine exhaust, when a multi-zone model of combustion is
applied. It is essential to take into account the non-equilibrium condi-
tions during the expansion stroke which lead to frozen chemical de-
composition: models of NO based on equilibrium relations do not lead
to a satisfactory performance.
The model calculated NO concentration is closely related to the
RAFR variations near the stoichiometric ratio condition, while the ex-
perimental data are related more to the maximum temperature trend; a
consequent deviation between the numerical predictions and experi-
mental results occurs for high BMEP values.
The calculated exhaust level of CO is highly consistent with the
experimental data, and leads to a significantly improved prediction
compared to that of the standard methods (such as the Bowman model)
based on simplified reaction mechanisms and chemical equilibriums.
The elementary path from HCO to CO contributes to most of the CO at
the engine exhaust, while a dissociation reaction of CO2 has been
(14)
proved not to be fundamental for fuel-rich methane combustion.
The calculations performed with the detailed chemical model show
a generally satisfactory agreement with the experimental data of the H2,
O2 and CO2 species. The deviation of the O2 concentration at
RAFR < 1, compared to the experimental values, is due to inaccuracies
in the evaluation of the unburned mass fraction in the multi-zone model
(the multi-zone model calibration procedure is in fact oriented toward
optimizing the prediction of the NO emissions and of the energy bal-
ance). Similarly, the level of hydrocarbons at the engine exhaust de-
pends on the unburned mass fraction, and its prediction has been shown
to be slightly inaccurate under fuel-rich conditions. The adjusted un-
burned mass fraction, which was obtained with a simple scaling based
on the ratio of the calculated chemical model O2 concentration to the
experimental one, leads to a better prediction of HC and CH4 under fuel-

6 6
Exp.
Exp.
5 5 Chem. ori. f u
Chem. ori. f u
Chem. mod. fu
HC mole fraction(×10-3)

Chem. mod. fu
CH4 mole fraction(×10-3)

4 4

3 3

2 2

1 1

0 0
0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6 0.7 0.8 0.9 1.0 1.1 1.2 1.3 1.4 1.5 1.6
RAFR RAFR
(a) (b)
Fig. 12. Methane (a) and HC (b) concentrations versus RAFRs; data of unburned mass fraction taken from the EOC procedure and Eq. (14).

402
M. Baratta et al.
rich conditions.
References
[1] Aruga K. The US shale gas revolution and its effect on international gas markets. J
Unconven Oil Gas Resour 2016;14:1–5.
[2] El-Sherif AS. Effects of natural gas composition on the nitrogen oxide, flame
structure and burning velocity under laminar premixed flame conditions. Fuel
1998;77(14):1539–47.
[3] Rousseau S, Lemoult B, Tazerout M. Combustion characterization of natural gas in a
lean burn spark-ignition engine. Proc Inst Mech Eng D J Automob Eng
1999;213(5):481–9.
[4] Korakianitis T, Namasivayam AM, Crookes RJ. Natural-gas fueled spark-ignition
(SI) and compression-ignition (CI) engine performance and emissions. Prog Energy
Combust Sci 2011;37(1):89–112.
[5] Kobayashi K, Sako T, Sakaguchi Y, Morimoto S, Kanematsu S, Suzuki K, et al.
Development of HCCI natural gas engines. J Nat Gas Sci Eng 2011;3(5):651–6.
[6] Kakaee AH, Paykani A, Ghajar M. The influence of fuel composition on the com-
bustion and emission characteristics of natural gas fueled engines. Renew Sustain
Energy Rev 2014;38:64–78.
[7] Cracknell R, Prakash A, Head R. Influence of laminar burning velocity on perfor-
mance of gasoline engines. SAE Technical Paper 2012–01-1742, 2012.
[8] Yousefzadeh A, Jahanian O. Using detailed chemical kinetics 3D-CFD model to
investigate combustion phase of a CNG-HCCI engine according to control strategy
requirements. Energy Convers Manage 2017;133:524–34.
[9] Benajes J, Novella R, Pastor JM, Hernández-López A, Hasegawa M, Tsuji N, et al.
Optimization of the combustion system of a medium duty direct injection diesel
engine by combining CFD modeling with experimental validation. Energy Convers
Manage 2016;110:212–29.
[10] Mattarelli E, Rinaldini CA, Golovitchev VI. CFD-3D analysis of a light duty Dual
Fuel (Diesel/Natural Gas) combustion engine. Energy Procedia 2014;45:929–37.
[11] Verhelst S, Sheppard CGW. Multi-zone thermodynamic modelling of spark-ignition
engine combustion–an overview. Energy Convers Manage 2009;50(5):1326–35.
[12] Fiveland SB, Assanis DN. Development of a two-zone HCCI combustion model ac-
counting for boundary layer effects. SAE Technical paper 2001–01-1028, 2001.
[13] Fiveland SB, Assanis DN. Development and validation of a quasi-dimensional model
for HCCI engine performance and emissions studies under turbocharged conditions.
SAE Technical Paper 2002–01-1757, 2002.
[14] Komninos NP. Assessing the effect of mass transfer on the formation of HC and CO
emissions in HCCI engines, using a multi-zone model. Energy Convers Manage
2009;50(5):1192–201.
[15] Kodavasal J, McNenly MJ, Babajimopoulos A, Aceves SM, Assanis DN, Havstad MA,
et al. An accelerated multi-zone model for engine cycle simulation of homogeneous
charge compression ignition combustion. Int J Engine Res 2013;14(5):416–33.
[16] Neshat E, Saray RK. Effect of different heat transfer models on HCCI engine simu-
lation. Energy Convers Manage 2014;88:1–14.
[17] Neshat E, Saray RK. Development of a new multi zone model for prediction of HCCI
(homogenous charge compression ignition) engine combustion, performance and
emission characteristics. Energy 2014;73:325–39.
[18] Neshat E, Saray RK, Parsa S. Numerical analysis of the effects of reformer gas on
supercharged n-heptane HCCI combustion. Fuel 2017;200:488–98.
[19] Bissoli M, Frassoldati A, Cuoci A, Ranzi E, Mehl M, Faravelli T. A new predictive
multi-zone model for HCCI engine combustion. Appl Energy 2016;178:826–43.
403
Fuel 231 (2018) 396–403
[20] Rakopoulos CD, Michos CN. Development and validation of a multi-zone combus-
tion model for performance and nitric oxide formation in syngas fueled spark ig-
nition engine. Energy Convers Manage 2008;49(10):2924–38.
[21] Rakopoulos CD, Michos CN, Giakoumis EG. Availability analysis of a syngas fueled
spark ignition engine using a multi-zone combustion model. Energy
2008;33(9):1378–98.
[22] Asgari O, Hannani SK, Ebrahimi R. Improvement and experimental validation of a
multi-zone model for combustion and NO emissions in CNG fueled spark ignition
engine. J Mech Sci Technol 2012;26(4):1205–12.
[23] Lavoie GA, Heywood JB, Keck JC. Experimental and theoretical study of nitric oxide
formation in internal combustion engines. Combust Sci Technol 1970;1(4):313–26.
[24] Miller R, Davis G, Lavoie G, Newman C, Gardner T. A super-extended Zel'dovich
mechanism for NOx modeling and engine calibration. SAE Technical Paper 980781,
1998.
[25] Bowman CT. Kinetics of pollutant formation and destruction in combustion. Prog
Energy Combust Sci 1975;1(1):33–45.
[26] Baratta M, Misul D. Development and assessment of a new methodology for end of
combustion detection and its application to cycle resolved heat release analysis in
IC engines. Appl Energy 2012;98:174–89.
[27] d'Ambrosio S, Misul D, Spessa E, Vassallo A. Evaluation of combustion velocities in
bi-fuel engines by means of an enhanced diagnostic tool based on a quasi-dimen-
sional multizone model. SAE Technical Paper 2005–01-0245, 2005.
[28] Fenimore CP. Formation of nitric oxide in premixed hydrocarbon flames.
Symposium on Combustion. Elsevier 1971;13(1):373–80.
[29] Catania AE, Misul D, Spessa E, Vassallo A. A diagnostic tool for the analysis of heat
release, flame propagation parameters and NO formation in SI engines (SI Engines,
Combustion Diagnostics). Proceedings of the International symposium on diag-
nostics and modeling of combustion in internal combustion engines, vol. 6.
Japanese Society of Mechanical Engineers; 2004. p. 471–86.
[30] Baratta M, Catania AE, d’Ambrosio S, Spessa E. Prediction of combustion para-
meters, performance, and emissions in compressed natural gas and gasoline SI
engines. J Eng Gas Turbines Power 2008;130(6):062805.
[31] Smith GP, Golden DM, Frenklach M, Moriarty NW, Eiteneer B, Goldenberg M,
Bowman CT, Hanson RK, Song S, Gardiner, Lissianski VV, and Qin Z. The GRI-Mech
3.0. Model for NG Combustion, Berkeley University, California, http://www.me.
berkeley.edu/gri_mech/.
[32] Liu CY, Chen G, Sipöcz N, Assadi M, Bai XS. Characteristics of oxy-fuel combustion
in gas turbines. Appl Energy 2012;89(1):387–94.
[33] Yousefi A, Birouk M. Investigation of natural gas energy fraction and injection
timing on the performance and emissions of a dual-fuel engine with pre-combustion
chamber under low engine load. Appl Energy 2017;189:492–505.
[34] Van Essen VM, Sepman AV, Mokhov AV, Levinsky HB. The effects of burner sta-
bilization on Fenimore NO formation in low-pressure, fuel-rich premixed CH4/O2/
N2 flames. Proc Combust Inst 2007;31(1):329–37.
[35] d’Ambrosio S, Finesso R, Spessa E. Calculation of mass emissions, oxygen mass
fraction and thermal capacity of the inducted charge in SI and diesel engines from
exhaust and intake gas analysis. Fuel 2011;90(1):152–66.
[36] Valério M, Raggi K, Sodré JR. Model for kinetic formation of CO emissions in in-
ternal combustion engines. SAE Technical Paper 2003–01-3138, 2003.
[37] Ramos JI. Internal combustion engine modeling. New York: Hemisphere publishing
corporation; 1989.
[38] Arsie I, Pianese C, Rizzo G. Models for the prediction of performance and emissions
in a spark ignition engine-a sequentially structured approach. SAE Technical Paper
980779, 1998.