This article is licensed under CC-BY 4.

pubs.acs.org/EF Article

Computational Co-optimization of Fuel and Spark-Ignition Engine

Published as part of Energy & Fuels special issue “Accelerating the Energy Transition: Advancements in
Electrofuels”.
Philipp Ackermann, Benjamin Auer, Patrick Burkardt, Bastian Lehrheuer, Philipp Morsch,
Karl Alexander Heufer, Stefan Pischinger, Alexander Mitsos, and Manuel Dahmen*
Cite This: Energy Fuels 2025, 39, 4079−4093 Read Online
See https://pubs.acs.org/sharingguidelines for options on how to legitimately share published articles.

ACCESS Metrics & More Article Recommendations *

sı Supporting Information
Downloaded via RWTH AACHEN HOCHSCHUL BIBL on March 4, 2025 at 15:09:47 (UTC).

ABSTRACT: Spark-ignition engine efficiency can be increased by co-

optimizing fuel and engine. First, computational fuel design can optimize
fuel molecules or composition using predictive fuel property models, e.g., for
high octane numbers. Then, the engine configuration can be optimized
experimentally to maximize the achievable efficiency. However, such a
sequential co-optimization based on fuel properties may yield suboptimal
fuels, as the fuel properties do not fully capture the complex fuel−engine
interaction. Therefore, we propose the computational, simultaneous co-
optimization of fuel and engine. To this end, we derive a thermodynamic
engine model that predicts the engine performance as a function of fuel
composition and engine configuration. We calibrate the engine model against
experimental data from a single-cylinder research engine, such that new candidate fuels require no model recalibration with
additional experimental engine data. As a case study, we select 10 possible alternative fuel components identified in previous studies
and create 39 binary and 60 ternary fuel mixtures. The composition of each fuel mixture is then co-optimized together with the
compression ratio and the intake pressure of the engine considering knock and peak pressure constraints to ensure smooth and safe
engine operation. The study reveals the small esters methyl acetate and ethyl acetate as promising fuel candidates for future spark-
ignition engines. For methyl-acetate-rich blends, the engine model predicts knock-free operation at compression ratios of up to 20
and boost pressures of up to 1.8 bar, rendering methyl acetate a promising alternative to methanol. Considering significant model
uncertainties, however, the findings require experimental validation.

■ INTRODUCTION
Spark-ignition (SI) engines running on liquid fuels are central
the early 2000s, the octane index (OI) was established as a
more accurate measure for ignition propensity of gasoline fuels
to today’s mobility. Despite their technological maturity, they in modern SI engines.11 The OI is defined as a weighted sum
can still be improved in terms of efficiency. 1,2 The of RON and the octane sensitivity (OS). OS is defined as the
thermodynamic efficiency of SI engines increases with the difference between RON and MON12 and its influence on OI
compression ratio and, in general, with boosting pressure.3 is weighted with an empirical factor K.
However, both high compression ratios and high boosting Alternative fuels can enable higher compression ratios and
pressures lead to higher in-cylinder temperature and pressure. boosting pressures than a typical RON 95 gasoline and thus
These extreme conditions may trigger autoignition of the fuel, higher thermodynamic efficiencies. A multitude of alternative
leading to engine knock, which impairs engine operation and fuels have been investigated in research engines and have
possibly damages the engine.4 Increasing the thermodynamic achieved higher engine efficiencies than RON 95 gasoline.
efficiency by means of increasing compression ratio and Examples for pure-component fuels include methylfuran,13 2-
boosting pressure requires lowering the autoignition propen- methylfuran,14 2-butanone,14−16 methanol,17−20 and etha-
sity of the fuel.5 nol.14−17,21,22 Examples of multicomponent fuels include a
Traditionally, fuel autoignition propensity is rated by octane
numbers like the research octane number (RON)6 and the
motor octane number (MON).7 For both numbers, stand- Received: September 30, 2024
ardization organizations have established minimal acceptable Revised: January 31, 2025
values for marketed gasoline fuels to ensure knock-free Accepted: February 5, 2025
operation in commercially used engines.8,9 For modern, highly Published: February 12, 2025
boosted SI engines, a high octane number alone does not
describe the ignition propensity of a fuel sufficiently well.10 In
© 2025 The Authors. Published by
American Chemical Society https://doi.org/10.1021/acs.energyfuels.4c04775
4079 Energy Fuels 2025, 39, 4079−4093
Energy & Fuels pubs.acs.org/EF Article

blend of acetone, butanol, and ethanol,23,24 a blend of ethanol, components and binary blends.43 The authors of the merit
2-butanone, cyclopentane, and cyclopentanone,15,16,25 and a function, however, discuss several limitations that compromise
six-component blend containing a ketone, esters, alcohols, and the utility of the function in computational co-optimization of
an alkane (KEAA).16,25 The experimental findings demonstrate fuel and engine.2 With individual terms for the contribution of
the potential of alternative fuels to exploit advanced engine each fuel property to the thermodynamic efficiency, the merit
concepts. In all cases, the efficiency increase can be at least function neglects the synergistic effects of the fuel properties
partly attributed to the lower autoignition propensity of the on each other, e.g., the effect of the enthalpy of vaporization on
alternative fuels, which is reflected in high values of RON and the RON44 or the OS.45 Furthermore, the inclusion of the OS
OS of these fuels. requires choosing a value for the K factor, introducing the
To unlock the full potential of both advanced engine same uncertainties as in the case of the OI, as discussed above.
concepts and alternative fuels, researchers have proposed Importantly, the merit function does not provide direct
systematic strategies to jointly develop engines and fuels, information about engine conditions necessary to maximize
resulting in a co-optimization.26−28 The maximization of the thermodynamic efficiency.
thermodynamic efficiency includes numerous degrees of Zero-dimensional thermodynamic engine models constitute
freedom that range from fuel constituents and composition a more rigorous option for predicting thermodynamic
to engine design and operation. Exploring this design space efficiency, as they build on mechanistic knowledge. Moreover,
solely by extensive experimental investigations is prohibitive. they provide insight in how engine design and operation affect
The number of degrees of freedom can be reduced by means engine performance. Most engine modeling efforts have been
of computer-aided product design (CAPD).29 CAPD identifies restricted to specific engine designs with specific alternative
promising fuel components and blends using predictive fuels such as methanol,46−49 hydrogen,50 or hydrogen−
property models. methane mixtures.51,52 Only few studies have used models to
Numerous studies have applied CAPD to fuels for advanced compare a limited number of fuels, but with calibrated
SI engines by targeting molecules and mixtures with high parameter data for each fuel based on dedicated engine
RON.27,30−34 Alternative fuels designed in this manner have data.53,54 The most comprehensive study has been published
been experimentally investigated in single-cylinder research by Gschwend et al., who developed a dedicated thermody-
engines with high compression ratios and under highly boosted namic engine model with an embedded optimization of the
conditions.14−16 The experimental investigations confirm the compression ratio55 to evaluate alternative fuels and screened a
hypothesis that fuels with higher RON values enable database of 50 possible alternative fuel candidates for
thermodynamic efficiency increases. However, the max- thermodynamic engine efficiency.56 However, the existing
imization of RON does not necessarily lead to the max- engine models are not suited for computational co-
imization of thermodynamic engine efficiency. For example, 2- optimization of fuels and engines maximizing engine efficiency
butanone exhibits a slightly higher RON than ethanol, but as a function of fuel composition as well as engine design and
engine experiments report equal or higher thermodynamic operation, as they are either tailored to specific fuels and
efficiencies for ethanol.14−16,25 engine designs46−52 or have not been validated for extreme
Fuel design can target OI instead of RON as an alternative engine conditions and alternative fuels.53−55
objective35 that incorporates the OS. The influence of the OS The present work provides a first approach to computational
on the OI, however, depends upon the engine design and co-optimization of multicomponent fuels and SI engine
operating conditions, which are captured by an empirical factor configuration at stoichiometric combustion conditions to
K. The K factor is linked to high uncertainty in highly boosted achieve maximum thermodynamic engine efficiency. As a
engine operation36 and itself depends upon the fuels used in its central building block, we derive a predictive one-zone
determination.37 Furthermore, autoignition propensity is not thermodynamic model of a single-cylinder research SI engine
the only fuel property that influences engine efficiency. A high and implement it in MATLAB R2020a.57 We calibrate and
enthalpy of vaporization can provide an additional efficiency validate the model with measurement data of different
increase through increased charge cooling during fuel alternative fuels and different compression ratios from a
evaporation. Charge cooling reduces both the wall heat losses single-cylinder research engine. Notably, the model can be
and the combustion temperature, and thus results in higher applied to new fuel candidates without the need for
thermodynamic efficiency and lower autoignition propensity.38 recalibration. Utilizing the model, we establish a computational
The charge cooling effect is especially pronounced in light co-optimization procedure of multicomponent fuels and SI
alcohol fuels such as methanol and ethanol.17 However, its engine configuration. First, we select few possible fuel
influence on the engine efficiency increase is challenging to constituents as candidates for blending based on previous
isolate.2 CAPD studies. Then, we maximize the thermodynamic
Several models attempt to incorporate the charge cooling efficiency of the engine by varying the compression ratio, the
effect and further fuel properties in the prediction of achievable intake pressure, and the fuel composition. The optimization is
engine efficiency. Leone et al.39 correlated the thermodynamic subject to knock and peak pressure constraints as well as
efficiency of gasoline−ethanol blends to octane number and volatility related fuel specifications.
ethanol content. The correlation has been used to design The remainder of this article is structured as follows. First,
gasoline−ethanol blends,40 but by its nature it is not applicable the thermodynamic engine model is described, including its
to alternative fuels other than ethanol. Szybist et al.2 validation against experimental data. Then, the co-optimization
introduced the so-called engine efficiency merit function that of fuel and engine is described, including problem formulation
correlates thermodynamic efficiency to fuel properties such as (scope, constraints, and solution procedure), the resulting
RON, OS, enthalpy of vaporization, and laminar burning blends and their corresponding engine designs are discussed,
velocity.2 The merit function has been used in database and the predicted efficiencies are compared to RON, OI, and
screenings33,41,42 and in a molecular design study of pure merit function values.
4080 https://doi.org/10.1021/acs.energyfuels.4c04775
Energy Fuels 2025, 39, 4079−4093
Energy & Fuels pubs.acs.org/EF Article

Figure 1. Left side of the figure shows a conceptual sketch of the single cylinder research engine. The piston moves between top dead center
(TDC) and bottom dead center (BDC). The red box depicts the time-dependent control volume V used for the governing equations that calculate
the average cylinder pressure p and the average cylinder temperature T. Mass and heat flows across the system boundary are shown. The right side
illustrates the thermodynamic engine cycle on which we base the model.

■ THERMODYNAMIC ENGINE MODEL

Model Scope and Assumptions. The engine model shall
require a vacuum behind the exhaust valve. To this end, we
introduce an idealized gas exchange model based on an
predict the net indicated efficiency and the occurrence of idealized thermodynamic cycle with fixed valve timing, as
knock in a single-cylinder research engine (see Figure 1, left) described in Heywood3 and in the following paragraph.
as a function of compression ratio, intake pressure, and fuel The thermodynamic cycle contains seven process steps (see
composition. In defining the fuel composition, we distinguish Figure 1) and can be divided into a high pressure cycle and a
between blend stocks and components: blend stocks can refer low pressure cycle. At point 1, intake of fresh air and fuel
to pure components, e.g., ethanol, or predefined mixtures, e.g., injection are completed, and the intake valve closes instantly.
gasoline; components refer to both the pure components and Here, the high-pressure cycle begins. Between point 1 and
the components used to model the predefined mixtures, e.g., point 2, the piston compresses the air−fuel mixture. Between
iso-octane, toluene, or n-heptane, in the case of gasoline. We point 2 and point 3, the fuel is oxidized. From point 3 to point
denote the mole fractions of blend stocks with zfuel,j, and the 4, the cylinder charge expands. At point 4, the exhaust valve
mole fractions of components with xfuel,j. opens. Here, the low-pressure cycle begins. From point 4 to
The net indicated efficiency ηi,n is a function of the net point 5, the pressure drops instantaneously to the pressure of
indicated work Wi,n and the amount of energy contained in the the exhaust manifold. From point 5 to point 6, the piston
fuel, which depends upon the molar amount of fuel injected movement purges the exhaust gas at constant pressure. At
per cycle nfuel,in and the lower heating value of the fuel LHV.3 point 6, the exhaust valve closes and the intake valve opens.
Wi,n Between point 6 and point 7, the cylinder pressure
i,n
= instantaneously takes the value of the pressure in the intake
n fuel,in LHV (1) manifold. Between point 7 and point 1, the piston takes in the
Wi,n is the area enclosed by the engine process in the pressure− fresh air at constant pressure and fuel is injected.
volume diagram.3 In the following, we list the most important model
equations. The remaining model equations and the derivations
Wi,n = p dV are given in the Supporting Information. Note that the
(2) equations are expressed in terms of molar amounts instead of
We base our model on the thermodynamic engine model by mass, in contrast to commonly published engine model
Gschwend et al.55 The model by Gschwend et al. is an one- equations.
zone model, i.e., the state variables are assumed to be spatially High Pressure Cycle. The heat transfer to and from the
uniform across the whole cylinder, also disregarding any walls is modeled with Newton’s law and utilizing the
differences between burned and unburned gas. Only to predict Hohenberg correlation59 for the heat transfer coefficient. The
knock, the unburned gas temperature is modeled separately. wall heat transfer term is further multiplied with an empirical
Because we made different assumptions in some cases, we heat transfer multiplier CHT that can take different values for
opted to derive the majority of model equations starting from the three process steps (compression, combustion, and
fundamental mass and energy balances following standard expansion), as commonly applied in thermodynamic engine
assumptions.3,50,58 Importantly, we relax Gschwend’s assump- models.48 The values for CHT are fitted to experimentally
tion of complete exhaust of the exhaust gas, which would measured pressure curves.
4081 https://doi.org/10.1021/acs.energyfuels.4c04775
Energy Fuels 2025, 39, 4079−4093
Energy & Fuels pubs.acs.org/EF Article

The model by Gschwend et al.55 defines the combustion which contains the known pressure p1 and volume V1 at the
using three subsequent idealized thermodynamic processes end of the intake stroke as well as the molar gas constant R,
(isochoric, isobaric, isothermal). To obtain more realistic and the unknowns T1 and n1, nair,in and nfuel,in can be calculated.
pressure curves for the combustion phase, we replace the The wall heat transfer term Qw,intake accounts for the fresh air
idealized process steps by predefining the fuel conversion rate. being heated up during the intake stroke, both in the intake
Specifically, we apply a Wiebe function,60,61 which specifies the manifold and in the cylinder. Because a higher amount of air
mass fraction burned ξ as a function of the crank angle θ using taken in leads to a higher heat transfer from the intake
the parameters a, b, m, θ0, and Δθ. manifold and the cylinder walls to the air, we assume Qw,intake
ÄÅ ÉÑ| to be a function of nair,in. We furthermore consider the cooling
l
o ÅÅ i (m + 1)ÑÑÑo
o y ÑÑo
o ÅÅ j o effect of the enthalpy of vaporization of the fuel Hvap,fuel on the
( ) = bm 1 expÅÅÅ ajjj 0 z
zzz
o
o ÑÑ} o
o fuel−air mixture and the resulting increase of wall heat transfer.
o Å Ñ
n ÅÅÇ k { ÑÑÖo
~ (3) We assume the influence of both effects to be linear, with the
parameters CQ,air determining the influence of the amount of
Note that the parameters a and Δθ are not independent: For air taken in, and CQ,vap determining the influence of the
any combination of parameter values {a, m, Δθ}, alternative evaporation cooling.
combinations {a*, m, Δθ*} can be found so that
Q w,intake = CQ ,airnair,in + CQ ,vapH vap,fuel (10)
1/(m + 1)
i *y
* = jjj a zzz Hvap,fuel is the product of nfuel,inj and the molar enthalpy of
ja z
k { (4) vaporization of the fuel hvap,fuel.
Therefore, we set a to 1. The parameter b is not part of the H vap,fuel = nfuel,inhvap,fuel (11)
original Wiebe function and was added by Yeliana et al.61 to
achieve a mass fraction burned of 1 within a finite crank angle. To estimate the parameters CQ,air and CQ,vap, the term Qw,intake is
The choice b = 1 would yield the original Wiebe function determined by using measured fuel mass flows mfuel , as
where full combustion is achieved at an infinite crank angle, explained in the section Parameter estimation. mfuel is related
whereas b > 1 yields full combustion at finite crank angles. Our to nfuel,in via the rotational speed N and the molar mass of the
choice for a then results in the following Wiebe function with fuel Mfuel.
four remaining parameters:
n fuel,inM fuelN
l ÄÅ É|
(m + 1)Ñ
mfuel =
o ÅÅ i ÑÑo
o ÅÅ j yz ÑÑo 2 (12)
o
m1 expÅÅÅ jjj
( ) = bo 0 zzz ÑÑo }
o ÑÑo
o
n
Å
ÅÅÇ k { ÑÑÖo
o
~ (5) For the exhaust stroke, we derive analogous mole and energy
balances to calculate the molar amount and temperature of the
Low Pressure Cycle. For the low-pressure cycle, the residual gas, but pragmatically neglect the heat transfer. The
pressure values are determined by intake and exhaust equations are given in the Supporting Information.
pressures. Further equations are needed to calculate the Knock Prediction. Knock occurs when the unburned air−
amount of substance in the cylinder n1, the in-cylinder fuel mixture autoignites. To predict the occurrence of knock,
temperature T1, and the molar amounts of air taken in nair,in, we need to estimate the temperature of the unburned gas Tub
and fuel injected nfuel,in. In accordance with the idealized during the combustion process. For this purpose, we follow the
modeling approach, we formulate a mole balance that equals n1 approach of Gschwend et al.55 who assumed identical pressure
to the sum of nair,in, nfuel,in, and the molar amount of exhaust gas p in burned and unburned gas and no heat transfer from
at state 6, n6. burned to unburned gas, reflecting typical assumptions,50 and
furthermore neglected the heat transfer from the unburned gas
n1 = n6 + nair,in + nfuel,in (6) to the cylinder wall, as the heat transfer is calculated in the
one-zone model using the temperature averaged over the
nair,in and nfuel,in are related through the stoichiometric whole cylinder. The temperature of the unburned zone can
coefficient of oxygen νO2 and the molar fraction of oxygen in
d

then be calculated with the following equation derived from

air xO2,air.
d
the energy balance for isentropic compression of the unburned
xO2,air gas,55 using the universal gas constant R and the heat capacity
nfuel,in = nair,in of the unburned gas c pub.
| O2| (7)
c pub dT ub 1 dp
We furthermore formulate an enthalpy balance for the intake ub
=
phase that equals the enthalpy of the fuel−air mixture at state RT dt p dt (13)
1, H1, to the sum of the enthalpy of the exhaust gas at state 6, Using the temperature of the unburned zone, the autoignition
H6, the enthalpy of the air taken in, Hair,in, the enthalpy of the delay time of the unburned fuel−air mixture is calculated using
fuel injected, Hfuel,in, and term for the wall heat transfer, the Livengood−Wu integral.62 It yields the value kI by
Qw,intake. integrating the reciprocal of the ignition delay time τ from
H1 = H6 + Hair,in + Hfuel,in + Q w,intake the time at which the combustion starts, t2, to the time
(8)
combustion is completed, t3.
Together with the ideal gas law, i.e. t3 1
kI = dt
p1 V1 = n1RT1 (9) t2 (p , T ub) (14)

4082 https://doi.org/10.1021/acs.energyfuels.4c04775
Energy Fuels 2025, 39, 4079−4093
Energy & Fuels pubs.acs.org/EF Article

The ignition delay time is commonly modeled with an model with measurement series of the alternative fuel
Arrhenius equation with the parameters A, TA, pref, and β. KEAA16,32 and conventional RON 95 E10 gasoline.
First, the heat transfer multipliers and the Wiebe function
i T yij p yz parameters are fitted to the pressure curve for each load point
= A expjjj A zzzjjjj zzzz separately. We discard load points with knock limitation, i.e.,
k T {jk pref z{ (15) points from the ethanol measurement series at intake pressure
63
Douaud and Eyzat proposed the values A = 0.01869 s(RON/ higher than 1.5 bar.16 During parameter estimation, the intake
100)3.4017, TA = 3800 K, pref = 1 × 105 Pa, and β = 1.7 for stroke is calculated in a different manner than described in the
gasoline fuels. However, alternative fuels may exhibit different previous section: We calculate nfuel,in from the fuel mass flow
autoignition behavior that is not adequately captured by the measured in the experiment. Then, we determine nair,in using
RON. Therefore, we parametrize Arrhenius models for every eq 7 and n1 using eq 6. T1 is then calculated using the ideal gas
blend component using kinetic ignition delay simulation data. law, and the enthalpy H1 is calculated using the molar amounts
For fuels that exhibit two-stage autoignition behavior, we use and T1. The wall heat transfer term Q*w,intake is then calculated
the 3-Arrhenius approach64 that models the ignition delay as a in the following manner:
function of three ignition delay times τ1, τ2, and τ3 for low, *
Q w,intake = H1 H6 Hair,in Hfuel,in (18)
medium, and high temperature regimes, respectively:
1 1 1 The heat transfer multipliers and the Wiebe function
= +
1 + 2 3 (16) parameters are fitted to minimize the absolute root mean
squared error of the simulated pressure p(θi) with regard to the
The ignition delay times for the specific regimes are each pressure measurements p( i) from θi = −175° to θi= 175°,
modeled with an Arrhenius equation. Details on kinetic where i is index for the measurement samples. Data for crank
mechanisms, software, simulated conditions, and fitting are angles between θi = −180° (BDC) to θi = −175° are discarded
listed in the Supporting Information. Ignition delay times of because the experimental pressure curve is affected by the valve
fuel blends can be subject to nonlinear mixing rules, depending closing right after BDC. Equally, data for crank angles between
upon kinetic interactions between the blend constituents and 175° (BDC) and 180° are discarded because the experimental
their combustion intermediates. However, dedicated rigorous pressure curve is affected by the valve opening right before
kinetic models are not available for most blend stock BDC. In summary, we minimize the least-squares error for
combinations. The mixture behavior can be idealized by every load point.
applying a linear-by-mole mixing rule to the logarithm of the
ignition delay time, as can be observed, e.g., for a mixture of min (p( i) p ( i))2
dimethyl ether and n-pentane.65 For blend stocks that are b, 0 , , m , C HT,Compr , C HT,Comb , C HT,Expan
i
predefined mixtures, such as gasoline, kinetic models are
available. Hence, we calculate the ignition delay of a fuel blend i { 175° , 175°} (19)
as a function the mole fractions of the blend stocks j, zfuel,j, Parameter estimation of dynamic nonlinear models can often
summing over the set of possible blend stock Nblend stocks. lead to suboptimal fits due to non-convexity.66 To avoid this,
ln = z fuel, j ln we use the MATLAB57 multistart optimization solver Global
j
j Nblend stocks (17) Search with fmincon as a local optimization solver. We provide
bounds for every parameter (see Table 2). The heat transfer
Parameter Estimation. Several parameters of the engine multipliers being correction factors to a semi-empirical heat
model must be estimated from experimental data: the heat
transfer multipliers CHT,Compr, CHT,Comb, and CHT,Expan, the Table 2. Generalized Parameter Values of Thermodynamic
parameters b, θ0, Δθ, and m of the Wiebe function (see eq 5), Engine Modela
and the parameters CQ,air and CQ,vap from the correlation that
models the heat transfer during intake (see eq 10). We use parameter lower bound upper bound value
measurement data obtained with a single-cylinder research heat transfer mulitplier
engine with a compression ratio of 16.4 described previously16 CHT,Compr 0.1 10 3.2
and perform regression using numerical optimization. Table 1 CHT,Comb 0.1 10 6.2
lists the most important specifications of the experimental CHT,Expan 0.1 10 0.1
engine. We estimate all model parameters using measurement switching point for heat transfer mulitplier
series of load variation for the alternative fuels ethanol and 2- ξHT,Expan,0 1
butanone under stoichiometric combustion conditions at a δHT,Expan −0.03
rotational speed of 2000 min−1. Subsequently, we validate the Wiebe function
b 1 1.1 1.04
Table 1. Specifications of the Single-Cylinder Research θ0 (deg) −11 0 −2.5
Engine16 Δθ (deg) 9 22 13.7
m 0.3 2.6 1.1
parameter value heat transfer during intake
CQ,vap 0 1 0.65
displacement volume (cm3) 500
CQ,air (J mol−1) 0 2000 920
bore (mm) 75
stroke (mm) 113.2 a
Note: For the heat transfer multipliers and the Wiebe function
connecting rod length (mm) 220 paramerters, the lower and upper bounds have been used during the
compression ratio 16.4 load point- and fuel-specific parameter estimation.

4083 https://doi.org/10.1021/acs.energyfuels.4c04775
Energy Fuels 2025, 39, 4079−4093
Energy & Fuels pubs.acs.org/EF Article

Figure 2. Parameters b, θ0, Δθ, m, CHT,Compr, CHT,Comb, and CHT,Expan for individual load points and fuels (denoted with blue markers) as well as the
averaged values, i.e., the deduced generalized parameter values (denoted with black lines). For CHT,Compr, the generalized parameter value is not
obtained by averaging, but by performing a sensitivity analysis of the knock calculation, resulting in a parameter value smaller than the average of
the load-point specific parameters. For the term Qw,intake, the black lines denote the values obtained with the load- and fuel-specific correlation (10)
for ethanol and 2-butanone.

transfer correlation are constrained to the interval [0.1, ..., 10].

The derivation of bounds for the Wiebe function parameters
are explained in section 3 of the Supporting Information.
The load-point specific parameters resulting from the
parameter estimation are plotted in Figure 2. To derive load-
independent parameters, we take the arithmetic average of the
load-point specific values for b, θ0, Δθ, m, CHT,Comb, and
CHT,Expan. For CHT,Compr, we make an exception, as the values
fitted to the pressure curves of ethanol alternate between very
high and very low values, reaching either the lower or the
upper bound for most load points. Apparently, the optimizer
tries to compensate for a structural model error by setting
CHT,Compr to either a very high or very low value. Instead of
taking the average of the load-point specific values, we use
CHT,Compr to fine-tune the prediction of knock, following
Vancoillie et al.48 To this end, we perform a sensitivity analysis
of CHT,Compr with respect to the knock prediction for the
highest load point of 2-butanone (which is not knock limited
in the experiment) and the second and third highest load
points of ethanol (one of which is knock limited, whereas the
other one is not). The first plot in Figure 3 shows the values of
the Livengood−Wu integral for the three selected load points Figure 3. Sensitivity of knock index kI with respect to the heat transfer
with varying CHT,Compr. It shows that knock is overestimated for multiplier during compression CHT,Compr (top) and influence of the
the highest load point of 2-butanone for CHT,Compr < 2.9 and heat transfer multiplier during expansion CHT,Expan on the root mean
knock is underestimated for the second highest load point of squared error (rmse) of the pressure curve during expansion
ethanol for CHT,Compr > 3.6. Hence, CHT,Compr should lie (bottom).
between 2.9 and 3.6 and we choose CHT,Compr = 3.2.
Furthermore, we note that all values for CHT,Expan are located
at the lower bound. A closer inspection suggests that the underestimated, which can be partly compensated for with a
optimizer tries to compensate for a structural model error by low value for CHT,Expan. To illustrate the effect on the calculated
setting CHT,Expan as low as possible. Specifically, a variation of net indicated work, we calculate the cumulative net indicated
CHT,Expan shows that the root mean squared error monoto- work as a function of θ (Figure 4, center).
nously decreases with decreasing CHT,Expan (see lower plot in V( )
Figure 3). Further clues can be derived from a comparison of Wi,n( ) = p( )dV ( )
V ( 6) (20)
the pressure curves of the experiment and the model. Figure 4
(top) shows the pressure curves for an exemplary load point of Whereas initially the cumulative net indicated work of the
2-butanone. Although, in general, the simulated pressure curve experiment is matched very well, a deviation becomes evident
(dashed line) matches the experimentally measured pressure for the expansion stroke after the combustion. Here, the
curve (solid line) well, the pressure during the expansion is pressure is underestimated an thus the net indicated work is
4084 https://doi.org/10.1021/acs.energyfuels.4c04775
Energy Fuels 2025, 39, 4079−4093
Energy & Fuels pubs.acs.org/EF Article

Figure 4. Experimental and simulated pressure curve (top) and cumulative net indicated work curve (center) during the high pressure cycle for 2-
butanone at intake pressure 1.94 bar, as well as simulated heat transfer (bottom). For the simulated values, the dashed lines denote the curves for
the model using the generalized parameters before correcting the switching point of the heat transfer multipliers CHT,Comb and CHT,Expan, whereas the
dotted line denotes the curves for the correction achieved with eq 22.

underestimated. This effect is reduced by choosing the lowest To implement a pragmatic correction of the underestimation
possible value for CHT,Expan. of the work, we modify the model by including an adjustable
A plot of the heat transfer over the crank angle (cf. Figure 4, switching point. To that end, we introduce a new parameter
bottom) shows very high values for the heat transfer during ξHT,Expan that determines the degree of fuel conversion at which
combustion and a discrete change to very low values when the the heat transfer multiplier is switched. First, we fit the value of
combustion is finished. The discontinuity is caused by the ξHT,Expan directly to the experimental values of the net indicated
discrete switch in the piece-wise constant heat transfer efficiency i,n for every load point, i.e., we solve
multipliers. The piece-wise definition of heat transfer multi- 2
min ( i,n )
pliers allows for a convenient calibration of the pressure curves HT,Expan
i,n
(21)
to experimental data, but it oversimplifies the physical
processes in the cylinder. CHT,Comb exhibits a high value to The load-point specific switching points are plotted in Figure
match the experimental pressure peak. However, with the 5. We observe an approximately linear correlation between
switching point to a lower value being determined by the intake pressure and the optimal switching point, resulting in
Wiebe function parameters, the heat transfer during later the following relation for determining ξHT,Expan:
phases of the combustion appears to be overestimated.
ji in zyz
p
HT,Expan j
Selecting a small value for CHT,Expan is not enough to HT,Expan = HT,Expan,0 + j z
compensate for the late switching. k 1 bar { (22)

4085 https://doi.org/10.1021/acs.energyfuels.4c04775
Energy Fuels 2025, 39, 4079−4093
Energy & Fuels pubs.acs.org/EF Article

efficiency, KEAA and 2-butanone perform similarly, and RON

95 E10 exhibits the lowest efficiency.
To evaluate the quality of the knock prediction, we compare
the intake pressure at which knock begins to occur in the
model with the intake pressure, at which the efficiencies
measured in the experiment decrease as a consequence of
spark retardation due to a knock limitation. For the model
prediction, we examine the values for the Livengood−Wu
integral kI, using both the Douaud−Eyzat model63 and a
kinetics-based model for each fuel. In Figure 7, the values for kI
are plotted over the intake pressure in one plot for each fuel,
and the areas that correspond to over- or underestimation of
knock are colored. The plots show that the Douaud−Eyzat
model only predicts knock of RON 95 E10 gasoline well. For
Figure 5. Optimal switching points ξHT,Expan for switching the heat the alternative fuels, the Douaud−Eyzat model overestimates
transfer multiplier of combustion to the heat transfer multiplier of the autoignition with increasing intake pressures, leading to
expansion. The markers denote the values obtained by fitting the net
indicated efficiency of every load point. The line denotes the derived
severe overestimation of knock already at throttled conditions.
general linear correlation (see eq 22). In contrast, the kinetics-based ignition delay model predicts
knock qualitatively well. However, we observe errors for the
fuels that have not been used in the calibration of the model.
The resulting pressure and cumulative pressure curve for one For KEAA, the model incorrectly predicts the absence of
exemplary load point are plotted in Figure 4 (dotted lines). knock, wheres in the experiment intake pressures greater than
Note that the prediction of the indicated work is corrected 1.7 bar lead to knock. Here, kI reaches a value of 0.83, which is
without altering the pressure curve peak. a significant but not severe deviation from a value greater than
Finally, the parameters CQ,air and CQ,vap are determined by 1. The underestimation of knock might be caused by the
minimizing the root mean squared deviation of the values of exclusion of nonlinear blending effects in the ignition delay
*
Qw,intake and Qw,intake, i.e., the values obtained during the time (see eq 17). Another factor could be the simplified fuel
estimation of load-point specific parameter and the values evaporation model that might lead to inaccurate in-cylinder
obtained with eq 10. temperatures. For RON 95 E10, the occurrence of knock is
slightly overestimated: the model predicts knock already at
min (CQ ,airnair,in, i + CQ ,vapH vap, i *
Q w,intake, i
)2 intake pressures of 0.5 bar, whereas in the experiment, a knock
CQ ,air , CQ ,vap
i limitation is observed at intake pressures higher than 0.6 bar.
(23) Unsurprisingly, the calculation of knock is correct for ethanol
The lower right plot in Figure 2 shows that the linear and 2-butanone, as data on these fuels have been used for the
correlation reproduces very well the values obtained for the calibration of the model. In summary, we consider the kinetics-
load-point specific simulations. The full set of the resulting based models a significant improvement over the Douaud−
model parameters is listed in Table 2. Eyzat model for the alternative fuels.
Validation. To validate the model, we compare the To validate the engine model for different compression
simulations against experimental data from our previous ratios, we compare its predictions with experimental data from
study16 on the fuels ethanol, 2-butanone, KEAA, and RON Wouters et al.19 using methanol as a fuel at compression ratios
95 E10, as shown in Figure 6 (left plot). The overall trends are of 10.8 and 17.7, as well as previously unpublished data using
very well reproduced by the model: ethanol has the highest ethanol at a compression ratio of 17.7 in the same engine (see

Figure 6. Efficiency calculation: the left plot shows net indicated efficiencies at varied intake pressures for RON 95 E10, ethanol, 2-butanone, and
KEAA, using single-cylinder research engine data from Ackermann et al.;16 the right plot shows indicated efficiencies for methanol with
compression ratios (CRs) of 10.8 and 17.7 using single-cylinder research engine data from Wouters et al.19 as well as previously unpublished data
for ethanol with CR 17.7, measured in the same engine. Note that for the right plot, the engine runs at a rotational speed of 2500 min−1, which
differs from the rotational speed of 2000 min−1 used in the measurements shown in the other plots. The markers denote the measurement values.
Nonfilled markers indicate that the ignition has been retarded due to knock limitation. The lines denote the simulation results. Dotted lines indicate
the occurrence of knock.

4086 https://doi.org/10.1021/acs.energyfuels.4c04775
Energy Fuels 2025, 39, 4079−4093
Energy & Fuels pubs.acs.org/EF Article

Figure 7. Knock calculation: The value of the Livengood−Wu integral kI is plotted over the intake pressure for the fuels ethanol (top left), 2-
butanone (top right), KEAA16,32 (bottom left), and RON 95 E10 (bottom right). Blue areas denote the regions where knock is overestimated,
green areas denote the regions where knock is underestimated. Dotted lines denote the values for kI calculated with the ignition delay model by
Douaud and Eyzat63 (15), dashed lines denote the values for kI calculated with Arrhenius ignition delay models fitted to data from zero-
dimensional kinetic simulations under isochoric conditions.

Figure 6, right plot) at a rotational speed of 2500 min−1, which cylinder mixture formation and thus driveability under cold
differs from the rotational speed of 2000 min−1 in the conditions.67 We tailor the volatility by imposing limits on the
experiments used of the fitting. Furthermore, the combustion vapor pressure and the enthalpy of vaporization, following
is phased differently, leading to burn profiles different from previous studies.31,32 Specifically, we use the bubble point
those used in the parameter estimation. Hence, we compare pressure pbubble as an approximation for the Reid vapor
the data in qualitative terms. The general trends are replicated pressure, as in previous fuel blend design studies,31,32,68 and
well: a higher compression ratio leads to higher indicated constrain it to the range [35 kPa, 100 kPa], following gasoline
efficiencies, and methanol yields a higher efficiency than standards8,9 and previous CAPD studies.31,32,68 The enthalpy
ethanol at the same compression ratio. At the low compression of vaporization is constrained to a value less than or equal to
ratio, the model does not predict knock for methanol at high 60 kJ/kgair, again following previous CAPD studies.30−32,43
loads where knock is observed in the experiment. At the high Because a high enthalpy of vaporization has a positive influence
compression ratio, the model slightly overpredicts knock for on engine efficiency due to charge cooling,38 the constraint
ethanol. potentially limits the maximum efficiency of the designed fuel.
We conclude that the engine model predicts trends in net The overall co-optimization problem reads
indicated efficiency and the onset of knock reasonably well for
fuels and compression ratios that were not used in the fitting. max i,n
( CR , pin , z)
CR [5,20], pin [1,2]bar, z [0,1]Nblend stocks
However, as we lack more validation data for a comprehensive
judgment of the model performance, in particular regarding the s.t. ppeak ( CR , pin , z) 200 bar
onset of knock, the fuels designed using the engine model must
be experimentally validated. kI( CR , pin , z) 0.99

■ CO-OPTIMIZATION OF FUEL BLEND AND ENGINE

Problem Formulation. We maximize the net indicated
pbubble (z)
pbubble (z)
35 kPa
100 kPa
efficiency ηi,n as a function of the compression ratio εCR, the h vap(z) 60 kJ/kg air
intake pressure pin, and the vector of molar blend stock
compositions z, whose size depends upon the number of z
i Nblend stocks i
=1
possible blend stocks. For the engine, we consider the
parameters of our single-cylinder research engine (see Table where ηi,n(εCR,pin,z), ppeak(εCR,pin,z), and kI(εCR,pin,z) are
1) and fix the rotational speed to 2000 min−1. Following obtained via integrating the engine model equations, and
Gschwend et al.,55 we allow for compression ratios between 5 pbubble(z) and hvap(z) are obtained via mixing rules described in
and 20, constrain the value of the Livengood−Wu integral kI to the Supporting Information. We first optimize the blend
less than or equal to 0.99 to avoid knock. The peak pressure composition z without the objective function and starting from
ppeak is constrained to less than or equal to the limit of our an equimolar mixture using the MATLAB57 local optimization
single-cylinder research engine, i.e., 200 bar.19 Moreover, we solver fmincon. If no feasible mixture is found, the blend stock
optimize the intake pressure, covering the range from fully system is discarded. If a feasible mixture is found, it is used to
opened throttle in naturally aspirated operation (1 bar) to initialize the blend composition in the co-optimization.
highly boosted engine operation (2 bar). The problem is nonlinear and potentially contains more
The fuel composition is constrained by fuel specifications than one local optimum. While we expect the net indicated
added to tailor the volatility of the fuel. An upper bound on the efficiency to increase monotonically with compression ratio
volatility is set to ensure safe handling and avoid cavitation in and intake pressure, no such statements can be made for the
the fuel pump system.67 At the same time, the fuel needs to fuel composition, given the nonlinear property relations.
exhibit a sufficient front-end volatility, which is crucial for in- Therefore, we use the MATLAB multistart optimization solver
4087 https://doi.org/10.1021/acs.energyfuels.4c04775
Energy Fuels 2025, 39, 4079−4093
Energy & Fuels pubs.acs.org/EF Article

Global Search with fmincon as a local optimization solver and optimized together with the engine using the above-described
set the maximum time to 2000 s. Because we do not provide procedure.
analytical expressions for the derivatives, fmincon approximates Resulting Blends and Engine Configurations. The co-
the derivatives with finite differences. Note that dynamic optimization yields 23 binary blends and 67 ternary blends.
models can be optimized more efficiently using dynamic Figure 8 shows an analysis of the top-rated binary blends. It
optimization solvers.69−71 For historical reasons, however, our can be noted that 7 of the 8 best-performing blends contain
model and optimization was implemented in MATLAB and we over 50 mol % of methyl acetate and a large fraction of ethyl
refrain from reimplemenation. We furthermore note that the acetate (41 mol %, blend 1) or a ketone (33−37 mol %, blends
degrees of freedom can be reduced by 1 by calculating the 4 and 7) or a small fraction of a volatility enhancer (7−13 mol
mole fraction of one component directly using the closure %, blends 2, 3, 5, and 6). These methyl acetate blends achieve
condition, which might be beneficial for some solvers. predicted net indicated efficiencies from 44.5 to 44.7%.
We select blend stocks based on our previous CAPD studies Considering the model uncertainties, they can be considered
for ultrahigh efficiency engines,31,77 where fuel blends were equally good. The efficiencies are achieved exploiting the
optimized with respect to production process performance. maximum compression ratio of 20 and intake pressures of 1.8
The selected blend stocks are listed in Table 3. Allowing 10 bar. In all cases, the intake pressure is limited by the maximum
possible blend stocks, the problem has 12 degrees of freedom. peak pressure, not by the occurrence of knock. Moreover, all
However, a blend containing all 10 blend stocks is unlikely to methyl acetate blends exhibit the maximum enthalpy of
emerge and probably undesirable as it would require a complex vaporization of 60 kJ/kgair.
production process. Instead, we restrict our attention to binary Only three blends (blends 6, 9, and 10) compete with the
and ternary blends and enumerate possible blend stock methyl acetate blends. They consist of a large ethyl acetate
combinations. fraction (85 to 95 mol %) and a small fraction of pentane,
To reduce the number of blends to be optimized, we cyclopentane, or gasoline. They achieve net indicated
categorize the blend stocks into low reactivity blend stocks and efficiencies of 44.1 to 44.5%, reaching compression ratios
volatility enhancing blend stocks. Table 3 shows that 6 of the between 16.8 and 19.7 and an intake pressures of 1.8 to 2.0
10 blend stocks lead to expected net indicated engine bar. They are both knock and peak pressure limited.
efficiencies between 43.7 and 45.8% (assuming a use as a After the top 10 blends, the next candidates utilize a ketone
pure-component fuel), whereas the other 4 blend stocks are as the main blend stock, with small fractions of cyclopentane,
interesting because of their volatility. Consequently, we refer to pentane, or gasoline as volatility enhancers. The achieved
the former blend stocks as low reactivity blend stocks and the efficiencies vary with the main blend stock: 43.2−43.4% (2-
latter blend stocks as volatility enhancers. Only allowing one butanone), and 42.7−43.2% (methyl isopropyl ketone), and
volatility enhancing blend stock per blend, we obtain 39 compression ratios range between 13.8 and 15.6, with the
possible binary systems and 60 possible ternary systems. Every maximum intake pressure of 2 bar. All of these blends are
possible binary and ternary blend stock system is then co- limited by knock but not by peak pressure. The bubble point
pressure always exhibits the minimum allowed value. A binary
Table 3. Blend Constituents with Boiling Point Tboil, Bubble blend of 2-butanone with n-hexane achieves a lower efficiency.
Point Pressure pbubble at 37.8 °C, Enthalpy of Vaporization Further blends contain ethanol or methanol as a main blend
hvap at 25 °C, and Maximum Achievable Engine Efficiency stock with gasoline, cyclopentane, or hexane as a second blend
ηi,n,maxa stock. The alcohol blends with cyclopentane or gasoline
achieve efficiencies of 41.0 to 42.1%, the alcohol blends with
hvap at pbubble at hexane achieve efficiencies below 40.0%. All alcohol-based
Tboil 25 °C 37.8 °C ηi,n,max
blend stock RON OS (°C) (kJ/kgair) (kPa) (%) blends contain a larger volatility enhancing fraction (27−43
mol %) than the ester and ketone-based blends. All nonmethyl-
low reactivity blend stocks
acetate blends are knock limited and do not achieve the
methanol 11074 274 65 181 32 45.8
methyl acetate 12027 027 57 67 49 44.8
maximum compression ratio.
ethyl acetate 11827 −227 77 52 23 44.6
All blends touch a volatility constraint. The blends
ethanol 10975 1975 78 103 16 44.5
containing ethyl acetate, 2-butanone, or MIPK use methyl
methyl 10927 727 94 38 12 43.9
acetate or a volatility enhancer to balance their unfavorably low
isopropyl bubble point pressure. The blends containing methyl acetate,
ketone methanol, and ethanol require other blend stocks to balance
2-butanone 11127 627 80 46 22 43.7 their unfavorably high enthalpy of vaporization. Considering
volatility enhancers the charge cooling effect of the light alcohols, the chosen limit
cyclopentane 10176 1676 49 27 68 40.7 for the enthalpy of vaporization might limit the maximum
gasoline 9676 776 25 59 38.9 achievable efficiency. Indeed, if we increase the upper limit on
n-pentane 6276 −176 36 24 108 the enthalpy of vaporization to 80 kJ/kgair, the two best blends
n-hexane 2576 −176 69 24 34 contain methyl acetate and either methanol or ethanol. Both
a
blends are predicted to achieve an efficiency of 44.9% with an
n-Pentane and n-hexane are no feasible pure-component fuels in the enthalpy of vaporization of 80 kJ/kgair. The third best fuel
engine model due to their high auto-ignition propensity, hence, no
would be pure methyl acetate achieving an efficiency of 44.8%.
value for ηi,n,max is given. Thermophysical properties are taken from
the DIPPR database,72 the sources for RON and OS are indicated in Interestingly, the ternary blends do not enable higher engine
the table. For gasoline, we use a surrogate.73 The molar composition efficiencies than the binary blends. Two ternary blends
of the gasoline surrgate and the mixing rules for hvap and pbubble are containing methyl acetate and ethyl acetate with a small
given in the Supporting Information; for RON and OS we use linear- share of either methanol or ethanol can compete with the best
by-mole mixing rules. binary blends. A graphical analysis of the top 25 ternary blends,
4088 https://doi.org/10.1021/acs.energyfuels.4c04775
Energy Fuels 2025, 39, 4079−4093
Energy & Fuels pubs.acs.org/EF Article

Figure 8. Binary blends resulting from the co-optimization. We sort the blends by their achievable net indicated efficiency ηi,n,max (plotted at the
top). We furthermore plot engine design and operation variables and the fuel properties in relation to their upper and lower bounds: the optimal
compression ratio εCR,opt, the optimal intake pressure pin,opt, the resulting value of the Livengood−Wu integral kI as a measure for knock, the
resulting peak pressure ppeak, the enthalpy of vaporization hvap, the bubble point pressure pbubble, and the molar composition zfuel.

Figure 9. Comparison of the engine efficiency predicted by our model with RON (left), OI11 (center-left), and merit function2 (center-right and
right) for all binary blends. Research and motor octane numbers of the blends were calculated using the linear-by-mole mixing rule. For the
gasoline-containing blends, the efficiency values calculated with the engine efficiency merit function are obtained with the gasoline surrogate73 used
throughout this study.

similar to the one for the binary blends (Figure 8), can be that rely on detailed kinetic mechanisms. Leppard12 links low
found in Figure S1 of the Supporting Information. octane sensitivity with low-temperature heat release observed
The co-optimization thus reveals small esters as the most in alkane combustion. However, methyl acetate and ethyl
promising blend stocks, outperforming the ketone and alcohol acetate do not exhibit low-temperature heat release.79 It is
blends. Several studies have pointed out the octane boosting therefore questionable whether octane sensitivity is a useful
qualities of methyl and ethyl acetate.27,32,78 In a previous measure for the knock resistance of small esters.
molecular design study,43 we also identified small esters as The engine model predicts methyl acetate blends to be
potential fuel candidates to maximize SI engine efficiency, but
limited by the peak pressure restriction, not by the occurrence
pointed to their low octane sensitivities and low laminar
of knock. This is an interesting finding because, in our previous
burning velocities as potential drawbacks. A presumable
influence of the laminar burning velocity is ignored by the experimental investigations, even highly knock-resistant fuels
engine model, as the burning progress is equal for all fuels. exhibited knock under extreme conditions, e.g., methanol in an
Thus it is conceivable that the engine model overestimates the engine with a compression ratio of 17.7.19 Assuming the
efficiencies for small esters. The octane sensitivity, however, is absence of knock is predicted correctly, the use of stronger
a measure for the influence of engine conditions on knock. Our engine materials could facilitate higher peak pressures and thus
model simulates the engine conditions and predicts the even higher efficiencies of methyl acetate blends. In general,
occurrence of knock using ignition delay surrogate models the material compatibility of small esters needs to be
4089 https://doi.org/10.1021/acs.energyfuels.4c04775
Energy Fuels 2025, 39, 4079−4093
Energy & Fuels pubs.acs.org/EF Article

investigated, in order to assess weather they are suitable for correlation. Note that the different engine configurations
engines and fuel systems. associated with each blend resulting from the co-optimization
Several top blends do not require a volatility enhancer. may be representative of different K values. Overall, we
Instead, they balance the volatility related properties of the conclude that RON is probably the best proxy for engine
low-reactivity blend stocks; i.e., they combine the high vapor efficiency as it is predicted by our model.
pressure of methyl acetate with the low enthalpy of
vaporization of ethyl acetate or a ketone. For blends that do
contain volatility enhancers, the co-optimization favors
gasoline, pentane, and cyclopentane over n-hexane.
■ CONCLUSION
We present an approach to computational co-optimization of
Cyclopentane and gasoline are predicted to not impede too fuel and engine. To this end, we derived a thermodynamic
strongly the knock resistance of a blend under rather extreme engine model based on an ideal engine cycle, largely following
engine conditions. It shall be noted though that the linear-by- standard assumptions and the approach by Gschwend et al.55
mole mixing rule for the ignition delay could distort that Importantly, our model is applicable to new fuel candidates
prediction. The knock resistance of the those blends must that have not been tested in an engine yet, as it relies solely on
therefore be confirmed or falsified in experiments. Pentane thermophysical property models and ignition delay models
exhibits a very high vapor pressure and therefore small fitted to detailed kinetic simulation data. Validation against
quantities of it suffice to meet the vapor pressure restrictions. experimental data from a single-cylinder research engine
Hexane is notably more reactive than pentane and at the same showed that the model is capable of reproducing trends in
time exhibits a lower vapor pressure, rendering it an inferior efficiency and, with some compromises, the occurrence of
choice. engine knock as a function of fuel composition, compression
Comparison of Predicted Maximum Efficiency to ratio, and intake pressure. We then selected 10 blend stocks on
RON, OI, and Merit Function. We compare the predicted the basis of previous fuel design studies and enumerated binary
engine efficiencies to RON and OI, which are often considered and ternary blend stock systems. Utilizing the model, we
proxies for achievable engine efficiency, and to the engine jointly optimized fuel composition, compression ratio for every
efficiency merit function of Szybist et al.,2 which is a linear blend stock system, and intake pressure under consideration of
combination of multiple fuel properties. Figure 9 shows the fuel volatility restrictions as well as knock and peak pressure
values of RON, OI, and the merit function in comparison to limitations.
the efficiency predicted with our model. For the evaluation of The small esters methyl acetate and ethyl acetate are
OI, we choose a K value of −1 as presumably representative for predicted to constitute particularly promising fuel components.
highly boosted conditions.10 For the evaluation of the merit They are competitive to small alcohols in terms of knock
function we use K values of both 0 and −1. The values for resistance, but exhibit better volatility characteristics. In
RON and MON are calculated using linear-by-mole mixing contrast to previous computer-aided fuel design studies, the
rules. co-optimization does not only identify fuels, but also
The merit function requires property values for RON, OS, corresponding engine configurations that may guide subse-
enthalpy of vaporization, and the laminar burning velocity quent first experimental investigations. For example, methyl-
(LBV) at a pressure of 1.013 25 bar, a temperature of 353 K, acetate-rich blends are predicted to achieve their maximum
and an air−fuel equivalence ratio of 1.1. For RON, OS, and engine efficiency at the maximum allowed compression ratio of
enthalpy of vaporization we use literature data as indicated in 20, with the boost pressure limited to 1.8 bar. Moreover, the
Table 3. For the LBV, we use values predicted with the method predicts whether the efficiency may be limited by peak
QSPR−ANN by Vom Lehn et al.80 The merit function pressure, knock, or both. The methyl-acetate-rich blends, e.g.,
requires the assumption of a baseline efficiency with properties are predicted to be not knock limited, even under extreme
of the corresponding reference fuel. Because we use the conditions, but peak pressure limited.
gasoline surrogate by Sarathy et al.73 in our model, we use both While our study improves upon previous computer-aided
its properties and its maximum achievable net indicated fuel design studies by following a fully integrated, computa-
efficiency as calculated by our engine model (see Table 3) to tional co-optimization approach, the results are prone to
define the baseline. Following Szybist et al.,2 the merit function uncertainties caused by the simplifications that were necessary
can then be computed as in deriving the engine model. To improve model accuracy, a
ÄÅ two-zone model50 could be implemented, and several
ÅÅÅ i
ÅÅ1 + 0.01jjjj RON 96 OS 7 submodels need to be improved, namely, the evaporation
merit = 38.9%Å ÅÅ j K
ÅÇ k 1.6 1.6 model, the combustion model, and the knock prediction. The
simple evaporation model might cause significant inaccuracies
0.0085(h vap 25 kJ/kgair)
+ in the calculated temperature profile. A reduced-order droplet
1.6 model, e.g., the one published by Pu et al.,81 might constitute
h vap 25 kJ/kgair LBV 44.4 cm/s yÑÉÑÑ an improvement, but it would also require a more detailed gas
zzÑÑ
+ + zÑÑ exchange model and thus increase the computational complex-
15.2 5.4 {ÑÖ (24) ity. The combustion model assumes identical burn profiles for
The plot shows a relatively strong correlation between the every fuel, although the combustion is influenced by fuel
efficiency predictions by our model and the RON. The chemistry.3 A turbulent flame model in combination with
correlation with the OI with a K value of −1 appears to be correlations for the laminar burning velocity50 could improve
weaker. Similarly for the merit function, the choice of K the predicted burn profiles. The knock prediction relies on a
impacts the correlation significantly: a K value of 0, denoting linear-by-mole mixing rule for calculating the ignition delay of
no influence of the octane sensitivity, leads to a moderately fuel blends, neglecting potential synergistic combustion kinetic
strong correlation, whereas a value of −1 leads to a rather weak effects altering the knock propensity of multicomponent fuels.
4090 https://doi.org/10.1021/acs.energyfuels.4c04775
Energy Fuels 2025, 39, 4079−4093
Energy & Fuels pubs.acs.org/EF Article

Further engine efficiency increases can be achieved by lean Stefan Pischinger − Chair of Thermodynamics of Mobile
combustion.19 This would however require considering air− Energy Conversion Systems (TME), RWTH Aachen
fuel-ratio-dependent models of burn profiles and ignition delay University, 52074 Aachen, Germany
time models as well as the lean burn limit of a fuel. Alexander Mitsos − JARA-ENERGY, 52056 Aachen,
Importantly, the model mimics a single-cylinder research Germany; Process Systems Engineering (AVT.SVT), RWTH
engine that is optimized for a single load-point. In the long run, Aachen University, 52074 Aachen, Germany; Institute of
computational co-optimization should consider an actual Climate and Energy Systems, Energy Systems Engineering
vehicle engine and consider a wide range of load points as in (ICE-1), Forschungszentrum Jülich GmbH, 52425 Jülich,
a driving cycle. Germany; orcid.org/0000-0003-0335-6566
In summary, our study constitutes an important first step in Complete contact information is available at:
the computational co-optimization of fuel and engine and https://pubs.acs.org/10.1021/acs.energyfuels.4c04775
yielded fuel candidates and recommended engine configu-
rations for subsequent experimental investigations. To further Author Contributions
advance the use of thermodynamic engine models in fuel
design, considerable modeling efforts will be required. Philipp Ackermann and Manuel Dahmen formulated the
research gap and scope of the study. Philipp Ackermann
■ ASSOCIATED CONTENT
Data Availability Statement
derived the engine model with conceptual input from Manuel
Dahmen, Patrick Burkardt, and Bastian Lehrheuer. Philipp
Ackermann and Benjamin Auer implemented the engine
The data underlying this study are openly available in our model. Philipp Morsch provided conceptual input for the
GitLab repository “0D SI engine model for fuel design” at ignition delay model and software for kinetic simulations.
https://git.rwth-aachen.de/avt-svt/public/0d-si-engine-model-
Patrick Burkardt provided the experimental data and
for-fuel-design.
conceptual input for parameter estimation and validation.
*
sı Supporting Information
Philipp Ackermann and Manuel Dahmen developed the
The Supporting Information is available free of charge at parameter estimation procedure with conceptual input from
https://pubs.acs.org/doi/10.1021/acs.energyfuels.4c04775. Alexander Mitsos. Philipp Ackermann and Benjamin Auer
Remaining engine model equations, details on ignition formulated the optimization problem with conceptual input
delay time simulations, details on estimation of ignition from Alexander Mitsos and Manuel Dahmen. Philipp
delay time model parameters, list of all ignition delay Ackermann derived the fuel requirements and selected the
time model parameters, derivation of bounds for Wiebe blend stocks with input from Manuel Dahmen. Alexander
function parameters, and illustration of top 25 ternary Mitsos, Karl Alexander Heufer, and Stefan Pischinger
blends (PDF) supervised the project. Philipp Ackermann wrote the original
draft. All authors reviewed and edited the manuscript and gave
■ AUTHOR INFORMATION
Corresponding Author
their comments for improvements.
Notes
Manuel Dahmen − Institute of Climate and Energy Systems, The authors declare no competing financial interest.
Energy Systems Engineering (ICE-1), Forschungszentrum
Jülich GmbH, 52425 Jülich, Germany; orcid.org/0000-
0003-2757-5253; Email: [email protected] ■ ACKNOWLEDGMENTS
The authors gratefully acknowledge funding by the Deutsche
Authors Forschungsgemeinschaft (DFG, German Research Founda-
Philipp Ackermann − Process Systems Engineering tion) under Germany’s Excellence Strategy, Cluster of
(AVT.SVT), RWTH Aachen University, 52074 Aachen, Excellence 2186 “The Fuel Science Center” ID 390919832.
Germany Manuel Dahmen received funding from the Helmholtz
Benjamin Auer − Process Systems Engineering (AVT.SVT), Association of German Research Centers. The authors kindly
RWTH Aachen University, 52074 Aachen, Germany thank Sophia Rupprecht for her help in implementing the
Patrick Burkardt − Chair of Thermodynamics of Mobile processing of the thermophysical property data and Sanket
Energy Conversion Systems (TME), RWTH Aachen Girhe for providing the laminar burning velocity predictions.
University, 52074 Aachen, Germany; orcid.org/0000- Furthermore, the authors kindly thank Adrian Schloßhauer for
0001-9433-4922 the fruitful discussions about thermodynamic engine modeling.
Bastian Lehrheuer − Chair of Thermodynamics of Mobile
Energy Conversion Systems (TME), RWTH Aachen
University, 52074 Aachen, Germany
Philipp Morsch − Chair of High Pressure Gas Dynamics
■ REFERENCES
(1) Kalghatgi, G.; Levinsky, H.; Colket, M. Future transportation
(HGD), Shock Wave Laboratory, RWTH Aachen University, fuels. Prog. Energy Combust. Sci. 2018, 69, 103−105.
(2) Szybist, J. P.; Busch, S.; McCormick, R. L.; Pihl, J. A.; Splitter, D.
52074 Aachen, Germany; Institute for a Sustainable
A.; Ratcliff, M. A.; Kolodziej, C. P.; Storey, J. M.; Moses-DeBusk, M.;
Hydrogen Economy, Process and Plant Engineering for Vuilleumier, D.; Sjöberg, M.; Sluder, C. S.; Rockstroh, T.; Miles, P.
Chemical Hydrogen Storage (INW-4), Forschungszentrum What fuel properties enable higher thermal efficiency in spark-ignited
Jülich GmbH, 52425 Jülich, Germany engines? Prog. Energy Combust. Sci. 2021, 82, 100876.
Karl Alexander Heufer − Chair of High Pressure Gas (3) Heywood, J. B. Internal Combustion Engine Fundamentals;
Dynamics (HGD), Shock Wave Laboratory, RWTH Aachen McGraw-Hill: New York, 1988.
University, 52074 Aachen, Germany; orcid.org/0000- (4) Wang, Z.; Liu, H.; Reitz, R. D. Knocking combustion in spark-
0003-1657-5231 ignition engines. Prog. Energy Combust. Sci. 2017, 61, 78−112.

4091 https://doi.org/10.1021/acs.energyfuels.4c04775
Energy Fuels 2025, 39, 4079−4093
Energy & Fuels pubs.acs.org/EF Article

(5) Splitter, D.; Pawlowski, A.; Wagner, R. A historical analysis of the a spark-ignition engine for passenger car applications. Fuel 2023, 331,
co-evolution of gasoline octane number and spark-ignition engines. 125579.
Front. Mech. Eng. 2016, 1, 16. (26) Hoppe, F.; Heuser, B.; Thewes, M.; Kremer, F.; Pischinger, S.;
(6) ASTM International. Materials Standard Test Method for Research Dahmen, M.; Hechinger, M.; Marquardt, W. Tailor-made fuels for
Octane Number of Spark-Ignition Engine Fuel; ASTM International: future engine concepts. International Journal of Engine Research 2016,
West Conshohocken, PA, 2018. 17, 16−27.
(7) ASTM International. Materials Standard Test Method for Motor (27) McCormick, R. L.; Fioroni, G.; Fouts, L.; Christensen, E.;
Octane Number of Spark-Ignition Engine Fuel; ASTM International: Yanowitz, J.; Polikarpov, E.; Albrecht, K.; Gaspar, D. J.; Gladden, J.;
West Conshohocken, PA, 2019. George, A. Selection Criteria and Screening of Potential Biomass-
(8) ASTM International. Materials Standard Specification for Derived Streams as Fuel Blendstocks for Advanced Spark-Ignition
Automotive Spark-Ignition Engine Fuel; ASTM International: West Engines. SAE International Journal of Fuels and Lubricants 2017, 10,
Conshohocken, PA, 2021. 442−460.
(9) German Institute for Standardization (DIN). EN (28) Leitner, W.; Klankermayer, J.; Pischinger, S.; Pitsch, H.; Kohse-
228:2012+A1:2017, Automotive Fuels�Unleaded Petrol�Require- Höinghaus, K. Advanced Biofuels and Beyond: Chemistry Solutions
ments and Test Methods; DIN: Berlin, Germany, 2017. for Propulsion and Production. Angewandte Chemie (International ed.
(10) Kassai, M.; Aksu, C.; Shiraishi, T.; Cracknell, R.; Shibuya, M. in English) 2017, 56, 5412−5452.
Mechanism Analysis on the Effect of Fuel Properties on Knocking (29) Zhang, L.; Mao, H.; Liu, Q.; Gani, R. Chemical product
Performance at Boosted Conditions. SAE Tech. Pap. Ser. 2019, 2019- design−recent advances and perspectives. Current Opinion in Chemical
01-0035. Engineering 2020, 27, 22−34.
(11) Kalghatgi, G. T. Fuel Anti-Knock Quality�Part I. Engine (30) Dahmen, M.; Marquardt, W. Model-Based Design of Tailor-
Studies. SAE Tech. Pap. Ser. 2001, 2001-01-3584. Made Biofuels. Energy Fuels 2016, 30, 1109−1134.
(12) Leppard, W. R. The Chemical Origin of Fuel Octane (31) Dahmen, M.; Marquardt, W. Model-Based Formulation of
Sensitivity. SAE Int. J. Fuels Lubr. 1990, 99, 862−876. Biofuel Blends by Simultaneous Product and Pathway Design. Energy
(13) Wang, C.; Xu, H.; Daniel, R.; Ghafourian, A.; Herreros, J. M.; Fuels 2017, 31, 4096−4121.
Shuai, S.; Ma, X. Combustion characteristics and emissions of 2- (32) König, A.; Siska, M.; Schweidtmann, A. M.; Rittig, J. G.; Viell,
methylfuran compared to 2,5-dimethylfuran, gasoline and ethanol in a J.; Mitsos, A.; Dahmen, M. Designing Production-Optimal Alternative
DISI engine. Fuel 2013, 103, 200−211. Fuels for Conventional, Flexible-Fuel, and Ultra-High Efficiency
(14) Hoppe, F.; Burke, U.; Thewes, M.; Heufer, A.; Kremer, F.; Engines. Chem. Eng. Sci. 2021, 237, 116562.
Pischinger, S. Tailor-Made Fuels from Biomass: Potentials of 2- (33) Vom Lehn, F.; Cai, L.; Tripathi, R.; Broda, R.; Pitsch, H. A
butanone and 2-methylfuran in direct injection spark ignition engines. property database of fuel compounds with emphasis on spark-ignition
Fuel 2016, 167, 106−117. engine applications. Applications in Energy and Combustion Science
(15) Burkardt, P.; Ottenwälder, T.; König, A.; Viell, J.; Mitsos, A.; 2021, 5, 100018.
(34) Cai, G.; Liu, Z.; Zhang, L. Transformation rule-based molecular
Wouters, C.; Marquardt, W.; Pischinger, S.; Dahmen, M. Toward co-
evolution for automatic gasoline molecule design. Chem. Eng. Sci.
optimization of renewable fuel blend production and combustion in
2022, 263, 118119.
ultra-high efficiency SI engines. International Journal of Engine
(35) Rittig, J. G.; Ritzert, M.; Schweidtmann, A. M.; Winkler, S.;
Research 2023, 24, 29−41.
Weber, J. M.; Morsch, P.; Heufer, K. A.; Grohe, M.; Mitsos, A.;
(16) Ackermann, P.; et al. Designed to Be Green, Economic, and
Dahmen, M. Graph Machine Learning for Design of High-Octane
Efficient: A Ketone-Ester-Alcohol-Alkane Blend for Future Spark- Fuels. AIChE J. 2023, 69, No. e17971.
Ignition Engines. ChemSusChem 2021, 14, 5254−5264. (36) Vuilleumier, D.; Huan, X.; Casey, T.; Sjöberg, M. Uncertainty
(17) Brusstar, M.; Stuhldreher, M.; Swain, D.; Pidgeon, W. High Assessment of Octane Index Framework for Stoichiometric Knock
Efficiency and Low Emissions from a Port-Injected Engine with Neat Limits of Co-Optima Gasoline Fuel Blends. SAE International Journal
Alcohol Fuels. SAE Int. J. Fuels Lubr. 2002, 111, 1445−1451. of Fuels and Lubricants 2018, 11, 247−270.
(18) Verhelst, S.; Turner, J. W. G.; Sileghem, L.; Vancoillie, J. (37) Zhou, Z.; Yang, Y.; Brear, M.; Lacey, J.; Leone, T. G.;
Methanol as a fuel for internal combustion engines. Prog. Energy Anderson, J. E.; Shelby, M. H. A Comparison of Four Methods for
Combust. Sci. 2019, 70, 43−88. Determining the Octane Index and K on a Modern Engine with
(19) Wouters, C.; Burkardt, P.; Pischinger, S. Limits of compression Upstream, Port or Direct Injection. SAE Tech. Pap. Ser. 2017, 2017-
ratio in spark-ignition combustion with methanol. International 01-0666.
Journal of Engine Research 2022, 23, 793−803. (38) Kasseris, E.; Heywood, J. B. Charge Cooling Effects on Knock
(20) Feng, H.; Lai, K.; Zheng, Z.; Lin, S.; Wu, X.; Tang, Q. Effects of Limits in SI DI Engines Using Gasoline/Ethanol Blends: Part 2-
methanol direct injection and high compression ratio on improving Effective Octane Numbers. SAE International Journal of Fuels and
the performances of a spark-ignition passenger car engine. Fuel 2024, Lubricants 2012, 5, 844−854.
357, 130052. (39) Leone, T. G.; Anderson, J. E.; Davis, R. S.; Iqbal, A.; Reese, R.
(21) Brinkman, N. D. Effect of compression ratio on exhaust A.; Shelby, M. H.; Studzinski, W. M. The Effect of Compression
emissions and performance of a methanol-fueled single-cylinder Ratio, Fuel Octane Rating, and Ethanol Content on Spark-Ignition
engine. SAE Tech. Pap. Ser. 1977, 770791. Engine Efficiency. Environ. Sci. Technol. 2015, 49, 10778−10789.
(22) Larsen, U.; Johansen, T.; Schramm, J. Ethanol as a Fuel for Road (40) Zhang, B.; Sarathy, S. M. Lifecycle optimized ethanol-gasoline
Transportation; Technical University of Denmark: Kongens Lyngby, blends for turbocharged engines. Applied Energy 2016, 181, 38−53.
Denmark, 2009; https://amf-tcp.org/app/webroot/files/file/ (41) Chen, Y.; Zheng, Z.; Lu, Z.; Wang, H.; Wang, C.; Sun, X.; Xu,
Annex%20Reports/AMF_Annex_35-1.pdf (accessed Aug 29, 2024). L.; Yao, M. Machine learning-based screening of fuel properties for SI
(23) Nithyanandan, K.; Zhang, J.; Li, Y.; Wu, H.; Lee, T. H.; Lin, Y.; and CI engines using a hybrid group extraction method. Applied
Lee, C.-f. F. Improved SI engine efficiency using Acetone−Butanol− Energy 2024, 366, 123257.
Ethanol (ABE). Fuel 2016, 174, 333−343. (42) Nagaraja, S. S.; Sarathy, S. M.; Mohan, B.; Chang, J. Machine
(24) Tang, Q.; Ren, K.; Xie, X.; Chen, T.; Jiang, P.; Zhang, D. The learning-driven screening of fuel additives for increased spark-ignition
Effect of Acetone-Butanol-Ethanol and Gasoline Blends on the engine efficiency. Proceedings of the Combustion Institute 2024, 40,
Knocking Performance of Spark-Ignition Engine. Thermal Science and 105658.
Engineering Progress 2023, 46, 102175. (43) Fleitmann, L.; Ackermann, P.; Schilling, J.; Kleinekorte, J.;
(25) Burkardt, P.; Wouters, C.; Pischinger, S. Beating neat fuels by Rittig, J. G.; Vom Lehn, F.; Schweidtmann, A. M.; Pitsch, H.;
dedicated blending: Performance analysis of bio-hybrid fuel blends on Leonhard, K.; Mitsos, A.; Bardow, A.; Dahmen, M. Molecular Design

4092 https://doi.org/10.1021/acs.energyfuels.4c04775
Energy Fuels 2025, 39, 4079−4093
Energy & Fuels pubs.acs.org/EF Article

of Fuels for Maximum Spark-Ignition Engine Efficiency by practical fuels at HCCI conditions: High-pressure shock tube
Combining Predictive Thermodynamics and Machine Learning. experiments and phenomenological modeling. Fuel 2012, 93, 492−
Energy Fuels 2023, 37, 2213−2229. 501.
(44) Foong, T. M.; Morganti, K. J.; Brear, M. J.; da Silva, G.; Yang, (65) Jiang, X.; Deng, F.; Yang, F.; Huang, Z. Ignition Delay
Y.; Dryer, F. L. The Effect of Charge Cooling on the RON of Characteristics and Kinetic Investigation of Dimethyl Ether/ n
Ethanol/Gasoline Blends. SAE International Journal of Fuels and -Pentane Binary Mixtures: Interpreting the Effect of the Equivalence
Lubricants 2013, 6, 34−43. Ratio and Dimethyl Ether Blending. Energy Fuels 2018, 32, 3814−
(45) Ratcliff, M. A.; Burton, J.; Sindler, P.; Christensen, E.; Fouts, L.; 3823.
McCormick, R. L. Effects of Heat of Vaporization and Octane (66) Singer, A. B.; Taylor, J. W.; Barton, P. I.; Green, W. H. Global
Sensitivity on Knock-Limited Spark Ignition Engine Performance. Dynamic Optimization for Parameter Estimation in Chemical
SAE Tech. Pap. Ser. 2018, 2018-01-0218. Kinetics. J. Phys. Chem. A 2006, 110, 971−976.
(46) Mohanan, P.; Babu, M. G. A simulation model for a methanol (67) Yanowitz, J.; McCormick, R. L. Review: Fuel Volatility
fueled turbocharged multi-cylinder automotive spark ignition engine. Standards and Spark-Ignition Vehicle Driveability. SAE International
SAE Tech. Pap. Ser. 1991, 912417. Journal of Fuels and Lubricants 2016, 9, 408−429.
(47) Shen, Y.; Bedford, J.; Wichman, I. S. Thermodynamic modeling (68) Yunus, N. A.; Gernaey, K. V.; Woodley, J. M.; Gani, R. A
of direct injection methanol fueled engines. Applied Thermal systematic methodology for design of tailor-made blended products.
Engineering 2009, 29, 2379−2385. Comput. Chem. Eng. 2014, 66, 201−213.
(48) Vancoillie, J.; Sileghem, L.; Verhelst, S. Development and (69) Hart, W. E.; Watson, J.-P.; Woodruff, D. L. Pyomo: modeling
validation of a quasi-dimensional model for methanol and ethanol and solving mathematical programs in Python. Mathematical
fueled SI engines. Applied Energy 2014, 132, 412−425. Programming Computation 2011, 3, 219−260.
(49) Nguyen, D.-K.; Suijs, W.; Sileghem, L.; Verhelst, S. Modeling of (70) Caspari, A.; Bremen, A. M.; Faust, J. M. M.; Jung, F.; Kappatou,
a methanol fuelled direct-injection spark-ignition engine with C. D.; Sass, S.; Vaupel, Y.; Hannemann-Tamás, R.; Mhamdi, A.;
reformed-exhaust gas recirculation. SAE Tech. Pap. Ser. 2021, 2021- Mitsos, A. DyOS�A Framework for Optimization of Large-Scale
01-0445. Differential Algebraic Equation Systems. 29th European Symposium on
(50) Verhelst, S.; Sheppard, C. Multi-zone thermodynamic Computer Aided Process Engineering; Computer Aided Chemical
modelling of spark-ignition engine combustion − An overview. Engineering; Elsevier: Amsterdam, Netherlands, 2019; Vol. 46; pp
Energy Conversion and Management 2009, 50, 1326−1335. 619−624, DOI: 10.1016/B978-0-12-818634-3.50104-1.
(51) Perini, F.; Paltrinieri, F.; Mattarelli, E. A quasi-dimensional (71) Andersson, J. A. E.; Gillis, J.; Horn, G.; Rawlings, J. B.; Diehl,
combustion model for performance and emissions of SI engines M. CasADi: a software framework for nonlinear optimization and
running on hydrogen−methane blends. Int. J. Hydrogen Energy 2010, optimal control. Mathematical Programming Computation 2019, 11,
35, 4687−4701. 1−36.
(52) Djouadi, A.; Bentahar, F. Combustion study of a spark-ignition (72) American Institute of Chemical Engineers (AIChE). DIPPR
engine from pressure cycles. Energy 2016, 101, 211−217. 801 Database, Version 12.3.0; AIChE: New York, 2018; http://www.
(53) Pourkhesalian, A. M.; Shamekhi, A. H.; Salimi, F. Alternative aiche.org/dippr.
fuel and gasoline in an SI engine: A comparative study of performance (73) Sarathy, S. M.; et al. Compositional effects on the ignition of
and emissions characteristics. Fuel 2010, 89, 1056−1063. FACE gasolines. Combust. Flame 2016, 169, 171−193.
(54) Caton, J. A. Implications of fuel selection for an SI engine: (74) Naegeli, D. W.; Yost, D. M.; Moulton, D. S.; Owens, E. C.;
Results from the first and second laws of thermodynamics. Fuel 2010, Chui, G. K. The Measurement of Octane Numbers for Methanol and
89, 3157−3166. Reference Fuels Blends. SAE Int. J. Fuels Lubr. 1989, 98, 712−722.
(55) Gschwend, D.; Soltic, P.; Edinger, P.; Wokaun, A.; Vogel, F. (75) Yanowitz, J.; Christensen, E.; McCormick, R. L. Utilization of
Performance evaluation of gasoline alternatives using a thermody- Renewable Oxygenates as Gasoline Blend Components; National
namic spark-ignition engine model. Sustainable Energy Fuels 2017, 1, Renewable Energy Laboratory (NREL): Golden, CO, 2011;
1991−2005. Technical Report NREL/TP-5400-50791, http://www.nrel.gov/
(56) Gschwend, D.; Soltic, P.; Wokaun, A.; Vogel, F. Review and docs/fy11osti/50791.pdf (accessed Aug 29, 2024).
Performance Evaluation of Fifty Alternative Liquid Fuels for Spark- (76) Derfer, J. M.; Boord, C. E.; Burk, F. C.; Hess, R. E.; Lovell, W.
Ignition Engines. Energy Fuels 2019, 33, 2186−2196. G.; Randall, R. A.; Sabina, J. R. Knocking Characteristics of Pure
(57) The MathWorks, Inc. MATLAB, Version 9.8.0.1359463 Hydrocarbons; ASTM International: West Conshohocken, PA, 1958.
(R2020a); The MathWorks, Inc.: Natick, MA, 2020. (77) König, A.; Neidhardt, L.; Viell, J.; Mitsos, A.; Dahmen, M.
(58) Pischinger, R.; Klell, M.; Sams, T. Thermodynamik der Integrated design of processes and products: Optimal renewable fuels.
Verbrennungskraftmaschine; Springer: Vienna, Austria, 2009; Comput. Chem. Eng. 2020, 134, 106712.
DOI: 10.1007/978-3-211-99277-7. (78) Ahmed, A.; et al. Small ester combustion chemistry:
(59) Hohenberg, G. F. Advanced Approaches for Heat Transfer Computational kinetics and experimental study of methyl acetate
and ethyl acetate. Proceedings of the Combustion Institute 2019, 37,
Calculations. SAE Trans. 1979, 2788−2806.
419−428.
(60) Ghojel, J. I. Review of the development and applications of the
(79) Morsch, P.; Döntgen, M.; Heufer, K. A. Kinetic investigations
Wiebe function: A tribute to the contribution of Ivan Wiebe to engine
on the high- and low-temperature chemistry of ethyl acetate. Combust.
research. International Journal of Engine Research 2010, 11, 297−312.
Flame 2022, 243, 111995.
(61) Yeliana, Y.; Cooney, C.; Worm, J.; Michalek, D.; Naber, J.
(80) Vom Lehn, F.; Cai, L.; Copa Cáceres, B.; Pitsch, H. Exploring
Wiebe function parameter determination for mass fraction burn
the fuel structure dependence of laminar burning velocity: A machine
calculation in an ethanol-gasoline fuelled SI engine. Journal of KONES
learning based group contribution approach. Combust. Flame 2021,
2008, 15, 567−574.
232, 111525.
(62) Livengood, J. C.; Wu, P. C. Correlation of autoignition
(81) Pu, Y.-H.; Dierickx, J.; Verhelst, S. Modelling the evaporative
phenomena in internal combustion engines and rapid compression
cooling effect from methanol injection in the intake of internal
machines. Fifth Symposium (International) on Combustion 1955, 5,
combustion engines. Fuel 2024, 372, 132131.
347−356.
(63) Douaud, A. M.; Eyzat, P. Four-Octane-Number Method for
Predicting the Anti-Knock Behavior of Fuels and Engines. SAE Trans.
1978, 294−308.
(64) Vandersickel, A.; Hartmann, M.; Vogel, K.; Wright, Y. M.; Fikri,
M.; Starke, R.; Schulz, C.; Boulouchos, K. The autoignition of

4093 https://doi.org/10.1021/acs.energyfuels.4c04775
Energy Fuels 2025, 39, 4079−4093