International Journal of Hydrogen Energy 89 (2024) 1310–1320

Contents lists available at ScienceDirect

International Journal of Hydrogen Energy

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

Development and validation of a predictive combustion model for

hydrogen-fuelled internal combustion engines
Andrea Piano a,*, Gianpaolo Quattrone a , Federico Millo a , Francesco Pesce b, Alberto Vassallo b
a
Politecnico di Torino, Energy Department, Torino, Italy
b
Dumarey Automotive Italia SpA, Torino, Italy

A R T I C L E I N F O A B S T R A C T

Handling Editor: Dr F Gallucci Internal combustion engines (ICEs) fuelled with hydrogen can play a major role in the short-term future
transportation sector since they abate all criteria pollutants at engine-out reducing tailpipe CO2 emissions to
Keywords: near-zero levels. However, optimizing hydrogen ICEs is a challenging task that can be addressed through the
Hydrogen development of a robust simulation tool capable to predict the H2 combustion process. In this study, a previously
Internal combustion engine
developed two-zone combustion model has been updated considering different laminar flame speed computa­
Alternative fuels
tions, both based on a detailed chemistry scheme: a polynomial correlation function and a tabulated approach.
Modelling
The predictive capabilities of the combustion model have been validated against experimental data coming from
a 0.5L PFI single-cylinder engine under several operating conditions. The tabulated approach for laminar flame
speed definition proved to be the best solution, leading to a combustion duration average error lower than 3 deg
over a dataset containing more than 45 different operating conditions.

1. Introduction losses typical of unthrottled operations [8]. Several studies in literature

[9,10] show that hydrogen has a good combustion stability even with
In order to achieve the highly ambitious carbon emission targets set high excess air ratios, retaining the coefficient of variation of the indi­
within the framework of the Net Zero Emissions by 2050 (NZE) scenario cated mean effective pressure (COV IMEP) lower than 3%. In addition,
[1], several OEMs in the transportation sector have introduced a range mixture enleanment is beneficial for reducing the NOx emission to
of innovative solutions. Besides the transition to battery electric vehicles extremely low levels [11], thanks to the low in-cylinder temperatures.
(BEVs) and hydrogen fuel cells (FCEVs) for light-duty applications, the Lean combustion strategies are commonly employed even at high loads,
utilization of low-emissions fuels such as biofuels, e-fuels, and hydrogen allowing to achieve higher levels of brake mean effective pressure
in internal combustion engines is gaining significant momentum [2,3]. (BMEP) without incurring in abnormal combustion events, compared to
Among them, hydrogen can be produced using renewable energy sour­ stoichiometric conditions [12]. However, despite the high auto-ignition
ces [4], and its complete combustion does not produce GHG emissions temperature and the high Research Octane Number (RON) [13], hot
due to its zero carbon content. Moreover, hydrogen has physical pecu­ spots or residuals in the combustion chamber can easily ignite
liarities that make its usage as a fuel for ICEs particularly appealing [5]. hydrogen-air mixtures due to their low minimum ignition energy (at
In particular, it is noteworthy that hydrogen is characterized by a higher stoichiometry, 0.02 mJ compared to 0.25 mJ of Gasoline [13]). In light
Lower Heating Value (LHV) in comparison with the traditional fossil of all the peculiarities that characterize the air-hydrogen mixtures
fuels commonly used for internal combustion engines (i.e., H2 = 120 combustion, the development of hydrogen-fuelled internal combustion
MJ/kg vs Gasoline = 44.1 MJ/kg [6]). Moreover, hydrogen features engine is gaining high interest, moving the focus of the OEMs towards
high flame speed propagation over a wide range of temperatures, the development of such concepts. In this framework, the availability of
pressure, and air-to-fuel ratios [7], which makes it particularly suitable combustion models able to predict the average pressure cycle for pre­
for lean combustion strategies, extending the usual flammability limits. mixed hydrogen-air mixtures can be extremely important to support the
Lean combustion strategies are indeed beneficial for engine thermal engine development process. To this aim, several predictive models have
efficiency, due to the reduced heat rejection [2] and the low pumping been proposed in literature. D’Errico et al. [14] presented a

* Corresponding author.
E-mail address: andrea.piano@polito.it (A. Piano).

https://doi.org/10.1016/j.ijhydene.2024.09.407
Received 6 August 2024; Received in revised form 19 September 2024; Accepted 28 September 2024
Available online 4 October 2024
0360-3199/© 2024 The Authors. Published by Elsevier Ltd on behalf of Hydrogen Energy Publications LLC. This is an open access article under the CC BY license
(http://creativecommons.org/licenses/by/4.0/).
A. Piano et al. International Journal of Hydrogen Energy 89 (2024) 1310–1320

Table 1
Main characteristics of the single cylinder engine.
Bore 83 mm
Stroke 90.4 mm
Displacement 0.5 L
CR 12
Piston Type Hemi-spherical bowl
Fuel System PFI with rail with relative feed pressure 6–8 bar

Table 2
Limits of the main operating parameters varied throughout the experimental
campaign.
Parameter Full Load Partial Load

Lower Upper Lower Upper

Limit Limit Limit Limit

Spark Advance [CAD >5 <35 >0 <35

bTDCf]
Lambda [− ] >1.3 <2.3 >1.3 <2.5
Boost Pressure [bar] >2.5 <3.5 >1.1 <1.8
EGR [%] 0 <15 0 <15

Fig. 1. Experimental test matrix.

quasi-dimensional predictive model for hydrogen combustion with

cryogenic port injection. In this study, the laminar flame speed was Fig. 2. Experimental data regarding two spark advance sweeps performed at
EGR "baseline" and EGR "baseline" +5% – full load conditions.
calculated through detailed chemistry calculations based on an in-house
reaction scheme over a wide range of engine-relevant operating condi­
tions. The laminar flame speed model was implemented into the com­ composition of the unburned gas, solving a steady-state reacting flow
bustion model through a correlation function whose coefficients were problem once per cylinder model timestep. This approach would require
tuned to minimize the standard deviation with respect to the detailed very high computational costs in the case of carbon-based fuels, whose
chemistry results. The combustion model provided results consistent reaction mechanisms are characterized by a large number of species.
with the experimental data but showed some discrepancies at operating The simplicity of hydrogen combustion mechanisms justifies this
points with (very) lean mixtures. Rezaei et al. [15] presented a methodology, but the laminar flame speed model could act as the bottle
phenomenological model for the prediction of hydrogen combustion. neck in terms of computational time. According to Krebs et al. a tabu­
The model is based on the renowned SITurb model, provided by the lated laminar flame speed model could be useful to reduce the compu­
commercial software GT-SUITE, largely employed in several studies [16, tational times. Moreover, even though reasonable trends were obtained,
17] to predict the combustion in spark ignition engines. This combustion the work does not provide any comparison with experimental engine
model is distinguished by a novel mathematical approach for the data, deferring this assessment to future studies. In Ref. [19], Millo et al.
hydrogen laminar flame speed calculation, which was implemented proposed a methodology for evaluating the potential of a
through a Gaussian Process Model (GPM). The combustion model was hydrogen-powered internal combustion engine by combining zero, one,
calibrated and validated over the experimental data coming from a 2 L and three-dimensional (0D/1D/3D) Computational Fluid Dynamics
single cylinder engine (SCE), obtaining a good agreement with the (CFD) approaches. In particular, the SITurb combustion model, devel­
experimental results. However, this combustion model has not been oped by Gamma Technologies, was updated with an innovative
validated on engine with a smaller displacement. Krebs et al. [18] pre­ approach for the laminar flame speed determination. The laminar flame
sented a multi-zone predictive combustion model for hydrogen com­ speed was calculated in a 0D-CFD simulation environment
bustion in a spark ignition engine with a direct-injection system. The (CONVERGE), employing a detailed chemical kinetic mechanism over a
hydrogen-air laminar flame speed is calculated online from the large number of engine-relevant operating conditions in terms of

1311
A. Piano et al. International Journal of Hydrogen Energy 89 (2024) 1310–1320

Fig. 3. Experimental data regarding two sweeps of spark advance performed at Fig. 4. Experimental data regarding three couples of spark timings performed
lambda “baseline” and lambda “baseline” + 10% – full load conditions. at lambda “baseline”, lambda “baseline” + 22% and lambda “baseline” +30% –
partial load conditions.
temperature, lambda, EGR and pressure. Then, the results obtained from
the 0D-CFD simulations were integrated into the 0D/1D simulation considering the wide range of thermodynamic and mixture composition
environment (GT-SUITE) through a correlation function of mixture conditions under investigation. The findings of this study build upon
equivalence ratio capable of fitting detailed chemistry results, accord­ previous studies by introducing a more comprehensive, time-efficient,
ingly with the approaches proposed in Refs. [14,20]. Since no experi­ and accurate method for predicting the combustion process, account­
mental data for the investigated engine were available at that time, the ing for hydrogen-air mixture detailed chemistry. This work lays the
0D/1D predictive combustion model was calibrated basing on the results groundwork for the development of comprehensive models capable of
from 3D-CFD combustion simulations, in which the SAGE detailed predicting not only the combustion process, but also the cycle-to-cycle
chemistry combustion model [21] was employed. variability and the knock tendency under various engine operating
In this framework, the present research study aims at improving the conditions.
accuracy of the previously developed combustion model [19] through a
novel approach for the definition of the hydrogen-air flame speed, 2. Case study
leveraging also to the availability of experimental data coming from a
hydrogen-fuelled 0.5L single-cylinder engine with a port fuel injection The research activity described in this paper relies on experimental
(PFI) configuration. Indeed, a dedicated experimental activity was data coming from a hydrogen spark-ignited single cylinder engine with a
conducted to support the validation of the combustion model consid­ port fuel injection (PFI) configuration, designed by Dumarey. The en­
ering variation of mixture composition, dilution and phasing of the gine is representative of the retrofitting of a state-of-the-art low
combustion process. Eventually, the results obtained from the combus­ compression ratio diesel engine designed for light duty applications.
tion model with the novel approach for laminar flame speed determi­ Diesel architectures provide better resistance to hydrogen infiltration
nation showed high accuracy in predicting the combustion process even [22], thanks to the rugged heavy-duty mission for which they have been

1312
A. Piano et al. International Journal of Hydrogen Energy 89 (2024) 1310–1320

Fig. 5. Experimental pressure cycle and burn rate: (1) – different spark timing, constant lambda; (2) – different lambda, constant MFB50.

designed and withstands high peak firing pressures typical of highly investigated points. The experimental average pressure cycle, which can
boosted lean strategies better than traditional SI engines. For these be considered as representative of the combustion process, was used as
reasons, diesel engines are particularly suitable to the adaptation to reference for the predictive combustion model calibration and valida­
hydrogen functioning through retrofitting techniques. In this case, the tion phases. More details about the engine, the test cell layout, and the
original central diesel injector was replaced by a spark plug without any experimental campaign can be found in Ref. [12].
substantial modification of the original cylinder head. In addition, the
diesel piston bowl was replaced with a hemi-spherical bowl. The cyl­ 3. Experimental data analysis
inder head, the intake manifold, and the piston have been specifically
machined through additive manufacturing techniques [12]. EGR was The experimental data were analyzed to evaluate and, if possible, to
controlled through a high-pressure recirculation circuit, equipped with a separate the effects of each calibration parameters swept (i.e., spark
water-cooled radiator. timing, boost pressure, lambda and EGR percentage) on the hydrogen
The pressure in the intake and the exhaust plenums were controlled combustion and the engine performance. It is worth to note that varia­
to reproduce a compressor-turbine matching without employing an tions in engine performances are caused by the impact of the calibration
actual turbocharger. This approach aimed to replicate different parameters not only on the combustion process, but also on gas ex­
compressor/turbine configurations of interest and to set boosting and change losses (for instance, usage of high-pressure cooled EGR can be
back-pressure to the appropriate levels, thereby exploring the hydrogen beneficial due to the pumping mean effective pressure reduction [23]).
combustion properties across a wide range. For this reason, gross quantities, such as the gross IMEP and the gross
The main characteristics of the engine are reported in Table 1. indicated efficiency, have been used since they focus on the thermody­
The engine performances were evaluated under two different load namics of combustion, while ignoring the variations in terms of gas
levels at 2000 RPM: the rated torque, critical due to high knock ten­ exchange, which also depend on the selected engine subsystems. The
dency, and a medium-load engine operating condition (i.e., 6 bar analysis considered only knocking-free operating conditions, defined as
BMEP). Since hydrogen can operate with very lean mixtures, the engine those characterized by a percentage of knocking cycles lower than 3%.
run unthrottled during the testing phase for both the engine operating Therefore, the impact of a calibration parameter on the experimental
conditions under investigation. results such as the combustion duration and the MFB50 is determined
A wide experimental campaign was carried out, considering several solely by variations of the front flame speed, and not by the effects of the
variations in terms of lambda, boost pressure, spark timing and EGR, knock occurrence.
within the limits shown in Table 2. The resulting IMEP720 at full load Fig. 2 depicts the combustion duration (a), the gross IMEP (b), and
conditions is included in the range 10–18 bar. the gross indicated efficiency (c), for two sweeps of spark advance tested
A representation of the combinations of the calibration parameters at different EGR levels at full load engine operating condition. The boost
tested is provided in Fig. 1, where the dimension of each bubble is pressure and the intake manifold temperature were kept constant during
proportional to the EGR ratio employed. these tests, as well as the air-to-fuel ratio. Moreover, the spark timing
Only knocking-free operating conditions were considered in this was gradually advanced until it reached the knock limit spark advance
study, with a total of 22 points at full load and 25 points at partial load. (KLSA) condition. As the spark timing is advanced, the combustion
For each experimentally tested condition, 100 consecutive cycles have duration decreases due to the more favorable conditions for the flame
been recorded. Despite the lean operations, the IMEP (Indicated Mean propagation, for both the EGR levels. However, the usage of EGR allows
Effective Pressure) variation coefficient was lower than 3% for all the water vapor to be recirculated in the cylinder, increasing the specific

1313
A. Piano et al. International Journal of Hydrogen Energy 89 (2024) 1310–1320

due to the longer combustion duration.

Fig. 3 depicts the combustion duration (a), the gross IMEP (b), and
the gross indicated efficiency (c), for two spark advance sweeps per­
formed at different levels of lambda under full load conditions. These
tests have been carried out keeping constant the boost pressure and the
EGR fraction. Also in this case, a reduction of the mixture reactivity due
to an increase of the air-to-fuel ratio reduces the knock likelihood,
allowing advanced spark timing with similar combustion anchor angle.
However, since lambda is controlled by the amount of hydrogen injec­
ted, the lower amount of fuel for the high air-to-fuel ratio case leads to
lower gross IMEP values, as shown in Fig. 4(b). This means that at full
load condition, the increment of efficiency thanks to lean operation
counterbalances only partially the reduction of available fuel energy
within the combustion chamber. Last but not least, even considering the
lower laminar flame speed at leaner operations [7], which leads to
longer combustion durations (a), similar MFB50 are achieved thanks to
the more advanced spark timings. This latter leads therefore to a slightly
increase of the gross indicated efficiency (c) also thanks to the lower
thermal losses of leaner operations.
As far as partial load condition is concerned, three spark advance
sweeps performed at three different lambda levels are depicted in Fig. 4,
where the combustion duration (a), the gross indicated efficiency (b)
and the engine-out NOx emissions (c) are shown. In these tests the same
EGR rate was employed. Differently from full load condition tests, at
partial load the engine load was targeted by varying the injected
hydrogen, therefore, except for a variation of the engine efficiency,
sweeping lambda corresponds to a boost pressure modification. As ex­
pected, increasing the mixture dilution leads to longer combustion du­
rations (a). However, if the combustion phasing is kept constant (by
advancing the spark timing), the impact of the lower thermal losses
related to the lower in-cylinder temperature prevails, resulting in higher
indicated efficiencies at higher lambda.
In addition, mixture enleanment is highly beneficial for NOx emis­
sions mitigation as illustrated by Fig. 4 (c). Moreover, also the com­
bustion phasing has an impact on NOx emission at constant mixture
composition, however its efficacy is less pronounced for highly diluted
mixtures, since the in-cylinder temperatures had already been lowered
by the mixture enleanment. These results are consistent with the avail­
able literature [22,23].
The impacts of dilution and combustion phasing can also be assessed
through the examination of the pressure traces and the average in-
cylinder gas temperature profiles shown in Fig. 5. Precisely, two con­
Fig. 6. NOx emissions at partial load conditions under three different levels of
ditions characterized by the different spark timings at the same lambda
EGR, for several combinations of lambda and MFB50. (red box “1” in Fig. 4(c)), and two conditions with the same combustion
phasing but different dilution levels (blue box “2” in Fig. 4(c)) are shown
in Fig. 5 left and right, respectively.
Table 3 The NOx emissions mitigation achieved postponing the spark timing
Simulation matrix for the 1D-CFD detailed chemistry laminar flame speed (Fig. 4 (c)) is attributed to the lower in-cylinder temperature due to an
calculations. unfavorable combustion phasing (Fig. 5 (b)). However, despite the
Minimum Maximum benefits in terms of pollutants emissions, postponing the spark timing is
detrimental for the indicted efficiency due to a delayed combustion
Equivalence ratio [− ] 0.2 1
Unburned temperature [K] 600 1100
phasing and a longer burn rate (Fig. 4(a and b)).
Pressure [bar] 10 160 As abovementioned, an alternative approach to mitigate the NOx
Burned Gas Content [%] 0 30 could be the increment of air-to-fuel ratio. In this scenario, advancing
the spark timing is necessary to keep the same MFB50, due to the longer
combustion duration given by the leaner mixtures. The increase of
heat capacity of the mixture, and resulting in lower mixture tempera­
trapped air within the cylinder leads to higher pressure levels (Fig. 5
tures [24], thus mitigating knock likelihood by reducing the mixture
(c)), but the diluting effect allows reducing the in-cylinder temperature
reactivity and allowing more advanced combustion phasing. It is worth
(Fig. 5 (d)), thereby reducing the NOx emissions (Fig. 4 (c)), and being
to point out that the gross IMEP for the sweep exploiting higher EGR rate
beneficial for the gross indicated efficiency due to the lower thermal
is lower than the baseline case due to the lower amount of hydrogen
losses (Fig. 4 (b)).
trapped within the cylinder and the slower combustion (~7 deg differ­
An alternative solution to air-to-fuel and combustion phasing aiming
ence in terms of MFB1075, as shown in Fig. 2(a)) resulting from the EGR
at reducing engine-out NOx emissions reduction is the exploitation of
usage. Moreover, even though the lower thermal losses thanks to the
EGR. Fig. 6 provides an assessment of the influence of these three pa­
EGR usage, the sweep of spark timing performed at the higher EGR level
rameters on the NOx emission at partial load engine operating condi­
is characterized by a lower gross indicated efficiency (Fig. 2(c)) mainly
tion. It shows the NOx emissions as a function of the air-to-fuel ratio and

1314
A. Piano et al. International Journal of Hydrogen Energy 89 (2024) 1310–1320

Fig. 7. Laminar flame speed percentage difference between 1D detailed chemistry calculations (reference) and polynomial correlation for different mixture con­
ditions (lambda, EGR, pressure and temperature).

the MFB50 for three distinct levels of EGR rate. High dilution levels
dMe
contributed to achieve ultra-low NOx emissions (<50 ppm), consistently = ρu Af (ST + SL ) (Eq.1)
dt
with other experimental results already available in literature [11,25].
Furthermore, retarding the combustion timing while retaining the same Where Me is the mixture entrained mass, Af is the flame area, ρu un­
mixture composition leads to a NOx emission reduction. However, the burned gas density, SL the laminar flame speed and ST the turbulent
impact of the air-to-fuel ratio on the NOx emissions is largely more flame speed.
significant in comparison with the combustion phasing. The speed up of flame propagation due to the flame wrinkling is
The analysis of the experimental campaign in terms of the different taken into account computing the turbulent flame speed as shown in Eq.
impact of each calibration variable on performance and emissions has (2).
been considered preparatory for the development of a predictive com­ ⎛ ⎞
bustion model, since the understanding of the impact of each calibration
parameters at different engine operating conditions is crucial for the
⎜ 1 ⎟
ST = CTFS uʹ⎜
⎝1 − ( )2 ⎟
⎠ (Eq.2)
validation of the proposed model. 1 + CFKG Rf
LT

4. Combustion model
Indeed, the turbulent flame speed computation depends, besides
from the flame radius (Rf ) and the turbulence integral length scale (LT ),
The SITurb predictive combustion model, already accessible in the
on the turbulent intensity (u’). CTFS and CFKG are tuning parameters to
commercially available software GT-SUITE, was used as the basis for the
adjust both the flame velocity and the initial growth rate of the flame
developed hydrogen combustion model. The SITurb is a phenomeno­
kernel, respectively. It is worth to note that during the flame kernel
logical two-zone model, widely used for gasoline homogeneous charge
growth phase, the laminar flame speed governs the entrainment rate.
spark ignition engines [16,17]. It is based on a traditional entrainment
This occurs because the flame radius is too small to make the flame
and burn-up approach in which the turbulent flame propagation is
wrinkling effects negligible during the early stages of the combustion.
initiated from a fixed location (i.e., spark) within the cylinder. Before the
The rate at which mass is transferred into the burned zone, known as
spark timing, the in-cylinder mass is contained into a single thermody­
the burn rate, is calculated in Eq. (3).
namic zone, the unburned zone. Once the combustion starts triggered by
the spark, the flame front starts propagating from the spark location dMb Me − Mb
= (Eq.3)
towards the combustion chamber walls, assuming an ideal spherical dt τ
propagation. During the flame propagation, the homogeneous mixture is
progressively entrained into the flame front accordingly to the Mb is the mixture burned mass. It is proportional to the amount of
entrainment rate, calculated in Eq. (1). unburned mixture behind the flame front, Me − Mb , divided by a char­
acteristic burning timescale τ, that represents the time needed by the
laminar flame speed to cover the Taylor microscale of turbulence (λ). τ is

1315
A. Piano et al. International Journal of Hydrogen Energy 89 (2024) 1310–1320

Table 4
Root Mean Square Errors (RMSE) on combustion and pressure-related parame­
ters resulting from SITurb model optimizations.
Root Mean Square Error (RMSE)

LFS correlation Tabulated LFS

MFB50 [CAD] 4.8 3.0

Burn Duration 10–75 [CAD] 3.4 2.7
Maximum Pressure [bar] 5.2 4.0
CA @ Max. Pressure [CAD] 2.7 1.7

Fig. 9. Experimental (black dashed), predicted with LFS correlation (red), and
predicted with tabulated LFS (blue) average in-cylinder pressure and burn rate
for two different conditions of spark timing at the same lambda, boost pressure
and EGR (full load). (For interpretation of the references to color in this figure
legend, the reader is referred to the Web version of this article.)

calculated by Eq. (4).

λ
τ= (Eq.4)
SL
The Taylor microscale of turbulence λ is defined as in Eq. (5), where
Ret is the Reynolds number and CTLS is a tuning parameter.
LT
λ = CTLS √̅̅̅̅̅̅̅ (Eq.5)
Ret
As it can be noticed, the combustion model relies on the turbulent
kinetic energy and the integral length scale. These quantities are
calculated by a predictive in-cylinder turbulence model based on the K-
k-eps approach [26]. The employed turbulence model was properly
calibrated to match the 3D-CFD simulation results.
As far as laminar flame speed is concerned, two different user sub­
routines have been defined: the first one based on a mathematical cor­
relation, the second based on LFS tables. Both approaches rely on the
Fig. 8. Correlation plots for combustion anchor angle (MFB50), combustion detailed chemistry 1D-CFD computations carried out considering wide
duration (MFB10-75), maximum in-cylinder pressure (Maximum Pressure) and ranges of pressure, temperature, equivalence ratio and residual gasses,
its crank angle (Crank Angle @ Maximum Pressure). representative of engine-relevant conditions. The detailed Zhang et al.

1316
A. Piano et al. International Journal of Hydrogen Energy 89 (2024) 1310–1320

Fig. 10. Experimental (black dashed) and predicted (red) average in-cylinder Fig. 11. Experimental (black dashed) and predicted (red) average in-cylinder
pressure and burn rate; 100 experimental consecutive cycles (grey). Two pressure and burn rate; 100 experimental consecutive cycles (grey). Two
different EGR conditions at same lambda, boost pressure and spark timing (full different spark timings at same lambda, boost pressure and EGR (full load). (For
load). (For interpretation of the references to color in this figure legend, the interpretation of the references to color in this figure legend, the reader is
reader is referred to the Web version of this article.) referred to the Web version of this article.)

mechanism [26] was used as reaction scheme for hydrogen oxidation. The coefficients of each polynomial expression were properly cali­
The simulation matrix is reported in Table 3, resulting in more than brated to minimize the differences between the correlation results (Eq.
1600 different operating condition. (6)) and the 1D-CFD detailed chemistry calculations. Fig. 7 shows the
The laminar flame speed model based on the correlation function percentage error between the laminar flame speed correlation and the
was inherited from the work described in Ref. [19]. Specifically, the results from the detailed chemistry calculations, as a function of the in-
correlation function employed was defined consistently with [20], as cylinder pressure and the unburned gas temperature, for different
shown in Equation (6). mixture dilution and compositions. The error is computed having as a
( )α ( )β reference the 1D results from the detailed chemistry calculations.
Tu pu Adopting the described mathematical correlation, although a satis­
SL (Φ, Tu , pu , xb ) = SL0 (1 − γxb ) (Eq.6)
T0 p0 factory agreement was achieved for large sweeps of pressure and tem­
perature, a not negligible overprediction can be highlighted especially
SL0 is the laminar flame velocity evaluated at the reference conditions
when EGR is exploited. More specifically, the second column is referred
(TU = T0 = 298 K; pU = p0 = 1 atm; residual gas fraction xb = 0), while
to EGR = 10% mixture condition and highlights up to 70% LFS over­
α, β, and γ are coefficients that quantify the sensitivity of the laminar
prediction, thus being potential source of error in the combustion pro­
flame speed to the unburned gas pressure, temperature, and residual gas
cess prediction. Due to this reason, in this study a novel approach was
fraction, respectively. The SL0 is defined accordingly with the approach
considered to minimize the error with the laminar flame speed detailed
proposed by Verhelst and Sierens [20], where a fifth-order polynomial
chemistry calculations, being the laminar flame speed fundamental for
function of the equivalence ratio Φ is used to describe the laminar flame
the model accuracy, as shown in Eq. (1) and Eq. (4). More specifically,
speed in reference operating conditions, as in Equation (7).
the laminar flame speed model was implemented in GT-SUITE by means
SL0 (Φ) = 30.009Φ5 − 87.697Φ4 + 88.628Φ3 − 33.781Φ2 + 5.489Φ of several look-up tables containing the results of the detailed chemistry
calculations. The tabulated approach enabled by a dedicated user sub­
− 0.301
routine provides the value of LFS at each timestep depending on pres­
(Eq.7) sure, temperature, equivalence ratio and residual gas fraction in the
Similarly, the coefficients α, β, and γ of Equation (6) were defined cylinder.
using second-order polynomial expressions dependent on the equiva­
lence ratio, as reported in Equations (8)–(10). 5. Results

α = 0.105Φ − 3.135Φ + 6.514

2
(Eq.8)
As previously described, the combustion model requires the tuning of
three calibration constants (i.e., CTFS CFGK and CTLS ). Therefore, an
β = 0.177Φ2 + 0.636Φ − 1.449 (Eq.9)
optimization process has been employed in GT-SUITE basing on the
NSGA-III genetic algorithm (GA) [27]. This algorithm is extensively
γ = − 0.306Φ2 + 0.418Φ + 2.134 (Eq.10)
utilized for optimizing predictive combustion models [17,28], as it is

1317
A. Piano et al. International Journal of Hydrogen Energy 89 (2024) 1310–1320

Fig. 12. Experimental (black dashed) and predicted (red) average in-cylinder Fig. 13. Experimental (black dashed) and predicted (red) average in-cylinder
pressure and burn rate; 100 experimental consecutive cycles (grey). Two pressure and burn rate; 100 experimental consecutive cycles (grey). Two
different lambda values at same boost pressure, EGR and spark timing (partial different spark timings at same lambda, boost pressure and EGR (partial load).
load). (For interpretation of the references to color in this figure legend, the (For interpretation of the references to color in this figure legend, the reader is
reader is referred to the Web version of this article.) referred to the Web version of this article.)

particularly effective for multi-modal problems of medium to high maximum in-cylinder pressure average errors are lower than 3 deg and
complexity. In particular, it is less likely to become trapped in local 4 bar, respectively.
minimum compared to other optimization tools provided by GT-SUITE, A further comparison between the results obtained from the two
thereby increasing the probability of reaching the global optimum so­ approaches for the LFS determination is given in Fig. 9, where a sweep of
lution [29]. As objective for the optimization process, the root mean spark timing at full load conditions was analyzed. In particular, the
squared error between the measured and the simulated burn rate was average in-cylinder pressure trace and the respective burn rate obtained
minimized, keeping unchanged the three calibration parameters moving from simulations were compared with the experimental data. Also in
from an engine operating condition to another. Moreover, to increase this case, the LFS model with the tabulated approach better matches the
the reliability of the model, without inducing any bias from the selection experimental data, especially as regards the in-cylinder peak pressure
of the training dataset, 16 operating conditions were randomly selected and the burn rate shape.
among the full load and partial load datasets to cover a wide range of Once assessed the differences among the two LFS models, the results
lambda, EGR and spark timings, and they were used as training dataset. obtained from the combustion model with the tabulated laminar flame
Therefore, the combustion model was validated over the remaining 31 speed were further analyzed since it provided more accurate predictions.
operating conditions. In addition, the calibration procedure was per­ Simulations results were compared with the experimental data consid­
formed for both the laminar flame speed models previously mentioned ering different variations of EGR percentage, boost pressure (lambda)
to highlight pros and cons of each approach and select the most reliable and spark timing in Figs. 10–12, and Fig. 13. Generally, a very good
one. agreement between the experimental traces and the predicted ones was
The correlation plots of MFB50, Burn Duration 10–75 (combustion- observed, both at full load and partial load. The combustion model
related parameters) and Maximum Pressure, CA of Maximum Pressure demonstrates to be able to properly calculate the burn rate varying spark
(pressure-related parameters) resulting from the optimization process timing and dilutions levels, both in terms of lambda and EGR, leading to
are shown in Fig. 8, where each circle represents a single operating a very good match between the results from the simulation and the
condition, and the color identifies the LFS model employed (i.e., red – experimental data.
tabulated approach, blue – polynomial correlation). Generally, the The assessment of the combustion model capabilities was finally
tabulated laminar flame speed model leads to a better correlation with carried out comparing the combustion model results in terms of com­
the experiments, since most of the points are much closer to the zero- bustion phasing (MFB50) with the experimental data considering
error line, highlighted with the black dashed line. several sweeps of spark timing, performed under various operating
To quantify the correlation level, the root mean square errors both conditions. Two spark timing sweeps carried out at full load with high
for combustion and pressure-related parameters were computed and are and low EGR levels are shown in Fig. 14. The experimental data are
shown in Table 4. Even if both models provide a more than satisfactory shown with their respective confidence bands, calculated using a ±σ
accuracy, the tabulated LFS approach shows lower errors demonstrating interval, where σ represents the standard deviation of MFB50. Also in
the better predictive capabilities compared to the polynomial correla­ this case, a very good agreement with experimental data was high­
tion. Indeed, thanks to this approach, the combustion duration and the lighted: the model is able to properly predict the MFB50 over the entire

1318
A. Piano et al. International Journal of Hydrogen Energy 89 (2024) 1310–1320

Fig. 14. Experimental (black) and predicted (red) MFB50 and burn duration Fig. 15. Experimental (black) and predicted (red) MFB50 and burn duration
for two different spark timing sweeps at different EGR levels (full load). (For for two different spark timing sweeps at different lambda levels (partial load).
interpretation of the references to color in this figure legend, the reader is (For interpretation of the references to color in this figure legend, the reader is
referred to the Web version of this article.) referred to the Web version of this article.)

sweep of spark timing for both EGR levels, demonstrating its good of a dedicated experimental campaign carried out on a 0.5L single-
sensitivity to both EGR and spark timing variations. cylinder engine, retrofitted from a diesel architecture and adapted to
The results indicate that the deviation from the MFB50 of the average the hydrogen functioning. The experimental data, widely used for the
experimental cycle due to cycle-to-cycle variation is higher than the validation of the model, include several sweeps of spark timing, lambda,
error of the predictive model. This confirms the model robustness in EGR, and boost pressure, performed under full load and partial load
capturing the average combustion behavior. The correlation of the operating conditions.
optimized combustion model with the experimental data was investi­ First, the experimental data have been analyzed to evaluate the
gated also for partial load operating conditions in Fig. 15. In particular, impact of each calibration parameter on the hydrogen combustion. The
two spark sweeps at two different lambda values were considered. analysis revealed that, as expected, calibration parameter variations
The developed predictive combustion model with the tabulated highly affects the combustion process, making the development of such
laminar flame speed results to properly estimate the MBF50 for different simulation tool challenging. Starting from this latter and considering
mixture compositions, confirming its good predictive capabilities also at hydrogen peculiarities, the air-hydrogen laminar flame speed has been
partial load engine operating conditions. It is worth to point out that a calculated employing a detailed chemistry scheme on a large number of
single set of calibration constants has been used avoiding a case-by-case operating conditions in terms of mixture composition, pressure and
calibration, thus confirming the reliability of the developed model. temperature. The laminar flame speed calculations were implemented
The optimized hydrogen combustion model can be combined with a considering two different approaches: a polynomial correlation function
knock model and a cycle-to-cycle variation model to create a compre­ of mixture composition, and a tabulated approach. The developed
hensive tool capable of predicting not only the combustion rate, but also combustion model has been calibrated through an optimization process
the cycle-to-cycle variability, the knock tendency and the percentage of with the NSGA-III Genetic Algorithm on a dataset of 16 operating con­
knocking cycles for each operating condition [30]. ditions and validated against the remaining 31 operating points. The
combustion model employing the tabulated laminar flame speed pro­
6. Conclusions vided a more than satisfactory accuracy level, demonstrating a good
agreement with the experimental data.
In this study a combustion model is proposed for the prediction of the The developed predictive combustion model, which provides an
combustion process of hydrogen fuelled internal combustion engines. accurate and time-efficient method for simulating the combustion pro­
The predictive combustion model relies on a novel approach for the cess in hydrogen-fuelled engines, can be further extended in future work
determination of the air-hydrogen flame speed, and takes the advantage through its integration with predictive knock and cycle-to-cycle

1319
A. Piano et al. International Journal of Hydrogen Energy 89 (2024) 1310–1320

variability (CCV) models. This integrated approach allows the devel­ Hydrogen Energy 2022;47:25044–56. https://doi.org/10.1016/j.
ijhydene.2022.05.250.
opment of a comprehensive model capable of predicting not only the
[12] Buzzi L, Biasin V, Galante A, Gessaroli D, et al. Experimental investigation of
combustion process but also the cycle-to-cycle variability and the per­ hydrogen combustion in a single cylinder PFI engine. THIESEL 2022 conference on
centage of knocking cycles under various engine operating conditions. thermo- and Fluid Dynamics of clean propulsion powerplants. 2022. https://doi.
Such a comprehensive model can serve as a highly powerful tool for org/10.4271/2022-01-0331.
[13] Yip HL, Srna A, Yuen ACY, Kook S, et al. A review of hydrogen direct injection for
significantly accelerating the development of hydrogen-fuelled engines. internal combustion engines: towards carbon-free combustion. Appl Sci 2019;9.
https://doi.org/10.3390/app9224842.
CRediT authorship contribution statement [14] D’Errico G, Onorati A, Ellgas S. 1D thermo-fluid dynamic modelling of an S.I.
single-cylinder H2 engine with cryogenic port injection. Int J Hydrogen Energy
2008;33:5829–41. https://doi.org/10.1016/j.ijhydene.2008.05.096.
Andrea Piano: Writing – review & editing, Visualization, Supervi­ [15] Rezaei R, Hayduk C, Fandakov A, Rieß M. Numerical and experimental
sion, Methodology, Conceptualization. Gianpaolo Quattrone: Writing investigations of hydrogen combustion for heavy-duty applications. SAE Technical
Paper, 2021-01-0522 2021. https://doi.org/10.4271/2021-01-0522.
– original draft, Visualization, Validation, Investigation. Federico [16] Millo F, Gullino F, Rolando L, et al. Methodological approach for 1D simulation of
Millo: Supervision, Project administration. Francesco Pesce: Supervi­ port water injection for knock mitigation in a turbocharged DISI engine. Energies
sion, Resources, Project administration, Conceptualization. Alberto 2020;13(17):4297. https://doi.org/10.3390/en13174297.
[17] Mirzaeian M, Millo F, Rolando L, et al. Assessment of the predictive capabilities of
Vassallo: Writing – review & editing, Supervision, Resources, a combustion model for a modern downsized turbocharged SI engine. SAE
Conceptualization. technical papers, 2016-01–0557. 2016. https://doi.org/10.4271/2016-01-0557.
[18] Krebs S, Biet C. Predictive model of a premixed, lean hydrogen combustion for
internal combustion engines. Transport Eng 2021;5. https://doi.org/10.1016/j.
Declaration of competing interest
treng.2021.100086.
[19] Millo F, Piano A, Rolando L, Accurso F, et al. Synergetic application of zero-, one-,
The authors declare that they have no known competing financial and three-dimensional computational fluid Dynamics approaches for hydrogen-
interests or personal relationships that could have appeared to influence fuelled spark ignition engine simulation. SAE International Journal of Engines
2021;15. https://doi.org/10.4271/03-15-04-0030.
the work reported in this paper. [20] Verhelst S, Sierens R. A Laminar burning velocity correlation for hydrogen/air
mixtures valid at Spark-Ignition engine conditions. Spring technical conference of
References the ASME internal combustion engine division. 2003. https://doi.org/10.1115/
ICES2003-0555.
[21] Convergent Science. “CONVERGE Manual v3.0.”. 2020.
[1] IEA. Net zero emissions by 2050 scenario (NZE). https://www.iea.org/reports/glob [22] Azeem N, Beatrice C, Vassallo A, Pesce F, et al. Review and assessment of the
al-energy-and-climate-model/net-zero-emissions-by-2050-scenario-nze. [Accessed material’s compatibility for rubbers and elastomers in hydrogen internal
22 July 2024]. combustion engines. SAE Technical Paper 2022. https://doi.org/10.4271/2022-
[2] Onorati A, Payri R, Vaglieco BM, Agarwal AK, et al. The role of hydrogen for future 01-0331.
internal combustion engines. Int J Engine Res 2022;23(4):529–40. https://doi.org/ [23] Thomas Koch D, Sousa A, Bertram D. H2-Engine operation with EGR achieving
10.1177/14680874221081947. high power and high efficiency emission-free combustion. SAE Technical Paper,
[3] Huang Q, Liu J, et al. Preliminary assessment of the potential for rapid combustion 2019-01–2178 2019. https://doi.org/10.4271/2019-01-2178.
of pure ammonia in engine cylinders using the multiple spark ignition strategy. Int [24] Chaichan MT. EGR effects on hydrogen engines performance and emissions. Int J
J Hydrogen Energy 2024;55:375–85. https://doi.org/10.1016/j. Sci Eng Res 2016;7(3):80–90.
ijhydene.2023.11.136. [25] Bao L, zhi, Sun, gang B, Luo, he Q. Experimental investigation of the achieving
[4] Bičáková O, Straka P. Production of hydrogen from renewable resources and its methods and the working characteristics of a near-zero NOx emission turbocharged
effectiveness. Int J Hydrogen Energy 2012;37:11563–78. https://doi.org/10.1016/ direct-injection hydrogen engine. Fuel 2022;319. https://doi.org/10.1016/j.
j.ijhydene.2012.05.047. fuel.2022.123746.
[5] Stępień Z. A comprehensive overview of hydrogen-fueled internal combustion [26] Zhang K, Banyon C, Bugler J, Curran HJ, et al. An updated experimental and
engines: achievements and future challenges. Energies 2021;14. https://doi.org/ kinetic modeling study of n-heptane oxidation. Combust Flame 2016;172:116–35.
10.3390/en14206504. https://doi.org/10.1016/j.combustflame.2016.06.028.
[6] Karim GA. Hydrogen as a spark ignition engine fuel. Int J Hydrogen Energy 2003; [27] Deb K, Jain H. An evolutionary many-objective optimization algorithm using
28:569–77. https://doi.org/10.1016/S0360-3199(02)00150-7. reference-point-based nondominated sorting approach, Part I: solving problems
[7] Sun D, Liu F. Research on the performance and emission of a port fuel injection with box constraints. IEEE Trans Evol Comput 2014;18:577–601. https://doi.org/
hydrogen internal combustion engine. International conference on computer 10.1109/TEVC.2013.2281535.
distributed control and intelligent environmental monitoring. CDCIEM; 2011. [28] Accurso F, Piano A, Millo F, Caputo G, et al. Numerical simulation of a
p. 963–6. https://doi.org/10.1109/CDCIEM.2011.523. 2011. prechamber-ignited lean-burn gas engine by means of predictive combustion
[8] Schefer RW, White C, Keller J. Lean combustion. 2008. p. 213–54. https://doi.org/ models. SAE International Journal of Engines 2023;16(5):623–41. https://doi.org/
10.1016/B978-012370619-5.50009-1. ISBN: 9780123706195. 10.4271/03-16-05-0037.
[9] Pauer T, Weller H. Hydrogen engines for future passenger cars and light [29] Gamma Technologies. “GT-SUITE reference manual”. 2022.
commercial vehicles. MTZ Worldwide 2021;82:42–7. https://doi.org/10.1007/ [30] Piano A, Millo F, Quattrone G, Pesce F, et al. Comprehensive quasi-dimensional
s38313-020-0603-1. model to predict combustion process, knock likelihood and cycle-by-cycle
[10] Ozdor N, Dulger M, Sher E. An experimental study of the cyclic variability in spark variability in a hydrogen-fueled internal combustion engine. JSAE/SAE
ignition engines. Reprinted from: Advances in Engine Combustion and Flow Powertrains, Energy and Lubricants International Meeting 2023;2023–32–0171.
Diagnostics, SAE Technical Paper, 1996-02–01 1996. https://doi.org/10.4271/ https://doi.org/10.4271/2023-32-0171.
960611.
[11] Sementa P, de Vargas Antolini JB, Tornatore C, Catapano F, et al. Exploring the
potentials of lean-burn hydrogen SI engine compared to methane operation. Int J

1320