Using on-board logging devices to study the long-term
impact of an eco-driving course
Beusen Barta,*, Broekx Stevena, Denys Tobiasa, Beckx Caroliena,b, Degraeuwe Barta ,
Gijsbers Maartena, Scheepers Kristofa, Govaerts Leena, Torfs Rudia, Int Panis Luca, b
aFlemish Institute of Technological Research (VITO), Boeretang 200, 2400 Mol, Belgium
bTransportation Research Institute, Hasselt University, Wetenschapspark 5 bus 6, 3590
Bart Beusen: Flemish Institute of Technological Research (VITO)
Address: Boeretang 200, 2400 Mol (Belgium)
e-mail address: firstname.lastname@example.org
Tel. +32 14 30 82 79
Fax +32 14 32 11 85
We have evaluated the long-term impact of an eco-driving training course by monitoring
driving behavior and fuel consumption for several months before and after the course.
Passenger cars were equipped with an on-board logging device that logged the position and
speed of the vehicle by means of a GPS tracking system as well as real time electronic
engine data extracted from the Controller Area Network (CAN). The CAN data included
information on mileage, number of revolutions per minute, position of the accelerator pedal
and instantaneous fuel consumption. Data gathered over a period of 8 to 10 months for 10
different drivers during real-life conditions enabled an individual drive style analysis. The
average fuel consumption for all drivers 4 months after the course was reduced by 5.8%.
Most drivers showed an immediate improvement in fuel consumption which was steady
over time. On the other hand some drivers tend to fall back into their original driving habits
for one or more aspects of driving behavior that were treated in the eco-driving course.
Fuel-efficient driving, Eco-driving, on-board logging device, Controller Area Network
CO2 emissions from vehicles in general and particularly from road transport are an
important fraction of total greenhouse gas emissions in most developed countries and are
often projected to rise further in the future. Among the policy options to reduce CO2
emissions from transport that have been studied eco-driving has always featured high
among the choices with a great reduction potential. Reducing fuel consumption significantly
by teaching drivers how to change their driving behavior is potentially a very cost-efficient
manner to reduce energy use and emissions (IEA, 2005).
A lot of studies report the short term impact of eco-driving on fuel consumption.
ECMT/IEA (2005) decided on an average estimate of 5% reduction for all OECD regions
based on an expert analysis of available literature. Few studies report on the long-term
impacts of fuel-efficient driving courses. Wahlberg (2007) monitored fuel consumption
reduction in busses and recorded 2% fuel savings during a 12 month period after training.
Zarkadoula et al. (2007) mention fuel savings on busses of 4.35% during a post training
monitoring period of 2 months. Both studies report on the fact that, after a certain time
period, drivers (partially) slip back into a less environmentally friendly driving habit
resulting in less fuel savings than originally attributed to the eco-driving course.
The aim of this paper is to present the results of a long-term passenger car monitoring
campaign that recorded driving patterns and fuel consumption from people participating in
an eco-driving course. A panel of drivers was followed up during 8 to 10 months to analyze
the impact on fuel consumption and on different driving parameters.
To analyze the impact of an eco-driving course on fuel consumption and driving behavior
on the long run, we developed a data logging device to monitor people’s driving behavior
accurately. In this section we will first describe this data collection device and then discuss
the set-up of the panel survey and the eco-driving course.
2.1 On-board data collection device
Tracking the amount of fuel bought by participants over a long period of time introduces
unwanted effects into an eco-driving experiment. First the burden on the participants to
record the details of every fuel purchase is high resulting in drop outs. Secondly bias may be
introduced by directly confronting participants with their fuel use long before the eco-
driving course. To obtain data in a long-term driving monitoring experiment without
affecting the burden on its participants or introducing bias, we have used an on-board device
to monitor and log the necessary driving parameters. The on-board device is equipped with
a memory card, a GPRS-modem, a GPS tracking system and is connected to the Controller
Area Network (CAN) of the vehicle. This configuration allows monitoring and logging of 2
types of data: the position and speed of the vehicle by means of the GPS tracking system
and electronic engine data extracted from the CAN-bus, which includes data on mileage,
number of revolutions per minute (RPM), position of the accelerator pedal, gear selection,
instantaneous fuel consumption and engine coolant temperature. The logging device is small
(approximately 10 by 10 cm) and was installed out of sight of the driver. Based on the
decoding of the CAN messages, the device was programmed to log the required CAN
parameters for a particular car. The logging device did not interfere with the engine
management system. Data were read from the CAN, stored on the internal memory card of
the on-board logger and transmitted to a central server via the GPRS-modem on a daily
Drivers could consult their recorded positions on a website, but not the data on their driving
behavior or fuel consumption. By no means was any information on their driving behavior
fed back to the drivers during the project, in order not to influence their driving behavior
during the monitoring period. Drivers were requested to add additional information on a
trip-by-trip basis concerning travel motives, the number of passengers and the actual driver
of the vehicle. Especially this last feature is important in this study as mostly multiple
drivers could use the same vehicle and the actual driver for a particular trip was not always
the same person that participated in the eco-driving course. Based on the trip declarations of
actual drivers this effect was filtered out. An analysis of fuel consumption and driving
behaviour as a function of trip motives and locations will be presented in a forthcoming
2.2 Experimental design
2.2.1 Participants and vehicle selection
Participants for the experiment were solicited both internally at VITO and externally using
an announcement in a car magazine (‘autogids’). In order to be eligible for the monitoring
campaign, the candidates had to be the owner of the vehicle, the vehicle had to be of a
model year later than 2001 (older vehicles generally do not have a CAN-bus) and had to be
at least 2 years old at the start of the experiment (because of warranty issues). In total, 30
vehicles were equipped with the on board logging device as described earlier. All of these
vehicles were registered in Belgium and predominantly used in the northern or Flemish
speaking part of the country.
The information available on the CAN-bus differs between vehicles from different
manufacturers, or even between different models from the same manufacturer. As
information on data protocols is not publicly available, signals had to be analyzed using a
case by case testing approach to find out how different parameters are transmitted. Not all
parameters were available for all cars to analyze because they were not found within the
data stream or were simply not present on the CAN. Fuel consumption, the critical
parameter within this trial, was registered for 19 vehicles. Throttle and RPM were registered
for 19 and 21 vehicles respectively. Further, for some cars an insufficient amount of data
was logged or too little trips were declared by the driver (for 2 months before and after the
course at least 100 km per month had to be declared by the driver taking the course). 2
drivers did not take the driving course. Due to all these issues, of the 30 cars equipped with
the monitoring device, only 10 could be used for further analysis in this paper. This sample
size is too small to learn lessons on the general impact of eco-driving. However, as results
will show, the sample size is large enough to indicate significant differences between
individual drivers and how drivers respond in time to an eco-driving course. Details on the
10 vehicles used for this analysis are shown in Table 1.
(Insert Table 1 here)
2.2.2 Eco-driving course
Approximately halfway the project, the participants were given a 4 hour course on fuel
efficient driving (see Table 1 for the exact dates). This course consisted of a drive with a test
vehicle prior to the course, a session on the rules of fuel-efficient driving and a second drive
with the same test vehicle and with guidance of an instructor
The main rules of fuel efficient driving (or e-positive driving (www.e-positief.be) as it was
called), can be summarized as follows:
1. Shift up as soon as possible (Shift up between 2000 and 2500 revolutions/minute)
2. At steady speeds use the highest gear possible and drive with low engine RPM
3. Try to maintain a steady speed by anticipating traffic flow
4. Decelerate smoothly by releasing the accelerator in time while leaving the car in
gear (this is called “coasting”).
Further, some additional driving style instructions were provided at the course:
5. Shut down the engine for longer stops, e.g. before a level crossing or when you pick
6. Do not drive faster than 120 km.h-1 (which is the legal speed limit on motorways in
Vehicle data were logged to study the impact of this training course on fuel consumption
and to check how well the different rules stated above were observed by the participants.
2.3 Data analysis
Over a 10 month period we collected 116.355 km and 2.026 hours of data for these 10
private vehicles. Based on this we calculated 10 relevant driving parameters. The selection
of these parameters was performed based on their relevance towards eco-driving. Table 2
presents an overview of the selected parameters, their units and the corresponding
abbreviations used in this paper. Two trip related parameters, the total distance and the
average speed, were also analysed to evaluate whether the travel pattern had changed
significantly over time. All the parameters were calculated on a weekly basis.
(Insert Table 2 here)
3.1 General effect of the eco-driving course
To get an overall picture of the effect of the course a 3-way ANOVA was performed. The
three factors included in this statistical analysis are the driver (1-10), the course (before or
after the course) and the week. Table 3 shows the p-values of the effects and interactions of
the three factors for the different parameters. From this analysis of the dataset as a whole,
we can conclude that:
- Neither week nor course have a significant effect on the total weekly distance and
the weekly average speed. This means that the drivers did not change their travel
patterns significantly. Differences can therefore be attributed to changes in driving
patterns rather than changes in travel patterns.
- The course has a significant effect on most driving parameters, except for average
shifting point and time idling.
- The course has a significant effect on the fuel consumption of the population as a
- The driver-course interaction has a significant effect on most behavioral parameters
except for the time idling. This indicates that the size of the effect of the course
depends on the driver.
- The week does not have a significant effect on fuel consumption. However the
interaction driver-week has a significant effect which indicates that the change of the
effect over time also depends on the driver.
(Insert Table 3 here)
Previous work from Vlassenroot et al. (2006) suggests that overall statistics for the
population as a whole provide insufficient information to assess the impact of Intelligent
Speed Adaptation on driving behaviour. Opposite individual impacts were averaged out if
only statistics for the total population were considered. The fact that within this experiment
the interactions of the factor course with the factor driver are very significant for most
driving parameters, suggests that the impact of the eco-driving course differs greatly
between drivers. Analyses on the impact of the eco-driving course on fuel consumption and
driving parameters are therefore performed per driver in the next section.
3.2 Individual effect of the eco-driving course
Table 4 presents the results from a t-test on the selected parameters to analyze the
differences before and after the eco-driving course for each driver. The effects, values and
confidence intervals of the effects are listed. The change in fuel consumption is expressed as
a relative change (%) of the average fuel consumption (l/100km) compared with the
observation before the course.
(Insert Table 4 here)
From Table 4 we can see that the trip parameters (average distance and average speed) did
not change significantly between the trips before and after the course, indicating that, on
average, the trips before the course were similar to the trips after the course. This confirms
the result from the 3-way ANOVA.
Average fuel consumption over all drivers decreased by approximately 6%. Per driver, these
changes varied between -12% and +3%. Fuel consumption decreased significantly after the
course for 7 out of 10 drivers.
The differences in fuel consumption are also reflected in a changing driving behaviour.
Parameters showing significant improvement for at least 7 out of 10 drivers are average
shifting point, distance coasting and distance driven with low rpm. The shifting point
decreased on average for all drivers with 97 rpm, with extreme values around 200 to 500
rpm. The percentage of distance coasting increased on average with 1.16% and the
percentage of time driven at low rpm increases on average with 12.2%. Drivers 2 and 4 who
did not achieve fuel savings after the course, also show significant improvement in the time
driven at low rpm. The average shifting point did not change significantly for these two
drivers indicating the importance of this parameter in achieving fuel savings.
The percentage of time during heavy acceleration or deceleration decreases by 0.26% and
0.22% respectively. These changes may seem small. However these kind of events occur
less frequently compared to coasting. The significant impact demonstrates that people did
change their acceleration and deceleration behaviour in most cases.
No significant impacts were observed in the time idling. In general the time idling decreases
but this impact is significant in only 2 cases. The amount of longer stops justifying engine
shutdowns is probably very limited. The fact that idling decreases is also not only due to
increased shutdown of engines during long stops. People showing increased coasting
distance will also spent less time standing still at crossroads and traffic lights.
The percentage of distance driven at speeds higher than 120 km/h decreases on average with
6.3 %. This decrease was significant for 3 drivers only.
3.3 Individual effects over time
We analyzed the different parameters on a weekly basis to check for a possible learning or
fading effect. Figures 1 and 2 show the impact for the individual drivers over time for the
parameters fuel consumption and shifting point. Regression lines were fitted before and
after the course to indicate trends over time. Table 5 shows the number of drivers for which
the slopes of these regression lines before and after the course decreased or increased
(Insert figures 1 and 2 here)
Some parameters already displayed a change over time before the eco-driving course took
place. Four drivers already realized a significant decrease in fuel consumption, indicating
that these drivers already practiced a more environmentally friendly driving behavior before
participating in the eco-driving course. The evidence for a changed driving style can be
found in the driving parameters where we observe decreasing average shifting points. The
average fuel consumption after the course gradually increases again for 2 out of 10 drivers,
while for 1 drivers fuel consumption significantly keeps improving after the course. For the
other 7 drivers, no significant change in fuel consumption is seen after the eco-driving
(Insert Table 5 here)
The temporal variations in the parameters after the eco-driving course demonstrate that for
some parameters most drivers continue to improve their eco-driving skills. This is the case
for the percentage of distance covered by coasting and the percentage of distance covered in
optimal gear. Other parameters like the heavy acceleration and idling tend to deteriorate
again after the course (gradual deterioration or increase for 4 drivers, no gradual
improvement for any driver). The impact on average shifting is very different between
drivers. While shifting points tend to increase again for 5 drivers, 3 drivers manage to
further reduce their average shifting points.
This analyses shows that the post-educational effects can vary strongly from one driver to
another. While some drivers use what they’ve learned to continue improving their eco-
friendly driving style, other drivers tend to forget what they’ve learned. This indicates that
drivers need more repetitive training to keep on applying eco-driving rules for shifting point
and traffic anticipation. The level of coasting however seems to stay the same or even
improve over time.
The technical set up of this project proves that on-board logging devices can be used to
study driving behaviour over long time periods. Collecting data through CAN messages
provided a large amount of additional information on fuel consumption, engine speed and
gear shifting on a second by second basis.
The mean change in fuel consumption for all drivers after their eco-driving course showed a
reduction of 5.8%. Differences between individual drivers however varied between -12%
and +3%. We may have underestimated the potential effect of the eco-driving course
because drivers participated on a voluntary basis and 4 out of 10 drivers had already reduced
their fuel consumption gradually during the monitoring period in the weeks before attending
The differences in fuel consumption are also reflected in changes in driving behaviour
parameters based on the eco-driving tips: shifting point during acceleration moved closer to
the optimal 2000 rpm, the distance driven while coasting increased, fewer heavy
accelerations and decelerations events occurred and the distance driven at steady speeds
using the optimal gear increased. However not all drivers realized significant improvements,
2 out of 10 drivers didn’t achieve any fuel savings.
This study was conducted as a part of the Feathers project which was partly financed by the
Flemish government through the Institute for the Promotion of Innovation by Science and
Technology in Flanders (IWT) under the SBO programme for strategic basic research.
ECMT/IEA, 2005. Making cars more fuel efficient: technology and policies for real
improvements on the road. OECD/ECMT and IEA.
IEA, 2005. Saving oil in a hurry. IEA publications, Paris, France. pp.164.
Vlassenroot, S., Broekx, S., De Mol, J., Int Panis, L., Brijs, T., Wets, G., 2007. Driving with
Intelligent Speed Adaptation: Final results of the Belgian ISA-trial. Transportation Research
Part A 41, 267-279.
Wahlberg, A.E., 2007. Long-term effects of training in economical driving: Fuel
consumption, accidents, driver acceleration behaviour and technical feedback, International
Journal of Industrial Ergonomics 37, 333-343.
Zarkadoula, M., Zoidis, G., Tritopoulou, E., 2007. Training urban bus drivers to promote
smart driving : a note on a Greek eco-driving pilot program. Transportation Research Part D
Table 1: Details on the vehicles used in the this eco driving experiment
Id Car model Fuel type Year Monitoring
1 Citroen Xsara diesel 2002 dec/2007 mar/2008 jun/2008
2 Citroen C5 diesel 2003 aug/2007 dec/2007 apr/2008
3 Renault Clio III 1.2 16V gasoline 2005 mrt/2008 jul/2008 nov/2008
4 Lexus IS 220 d diesel 2006 feb/2008 jun/2008 okt/2008
5 Renault Vel Satis 2.2 cdi diesel 2003 feb/2008 jun/2008 nov/2008
6 VW golf diesel 2000 okt/2007 dec/2007 apr/2008
7 Audi A3 diesel 2007 feb/2008 jun/2008 nov/2008
8 Peugeot 206 1.4 Hdi diesel 2002 nov/2007 mar/2008 jul/2008
9 Ford mondeo 2.0 tdc diesel 2002 dec/2007 mar/2008 jun/2008
10 Audi A4 2.5d automatic diesel 2005 dec/2007 mar/2008 jun/2008
Table 2: Selection of the parameters used for this study. The relevance of each
parameter to one or more eco-driving rules is indicated in the last column.
Parameter Unit Abbreviation Description Rule
Total distance km
Tot_dist Total distance covered
Average speed Avg_speed Average driving speed (time based )
Average fuel consumption Avg_fc Average fuel consumption
Average shifting point Rpm Avg_sp Average engine speed reached before
shifting to a higher gear during acceleration
Percentage distance coasting
% Dist_coast % distance covered during prolonged
coasting actions (prolonged coasting is
defined as period of at least 3 seconds while
fuel consumption = 0, and speed > 0)
% time driven at accelerations > 1.5m.s2
Percentage time heavy
% Time_acc 3
Percentage time heavy
% Time_dec % time driven at decelerations > 2.5m.s2 3 & 4
Percentage time idling % Time_idl % time standing still (v < 3 km.h-1) with the
engine running (idling)
Percentage distance in
% Time_rpm % distance covered with engine speed
between 1100 and 1700 rpm
(optimal engine speed for steady speeds)
2 & 3
Percentage distance at more
than 120 km.h-1
% Dist_120 % distance covered at speeds higher than
Table 3: Significance -values of the effects and 2-factor interactions of a 3-way Download full-text
Parameter course driver week course*driver course*week driver*week
Tot_dist *** ***
Avg_speed *** **
Avg_fc *** *** **
Avg_sp *** *** **
Dist_coast ** *** *** *** ***
Time_acc *** *** ** *** ** ***
Time_dec *** *** **
Time_rpm *** *** * *** **
Dist_120 *** *** *** *
* = p<0.05; ** = p<0.01; *** = p<0.001