Energy Conversion and Management 92 (2015) 251265

Contents lists available at ScienceDirect

Energy Conversion and Management

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

Establishing bonds between vehicle certication data and real-world

vehicle fuel consumption A Vehicle Specic Power approach
G.O. Duarte , G.A. Gonalves, P.C. Baptista, T.L. Farias
LAETA, IDMEC, Instituto Superior Tcnico, Universidade de Lisboa, Av. Rovisco Pais, 1, 1049-001 Lisboa, Portugal

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

Article history: A method to perform the energy characterization of a vehicle according to the specic power required
Received 17 May 2014 while driving was developed using public vehicle certication data. Using a portable emission measure-
Accepted 14 December 2014 ment system, fuel consumption was quantied in a second-by-second basis under on-road conditions for
19 vehicles (spark-ignition, compression-ignition and hybrids). This data allowed building generic curves
of fuel consumption as a function of the specic power, according to Vehicle Specic Power methodology.
Keywords: Comparing on-road measurements and the model estimates, a R2 higher than 0.9 for conventional and
Certication data
hybrid vehicles was obtained regarding modal fuel consumption. Comparing the fuel consumption
Vehicle Specic Power
Portable emission measurement system
measured on the drive cycles performed by each vehicle and the correspondent estimates, an absolute
Modeling of fuel use deviation of 9.2% 9.2% was found for conventional vehicles and 4.7% 1.8% for hybrids vehicles. This
methodology was validated and applied to estimate the energy impacts of the best-selling vehicles in
Portugal for different driving cycles. This prompt method, that does not require vehicle monitoring,
can estimate curves of fuel consumption in g/s, as a function of specic power, which allows quantifying
the absolute fuel use for any driving cycle.
2014 Elsevier Ltd. All rights reserved.

1. Introduction laboratory conditions, CO2 values exceeded 15 10% compared to

In Europe, vehicle certication data regarding fuel consump-
tion, CO2 and regulated pollutant emissions is public, easily avail-
able and is obtained from vehicle testing on a standard driving
cycle (the New European Driving Cycle NEDC). However, NEDC
is recognized as a low load cycle and hardly representative of real
world driving patterns, where vehicles may be optimized to the
test driving conditions [1]. The gaps between laboratorial tests
and on-road measurements have been addressed in a study carried
by the European Commission Joint Research Center using a porta-
ble emission monitoring system (PEMS) on 12 vehicles ranging
from EURO 3 to EURO 5 [2], concluding that CO2 emissions are over
21 9% higher on-road tests on typical rural and highway routes
when comparing with NEDC. Moreover, performing NEDC under
Abbreviations: CO, carbon monoxide; CO2, carbon dioxide; HC, hydrocarbons;
ICE, internal combustion engine; NEDC, New European Driving Cycle; NO, nitrogen
oxide; O2, oxygen; OBD, on-board diagnostic port; PEMS, portable emission
monitoring system; SAE, Society of Automotive Engineers; SFC, specic fuel
consumption; SOC, state-of-charge; TTW, Tank-to-Wheel; VSP, Vehicle Specic
Power; WTT, Well-to-Tank; WTW, Well-to-Wheel.
Corresponding author. Tel.: +351 21841955; fax: +351 218417640.
E-mail address: goncalo.duarte@ist.utl.pt (G.O. Duarte).
certication values.
The European Commission Joint Research Center has also com-
pared type-approval certication data and real-world fuel con-
sumption from several sources (including studies from ADAC
organization in Germany, Artemis project, automotive journals
and vehicle owners data), in order to establish correlations
between these two indicators [3]. This study conrms that
certication data underestimates fuel consumption by 1015% on
gasoline vehicles and 1220% on diesel vehicles. Correlations were
developed to predict real-world fuel consumption (in l/100 km)
based on vehicle characteristics, such as engine displacement,
vehicle mass and fuel consumption on certication cycle, but these
do not consider the driving cycle.
A study carried in England by Rhys-Tyler and Bell [4] used data
from remote sensing surveys carried in 2008 as input to ll Vehicle
Specic Power (VSP) modal bins with real vehicle operation
emission rates and use them to reconstruct the New European
Driving Cycle and provide pollutant emissions estimates. Vehicle
Specic Power methodology consists of a road load modal analysis
based on vehicle dynamics and road topography, which groups
experimental data points with similar power conditions, allowing
the comparison of fuel and emission rates, independently of the
driving cycle [5].

http://dx.doi.org/10.1016/j.enconman.2014.12.042
0196-8904/ 2014 Elsevier Ltd. All rights reserved.
252 G.O. Duarte et al. / Energy Conversion and Management 92 (2015) 251265

Table 1
VSP binning and ranges of W/kg for each mode.

VSP W (kg) VSP W (kg) VSP W (kg) VSP W (kg) VSP W (kg) VSP W (kg)
mode mode mode mode mode mode
21 VSP < 20 12 12 6 VSP < 11 3 3 6 VSP < 2 6 5 < VSP 6 6 15 14 < VSP 6 15 24 23 < VSP 6 24
20 20 6 VSP < 19 11 11 6 VSP < 10 2 2 6 VSP < 1 7 6 < VSP 6 7 16 15 < VSP 6 16 25 24 < VSP 6 25
19 19 6 VSP < 18 10 10 6 VSP < 9 1 1 6 VSP < 0 8 7 < VSP 6 8 17 16 < VSP 6 17 26 25 < VSP 6 26
18 18 6 VSP < 17 9 9 6 VSP < 8 0 VSP = 0 9 8 < VSP 6 9 18 17 < VSP 6 18 27 26 < VSP 6 27
17 17 6 VSP < 16 8 8 6 VSP < 7 1 0 < VSP 6 1 10 9 < VSP 6 10 19 18 < VSP 6 19 28 27 < VSP 6 28
16 16 6 VSP < 15 7 7 6 VSP < 6 2 1 < VSP 6 2 11 10 < VSP 6 11 20 19 < VSP 6 20 29 28 < VSP 6 29
15 15 6 VSP < 14 6 6 6 VSP < 5 3 2 < VSP 6 3 12 11 < VSP 6 12 21 20 < VSP 6 21 30 29 < VSP 6 30
14 14 6 VSP < 13 5 5 6 VSP < 4 4 3 < VSP 6 4 13 12 < VSP 6 13 22 21 < VSP 6 22 31 VSP > 30
13 13 6 VSP < 12 4 4 6 VSP < 3 5 4 < VSP 6 5 14 13 < VSP 6 14 23 22 < VSP 6 23

The impacts of a vehicle trip can only be quantied in a rigorous
way by doing on-road measurements using a portable emission
monitoring system (PEMS) to collect vehicle dynamics, engine
data, road topography and tailpipe gas concentration of pollutants
during operation. However, it is not feasible to measure every vehi-
cle technology performing selected driving cycles and, as a result,
numerical tools are commonly used to simulate vehicle operation.
Numerical tools can perform micro-simulation of vehicle use, such
as Advisor [6], CMEM [7], EcoGEST [8] or based on activity in a
macro perspective, such as Copert [9,10]. Some of the numerical
tools referred are too complex, in order to account for the exact
vehicle characteristics and driving prole, but some are too simple
and too generic to provide an accurate analysis.
Considering that all new vehicles are certied over a standard
driving cycle, homologation data should intrinsically reect both
vehicle technology, as well as properties, such as mass, aerody-
namics and powertrain characteristics. As stated, it is not feasible
to characterize all vehicles under real-world usage. Therefore, this
research work main purpose was to develop a methodology that,
using individual public certication data, could provide fuel
consumption and pollutant emissions proles, independently of
vehicle technology (spark-ignition, compression-ignition, hybrids).
Using this information as input, it is possible to evaluate driving
cycles without the need to perform eld measurements and estab-
lish links between real-world usage of vehicles and certication
data, but also between laboratorial and simulation elds of scien-
tic application.
software to receive, integrate, synchronize and record the data
along the trips.
The information provided by the vehicle sensors is collected via
the OBD port reader including, vehicle speed, engine speed and
load, manifold absolute pressure, airow mass, intake air temper-
ature, coolant temperature and throttle position.
A GPS receiver with an integrated barometric altimeter is also
used to collect vehicle position and altitude along the trip for sub-
sequent calculation of road grade. Tailpipe emissions were mea-
sured continuously with a portable ve gas analyzer, which
provides simultaneous information on carbon dioxide (CO2), car-
bon monoxide (CO), hydrocarbons (HC), nitrous monoxide (NO)
and oxygen (O2). The data collected allows estimating ICE fuel con-
sumption and mass of pollutants in the exhaust gas in a second by
second basis [11].
2.2. Data analysis Vehicle Specic Power methodology
The Vehicle Specic Power (VSP) methodology is based on a
simplication of the forces applied to a vehicle. Hence, data col-
lected during on-road measurements is analyzed using VSP meth-
odology to perform an energy and environmental characterization
of the vehicle monitored [5,12,13]. This analysis provides an esti-
mate of the power per mass unit due to a combination of vehicle
dynamics (speed, acceleration, rolling and aerodynamic resistance)
and road grade. To each point of the trip, the correspondent spe-
cic power is assigned according to Eq. (1) [5].

2. Materials and methods VSP v  1:1  a 9:81  grade 0:132 3:02  104  v 3 1

Fuel consumption was assessed for 19 EURO 5 vehicles sold in
the European market, covering the most representative segments
in vehicle sales, comprising the following vehicle technologies:
conventional gasoline spark-ignition; conventional diesel com-
pression-ignition; gasoline parallel hybrid; and gasoline parallel/
series hybrid.
Fuel consumption was measured through on-road measure-
ments, performed in the Lisbon Metropolitan Area, Portugal, using
a portable emission measurement system (PEMS) collecting data at
1 Hz. Two types of routes were chosen: a primary itinerary which
includes urban, extra-urban and highway roads; and a secondary
itinerary almost exclusively in urban driving context.
2.1. PEMS description
A PEMS system was installed in all vehicles to collect data in a
second-by-second basis, namely engine parameters, exhaust gas
composition, road topography and vehicle dynamics. This portable
laboratory was already used in previous studies [11]. On-road data
is collected from the numerous equipment installed on-board the
vehicle and acquired by a laptop, using developed Labview
where v, vehicle speed (m/s); a, vehicle acceleration (m/s2); grade,
road slope.
A modal analysis is used to group points of similar power per
mass demand (W/kg). For light-duty vehicles, VSP is commonly
divided in 14 modes [14,15]. However, in the scope of this work,
to obtain a higher level of resolution of the specic power demand,
VSP was divided in modes consisting of 1 W/kg, according to
Table 1.
2.3. Vehicles description
Using the portable laboratory, 14 EURO 5 vehicles with conven-
tional gasoline and diesel technologies were measured under on-
road conditions, as well as 5 EURO 5 hybrid vehicles (3 with paral-
lel/series conguration and 2 with parallel conguration). Table 2
presents the summary of the characteristics of the conventional
vehicles measured and the total data collected (in seconds). The
average measurement time was of around 6900 s (almost 2 h of
on-road data). Table 3 summarizes the characteristics of the hybrid
vehicles studied, with an average measurement time of around
9700 s.
 G.O. Duarte et al. / Energy Conversion and Management 92 (2015) 251265 253

Table 2
Summary of the characteristics of the conventional gasoline and diesel vehicles measured.

Vehicle classa Supermini Small family car Large family car Small multi-
purpose vehicles
Vehicle A B D G J L N E K F H I C M
Technology SI CI SI-DI CI CI CI SI CI CI CI CI CI CI CI
ICE displacement (cc) 1598 1995 1198 1461 1120 1248 998 1686 1995 1560 1598 1685 1248 1796
ICE compression ratio 10.5:1 16.5:1 13.0:1 15.2:1 16.0:1 16.8:1 10.5:1 18.0:1 16.5:1 N/A 16.5:1 17.0:1 16.8:1 16.2:1
ICE Power (KW/rpm) 135/ 105/ 72/ 80.5/ 55.2/ 55.2/ 50.7/ 96.9/ 85/ 82/ 77.2/ 84.3/ 70/ 79.8/
5500 4000 5600 4000 4000 4000 6000 4000 4000 3600 4400 4000 4000 3200
ICE Torque (Nm/rpm) 240/ 305/ 142/ 240/ 170/ 190/ 93/ 300/ 260/ 270/ 250/ 260/ 180/ 250/
1600 1750 4400 1750 1500 1750 3600 2000 1750 1750 1500 1250 1750 1400
Vehicle mass (kg) 1205 1225 980 1285 1191 1140 950 1503 1385 1430 1499 1713 1360 1475
Power-to-Weight ratio 112 86 73 63 46 48 53 64 61 57 52 49 51 54
(W/kg)
Measurements total 10,696 10,705 4733 5067 5418 7491 5945 4158 7213 4508 7588 8245 7260 7359
time (s)
a
According to European New Car Assessment Programme (Euro NCAP).

Table 3
Summary of the main characteristics of the hybrid vehicles studied.

Vehicle classa Supermini Small family car Large family car

Vehicle O P Rb Sb Qb
Technology HEV-SI HEV-SI HEV-SI HEV-SI HEV-SI
Conguration Parallel Parallel Full Full Full
ICE displacement (cc) 1496 1281 1798 1798 1798
ICE compression ratio 10.4:1 10.8:1 13.0:1 13.0:1 13.0:1
ICE Power (kW/rpm) 84/6000 64.9/5800 73/5200 73/5200 73/5200
ICE Torque (Nm/rpm) 145/4800 121/4500 142/4000 142/4000 142/4000
Vehicle mass (kg) 1130 1225 1495 1410 1725
Electric Power (kW/rpm) 10/1500 10.3/ 60/ 60/ 60/
Electric Torque (Nm/rpm) 78/1000* 48.5/1000* 207/ 207/ 207/
Combined power (kW) 91 66 100 100 100
Total power-to-weight ratio (W/kg) 81 54 67 71 58
Measurements total time (s) 8069 10,937 9463 4934 15,051
a
According to European New Car Assessment Programme (Euro NCAP).
b
Vehicles R, S and Q share a similar powertrain.
*
Constant up to 1000 rpm.

 
Hybrid vehicles usually have different strategies regarding vehi- Fueli gfuel =s
SFC ; i being the VSP bin 2
cle use: eco mode, normal, fully electric (EV), etc. Under on-road jVSPi j W=kgvehicle
conditions, the normal mode was always used since the objective
was not to study the difference between modes, but to capture Fig. 1 presents the specic fuel consumption for 14 vehicles as func-
the most natural powertrain management by the vehicle. The tion of VSP. Fig. 1 shows that under low loads (between 1 and 10 W/
possibilities of each system to operate/manage the powertrain, kg) the magnitude of specic fuel consumption is greater than under
combining ICE with electric motor, ICE only, electric motor only, higher power conditions (over 10 W/kg), which is according to typ-
etc, introduce several difculties to this analysis. For the scope of ical internal combustion engine specic fuel consumption maps.
this research work, hybrid vehicles were considered as black The results presented in Fig. 1 show that there is coherence
boxes in powertrain management, using the same experimental between engine physics and the on-road data collected. As a result,
apparatus as for conventional vehicles. This approach is supported the following approaches were made:
by the fact that liquid fuel is the only external source supply to the
vehicle and, under long-term operation, fuel consumption rates (a) VSP 010 W/kg Integrating SFC t from Fig. 1:
would tend to an average value, consequence of the vehicle elec- Z Z
tric/ICE powertrain management. Fuel SFC dVSP  VSP0:5 dVSP  VSP0:5 C 3

3. Methodology development (b) VSP 10 to 0 W/kg Integrating SFC t from Fig. 1:
Z Z
Fuel SFC dVSP  eVSP dVSP  eVSP C 4
3.1. Conventional vehicles
(c) VSP >10 W/kg Integrating SFC t from Fig. 1:
In order to dene a generic trend of fuel consumption as func- Z Z
tion of VSP, a specic fuel consumption approach was adopted. Fuel SFC dVSP  const dVSP  const VSP C 5
Assuming VSP as a measure of required power, each value of fuel
consumption was divided by the correspondent VSP bin value, to (d) VSP <10 W/kg Integrating SFC t from Fig. 1, observing
obtain an indication of specic fuel consumption (SFC), according measurement data and knowing that fuel cut-off occurs under
to Eq. (2). these conditions, it was assumed that:
254 G.O. Duarte et al. / Energy Conversion and Management 92 (2015) 251265

0.5

0.4

SFC (g/s)(W/kg)
0.3

SFC = 0.253xVSP-0.476
0.2 R = 0.8556
SFC = 0.1126xe0.3584xVSP
R = 0.7029 SFC= -0.0002xVSP + 0.0900
R = 0.0112

0.1
SFC = 0.0004xVSP + 0.0085
R = 0.1457
0

VSP Mode (W/kg)

Fig. 1. Specic fuel Consumption for 13 vehicles: N VSP mode lower than 10 W/kg; h VSP mode between 10 and 0 W/kg; low power demand with VSP mode
between 1 and 10 W/kg; higher power requirements with VSP mode between 1 and 10 W/kg.

8
> X
21 X1
Fuel const 6 >
> aVSP2 bVSPct VSP
> FuelECE-15
> f t VSP
>
>
Having in mind the previous assumptions and considering the data >
>
VSP11 VSP10
>
>
available (only from certication), a generic trend of fuel consump- >
> daVSP2 bVSPctVSP0
>
>
tion can only have a limited number of variables. Therefore, the four
>
> X10 X 31
>
> 2
>
> aVSP bVSPct VSP mVSPdtVSP
zones described earlier were simplied in three main areas: >
>
>
> VSP1 VSP11
>
>
>
> X21 X1
(1) VSP <10 W/kg Fuel consumption rate was assumed to be >
>
> FuelEUDC
> f tVSP aVSP2 bVSPct VSP
a constant value, as conrmed in on-road measurements: >
>
>
> VSP11 VSP10
Fuel g=s f 7 >
>
>
> X10
>
> daVSP2 bVSPctVSP0 aVSP2 bVSPc
(2) 10 6 VSP 6 10 W/kg The integration of SFC into Fuel, >
>
>
< VSP1
shown in (a) and (b), was simplied and a second order
X31
equation was adopted: >
> t mVSPdt VSP
>
> VSP
Fuel g=s aVSP2 bVSP c 8 >
>
>
>
VSP11
>
>
(3) VSP >10 W/kg Fuel was assumed to increase linearly with
>
> aVSP2 bVSPc f j VSP10 ) 100a10b
>
>
>
> c f continuity at VSP 10W=kg
VSP, as described in (c): >
>
>
>
Fuel g=s mVSP d 9 >
> daVSP 2
bVSPc
0j VSP10 ) b 20a continuous derivative at VSP
>
> dVSP
>
>
The variables a, b, c, d, m and f were determined using a system >
> 10 W=kg
>
>
of equations with 2 certication data inputs, the ECE-15 and EUDC >
> 2
>
> aVSP bVSPc mVSPdj VSP10 ) 100a10bc 10m
certication values of fuel consumption (in liters per 100 km), >
>
>
> dcontinuity at VSP 10 W=kg
which can be found in specic databases, such as the United King- >
>
>
>
> daVSP2 bVSPc mj
>
dom government based Vehicle Certication Agency database [16] >
> dVSP VSP10 ) m 40acontinuous derivative at VSP
:
and 4 main conditions, which are continuity and continuous deriv- 10W=kg
ative in VSP = 10 and VSP = 10 W/kg. The total fuel consumption 11
and pollutant emission results from the sum of fuel and pollutant
outcomes of each point of the driving cycle, according to Eq. (10)
Certication speed proles can be found in software such as
which is an approach similar to the one described in the MOVES
EcoGEST [8] databases and, therefore, VSP time distribution and
model.
total distance can be calculated. The parameter d present in Eq.
X
31 (11) is related with fuel consumption rate at idling, which varies
Total Fuel Fueli  ti 10 from vehicle technology (SI or CI) or the existence or not of Stop/
i21
Start technology, for instance.
Fig. 2 presents the generic trend of fuel consumption used in This approach, using individual vehicle certication data as
this research work, regarding the areas of application of Eqs. (3) inputs, aims at including characteristics of each vehicle, such as
(5). The generic trend presented here, with its 6 variables, was used weight, propulsion technology, etc. However, some prior consider-
in certication driving cycles in a reverse manner: the total fuel ations must be explained:
consumption was known (and its VSP time distribution) and the
fuel consumption variables were dened according to the certica- (1) The results obtained with this methodology are dependent
tion values for each specic vehicle. on the assumptions made regarding the generic trend of fuel
Eq. (11) presents the system of equations used to determine the consumption.
variables. The rst 2 equations represent Eq. (10) applied to ECE-15 (2) The results are more reliable as more representative the cer-
and EUDC driving cycles, respectively. The remaining 4 equations tication values of fuel consumption are.
are conditions of continuity and derivative continuous at points (3) Not all VSP modes are covered in the cycles used, hence
VSP 10 and 10 W/kg. extrapolation is done using the generic trend variables.
 G.O. Duarte et al. / Energy Conversion and Management 92 (2015) 251265
Fuel ( g / s ) = aVSP 2 + bVSP + c
Fuel ( g / s ) = f
Fig. 2. Generic trend of fuel consumption as function of VSP mode, using three generic equations and 6 variables.
(4) In the possible case of negative values of fuel consumption in
negative VSP, fuel cut-off (or zero grams per second) is
assumed, since that is usually the case assuming the vehicle
is driven correctly, by keeping the car in an adequate gear.
The variable d presented for VSP 0 is related with fuel consump-
tion rate at idling. This parameter is very important as idling
represents around 32% of total time in ECE-15 and 9% in EUDC driv-
ing cycles. If a vehicle is equipped with Stop/Start, the engine can
be shut off under idling conditions (VSP equal to 0 W/kg). Stop/
Start operation is dened by ECU analysis of several vehicle
sensors. However, if the battery level is low, Stop/Start may be
electronically inactive [17]. Variable d is dened according to
Eq. (12) and assumes values between 0 and 1. Furthermore, fuel
consumption rate at idling has a range of typical values that do
not change considerably on SI and CI engines. Therefore, it is
important that, if the model is over-estimating idling fuel con-
sumption rate, this value can be corrected. Consequently, d is a
parameter that adjusts the predicted modal fuel consumption rate
at idle (with and without Stop/Start) in accordance with the mea-
surements at VSP mode 0.
Predicted FuelS=S Predicted FuelNoS=S
d or d
Measured FuelS=S Measured FuelNoS=S
Hence, d is related with:
Correcting the estimate of idling fuel consumption rate for vehi-
cles equipped with Stop/Start.
Correcting the possible overestimates of idling fuel consump-
tion rate for vehicles without Stop/Start.
Table 4 presents the values for d that correspond to a better esti-
mate of the fuel consumption rate at idling, whether the vehicle is
equipped with Stop/Start or not. Therefore, for vehicles with Stop/
Start d was set at 0.330, while for the remaining, d was set at 0.664.
These values were used to correct parameter c, which is directly
dependent of the certication inputs.
3.1.1. Sensitivity of parameters
For each of the 14 conventional gasoline and diesel vehicles the
variables a, b, c, d, f and m were found using certication values
and using d values described in Table 4. Table 5 presents a sum-
mary of the results achieved. Negative values of f are not accept-
able so, when that occurs, they are assumed to be zero according
to assumption (4).
Using only certication data and the generic fuel consumption
trend tted with the variables found for each vehicle, it should
be noted that each vehicle has its own characteristic values. If
255
Fuel ( g / s ) = mVSP + d
Table 4
Final values of d used further in this work.
Fuel consumption rate Measured Predicted, corrected with d d
Without Stop/Start (g/s) 0.120a 0.122b 0.664
With Stop/Start (g/s) 0.064c 0.064d 0.330
a
Value determined for idling conditions accounting 1 SI vehicle and 4 CI vehicles
(3 CI vehicles had S/S turned off to perform the measurements).
b
Average accounting 1 SI vehicle and 1 CI vehicles.
c
Value determined accounting 1 SI vehicle and 8 CI vehicles.
d
Average accounting 1 SI vehicle and 8 CI vehicles.
certication values are the same, the methodology would result
in similar characteristic values. That does not occur in the case of
Vehicles C and G (with the same input values) due to Stop/Start
equipped in the last, which affects the value of d.
The presented approach aims at being as generic as possible,
considering the propulsion systems studied (spark-ignition, com-
pression-ignition and hybrid vehicles) and the limitations regard-
ing the number of variables (which are a consequence of the
limited number of inputs from certication). In some vehicles, f
has negative values, but only for VSP modes below 10 (typically
hard braking situations), which present a small percentage of time
12
on certication cycles: only 7% in EUDC and none in ECE-15. Due to
the nature of the formulation, the time weight of these conditions
has a less representativeness when comparing with the remaining
areas of the generic trend. When the model ts the variables to nd
modal fuel consumption rates that are consistent with certication
fuel use, the areas that are more representative in respect with the
VSP modal time distributions of certication cycles are best tted
than the ones that have limited representation.
Fig. 3 presents a comparison between the estimated modal fuel
consumption rates for each VSP mode (obtained from the method
presented in this research work based on NEDC public data) with
on-road measurements for 14 vehicles. The proposed methodology
follows the trend of the on-road measurements, providing similar
dispersion in fuel consumption for the same VSP bin, according
to the vehicle characteristics, as occurs with on-road data.
3.1.2. Model development using on-road data
As referred in the Section 1, there is a considerable gap between
type-approval certication data and real-world fuel use. Regarding
the differences found using on-road data as input to NEDC driving
prole and certication values, a correction was implemented in
the generic trend using on-road data. In order to maintain an indi-
vidual vehicle asset to the correlations, a nondimensional approach
was chosen, by dividing fuel consumption (of each VSP bin) by the
respective vehicle average fuel consumption (in g/s) of the NEDC,
256 G.O. Duarte et al. / Energy Conversion and Management 92 (2015) 251265

Table 5
Summary of the results obtained for 14 gasoline and diesel conventional vehicles using certication inputs and d equal to 0.33 and 0.664 for vehicles with and without Stop/Start,
respectively.

Inputs Outputs
h i h i h i
ECE-15 (l/100 km) EUDC (l/100 km) a g=s
b g=s c (g/s) d (g/s) f (g/s) m g=s
2 W=kg W=kg
W=kg

Vehicle A SI 7.3 5.0 2.10E03 4.21E02 2.60E01 5.02E02 5.02E02 8.41E02

Vehicle B CI 5.1 3.9 2.09E03 4.17E02 1.94E01 1.49E02 1.49E02 8.35E02
Vehicle C CI 5.9 4.3 2.34E03 4.67E02 2.00E01 3.31E02 3.31E02 9.34E02
Vehicle D SIDI 5.2 3.7 1.62E03 3.23E02 1.81E01 1.94E02 1.94E02 6.46E02
Vehicle E CI 4.3 3.4 1.87E03 3.74E02 1.59E01 2.77E02 2.77E02 7.47E02
Vehicle F CI 5.1 4.0 2.18E03 4.37E02 1.90E01 2.85E02 2.85E02 8.74E02
Vehicle G CI 5.9 4.3 2.21E03 4.42E02 2.32E01 1.15E02 1.15E02 8.83E02
Vehicle H CI 5.2 3.8 1.96E03 3.91E02 2.04E01 8.77E03 8.77E03 7.82E02
Vehicle I CI 5.3 3.7 1.83E03 3.65E02 2.15E01 3.24E02 3.24E02 7.30E02
Vehicle J CI 3.9 3.3 1.91E03 3.81E02 1.36E01 5.45E02 5.45E02 7.62E02
Vehicle K CI 5.4 3.9 1.99E03 3.97E02 2.14E01 1.54E02 1.54E02 7.94E02
Vehicle L CI 5.1 3.6 1.79E03 3.59E02 2.05E01 2.60E02 2.60E02 7.20E02
Vehicle M CI 5.4 3.8 1.89E03 3.78E02 2.18E01 2.89E02 2.89E02 7.55E02
Vehicle N SI 5.8 4.3 2.06E03 4.11E02 1.69E01 3.64E02 3.64E02 8.22E02

Fig. 3. Comparison of fuel consumption rates for each VSP mode between estimates from NEDC and on-road measurements.

Fig. 4. Analysis of modal fuel consumption rate, nondimensional by NEDC average fuel consumption rate, with the comparison between estimate and measured values: (a)
for positive VSP modes only; (b) for negative VSP modes, using the average of the 14 vehicles for each VSP mode.
 G.O. Duarte et al. / Energy Conversion and Management 92 (2015) 251265 257

Table 6 higher fuel consumption rates under idling conditions. Full hybrids
Analysis of d to use in the model, according to hybrid conguration. present a low d value as consequence of high percentages of ICE Off
Fuel consumption rate ICE Stop/Start (g/s)a ICE Stop/Start (g/s)b d under idling conditions. For both congurations, additional mea-
Parallel hybrid 0.067 0.067 0.429 surements of other available vehicles and more data could provide
Full hybrid 0.010 0.010 0.090 an improvement of the results presented on Table 6.
a
Table 7 presents the outputs of the model that allow building
Time weighted average of measured vehicles, excluding Vehicle S.
b
Time weighted average of model estimates, corrected with d, excluding Vehicle S.
the curves of fuel consumption rate as function of VSP. The vari-
ables obtained differ from vehicle to vehicle, according to the cer-
tication inputs.
according to Eq. (13), where 10.96 correspond to total NEDC trav-
As seen in Fig. 5, the proposed methodology presents estimates
elled distance (in km) and 1180 to the total time (in seconds).
that are within the range of on-road measurements found for these
CertificationNEDC l=100 km vehicles. For lower positive VSP modes, on-road measurements
100
 10:96  qfuel  1000
FCNEDC g=s 13 present lower fuel consumption rates than the estimates, which
1180
is related to the full electric operation of parallel/series hybrids.
Fig. 4a depicts the correlation for positive VSP modes, present- Under certication, according to a study by SAE [19], specic
ing the relation between estimate values and on-road measure- rules were considered to provide consistent fuel consumption
ments, nondimensional by each vehicle fuel consumption rate in and emission results from testing hybrid congurations. For the
NEDC. For positive VSP modes, a correction factor of 1.0541 and vehicles studied, without external charging, the tests are carried
an offset of 0.1302 were found to relate estimate and on-road mea- out under charge sustaining conditions, even if instantaneously
surements. This correction was introduced in fuel consumption the battery is charge depleting or charge increasing. Under on-road
estimate based on NEDC data, to cope with off-cycle conditions, conditions, battery State of Charge (SOC) was available only for
relating individual certication and eet average behavior. Vehicle Q [11], and collected at 1 Hz via the OBD port. On the
Situations of deceleration, mostly corresponding to negative 3 days of measurements, initial and nal SOC were different for
VSP modes, are responsible for less than 10% of total fuel consump- single trips with a maximum difference between initial and nal
tion [18]. Regarding these situations, Fig. 4b presents an analysis SOC never exceeding more than 10%. Regarding Vehicles O, P, R
that compares the estimates based on NEDC with the on-road mea- and S, battery SOC was only visually available on the dashboard
surements. Since each vehicle presents signicant different values, display. From the records taken during measurements, differences
nondimensional values were used (following the approach of Eq. between initial and nal SOC were veried on Vehicles O and S. On
(13)) and an average of values was chosen to represent each VSP Vehicles P and R, battery initial and nal SOC were different on sin-
mode for the 14 vehicles measured. gle test trips, but initial and nal SOC between measurement peri-
Summarizing, the previous values obtained by the methodology ods were similar, although with some constraints considering the
based on certication data were underestimated and, conse- level of denition of the vehicles displays.
quently, correction factors were introduced obtaining the method- Fig. 6 presents the correlations between the estimate values and
ology nal equations (Eqs. (14) and (15)). the on-road measurements, nondimensional by each vehicle fuel
FCcorrected g=si FCNEDC estimate g=si consumption rate in NEDC. Regarding parallel hybrid conguration
1:0541 0:1302; see Fig. 6a , the model estimates better agree at lower VSP
FCNEDC g=s FCNEDC g=s
modes (lower fuel consumption rates) and over-estimate at higher
i correspond to VSP > 0 14 VSP modes (higher fuel consumption rates), which is in accordance
with the vehicle operation: it uses the ICE as the main power
FCcorrected g=si FCNEDC estimate g=si source, using electric motor to supply the reminiscent power dur-
0:7864 0:1117;
FCNEDC g=s FCNEDC g=s ing standing starts or acceleration, which are conditions related
i correspond to VSP < 0 15 with higher VSP modes.
Fig. 6b proposes a correction to the model based on real-world
data collected on full hybrid vehicles, since there is an over-esti-
3.2. Hybrid vehicles mate due to ICE on/off operation on lower VSP modes (or lower
fuel consumption rates). After this improvement, estimates for
The approach developed for hybrid vehicles was similar as the lower VSP modes are corrected to include the effect of ICE on/off
one for conventional CI and SI vehicles. Therefore, the analysis of operation, increasing the estimates of fuel consumption rates as
ICE Stop/Start under idling conditions was the rst step to better VSP mode increases.
determine estimates of fuel consumption rates as function of VSP Fig. 7 presents the corrections to the model estimates devel-
mode. Table 6 presents the d values introduced in the model that oped using the on-road data. Considering that in most of these
more properly t with the on-road measurements. Regarding vehicles the variable f of the model was negative and set as zero
parallel hybrids, d value is slightly higher than the one used for according to the model formulation, the vehicles averages of fuel
conventional vehicles with Stop/Start, which is in agreement with consumption rate nondimensional by NEDC data were used.

Table 7
Summary of the results obtained for 5 parallel and full hybrid vehicles using certication inputs and d equal to 0.429 and 0.090 for parallel and full hybrids, respectively.

Inputs Outputs
h i h i h i
ECE-15 (l/100 km) EUDC (l/100 km) a g=s
b g=s c (g/s) d (g/s) f (g/s) m g=s
2 W=kg W=kg
W=kg

Vehicle O 6.1 4.4 1.98E03 3.97E02 2.01E01 2.48E03 2.48E03 7.93E02

Vehicle P 4.6 4.4 2.40E03 4.81E02 1.17E01 1.23E01 1.23E01 9.62E02
Vehicle Q 3.9 3.7 1.94E03 3.88E02 1.18E01 7.56E02 7.56E02 7.76E02
Vehicle R 3.8 3.8 2.06E03 4.12E02 1.08E01 9.78E02 9.78E02 8.23E02
Vehicle S 4.1 4.0 2.11E03 4.22E02 1.26E01 8.45E02 8.45E02 8.44E02
258 G.O. Duarte et al. / Energy Conversion and Management 92 (2015) 251265

vehicles
vehicles

Fig. 5. Fuel consumption rate as function of VSP measured on road and estimated by NEDC certication data.

Fig. 6. Correction applied to the model estimates, regarding positive VSP modes, based on real-world data. (a) Parallel hybrid conguration; (b) full hybrid conguration.

According to the data presented, Eqs. (16) and (17) summarize

FCcorrected g=si FCNEDC estimate g=si
the corrections to the model regarding parallel hybrid vehicles, 1:183  0:4701;
while Eqs. (18) and (19) correspond to full hybrid congurations. FCNEDC g=s FCNEDC g=s
i correspond to VSP > 0 18
FCcorrected g=si FCNEDC estimate g=si
0:8233 0:2899;
FCNEDC g=s FCNEDC g=s !2
FCcorrected g=si FCNEDC estimate g=si
i correspond to VSP > 0 16 1:7928
FCNEDC g=s FCNEDC g=s
!
FCcorrected g=si FCNEDC g=si FCNEDC estimate g=si
0:5177
estimate
0:2108;  0:7478 0:2700;
FCNEDC g=s FCNEDC g=s FCNEDC g=s
i correspond to VSP < 0 17 i correspond to VSP < 0 19
 G.O. Duarte et al. / Energy Conversion and Management 92 (2015) 251265 259

Fig. 7. Correction applied to the model estimates, regarding negative VSP modes, based on real-world data. (a) parallel hybrid conguration; (b) full hybrid conguration.

Table 8 Table 9
Summary of coefcient of determination (R2) found for the 14 vehicles, comparing Summary of R2 for the 5 hybrid vehicles, comparing certication based estimates vs
certication based estimates vs on-road measurements and estimates provided by on-road measurements and estimates provided by certication based corrected by
certication based corrected by on-road data (model) vs on-road measurements. on-road data (model) vs on-road measurements.

Vehicle R2 Vehicle R2
Vehicle A Measured vs cert based estimates 0.974 Vehicle O Measured vs cert based estimates 0.946
Measured vs model estimates 0.989 Measured vs model estimates 0.986
Vehicle B Measured vs cert based estimates 0.976 Vehicle P Measured vs cert based estimates 0.94
Measured vs model estimates 0.961 Measured vs model estimates 0.988
Vehicle C Measured vs cert based estimates 0.788 Vehicle Q Measured vs cert based estimates 0.975
Measured vs model estimates 0.645 Measured vs model estimates 0.979
Vehicle D Measured vs cert based estimates 0.709 Vehicle R Measured vs cert based estimates 0.954
Measured vs model estimates 0.784 Measured vs model estimates 0.975
Vehicle E Measured vs cert based estimates 0.948 Vehicle S Measured vs cert based estimates 0.953
Measured vs model estimates 0.978 Measured vs model estimates 0.987
Vehicle F Measured vs cert based estimates 0.979
Average Measured vs cert based estimates 0.954
Measured vs model estimates 0.964
Measured vs model estimates 0.983
Vehicle G Measured vs cert based estimates 0.983
Measured vs model estimates 0.992
Vehicle H Measured vs cert based estimates 0.968
Measured vs model estimates 0.986
Vehicle I Measured vs cert based estimates 0.910
3.3.1. Modal fuel consumption analysis
Measured vs model estimates 0.950
Vehicle J Measured vs cert based estimates 0.920 Table 8 presents the summary of the coefcient of determina-
Measured vs model estimates 0.961 tion between modal fuel consumption estimates and on-road mea-
Vehicle K Measured vs cert based estimates 0.971 surements for the 14 vehicles studied. Most of the vehicles studied
Measured vs model estimates 0.985 have R2 near or higher than 0.9, resulting in similar average coef-
Vehicle L Measured vs cert based estimates 0.983
Measured vs model estimates 0.988
cient of determination of 0.930 for the estimates based only on cer-
Vehicle M Measured vs cert based estimates 0.986 tication data and 0.931 including the corrections made to
Measured vs model estimates 0.968 introduce real-world vehicle operation in the estimates of fuel con-
Vehicle N Measured vs cert based estimates 0.925 sumption rates. If vehicles C and D were not considered, coefcient
Measured vs model estimates 0.881
of determination would increase to 0.960 for certication based
Average Measured vs cert based estimates 0.930 estimates and 0.967 for the nal model.
Measured vs model estimates 0.931
Table 9 presents the summary of the coefcients of determina-
tion for hybrid vehicles using the developed model described in
3.3. Methodology validation this work. All vehicles present a good agreement with on-road
data, with R2 values higher than 0.9 both for certication based
This section presents the validation of the developed methodol- estimates and for the nal model, which include certication val-
ogy, using the certication data as input, and the improvements ues corrected by the on-road data, although this last approach pro-
established using on-road measured data. vides better results.
260 G.O. Duarte et al. / Energy Conversion and Management 92 (2015) 251265

Table 10
Comparison of integrated second-by-second measured fuel consumption with modal VSP approach using on-road data, estimates provided by certication and nal model
estimates.

Inputs On-road trips (g) Measured Model

ECE-15 (l/100 km) EUDC (l/100 km) Measured Cert. based Model Deviation (%) Deviation (%) Abs. dev. (%)
Vehicle A 7.3 5 10413.6 8926.2 9798.2 14.3 5.9 5.9
Vehicle B 5.1 3.9 9121.6 8272.2 9090.3 9.3 0.3 0.3
Vehicle C 5.9 4.3 2786.1 3025.1 3391.4 8.6 21.7 21.7
Vehicle D 5.2 3.7 3707.6 2234.7 2478.2 39.7 33.2 33.2
Vehicle E 4.3 3.4 2424.6 1938 2153.7 20.1 11.2 11.2
Vehicle F 5.1 4 2372.2 2300.6 2549.9 3.0 7.5 7.5
Vehicle G 5.9 4.3 2377.6 2163.3 2433.4 9.0 2.3 2.3
Vehicle H 5.2 3.8 4084.3 3758 4185.7 8.0 2.5 2.5
Vehicle I 5.3 3.7 4435.6 3589.2 3984.9 19.1 10.2 10.2
Vehicle J 3.9 3.3 3152.7 2378.4 2650.6 24.6 15.9 15.9
Vehicle K 5.4 3.9 4305.3 3888.2 4309.5 9.7 0.1 0.1
Vehicle L 5.1 3.6 3444.2 3369.6 3759.9 2.2 9.2 9.2
Vehicle M 5.4 3.8 3795.4 3619.8 4019 4.6 5.9 5.9
Vehicle N 5.8 4.3 3290.7 2864.7 3186.4 12.9 3.2 3.2
Average 12.0 2.2 9.2
St. dev. 11.6 13.1 9.2
Excluding C and D Average 11.4 1.6 6.2
St. dev. 7.0 7.9 4.8

Table 11
Comparison of integrated second-by-second measured fuel consumption with modal VSP approach using on-road data and estimates provided by certication and nal model
estimates.

Inputs On-road trips (g) Measured Model

ECE-15 (l/100 km) EUDC (l/100 km) Measured Cert. based Model Deviation (%) Deviation (%) Abs. dev. (%)
Vehicle O 6.1 4.4 3709.6 3612.7 3482.9 2.6 6.1 6.1
Vehicle P 4.6 4.4 3971.8 4307.8 4240.2 8.5 6.8 6.8
Vehicle Q 3.9 3.7 6587.2 6170.4 6433.7 6.3 2.3 2.3
Vehicle R 3.8 3.8 3993.3 3624.2 3809.5 9.2 4.6 4.6
Vehicle S 4.1 4 2794.7 2534.4 2694 9.3 3.6 3.6
Average 3.8 2.0 4.7
St. dev. 7.4 5.1 1.8

Fig. 8. Comparison between measured and model estimates of fuel consumption rates according to VSP mode.

3.3.2. Global fuel use
On-road measurements were used to verify the applicability of
the methodology proposed in this work. Three different
approaches were compared with the second-by-second data inte-
grated along the route: modal data from on-road measurements
(on-road); modal NEDC estimates (cert. based); and modal NEDC
estimates corrected by on-road data (model).
Table 10 presents a summary of the comparison between the
integrated second-by-second measured fuel consumption with
the modal VSP approach using on-road data and estimates pro-
vided by certication, using all the data points for the 14 vehicles
studied. The VSP modal approach using on-road measurements
data provides the same results as the second-by-second sum of
fuel-consumption data. Using only certication data, an
 G.O. Duarte et al. / Energy Conversion and Management 92 (2015) 251265 261

under-estimation of fuel consumption of 12% in average is Table 12

observed. The nal model estimates using the on-road data pro- Summary of the vehicles evaluated with the model proposed in this work, considering
different vehicle classes.
vides robust results, with an average deviation of 2.2% and an
absolute deviation of 9.2%. This method appears not to be biased, Inputs
as it under-estimate and over-estimate results, depending on the ECE-15 (l/100 km) EUDC (l/100 km)
vehicle. Supermini A 4.5 3.9
If vehicles C and D were considered as outliers, and by apply- Supermini B 4.3 3.2
ing the model, an average deviation of 1.6% with standard Small Family Car C 3.9 3.2
deviation of 7.9% was obtained. The absolute average deviation Large Family Car D 5.6 3.7
Executive E 5.7 4.2
was of 6.2% with standard deviation of 4.8%. Vehicle C was only
monitored under urban environment, resulting in limited time
under higher VSP modes and vehicle D (which had a direct-
injection spark-ignition engine) presented fuel consumption Table 11 presents a summary of the comparison between the
rates higher than expected, also conrmed by the on-board con- integrated second-by-second measured fuel consumption with
sumption information. These results show that the model devel- modal VSP approach using on-road data and the estimates pro-
oped provides satisfactory results using the few data available vided by certication, using all data points for the 5 hybrid vehicles
from certication. Using only known data was an objective of studied. Although the number of vehicles is low and, consequently,
this work, so obtaining an average deviation lower than 10% few data is available, the model provides good estimates compared
for all vehicles and of 6% excluding two possible outliers is with the measured data. Using only certication inputs, the aver-
an adequate result. age deviation is lower than 4% (with 7.4% of standard deviation).

Fig. 9. Schematics of algorithm of the methodology proposed.

3
Vehicle Class
Supermini - A (SI)
2.5 Supermini - B (CI)
Small Family Car - C (CI)
Large Family Car - D (CI)
2 Executive - E (CI)
Fuel (g/s)

1.5

0.5

VSP Mode (W/kg)

Fig. 10. Estimates of fuel consumption rates for ve vehicle class vehicles according to VSP mode. Vehicle A uses a Spark Ignition engine and the remaining use Compression
Ignition engines.
262 G.O. Duarte et al. / Energy Conversion and Management 92 (2015) 251265

Fig. 11. VSP modal time distribution of drive cycles collected under real conditions, typical for: (a) urban trip; (b) urban and extra-urban; and (c) highway, extra-urban and
urban.

When considering the corrections applied to the model, the abso-
lute deviation is lower than 5% with 1.8% of standard deviation.
4. Methodology application and potentialities
4.1. Comparison between model estimates and real-world
measurements
The developed model was applied to a vehicle that was mea-
sured under on-road conditions, but that was not included in the
development of the model. Using only certication data as input,
Fig. 8 presents the comparison between the model estimates and
the on-road measurements. The estimates present a good agree-
ment with the measured VSP modal fuel consumption rates, with
an R2 higher than 0.9. Due to the high engine power of this vehicle,
the points collected in VSP mode 31 (which include all points over
VSP 30) correspond to very high power demand conditions (largely
exceeding the 30 W/kg), hence the considerably higher VSP fuel
consumption rate measured.
This type of analysis is useful to predict what would be the
expectable results from on-road monitoring tests. Usually on-road
 G.O. Duarte et al. / Energy Conversion and Management 92 (2015) 251265
Fig. 12. Estimates of fuel consumption on different trips for different vehicle
classes, comprehending SI and CI powertrains.
tests are design to cover a broad range of power requirements (VSP
modes), however, since these tests are made on public roads (with
trafc) not all VSP modes are covered with enough points to have a
good representation of fuel consumption, consequently this tool
can provide a solution to ll some data gaps.
4.2. Application to best-selling vehicles in Portugal in a fuel life-cycle
approach
The data provided by the model can be used to evaluate the
energy use impacts on different driving cycles with a dened VSP
time distribution. Fig. 9 schematically presents a possible applica-
tion of the developed model. Using only public data from certica-
tion fuel consumption, it is possible to build the modal fuel
consumption for a given vehicle. Combining this information with
VSP modal time distribution of a desired drive cycle, it is possible
to present an estimate of the fuel consumption that a dened vehi-
cle would have if it was performing any desired driving prole.
An analysis of the fuel consumption was performed to the best-
selling vehicles in Portugal from January to October 2012, accord-
ing to data from Portugal Automotive Association [20]. The vehi-
cles with highest sales in the 5 segments in the referred period
were selected, with the main characteristics summarized in
Table 12.
Fig. 10 presents the estimates of fuel consumption rates given
by the developed methodology for the vehicles selected according
to the VSP mode, given the respective certication inputs. The
vehicles with higher ECE-15 values are the ones with highest fuel
consumption rates at lower VSP modes. Vehicle A presents low fuel
consumption rates at low VSP modes increasing considerably at
higher VSP modes.
4
TTW
3 Supermini A (SI)
Energy (MJ/km)
0
UDC
EUDC
HDC
Fig. 13. Tank-to-Wheel and Well-to-Tank energy use estimates for ve vehicles from different segments and three driving cycles. UDC Urban Driving Cycle; EUDC Extra-
Urban Driving Cycle; HDC Highway Driving Cycle.
263
The data provided by the model can be used as input to evaluate
different driving cycles with a dened VSP time distribution.
Therefore, 3 case-study drive cycles collected under on-road condi-
tions were considered to evaluate the vehicles presented on
Table 12, using Eq. (10).
Combining the VSP modal time distribution of the case-study
trips (under urban, rural or extra-urban and highway conditions),
shown in Fig. 11, with the estimates of fuel consumption rates pro-
vided by the model developed in this work, the performance of
each vehicle on each trip was evaluated.
Fig. 12 presents the estimates of fuel consumption (in liters per
100 km) for the trips presented in Fig. 11. As expected, fuel con-
sumption tends to increase with the vehicle class, as a result of
vehicle characteristics, namely higher mass. Vehicles D and E are
strongly penalized in urban environments, but their performance
is fairly good in extra urban and highway trips. In more detail,
Vehicle D presents a 33% higher fuel use and Vehicle E a 40% higher
fuel use compared to Vehicle C (which has the lowest fuel con-
sumption in urban trips). For extra urban and highway driving con-
ditions, having as a reference Vehicle B, Vehicle C has 1% and 4%
higher fuel use, respectively on extra urban and highway, while
Vehicle D spends 16% and 10% more fuel and Vehicle E 31% more
fuel in both extra urban and urban trips.
The energy impacts of a vehicle include both the usage stage
(Tank-to-Wheel, TTW) but also the energy required to obtain the
fuel used (Well-to-Tank, WTT). The combination of both stages is
called Well-to-Wheel (WTW). The energy required to obtain 1 MJ
of fuel depends on the type of fuel (gasoline or diesel), since their
transformation processes differ and will be increasingly important
with the use of electricity in the road transportation sector. The
coefcients used were obtained from the work by Baptista [21].
Fig. 13 presents the estimates of the energy use on the TTW and
WTT stages for the ve vehicles selected from different segments
and different driving conditions. The results reinforce that urban
driving is where energy consumption is higher. For the vehicles
and trips considered, urban driving represents 42% and 43% higher
WTW energy consumption comparing with extra-urban and high-
way driving, respectively. On average, the extra-urban trip consid-
ered presents 0.2% higher fuel consumption than highway, but
depends on the vehicle considered (A and C have less fuel con-
sumption in extra-urban context).
The sales of these vehicles in 2012 in Portugal were of 1437,
3664, 4118, 1828, 1212 units for Vehicle A, B, C, D and E, respec-
tively. This represents a WTW energy requirement of 33850 MJ/
km for the urban trip and the ve vehicles considered,
23,944 MJ/km for the extra-urban route and 23,971 MJ/km for
the highway trip. Although the energy requirements per kilometer
WTT Large Family Executive E (CI)
Car D (CI)
Small Family
Supermini B (CI) Car C (CI)
UDC
EUDC
HDC
UDC
EUDC
HDC
UDC
EUDC
HDC
UDC
EUDC
HDC
264 G.O. Duarte et al. / Energy Conversion and Management 92 (2015) 251265
are circa 40% higher in this urban trip, due to its short distance
(14.6 km, as presented in Fig. 11a), the extra urban route represent
237% more energy (it has 69.6 km), while highway route
(130.4 km) has more 532% WTW energy requirements. It can also
be observed that the vehicles with highest share of sales in Portu-
gal (B and C) have, on average 11% lower energy requirements
when compared with the other vehicles in a WTW analysis, which
is in accordance with the recognized low power motorization vehi-
cle eet in Portugal.
5. Conclusions
This research work focused on the development of an innova-
tive methodology to estimate the Vehicle Specic Power (VSP)
modal fuel consumption based only on public available data,
namely the European certication cycle information. Therefore, a
novel simple and fast method was developed for different vehicle
technologies (SI, CI and hybrids) to enable an energy characteriza-
tion of a vehicle according to the specic power required while
driving, without the need to perform on-board vehicle measure-
ments of individual vehicles.
A portable laboratory was used to measure fuel consumption,
pollutant emission and vehicle dynamics under on-road conditions
of 14 conventional vehicles (spark-ignition and compression-igni-
tion) and 5 hybrid vehicles. For each second of driving, according to
the power demand resulting from the combination of speed, accel-
eration and road grade as dened by the Vehicle Specic Power
methodology, the correspondent VSP mode was calculated. Using
the data collected on-road, a generic trend of fuel consumption
as a function of VSP mode was observed, which was dened by 6
coefcients and adjustable from vehicle to vehicle according to
the certication inputs.
Considering all the 14 conventional vehicles, the average R2
between measured and estimates is around 0.93 and if two outlier
vehicles were not considered, this result increases to 0.967. In this
case the lowest R2 was found to be 0.881 for a SI vehicle and the
highest, 0.992, for a CI vehicle. Regarding the 5 hybrid vehicles
studied, R2 was found to be on average 0.983, with the lowest value
of 0.975 and the highest of 0.988. Comparing the integration of sec-
ond-by-second measured fuel consumption on the drive cycles for
each vehicle and the modal time distribution of the same drive
cycle, using the estimates provided by the model shows that, for
the eet of 14 conventional vehicles studied, an absolute deviation
of 9.2% 9.2% was observed (6.2% 4.2% if two outlier vehicles
were not considered) and the average deviation was of
2.2% 13.1% (1.6% 7.9% if two outlier vehicles were not con-
sidered), suggesting the accuracy and precision of the model. Using
the same approach for the 5 hybrid vehicles, an absolute deviation
of 4.7% 1.8% was observed, while the average deviation was
2.0% 5.1%. These results indicate that the model estimates
(using only certication data) are coherent with on-road
measurements.
The validation and verication of the developed methodology
was also done with experimental data from a vehicle measured
under on-road conditions, but not included in the development
of the model, indicating a good agreement between the estimated
and the measured fuel consumption rates (R2 higher than 0.9).
Additionally, a possible application of the model was presented.
The model was applied to estimate the real world fuel consump-
tion rates of best-selling vehicles in each vehicle segment in Portu-
gal. These fuel rates were applied to 3 different driving conditions
to conclude that for urban driving conditions the top selling vehicle
in segment C is the most adequate (average reduction in WTW
impacts of 15% compared to remaining segments). For extra-urban
and highway conditions, Vehicle B is the most adequate (average
reduction in WTW impacts of 11% and 12% for extra-urban and
highway conditions, respectively, compared to the remaining seg-
ments). These results reinforce that vehicles from segments D and
E are highly penalized under urban driving conditions. Moreover,
for extra-urban and highway driving patterns, the difference in
energy consumption for the best vehicle tends to decrease.
Potential applications of the presented methodology include
the prediction of the real world energy consumption impacts from
on-road monitoring tests to complement laboratorial data, which
is crucial taking into consideration the considerable gap between
certication and real world results.
Acknowledgements
Thanks are due to Fundao para a Cincia e Tecnologia for Doc-
toral (SFRH/BD/61109/2009) and Post-Doctoral nancial support
(SFRH/BPD/62985/2009 and SFRH/BPD/79684/2011).
References
[1] Kageson P. Cycle-beating and the EU test cycle for cars. European Federation
for Transport and Environment, Report n T&E 98/3; 1998.
[2] Weiss M, Bonnel P, Hummel R, Manfredi U, Colombo R, Lanappe G, et al.
Analyzing on-road emission of light-duty vehicles with Portable Emission
Measurement Systems (PEMS). Luxembourg: European Commission Joint
Research Centre Institute for Energy, Report n EUR 24697 EN; 2011. http://
dx.doi.org/10.2788/23820.
[3] Keller M, Hausberger S, Samaras C, Mellios G, Nziachristos L. Parametrization
of fuel consumption and CO2 emissions of passenger cars and light commercial
vehicles for modelling purposes. Joint Research Center, Report n EUR 24927
EN; 2011. http://dx.doi.org/10.2788/58009.
[4] Rhys-Tyler G, Bell M. Toward reconciling instantaneous roadside
measurements of LDV exhaust emission with type approval driving cycles.
Environ Sci Technol 2012;46:105328. http://dx.doi.org/10.1021/es3006817.
[5] Jimnez-Palacios J. Understanding and quantifying motor vehicle emissions
with vehicle specic power and TILDAS remote sensing. Ph.D. Thesis,
Massachusetts Institute of Technology, Dept. of Mechanical Engineering; 1999.
[6] Wipke K, Cuddy M, Burch S. Advisor 2.1: a user-friendly advanced powertrain
simulating using a combined backward/forward approach. IEEE Trans Veh
Technol 1999;48:175161. http://dx.doi.org/10.1109/25.806767.
[7] Barth M, An F, Younglove T, Scora G, Levine C, Ross M, et al. NCHRP Project 25-
11 nal report: development of a comprehensive modal emission model.
Prepared for National Cooperative Highway Research Program, Transportation
Research Board, National Research Council; 2000.
[8] Silva CM, Farias TL, Frey HC, Mendes-Lopes JMC. Evaluation of numerical
models for simulation of real-world hot-stabilized fuel consumption and
emissions of gasoline light-duty vehicles. Transp Res Part D 2006;11:37785.
http://dx.doi.org/10.1016/j.trd.2006.07.004.
[9] Gkatzoias D, Couridis C, Ntziachristos L, Samaras Z. Copert 4 Computer
programme to calculate emissions from road transport. Laboratory of Applied
Thermodynamics Mechanical Engineering Department Aristotle University
Thessaloniki; 2007.
[10] Ntziachristos L, Gkatzoias D, Kouridis C, Samaras Z. COPERT: A European
Road Transport Emission Inventory model. Book series: Environmental Science
and Engineering; 2009. p. 49104. http://dx.doi.org/10.1007/978-3-540-
88351-7_37.
[11] Duarte GO, Varella RA, Gonalves GA, Farias TL. Effect of battery state of charge
on fuel use and pollutant emissions of a full hybrid electric light duty vehicle. J
Power Sources 2014;246:37786. http://dx.doi.org/10.1016/
j.jpowsour.2013.07.103.
[12] Zhai H, Frey H, Rouphail N. A vehicle-specic power approach to speed- and
facility-specic emissions estimates for diesel transit buses. Environ Sci
Technol 2008;42:798591. http://dx.doi.org/10.1021/es800208d.
[13] Frey H, Rouphail N, Zhai H, Farias T, Gonalves G. Comparing real-world fuel
consumption for diesel- and hydrogen-fueled transit buses and implication for
emissions. Transp Res Part D 2007;12:28191. http://dx.doi.org/10.1016/
j.trd.2007.03.003.
[14] Frey H, Unal A, Chen J, Li S. Modelling mobile source emissions based upon in-
use and second-by-second data: development of conceptual approaches for
EPAs New MOVES Model. In: Proceedings of annual meeting of the air and
waste management association, Pittsburgh, PA; 2003.
[15] EPA US Environmental Protection Agency. Methodology for developing
modal emission rates for EPAs multi-scale motor vehicle and equipment
emission system. Report n EPA420-R-02-027, Ofce of Transportation and Air
Quality, U.S. Environmental Protection Agency; 2002.
[16] UK Government Directgov. Directgov nd fuel consumption and emissions
information on a new or used car. <http://carfueldata.direct.gov.uk/search-
new-or-used-cars.aspx>; 2012 [accessed 05.09.12].
 G.O. Duarte et al. / Energy Conversion and Management 92 (2015) 251265 265

[17] Bosch GmbH. Milestone for fuel-saving technology Five million Bosch start/ [20] ACAP. AutoInforma Homologaes, Estatsticas, Estudos e Servios do Sector
stop starters. <http://www.bosch.com.tw/en/tw/newsroom_10/news_10/ Automvel. <http://www.autoinforma.pt/>; 2012 [accessed 13.12.12].
news-detail-page_6464.php>; 2012 [accessed 12.09.12]. [21] Baptista P, Silva C, Farias T, Heywood J. Energy and environmental impacts of
[18] Frey H, Unal A, Rouphail N, Colyar J. On-road measurement of vehicle tailpipe alternative pathways for the Portuguese road transportation sector. J Energy
emissions using a portable instrument. J Air Waste Manag Assoc 2003;53: Policy 2012;51:80215. http://dx.doi.org/10.1016/j.enpol.2012.09.025.
9921002.
[19] Society of Automotive Engineers. Recommended practice for measuring the
exhaust emissions and fuel economy of hybrid-electric vehicles, including
plug-in hybrid vehicles. Standard n J1711. SAE International; 2010.