Professional Documents
Culture Documents
Energy
journal homepage: www.elsevier.com/locate/energy
a r t i c l e i n f o a b s t r a c t
Article history: CO2 emissions of light-duty vehicles are certified over standardised, laboratory-based conditions and
Received 6 September 2018 reported to the consumers. Such tests reflect specific operating conditions that differ from what an in-
Received in revised form dividual driver experiences. Vehicle simulation can bridge the gap and help provide customised, vehicle
25 November 2018
and trip-specific values. This study investigates the potential of using a simulation-based approach for
Accepted 11 December 2018
Available online 19 December 2018
calculating CO2 emissions over real-world operation, when limited information and test-data are
available. The methodology introduced in the European vehicle certification regulation since 2017 is used
as a basis. Seven vehicles were tested over multiple on-road trips and in some cases on a chassis dyno.
Keywords:
CO2 emissions
First, the analysis focused on the accuracy of the simulations when only limited information for the
Fuel consumption vehicle and its components are used. Subsequently, the model was calibrated on test data. The first case
Real-world operation presented an error between 1.0% and 4.4% depending on the test, while the standard deviation was 10.0%.
WLTP When using WLTP for calibration, the average error dropped to 2.9% to 0.2%, and the standard de-
Vehicle simulation viation decreased to 2.0%. When calibrating over on-road tests, the average error was 0.7% for the on-
road tests and 4.5% for the WLTP.
© 2018 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license
(http://creativecommons.org/licenses/by/4.0/).
https://doi.org/10.1016/j.energy.2018.12.063
0360-5442/© 2018 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/4.0/).
1154 S. Tsiakmakis et al. / Energy 169 (2019) 1153e1165
increasing over time [9] and has reached approximately 40% in [20]; still the information communicated with each individual
2016 [10,11]. The same problem is reported in China with a gap vehicle includes the disclaimer that “Actual results will vary for
value of 30% being calculated in 2017 [12]. Due to an effective many reasons, including driving conditions and how you drive and
correction procedure the gap reported for the USA was less than maintain your vehicle” [21]. In Europe, some first initiatives for
20% in 2012 [13] and possibly much smaller today. A series of fac- consumer information have been presented by vehicle manufac-
tors responsible for this gap [8,14] have been identified. The factors turers and other companies or non-governmental organisations
include the deficiencies of the NEDC test procedure to represent who are providing estimates on the fuel consumption when the
realistic on-road driving conditions, the inherent variability of the vehicles are used under realistic operating conditions [24e26].
real-world operating conditions, the effect of the driving style, the However, the information provided is still very fragmented and
unrealistic cycle composition of urban, rural and motorway driving partly contradictory due to the different methods used to deter-
compared to real life, the use of a vehicle's inertia class and not its mine these emissions. The values measured in various standardised
actual mass, the exclusion of all additional equipment in the or non-standardised test protocols (partly remaining the property
determination of mass, and the exploitation of test margins by the of their developers) cannot be linked directly to the official type
manufacturers. The CO2 emissions were always determined for the approval values or other standard tests foreseen by the EU regu-
lightest vehicle (excl. any optional equipment) and, also, the vehicle lation. Furthermore, the impact of exogenous factors - environ-
was tested for an inertia class, not its actual mass. The situation is mental conditions, traffic, usage-specific characteristics - on the
expected to improve with the introduction of WLTP [15e17] but not fuel consumption and CO2 emissions is not taken entirely into ac-
completely be alleviated as no laboratory test can capture all factors count. These factors can vary stochastically and are thus difficult to
and the inherent variability of fuel consumption in real-world reproduce under laboratory conditions. Most of these approaches
operation. Some first studies estimate a gap of 15e20% between are based on statistical corrections. Ntziachristos et al. [27] pro-
real-world and WLTP CO2 emissions [6,18], but it is still too early for posed a simple regression model with three parameters: type
deriving solid conclusions as the WLTP is still in its introduction approval fuel consumption, reference mass, and engine capacity,
phase. which were found to account for more than 85% of the variance in
To achieve higher CO2 emissions reductions in real-world real-world fuel consumption. As the model uses publicly available
operation and pave the way for new innovative technologies that vehicle data, it can be used to estimate real-world fuel consumption
will improve the sustainability of the road transport sector, it is on and CO2 emission values for large data sets and fleet analysis. Lig-
the one hand essential to communicate to the consumers realistic terink et al. [28] proposed four regression models, using a combi-
CO2 emissions values [22], and on the other hand to develop the nation of type-approval fuel consumption values, vehicle mass,
necessary instruments to increase the effectiveness of policy and frontal area, engine power, maximum vehicle speed, and vehicle
monitoring tools [23]. In the US the Environmental Protection build year. When externally validated these models achieved co-
Agency (EPA), introduced a weighing procedure based on the re- efficients of determination which ranged from 0.42 to 0.56. Duarte
sults of five tests cycles, that are employed for deriving the fuel et al. [29] used a regression model based on vehicle specific power
economy value that is communicated to the consumers (window- to predict fuel consumption. The model achieved coefficients of
sticker gas mileage) [19]. Remaining gaps between laboratory determination higher than 0.9 and was used to estimate the on-
measurements and real-world emissions (e.g. due to road condi- road fuel consumption of best-selling vehicles in Portugal. Greene
tions) are addressed by applying a mathematical adjustment factor et al. [30] applied a simple regression model to US on-road fuel
S. Tsiakmakis et al. / Energy 169 (2019) 1153e1165 1155
economy data and found that 56e68% of the variance in on-road a. Group A: The first group of cars consists of three Diesel Euro 6
fuel economy figures could be explained by test-cycle fuel con- vehicles which were tested on the road in several repetitions
sumption values alone. The previous models provide some robust over four different routes in Germany, using PEMS (vehicles 1e3,
approaches for estimating on-road fuel consumption, but they also routes 1e4);
have a series of limitations. Even though they offer good accuracy b. Group B: The second group consists of two Gasoline and one
when averaging over a large number of vehicles their uncertainty Diesel Euro 6 vehicles which were tested both in the lab (WLTP)
increases for individual vehicles. In addition, they require pre- and over two real-world routes in Italy, using PEMS (vehicle
existing calibration datasets which may vary from year to year. 4e6, routes 5e6);
Finally, they have limited capacity to capture additional factors such c. Group C: The third group consists of one Diesel Euro 6 vehicle
as the exact vehicle characteristics, driver behaviour or extreme which was tested both in the lab, over the WLTP, and over one
operating conditions. real-world route in Italy, using PEMS and onboard data logger
In the present study, an alternative approach to simulate (vehicle 7, route 7), and over four routes in Italy and Germany,
vehicle-specific real-world vehicle CO2 emissions is investigated. using only the onboard data logger (vehicle 7, routes 8e11).
The simulation model accepts both detailed vehicle characteristics
and operating conditions into account. The simulation tool used for More detailed information regarding the measurement
this study is PyCSIS, which is presented in detail in Ref. [31]. PyCSIS campaign of the first group of vehicles (1e3) can be found in
employs the simulation methodology introduced in the CO2MPAS Ref. [22], while the measurements performed with the rest, both
tool [4,32] and can use as input vehicle specific parameters and, on-road and in the laboratory, are summarized in Ref. [33].
when available, data from chassis-dyno WLTP tests or real-world
measurements. In the latter case, the model can calibrate inter- 2.2. Simulation tool
nally generic component operation and efficiency functions and
sub-models to reflect the performance of the specific vehicle better. The tool used to perform the simulations in this study is PyCSIS
The suitability and accuracy of PyCSIS for simulating fleet-average [31]. The suitability and accuracy of the tool for simulating fleet-
CO2 emissions have been thoroughly tested and validated [31]. average CO2 emissions have been tested thoroughly and vali-
The present work presents the first attempt to use laboratory data dated, as discussed in Ref. [31]. PyCSIS uses the same underlying
to calibrate the tool and use it for the estimation of real-world simulation methodology as the CO2MPAS [32], which has been
emissions, and/or vice versa, use real-world data to calibrate the extensively validated using data of vehicles representing the
tool and use it for the estimation of laboratory CO2 emissions. complete EU fleet. Similarly to CO2MPAS, PyCSIS consists of three
Furthermore, as of today, to the best of the authors' knowledge, no main modules.
other work demonstrates the increased accuracy which could be The Inputs Module combines the various inputs together,
achieved in the estimation of vehicle specific CO2 emissions by namely: (a.) the user-defined inputs, i.e. the vehicle specifications
using only publicly available vehicle data and tests only at the (Table 4) and the mission profile to be simulated (velocity profile),
vehicle level. (b.) the default inputs, i.e. default parameters of the tool which are
The paper is structured as follows. The methodology section not necessarily demanded from the user, and (c.) the test data
presents the simulation methodology and the different types of which are used to calibrate the calculated inputs.
tests performed on real vehicles. Data from tests conducted on The Energy Demand Module calculates the energy demand from
seven vehicles are divided into three groups and are used to the engine to follow a predefined velocity profile using longitudinal
quantify the accuracy and uncertainty of reproducing the CO2 vehicle dynamics. The equation of the instantaneous drivetrain
emissions on the road and in the laboratory when using different energy balance is:
input data and model calibration combinations. Initially, generic
.
vehicle inputs are employed to calculate emissions over real-world
Pdtr ¼ F0 *cosðfÞ þ F1 *v þ F2 *v2 þ m*a þ m*g*sin f *v h
trips. Subsequently, measurement data is used to calibrate vehicle dtr
specific sub-models and parameters. The calibrated vehicle models (1)
are used to simulate the CO2 emissions for different real-world and
chassis dyno speed profiles. The paper concludes with a discussion where the numerator represents the power at the wheels required
on the possible applications of the proposed methodology and on to move the vehicle along the defined velocity profile, and the
specific extensions of the tool to better reflect the effect of external denominator represents the efficiency of the complete drivetrain,
factors on real-world driving CO2 emissions. i.e. from the wheels to the clutch. The variables m, g, and 4 are the
vehicle mass, the acceleration of gravity and the road gradient,
2. Methodology respectively. The drivetrain efficiency, hdtr, is considered as a
function of torque, engine rotational speed and constant empirical
2.1. Experimental tests and vehicle data factors, which are derived from detailed generic gearbox models.
The factors F0, F1, F2, represent the road load coefficients of the
For the needs of this study, seven passenger cars were tested vehicle and are calculated by PyCSIS for the respective test pro-
over various (real-world and chassis dyno) conditions (vehicle in- cedures, i.e. NEDC, WLTP, from publicly available information.
formation presented in Table 1). Real-world tests were performed The Fuel Consumption Module calculates the instantaneous
on public roads using Portable Emissions Measurement System engine speed, power, coolant temperature and fuel consumption,
(PEMS) and onboard data loggers while chassis dyno tests were starting from the engine speed and power demand from the
performed in the Vehicle Emissions Laboratory (VELA) of the Joint drivetrain to follow the vehicle speed profile. The engine power is
Research Center (JRC), in Ispra, Italy. Table 2 summarises some defined as the sum of the drivetrain power demand, as calculated in
essential information regarding the real-world routes: number of the Energy Demand Module, the power demand from the vehicle
repetitions, average velocity, and distance driven. The cars and electrical system, e.g. the alternator power demand, and the power
respective measurement data are divided into three groups demand of auxiliary systems connected to the engine, e.g. pumps.
depending on the test locations and the type of data measured in Some state-of-the-art energy-saving technologies, such as start-
each case (Table 3). stop and advanced alternator control for partial brake energy
1156 S. Tsiakmakis et al. / Energy 169 (2019) 1153e1165
Table 1
Vehicles specifications used in the sample.
Body Type 4-door Sedan 5-door Station Wagon 5-door Station Wagon 5-door 5-door SUV 5-door 5-door Hatchback
Hatchback Hatchback
Fuel Type Diesel DI Diesel DI Diesel DI Gasoline Diesel DI Gasoline Diesel DI
No of Cylinders 4 4 4 3 4 4 4
Type of Aspiration Turbocharged Turbocharged Two-Stage Turbocharged Turbocharged Naturally Turbocharged
Turbocharged Aspirated
Capacity 1968 cc 1995 cc 2191 cc 999 cc 1956 1242 cc 1968 cc
Power 103 kW @ 4200 120 kW @ 4000 RPM 110 kW @ 4500 RPM 70 kW @ 103 kW @ 3750 RPM 51 kW @ 110 kW @ 3500 RPM
RPM 5000e5500 5500 RPM
RPM
Max Torque 320 Nm @ 1750 380 Nm @ 1750e2750 380 Nm @ 2000 RPM 160 Nm @ 350 Nm @ 1750 102 Nm @ 340 Nm @ 1750e3000 RPM
e2500 RPM RPM 1500e3500 RPM 3000 RPM
RPM
Gearbox Type Dual Clutch Torque converter Manual Transmission Manual Torque Converter Manual Dual Clutch Automated
Automated automatic Transmission Transmission Automatic Transmission Transmission
Transmission Transmission
No of Gears 6 8 6 5 9 5 6
Driven Axles Front Rear Front Front AWD Front Front
Emission Particulate Filter NOx Storage Catalyst & Low compression ratio & 3-way NOx Storage 3-way NOx Storage Catalyst &
Reduction & SCR & High- Particulate Filter & Particulate Filter & High- Catalyst Catalyst & Catalyst Particulate Filter & Low-
Systems Pressure EGR High-Pressure EGR Pressure EGR Particulate Filter & and High-Pressure EGR
Dual EGR
Mass in Running 1599 kg 1575 kg 1578 kg 1210 kg 1645 kg 1015 kg 1413 kg
Ordera
Maximum Mass 2110 kg 2050 kg 2090 kg 1680 kg 2100 kg 1420 kg 1880 kg
PEMS Test Load 152 kg 146 kg 150 kg 140 kg 110 kg 140 kg 107 kg
Wheels 205/50 R17 205/60 R16 225/55 R17 215/40 R17 225/45 R18 175/65 R14 225/40 R18
Type Approval CO2 131 g/km 112 g/km 116 g/km 98 g/km 144 g/km 119 g/km 117 g/km
(NEDC)
Max Velocity 212 km/h 222 km/h 210 km/h 186 km/h 190 km/h 164 km/h 216 km/h
Initial Mileage 75,806 km 68,753 km 82,760 km 2539 km 10,457 km 2336 km 25,602 km
Production Year 2013 2013 2012 2016 2016 2016 2015
a
Empty weight of the vehicle with manufacturer defined standard equipment, including 75 kg for driver and luggage.
recuperation, are incorporated by a general operating strategy. The or calculated based on empirical functions and sub-modules
fuel consumption is calculated by an extended Willan's Model implemented in PyCSIS. A comprehensive list of all PyCSIS model
[34,35] as follows: parameters is provided in the Appendix.
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 k
a þ b*CmðtÞ þ c*CmðtÞ2 þ a þ b*CmðtÞ þ c*CmðtÞ2 4*a2* TðtÞ=Ttrg * l þ l2*CmðtÞ2 BMEPðtÞ
FMEPðtÞ ¼ (2)
2*a2
Table 2
Routes and trips specifications.
Route Min-Max Avg. Trip Avg. Trip Avg. Coolant Avg. Dynamicity Ambient Min/Max/Avg
Distance Altitude [m] Duration [min] Velocity [km/h] Temperature at Start [oC] V*A [m2/s3] Temperature [oC]
[km]
Table 3
Outline of the three different test groups.
A 1 to 3 1 to 4 On-road (PEMS)
B 4 to 6 5 to 6 On-road (PEMS) and chassis dyno (WLTP)
C 7 7 to 11 On-road (PEMS, route 7), on-board fuel consumption, (routes 8 to 11), and chassis dyno (WLTP)
Table 4
User-defined inputs of the Vehicle Simulation Tool (Vehicle Specifications).
different way of calibrating the simulation tool: Lab Data Calibration (LDC): This approach was applied to the
second and third group of cars, for which laboratory WLTP
Generic Data Calibration (GDC): The GDC approach was applied measurements were available. As in the GDC approach, the
to the first and the second group of vehicles and measurements. model for each vehicle was pre-calibrated based on its technical
For each one of the vehicles, PyCSIS was set-up based on the parameters listed in Table 1. Whereas in GDC approach generic
vehicle specific parameters listed in Table 1, completed by and calculated values were used to estimate unavailable or
generic or calculated parameters where the vehicle specific ones empirical calibration parameters, in the LDC approach, the
were not available (e.g. gearbox ratios, road loads). Those measurement data collected during the vehicle testing was used
generic parameters were extracted from the vehicle database to calibrate the pre-calibrated submodels. During the calibration
described in Ref. [34] or estimated as described in the same process the Willans coefficients, the drivetrain losses as well as
document; the thermal system and electric system parameters were
1158 S. Tsiakmakis et al. / Energy 169 (2019) 1153e1165
2.4. Simulations
The models were calibrated using specific test data and were
subsequently used to simulate the remaining tests in each data
group. For defining a test to be simulated, the velocity profile as
recorded in the GPS was directly used, while the altitude profiles
were derived for each trip from the GPS position data and the
digital elevation model database of gpsvisualizer.com [37]. The
latter is based on the NASA's SRTM3 and the US Geological Survey's
NED elevation datasets. The altitude profiles are used to estimate
the road grade/slope and to calculate the instantaneous air density;
both directly affecting the power demand.
When the road load factors of the simulation tests were
different from the calibration test (as in the case of the LDC
approach for example), the rolling and aerodynamic resistances
were directly calculated by the tool, considering the additional
mass of the PEMS and its effect on the aerodynamic resistance
(similar to the calibration methodology). Only for the simulation of
the WLTP test (RWDC approach), where all test-specific boundary
conditions and provisions were set, the official road loads of the
vehicle were used.
The simulated total CO2 emission of each trip was compared to
the target value, i.e. the value measured by the PEMS. For a more
detailed analysis, the accumulated simulated CO2 emissions over
time and other instantaneous signals were also compared to the
Fig. 1. Outline of PyCSIS and its sub-models. The various sub-models are grouped into respective signals based on the PEMS measurement, for selected
the Energy Demand Module and the Fuel Consumption Module. The various inputs e trips. Results are presented in detail in the following section.
user-defined inputs, i.e. vehicle specifications & mission profile, default inputs, and
calibrated inputs e are separately grouped under the Inputs Module.
Table 5
Generic data calibration approach simulation results for the first group of cars - Statistics.
Mean Error/Std Mean Abs Error/Std Mean Error/Std Mean Abs Error/Std Mean Error/Std Mean Abs Error/Std Mean Error/Std Mean Abs Error/Std
Fig. 4. Cumulative CO2 emission simulation errors of the generic data calibrated
models for the three vehicles of the first group of cars, over the four routes. As ex-
pected the simulation accuracy is more influenced by the vehicle than by the route.
Fig. 6. Measured vs simulated CO2 emissions & cumulative error of the lab data
calibration approach for the three vehicles and six tests of the second group. The time
series data show a good correlation, and the overall average error is 0.15%, with a
standard deviation of 1.89%.
Fig. 8. Effect of the calibration method on the simulation model prediction quality: the Fig. 9. Real-world simulation results of the lab data calibration approach for vehicle 7
average simulation error drops from 4.4% to 0.2%, while the standard deviation re- on route 7. Simulated results follow the measured signals closely, resulting in an
duces from 9.8% to 1.9%. overall error on the cumulative CO2 emissions of 2.6 g/km, equivalent to 1.9%.
1162 S. Tsiakmakis et al. / Energy 169 (2019) 1153e1165
Fig. 10. Engine Speed vs Engine Load for the PEMS and the WLTP tests. The PEMS test
covers a more extensive area of the engine operation range than the WLTP.
4.1. Conclusions
detail of the data used for the model calibration, the achieved ac- benchmark different vehicle configurations and models to identify
curacy and overall performance of the proposed methodology is the most efficient one for the specific use case.
promising. The results show that the investigated calibration ap- Seen from a regulator's perspective, the data collected during
proaches, with certain improvements, are suitable for simulating on-road driving can be used to calibrate the respective vehicle
real-world CO2 emissions. model, as demonstrated in this paper. The later could then be used
The present work also revealed that not all parameters that have to simulate the vehicle type approval CO2 emissions; in this use
an influence on the CO2 emissions during real-world driving could case, significant divergences between the simulated CO2 value and
be fully captured. In order to further improve the methodology's the official one could trigger further investigation.
predictive capabilities, more comprehensive measurement data Finally, seen from a more scientific perspective, data collected
would be needed for the model calibration, mainly covering the from a specific number of on-road driving and laboratory tests can
usage of auxiliaries and the varying environmental conditions. Also, be used to calibrate specific vehicle models, which could then be
the calibration model needs to be extended by appropriate func- used to produce realistic data, representative of extended, or less
tions that can reproduce these dependencies as demonstrated also ordinary, operating and environmental conditions. Those could be
by [38]. used as input to existing CO2 emission inventories and projection
It is expected that the introduction of onboard fuel consumption tools or for updating and extending existing emission factors
monitoring systems in European light-duty vehicles will help without the need for costly additional testing. Building on the same
further improve the simulation method described in this paper and concept, the methodology presented could be used to normalise
could increase its application potential. The measurement of fuel CO2 measurements of different vehicles over different conditions in
consumption under actual operating conditions will become more order to be used as a reference for extending consumer information
straightforward and approachable to the average non-expert driver, material, similar to the window-sticker applied in the USA or for
and the improvements in the quality and accuracy of the measured providing qualified estimates of the CO2 emissions gap.
signal are also likely to improve the results of the method.
Acknowledgements
4.2. Outlook and potential applications
The authors would like to acknowledge the help of all the VELA
The developed methodology could become the basis for a series staff who contributed to the test measurements. Authors would like
of applications. to thank Dr. Jelica Pavlovic for her valuable feedback and support
On a driver and consumer level, the first possibility is the regarding the WLTP certification procedure, and Dr. Vincenzo
development of customizable tools that could calculate vehicle and Arcidiacono and Mr Konstantinos Anagnostopoulos for their help
user-specific CO2 emissions, for specifically defined use cases. on the CO2MPAS model and the data structuring. In addition, the
Those could be based on either publicly available vehicle infor- authors would like to thank Mr Cosmin Codrea for providing
mation or simple, low-cost test data, e.g. collected by the driver constructive feedback during the writing of the paper. Part of the
with a basic OBD data logger, or on a combination of the two. work presented in this paper has been performed within the
Building on the same concept another potential use-case could be framework of Dr. Tsiakmakis' doctoral research that was funded by
the application of the methodology to analyze the real-life fuel the European Commission.
consumption and driving performance of a specific driver over
specific driving conditions. This information could, in turn, be used Appendix
to provide feedback regarding the environmental and economic
effects of driving style and habits, the use of auxiliary equipment, or
Table 6
Default Inputs of the Vehicle Simulation Tool.
Table 7
Calculated / Derived Inputs of the Vehicle Simulation Tool.
duty vehicle CO2 certification process. Appl Energy 2018;226:784e96. regard to type approval requirements for masses and dimensions of motor
https://doi.org/10.1016/j.apenergy.2018.06.009. vehicles and their trailers and amending Directive 2007/46/EC of the Euro-
[33] Clairotte M, Valverde V, Bonnel P, Giechaskiel B, Carriero M, Otura M, et al. pean Parliament and of the Council. 2012 [n.d].
Joint Research Centre 2017 light-duty vehicles emissions testing. 2018. Ispra. [37] GPS Visualizer: Assign elevation data to coordinates n.d. http://www.
[34] Application of Willans Line Method for Internal Combustion Engines Scal- gpsvisualizer.com/elevation (accessed May 20, 2018).
ability towards the Design and Optimization of Eco-Innovation Solutions n.d. [38] Review of in use factors affecting the fuel consumption and CO2 emissions of
http://papers.sae.org/2015-24-2397/(accessed November 28, 2016). passenger cars - EU Science Hub - European Commission. EU Sci Hub; 2016.
[35] Introduction to modeling and control of internal combustion | lino guzzella | https://ec.europa.eu/jrc/en/publication/eur-scientific-and-technical-research-
Springer. [n.d]. reports/review-use-factors-affecting-fuel-consumption-and-co2-emissions-
[36] European Commission. Regulation (EC) No 1230/2012 implementing Regu- passenger-cars. [Accessed 28 November 2016].
lation (EC) No 661/2009 of the European Parliament and of the Council with