Proceedings

22nd Annual Congress

Firenze 2013

English Version

ISBN -------------
Published by IT.Comm for:
EVU Italia
European Association of Accident Research and Analysis

Chairman of the Board: Prof. Dr. Hermann Steffan

Committee EVU Headquarters
Inffeldgasse 11/II
8010 Graz (A)
http://www.evuonline.org

Scientific Board of the Conference

Unarski Jan Prof.
Hugeman Wolfgang PhD
Lugner Peter Prof.
Rivano Virginio Dr.
Vangi Dario Prof.

Chairman of the Organizing Committee

Virginio Rivano Dr.

EVU Italia
Via Roma, 12
50012 Bagno a Ripoli (Firenze)
http://www.evuitalia.eu

© EVU - 2013
All rights reserved
Partial reproduction is only permitted under citation of the source
ISBN --------
It is a great pleasure to welcome you to Florence for the 22nd EVU Annual Congress.

For the first time EVU Italia is in charge of the organization of this important event and
I am sure Florence and Tuscany will be able to offer a warm and elegant welcome to all
our friends from all over the world.

More than thirty reconstructors and researchers will discuss the results of their work on
accidents with two-wheels vehicles (reconstruction and safety), human factors in road
accidents, use of EDR (Event Data Recorder) data in the reconstruction process.
Moreover, an Open Forum session will be held to discuss some topics which are
particularly interesting and relevant.

The Congress will be a stimulating and challenging workshop but it will also offer you a
great chance to visit this fascinating city after a day of hard work, enjoying the beauty,
culture and food of Tuscany.

Once again, many thanks for taking part in this event.

I hope you will come back home hoping to come back soon to Florence.

Virginio Rivano
 Acknowledgements
First of all, I would like to express my thanks to the 31 speakers who contributed with
their hard work to the contents of this Congress. My gratitude also goes to another 20
colleagues who submitted their abstracts to the Scientific Committee, but due to the
limited time for the delivering the speeches, were not selected. All proposals deserved
attention and choosing was really hard.
Secondly, I would like to thank the Scientific Committee itself, and in particular its
chairman, Professor Unarsky, for the important support to guarantee the quality of the
selected papers.
The translation of the articles was carried out by 15 German volunteers, coordinated by
Stephan Shall, while 15 Italian volunteers, with the help of Anna Caldini (for German)
and Francesca Rivano (for English), have made the publication of these multilingual
Proceedings possible. Thank you all for supporting and helping me in this challenging
task of organizing the Congress. Another special thanks goes out to Tommaso Morandi
who brilliantly and professionally managed the layout and publication of the papers.

For the general organisation of the Congress, my heartfelt thanks go to my assistant,

Lucia Pagni and the Scaramuzzi Team, particularly Laura Rombai and Francesca
Vannucchi, who made my dreams reality, taking care of every single detail and helping
me to be practical and down to earth.

In conclusion, I recognize that this congress would not have been economically
sustainable without the contribution of our sponsors and express my gratitude to them
for helping despite the tough economic times.
 Table of Contents

Paper # Pag

2013 - 01 Experimental Analysis of Post-Accident Motion of Motorcyclist 11

Dario Vangi, Antonio Virga, Jarno Zaffelli,

2013 - 02 Motorcycles Slide Tests Analysis 17

Filippo Begani, Carlo Cialdai, Dario Vangi, Antonio Virga

2013 - 03 Sliding Decelerations of Light Weight Scooter Models (50 cc / weight 60-100 kg) 27
Jan Meuwissen

2013 - 04 Motorcycle to Car Collisions: Evaluation of Energy Loss and Relative Impact 35
Velocity
Filipo Begani, Carlo Cialdai, Dario Vangi, Antonio Virga

2013 - 05 On the Lean Angle of Motorcycles Driving through a Courve 45

Christian Hädrich, Wolfgang Hugemann,

2013 - 06 Speed Loss During a Motorcycle Stoppie 53

Roy Strzelec, Wolfgang Hugemann

2013 - 07 Comparision of Computer Simulations and Reconstruction Methodologies for 61

Motorcycle Collisions
Charles Funk, Christian Sax,

2013 - 08 Use of Computer Simulation in Motorcycle Traffic Accident Analysis 71

Senad Omerovic, Simon Krasna, Ana Trajkowski, Ivan Prebil,

2013 - 09 Determining Visibility Distance Based on Measurements with LMK System 85

Jan Unarski, Wojciech Wach, Piotr Ciępka

2013 - 10 Driving simulator analysis of drivers' behaviour in tunnels with obstacles 95

Lorenzo Domenichini, Francesca La Torre, Dario Vangi, Valentina Branzi, Carlo
Cialdai, Monica Meocci, Antonio Virga

2013 - 11 Hazard detection in driving context and breaking reaction time 101
Rita Ciceri, Federica Confalonieri, Dario Vangi, Antonio Virga

2013 - 12 Analysis of Drivers’ Conduct while Driving over modern Pedestrian Crossings 107
Robert Kledus, Marek Semela, Pavel Maxera, Martin Kunovský

2013 - 13 Recording and Processing of Data Crash Tests Performed by Institute of Forensic 119
Engineering University of Zilina
Gustav Kasanicky, Pavol Kohut

2013 - 14 Vehicle Accelerations and Time of Impacts based on a Data Analysis of Accident 125
Data Recorders
Marvin Lammert

2013 - 15 The Usage of Smartfones for Recording Accidents and Incidents from the Critical 131
Situation up to the Post-crash Phase
Lars Hannawald, Mario Marschner, Henrik Liers
2013 - 16 The Finite Element Calculation Module in PC-CRASH 10.0 Application of FE 137
calculations in Accident Reconstruction
Andreas Moser, Hermann Steffan

2013 - 17 Bypassing Manoeuvre Driving a Car with Reduced and No Tire Pressure 145
Jakub Zebala, Wojciech Wach, Piotr Ciępka, Robert Janczur

2013 - 18 TACOT project, toward a Trusted GNSS function 157

Steeve Favre, Ralf-Roland Schmidt-Cotta

2013 - 19 Reconstruction of an accident involving a Hummer driving at high speed and a 163
Nissan turning in based on a crash test using identical vehicles
Manfred Becke, Joost Wolbers

2013 - 20 Viewing Analysis of Experienced vs. not Experienced Motorcyclists and 171
Conclusions about Traffic-safety
Ernst Pfleger

2013 - 21 Rollover of Mountain Bikes which are equipped with disc brakes and suspension 175
forks
Christian Hittinger

2013 - 22 Bicycle Tire Friction Coefficient Variance in Wet and Dry Conditions Across 183
Multiple Surface Pavement Types
Daniel. L. Melcher, Christian R. Sax, Rachel E. Keller

2013 - 23 Simulation and Analysis of Quad Bike Rollovers using PC-CRASH to Evaluate 193
Alternative Safety Systems
S.Richardson, T.Orton, C. Jones, A. Sandvik, T.Pok

2013 - 24 Light reflective Attributes of Selected Materials 203

Jan Schejbal, Arnost Kure, Jakub Motl, Jan Skoda, Michal Belak, Albert Bradac,
Martin Bilik

2013 - 25 Driver Response to Roadway Intrusion at Night 211

John Grindey

2013 - 26 Consideration of Human Factors in Road Accident Investigation 221

Sibille Birth, Lorenzo Domenichini

2013 - 27 A Review of Speedometers and the Criteria to Be Considered Before Accepting 229
Frozen Readings and After Marks
Ch. Goddard, D.Price

2013 - 28 Muscular activity in normal driving and in rear end low speed crash tests 239
Dario Vangi, Antonio Virga, Carlo Cialdai

2013 - 29 Impact on roadside trees - not only German problem 247

Alexander Berg, Jörg Ahlgrimm, Stephan Schlosser, Hermann Steffan

2013 - 30 The Importance of Active Control Systems on Accident Avoidance 261

Leonidas Kakalis, Federico Cheli, Edoardo Sabbioni

2013 - 31 Traffic accidents caused by electronic failure 269

Uzi Raz
© 2013 by EVU
EVU 2013 – 01
Experimental analysis of post-accident motion of motor-
cyclists
Dario Vangi (University of Florence)
Antonio Virga (University of Florence)
Jarno Zaffelli (Dromo - Applied Italian Circuit Design)

Abstract
A common feature in road accidents is the total loss of control of the vehicle. Dealing with motorcycle accidents this
means that the rider usually abandons the vehicle and follows a different trajectory. From the point in which the control
loss happens and the point in which the body is at rest, usually the latter bounces, rolls and slides on surfaces that can be
asphalt, grass or gravel. In technical literature can be rather easily found data regarding coefficients of friction of sliding
motorcycles (with or without fairing) as in [1], or projected pedestrians [2]. Little can be found about sliding
motorcyclists: in [2] coefficients of friction for a motorcyclist wearing normal clothes and sliding on asphalt or grass can
be found. The aim of this work is the characterisation of different surfaces as regards the behaviour of motorcycle riders
after having lost the control of their vehicles, aiming at reconstructing road accidents and predict sliding distance in case
of a fall. The experimental activity here described, carried out at different sites, on different surfaces and with different
speeds, allows the prediction of sliding distances from speeds up to 240 km/h, typical of road racing (like Isle of Man
TT) and circuit racing. Thus the activity here presented can yield useful data for the purpose of accident reconstruction
and also for the design of run-off areas along roads and racing circuits.

State of the Art position of the dummy and the launch system was
not adequate to represent real world cases.
Very few tests and proper research has been done
until now, if any, about the analysis of post-accident Maybe also for this reason, the tests performed was
motion of motorcyclists. never applied in real world, or undisclosed until
recently.
Initially the importance was recognised in circuit
racing; a first known test on lightweight aggregates
was done by Laterilite, with Leca Stopsafe®, in
Anderstorp Racetrack (Sweden), 27 October 1999.
The test used cars and bikes to understand the
stopping power of the vehicles inside the arrestor
beds.

The results were encouraging. Only 29 metres to

stop a 170 Kg motorcyle at 70 Kph and 40 metres at
120 Kph.

But no information was knwon about the rider. The

first known tests were done by the Federation
Internationale de Motocyclisme.
In 2002 a series of tests were conducted by the
Transport Research Laboratory in UK,
commissioned by FIM. The tests were done at 100
Kph, 150 Kph on dry asphalt and 200 Kph on wet
tarmac, with a dummy dressed by leather suit, helmet
and gloves. These tests were very superficial and
done just to have an idea of the distance travelled at
such speed from a rider. Unfortunately the rigid
Figure 1 – Dummy used for the test at Transport Research
Laboratories in UK. ©TRL
The regulations - run off dimensioning
If we look at motorcycle racing, no rules are
disclosed about the run off area necessary in a
modern racetrack, nor a performance calculation
about the amount of space needed from a rider to
stop himself in case of fall or accident. Different is
the car perspective. The Federation Internationale de

11
 Dario Vangi, Antonio Virga, Jarno Zaffelli
l'Automobile (FIA) has, since the late nineties,
published the Internal Circuit Guidelines for Circuit
Design and Construction, which contains a formula
accepted by the governing body to assess the runoff
space, albeit with some limitations.
There are numerous studies about the material used
for clothing, but nothing is known about studies of
what happens to a motorcyclist post-accident.
This gap in knowledge was at the base of the
presented research.
Motorcycle rider post accident dynamic
Figure 2 – A typical sliding phase during a racetrack
accident. ©MotoGP.com
An accident could happen from a crash/collision or a
fall, either a low side (front and/or rear wheels loose
grip) or a high side (rear wheel looses grip before
regaining it, projecting the rider vertically).
The difference between racing and everyday is the
post accident environment.
While in racingwe will encounter tarmac, gravel,
grass and, eventually, the line of protection, on roads
we have tarmac and line of protection, if not another
obstacle before it.
The idea behind the research
In 2008 the Misano World Circuit management
asked Jarno Zaffelli to conduct investigations about
the new run off areas, built in 2007 during the
facility's revamping.
The target was to understand the performance of the
run off to decelerate riders in case of acrash/fall, and
to look for the improvements which could be
achieved with different materials.
The research team was joined by Dario Vangi and
Antonio Virga, from University of Florence.
12
Selected tests were to be done on three different
materials: road asphalt, gravel and lightweight
aggregate.
Due to budget restrictions, we decided to run tests
throwing a dummy in a proper position launched on
a sled propelled by aeronautic bungees.
The sled
A specially designed launch device was used during
the tests (fig. 4) which aimed to accelerate the
dummy to the desired velocity and launch it on the
surface under test.
Figure 3 – The dummy collocated on the sled. ©
jarnozaffelli.it
More specifically, the launching device is composed
by a steel double rail on which a lightweight cart
made of aluminum and plexiglas can move with its 4
wheels engaged in the rails. The cart is pulled by an
elastic cord, connected by means of a steel rope. The
elastic cord was made of an array of 3 or 4 cords
(Sandow Technic model Avia, each with 30 mm
diameter, length about 28 m, maximum load 400
daN) joint in parallel. At the opposite end the cord
was tied to a plinth fixed to the ground. The elastic
system was put in tension by means of an electric
winch (Ramsey model Rep 8000) connected to the
rear part of the cart with a steel rope. Before the test
(see fig. 5) the dummy was simply laid down on the
cart, with its head forward; a restraint device allows
the dummy to leave the cart only from the front side.
By operating the electric winch the elastic cords
were pulled to the desired tension, then the cart was
released by means of a hook operated from a proper
distance. The cart accelerated till the end of the rail
was reached, where a restraint device stopped it
suddenly, leaving the dummy free to continue its
trajectory on the testing surface.
 Experimental analysis of post-accident motion of motorcyclists
Figure 4 - Layout of the launch device.
The dummy
For the experimental activity an anthropometric
dummy was used (1.8 m tall and with 80 kg mass)
dressed with a complete motorcyclist apparel,
including helmet, racing suit, gloves and boots. The
position of the dummy (fig. 2) was especially
selected to be adequate to represent real world cases.
In the worst case scenario of fall the rider will slide
on the surfaces, not rolling, as a dead weight, due to
the loss of consciousness. Our dummy excellently
replicated this scenario.
Figure 5 - The dummy put on the cart before the
beginning of the testing procedure.
The instruments
During the tests the dummy launch speed (i. e. the
velocity at which the dummy left the cart and entered
the testing bed) was measured by means of a laser
photoelectric system. During the sliding trajectory of
the dummy, its velocity was measured using a radar
gun (Stalker Pro model Ats) and a 10Hz Gps data
logger (Racelogic model Video Vbox Lite).
Alongside, a high speed video camera was used
(Fastec Inline, 250 fps), being positioned laterally or
above in respect of the trajectory. Having positioned
several markers at known distance along the path, in
processing the frames of the video it was possible to
obtain, as a function of time, the position and
velocity of the dummy. Then, after each test, the
total distance travelled by the dummy was measured
by means of a laser telemeter (Leica Disto D8). As
said, in several launches we also tested a GPS data
acquisition system from Racelogic, at 10 Hz, battery
powered (Fig. 6). Accuracy issues were found in
some cases, so others data sources were considered
in those cases.
Figure 6 – The dummy with the GPS antenna collocated
inside the helmet visor.
Results
Three tests were carried out at Fornovo di Taro,
making the dummy slide on expanded clay; the
following measurements were performed:
- initial velocity of the dummy by means of a
radar gun;
- total distance travelled by the dummy using
laser telemetry;
- dummy’s position as a function of time by
means of a high speed video camera,
positioned on a metallic gangway above the
testing bed. By analysing the recordings it
was possible to obtain velocity as a function
of distance, as, for instance, in fig. 3.
In table 1 summary data of the test are shown; by
successive elaboration coefficients of deceleration
were obtained (fig. 11). More specifically, by
applying energy conservation, the formula for a
sliding to stop test, and assuming a constant
deceleration, can be written as:
Vi 2
f 
2 g  X
where Vi is the initial velocity, X the distance
travelled, and g acceleration due to gravity. The
coefficient f depends on friction but also on bounces
and impacts that occurred during the motion of the
13
 Dario Vangi, Antonio Virga, Jarno Zaffelli

dummy. As can be observed in figure 8, deceleration Test Site 2 - Misano World Circuit "Marco
rate is not constant, being also influenced by bounces Simoncelli"
and impacts. Location - GPS 43.962913,12.68392
Material 1 - Round washed gravel 8-16
Material 2 - Dry Asphalt
Test Site 1 - Laterlite Leca® production site
Location - GPS 44.682376,10.068079 Six tests were carried out at Misano World Circuit, 5
Material 1 - Laterlite Leca® Stopsafe 8-20 LWA on gravel and 1 on asphalt and gravel; the following
measurements were performed:
- initial velocity of the dummy by means of a
radar gun;
- initial velocity of the dummy by means of a
laser photoelectric system;
- dummy’s position as a function of time by
means of a high speed video camera,
positioned laterally. By analysing the video
recordings, velocity as a function of distance
was obtained, as shown in fig. 10.

Figure 7 – The trajectory of the dummy, as recorded by In test number 6 the trajectory of the dummy
GPS system, for a test on expanded clay. consisted in a gravel part (12 m) followed by asphalt.

140,0 In table 2 summary data of the tests are shown;

120,0
coefficients of deceleration are shown in figure 11.
Velocity (km/h)

100,0
Table 2 - Summary data of the tests at Misano World
80,0
Circuit (gravel: tests 1 to 5; asphalt and gravel: test
60,0 number 6).
40,0 max velocity (km/h) Distanc
20,0 e
0,0 (radar) (laser) (video (Gps) (m)
0,0 5,0 10,0 15,0 Test camera)
Distance (m) 1 56,3 55,7 52,4 n.a. n.a.
Figure 8 – Velocity of the dummy as a function of 2 65,5 64,6 62,9 n.a. n.a.
distance (test 2 on expanded clay). 3 57,6 54,7 n.a. 56,8 11,0
4 107,6 101,0 n.a. 99,3 44,2
5 95,0 96,0 n.a. n.a. 36,0
6 105,9 n.a. n.a. n.a. 60,1

120,0

100,0
Velocity (km/h)

80,0

60,0

40,0

20,0

0,0
0,0 5,0 10,0 15,0 20,0
Figure 9 – The dummy sliding during a test on expanded Distance (m)
clay. © jarnozaffelli.it
Figure 10 - Velocity of the dummy as a function of
Table 1 – Summary data of the tests at Fornovo di Taro (expanded clay). distance (test 4 on gravel).
initial velocity (km/h) Distance
Test (radar) (video camera) (m) The results
1 88,9 95,4 30,0 Even though they are not exhaustive, the results were
2 112,8 121,2 45,0 extremely interesting, as a base for further analysis.
3 112,1 n.a. 47,0
14

It was demonstrated that, at the same speed, the
rider's stopping performance on dry tarmac is less
than gravel, which is less than lightweight aggrega-
tes.
expanded clay
1,40
asphalt
1,30 gravel
Deceleration coefficient
1,20
1,10
1,00
0,90
0,80
0,70
0,60
80 85 90
Figure 11 – Summary of deceleration coefficients as a
function of velocity for various surfaces.
DroCAS™ crash case history - Heber
Pedrosa - Imola
DroCAS™ (Dromo Circuit Analysis System), is
purpose built simulation software that relies on the
result of this and following research, to establish the
dimension of run off areas in racetracks. One of the
highest speed crashed recorded happened in 2012 in
Autodromo Internazionale Enzo e Dino Ferrari,
Imola, during the World Superstock Championship.
Heber Pedrosa lost the front of his bike during the
race at more than 200 Kph. The rider stopping
prediction made by DroCAS™, compared to the
actual stopping distance of Mr. Pedrosa, was
extremely accurate, demonstrating the steps taken in
the professional rider's safety thanks to the research
presented in this paper.
Figure 12 – Heber Pedrosa accident at Imola in 2012.
©jarnozaffelli.it
The safety calculation in such a high speed crash has
been demonstrated to be with a margin of 4.5% in
excess, due to the need to keep an additional safety.
The rider fell and slid until stopping, 25 meters
before the existing protection (TecPro barrier custom
built and positioned for a motorcycle with higher
performance, MotoGP class), with a deviation of
Experimental analysis of post-accident motion of motorcyclists
some degrees from his bike, without any physical
consequences.
Conclusion and Next steps
Further analysis in the last few years has produced
consistent results. By analysing more than 1000
accidents and continuosly gathering data from
hundreds of those crashes, a robust coefficient data-
base for many surfaces over 240 Kph has been
produced.
The first application of these results in everyday road
accidents has been to start to develop different
friction materials to increase the protective clothing's
95 100 105 110 115 120 125 braking performances, using this experience as a re-
Velocity (km/h) ference.
However, much has to be explored yet, and our re-
search efforts are focused on filling the gaps in the
knowledge of post-accident motion of motorcyclists.
References
[1] Medwell C. J., McCarthy J. R., Shanahan M. T., Motor-
cycle Slide to Stop Tests, (1997) SAE Technical Paper
970963.
[2] Searle J. A., Searle A., The trajectories of Pedestrians,
Motorcycles, Motorcyclists, etc., Following a Road Acci-
dent, (1983) Sae Technical Paper 831622.
[3] Corcelle, F. and Wright, P., "Circuit and Safety Analysis
System (CSAS)" SAE Technical Paper 2000-01-3573,
2000, doi:10.4271/2000-01-3573.
[4] Wright, P., "The Analysis of Accident Data Recorder
(ADR) Data in Formula 1" SAE Technical Paper 2000-01-
3551, 2000, do-i:10.4271/2000-01-3551.
[5] Patent Itrm20050278 (A1), 2005-09-01, Inventors: Cos-
salter Vittore, Roberto Lot, Ambrogi Andrea; Bellati Ales-
sandro, Metodo Di Attivazione Per Airbag Da Motociclisti,
2005.
[6] Patent Ep2079613, 2009-07-22, Inventors: Cossalter
Vittore, Roberto Lot, Ambrogi Andrea; Bellati Alessandro,
Method And Device For The Prediction Of A Fall Of A
Person From A Vehicle Or The Like, Applicant: Dainese
Spa [It] Ec: B60r21/0132, Ipc: B60r21/0132; B60r21/00;
B60r21/0132, 2009.
[7] V. Cossalter, A. Aguggiaro, D. Debus, A. Bellati, A.
Ambrogi, Real Cases Motorcycle And Rider Race Data In-
vestigation: Fall Behavior Analysis, 20th Enhanced Safety
Of Vehicles Conference, Innovations For Safety Opportuni-
ties And Challenges Lyon, France, June 2007.
15
 Dario Vangi, Antonio Virga, Jarno Zaffelli

[8] A. Bellati, V. Cossalter, R. Lot, A. Ambrogi, Prelimi-

nary Investigation On The Dynamics Of Motorcycle Fall
Behavior: Influence Of A Simple Airbag Jacket System On
Rider Safety, 6th International Motorcycle Conference, Ifz
Institute For Motorcycle Safety, Cologne, 9 -10 October,
2006.
[9] Vittore Cossalter, Alessandro Bellati, Vittorio Cafaggi,
Exploratory Study Of The Dynamic Behaviour Of Motor-
cycle-Rider During Incipient Fall Events Paper Number 05-
0266, 19th International Technical Conference On The En-
hanced Safety Of Vehicles Conference (Esv) In Washing-
ton, D.C. 6-9 June 2005.

[10] A. Bellati, V. Cossalter, R. Lot, A. Ambrogi. 2006.

“Preliminary investigation on the dynamics of motorcycle
fall behaviour: influence of a simple airbag jacket system
on rider safety”, Proceeding of 6th International Motorcycle
Conference, IFZ Institute for Motorcycle Safety, Cologne, 9
-10 October 2006.

[11] Cossalter, V., R. Berritta. 1999. “Analysis of the dy-

namic behaviour of motorcycles in dangerous manoeuvres
using the multi-body code MSC Working Model.” Proceed-
ing of the 1999 Hightech Engine and Cars Conference
(Modena, Italy).

[12] Cossalter, V. 2006. “ Motorcycle Dynamics”, ISBN:

978-1-4303-0861-4, Publisher: Lulu.com.

[13] TRL-FIM Dummy launch Test 19th November 2002.

[14] Woods, R.I. (1996a), Specification of motorcyclists'

protective clothing designed to reduce road surface impact
injuries . Performance of Protective Clothing: Fifth Volume
ASTM STP 1237 , James S. Johnson and S.Z. Mansdorf,
Eds., American Society for Testing and Materials, Philadel-
phia, 1996. pp 3-22.

[15] Woods, R.I. (1996b), Testing of protective clothing for

motorcyclists: validation of laboratory tests by accident

[16] studies , Performance of Protective Clothing: Fifth

Volume ASTM STP 1237, James S. Johnson & S.Z.
Mansdorf, Eds., American Society for Testing and Materi-
als, Philadelphia, pp 43-56.

Contact
Jarno Zaffelli
via 4 Novembre 12
42121, Reggio Emilia, Italy
e-mails:
jarno@studiodromo.it
tel: +39 3356699805

16
© 2013 by EVU
EVU 2013 – 02
Motorcycles Slide Tests Analysis
Dario Vangi
FilippoBegani
Carlo Cialdai
Antonio Virga
University of Florence

Abstract
This paper presents analysis of the phase of the fall-over to side and the sliding of two-wheeled vehicles, which occurs
typically in road accidents involving two-wheeled vehicles.
Knowing the coefficient of deceleration in the phase of sliding of two-wheeled vehicles allows calculating the kinetic
energy dissipated and the speed of the vehicle just before the fall to the ground. This parameter is very important for the
analysis and reconstruction of accidents.
The work presented in this paper has been developed in two experimental test sessions on vehicles of the fully faired,
scooter type, widely used in urban areas. In the first session, sliding tests were carried out, with speed range 10 - 50
km/h, on three different types of road surface. The analysis of the evidence allowed identification of the dissipative main
phases of motion of the vehicle (ground impact and rebounds, stabilised swiping) and the study of some factors that
affect the phenomenon. The coefficient of average deceleration during the entire sliding on the ground of the vehicle was
strongly related to the type of vehicle, faired or naked, and the type of road surface, while the mass of the vehicle and the
speed of fall seem to have less influence. The second session of tests has been conducted with drag tests, in order to
verify the possible use of an expeditious method for on-site evaluation of the coefficient of deceleration of the motor
vehicle and road surface, even during the inspection of the road accident by the authorities. In these tests, the vehicle,
previously laid on a side, was dragged for a few metres at a constant speed of about 20 km/h, while measuring the drag
force. The comparison of the results obtained in these tests, with those obtained in the sliding tests has highlighted the
limits of applicability of the expeditious method.

1. Introduction and side sliding on the ground and quantify

The analysis of car accidents is the starting point to
investigate and mitigate crash phenomena and their
consequence. Over recent years, the science
community has come up with models and
correlations to improve the accident reconstruction
[1-9] and comprehension, in terms of kinematics,
human behaviour and risk factors inherent in
infrastructure and in interactions with the driver and
the vehicle. Despite the diversity between the various
types of accident and diversity of approach used for
their analysis, in the majority of road accidents it is
possible to identify some typical phases that can be
addressed separately in the reconstruction: pre-
impact phase, impact phase and post-impact phase.
In most accidents involving two-wheeled vehicles,
they fall to the ground and subsequently slide on the
pavement. This phase, which can happen either
before or after the collision, is characterised by
motion whose description is not always easy to
model and it is not always possible to reach an easy
evaluation of the kinetic energy dissipated.
In relation to two-wheeled vehicle behaviour in the
post-impact phase, there exists in literature some
studies which present experiments designed to
evaluate the behaviour of motorcycles during the fall
deceleration forces [10-16].
When a motor vehicle falls to the ground, typically
three distinct stages of its motion can be identified
[10]: 1) loss of control, 2) ground impact and
rebounds, 3) stabilised swiping.
In the first stage, it is not possible to evaluate
objectively an average deceleration and the
subsequent dissipation of kinetic energy of the
vehicle. The second stage is between the first contact
to the ground of the motor vehicle and the
stabilisation of the sliding, and may be associated
with the phase of impact in which the motorcycle
undergoes rebounds and decelerates more rapidly
compared to subsequent portions of swiping.
In some research [11] the deceleration in the first
instants of the slipping (the phase of impact) has
been been measured to be about 0.8 ÷ 1.5 g. In the
phase of stabilised sliding the deceleration decreases,
reaching values of about 0.35 g.
In [12] is observed that the average deceleration is
greater during the phase of initial contact due to the
effect of incision into the road surfaces that occurs
due to protruding parts of the vehicle. In the impact
on the ground a contact force greater than the weight
of the vehicle is generated, and therefore a greater
deceleration. The initial higher deceleration phase
has an average duration of approximately 0.5

17
 Dario Vangi, Filippo Begani, Carlo Cialdai, Antonio Virga
seconds, so it is significant for short marks and
negligible for long ones.
In cases of initial speed lower than 10 km/h is often
impossible to assess the deceleration and the
equivalent friction coefficient since the vehicle does
not leave a defined and measurable trace on road
surface, but practically stops at the position where it
impacts on the ground: the only source of energy
dissipation is due to the collision with the ground,
not to the sliding.
In such events, in literature [11] it is suggested that a
friction coefficient between 0.75 and 1 is adopted
when a measurable sliding trace is identifiable. If no
traces on road surface are detected, it is possible to
state, given the results of the experiments conducted,
that the driving speed of the vehicle is definitely less
than 10 km/h.
The third phase of the phenomenon, the stabilised
sliding, is between the time when the stabilisation of
the sliding begins, with no more bumps and
significant rebounds, and the motor vehicle coming
to a stop. In this phase, rotations of the vehicle may
be generated, affecting the length of the marks. The
analysis of the films of the experimental tests with
such rotations indicates that they do not occur in
precise points of the sliding path. In some of the tests
performed, the rotation occurs immediately after the
impact, while in others, shortly before the vehicle
stops. In the first case, it is almost always the initial
configuration of the motorcycle that generates the
rotation’s phenomenon, since, if at the moment of
impact on the ground, it is not oriented in the
original travelling direction, or if it simply impacts
with steered handlebars, rotation may be produced.
In cases when rotations are generated in the final
phase of the sliding, the causes are to be found in the
profile of the road surface or in the characteristics of
the vehicle. Due to transfer of inertial load, the
contacts between the side of the vehicle and the
asphalt may vary over time during the sliding phase,
generating rotations which, in turn, trigger new
transfers of inertial load. The result is that the vehicle
may show rotations towards different directions
during thesame phase of sliding.
In any case, tests and literature data indicate that they
are not so frequent and generally small, i.e., the
angle that is formed between the correct position of
travel and the final position rarely exceeds 90°.
The above-mentioned literature studies differ in the
way the motor vehicle is thrown to perform the
experimental tests. In [11], for example, the authors
used equipment capable of supporting the front
wheel of the vehicle at a certain distance above the
ground level, while the desired speed is reached. In
[12] authors used two different throwing systems.
The first uses equipment similar to that used in [11],
18
and provides a connection of the cables to the
handlebars in order to manage the time of release
and force the vehicle to crash on one side rather than
on the other. The second is based on the realisation
of a bent towing platform, on which the vehicle is
laid on side position. The motorcycle is held by
ropes. Once it reaches the desired speed, it is
released. In [13] a particular platform mounted on
the side of a truck that holds the vehicle to a very
low height from the ground was used: the distance
between the tyres and road surface was about 90
mm. The motorcycle was kept in upright position by
an operator positioned on the body of the truck itself,
by means of support cables.
In the cited articles, speeds are included in the range
of 10-95 km/h and the average measured
deceleration coefficient varies between 0.25 and
0.79.
In this work, the phase of ground sliding of two-
wheeled vehicles is analysed taking into account the
scooter type. Vehicles used in the urban environment
that have not been tested in the literature are
compared with motorcycles or mopeds. The scooters
generally have the characteristic of being completely
covered with plastic fairings, have low height and
gravity centre and wheels are of smaller size
compared to motorcycles. The scooter testing was
carried out in a range of typical urban speeds, up to
50 km/h. In the study, the interaction between rider
and vehicle, typical of the first stage of loss of
control of the vehicle, is ignored. In addition, the
study neglected the influence of wear of the vehicle
on the deceleration coefficient, due to the repetition
of more tests on the same vehicle. In [12] a clear
trend between deceleration and state of wear relating
to natural deterioration due to the succession of
experimental tests carried out on the vehicles
themselves was not found.
The analyses have been conducted using two types
of tests, the so-called "sliding" and "drag". In the
slide tests, the way to launch vehicles has been
designed so as to reproduce, as realistically as
possible, a real motorcycle accident that results in
the fall and subsequent sliding of the vehicle on the
road.
Marks, abrasions and cuts, produced by various
elements of the motor vehicle are usually generated
on the road surface. Analysing and measuring these
tracks, for the macroscopic description of the motion
and the assessment of the initial velocity of the
vehicle, generated a suitable model of uniformly
decelerated motion which identified a deceleration
coefficient that takes in account the effects of
rebound and subsequent roto-translations generated
in the first temporal phase of the fall.
 Motorcycles Slide Tests Analysis

Dragging tests were aimed at verifying the cm3 has also been tested. In total 10 different models
applicability of a simplified and expeditious method of vehicles were used, for a total of 72 sliding tests:
for in situ evaluation, even during the survey of the 33 on naked vehicles and 39 on faired vehicles.
road accident by the authorities, of the deceleration
coefficient between motor vehicle and road surface. Table 2 - Vehicles types used for testing.
Make Model Fairing Weight (kg)
2. Sliding test Piaggio Si no 51,0
Garelli Noi no 48,0
2.1 Materials and methods Aprilia ST no 106,0
MBK Target yes 69,5
The tests were carried out reproducing realistically MBK Doodo yes 113,5
the phenomenon of the impact on the ground and Piaggio Vespa yes 86,5
rebound that occurs during an accident. Piaggio Zip yes 69,5
The motorcycles were towed to the desired throwing Honda Sky yes 71,0
speed with a truck that has been connected to the Kymco DJ yes 73,5
handlebar with a quick release attaching system Gilera Storm yes 81,0
(figure 1).
Dragging is done by lifting the front wheel a few
centimetres of so as to allow only the rolling of the
rear wheel. When the desired speed is reached, the
vehicle is released, it loses his balance and falls to
the ground producing sliding on the side.
The tests were carried out on three different types of
road surfaces (Table 1) and at different speed values,
included in the range10-50 km/h.
Figure 2 – Faired and naked motorcycle.

The tests, on each vehicle, were conducted starting

from the lowest speeds, using alternate making sides
of the motorcycle. The tests were repeated at
gradually higher speeds.
The measure of the throwing speed of the vehicle
was conducted with a tachymeter wheel connected to
the truck and equipped with an encoder Peiseler that
ensures an accuracy of 0.1 km/h.
The tests were recorded with high-speed camera
(250 frames/s) in b/w.
Figure 1- Method used to launch the motorcycle. The commonly used method to estimate the
deceleration as a function of the sliding distance and
initial speed of the vehicle is based on the
Table 1 - Different types of road surface on which the
tests were carried out.
application of energy theorem [12, 14, 15].
In some of the mentioned articles the application is
Road surface ID aimed at the research on the friction of the tyres, but
the authors confirm the possibility of extending the
New asphalt A method to traces of side sliding of motor vehicles,
Worn asphalt B sometimes accompanied by an adjustment factor.
Concrete C The work done by sliding friction forces has been
considered the only source of variation of the kinetic
The experimental work was conducted on two energy, ignoring the aerodynamic forces. For the
different types of vehicles: faired motorcycles, like initial kinetic energy only the contribution of the
scooters, and naked motorcycles (see Figure 2 and translational motion of the initial centre of gravity
Table 2). has been considered, ignoring the motion associated
In addition, for purposes of comparison with the data with the wheels rotation.
available in the literature, a naked motorcycle of 125 Since the final state of the vehicle is of rest, we have:

19
 Dario Vangi, Filippo Begani, Carlo Cialdai, Antonio Virga

1
mv 2  fmgL 1
2

where:
f = average friction coefficient between vehicle and
asphalt along the sliding path;
L = sliding length identifiable on the road surface
between the first sign of sliding detected and the
centre of gravity of the vehicle in stop position;
m = vehicle mass; Figure 3 - Stop position of the vehicle, outside of the
g = gravitational acceleration; asphalt pavement.
v = throwing speed of the vehicle. Table 3 – Sliding test of naked motorcycle. In red,
Obtaining, for each experimental test, the length of discarded tests.
the traces on the ground (L) as a function of the Speed Slide
speed (v), it follows that L=f(v2). Introducing the ID Medel
(m/s) distance (m)
Fairing f
values of v and L on a Cartesian plane respectively A Garelli 14,6 15,85 no 0,69
as abscissa and ordinate, we expect a parabolic curve A Si 10,7 13,40 no 0,43
passing through the origin of the axes. A Noi 5,3 4,40 no 0,32
A Noi 3,8 1,90 no 0,38
To analyse data the above-mentioned diagram (slide A Si 5,7 3,40 no 0,49
distance/speed of throw) is used, making a best fit of A Si 7,5 6,86 no 0,42
the data with a parabola passing through the origin. A Si 10,4 13,52 no 0,41
The deceleration of the vehicle during the test is not A Si 4,8 3,30 no 0,36
uniform, and the above estimated average friction A Noi 8,1 7,20 no 0,46
coefficient can be considered as an equivalent A Noi 9,5 9,20 no 0,50
A Si 8,0 8,50 no 0,38
friction coefficient to be used with a uniformly A ST 7,9 7,00 no 0,46
decelerated motion model that allows for the fall to A ST 8,0 6,10 no 0,53
the ground on the side, of rebounds and of the A ST 9,4 7,80 no 0,58
subsequent sliding on the ground with roto- B Noi 3,7 1,00 no 0,70
translational motion of the vehicle. B Noi 10,0 11,00 no 0,46
B Noi 14,0 16,50 no 0,61
B Si 3,8 1,60 no 0,45
2.2 Results and discussion B Si 9,3 11,35 no 0,38
B Si 15,0 25,00 no 0,46
Initially, we proceeded by analysing the individual B ST 3,9 1,50 no 0,52
tests and, with the help of video and photographic B ST 10,3 6,80 no 0,79
B ST 6,9 4,20 no 0,59
reports, a selection of the tests was made. B ST 10,0 7,70 no 0,66
In the data analysis some of the tests were not B ST 9,7 6,80 no 0,71
considered, that is tests in which the motor vehicle B Si 6,9 4,90 no 0,50
held very different configurations from those of B Si 13,9 16,80 no 0,59
normal driving (vehicle already falling, vehicle not C Noi 7,0 3,80 no 0,66
aligned to the longitudinal development of the road, C Noi 10,0 7,40 no 0,69
C SI 6,0 3,65 no 0,50
etc..), nor the test in which, during the sliding, the
C Noi 6,0 3,80 no 0,48
vehicle skidded off the road (figure 3). C Noi 10,3 8,90 no 0,60
Tables 3 and 4 shows the data for all the tests, C SI 10,0 4,80 no 1,06
highlighting the discarded ones.

Table 4 - Sliding test of faired motorcycle. In red,

discarded tests.
Speed Slide
ID Model Fairing f
(m/s) distance (m)
A Target 9,0 11,00 si 0,38
A Zip 14,2 31,30 si 0,33
A Vespa 11,4 19,60 si 0,34
A DJ 7,1 7,80 si 0,33
A Target 4,6 2,10 si 0,52
A Zip 4,6 2,45 si 0,44
A Vespa 4,7 1,60 si 0,70

20
 Motorcycles Slide Tests Analysis

A Target 9,1 7,80 si 0,54

A Storm 8,3 10,70 si 0,33
A Storm 9,7 12,10 si 0,40
A Zip 8,8 9,70 si 0,40
A Zip 8,4 10,10 si 0,36
A Zip 8,9 12,10 si 0,34
B Doodo 10,0 11,50 si 0,44
B Doodo 15,0 20,00 si 0,57
B Target 4,0 2,10 si 0,39
B Target 9,5 8,75 si 0,53
B Target 14,5 30,40 si 0,35
B DJ 4,6 3,30 si 0,33
B DJ 9,8 9,80 si 0,49
B DJ 15,0 26,60 si 0,43
B Sky 4,1 1,20 si 0,71 Figure 4 - Comparison between faired and naked vehicles
B Sky 10,5 14,50 si 0,39 on the pavement B.
B Sky 15,0 30,20 si 0,38
B Sky 6,9 6,80 si 0,36
B Sky 9,7 12,20 si 0,39
Figure 5 shows the trends for the two vehicles in the
B Zip 13,9 15,30 si 0,64 case of worn asphalt. It is observed that on the
B Zip 13,9 30,70 si 0,32 pavement with worn asphalt, the two curves start to
B Zip 7,2 6,50 si 0,41 separate clearly for higher speed values, around 30
B Zip 6,9 6,40 si 0,38 km/h.
C Doodo 3,2 1,40 si 0,37
C Sky 4,0 2,60 si 0,31
C Sky 7,0 4,70 si 0,53
C Doodo 7,0 3,90 si 0,64
C Doodo 8,5 6,85 si 0,54
C Sky 9,0 6,50 si 0,64
C Target 9,0 7,40 si 0,56
C Doodo 6,5 3,80 si 0,57
C Target 10,3 5,30 si 1,01

The tests, carried out on different types of vehicle

and road surface, showed different values of the
coefficients of deceleration. In the following section
the data are analysed in order to highlight the
influence of individual factors on the results. Figure 5 - Comparison between faired and naked vehicles
on the pavement A.
2.3 Influence of vehicle type
The comparison between the results obtained for
On the surfaces with new asphalt a different naked and faired vehicles on a concrete road surface,
behaviour for the two categories of vehicles: faired show a behaviour largely similar for the two types of
and naked, can be noticed. vehicle, with overlapping parabolic trend, as can be
The curves slide distance/speed of throw (figure 4) seen from figure 6.
for the two types of vehicles, begin to separate at the
speed of about 20 km/h.
Below this value the difference is less evident: at low
speed different characteristics of the two different
types of vehicle have less influence than the effect of
the ground impact phase. This phase can be
considered to have a similar influence on the two
categories of vehicles.

Figure 6 - Comparison between faired and naked vehicles

on the pavement C.

21
 Dario Vangi, Filippo Begani, Carlo Cialdai, Antonio Virga

The reason for the difference in behaviour of the two

categories of vehicles lies in the characteristics of the
pavement surface. In asphalt pavements, generally,
the macrotexture is high and roughness facilitates the
engagement of the projecting parts of vehicles:
tripod; pedal brake; and muffler, both for faired and
naked vehicles.
Indeed the concrete surface shows a very low
macrotexture and high surface hardness. These
features minimise the effects of etching and
engagement of the projecting parts of the vehicle,
and make the difference in behaviour of the vehicles
tested negligible. Figure 7 – Relation between slide distance and speed in
the three road surfaces, for faired motorcycles.
The worn asphalt, characterised by a macrotexture
lying between new asphalt and concrete, also shows
an intermediate behaviour.
In general terms we can say that in pavements where
the macrotexture is lower, the curves tend
overlapped more: the influence of the projecting
parts decreases, and then the type of vehicle has low
influence on the phenomenon.
In analysing each road surface and each vehicle type
tested (faired and naked) a minimum correlation
coefficient R2 of 0.89 is observed with an estimated
standard deviation of the error (s) of always less than
2.2 m.
By calculating the coefficient of deceleration,
Figure 8 - Relation between slide distance and speed in
independently of the surface type, for the 33 tests
the three road surfaces, for naked motorcycles.
carried out on naked vehicles and for the 39 tests
carried out on faired vehicles it is found that, on
average, a higher value for the category of motor The average coefficient of deceleration, obtained
vehicles which do not present any kind of fairing from the equation (1), is reported in Table 5 for each
have a higher value: f = 0.52 for naked vehicles, than road surfaces and is higher for naked vehicles.
faired vehicles f = 0.44.
Table 5 - Average deceleration coefficient and standard
2.4 Influence of road surface type deviation, calculated for each category of vehicles and on
each road tested surface.
Figures 7 and 8 show, for various road surfaces, f St. f St.
curves of best approximation for slide distance/speed ID
faired dev. naked dev.
of throw for the two categories of motor vehicles
A 0,36 0,12 0,48 0,10
tested. For faired vehicles it is possible to see that the
B 0,39 0,10 0,55 0,12
curve relative to the road surface C is the most
C 0,55 0,11 0,60 0,09
flattened, while the curves relating to the road
surfaces A and B generally show a coefficient of
deceleration lower than concrete. Worn asphalt (A)
proved to be the road surface with the lowest 2.5 Influence of speed and weight
coefficient.
For naked vehicles (figure 9), in the speed range From the test a clear correlation between speed of
between 10 and 40 km/h, the curves have a similar throw of the motor vehicle and coefficient of
trend. The curve of the worn asphalt pavement (A) is deceleration has not been found. Figure 9 shows the
moved upwards, showing a smaller coefficient of value of the coefficient of deceleration (f) as a
deceleration, while it is not possible to notice a clear function of speed for naked motorcycles tested on
difference between concrete (C) and new asphalt (B). pavement B. The trend is substantially in agreement
with the parabolic trend shown above between speed
and slide distance. The dispersion of the data (0.30to
0.70 for faired vehicles, 0.35 to 0.70 for naked

22
 Motorcycles Slide Tests Analysis

vehicles) is due to the intrinsic non-repeatability of

the phenomenon.
Contrary to what reported in [11], in which the
authors show that for short slide (low speed), the
friction coefficient is higher than the one detected for
the same type of motor vehicle for long sliding (in
the speed range between 44 and 62 km/h), from the
analysis of the tests carried out, the same conclusions
cannot be drawn.
Data dispersion generated by this phenomenon does
not allow identifying substantial differences in
behaviour and in the coefficients of deceleration.
Figure 10 - Deceleration coefficient variation as a function
of vehicle mass on the pavement C.

2.6 DOE analysis

A more detailed analysis of the influence of various

factors on the coefficient of deceleration was done
by applying the Design Of Experiment (DOE), with
the software "Minitab" (www.minitab.com). A full
factorial plan composed of 4 factors, each with two
levels of reference has been used (table 6).

Figure 9 – Correlation between speed and deceleration

coefficient. Table 6 - Factors and levels set in the analysis Design
Of Experiment (DOE).
To evaluate the influence of the vehicles’ mass on Worn
the coefficient of deceleration, data derived from the Road surfaces
New
experiments carried out and those found in literature Faired
were analysed [11, 13]. Tests were compared which Vehicle type
Naked
were carried out on different motorcycles but Light (<80 kg)
characterised by: Weight
Heavy (>80 kg)
Low (< 9 m/s)
 same type; Speed
High (>9 m/s)
 same throwing method;
 same asphalt; In the analysis the concrete road surface was not
 same speed; taken into consideration, since it is barely used in
 different mass. European roads. The results of factorial analysis
reflect the considerations previously made. Figure 10
It is possible to observe the variation of the shows the main effects of the various factors, taken
coefficient of deceleration as a function only of the separately, plotted on the same scale. It is observed
variation of the mass of the vehicle. that the type of vehicle (faired or naked) is the factor
Data from technical literature and from the that most influences the phenomenon. Mass and
experimental tests carried out by the authors, show speed are negligible in relation to the coefficient of
that the mass of motorcycles has a minimal influence deceleration. Their variability may be due to non-
on the coefficient of deceleration. In all cases the repeatability of the experimental tests carried out.
coefficient of deceleration, relating to various The effects of interaction between the factors
motorcycles sliding on the same asphalt, does not considered are negligible.
change, or varies slightly in a small range (Figure
10).

23
 Dario Vangi, Filippo Begani, Carlo Cialdai, Antonio Virga
Figure 11 - Main effects of individual factors on the
deceleration coefficient.
3. Drag test
3.1 Materials and methods
The sliding test described above, is difficult to
reproduce at the site of an accident in order to
estimate the coefficient of deceleration of the
vehicle. More practical and expeditious are the drag
tests that in [14] are used to estimate the speed of the
motor vehicle. The simplicity of execution of these
tests at the scene of the accident, led to the
verification, by means of experimental tests, of the
correspondence between the values of the average
deceleration coefficient resulting from the drag test,
and those resulting from the slide test.
The substantial difference between the two modes of
test is in the initial configuration: in the slide tests
the motorcycle is in running position and then
undergoes a bump on the ground before sliding on
the road surface, in the drag tests the side of the
vehicle is already in contact with the ground at the
beginning of towing. In the drag tests the vehicle
tested was positioned on one side with the front
wheel oriented in the running direction, and pulled
by a truck through a cable (figure 11). In series with
the tow cable has been inserted a load cell with a
resolution of 10 N.
As with the sliding test described in this paper, the
truck was equipped with tachymeter wheel.
24
Figure 12 - Method used for the execution of the
performed drag tests.
The models tested were two faired vehicles and two
naked, differentiating in some tests the side in
contact with the road surface in order to observe the
differences in coefficient of deceleration as a
function of the fairing and projecting parts. Six drag
tests were carried out on the road surface with new
asphalt.
The towing of the vehicle has been carried out
starting from zero initial speed until reaching the
velocity v = 20 km/h, truck kept a constant velocity
for few seconds, in order to acquire a value of force
approximately constant and not affected by inertial
effects. The tests were performed at a single velocity
value, in accordance with [14], in which dependence
between the coefficient of friction and speed is not
observed.
3.2 Results and discussion
The deceleration coefficient in the drag tests was
obtained by using the equation that relates the
vertical force, pressing the ground, with the
tangential force, which opposes the movement. By
knowing the vehicle mass and the force required for
the dragging it is possible to obtain the coefficient of
deceleration f.
The typical trend in the force (friction force and
inertia force), measured by the load cell, is shown in
Figure 12 together with the speed signal in function
of time.
 Motorcycles Slide Tests Analysis
Figure 13 - Friction force, inertia force and speed in
function of time. Curves related to the test on the vehicle
MBK Doodo: left side.
In the drag tests, the chance of causing irregular
rotation while sliding on the ground is prevented by
towing constraint.
In the two tests the coefficient of deceleration is
comparable when sliding traces are approximately
continuous. When the track, after the throw from the
running position, is irregular, characterised by
discontinuous portions and non-rectilinear trajectory,
the two types of tests provide results are not
comparable.
The coefficient of deceleration is also comparable in
the two types of tests in the cases in which the
effects of the impact on the ground is minimal, that
is the trace result of considerable length (greater than
10 meters). In the tests at low speeds, however, for
which the trace produced is short, the coefficients of
deceleration are substantially different in the two
types of tests.
Table 7 shows the comparison between the average
deceleration coefficients obtained in the two test
modes, for the two types of vehicle (faired, naked).
The coefficients of deceleration for the sliding tests
were calculated by averaging the tests with tracks
greater than 10 m, to minimise the influence of the
effect of impact.
Table 7 - Comparison between average deceleration
coefficient in drag and sliding tests; differences between
faired and naked vehicles.
f drag f slide
Faired 0,43 0,42
Naked 0,43 0,54
In the case of faired vehicles the coefficient of
deceleration is similar to that obtained with the
sliding tests (0.43 and 0.42), while in the case of
naked vehicles the differences are more evident. In
sliding tests the projecting parts on the sides of
naked vehicles work as a handhold on asphalt
surface, especially in the initial phase in which the
vehicle impacts the ground. For this reason a greater
average deceleration coefficient was obtained,
compared to dragging tests in which the vehicle has
a forced advancement that prevents hooking between
vehicle and road surface, which can slow down the
vehicle.
3.3 Applicability of the method
From the above results it emerges that the drag test
can be applied as an expeditious method to evaluate
the average coefficient of deceleration in a given
place where a road accident occurred, only in those
cases in which the crashed vehicle is faired and the
track is longer than 10 m and substantially uniform
and continuous.
For the practical applicability of the method, police
have to put the crashed vehicle on the interested side,
and tow it, using the emergency hook on the rear of
the service vehicle and tow it a few metres at a
constant speed of about 20 km/h.
The instrumentation required for the determination
of the coefficient of deceleration that the police must
have, consists in:
 a towing steel cable;
 a load cell with memorisation of the values
read, to be inserted in series to the cable.
4. Conclusions
This paper shows the results of experimental analysis
of the sliding on the ground of two-wheeled vehicles,
typical of accidents involving such vehicles. Fully
faired vehicle (such as scooters), in the range of
speeds typical of urban environments have been
investigated. These results complete results, in which
can be found similar studies, conducted on two-
wheeled vehicles (motorcycles or mopeds). A similar
test session on naked vehicles was also undertaken
and a comparison of the results was carried out.
In the study sliding and drag tests were conducted in
order to try and reproduce real accidents.
The study allowed the main phases of the motion of
the vehicle to the ground (ground impact and
rebounds, stabilized sliding) to be identified and for
the influence of some factors that influence the
phenomenon: type of asphalt; the type of vehicle
(faired or naked); fall speed; and mass of the vehicle,
to be studied.
The analysis of the experimental results
demonstrated a greater average deceleration
coefficient for naked vehicles. In particular, a mean
coefficient, averaged over all types of ground tested,
equal to 0.44 for faired vehicles and 0.52 for naked
25
 Dario Vangi, Filippo Begani, Carlo Cialdai, Antonio Virga

vehicles was calculated. The difference can be

explained by the characteristics of the two categories [6] Vangi D., Energy loss in vehicle to vehicle oblique
of tested vehicles: in naked vehicles the presence of impact,Int. Journal of Impact engineering vol. 36, issue 3,
protruding parts increases the friction with the 2009, pp. 512 – 521. ISSN 0734-743X.
ground, increasing the deceleration coefficient. For
[7] Campbell K.E., Energy basis for collision severity,
both types of tested vehicles (faired and naked Environmental Activities Staff, General Motors Corp., SAE
vehicles) the coefficient of deceleration is greater paper 740565, 1974.
with the bituminous pavement surfaced with new
asphalt than with worn asphalt. The deceleration [8] Backaits S. H., Accident reconstruction
onto the concrete road surface for both categories of technologies: Pedestrian & motorcycle in automotive
tested vehicles was higher than on bituminous collision, SAE PT-35, 1990.
pavements.
In the speed range of 10-50 km/h no significant [9] Limpert R., Motor vehicle accident reconstruction
correlation between the vehicle mass and the and cause analysis, Lexis Publishing, Charlottrsville (VA)
1999.
deceleration coefficient was found, nor between the
vehicle running speed and the deceleration [10] WoodD. P.; Alliot R.; Glynn C.; et al,
coefficient. Confidencelimits for motorcyclespeed slide distance,
In conclusion, the applicability of a possible Journal of Automobile Engineering, Vol. 222, No. 8.
expeditious method, to assess, even during the (2008), pp. 1349-1361.
inspection of road accident sites by the authorities, [11] McNallyB. F.; Bartlett W.,
the deceleration coefficient of the motor vehicle and MotorcycleSlidingCoefficient of FrictionTests, Accid.
the road surface has been verified. The method is Reconstr. J., 2007, 17(2), 47–49.
based on dragging the vehicle already on the ground
and provides results comparable to that of the tests [12] LambournR. F., The calculation of
motorcyclespeeds from slidingdistances, SAE Paper
that reproduce the phase of fall to the ground, only in
910125, 1991.
the case of faired vehicles (scooters), in which the
trace of sliding on the ground is substantially [13] MedwellC. J.; McCarthyJ. R.; M, T, Shanahan,
continuous and greater than about 10 m. Motorcycle Slide to Stop Tests, SAE Paper 970963, 1983.

[14] DayT. D.; SmithJ. R., FrictionFactors for

References MotorcyclesSliding on VariousSurfaces, SAE Paper
840250, 1984.
[1] Vangi D., Simplifiedmethod for
[15] NeptuneJ. A.; FlynnJ. E.; ChavezP. A.;
evaluatingenergyloss in vehiclecollisions, Accident
UnderwoodH. W., Speed from Skids: a modernapproach,
Analysis and Prevention, Volume 41, Issue 3, May 2009,
SAE Paper 950354, 1995.
Pages 633-641 (2009).
[16] Carter; Thomas J., et al, Measurement of
[2] Vangi D.; Begani F., Performance of
Motorcycle Slide Coefficients, SAE Paper 961017, 1996.
trianglemethod for evaluatingenergyloss in
vehiclecollisions, Journal of Automobile Engineering,
Volume 226 Issue 3 pag. 338-347 (2012). Contact
Dario Vangi, Professor, Antonio Virga, PhD, Carlo Cialdai,
[3] Vangi D.; Begani F., Energy loss in PhD, Filippo Begani, PhD
vehiclecollisions from permanentdeformation: an extension Via di S. Marta, 3
of the “Triangle Method”,Vehicle System Dynamic, 2013 50139, Florence, Italy
http://dx.doi.org/10.1080/00423114.2013.770538. dario.vangi@unifi.it, antonio.virga@unifi.it,
[4] Vangi D., A fuzzyapproach to carlo.cialdai@unifi.it,
reconstructingvehicle- filippo.begani@unifi.itdario.vangi@unifi.it,
pedestriancollisions,Vehiclesystemdynamics, vol. 47, iss. 9 antonio.virga@unifi.it, carlo.cialdai @unifi.it,
pag. 1115-1135 DOI: 10.1080/00423110802460689, 07 filippo.begani@unifi.it
January 2009. tel: ++ 39 055 4796505
[5] Vangi D.;Virga A., Evaluation of
vehiclebrakingdeceleration in
accidentreconstruction,Vehicle System Dynamics, Volume
45, Issue10 October 2007 , pages 895 – 910.

26
© 2013 by EVU
EVU 2013 - 03

Sliding decelerations of light weight scooter models (50cc

/ ± 60 - 100 kg)
Jan Meuwissen
Joos Spätjens
MVOA

Abstract
In this paper a research and test project is presented to measure and calculate an average sliding deceleration of a sliding
scooter. Seventy-one experimental sliding tests were perfromed with 50cc light weight (60 - 100 kg) scooter models on
three different asphalt road surfaces and one concrete road surface. The test results showed a significant difference in the
minimal value range compared with the results of average sliding decelerations tests done in the past with 50cc scooter
models. The test results also showed that a significant difference in average sliding decelerations can be expected due to
different shape of body and frame parts, bodypart materials and how the scooter frame is covered by bodyparts. For less
covered or naked scooter models there's a higher change that frame parts get stuck on the road surface during the sliding
movent which results in high peak decelerations.

Introduction When two wheeled vehicles are involved in a traffic

accident, they often slide along the road surface
during the pre, crash or post-crash phase. Under
these sliding motions the scooter decelerates
depending on the coefficient of friction between the
two wheeled vehicle and the road surface. The body
and frame parts of these two wheeled vehicles
scratch/scrape during the sliding motions, leaving
scratch marks on the road surface. The sliding
distance and the coefficient of friction between the
two wheeled vehicle and the road surface can be
used to derive a vehicle speed using the equation of
kinetic energy. However, in practice you only have
the length of these scratch marks. The coefficient of
friction between the two wheeled vehicle and the
road surface is estimated from the average sliding
decelerations resulting from previous experimental
Figure 1: Example 50cc modern scooter model.
Research and experimental tests conducted in
the past
The first research and experimental sliding tests with
two wheeled vehicles were done in the late 70’s and
the 80’s in [1], [2] and [3]. Tests were conducted
with heavyweight two wheeled motorcycles, mopeds
tests. These tests where mainly conducted with
heavyweight two wheeled motorcycles and
lightweight 50cc and 125cc scooters. In the
Netherlands the average scooter is a 50cc light
weight (60 - 100 kg) as shown in figure 1. To see if
there is a significant difference in average sliding
decelerations between the 50cc modern design
scooter models and the experimental tests conducted
in the past, in 2010 MVOA started a research and
test project. In 2010 and in 2011 (in cooperation with
IbB Engineering) 79 experimental sliding tests
mainly with 50cc modern design scooter models
were conducted. From the 79 tests 71 were
successfully completed and evaluated.
and bicycles. These tests were performed by turning
or tipping the motorcycle sideways from the back of
a car, with vehicle speeds in range of 20 to 120
km/h, on different road surfaces and conditions. In
the late 90’s and in the early 2000s there were some
experimental sliding tests conducted with
lightweight scooter models, 50- and 125cc, on dry
and wet asphalt. These tests were also performed by
turning or tipping the scooter sideways from the back
of a car [4] or trailer and by towing the scooter over
the road surface [5]. Using the force required to tow
the scooter over the road surface and the towing
distance, it was possible to derive an average sliding
deceleration. As a result of these latest experimental
tests with scooters, average sliding decelerations in
the range of ± 3.5 to 5 m/s2 were calculated.
Nowadays the body and frame part shapes and
materials of the scooter models are different
compared with the scooters models used in these
27
 Jan Meuwissen Joos Spätjens

latest tests. During our research and tests project we

found a recent published paper (August 2012) [6]
where the results of experimental sliding tests with
modern 125cc scooter were presented. The aim of
these experimental tests was to estimate the
coefficient of frictions on a dry and wet asphalt road
surface at different sliding speeds, as well as to see if
the coefficient of friction is dependent on the sliding
speed. The calculated friction coefficients on dry
asphalt were in the range of 0.36 to 0.53 (± 3.5 to 5
m/s2 ). The authors found that the friction coefficient
is independent of the sliding speed.

Methods and used equipment

The purpose was to create a test setup which is
comparable with the situation and conditions when a
scooter ends up in a sliding movement in practice. Figure 2: Sliding distance.
Therefore the scooters were accelerated in an upright
driving position to a speed of ± 60 km/h, released Acceleration, release and tipping process
and forced to tip onto the left or right side of the bo- The scooters were accelerated with a specially
dy and frame. As is the practice at a accident scene, designed trailer which was connected to a car. The
after the test the first contact with the road surface scooter is connected with the trailer by the steering
was found. By measuring between this point and wheel as shown in figure 4.
end position of the slide, as shown in figures 2, the
sliding distance was known.

To derive an average sliding deceleration the speed

of the scooter during the first contact with the road
surface has to be known. The speed of the scooter
was recorded with a GPS sensor from the PIC-DAQ
developed by DSD which was mounted on the tested
scooter. The GPS-sensor had a sampling rate of 5
Hz. For the tests conducted in 2011, a new type GPS
sensor was used with a sampling rate of 10 Hz. The
GPS x and y coordinates were also logged in the files
generated. These weren't used to determine the speed
of the scooter during the first contact with the road Figure 4: Special designed trailer.
surface because of the known and possible accuracy
faults of logged GPS x and y coordinates. The The steering wheel is lifted with two pins and the
sliding distance in the file was determined by front wheel of the scooter is fixed by an adjustable
integrating the logged GPS speed. Using the physical fork. The rear wheel can roll freely during the
sliding distance measured after the test, the speed of acceleration. With the trailer the scooters could be
the scooter during the first contact with the road accelerated in a stable condition up to speeds of 70
surface could be read out of the file. Using equation km/h and higher. To release the scooter the trailer is
1, the average sliding deceleration could be equipped with adjustable hinge pins and a special
calculated. designed release mechanism which can be activated
by a pneumatic cylinder as shown in figure 5. The
(1) pneumatic cylinder can be operated electronically by
a push button from the towing vehicle.

28
 Sliding decelerations of light weight scooter models (50cc / ± 60 - 100 kg)

Figure 7: Locked front wheel by a chain.

Figure 5: Release mechanism.

The tipping process of the scooter on a pre-defined

side, left or right, was realised by fixing the steering
wheel to the opposite way. To ensure that the scooter
tips shortly after the release process, the front brakes
were activated and/or the front wheel was locked by
a chain as shown in figure 6 and 7.
Before the test, pre-defined side parts of the body
and frame were painted so the first contact with the
road surface could be found easier as shown in figure
8 and 9.

During the tests the scooters were accelerated up to a Figure 8: Painted body and frame parts.
speed of ± 60 km/h. During the moment of release
and the first contact with the road surface, the
scooters decelerated by 5 – 15 km/h. This
deceleration was caused by the activated front brake
and/or the locked front wheel. The speed of the
scooters during the first contact with the road surface
was about 45 – 55 km/h. This is a typical maximum
speed for scooters in the Netherlands. Figure 10
shows an overview shown of the release, tipping and
sliding process during a test.

Figure 9: First contact mark.

Figure 6: Activated front brake.

29
 Jan Meuwissen Joos Spätjens

Used scooters models and test road surfaces

Most of the tested scooters had similar body part
shapes and body parts which were made of plastic.
Figure 11 illustrates an overview of the scooters
which were used for the sliding tests. The Piaggio
Vespa and Piaggio Velofax have different body part
shapes and the body parts are mainly made of metal.
The Piaggio Vespa for instance has a so called "retro
design" and is at the moment, a very popular model
in the Netherlands.

Figure 11: Overview of the tested scooters.

Figure 10: Release, tipping and sliding process during a
test. The tests were performed on 3 different dry asphalt
road surfaces and 1 dry concrete road surface. The
30
 Sliding decelerations of light weight scooter models (50cc / ± 60 - 100 kg)
different asphalt road surfaces and the concrete road
surface are shown in the figures 12 - 15.
Figure 12: Asphalt 1 Maasbracht (NL).
Figure 13: Asphalt 2 Maasbracht (NL).
Figure 14: Asphalt 3 Weeze (D).
Figure 15: Concrete Weeze (D).
Test results
All the successfully completed tests were filmed
with normal and high speed video cameras (Casio
Ex-F1). The damaged scooters and the scratch marks
on the road surface were extensively
photographically documented. The tested scooters
slid on both the left and the right side of the body
and frame. On each side a maximum of four tests
were conducted, after three or four sliding tests the
damage was no longer representative of conditions
in practice. The calculated average sliding
decelerations were evaluated in an excel-file and
illustrated in different diagrams.
Figure 16 shows an overview of the calculated
average sliding decelerations of the tests with the
"normal" scooter models on the asphalt road
surfaces. The normal scooters are those models
shown at the bottom of figure 16. Between the
different asphalt surfaces there wasn’t a noticeable
significant difference. In the tests with the normal
scooter models (n41) the calculated average sliding
deceleration was in the range of 2.4 – 4.4 m/s². With
respect to which side these scooter slid, the left or
the right, there wasn’t a noticeable significant
difference. For a few of the scooter models an almost
equal average sliding deceleration was calculated for
both the right and the left side.
Figure 17 shows an overview of the calculated
average sliding decelerations for the Piaggio Velofax
and the retro designed Piaggio Vespa on the asphalt
road surfaces. These tests were conducted on the
asphalt types 1 and 2. for each of these scooter
models there wasn‘t a noticeable significant
difference between the different asphalt surfaces and
or between the sides on which the scooter slid, the
left or the right. The calculated average sliding
decelerations for the Piaggio Vespa were in the range
of 3.7 to 5.0 m/s² and for the Piaggio Velofax in the
range of 3.7 to 6.0 m/s². For the Piaggio Vespa and
Piaggio Velofax we calculated significantly higher
average sliding decelerations then for the other
normal scooter models as shown before in figure 16.
This difference is mainly caused by the difference in
shape of the body, the frame parts and body part
materials. It was also due to the fact that the frame
parts, for example in the Piaggio Velefax are largely
naked or less covered. The chance that these parts
will become embeded into the road surface during
the sliding movent is higher, resulting in higher peak
decelerations.
Figure 18 shows an overview of the calculated
average sliding decelerations of the tests with the
"normal" scooters models on the concrete road
surface. We completed 21 tests, five of these tests
were not valid because of the lack of the GPS-signal.
The calculated average sliding deceleration of the
successfully completed tests (n16) were in the range
of 2.1 to 4.3 m/s². With respect to which side these
scooters slid, the left or the right, on the concrete
road surface there was no noticeable significant
difference. The scooter models like the Piaggio
Velofax and the retro designed Piaggio Vespa were
not tested on the concrete road surface.
31
 Jan Meuwissen Joos Spätjens

Figure 16: Overview of the calculated average sliding decelerations of the tests with the normal scooters on the asphalt
road surfaces.

Figure 17: Overview of the calculated average sliding decelerations for the Piaggio Velofax and the retro designed Piaggio
Vespa on asphalt road surface.

32
 Sliding decelerations of light weight scooter models (50cc / ± 60 - 100 kg)

Figure 18: Overview of the calculated average sliding decelerations of the tests with the "normal" scooters models on the
concrete road surface.

Conclusions and outlook Futhermore expertimental sliding tests need to

For the tests with the normal scooter models we
calculated average sliding decelerations on dry
asphalt in the range of 2.4 to 4.4 m/s² and on
concrete in the range of 2.1 to 4.3 m/s². A
comparison of these results with those from the r
most recent studies using the 50cc scoote rmodels on
asphalt whch found decelerations of 3.5 to 5 m/s²,
shows there is a significant difference in the minimal
value range.
conducted under different conditions, for example,
wet road surfaces and grass slopes.
References:
[1] Becke, M.: Zweiradrutschverzögerungen bei hohen
Geschwindigkeiten, Verkehrsunfall und Fahrzeugtechnik,
Februar 1985 (37-40)

The tests with the Piaggio Velofax and the retro
designed Piaggio Vespa showed that higher values
can be expected due to the different shape of body
and frame parts and body part materials. It was also
due to the fact that the frame parts are largely naked
or less covered. The chance that these parts will
become embeded into the road surface during the
sliding movent is higher, resulting in higher peak
decelerations. In order to gain more knowlegde
relating to the sliding behaviour and average sliding
decelerations of the popular retro design scooter
models, more experimental tests are needed.
[2] Becke, M.; Golder, U.: Rutschversuche mit
Zweirädern auf nasser Fahrbahn und auf Gras,
Verkehrsunfall und Fahrzeugtechnik, April 1986 (91-96)
[3] Golder, U.; Becke M.: Rutschverzögerungen von
vollverkleideten Motorrädern, Verkehrsunfall und
Fahrzeugtechnik, September 1990 (237-240)
[4] Wiek, A.: Motorroller-Rutschverzögerungen auf
trockener Asphaltfahrbahn, Verkehrsunfall und
Fahrzeugtechnik, Februar 1998 (51-54)

33
 Jan Meuwissen Joos Spätjens

[5] Golder, U.: Kippen und Rutschen von Motorrädern,

Verkehrsunfall und Fahrzeugtechnik, Mai 2007 (143-147)

[6] Chih-Yung Lin; Chun-Chsia Hsu; Chin-Ping Fung:

The study of coefficient of friction for light motorcycle
sliding on asphalt road, International Journal of the
Physical Sciences, Vol. 7(30), pp. 5167-5174, 9 August,
2012

Contact
Jan Meuwissen / Joos Spätjens
MVOA
Postbus 7030
6050 AA Maasbracht (NL)
www.mvoa.nl
info@mvoa.nl
tel: ++31 475 559027

34
© 2013 by EVU
EVU 2013 - 04

Motorcycle to Car Collisions: Evaluation of Energy

Loss and Relative Impact Velocity
Dario Vangi,
Filipo Begani
Carlo Cialdai
Antonio Virga
University of Florence

Abstract
In this paper an analysis of collisions between motorcycles and cars is presented, with the limitation that the motorcycle
is involved in its frontal part, with wheelbase shortening. The work focuses on finding an empirical formulation for the
relative velocity, allowing for the presence of the motorcycle's rider at the moment of collision. In addition, attention is
givent to the crash energy loss evaluation, which is an essential parameter for estimating the relative velocity of the vehi-
cles. Often in this field, as is remarked upon in technical literature, the main difficulty lies in the evaluation of the energy
loss for the car, on which localised deformations are usually produced, involving zones, for example near the wheels,
which have a stiffness completely different from those areas on which crash tests are normally carried out. In this paper a
simplified equation for calculating impact relative velocity is presented, which is applicable in the case where the motor-
cycle driver is still on his vehicle, that is to say when three different bodies are present (rider, motorcycles, car), each
characterised by its post-collision motion. Motorcycle to vehicle crash tests carried out, highlight that with the formula-
tion here proposed a good estimation of relative velocity can be obtained. In addition, a semi-empirical approach is in-
troduced for the evaluation of the kinetic energy lost by the two vehicles in the collision. when the front part of the mo-
torcycle is involved in the collision. It aims at providing a method for calculation that, unlike those already presented in
technical literature, is independent of the zone of the car involved in the impact.

1.Introduction vehicles under analysis, and these are not always

Road accident reconstruction represents the first fun-
damental step towards the comprehension of risk
factors, as well as vehicles’ and drivers’ behavior in
hazard situations. It is a preparatory step to the cor-
rect design of vehicles and road infrastructures. Re-
sidual deformations observable on involved vehicles,
are an essential parameter to reconstruct the accident,
being correlated to kinetic energy loss or velocity
variation during the impact [1-3].
Unfortunately in motorcycle to vehicle collisions,
energy loss determination is often a difficult task.
Kinetic energy loss related to the deformation of the
motorcycle is difficult to quantify, and for the car as
well, on which localised deformations are usually
produced, involving parts, such as near the wheels,
where stiffness is completely different from the parts
usually involved in crash tests. In such zones usual
methods, like those derived from Campbell’s ap-
proach [3-7], cannot be applied due to the lack of
stiffness parameters.
The so called Energy Equivalent Speed (EES) meth-
ods [8-10], cannot always be applied in case of mo-
torcycle to vehicle collisions because reference de-
formations are required. They need to be similar to
the type and seriousness of those observed on the
easily available. In fact, in technical literature, a suf-
ficiently complete database cannot be found for mo-
torcycles and cars, because two-wheeler crash tests
are not common. The first research in this field ap-
pears to be the one carried out by Severy [11], who
found an experimental correlation between impact
velocity and the motorcycle’s wheelbase defor-
mation.
More recently tests relating to motorcycles to barri-
ers [12-14] and motorcycles to cars [13, 15], were
carried out to define more precisely the empirical re-
lation between EES and wheelbase shortening, and
the estimation of collision speed in a collision
against a car. The study of the behaviour of the
frontal part of motorcycles is important since defor-
mation in the forks can absorb a large amount of en-
ergy. In the work presented by Niebor [16] it
emerged that front fork deformation occurs, in im-
pact against rigid barriers, for speeds greater than
32.2 km/h.
In the paper presented the impact of a two-wheeler
against a car is analysed by means of an impulse
model, allowing for the presence of the rider and his
mass when not rigidly connected to the motorcycle.
A simplified equation for estimating the relative im-
pact velocity is defined, and is suitable for the cases

35
 Dario Vangi, Filipo Begani, Carlo Cialdai, Antonio Virga
in which the motorcycle’s rider is still on board at
the moment of the collision. That is to say three dif-
ferent bodies are present (car, motorcycle, rider),
each with its own post-collision motion. The valida-
tion of the proposed formula was carried out by
means of some crash tests between motorcycles with
dummies and cars. In the work presented moreover,
semi-empirical methods are introduced, aimed at
evaluating the energy loss for both car and motorcy-
cle, when the front part of the motorcycle is involved
in the collision. It provides a computation method
that, unlike those proposed in technical literature, is
independent from which part of the car is involved.
2. Relative speed
The speed of the vehicles before the collision is usu-
ally the most important information in road accident
reconstruction, whenever a driver’s behaviour must
be determined. In motorcycle to car collisions, rela-
tive impact velocity can be obtained by analysing the
collision stage by means of impulse models. In [15]
relative velocity is calculated by assuming that at the
collision, only two bodies are involved (motorcycle
and motor vehicle), applying the following equation:
mM  mC
VR  2Ed (1)
mM mC
where Ed is the sum of energy loss for the motorcy-
cles and the car, and mM and mC the reduced mass for
motorcycles and car respectively. If energy loss for
both vehicles is known, the equation allows the cal-
culation of impact relative velocity. Equation (1)
does not allow for the restitution coefficient, that can
be approximated to zero only for high speed colli-
sions, causing great deformations. Besides it can ap-
plied, after having found the energy loss, only for
those accidents in which the rider has already aban-
doned the motorcycle when the impact occurs. If the
rider is still on the motorcycle, for relative velocity
calculations the rider dynamics must be allowed for
in the pre and post-collision phases. In the following,
an equation suitable for calculating the relative ve-
locity, valid in the case that three bodies are in-
volved, will be introduced.
2.1 Relative speed: 3 bodies
Pre-impact velocity of the motorcycle, rider and car,
can be determined by solving the following linear
system of three equations and three unknown factors,
36
obtained by momentum conservation equations
along two reference orthogonal directions, and by the
equation expressing the restitution coefficient be-
tween two-wheeler and car:
mC cos α mM cos β mR cosγ  VC  mCVC cos α  mM  mR VM cos β 
    
 mC sin α mM sin β mR sin γ  VM    mCVC sin α  mM  mR VM sin β  2
  cos α
 cos β 0  VD   ε VC cos α  VM cos β  
where m indicates the mass, α, β, ϒ, α , β e γ pre and
post-impact angles, V and V pre and post-impact
velocities, ε the restitution coefficient, and with de-
ponent C, M and R variables referring to the car, to
the motorcycle and to the rider, respectively. Solving
the system in accident reconstruction is usually diffi-
cult, because the information necessary to study the
post-impact motion of the three bodies is not always
available, in particular the estimation of the motor-
cycle and rider post-impact angles. To solve colli-
sions with a rider, a formulation similar to (1) was
sought, to determine the relative velocity between
motorcycle and car by establishing the total kinetic
energy loss.
The analytical solution of the equations of conserva-
tion of momentum and energy, in the case of a colli-
sion between two bodies leads to (1). However, in
the case of three bodies, it cannot be expressed as a
function of initial relative velocity between car and
motorcycle, and the solution must be sought numeri-
cally. For the purpose of finding a formula analyti-
cally simple and comparable to (1), numerical simu-
lations have been carried out, solving the system (2),
where the pre-impact data (masses, speeds, angles)
and post-impact angles were assumed, making them
vary within realistic ranges. In particular, 7 different
crash configurations between cars and motorcycles
were simulated, summarised in Figure 1. The varia-
tion of the coefficient of restitution as a function of
the impact velocity was assumed following the indi-
cations given in [17], using slightly higher values
because the elasticity of the front suspension of the
motorcycle typically produces higher values of the
coefficient of restitution, in comparison to collisions
between two cars.
 Motorcycle to Car Collisions: Evaluation of Energy Loss and Relative Impact Velocity
Figure 1 – Simulated collision configurations between car
and motorcycle with rider.
After the determination of post-impact velocities for
the three bodies in each simulation, the energy loss
in the single impact was calculated by using the
energy conservation equation. By correlating the
values of the initial relative velocity to the kinetic
energy loss calculated for each simulated case, an
equation was found that best approximates the
relative velocities calculated to those set in the
simulation.
Such equation is:
 3
2 Ed
VR 

m 1  2
*
formally similar to (1), in which m* is given by:
mC mM  ηmR 
m* 
mC  mM  ηmR 
where , rider’s mass reduction factor, varying from
0 to 1, is the ratio of rider’s mass to the sum of the
masses of rider and motorcycle: R/( mR.+ mM)
In figure 2 the values of relative velocities calculated
by using equation (3) for 350 simulated collisions are
shown. Data are well distributed around 45 degrees
line, with standard deviation equal to 3.1 km/h.
Figure 2 - Comparison between the relative speed
calculated by using equation (3) and relative speed of the
simulated crash test.
The formulat found shows that only a fraction of
the mass of the driver of the motor vehicle
contributes to the estimation of the relative velocity.
Such a result fully reflects the dynamics of these
kind of collisions, in which the driver, being able to
move, slides forward and then rises from the
motorcycle, contributing only partially to the plastic
deformation of the two vehicles.
3. Validation
In order to validate the relationships found by
numerical simulation with an impulse impact model,
a series of experimental tests was carried out, in
which impacts between a motor vehicle and
motorcycle with a dummy on board were
reproduced. All tests were performed with the
vehicle initially stopped and placed in contact with a
rigid barrier, so as to prevent any post-impact
motion. In these tests, the launch system of the
motorcycle consisted of a tense elastic cord, attached
to the front wheel hub of the motorcycle. The
motorcycle was held in upright position by means of
two side wheels fixed to the rear of the vehicle (see
figure 3).
Figure 3 - Motorcycle connected to the launch system and
equipped with two side wheels fixed to the rear of the
vehicle.
By means of a suitable device, one meter before the
impact with the vehicle, the side wheels were disen-
gaged from the motorcycle and the tension of the
elastic cord released, so that the motion of the vehi-
cle was completely free at the moment of impact
(figure 4). The motorcycle velocity, before and after
the impact, was measured by means of a laser photo-
electric system, (resolution 0.1 km/h), a radar gun
(Radar Stalker Pro: Speed Range 1-480km/h,
accuracy ±1.6km/h) and by analysing the movie
recorded by high speed video camera Fastec InLine
(250 fps, 8.8 s storage capacity). A non instrumented
dummy woth a mass of 80kg was placed on the two-
37
 Dario Vangi, Filipo Begani, Carlo Cialdai, Antonio Virga

wheeler. It was held in position, during the launch, For each test the kinetic energy loss was calculated
by using adhesive between hands and handlebars and by using the following:
a lightweight wooden support, thus not influencing
the motion during impact and post-impact phases.
1
mM  mR VM 2  1 mMVM 2  1 mRVR 2  Ed 4
2 2 2
The tests were conducted on motor scooters at Equation (3) was then applied to evaluate the impact
different impact speeds, hitting the car in different relative velocity and compare it to the actual test
positions: door; front wing; and front area. Two speed. As can be seen in table 2 and figure 5, the
markers at known distance were placed on the two- equation yields a satisfactory estimation of relative
wheeler in order to obtain the post-impact velocity velocity of the vehicles involved. Maximum error
from the high-speed movie. Before each crash test does not exceed 3 km/h, and the best-fit line shows a
the motorcycles were weighed and their wheelbase correlation coefficient equal to 0.905.
was measured. After the end of the test the
wheelbase shortening was evaluated, by performing Table 2 - Relative speed verification in the experimental
tests conducted.
a measurement on both sides, and finally the
maximum deformation on the motor vehicle was VR VR
Varia Varianc
recorded. ID crash calculated
nce e%
(km/h) (km/h)
1 33,0 35,8 -2,8 8,48
2 41,0 43,8 -2,8 6,83
3 41,8 41,2 0,6 1,44
4 49,4 51,3 -1,9 3,85
5 43,9 45,7 -1,8 4,10
6 40,0 39,1 0,9 2,25

Figure 4 – Crash test with motor vehicle and motorcycle

with dummy, carried out at the University of Florence by
the authors.
Figure 5 – Comparison between the relative speed
A total of 6 tests were carried out with 6 different calculated by using equation (3) and relative speed of the
scooters in the speed range between 30 and 50 km/h 6 crash tests carried out.
(table 1).

Table 1 - Car-motorcycle crash tests carried out with

4. Energy dissipated
dummy rider.
Weight Speed To calculate the relative velocity in a collision
ID Motorcycle Car
(kg) (km/h)
between a car and a motorcycle by applying (3), it is
Yamaha Fiat
1
Majesty
160,0 33,0
500
necessary to estimate the kinetic energy dissipated.
Aprilia Fiat The overall kinetic energy dissipated in a collision
2 135,0 41,0 between acar and a motorcycle is the sum of the
Leonardo 500
Piaggio Fiat individual energy contributions dissipated by the two
3 109,0 41,8 vehicles involved in the collision:
Liberty Punto
Lambretta Opel
4
151-N
111,0 49,4
Corsa Ed  Ed ,M  Ed ,C 5
Opel
5 Outlook 150 116,0 43,9
Corsa
where Ed,M is the kinetic energy dissipated by the
Peugeot Opel
6 101,0 40,0 motorcycle and Ed,C that dissipated by the car. In the
Looxor Corsa
following some methods will be introduced, intend-
38
 Motorcycle to Car Collisions: Evaluation of Energy Loss and Relative Impact Velocity

ing to evaluate kinetic energy loss for the two vehi-

cles when the motorcycles presents deformations in
its front part.

4.1 Motorcycle kinetic energy dissipated

The kinetic energy dissipated by the motorcycle is

determined typically through the estimation of the
value of EES for the particular damage:

mM EES  6
1
E d ,M 
2

In the cases in which motorcycle’s front forks de-

formation occurs, without fracture of the frame or of
the steering tube, EES evaluation is made by means
of semi-empirical correlations with total wheelbase Figure 6 - Methods used in literature to make the vehicles
shortening. Such relations were obtained, for bikes collide the rigid barrier.
and scooters, by means of crash tests against a rigid
barrier [12]. Correlation between wheelbase shorten-
Afterwards, with the aim at extending the data from
ing and EES seems better, with less dispersion, in
literature, a number of crash tests between motorcy-
comparison to that between total vehicle shortening
cles and rigid barrier were carried out at the testing
and EES, moreover it results in an easy application
facility run by the University of Florence. Tests were
of the motorcycles’ energy loss calculation.
conducted with a double objective: a) evaluating the
influence on wheelbase shortening of forks extension
Most of the experimental data that can be found in
during the impact; b) identifying, in the tests at high-
technical literature, derives from the research de-
er speed, where the contact occurred and how forces
scribed in [12-14]. Tests described in [12-14] regard
were transferred between motorcycle and rigid barri-
collisions between two-wheelers and rigid barriers.
er.
They differ in the method used to make the vehicle
For the two-wheeler the same method used for mo-
collide the fixed barrier: in [13] the motor cycle is
torcycle to car impact was used (figure 3), this time
towed by another vehicle, while in [12] and [14] a
without using the dummy to allow for the rider's
pendulum system is employed (figure 6). Literature
mass. The instrumentation used to measure the im-
data record values of motorcycles’ wheelbase short-
pact velocities is the same described in paragraph 3.
ening up to 0.4 m. In the case in which the tubes of
A total of 17 crash tests against rigid barriers were
the front fork present a different deformation, wheel-
executed (table 3), using both motor scooters and
base variation was calculated as the average between
motorcycles. Impact speed was varied in the range
the measurements on both sides.
15-80 km/h.
The literature data cited above appear not to be ho-
mogeneous for both the different test modes and for
the presence of some cases, considered abnormal, in Table 3 – Motorcycles used in crash test against the rigid
which the vehicle arrives at the collision in such a barrier.
position that the results sought are partially altered. Engine Weight
In order to define a better correlation between mo- Model
size (cm3) (kg)
torcycle’s wheelbase shortening and EES, the indi- Aprilia
vidual crash tests reported in the literature were first 125 150,0
Scarabeo
analysed and, with the help of video and photograph- Aprilia
ic reports, a selection of the tests was made, exclud- 200 148,0
Scarabeo
ing those tests in which the motorcycle's behaviour Aprilia
varied significantly from that in the rest of the tests. 500 178,0
Scarabeo
Tests in which the vehicle presented a fracture of the
Aprilia
frame or steering tubes, causing a wheelbase shorten- 850 214,0
Mana
ing not solely due bending of the front forks were al-
Aprilia
so discarded. 650 186,0
Pegaso

39
 Dario Vangi, Filipo Begani, Carlo Cialdai, Antonio Virga
In front of the rigid concrete barrier an additional
steel barrier was fixed (80x120 cm), holding, in its
rear part, three strain gage load cells, positioned at
the vertices of a triangle, with one vertex at the top
(figure 7). Such positioning allows the evaluation of
the modulus and position of the resulting force that
the motorcycle transfers to the barrier, by analysing
the measurement yielded by the three load cells.
Figure 7 - Steel barrier equipped with three strain gage
load cells.
In the tests aimed at the evaluation of the influence
of front suspension position, front forks were retract-
ed and tied with a rope at their maximum compres-
sion position, in order to simulate the case of a brak-
ing maneouvre (emergency braking) before a colli-
sion. Wheelbase shortening was measured by means
of measurements done before and after each crash
test.
In figure 8 the data obtained are presented, together
with data from literature previously selected. As it
can be seen, two well separated zones can be identi-
fied: for Δp lower than 0.4 m and for Δp greater than
0.4 m.
The first area includes tests performed on scooters
and motorcycles with front forks retracted and not
retracted.
Wheelbase shortening occurs because of deformation
of one or more plastic hinges, located in the weakest
and more stressed points of frame and fork, that is to
say because of localised deformation, rather than a
distributed plastic bending. In the case of the motor
scooter, u the plastic hinge is usually located near the
steering tube, where the front fork bifurcates. In the
case of the motorcycles two plastic hinges are usally
placed on the two tubes of the fork, near the steering
stem, and/or one plastic hinge is located near the
connection of the steering tube to the frame.
Motorcycles and scooters show the same behaviour,
and the relation between wheelbase shortening and
EES appears to be independent from the mass of the
vehicle.
The best fit line is:
40
EES  b0  b1  p  2,91  37,19  p 7
showing a coefficient of correlation R2 equal to
0.962 and a standard deviation of 0,75 m/s.
Figure 8 - Relation between wheelbase shortening and
EES obtained from rigid barrier crash test. In rhomboidal
shape the literature data, in triangular and round shape
data from tests conducted by the authors.
On a qualitative basis, theere is no observed
influence of the excursion of the forks in the
correlation between wheelbase shortening and EES.
During the collision, even in cases where the front
forks were not already precompressed, an almost
complete compression of the front suspension occurs
before plastic deformation begins.
So, in actual incidents, whether or not the motorcycle
is already under braking before the impact,
deformation of the fork begins when the suspensions
are already well compressed.
Regardless of fork excursion, vehicle kinetic energy
variation, equal to the work of external forces, is also
equal to the internal work done by stress and strain in
the plastic hinge; so that, with equal EES, an equal
rotation angle is developed by the fork around the
plastic hinge. Thus the expected wheelbase shorten-
ing is independent from fork’s compression condi-
tion during the impact.
The second zone of the graph shown in figure 8 in-
cludes tests carried out at speed greater than 65
km/h; in this area the linearity between wheelbase
shortening and EES previously found is lost. Analy-
sis of high speed videos and signals recorded by the
instrumented barrier, shows that the two-wheeler be-
haved differently during the impact. After the contact
between wheels and barrier begins, the following
phases can be identified:
 deformation of front fork until the rear part of
the wheel comes in contact with the engine
block (this phase can have different amplitude
 Motorcycle to Car Collisions: Evaluation of Energy Loss and Relative Impact Velocity
depending on whether the wheel remains
unbroken or not);
 rotation of the two-wheeler around wheel-
barrier contact point;
 contact between the upper part of the vehicle
and the instrumented barrier.
In figure 9 the evolution of contact forces as a func-
tion of time can be observed; in particular it can be
observed that after the first force peak (A), produced
when the contact between wheel and barrier begins
and revealed by both the cells positioned in the lower
part of the barrier, the following sequence can be ob-
served: a second peak (B), recorded by lower cells,
corresponding to the moment in which the rear part
of the wheel touches the engine block; a third peak
(C), revealed by the upper cell, produced by the con-
tact between the upper part of the vehicle and the
barrier.
For those collisions in which the cited feature occurs,
it is clear that a share of the vehicle’s kinetic energy
is dissipated by mechanisms different from front fork
bending, so that relation (7), given a value for
wheelbase shortening, underestimates the energy
loss.
Figure 9 - Evolution of contact forces as a function of time
as measured by the three load cells. The gray and black
curves they relate to the cells positioned in the lower part
of the barrier, while the white curve is the signal derived
from the cell at the top of the barrier.
4.2 Car kinetic energy dissipated
In technical literature the proposed methods for the
evaluation of the kinetic energy loss for motor vehi-
cles, are based on correlations among impact veloci-
ty, motorcycle shortening and car deformation [9].
Nonetheless, such methods yield results with large
dispersion, basically due to input data, obtained
without allowing for the difference in stiffness be-
tween the car and the motorcycle.
Another method commonly employed for motor ve-
hicle kinetic energy loss estimation, is based on the
evaluation of EES value for the damage under con-
sideration. This is done by means of a comparison
with documented crash tests, carried out on the same
vehicle and with deformation on the same area, for
which energy loss is known, together with the extent
of the deformation. The limitation of such an ap-
proach lies in the availability of documented crash
tests for the numerous types of vehicles in circulation
and for the required type of deformation; moreover
the method reveals to be only qualitative, and strong-
ly dependent on the technician’s experience.
Recently an empirical approach was presented [15],
based on a statistical correlation between defor-
mation and energy loss on the car only. In [15] 31
crash tests looking at motor scooters and motorcy-
cles impacting against stationary cars were analysed.
A relation was obtained by correlating the orthogo-
nal maximum deformation on the car to the energy it
dissipated, independent from the mass of the bodies.
This approach for energy loss evaluation does not al-
low for the motorcycle and its wheelbase shortening,
nor the possible difference in stiffness of the differ-
ent part of the car.
Car kinetic energy loss evaluation can also be carried
out by applying the Triangle Method [18-19] to these
kinds of collisions. The Triangle Method, derived
from the schematisation of force and deformation
proposed in [3], is based on the approximation of any
vehicle’s deformation, with linear profiles with rec-
tangular and triangular shapes. Application of the
Triangle Method allows the energy loss of the car to
be expressed as a function of the force (FR) that car
and motorcycle exchange, during the compression
phase, through the contact between front wheel and
deformed area:
Ed ,C 
1
FR (kC  δ ) 8
2
where C is the maximum car deformation, k the
shape coefficient depending on the type of defor-
mation observable on the car, and δ the elastic de-
formation.
41
 Dario Vangi, Filipo Begani, Carlo Cialdai, Antonio Virga
By regarding the two-wheeler as a spring-mass sys-
tem, it is possible to calculate the force that the two
vehicles exchange during the impact, by knowing the
wheelbase shortening and parameters b0 and b1 ex-
pressed in equation (7). Once the force is known, the
relation of the Triangle Method expressing the ener-
gy loss for the car becomes:
E d ,C 
1
2
 
M M b0 b1  b1 p (kC  δ )
2
9
where MM is the mass of the motorcycle only, b0 and
b1 are the stiffness coefficients shown in (7), Δp is
the motorcycle wheelbase shortening, k a form fac-
tor, equal to 0.564, allowing to pass from a triangle
shaped damage to a rectangular one with the same
energy loss, C the maximum deformation on the car,
and δ is elastic deformation. To better understand the
concept of elastic deformation it can be explained as
follows: δ is the ratio between A and B where A and
B are stiffness coefficients of the motor vehicles de-
pending on the part of wheelbase which is relevant,
according to Campbell’s method [3] (mean value for
coefficients A and B are, as published by NHTSA: δ
= 0.0712 m for the front zone, δ = 0.0798 m for the
rear zone, δ = 0.0364 m for the side zone).
Equation (9) express the energy loss for the car in the
case in which the deformation, typically with a trian-
gular shape, is limited, as regards its height, to a lev-
el comparable to that of the wheel rim of the motor-
cycle. In the cases in which the deformation on the
car exceeds such level, due to vehicles geometry
and/or to high impact velocity causing a rotation of
the motorcycle, the kinetic energy loss is not totally
ascribed to bending of front forks. Thus the equation
(9) will yield an underestimated value for the car’s
energy loss.
By analysing crash tests reported in technical litera-
ture [12, 14] it was possible to identify a correction
coefficient (λ) to be applied to (9) in order to com-
pensate energy loss calculation for the car:
1
 
E d ,C  λ  M M b0 b1  b1 p (kC  δ )
2
2
10
with λ = C + 1 (C expressed in meters).
5. Observations
The methods for energy loss evaluation here present-
ed, generally yield results that are comparable to
each other, and the choice among them basically de-
pends on available data. All the methods yield an ap-
proximated estimation of the energy loss, mainly due
to the dispersion of the experimental data from
42
which the correlation between wheelbase shortening
and energy loss for the two-wheeler is obtained.
Such dispersion ends up at around ±6 km/h. Unlike
the method presented in [15], based only on experi-
mental tests, the Triangle Method is based on a phys-
ical model of the phenomenon, allowing its applica-
tion in cases very different from those typically re-
produced in crash tests. Under this perspective the
use of the Triangle Method is preferable and can be
considered more accurate when the motorcycle col-
lides with the car in areas not usually subjected to
crash tests and for which the stiffness is unknown.
The use of the Triangle Method or EES, yields the
energy actually dissipated by the vehicles and, in
those cases in which at the moment of the collision
the rider is still on the motorcycle, such values have
to be used with equation (3). If the kinetic energy
dissipated in deformation is estimated by means of
the method proposed in [15], if the rider is still on
the motorcycle, the value obtained does not represent
the actual energy loss, since the statistical correlation
between energy loss and deformation was found by
using the (1). So, to obtain a correct value for the
relative velocity of the vehicles at the moment of
collision, the value for energy loss found with the
method presented in [15] must be used with equation
(1) and not with (3).
6. Conclusions
The paper presented analyses, by means of an impul-
sive method, the collision of motorcycles against
cars, allowing for the presence of the rider and its
mass, not rigidly bound to the motorcycle. Firstly a
simplified equation was defined, aimed at the calcu-
lation of the relative velocity between motorcycle
and car at the moment of collision, suitable for those
cases in which the rider is still on his seat. The vali-
dation of such equation, carried out by comparing
the results of a motorcycle to car crash tests in which
a dummy used as rider, and the speed ranged from
30 to 50 km/h, yielded a maximum error on relative
(6) Finally,
velocity calculation not greater than 3 km/h.
the methods for impact energy loss evaluation were
revisited, this being the main input parameter in the
equations for relative velocity calculation.
With regard to the energy loss for the two-wheeler,
data from literature, correlating the empirical corre-
lation between motorcycle’s energy loss and wheel-
base shortening, were completed by means of crash
tests against a rigid barrier. In the tests carried out,
the influence on the motorcycle of the shortening of
front forks extension was investigated as well, which
identified, in tests at higher speed, contact areas and
 Motorcycle to Car Collisions: Evaluation of Energy Loss and Relative Impact Velocity
the consequent force exchange between motorcycle
and rigid barrier.
As regards the energy loss for the motor vehicle, be-
sides the methods already available in technical liter-
ature, the use of the Triangle Method was proposed.
All the methods for calculating the energy loss here
surveyed, being comparable to each other, produce
results affected by some uncertainty, due to the dis-
persion of the experimental data from which they
were developed.
References
[1] Emori R.I., Analytical approach to automobile
collisions, SAE Technical Paper No. 680016, 1968.
[2] Mason R.P., Whitcomb DW., The estimation of
accident impact speed, Cornell Aeronautical Laboratory,
ReportNo. YB-3109-V-1, August 1972.
[3] Campbell K.E., Energy basis for collision severity, SAE
paper 740565, Society of Automotive Engineers, Inc.,
Warrendale, Pennsylvania, USA (1974).
[4] Fonda A. G., Principles of crush energy determination,
SAE 1999-01-0106, Society of Automotive Engineers, Inc.,
Warrendale, Pennsylvania, USA (1999).
[5] Tumbas N.S., Smith R.A., Measurement protocol for
quantifying vehicle damage from an energy basis point of
view, Tumbas and Associates - U. S. Naval Academy, SAE
paper 880072, Society of Automotive Engineers, Inc.,
Warrendale, Pennsylvania, USA (1988).
[6] Nystrom G.A., Kost G., Werner S.M., Stiffness parame-
ters for vehicle collision analysis, Failure Analysis Associ-
ates, Inc., SAE paper 910119, Society of Automotive Engi-
neers, Inc., Warrendale, Pennsylvania, USA(1991).
[7] Neptune J.A., Blair G.Y., Flynn J.E., A method for
quantifying vehicle crush stiffness coefficients, Blair,
Church & Flynn Consulting Engineers, SAE paper 920607,
Society of Automotive Engineers, Inc., Warrendale, Penn-
sylvania, USA(1992).
[8] Zeidler F., Schreier H.H., Stadelmann R., Accident re-
search and accident reconstruction by the EES-accident re-
construction method, SAE paper 850256, Society of Auto-
motive Engineers, Inc., Warrendale, Pennsylvania, USA
(1985).
[9] Schreier H.H., Nelson W.D., Applicability of the EES-
accident reconstruction method with MacCAR©, SAE
870047, Society of Automotive Engineers, Inc.,
Warrendale, Pennsylvania, USA(1987).
[10] ISO 12353-1, Road Vehicles – Traffic Accident Anal-
ysis – Part 1: Vocabulary – First Edition, 2002.
[11] Severy D.M., Brink H.M., Blaisdell D.M., Motorcycle
collision experiments., Proceedings of the fourteenth Stapp
car crash conference. Warrendale, PA: Society of Automo-
tive Engineers; 1970.
[12] Kasanicky G., et al.,Analysis of single-track vehicle
accidents, University publishers Zilina (2003).
[13] Adamson K. S., et Alii. Seventeen motorcycle crash
tests into vehicles and a barrier, SAE 2002-01-0551, Socie-
ty of Automotive Engineers, Inc., Warrendale, Pennsylva-
nia, USA(2002).
[14] Priester J., Weyde M., Tests carried out in Berlin. CD-
ROM, Ingenieur- und Kfz- Sachverstӓndigenbüro Priester
und Weyde, Saarbrücken, Berlin (2005).
[15] Wood D. P., Glynn C., Walsh D., Motorcycle-to-car
and scooter-to-car collisions: speed estimation from perma-
nent deformation, Proc. IMechE Vol. 223 Part D: J. Auto-
mobile Engineering, PP. 738-759, DOI:
10.1243/09544070JAUTO1069(2009).
[16] Nieboer J.J., Wismans J., Versmissen A.C.M., van
Slagmaat M.T.P., Kurawaki I., Ohara N., Motorcycle crash
test modeling, SAE Paper No. 933133, 37th Stapp car crash
conference proceedings, 1993.
[17] Antonetti V.W., Estimating the coefficient of restitu-
tion of the vehicle-to-vehicle bumper impacts, SAE paper
980552, 1999.
[18] Vangi D., Simplified method for evaluating energy
loss in vehicle collisions, Accident Analysis and Prevention
41 (2009).
[19] Vangi D., Begani F., Performance of triangle method
for evaluating energy loss in vehicle collisions, Journal of
Automobile Engineering, Volume 226 Issue 3 pag. 338-347
(2012).
[20] Vangi D., Begani F., Energy loss in vehicle collisions
from permanent deformation: an extension of the “Triangle
Method”, Vehicle System Dynamic, 2013
http://dx.doi.org/10.1080/00423114.2013.770538.
Contact
Dario Vangi, Professor, Antonio Virga, PhD, Carlo Cialdai,
PhD student, Filippo Begani, PhD
Via di S. Marta, 3
50139, Florence, Italy
dario.vangi@unifi.it, antonio.virga@unifi.it,
carlo.cialdai@unifi.it, filippo.begani@unifi.it
tel: ++ 39 055 4796505.
43
 Dario Vangi, Filipo Begani, Carlo Cialdai, Antonio Virga

44
 © 2013 by EVU
EVU 2013 - 05

On the lean angle of motorcycles driving through a

curve
Christian Hädrich
Wolfgang Hugemann

Abstract
The lean angles of motorcycles driving through a sequence of curves were measured by means of a GPS logger. All
participants had to ride on a standardised pre-set route, about 20 km long. From the logged GPS data – speed,
geographic coordinates and time – the motorcycle lean angles were calculated by means of a mathematical model that
accounted for individual motorcycle parameters, especially the tyre width.
The test route had 39 curves whose radii were evaluated by means of aerial photographs. The lean angles at the curves’
vertices were derived from the GPS data and grouped into radii categories – first without any additional criteria and in
a second step depending on the characteristics of the rider and the motorcycle. Typical rider characteristics are the
personal risk-taking behaviour and the riding experience.
The results show that the standard values published in 1981 do not reflect the achievements of modern motorcycle
technology. Furthermore, a categorisation by the driving radius is highly recommendable.

simplifies a lot of calculations, while on the other

Introduction
Investigations on the behaviour of passenger car
drivers have shown that the average driver hardly
pushes his car to the stability limit when negotiating
a turn [4]. This holds all the more when the curve
radius becomes large: the accepted centrifugal
acceleration of the average driver drops with
increasing radius, i.e. rising speed.
The same is to be expected for motorcyclists: the
physical limit, i.e. the critical curve speed, is one
thing – the curve speed actually accepted by the rider
is another thing. Up to now there has been no
detailed investigation on this and thus the values
postulated in [1] are cited on and on, without
knowing about their reasons [2].
The maximum curve speed of a certain motorcycle
(or rather its rider) is a valuable information when
reconstructing a motorcycle accident that happened
after a turn in the road. A typical scenario would be a
car turning slowly in from a minor road, the driver
claiming that the motorcycle had still been invisible
when he decided to turn in. The answer to the
question whether this is just an excuse or indeed a
partial cause of the accident, severely depends on the
motorcyclist’s approach speed, i.e. the accepted lean
angle in the curve.
Experimental Setup
For two-wheelers, the lean angle is strictly coupled
to the centrifugal acceleration. On the one hand, this
hand it impedes the simplest measuring approach by
accelerometers: in the steady state, an accelerometer
mounted to the motorcycle frame will measure no
lateral acceleration. Any approach by means of
accelerometer would thus require the use of an
inertial platform mounted to the test vehicle, which
would make the measurements tedious and costly.
Another possible measuring approach would be an
angular rate sensor, more specific a roll rate sensor,
as it is used in the BMW S 1000 RR. This would
require a numeric integration on the lean angle,
starting at the beginning of a turn, in order the
prevent data deterioration by excessive integration
times. This approach would allow to measure (or
rather calculate) the motorcycle’s lean angle in the
literal sense – not the lean angle of the entire system,
including the rider.
Although this might be an interesting concept for
future investigations, we opted for the simplest
approach by means of GPS receivers, logging the
speed and the position.
From the position data we can conclude on the local
curvature, which allows us to calculate the
centrifugal acceleration when combined with the
speed. This in turn will give us the lean angle when
combined with geometrical data of the motorcycle.
The approach via GPS position measurements might
seem too coarse at the start, as its accuracy is
typically only ±15 m. The calculation of the local
curvature does however only call for a local relative
45
 Christian Hädrich, Wolfgang Hugemann

accuracy of the position data, which GPS provides cog e

quite reliably. mg · a
cog g
After some consideration, we opted for a pre-defined
route for the driving tests: this allowed us to compare
the data of different drivers more easily. Moreover,
mg · g
we could avoid excessive shadowing of the satellite
signals by an appropriately chosen test track: 
otherwise a lot of curvy roads would have led e
through deep woods and would have deteriorated the
GPS measurements. Furthermore, the choice of a '
pre-defined test drive allowed us to control the
influence of speed limits: on the chosen test route, r
the allowed speed did not limit the cornering speed.
Prior to the test drives, the riders were instructed to
drive normally, avoiding excessive risk. To prevent
too sluggish drives (leading to an underestimation of
the normal lean angles), the riders were meticulously
instructed about the route, whose turns were also
marked with arrows in advance.
By means of Internet forums we could acquire 29
volunteers of varying age, riding a variety of
motorcycles.

Mathematical model
The basic relationship between the speed , the local
curvature R and the motorcycle lean angle e is
simple,
 v²  Figure 1: Influence of tyre width on the lean angle
 e  arctan   (1)
 Rg 
This relationship holds for infinitesimally narrow We can thus split the lean angle into a portion that
tyres. The real lean angle is somewhat larger, as the can be calculated by the simple relationship eq. (1)
wheel standing point moves inward for wider tyres, and an additional portion which results from the tyre
Figure 1, which can easily be accounted for in the width
calculation    e   (3)
These relationships allow us to calculate the lean
 r  sin   e   angle from the speed, the local curvature of the
  arctan   (2) motion curve and two motorcycle parameters. This
 hr  lean angle will refer to the entire system of the
motorcycle plus the rider and cannot be split into the
with h as the centre of gravity (cog) height of the actual lean angles of the motorcycle and the rider,
entire system (motorcycle and rider) and r as the tyre which can deviate from this angle, depending on the
tube diameter. rider’s curve negotiation strategy.

Data processing
The GPS speed is calculated from the receiver’s
speed relative to the single satellites, which in turn is
calculated by the detuning of the signal’s carrying
frequency. This low-level signal processing is
performed by the GPS receiver and leads to an
accuracy of ± 0.1 km/h. This value from the
receiver’s technical specification can easily be
46
 On the lean angle of motorcycles driving through a curve

checked by recording the speed signal of the on 21 vertices are Ci  i with N = 770. For the
stationary receiver and calculating its RMS. second order derivative the coefficients can be taken
The local curvature can be calculated by from the tables in [5]:
3 i -10 -9 -8 -7 -6 -5 -4
 dx  2  dy  2  2
      Ci 190 133 82 37 -2 -35 -62
 dt   dt  
R  (4)
dx d 2 y d 2 x dy i -3 -2 -1 0 1 2 3
  
dt dt 2 dt 2 dt Ci -83 -98 -107 -110 -107 -98 -83
i 4 5 6 7 8 9 10
which calls for the numerical computation of the first
and the second derivative of the GPS position Ci -62 -35 -2 37 82 133 190
measurements. These are, however, affected by
noise, due to the short-term variations in the satellite
signals’ running times. As differentiation has a The scaling factor for the second order derivative is
“roughening” effect on the signal, we have to apply N = 33649.
some means of levelling the data. The simplistic When using the estimates for the derivatives in eq.
approach would be to apply the usual formulae for (4), the sample interval T cancels out.
numerical differentiation and then apply a low-pass
filter to their result. Motorcycle characteristics
There is, however, a more direct solution to the In order to calculate the additional lean angle caused
problem: we interpolate the data points locally by a by the finite tyre width, we need to determine the
low-order interpolation polynomial which is then tyre tube diameter and the cog height of the
differentiated in order to gain the local derivatives. individual motorcycle. The first task was performed
This makes levelling an intrinsic step of the by means of a contour template, which we used to
differentiation and achieves better results. For the transfer the tyre contour on a sheet of paper which
first order derivative, the data points have to be was then approximated by a circle segment. The
approximated by a straight line, whose slope outer surface of a real motorcycle tyre is however no
provides the sought parameter. For the second order torus, as the middle part, which is used when running
derivatives, the same points are approximated by a straight, is flattened. With our approximation we
second order polynomial. In practice, this way of thus concentrated on the outer parts, which only
proceeding leads to finite differences equations of come into road contact within curves, Figure 2.
the structure
 d nu  1 m

 n     Ci  n, m   u j i (5)
 dt  j N (n, m)  T i  m
n

The data points uj+i have to be recorded at a constant

rate, with T as the sample interval. The term left of
the equal sign denotes the estimate for the nth order r
derivate at index j. In our case we have two input
signals xj,yj, which have to be substituted for uj.
Figure 2: Approximation of the tyre contour
The following equation provides an estimate for the
speed vx, based on the five data points around the
index 25, taken with 10 Hz sampling rate: The cog height of the entire system – motorcycle
 dx  1 with rider in normal driving position – is not that
     2  x23  x24  x26  2  x27  (6) easy to determine. We tried to perform this by
 25
dt 10  0,12
measuring the wheel standing forces in horizontal
The needed coefficients can be readily taken from driving position and with elevated front wheel,
tables [5]. Testing the results of eqs (4, 5) with Figure 3, but this yielded only rough estimates, as
various numbers of data points, m = 10 seemed to the errors in this differential measurement turned out
give the best results for our 10 Hz data, as too high (± 25%). As the additional angle ’ is
corresponding to a local levelling stretching over 2 s. only a tenth of the effective lean angle e, this high
The coefficients for the first order derivative based error span still seemed to be acceptable.

47
 Christian Hädrich, Wolfgang Hugemann
Data acquisition and processing
The position and speed data was recorded with 10
Hz by a QStarz BT-Q1000EX Race Logger in
NMEA format. The data was recorded without
EGNOS correction, as this correction data is
considered as problematic for ground navigation in
Europe, especially in a forested environment. Post-
processing of the data with EGNOS tables supplied
on the Internet (EMS) would have been an option.
Some experience with the DL1 data recorder [3],
whose software offers this correction, however led to
questionable results.
Figure 3: Measuring the wheel standing forces with
elevated front wheel
Raw data x t  Equidistant time signal
x t  y t 
Figure 4: Overview of a DaDisp worksheet
48
The geographic coordinates
x, y
by a simple sphere model of the earth
xi  R  cos  i    i  0  (7)
yi  R   i  0  (8)
with R as the earth radius.
Overlaying an aerial photograph with the entire track
computed from eqs (7, 8) showed that this rough
approximation is sufficiently accurate for our
purposes, especially for the local evaluations that we
are after.
Each motorcycle carried a camcorder which recorded
the entire test drive, such that we could check
whether any specifics (e.g. oncoming traffic, slow
traffic ahead) had to be accounted for in the
negotiation of a certain curve. If that was the case,
the data for that curve drive was dropped in the
evaluation.
For each curve and driver we picked the peak lean
angle, such that we yielded 29 test drives for each
curve, if no data had to be dropped. As eq. (4)
incorporates 2 s of measurement, the peak lean angle
is to be concerned as a short-term average rather than
a single value.
Reduction of raw data
(90 percent)
Visualisation of the
Radius y t  raw data (track)
 On the lean angle of motorcycles driving through a curve

Older experiments with passenger cars had shown

that the centrifugal acceleration accepted by average n = 25,9°
drivers is coupled to the speed: the higher the speed,
the lower the accepted centrifugal acceleration [4]. g = 29,0°
As we expected similar effects for motorcyclists, the
curves on the test drive were grouped into radius
categories, such that the lean angles could be
associated with a certain curve radius. The
boundaries of the radius categories were chosen such
that each category contained an equal count of
curves.
For the data processing we used DaDisp as a
dedicated signal processing software, which proved
to be more flexible and significantly faster than
comparable computations in a spread sheet, Figure 4.
Figure 5: Lean angle by video analysis
Precision For a single curve, we determined the actual lean
In the beginning we had our doubts about the angle by means of video recording, compared it to
computation of the local curvature from the GPS the lean angle computed from the data and again
data. We therefore compared the computed radii of found them in good accordance, Figure 5.
the curve drives with the curve radii determined from
aerial photographs and found them to be in good
accordance.
45°
90%
number off evaluated runs 75%
40° median
25%
10%
35°
178

30° 171 154

132 209
lean angle

25°
134
20°

15°

10°

5°

0°
10 - 35 m 35 - 55 m 55 - 80 m 80 - 125 m 125 - 200 m 200 - 300 m
driving radius

Figure 6: Overall evaluation of the data, grouped by the curve categories

49
 Christian Hädrich, Wolfgang Hugemann

40°
sport enduro street 80er
55 f - sport f - enduro f - street f - 80er
35°
36
75 54 35 49
28 number of evaluated runs
30° 42
68
69
25 54 58 46 93
25° 13 47
9
lean angle

12
25 54
20° 14
7
10
15°

10°

5°

0°
10 - 35 m 35 - 55 m 55 - 80 m 80 - 125 m 125 - 200 m 200 - 300 m
driving radius

Figure 7: Lean angle by type of motorcycle

40°
a <= 5 5 < a <=10 a > 10

50 f-a≤5 f - 5 < a ≤ 10 f - a > 10

35°
55
53
73 53
number of evaluated runs
30° 57 45
60
58
44 73
43 44 63
25° 73
39
lean angle

40
55
20°

15°

10°

5°

0°
10 - 35 m 35 - 55 m 55 - 80 m 80 - 125 m 125 - 200 m 200 - 300 m
driving radius

Figure 8: Lean angle by driving experience

50
 On the lean angle of motorcycles driving through a curve

0-3
4-7
8-10
>10
0,75

0,5

0,25

0
0 5 10 15 20 25 30 35 40 45 50
lean angle

Figure 9: Lean angle by driving experience (without grouping by the radius)

40°
high medium low
60 f-high f-medium f-low
35°
102 60 number of evaluated runs
61 44
30° 78 77
92 42
16 15
77
25° 19 110
11
lean angle

22 80 12
20°

15°

10°

5°

0°
10 - 35 m 35 - 55 m 55 - 80 m 80 - 125 m 125 - 200 m 200 - 300 m
driving radius

Figure 10: Lean angle by risk-taking behaviour

51
 Christian Hädrich, Wolfgang Hugemann

Results References
Figure 6 shows an overall evaluation of the data
grouped by the curve categories. As predicted, the [1] Danner, M.; Halm, J.: Technische Analyse von
accepted lean angle drops with growing curve radius. Straßenverkehrsunfällen, Kraftfahrzeugtechnischer
For the higher percentiles of the larger radii this Verlag, München, 1981.
might however partially result from the speed limit
(100 km/h).
[2] Personal information given to us by Johannes Halm.
The lean angles are generally larger than postulated
in [1], especially for the higher percentiles. This had [3] Rösrather Crashtage 2010, Crash 18c.
to be expected from the technical achievements in
regard to driving stability and tyre grip.
[4] Schimmelpfennig, K.-H.; Nackenhorst, U.: Bedeutung
If we concentrate on influencing factors which can der Querbeschleunigung in der
be technically objectified, it suggests itself to group Verkehrsunfallrekonstruktion, in: Verkehrsunfall und
the data additionally by the types of motorcycles, Fahrzeugtechnik 23 (1985), pp. 94 – 96.
Figure 7. The effect of motorcycle design is however
smaller than might have been expected; sporty [5] Savitzky, A.; Golay, M. J. E.:
motorcycles hardly differ from enduros and cruisers. Smoothing and Differentiation of Data by Simplified
The older motorcycles (1980s) seem to fall Least Squares Procedures, in: Anal. Chem. 36, 1964, p.
somewhat behind the newer ones. The difference 1627 – 1639.
between street bikes and bikes from the 80s is about
6,5° over all categories. If we come to rider’s Contact
parameters, the riding experience in years is Dipl.-Ing. Christian Hädrich
probably the quality which is easiest to objectify, haedrich@unfallrekonstruktion.de
Figure 8. In contrast to the findings in [1], the Ingenieurbüro Morawski + Hugemann
influence of riding experience seems to be weak. von-Diergardt-Str. 19
This may however also be attributed to our coarse 51375 Leverkusen
grouping of the riders, trying to establish equal Germany
counts in each group. However, a finer grouping,
leaving the effects of the curve radius aside, could
still not backup the former findings, Figure 9. Dipl.-Ing. Wolfgang Hugemann
hugemann@unfallrekonstruktion.de
Although it is probably hard to objectify at court, we
can group the riders by their risk-taking behaviour, Ingenieurbüro Morawski + Hugemann
which shows significant correspondence with the von-Diergardt-Str. 19
accepted lean angle, Figure 10. 51375 Leverkusen
Germany
Conclusion
The maximum lean angle accepted by the average
motorcyclist is smaller than the physical limit and it
lowers with growing curve radius. Due to the
technical development of motorcycles, especially
their tyres, the lean angle is, however, significantly
larger than postulated in the early 1980s.
Interesting enough, the type of machine – sporty,
leisure-riding or enduro – seems to have little
influence on the accepted lean angle. It surely has
less influence than the risk-taking behaviour of its
rider.
Measuring the local curvature of a driving curve by
means of common GPS receivers proved to be more
reliable than expected, which paves the way for
simple experimental set-ups in the future.

52
© 2013 by EVU
EVU 2013 - 06

Speed drop during a Motorcycle Stoppie

Roy Strzeletz
Wolfgang Hugemann

Abstract
We present an analytical approach by means of which the speed drop during a ‘stoppie’ can be computed numerically.
This approach is checked by the evaluation of video footage from motorcycle braking tests. It turns out that the time
interval up to the final pitchover is about 0.4 – 0.7 s and that the initial deceleration must be about 11 – 13 m/s2,
depending on the geometry of the motorcycle. This results in a speed drop of about 5 – 10 m/s during a stoppie, which
may not be neglected when calculating the initial speed.

Introduction wheel. In contrast to the accidental stoppie, these

In motorcycle accidents the rider will frequently lock competitive stoppies aim on long rides on the front
up the front wheel and thereby induce tilt prior to wheel. This requires a fast lift-off of the rear wheel,
collision. This will cause scratch marks on the road- such that the unstable equilibrium can be controlled
way, which will be used when calculating the initial by minor shifts of the rider’s body centre of gravity.
speed. Very intensive braking manoeuvres will how-
ever lead to a so-called stoppie, in which the rear
wheel lifts off and the motorcycle rotates around the
front wheel standing point. This manoeuvre will
leave no skid mark, as this would instantly cause the
end of this manoeuvre. If at all, this manoeuvre will
only document in a premature break-off of the rear
wheel skip mark.
The manoeuvre as such will however document in a
significantly modified impact configuration: The
damage to the passenger car will be located extraor-
dinarily high above the ground, the motorcycle may
fly over the car, even farther than its rider. (In a
stoppie there is only the rider on the machine, other-
wise it would be impossible.) In the present paper we
will investigate the minimum speed drop during a
stoppie and we will show that its value is significant,
such that it has to be accounted for during the recon-
struction.We will first address the issue empirically
and later by numerical simulation.

Video footage
Competitions and hazards
The search term ‘stoppie’ will instantly produce a lot
of results when entered into the Youtube search field.
Besides videos that demonstrate the basic technique, Figure 1: The drop onto the pillion propergates the lift-off
there are amateur videos and complete DVDs from the rear wheel
competitive events that show the ride on the front
During the video evaluation it turns out that the fast ing the rear suspension to dive. The resulting bounce
rise of the rear wheel is achieved by a special tech- of the suspension will then assist the lift-off, which is
nique: Just before starting to brake, the rider further propagated by the rider straightening again
straightens and then falls back onto the pillion, caus- and leaning forward. Figure 1 illustrates this phe-
53
 Roy Strzeletz, Wolfgang Hugemann

nomenon which is noticeable in almost any video

from a competitive stoppie. In regard to our question
we have to conclude that the lift-off of the rear
wheel, as propagated by this manoeuvre, will be
much faster in a competitive stoppie than in an real
accident. Just after the start the angular speed around
the front wheel will rapedly drop, as the intension is
to ride on the front wheel as long as possible, avoid-
ing the pitchover.
From the video footage we determined the time span
between lift-off of the rear wheel and the cease of the
rotation. In total we could evaluate 13 competitive
stoppies based on Internet video footage, giving a
time span of 0.52 – 1.86 s, table 1. (The problem is
that the viewing angle of most videos is chosen
based on dramaturgical aspects and does not allow
for an evaluation.)

Test Zeit [s]

1 0.84
2 0.64
3 0.52
4 0.72
5 0.88
6 0.52
7 0.68
8 1.08
9 0.72
Figure 2: Accidental stoppie of an amateur (video 14 in
10 0.84
table 1)
11 0.80
12 1.48
Stoppie number 16 in table 1 resulted in a pitchover.
13 0.84
In order to allow the comparison to the other time
14 0.667
15 1.334
spans listed in the table, it was only evaluated up to a
16 0.80
motorcycle pitch angle of 90°. The total pitchover
then lasted about 1.35 s, figure 3.
Table 1: Time to reach the unstable equilibrium

Besides that we evaluated the amateur videos which

show accidental stoppies. In these videos the drop
onto the pillion is not as pronounced as in the com-
petitive manoeuvres and thus the rise times tend to
be longer, in the range of 0.667 – 1.334 s, figure 2.

54
 Speed drop during a Motorcycle Stoppie
Figure 3: Stoppie with pitchover (video 16 in table 1)
Collision experiment
Figure 4(a) shows a misscarried collision experi-
ment, in which the front wheel lock-up of the motor
scooter led to a pitchover prior to collision.
This experiment was provided by DEKRA
Automobil GmbH and AXA Winterthur (WH 08.04).
The motor scooter was propelled by a sledge up to
51 km/h and then launched, figure 4(b). By means of
high-speed video the time span needed up to the
pitchover could be determined. The markings on the
test ground also allowed to determine the
accompaning drop of the ground speed. It took 0.42 s
from the first contact of the front wheel and the
roadway up to the point where the motorcycle be-
came totally airborne, as the front wheel lifted off the
ground. The entire manoeuvre was accompanied by a
speed drop of nearly 12 km/h.
When applying this result to real accidents, we have
to keep in mind that in this test setup, the front wheel
was lifted by the sledge and then dropped onto the
test ground. This caused the front forks to dive in se-
verely, causing a higher wheel load, which in turn al-
lowed for more braking force than in a usual braking
manoeuvre. As a consequence, this test probably
marks the lower bound for the time needed for a
pitchover.
Figure 4: Miscarried collision experiment [9]
Braking tests
In an series of concluding tests we took videos from
stoppies performed by an instructor, figure 5. These
tests were conducted on dry ground at 20°C; the ini-
tial speed was about 50 km/h. The motorcycle was
equipped with an anti-lock brake system, which
forced the rider to initiate the manoeuvre very
smoothly in order to prevent the ABS from taking
control and preventing the rear wheel lift-off.
The tests were documented by full-HD video footage
from about 30 m lateral distance with 50 fps and
300 fps. The two targets placed on the front axle and
the passenger pillion could be traced in the video and
their trajectories transformed to scale by means of
the surveyor’s rods placed next to the starting point.
Translational and angular speeds could then be com-
puted from these trajectories.
55
 Roy Strzeletz, Wolfgang Hugemann
Figure 5: Evaluation of the braking tests
Figure 6(a) shows the vertical motion of the target
placed on the rear pillion, which raised from 0.92 m
to 1.32 m. In another test the rising distance was
about 0.49 m.
More interessting than the absolute height is howev-
er its progression in time, figure 6(b): Due to the
smooth start of braking manoeuvre, the rear wheel
lift-off starts very slowly. The pitch motion only ac-
celerates when the horizontal motion arroaches its
end. The maximum vertical speed reaches about
1.5 m/s.
This value is mirrored by the angular speed reaching
100 – 130 °/s in both tests, figure 6(c). Based on the
maximum value of the angular speed, the time need-
ed to reach a pitch angle of 90° computes to 0.7 –
0.9 s.
56
Figure 6: Evaluation oft he braking tests, target at the rear
pillion
Speed drop
The videos from the miscarried motor scooter test
and our braking tests allowed an evaluation of the
linear speed based on the test ground markings (mo-
tor scooter) or the targets (motorcycle), figure 7. This
resulted in a speed drop of 12 km/h for the motor
scooter and 20 km/h for
the motorcycle.
Figure 7: Speed drop in a stoppie
Theory
In order to initiate a partial pitchover, the brake pres-
sure at the front wheel must be raised gradually in
the beginning, such that the front forks spring deflec-
tion is exhausted. Only when the front wheel is stiff-
ly pressed onto the ground, the full brake pressure
may be applied. If however the brake pressure is ap-
plied too abruptly in the beginning, the front wheel
 Speed drop during a Motorcycle Stoppie
will just start to skid and the pitch motion will be
avoided.
Therefore even the accidental stoppie requires a con-
trolled rise of the brake pressure, at least in the be-
ginning. In contrast to the competitive stoppie, any
tricks that typically initiate the stoppie (dropping on-
to the pillion, straightening, leaning forward) will
surely be ommitted in an accident. In a rough analyt-
ical approach, the motorcycle and the rider can thus
be considered as a rigid body turning around the
front wheel standing point.
Static treatment
Figure 8: Static condition for rear wheel lift-off
The pitchover condition results from the moment
equlibrium around the front wheel standing point,
figure 8,
mha  msg
If we denote the motorcycle with index 1 and the
rider with index 2, we arrive at
(m1 h1  m2 h2 )a2  (m1s1  m2 s2 ) g
with the centre of gravity heights hi, the horizontal
lever arms si, the deceleration a and the gravitational
acceleration g. The deceleration threshold as, at
which the stoppie is initiated can thus be calculated
by
m s  m2 s 2
as  1 1 g (3)
m1h1  m2 h2
The value of as depends on the masses incorporated
and the geometry of the two-wheeler. For bicycles
there are actual measurements [2,1,5,7], leading to a
threshold of about 5.5 m/s2.
For sporty motorcycles with their mostly male riders
the masses are about m1 = 200 kg and m2 = 85 kg
(with clothing and helmet).
For a sporty motorcycle the centre of gravity is lo-
cated roughly in the middle between the wheels, as
acceleration and deceleration reach about the same
maximum values during normal use and both wheel-
ie (lift-off the front wheel) and stoppie have to be
counteracted. For race machines up to 57% of the
mass weighs upon the front wheel in order to prevent
the wheelie when accelerating and to counteract the
weigh shift due to the aerial drag. The rider will shift
the centre on gravity somewhat to the rear. The cen-
tre of gravity height is about 0.4 – 0.55 m for the
sheer motorcycle and will rise to 0.5 – 0.7 m when
the rider is placed on the machine [3]. For our model
system we will use the data set shown in table 2. The
pitchover deceleration thereby calculates to
11.6 m/s2.
motorcycle rider
mass [kg] 200 85
Radius of gyration [m] 0.5 0.25
CoG – front wheel [m] 0.65 0.9
CoG height [m] 0.5 0.88
Front wheel radius [m] 0.3
Table 2: Data oft he model sysem
This calculation neglected the dive-in caused by the
braking manoeuvre, which however does not greatly
affect the calculation: As the motorcycle dives in at
the front and rises at the rear and the centre of gravi-
ty lies in the middle between the wheels, its height is
hardly affected by the dive.
Dynamic treatment
(1)
(2)
Figure 9: Lever arms and firces during a stoppie
When the threshold has been passed, the pitchover
will gain speed, as the changing lever arms will
propagate it, figure 9. In a competitive stoppie the
rider will reduce the brake pressure in order to keep
the balance. The rotation around the front wheel
standing point can be described by
I  I  m(ha  sg) (4)
with I as the moment of inertia of the entire system
around the wheel standing point. At the start of the
manoeuvre this calculates to
57
 Roy Strzeletz, Wolfgang Hugemann
I  m1 (i1  s1  h1 )  m2 (i2  s2  h2 )
2 2 2 2 2 2
with ii as the radii of gyration. The lever arms si, hi
and (to less amount) also the moment of inertia I are
functions of the pitch angle, namely
si ( )  si 0 cos   hi 0 sin   r sin  (6a)
hi ( )  si 0 sin   hi 0 cos   r (1  cos  ) (6b)
I i ( )  mi [ii  xi ( ) 2  yi ( ) 2 ]
2
(6c)
and thus
1 v[km / h]
 mi [hi ( )a(t )  si ( ) g ]
 
I ( )
This differential equation is explicit in  and may be
solved – with fair accuracy for our purposes – by
simple forward step integration, which can easily be
set up in a spreadsheet [6]. Analytical solutions may
be possible for special cases a(t), especially for the
case a(t) = const. The speed drop during the stoppie
then results from a simple integration over a(t).
In accident reconstruction we are however mostly in-
terested in the extremes of a consideration, in this
special case in that time propagation a(t) which min-
imises (or maximises) the speed drop gained when a
certain pitch angle  is reached. This pitch angle
may be read from the impact configuration. This
question results in a problem of calculus of variation
[10]. It is questionable whether this can be solved
analytically. (At least this solution is far beyond the
authors’ mathematical capabilities.) However, the
numerical solution by means of spreadsheet calcula-
tion easily allows to consider various progressions
a(t).
For the computation we need the radii of gyration of
the motorcycle and the rider. The human radius of
gyration can be calculated [8, 4] by
i  0.0329  L  0.1776m
 0.0329  1.78m  0.1776m
 0.2362m
This equation however holds for the straightened na-
ked body. The riders pitch radius of gyration is de-
noted in [3] to 0.23 – 0.28 m. It seems that the
stretch of the radius of gyration by the clothing –
mainly by the helmet – and its compression by the
crouched posture counterbalance one another. The
motorcycle’s pitch radius of gyration is about 0.45 –
0.55 m [3], which corresponds to about a third of the
wheelbase and turns out to be considerably smaller
than the one that would be calculated by regression
formulae developped for passenger cars. Our for-
ward calculations will be based on a radius of gyra-
58
tion of 0.25 m for the rider and 0.50 m for the motor-
(5)
cycle.
Practical computation
The differential equation (7) will now be solved in a
spreadsheet by a simple forward step method. At the
starting point, the initial conditions are  (t  0)  0
and  (t  0)  0 . Thereby we calculate the time T
it takes to reach a certain pitch angle, namely 90°. At
the start we will consider constant decelerations, ta-
ble 3.
a[m / s 2 ] T [s]
(7)
12,0 1.49 64.4
12,5 1.22 54.9
13,0 1.07 50.1
13,5 0.97 47.1
14,0 0.90 45.4
Table 3: Simulation with constant deceleration
Unsurprisingly it turns out that the pitchover time
lowers with rising deceleration. It is however quite
astonishing that the total speed drop  v = aT also
lowers with rising deceleration, table 3.
As we draw from figure 10, a significant part of the
total pitchover time is spend on small pitch angles at
the start of the manoeuvre, before the pitchover then
accelerates for larger pitch angles. The elongation of
the manoeuvre going along with this effect then
overcompensates the lower deceleration, such that
the total speed drop rises.
(8)
Figure 10: Propagation oft he pitch angle (a = 12m/s2
=const.)
Scrutinising videos of stoppies it turns out that this
extreme slow rise of the rear wheel is not realistic, as
this effect is not that strong, even in accidental
stoppies. The problem can be located in the initial
condition  (t  0)  0 : In a real stoppie, the mo-
torcycle has an initial angular speed due to the pitch
motion at the start of the braking manoeuvre.
 Speed drop during a Motorcycle Stoppie
The natural frequency of this pitch motion is about
3 Hz [3], i. e. somewhat higher than for a passenger
car. This value will propably not be exhausted when
a stoppie is initiated, as even skilled riders will be
unable to brake exacly at the limit. If we assume that
the deceleration buildup time is at least 0.2 s and the
spring deflection is 7 cm at the front and rear wheel,
the pitch angular speed calculates to 0.5 s−1. Simula-
tions with this altered initial condition then lead to
significantly shorter pitchover times and, going
along with that, a lower speed drop during the
stoppie, table 4.
a[m / s 2 ] t[s ] v[km / h]
12,0 0.77
12,5 0.73
13,0 0.70
13,5 0.67
14,0 0.64
Table 4: Simulation with constant deceleration and
0  0,5s 1
Searching for progressions a(t) that lead to the least
speed drop we should especially consider a cease of
the brake force during the manoeuvre: By this way
of proceeding, the motorcycle rear wheel will quick-
ly lift-off, but the motorcycle will ‘profit’ from the
lower deceleration at larger pitch angles (apart from
the fact that this braking behavior of the rideris high-
ly probable in an accident). We will use a simple lin-
ear approach
t
a(t )  ao  (ao  au )
T
It turns out that the speed drop can be lowered to
28.5km/h by this (a0 = 14 m/s2, au = 5 m/s2),
figure 11.
Figure 11: Propagation oft he pitch angle ( ao  14m / s 2
, au  5m / s und o  0,5s )
2 1
Conclusions
If the motorcyle stands on its front wheel at impact,
this allows to conclude on a significant speed drop,
independently of other evidence at the spot. Our
simple simulations based on a rigid body model lead
to a speed drop of 30 km/h for 90° pitch angle. Val-
ues found on experiments are somewhat lower, about
20 km/h, whereby we have to consider that in com-
petitive stoppies the initial phase is intentionally
shortened by special manoeuvres, which will hardly
be the case in an accident.
These results are relevant, as the stoppie typically
does not leave a skid mark, as this would mean the
instantaneous end of the manoeuvre. The start of the
33.3 stoppie will, if at all, only document in a premature
32.9 break-off of the rear wheel skid mark. If the effects
32.8 investigated in this paper would be neglected, the
32.6 initial speed of the motorcycle would be significantly
32.3 underestimated.
It is difficult to address the issue experimentally,
such that our simple numerical computations should
be enhanced by more sophisticated simulations in the
future. It would also help, if EVU members would
supply thoroughly documented accident examples of
stoppies prior to collision. These would allow to
compute the actual pitchover time by the distance be-
tween the end of the rear wheel skid mark and the
impact point.
References
(9) [1] Bäumler, H. Bremsverzögerungen von modernen
Fahrrädern Deceleration of up-to-date bicycles
Verkehrsunfall und Fahrzeugtechnik 47 (2009), pp. 338 –
346 (#11).
[2] Bretting, G.; Jansen, H.; Callahan, M.; Bogler, J. et al.
Analysis of Bicycle Pitch-Over in an Controlled
Environment SAE Int. J. Passeng. Cars – Mech. Syst.
3(1): 57 – 71, 2010
[3] Cossalter, V. Motorcycle Dynamics, 2nd Ed., Lulu,
NC, USA, 2006.
[4] Döschl, R. Schwerpunktshöhe und Trägheitsmoment
des menschlichen Körpers Centre of gravity height and
moment of inertia oft he human body Dissertation,
München 1974
[5] Hittinger, C. Rollover of Mountain Bikes which are
equipped with disc brakes and suspension forks 22nd EVU
conference (2013), Florence, Italy
[6] Hugemann, W. Standardprogramme. In: Hugemann,
W. (ed.) Standard software. In: Hugemann, W. (ed.)
Unfallrekonstruktion, autorenteam, Münster 2007.
59
 Roy Strzeletz, Wolfgang Hugemann

[7] Metz, L. Road Bicycle Dynamics in the Presence of

Idealized Roadway Irregularities SAE Technical Paper
2010-01-0053 (2010).

[8] Rohm, M. Der Fußgängerunfall. In: Hugemann, W.

(ed.) The pedestrian accident. In: Hugemann, W. (ed.) Un-
fallrekonstruktion, autorenteam, Münster 2007.

[9] Crashtest WH08.04 by courtesy of DEKRA Automo-

bile GmbH und AXA Winterthur.

[10] http://en.wikipedia.org/wiki/Calculus_of_variations.

Contact
Dipl.-Ing. (TU) Roy Strzeletz
Unfallanalyse Berlin
Schönhauser Allee 10-11
D-10119 Berlin
strzeletz@unfallanalyse.de
Dipl.-Ing. (TU) Wolfgang Hugemann
IB Morawski + Hugemann
von-Diergardt-Str. 19
51375 Leverkusen, Germany
hugemann@unfallrekonstruktion.de
English version suplied by the author(s).

60
© 2013 by EVU
EVU 2013 – 07

Comparison of computer simulations and rotational

mechanics for motorcycle collisions
Charles O. Funk, Ph.D., P.E.
Christian R. Sax, P.E., PTOE
Armstrong Forensic Engineers

Abstract
Motorcycles and passenger cars and trucks have well-known weight and post-collision trajectory disparities that make
traditional reconstruction methodologies such as momentum and energy sensitive to user inputs, which may result in
erroneous reconstructions. Rotational impact mechanics and other reconstruction calculation methodologies have been
developed to address these disparities with promising results.
In this work, numerous staged motorcycle into vehicle collisions were studied and evaluated using the SI-
MON/DYMESH computer simulation models. The post-impact trajectory, rotation, and vehicle speed changes from the
simulation were compared with the measured values in the staged collisions and to the simulation results from the
EDSMAC4 simulation model validation study published by Deyerl and Cheng (2008). A sensitivity analysis was
conducted to determine the effect of speed, roadway friction, and wheel percentage lockup on the collisions.
The results showed that the use of SIMON/DYMESH in the reconstruction of a motorcycle and a passenger vehicle
collision is a valid methodology. The post-impact translation and change in heading angle for the Ford Thunderbird
using SIMON/DYMESH correlated well with the crash data. The Delta-V values for the motorcycle in the simulation
results were comparable to the published crash test results. The percent difference in the Delta-V for the Ford
Thunderbird between the simulations and the crash tests were greater than the motorcycle Delta-V. The rotational impact
mechanics methodology provided reasonable impact speeds for the motorcycle, but the Delta-V for the motorcycle and
Ford Thunderbird were notably less than the measured values and the values calculated in the SIMON/DYMESH simu-
lation models.

front of a motorcycle where the front end of the

Introduction
Collisions involving a motorcycle and a passenger
vehicle commonly involve three pieces of data: the
rest position of the motorcyclist, the rest position of
the motorcycle, and the rest position of the passenger
vehicle. Vault and Slide equations have typically
been used to solve for the impact speed of the
motorcycle based on the total throw distance of the
motorcyclist (Fricke (1990)). Traditional accident
reconstruction methodologies such as momentum
can be implemented, but due to the weight
discrepancy between the motorcycle and passenger
vehicle, the separation of the motorcyclist, and the
potential unpredictable post-collision behavior of the
motorcycle, the impact speed range for the
motorcycle can vary significantly.
Rotational mechanics has been applied to analyzing
motorcycle and passenger vehicle collisions. Daily et
al. (2006) developed equations for rotational
mechanics involving motorcycle and passenger
vehicle collisions. The derivations are listed in Daily
et al. (2006) and Keifer et al. (2007) and are not
restated here. One common application of rotational
mechanics involves a passenger vehicle turning in
motorcycle impacts the fender, door, or quarter panel
areas of the passenger vehicle. The location of
impact, inertial properties of the passenger vehicle,
weights of the motorcycle and passenger vehicles,
and the amount of rotation by the passenger vehicle
are input to determine the motorcycle impact speed,
and Delta-V for the motorcycle and passenger
vehicle. The equations to compute the motorcycle
impact speed and Delta-V for both vehicles are
restated from Daily et al. (2006), Keifer et al. (2007),
and Ogden et al (2012) as:
where:
Vm= motorcycle impact speed (ft/s) (m/s)
g = gravity (32.2 ft/s2) (9.81 m/s2)
Wmc = Motorcycle weight (lbs) (kg)

61
 Funk, Charles, Sax, Christian
Wcar = Passenger vehicle weight (lbs) (kg)
dI = moment arm from CG to impact (ft) (m)
Icar = Yaw moment of inertia for passenger vehicle
(lb-ft-s2) (kg-m-s2)
WBcar = Wheelbase of passenger vehicle (ft) (m)
= Delta-V motorcycle (ft/s) (m/s)
MC
CAR = Delta-V passenger vehicle (ft/s) (m/s)
The rotational friction, , is based on the degrees
of rotation for the passenger vehicle during its post-
collision trajectory relative to its original impact
configiration and the roadway friction. The rotational
friction versus degrees curve was shown in Ogden et
al. (2012) and Keifer et al. (2007).
Empirical relationships have also been developed to
relate motorcycle wheelbase deformation and pass-
enger vehicle crush deformation to impact speed
(Severy (1970), ARC-CSI (2008)). No comparison
between passenger vehicle intrusion, wheelbase
reduction, and motorcycle impact speed is presented
in this work. Deyerl et al. (2008) noted that
increasing or decreasing the stiffness coefficients did
not appreciably affect the post-impact trajectory and
rotation of the the passenger vehicle.
While traditional collision reconstruction techniques
such as rotational mechanics can be useful, often
there is not enough documented scene evidence to
complete these techniques. However, through
testimony, vehicle inspections, Event Data Recorder
information, and other physical evidence including
photographs, a simulation model can be used to
determine the impact speed, Delta-V, Principal
Direction of Force (PDOF), and accelerations for the
passenger vehicle and motorcycle in the subject
collision.
Engineering Dynamics Corporation offers several
collision simulation programs as part of their HVE
(Human, Vehicle, Environment) software package.
EDSMAC4 and SIMON/DYMESH are two of the
simulation packages contained within HVE. SI-
MON/DYMESH is a three-dimensional vehicle
collision simulation tool compared to the two-
dimensional collision simulation tool, EDSMAC4,
and the physics and engineering models are indepen-
dent between the two models. The validation papers
for these simulation packages and the physics and
vehicle engineering methodologies have been
published and are publically available on the EDC
website (http://www.edccorp.com). The details of the
collision models for EDSMAC4 and SI-
62
MON/DYMESH are not discussed here. The reader
is directed to Day et al. (2001), Day et al. (2004),
and York et al. (1999) for the model information
contained within SIMON/DYMESH and Day
(1999) for collision model details for EDSMAC4.
Previous investigators have evaluated EDSMAC4
and SIMON/DYMESH for motorcycle collisions.
Deyerl et al. (2008) used EDSMAC4 to simulate
motorcycles colliding with passenger vehicles and
compared the simulations with the measured crash
test results. The crash tests were conducted during
the WREX 2000 Crash Conference and published by
Adamson et al. (2002). The tests involved a
motorcycle impacting a stopped passenger vehicle at
various speeds and in various locations on the front,
side, and rear of the passenger vehicle. Deyerl et al.
(2008) concluded that EDSMAC4 provided a valid
means for analyzing motorcycle collisions with
configurations similar to the crash tests from the
WREX 2000 Crash Conference. However, Deyerl et
al. (2008) did not use SIMON/DYMESH in their
work.
Frank et al. (2012) conducted several moving
motorcycle into moving passenger vehicle crash tests
and used both EDSMAC4 and SIMON/DYMESH to
compare with the data collected during the testing. In
all of the tests, the motorcycle impacted the side of
the vehicle. They found that EDSMAC4 and SI-
MON/DYMESH provided reasonable agreement
with the test results with general agreement in the
post-impact travel speeds and trajectory. However,
the authors cautioned the readers about using SI-
MON/DYMESH due to the limited data set of
motorcycle versus passenger vehicle crash tests used
in their study.
This work addresses the need for further validation
of the three-dimensional physics models of SI-
MON/DYMESH for motorcycle-vehicle collisions.
The procedure is similar analysis to the work from
Deyerl et al.(2008) ), but uses the staged crash data
set to validate SIMON/DYMESH rather than
EDSMAC4. The SIMON/DYMESH results from the
current work are compared to the EDSMAC4 simu-
lation results from Deyerl et al. (2008) as well as the
physical data published by Adamson et al. (2002).
Additionally, a rotational mechanics comparison for
the test data is presented and compared to the simu-
lation models and measured results.
Methodology
Adamson et al. (2002) conducted seventeen staged
motorcycle collisions into a barrier and into a 1989
Ford Thunderbird during the WREX 2000 Crash
Conference. The motorcycle used in each test was a
 Comparison of computer simulations and rotational mechanics for motorcycle collisions
Kawasaki 1000. The details of the testing
methodology are described in the Adamson et al.
(2002) paper. The raw data from the conference was
obtained through the Australasian & South Pacific
Association of Collision Investigators (ASPACI).
Accelerometer data, pre- and post-collision vehicle
locations depicted in scaled CAD files, photographs,
and videos were provided. Seven tests were
conducted by propelling a motorcycle into a non-
fixed concrete barrier. These collisions are not
evaluated in this work. In addition to the barrier
collisions, there were ten motorcyle crash tests into a
passenger vehicle labeled Test 8 through Test 19. No
tests were conducted for test numbers 15 and 17. The
motorcycle impacted the passenger vehicle in the si-
de, front, and rear body structures. The distribution
of tests was:
 Side: Tests 8, 9, 11, 12, 13, 16, 19
 Front: Tests 14, 18
 Rear: Test 10
It is noted that there was an error in the diagram
created for Test 16. Based on a review of the video
and post-collision photographs, the relative
measurements were made to the incorrect side of the
vehicle. A similar observation was noted by Deyerl
et al. (2008). In this work, the diagram was adjusted
by shifting the left side of the vehicle onto the mar-
kers instead of the right side.
Several factors from the testing that are relevant to
the current work include:
 The Ford Thunderbird was stopped during
each test with the transmission in Park.
 The nominal tire-roadway friction was
measured as 0.72.
 The impact and rest position measurements
for the motorcycle and Thunderbird were
measured relative to various pre-determined
locations.
 The weight distributions for each vehicle
were measured.
As noted, only the collisions between the
motorcycles and Ford Thunderbird are analyzed in
this work. Deyerl et al. (2008) simulated this data set
using EDSMAC4, which is a commercially available
forward-projection two-dimensional collision simu-
lation package where the initial speeds, vehicle
orientation, steering inputs, throttle inputs, braking
inputs, wheel damage, stiffness coefficients, and
inertial properties are input to determine the resulting
vehicle trajectories and dynamics. In contrast, SI-
MON and DYMESH are three-dimensional vehicle
dynamics and collision programs within the
commercially available HVE simulation package
from Engineering Dynamics Corporation. Similar to
EDSMAC4, SIMON is a forward-projection simula-
tion program where the initial speeds, vehicle orien-
tation, steering inputs, throttle inputs, braking inputs,
wheel damage, three-dimensional stiffness
coefficents, and inertial properties are entered to
determine the resulting vehicle trajectories and dy-
namics. DYMESH is the three-dimensional collision
algorithm that uses methods from finite element mo-
deling in calculating collision forces. EDSMAC4
and SIMON/DYMESH use different vehicle dyna-
mics and collision algorithms to compute the
collision forces and vehicle/tire dynamics.
Additionally, SIMON/DYMESH accounts for pitch,
yaw, and roll effects as well as three-dimensional
collision effects whereas EDSMAC4 accounts only
for yaw effects and assumes a vertically homogenous
deformation profile due to the two-dimensionality of
the program.
Currently, HVE does not have a motorcycle in their
vehicle library. This issue was addressed in Deyerl et
al. (2008) by modifying a generic four-wheeled
vehicle with a track width of 0.02 inches (.05 cm)
and approximating an overall width of 18 inches
(45.7 cm). The dimensions of the motorcycle were
taken from technical data as:
 Overall length: 90.2 inches (2.29 meters)
 Overall height: 50.6 inches (1.29 meters)
 Wheelbase: 60.4 inches (1.53 meters)
 Tires: P155/80R13
The front and rear overhang for the motorcycle were
assumed equivalent.The weight distribution of each
motorcycle was measured by Adamson et al. (2002)
in each test and the horizontal location of the center
of gravity was adjusted in each test. The frontal
stiffness coefficients for the motorcycle were taken
as A=333 lb/in (583 N/cm) and B=155 lb/in2 (107
N/cm2), as calculated by Deyerl et al. (2008) from
the motorcycle/barrier crashes conducted by
Adamson et al. (2002). In this work, the pitch, roll,
and yaw moments of inertia for the motorcycle were
determined assuming the motorcycle was a
homogenous rectangular object.
The HVE vehicle database contained a Ford Thun-
derbird consistent with the vehicle used in the test-
ing. Two different Ford Thunderbirds were used in
the testing and the weight distributions for each were
documented. The frontal and side stiffness coeffi-
cients for the Ford Thunderbord were determined in
this work from NHTSA crash tests (1282 and 2218)
as:
 Front: A= 274 lb/in (480 N/cm) and B=94
lb/in2 (65 N/cm2)
 Side: A= 91 lb/in (159 N/cm) and B=48
lb/in2 (33 N/cm2)
63
 Funk, Charles, Sax, Christian
HVE contains generic stiffness coefficients adopted
from the study by Siddell and Day (1996). Deyerl et
al. (2008) used the default stiffness coefficients in
their analysis, which were similar to the calculated
values for the front and side. There was no rear end
NHTSA test for the subject Ford. Thus the default
stiffness coefficients of A= 257 lb/in (450 N/mm)
and B=82 lb/in2 (57 N/cm2) were used for the rear in
the subject work.
The dimensions and curb moments of inertia were
adopted from Expert Autostats information for the
1989 Ford Thunderbird. The curb weight and mo-
ments of inertias were input into HVE. HVE scales
the moments of inertia based on the test weights for
each vehicle compared to the curb weight. The tires
used in the simulation were P195/75R14, which were
the recommended tire size for the 1989 Ford
Thunderbird.
As noted, the Ford Thunderbirds in the subject tests
were stopped with the transmission selector in the
Park position. The Ford Thunderbird was a rear-
wheel drive vehicle. Thus, the rear drive wheels were
locked in all runs and the front steer axle wheels we-
re allowed to roll. There was no steering input
initially input for the Ford Thunderbird in the runs.
However, Tests 10 and 18 required a steering input
of 200 degrees (right) and -90 degrees (left),
respectively, to properly represent the rotation
measured in the testing.
The layout diagrams constructed for each crash test
indicated the point of impact between the motorcycle
and Ford. The following constraints were applied in
the subject work:
 A variation of 3 inches (7.6 cm) to account
for potential variability from the nominal
impact location was used.
 A friction value of 0.72 was noted for the
concrete test surface. The test methodology
to obtain the friction value was not listed in
Adamson et al. (2002). Additionally, only a
nominal value of 0.72 was provided. In the
subject work, a variability of 10%,
accounting for a friction range of 0.65 to
0.79 between the Ford Thunderbird tire and
the concrete test surface was used. A range
of friction values is commonly used in
accident reconstruction analyses and the sub-
ject range is consistent with published values
in Fricke (1990).
 The accuracy of the speed was noted to be
within 1.6 mph (2.6 kph). Thus, the nominal
motorcycle impact speed value noted by
Adamson et al. (2002) was adjusted within
the speed accuracy, if necessary, to closer
64
match the simulation post-collision
trajectory of the Ford to the test data.
The goal of the subject work is to compare the
SIMON/DYMESH simulation runs with the test data
from Adamson et al. (2002) and the EDSMAC4
work conducted by Deyerl et al. (2008). Specifically,
the collision pulse duration, post-collision rotation
angle of the Ford, translation of the Ford's center of
gravity, and the change in velocity for the motorcy-
cle and Ford were evaluated. The stiffness coeffi-
cients for the Ford and motorcycle were not
modifieded. The crush deformation was not consid-
ered in this analysis. As noted, modifying the stiff-
ness coefficients affected the crush depth, but had a
minimal effect on the rotation, translation, and Delta-
V for the Ford Thunderbird and motorcycle, even in
a three-dimensional analysis. Similar results were
found by Deyerl et al (2008) in the two-dimensional
analysis using EDSMAC4. Due to the post-collision
movement of motorcycles and their inherent instabil-
ity, matching the rest positions of the motorcycles
was outside the scope of the subject work, but the re-
sults from the simulations are presented.
The results from the SIMON/DYMESH simulations
were compared to the test data. Additionally, a rota-
tional mechanics analysis was performed for each
staged collision to analyze the motorcycle impact
speed and Delta-V values for the motorycle and
Ford. In the rotational mechanics analysis, the fric-
tion range and damage location range used in the
simulation analysis were applied. The yaw moment
of inertia for the Ford Thunderbird was given a vari-
ability of 25 lb-ft-s2 (34 N-m-s2 ) to account for the
different methodologies for determining its value.
Additionally, the post-collision amount of rotation
by the Ford Thunderbird was given a variability of
10% from the measured value. A Visual Basic macro
was written in Microsoft Excel to assess the range of
motorcycle impact speed and Delta-V values. The
range of values was compared to the
SIMON/DYMESH simulation results, the
EDSMAC4 results, and measured values from the
staged collisions.
Results
The SIMON/DYMESH simulation results are
compared to the test data results and EDSMAC4 si-
mulation results in Table 1. The rotational mechanics
(RIM) values are also shown in Table 1. The final
rest positions of both vehicles in the SI-
MON/DYMESH simulations are compared to the
field measured rest positions in Figure 1 through
Figure 10.
 Comparison of computer simulations and rotational mechanics for motorcycle collisions
Collision Pulse
The SIMON/DYMESH simulations reported a shor-
ter collision pulse duration than the field
measurements in seven of the ten tests. The SI-
MON/DYMESH simulation reported a duration
within 0.006 seconds of the field measurements in
Tests 8, 16, and 19. SIMON/DYMESH determined
the collision duration based on a 1 g acceleration
requirement to start and end the collision sequence.
The requirements for the measured data were not
stated in the Adamson et al. (2002) work to
determine the start and end points of the simulation,
which may explain the observed difference.There
was no report of the collision pulse duration in the
EDSMAC4 runs conducted by Deyerl et al (2008)
for comparison.
Ford Thunderbird post-impact motion
The SIMON/DYMESH simulated post-impact shift
in center of gravity (CG) and change in heading
direction of the Ford were compared to the field
measurements. The resultant motion of the CG was
measured, not the individual vector components. In
all the SIMON/DYMESH simulations, the post-
collision distance that the Fords center of gravity
moved from its pre-impact position differed from the
test-measured distance by an average of 4 inches (1.6
cm). The largest variation was 10 inches (3.9 cm) in
Test 18 and the smallest variation was 0.5 inches
(0.2 cm) in Test 12. As noted, the post-collision
measurements of the Ford rest position in Test 16 as
reported in Adamson et al. (2002) were not correct.
Once corrected, the heading angle and CG translati-
on of the Ford in the adjusted diagram (Figure 8) we-
re consistent with the SIMON/DYMESH and
EDSMAC4 positions and heading angles.
The change in heading angle in the SI-
MON/DYMESH simulations differed from the
measured distance by an average of 1.6 degrees. The
maximum variation was found in Test 18, where it
differed by 4.7 degrees.
In nine out of the ten tests, the post-impact shift in
the center of gravity had less variation from the
measured values in the SIMON/DYMESH simulati-
on then the EDSMAC4 simulation. The only
exception was Test 19 where EDSMAC4 was 6 in-
ches (2.4 cm) closer to the measured value. In eight
out of the ten tests, the change in heading direction
had less variation from the measured values in the
SIMON/DYMESH simulation than in the
EDSMAC4 simulation. In Test 8, the two simulation
programs differed by the same amount, and in Test
10, EDSMAC4 was nearer to the measured value by
less then 1 degree.
Impact speed
As noted, the accuracy of the speed measurements
was 1.6 mph (2.6 kph) in the work from Adamson et
al. (2002). The SIMON/DYMESH results were
comparable to the accuracy limits and the average
value for the motorcycle impact speed differed by
1.2 mph (1.9 kph) from the nominal report value.
Deyerl et al (2008) used the nominal values in their
work. If Deyerl et al. (2008) had included a speed
various in their analysis within the measured
accuracy limits, their results may have decreased the
variabiity in the EDSMAC4 values from the test data
and the SIMON/DYMESH simulation results.
Vehicle change in speed
The SIMON/DYMESH simulations resulted in an
average Delta-V for the motorcycle within 1.8 mph
(2.9 kph) of the field measurements. The SI-
MON/DYMESH simulated Delta-V for the Ford we-
re on average within 1.7 mph (2.7 kph) of the
measured values. Some of the Delta-V values for the
Ford in the simulations were greater than the
measured values and others were less than the
measured values. There was no noted trend between
the simulation data and the measured data. Although
the average Delta-V for the motorcycle and passen-
ger car were similar, the percentage difference from
the published crash test data was notably less for the
motorcycle.
In six out of the ten tests, the Delta-V for the
motorcycle had less variation from the measured va-
lues in the SIMON/DYMESH simulation than the
EDSMAC4 simulation. In the four tests where the
EDSMAC4 Delta-V was closer to the measured va-
lue, the SIMON/DYMESH value differed from the
EDSMAC4 value by a maximum of 2.1 mph (3.4
kph). In four out of the ten tests the Delta-V for the
Ford had less variation from the measured valued in
the SIMON/DYMESH simulation then the
EDSMAC4 simulation. In the six tests where the
SIMON/DYMESH Delta-V was closer to the
measured value, the SIMON/DYMESH value
differed from the EDSMAC4 value by less than 1.4
mph (2.3 kph).
Rotational impact mechanics
In order to evaluate the effects of ranging several in-
put variables in the collision scenario, the method of
Rotational Impact Mechanics (RIM) was utilized and
a range was applied to each variable as shown in
Table 2. The average result of the variation in the in-
put variables and the standard deviations are shown
in Table 1. The RIM methodology was not applied to
the frontal and rear impacts on the Ford Thunderbird
(Tests 10, 13, 14, and 18).
65
 Funk, Charles, Sax, Christian

all of the crash tests. The difference was as much as

In the remaining 6 tests, the motorcycle impact speed
results were within the accuracy of the nominal
measured speed values except for Test 16 where the
upper bound of the RIM impact speed was 5.5 mph
(8.9 kph) less than the reported impact speed. The
reasons for the innacuracy of the RIM model when
applied to the data of Test 16 have not yet been
determined.
The Delta-V for the motorcycle in the RIM
calculations was notably lower than the measured
values and the calculated values in the simulations in
46% less than the measured value (Test 16). On
average, the Delta-V was approximately 30%
underreported.
Contrary to the motorcycle Delta-V, the Delta-V for
the Ford Thunderbird in the RIM calculations were
consistent with the simulation results in Tests 8, 11,
and 12. As with the motorcycle Delta-V, the Delta-V
for the Ford was lower than the published values.
The primary reason for this is the direct scalar
relationship adopted between the Delta-V of the
motorcycle and the Ford.

Table 1: Measured test data and results of EDSMAC4 simulation, SIMON simulation, and Rotational Impact Mechanics
(RIM).

66
Comparison of computer simulations and rotational mechanics for motorcycle collisions

Table 2: Rotational Impact Mechanics (RIM) variable ranges used.

Figure 1: Measured and SIMON simulated vehicle positions – Test 8

Figure 2: Measured and SIMON simulated vehicle positions – Test 9

67
 Funk, Charles, Sax, Christian

Figure 3: Measured and SIMON simulated vehicle positions – Test 10

Figure 4: Measured and SIMON simulated vehicle positions – Test 11

Figure 5: Measured and SIMON simulated vehicle positions – Test 12

Figure 6: Measured and SIMON simulated vehicle positions – Test 13

68
Comparison of computer simulations and rotational mechanics for motorcycle collisions

Figure 7: Measured and SIMON simulated vehicle positions – Test 14

Figure 8: Measured and SIMON simulated vehicle positions – Test 16

Figure 9: Measured and SIMON simulated vehicle positions – Test 18

Figure 10: Measured and SIMON simulated vehicle positions – Test 19

69
 Funk, Charles, Sax, Christian

Conclusions
Based on a comparison with crash test data, the use of
SIMON/DYMESH in the reconstruction of a motor-
cycle and a passenger vehicle is a valid methodology.
These results build upon the validation study reported
in Frank et al. (2012) using a different, limited set of
test data.
The post-impact translation and change in heading
angle for the Ford Thunderbird using
SIMON/DYMESH correlated well with the crash da-
ta. The Delta-V values for the motorcycle in the
SIMON/DYMESH simulation results were compara-
ble to the published crash test results. The percent dif-
ference in the Delta-V for the Ford Thunderbird be-
tween the simulations and the crash tests were greater
than the motorcycle Delta-V difference.
The rotational impact mechanics methodology pro-
vided reasonable impact speeds for the motorcycle,
but the Delta-V for the motorcycle and Ford Thunder-
bird were notably less than the measured values and
the values calculated in the SIMON/DYMESH simu-
lation models. Additional study of this phenomenon
may be warranted.
References
[1] Adamson, K., et al. Seventeen Motorcycle Crash Tests
into Vehicles and a Barrier. SAE Technical Paper. 2002-
01-0551.
[2] ARC-CSI 2008 Crash Conference proceedings, Colli-
sion Magazine, Volume 3, Issue 2, Fall 2008.
[3] Australasian & South Pacific Association of Collision
Investigators (ASPACI), World Reconstruction Exposition
(WREX 2000), College Station, Texas, September 2000.
[4] Daily, J., et al. Fundamentals of Traffic Crash Recon-
struction, Vol. 2. Institute of Police Technology and Man-
agement, 2006.
[5] Day, T.D., An Overview of the EDSMAC4 Collision
Simulation Model, SAE Technical Paper 1999-01-0102.
[6] Day, T.D., and Roberts, S.G., SIMON: A New Vehicle
Simulation Model for Vehicle Design and Safety Research,
SAE Technical Paper 2001-01-0503.
[7] Day, T.D., Validation of the SIMON Model for Vehicle
Handling and Collision Simulation - Comparison of Results
with Experiments and Other Models, SAE Technical Paper
2004-01-1207.
[8] Day, T.D., and Siddall, D.E. Validation of Several Re-
construction and Simulation Models in the HVE 3-D Envi-
ronment, SAE Technical Paper 960891.
[10] Expert Autostats, version 5.2.0 © 2012.
[11] Frank, T.A., et al. Simulating Moving Motorcycle to
Moving Car Crashes. SAE Technical Paper 2012-01-0621.
[12] Fricke, L.B., Traffic Accident Reconstruction: Vol-
ume 2 of the Traffic Accident Investigation Manual,
Northwestern University Center for Public Safety, 1990.
[13] Keifer, O. et al. Linear and Rotational Motion Analysis
in Traffic Crash Reconstruction. Institute of Police Tech-
nology and Management, 2007.
[14] Neptune, J.A., Flynn, J.E., 'A Method for Determining
Accident Specific Crush Stiffness Coefficients,' SAE Tech-
nical Paper 940913.
[15] New Car Assessment Program Frontal Barrier Impact
Test - 1989 Ford Thunderbird, NHTSA Test 1282.
[16] Ogden J., Kloberdanz K., Forensic Analysis of Motor-
cycle Impacts Using Rotational Mechanics and
Fork/Vehicle Deformation, defended before the National
Academy of Forensic Engineers July 2012, peer reviewed
and pending publication in The Journal of the National
Academy of Forensic Engineers.
[17] Safety Compliance Testing for FMVSS 214 "Side Im-
pact Protection - Passenger Cars." - 1995 Ford Thunderbird,
NHTSA Test 2218.
[18] Severy, D. et al. Motorcycle Collision Experiments.
SAE Technical Paper 700897.
[19] Siddall, D.E., Day, T.D., ‘Updating the Vehicle Class
Categories,’ SAE Technical Paper 960897.
[20] York, A., and Day, T., The DyMesh Method for
Three-Dimensional Multi-Vehicle Collision Simulation,
SAE Technical Paper 1999-01-0104.
Contact
Charles O. Funk, Ph.D., P.E.
1042 North Milford Road, Suite 208
Milford, MI 48381, USA
Email: c.funk@armstrongforensic.com
Telephone: +1 (734) 730-0292
Christian R. Sax, P.E., PTOE
8500 Shoal Creek Boulevard, Suite 201B
Austin, Texas 78757, USA
Email: c.sax@armstrongforensic.com
Telephone: +1 (512) 765-9532

[9] Deyerl E. and Cheng, L. Simulation of Staged Motorcy-

cle-Vehicle Collisions using EDSMAC4. HVE White Paper
2008-3.

70
© 2013 by EVU
EVU 2013 – 08

Use of computer simulation in motorcycle traffic

accidents analysis
Senad Omerović, dr. Simon Krašna, Ana Trajkovski,
prof. dr. Ivan Prebil
University of Ljubljana

Abstract
This paper presents the use of computer simulations as the approach of determining the possible causes of various
injuries sustained by motorcyclists in traffic accidents. The main advantage of computer simulations is their ability to
reconstruct the traffic accidents in three-dimensional space with the use of mathematical models of the vehicles and
occupants involved. In this research several typical motorcycle accidents were simulated. For each of them, with the
analysis of simulation results possible injuries of the motorcyclist were determined. These simulation results can be
helpful for forensic experts in determining the cause of various impact injuries in motorcyclist victims, as well as for
improving road safety. In this research, the FEM model of the motorcycle Kawasaki ER-5, car Suzuki SWIFT and
Hybrid III dummy were used for simulating the motorcyclists’ accidents. Each of the dummy’s body segments was
equipped with an accelerometer and load cell for measuring the accelerations and impact forces during the impact phase.
Based on the literature review, several of most common motorcyclists’ accidents scenarios were determined and
simulated with the use of LS-Dyna FEM software. Simulation results were then analysed to determine the loads and
possible injuries on the individual body segments of motorcyclist. Simulation results were also used to determine
damage to the vehicles involved in the accident and to correlate it to the impact speed, which in many cases it is difficult
for forensic crash investigators to determine.

- Large truck (7%)

1. Introduction
In the past decade the number of road accidents in
European Union (EU) has decreased significantly
due to the road safety measures introduced by
European Commission. Although motorcycle safety
was also improved, motorcyclist fatalities numbers
were only decreased slightly. This is mainly
explained by the raised popularity of motorcycles in
recent years and consequently increased number of
motorcyclists on the roads [1].
Resent EU research shows that motorcyclists have
approximately a 28-times higher risk of death during
the road accidents, compared to occupants in other
types of motor vehicles. Their vulnerability is best
presented by vehicle traffic statistics where
motorcyclists make less than 1% of vehicle traffic,
but their riders suffer 14% of total deaths in road
accidents [2].
Research in EU shows that nearly 70% of
motorcycle accidents involve a second vehicle (two
vehicle accident) such as: car, truck or bus, and that
approximately 55% of those accidents occurred at
junctions. There are numerous ways of classifying
motorcycle accident, one being by the type of
vehicles involved::
- Other motorcycle (4%)
- Bus (<1%)
- Other (Unknown) (2%)
From this statistic it is clear that in two vehicle
accidents, the passenger vehicles are predominant,
comprising(86%) of all two vehicles accidents.
A recent motorcycle accident study in five European
countries found that in 37% the motorcyclist was the
cause for the accident, while in 50% the other
motorist was responsible. Of this 50%,72% are so-
called detection failures (car driver failed to see the
motorcycle), while 20% are decision errors (they
have seen the motorcycle, but decided to proceed
with intended manoeuvre as a result of
speed/distance misjudgement) [3].
In motorcycle-car accidents there are numerous
configurations in practice. In ISO 13232 [4], some
200 configurations for motorcycle-car accident are
recommended for the evaluation of rider’s protective
devices. In addition in motorcycle-car accidents it is
likely that the motorcyclist will experience a
secondary impact which in car to car accidents is
usually not the case.

- Passenger vehicles (86%) In this study simulation results for seven typical
(motorcycle-car) accidents scenarios according to
71
 Senad Omerović, dr. Simon Krašna, Ana Trajkovski
ISO 13232 are presented. The developed (finite
element method) FEM model incorporates a medium
size car, a medium size motorcycle and a male
dummy as the motorcyclist.
The study showed that the developed FEM model
can be used in determining the level of possible
injuries to the motorcyclist. Beside that, the detailed
analysis of the model also showed that the simulation
results can be easily altered with slight changes to
the motorcyclist behaviour before the impact.
Although the use of the FEM model for investigation
of motorcycle-car accidents has numerous
difficulties, it is certain that use of such a model can
Figure 1. Seven basic motorcycle-car accidents according to the ISO 13232
Each of the seven basic accidents is coded with
numbers such as “413-6.7/13.4”. The first three
digits describe the contact points for car and
motorcycle and the angle of motorcycle’s direction
of travel relative to the car axis. The digits following
the hyphen (-) are the car and motorcycle impact
speed respectively in [m/s].
2.1 Model description
The developed FEM model includes the FEM model
of passenger car Suzuki Swift, motorcycle
(Kawasaki ER-5) and 50th percentile male dummy as
the motorcyclist.
72
be useful to forensic crash investigators. Such a FEM
model can also be used for studies and research to
increase the motorcyclist’s safety.
2. Methods
For the analysis of motorcycle-car accident FEM
explicit dynamic analysis was chosen as the
simulation method with the use of LS-Dyna 971
solver. The motion of the vehicles just prior to the
collision was taken from the ISO 13232 for seven
different basic (motorcycle-car) accidents scenarios
as depicted in Figure 1.
2.2 Suzuki Swift, m=900kg
For the simulations of vehicle collisions a publicly
accessible FEM model of passenger vehicle papered
by NCAC [5] was used. Figure 2 shows the Suzuki
Swift FEM model composed of 242 component
parts. The vehicle model is comprised of 16,100
finite elements. SHELL elements are used for sheet-
metal parts, SOLID elements are used for parts (such
as engine components, brake discs, brake drums,
etc.), BEAM linear elements are used for steering
rods, DISCRETE linear elements are used for
springs and dampers.
 Problems of uncertainty in time-distance analysis of pedestrian accident

For the simulation of motorcyclist’s response during

the accident a FEM model of a 50th percentile crash
test dummy developed by LSTC [7] was used.

Figure 2. FEM model of Suzuki Swift

2.3 Kawasaki ER-5, m=200kg
Figure 4. FEM model of 50th percentile dummy
Since the usable FEM model of a motorcycle is not The dummy model presented i Figure 4 is equipped
publicly available, we opted to develop our own with accelerometers for recording the head, chest and
model showed on Figure 3. There are many pelvic acceleration during the accident. Outputs from
classification types of the motorcycles, depending on these sensors are mainly used for the calculations of:
their performance, size, use, etc.
- HIC (Head Injury Criterion),
- CSI (Chest Severity Index),
- Chest Deflection.
Beside the accelerometers, the model also includes
several force and bending moments load cells.
Outputs from these cells are used for calculations of:
- Femur Forces,
- Upper Neck Forces,
- Upper Neck Moment at Occipital Condyle,
Figure 3. Model of standard (naked bike) motorcycle - Lower Tibia Forces,
Kawasaki ER-5 - Lower Tibia Moments,
Six main categories are mostly in use: cruiser, sport - Lower Lumbar Forces,
bike, touring, sport touring, standard (naked), dual-
sport, scooter and moped. Standard naked Kawasaki - Lower Lumbar Moments.
ER-5 motorcycle was chosen as one of the most The FEM model of the standing 50th percentile crash
popular in its category among beginners and young test dummy was positioned with the use of limbs
drivers in general. The model was developed with operations in LS-PrePost 2.4 [8] into the motorcycle
the use of available literature [6] and on-site riding position showed in Figure 5.
measurements.
2.4 Motorcyclist dummy, m=78.6kg

Figure 5. Developed FEM model (ISO 13232 impact case no.”114-6,7/13,4”)

73
 Senad Omerović, dr. Simon Krašna, Ana Trajkovski

Sensor data output was set at every 0.0001s. 3. Results

Simulations time step ( dt ) was automatically For each of the seven motorcycle-car crash scenarios
calculated by LS-Dyna solver according to the simulation results were analysed and presented in
equation: form of graphs. In addition, two pictures are added,
l corresponding to the primary and secondary
dt   7, 08 104 ms ,
c motorcyclist’s impact. All of the simulations are
conducted for motorcyclist not wearing a protective
where l is the minimum finite element length and crash helmet.
c -speed of deformation wave propagation through The timeline of the kinematic and kinetic quantities
the finite element calculated in this case for SHEEL is depicted on the following diagrams. The graphs
element according to the equation: shown can be used to determine the possible injury
E to the motorcyclist using the available injury criteria.
c , where E is module of elasticity
1   
2 From the analysis of the simulation results it is clear
that the head and the neck are generally the most
and  is Poisson’s ratio. vulnerable body regions in each of the impact cases.
In motorcycle accident motorcyclists are likely to It should also to be noted that the simulation results
experience secondary impact after the primary one are rather sensitive to the input parameters, due to
with the vehicle. In generally complete analysis of the highly non-linear and abrupt dynamic response
motorcycle accident would cover three phases: of the model during the accident.
I Phase (Pre Impact Phase):
Pre Impact Phase covers the period from the
beginning of the simulation, until the first contact
between the vehicle or the motorcyclist and the
vehicle occurs. In this phase it is usual to analyse the
accuracy of simulation’s initial conditions such as:
the positions and velocities of the vehicles, predicted
pre-impact vehicle trajectories, point of impact
together with behaviour of the motorcyclist prior to
the accident (braking or no braking, avoiding
manoeuvres, time of releasing the handlebars, etc.).
II Phase (Primary Impact Phase):
This includes the period in simulation from the first
contact between the vehicles (motorcycle-car) until
the time when the motorcyclist is no longer in
contact with the vehicle. This is usually the phase
during which the biggest injuries to the motorcyclist
occur. Aggregated duration time for first and second
phase in most cases is 0.5 s.
III Phase (Secondary Impact Phase):
The third phase covers the period from the end of the
second phase, until the moment when motorcyclist’s
body stops. During the third phase secondary impact
of the motorcyclist is likely to happen (ground, guard
rails, etc.). Depending on the impact conditions and
surrounding environment, injuries in third phase can
in some cases be more severe than in second one.
The third phase proved to be difficult to accurately
predict and evaluate with the developed model, due
to its high sensitivity to the simulated initial
conditions defined in first phase. Therefore the
simulation calculation time of 0.5s that covers the
first two phases of the accident was chosen.

74
 Problems of uncertainty in time-distance analysis of pedestrian accident

Case No.1 (143-9,8/0)

Figure 6. Simulation results for stationary motorcycle impacted sidewise by car traveling at 35km/h.

75
 Senad Omerović, dr. Simon Krašna, Ana Trajkovski

Case No.2 (114-6,7/13,4)

Figure 7. Simulation results for motorcycle hitting the car’s front end at 48km/h, while the car is traveling at 24 km/h.

76
 Problems of uncertainty in time-distance analysis of pedestrian accident

Case No.3 (413-6,7/13,4)

Figure 8. Simulation results for motorcycle hitting the car sidewise at 48km/h, while the car is traveling at 24km/h.

77
 Senad Omerović, dr. Simon Krašna, Ana Trajkovski

Case No.4 (412-6,7/13,4)

Figure 9. Simulation results for motorcycle hitting the car sidewise at an angle, traveling at 48km/h, while the car is
traveling at 24 km/h.

78
 Problems of uncertainty in time-distance analysis of pedestrian accident

Case No.5 (414-6,7/13,4)

Figure 10. Simulation results for motorcycle hitting the car sidewise at an angle, traveling at 48km/h, while the car is
traveling at 24 km/h.

79
 Senad Omerović, dr. Simon Krašna, Ana Trajkovski

Case No.6 (225-0/13,4)

Figure 11. Simulation results for motorcycle hitting the left front corner of stationary car at 48km/h

80
 Problems of uncertainty in time-distance analysis of pedestrian accident
Case No.7 (413-0/13,4)
4 Discussion
Poor motorcyclist safety and increased number of
motorcyclists on the road increases the need for
usable methods for the investigation of motorcycle
accidents in order to prevent them, and to increase
motorcyclist road safety. The main advantage of
computer simulation is its ability to reconstruct
traffic accidents in three-dimensional space. FEM
enables calculation of all the relative parameters that
can give detailed insight into the behaviour of
vehicles and occupants during the accident.
In this research, the developed FEM model of the
motorcycle was incorporated into the publicly
available FEM model of the car and the crash test
dummy. For seven basic accidents scenarios, the
simulation results were processed and presented in
form of graphs, depicting the loads on motorcyclist’s
various body regions. The injury criteria index for
different body regions are automatically calculated
within the post processing software LS-PrePost [8].
Possible level injuries can be determined for
different body regions using the various injuries
criteria such as HIC36 presented in Table 1.
Table 1. Head Injury Criterion
Head Injury Level Of Brain Concussion and Head Inju-
Criterion HIC ry
135-519 Headache or dizziness
520-899 Unconscious less than 1 hour-linear fracture
900-1254 Unconscious 1-6 hour-depressed fracture
1255-1574 Unconscious 6-24 hour-open fracture
81
 Senad Omerović, dr. Simon Krašna, Ana Trajkovski
1575-1859 Unconscious greater than 25 hours-large
hematoma
>1860 Non survivable
Figure 12 depicts the probability of foot/ankle injury
based on the person’s age and the intensity of Tibia
axial force.
Figure 12. Foot/Ankle Injury Risk Curves for 25, 45 and
65 Years Old [9].
Figure 13.Injury Tolerance Curves for Axial Neck
Compression Force when Using a Hybrid III 50th
Percentile Male Crash Test Dummy [10]
Figure 13 defines risks of possible injuries regarding
the neck compression force and its duration time.
In many cases forensic crash investigators have to
determine vehicle velocities and trajectories just
before the impact. In order to solve this problem,
reverse engineering is possible using the developed
FEM model. This method is more time consuming
since it requires several simulations to be run with
different initial conditions (vehicle direction and
velocity) before optimal simulation results regarding
the vehicle damage are obtained.
82
5 Conclusion
The complexity of traffic accidents is usually
difficult to understand without the assistance of a 3D
visualisation, especially in a case of a motorcycle
accident. As was shown in this research, FEM
presents one of the computer methods suitable for
traffic accident reconstruction. It can also be used for
analysis and improvement of different road and
vehicle safety features.
In order to get realistic simulation results, every
FEM model must be validated experimentally.
Initially these tests are very expensive and time
consuming, but once the FEM model of a vehicle,
safety feature or a human is validated, it can be used
for numerous different traffic accident scenarios with
great accuracy and practical simulation results.
In the presented research, a crash test dummy was
used as the motorcyclist. This means that the
developed model basically predicts the behaviour of
the crash test dummy and not the actual human,
during the traffic accident. Additionally, the used
dummy (Hybrid III) was primarily developed and
validated for the seated position of a car occupant.
Further work will include the addition of acrash
helmet to the model, together with the
implementation of a detailed anatomical FEM model
of human neck. This would enable more insight into
the mechanism of neck injuries, which is crucial for
road safety improvements as the neck is of the most
injured region of human body in traffic accidents
[11].
References
[1] Motorcycle Related Injuries and Fatalities. Network of
Employers for Traffic Safety; 2008.
[2] David D. Clarke PW, Craig Bartle, Wendy Truman. In-
depth Study of Motorcycle Accidents. Road Safety
Research Report No 54. London: Department for Transport;
November 2004. p. 69.
[3] Chih-Wei Pai. Motorcycle right-of-way accidents—A
literature review. Accident Analysis & Prevention.
2011;43:971-82.
[4] Internations Standards Organisations. Motorcycles-Test
and analysis procedures for research evaluation of rider
crash protective devices fitted to the motorcycles.
Geneva1996.
[5] NCAC. Geo Metro Reduced Model.
[6] Kawasaki ER-5 Motorcycle Service Manual: Kawasaki
Heavy Industries, Ltd.; 2004.
[7] LSTC. Hybrid III 50th Male Standing. Livermore
 Problems of uncertainty in time-distance analysis of pedestrian accident

Software technology Corporation; 2008.

[8] LS-PrePost. 2.4 ed. Livermore, California, USA:

Livermore Software Technology Corporation; 2010. p. LS-
PrePost is an advanced pre and post-processor that is
delivered free with LS-DYNA.

[9] Yoganandan N, Pintar, F.A., Boyton M., Begeman, P.,

Prasad, P., Kuppa, S.M., Morgan, R.M. and, Eppinger RH.
Dynamic Axial Tolerance of the Human Foot-Ankle
Complex. Warrendale, PA, USA: Society of Automotive
Engineers; 1996.

[10] Mertz HJ, Hodgson, V.R., Murray Thomas, L. and

Nyquist, G.W. An assessment of Compressive Neck Loads
Under Injury-Producing Conditions. The Physician and
Sport Medicine. 1978.

[11] Vanama AOGM. Modelling of Whiplash Trauma

Parametric study of rear-end impacts using FEM and CFD.
Göteborg, Sweden: Department of Applied Mechanics
Division of Fluid Dynamics/Division of Vehicle Safety
Chalmers University of Technology; 2010.

Contact
Senad Omerović

Faculty of Mechanical Engineering,

University of Ljubljana

Aškerčeva 6,
1000 Ljubljana, Slovenia

e-mail: senad.omerovic@fs.uni-lj.si
tel: ++386 40 181669

83
 Senad Omerović, dr. Simon Krašna, Ana Trajkovski

84
© 2013 by EVU
EVU 2013 - 09

Determining Visibility Distance Based on

Measurements with LMK System
Jan Unarski, Wojciech Wach, Piotr Ciępka
Institute of Forensic Research, Kraków, Poland

Abstract
A new LMK system was used to measure the luminance of non-illuminated obstacles at night and the background. Next, a
calculation of the parameters of visibility, i.e. the threshold luminance difference, at which is necessary for an object to be visible in
the conditions of dynamic observation (i.e. driving at night-time) was performed. The calculations were done in various ways taking
into account the effect of the glare from oncoming vehicles. The results were compared with the results of other empirical tests. A
visibility study using the LMK system was proposed and an original computer program for multi-parameter calculation of predicted
visibility distance was developed. The system was found to be particularly useful in complex road-atmospheric conditions.

1. Introduction which have been covered by the so-called praxifactor

Determination of the distance from which the driver
was able to and should perceive an obstacle on the
road in conditions of limited visibility is a key
element in the reconstruction of road accidents. In
the field of the expert witness, different countries
have a number of systems to ensure establishment of
this distance. The earliest studies were based on
measuring the luminance difference between the
object and background [0], while other studies used
the definition of contrast [0] and attempted to define
for human population the so-called threshold contrast
necessary to distinguish the object against its
background. With time, more advanced theories of
perception were presented including many important
factors such as the object size or the viewer’s age [0,
0, 0, 0, 0]. Another factor was included, i.e. other
sources of light in the observation field [0, 0]. It was
proved that the minimal luminance difference
necessary for perception or contrast threshold
depends to a large extent on whether the driver was
told of the aim of the tests beforehand, and on how
long the eyes rested on the object. In actual road
traffic, the moment of the first recognition of the
obstacle (i.e. directly outside the visibility boundary)
usually indicates a short look in the initial part of
reaction time called detection. which usually lasts
about 0.2 s. This is diametrically opposed to the way
of performing laboratory tests, and even more so
from the experiment carried out at the accident
location with the participation of the parties
involved. This is the reason why in the mathematical
description of the process of detection it was found
necessary to introduce various corrective factors,
(in German called Praxisfaktor).
The strictly empirical approach was based, in turn,
on the statistical description of the results of tests
involving a large number of participants. Then, what
the expert reconstructionist could adopt as the
minimum perception distance. This was the value
that was met by a sufficiently large percentile (e.g.
90-95%) of the sampling population [0]. In these
studies a significant scatter of results was noticed
resulting from, for instance, changing vehicle body
inclination the effect of which was momentary
changes in the range of the field illuminated by the
car headlights, and the driver’s random focusing on
certain parts of the driving foreground. The large
scatter of results unfavourably affected the court of
law’s decision-taking process, often leaving two
opposing answers as to the possibility of perception.
When more advanced technologies of monitoring the
points of a tested person’s gaze (oculography) was
introduced in measurements it was possible to
investigate the real dynamics of the perception
process [0].
All these perception theories required the
introduction of measurements of luminance which is
a light parameter not easily measurable, mainly
because of the requirement of equipment sensitivity
and very small angular measuring range necessary.
Hence studies were undertaken to simplify
measurements by replacing luminance for
illumination and altering the object’s colours and
type of vehicle lights [0].
It is only owing to the latest achievements in the
sensors and microprocessors technologies was it
possible to simplify the measurement procedure

85
 Jan Unarski, Wojciech Wach, Piotr Ciępka
(improving its precision at the same time) and we
could propose a method of calculation of visibility
parameters and sighting distance. By studying the
literature on the subject for many years we have
learnt to be slightly distrustful of the formulae
published in literature, often quoted inaccurately in
subsequent papers and difficult to verify due to their
empirical character.
2. TechnoTeam LMK System
TechnoTeam have been specializing for many years
in the manufacture of equipment and software for the
measurement of light and colorimetric parameters.
With the specific nature of field measurements in
mind, the company has developed a mobile system
of luminance measurements, LMK Mobile
Advanced. Its operation is based on Canon 550 D
camera, equipped with a light-sensitive matrix
CMOS APS-C of high definition of picture
(5184x3456 pix) and high definition for luminance
measurements (2592x1728 pix). The system has a
Canon lens of focal length in the range of 70 to 200
mm, which enables convenient measurements of
objects at the distance of 100 m and more. This is
important because it is sometimes at this distance in
front of the car that an object appears, for example,
the narrow wheels of a bicycle whose luminance
cannot be measured by means of devices such as
Minolta LS-110. The system has been calibrated and
certified by TechnoTeam. Its integral part is LMK
LabSoft software for luminance measurement and
contrast calculations in various areas of the
photograph together with the possibility of averaging
the results. The system also enables determination of
illumination parameters.
In the tests EXTECH Instruments model EA31 was
also used for measurements of illuminance.
3. Aim and course of experiments
The ultimate aim of paper was to propose a relatively
easy method of measurement and calculation of
sighting distance that could be widely used in the
actual practice of road accident reconstruction. To
achieve this aim it was necessary to perform a
number of experiments the course of which has been
briefly described below.
It was assumed that it was necessary to perform
measurements for selected colours of obstacles or
86
their combinations in different meteorological
conditions (dry, wet and snow-covered road surface).
Next, the distance of likely sighting of obstacles was
calculated following Berek [0], Adrian [0] and
Carraro [0] procedures. The results of calculations
were compared with subjective impressions and the
results of previous empirical investigations [0, 0, 0,
0, 0].
It was decided that the test object would be a
pedestrian moving at a slight distance from the right-
hand edge of the road, looking in the direction of
travel of the car. His legs are an obstacle of about
0.25 m in width. This size was selected since it was
necessary to compare the results with those of other
experiments discussed in literature, where the
obstacle was also a pedestrian or a square of 0.25 m.
Next, clothing (a pair of trousers and a jacket) was
prepared in five colours (shades of grey) from white
1 to black 5. The reflection index/coefficient ρ was
measured for particular colours and the clothes made
from black velvet, which absorbed light to the
maximum, as colour 6. The results have been shown
in Table 1.
Table 1. Colours of clothing and their reflection coeffi-
cients
Colour 1 2 3 4 5 6
ρ 0.44 0.22 0.12 0.095 0.05 0.04
A straight section of road of 5 m in width with no
shoulders was selected because this is the type of
road in which pedestrians are most frequently met
walking along the road edge. The road was in a
forested area to reduce the effect of other light
sources to the maximum. Three types of road surface
were selected: dry, wet and snow covered,
representing different reflection coefficients.
The illumination source were the lights of Renault
Laguna (one year old), of ellipsoidal lens, adjusted
correctly.
Following the procedure of the LMK system,
photographs were taken with a camera placed within
the car, in the position of driver’s vision. For each
obstacle a series of photographs was taken from the
distance of 120 m to 10 m in front of the car every
ten metres. Each photograph contained a set of
measurement which was processed in LMK LabSoft,
which automatically separated areas of the same
luminance (Figure 1).
 A new LMK System for Measuring the Luminance and Determining a Visibility Distance

Figure 1. Measurement in LMK LabSoft (colour of clothing 4, distance 40 m)

Measurements were taken for colours 1 to 6 and

combinations:
 trousers colour 5 + jacket colour 1,
 trousers and jacket colour 5 + shoes colour 1,
 trousers and jacket colour 5 + hat colour 1,
 trousers and jacket colour 5 + colours scale.
The cumulative results together with visibility
distances for praxifactor 5 and 10 have been shown
in Table A, B and C in the Appendix.
4. Methods of visibility assessment
4.1. Graphic method after Berek’s curves (Din
5037)
Berek’s curves are usually presented in a logarithmic
system of coordinates and they indicate the
luminance difference, required for visual perception,
between the object and its background, depending on
the object’s angular size given in ['] (arcminute).
They enable graphic determination of visibility
distance.

87
 Jan Unarski, Wojciech Wach, Piotr Ciępka
Two values of the praxifactor, 5 and 10, have been
considered. The angular size of the object is
calculated based on the object’s distance from the
viewer. The X-axis on top of the plot indicates this
particular distance in [m].
In Figure 2 the plotted visibility distance for a dry
road and pedestrian wearing clothes in colour 4 has
been shown. The value of luminance of the
background and object was read by LMK LabSoft.
For the same object (colour 4) visibility distance on a
snow covered road has been plotted in Figure 3.
From a comparison of visibility distance for various
road conditions it follows that in the case of a snow
covered road, an object is visibilefrom a slightly
smaller distance than in the case of dry asphalt road.
This paradox can be accounted for by the fact that
pupil’s diameter diminishes in the conditions of
higher background luminance (high coefficient of
Figure 2. Determination of distance of sighting a pedestrian wearing clothes in colour 4 on a dry road (44–52 m)
88
light reflection from snow), and as a result the
amount of light reflected from the obstacle that
reaches the eye is reduced.
4.2. Calculation methods after Adrian and Carraro
Software, of a working name Marlo, for calculating
object sighting distance by means of two methods
called Model 1 and Model 2 was developed. The
Models are based on theoretical-empirical
relationships presented by Adrian [0] and Carraro
[0]. Since the set of data in both cases is very similar,
certain generalisations have been made, a common
table for introducing data into the model is shown in
Figure 4.
 A new LMK System for Measuring the Luminance and Determining a Visibility Distance

Figure 3. Determination of distance of sighting a pedestrian wearing clothes in colour 4 on a snow covered road (34–49 m)

Figure 4. Table for introducing data in Marlo software

In both cases it was accepted, after Berek [0], that

the luminance threshold difference required for an (1)
obstacle to be sighted is given by formula
where:

89
 Jan Unarski, Wojciech Wach, Piotr Ciępka

– luminous flux function, characteristic for the Effect of glare

Ricco-process,
L – luminance function reflecting Weber’s law,
f – factor (i.e. a constant covering coefficients of
precise names mentioned below),
– obstacle’s angular size in ['] (arcminutes).
Auxiliary functions and in the formula (1)
are described for three ranges by empirical formulae
quoted in [0], which depend on background
luminance Lb [cd/m2]. The conditions of obstacle
observation are included in factor f. The way of
calculating it is a basic difference between Models 1
and 2.
4.2.1. Model 1 – Adrian
In case the of no glare
(2)
was adopted. Factor f is given by formula
Glare of the observer by outside light sources is
described using the so-called veiling luminance.
(5)
where:
j – number of glare sources,
– viewer’s age factor (given by a formula other
than ) and valid for the range of 25 to 80
years of age,
– angle of glare source from the vision fixation
in [°], which should lie within the range of
1.5° <  < 30°.
Following [0], to take glare into account it is
sufficient to use background luminance enlarged by
veiling effect in all the previous formulae (related to
visibility with no glare)

(3) . (6)

with Since in this case there are large differences

between calculation results after Adrian and Carraro,
(4) the problem of glare needs further analysis.

making it possible to include the effect of 4.2.2. Model 2 – Carraro

observation time, negative contrast and age (covered
by , and respectively). Regardless of glare or its lack

Effect of observation time (7)

For road traffic conditions observation time t = 0.2 s
can be adopted. For t > 2 s, in practice, observation was adopted. Factor f is given by formula
time does not affect the value of , or luminance
threshold difference . , (8)

Effect of negative contrast where:

K – coefficient of sighting probability; K = 3.1 for
To take into account the difference between for
nearly 100% of sighting likelihood (such value
positive and negative contrasts the so-called contrast was adopted in the software),
polarity factor .was introduced. In the case of
C – praxifactor, which according to various
positive contrast, i.e. , Otherwise,
authors ranges from 4 to 40; others indicate
empirical dependencies as functions of background
the range of 3–10; in laboratory conditions
luminance and angular size of obstacle are
valid. C = 0, Carraro recommends for an unexpected
obstacle in road traffic the value of C = 10.
Effect of age From the calculations done with the software
it follows that for the results reached in
Factor of viewer’s age is calculated in three
Models 1 and 2 to be (almost) identical, C
ranges: under 23 years, 23–64 years and 64–75
years. should be about 3,
– threshold increment related to the case of
glare.

90
 A new LMK System for Measuring the Luminance and Determining a Visibility Distance

Effect of glare In conclusion, comparing the three methods of de-

The effect of glare is covered only by , which in-
creases the threshold value of luminance indispensa-
ble for an obstacle to be detected . When there is
no glare , in case of glare and is ex-
pressed by the formula dependent on veiling lumi-
nance and mean luminance of the surroundings
of the area within the angle (conical) ±30° around
termining visibility distance it can be stated that the
shortest distance is found with the graphic method
based directly on Berek’s curves (44–52 m), while
the calculations after Adrian and Carraro models
give the visibility distance in the range of 58–59 m.
a)
1

the line of vision in [cd/m2]. .was

0,1
adopted.

Luminance [cd/m2]
0,01
4.2.3. Calculation of visibility range
The distance from which an obstacle can be noticed 0,001
is defined by comparing the threshold luminance Actual dL=|Lo-Lt|
with with the real luminance : 0,0001
Threshold dLp, Model 1
Threshold dLp, Model 2

obstacle is invisible, 0,00001

(9)
obstacle is visible, 0 20 40 60 80 100
Distance to object (without glare) [m]
on condition b)
0,6
. (10) 0,5
Visibility curve kw

0,4
When a series of measurements is performed for Model 1
0,3
objects shifting in subsequent points (see Table 1), it Model 2
is most convenient to check this condition by the so- 0,2

called visibility curve 0,1

0
(11) -0,1
0 20 40 60 80 100
where: i – number of measurement. The simplest Distance to object (withhout glare) [m]
answer results: Figure 5. Determination of visibility distance by compar-
ing the differences between actual and threshold lumi-
obstacle is invisible, nance, represented by: a) luminance, b) visibility curves
(12)
obstacle is visible.
Visibility distances determined in such ways can be
compared with the photographs taken at the tests lo-
4.2.4. Results of theoretical calculations
cation (colour 4, distances 60 m, 50 m and 40 m),
shown in Figure 6.
To provide an example, using the measured lumi- On the basis of their experience in studying obstacle
nance of the obstacle and background, theoretical sighting, the authors of the present paper would
sighting distances were calculated for an obstacle in adopt the results determined by the Berek‘s curves
colour 4. based graphic method as more reliable.
The sighting distance calculated after Models 1 and 2 Cumulative results of obstacle sighting distances
is ca. 59 m and 58 m, respectively. In Figure 5a it is thus obtained have been given in Table 2.
shown at the spot where the threshold luminance
curve intersects the curve of the actual lumi-
nance difference (determined in measurements at Table 2. Results of visibility measurements
different distances of the vehicle from the obstacle).
In Figure 5b the object is visible in the section Colour 1 2 3 4 5 6
.
ρ 0.44 0.22 0.12 0.095 0.05 0.04
The two distances were calculated for a viewer 50
years of age and praxifactor C = 7.

91
 Jan Unarski, Wojciech Wach, Piotr Ciępka

dry
77-69 68-63 58-52 55-48 53-46 42-33 results in light intensity increase in the range of 0.2
m m m m m m
to 0.8 lx.
64-56 49-38 39-31 49-34 50-34 51-28 The calculations point to sighting distance of ca. 53
snow
m m m m m m
m and 44 m after Models 1 (Adrian) and 2 (Carraro),
63-55 41-37 27-24
wet – – – respectively, which has been shown in Figure 7.
m m m
Notice: longer distances for C = 5, shorter for C = 10
a)
1

Luminance [cd/m2]
0,1

0,01

0,001

0,0001 Actual dL=|Lo-Lt|

60 m
0,00001
0 20 40 60 80 100
Distance to object (with glare) [m]
b)
0,6
0,4
Visibility curve kw

0,2
0
-0,2
50 m
-0,4
-0,6 Model 1
-0,8 Model 2
-1
0 20 40 60 80 100
Distance to object (with glare) [m]
Figure 7. Visibility distances for an obstacle in colour 4
with theoretical effect of an oncoming vehicle considered,
represented by : a) luminance, b) visibility curves
40 m
This drop can be expressed as 11–24%, which seems
Figure 6. Photographs of an obstacle in colour 4 from the to be an expected value, compatible with the data in
distance of 60 m, 50 m and 40 m the literature.

It should be stressed that in the case of three darkest
colours their sighting distances for C = 5 are even
greater than for an object in colour 3, which points to
a strong effect of better perception in negative con-
trast (dark clothing light background).
4.2.5. Influence of oncoming vehicles.
The situation of glare with the lights of an oncoming
vehicle was preliminarily included in the software.
However, we have not verified it in experiments yet
(apart from some general cases).
It was found that the oncoming vehicle in the area of
the viewer’s face, an increase of the intensity of light
entering his eyes increases as the oncoming vehicle
approaches. The change in the distance between the
vehicle and the viewer in the range of 100 to 40 m
92
5. Other observations resulting from tests
During the tests it was noticed that the pedestrian
wearing a combination of colours, i.e. the trousers in
colour 5 (black) and a jacket in 1 (white), was per-
ceived from a distance similar to the sighting dis-
tance of an object in colour 3. This can be accounted
for by the jacket’s large angular size (improvement
of sighting) on the one hand, and its lower illumina-
tion than of white trousers due to its higher location
on the other hand. The pedestrian wearing clothes in
colour 5 (black) but wearing white shoes was detect-
ed from a large distance, not much smaller than the
pedestrian all in white. This results from the fact that
light shoes (white) (colour 1) are located right by the
ground, i.e. where the vision and vehicle lights fall.
Moreover, the pedestrian’s feet are in bio-motion.
 A new LMK System for Measuring the Luminance and Determining a Visibility Distance
The final observation relates to a pedestrian in black
(colour 5) and wearing a white cap (colour 1). Unfor-
tunately, the cap did not significantly improve the
visibility distance (only 15–25%) because it was
small, located high, i.e. far from the light flux from
the headlights and appeared in a place unexpected by
the viewers.
It was also found that the most difficult to see is
standard black clothing, while the black clothing
which is even darker as it is made from velvet ab-
sorbing light, on a dry road is visible from a larger
distance because the deep black creates a stronger
contrast with the grey roadway.
The tests on a snow covered road showed that the
visibility distances for three darkest colours (4, 5 and
6) were close to one another and the objects were
visible to the viewers at almost the same time. It
seems that this is strongly affected by the glare expe-
rienced by the driver, caused by his headlights being
reflected against the white surface of snow. In the
conditions of the snow covered road the road itself is
very clearly visible.
6. Comparison with results of other stud-
ies
The ranges of visibility distances on a dry road ob-
tained by R. Kledus [0] were 29 to 41 m for the
darkest clothing (equivalent to colours 5 or 6), while
for light clothing (colours 2 or 3) was 47–61 m. This
makes them basically consistent with the results we
have obtained in our tests and calculations.
Similar studies were performed by Olson in 1983
[0], but he used a statistically large number of view-
ers. For the European system of lighting and bulbs
H4 he gives in [0] a diagram of distances of sighting
a pedestrian in dark clothing and standing on the
right-hand side of the road. In fact it corresponded to
dark-blue denim trousers and a black T-shirt (Figure
8).
Figure 8. Visibility distance of a dark-clad pedestrian with
percentage of population, after [0]
The visibility distance consistent with that reached in
our study is found in about 50-70% of test partici-
pants, which corresponds to praxifactor C = 5, and
the distance below 30 m is found in ca. 10% of the
participants (C ≥ 10).
7. Further research
In view of the fact that our tests are rather arduous
since it is necessary to arrange an appropriate place
and combine it with appropriate weather conditions,
not to mention holding up the traffic, and the time of
performing them, the authors reserve the right for
their results to be treated as preliminary. However,
our results have proved a general consistency with
the results obtained in other research centres. We
hope to complete our work soon and implement the
LMK system in determining the visibility distances
in difficult conditions, e.g. when obstacles appear in
parts of roadway of different luminance (e.g. a wet
road in a city with reflected) light).
In the case a standard pedestrian plus an illuminating
car, it seems sufficient to use cumulative tables or
test results related to sighting distance.
Notice:
These results are valid only for the headlights and
road used in the tests. On other conditions results
may vary, therefore using LMK System makes pos-
sible to obtain appropriate results.
References
[1] Berek M., Zeitschrift für Instrumentenkunde 1943,
63 297 (in German language).
[2] Blackwell H.R., Contrast Thresholds of the Human Eye,
Journal of the Optical Society of America vol.36, No 11,
Nov 1946.
[3] Aulhorn E., Graefe’s Archiv für Ophthalmologie 1964,
167(1) 4 (in German language).
[4] Carraro U., C13 Dunkelheitsunfälle, [in:] Burg H., Mo-
ser A. [ed.], Handbuch Verkehrsunfallrekonstruktion, 2.
Auflage, pp. 875–889, Vieweg + Teubner Verlag, Wiesba-
den 2009 (in German language).
[5] Eckert M., Lichtechnik und optische Wahrnehmungs-
sicherheit in Strassenverkehr, Verlag Technik GmbH, Ber-
lin-München 1993 (in German language).
[6] Kosmatka W. J., UDC – Uniform Detection Character-
istic for Detecting Roadway Obstacles, SAE Technical Pa-
per Series 2006-01-0948.
93
 Jan Unarski, Wojciech Wach, Piotr Ciępka
[7] Olson P. L., Forensic Aspects of Driver Perception and
Response, Lawyers & Judge Publishing Co, Tuscon, 1996.
[8] Kledus R.,Bradac A., Semela M., Comparative percep-
tion of objects by drivers from stationary and moving vehi-
cles in regular road traffic, Proceeding from EVU Annual
Conference, Prague 2010, pp. 9 – 28.
[9] Muttart J. W, Kauderer Ch., Unarski J., Barlett W,
Johnston G., Barshinger D., Romoser M., Determining
when an object enters the headlight beam pattern of a vehi-
cle, SAE Technical Paper 2013-01-0787.
[10] Olson P. L., Sivak M., Improved Low Beam
Photometrics, Transportation Research Institute, Ann Arbor
Michigan 1983.
[11] Unarski J., Wach W., Zębala J., Świder P., Pilotażowe
badania widoczności przeszkód w światłach mijania w
warunkach jezdni ośnieżonej (Discernibility of non-lit ob-
stacles by car low beam lights on snow covered road. Re-
sults of pilot experimental research), [in:] Zbiór referatów z
VIII Konferencji “Problemy rekonstrukcji wypadków
drogowych”, Institute of Forensic Research Publishers,
Kraków 2002, 309–318 (in Polish language).
[12] Unarski J., Zębala J., Wach W., Establishing Visibility
Distance at Night, [in:] The Institute of Traffic Accident In-
vestigators, Proceedings of the 3rd National Conference,
1997, Telford (UK), 142–148.
[13] Adrian W., Visibility of targets: Model for calculation,
Lighting Research and Technology 1989, Vol. 31(4), 181–
188.
Further literature
[14] Farber E. Revising the DETECT seeing distance model
– Paper no. 880716, SAE Society of Automotive Engineer-
ing, Inc. 1988.
[15] Hugemann W.(Hrsg.) – Unfallrekonstruktion, Auto-
renteam GbR 2007, ISBN 3-00-019419-3.
[16] Wördenweber B., Wallaschek J., Bozce P., Hoffman
D.D., Automotive Lighting and Human Vision, Springer
Verlag 2007, ISBN 978-3-540-36696.
[17] Gramberg-Danielsen B., Hartmann E, Giehring H.,
Der Dunkelheitsunfall, Ferdinand Enke Verlag 1984, ISBN
3 432 94411 X.
[18] Allen M.J., Abrams B.M., Ginsburg A.P., Weintraub
L., Forensic Aspects of Vision and Highway Safety Law-
yers & Judge Publishing Co, ISBN 0-913875-24-4.
[19] Rivers R.W. (Bob), Hochgraf F. G. Traffic Accident
Investigators’ Lamp Analysis Manual, Charles C. Thomas
Publisher LTD, ISBN 0-398-07134-9.
94
[20] MacInnis D. D., Williamson P. B., Nielsen G. P. Sun
Position and Twilight Times for Drivers Visibility Assess-
ment, SAE Technical Paper Series, No. 959359.
[21] Schmedding K., Becke M., Das SBU-diagram und die
Bestimmung von Erkennbarkeitsentfernungen mit Hilfe des
SI-Diagrams, Verkehrsunfall und Fahrzeugtechnik 1/1990.
[22] Ising K. W., Fricker T.C., Lawrence J. M., Siegmund
G. P., Threshold Visibility Levels for the Adrian Visibility
Model under Nighttime Driving Condition, SAE Technical
Paper Series, No. 2003-01-0294.
[23] Schmedding K., Erkennbarkeitsenfernung von Fuss-
gänger unter besonderer Berücksichtigung von Gegenver-
kehr, Verkehrsunfall und Fahreugtechnik 9/1992.
[24] Schmedding K., Die Bedeuntung des Praxisfaktors in
der lichtechnischen Analyse: Der versuch einer Eingren-
zung, Verkehrsunfall und Fahreugtechnik 7-8/1992.
[25] Ekrias A., Eloholma M., Halonen L., The contribution
of vehicle headlights to visibility of targets in road lighting
environments, International Review of Electrical Engineer-
ing, vol.3 N. 1, pp.208-217, 2008.
[26] Flannagan M.J., Sivak M., Traube E.C., Kojima S., Ef-
fect of overall low-beam intensity on seeing distance in the
presence of glare, UMTRI-96-26, July 1996.
[27] Roze C.,Maheu B., Grehan G.,Menard J., Evaluation
of the sighting distance in a foggy atmosphere by Monte
Carlo simulation, Atmospheric Environment Vol.28, No.5,
pp.769-775. Pergamon 1994.
[28] Paulmier G., Carta V., The luminance of daytime fog:
Visibility alteration for the motorist, Bulletin des Laborato-
ries des Ponts et Chaussees, Mars-Avril 2003 – Ref. 4477,
pp.71-81.
Contact
Jan Unarski, PhD
Institute of Forensic Research
ul. Westerplatte 9
31-033 Kraków, Poland
e-mail: junarski@ies.krakow.pl
Phone: ++48/12 61 85 763
© 2013 by EVU
EVU 2013 - 10

Driving simulator analysis of drivers’ behaviour in

tunnels with obstacles
Lorenzo Domenichini, Francesca La Torre, Dario Vangi, Valentina Branzi, Carlo Cialdai,
Monica Meocci, Antonio Virga
University of Florence

Abstract
The behaviour of drivers on a highway track was studied, in particular when meeting an unexpected obstacle, consisting
on a stationary car inside a tunnel. The research was carried out using the newly installed full scale dynamic driving
simulator at the University of Florence. For this purpose a special virtual environment was produced, based on an actual
Italian highway, also simulating two types of tunnel lighting: LED lights and sodium vapour lights. Twenty-one young
subjects voluntarily participated to this experiment and, in the two scenarios, 36 tests were completed. The tests showed
some differences in the manoeuvres the subjects performed to avoid the obstacle. A statistically significant difference
was revealed as a function of the type of lighting, simulated LED lights often inducing a better driving behaviour in
several respects.

a driving simulator allows the assessment of the rela-

Introduction
Driving simulators are powerful tools that allow the
experiment to reproduce dangerous and difficult sit-
uations, without any danger for subjects and opera-
tors, in controllable and repeatable virtual scenarios.
In past research [1, 3, 5] simulators were usefully
employed to test many aspects of real driving, such
as drivers' behaviour under different conditions, in-
frastructure and environment influences on road
safety, reaction time, distraction, etc. The usefulness
of using more realistic, full scale simulators was
evaluated [4], as well as the difference in driver’s
behaviour in real and simulated environments [6].
The research here presented aims at evaluating, by
means of an advanced dynamic driving simulator,
the driving behaviour of road users when facing an
unexpected obstacle. In particular attention was fo-
cused on reaction time analysis and on the emergen-
cy manoeuvres that drivers carry out in the moment
of perception of an unexpected event inside highway
tunnels. Driving simulations were carried out in both
simulated highway tunnels with usual sodium vapour
lighting and tunnels with LED lighting. This latter
type of illumination is particularly promising and its
use is spreading quickly because of its low energy
consumption and for the possibility of modulating
light intensity depending on several other factors,
such as external light, traffic intensity, etc. Nonethe-
less research and comparisons with usual lighting are
still needed, in order to demonstrate the effectiveness
of LED lighting as regards to road safety. The use of
tionship between infrastructural geometry, how it is
used and the behaviour it induces on the users in the
peculiar environment (for instance) of tunnels, with-
out creating danger for the drivers and by exploiting
the important contribution that simulation virtual re-
ality techniques can offer while studying human fac-
tors.
The goals proposed were pursued by means of an
experimental campaign carried out with volunteers
on the dynamic full scale driving simulator recently
installed (2012) at the University of Florence. Twen-
ty one subjects were enrolled for the purpose, select-
ed among young people with enough driving experi-
ence. They were asked to drive on a highway track,
in which an unexpected hazard was placed inside a
tunnel: an obstacle was placed in the first lane and
consisted of a stationary vehicle preceded by an
emergency triangle.
The simulator
The driving simulator used in the experiment here
described, is the model AS 1200 (figure 1), supplied
by AutoSim (Norway), and it is running on the soft-
ware SimWorld version 2.8.2. It is a full scale, dy-
namic simulator, with a complete Lancia Ypsilon
cabin installed on a 6 axes Stewart platform, capable
of reproducing the typical movements of a travelling
vehicle and the forces caused by road and environ-
ment. The vehicle interior is identical to the com-
mercial version and includes all the controls normal-
ly available in these cars, with steering wheel with
95
 Lorenzo Domenichini, Francesca La Torre, Dario Vangi, Valentina Branzi, Carlo Cialdai, Monica Meocci, Antonio Virga
force feed-back. The cabin is surrounded by a cylin-
drical screen about 200 degrees wide on which 4
projectors (with resolution of 1920x1200 pixels) re-
produce the driving environment; rear mirrors are
replaced by 6.5” LCD monitors. Sounds and noise
are generated by a multichannel audio system. All
functions are supervised by a network of 5 comput-
ers, including operator’s station from which the sim-
ulation is managed.
Figure 1 - A detail of the driving simulator AS 1200.
The terrain
For the purpose of this experimentation, a special
virtual road terrain was set up, including a graphical
3d database (made with 3D Studio Max) and a tech-
nical database (made with Road Tools, supplied by
AutoSim). These databases contains all graphical,
geometrical and technical information regarding the
terrain on which the tests will take place. More spe-
cifically, a 18 km section of southward A1 Italian
highway near Florence was reproduced; the section
includes four open road sections and four tunnels.
The carriageway in the open section comprises three
lanes plus an emergency lane; on the tunnel seg-
ments only three lanes are present, without emergen-
cy lane. in the first two tunnels a LED lighting was
simulated, while the last two had sodium vapour
lights. Particular care was taken in reproducing the
tunnel environment, with signals, emergency exits,
ventilation and illumination systems, etc. (figure 2).
In the second tunnel of each pair of tunnels, an ob-
stacle was inserted; it consisted on a vehicle stopped
96
on the first lane, preceded by an emergency triangle
at a distance of 100 m, put halfway into the first lane
(figure 3). Triangles and vehicles were rendered with
the proper tunnel lighting. Thus two terrains were
set-up: terrain A, with the obstacle put on the second
tunnel (with LED lighting) and terrain B, with the
obstacle on the fourth tunnel (with sodium vapour
lighting).
Figure 2 - The highway tunnel 3D model.
Figure 3 - The emergency triangle and the stationary
vehicle placed in the tunnel with LED lighting.
Participants
The subjects' sample was made by selecting among
young people (most of them were actually university
students), with at least three years driving experi-
ence; drivers were selected after having filled in a
preliminary form, containing biographical and medi-
cal questions. Twenty one subjects were selected, 2
female and 19 male, with a mean age of 25.3 years
and having owned a driving license for 7.7 years.
Testing scenario
Two scenarios were set-up starting by the terrains
already described: one using terrain A (scenario A),
 Driving simulator analysis of drivers’ be-haviour in tunnels with obstacles
and one using terrain B (scenario B). Both the
scenarios were characterszed by autonomous traffic,
made with 10 different vehicles, organized as
“swarm” around the interactive vehicle; the swarm is
a technique aimed at having enough traffic even with
few vehicles travelling in the scenario: as soon as a
vehicle exits from a circle centred on the interactive
car, whose radius is selectable, it is immediately
made enter the circle again, so that the few vehicles
are equivalent to a larger number. To avoid
excessive traffic when approaching the obstacles, the
swarm was switched off in those positions. In
addition, cars belonging to the swarm were forced to
change lane (moving to the next free lane on the left)
to avoid hitting the obstacle or forming a queue
behind the obstacle itself. The scenario was
programmed to acquire and save to file the following
values, with the sampling time set to 0.05s:
- relative time (s);
- position coordinates X and Y (m);
- offset from the centre of the lane (m),
negative number indicating leftward offset;
from this signal the distance from the centre
of the first lane was then calculated;
- speed (m/s);
- position of pedals (accelerator, brake and
clutch, 0 to 1);
- wheel position (radians);
- gear engaged;
- acceleration along X and Y axes (m/s²).
Half of the subjects were put through scenario A
(with the obstacle put in the second tunnel, with LED
lighting) first followed by scenario B (the obstacle in
the fourth tunnel, with sodium vapours lighting), for
the other half of the participants the scenarios were
reversed, to allow for the influence of the order in
which the scenarios are used.
Testing procedure
Before sitting in the car, each subject was given
some basic information on the use of the simulator
and the behaviour to maintain during the test; in par-
ticular they were asked to wear safety belt and drive
normally, trying to respect driving regulations, espe-
cially in regard to the use of lights and highway driv-
ing. The subjects were warned about simulator sick-
ness and told they could stop the test in any moment.
Before the beginning of the test, a training scenario
was submitted, aimed at making the subject ac-
quainted with the simulator and its behaviour, espe-
cially as regards the use of pedals, steering wheel,
platform motion (during curves, acceleration and
braking) and the use of commands, instrumentation
and sound. This training was carried out on a stand-
ard highway terrain, with moderate traffic; it was be-
gun with daylight and was gradually shifted to dark-
ness by changing the time of day. The training lasted
about 10 minutes. After the training course was suc-
cessfully completed, the subject was asked to get
down from the cabin and fill a post-training form; af-
terwards the particpant underwent the first testing
scenario (driven always by night). After its comple-
tion, the second scenario was loaded; at the end the
subject was asked to fill a post-test form. The sub-
jects were unaware of the aims of the test, so that the
obstacles (at least the first one), were unexpected.
Data elaboration
From the data acquired the distance from the obsta-
cle was calculated as a function of time, the moment
in which the vehicle reached the obstacle was taken
as new origin of the axis of time. A subset of the data
set was then extracted. In the defined time interval it
was generally possible to detect three different types
of driver's reaction:
- reaction on the accelerator pedal (figure 4),
when a sudden reduction of its activation oc-
curs;
- reaction on the brake pedal (figure 4), when
a sudden rise of the signal is detected;
- reaction on the steering wheel (figure 5),
when a sudden steering manoeuvre (usually
leftward) is detected; as a consequence, also
a lane change is generally detectable by ob-
serving the offset signal.
1,0
0,9
accelerator
0,8
brake
pedals position
0,7
0,6
0,5
0,4
0,3
0,2
0,1
0,0
-12 -10 -8 -6 -4 -2 0
relative time (s)
Figure 4 - Identification of reaction on accelerator and
brake while approaching the obstacle.
97
 Lorenzo Domenichini, Francesca La Torre, Dario Vangi, Valentina Branzi, Carlo Cialdai, Monica Meocci, Antonio Virga

0,50 1,5 reaction was detected and a collision occurred, all

0,40 wheel position 1,0 those who had no reaction on the accelerator pedal,

position on the road (m)

0,30 position on the road 0,5
hit the obstacle. Even when it was not strictly neces-
wheel position (rad)

0,20 0,0

0,10
-0,5 sary, because the vehicle was travelling in the fast
-1,0 lane, a reaction on accelerator pedal was detectable.
0,00
-1,5
-0,10
-2,0
In scenario B, the driver hit the obstacle in spite of a
-0,20 -2,5 reaction (on accelerator and brake or steering wheel)
-0,30 -3,0 in three cases.
-0,40 -3,5
Some of the drivers chose not to brake but steered
-0,50 -4,0
-12 -10 -8 -6 -4 -2 0
only, having already reacted on the accelerator; it
relative time (s) seems that this was done because it was necessary,
the participant having deemed themselves to be too
Figure 5 - Identification of reaction on steering wheel and close to the obstacle to attempt to brake; there is in
subsequent position on the road while approaching the fact a statistically significant difference in d1 for the
obstacle. two groups (those who chose to brake and those who
chose otherwise, figure 7): mean = 63.9 and SD =
In each instant in which a reaction is detected (the 21.9 for the first group, mean = 44.3 and SD = 15.0
points of interests that are shown in figure 6), the for the second group, yielding a P-value equal to
corresponding speed and distance from the obstacle 0.007.
were identified, and the time to collision was calcu-
lated [2, 5] as the time needed to reach the obstacle
from that position if maintaining the same speed.
Generally, the order with which reactions were de-
tected is the one shown in figure 6: reaction on ac-
celerator pedal is always the first action, while the
successive actions were not always detected, and
moreover their sequence was not always the same. In
figure 6 d0 is the distance from the obstacle from
which is theoretically possible to see it, the moment
in which it was actually perceived is not known. The
distance d0 was evaluated in a qualitative way: the
first perceivable object was the stationary car; in the
two scenarios (A and B) the distance was about 165
m and 140 m respectively; the emergency triangle
was visible a little later (133 m and 105 m, respec-
tively).

Figure 6 - Points of interest while approaching the

obstacle. Figure 7 - How distance of first action d1 influenced the
choice of second action.

Results Thus if the time to collision of the first reaction (ttc1)

is considered, two different types of behaviour can
From the 21 subjects it was possible to obtain 18 val- be indentified (figure 8, where all the tests with de-
id tests on scenario A and 18 on scenario B; while tectable reactions on scenario A and B are shown):
the interactive vehicle approached the obstacle, driv- drivers of type X, who chose to brake as second ac-
er’s reactions were sought: in 6 cases out of 36, no tion, and drivers of type Y, who decided to steer in-

98
 Driving simulator analysis of drivers’ be-haviour in tunnels with obstacles
stead. It can be clearly seen that on average, those
who chose to steeri instead of brake (black lines), are
those who were closer to the obstacle, while those
who were more distant (gray lines), had enough time
to use also the brake. For these two groups of driver
a statistically significant difference is found, being
mean ttc1 equal to -2.11 s and -1.37 s respectively,
with SD equal to 0.61 and 0.36, yielding a P-value
less than 0.002. In figure 8 the approximated fre-
quency density functions of ttc1 are also shown, for
the two groups of drivers.
Figure 8 - How time to collision of first action ttc1
influenced the sequence of manoeuvres.
With regard to the difference between the two sce-
narios, in figures 9 and 10 the distance of first action
(i. e. reaction on the accelerator pedal) is shown. For
scenario A the average first action distance was 56.8,
with a standard deviation of 22.3; on scenario B it
was 46.9 and 18.4. In scenario A such distance
proved to be only slightly significantly higher than
scenario B (P-value was 0.10).
Figure 9 - Distance from the obstacle at which the first
reaction occurred, scenario A.
Figure 10 - Distance from the obstacle at which the first
reaction occurred, scenario B.
Nonetheless, it must be noted that in three of the
tests carried out on scenario A and in one on scenario
B the vehicle was travelling on the fast lane, so that
the manoeuvre was not urgent; by eliminating those
tests, mean values become 60.4 and 46.8 respective-
ly, while standard deviations are 20.9 and 19.3; thus
P-value is less than 0.05, indicating a significant dif-
ference. When detecting the first action, travelling
speed was in average 117 km/h in scenario A and
121 km/h in scenario B, ranging from 93 to 137
km/h; given such variability, the time to collision pa-
rameter, allowing for both reaction distance and
speed, was 1.75 s for scenario A, and 1.38 s for sce-
nario B, showing a significant difference (P-value =
0.04).
As said, each participant met the obstacle twice on
the two different scenarios, so that a possible influ-
ence can be suspected: the overall average distance
for the first action to the first obstacle was 52.6 m
and 53.0 m for the second, a non statistically signifi-
cant difference.
As already said, the second action was sometimes
braking, in the other cases it was steering, instead.
Distance to the second action d2 was higher for sce-
nario A (mean value = 37.3 m) than scenario B (25.8
m), with a slightly significant difference (P-value =
0.09). Referring again to figure 6, the difference t2-t1
is the time interval between the first and the second
action; it can be considered a type of decision time,
since it is the time the driver took to decide the suc-
cessive action. This time was a little longer for sce-
nario A (mean = 0.73 s; SD = 0.35) than scenario B
(mean = 0.62 s; SD = 0.3), suggesting that the driv-
ers in scenario A, having spotted the obstacle earlier,
behaved with less haste.
99
 Lorenzo Domenichini, Francesca La Torre, Dario Vangi, Valentina Branzi, Carlo Cialdai, Monica Meocci, Antonio Virga

Conclusions [5] Lechner D., Malaterre G., Emergency manoeuvre exper-

imentation using a driving simulator, SAE Technical Paper
910016.
Thirty-six virtual driving tests were carried out on
the simulator installed at University of Florence,
thanks to the participation of 21 volunteers. Two dif- [6] Törnros J., Driving behaviour in real and simulated road
ferent scenarios were setup, based on a terrain that tunnel - A validation study, Accident Analysis and Preven-
tion, Vol. 30, No. 4, pp. 497–503, 1998.
was specially produced for this experiment. The task
consisted in driving on a highway track on which an
unexpected obstacle was placed inside tunnels with Contact
two different types of lighting. In 9 cases out of 36 LORENZO DOMENICHINI, Professor
the task failed, and a collision with the obstacle was Department of Civil and Environmental Engineering
caused. By analysing the data recorded by the simu-
DARIO VANGI, Professor
lator control software, the behaviour of the drivers
Department of Industrial Engineering
while approaching the obstacle was observed; in par-
ticular two types of behaviour were detected, with Via di S. Marta, 3,
one group of subjects that, having spotted the hazard 50139, Florence, Italy
earlier, had enough time to react with manoeuvres on e-mails:
the accelerator, brake and steering wheel; the second lorenzo.domenichini@unifi.it, dario.vangi@unifi.it
group, seeing the obstacle once too close to it, decid- tel: +390554796315, +390554796505
ed to release the accelerator and steer to avoid the
obstacle, without braking. Since the obstacles in the
two scenarios were placed in tunnels with different
type of lights (LED and sodium vapour), a difference
in driver’s behaviour depending on lighting was as-
sessed: several aspects were analysed, and most of
them revealed a better driver behaviour when facing
the obstacle in the tunnel with LED lighting. In par-
ticular, the obstacle was perceived earlier, and the
consequent manoeuvres were carried out in a more
effective way (only one collision in the tunnel with
LED lights, compared to 8 collisions with sodium
vapour lights); the distance from the obstacle of first
(always the release of accelerator pedal) and second
action (on brake or steering wheel), was significantly
higher with LED lights.

References

[1] Dutta A., Fisher D. L., Noyce D. A., Use of a driving

simulator to evaluate and optimize factors affecting under-
standability of variable message signs, Transportation Re-
search Part F, 7 (2004) 209–227.

[2] Hoffmann E. R., Mortimer R. G., Drivers' estimates of

time to collision, Accident Analysis and Prevention, Vol-
ume 26, Issue 4, August 1994, 511-520.

[3] Horberry T., Anderson J., Regan M. A., The possible

safety benefits of enhanced road markings: A driving simu-
lator evaluation, Transportation Research Part F, 9 (2006)
77–87.

[4] Jamson S. L., Jamson A. H., The validity of a low-cost

simulator for the assessment of the effects of in-vehicle in-
formation systems, Safety Science 48 (2010) 1477–1483.

100
© 2013 by EVU
EVU 2013 - 11

Hazard detection in driving context and braking

reaction time
Ciceri M., Ruscio D., Confalonieri F. (Catholic University of Milan)
Vangi D., Virga A. (University of Florence)

Abstract
Present research aims to understand how drivers behave when approaching a hazard. The necessity for this work arose
from the number of road accidents, in particular the ones involving pedestrians. By using a simulator, driving scenarios
were created by manipulating the following variables: presence/absence of zebra crossing; the presence/absence of
pedestrian road signs; and the position of the hazard. Results confirm the hypothesis that pedestrian sighting depends on
the experimental variables. Road signs guarantee an optimal condition to detect a potential hazard,but it is the
predictability of the crossing itself that makes a difference in the time needed for the driver to realize when a hazard is
already present.

Introduction cally collects only response times to the appearance

Walking is essential to the mobility of road users, not
only for carrying out daily living tasks, but also for
social interaction and physical exercise. However,
pedestrians, are vulnerable road users and at high
risk of death or serious injury. International crash
statistics indicate that pedestrians are an extremely
vulnerable road-user group. In Italy, in the year
2010, 15% of road deathes were pedestrians. The
most common causes of accidents are traceable to
the road condition, the vehicle and the driver. How-
ever, what happens when drivers meet pedestrians on
the road? The present study will focus on the driver
behaviour during the approach to a specific road risk:
the pedestrian.
One of many ways to study this phenomenon is by
using a driving simulator. Simulators are essential
tools of driver assessment, most especially for ethical
reasons, in any task where drivers may be exposed to
actual driving hazards such as the high probability of
collision. Simulators and virtual reality techniques
have proven to be powerful training devices in pre-
venting pedestrian injury (Schwebel, Gaines, & Sev-
erson, 2008) and teaching basic driving skills (e.g.,
De Winter, Wieringa, Kuipers, Mulder, & Mulder,
2007; Kappé & Emmerik, 2005). Simulators provide
feedback, allow for graduated levels of task difficul-
ty, and make it possible to adapt the training to each
individual’s abilities, in such a way that effective and
personalised skill learning or relearning can be
achieved (Weiss et al., 2003).
There are a number of advantages that simulators
have over other hazard perception tests. Perhaps the
most important is the addition of interactivity.
Whereas a video-based hazard perception test typi-
of a hazard, simulators can record a much more
complex behavioural response, including preparatory
behaviours (slowing in anticipation of a hazard,
changing lane position to avoid a potential hazard) as
well as emergency manoeuvres to avoid the actual
hazard. Speed, braking, steering angle and lane posi-
tion can provide multiple measures on the approach
to a hazard, providing a behavioural signature that
not only indicates that a driver has spotted the hazard
but also what behaviour they have chosen to avoid it.
A second argument for valuing the interactivity of a
simulator over the passive nature of video clips is
that it places a more realistic level of demand upon
the visual system. Certain types of visual cues be-
come extremely important for car handling, such as
the suggestion that drivers need to fixate the tangent
point when steering around curves (Land & Lee,
1994).
The requirement to attend to these cues when con-
trolling a car will invariably interrupt a visual search
for hazards (Schieber, Schlorholtz, & McCall, 2009).
Video-based hazard perception clips place no such
demands on the viewer, and may therefore over-
estimate the hazard perception abilities of individu-
als. To study hazard perception through simulation
however, we need to assume some level of corre-
spondence between the way we move our eyes in the
simulator and the way we move our eyes while on
the road in a real driving situation,
(Konstantopoulos, Crundall, & Chapman, 2010).
Accepting this correspondence it becomes possible
understand how subjects oriented their attention dur-
ing the driving experience. Attention is a limited in-
dividual ability, the driver can look at one environ-
101
 Ciceri M., Vangi D., Ruscio D., Virga A., Confalonieri F.

mental aspect but not at all hazards, which is why a
number of accidents happen.
The aim of this research is to collect data about driv-
ers’ eye movements and the parameters of their be-
haviour when approaching a hazard in order to de-
fine a behavioural model to understand the limits and
the difficulties which impinge on safe driving. The
final aim is exploiting this knowledge to empower
drivers ability and to reduce accidents involving pe-
destrians.
Independent variables
In each scenario the following variables had been
manipulated:
A. road signs:
A 1. with zebra crossing and with the pedestrian road
signs (the pedestrian is near the traffic signs when
starting to cross), (Figure 1);
A 2. without zebra crossing and without pedestrian
road signs (the pedestrian is at the roadside and he is
starting to cross), (Figure 2).).

Method

Aims and hypothesis

The main aim of the research is understand how
drivers approach a to hazard and thereby to under-
stand how accidents can occur. In particular the fol-
lowing behavioural aspects will be considered:
1) analyze drivers eye movements preceding
anaccident scenario;
2) record drivers’ behaviour highlighting: 1) the de- Figure 1. Pedestrian crossing with road signs
tection of the hazard, 2) the reaction time process.
The resulting hypothesis are:
1) It is assumed that different accidents correspond
to different eye movements pattern;
2) It is assumed that different accidents correspond
to differenthazard detection and reaction times.

Subjects
Thirty-one drivers (25 males, 6 females) participated
voluntarily in this study. Each has held adriving li-
cense for more than three years (M = 8.31, SD =
3.72), and all were aged between 21-31 years (M = Figure 2. Pedestrian crossing without road signs
26.39, SD = 3.84).

The simulator
For this experiment, the driving simulator recently B. Predictability:
B 1. unexpected with zebra crossing (the pedestrian
acquired (2012) by the University of Florence was
is near the traffic signs but is hidden by a car/tree),
used. It is a full scale dynamic simulator provided by
(Figure 3);
AutoSim (Norway), consisting of a full vehicle cabin
B 2. unexpected without zebra crossing (the pedes-
(Lancia Ypsilon) installed on a 6 axes Stewart plat-
trian is near the traffic signs butis hidden by a
form. The scenario is projected by 4 projectors on a
car/tree), (Figure 4);
200 degrees cylindrical screen, while rear vision is
B 3. unexpected without zebra crossing and with
obtained by means of three LCD monitors, replacing
something else that will attract the driver’s attention
rear mirrors. A multichannel sound system produces
(the pedestrian is near the traffic signs but is hidden
sounds and noise. A city terrain was chosen among
by a car/tree, there are also other pedestrians on the
those available, in which autonomous traffic and pe-
road), (Figure 5).
destrians were added; with the latter used to create
the hazard situations under study. As the test vehicle
entered a circle with a radius of 30 m centred on the
pedestrian, the figure begins to cross the road. Dur-
ing the tests, information on a number of parameters
relating to the vehicle (position, speed, accelerations,
pedals' position, steering wheel position, gear en-
gaged, etc.) was acquired every 20 seconds?

102
 Hazard detection in driving context and braking reaction
Figure 3. Pedestrian Crossing with road signs and
partial view of the danger.
Figure 4. Pedestrian Crossing without road signs
and partial view of the danger
Figure 5. Pedestrian Crossing without road signs, partial
view of the danger and a distractor on the left side.
Procedure
The subjects were invited to sit in the simulator, to
set the seat and to fasten their safety belt. Instruc-
tions were given about the general study on the driv-
ing simulator. Participants were asked to drive nor-
mally, as they would do in reality, respecting the
road rules and laws. All the scenarios are character-
ized by a medium levelroad traffic with each hazard
appearing after two minutes of simulated driving.
The sequence of the road interactions was random-
ised for each participant, in order to avoid sequence
effects. Each driving session lasted about 13-22
minutes, and during which each participanthad to
face all of the hazard situations which were placed in
specific spots within the town where they were
bound to pass.
During the driving session, the drivers’ eye move-
ments were recorded by an eye tracker (Tobii x 120),
supported by a pc Acer Aspire 5930G (Intel Core
Duo 2.26GHz, 4 GB DDR2 e NVIDIA GeForce
9600M GT 512 MB RAM). In addition Some others
aspects of their behaviour were monitored:
- Driving speed;
- Pressure on the brake.
The first few minutes were used to define the sub-
jects behavior baseline. The scenarios were then ran-
domly administered to the subjects. At the end of the
session the participants were debriefed, and a ques-
tionnaire was administrated to gain information
about:
- Personal information (age, gender, driving
experience)
- Driving Behaviour Questionnaire (DBQ)
(Cicognani & Zani, 2002)
- Italian version of The Barratt Impulsiveness
Scale (BIS11) (Fossati, Di Ceglie, Acquarini,
and Barratt, 2001)
Measures
In relation to each hazard, two Areas of Interest
(AOIs) were selected for the statistical analysis. For
each AOI, drivers’ fixation was calculated using
Tobii Studio 2.0 eye-tracking software. Analyses
were conducted on the seconds preceding the partic-
ipants’ approach to each pedestrian interaction,
measuring whether the subject was looking at the
pedestrian approaching the crosswalk (Pedestrian
Sighting) and the Time to First Fixation (TTFF), that
is the time it took for the driver to shift his/her gaze
towards the moving pedestrian (Figure 6).
Figure 6. Calculation of the time needed from the begin-
ning of the movement of the pedestrian, to the shift of
driver’s gaze direction toward the pedestrian (Time to
First Fixation)
103
 Ciceri M., Vangi D., Ruscio D., Virga A., Confalonieri F.

Results crossing were significantly less frequent, with only

1. Pedestrian sighting
Analyses were conducted on the seconds preceding
the participants’ approach to each pedestrian cross-
ing considered. As shown in Table 1, the frequency
of drivers’ attention towards the pedestrian was dif-
ferent in the different scenarios. In the A1 condition,
almost every driver (78%) was looking at the pedes-
trian near the crosswalk, even before the pedestrian
actually started crossing. On the other hand in the B2
scenario, the fixations at the pedestrian before the
21% of drivers looking at the pedestrian before the
crossing. In the A2 and B1 conditions the probability
that the driver would look at the pedestrian near the
crosswalk before crossing, was comparable to the
probability of the driver not to look at the pedestrian
before the crossing. In the B3 condition the driver
was more likely to look at the pedestrian (73%) than
not (27%). The difference was not statistically sig-
nificant, yet the Odds Ratio for the driver to not look
at the pedestrian before the crossing, is still higher
than the probability of the driver to look.

Table 1: Driver looking at the pedestrian near the crosswalk, in the different conditions

Pedestrian sighting before its movement Percentage

Yes No χ2 Odds Ratio
A1: With road signs - Predictable 78% 22% 7,35 ** 12,96
A2: Without road signs - Predictable 65% 35% 2,13 3,52
B1: With road signs - Unpredictable 50% 50% 0,00 1,00
B2: Without road signs - Unpredictable 21% 79% 6,36 * 0,07
B3: Without road signs - Unpred. - Distractor 27% 73% 2,27 0,14

2. Delay of pedestrian Sighting tion, to a maximum of 0.53 seconds in B2 condition,

In the case of the driver not looking at the pedestrian
approaching the crosswalk, the time it took for the
driver to shift his/her gaze towards the moving pe-
destrian once beginning to cross, was dependent on
the experimental conditions. The means of time
needed to shift the attention were significantly dif-
ferent (Table 2). The delay in the sighting was short-
er in the A1 condition, with a mean of detection of
the pedestrian’s movement of 0.06 seconds. Sighting
was slightly slower in the A2 condition (predictable
crossing without road signs). In the condition where
the crossing was not predictable, the time needed
was significantly longer: 0.20 seconds in B1 condi-
almost half a second longer than the optimal A1
condition.
The difference in the time to the first fixation to-
wards the pedestrian was analyzed using SPSS
through General Linear Model 2x2 (Road Signs x
Predictability). The ANOVA analysis reported a
main effect for the Predictability (F1,30 = 17.835; p =
.001), but not for the Road Signs (F1,30 = 3.463; p =
.087), nor for the interaction effect (F1,30 = 16.04; p
= .229), as can be seen in Figure 7.

Table 2: Time needed to shift the driver’s gaze towards the moving pedestrian

Time to First Fixation of the Moving Pedestrian Delay (sec.)

M SD t p
A1: With road signs - Predictable 0,06 0,14 1,99 0,06
A2: Without road signs - Predictable 0,10 0,17 2,73 0,01
B1: With road signs - Unpredictable 0,21 0,30 3,34 0,00
B2: Without road signs - Unpredictable 0,53 0,42 5,61 0,00
B3: Without road signs - Unpred. - Distractor 0,47 0,33 4,72 0,00

104
 Hazard detection in driving context and braking reaction
Figure 7: Time for driver to shift gaze towards the moving
pedestrian in the different conditions.
Conclusions
Results confirm the hypothesis that pedestrian sight-
ing depends on the experimental variables.
The more complex the scene in terms of road signs
and expectability, the less likely it is that the driver
will look at a pedestrian before crossing, and the
longer it takes to be aware of the movement of the
pedestrian.
Road signs guarantee an optimal condition to detect
a potential hazard, while the predictability of the
crossing itself makes a difference in time needed to
realize when an hazard is already present.
The scenario can be improved, by introducing more
traffic on the road in order to test if the pedestrian
sighting is also dependent on traffic conditions, as
previously reported in literature (Underwood, 2005).
Moreover the sample for further study can be im-
proved in order to allow other statistical analysis,
and could be controlled for gender and other varia-
bles that can impact on reaction times (Green, 2000,
Summala, 2000).
Nevertheless the experimental setting built and tested
in this study seem to warrant further research in or-
der to obtain concrete data and important infor-
mation for specific expectation training for both
drivers and pedestrians, and to build safer pedestrian
crosswalks.
Bibliografia
[1] De Winter, J. C. F., Wieringa, P. A., Kuipers, J.,
Mulder, J. A., & Mulder, M. (2007). Violations and er-
rors during simulation-based driver training. Ergonomics,
50(1), 138–158.
[2] Kappé, B., & Emmerik, M. L. (2005). The use of
driving simulators for initial driver training and testing.
Report No 75151.01. The Hague: TNO Defence, Security
and Safety.
[3] Konstantopoulos, P., Crundall, D., & Chapman, P.
(2010). Driver’s visual attention as a function of driving
experience and visibility. Using a driving simulator to
explore visual search in day, night and rain driving. Acci-
dent Analysis and Prevention, 42, 827–834.
[4] Land, M. F., & Lee, D. N. (1994). Which parts of the
road guide steering? Nature, 369, 742–744.
[5] Schieber, F., Schlorholtz, B., & McCall, R. (2009).
Visual requirements of vehicular guidance. In C. Castro
(Ed.), Human factors of visual and cognitive performance
in driving (pp. 31–50). Boca Raton, FL: CRC Press.
[6] Schwebel, D. C., Gaines, J., & Severson, J. (2008).
Validation of virtual reality as a tool to understand and
prevent child pedestrian injury. Accident Analysis and
Prevention, 40(4), 1394–1400.
[7] Weiss, P. L., Naveh, Y., & Katz, N. (2003). Design
and testing of a virtual environment to train stroke pa-
tients with unilateral spatial neglect to cross a street safe-
ly. Occupational Therapy International, 10(1), 39–55.
[8] Cicognani, E. & Zani, B. (2002). Determinanti psico-
sociali dei comportamenti stradali a rischio. Psicologia
della salute, 3, 5-17.
[9] Fossati A, Di Ceglie A, Acquarini E, and Barratt ES
(2001).Journal of Clinical Psychology, 57, 815-828.
[10] Underwood, G. (A cura di). (2005). Traffic and
Transport Psychology. Theory and Application: Proceed-
ings of the ICTTP 2004. Oxford: Elsevier Ltd.
[11] Green, M. (2000). 'How long does it take to stop?'
methodological analysis of driver perception-brake times.
Transportation Human Factors , 2, p. 195-216.
[12] Summala, H. (2000). Brake Reaction Times and
Driver Behavior Analysis. Transportation Human Factors
, 2 (3), p. 217–226.
Contact
Maria Rita Ciceri, Professor, PhD.
Daniele Ruscio, PhD student.
Federica Confalonieri.
Università Cattolica del Sacro Cuore, Milano
Largo A. Gemelli, 1
20123 Milano, Italy
e-mails: maria.ciceri@unicatt.it,
daniele.ruscio@unicatt.it,
psicotraffico@unicatt.it
tel: +39 02 72345931
Dario Vangi, Professor, PhD.
Antonio Virga, PhD.
Dipartimento di Ingegneria Industriale
Via di Santa Marta, 3
50139, Firenze, Italy
e-mails: dario.vangi@unifi.it,
antonio.virga@unifi.it
tel: +39 055 47965
105
 Ciceri M., Vangi D., Ruscio D., Virga A., Confalonieri F.

106
© 2013 by EVU
EVU 2013 - 12

Analysis of Drivers’ Conduct while Driving over modern

Pedestrian Crossings

Robert Kledus , Marek Semela , Pavel Maxera , Martin Kunovský

(Institute of Forensic Engineering VUT Brno)

Abstract
The Study analyses the conduct of eight drivers while driving over pedestrian crossings at dusk and at night. The
monitored vehicle drives over a modern, horizontally and vertically marked pedestrian crossing in a town. The
pedestrian crossing is illuminated from the left side of the vehicle’s direction of travel. There is a pedestrian refuge
island in the middle of the road, marked with a Keep Right sign.
The drivers’ optical responses to various impulses during test trips are monitored using a special device named
Viewpointsystem®, GmbH that tracks changes in the driver’s focus. The individual driver’s behavioural strategy when
driving over the pedestrian crossing is thereupon evaluated focusing on the mode of the driver’s navigation and his
reaction to crossing pedestrians. The degree of danger of each traffic scenario is assessed and juxtaposed to the strategy
that driver has chosen. The results of the analysis are then compared to the results of extensive driving tests conducted in
close collaboration with the Institute of Forensic Engineering of the University of Technology in Brno (VUT) and the
EPIGUS Research Institute – Institute for Holistic Accident and Traffic Safety Research of Austria, in 2010.

ern methods and procedures aiming to facilitate

1. Preface
Pedestrians are among the most vulnerable road us-
ers, as collision with a vehicle nearly always has
very grave consequences for them. Accidents involv-
ing pedestrians often occur under poor visibility,
where the driver’s chance of a timely recognition of
a pedestrian is significantly affected by the light
conditions. The ability of the pedestrian to recognise
the position and speed of vehicles is also impacted
upon.
According to the statistics of the Czech Republic Po-
lice, 110 pedestrians were killed nationwide due to
collisions with personal vehicles during the period
from January to October 2012, i.e., one person every
three days, of these 66 % (73) were killed in urban
traffic and 61 % (67) under poor visibility. The sta-
tistics also indicate that the most common cause was
the pedestrian’s inattention or sudden stepping into
the road from a sidewalk or road shoulder.
2. Status quo analysis
Analyses of the intervals of drivers’ responses to var-
ious stimuli and the influence of the light conditions
and other factors have been conducted by many re-
searchers in the field of traffic safety and traffic ac-
cidents. Pfleger 0 analyzed drivers’ ability to recog-
nize objects while driving a vehicle, with regard to
multiplicity of stimuli. Similar issues were examined
by Weyde [2], who presents certain aspects of mod-
recognition of pedestrians by drivers from a moving
vehicle. Other specialists, e.g., Šiška [3], concentrat-
ed on the visibility of pedestrians and cyclists on the
side of the road and on the visibility variances due to
diverse illumination or reflective elements. Kaplánek
[4] compared drivers’ response intervals and the in-
fluence of using a mobile (cell) phone while driving.
Kropáč [5] focused on the ability to determine the
distance necessary for the driver’s recognition of a
pedestrian from a vehicle in motion. Similar issues
are dealt with by the authors of Article [6]. Sullivan
[7] evaluated various causes of accidents, using data
from police traffic reports involving pedestrians,
comparing accidents that occurred during daytime
and during the night, the effects of symmetrical and
asymmetrical headlights on drivers’ ability to see
pedestrians, and the effect of the point where the pe-
destrian steps into the roadway and the direction of
his crossing. Clifton [8] studied the influence of per-
sonal and environmental characteristics on the rate of
collisions with pedestrians and the gravity of their in-
juries. However, no detailed analysis of drivers’
conduct while driving over a pedestrian crossing has
been made.
3. The study’s objectives
The aim of the Study is to analyse the behaviour of
eight drivers in a situation resembling a drive over a
modern pedestrian crossing while being crossed by a

107
 Robert Kledus , Marek Semela , Pavel Maxera , Martin Kunovský

pedestrian. The Study focuses on evaluation of the
following circumstances:
 How does the driver who is driving over the
crossing react to traffic signs (informing him of
a potential danger) and the crossing pedestrian?
 Does the driver – after realising that he is about
to drive over a pedestrian crossing – react in a
learned manner or does he respond to individual
stimuli spontaneously?
 How much attention does the driver pay to the
pedestrian who is crossing the roadway and/or
is his attention affected by other stimuli?
sign has a yellow reflective frame. The crossing is il-
luminated from the left side with a street light, see
Fig. 6. A plan showing the driving direction of vehi-
cles during the tests and the symbols describing the
traffic signs is presented in Fig. 7.
The road at the point of the pedestrian crossing is
10.4 m wide and is divided into two lanes. The width
of the road shoulders on both sides is 0.6 m; the
width of each driving lanes is 3.2 m, and width of the
island is 2.8 m.

4. Solution
The Study is based on the results of extensive
driving tests realised in 2010. During those tests,
eight drivers-volunteers kept driving back and forth
on a road stretch of approximately 30 km between
the towns of Lednice, Břeclav, Valtice, and
Hlohovec. During those trips, the location and speed
of their vehicles was recorded and a special device
called Viewpointsystem®, GmbH, was used to
monitor the driver’s perspective and his optical Figure 1 The point of arrival from Lednice, view of the
reactions to various stimuli in similar simulated town’s information sign and the Keep Right traffic sign
traffic scenarios.
The measuring was realised in collaboration with the
Institute of Forensic Engineering of the University of
Technology in Brno and the EPIGUS Research
Institute – Institute for Holistic Accident and Traffic
Safety Research of Austria, as part of a project
named “Experience and Good Practice Exchange in
Analyzing and Prevention of Road Traffic
Accidents,” CZ0069, 2010 – 2011. For more
information on the measurements, see
publication [6].
4.1. Description of the measured location
The measuring point was situated on the edge of the
Figure 2 Section after the deceleration island and view of
town of Lednice coming from the town of Valtice the No Parking zone
along Valtická Street. After entering Lednice, the
speed of vehicles is limited by means of a decelera-
tion island, see Fig. 1. The island is marked with a
Keep Right traffic sign. After passing the island, in a
section of about 90 meters, the road widens for bus
stops on both sides, see Fig. 2 and 3. On the right is a
clearly visible traffic sign marking the bus stop, with
a waiting area for passengers. Next comes an inter-
section with Nádražní Street, see Fig. 4, and about 50
metres thereafter, where the road turns left, is the
monitored pedestrian crossing, see Fig. 5. The cross-
ing is protected by another deceleration island,
marked with a Keep Right traffic sign. The crossing
is marked with clearly visible traffic signs, both hor-
izontal and vertical. The vertical pedestrian crossing

108
 Problems of uncertainty in time-distance analysis of pedestrian accident

Figure 3 View of the bus stop, the bus stop sign, and Figure 4 View of the intersection of Valtická and
main road way Nádražní Streets with partly offset shoulders

Figure 5 View of the deceleration island, marked with Figure 6 Detailed view of the monitored point, with hor-
Keep Right and Pedestrian Crossing signs izontal and vertical signs marking the pedestrian cross-
ing, illuminated from the left

Figure 7 Plan of the monitored point. Source [9]

109
 Robert Kledus , Marek Semela , Pavel Maxera , Martin Kunovský

4.2. The use of actors (figurants)

The actor was dressed in dark clothing (black shoes
and trousers, dark-brown coat, black hat), see Fig. 8.
For the purposes of the test, he waited near the pe-
destrian crossing (on the left from the perspective of
the vehicle). He had instructions to cross the street in
front of the test vehicle, giving consideration to the
current traffic. Usually, he would start to cross the
street just as the test vehicle was passing by the bus
stop, shown in Fig. 8.

Figure 9 Test vehicle Škoda Octavia Combi with H7 light

bulbs

Figure 8 The actor’s standpoint at the point of measuring

4.3. Measuring technology and vehicles

The tests were conducted using two comparable mo-
tor vehicles, type Škoda Octavia Combi 4x4, 2.0
TDi. The only substantial difference between them
was the headlights. The first vehicle’s headlights
were fitted with light bulbs type H7, see Fig. 9; the
second vehicle was fitted with Xenon lights.
A special device, made by Viewpointsystem®,
GmbH, was used for measuring the angle of the
driver’s view. The device consists of eyeglass-like
frames fitted with two cameras, see Fig. 10. With the
aid of Viewpointsystem® evaluation software, the
recordings of both cameras were evaluated from the
driver’s perspective. In order to perfect the evalua-
tion at the moment of the driver’s passing the actor,
another camera was installed firmly in the vehicle,
and took images of the vehicle’s front and immediate
area. The vehicle’s momentary speed (i.e., relative
change of the vehicle’s position) was measured
at frequency 1 Hz using a GPS device with high-
sensitivity chip Sirf III. All these devices were con-
nected to data logger HOLUX M-241, which simul-
taneously ensured synchronization of all records at a
single time.
Figure 10 Driver with eyeglass frames fitted with 2 cam-
eras for measuring the angle of driver’s view
4.4. Drivers and conditions during individual
trips
Eight drivers participated in the tests, seven men and
one woman. From optical point of view, they all met
the criteria for driving motor vehicles, group B.
Their assignment called for driving normally through
the test route of 30 km, while wearing the eyeglasses
for monitoring their eye movements and observing
road traffic rules.
The conditions of individual test trips are summa-
rised in Tab. 1. The trips were divided according to
the number of other (random) participants as stand-
ard and special (see further).

110
 Problems of uncertainty in time-distance analysis of pedestrian accident

Tab. 1 Testing conditions

Special condi-
No special circumstances Special conditions
tions
Test P8 P6 P2 P4 P3 P5 P7 P1
Night Night Night Night Night Dusk Night Night
Driving condi- dry dry after rain after rain after rain dry dry raining
tions dry dry wet wet wet dry dry wet
roadway roadway roadway roadway roadway roadway roadway roadway
Young Young Young Old Young Young
Driver Old man Old man
man man man woman man man
Vehicle 1 Vehicle 2 Vehicle 1 Vehicle 2 Vehicle 1 Vehicle 2 Vehicle 1 Vehicle 1
Vehicle
(H7) (XE) (H7) (XE) (H7) (XE) (H7) (H7)

Evaluation of the test trips is based on methodology

5. Processing of results
5.1. The mode of processing the results
The driver responds to various stimuli situated out-
side the focus of his vision area (traffic signs, infor-
mation boards, and other traffic participants, includ-
ing pedestrians), by quickly changing the angle of
his view to be able to see the noticed object within
his focal area. If the change of his view angle is ef-
fective enough, he is able to recognise with high
probability the first optical reactions to selected ob-
jects.
published in Source [6]. Firstly, the evaluation con-
centrates on optical reactions (henceforth only FOR)
to monitored stimuli from the perspective of the
driver’s view using the Viewpointsystem® by
measureing the direction of the view angle. Exam-
ples of trips P8, P7, and P5 are shown in Fig. 11
through 16. Evaluation of the vehicle’s distance from
the monitored crossing is done by means of continu-
ous measuring of the vehicle’s momentary speed and
the time that the vehicle needed to cover the distance
between the vehicle’s position when the driver first
reacted to the object and the position when the vehi-
cle rolled over the crossing.

Figure 11 Trip P8 – Beginning of the driver’s optical Figure 12 Trip P8 – Change of the angle of the driver’s
reaction to the actor (12.24 sec and 149.5 m before the view to the actor – the actor in focus (12.20 sec and
crossing) 148.8 m before the crossing)

111
 Robert Kledus , Marek Semela , Pavel Maxera , Martin Kunovský
Figure 13 Trip P7 – Beginning of the driver’s optical
reaction to the actor (16.12 sec and 175.1 m before the
crossing)
Figure 15 Trip P5 – Beginning of the driver’s optical
reaction to the actor (9.36 sec and 113.0 m before the
crossing)
5.2. Presentation of the results of measuring
The results of the evaluation of individual tests are
shown in Fig. 17 through 24 in the form of route-
time diagrams (henceforth only s-t diagrams). Each
of the Figures is divided into two parts. On the left
are the coordinates of the route-time diagram, show-
ing the progress of the vehicle’s trip, as it moves
along the roadway (henceforth only s-t diagram of
the vehicle). On the right are the coordinates of the
route-time diagram showing the actor’s movement,
as he moves perpendicularly to the roadway (hence-
forth only s-t diagram of the pedestrian). The begin-
ning of the system of coordinates of both diagrams
was chosen to be on the right-side edge of the road-
way, at the point where the actor leaves the roadway
(see Fig. 6). The vehicle’s route before it rolls over
the crossing and the actor’s route before he leaves
the roadway are presented as negative values. (The
112
Figure 14 Trip P7 – Change of the angle of the driver’s
view to the actor – the actor in focus (16.04 sec and
174.1 m before the crossing)
Figure 16 Trip P5 – Change of the angle of the driver’s
view to the actor – the actor in focus (9.32 sec and
112.4 m before the crossing)
diagram’s articulation is similar to collision solu-
tions, where movements before collision are present-
ed as negative values (minus), movements after col-
lision as positive values (plus), and the point of colli-
sion as the beginning of the system of coordinates.)
Time t=0 is always the moment when the vehicle
rolls over the crossing, or more accurately, crosses
the trajectory of the pedestrian’s movement. Accord-
ing to common practice in the CR – and also for
practical reasons – the route is always presented as
x-axis and time as y-axis.
In the vehicle’s s-t diagram the vehicle’s route is
drawn as a blue line. The relevant pictograms mark
the moments of the driver’s first optical reactions to
traffic signs and the actor. The vertical red line on t-
axis indicates the time interval (derived from the pe-
destrian’s s-t diagram), during which the pedestrian
passes through the corridor of the vehicle’s trip. The
 Problems of uncertainty in time-distance analysis of pedestrian accident

s-t diagram shows clearly the distance between the margin of the roadway (henceforth only LRM and
vehicle and the actor at the moment when the actor RRM), and the left and right edge of the island
entered the corridor and then left the corridor of the (henceforth only LIM and RIM). The thin uninter-
vehicle’s trip. Since practically all the drivers slowed rupted line shows the road shoulders and the blue
down on seeing the actor, their trips are marked in a dashed lines mark the corridor of the vehicle’s trip,
thinner orange line, on the presumption that the driv- whose width corresponds to the width of the ana-
er would not react by slowing down the moment of lysed vehicle, including the rear view mirrors. The
the first optical reaction to the actor and would con- actor’s movement is marked in green, with the time
tinue to proceed at the same speed. intervals when the driver directly observes the actor
marked. The distance to the vehicle during these
In the pedestrian’s (actor’s) s-t diagram the roadway
time intervals can be derived from the vehicle’s s-t
is marked at the point of crossing. Thick vertical
diagram.
lines indicate fixed obstacles, i.e., the left and right

-15 s -15 s
time [s]

time [s]

-15 s -15 s

time [s]
time [s]
-10 s -10 s
-10 s -10 s

-5 s -5 s
-5 s -5 s

0s 0s
distance [m] distance 0s 0s
distance [m] distance
[m]
[m]
5s 5s
-200 m -150 m -100 m -50 m 0m 50 m -15 -10 -5 0 5 5s 5s
-200 m -150 m -100 m -50 m 0m 50 m -15 -10 -5 0 5

Figure 17 Trip P3 – standard Figure 18 Trip P8 – standard

-15 s -15 s -15 s -15 s

time [s]

time [s]

time [s]
time [s]

-10 s -10 s -10 s -10 s

-5 s -5 s -5 s -5 s

0s 0s 0s 0s
distance distance [m] distance
distance [m]
[m] [m]

5s 5s 5s 5s
-200 m -150 m -100 m -50 m 0m 50 m -15 -10 -5 0 5 -200 m -150 m -100 m -50 m 0m 50 m -15 -10 -5 0 5

Figure 19 Trip P6 – standard Figure 20 Trip P2 – random pedestrian at LRM

-15 s -15 s -15 s -15 s

time [s]
time [s]

time [s]

time [s]

-10 s -10 s -10 s -10 s

-5 s -5 s -5 s -5 s

0s 0s 0s 0s
distance [m] distance distance [m] distance
[m] [m]

5s 5s 5s 5s
-200 m -150 m -100 m -50 m 0m 50 m -15 -10 -5 0 5 -200 m -150 m -100 m -50 m 0m 50 m -15 -10 -5 0 5

Figure 21 Trip P4 – standard Figure 22 Trip P1 – three random vehicles driving in

opposite direction, random pedestrian at LRM

113
 Robert Kledus , Marek Semela , Pavel Maxera , Martin Kunovský

-15 s -15 s -20 s -20 s

time [s]
time [s]

time [s]
time [s]
-15 s -15 s
-10 s -10 s

-10 s -10 s
-5 s -5 s
-5 s -5 s

0s 0s
distance [m] distance 0s 0s
[m] distance [m] distance
[m]
5s 5s 5s 5s
-200 m -150 m -100 m -50 m 0m 50 m -15 -10 -5 0 5 -200 m -150 m -100 m -50 m 0m 50 m -15 -10 -5 0 5

Figure 23 Trip P5 – random vehicle driving in front of Figure 24 Trip P7 – random vehicle at the intersection
the analyzed vehicle, another pedestrian at LRM before the crossing

The basic evaluation of the results is presented driver’s trip in the monitored section of the road was
in Tab. 2. On average(see Tab. 2, Part c), most driv- not substantially influenced by other motor vehicles.
ers reacted to pedestrians optically at a distance of All of the test trips took place at night.During P5, P7,
119 m (distance s1), i.e., just about 10.9 sec before and P1 trips, the drivers were forced to react to other
the vehicle’s driving over the crossing (beginning of vehicles, which appeared on the road randomly –
interval t1). these scenarios were therefore classified as special
The pedestrian’s intention to cross the roadway may circumstances.
have been apparent to the driver approximately (4.0 The P5 trip was also classified as special circum-
+2.7 + 2.5 =) 9.1 sec before the vehicle’s driving stances, because the driver entered the town of
over the crossing, on average, i.e., the moment when Lednice while driving behind another vehicle. Thus,
the actor was about to enter the roadway (beginning he paid attention to the vehicle driving in front of his
of interval t2). vehicle before noticing the actor. This was the only
The actor entered the roadway at the beginning of in- trip that took place at dusk. The P7 trip was also rec-
terval t3, i.e., approximately (2.7 + 2.5 =) 5.2 sec, on orded as having special circumstances, due to the
average, before the vehicle’s driving over the cross- fact that just after the driver had noticed the actor,
ing. At this time, the average vehicle was at a dis- another vehicle began to approach Valtická Street
tance of 48 m from the crossing (distance s2). From from Nádražní Street on the right (see intersection in
the FOR moment to the actor, the average driver no- Fig. 5). This vehicle yielded on the roadway in front
ticed the actor approximately (11.3 s * 52 % =) 5.9 of the test vehicle, so that the test driver had to di-
sec. vide his attention between the actor and the yielding
The drivers paid the most attention to the actors im- vehicle.
mediately before they entered the roadway (63 % of The P1 trip was also recorded as special circum-
1.8 sec), and the least attention after they left the stances due to relatively heavy traffic in the opposite
roadway (38 % of 2.5 sec). direction. The driver met three oncoming vehicles.
The progress of individual trips can be seen on lines The actor stepped into the roadway after the last on-
P1 through P8. Similar conditions existed when coming vehicle and was at that moment hidden from
measuring trips P8, P6, P2, P4, and P3, where the the test driver’s view.

114
 Problems of uncertainty in time-distance analysis of pedestrian accident

Tab. 2 Evaluation of test trips

Part a) Standard trips (driver’s attention not substantially affected by other vehicles)
Pedestrian’s time on the road from driver’s FOR Vehicle’s distance from the crossing
Trip t1 t2 t3 t4 t s1 s3
P8 3,4 s 4,0 s 2,6 s 2,3 s 12,3 s 149 m 48 m
P6 2,6 s 4,1 s 2,1 s 1,8 s 10,6 s 111 m 35 m
P2 3,4 s 4,2 s 2,6 s 3,2 s 13,4 s 125 m 38 m
P4 0,0 s 3,8 s 3,5 s 2,6 s 9,9 s 110 m 61 m
P3 0,6 s 4,5 s 2,7 s 2,0 s 9,8 s 104 m 44 m
Average 2,0 s 4,1 s 2,7 s 2,4 s 11,2 s 120 m 45 m
Time of the driver’s direct attention paid to the actor Average speed of the vehicle at the begin-
(average) ning of the interval
% of t1 % of t2 % of t3 % of t4 % of t t1 t3
Average 78% 57% 59% 24% 54% 51 km/h 36 km/h
Part b) Non-standard trips (driver’s attention distracted by other vehicles)
Pedestrian’s time on the road from driver’s FOR Vehicle’s distance from the crossing
Trip t1 t2 t3 t4 t s1 s3
P5 0,7 s 4,3 s 2,8 s 1,5 s 9,3 s 112 m 47 m
P7 3,3 s 5,7 s 2,7 s 4,5 s 16,2 s 175 m 60 m
P1 0,0 s 4,6 s 2,5 s 1,9 s 9,0 s 68 m 49 m
Average 1,3 s 4,9 s 2,7 s 2,6 s 11,5 s 118 m 52 m
Time of the driver’s direct attention paid to the actor Average speed of the vehicle at the begin-
(average) ning of the interval
% of t1 % of t2 % of t3 % of t4 % of t t1 t3
Average 26% 45% 57% 58% 49% 47 km/h 46 km/h
Part c) Comprehensive evaluation - standard and non-standard trips
Pedestrian’s time on the road from driver’s FOR Vehicle’s distance from the crossing
t1 t2 t3 t4 t s1 s3
Average 1,8 s 4,4 s 2,7 s 2,5 s 11,3 s 119 m 48 m
Time of the driver’s direct attention paid to the actor Average speed of the vehicle at the begin-
(average) ning of the interval
% of t1 % of t2 % of t3 % of t4 % of t t1 t3
Average 63% 52% 58% 38% 52% 49 km/h 40 km/h

Symbols used in Tab. 2: s1 - vehicle’s distance from the crossing at the

t1 - time interval, during which the actor appeared
on the left sidewalk (from the driver’s FOR to
the actor stepping into the roadway)
t2 - time interval, during which the actor was
crossing the left driving lane and island (from
the actor entering the roadway to his entering
the right driving lane)
t3 - time interval, during which the actor was
crossing the right driving lane and began to
cross the vehicle’s corridor (from the actor en-
tering the right driving lane and his leaving the
roadway)
t4 - time interval, during which the actor stepped
onto the right sidewalk (from the actor leaving
the roadway to the vehicle driving over the
crossing)
t- total time from driver’s FOR to the vehicle
driving over the crossing
moment of the driver’s FOR (at the beginning
of interval t1)
s3 - vehicle’s distance from the crossing at the
moment of the actor’s entry into the right driv-
ing lane, where the test vehicle was driving (at
the beginning of interval t3)
6. Conclusion
In the Czech Republic, like in many other countries,
pedestrians have the right of way before moving mo-
tor vehicles. Wherever necessary, drivers are obliged
to come to a full stop before a pedestrian crossing.
One of the objectives of this Study was to find out
how drivers react on modern pedestrian crossings, in
terms of their response to traffic signs, and how they
learn of potential risks arising from the possibility of
encountering pedestrians crossing the roadway.
The graphic rendition of the analysis shown in Fig.
17 through 24 indicates that the most significant
stimulus in this particular case was the Keep Right
115
 Robert Kledus , Marek Semela , Pavel Maxera , Martin Kunovský
sign which marks the obstacle, i.e., the island, in the
roadway (see detail in Fig. 6). Of the eight drivers,
six noticed the Keep Right sign first, only two no-
ticed and reacted to the Pedestrian crossing sign first,
despite the fact that it was posted in a reflective
frame.
Drivers who reacted to the Keep Right sign first, no-
ticed the sign approximately 118 m to 185 m before
driving over the crossing.
Drivers who saw the Pedestrian Crossing sign first
(P1 and P2 trips), noticed the sign approximately 110
m to 132 m driving over the crossing. However, trip
P1 took place while it was raining and trip P2 took
place shortly thereafter.
The evaluation therefore included comparison of
trips P2, P3, P4, P6, P7, and P8, where the moment
of the first spotting of the actor was not substantially
influenced by any other participants. The distance, at
which the drivers first noticed the actor, ranges from
104 m to 175 m before driving over the crossing.
During the evaluation, special circumstances were
recorded with trips P1 and P5, where
 in trip P1, the actor was first noticed by the
driver when he stepped into the roadway from a
point where he was hidden from the driver’s
view behind the last oncoming vehicle, so that
the driver saw him only from a distance of 68 m
from the crossing,
 in trip P5, the driver was paying attention to the
vehicle driving in front of him, so that he no-
ticed the actor only at a distance of 112 m.
Furthermore, the aim of the evaluation was to estab-
lish whether the driver, on realising that he is about
to come to a pedestrian crossing, acted in a learned
manner or reacted to individual stimuli spontaneous-
ly, instead. The evaluation did not let us conclude,
whether the drivers associated the Keep Right sign
that marked the island with a pedestrian crossing,
i.e., whether they were consciously looking for a Pe-
destrian Crossing sign. Of the eight drivers, two no-
ticed the Pedestrian Crossing sign first, i.e., before
noticing the Keep Right sign. Of the remaining six
drivers, three (trips P4, P7, P8) changed the direction
of their view to the right, where they noticed the Pe-
destrian Crossing sign and immediately thereupon
reacted to the actor appearing on the left side of the
roadway. The other two/three drivers reacted to the
actor immediately after noticing the Keep Right sign,
without noticing the Pedestrian Crossing sign at all.
There was no indication that drivers, after noticing
the Pedestrian Crossing sign, would have checked
the right side of the road, or the right edge of the is-
land, as the point where the actor would have had the
least time to enter the vehicle’s corridor, and then,
for example, the left side of the road. In other words,
116
drivers reacted spontaneously to the most obvious
stimuli.
In this context, it is perhaps appropriate to emphasise
that all drivers noticed the actor from a safe distance.
No dangerous situation occurred, not even when the
actor stepped into the roadway. On average, the ve-
hicle was 43 m before the crossing and was moving
at a speed that would have allowed the driver to stop
in 17 meters. No dangerous situation occurred at the
moment of the actor leaving the driver’s driving cor-
ridor.
Furthermore, the aim was to assess the degree of at-
tention that the drivers paid to the actor who was
crossing the road, or how other stimuli impacted on
the drivers’ attention. In this context, the valuation
compared trips P3, P4, P5, P6, and P8, where the
time of the driver’s attention was not distracted by
other participants. During these trips, the actor
crossed the road in 6.2 to 7.3 seconds and the drivers
observed him for 2.24 to 6.16 seconds, i.e., 32 % to
86 % of the actor’s presence in the roadway.
Trips P1, P2, P5, and P7 were classified as special
circumstances.
 In trip P1, the actor crossed the roadway in 7.04
seconds, while the driver was observing him for
2.12 seconds, i.e., 33 % of the actor’s presence
in the roadway. The short time interval of the
driver’s attention was accountable to the fact
that the actor was hidden behind an oncoming
vehicle part of the time.
 In trip P2, the actor crossed the roadway in 6.80
seconds, while the driver was observing him for
1.84 seconds, i.e., 23 % of the actor’s presence
in the roadway. While the actor was crossing
the road, the driver was observing another actor
on the left side of the roadway, who did not in-
dicate clearly whether or not he intended to
cross the road. The driver kept watching this ac-
tor for 3.32 seconds, i.e., 48 % of the actor’s
presence in the roadway. The ambiguous behav-
ior of the second actor was evidently very im-
portant to him.
 In trip P7, the actor crossed the roadway in 8.4
sec. The actor was crossing the road in front of
a vehicle that yielded into the road in front of
the test vehicle. Thus, the test driver paid atten-
tion to both the actor and the yielding vehicle.
The driver kept observing the actor for 5.2 se-
conds, i.e., 63 % of the actor’s presence in the
roadway and the yielding vehicle for 1.0 se-
conds, i.e., 11 % of the actor’s presence in the
roadway.
Six of the eight drivers noticed the actor while still
on the left sidewalk, over a time interval ranging
from 0.3 s to 2.8 seconds.
 Problems of uncertainty in time-distance analysis of pedestrian accident
The study has led to interesting conclusions, which
however cannot be generalised, as the measurements
were taken in a single traffic situation. Subsequent
studies will therefore focus on measuring and com-
paring other traffic scenarios.
7. Acknowledgements
The study was conducted as part of a project entitled:
“Experience and Good Practice Exchange in Analyz-
ing and Prevention of Road Traffic Accidents,”
CZ0069, 2010 – 2011.
The project was co-financed by the European Un-
ion’s European Fund for Regional Development,
Small Projects Fund (Goal: European Regional Co-
operation Austria – Czech Republic 2007 – 2013),
with the support of the Regional Development
Agency of South Moravia.
The project was elaborated in collaboration with the
Institute of Forensic Engineering of the University of
Technology in Brno and the EPIGUS Research Insti-
tute – Institute for Holistic Accident and Traffic
Safety Research of Austria (Institut für ganzheitliche
Unfall – und Sicherheitsforschung).
Special gratitude is extended to Prof. Dipl.-Ing. Dr.
Ernst Pfleger and his whole team, which performed
the extensive measuring activities and evaluations of
the tests that were used in the preparation of this
Study. We thank them for their willingness to pro-
vide us with the necessary instruments, equipment,
knowledge, and time.
Furthermore, we thank the company Autonova Brno,
spol. s r.o., for lending us the test vehicles.
References
[1] PFLEGER, E., Hazard Recognition and Reaction in
Practice – Exact Time Proof by Visualization Analysis.
Brashov (Romania), 2012. Collected Studys from Annual
EVU Conference 2012. ISBN 978-973-0-13537-4
[2] WEYDE, M. Reconstruction of the Visibility of Pe-
destrians in Darkness under Dynamic Circumstances.
Prague: TRIBUN EU, 2010. Collected Studys from Annu-
al EVU Conference 2010. ISBN 978-80-7399-128-9
[3] ŠIŠKA, A., Visibility of Cyclists or Pedestrians Along
Roads. Brno, 2012. Collection of Studys from Junior Fo-
rensic Scientists Conference Brno 2012. ISBN 978-80-
214-4485-0
[4] KAPLÁNEK, A., Analysis of Drivers’ Reactions to
Combined Stimuli. Brno, January 2011. The 20th Interna-
tional Scientific Conference of Forensic Engineering.
ISBN 978-80-214-4238-2
[5] KROPÁČ, F., Issues Concerning Expert Assessment
of Collisions of Vehicles and Pedestrians under Poor Vis-
ibility Conditions, Brno: University of Technology in
Brno, Institute of Forensic Engineering, 2002
[6] KLEDUS, R.; BRADÁČ, A.; SEMELA, M.; CUPAL,
M. Experimental Research of Differences in Drivers' Per-
ception of Objects from Stationary and Moving Vehicles.
In 13. EAEC 2011 European Automotive Congress 13. -
16. 6. 2011 (Valencia) - Programme and proceedings
(DVD-ROM). EAEC. Valencia (Spain): EAEC 2011,
2011. s. 1-11. ISBN: 978-84-615-1794-7
[7] SULLIVAN J, FLANNAGAN M. Differences in the
Geometry of Pedestrian Crashes in Daylight and Dark-
ness. Journal Of Safety Research [serial on the Internet],
2011, [cited April 30, 2013]; 42(1): 33-37. Available
from: Academic Search Complete
[8] CLIFTON K, BURNIER C, AKAR G. Severity of In-
juries Resulting from Pedestrian–Vehicle Crashes: What
Can We Learn from Examining Urban Environment?.
Transportation Research: Part D [serial on the Internet],
2009, [cited April 30, 2013]; 14(6): 425-436. Available
from: Academic Search Complete
[9] Google Maps [Online]. 2012 [cit. 2013-01-02].
Available at: <maps.google.com>
Contact
Doc. Ing. Robert Kledus, Ph.D., e-mail: rob-
ert.kledus@usi.vutbr.cz
Ing. Bc. Marek Semela, Ph.D. e-mail:
marek.semela@usi.vutbr.cz
Ing. Pavel Maxera, e-mail: pavel.maxera@usi.vutbr.cz
Ing. Martin Kunovský, e-mail: mar-
tin.kunovsky@usi.vutbr.cz
Institute of Forensic Engineering VUT Brno, Údolní
244/53, (CZ)
117
 Robert Kledus , Marek Semela , Pavel Maxera , Martin Kunovský

118
© 2013 by EVU
EVU 2013 - 13

Recording and processing of data from crash tests per-

formed by Institute of Forensic Engineering Univerzity
of Zilina
Gustáv Kasanický - Pavol Kohút
Institute of Forensic Engineering Univerzity of Zilina

Abstract
For proper analysis of traffic accidents, it is important to properly evaluate the work on vehicle deformation based on the
extent of damage of the vehicle. The Institute of Forensic Engineering, University of Zilina, carrried out a series of
vehicle crash tests focused on evaluation of the vehicle deformation work. This article focuses on the recording and
processing data from crash tests and the evaluation of the vehicle deformation work, of the EES (energy equivalent
speed), the comparison of stiffness tests of cars (production year around 2000) and older generation of vehicles (which
were subject of tests - Crash tests DSD LINZ 1996).

CONDITIONS OF CRASH TESTS

Crash vehicles
For crash tests 5 vehicles with age structure shown in
the table 1 were used. By The yellow highlights the
period, when a given type of vehicle was produced,
while the red dot marks the production year of the
vehicle used in the crash test.
Tab. 1 - Age composition of vehicles
Figure 1 - Photo documentation of passenger cars (before
crash test)
Crash barrier
In crash tests the vehicles were accelerated to the
impact velocity by the Mercedes trailer and a pulley
system, before they were crashed into a solid con-
crete barrier of weight 10,000 kg (see fig. 2).

Photo documentation of individual vehicles before

relevant crash test, including EURONCAP evaluati-
on of the given vehicle, is shown in fig. 1.

Figure 2 - Concrete solid barrier

The weight of the barrier was not sufficient to ensure

that no significant displacement of the barrier
occured, and therefore the barrier was fixed using a
blade of a recovery tank of weight 35,000 kg (see
fig. 3).
119
 Gustáv Kasanický, Pavol Kohút

Fig. 3 - Recovery tank

THE RESULTS OF CRASH TESTS

The following photos (fig. 4, 5, 6, 7, 8) show the
damage to the vehicles after the crash tests, with the
accompnaying description of the photo including the Fig. 6 - Damage of the car Renault Megane Scenic (EES =
EES value, corresponding to the given damage. 72 km/h)

Figure 4 - Damage of the car Nissan Primera (EES = 54 Figure 7 - Damage of the car Honda Accord (EES = 58
km/h) km/h)

Figure 5 - Damage of the car Renault Laguna (EES = 62

km/h)
Figure 8 - Damage of the car Audi (EES = 18 km/h)

The course of vehicle deceleration was captured by

the crash meter, while the deceleration over time of
Renault Laguna vehicle is shown in the fig. 9.

120
 Recording and processing of data from crash tests performed by Institute of Forensic Engineering Univerzity of Zilina

v(t) - time-depending course of the velocity.

Consequently, from obtained data of the path and the

deceleration was constructed a path - deceleration
diagram (during respective tests). In following fig-
ures (fig. 11, 12, 13) individual diagrams for respec-
tive vehicles are shown. For the Renault Megane
Scenic vehicle, the crash meter broke down and a
correct record was not captured.

Figura 9 - Time - deceleration diagram

In the figure above is a considerable oscillation can

be seen, caused by natural frequency of the
measuring device due to the fixing method of the
measuring device, as well as by the oscillation of the
measuring point (trunk floor of the vehicle) during
impact. For further evaluation it was therefore
necessary to filter out the resulting oscillation, to
obtain (for each vehicle) t course of deceleration
against time of the given vehicle. Fig.10 ishows the Figure 11 - Deformation depth - Deceleration diagram
filtered course for Renault Laguna vehicle. (Renault Laguna – EES = 62 km/h)

Figure 10 - Time - deceleration diagram (Renault Laguna)

Figure 12 - Deformation depth - Deceleration diagram
(Honda Accord – EES = 58 km/h)
The time-dependent course of deceleration was inte-
grated according to following formula:

where:
a(t) - time-dependent course of deceleration.

By the aforementioned integration the velocity over

the time was obtained. This was integrated by the
following formula:
Figure 13 - Deformation depth - Deceleration diagram
(Nissan Primera – EES = 54 km/h)

All three diagrams show the course of deceleration

depending of the deformation depth, divided into
where:
121
 Gustáv Kasanický, Pavol Kohút

three approximately linear areas with characteristic
points A and B (see fig. 14).
for vehicles produced around 2000, there is almost
linear increase between points marked as A and B.
Consequently, from point B there is a significant in-
crease of deceleration, from a significant increase in
stiffness at the deformation depth of 0.5 m.
In the Ford Escort it was possible to identify an area,
where the deceleration was almost constant within
relatively large range of the deformation depth. This
effect was not detected in the vehicles tested by In-
stitute of Forensic Engineering, University of Žilina.
A significant increase of deceleration (and hence
stiffness) for the Ford Escort occurs only at the de-
formation depth of about 0.85 m.

Figure 14 - Deformation depth - Deceleration diagram

(Nissan Primera – EES = 54 km/h)

From a comparison of the diagrams the following

can be concluded:
1. Position of point B (approximately 0.1 m and 150
m/s 2 ) is almost identical for all three vehicles (Nis-
san Primera, Honda Accord and Renault Laguna).
2. Position of point B in terms of achieved decelera-
tion (approximately 300 m/s2) is almost identical.
3. Position of point B in terms of deformation depth
is similar for the Honda Accord and Renault Laguna
(about 0.53 meters), and different in comparison to
theNissan which was about 0.43 m.
Figure 16 - Comparison Deformation depth - Deceleration
A comparison of between the diagrams shown with diagrams of cars Ford Escort (Crash tests DSD LINZ
older generation of vehicles was performed using 1996) and Renault Laguna (Crash tests Institute of Foren-
analogous procedure conducted by DSD LINZ (Aus- sic Engineering, University of Žilina 2012)
tria) in 1996 with 4 crash tests of a Ford Escort vehi-
cle. CONCLUSION
. In fig. 15 the course of deceleration, deformation
depth and impact velocity for the four crash tests are From the tests performed is concluded, that for the
shown. newer generation of vehicles - production year
around 2000, it will be more difficult to properly
identify the EES of a given vehicle (under conditions
where EES exceeds 50 km/h), when compared to the
older generation of vehicles (which were subject of
tests - Crash tests DSD LINZ 1996). This is due to
the significantly different course of deceleration and
therefore also due to course of stiffness of vehicles,
at crash velocities above 50 km/h. This difference is
evident also from a comparison of the extent of dam-
age, as shown in fig. 17.

Figure 15 - Deformation depth - Deceleration diagram

(Ford Escort – DSD LINZ 1996)
From a comparison of the deceleration courses and
the deformation depth for thes Ford Escort (year of
manufacture about 1986) and the crash tests per-
formed by the Institute of Forensic Engineering,
University of Žilina (see fig. 16) it can be stated, that

122
 Recording and processing of data from crash tests performed by Institute of Forensic Engineering Univerzity of Zilina

[5] Ikrinský,A. - Patek,P. - Tichý,J.: Teória

dopravných prostriedkov (Theory of transportation
means), Bratislava,1987.

[6] Kasanický, G. – Kohút, P. – Lukášik, M.: Impact

dynamics theory for the analysis and simulation of
collisions, Žilinská univerzita in Žilina, Žilina, ISBN 80-
8070-312-4, 2004.

Contact
prof. Ing. Gustáv Kasanický, CSc.,
doc. Ing. Pavol Kohút, PhD.
Figure 17 - Comparison damage of cars Ford Escort Institute of Forensic Engineering
(Crash tests DSD LINZ 1996) and test car Nissan Primera University of Žilina
(Crash tests Institute of Forensic Engineering, University ul. 1. maja c. 32
of Žilina 2012) 010 01, Zilina, Slovakia
e-mails:
Significant differences between vehicles Ford Escort gustav.kasanicky@usi.sk,
and vehicles tested by Institute of Forensic Engineer- pavol.kohut@usi.sk
ing, University of Žilina are also reflected also in
terms of a shorter wheelbase and impact velocity tel: ++ 421 41 513 69 00
(see fig. 18).

Figure 18 - Comparison Change in wheelbase for cars

Ford Escort (Crash tests DSD LINZ 1996) and for test
cars Honda Accord, Nissan Primera, Renault Laguna and
Renault Megane Scenic (Crash tests Institute of Forensic
Engineering, University of Žilina)

References

[1] Appel, H. - Krabbel, G.: Unfallforschung, Un-

fallmechanik und Unfallrekonstruktion, Berlin, 1994.

[2] Burg, H.; Moser, A. (ed.): Handbuch Verkehrs-

unfallrekonstruktion, Unfallaufnahme – Fahrdynamik –
Simulation, GWV Fachverlage GmbH, Wiesbaden 2007,
ISBN 978-3-8348-0172-2.
[3] Burg, H – Rau, H.: Handbuch der Verkehrsun-
fallrekonstruktion Verlag Information Ambs GmbH,
Kippenheim, 1981.

[4] CD DSD LINZ, 1996.

123
 Gustáv Kasanický, Pavol Kohút

124
© 2013 by EVU
EVU 2013 - 14

Vehicle Accelerations and Times of Impact based on a

Data Analysis of Accident Data Recorders
Dipl.-Ing. (FH) Marvin Lammert
Dipl.-Ing. Armin Kast
Kast GmbH

Abstract
In this investigation, vehicle accelerations and the times of impact are statistically analysed. The data forming the base of
this investigation is the acceleration data from accident data recorders recorded in real traffic accidents. In these
accidents one party in the collision was equipped with an accident data recorder. To analyse the acceleration data of
accident data recorders in a way that was reproducible, an impact model was developed. As part of the investigation it
will be differentiated between the time of impact and the time of main load. The values of these two time parameters are
calculated and compared with each other. In addition, the mean vehicle accelerations during the time of impact and
during the time of main load are calculated and compared with each other.

Introduction
In 1973, the “Deutsche Verkehrsgerichtstag” called
for the introduction of accident data recorders in road
vehicles with the intention of gaining accurate data
for accident research and for accident reconstruction
[2].
Most of the currently used accident parameters in
accident reconstruction and in accident research,
such as acceleration and time parameters, were
established by the analysis of crash tests. Crash tests
are essential for accident research, accident
reconstruction and for vehicle development despite
modern simulation methods. This is because they
allow us to achieve comparable results under
precisely defined conditions, which are as realistic as
possible [3].
In accident research and in accident reconstruction
especially the knowledge of the real values of
accident parameters is of a great importance. It
allows future vehicles to be constructed so as to
safer, and more accurate reconstruction of traffic
accidents.
The aim of this study is to develop an impact model
by which the recorded acceleration-time-graphs from
accident data recorders could be analysed in way that
was reproducible. With the developed impact model,
vehicle accelerations and times of impact are
determined.
Database
The basis of this study are longitudinal and lateral
acceleration-time-graphs of real accidents recorded
by accident data recorders. At these accidents one
party in the collision was equipped with an accident
data recorder.
Overall, acceleration data from 200 accidents were
analysed. In collisions with slide impacts the time of
impact is usually extremely long. In collisions with
impact velocity changes less than 5 km/h there times
of impact is generally extremely short. Therefore
collisions with slide impacts and collisions with an
impact velocity change of less than 5 km/h need to
be considered separately. Therefore the acceleration
data from 82 traffic accidents was included in the
statistical analysis.
The impact velocities of the vehiclesequipped with
the accident data recorder, vary between 0 and
105 km/h. An impact velocity of 0 km/h indicates
that the vehicle equipped with the data recorder was
stationary at the beginning of the collision.
In this study traffic accidents with different types of
impacts and involving different types of vehicles are
analysed, impacts from the front, front lateral,
lateral, rear lateral and rear, collisions with passenger
cars, transporters, trucks and buses. As far as it is
known, most of the collision partners are cars, but a
collision with a motorcycle, a guardrail and a tree are
included in the evaluation as well.
Table 1 provides an overview of the different types
of impacts and of the different types of vehicles.
125
 Dipl.-Ing. (FH) Marvin Lammert, Dipl.-Ing. Armin Kast

Type of Impact Frequency (Σ 82) the positive lateral acceleration during the impact
front 13 shows an impact from the front right. This can be
seen in Figure 1, as well.
front lateral 36
lateral 27
rear lateral 1
Rear 5
Type of Vehicle Frequency (Σ 82)
Passenger car 39
Transporter 38
Truck 1
Bus 4

Table 1: Types of impacts and types of vehicles

A large number of the accident data recorders are

installed in agency and emergency vehicles. When
these vehicles are involved in a response, they are
often involved in junction accidents. Therefore the
front lateral and lateral types of impact are
represented in high numbers.
When using data from real traffic accidents, some
factors must be considered in the investigation. In
reality before a collision there is: a brake application;
an acceleration application; an avoid application;
and/or skidding occurs. These movements can be
identified by the recorded longitudinal and lateral
acceleration. These driving operations sometimes
Figure 1: Ambulance vehicle and passenger car
occur during and after the collision and are reflected
in the impact pulse. In some acceleration graphs 80

there are short resonance vibrations during the Impact from the front right
a-längs a-quer

impact pulse. 60

These factors must be considered in the 40

determination of the start of the crash, the crash end 20

and acceleration values such as maximum
a[m/s²]

acceleration during the collision. In order to 0

determine a reproducible and accurate result, the

-20
impact model by Kast has been developed.
-40
Braking
Impact Model by Kast -60
27,5 27,6 27,7 27,8 27,9 28 28,1 28,2
The impact model by Kast will be demonstrated in t [s]
an accident in which an ambulance and car collided
(Figure 1). Figure 2: Acceleration-time-graph
The longitudinal and lateral acceleration-time-graphs
of the ambulance vehicle recorded by the accident
data recorder are illustrated in Figure 2. This investigation is only concerned with the impact
pulse. When brake, acceleration, avoidance
At the beginning of the collision the ambulance applications and/or skidding before the collisions
vehicle was travelling at a speed of 26 km/h. The have occurred, the acceleration graphs have to be
negative longitudinal acceleration before the impact corrected. For this purpose, the longitudinal
shows that the brake of the ambulance vehicle was acceleration and/or the lateral acceleration at the
operated. The negative longitudinal acceleration and beginning of the collision have to be corrected to the
126
 Vehicle Accelerations and Times of Impact based on a Data Analysis of Accident Data Recorders

value zero. For the present accident that means the
longitudinal acceleration has to be set at zero at the
beginning of the collision due to the deceleration
resulting from braking.
In order to analyse the time of impact and the time of
main load of the collision the resultant acceleration
graph is calculated from the longitudinal- and lateral-
acceleration graphs. Time of impact is the total time
of the collision pulse. The time of main load is the
time range in which the resultant accelaration is
greatest [2]. In the time of main load on average
59 %, in an extreme case up to 91 %, of the total
impact velocity change occurs during the time of
main load.
The impact model by Kast is essentially described by
four points.
 Point P 1: start of the collision
 Point P 2: start of the time of main
load
 Point P 3: end of the time of main
load
 Point P 4: end of the collision
At the points P 1 and P 4, the acceleration values of
the impact model are set at zero. Point P 2 or point
P 3 describes the maximum acceleration of the
impact model. Because of resonance vibrations the
maximum acceleration of the impact model (P 2 or
P 3) the average of the acceleration values from the
resultant acceleration graphs.
In Figure 3 is shown the corrected longitudinal
acceleration graph (blue), the lateral acceleration
graph (pink), the resultant acceleration graph
(orange) and the linear impact model graph (grey)
with its four points identified.
80
Time of Main Load
60
P2
40
The four points of the impact model are connected
by three straight lines. The area below the impact
model respectively the area below the resultant
acceleration graph equates to the impact velocity
change. The area below the impact model between
the points P 1 and P 4 equates exactly to the area
below the resultant acceleration graph between the
points P 1 and P 4. The impact velocity change
serves here to verify the impact model by Kast.
To investigate the accurateness of the impact model
by Kast, the individual areas below the three straight
lines of the impact model: from point P 1 to point
P 2; from point P 2 to point P 3; and from point P 3
to point P 4, are compared to the areas below the
resultant acceleration graph.
On average there is a deviation in the individual
areas of the impact velocity change between the
impact model and the resultant acceleration of
± 0.086 m/s (± 0.31 km/h).
As the impact model by Kast, describes the impact
pulse with four points connected by three straight
lines, the difference of the impact velocity change is
low. So it is possible to describe and analyse the
impact pulses and to reproduce the results by using
the impact model by Kast.
In addition to the times of impact and the times of
main load, the mean acceleration during the time of
impact and during the time of main load are
statistically analysed using the impact model by
Kast.
In table 2, accident parameters of the present
accident are shown.
Transporter
Type of Vehicle (EDR)
(Ambulance Vehicle)
Type of Collision Partner Passenger Car
Type of Impact front laterally
Impact Velocity (Ambulance
26 km/h
Vehicle)
P3
Time of Impact 105 ms

20
Time of Main Load 82 ms
a[m/s²]

Maximum Acceleration 38,0 m/s² (3,9 g)

0

Mean Acceleration Time of

29,4 m/s² (3,0 g)
-20 Impact
P1
-40 Mean Acceleration Time of
32,9 m/s² (3,4g)
P4 Impact
-60
Time of Impact
27,75 27,8 27,85 27,9 27,95
t [s] Table 2: Accident parameters
a-längs a-quer a-res a-m od

Time Parameters
Figure 3: Acceleration-time-graph with impact model by
In figure 4 the frequency distribution of the times of
Kast
impact is shown in a histogram.

127
 Dipl.-Ing. (FH) Marvin Lammert, Dipl.-Ing. Armin Kast

16 σ ±20 ±19
14
min 70 12
12
max 156 102
10
Frequency

8
Table 3:Statistical and extreme time parameters
6

2 Acceleration Parameters
0 30
70 - 80

80 - 90

90 - 100

100 - 110

110 - 120

120 - 130

130 - 140

140 - 150

150 - 160
25

Time of Impact [msec] 20

Frequency
15

Figure 4: Frequency distribution of the times of impact 10

The times of impact are between 70 msec (min) and 5

156 msec (max). The arithmetic mean (µmean) of all 0

82 times of impact is 105 msec, the median (µmed) is

100 - 110
10 - 20

20 - 30

30 - 40

40 - 50

50 - 60

60 - 70

70 - 80

80 - 90

90 - 100
0 - 10
102 msec and the standard deviation (σ) ±20 ms.
Within the standard deviation around the mean Mean Accceleration Time of Impact [m/s²]

value, the times of impact varies from 85 to

125 msec. The calculated statistical parameters of the Figure 6: Frequency distribution of the mean acceleration
times of impact are consistent with the distribution in during the time of impact
the histogram. The histrogram in Figure 6 shows the frequency
The frequency distribution of the times of main load distribution of the mean acceleration during the time
is shown in figure 5 in a histogram. of impact.
18
The smallest value of the mean acceleration during
16 the time of impact is 5.6 m/s² (0.6 g). The highest
14 value is 102.4 m/s² (10.4 g). The arithmetic mean
12 value of all 82 accidents is 36.3 m/s² (3.7 g), the
Frequency

10 median is 29.0 m/s² (3.0 g) and the standard

8 deviation is ±23.3 m/s² (±2.4 g). Within the standard
6
deviation around the mean value, the mean
4
acceleration during the time of impact varies from
2
13.0 m/s² (1.3 g) to 59.6 m/s² (6.1 g).
0
100 - 110
10 - 20

20 - 30

30 - 40

40 - 50

50 - 60

60 - 70

70 - 80

80 - 90

90 - 100

In figure 7, the mean acceleration during the time of

main load is shown in a histogram.
Time of Main Load [msec]

20
18
Figure 5: Frequency distribution of the times of main load
16

The times of main load are between 12 msec and 14

Frequency

12
102 msec. The arithmetic mean of all 82 times of
10
main load is 46 msec, the median is 43 msec and the 8
standard deviation is ±19 msec. Within the standard 6

deviation around the mean value, the times of main 4

load varies from 27 msec to 65 msec. The calculated 2

0
statistical parameters of the time of main load are
100 - 110

110 - 120

120 - 130

130 - 140

140 - 150

150 - 160

160 - 170
10 - 20

20 - 30

30 - 40

40 - 50

50 - 60

60 - 70

70 - 80

80 - 90

90 - 100
0 - 10

consistent with the distribution in the histogram.

Mean Acceleration Time of Mean Load [m/s²]
In table 3, the calculated statistical values and
extreme values of the time of impact (timpact) and of
the time of main load (tmain) are shown. Figure 7: Frequency distribution of the mean acceleration
during the time of main load
Parameter timpact [msec] tmain [msec]
µmean 105 46 The margin of the mean acceleration during the time
µmed 102 43 of main load is between 7.1 m/s² (0.7 g) and
167.4 m/s² (17.1 g). The arithmetic mean value of all
128
 Vehicle Accelerations and Times of Impact based on a Data Analysis of Accident Data Recorders

82 accidents is 51.1 m/s² (5.2 g), the median is time of main load lies between 27 % and 61 % of the
38.9 m/s² (4.0 g) and the standard deviation is time of impact.
±33.8 m/s² (±3.4 g). Within the standard deviation
The ratio can be expressed by the following
around the mean value, the mean acceleration during
equations. Equation 1 describes this ratio in general.
the time of main load varies from 17.3 m/s² (1.8 g) to
The arithmetic mean was set with the tolerance of
84.9 m/s² (8.7 g).
the standard deviation for the factor ft into
Equation 2.
Looking at the distributions of the mean acceleration
tmain ═ ft ∙ timpact (1)
values during the time of collision and during the
time of main load, these are almost normally tmain ═ (0,44 ± 0,17) ∙ timpact (2)
distributed between 10 m/s² and 50 m/s² and between In figure 9, the frequency distribution of the factor fa
10 m/s² and 70 m/s² respectively. The acceleration is shown in a histogram. The factor fa describes the
values for both have a normal distribution. This can ratio of the mean acceleration during the time of
be explained by the fact that in this investigation main load to the mean acceleration during the time
different types of impacts with different types of of impact.
vehicles and different impact velocities were
analysed. 20
18
Table 4 shows the calculated statistical and extreme 16

values of the mean acceleration during the time of 14

Frequency
12
impact (aimpact) and during the time of main load
10
(amain). 8
6
Parameter aimpact [m/s² (g)] amain [m/s²(g)]
4
µmean 36,3 (3,7) 51,1 (5,2) 2
0
µmed 29,0 (3,0) 38,9 (4)
1,0 - 1,1

1,1 - 1,2

1,2 - 1,3

1,3 - 1,4

1,4 - 1,5

1,5 - 1,6

1,6 - 1,7

1,7 - 1,8
σ ±23,3 (±2,4) ±33,8 (±3,4)
Faktor fa
min 5,6 (0,6) 7,1 (0,7)
max 102,4 (10,4) 167,4 (17,1) Figure 9: Ratio of acceleration parmeters

Tabelle 4: Statistical and extreme acceleration parameters On average, the mean acceleration during the time of
main load is 40 % higher than the mean acceleration
during the time of impact. The standard deviation of
±17 % for the mean acceleration during the time of
Ratio of Time- and Acceleration Parameters main load is between 23 % and 57 %, higher than the
25 mean acceleration during the time of impact.
20 The ratio can be expressed by the following
equations. Equation 3 describes this ratio in general.
Frequency

15
The arithmetic mean was set with the tolerance of
10
the standard deviation for the factor fa into
Equation 4.
amain ═ fa ∙ aimpact
5
(3)
0
amain ═ (1,4 ± 0,17) ∙ aimpact (4)
0,2 - 0,3

0,3 - 0,4

0,4 - 0,5

0,5 - 0,6

0,6 - 0,7

0,7 - 0,8

0,8 - 0,9
0,1 - 0,2

Factor ft
Summary
The present investigation indicates that acceleration
Figure 8: Ratio of time parameters
data recorded by accident data recorders could be
analysed in a manner which could be reproduced.
Respecting the specified factors and tolerances the
In figure 8, the frequency distribution of the factor ft
impact model by Kast could be used for such an
is shown in a histogram. The factor ft describes the
analysis.
ratio of the time of main load to the time of impact.
Within the 82 analysed traffic accidents, the time of
On average, the time of main load is 44 % of the
impact is on average 105 msec, the time of mean
time of impact. The standard deviation of ±17 % the
129
 Dipl.-Ing. (FH) Marvin Lammert, Dipl.-Ing. Armin Kast

load is on average 46 msec, the mean acceleration Vereinigung für Unfallforschung und Unfallanalyse,
during the time of impact is on average 36.3 m/s² Brasov 2012.
(3.7 g) and the mean acceleration during the time of
main load is on average 51.1 m/s² (5.2 g). On [2] Von Aswegen A., Die biomechanische Belastung bei
average the time of main load is 44 % of the time of leichten Pkw-Seitenkollisionen – Eine experimentelle
impact. The mean acceleration during the time of Grundlagenuntersuchung mit Freiwilligen und Dummys,
main load is 40 % higher than the mean acceleration Diplomarbeit an der Bergischen Universität Wuppertal in
during the time of impact. Zusammenarbeit mit dem Ingenieurbüro Schimmelpfennig
und Becke, Wuppertal 1998.
A further investigation could examine the relevance
that these parameters have in terms of biomechanical
[3] Von Panajott A., Fahrzeugerprobung,
questions. Vorlesungsmanuskript Hochschule für angewandte
Collisions with slide impact and collisions with Wissenschaften München, München 2010.
impact velocity change less than 5 km/h should be
considered separately because of the extremely long Contact
and extremely short times of impact respectively. Dipl.-Ing. (FH) Marvin Lammert
Dipl.-Ing. Armin Kast
The wide scatter of mean acceleration shows that
more classifications are required. In another study, Kast GmbH
the mean acceleration should be classified according In den Seewiesen 80
to the impact velocity change, to the type of vehicle, 89520 Heidenheim an der Brenz
and to the type of impact. Germany

e-mail: marvin.lammert@kast-gmbh.de
References tel: +49 7321 96 32 0
[1] Von Glasner C., Die Geschichte der EDR-Entwicklung,
Tagungsband 21. Jahrestreffen der Europäischen

130
© 2013 by EVU
EVU 2013 - 15

The Usage Of Smartphones For Recording Accidents

And Incidents From The Critical Situation Up To The
Post-Crash Phase
Dr. Lars Hannawald
Mario Marschner
Henrik Liers
Verkehrsunfallforschung an der TU Dresden
Germany

Abstract
Smartphones are becoming more and more popular and not only for younger people. Contrary to traditional mobile
phones they are often equipped with sensors for acceleration and yaw rates, they have GPS modules as well as cameras
with high definition resolution. Additionally they have high-performance processors that enable the execution of CPU-
intensive tools directly on the phone. The wide distribution of these smartphones enables researchers to get a high
numbers of users for such studies.
The paper shows and demonstrates a software app for smartphones that is able to record different driving situations
leading up to crashes. Therefore all relevant parameters from the sensors, camera and GPS device are saved for a given
duration if the event is triggered. The complete configuration is independently adjustable to the relevant driver and all
events were sent automatically to the research institute for further processing. Direct after the event, interviews with the
driver can be carried out and important data regarding the event itself can be documented.
The presentation shows the methodology and gives a demonstration of the working progress as well as first results and
examples of the current study. In the discussion the advantages of this method will be discussed and compared with the
disadvantages.

Introduction
Changes in technologies from passive/secondary to
active/primary safety have become more and more
important. Due to that, the used data used will also
change from conventional impact and injury data to
all information prior to the crash. In figure 1 the real
accident database GIDAS is compared to naturalistic
driving data of VUFO.
If an accident occurs, the accident investigation team
Figure 1. Comparison of accident and incident data.
will be informed by the police or rescue services, so
that they can investigate the real data on the spot
immediately. Investigation of incidents
If an incident occurs, it will be much more difficult For those reasons, VUFO began to develop a new
to make sure that the team are informed. A tool for the investigation of incident data with the
completely new method of investigation of this data following boundary conditions:
is necessary to ensure an interview with the
participant occurs as fast as possible. A normal event  minimum installation effort in the vehicle
data recorder can detect crashes and strong near mis-  (preferably) no influence on the driver
ses, but a video-based analysis of the situation is not
applicable.  Tool should record video-, speed-, accelera-
tion-, gyro- and global position-data
 events should be triggered automatically

131
 Name Prename, Name Prename

 triggering should be possible depending on

position, by the exceeding of physical thresh-
olds or manually
 tool should be centrally configurable
To achieve a high number of participants in a
representative manner the tool should be easy to
handle for consumers as well as the study operators.
For an easier analysis of the data, the coding of the
parameters should be analog to the GIDAS database.
This also implements the simulation process analog Figure 3. Server environment.
to GIDAS with a Pre-Incident-Matrix (PIM). This
allows the use of the same tools and simulation fra-
mework that already exist for a real accident databa- To ensure a complete and independent investigation
se like GIDAS. of all parameters, the tool can be configured via a fi-
le transfer server. The complete data exchange is al-
In figure 2 the setup method is shown. so managed via a file transfer server. This guarantees
that the triggered events will be available as fast as
possible for the subsequent investigation of the other
relevant data in relation to the event. The study
participant can be also be interviewed concerning the
event by our experts within a short period of time.
In figure 4 the basic functionality of the VUFO
NDS-APP is shown.

Figure 2. Setup for investigation of naturalistic driving

data using smartphones.

Modern smartphones are equipped with

Figure 4. Basic functionality of VUFO NDS-APP.
 camera
 acceleration sensor The software records the GPS positions, all the
 gyro-sensor moving parameters and the video stream in a
 GPS sensor circular buffer. If an event is triggered, a sequence
of maximum 60 seconds prior to the event will be
 transceiver saved to the RAM of the smartphone. This event
 CPU and memory file will be sent via UMTS or WLAN connection
to the file transfer server immediately.
The described method is based on an application
for these smartphones which will record all the
necessary parameters in a circular buffer. In the next step VUFO will analsze this event and
call the participant to get further information, and to
For the central configuration of the application, establish if the event is of interest for VUFO NDS.
especially for the individual triggering parameters
the method bases on a server environment as In figure 5 the further tasks are shown.
shown in figure 3.

132
 Problems of uncertainty in time-distance analysis of pedestrian accident

Figure 5. Further tasks for events in VUFO NDS.

The event data will be combined with a scaled

sketch, and all available template data of the
vehicle and the participant and the interview data
will be used to produce a simulation of the event.
This file is called Pre Incident Matrix (PIM). All
parameters will be coded into the VUFO NDS da-
tabase additionally.

Data of VUFO NDS

With the described process VUFO NDS is able to
collect data in the following manner:
 driving behaviour
 incidents
 accidents
 manually triggered records
 position-based records
It is especially important to know the driving Figure 6. Recorded individual comfort zone for
behaviour of the participants.In addition, VUFO acceleration in x- and y-direction for the individual
participant.
NDS is using this data for an individual triggering
threshold for incidents involving this participant.
Incidents
Driving Behavior Incidents as well as more or less critical situations
Figure 6 shows a recording of a participant with are of interest in this project. As it is rare that
his individual comfort zone regarding longitudinal collisions occur, it is very important to analyse the
pre collision and incident phase.
and lateral acceleration. This comfort zone is ba-
sed on a record of all moving parameters for All warning strategies of ADAS are only
several days. appropriate when they are accepted. If acceptance
could be assumed in high critical situations, it will
become more and more difficult for less and non-
critical situations.

133
 Name Prename, Name Prename

Accidents

Accidents are very rare events. Only once every

500.000 km will one occur. Nevertheless if an
accident happens, the VUFO APP will record all
the relevant parameters prior to the crash. This da-
ta could be used by experts to prove the innocence
of the participant or at least provide data similar to
an EDR. Figure 8. Scenario of the test drive

The test car was equipped with a KIENZLE event

Manually triggered records
data recorder and the VUFO NDS APP. Both
measuring devices recorded the longitudinal and
Not all situations of interest can be triggered by lateral acceleration as well as the speed.
exceeding a moving parameter or by passing of
GPS positions. In that case the participant can
trigger an event manually. These events are Figures 9 and 10 show the validation results of
longitudinal and lateral acceleration.
recorded on the same basis as all other records by
the VUFO APP.

Position based records

It will be of particular interest to record individual
sequences at accident hotspots by vehicles passing
the spot. These records can be analysed via video
and moving parameters. In figure 7 a special ana-
lysis regarding loss of control accidents on an
accident hotspot is depicted.

Figure 9. Validation of the longitudinal acceleration.

Figure 7. GPS position based accident hotspot recor-

ding.

Validation of VUFO NDS

To validate the measurement results of the VUFO

NDS APP different test drives were initiated.
Figure 8 shows the scenario on the test track.
Figure 10. Validation of the lateral acceleration.
Both measurement results of the VUFO NDS APP
correlate with the accredited and calibrated results
from the event data recorder.

134
 Problems of uncertainty in time-distance analysis of pedestrian accident
Consumer benefit of VUFO NDS
Event Data Recorder
For the consumer and study participants of this
project the VUFO NDS APP will work as a nor-
mal event data recorder with additional video in-
formation. (see figure 11) This data could be used
by experts to prove the innocence of the
participant.
Figure 11 VUFO APP as event data recorder
A normal upgraded event data recorder costs
around 1000€ while the VUFO NDS APP is free
of charge.
Hazard Warning
The VUFO NDS APP could also be used as a
hazard warning of different daily driven situations.
Figure 12 shows show some of these possibilities.
Figure 12. Hazard warning functionality.
The warning threshold could be easily adjusted be
using the driver behaviour results as described
before.
Economic Driving
Economic driving is becoming more and more
important. VUFO NDS APP could help to drive as
economically as possible by measuring the real si-
tuation and comparing it to the average or most
effective one in the same situation. This will help
to reduce unnecessary expense. Figure 13 shows
the principle setup.
Figure 13. Economic driving functionality.
Conclusions
Detailed information about the pre-crash or pre-
incidence phase needs to be investigated with new
methods. The paper shows an application for
smartphones which is able to detect critical scena-
rios as well as record moving parameters of the
participant.
At the Accident Research Unit VUFO, this app is
used to build up a naturalistic driving database
alongside existing real world accident databases
(e.g. GIDAS).
The comparability to this database guarantees the
use of the same methods and simulation tools for
all future users.
The VUFO NDS APP could also be used by the
consumer as an event data recorder for crashes, to
provide warnings, information about hazards and
to provide information on economic driving.
Contact
Dr.-Ing. Lars Hannawald, lars.hannawald@vufo.de
Dipl.-Ing Mario Marschner mario.marschner@vufo.de
Dipl.-Ing. Henrik Liers, henrik.liers@vufo.de
Institute for Traffic Accident Research
Semperstraße 2a
01069 Dresden Germany
tel: +49 351 4389890
135
 Name Prename, Name Prename

136
© 2013 by EVU
EVU 2013 - 16

The Finite Element calculation module in PC-Crash 10.0

application of FE calculations in accident reconstruction
Dr. Andreas Moser, Dr. Hermann Steffan
DSD Linz - Austria, VSI Graz - Austria

Abstract
This paper presents the new finite element (FE) module for structural mechanics calculations in PC-Crash 10.0. The
theory of the models and typical applications are discussed. The model validation is presented as well using a set of
well-defined crash tests.
Currently most programs used in accident reconstruction cannot produce detailed information about the deformation of
vehicles and other parts well. Thus predictions about deformation energy in relation to deformations can be given only
very limited.
Finite Element calculations can close this gap. Using FE calculations, contact forces and deformations can be calculated
at a high resolution for a known geometry and material contribution. PC-Crash 10.0 closes this gap with integrating FE
calculations into the accident reconstruction program and work chain.

Introduction
Currently most programs used in accident
reconstruction cannot produce detailed information
about the deformation of vehicles and other objects
well. Thus predictions of the deformation energy in
relation with deformations can be given only very
limited.
Finite Element (FE) calculations can close this gap.
Using FE calculations, contact forces and
deformations can be calculated at a high accuracy for
a known geometry and material contribution 0. PC-
Crash 10.0 closes this gap with integrating FE
element calculations into the accident reconstruction
program and work chain. 0
This paper describes the different models, their
application and validation.
The finite element calculation method extends the
possibilities of PC-Crash in the analysis of vehicle
accidents into areas which can not be dealt with
currently. Questions like how much energy went into
or what force was produced in the deformation of a
trailer hitch or pole with a well-known geometry and
material can be answered in detail now. This opens
vast possibilities for further analysis in defining new
models and gaining more detailed information.
Figure 1: Deformed pole and vehicle after a crash test
(AREC 2011.09 impact velocity 10.8 km/h) 0
In the accident reconstruction process very often
detailed information about the geometry, material
and deformation of objects and vehicle parts is
known or can be collected (Figure 1 to Figure 4).
However, it is unsatisfying that this information can
not be used or evaluated in most current accident
reconstruction programs.
Crash tests like in Figure 1 can be used to compare
the real accident and the test to gain information
about the impact velocity, velocity change of the
vehicle and vehicle acceleration in a real accident.
However, this would only provide reasonable results
if the configuration (diameter of the poly, wall
thickness, material, bumper structure) is very similar.
Although crash tests give very detailed and accurate
information extrapolation to different configurations
137
 Andreas Moser, Hermann Steffan
have to done carefully and may not be possible. To
gain reliable information in crash tests many tests
have to be performed.
Crash tests are of course extremely useful to
understand the collision and behaviour of the
colliding objects and the damage generation in all its
details. Also for model validation crash tests are
essential.
Figure 2: Deformed front bumper structure after a colli-
sion against a trailer hitch1
Figure 3: Deformed trailer hitch2
Figure 4: Deformed structure after a minor collision
Model description
The finite element code in PC-Crash uses an explicit
solver to calculate element stresses, strains and
displacements. This code can solve problems in
linear and nonlinear transient dynamic analysis. A
contact detection module using contact elements is
integrated as well.
1
Source: Förtig GbR, Ingenieur-Büro für Kfz-Technik
2
Source: Ingenieurbüro für Fahrzeugtechnik von der
Wehd
138



Figure 5: Explicit finite element procedure (central differ-
ence method) 0
The main calculation procedure for the explicit finite
element method is outlined in Figure 5. Once the
nodal displacements for a given timestep have
been calculated () the nodal strains can be cal-
culated, the stress/strain law of the material used de-
termines the relationship between stress and strain
and hence the stresses can be calculated (). As
a results the nodal accelerations are calculated
based on the internal and external nodal forces ().
Based on the nodal forces the nodal accelerations are
calculated, the new nodal velocities and dis-
placements are then calculated through integration
(). These steps are then repeated for all integration
steps.
In the finite element method complex problems in
terms of geometry and formulation are subdivided
into smaller and simpler finite regions. This leads to
the following advandages:
 Accurate representation of complex geome-
try
 Inclusion of dissimilar material properties
 Easy representation of the total solution
 Capture of local effects.
For the calculation a 3D mesh describing the object
geometry to be modelled needs to be created. The
nodes and faces of the mesh represent the nodes and
elements of the model. In the finite element method
elements can have a varying number of nodes with a
different number of degrees of freedom for surface,
volume and line elements. Each element may then
have a different material and different parameters for
contact detection and friction.
 The Finite Element calculation module in PC-Crash 10.0 – application of FE calcula-tions in accident reconstruction
Elements
The following element types are implemented:
 Shells are used to model thin-walled struc-
tures where the thickness is less then approx.
10 % of the element dimension. 3 node shell
elements according to the classical C0 for-
mulation (Shell_C0_3) and 4 node shell el-
ements according to the Belytschko-Tsai
formulation (Shell_BT_4) are available 0, 0,
0 (Figure 6). Shell elements can have up to 3
degrees of freedom in their nodes and do
have integrated contact elements.
 Solids are used to model volumes. These el-
ements can have 4 to 8 nodes (Figure 7),
each node has 3 degrees of freedom (dis-
placements).
 Rods are 2 node elements to model lattice
structures or to connect parts (spotwelds).
These elements have integrated contact ele-
ments, each node has 3 degrees of freedom.
 Spring damper elements are connection ele-
ments to connect different parts of the mod-
el. Translational and rotational stiffness and
damping can be specified.
Figure 6: Shell elements
Figure 7: Different solid elements
Nodes
The nodes in the model form the elements. In the
calculation the mass and moment of inertia of the el-
ements is concentrated at the nodes (lumped mass
approach). Moreover nodes can be assigned an addi-
tional mass and moment of inertia. Loads, boundary
conditions and constraints can be applied to nodes.
Constraints
Two different contraints can be specified for the
nodes:
 Boundary conditions are used to define ini-
tial conditions like initial velocity or acceler-
ation to nodes. Linear and angular accelera-
tions and velocities can be specified in all 3
directions either in the inertial frame or in a
local coordinate frame. A time dependent
function can be specified. In this way a ve-
locity for nodes can be defined or certain
nodes can be fixed if they are fixed on the
road. Also kinematic constraints can be
specified.
 Rigid body contraints are used to combine a
number of nodes as part of one rigid body
connected to a master node. This can be used
if parts of a surface should be fixed to the
center of gravity of the model.
Loads
To apply loads to the model forces, torques, linear
and angular accelerations as well as pressures can be
defined either as constant or time dependent values
which will be applied to the nodes.
Contact calculation
The module in PC-Crash uses a penalty based con-
tact algorithm. Surface and line contact algorithms
are available. Some element types have integrated
contact elements (shells and rods) for others like sol-
ids explicit contact surfaces have to be defined, as
only a part of the solid elements surfaces would act
as contact surfaces (e.g. for the engine block consist-
ing of several solid elements).
Mesh generation
The mesh generation if one of the key steps in creat-
ing the finite element model. In the mesh generation
step the mesh for the model is created based on the
object geometry. Depending on the element types to
be used for the analysis the mesh can consist of 3 or
4 node surface elements, volume elements and line
elements.
139
 Andreas Moser, Hermann Steffan
Figure 8: Mesh generation of a structural element in gmsh
0
Different programs are available to perform this task
also free programs like gmsh 0 are available (Figure
8). Many meshing programs as gmsh can use param-
eters to describe the model geometry so changes to
the dimensions of an object can be done easily with-
out the need of regenerating the whole geometry.
The finite element preprocessor in PC-Crash can
read gmsh and GID 0 files directly, material and el-
ement properties as well as boundary conditions and
constraints can then be specified from within PC-
Crash.
Material models
Three different isotropic material models (material
properties are the same in all directions) with the fol-
lowing parameters are implemented:
 Elastic material model:
o E: Young’s modulus
o ρ: material density
o v: Poisson ratio
o σF, εF: failure stress and failure strain
 Elastoplastic material model: In addition to
the above parameters the platic behaviour of
the material can be specified in one of the
following three ways
o σY, EP: yield stress and strain hard-
ening coefficient
o σ=f(ε): stress/strain curve (as in Fig-
ure 9)
o σ=f(ε,έ): strain rate dependent
stress/strain diagrams
 Thermoelastoplatic material: In addition to
the elastoplastic material parameters the fol-
lowing values can be specified
o α: thermal expansion coefficient
140
o T: nodal temperatures
Figure 9: Stress/Strain curve showing typical yield behav-
ior with true elastic limt (1), proportionality limit (2), elas-
tic limit (3) and offset yield strength (4) 0
Depending on the material model used all of the
above parameters have to be specified or only a sub-
set of the parameters. PC-Crash provides a library of
different material parameters which can be used as
an alternative to obtaining the parameters from man-
ufacturer of the object.
Using a failure stress or failure strain value material
failure can be modeled as well, once the failure
threshold is exceeded for an element this element is
removed for the further simulation.
Results
The results of the simulation can be visualized (post-
processed) in different ways directly in PC-Crash.
The results are also save in GiD 0 format so external
post processing programs can be used as well.
The whole model with color coded node and element
values for stress, strain and displacement can be
viewed for a given point of time in the simulation,
animations can be generated as well so the dynamics
can be seen easier (Figure 10).
Diagrams for all relevant nodal values can be gener-
ated for selected nodes or groups of nodes. (e.g. nod-
al accelerations as in Figure 15). The following dia-
grams are available
 Contact forces
 Energy (internal and external)
 Node acceleration, displacement, distance,
forces, moment, velocity
 Rod forces
 Section forces (e.g. forces for a certain cross
section)
 The Finite Element calculation module in PC-Crash 10.0 – application of FE calcula-tions in accident reconstruction

Figure 11: Impact configuration in AREC 2011.09 (10.8

km/h) and AREC 2011.12 (19.4 km/h)
Vehicle
Make Ford
Model Focus Turnier
Figure 10: Color coded element stresses and nodal dis- Test weight 1270 kg
placements are displayed for a contact between a contact
surface and a trailer hitch at contact and 3, 6 and 9 ms af-
Pole
ter first contact.
Diameter 76.1 mm
Calculation time Wall thickness 3.6 mm
As the number of calculations needed for the finite Material Steel S235JR
element model is significantly higher than for a rigid Elements (Shell_C0_3) 2098
body vehicle simulation the simulation time for the Nodes 1107
finite element model is higher. However, with the
Mesh resolution 40 mm
calculation power of modern computers with multi-
core or multi-processor architecture simulations, as Figure 12: Vehicle and pole data for the crash tests
shown in the validation part of this paper, can be per-
formed in approx. 400 s for 600 ms of simulation As only very little deformation has been recorded
time on an Intel i7 processor. PC-Crash also provides during the crash tests the vehicle surface was mod-
batch processing and support for multi-core and mul- eled only. Deformation on the surface is allowed,
ti-processor systems. Using batch processing a num- however, if the deformation were large the structure
ber of simulations can be configured first (e.g. dif- underneath should be modelled as well.
ferent impact velocities, boundary conditions etc.)
and then all these different simulations can be simu-
lated in batch mode so no user interaction is needed
during the calculation. Once the calculations are fin-
ished the results can be imported into the program
again. In this way the workflow can be optimized.

Validation
During the AREC conference in 2011 several crash
tests between a car and a pole have been performed.
The vehicle movement, velocities and accelerations
have been recorded using accelerometers and high
speed video. The tests where the pole was hit in the
middle of the front or rear of the vehicle are used for
the following validation of the FE model in PC-
Crash.
The impact configuration and vehicle and pole data
are shown in Figure 11 and Figure 12.
Figure 13: Simulation model for the pole impact simula-
tion, only the vehicle surface is modeled as the defor-
mations on the vehicle are not very deep (2098 elements,
1107 nodes)
The comparison of the simulation results for the
movement of the vehicle and the pole in Figure 14 as
well as the comparison of the vehicle acceleration in
the test and in the simulation in Figure 15 show a
very good correlation between the simulation and the
crash test.
141
 Andreas Moser, Hermann Steffan

The simulation model is able to represent the effects

of the pole impact occurring during the test. In addi-
tion using the simulation model variations for the
impact parameters can be done very easily. In this
way the unknown areas between the test performed
can be filled using simulation.
The velocity vs time diagram in Figure 16 shows
from which impact velocity the vehicle will drive
through the pole and the pole is fully bent to the
road. There is a certain velocity change for the vehi-
cle but the vehicle still has a forward velocity after
the collision.
Parameters for the momentum based impact analysis
like delta-v and the coefficient of restitution can be
derived as well as shown in Figure 17 and Figure 18.

Figure 14: Comparison of calculation results (right) and

test data (left) for test AREC 2011.09 (10.8 km/h)
AREC 2011.09 vehicle acceleration vs FE calculation
25
res40 3.6mm 10.8 kph 280
res40 3.6mm 10.8 kph 235
20
res40 3.6mm def 10.8 kph 235
Acc x [m/ss]
Acceleration [m/s²]

15

10

-5
-0.1 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
Time [s]

Figure 15: Comparison of calculation results (vehicle ac-

celerations for yield stress of 280 and 235 N/mm², with
and without vehicle deformation) and test data for test
AREC 2011.09 (10.8 km/h)
Vehicle velocity - FE simulation
25

20

15
velocity [km/h]

10

-5
0 0.1 0.2 0.3 0.4 0.5 0.6
Time [s]

Figure 16: Velocity vs time diagram for different impact

velocities (simulation)

142
 The Finite Element calculation module in PC-Crash 10.0 – application of FE calcula-tions in accident reconstruction

Delta-v conduction can be spend in developing and validat-

16
AREC2011
ing additional models.
14 RearL-11.11
FE calculation

12 RearM-11.09 FrontM-11.16
Even if many crash tests are performed parameter
RearL-11.10 variation is difficult to do, e.g. if the wall thickness
Delta-v [km/h]

10

8
FrontL-11.14 of the pole should be varied a complete new set of
6 RearR-11.12b
tests has to be performed. Here the strength of the
4
RearM-11.12 simulation can be used to do these variations based
2
on a model validated through a limited number of
crash tests.
0
0 5 10 15 20 25 30 35 40 45
vk [km/h] The finite element model in PC-Crash now provides
the means for a complete new set of possiblities for
Figure 17: Comparison of calculation results (vehicle ve- the analysis of accidents. Many applications in the
locity change) and test data
field of accident reconstruction where deformation
behaviour is to be analysed can be considered.
The more complex the models are of course the more
Restitution
validation work is needed. Additional models and
0.60 their validation will be performed by the autors but
0.40
AREC2011
FE calculation
also validated finite element models are available
0.20 RearR-11.12b RearL-11.11 through the National crash analysis center (NCAC)
Coefficient of Restitution

0.00 RearM-11.09 0. Some of these validated models are shown in Fig-

-0.20 ure 20, models for different vehicle types and cate-
-0.40 gories as well as models for guardrails, barriers,
-0.60 RearL-11.10 FrontM-11.16
impactors and other object on the roadside are pro-
-0.80 RearM-11.12 FrontL-11.14 vided. In interface to these models will be included
-1.00
in PC-Crash.
0 5 10 15 20 25 30 35 40 45
vk [km/h]

Figure 18: Comparison of calculation results (coefficient

of restitution) and test data
AREC 2011.12 vehicle acceleration vs FE calculation
30
res40 3.6mm 20 kph 280
25 Acc x [m/ss]

20
Acceleration [m/s²]

15

10

-5
-0.1 -0.05 0 0.05 0.1 0.15 0.2 0.25 0.3 0.35 0.4
Time [s]

Figure 19: Comparison of calculation results (vehicle ac-

celerations) and test data for test AREC 2011.12 (19.4
km/h)
Figure 20: Validated finite elements models from NCAC
0
Summary/Outlook
The simulation model has shown a very good corre-
lation with the test data in the above described crash
tests. Using finite element calculation in conjunction References
with crash tests helps to validate the model on the [1] Nasdala L., FEM-Formelsammlung Statik und
one hand and to provide additional data on the other Dynamik, 2. Auflage 2012, Springer Vieweg
hand, so the number of crash tests can be reduced. [2] Shen R. Wu, Lei Gu, Introduction to the explicit finite
This combination is very effective as performing element method for nonlinear transient dynamics, John
crash tests is expensive and time consuming and a Wiley & Sons, Inc. 2012
part of the workload for crash test preparation and
143
 Andreas Moser, Hermann Steffan

[3] AREC – Accident reconstruction conference group,

www.arecgroup.info
[4] NCAC - National Crash Analysis Center, 45085 Uni-
versity Drive, Ashburn, VA 20147, www.ncac.gwu.edu
[5] PC-Crash 10.0 operating and technical manual, DSD
Linz, Austria
[6] Gmsh: a three-dimensional finite element mesh genera-
tor with built-in pre- and post-processing facilities
(www.geuz.org/gmsh/)
[7] Ted Belytschko, Wing Kam Liu, Brian Moran, Nonline-
ar Finite Elements for Continua and Structures, 2009, John
Wiley & Sons Ltd.
[8] Klaus Jürgen Bathe, Finite Element Procedure, 2005,
Prentice Hall
[9] GiD: a universal, adaptive and user-friendly pre and
postprocessor (www.gidhome.com)
[10] Wikipedia
http://en.wikipedia.org/wiki/Stress%E2%80%93strain_curv
e

Contact
Andreas Moser, PhD
DSD – Dr. Steffan Datentechnik GmbH
Salzburgerstr. 34
A-4020 Linz, Austria
e-mails: moser@dsd.at
tel: ++43 732 343200

Hermann Steffan, PhD, Univ. Prof.

DSD – Dr. Steffan Datentechnik GmbH
Salzburgerstr. 34
A-4020 Linz, Austria
e-mails: h.steffan@dsd.at
tel: ++43 732 343200

144
© 2013 by EVU
EVU 2013 - 17

Bypassing manoeuvre driving a car with reduced and no

tire pressure
Jakub Zębala, Wojciech Wach, Piotr Ciępka
Institute of Forensic Research, Kraków Poland
Robert Janczur
Cracow University of Technology, Poland

Abstract
The objective of the paper is to present the results of investigations to determine the effect of reduced tyre pressure on
car lateral dynamics in an obstacle bypassing manoeuvre. This was done by performing the bench and road tests. The
bench tests were carried out for parameterisation of the mathematical model of the tested car. The road tests covered the
vehicle motion with reduced and no tyre pressure on a curvilinear track adequate for bypassing an unexpectedly
appearing obstacle. Next, simulations in PC-Crash were performed, and the results were compared with those obtained
in experiments.

AIM AND SCOPE OF THE PAPER

INTRODUCTION
The aim of the research was to determine the effect
Road accident reconstruction includes the recon- of reduced tyre pressure on car lateral dynamics in
struction of the car motion after the crash. This mo- an obstacle bypassing manoeuvre.
tion is often disturbed by pressure reduction in one This aim was achieved by means of bench and road
of the wheels resulting from tyre damage resulting tests during which the vehicle motion parameters
from the impact, the vehicle driving over elements of were recorded. Next, simulations were performed,
the damaged vehicles, or the tyre flange slipping off and the results were compared with those obtained in
the rim during the accident. In an extreme case the experiments.
tyre loses its air-tightness completely, which causes The object tested was a 1999 Deawoo Nubira pas-
the vehicle tilt, decreases the cornering stiffness and senger car with Pirelli P6000 185/65 R14 86H tires.
radial tyre stiffness, as well as increasing rolling re-
sistance, which in turn affects the vehicle dynamic BENCH TESTS
characteristics, stability and steerability [1,10]. A car
on straight track motion with reduced tyre pressure The aim of the bench tests was parametrisation of the
has been investigated thoroughly and the research re- car’s model in the simulation programs. These in-
sults published seem sufficient for the needs of acci- cluded determination of:
dent reconstruction [2,3]. The tests presented herein,  mass distribution,
therefore, covered the vehicle motion with reduced  centre of gravity (CG) location in 3D,
and no tyre pressure on a curvilinear track adequate  moments of inertia about the vertical axis,
for bypassing an unexpectedly appearing obstacle. In  characteristics of suspension stiffness and shock
the motion on a curvilinear track lateral forces occur, absorber damping,
which make the damaged tyre slide along the rim.  steering ratio.
This changes the adhesion coefficient in relation to The data determined in the bench tests and adopted
both tyre-road surface and tyre-steel rim mutual re- for calculations in PC-Crash program have been
action. The vehicle lateral dynamics is heavily af- listed in Table 1.
fected by the reduction of tyre resistance to lateral
slip. All these phenomena do not allow it to be stated Table 1. Technical data of tested car
definitely whether driving a car with a damaged tire Variable Value
the driver can efficiently perform an obstacle bypass- Vehicle mass 1374 kg
ing manoeuvre. Distance of CG from front axle 1.108 m
CG height 0.54 m
Steering system ratio 16.9:1
Stiffness coefficient – front right
16500 N/m
wheel suspension
Stiffness coefficient – front left wheel 23800 N/m
145
 Jakub Zębala, Wojciech Wach, Piotr Ciępka, Robert Janczur

suspension
Stiffness coefficient – rear right wheel
23400 N/m
suspension
Stiffness coefficient – rear left wheel
22900 N/m
suspension
Moment of inertia – roll 588 kgm2
Moment of inertia – pitch 2380 kgm2
Moment of inertia – yaw 2478 kgm2
Figure 2. The order of the reduction and
Moments of inertia roll and pitch were calculated us- removal tire pressure on the wheels
ing Bixel’s formulas [4].

ROAD TESTS TESTING EQUIPMENT

The road tests were performed on a flat and horizon- In Table 2 the parameters of the devices and sensors
tal section of a concrete surface in summer condi- used in the road tests have been shown.
tions. The track for the lane change, shown in Figure Table 2. Measurement unit used in road tests
1, was laid following the recommendations of norm Measure-
Measure-ment Measure-
project BN-90/3615- /06, the aim of which was to device
Parameter
men range
ment preci-
determine the properties of a passenger car in by- sion
passing an unexpected obstacle on a road straight 2-Axis Optical Longitudinal 0–97 m/s
0.03 m/s
Sensor velocity 0–350 km/h
section [5]. In addition, posts indicating the Correvit Lateral ± 62 m/s
trajectory were numbered for precisely describing 0.03 m/s
Corrsys®S-CE Velocity ± 225 km/h
which of them was knocked by the car during the Crossbow’s Acceleration ± 4,0 g < 0,5 mg
bypassing manoeuvre. 440 Series
Inertial Angular Rate ± 200 º/s < 0,02 º/s
System
Universal Steering
± 1250º ≤ 0,1º
Measurement angle
Steering Wheel Steering
± 10/50 Nm ± 0,15%
Kistler Group Moment
Linear 0.03 m/s
GPS data 0.1-447 m/s
velocity 0.1 km/h
logging
Figure 1. Lane change trajectory after norm BN-90/3615- Heading 0–360º 0.1º
System (RLV
/06 adopted in tests 0.05 %
B3iR10G10) Distance
<50 cm/km
Initially, the pressure in all the wheels of the tested Acceleration:
car was of nominal value of 2.0 as recommended by Longitudinal,
Inertial ± 1.7 g 0.01 g
Lateral,
the manufacturer. In the following tests, before the Measurement
Vertical
car motion started, the air pressure was reduced in Unit
Angular
(RLVBIMU03)
one wheel to the values of 1.5 and 1.0 and finally to Rate: Roll, ± 150 º/s 0.01 º/s
0 in successive tests. The pressure in one wheel was Pitch, Yaw
reduced before test commenced by a gradual reduc-
tion of the pressure while the car was approaching The movement of the vehicle and selected wheel was
the beginning of the measurement track [6]. The runs documented by three digital cameras CASIO EX-F1
with reduced pressure were repeated successively for with the frequency 25 and higher 300 fps. After each
all the wheels (Figure 2). In the final tests the pres- test the wheels were photographed.
sure was completely eliminated in all the wheels of The friction coefficient of the concrete surface was
the tested car. established based on series of braking tests of the
In each test the following values were measured: lin- tested vehicle with two wheels locked according to
ear and angular velocities, linear acceleration, steer- the Regulation No. 13 UNECE [9]. In order to estab-
ing wheel angle, force moment on the steering lish the acceleration, two independent systems were
wheel. In total 10 parameters of tested vehicle used:
movement were recorded.  the Crossbow DMU-FOG (direct measurement of
accelerations and angular velocities in three di-
rections),
 the Correvit Corrsys® S-CE gauge head (indirect
establishing of acceleration based on numerical
differentiation of measured velocity).

146
 Bypassing manoeuvre driving a car with reduced and no tire pressure
VEHICLE MOTION VISUALISATION IN
PC-CRASH PROGRAM
A GPS-based Racelogic VBOX measurement system
used in the road tests enabled recording of the
motion trajectory and velocity of the tested vehicle
[8]. The GPS antenna fixed on the vehicle roof above
its CG. recorded the geographic position of the CG,
the velocity as well as its deviation from the north
Figure 3. Vehicle motion visualisation in PC-Crash program
SIMULATIONS IN PC-CRASH
The vehicle motion was simulated in PC-Crash 9.2
for a linear tyre model [7,11]. The TMeasy model
was disregarded because no actual tyre
characteristics were available. In the linear tyre
model the maximum slip angle is the equivalent of
cornering stiffness. In the simulations both the
default linear tyre characteristic for slip angle of 10o,
and a modified characteristic obtained by changing
the slip angle value have been used. The angle range
was changed so as to meet the following criteria:
 the vehicle moving along the assigned trajectory
in the manner recorded in tests including driving
into the posts marking the trajectory (criterion 1),
 deviation of the vehicle motion trajectory in the
simulation from the trajectory recorded in a test
(criterion 2).
To evaluate quantitatively the compatibility of vehi-
cles positions obtained in simulations with the results
of real ones, a quality function was selected, de-
scribed by formula:
n
 w  q  i i
2
Q i 1 ,
n  analysis of the criteria of simulation assessment
wi 1
2
i by giving the number of the knocked post a value
where:
qi relative difference between the real value of pa-
rameter i and that obtained in simulation,
wi weighting coefficient of parameter i.
The function enables comparison of the values of
different physical quantities. The relative difference
qi was determined for x, y coordinates for each inte-
grated step. In the authors’ opinion, the position of
vehicle is a crucial indicator of the quality of the
simulation, hence the maximum value of weighting
direction. To use the obtained data the co-ordinates
had to be rotated by the angle formed by the
longitudinal axis of the test trajectory and north
direction. The resulting table of data was imported in
text format to PC-Crash in which the vehicle motion
trajectory was visualised. An example of such
trajectory with tyre pressure removed from all the
wheels has been shown, from the start to the final
stop (Figure 3).
coefficients of these parameters, equal to 1
(qx=qy=1), was adopted.
In the simulations with a vehicle whose all wheels
had a nominal pressure the slip angle was reduced to
a value at which the simulation course was found to
be satisfactory. The determined values of the slip
angle were used in subsequent simulations for
wheels with nominal pressure.
The steering wheel angle changes in time were
approximated by linear functions to perform the
simulation. Each linear section of the slip angle was
a single sequence of the vehicle motion.
The simulation started a few seconds before entering
the assigned motion trajectory. The vehiclse initial
velocity corresponded to the velocity measured in
the test and the velocity in subsequent sequences of
the vehicle motion was corrected and also
corresponded to the recorded velocity. The
simulation was considered completed when the
vehicle stopped or its motion was stabilised.
In all the simulations a global adhesion coefficient of
0.78 was adopted, value and was a mean value
reached in tests.
The simulation results covered:
 displacement of vehicle CG in experiment and
motion simulation,
of function Q.
The results have been shown in a graphic form (see
the appendices).
For a simulation that met the criteria the following
was given in detail: motion trajectory visualisation,
front wheels steer angle as a function of time, linear
velocity as well as yaw and lateral acceleration.
TEST RESULTS
The recorded tests covering the assigned trajectory
have been tabulated (Table 3) including number of
tests, the vehicle’s initial velocity on entering the
147
 Jakub Zębala, Wojciech Wach, Piotr Ciępka, Robert Janczur

trajectory as well as the number of knocked posts Left front wheel – 1.0 at
and skidding (or its lack) following bypassing Simulations of a vehicle with the pressure in the left
manoeuvre. front wheel of 1 and conducted with the default
characteristics failed to meet criterion 1 (Table 8). In
Table 3. Measurement results
these simulations the vehicle CG deviated the most
No. Test Pressure Velocity Remarks from the trajectory assigned in tests. Reduction of
1 All wheels 2.0 at 47 km/h -
2 All wheels 2.0 at 61 km/h Post 6 slip angle of all the wheels did not diminish the
3 All wheels 2.0 at 51 km/h - discrepancies. The lowest values of function Q were
4 Left front wheel 1.5 at 52 km/h - obtained after individual corrections of the slip angle
5 Left front wheel 1.5 at 51 km/h -
6 Left front wheel 1.0 at 51 km/h - of the wheel in which the pressure was reduced to
7 Left front wheel 0.0 at 60 km/h Post 6 the value
8 Right front wheel 1.5 at 51 km/h -
9 Right front wheel 0.0 at 48 km/h Post 6
of about 10°.
10 Right front wheel 0.0 at 48 km/h -
11 Right front wheel 0.0 at 60 km/h Post 6 Left front wheel – 0.0 at
12 Left rear wheel 1.5 at 52 km/h Post 6 Simulations of the motion of a vehicle with the
13 Left rear wheel 1.5 at 52 km/h -
14 Left rear wheel 0.0 at 50 km/h - pressure in the left front wheel eliminated
15 Left rear wheel 0.0 at 61 km/h
Post 6 completely and with an identical slip angle in all the
Skidding wheels also failed to meet the criteria (Table 9). The
16 Right rear wheel 1.5 at 48 km/h -
17 Right rear wheel 1.5 at 52 km/h - criterion 1 was met after the reduction of slip angle
18 Right rear wheel 0.0 at 51 km/h - for wheels with nominal pressure and increasing the
Post 3,6
19 Right rear wheel 0.0 at 60 km/h
Skidding
angle to the value of 11° for the wheel with no
Post 6 pressure (Figure 25). The higher value than 11°
20 All wheels 0.0 at 37 km/h
Skidding resulted in a knocking of post number 6 by the front,
Post 6,7
21 All wheels 0.0 at 46 km/h
Skidding
not by right side of the car as occurred during
experiment. The lowest value of the function Q
obtained, about 30°, for the wheel with no pressure
SIMULATION RESULTS (Figure 26).
Right front wheel – 0.0 at
All wheels – 2.0 at (nominal value)
As in the case above, the criteria were met in the
Three runs of a vehicle with all wheels at the
simulation when the pressure in the right front wheel
nominal pressure were performed. When default
was completely eliminated, with reduced slip angles
linear tyre characteristic was applied, the simulations
in wheels with nominal pressure and the angle for
failed to meet the adopted criteria. In each simulation
the wheel with no pressure at about 32° (Table 10).
for the angle of 10° the vehicle drove into post no 6
and a significant deviation of the vehicle’s CG Left rear wheel – 0.0 at
trajectory from the position defined during the Simulation of the motion of a car with pressure
experiment was observed. The most favourable eliminated from the left rear wheel was the most
results were obtained for modified linear difficult. The problem was that the car skidded and
characteristics in the slip angle range of 7° to 3°. In in order to reach a stable motion, the driver was
this range the value of function Q reached the lowest required to perform a number of corrections of the
values (Tables 4, 5 and 6). Within this slip angle motion trajectory. Slip angle change at the level of
range the vehicle’s motion dynamic parameters such hundredth parts of a degree caused significant
as lateral acceleration and yaw velocity were changes of the vehicle motion trajectory. The result
convergent with the results obtained experimentally, of the simulation performed in order that the CG
to an extent sufficient for road accident trajectories in the simulation and experiment were
reconstruction (Figures 10 and 11). close to each other has been shown below. In this
Left front wheel – 1.5 at simulation the slip angles were: 5.79–5.79–21.88–
5.79 (Figure 37).
Simulations of the motion of a vehicle with the
pressure in the left front wheel reduced to 1.5 Right rear wheel – 0.0 at
showed that for default characteristics the criteria are It was much easier to simulate the motion of a
not met (Table 7). Satisfactory results were obtained vehicle with pressure eliminated from the right rear
after the slip angles of all wheels were reduced to wheel. The criteria were met in a simulation when
3.5°. more rigid characteristics of wheels with nominal
pressure, and reduced rigidity of the wheel with no

148
 Bypassing manoeuvre driving a car with reduced and no tire pressure

pressure, and a slip angle in the range of 10° were

adopted (Table 11).
All wheels – 0.0 at
None of the simulations with pressure eliminated
from all the wheels met the adopted criteria (Figures
48 and 49). Large deviation of the vehicle’s CG
motion trajectory in simulations and in the
experiment allows us to say that a simulation of the
motion of a vehicle with no pressure in the wheels,
using the tyre linear model in PC-Crash, is not
reliable.
CONCLUSION
1. During a bypassing manoeuvre on a dry concrete
road, pressure reduction in one wheel to the value
of 1.0 at does not affect vehicle stability and
steerability.
2. Stability and steerability are strongly affected by
pressure reduction below 1.0 at and a lack of
pressure in the rear wheels.
3. The simulations on a curvilinear trajectory of a
vehicle with nominal, reduced and completely
eliminated wheel pressure, using default linear
characteristics of tyre, did not meet the adopted
criteria.
4. In the simulations of a vehicle with nominal
wheel pressure the most favourable results were
obtained for modified linear characteristics in the
slip angle to a range of 3° to 7°.
5. In the simulations of a vehicle with reduced and
completely eliminated pressure the criteria were
met when the slip angle of the unserviceable
wheel was in the range
of 10°–32°.
6. In the simulations that met the criteria the vehicle
motion dynamic parameters such as lateral
acceleration and yaw velocity were convergent
with the results obtained experimentally to an
extent sufficient for road accident reconstruction.
7. It is not possible to simulate bypassing of an
obstacle by a vehicle with no pressure in all the
wheels, using linear tyre model in PC-Crash.
8. In the simulations in which the vehicle velocity
was about 60 km/h, skidding occurred, did it
failed to meet the criteria.
[4] Robinette R., Deering D., Fay R. J., Drag and Steering
Effects of Under Inflated and Deflated Tires, SAE Tech-
nical Paper 970954, 1997.
[5] Bixel R. A., Developments in Vehicle Center of Gravity
and Inertial Parameter Estimation and Measurement, SAE
Technical Paper 950356, 1995.
[5] BN-90/3615- /06, Project of standard: "Motor Vehicle.
Stability and Steerability. Lane Change Maneuver and Go-
ing a Car into a Bend".
[6] Andreatta D., Bixel R. A., Guenther D. A., An Analysis
of the Time to Deflate a Ruptured Tire, Accident Recon-
struction Journal, Vol 17, No.6, November/December 2007.
[7] Brach R. M., Brach M. R., Tire Models for
Vehicle Dynamic Simulation and Accident Reconstruction,
SAE Technical Paper 2009-01-0102, 2009.
[8] www.racelogic.co.uk., 2013.
[9] Regulation No. 13 (UNECE), Uniform provisions
concerning the approval of vehicles of categories M, N and
O with regard to braking.
[10] Lozia Z., Simulation Tests of Biaxial Vechicle
Motion after a Tire Blow-out, SAE Technical Paper 2005-
01-0410, 2005.
[11] Wach W., Simulation of Vehicle Accidents using PC-
Crash, Institute of Forensic Research Publishers, Krakow
2011.
Contact
Jakub Zebala, PhD
Westerplatte 9
31-33, Krakow, Poland
e-mail: jzebala@ies.krakow.pl,
tel: +48 12 6187523
Wojciech Wach, PhD
Westerplatte 9
31-33, Krakow, Poland
e-mail: wwach@ies.krakow.pl,
tel: +48 12 6187523

References
[1] Blythe W., Day T. D., Grimes W. D.,
3-Dimentional Simulation of Vehicle Response to Tire
Blow-outs, SAE Technical Paper 980221, 1998.

[2] Robinette R. D., Fay R. J., Drag and Steering

Effects From Disablements of Run Flat Tires, SAE Tech-
nical Paper 2000-01-1316, 2000.

149
 Jakub Zębala, Wojciech Wach, Piotr Ciępka, Robert Janczur

Appendix

All wheels – 2.0 at (nominal value) (Tests No.

1,2,3)

Figure 6. Displacement of vehicle CG in experiment and

motion simulation

Table 6. Analysis of adopted criteria

Criterion 1 Q (Criterion 2)
Linear 3-3-3-3 Yes 0.21
Linear 3.5-3.5-3.5-3.5 Yes 0.26
Figure 4. Displacement of vehicle CG in experiment and Linear 4-4-4-4 Yes 0.37
motion simulation Linear 10-10-10-10 No (post 6) 0.27
Experiment - -
Table 4. Analysis of adopted criteria
Criterion 1 Q (Criterion 2)
Linear 6-6-6-6 Yes 0.26
Linear 7-7-7-7 Yes 0.19
Linear 10-10-10-10 No (post 6) 0.18
Experiment - -
Figure 7. Measured (blue silhouettes) and simulated (red
silhouettes) movement trajectories of the vehicle (Linear
3.8–3.8–3.8–3.8)

Figure 5. Displacement of vehicle CG in experiment and Figure 8. Measured and approximated steering wheel an-
motion simulation gle
Table 5. Analysis of adopted criteria
Criterion 1 Q (Criterion 2)
Linear 3-3-3-3 Yes (post 6) 0.36 cut
Linear 3.8-3.8-3.8-3.8 Yes (post 6) 0.15
Linear 5-5-5-5 Yes (post 6) 0.50
Linear 10-10-10-10 No (post 6, 7) 0.34 cut
Experiment Post 6 -

Figure 9. Time history of velocity

150
 Bypassing manoeuvre driving a car with reduced and no tire pressure

Figure 10. Time history of lateral acceleration

Figure 14. Measured and approximated steering wheel

angle

Figure 11. Time history of yaw velocity

Left front wheel – 1.5 at (Test No. 5)

Figure 15. Time history of velocity

Figure 12. Displacement of vehicle CG in experiment and Figure 16. Time history of lateral acceleration
motion simulation

Table 7. Analysis of adopted criteria

Criterion 1 Q (Criterion 2)
Linear 3.5-3.5-3.5-3.5 Yes 0.08
Linear 5-5-5-5 Yes 0.72
Linear 10-10-10-10 No (post 6) 0.81
Experiment - -

Figure 17. Time history of yaw velocity

Figure 13. Measured (blue silhouettes) and simulated

(red silhouettes) movement trajectories of the vehicle
(Linear 3.5–3.5–3.5–3.5)

151
 Jakub Zębala, Wojciech Wach, Piotr Ciępka, Robert Janczur

Left front wheel – 1.0 at (Test No. 6)

Figure 22. Time history of yaw velocity

Figure 18. Displacement of vehicle CG in experiment and

motion simulation
Figure 23. Time history of lateral acceleration
Table 8. Analysis of adopted criteria
Criterion 1 Q (Criterion 2)
Linear 20-3.8-3.8-3.8 No (post 6) 0.32
Linear 10-3.5-3.5-3.5 Yes 0.59 Left front wheel – 0.0 at (Test No. 7)
Linear 10-3.8-3.8-3.8 Yes 0.60
Linear 3.8-3.8-3.8-3.8 Yes 0.49
Linear 6-6-6-6 Yes 0.99 cut
Linear 10-10-10-10 Yes 0.72 cut
Experiment - -

Figure 19. Measured (blue silhouettes) and simulated

(red silhouettes) movement trajectories of the vehicle
(Linear 10–3.8–3.8–3.8)

Figure 24. Displacement of vehicle CG in experiment and

motion simulation

Table 9. Analysis of adopted criteria

Criterion 1 Q (Criterion 2)
Linear 30-3-3-3 No (Post 6) 0.95
Linear 19-3.8-3.8-3.8 No (Post 6) 1.70
Linear 11-3-3-3 Yes (Post 6) 2.30
Linear 3.8-3.8-3.8-3.8 No (Post 6,7) 0.66 cut
Linear 10-10-10-10 No (Post 6,7) 1.11 cut
Experiment Post 6 -

Figure 20. Measured and approximated steering wheel

angle

Figure 25. Measured (blue silhouettes) and simulated

(red silhouettes) movement trajectories of the vehicle
(Linear 11–3–3–3)

Figure 26. Measured (blue silhouettes) and simulated

(red silhouettes) movement trajectories of the vehicle
(Linear 30–3–3–3)
Figure 21. Time history of velocity
152
 Bypassing manoeuvre driving a car with reduced and no tire pressure

Right front wheel – 0.0 at (Test No. 10)

Figure 27. Measured and approximated steering wheel

angle
Figure 31. Displacement of vehicle CG in experiment and
motion simulation

Table 10. Analysis of adopted criteria

Criterion 1 Q (Criterion 2)
Linear 4.5-32-4.5-4.5 Yes 0.33
Linear 3,8-20-3.8-3.8 No (post 7) 1.14
Linear 3.8-10 3.8-3.8 No (post 3, 5) 2.09
Linear 3.8-3.8-3.8-3.8 No (post 3, 5) 2.44
Linear 10-10-10-10 No (post 5) 3.17
Experiment - -

Figure 28. Time history of velocity

Figure 32. Measured (blue silhouettes) and simulated

(red silhouettes) movement trajectories of the vehicle
(Linear 4.5–32–4.5–4.5)

Figure 29. Time history of lateral acceleration

Figure 30. Time history of yaw velocity

Figure 33. Measured and approximated steering wheel
angle

Figure 34. Time history of velocity

153
 Jakub Zębala, Wojciech Wach, Piotr Ciępka, Robert Janczur

Figure 35. Time history of lateral acceleration Figure 40. Time history of yaw velocity

Figure 36. Time history of yaw velocity Figure 41. Time history of lateral acceleration

Left rear wheel – 0.0 at (Test No. 15) Right rear wheel – 0.0 at (Test No. 19)

Figure 37. Measured (blue silhouettes) and simulated

(red silhouettes) movement trajectories of the vehicle
(5.79–5.79–21.88–5.79)

Figure 42. Displacement of vehicle CG in experiment and

motion simulation

Table 11. Analysis of adopted criteria

Criterion 1 Q (Criterion 2)
Figure 38. Measured and approximated steering wheel Linear 6-6-6-9.5 Yes (Post 3,6) 0.21
angle Linear 6-6-6-9 Yes (Post 3,6) 0.30
Linear 6-6-6-6 No (post 3,6,8) 0.39
Linear 10-10-10-10 No (post 6,8) 1.12
Experiment Post 3,6 -

Figure 43. Measured (blue silhouettes) and simulated

(red silhouettes) movement trajectories of the vehicle
(Linear 6–6–6–9.5)

Figure 39. Time history of velocity

154
 Bypassing manoeuvre driving a car with reduced and no tire pressure

All wheels – 0.0 at (Test No. 20)

Figure 44. Measured and approximated steering wheel

angle
Figure 48. Displacement of vehicle CG in experiment and
motion simulation

Figure 49. Measured (blue silhouettes) and simulated

(red silhouettes) movement trajectories of the vehicle
(Linear 10–10–10–10)
Figure 45. Time history of velocity

Figure 46. Time history of yaw velocity

Figure 50. Measured and approximated steering wheel

angle

Figure 47. Time history of lateral acceleration

Figure 51. Time history of velocity

155
 Jakub Zębala, Wojciech Wach, Piotr Ciępka, Robert Janczur

156
© 2013 by EVU
EVU 2013 - 18

TACOT Project, toward a Trusted GNSS function

Steeve Favere FDC
Ralf-Roland Schmidt-Cotta
Continental

Abstract

The need for robust and trustable GNSS information is constantly growing since many years and far beyond traditional
safety of life domains. Today, if few technical solutions can actually cover this need, they are too complex or too costly
for most applications.
TACOT is offering an innovative technical alternative which is able to provide a good confidence and robusteness in
Position, Velocity and Time (PVT) at low cost. The proposed solution (Trusted PVT function) relies mainly on the
merge and the consistence of several sources of information to provide a high confidence level. This new concept
notably relies on Bayesian technic and is particularly difficult to fool.
This solution also takes advantage of GNSS multi-constellations and EGNOS to increase the quality of its outputs. As
such, it also prepares markets for Galileo introduction.
The approach of TACOT is to demonstate the assets of this innovation by improving Digital Tachographs (then called
Augmented Digital Tachograph or ADT) in the frame of anticipating an envisaged amendement of the EC Regulation
3821/85. The expected improvement is thus first to provide a second motion information and second a GNSS position
for the start/end of the travels in a trusted and secured manner to the ADT. TACOT is then able to support the use of
trusted PVT data by any third party ITS application, through the definition and implementation of an external API in the
ADT, and making these data available on the CAN bus.
To fully demonstrate the TACOT’s achievements, the prototypes developed in the project will transmit their data to a
commercial Fleet Management System (FMS). This demonstration will notably highlight the GNSS improvements
(accuracy, availability…), the plus in confidence with regard to potential threats and weaknesses of GNSS, and the data
used by the ADT and the FMS.

 GNSS signals are vulnerable to interferences

GNSS assets and drawbacks and also may be subject to
malicious attacks such as
GNSS systems are today widely used in a multitude jamming, meaconing or
of applications everywhere in the World. spoofing.
This is particularly true within the road transport
domain in which GNSS already enables applications
such as car navigation or fleet management but also
provides the ground for developing advanced appli-
cations in the ITS domain.

The success of GNSS is due to the high performance
of their services which allow any user to get an accu-
rate position, velocity and time (PVT) more or less
everywhere in the World at low cost.
Nevertheless these systems do have some draw-
backs:
 GNSS signals are either not available in “in-
doors” environments (buildings, tunnels…)
or only partially available in masked
environments (urban areas, mountains…)
Figure 1 Low cost GNSS jammer
The GNSS signal vulnerability is met in everyday’s
life. For example jamming has been experienced in
US and Taiwan airports, while South Korea reported
a loss of GNSS signals near its border with North
Korea.
A proof of concept of Spoofing was
also performed by a US university
which could land a drone by fooling
157
 Steeve Favere
its GPS receiver.
From users point of view these issues may lead to
either:
 A lack of availability of the GNSS service,
 A GNSS-like misleading information
 And/or degradation of the performances
These issues hinder or slow down the adoption of
GNSS in some applications which requires a high
availability of the positioning / velocity or time
(PVT) services, even in constringent environments
(such as urban areas) or which requires a good level
of confidence in PVT information.
Project Context and Objectives
TACOT is a project co-funded by the European
Commission in the frame of FP7 (Galileo area) and
managed by the European GNSS Agency (GSA).
TACOT duration is 2 years and will end in January
2014.
The main objectives of TACOT are twofold:
 It intends to demonstrate the possibility to
support the penetration of European GNSS
(EGNOS and Galileo) in road transport
through the vector of digital tachographs.
 It anticipates the amendment of the EC-
Regulation 3821/85 on recording equipment
in road transport: recording of location data
& enhanced security.
To reach these objectives, TACOT develops a low
cost Trusted GNSS function able to detect major
GNSS related threats.
The approach retained in the project is the integra-
tion of this function in the Digital Tachographs (DT);
but it can fit to any ITS applications.
TACOT technical framework
TACOT is producing a Trusted PVT function
prototype that will be included in a Digital
Tachograph forming an Augmented Digital
Tachograph (ADT) prototype.
This innovative function as well as the prototypes
will be entirely developed in the project.
The ADT is developed from an existing Digital
Tachograph enhanced with new interfaces to enable
158
the full integration of the Trusted PVT function and
its new features.
The ADT is also designed to transmit the data
provided by the Trusted PVT function on CAN bus
to offer the possibility, for third parties ITS
applications, to take advantage of developed new
functions and data.
To demonstrate this last feature, TACOT also
upgrades an existing commercial Fleet Management
System to benefit from Trusted PVT function output.
The following scheme illustrates the main develop-
ments / achievements of the project:
Fugure 2 TACOT’s technical framework
Trusted PVT function: Overview
The solution proposed and developed in TACOT, is
a function which takes advantage of the redundancy
and the independency of several sources of infor-
mation which are either internal or external sensors,
such as:
 GNSS receivers (GPS, Galileo, Glonass,
EGNOS)
 Integrated MEMs1 sensors
 Vehicle on-board sensors (odometer
connected to the digital tachograph, secure
clock)
 Other sensor (time source)
The sensors information gathered by the PVT
function is combined to ensure a smooth availability
of the service provided, even when the GNSS
services are unavailable for a period of time.
Thus the PVT function provides:
 A measure of position, velocity and UTC
time (PVT)
 The accuracies associated with these
measures
 The level of confidence, which is an
evaluation of the risk of false positioning
The confidence level is elaborated through the moni-
toring of metrics and an evaluation of the statistical
behaviour of the inputs using Bayesian techniques.
The PVT function will use secure certified
technologies and an innovative sensor fusion
approach based on Bayesian techniques to detect
1
Microelectromechanical systems
 TACOT Project

GNSS attacks. The Bayesian techniques are already

successfully applied in several fields such as fraud Services: reliable location data and detection of PVT
detection for banking electronic transaction, threat manipulation
identification and detection in aerial defence.
* Jamming mitigation (sensor fusion)

Their main assets are to reduce the sensors fusion * Meaconing & spoofing detection (sensors outputs
complexity, to match with security constraints and to consistency)
be cost effective regarding other techniques.
* Exclusion of unhealthy GNSS satellites
The PVT function solution concept is illustrated in
the following figure:

Security policy

* PVT function has its own Public Key Infrastructure

(PKI)

* Secured transmission of data

Trusted PVT function: data security

From a security point of view, the trusted PVT

function will provide independent security features.
It will manage its own keys – issued from EU (DT)
or other key hierarchy – and will deliver signed PVT
data to the DT.

The trusted PVT function is planned to reach

Evaluation Assurance Level up to EAL4+ in its
commercial version.

For this purpose, a Protection Profile (PP) document

is developed in parallel of the prototypes to propose
a basis to enable future equipment to be accredited
by a national certification authority for regulated
applications.
This PP defines the generic scope of essential
Figure 3 Trusted PVT function
security functions to be covered by the Target of
Evaluation (i.e. the PVT function) necessary to
enable liability critical / legally binding applications.
Trusted PVT function: outputs/services
TACOT’s ADT prototype
The following tables summarise data and services of
the PVT function: TACOT project develops an Augmented Digital
Tachograph (ADT) which takes advantage of the
Outputs: PVT related data benefits enabled by the Trusted PVT function.
* Position (LLH coordinates)
This ADT is an evolution of an existing Digital
* Velocity (NED coordinates)
Tachograph which integrates the Trusted PVT
* UTC Time function.
* Heading
It uses the Trusted PVT function output to improve
* Quality indicators (Circular Error Probability)
the qualiy of service of the DT within its mission
* PVT confidence level (percentage of reliability) with regard to the regulation.

159
 Steeve Favere

The following diagram illustrates the main

architectural entities of the ADT and surrounding
elements.

Figure 5 Overview of the External API Services

There will be two data services available:

 PVT Data And Confidence Level:

Figure 4 The Augmented Digital Tachograph
The ADT prototype will benefit from the trusted
PVT function to respond to the Regulation
applicable to Digital Tachograph notably for:
 A second source of motion
 Position recording on specific events (travel
start/end, …)
Furthermore, it is expected that the trusted PVT
function will improve of the operational use of the
Digital Tachograph through:
 The automatic positioning (in the
beginning/end of a working day)
 The simplification of the maintenance of
Digital Tachograph (thanks to GNSS time
adjustment)
 An evaluation of the risk of false positioning
This service broadcasts periodically the set
of PVT-data and confidence level to the
vehicle bus. All other CAN applications
connected to the same bus can use this
service.
 PVT Signature Segment:
The ADT creates a digital signature for
selected sets of PVT data. Due to the
relatively high amount of data the
signature is divided into a particular
number of segments. This service
broadcasts the single segments of the
PVT signature at the vehicle bus.
TACOT’s benefits
The TACOT project will demonstrate the assets of
the trusted PVT function when embedded in the
ADT.

Nevertheless, as suggested previously when

The ADT also possesses an external interface (API)
describing the external API, the benefit of the trusted
to provide the trusted PVT function data to other ITS
PVT function can be extended to many other ITS
applications via a CAN bus. In the frame of TACOT,
applications which would take advantage from a
the possibility for a third party ITS application to
level of confidence on data provided.
gather and use Trusted PVT outputs will be
demonstrated through a commercial Fleet
TACOT will consequently also show how such a
Management System.
function can benefit to any other ITS applications by
its implementation and real time demonstration in a
The external API acts as "data-source" only. No
Fleet Management System.
input services via the external interface are specified.
TACOT’s has identified three categories of
The following picture shows an overview of the
applications which would benefit directly from the
different services of the external interface:
Trusted PVT services:

160
 TACOT Project

Note: This list is not exhaustive and the expected

Secured business
benefits for each are to be further consolidated
Trusted PVT information helps to mitigate the risk of
economic loss. This can be either direct loss (correct charging
in PAYD models) or indirect (time loss, degradation of service Accident reconstruction needs
quality which lead to benefits loss…)
Vehicle accident reconstruction aims at making
Improved safety business roads safer by investigating vehicle collision events.

The Trusted PVT function is not designed to be used as a life To that end it is often needed to conduct an in-depth
saving device. Nevertheless, trusted information can actually collision analysis to
benefit to secure operations or activities which can be harmful identify the collision
to the society or to the environment like FMS applications for
hazardous goods transports. causation and
contributing factors,
including the role of
Law enforcement
the driver(s),
vehicle(s), roadway
Trusted PVT information helps to mitigate the risk of frauds and the
with regards to law rules. As regulations are felt as constraints
to the business, the adoption of the PVT function by users environment…
enforces the compliance with these regulations.
The access to highly reliable vehicles’ position,
velocity and time during the last few seconds before
Applications’ categories for Trusted PVT the collision constitutes one important element to
understand the causes.
The following table indicates examples of
applications where the trusted PVT function benefits The Trusted PVT function developed in TACOT,
are relevant: offers a unique added value to these information for
the following reasons:
Secured Safety
ITS Applications Law Enf.
bus. Imp.
Speed enforcement
First, it enhances significantly the quality of
traditional GNSS devices as it combines the
Electronic fee collection (road following assets:
pricing)
Freight tolling  Takes advantage of GNSS multi-
constellations (GPS, Glonass, Galileo)
Street parking pricing
 Uses the European SBAS2 system (EGNOS)
Regulated Fleets in Urban Areas which provides information to minimize
GNSS main error sources
Pay as you pollute
 Integrates in-vehicle motion and time
Eco-driving systems sources
Pay-per-use / Pay as you drive
insurance This will lead to improved accuracy and a better
Vehicle Tracking availability of the data, in particular in “masked”
Fleet management environment such as urban areas or in mountains.
Anti-Theft Systems Secondly, the PVT function provides a UTC time,
Car/Ride Sharing Applications quality indicators and a confidence level:
Traffic Travel Information
 The access to UTC time is important in
Europe as it is the recognised time in the
E-call / Emergency Services legal domain; the UTC time can easily be
Accident reconstruction, Event converted into local time
Data Recording  The quality indicators can be useful to
interpret the values of PVT data
Applications’ exemples for Trusted PVT

2
Satellite Based Augmentation System
161
 Steeve Favere

 The confidence level indicates the

consistence of the different sources of data
(and the probability to face hostile
manipulations)

Furthermore, the Trusted PVT function and data

transmission are secured and are the objects of a
protection profile document elaborated in TACOT
which objective is to propose a base for future
equipment needed to be homologated by a national
security agency for regulated application.

Thus, TACOT achievements offer an interesting

ground for Accident Reconstruction improvements
from both operational and legal point of views.

Contacts
Steeve Favre
10 Cours Louis Lumière
94300, Vincennes, France
e-mails: steeve.favre@fdc.eu
tel: +33 1 53 66 11 11

Ralf-Roland Schmidt-Cotta
Heinrich-Hertz-Strasse 45
D-78052, Villingen-Schwenningen, Germany
e-mails: Ralf-Roland.Schmidt-Cotta@continental-
corporation.com
tel: +49 7721 67 2570

TACOT’s Consortium

162
© 2013 by EVU
EVU 2013 - 19

Reconstruction of an accident involving a Hummer

driving at high speed and a Nissan turning in based on a
crash test using identical vehicles
Manfred Becke
Joost Wolbers

Abstract
We present an extraordinary accident reconstruction for which two experts have previously offered different conclusions
beforehand. A Hummer H2 crashes, at high speed, into the side of a Nissan Primera turning into its road.
There were large differences in the evaluation of the EES values due to the different structural stiffness of the vehicles
and the extreme deformation of the Nissan, which was hit in the side.
A crash test using vehicles identical to the ones involved in the accident was carried out at considerable financial
expense, not only to determine the total deformation energy but also to obtain additional information about the exact
collision location. The insight gained through the test led to distance-to-time correlations that differ decisively from the
conclusions of the two previous experts.

Assignment Final situation

An insurance company commissioned us to analyse a In his expertise, expert N. evaluated the marks left
traffic accident for which two other experts related to on the road (fig. 1).
two different parties had already presented their
expertises. The conclusions of the two experts
differed from each other decisively. Therefore, a
crash test using vehicles identically built to the ones
involved was to be carried out.
The accident occurred in a built-up area at daylight
in summer. A Hummer H2 was driving on a priority Figure 1: Marks left on the road
road when, from a secondary road to its right, a
Nissan Primera turned into the main road and drove We have to examine one scratch mark in particular
towards the Hummer. This resulted in a collision that both experts assigned to the right front wheel
causing substantial material damage. The driver of suspension in the collision position. Fig. 2 illustrates
the Nissan suffered lethal injuries, the front the final position of the vehicles.
passenger suffered severe injuries.
In the following, we will present a short survey of
the main differences between the two prior
expertises. While expert N. sets the Hummer’s speed
at the moment of collision at approx. 85 to 95 km/h,
expert R. sets it at about 70 km/h. Both experts
estimate the Nissan’s collision speed at 10 to 15
km/h. Also, both experts reconstructed the same
collision location which results in a start-up distance
for the Nissan of 5 m. This leads to an approach
speed for the Hummer equalling the collision speed
since there remained no possibility of a defensive Figure 2: Final situation after the accident
action.

163
 Manfred Becke, Joost Wolbers

Crash test preparation

Naturally, in preparation for a crash test the
velocities have to be determined. Since the test
vehicles were very expensive it was impossible to
approach the correct result by repeating the test.
The crucial point was to determine the collision
speed of the Hummer precisely so that the severity of
the damage to both vehicles would be comparable to
that of the original accident. In order to achieve this,
the test was prepared by means of a preceding
accident analysis using both existing test results and
PC Crash simulations to investigate the deformation
energy to be expected.
The particulars of the preparatory accident analysis
Figure 3:
will not be presented here. In order for the Hummer Comparison of the crash result (Hummer)
to hit the Nissan in a manner comparable to the
original accident, the Nissan had to be left standing
still at the moment of collision. Since the Nissan in
the real accident was moving and was, hence,
contributing kinetic energy this circumstance had to
be balanced by choosing a higher collision speed for
the Hummer in the test in order to achieve the
suitable total deformation energy. The preparatory
collision analysis assumed that the ratio between the
deformation depths of the vehicles was 1:10, which
leads to a deformation energy of the vehicles of the
same ratio.

Experimental setup and damage to the vehicles

In the crash test, the Hummer drove into the standing

Nissan Primera at a collision speed of 83.5 km/h. It
must be pointed out, that this was not the Hummer’s
collision speed in the real accident, in which the
Nissan was moving and, thus, contributing energy as
well.
Since, for technical reasons, the Nissan was standing,
deceased post-collision movements were to be
expected. The films showing the motion sequence
confirm this.
Fig. 3 and 4 illustrate the differences in damage the
vehicles obtained in the original accident and in the
test, respectively.
The main difference in the damage of the Hummer is
that the identically built rim did not break. We found
that the deformation depth of the Hummer was
considerably less than that of the Nissan, which
correlates with the fact that the Hummer absorbed a
much lower deformation energy than the Nissan.
Figure 4: Comparison of the crash result (Nissan)
In a comparison between the damages of the original
Nissan and the Nissan used in the crash test there
seems, at first sight, to be a good correspondence. A
closer look at the last pair of photos, however,
reveals that the intrusion was deeper in the test than
in the original accident.
Marks on the road in the crash test, start-up
distance of the Nissan
In their prior expertises, the experts N. and R.
assigned noticeable scratch marks on the road (view
fig. 1 to the left) to the right front wheel or, rather, its
suspension. This led them to conclude the collision
position as presented in their expertises.
There were doubts about this assumption even before
the crash test because the Nissan’s right front wheel
was not de-aerated. At best, the scratch marks could

164
 Rekonstruction of one exceptional accident

have been assigned to the wheel suspension or to the

engine/transmission unit.
The crash test caused scratch marks that are
comparable to the original ones (view fig. 5).

Figure 6: Underbody of the Nissan

Figure 5: Scratch marks

By marking the position of the Nissan at the moment

of collision with chalk we were able to discover the
origin of these marks. After lifting the Nissan, we
found that the lower door sill was folded and bent
downwards during the collision.
This sharp bend formed a few centimeters to the left
of the door handle on the driver’s side.
We were able to ascertain that the experts had not
only partly disagreed about the deformation energies
absorbed by the vehicles but that they had also
incorrectly assigned the noticeable scratch marks to
the right front wheel of the Nissan whereas these
marks were really caused by the door sill on the
driver’s side being bent downwards in the area of the
b-pillar. This fact also leads to a different collision
position on the site of the accident (view fig. 7).
The start-up distance of the Nissan increases
considerably to 8.3 m as opposed to 5 m as assumed
in the earlier expertises.
Figure 7: Start-up distance /collision position
Total deformation energy in the crash test
The velocities of the crash test cannot be transferred
directly to the case as the Nissan in the test was
standing. Accordingly, we first calculated the total
deformation energy absorbed by the two vehicles
using PC Crash simulations to emulate the test
results. The changes in speed due to the collision as
given by CTS refer to the positions of the crash data
recorders used. Especially by varying the k-factor,
the differences between the test results and the
calculations could be diminished. Increasing the k-
factor results in a more partly-elastic collision and in
greater changes of velocities caused by the collision
while, simultaneously, the total deformation energy
decreases. Tab. 1 presents a survey of four
calculations executed using various k-factors.
Calculation no. 3 achieved the best approximation
with a total deformation energy of 242,400 Nm.
165
 Manfred Becke, Joost Wolbers
Based on the other calculations a bandwith of ±
2,000 Nm was taken into account, which lead to the
next interim result, which is that the total
deformation energy absorbed by both vehicles in the
test was between 240,000 and 244,000 Nm.
Tab. 1: Analysis of the total deformation energy
Analysis of the collision velocities in the
accident
To determine the collision velocities we carried out
various PC Crash simulations. The intention was to
emulate as exactly as possible the post-collision
movements while taking into account the total
deformation energy of 240,000 Nm measured during
the crash test. Using the total deformation energy of
222,000 Nm for the simulation resulted in the best
solution. In this scenario the Hummer’s collision
speed was 68 km/h and the Nissan’s was 18 km/h
(view fig. 8).
166
Figure 8: PC Crash simulation
Changing the collision speed of the calculations to
v(k) = 65-70 km/h for the Hummer and v(k) = 15-20
km/h for the Nissan leads to total deformation
energies between 193,000 Nm and 240,000 Nm.
Following this, a bandwidth of collision velocities
between 65 and 70 km/h for the Hummer and
between 15 and 20 km/h for the Nissan were taken
into account. Of course, the lowest and the highest
velocities were not to be used simultaneously.
Deformation energies applied by the prior
experts
The differing collision velocities stated in the earlier
expertises are surely attributed to the total
deformation energies applied. Expert R. expresses all
the deformation energies using EES values.
Converting these, one obtains a total bandwidth of
approximately 166,000 to 220,000 Nm. The upper
limit is in accordance with our optimal simulation.
Consequently, it is not surprising that expert R.
determines a collision speed for the Hummer of
approx. 70 km per hour.
In the other earlier expertise, expert N. describes the
total deformation energies applying EES values and
certain masses for the vehicles. This approach results
in a total deformation energy of approximately
255,000 to 385,000 Nm. The upper limit derives
from the highly overestimated deformation energy
expert N. assigns to the Hummer (view tab. 2).
.
 Rekonstruction of one exceptional accident
Tab. 2: Total deformation energy
Estimating the deformation energies of the respective
vehicles the expert made the severe mistake of
choosing EES values for the Hummer which were
much too high. Due to the strongly differing
deformation depths in conjuction with Newton’s law
“actio = reactio” one can deduce that the absorbed
deformation energies at similar deformation force
progression is relative to the deformation depth ratio.
Expert N. stated EES values of approx. 45 to 52
km/h for the Nissan and approx. 35 to 42 km/h for
the Hummer. Expressing this using deformation
energies, we obtain values of 117,188 to 166,914
Nm for the Nissan and 137,529 to 217,778 Nm for
the Hummer. The ratio of the deformation energies
of the Hummer and the Nissan is 1:0.8.
Measuring the deformation depths of the vehicles
used in the crash test resulted in a ratio between the
Hummer and the Nissan of 1:6 up to 1:11 depending
on the exact location of deformation.
Distance-time analysis
There is no way of determining the driving process
of the Nissan before it reaches the point of collision
using technical means. There are three possible
choices for the driver of the Nissan:
1) driving through without stopping,
2) starting from a position in which the front of the
vehicle just reaches the road side markings,
3) starting after having gradually approached a
position 1.1 m beyond the road side markings from
which he has an unobstructed view to the left.
All these three scenarios were checked using
distance-time diagrams.
The version most in favour of the Hummer driver
results from the Nissan driving through without
stopping, a low collision speed of 65 km/h as well as
a suitable low braking deceleration of 5 m/s². This
variant is illustrated in fig. 9.
A spontaneous reaction to a signal position of the
Nissan at the moment of its crossing the road side
markings would result in an approach speed between
74 and 83 km/h. Based on an approach speed of 74
km/h of the Hummer, the accident could have been
avoided, not spacially but temporally, if the Hummer
had kept within the speed limit of 50 km/h and the
Nissan had continued without stopping. The Nissan
would, in that case, have proceeded approx. 6.5 m
further at the time when the Hummer would have
reached the collision point.
If we assume an approach speed of 83 km/h the
Hummer would have stopped 10.5 m before the
collision point.
In another scenario in favour of the Nissan driver
(fig. 11) we assume that he proceeded gradually until
his view to the left was unobstructed and he only
started up then. (In the previous scenario he could be
blamed that he would have been obliged to progress
carefully until his view was unobstructed.) This
“new” scenario reduces the start-up distance to
approximately 7.2 m. If we, again, assume a signal
position at a distance of 1 m away from the start-up
position and a collision speed of 70 km/h for the
Hummer we can calculate an approach speed of the
Hummer between 84 and 98 km/h. If the Hummer
had kept within the speed limit of 50 km/h he could
have avoided the accident in terms of space.
From the Nissan driver’s point of view at the
moment of starting up, the Hummer would have
been about 60 to 90 m away at a speed of 84 to 98
km per hour. Driving constantly at the maximum
permissible speed of 50 km/h the Hummer would
(without braking) have reached the collision point
between 1.7 and 3.1 seconds later. Assuming that the
Nissan would have accelerated continuously the
Nissan would, in that case, have advanced between
approx. 12 and 18 m further. This illustrates that,
under these specific conditions, the Hummer could
easily have prevented the accident. The Nissan
driver, on the other hand, could only have avoided
the accident if he was expected not to trust in the
Hummer driver’s abiding by the speed limit.
Additionally, if we can assume that the Nissan driver
advanced onto the priority road gradually this
behaviour could have prompted the Hummer driver
167
 Manfred Becke, Joost Wolbers

to reduce his speed preventively even before however, the deducible approach speed would be
initiating an emergency stop. If this was the case, even higher.

Figure 9: Distance-time diagram

Figure 10: Distance-time diagram

168
 Rekonstruction of one exceptional accident

Figure 11: Distance-time diagram

Contact

Dr. Manfred Becke

becke@ureko.de

Dipl.-Ing. Joost Wolbers

wolbers@ureko.de

Ingenieursbüro Schimmelpfennig + Becke

Münsterstrasse 101
D-48155 Münster
Deutschland

169
 Manfred Becke, Joost Wolbers

170
© 2013 by EVU
EVU 2013 - 20

Viewing analysis of experienced vs. not experienced

motorcylists and conclusions about traffic-safety
Univ.Prof. DI Dr. Ernst PFLEGER

Abstract
In the Austrian lands of Carinthia and Lower Austria, extensive viewing analyses with motorcycle drivers were executed
at accident sites and risk areas on selected routes.The viewing analyses revealed incorrect viewing strategies and faulty
priority order as main causes for accidents. Many accidents are caused by wrong viewing behaviour, missing viewing
points, misjudgements and complexities (uncoordinated viewing strategies caused by excessive information) that are too
high, especially at high velocities. In the presentation the proof is shown with selected movie sequences.

Use of eye examinations to improve the road

safety of motorcyclists
Using the TÜV-certified viewpointsystem® viewing
analyses, the objective evaluation of real viewing
behaviour, the danger recognition and reaction
behaviour of motorcyclists were investigated for the
first time.

Viewing analyses allow the complete observation of

detail recognition and the perception of
surroundings. It became apparent that the choice of
driving line is significantly influenced by the
viewing behaviour. In bends in particular incorrect
posture of the head results in perspective distortions Figure 1: No navigation to the guidance marks in shadow-
areas - longer fixations of the drain, no measuring of the
and misjudgements. Especially at high driving speed
curve [screenshot of a movie]
the angular velocities, and thus the error magnitude,
increase. Over the course of the analyses, the 3
following results can be derived for the accident and
danger sites: 2. Accident cause: navigation gaps (naviga-
tion lapses) in dangerous bends

Another important result of the study is the finding

1. Accident cause: incorrect priority orders
(accident-causal reversal of order in percep-
tion)
This means that there are constant causal
relationships between distinctive visual features or
movements (see also extensive viewing studies and
works on viewing research by the author).
of so-called navigation lapses on the taken routes,
i.e. points at which too few points of guidance are
given and a ‘loss of orientation’ becomes evident.
Ordered viewing strategies are not possible at these
points, and incorrect reactions of the motorcyclists
are a given.

Bright areas, movements, conspicuous objects are

always perceived earlier than inconspicuous ones.
This explains the masking of safety-relevant objects.

171
 Univ.Prof. DI Dr. Ernst Pfleger
Figure 2: Missing guidance marks - no perception of curve
– search for navigation points in the curve area (shown by
yellow lines) [screenshot of a movie].
3. Accident cause: Finding of perspective
misguidance
The analyses of the perspective visualisations at
accident hot-spots show potential areas for
misjudgements and allow specification of unsafe
navigations of curves.
Figure 3: Deficient perspective guidance, the following
left turn is not yet recognised by the motorcyclist
[screenshot of a movie].
4. Accident cause: Wrong head position in
narrow curves
Locations with real perspective misguidance are
usually present when misjudgements caused by a
wrong head posture are given from the road layout –
mostly in double curves. For less experienced
drivers, the visual horizontal plane often shows an
angle of up to 20°, in some cases even greater, so
that angular changes can no longer be associated
172
within each time/distance segment and physiological
real errors in estimations must occur.
Figure 4: Wrong head position in a right curve (18°)
[screenshot of a movie].
These four topics, as presented by the author, ex-
plain the singularity of traffic accidents with mo-
torcycles, which was recognised in the viewing
analyses, and require completely new approaches
by the traffic authorities.
Classification of errors based on detailed
viewing analyses:
A. Analysis in dangerous situations - The
following (negative) abnormalities are present:
 Inexact choice of driving line and cutting
of curves – sudden corrections of the driving
line become necessary during navigation of
the curve. Increased danger of overreaction
(sudden braking etc.) when the driver is
startled.
 Errors in navigation behaviour caused by:
o Information defects
o Perspective misjudgements
o Static / dynamic viewing
obstructions
o Distractions e.g. by road signs in
unfavourable positions just before
entry into the bend and overly long
viewing fixations to these elements.
o Inadequate equipment of guidance
elements leads to increased number
of searching processes.
 Unclear interactions with other road users
 Incorrect danger recognition – neglecting
safety relevant control gazes
 Blickanalysen bei routinierten / unroutinierten Motorradlenkern
 Reversal of priorities in the viewing
strategies of drivers. Relevant information is
masked by less important, but more
conspicuous stimuli.
(compare: distractions from forced viewing
attraction by stimuli appearing in the
periphery)
B. Analyses at accident sites - The viewing-related
and safety-related detail analyses of accident sites
(singular accidents as well as accumulations)
show that often the following environmental
conditions were present:
 Missing or incomplete equipment with
vertical guidance elements complicates the
optical guidance considerably.
 Frequent changes between light and shade
necessitate constant adaptation, during
which no ordered perception occurs.
 Reversal of priority order especially in
complex traffic situations (parked vehicles
or entrances to touristic destinations on
overland roads).
 Partially faulty optical guidance or
expansions of the road surface, such as wide,
undefined road surfaces or widening of the
road in curves etc.
 Narrowing curves (three-centre curves,
clothoids, et al.)
 Wrong or unfavourable cross slope
(“hanging curves“)
 Inhomogeneous friction of the surface layer
from maintenance work
 Unfavourable road markings: lack of solid
line / border line in dangerous segments
 The accident circumstance “sudden braking“
often noted in accident records, gives a
potential clue to overreactions of drivers in
hot spots.
C. Analysis in perspective interactions - Using
detailed perspective analyses on selected
examples, the following interrelations could be
verified:
 Gaps in information transfer: These gaps
explain unstructured navigation behaviour
because of missing navigation points. This
causes search processes and distractions from
relevant navigation processes. The result is a
negative influence on the choice of driving line
or delayed recognition of bends.
 Distortion / wrong perspective appearance:
Unfavourable head posture leads to angular
deviations of the eye axis from the horizontal of
up to 16° or more. This complicates the
estimation of curves and leads to information
defects (misestimations).
Result: Safety analyses must increasingly take
into account perspective interrelations.
Summary
Reflection of the results for traffic safety
Equipment for optical guidance, especially at
accident sites, needs to be considered. The current
strategies for accident prevention have to be re-
examined when looking at the re-design of accident
sites.
It is a grave mistake to insufficiently consider the
approaches from the driver’s point of view in the
analysis of accident situations. Viewing analyses in
the relevant approach directions may uncover faults
in the road equipment and optical guidance (see:
accident- and danger sites) in many cases.
It is professionally incorrect to apply mechanic
passive safety elements without viewing analyses
and with no knowledge of perspective interrelations.
Many accidents are explained by incorrect viewing
behaviour, missing viewing points, misestimations,
and too high a level of complexity (uncoordinated
viewing movements) especially at high driving
velocities.
Viewing analyses with motorcyclists show areas for
misestimations and show the unsafe navigation of
curves. They reveal the high relevance of faulty
viewing strategies and a constant wrong order of
priorities. The choice of driving line is significantly
influenced by the viewing behaviour.
Particularly in curves, an incorrect head posture
leads to perspective distortions and misjudgements;
especially at high speed and great radial velocities,
errors are increased. Because of unfavourable
viewing lines, less experienced drivers are not able
to practice avoidance strategies and tend towards
overreactions (overbraking of the front wheel, strong
brakes in curves, etc.) Especially in open land curve
173
 Univ.Prof. DI Dr. Ernst Pfleger

segments this leads to severe or fatal injuries in run- c. Road equipment – guidance
off road accidents. elements
d. Road environment design
In the system human-street-technics it becomes
e. Environment conditions (vertical
apparent that interrelations in human information
perception and the road equipment are present. In curves, three center curves,
technical aspects, the head posture and driving clothoids, et al.) etc.
technique in combination with driving speed are 3. Technics
relevant. Of particular scientific interest is that for a. Driving technique: Body and head
the first time over the course of a practical scientfic posture in curves etc.
study, by using viewing analyses it was possible to b. Time-distance sequences
identify particluar aspects of this perspective.
c. Driving velocities etc.
In this way it was possible to verify both gaps in the
perspectivic association of navigation points and
distortions and incorrect perspectives leading to
misjudgements.

Systematic Interactions between Human / Street /

Technics

The results of the detailed accident studies and

viewing studies of motorcyclists show that in the
system human-street-technics, human information
perception has a defining influence on the traffic
safety.
The factors are in direct relation to each other, Figure 5: risk-triangle
controlled information perception is continuously
required:
Result:
1. Human factors The human information perception (human factor)
a. Order of priorities is in constant direct relation to the vehicle and the
road.
b. Perception / navigation of curves
c. Perspective interrelations depending
on the head posture
d. Danger recognition and danger esti- Contact
mation Univ.Prof. DI Dr. Ernst PFLEGER
e. Driver experience: “Beginner” vs. Chairman of EVU-Austria
EPIGUS-Institute of holistic accident and safety research
“Seasoned driver“ etc. Schmerlingplatz 3/7
1010 Wien
2. Street e-mail: ernst@pfleger.cc
a. Surface conditions tel: +43 664 20 20 234
b. Road markings

174
© 2013 by EVU
EVU 2013 - 21

Rollover of Mountain Bikes which are equipped with

disc brakes and suspension forks
Christian Hittinger
München/Deutschland

Abstract
This paper deals with the reconstruction of an accident in which a bicycle rider performed a rollover with a mountain bike.
The bike was equipped with disc brakes and a suspension fork. In the accident, the driver suffered serious spinal injuries
and has been paralysed since the accident from the head downwards. As part of the reconstruction of the accident it was
clarified which mechanisms lead to a rollover of a mountain bike, and the influence of a faulty suspension fork on rollover
propensity is shown. The braking performance achievable on the dry asphalt pavement was investigated and documented.
The driving technique “maximum braking”, to achieve maximum deceleraton in emergency situations, is presented.

Keywords:
Mountainbike, Hydraulic Disk BrakesSuspension
fork, Rollover

1. Introduction
In April 2009, a traffic accident in which the driver
of a mountain came to fall and suffered severe inju-
ries to the spine occurred. Initially, the accident was
not investigated by an expert, since no further per-
sons where involved in the incident. Because of the
serious injury to the cyclist it needed to be clarified
later, whether defects existed on the bike that may
have caused or encouraged the fall. As facts we had Figure 1 Way to the roundabout
the final position of the bicycle rider, the bike and
the helmet.
The accident site is located on a roundabout, which
the rider wanted to pass. The road travels downhill Rest position
towards the roundabout (Figure 1).
The final position of the cyclist was near by to an
electrical junction box, which was located approxi-
mately opposite the entrance point to the roundabout
(Figure 2).

Figure 2 Rest position of the driver

175
 Christian Hittinger, München/Deutschland
The bicycle was a mountain bike, which was
equipped with a suspension fork and disc brakes.
The fork had a lockout feature that should prevent
the fork movement if set by the rider.
Figure 3 Bycicle
Figure 4 Lockout-function
The helmet showed peculiar accident tracks that
were at the center of the helmet and ran at a right an-
gle to the direction of travel.
Figure 5 Helmet with notch and scratches
The investigation of the bike revealed that it was
largely new. The only circumstance that stuck out
176
was that the fork very much immersed in traction
and had low resistance.
A spongy ride was clearly detectable. The
investigation of the accident site revealed that the
loading damage on the helment could be assigned to
a chain that was stretched nearby the final position of
the rider.
Figure 6 Chain
Figure 7 Measuring the notch
The impact angle of the helmet to the chain and the
injuries to the spine revealed that the cyclist had per-
formed a rollover when braking. For a more detailed
evaluation, test runs was made on the accident scene.
The investigation showed that the end position of the
cyclist was directly in the extension of the point of
greatest slope when driving through the roundabout.
The bike moved tangentially from the curve, the rid-
er obviously performed heavy braking, and the de-
scribed rollover occurred.
 Rollover of Mountain Bikes which are equipped with disc brakes and suspension forks
Figure 8 Comparision of the tracks at the helmet and the
chain
Figure 9 Maximale Schräglage 19° bei 25 km/h
2. Brake system
The brake system was a disk brake with hydraulic
transmission.
Figure 10 Disk brake
The manual gave a maximum decelaeration of 7.2
m/s²
The advantages of this type of brake are
• great wet and dry braking performance
• good controllability
• low actuation force required
Disadvantage: High decelerations favor the rollover
propensity. The brake system was nearly new, when
it was tested.
3. Suspension fork
The fork was a TPC (Twin Piston Chamber) design.
This system has two pistons, one affects the
deflection and the other affects the rebound. The
suspension is solved via an air spring that is adjusted
to the driver's weight by more or less air pressure
brought by a pump. The rebound is achieved through
a combination of air and oil. The oil flows through
an adjustable valve between two chambers back and
forth and thus allows an adjustable damping. The
lockout function can block this overflow, causing the
fork to be rigid. This has advantages when driving on
paved roads where the suspension fork is not desired.
The review of the fork revealed that the lockout fea-
ture was disabled. In the assembly the bowden cable
was too long, so that the valve was not actuated.
Figure 11 Suspension fork TPC type (Quelle: [1])
During a test drive we had a very spongy ride with
pulsed deflection. The air pressure was checked with
a special air pump.
177
 Christian Hittinger, München/Deutschland

m  g  ls  m  a  hs  FBy  lR , (1)

Contact Forces:
Equilibrium of forces in x-direction:

FAx  FBx  m  a , (2)

Equilibrium of forces in y-direction:

FAy  FBy  m  g , (3)

Contact forces general:

Figure 12 Special air pump for adjusting the pressure
Front:
The air pressure was 50 psi, 3.5 bar. The rider’s FBy
weight was 100 kg and the manual told us that the %v   100 (4)
pressure should be 7.5-10.5% of the weight. So the FAy  FBy
pressure should have been 7.5-10.5 bar, and was Rear:
therefore far too low. To investigate the influence of FAy
too low air pressure tests have been conducted, %h   100 (5)
FAy  FBy
which are described in chapter 5.

4. Braking Braking forces: (Equilibrium in x-direction)

Before the mechanisms involved in rollover of a bi- The mass of the rider was 100 kg. The deceleration
cycle are considered in more detail, the brakes must was assumed with 5.0 m/s² to perform the following
be examined first. Contact forces must be investigat- calculations.
ed. Front and rear wheel contact forces are different,
they are significantly higher at the front than at the The dimensions of the bicycle are derived from this
rear. The centre of gravity of the system of cyclists photograph:
and bicycles can be assumed with good approxima-
tion in the area of the pelvis above the bottom brack-
et at the end of the seatpost.

Figure 14 Dimensions of the bicycle

FAx  FBx  m  a , (6)

Figure 13 Contact forces FAx=0, if no braking at the rear brake.

Moment equilibrium at point A:

178
 Rollover of Mountain Bikes which are equipped with disc brakes and suspension forks

FBx  100kg  5,0m / s ²  500N , The tipping security  is calculated on the relation-
Contact forces: (Equilibrium in y-direction) ship:

m  g  ls  m  a  hs
FBy  , (7) MS
lR  , (9)
100  9,81m / s ²  0, 4m  100kg  5,0m / s ² 1,1m
MK
FBy   942, 40 N where:
1,0m
 MS Standmoment
 MK - Tiltmoment
FBy  942,40 N ,
M S  m  a  (lR  lT ) (10)
FAy  m  g  FBy , (8)
M K  m  g  hS (11)
FAy  100kg  9,81m / s²  942,40 N  39 N
100kg  9,81m / s ²  0,6m
  1,00 (11)
Contact forces: 100kg  5,35m / s ²  1,1m

Front:
942, 40 N
%v   100  96,03%
942, 40 N  39 N The system does not tilt until a deceleration of 5.35
Rear: m/s² is reached. At larger retardments, the rider goes
%h  3,97% over the handlebars, unless he has moved the centre
of gravity behind the saddle.
The contact force at the rear wheel is about 4%, the
contact force at the front wheel is about 96%. A test run was made with a deceleration of 5.4 m / s
². The suspension behaved as proposed and the bike
The main power transmission when braking is done overturned. Both brakes were simultaneously operat-
at the front wheel. The uneven distribution of forces ed abruptly with maximum force.
favours rollovers at high retardments in emergency
situations. The trial showed that before the attainment of the
maximum deceleration on the front is reached, the
5. Rollover rear wheel lifts off the ground, 0.3 s after the initia-
tion of braking.
The system performs a rollover when the tilt moment
is larger than the stand moment [1]. The time from pressing both brakes up to the time of
loss of control was approximately 0.8 s The test also
shows that the loading of both brake actuations does
not prevent a rollover.

Generally, a rollover is also possible with a rigid fork

with correspondingly high retardment. The early lift-
ing of the rear wheel, however, favoured the rollo-
ver.

Figure 15 Tilt- and standmoment

179
 6.1 Rear wheel braking

Figure 16 Driving test with rollover

6. Achievable deceleration
The achievable braking deceleration of this bike type
have been studied. Three braking tests on dry asphalt
were conducted. The first attempt was decelerated
only with the rear brake. In the second experiment
only the front brake was pulled. On the third try both
brakes have been applied and the centre of gravity
was simultaneously moved behind the saddle. One
calls this driving technique Maximum Braking
Technic.

180
 Rollover of Mountain Bikes which are equipped with disc brakes and suspension forks

Figure 17 Rear wheel braking

The braking with the rear brake alone led to a lock-

ing of the rear wheel with a strong track mark about
12 m. The deceleration was about 3 m / s ²

Figure 20 Front wheel braking

6.3 Maximum Braking Technic

In this braking technique, the centre of gravity is
shifted behind the saddle such that the maximum
deceleration can act without rollover.

Figure 18 Skidmark

6.2 Front wheel braking

Braking with the front wheel alone shows signifi-
cantly higher values of deceleration, but the rear
wheel lifted of the ground. The system was unstable
and near to a rollover.

Figure 21 Maximum Braking

The delays are achievable with this technique are

between 5-6 m / s²

The maximum delay can be maintained thanks to this

technology throughout the braking phase. about A
bike travelling at 25 km / h can be stopped in this
way in about 5 m.

Figure 19 Front wheel braking

181
 Christian Hittinger, München/Deutschland

[5] Green, J. et. al., Bicycle Accident Reconstruction and

Litigation, Lawers & Judges Publishing Company, Tuc-
son, USA, 4. Auflage, 1996.

[6] http://www.ipmba.org
http://www.youtube.com/watch?v=z820UnNKVHc.

Contact
Christian Hittinger, Dipl.-Ing. (FH)
Sachverständigenbüro Max Bauer
Flößergasse 7
81369 München
www.unfallanalytik.com
e-mail: hittinger@unfallanalytik.com

Figure 22 Maximum Braking

7. Conclusion
The investigation of this case has shown that a
faulty fork promotes rollover propensity, as the
rear wheel lifts, caused by the increased contact
force when the maximum braking power is not
reached. Thus, the centre of gravity is raised and
a rollover is more likely. Suspension forks are
useful facilities, which - when properly setting –
maintain the contact of the tyre with the road.
However, incorrect settings can lead to
accidents. During braking of a bicycle, the brake
force is transmitted substantially over the front
wheel. At full utilisation of the braking power of
a hydraulic brake, in conventional driving
technique, a rollover, even when pressing both
brakes can not be avoided.
The Maximum Braking technique makes it
possible to maintain the maximum possible
deceleration during the entire braking phase.

References
[1] Michael Gressmann et. al., Fachkunde Fahrradtechnik,
Verlag Europa Lehrmittel, Han-Gruiten, 3. Auflage, 2009.

[2] Michael Gressmann, Fahrradphysik und Biomechanik,

Delius Klasing Verlag, Kiel, 10. Auflage, 2009.

[3] Wilson, D., Bicycling Science

Massachusetts Institute of Technology, Cambridge, Mas-
sachusetts, 3. Auflage 2004.

[4] Green, J., Bicycle Accident Reconstruction for the Fo-

rensic Engineer, Trafford Publishing, Victoria, Canada, 5.
Auflage, 2001.

182
© 2013 by EVU
EVU 2013 - 22

Bicycle tire friction coefficient variance in wet and dry

conditions across multiple surface pavement types
Daniel J. Melcher, P.E.
Christian R. Sax, P.E., PTOE
Rachel E. Keller, P.E.
Armstrong Forensic Engineers

Abstract
Bicycle incidents occur in dry and wet conditions, and in a wide variety of locations, including public roadways,
dedicated bicycle paths, and private parking lots or driveways. The capabilities for bicycle acceleration, braking, and
lateral movement depend on the available friction between the bicycle tire and the roadway surface. When analyzing the
sequence of collision events, bicycle speeds, bicycle handling, and potential bicyclist avoidance capabilities, it is
important to understand the available coefficient of friction at the time and place of the incident. A literature search
revealed useful information regarding engineering design values for the minimum safe pavement friction characteristics
for bicycle operations, but little published data for accident reconstruction purposes regarding the typical values of
available peak bicycle tire-roadway friction coefficients for a variety of surface types in real world situations. This
research collected coefficient of friction data in wet and dry conditions on multiple pavement types, including asphalt
and concrete public roadways, concrete sidewalks, dedicated bicycle paths, and private property sealed parking lots and
driveways.

Introduction
An accident reconstruction analysis may be called
for in a variety of bicycle collision scenarios, includ-
ing single-bicycle crashes, bicycle vs bicycle colli-
sions, bicycle vs pedestrian accidents, or bicycle vs
motor vehicle impacts. Though the scenarios may
vary, all reconstructions share a common goal of de-
termining vehicle speeds, paths of travel, timelines
of events, causal factors, and an evaluation of avoid-
ance capabilities by all involved parties. To accom-
plish these evaluations, the analyst must study issues
related to the motion of the bicycle in terms of posi-
tion, velocity, and acceleration. Fundamentally, all
motion or changes of motion for bicycles depend on
the interaction of the two bicycle tires with the riding
surface. Steering, lane changes, cornering, turn-
away, braking, and acceleration all involve forces
exchanged at the tire-surface contact patch. For this
reason, the availability and magnitude of friction
forces are paramount to bicycle accident reconstruc-
tion. This research presents values for the peak
available coefficient of friction between bicycle tires
and various surfaces, which can be used as part of a
comprehensive bicycle collision reconstruction
Road friction fundamentals
The Coefficient of Friction is a physical property of
any two materials in contact with one another, and is
expressed as the ratio of friction force between two
the bodies relative to the force pushing them togeth-
er.
The coefficient is unitless, but is commonly ex-
pressed in terms of g-units, relative to the gravita-
tional constant.
The available peak coefficient of friction for a given
tire will vary based on the material properties of the
roadway, the texture of the roadway, and the possible
presence of foreign substances on the roadway.
Common materials used to construct roadways and
bicycle facilities include asphalt, concrete, gravel, or
graded dirt. There are a variety of compositions of
each of these materials, with different aggregate siz-
es, binders, and manufacturing processes, all of
which result in slightly different friction characteris-
tics. The macrotexture and microtexture of a surface
can play a significant role in interaction with rubber
tires in dry and wet pavement conditions, so even
among asphalt surfaces, different compositions can
have varying properties (Figure 1). Similarly, aging
and maintenance of all surfaces will affect their fric-
tion characteristics, with important variables includ-
ing weather, axle loading, and exposure to chemi-
cals. For these reasons, accident reconstruction ana-
lysts usually rely on a range of values for tire-surface
friction in order to account for variability and uncer-
tainty.

183
 Melcher, Daniel; Sax, Christian; Keller, Rachel
The peak tire-roadway friction available is an im-
portant concept in vehicle dynamics, as it forms the
outline of the Friction Circle within which all vehicle
dynamic events can potentially occur (Figure 2). If
friction is used in longitudinal demand such as brak-
ing or acceleration, then less is available for lateral
acceleration, and vice versa. Vehicle or bicycle tires
are most likely to reach peak coefficient of friction
during braking events. As wheel slip increases as a
result of braking effort, the effective deceleration in-
creases until reaching the theoretical peak coefficient
of friction value. However, once full wheel lockup
occurs, the effective sliding friction coefficient of the
tire against the surface is usually lower than the peak
available friction value. Vehicle ABS systems are
designed to modulate braking effort to keep the
wheel slip as close to optimal as possible, resulting
in deceleration very near to the peak value. Bicycles
are not equipped with ABS systems, so braking
modulation depends on rider inputs.
Ranges of peak available coefficients of friction have
been published for passenger vehicles and commer-
cial vehicles based on testing for many surface types.
Table 1 presents widely used published available co-
efficients of friction for passenger vehicle rubber
tires on common categories of roadway surfaces.
The values in Table 1 are for clean dry surfaces.
Friction capabilities for virtually all surfaces
decrease if wet, covered in ice or snow, or covered in
a foreign substance such as oil or grease. Table 2
presents published values for the avaiable coefficient
of friction between passenger vehicle rubber tires
and categories of surfaces when wet. These results
indicated water reduced the coefficient of friction by
as much as 43% below the dry pavement values.
Bicycle braking fundamentals
A bicycle is made of up of four primary systems: the
frame, the wheels, the braking mechanism and pedal-
ing / gear mechanism (Figure 3). The principles of
operation are generally shared across bicycle types.
The scope of this paper focuses on the wheel assem-
blies and braking systems.
The wheel assembly of a bicycle comprises a hub,
spokes, rim, and tire (Figure 4). The hub, which
contains an axle, connects the center of the wheel to
the frame of the bicycle. Hubs may also be linked to
the gearing mechanism, if the bicycle is so equipped.
Spokes provide tension between the hub and the rim
of the wheel. The rim, which ranges in composition
from carbon fiber to aluminum alloy, provides a
framework for the tire. Most modern bicycle tires
are either tubed or tubeless tires. If the bicycle has a
tubed tire, the inflated tube between the tire and the
rim helps the tire to remain firmly engaged to the rim
(Figure 5). Tubeless tires make contact with the rim
on its inner edges, and air is pumped directly into the
space between the tire and the rim to create tension
to hold the wheel onto the rim. Generally, on-road
bicycles have thinner tires with minimal tread so that
when the bicycle is in motion there is a narrow con-
tact patch with the roadway, helpful for high speed
operation. Off-road tires are typically wider and
have more tread depth so that stability is easier to
maintain in rugged terrain.
Most common braking systems on bikes utilize rim
braking (Figure 6). Rim brakes are a mechanical-
friction system applying a shoe to contact the rim of
the bike. For bicycles with hand brakes, a cable
connects the handle to a caliper, which houses brake
shoes on either side of the rim. When the handle is
actuated, the shoe is squeezed onto the rim creating a
friction surface between shoe and rim and subse-
quently slowing the wheel. Each hand brake will ac-
tuate a separate wheel and can be modulated utilizing
varying amounts of pressure. With well adjusted
brakes and a hard brake application, braking can be
applied to the point that the subject wheel locks and
begins to skid. High-end performance or competi-
tion units may be equipped with hydraulic rim brakes
in which fluid is pumped through hoses to actuate
the brake shoes.
Generally the left hand-brake actuates the front brake
shoes and the right hand-brake actuates the rear
brake shoes. As the brakes of a bicycle are applied
the weight of the bicyclist will shift to more heavily
load the front wheel. Because of this factor, the most
efficient braking usually can be attained by utilizing
the front brake with modulated rear wheel braking to
help maintain stability.
One of the largest limiting factors in real-world bicy-
cle braking deceleration capabilities is pitchover, a
phenomenon that can result in the rider being pro-
jected forward over the handlebars. This occurs if
the braking at the front wheel results in a rapid de-
celeration of the bicycle that creates a moment
around the front wheel contact patch. In theoretical
terms, the deceleration rate at the front wheel has to
exceed the ratio of the height of the rider center of
gravity divided by the wheelbase of the bicycle if
both of the wheels are in contact with the ground
(Figure 7). The basic theoretical equation for calcu-
lating the maximum deceleration of a bicycle before
reaching pitchover is:
Height of Center of Gravity
a= Wheelbase
184
 Bicycle tire friction coefficient variance in wet and dry conditions across multiple surface pavement
However, real-world variables including bicycle
mechanical construction and adjustment, rider
posture, body size, and roadway slope affect the
theoretical threshold for pitchover. In Metz’s work,
he studied how roadway irregularitys contribute to
pitchover, which is one of the limiting factors for bi-
cycle deceleration1. Bretting indicated that pitchover
thresholds typically occur between 0.6 and 0.7 g for
most road bicyclists’ bicycle geometry, seating posi-
tion, and physiology (Figure 8)2. Forester found that
pitch over will typically occur when there is a
deceleration of 0.67 g3. Green suggested that a
bicyclist can risk pitchover on level ground in a
crouched position at deceleration rates as low as 0.5
g4. If the bicyclist were to exceed the real-world
threshold for that rider and bicycle combination, the
bicyclist will pitch over the handlebars and front
wheel (Figure 7).
The experience level of a rider plays a large part in
the actual deceleration that is achieved. A highly
experienced rider will learn to shift his weight
toward the rear wheel and keep his center of mass as
low as possible while modulating the brakes in such
a way as to reach near lockup for the front wheel.
Less experienced riders may not be as skilled in
physical positioning and brake modulation, and thus
may not achieve as optimal of a deceleration rate.
Whitt and Wilson suggested that the rear brake is
about 75% as effective as the front brake5. Green
states that 90% of the retarding force in most braking
applications comes from the front brake4.
Transportation design standards
According to the 2012 AASHTO Guide for the De-
velopment of Bicycle Facilities there are five
categories of facilities that are designed for bicycles:
shared lanes, paved shoulders, bike lanes, bicycle
boulevards, and shared use paths. Of these, only
shared use paths have a surface course that is
designed specifically for the characteristics of
bicycles. The other facility types have a surface
course that is designed for automobiles but that can
also be expected to be used by bicyclists. Private
driveways and parking lots are also areas designed
for automobiles that may have bicycle uses.
AASHTO lists various performance criteria for a
typical adult on an upright bicycle and on a
recumbant bicycle, including acceleration, braking,
and lateral acceleration (Table 1).
When designing a shared use path, one method relies
on the facility‘s superelevation and the available
coefficient of friction. For design purposes,
AASHTO recommends friction factors for paved
shared use paths ranging from 0.34 g at 6 mph to
0.21 g at 30 mph. As design values allowing for a
margin of safety, they are well below the peak
capabilities for most surfaces.
More documents have been published regarding the
minimum required automobile tire / wet pavement
friction for automobiles. NCHRP 108: Guide for Pa-
vement Friction found that 15 states use a wet
coefficient of friction 0.20 to 0.37 g as a threshold
for deficiency6. NCHRP 108 cited Arizona policy
that, “research by others has indicated a coefficient
of friction of 0.35 g reflects an accident breakpoint”.
California policy noted that “a CALTRANS study
found that there was a relationship between a friction
coefficient of 0.25 g and a high frequency of
accidents”. A report by Ivan et al. recommends that
measures should be taken to correct a pavement
surface when the wet coefficient of friction is less
than 0.30 g7. A 2007 article by Snyder concluded
that coefficients of friction between 0.30 and 0.40 g,
measured at 40 mph on wet surfaces using a ribbed
tire, are generally considered acceptable for intersta-
te highways or other roads with design speeds
greater than 40 mph8.
Literature review on bicycle deceleration
rates
The United States Consumer Product Safety Com-
mission (CPSC) requires that bicycles with front and
rear brakes be capable of stopping in 15 feet from 15
mph with a 150 to 200 lb rider, the equivalent of an
average 0.5 g deceleration9. The Florida Bicycle
Planning and Design Handbook also indicated that
the maximum deceleration for a bicycle is 11 mph
per second (0.5 g)10.
However, bicycle brake design capability and real
world performance may not coincide in all cases.
Broker and Hill suggest that a typical bicyclist
should be able to decelerate at the AASHTO design
deceleration rate of 0.35 g11. Beck indicated the
braking and accelerating capabilities of mountain
bikes tested on dry dirt, asphalt, and concrete road-
ways12. He did not attempt to test wet roadways or
road bicycles. On dry surfaces he found that the
range of overall average deceleration values was
0.37 to 0.50 g, higher than the results of prior testing
done by Eubanks and Haight who determined a
range of 0.24 to 0.28 g for a roadway with a passen-
ger vehicle coefficient of friction of 0.70 g13. Beck’s
testing substantiated the understanding that a more
expert rider has the ability to accelerate and deceler-
ate more aggressively than novice riders.
However, even Beck’s numbers are likely underre-
porting a maximum available braking capability, as
185
 Melcher, Daniel; Sax, Christian; Keller, Rachel
he took an average deceleration from the beginning
of negative deceleration, as opposed to the testing
presented in this paper in which only maximum sus-
tained peak coefficient of friction has been reported.
As with other vehicles, bicycles achieve lower decel-
erations on wet surfaces. This occurs because of re-
duced friction between the rubber tire and the pave-
ment, as well as the buildup of water on the bicycle
rim reducing the effective friction against the brake
pads. The Florida Bicycle Planning and Design
Handbook estimates that braking performance is re-
duced by 50-80% when the rims are wet10. This is
consistent with the AASHTO dry coefficient of fric-
tion being 0.32 g as compared to a wet coefficient of
friction of 0.16 g for design values14.
In Green’s analysis of braking during wet conditions,
he notes that water between the rim and the brake
shoe is dissipated in 50 rotations of the wheel4.
Therefore, the braking capabilities of a bicycle in
wet conditions increase over time during a braking
application; however, it never reaches equivalence
with dry braking capabilities. Whit and Wilson’s
testing confirmed the number of revolutions to in-
crease braking efficiency when the rims of a bicycle
are wet5.
Methodology
There is a discontinuity between the design standards
for roadways and bicycle facilities, the published
values for bicycle friction, and the fundamentals of
bicycle braking. Design values typically refer to the
peak available friction in any direction (cornering,
acceleration, or braking). However, bicycle testing
values that have been reported previously have pre-
sented average braking values only, from initiation
of braking to zero velocity, which depends largely on
the actions and input modulations of the rider. For
this reason, a research plan was developed to attempt
to quantify the peak sustainable coefficient of fric-
tion between bicycle tires and the various test surfac-
es. Passenger vehicles and bicycles were equipped
with a Video VBOX Lite data acquisisition system.
This system is GPS-based and has been validated
and widely used in a variety of vehicle dynamics, tire
testing, roadway testing, and similar applications15.
Friction testing was conducted using three test set
ups: two road bicycles, and a mountain bicycle. The
first road bicycle was a Kona Honky Tonk equipped
with Vittoria Zaffiro 700x25c tires. The second road
bicycle was a Trek 1500 equipped with Bontrager
Select B 700x25c tires. The mountain bicycle was a
Trek 6000 equipped with iRC Mythosxc 26x2.10
tires.
Testing locations were identified in multiple areas of
Florida and Texas, USA, and included public
roadways, concrete sidewalks, paved dedicated
bicycle trails, and sealed parking lots and driveways
of private properties. All surfaces were essentially
flat in order to eliminate the influence of grade on
the results. The same bicyclist was used in all tests to
reduce the influence of rider skill variability.
Both dry conditions and wet conditions were tested
and evaluated. The reported values do not include
build-up at the beginning of braking or rider modula-
tion. The authors analyzed each data set and
identified the largest deceleration rate achieved by
the bicycle that was sustained over a duration of at
least 200 milliseconds.
Results
The results of all of the tests completed using the
three bicycles are summarized in Table 2. All tested
surfaces provided available friction characteristics
well greater than published minimum standards for
bicycle facilities. The peak sustainable friction be-
tween all types of bicycle tires and all of the wet and
dry tested surfaces were found within a relatively
narrow range between 0.40 and 0.66 g. The mean
peak deceleration rate for the 20 different combina-
tions of bicycle type, surface types, and surface con-
ditions tested was 0.52 g. No pitchover incidents oc-
curred during the testing process. There were some
distinct trends to assist the analyst in reconstructions
involving specific variables or scenarios:
 WET/DRY: The average bicycle tire peak
coefficient of friction on all wet surfaces was
0.48 g and was 0.54 g for dry surfaces. The
presence of water on the pavement and/or
bicycle tire and rim was found to decrease
the average bicycle peak coefficient of fric-
tion by 11%, much less of an effect than
published in some previous studies.
 MOUNTAIN/ROAD: The mountain bike
tire’s average peak coefficient of friction
across all surface conditions was 0.54 g,
compared to 0.51 g for road bike tires across
all surface conditions. This indicated that
there is not likely a significant difference be-
tween these tire types in most roadway and
bike path scenarios.
 BICYCLE / MOTOR VEHICLE: Given the
differences in mechanical brake system de-
sign and tire rubber compositions, it was ex-
pected that bicycle tires would achieve fric-
tion values less than motor vehicles. Our
testing found a range of peak coefficient of
friction of 0.46 g to 0.66 g for bicycles on
dry surface. This is 22 % to 34% less than
186
 Bicycle tire friction coefficient variance in wet and dry conditions across multiple surface pavement
the published values for passenger vehicle
tires on dry concrete or asphalt roadways.
Our testing found a range of peak coefficient
of friction of 0.40 g to 0.61 g for bicycles on
wet surfaces. This is 0% to 13% less than
the published values for passenger vehicle
tires on wet asphalt or concrete surfaces.
Summary and recommendations
In motor vehicle reconstruction applications, pub-
lished tire-roadway friction values tabulated from
testing usually report the peak sustained friction ca-
pability available for that surface. This value is use-
ful, but requires cautious application to a specific
scenario. Motor vehicles are usually not capable of
accelerating at a peak coefficient of friction due to
engine and drivetrain limitations. In braking, the an-
alyst must determine how many wheels were locked,
pinned, or rolling, and what degree of driver braking
input was involved. Many reconstructions utilize
less than peak values when no locked tire marks are
left on the roadway, or may require adjustments for
ABS equipped systems. Motor vehicles are also
rarely cornered with lateral acceleration values ap-
proaching the peak available friction, primarily due
to variables of suspension design and driver experi-
ence and comfort. When both lateral and longitudi-
nal accelerations are demanded, the friction circle
dictates how much of each may be available for the
tire to use.
Similarly, the results of this bicycle testing are in-
tended for use as a peak available friction value that
may not represent the average deceleration, accelera-
tion, or cornering lateral acceleration used or
achieved by the rider in any one situation. However,
the results are useful and insightful as a basis for de-
veloping a friction circle within which the bicycle
can operate. An experienced rider with optimal
braking inputs can accomplish braking deceleration
approaching these values. Lateral acceleration capa-
bilities are useful for analysis of lane changes, cor-
nering, or avoidance maneuvers. Combinations of
lateral and longitudinal accelerations can be analyzed
in the context of total available friction.
With modern data acquisition equipment, analysts
may have the opportunity to test a specific bicycle, a
specific road surface, or a combination of bicycles
and surfaces to assist them in their reconstruction.
However, in the absence of site-specific testing, pub-
lished tabular values such as presented in this report
constitute a valid starting point for reconstruction
analysis.
The research initiated for this report has generated
ideas for future studies to continue developing the
body of knowledge in the area of bicycle-roadway
coefficients of friction. Further testing and analysis
possibilities include evaluation of average drag fac-
tor over the entirety of braking events, additional
surface types or bicycle configurations, or variations
among riders.
References
[1] Metz, D., Road Bicycle Dynamics in the Presence of
Idealized Roadway Irregularities, SAE technical paper
2010-01-0053 (2010).
[2] Bretting, G., Analysis of Bicycle Pitch-Over in a Con-
trolled Environment, SAE technical paper 2010-01-0064
(2010).
th
[3] Forester, J.; Effective Cycling, 6 Edition; MIT Press
(1993), Cambridge, Massachusetts.
[4] Green, J., Bicycle Accident Reconstruction for the
Forensic Engineer, 2001; pg 130-108.
[5] Whitt, F. and Wilson, D., Balance and steering,
Bicycling Science (1993), Cambridge, Massachusetts.
[6] Hall, J., Smith, K., Titus-Glover, L., Wambold, J.,
Yager, T., Rado, Z., NCHRP Web-Only Document 108:
Guide for Pavement Friction (2009), National Cooperative
Highway Research Program, Champaign, Illinois, State
College, Pennsylvania, Hampton, Virgina, University
Park, Pennsylvania.
[7] Ivan, J., Ravishanker, N., Jackson, E., Aronov, B.,
Guo, S., Incorporating Wet Pavement Friction into Traffic
Safety Analysis (2007), University of Connecticut, Storrs,
Connecticut.
[8] Snyder, M., Pavement Surface Characteristics: A Syn-
thesis and Guide (2006), American Concrete Association,
Skokie, Illinois.
[9] www.cpsc.gov.
[10] Florida Department of Transportation, Bicycle Facili-
ties Planning and Design Handbook, Pedestrian and Bicy-
cle Program (2000), Tallahassee, Florida.
[11] Broker, J. and Hill, P., Bicycle Accidents Biome-
chanical, Engineering and Legal Aspects, Lawyers and
Judge (2006), Tucson, Arizona.
[12] Beck, R., Mountain Bicycle Acceleration and Brak-
ing Factors; Proceedings of the Canadian Multidiscipli-
nary Road Safety Conference XIV (2004), Ottawa Ontar-
io.
187
 Melcher, Daniel; Sax, Christian; Keller, Rachel

[13] Eubanks, J. and Haight, W, Accident Reconstruction Contact

for Bicycle-involved Traffic Collisions; Texas A&M Daniel J. Melcher, P.E. (Presenter)
Pedestrian/ Bicycle Accident Reconstruction Course 624 Davis Street, 2nd Floor
(1991). Evanston, IL 60201, USA
Email: d.melcher@armstrongforensic.com
[14] Guide for the Development of Bicycle Facilities, 4th Telephone: +1 (847) 906-2860
Edition (2012), American Association of State Highway
and Transportation Officials, Washington, District of Co- Christian R. Sax, P.E., PTOE
lumbia. 8500 Shoal Creek Boulevard, Suite 201B
Austin, TX 78757, USA
[15] Ruth, R. and Daily, J., "Accuracy of Event Data Re- Email: c.sax@armstrongforensic.com
corder in 2010 Ford Flex During Steady State and Braking Telephone: +1 (512) 765-9532
Conditions," SAE Int. J. Passeng. Cars – Mech.
Syst. 4(1):677-699, 2011, doi:10.4271/2011-01-0812. Rachel E. Keller, P.E.
17844 North US Highway 41
[16] Shaffer, S., Christiaen, A., and Rogers, M., Assess- Lutz, FL 33549, USA
ment of Friction-based Pavement Methods and Regula- Email: r.keller@armstrongforensic.com
tions, National Transportation Research Center Incorpo- Telephone: +1 (813) 279-2155
rated (2006), U.S. Department of Transportation

Type of Roadway Coefficient of Friction

Asphalt 0.70 to 0.85 g
Concrete 0.70 to 0.85 g
Gravel 0.40 to 0.85 g
Table 1. Typical coefficient of friction values between automobile tires and common dry surfaces (source: Fricke (2010)
and Fricke (1990))

Type of Roadway Wet Coefficient of Fric-

tion
Asphalt 0.40 to 0.70 g
Concrete 0.40 to 0.70 g
Gravel 0.40 to 0.80 g
Table 2. Typical coefficient of friction values between automobile tires and common wet surfaces (source: Green and
Fricke(1990))

Table 3: Bicycle performance criteria (source: AASHTO Guide for the Development of Bicycle Facilities, 2012

188
 Bicycle tire friction coefficient variance in wet and dry conditions across multiple surface pavement

Table 4: Summary of peak friction coefficient measured with bicycles at each test location

Figure 1. Illustration of textured surfaces of roadways (National Transportation Research Center)

Figure 2. Illustration of friction circle concept

189
 Melcher, Daniel; Sax, Christian; Keller, Rachel

Figure 3. Parts of a typical bicycle

Figure 4. Bicycle wheel assembly

Figure 5. Diagram of standard versus tubeless tires (bicyclesportshop.com)

190
Bicycle tire friction coefficient variance in wet and dry conditions across multiple surface pavement

Figure 6. Rim Brake Assembly

Figure 7. Free body diagram of bicycle and rider (Broker and Hill)

Figure 8. Pitchover scenario (Beck)

191
Melcher, Daniel; Sax, Christian; Keller, Rachel

192
© 2013 by EVU
EVU 2013 - 23

Simulation and Analysis of Quad Bike Rollovers using

PC-CRASH to Evaluate
Alternative Safety Systems
Shane Richardson, Andreas Sandvik, Chris Jones, Tia Orton,
Nikola Josevski, Wei Pei (Tandy) Pok and Thomas Emmett
Delta-VExperts, Australia

Abstract
From 2001 to 2012, over 160 Australians have died in quad bike (ATV) incidents. Annually in the USA there will be at
least 700 fatalities from quad bike related incidents. The options which have been considered to protect riders in the
event of a quad bike rollover are: 1) Personal protective equipment; 2) Active riding; 3) Crush Protection Devices
(CPD); 4) Roll Over Protective Structure (ROPS) without restraint; and 5) ROPS with restraint. The Institute for Safety
Compensation and Recovery Research identified: “…serious issues with the simulation methods used and the nature of
incidents tested to predict the effect of crush protection devices on Quad bike roll over injuries and fatalities”.
PC-Crash is a commercially available collision simulation tool, which can output simulated forces of a simulated riders
body parts. A PC-Crash model of a motorcycle and rider was adapted to create an exemplar Original Equipment
Manufacturer (OEM) quad bike and quad bike with CPD. 18 riderless quad bike rollover tests (7 OEM quad bikes and
11 quad bikes with CPD) were used to validate the PC-Crash models of OEM quad bike and quad bike with CPD. In a
serious or fatal quad bike rollover, injuries could result from impact or crush to either the torso, neck, head or
combination. Entrapment, specifically of the torso, can result in traumatic or mechanical asphyxia. 1,600 simulations (4
x 400) were used to comparatively evaluate impact and crush injuries of quad bike riders. The rollover simulations
identified that the rider of: OEM quad bike and unrestrained rider, quad bike with CPD and unrestrained rider, quad bike
with ROPS and unrestrained rider and quad bike with ROPS and restrained rider could be traumatically or mechanically
asphyxiated 46, 28, 0 and 8 times respectively. Where there is an identifiable risk of serious or fatal injury from quad
bike rollover, consideration should be given to fitting either: CPD, ROPS or ROPS with rider restraint; to mitigate the
potential for serious and/or fatal injury due to torso impact, crush or entrapment during a quad bike rollover.

US Consumer Product Safety Commission, Commis-

Background
A Safe Work Australia discussion paper 0detailed
research of Lower et al. 0:
1. Since 2001, over 160 Australians have died in
quad bike incidents.
2. 45% of fatalities are work related and 55% are
from recreational use. In 2011 there were 23
quad bike related fatalities and 18 of these oc-
curred on a farm.
3. Deaths due to roll over account for about half of
the quad bike fatalities and 90% of rollover
deaths occur on farms.
4. Quad bikes are the leading cause of injury and
death on Australian farms.
5. Children under 15 years account for almost 20%
of all quad bike deaths.
6. In 2011 the youngest and oldest quad bike riders
killed were 4 years old and 94 years old respec-
tively.
7. 45% of deaths are those aged 45 years and over.
sioner Mr Robert Adler 0 stated: “Each year there
will be at least 700 funerals because of a (quad bike)
ATV-related incident. Since 1982, the US CPSC has
received more than 11,000 reports of ATV-related
fatalities. Almost 3,000 have been children…”
The options which have been considered to mitigate
and/or protect quad bike riders in the event of a roll-
over are:
1. The Original Equipment Manufacturer (OEM)
quad bike, i.e. without any structural modifica-
tion;
2. Active riding;
3. Personal protective equipment (specifically
helmets);
4. Crush Protection Device (CPD) 0, refer to Fig-
ure 1;
5. Roll Over Protective Structure (ROPS) without
occupant restraint;
6. ROPS with occupant restraint 0, refer to Figure
2.

193
 Shane Richardson, Andreas Sandvik, Chris Jones, Tia Orton, Nikola Josevski, Wei Pei (Tandy) Pok and Thomas Emmett
Figure 1: A Quadbar 0 CPD.
Figure 2: ROPS with occupant restraint 0 (seatback and
seatbelt).
1
Lower et al. 0 identified that /3 of Australian quad
bike fatalities involve head injury as the primary
cause. The quad bike industry does actively promote
the wearing of helmets, which will most likely
mitigate the number of head injuries.
The quad bike industry appears to have adopted
ROPS with occupant restraints for side by side’s (
Figure 3) 0. However the quad bike industry appears
opposed to and advertises against CPD’s or ROPS
for quad bikes 0 (refer to Figure 4).
Figure 3: Side by side with ROPS and occupant restraints
(3 point seatbelt).
194
Figure 4: The Australian Federal Chamber of Automotive
Industries advertisement for BRP (Can-am), Honda, Ka-
wasaki, Kymco, Polaris, Suzuki, and Yamaha 0.
This paper is based previous work published by Ri-
chardson et al. 0 and extends the number of quad bi-
ke rollover simualtions from 400 to 1600.
Technical Engineering Group TEG
The Heads of Workplace Health and Safety (HWSA)
Trans-Tasman Working-Party on ‘Quad Bike Safety’
tasked a Technical Engineering Group (TEG) 0 to
identify if it was: “…reasonably practicable for a
device to be developed that when fitted to Quad Bike
reduces the potential for death and/or serious injury
caused by entrapment beneath an overturned
vehicle?”.
Lower et al. 0 from the Australian Centre for
Agricultural Health and Safety presented to the TEG,
data on quad bike rollovers in Australia that:
1. “Of the 127 quad bike fatalities, eight cases could
not be classified as either a rollover or non-
rollover incident. Of the remaining 119 cases, 56
(47%) were rollovers.”
2. “Rollovers accounted for 59% of on-farm and
18% of non-farm deaths.”
3. “Rollover deaths were primarily associated with
asphyxiation or respiratory difficulty (n=14),
head injury (n=11), chest (n=6) and spine
injuries (n=4). This compares with injuries from
non-rollovers where multiple injuries (n=13),
 Simulation and Analysis of Quad Bike Rollovers using PC-CRASH to Evaluate Alternative Safety Systems

head (n=10) and brain injuries (n=4)

predominate”.
4. “The data clearly indicate that the major risk for
use of quad bikes on Australian farms in
undertaking the current range of tasks, is death
due to rollover and being trapped and crushed
under the machine.”
The fundamental criterion posed by the HWSA to
the TEG was to prevent entrapment. Entrapment Figure 7: Quad bike with ROPS and unrestrained rider.
could result in crush injury to the either the torso, Figure 8 illustrates the multi-body quad bike with
neck, head or combination. Entrapment of the torso ROPS and restrained rider. The restraints are
could result in traumatic or mechanical asphyxia. provided via a seat back and two over shoulder seat
belt loops.
Quad Bikes and Rider
The mass of the multi-body rider is 77kg.
The rider multi-body was initially held in position by
50N or 100N tension only tethers between the: head
and seat; torso and seat; feet and footplate; and hands
and handlebar. The tension only tethers prevented
the multi-body rider from initially flopping forward.
Figure 5 illustrates the multi-body quad bike and Figure 8: Quad bike with ROPS and restrained rider.
unrestrained rider.
Simulation Senarios
A comparative evaluation matrix of 400 simulations
for the quad bike combinations (i.e. 1,600
simulations) were conducted at ramp angles of 20⁰ to
29⁰ (in 1⁰ intervals) and at lateral speeds of 6km/h to
15km/h (in 1km/h intervals) and longitudinal speeds
of 0km/h, 10km/h, 20km/h and 30km/h. The forces
on the torso, neck and head were evaluated. The raw
Figure 5: Quad bike and unrestrained rider. data from PC-Crash was filtered (refer to Figure 9).

Figure 6 illustrates the multi-body quad bike with

CPD and unrestrained rider.

Figure 6: Quad bike with CPD and unrestrained rider.
Figure 7 illustrates the multi-body quad bike with
ROPS and unrestrained rider. The width of the fra-
mes at the front and rear of the quad has been
increased by 200mm to have the frames engaged
with ground in a rollover.
Figure 9: Example of the raw and filtered data (quad bike
and unrestrained rider rollover at a ramp angle of 26⁰ and
lateral speed of 13km/h).
The torso impact force injury tolerance for a PC-
Crash multi-body is not known. The 77kg rider has
a torso mass of 22kg, hence a torso force of 1,500N
and 3,000N could equate to a torso acceleration of
6.9g and 13.9g. The peak torso impact force in the
first 1.5s was plotted for all the 400 PC-Crash simu-

195
 Shane Richardson, Andreas Sandvik, Chris Jones, Tia Orton, Nikola Josevski, Wei Pei (Tandy) Pok and Thomas Emmett

lations. (Note that torso impacts of 1,500N or

3,000N are unlikely to fatally injurious. The peak
torso impact force in the first 1.5s is used to compare
the different configurations.)
Zellner et al. 0 using a different simulation tool
(Articulated Total Body) utilises a compressive force
of 490N as potential for “…significant breathing
difficulty beyond 1 hour”. Zelllner et al. present that
490N: “…might be relevant to assessing potential
for hypothetical mechanical/traumatic (compressive)
asphyxia phenomena.” From the 1600 simulations Figure 11: Peak initial impact torso forces in the first 1.5s
conducted using PC-Crash it was possible for the for a range of simulated quad bike and unrestrained rider
rollovers. 41 of the 100 simulations exceed 1,500N im-
ejected rider to have a resting torso force greater than
pact torso force and 11 of the 41 simulations exceeded
490N without being beneath a rolled quad (i.e. the 3,000N.
body orientation on the ground can result in a resting
torso force of greater than 490N). Zellner et al. cites
Hopkins et al 0 and presents Figure 10.

Figure 12: Resting torso forces for a range of simulated

quad bike and unrestrained rider rollovers. 32 of the 100
simulations exceed 1,000N resting torso force.
Figure 10: Extracted from Zellner et al. p52, Figure 8:
“Estimate of Human Tolerance to Mechanical Asphyxia
With Fatal and Survival Data”. This is identified as being
sourced from Hopkins et al. The green line is at 1000N
and the orange line is at approximately 7 minutes.

Based on Figure 10 a quad bike rider could be

traumatically or mechanical asphyxiated if the
average resting force on the rider torso was greater
than 1,000N for greater than 7 minutes. The resting
average torso force, from 1.5s to 4.0s after the
simulation started, was plotted for all the 1,600 PC-
Figure 13: Peak initial impact torso forces in the first 1.5s
Crash simulations. for a range of simulated quad bike with CPD and
The ramp angle, speed, forces [peak (torso, neck and unrestrained rider rollovers. 2 of the 100 simulations
head) and resting torso)] are plotted for all the 1,600 exceed 1,500N impact torso force. Neither of the 2 simu-
simulated configurations. lations exceeded 3,000N.

Figure 11 to Figure 18 illustrate the 0km/h

longitudinal results.

196
 Simulation and Analysis of Quad Bike Rollovers using PC-CRASH to Evaluate Alternative Safety Systems

Figure 14: Resting torso forces for a range of simulated

quad bike with CPD and unrestrained rider rollovers. 17
of the 100 simulations exceed 1,000N resting torso force.

Figure 18: Resting torso forces for a range of simulated

quad bike with ROPS and restrained rider rollovers. None
of the 100 simulations exceed 1,000N resting torso force.

Figure 15: Peak initial torso forces for a range of Figure 19 to Figure 26 illustrate the 10km/h
simulated quad bike with ROPS and unrestrained rider longitudinal results.
rollovers. None of the 100 simulations exceed 1,500N
impact torso force. (Hence none of the 100 simulations
exceeded 3,000N impact torso force.)

Figure 19: Peak initial impact torso forces in the first 1.5s
for a range of simulated quad bike and unrestrained rider
rollovers. 1 of the 100 simulations exceeded 1,500N.

Figure 16: Resting torso forces for a range of simulated

quad bike with ROPS and unrestrained rider rollovers.
None of the 100 simulations exceed 1,000N impact torso
force

Figure 20: Resting torso forces for a range of simulated

quad bike and unrestrained rider rollovers. 13 of the 100
simulations exceed 1,000N resting torso force.

Figure 17: Peak initial torso forces for a range of

simulated quad bike with ROPS and restrained rider roll-
overs. None of the 100 simulations exceed 1,500N impact
torso force. (Hence none of the 100 simulations exceeded
3,000N impact torso force.)

197
 Shane Richardson, Andreas Sandvik, Chris Jones, Tia Orton, Nikola Josevski, Wei Pei (Tandy) Pok and Thomas Emmett

Figure 21: Peak initial impact torso forces in the first 1.5s
for a range of simulated quad bike with CPD and
unrestrained rider rollovers. None of the 100 simulations
exceed 1,500N impact torso force.

Figure 25: Peak initial torso forces for a range of

simulated quad bike with ROPS and restrained rider roll-
overs. None of the 100 simulations exceed 1,500N impact
torso force. (Hence none of the 100 simulations exceeded
3,000N impact torso force.)
Figure 22: Resting torso forces for a range of simulated
quad bike with CPD and unrestrained rider rollovers. 10
of the 100 simulations exceed 1,000N resting torso force.

Figure 26: Resting torso forces for a range of simulated

quad bike with ROPS and restrained rider rollovers. None
of the 100 simulations exceed 1,000N resting torso force.
Figure 23: Peak initial torso forces for a range of
simulated quad bike with ROPS and unrestrained rider Figure 27 to Figure 34 illustrate the 20km/h
rollovers. None of the 100 simulations exceed 1,500N longitudinal results.
impact torso force. (Hence none of the 100 simulations
exceeded 3,000N impact torso force.)

Figure 24: Resting torso forces for a range of simulated
quad bike with ROPS and unrestrained rider rollovers.
None of the 100 simulations exceed 1,000N resting torso
force
Figure 27: Peak initial impact torso forces in the first 1.5s
for a range of simulated quad bike and unrestrained rider
rollovers. 1 of the 100 simulations exceeded 1,500N im-
pact torso force and none of the 100 simulations exceeded
3,000N.

198
 Simulation and Analysis of Quad Bike Rollovers using PC-CRASH to Evaluate Alternative Safety Systems

Figure 28: Resting torso forces for a range of simulated

quad bike and unrestrained rider rollovers. None of the
100 simulations exceed 1,000N resting torso force.

Figure 32: Resting torso forces for a range of simulated

quad bike with ROPS and unrestrained rider rollovers.
None of the 100 simulations exceed 1,000N impact torso
force

Figure 29: Peak initial impact torso forces in the first 1.5s
for a range of simulated quad bike with CPD and
unrestrained rider rollovers. None of the 100 simulations
exceed 1,500N impact torso force. Neither of the 100 si-
mulations exceeded 3,000N.

Figure 33: Peak initial torso forces for a range of

simulated quad bike with ROPS and restrained rider roll-
overs. None of the 100 simulations exceed 1,500N impact
torso force. (Hence none of the 100 simulations exceeded
3,000N impact torso force.)

Figure 30: Resting torso forces for a range of simulated

quad bike with CPD and unrestrained rider rollovers. 1 of
the 100 simulations exceed 1,000N resting torso force.

Figure 34: Resting torso forces for a range of simulated

quad bike with ROPS and restrained rider rollovers. None
of the 100 simulations exceed 1,000N resting torso force.

Figure 35 toFigure 42 illustrate the 30km/h

longitudinal results.
Figure 31: Peak initial torso forces for a range of
simulated quad bike with ROPS and unrestrained rider
rollovers. None of the 100 simulations exceed 1,500N
impact torso force. (Hence none of the 100 simulations
exceeded 3,000N impact torso force.)

199
 Shane Richardson, Andreas Sandvik, Chris Jones, Tia Orton, Nikola Josevski, Wei Pei (Tandy) Pok and Thomas Emmett

Figure 35: Peak initial impact torso forces in the first 1.5s
for a range of simulated quad bike and unrestrained rider
rollovers. 30 of the 100 simulations exceed 1,500N im-
pact torso force and none of the 100 simulations exceeded
3,000N.

Figure 39: Peak initial torso forces for a range of

simulated quad bike with ROPS and unrestrained rider
rollovers. None of the 100 simulations exceed 1,500N
impact torso force. (Hence none of the 100 simulations
exceeded 3,000N impact torso force.)
Figure 36: Resting torso forces for a range of simulated
quad bike and unrestrained rider rollovers. 1 of the 100
simulations exceeded 1,000N resting torso force.

Figure 37: Peak initial impact torso forces in the first 1.5s
for a range of simulated quad bike with CPD and
unrestrained rider rollovers. None of the 100 simulations
exceed 1,500N impact torso force. (Hence none of the
100 simulations exceeded 3,000N impact torso force.)
Figure 40: Resting torso forces for a range of simulated
quad bike with ROPS and unrestrained rider rollovers.
None of the 100 simulations exceed 1,000N impact torso
force

Figure 38: Resting torso forces for a range of simulated

Figure 41: Peak initial torso forces for a range of
quad bike with CPD and unrestrained rider rollovers. No-
simulated quad bike with ROPS and restrained rider roll-
ne of the 100 simulations exceed 1,000N resting torso for-
overs. 2 of the 100 simulations exceed 1,500N impact
ce.
torso force. None of the 100 simulations exceeded
3,000N impact torso force.

200
 Simulation and Analysis of Quad Bike Rollovers using PC-CRASH to Evaluate Alternative Safety Systems
Figure 42: Resting torso forces for a range of simulated
quad bike with ROPS and restrained rider rollovers. 8 of
the 100 simulations exceed 1,000N resting torso force.
Simulation Scenarios Analysis
From the 1600 simulations for each of the four quad
bike configurations:
1. The rider’s torso was impacted with force greater
than 1,500N:
a. 48 times for the quad bike and unrestrained
rider.
b. 2 times for the quad bike with CPD and
unrestrained rider.
c. Zero times for the quad bike with ROPS and
unrestrained rider.
d. 2 times for the quad bike with ROPS and
restrained rider.
2. The rider’s torso was impacted with force greater
than 3,000N:
a. 11 times for the quad bike and unrestrained
rider.
b. Zero times for the quad bike with CPD and
unrestrained rider.
c. Zero times for the quad bike with ROPS and
unrestrained rider.
d. Zero times for the quad bike with ROPS and
restrained rider.
3. The rider could have been traumatically or
mechanically asphyxiated with a resting torso
force of greater than 1,000N if trapped beneath
the quad for more than 7 minutes:
a. 46 times for the quad bike and unrestrained
rider.
b. 28 times for the quad bike with CPD and
unrestrained rider.
c. Zero times for the quad bike with ROPS and
unrestrained rider.
d. 8 times for the quad bike with ROPS and
restrained rider.
Based on the 1,600 simulations (4 x 400 comparative
simulations) where there is an identifiable risk of
serious or fatal injury from quad rollover,
consideration should be given to fitting either: CPD,
ROPS or ROPS with rider restraint to mitigate the
potential for serious or fatal injury due to torso
impact or entrapment during a quad bike rollover.
Limitations
The use of PC-Crash as a quad bike rollover analysis
tool needs to be further validated with respect to:
1. Different sized quad bikes i.e. 150kg to 200kg
and 300kg to 450kg;
2. Quad bikes with ROPS and unrestrained and
restrained rider.
Future simulations should include:
1. Backfilps;
2. The effect of helmet use;
3. Torso impact injury consideration;
4. Restrain options;
5. A range of rider sizes and masses;
6. Improved shape of the quad bike multi-body.
Conclusions
Using the multi-body feature within PC-Crash can be
used to:
1. Simulate 250kg quad bike rollover crashes.
2. Comparatively evaluate the effectiveness of
alternative safety systems for quad bikes.
Based on the presented research where there is an
identifiable risk of serious or fatal injury from quad
rollover, consideration should be given to fitting
either: CPD, ROPS or ROPS with rider restraint; to
mitigate the potential for serious and/or fatal injury
due to torso impact or entrapment during a quad bike
rollover.
References
[1] Safe Work Australia, Public Discussion Paper, ‘Review of
Design and Engineering Controls for Improving Quad Bike
Safety’. http://www.safeworkaustralia.gov.au/sites/SWA/whs-
information/agriculture/quad-watch/Documents/Discussion-
paper-PC/Discussion-
Paper/Discussion_paper_quad_bike_safety.pdf
[2] Lower T., Herde E., and Fragar L., ‘Quad bike deaths in
Australia 2001 to 2010’, Journal of Health Safety and
Environment 28(1) 2012.
[3] http://ministers.deewr.gov.au/shorten/quad-bike-crush-
protection-devices-be-mandatory-federal-employers
[4] http://www.quadbar.com.au/ Quadbar is a commercially
available retrofit product which is sold as a Crush Protection
Device. “The Quadbar is a small unobtrusive, hairpin shaped
hoop mounted on the quadbike behind the rider designed to
counter some of the risks associated with rollovers. The
Quadbar acts as a Crush Protection Device (CPD) as opposed
to the more commonly known roll over protective structures
(ROPS). A typical ROPS would require a full cage and driver
201
 Shane Richardson, Andreas Sandvik, Chris Jones, Tia Orton, Nikola Josevski, Wei Pei (Tandy) Pok and Thomas Emmett

restraint, which are not feasible on a rider active vehicle such as

a quadbike.
The Quadbar (aluminium tube) is mounted on the tow bar and is
telescopically adjustable at the base. A support mount is
attached to the rear rack, where the bar passes through sliding
collars (bushes), which subsequently allow the suspension to
move freely. The use of the tow bar is not affected by attachment
of the Quadbar. The Quadbar is a device designed to reduce the
risk of injury caused by quadbike rollovers. Research has
identified injury caused by the pinning of the rider to be of
particular concern.”
[5] Richardson S., ‘Performance Criteria for Effective Structural
Rollover Protective Systems for Light Passenger Vehicles
Performance Criteria for Effective Structural Rollover
Protective Systems for Light Passenger Vehicles’, PhD Thesis,
Department of Civil Engineering, Monash University, 2009.
[6] Draft Standard in the United States of America Recreational
Off-Highway Vehicle Association, ANSI/ROHVA 1 – 201X
http://www.atvea.org/CobraManagedFiles/110628-
US_ANSI_ROHVA_1-
201X_Utility_draft_standard_Recirculation_Ballot_Draft_4_29_
11_1.pdf
[7] Australian Federal Chamber of Automotive Industries
http://www.atvsafety.com.au/
[8] Richardson S. Orton T, Sandvik A. Jones C. Josevski N. and
Pok W P., ‘Simulation of Quad Bike (ATV) Rollover Using PC-
Crash to Evaluate Alternative Safety Systems’, 13-0286, 23rd
Enhnaced Safety of Vehicles, Seoul, South Korea, May 27-30,
2013.
[9] Lower T., Fragar L. and Herde E., ‘Quad bike rollover
deaths in Australia (2001-09)’, Report prepared for the Trans-
Tasman Quad Bike (Engineering) Group, October 5-6, 2010.
[10] Anon. ‘Report of the Technical Engineering Group (TEG)
to the Heads of Workplace Health and Safety Authorities
(HWSA)’, Trans Tasman Working Party on “Quad Bike Safety”
Meeting, October 5-6, 2010, Australia.
[11] Zellner J. W., Kebschull S. A. and Van Auken R. M.,
‘Updated Injury Risk/Benefit Analysis of Quadbar Crush
Protection Device (CPD) for All-Terrain Vehicle (ATVS)’, DRI-
TR-12-06, updated 6 August 2012.
[12] Hopkins I. H. G., Pountney S. J., Hayes P., and Sheppard
M. A., ‘Crowd Pressure Monitoring’, In: ‘Engineering for
Crowd Safety’, Smith R. A. and Dickie J. F. (editors), Elsevier
Science Publishers B.V.,

Contact
Dr Shane Richardson
Delta-V Experts
377 St Georges Road
Fitzroy North 3068
Australia
e-mail: srichardson@dvexperts.net
tel: ++ 61 4 0901 1362

202
© 2013 by EVU
EVU 2013 - 24

Light reflective attributes of selected materials

Albert Bradáč, Jan Schejbal, Arnošt Kuře,
Jakub Motl, Michal Belák, Martin Bilík, Marek Semela
Institute of Forensic Engineering, Brno University of Technology
Jan Škoda, Stanislav Sumec
Faculty of Electrical Engineering and Communication
Brno University of Technology

Abstract
This article is a summary of the specific research project, conducted at the Institute of Forensic Engineering, Brno
University of Technology (IFE BUT) in order to verify properties of selected light reflective and fluorescent materials
and their properties compared with those obtained with conventional textile materials. The article summarises the results
of laboratory and exterior measurements of selected light properties on a series of samples, including reflective materials
used in the clothing industry, on warning garments and on various promotional items and at the same time including the
most used materials and colors for outerwear. Within the laboratory measurements, inter alia, measurements of the
diffuse and specular components of the reflected light, which is important for the visibility of these elements, were
carried out.
Following this measurement practical night measurements in the exterior were performed which showed that the results
do not always correspond to those obtained from the laboratory measurements. The measurements also revealed major
problems in the quality of some reflective components available on the market. The aim of the research is to demonstrate
a fundamental difference in the visibility of pedestrians in a casual dress under reduced visibility compared to clothing
equipped with reflective materials and to find the correlation between the laboratory results and the actual subjective
perception of these surfaces in real life. The result of the project, after its completion, should be recommendations for the
selection and placement of reflective components for pedestrians moving under poor visibility on the road, both in terms
of color and size, and in terms of location for the best visibility.

Introduction yellow, orange or green, designed primarily as vests

The current status of the use of reflective compo-
nents in the Czech Republic is very disappointing.
The reflective components are used only in very lim-
ited quantities, they are taken rather as a part of the
clothing from the manufacturer, than as a required
safety feature. Although some people have been de-
liberately demanding them, others consider them un-
necessary. The reflective materials, however, can
make pedestrians several times more visible in traffic
and thereby protect them against potential hazards.
In the Czech Republic, in the accidents outside town
limits, for example in 2010, there were killed 54 pe-
destrians killed, out of which 39 pedestrian in the ac-
cidents at night time (i.e. almost three quarters!!!) -
mostly on the 1st and 2nd class roads (27, resp. 8 pe-
destrians killed), but also on motorways (2 killed -
mostly drivers repairing vehicles, etc.).
Actual Status
Currently, there are two most widely used types of
materials which make pedestrians or cyclists more
visible. In addition to reflective materials there are
also so called fluorescent colors, which are bright
or jackets. These fluorescent colors create a good
contrast against the area where they are not common,
and they are therefore easy to see in daylight. How-
ever, at night and under a poor visibility they are
practically invisible. And in these situations, the
benefits of reflective materials, which work using the
principle of light reflection (thus returning the big-
gest proportion of the coming light back to the
source) are proved. The amount of reflected light is
dependent inter alia on the light intensity and the
type of light source. According to the current
knowledge a two-inch strip of reflective material is
sufficient to make the pedestrian substantially visi-
ble. With a high-quality reflective material the driv-
ers have the chance to detect pedestrians at a sub-
stantially greater distance ahead of their vehicles and
therefore have more time to adapt their driving to the
specific circumstances.
It is well known that various types of commonly
used colors of clothes are very difficult to see under
conditions of poor visibility and it is recommended
to use reflective elements to increasing the visibility
of pedestrians (cyclists). However, what is not being
considered is how the different types of reflective
materials are visible, i.e. what is their retro-
203
 Albert Bradáč, Jan Schejbal, Arnošt Kuře, Jakub Motl, Michal Belák, Martin Bilík, Marek Semela
reflectivity, and at what distance is the driver able to
register such designated material. Furthermore, the
difference between homologated and non-
homologated reflective elements has not been quanti-
fied.
Another problem of nowadays is the fact that in
Czech legislation there is no compulsory use of re-
flective materials for pedestrians and cyclists at
night, or in a case of reduced visibility, as is the case
in other countries (e.g. in the Slovak Republic). In
contrast to that, motorists are obliged to have a re-
flective vest in their cars for circumstances when
they have to move around their vehicle on the road.
However, the likelihood of unexpected movement of
the driver outside the vehicle is substantially lower
than the probability of moving pedestrians or cyclists
(often unlit) on an unlit road at night.
This research has been partially addressed by the Na-
tional Road Safety organization “BESIP” with which
a close cooperation was established within this pro-
ject. From the results of the organization BESIP we
can learn, for example, a sight distance of pedestri-
ans in different colours of clothing. However, for our
purposes resolution of the colour or colours of
clothes is not enough. The actual distance at which
the object can be seen also depends on the type of
paint and the surface material. Nevertheless, it is ob-
vious from the following picture that reflective mate-
rial is visible at night from a distance about 4 times
greater than white clothing and more than 10 times
greater distance than blue clothes. At a speed of 90
km/h the driver needs at least 25 m (1 second) to re-
alise the danger and react accordingly. After 1s he
starts to slow down, or initiates an action preventing
the accident.
Figure 1 – Pedestrian clothes visibility dependence
(www.ibesib.cz)
From the Fig 1, it is clear that the driver would not
even have time to react to pedestrians in blue or red
clothes before getting level with their position. The
source from which this pictures is taken, did not take
204
into consideration what colour the tested reflective
element was, if it had been approved or not, and
what the weather conditions were.
The issue of reflective elements
The reflective vest usually consists of two materials
which are knitwear and reflecting components. Each
of the materials fulfils a different function on the
vest, and therefore it is necessary to approach them
differently in their assessment. The knitted fabric is
mostly of a noticeable color, which is characterised
by fluorescence, i.e. an action in which the short-
wave radiation (UV, blue color, ...) transforms into
radiation with longer wavelengths (e.g., green or or-
ange colour), which adds to the reflected light creat-
ing an impression of a highly noticeable (reflective,
rich) colour. It should be noted, however, that for the
realisation of this phenomenon incident radiation
containing short wavelengths, which will be trans-
formed into the longer ones is needed.
100
80
60
40
20
0
400 450 500 550 600 650 700 750
 [nm]
Figure 2 – Daylight color spectrum CIE D65
This process can be seen clearly during daylight, be-
cause its spectrum is very rich in blue light compo-
nents (Fig. 2). In contrast, when illuminated by arti-
ficial light, such as the street lighting system imple-
mented in the Czech Republic which mainly utilises
high-pressure sodium lamps, this phenomenon is not
so visible (Fig. 3) due to the low proportion of blue
components. Therefore, for a reflective vest to serve
well as active protection for pedestrians on the road
at night as well, it must be equipped with reflective
elements that reflect as much light back in the direc-
tion of impact as possible, i.e. preferably in the eye
of the approaching driver. From the above it is clear
that the knitwear material plays a role mainly during
the day and the reflective elements at night, or in a
situation where the pedestrians are illuminated by the
headlights of oncoming vehicle. This also implies the
requirements for the individual vests elements.
 Light reflective attributes of selected materials
100
80
60
40
20
0
400 450 500 550 600 650 700
 [nm]
Figure 3 – Sodium lamp color spectrum
The knitwear should have, with respect to the disper-
sion of natural light, properties of a diffusion materi-
al (Fig. 4), i.e. material which will reflect light in all
directions regardless of the angle of its impact and
should be of a colour (see above) which ensures the
best possible contrast for the person with the sur-
rounding environment (i.e., a significant differentia-
tion from the vicinity).
Figure 4 – Mirror and diffuse reflection
The reflective elements which are placed on the vest
mostly in the form of stripes should , on the contrary,
serve as a mirror, i.e. an element which would not
disperse the light, but reflect it back toward the
source, or the driver of an approaching vehicle. Such
reflection is, in practice, developed mostly in two
ways, namely by a set of corner mirrors (a prism or
three planar mirrors arranged so that they form a
corner) or by a group of spherical mirrors mostly
made of microspheres.
Laboratory Measurements
The reflectivity measurements were performed on
the sample set, which contained the selected fabric,
reflective and fluorescent materials – the mostly
commonly available materials on the Czech market.
The set includes reflective vests (branded and un-
branded), reflective tapes and stickers, hats with re-
flective elements, reflective pendants, but also purely
textile samples without reflective or fluorescent
properties. From each material a sample was taken,
of a size 50x50 mm square, which was then glued to
a white cardboard (Fig. 5).
Figure 5 – Prepared samples for lab testing
Given that the diffuse or reflective materials need to
750 be evaluated for the degree of reflected light, it is
necessary to use appropriate measuring apparatus.
This measuring assembly should be able to measure
the diffuse and mirror component of materials. Those
requirements are met by the Spectrophotometer Kon-
ica Minolta CM-3600d we used, whose internal
structure is designed so that it is possible to measure
both required parameters. When measuring the spec-
ular reflection it is paramount to measure light re-
flected back to the source, so the source and the sen-
sor should be as close to each other as possible. The
above mentioned spectrophotometer can achieve this
by a set of optical components, by which means a
deviation from the normal 8° when measuring the
specular component can be achieved.
The machine does not evaluate integral reflectivity,
but the spectral reflectance (in the visible area) i.e.
the degree of reflected light at different wavelengths
from 360 nm to 740 nm with a step of 10 nm. The
advantage of the spectral reflectance to the integral
one is that it is independent of the incident light.
Example: If a plate reflecting only red light receives
blue light, there will not be any reflected energy and
all energy will be absorbed. Integral reflectance fac-
tor will be zero. If, contrary to this, the board re-
ceives red radiation, ideally everything will be re-
flected. The reflectance factor of the red light will
therefore be one hundred percent.
In this way we can evaluate whether this or that ma-
terial will reflect light well from halogen or xenon
headlights.
The measurement is performed by a comparative
method, with the test samples compared with a cali-
brated highly reflective white material, for which the
manufacturer of the measuring devices have accu-
rately measured spectral characteristics. Mutual
comparison of the tested material with this sample
can then determine its spectral reflectance.
An interesting fact arises when measuring materials
exhibiting fluorescence. In order to determine spec-
tral reflectance of the material within the above
range, it is necessary to illuminate the sample tested
by a broad-spectrum light source. Due to the pres-
ence of short wavelengths in the used light, fluores-
cent materials are transformed to light with longer
wavelengths and the material in certain areas of the
spectrum shows a reflectivity greater than 100%, i.e.
at certain wavelengths there is reflected, or emitted,
more light than that which had been originally re-
205
 Albert Bradáč, Jan Schejbal, Arnošt Kuře, Jakub Motl, Michal Belák, Martin Bilík, Marek Semela

ceived. However, as this process the actual meas- combination of yellow fluorescent substrate area and
urement depends on the spectrum of the incident reflective stripes. As seen from the graph in a partic-
light, it is necessary that all the samples that we want ular colour spectrum the fluorescent material over-
to compare, were illuminated by the same type of comes, with its properties, the reflective element.
light. This is achieved by manufacturer prescribed However, the reflective element is in more stable
calibration before each measurement. In this way, over the entire light spectrum.
relevant data can be obtained which can be used for
further processing.
In the case where we were measuring the spectral
characteristics without fluorescence, monochromatic
light source was used and the amount of reflected
light emitted on the same wavelength as the wave-
length of the source was tested.
If we measure these wavelengths, we can easily de-
tect the fluorescent property of the material. The Fig
6 shows the response to monochromatic radiation of
457 nm at which within the band from 470 nm to
700 nm there is reflected radiation caused by fluo-
rescence. Figure 10) illustrates the spectral wave-
forms of the same sample for cases where there is
considered a fluorescent component and when it is Figure 7 – Gerdhard vest – retro-reflectance
excluded.
Exterior Measurement
-3
x 10
1.2
Dopadající záření The external measurements were carried out in a re-
X: 457
Y: 0.00116 Odrazené záření
1 mote place without light pollution. It was a moonless
night, partly cloudy and raining immediately prior to
0.8 measurement.
For the measurements two cars were used as the light
W/sr•m 2•nm

0.6
0.4
0.2
0
400 450 500 550 600
 [nm]
Figure 6 – Monochrom. radiation 457 nm response
For comparison of samples and their evaluation, it is
appropriate the fluorescence property is included in
spectral characteristics, as it is a desirable condition.
In the laboratory measurements 154 reflective, fluo-
rescent and clothing materials of different colors and
compositions were measured. Data from the per-
formed laboratory measurements, which were carried
out in collaboration with the Faculty of Electrical
Engineering and Communication Technologies of
BUT in Brno, served as the source for the following
graphs of reflectance values of individual elements.
The following Fig 7 shows the dependence of the re-
flectance of the measured material (including the
specular component) on the wavelength, i.e. "behav-
iour" in the visible wavelength spectrum of the light
spectrum. The graph shows the measurement of the
vest material Gardhard. This is a jacket in the colour
206
source, one with halogen, the other with xenon head-
lights.
The actual measurement was carried out so that the
test dummies were lit from a distance of 80 m by
each light source (halogen, xenon), by both low and
high beams (80 m is an approximate distance to stop
650 700 750
the vehicle safely from a speed of 90 km/h incl. reac-
tion of the driver at 1.0sec.). Various fabrics of dif-
ferent colors were gradually installed on the dum-
mies and, using a special device Lumidisk luminance
maps were taken. This procedure was also applied to
the reflective elements and vests.
The luminance maps are the output from the external
measurements from which it is possible to determine
the amount of brightness that the element reflects
back. The individual situations were also photo-
graphed by a classic camera (Fig. 11, Fig 12).
Measured Data Assessment
For more a illustrative comparison of the parameters
of individual elements, the measured materials were
divided into three main categories, namely conven-
tional fabric, reflective vests, and reflective accesso-
ries. From each category six elements that best rep-
resented the group were selected. The resulting val-
 Light reflective attributes of selected materials

ues of these materials are summarised in the charts flective vests gathered from various stalls and mar-
from laboratory measurements (Fig. 13) and exterior ketplaces. These jackets show a great problems in
measuring (Fig. 14). From these graphs, especially the quality of the retro-reflective element of the vests
with some reflective elements, a huge difference be- - the exterior measurement almost zero, which is ev-
tween light from the standard halogen and xenon ident from Fig 12, the fifth vest from left.
headlights is evident. The results from laboratory
measurements clearly do not correspond with meas-
urements in the exterior setting, which confirms that
the values measured in the laboratory do not always
reflect the practical use.
The first category of ordinary fabric is represented
by samples of the six most commonly used basic
colours, i.e. white, yellow, red, green, blue and
black. The differences between colours are evident
both in the exterior measurements (Fig. 8) and in the
Figure 9 – High visible vests
laboratory measurements (Fig. 13), where in both
measurements, the white and yellow colour in par- The third category of reflective accessories achieved
ticular have several times higher values than the val- the highest values. The best values were given by
ues of other colours. In comparison with the reflec- the reflective yellow bracelets which showed a rela-
tive elements in the exterior the value of their bright- tively balanced and on average, highest values of all
ness is almost insignificant (Fig. 14). measured materials during all modes of measure-
ment.

Figure 10 – High visible accessories

Figure 8 – Classic textile colors - exterior
In the second category which included the reflective
vests, the greatest difference between laboratory and
exterior measurements can be seen, especially in re-

207
 Albert Bradáč, Jan Schejbal, Arnošt Kuře, Jakub Motl, Michal Belák, Martin Bilík, Marek Semela

Figure 11 – Classic textile (top – high beam, bottom - low beam)

Figure 12 – High visible vests (left – high beam, right – low beam)

208
 Light reflective attributes of selected materials

Figure 13 – Laboratory measurements – comparison of selected materials

Figure 14 – Exterior measurements – comparison of selected materials

ous stalls and marketplaces are not actually reflective

Conclusion
From all the 154 measured materials, 18 major mate-
rials were selected to illustrate the evaluation, the re-
sults from which are shown in Fig 13 and Fig 14. It
is apparent here that the laboratory measurements do
not always correspond to the results of measure-
ments of the brightness and thus the associated visi-
bility of various materials taken outside at night
with the given lighting. There were quite substantial
differences between the visibility of individual mate-
rials are, example reflective vests available in vari-
at all. Such a poor quality reflective vest cannot be
distinguished at a glance, and its user may thus be
unwittingly be exposed to risk, which may even ag-
gravated if the user is to move among users with
high-quality jackets, and subjectively "disappears"
for any driver. On the contrary, commercially avail-
able yellow reflective bracelets made of material
from the company 3M proved to be one of the best
and achieved excellent results in all test parameters.
Another part of the research will be dynamic meas-
urements in the exterior, when the already measured

209
 Albert Bradáč, Jan Schejbal, Arnošt Kuře, Jakub Motl, Michal Belák, Martin Bilík, Marek Semela

material properties will be compared with subjective

visibility of their users while driving.
The aim of the research is to demonstrate a funda- Special Greetings
mental difference in the visibility of casual clothing Klimatex a.s. (www.klimatex.cz)
of pedestrians under reduced visibility conditions BESIP (www.ibesip.cz)
when compared to clothing equipped with reflective
materials, and to find the correlation between labora- The research was conducted with the support of BUT
tory results and the actual subjective perception of specific research and the Center for Research and use
these surfaces in real life. renewable energy sources (European Regional De-
The result of the project, after its completion, should velopment Fund under Project No.
be to produce recommendations for the selection and CZ.1.05/2.1.00/01.0014)
placement of reflective elements on pedestrians mov-
ing in poor visibility conditions on the road, both in
terms of colour and size, and in terms of location, for
the best visibility.

Contact

Albert Bradáč, PhD, Marek Semela, PhD

Údolní 53
602 00, Brno, Czech republic
E-mails:
ing.bradac@usi.vutbr.cz
marek.semela@usi.vutbr.cz
tel.: +420 54114 6011

Stanislav Sumec, Ph.D., David Škoda, Ph.D.

Technická 2
61600, Brno, Czech republic
E-mails:
skoda@feec.vutbr.cz
sumec@feec.vutbr.cz
tel: +420 54114 2813

210
© 2013 by EVU
EVU 2013 - 25

Driver Response to Roadway Intrusion at Night

John M. Grindey, Christopher D. Armstrong, Stein E. Husher,
John C. Steiner, Michael S. Varat, KEVA Engineering, LLC
Camarillo, California, USA

Abstract
A dark object in the path of a vehicle at night is a challenge for a driver to detect and respond to. Many factors influence
the driver’s ability to perform this task. These include: roadway configuration, environmental conditions, driver
physiology and object characteristics. This study analyzes the relative distance between a driver discerning an object in
the roadway during nighttime conditions and the driver‘s response.

This study involves 25 un-alerted, but attentive drivers approaching a dark object in their path at speeds ranging from 63
to 113 kph (39 to 70 mph). The test vehicle was driven on a closed course with drivers participating in an unrelated night
driving exercise. During the exercise a dark object was placed in the path of the subject driver’s vehicle. The driver‘s
responses, response times and distances from the object were monitored and reported.

Introduction factors that influence the detection and avoidance of

Object detection and recognition are dependent on
many factors. The mnemonic CLAPS gives an
indication of some of the factors that influence object
detection. Contrast, Lighting, Adaption, Pattern and
Size all influence a driver’s ability to discern an
object in the roadway [11].
Detection of an object is influenced by the luminance
contrast. Weber’s Law states that the luminance
difference needed for a target to be detected is a
constant proportion of the background luminance
[2].
An often utilized threshold value for illuminance is
3.2 lux which is the level of ambient illumination at
the end of civil twilight [16].
The distance at which a driver may recognize an
object as a hazard, as opposed to seeing an object, is
the threshold discernability distance (TDD). TDD is
defined as “the distance at which 50% of subjects
will be able to identify the object” [8].
This study was developed after the investigation of
numerous crashes involving a darkened, disabled
vehicle in the roadway at night which had been
conducted. Some drivers fail to detect the object, or
detect it too late to execute a successful avoidance
maneuver. It was the intent to examine some of the
drivers encountering a dark object in their path.
The investigation of these types of crashes can be
difficult due to the complex human factors issues
involved. Without careful consideration and a good
understanding of the human factors issues involved,
an incorrect assessment of the cause of the crash is
possible.
Some of the factors that must be evaluated in order
to determine if the crash is avoidable include:
 Observability threshold (TDD)
 Available maneuvering distance (speed
dependent)
 Available avoidance paths
 Approach speed
The difference between the observability threshold
and the available maneuver distance is the time
available for perception and reaction. Determining
the speed of the approaching vehicle and the time
and distance required to avoid a hazard in the
roadway can be accomplished through the use of
standard accident reconstruction techniques. If the
observability threshold can be established, then the
available perception-response distance can be
determined.
Investigators at a collision scene or at a nighttime
reenactment can estimate the threshold discernibility
211
 John M. Grindey, Christopher D. Armstrong, Stein E. Husher, John C. Steiner, Michael S. Varat
distance, or the distance at whch investigators could
first discern a darkened out exemplar vehicle in the
traffic lane. While this threshold discernibility dis-
tance has some value, it cannot be applied directly to
the collision being investigated because it does not
account for driver expectation.
A method for correcting for expectancy is discussed
by Olson [13]. This method involves reducing the
threshold discernibility distance by 50 percent to
correct for expectancy. The 50 percent correction
factor for expectancy is also discussed by Muttart
and Rossmer [10].
Methodology
The test subjects were 25 California Highway Patrol
(CHP) cadets at the CHP Academy. There was one
female test subject. The ages ranged from 23 to 38
years old. Background information collected for
each test subject included: driving experience
(years), vision, age and weight (appendix 1). All of
the test subjects had been employed by the CHP as
cadets for four months.
The testing was conducted in March of 2011. The
test facility was the CHP Academy Defensive
Driving Emergency Vehicle Operations Course
(EVOC) facility (appendix 2). The test subjects we-
re told that the purpose of the test was to evaluate a
driver’s behavior through a predefined maneuver un-
der nighttime conditions.
Figure 1: Driver’s position and data aquisition equipment,
GPS data logger circled
212
Figure 2: Rear investigator’s position
The test vehicle was a 2008 Ford Crown Victoria Po-
lice Interceptor. Data was obtained from the Power-
train Control Module (PCM) using a Bosch CDR
Tool. The test vehicle was also equipped with Mobi-
le Video/Audio Recording System (MVARS). Each
test run was recorded for later evaluation and analy-
sis. A Racelogic VBOX II GPS 20 Hertz data logger
was used to capture velocity and acceleration data
from each test. The GPS antennae was mounted on
the roof of the Ford above the center of gravity. The
data logger was located on the passengers side near
the center console (Figures 1 and 2).
The test vehicle‘s low beam type 9007 headlights
were mapped with a Leica TPS 1100 Total Station
Survey System and a Spectra Candela II Luminance
Meter, model C-2010EL (appendix 3). A 3.2 Lux
isolumen line was measured at a height of 0.6 meters
[11,16]. There was no artificial lighting near the
EVOC facility test location.
Maximum braking performance was evaluated
utilizing a Vericom VC3000 and VBOX II. The
average deceleration was 0.74 g‘s with the Antilock
Braking System (ABS) active (appendix 4).
During the testing an EVOC instructor was in the
front seat and an observer (investigator) was in the
rear seat to collect data. The EVOC instructor
provided detailed instructions to the test subject.
The instructions were unrelated to the true purpose
of the testing. Each test subject was also advised to
keep the test vehicle‘s headlights on low beam as
they traversed the course.
The test subjects were tasked with driving through
the EVOC course and negotiating a series of
predefined maneuvers. After traveling through the
designated course a minimum of two times, the
EVOC instructor advised the test subject to stop at a
pre-determined location to “collect data.” The test
 Driver Response to Roadway Intrusion at Night
subject was advised to “hustle” back to the EVOC
facility. This was to keep the test driver from
attempting to visually maintain a target speed. The
actual test speeds ranged between 63 to 113 kilome-
ters per hour (39 to 70 miles per hour) as the drivers
traversed the EVOC course.
As the test subject began the return trip to the EVOC
facility well concealed investigators stationed along-
side the testing route placed the object in the road-
way. The object, a 0.9 x 1.8 meter (3 x 6 feet), flat
black painted rectangular box was placed perpendic-
ular to the roadway (Figures 3 and 4). Two objects
were utilized blocking a portion of the available
roadway.
Figure. 3: The simulated object in roadway
Figure 4: View from MVARS video on initial approach
After the test subject came to a stop the scene was
documented and data was obtained. The data
included: the distance from the test vehicle‘s positi-
on at rest to the object, the PCM data and the VBOX
data. At the completion of all testing, the MVARS
video recordings were downloaded to a compact
disc.
After the test data was taken, the test subject was
tasked with driving back through the course and ma-
king a slow approach toward the object to identify
the TDD. The TDD was measured from the front of
the test vehicle to the object using a Lasar Technolo-
gy UltraLyte 200 LIDAR measuring device. The
reported TDD in this study is a single measurement
for each individual test subject.
Discussion
Of the 25 test subjects that completed testing, 15
tests were considered viable and included a complete
data set.
Five test were not included in the data set due to in-
consistent test subject determination of the TDD.
Their determined TDD’s were significantly shorter
than the test subjects actual braking response dis-
tance. This incorrect TDD assessment merits further
study.
The reported 50th percentile TDD for all test sub-
jects was measured at 67 meters (220 feet).
Each test subject was in active training with an in-
structor and observer during the test. This configura-
tion creates a more attentive test subject, which may
not be directly comparable to a driver in a normal
traffic situation.
Test protocol was not adhered to for four test sub-
jects. Inconsistent data occurred for six test runs.
Therefore this data was not included in the reported
data set.
The speed of the test runs ranged from 63 to 113 kil-
ometers per hour (39 to 70 miles per hour) as they
approached the object and began braking. The aver-
age speed at the beginning of braking for the test
runs was 84 kilometers per hour (52 miles per hour).
The evasive maneuver of each of the test subjects
was evaluated using MVARS. It was noted that
about one third of the test subjects made some audi-
ble sound when they first recognized the object in
their path before initiating their evasive maneuver.
Eight test subjects came to a stop prior to the object.
One test subject steered to the left and four test sub-
213
 John M. Grindey, Christopher D. Armstrong, Stein E. Husher, John C. Steiner, Michael S. Varat
jects steered to the right around the object. Two test
subjects struck the object. Of the test subjects that
steered, each had applied the brakes of the test vehi-
cle prior to the steering motion.
The total distance from the beginning of braking to
the end of the test vehicles motion for each subject
averaged 61 meters (200 feet) with a range from 110
to 350 feet. The resulting average was 49 meters
(161 feet) of braking prior to the object with a range
of 110 to 205 feet. The average deceleration during
their evasive maneuvers was 0.50 g‘s with approxi-
mately half of the test runs activating the ABS of the
test vehicle.
The average TDD for this study was 71 meters (234
feet) from the object, with a range of 170 to 287 feet.
This equated to an average time of 1.0 seconds from
TDD to brake application.
Conclusions
 Some test subject incorrectly identified the
TDD
 About half of the test subject did not utilize
the maximum braking performance
capabilities of the test vehicle
 The attentive test subjects may perform
differently than real world drivers
 The size of the data set is insufficient to
generalize to the general population
 The TDD is greater that the maximum for-
ward projection of the 3.2 Lux isolumen line
 This is a complex driving task and merits
further study
References
[1] Badger, J. E., “Human factors affecting perception.”
Law and Order Magazine. (1996).
[2] Boyce, P. R., “Lighting for Driving – Roads, Vehicles,
Signs, and Signals“. Boca Raton, FL: CRC Press – Tay-
lor & Francis Group. (2004).
[3] Green, J.M., Forensic engineering analysis of total
reaction time. National Academy of Forensic Engineers,
Volume 20, #1. (2003).
[4] Green, M. “Driver Reaction Time.” Visual Expert
Human Factors: Driver Reaction Time. Web. (2011).
[5] Ising, K. W.; Fricker, T. R. C.; Lawrence, J. M.;
Seigmund, G. P., “Threshold visibility levels for the Ad-
rian Visibility Model under nighttime driving conditions“.
(SAE Paper No. 2003-01-0294) Warrendale, PA: Society
of Automotive Engineers. (2003).
214
[6] McGehee, D. V.; Mazzae, E.N.; Baldwin, G.H.S., Dri-
ver reaction time in crash avoidance research: Validation
of a driving simulator study on a test track. IN, 320-323.
Proceeding of the IEA 2000/HFES 2000 Congress.
(2000).
[7] Muttart, J. W., Development and evaluation of driver
response time predictors based upon meta analysis. (SAE
Paper No. 2003-01-0885) Warrendale, PA: Society of Au-
tomotive Engineers. (2003).
[8] Muttart, J.W., “Estimating Driver Response Times.”
Handbook of Human Factors in Litigation. (Noy &
Karkowski Ed.) (Ch. 14). Boca Raton, FL: CRC Press
(Taylor & Francis) 14-1 – 14-25. (2004).
[9] Muttart, J. W., Quantifying driver response times ba-
sed upon research and real life data. Proceedings of the
3rd International Driving Symposium on Human Factors
in Driver Assessment, Training and Vehicle Design.
(2005).
[10] Muttart, J. W., Messerschmidt, W. F., Gillen, L.G.,
Relationship between relative velocity detection and dri-
ver response times in vehicle following situations. (SAE
Paper No. 2005-01-0427) Warrendale, PA: Society of Au-
tomotive Engineers. (2005).
[11] Muttart, J., Romoser, M., Evaluating driver response
and ability to avoid a crash at night, Proceedings of the 1st
joint ITAI-EVU conference. (2009).
[12] Okawa, M., “The development of visibility evaluati-
on method for automotive headlamps in motion“. (SAE
Paper No. 2003-01-0928) Warrendale, PA: Society of Au-
tomotive Engineers. (2003).
[13] Olson, P. L., “Visibility Problems in Nighttime
Driving“. (SAE Paper No. 870600) Warrendale, PA:
Society of Automotive Engineers. (1987)
[14] Olson, P. L., “Driver perception response time“.
(SAE Paper No. 890731) Warrendale, PA: Society of Au-
tomotive Engineers. (1989).
[15] Olson, P. L.; Farber, E., “Forensic Aspects of Driver
Perception and Response, second edition.” Tucson, AZ:
Lawyers and Judges Publishing Company, Inc. (2003).
[16] Owens, D. A.; Francis, E. L.;, Leibowitz, H. W.,
“Visibility distance with headlights: A functional app-
roach“. (SAE Paper No. 890684) Warrendale, PA: Socie-
ty of Automotive Engineers. (1989).
[17] “Geometric Design of Highways and Streets”, 5th
Ed., Washington, D.C.: American Association of State
Highway Transportation Officials. (2004)
Contact
Mr. John M. Grindey
KEVA Engineering, LLC
840 Calle Plano
 Driver Response to Roadway Intrusion at Night

Camarillo, California 93012

USA
E-mail:
jgrindey@kevaeng.com
tel: +1(805) 388-6016

Acknowledgements

The authors would like to thank Mr. Donald Karol,

Director of Highway Safety, National Transportation
Safety Board, for his assistance and direction in this
study. In addition, the authors would like to thank
the California Highway Patrol for their assistance
with the overall success of the testing.

215
 John M. Grindey, Christopher D. Armstrong, Stein E. Husher, John C. Steiner, Michael S. Varat

Appendix 1

216
 Driver Response to Roadway Intrusion at Night

Appendix 2

Maneuver Area

Run Start

Object

Aerial View – CHP Academy Defensive Driving Couse (VBOX GPS path data in red)

217
 John M. Grindey, Christopher D. Armstrong, Stein E. Husher, John C. Steiner, Michael S. Varat

Appendix 3

Headlight mapping – TDD and braking distances - Test subjects’ final position

218
Driver Response to Roadway Intrusion at Night

Appendix 4

Run 4 Test Data Graph

219
 John M. Grindey, Christopher D. Armstrong, Stein E. Husher, John C. Steiner, Michael S. Varat

Appendix 5

220
© 2013 by EVU
EVU 2013 - 26

Consideration of Human Factors in Road Accident

Investigation
Dr. Sibylle Birth
Intelligenz System Transfer GmbH – D
Prof. Lorenzo Domenichini
University of Firenze - IT

Abstract
Road Accident Investigations procedures can benefit to a great extent from the introduction of Human Factors (HF)
principles in the analysis activities. This is only possible if a cause-oriented method of investigation is used during the
on-site accident data acquisition.
The paper aims to discuss the benefits achievable from a HF oriented approach in the implementation of Road Accident
Investigation (RAI), and the criteria to be applied to allow performing such an analysis “on-the-spot” or in a post-
accident approach (“HF-Accident Profiling”).
The paper highlights the principles of HF as defined by the PIARC Human Factors Guideline “Human Factors for a
Safer Road Infrastructure” – 2007. Three HF criteria describe three possible classes of accident provoking road features:
1. the relationships between sight distances and driver reaction capabilities, 2. driver’s limitations of spatial perception
and 3. the characteristics of managing driver’s expectations. They are discussed from the point of view of the road
accident investigator. The “HF-Accident Profiling” of an accident spot allows the identification of accident-provoking
road features such as misleading stimuli that create confusion, misperception or wrong expectations and lead to delayed
or inaccurate driver reactions.
A set of practical examples taken from the analysis of specific high accident concentration points are presented and
discussed to underline the benefits of the HF-related analysis.

occurrence and possible infrastructure deficiencies

Introduction (figure 1) [4,5].
The identification of road deficiencies influencing At present, the Working Group “Accident
the occurrence of accidents is a necessary part of Investigation Guidelines” of PIARC TC3.2 “Design
increasing traffic safety on the existing road network. and Operation of Safer Road Infrastructures” is
Both Road Inspections (RI) and Road Accident working on updating the document “Road Accident
Investigations (RAI) are diagnostic activities Investigation Guidelines for Road Engineers” [2] in
implemented to identify the appropriate order to introduce the HF principles and logics to the
improvement measures to be applied on road investigation methodology.
sections with high concentration of accidents.
Since 2004, the PIARC Technical Committee for
Safer Roads Infrastructures has conducted
multidisciplinary studies to identify the
methodologies to be applied to perform both RI [1]
and RAI [2] in order to identify the most appropriate
safety countermeasures to be applied to enhance the
traffic safety of both newly constructed and existing
road infrastructures [3].
During the last 10 years special attention has been
paid to the topic concerning the influence of Human
Factors (HF) on road safety. The aim was to enhance
the knowledge of a driver oriented approach in
analysing the relationships existing between accident
Figure 1: The road safety influencing elements
(source: [4])

221
 Sybille Birth, Lorenzo Domenichini
The present paper gives an insight into the possible
benefits that the introduction of HF logic into RAI
can offer, especially in terms of a better
understanding of accident causation.
The Human Factors in Road Safety
Road standards have always been drafted with traffic
safety as the principal design criterion. High level
research has been, and is still being devoted to
examining factors such as vehicle stability and driver
perception with the aim of delivering a high quality
traffic system with a low risk of accidents.
Nevertheless accident statistics and the road safety
management procedures introduced by the EU
Directive 2008/96/CE [6] highlight that both along
the existing road network and, although to a less
extent, along newly constructed infrastructure, high
accident concentration sections are present. The
nature of road safety inspections and the accident
investigation activities performed in these locations
didn’t, in some cases, identify the deficiencies
causing the occurred accidents, and these were
therefore considered as technically “unexplainable”.
In Germany, over the past 10 years, about 1,500
technically “unexplainable” accidents
have been investigated accounting for the
contributions of psychological/physiological factors.
It was found that about 70% of these accidents were
caused by: misperception; the drivers’ ability to
perceive and react being exceeded; deficiencies in
optical orientation and guidance; and inconsistencies
between drivers’ expectations or habits and road
design and signage. The results were summarised in
the first Human Factors Guideline of Brandenburg,
2004 [7].
All these aspects fit in the subject concerning the
study of HF in driving activity, which has been
recognised to be deeply involved in accident
causation.
“Human Factors” has been a professional term since
around 1930. Those who are familiar with the jargon
of industrial safety engineers know that “Human
Factors” is a terminus technicus. In this case the term
means a number of psychological and
physiological threshold limit values that are
relevant in operating machines, cars and technical
facilities. The term Human Factors (HF) includes all
qualities that are generally found as
characteristical human limitations of perception,
thinking, language and memory in operating
technical facilities. Only general human qualities
that are typical and stable in all people,regardless of
race, culture or age etc., are included. Individual
qualities as well as temporary affectivity, illness,
222
medication or special motivations are excluded. For
instance drunkenness, over-ambition or heart failure
would not count as “Human Factors” because man
as such is not generally drunk, over-ambitious or
suffering from heart failure [4].
Since 1930 a great deal of scientific literature and
technical guidelines have allowed us to improve the
theoretical knowledge about the influence of the
physiological, neurological and cognitive limitations
of drivers. Much practical advice has been developed
for road designers and safety experts willing to take
this knowledge into account in their activities.
A recently proposed approach for an ergonomic road
design, taking into account the HF logic in the
planning or evaluation of the road alignment or cross
section, identified the main HF-related road
characteristics in the following road features:
1. The road should give users enough time -
Perceptibility (6-Second Rule): The road should
give the driver approaching a critical point
(junction, sharp bend, pedestrian crossing, bus
stop, bicycle path intersection etc.) enough time
to allow him to understand that the road situation
requires the changing of his driving program
(recognise the need to reduce the speed, to turn
right at the junction, to give priority to another
road user etc.). The average driver needs 4–6
seconds to change his driving program
completely. At a speed of 100 km/h this results in
a distance of up to 170 m.
This implies that the critical point has to be
visible and recognisable at the appropriate
distance: all potential sight barriers should be
removed and the critical point should be designed
so as to allow it to be correctly perceived.
The 6-sec rule is a different concept than the
“stopping sight distance” usually included in road
design standards, requiring a visibility of 1.5 –
2.5 sec. In fact, the stopping sight distance refers
to an emergency situation instinctively managed
by the drivers, while the 6-sec rule refers to a
cognitive process which the driver is required to
perform.
2. The road leads drivers to appropriate speed
and stabilises lane tracking (Field of Vision-
Rule): Monotonous, un-clear, deceptive or
distracting impressions affect the quality of
driving. The road, together with its surrounding
environment, offers an integrated field of vision.
Depending of driver’s field dependency this can
either stabilise or destabilise drivers; it can tire or
misguide them. Monotonous environments
decreases alertness and results in the
subconscious speeding up. The need to make a
 Consideration of Human Factors in Road Accident Investigation
decision can overload the driver’s information
processing capacity.
A good-quality field of vision keeps the driver
from drifting to the edge of the lane, roadside
facilities are parallel and symmetrical to road’s
edge to avoid optical illusions (e.g. safety
barriers, lighting fixtures, pillars of over-passing
structures etc.). It ensures that there is closed
optical framing of the outer curves of bends, with
no gaps present, to ensure that
the driver is not misinformed about the sharpness
of the curve.
3. Road design should pre-program the driver’s
actions correctly (Logic Rule): Drivers follow
the road with their expectation logic formed by
their recent perceptions of about the last 10
minutes. The road characteristics have to be
organised so as to respect this driving
motivational approach. Therefore the road layout
and its relationship with the characteristics of the
environment through which it passes has to be
coherent with the road functional class perceived
by the driver, the modulation of the road plano-
altimetric alignment and the smooth transition
between road sections of different characteristics
have to follow driver’s expectation logics.
The lane dimensions, its margins and the
peripheral elements are the key identification
elements perceived by the driver; unexpected
objects disturb the automated sequence of
operations.
Designers therefore should try to keep road
characteristics flowing in a logical sequence.
They should introduce inevitable changes as
clearly as possible.
Not following one of the above listed HF principles
can lead to, incorrect mostly subconscious reactions
by the driver, triggering operational errors which can
even evolve into an accident.
A recent international audit aimed at identifying the
extent to which the HF principles are included in
existing road standards showed that the 6-second and
the logic rules are, to a greater or lesser extent,
generally considered in the standards. However, the
subject of spatial perception of the road environment,
is generally not clearly covered.[5].
Damage oriented and prevention oriented
approach in accident analysis
Road accident investigation practices were examined
in depth by the ROSAT EC Working Group, and
four levels of accident investigations [8] were
identified:
 statistical data collection
 intermediate level investigations
 in-depth investigations (independent as well as
non-independent)
 special accident investigations
The ROSAT group acknowledged that all these
levels of investigation are important in making up a
national investigation system, but that in-depth
multidisciplinary investigations are required in
addition to the collection of statistics and
intermediate level data, in order to fully learn from
road accidents.
This means that the statistics are not enough; police
or other intermediate-level investigation cannot be
enough; in-depth independent technical
multidisciplinary investigation should instead be the
core ingredient of road traffic safety policies [9].
In nearly 80 years of accident analysis, a difference
can be seen between the damage-oriented approach
(e.g. Insurance Companies) and the prevention
approach (Human Factors professionals) (figure 2).
Due to the fact that the insurance business depends
on contracts and payments, insurance interests are
focused on the point of the crash. This is the point of
interest of the post-accident approach.
Contrary to this line, the prevention experts are
focused on finding the stimulus of a mistake that
starts the chain of action from driver’s mistake up to
the accident. This is the point of interest in the pre-
accident approach.
Crash point Point of interest of the post-accident,
(e.g.: crash with a tree) damage-oriented, approach
(Judiciary system, Insurances)
(analysis of the damages and
consequences of the crash)
Process
(e.g.: the vehicle is swerving)
Point of interest of the pre-accident,
Stimulus causing an road safety oriented, approach
operational mistake (Human Factors professionals)
(e.g.: optical illusion) (analysis of the road features stimulating
the accident)
Figure 2: Pre-accident and post-accident approach
(source: according to [4])
The prevention approach considers the operational
mistake as the first step of a chain of actions that
may lead to an accident (figure 3).
Operational mistakes are an unintended or missed
action. An operational mistake is the initial, yet
unintended action which may result in a driving
error. It is caused by a lack of information in the road
environment, or a misinterpretation of the
information by the driver. In most cases, the
operational mistake can be corrected instinctively.
Otherwise, an operational mistake, left unchecked,
can become a driving mistake.
223
 Sybille Birth, Lorenzo Domenichini
A driving mistake follows on from an uncorrected
operational mistake. E.g. the vehicle is in an unstable
situation. Often the driver is able to correct it by
steering, speeding up or braking. In this case, the
driving mistake has no consequences. Otherwise, the
mistake leads to an accident.
An accident is the consequence of an unsuccessful or
missed correction of a driving mistake.
Driving action
Intended Not intended result:
result OPERATIONAL MISTAKE
Not corrected
Corrected operational mistake:
DRIVING MISTAKE
Not corrected
Corrected operational mistake:
ACCIDENT
Figure 3: The chain of actions to a possible accident
(source: according to [4])
The prevention approach in accident investigation
should be considered if the ultimate goal of the
investigation is enhancing road safety. This type of
an approach is in agreement with the
recommendations of the European Commission 2003
paper “Saving 20 000 lives on our roads” [10] and
the Safety Oriented Road Accident Investigation
concept developed in the SAFETYNET [10]
research project.
A Safety Oriented Road Accident Investigation
acquires all relevant information to identify accident
causes and contributing factors, and helps in
identifying how the accidents injuries could be
prevented by adopting appropriate countermeasures.
Under this perspective, the investigation needs to
adopt an holistic view of accident analysis, paying
attention to aspects that a judicial enquiry might not
consider: physical, psychological, social, political,
economic or technical issues.
During accident investigations, the data collected
should allow investigators to understand as much
about the accident as possible.
There are many different data collection
methodologies current employed in Europe and as
yet there is little consensus about which is the best
method. However it is possible to identify best
practice in collecting data which corresponds to the
three components of an accident: road user, vehicle
and environment. This includes: general information
about the accident as a whole (e.g. date and time of
accident); information about the road environment
(e.g. carriageway information, weather conditions);
224
the vehicles (e.g. physical characteristics,
information on damage); and the road users involved
(e.g. behaviour, details of injuries).
The interest in harmonising the investigation
methodologies at an EU level has been considered
since the first years of ’90s and a great deal of
research activity has been performed since then
(figure 4).
Figure 4: the development of a Pan-European in-depth
Accident Data System (source: [12])
The development of a new Pan-European in-depth
accident investigation protocol was the scope of
WP2 of the DaCoTa research project and includes
[13]:
- accident scene and road examinations;
- vehicle examinations;
- vulnerable road user investigations;
- road user behavioural data;
- medical data collection and analysis;
- analysis of the information to:
 calculate speeds and trajectories;
 identify the cause of the accident;
 identify the cause of injuries.
An on-line manual for in-depth road accident
investigators has been developed and has been made
available through the DaCoTa project web site [14].
The manual is a very comprehensive and detailed
one. A lot of accident data is required and numerous
forms must be filled in. As far as the road variables
to collect, information concerning traffic, speed
limits, cross-section dimensions, horizontal and
vertical geometry, signing, restricted sight lines,
restraint system characteristics, vulnerable road users
infrastructure etc. is required. Scene and road
sketches and instructions on how to compile the
forms are included.
Looking at the general planning of the manual, it is
to be noted that, despite the great amount of data
required, not so much information is required to be
collected on the stretch of the road prior to the
 Consideration of Human Factors in Road Accident Investigation
accident locus, which is the main focus of the HF
analysis.
To answer the questions of how and why accidents
occur, the new methodology looking at the HF
influence must be addressed too. This is in line with
the DREAM (Driving Reliability and Error Analysis
Method) approach, characterised by a human-
technology-organisation perspective, which implies
that accident happens when the dynamic interaction
between people, technologies and organisations fails
in one way or another, and that there are a variety of
interacting causes creating the accidents [15].
Human Factors in road accident
Operational and driving mistakes represented in the
chain of actions of figure 3 are “incidents”, most of
them without serious effects. A few of them lead to
an accident.
As an order of magnitude, statistical analysis show a
ratio of 10/1 between operational mistakes and driv-
ing mistakes and a ratio of 25/1 between operational
mistakes and accidents.
In the application of the HF logic to road accidents,
the safety expert looks for the reasons causing a
driver’s operational mistake, which could lead to an
accident.
Many of the often observed operational errors result
from a direct, subconscious interaction between road
characteristics and road users’ threshold limit values
of perception. Since the human reaction is unswaya-
ble, attention should be focused on the road feature.
So, it is important to consider the laws of human
perception, the processing of information and the
regulation of action programmes, wherever an acci-
dent investigation activity is performed.
The limitations and the physiological, sensorimotor
and cognitive principles of spatial perception most
often can reliably explain accidents. They provide a
solid basis for developing effective remedies.
During the time period 2001 – 2006 in Brandenburg,
Germany, there were more than 1,400 accidents
analysed on the basis of these principles. The results
obtained, reported in Table 1, showed that about
68% of the analyzed accident have been likely pro-
voked by HF design mistakes in the road. These re-
sults are a strong indicator for the sustainable inte-
gration of the HF concept in road planning, construc-
tion and maintenance [4] and in the safety oriented
road accident investigations.
Should an HF problem be identified, expensive full
design and construction is often not required. In
many cases, it is sufficient to provide a gradual tran-
sition for the driver, or to improve optical guidance,
to manage the field of view or remove the inconsist-
encies between drivers’ expectations and de-
sign/signage. A new kind of optical guidance and
some innovative low-cost remedies proved to be sur-
prisingly effective as supports for drivers’ spatial
perception, anticipation and orientation [4].
Table 1: HF design mistakes likely leading to accidents
(source: [4])
In four human factors training courses for road plan-
ners, constructors and accident investigating special-
ists it was observed that road engineers encounter
varying difficulties in identifying the HF classes of
road mistakes possibly provoking road accidents.
The accident investigators and experts should there-
fore be trained to fully understand the HF principles
which can trigger an accident and to recognise HF
road mistakes when performing on site road inspec-
tions.
The training courses should present the HF princi-
ples and describe the survey methodologies to be ap-
plied, focusing attention on the road stretch before
the accident locus.
Once the location and the type of the accident is
known, the site survey of a section of road before the
accident location, having a length depending on the
road and vehicles speed, must be performed, looking
at the possible presence of driver-misleading stimuli.
Photos and videos are useful to record the situation
and allow a post-accident analysis.
A set of data sheets, including the main attention
points to be considered by the investigator, has been
developed and tested [4,5]. These can be used as a
kind of checklist for the “on-the-spot” investigation
of accident points, helping the judgment of accident
causations linked to HF principles.
Case stories
Some practical examples on how to perform a site
visit to acquire the information necessary to allow an
on-the-spot or a post-accident analysis of the acci-
dent road environment to understand if a Human
Factor-related problem exists will be presented.
The HF Accident Profiling is based on the “IST-
Checklist 2008”. This is a complex diagnostic
instrument for the prediction of the accident
probability of road sections. It contains some 100
reliable items with a correlation of 0.65 between
accidents and design mistakes regarding HF [16].
225
 Sybille Birth, Lorenzo Domenichini

signs without gaps to ensure continuous guiding at a

The methodology follows three steps: In the minimum 1m above the ground level.
first step the general characteristic of the road
section and the description of accidents is
analysed to classify the accidents into two
classes:
 HF related accidents (influenceable by
road design).
 Accidents that are caused by technical or
other conditions (e.g. weather, animals,
driver behaviours like alcohol use) and are
not influenceable by design.
The second step is an on-the-spot investigation.
After a driven inspection of the section conducted at
traffic speed to gain an experience of the road, a
Figure 5: Optical distance illusion created by a diverging
detailed inspection is conducted. This contains an
safety barrier non-parallel to road’s edge (red) at a
examination of all road and roadside characteristics, location with severe run of road accidents (source:
especially of the possible accident provoking road [17])
features every 50m (Storyboard-Method).
In the third step the documented accident Case 2
provoking road features are compared with In Brandenburg, Germany, , a sharp accident curve
the accident provoking road features of the with inconsistent radius has been investigated to
“IST-Checklist 2008”. clarify the possibilities of effective countermeasures
The important consequences from the point of view to avoid fatal accidents. The curve is located in a dip
of HF are pointed out so that a list of ranked without sufficient perceptibility from the approach-
measures completes the report. ing sections. The sharpness of the curve is not relia-
ble predictable for drivers. Several countermeasures
Case 1 of the accident authority were ineffective.
In the Netherlands the Rijkswaterstaat Dienst In a field experiment engineers and psychologists to-
Verkeer en Scheepvaart has 15 highway sections gether observed the driving behaviour in response to
with higher-than average numbers of single vehicle the height of an optical guiding frame in the outer
accidents. In four curves the accident investigators curve. Best results were achieved with an optical
could not find any technical problems or deviations guiding frame (2.75m above ground) with alternating
from the guidelines. For one of these curves we want yellow-orange colors of different length that work
to point out the results of the HF Accident Profiling. additionally like an optical break.
Out of 14 single vehicle accidents, 8 were related to
HF provoking road features of a distance illusion, 6
were related to technical problems of cars, icy
conditions and alcohol. The demands of the 6-
Seconds-Rule and the Logic-Axiom were
sufficiently fulfilled. But the structure of the field of
vision was accident provoking. The outer curve had
no optical framing. The lack of optical guidance
facilities led to an impression of an unlimited lateral
field of view. This causes destabilisation of the
driver and disturbances in lane tracking.
Additionally, the safety barrier in the inner curve
diverged from road’s edge and caused an optical
distance illusion - the apex of the inside bend seemed Figure 6: Optical guiding frame in a curve with
to be nearer than it really was. These features led to fatal accidents (source: [18])
irritations and wrong steering - especially under
twilightand nighttime conditions. The speed decreased by an average of 15km/h. The
It was recommended to dramatise the outer curve greatest decrease in speed was from 80km/h to
with bright safety barriers with retro reflecting 55km/h without the presence of any other signing.
delineation strips combined with supersized chevron

226
 Consideration of Human Factors in Road Accident Investigation
Figure 7: Speed before and after the construction of an
optical guiding frame in an accident curve
(source: [18])
Case 3
In Brandenburg, Germany, several city bypasses
have been built in the last ten years. After opening
the new city bypass fatal accidents were found to be
occurring, although the road design was in line with
the guidelines. Accidents frequently occur, when a
dominant eye-catching view axis guides the attention
in the wrong direction against the road’s course. Be-
cause tracking is largely performed subconsciously,
the probability of misinterpretation of the road’s
course is high. After the implementation of mandato-
ry optical guiding principles by planted embank-
ments and an improved design solution for the new
intersection, the accidents ceased.
Figure. 8: Wrong and corrected optical impression
corrected by an eye-catching planted
embankment (source: [16])
References
[1] PIARC C3, “Road Safety Inspection guideline for safety
checks of existing roads”, Cycle 2004-2007, 2012R27EN,
ISBN 978-2-84060-259-8.
[2] PIARC TC3.1 “Road Accident Investigation Guidelines
for Road Engineers”, Cycle 2004-2007, 2013R07EN, ISBN
978-2-84060-321-4.
[3] PIARC TC C3.1, “PIARC catalogue of design safety
problems and potential countermeasures”, 2009R07, ISBN
2-84060-227-X, 2009.
[4] Intelligenz System Transfer GmbH [Birth, S.] (2008).
“Human Factors Guideline for Safer Road Infrastructure”.
Report 2008R18 of Technical Committee C3.1 Road
Safety, Paris: PIARC. 2008R18, ISBN 2-84060-215-6,
2008.
[4] PIARC TC 1, “Human Factors in Road design. Review
of design standards in nine countries ”, 2012R36EN, ISBN
978-2-84060-301-1.
[5] EU Directive 2008/96/CE, “Road Infrastructure Safety
Management”, 19 nov. 2008.
[6] Birth, S., Sieber, G. & Staadt, H. (2004). “Straßenpla-
nung und Straßenbau mit Human Factors”. Ein Leitfaden.
Potsdam: Ministerium für Infrastruktur und Raumordnung.
Abt. 5, Straßenwesen, Straßenverkehr.and Regional Deve-
lopment, Brandenburg. Potsdam, 2004.
[7] RO-SAT (ROad Strategy for Accidents in Transport)
Working Group within the Group of Experts to advise the
Commission on a Strategy to Deal with Accidents in the
RO-SAT Working Group, “Road Accident Investigation in
the European Union - Review and Recommendations”,
Group of Experts to advise the Commission on a Strategy to
Deal with Accidents in the Transport Sector, May 11th,
2006.
[8] J. Monclus, L.G. Löwenadler, R. Maier, “Independent
In-depth Road Accident Investigation in the EU”, presenta-
tion at the SafetyNet Workshop, March 27th 2007, Brus-
sels.
[9] European Commission, “Saving 20 000 Lives on our
Roads - A shared responsibility”, COM(2003) 311 final,
2003, ISBN 92-894-5893-3.
[10] SafetyNet (2008), “Recommendations for Transparent
and Independent Road Accident Investigation”, Deliverable
D4.5 of the EU FP6 project SafetyNet.
[11] J. Hill, “The Pan-European In-depth Accident Investi-
gation Network”, DaCoTA Conference, Athens, 22-23 No-
vember 2012.
[12] J. Hill, “DaCoTa – Work Package 2: The Pan-
European In-depth Accident Investigation Network”,
presentation at TRA2012, Special Session 15: Development
of the European Road Safety Observatory – The DaCoTa
project, 25 April 2012.
[13] R.K. Elliman, H. Jähi, L. Persia, M. Jänsch, D. Otte, G.
Giustiniani, D. Usami, H. Fagerlind, K. Parkkari, L.K.
227
 Sybille Birth, Lorenzo Domenichini

Rackliff, A.P. Morris, G. Vallet, “Recommendations for es- [17] Birth, S., Demgensky, B. (2009). „HF-Unfallanalyse
tablishing Pan European Transparent and Independent B158 Abschnitt 130“. Bericht an die Unfallkommission
Road Accident Investigations”. Märkisch Oderland. Potsdam.

[14] M. Ljung, “DREAM- Driving Reliability and Error

Analysis Method”, Master’s thesis, Linköping University, Contact
2002.
Sibylle Birth, Dr.
Große Weinmeisterstraße 29a.
[15] Birth, S., Pflaumbaum, M. (2006). “Human Factors in
14469 Potsdam
Road Design: A way to self-explaining roads. Validation of
Email: sibylle.birth@ist-potsdam.de
the „IST-Checklist 2005“: Project Report of Work Package
3: Expert Assistance for Safety Review of Rural and Urban tel: + 49 331 280 39 79
Roads (single carriageway roads). In: Ranking for European
Road Safety (RANKERS) Research Project funded under
the 6th Framework Program of the European Community. Lorenzo Domenichini, Prof.
Via s. Marta, No. 3
50139, Firenze, Italy
[16] Birth, S. (2008). “Human Factors Accident Profiling Email: lorenzo.domenichini@unifi.it
of four accident bends in The Netherlands”. Project Report tel: +39 055 4796315
for Rijkswaterstaat Dienst Verkeer en Scheepvaart. Pots-
dam.

228
 © 2013 by EVU
EVU 2013 - 27

A review of speedometers and the criteria to be

considered before accepting ‘frozen’ readings and other
marks
Christopher Goddard (BSc IEng MIET)
David Price (B.Tech MITAI)

Abstract
Various mechanisms have been used to drive speedometers and other instrument gauges. This paper reviews the
mechanisms used; in particular stepper motors which have become the most common type in the last decade or so.
Stepper motors require power to drive the needle to any indicated position, including to return it to zero. Hence if power
to the instrument is lost as a result of a collision there is no power to move the needle and it should be left at the reading
shown at the moment the power was lost. However, not all stepper motor instruments are the same and before accepting
the reading a number of criteria need to be considered to give a level of confidence in the result.
Keywords
Collision Investigation, Stepper Motor, Speedometer

Introduction The instruments fall into two main groups:

Speedometers have been essential instruments since
cars started to travel above walking-speed. As well
as having an instrument to inform the driver of their
speed, in some countries early cars were required to
have a large external speedometer to inform other
road users (and most notably the police) of the speed
of the car. Not surprisingly, those external
speedometers were soon abandoned!
In the vast majority of instruments since introduced,
the speedometer and other main instruments have
been of the analogue type, having a circular dial with
a centre-mounted needle (or ‘pointer’). There have
been many types of mechanism used to drive the
needle but the greatest change has been in the last
decade during which there has been an almost uni-
versal adoption of instruments driven by stepper mo-
tors.
The types of instruments that are likely to be
encountered are described in this paper and the
likelihood that residual (‘frozen’) readings are
representative of the pre-impact reading is
considered.
Types of instrument
There have been many types of mechanisms used to
drive speedometers as well as those of other
analogue instruments, such as rev. counters (tacho-
meters), water temperature and fuel gauges.
mechanically-driven types, and electronic/electrical
types.
i. Early speedometers were mechanically-
driven. Initially, complex chronometric
(clock-type) movements were used.
However, for many years up until the 1990s,
an eddy-current type of instrument was used,
based on a design by Otto Schulze dating
from 1905. Many instruments of that type
are still in use today. They are mechanically
driven via a cable that is turned by the
transmission, or by the front wheel in the ca-
se of many motorcycles. At the other end of
the cable, within the instrument, is a magnet
(generally housed within a steel cup). The
magnet rotates within an aluminium
(‘reactor’) cup which is attached to the
needle’s spindle. The eddy-currents created
by that rotation cause the aluminium cup to
rotate against the resistance of a hair-spring.
The faster the magnet rotates, the greater the
rotational force applied to the cup and hence
the reading increases.

229
 Christopher Goddard, David Price
Figure 1 Eddy-current mechanical speedometer,
showing the magnetic coupling (arrowed)
ii. In more recent times, electronic/electric
powered instruments have been used instead
of mechanically-driven units. In the transiti-
on period, some manufacturers used cross-
coil air-core instruments. In that type, sen-
sors in the transmission provide information
to the instrument which in turn causes
current to flow through fixed electrical coils
that surround the central rotor, creating
magnetic flux. The central rotor shaft is
fitted with a small magnet which reacts to
the flux and turns, against the resistance of a
hair-spring. Cross-coil air-core instruments
can easily be distinguished from stepper mo-
tors by their partly exposed copper coils that
surround the central rotor, there are two coils
which cross over each other. Stepper motors
normally have their coils situated to one side
of the rotor.
Figure 2 Cross coil instrument,
Cross-coil instruments were used by some
vehicle manufacturers around the 1990s but
have since been superseded by instruments
powered by miniature stepper motors. In
fact many manufacturers by-passed the use
of cross-coil instruments and changed
directly from eddy-current to stepper motor
instruments around the 1990s, at least for the
speedometer and rev. counter. The design of
stepper motors is considered in detail below.
230
Stepper motors
A stepper motor is an electromechanical device
which converts electrical current into discrete
mechanical movements. It is a permanently powered
device that constantly corrects the position of the
needle when in use. The needle is powered to its rest
position when the ignition is switched off and is held
there, hence the power is maintained to the instru-
ment to allow the instruments to return to the zero
position even after the ignition is turned off
(albeit some instrument panels remain permanently
‘live’ whereas others are programmed to be switched
off automatically after several seconds).
Stepper motors comprise a shaft bearing a permanent
magnet, called the rotor, surrounded by a number of
electromagnets on the stationary portion, called the
stator. The electromagnets are energized by an
external control circuit.
The output shaft of a stepper motor rotates in
discrete step increments when electrical command
pulses are applied to it in the proper sequence, unlike
an AC or DC motor which rotates continuously.
This makes the motor suitable as a digitally
controlled device. An integer number of steps makes
a full rotation. The number of full steps depends on
the number of external coils and the number of
magnetic poles within the rotor. To make the motor
shaft turn, first one electromagnet is given power,
which makes the rotor’s magnet align to the stator’s
magnetic field. Further motion is created when the
next electromagnet is turned on and the first is turned
off, at which time the rotor rotates to align with the
next magnet, and from there the process is repeated
(Fig 3). In this way, the motor can be turned by a
precise angle, although only to an angle that
corresponds to one or more steps. In the example
shown in Fig 3 there are 4 steps to one full rotation.
This disadvantage is overcome by the use of ‘micro-
steps’ which will be discussed later in this paper.
Figure 3 Diagram showing how a stepper motor rotates1.
A common design of stepper motor found in automo-
tive instrument clusters uses a rotor with 5 magnetic
pole pairs and 4 electromagnetic stator poles giving
20 incremental steps for a full rotation. This results
in a single step angle of 18° of the motor shaft. The
4 poles are created by using only 2 electromagnetic
coils. If the current is reversed in either coil, then
the polarity of the magnetic field is also reversed.
While the principle of all stepper motors is the same,
the design of the linkage to the instrument needle
 A review of speedometers and the criteria to be considered before accepting ‘frozen’ readings and other marks

varies considerably. The output shaft from the motor

may form the spindle on which the needle is fitted,
or the output may pass through a reduction gearbox.

The design of the gearbox or the absence of it has a

significant impact on the ‘static torque’ of the
needle. The ‘static torque’ is the torque required to
move the needle when the instrument is un-powered,
i.e. as we would expect to find in a post collision
vehicle. However, when power is connected to the
motor and the electromagnets are energised the
torque is much higher, because the magnetic flux
Figure 4 A spur-geared stepper motor
adds resistance to the movement. That torque is manufactured by Sonceboz (with black, blue and
known as the ‘holding torque’. For example, in a white gears), typical of group ‘B'
Sonceboz motor the holding torque is over 4 times ’
greater than the static torque.
c) Another type of stepper motor is currently
The common methods of linking the motor to the used for a number of vehicles, currently
needle are described below. mainly within the Volkswagen/Audi Group.
It utilises a motor and gearbox manufactured
a) A number of vehicles have non-geared step- by Continental Automotive (formerly Sie-
per motors. In these the needle is mounted mens VDO). This arrangement has a worm
directly on the rotor shaft. Such instruments gear connecting the rotor shaft to the needle
are manufactured by NMB/Minebea (Thai- (Fig 5). Employing a worm gear results in a
land) and are used in Magneti-Marelli in- high static torque and so provides a high
strument clusters. Since there is no interme- degree of resistance to free movement of the
diate gearbox, the needle has little resistance needle when the instrument is not powered.
to movement in the event of external forces
being applied when there is no power to the
instrument to hold the reading (i.e. such in-
struments have low ‘static torque’).

b) Currently, by far the most common method

of linking the motor to the needle is to
incorporate a small gearbox within the body
of the motor (Fig 3) that has a series of spur
gears and typically provides a reduction of
around 36:1. Coupled with a 20-step motor,
gearing of that ratio produces a 0.5 degree
step angle resolution of the needle.
Compared to the non-geared type, there is
greater resistance to free movement of the Figure 5 A Continental (Siemens VDO) motor driving a
needle when not powered (see later). worm gear
Examples of this type are made by
Sonceboz, Switec, Fraen and JST (Juken). A few instrument panels, such as those of various
models of Mercedes, do not have a central needle.
Instead the pointer is part of a large ring that
surrounds the dial face. These rings are driven by
stepper motors fitted with a spur gear on the spindle.
This engages with gear teeth on the outer edge of the
ring.

231
 Christopher Goddard, David Price
Figure 6 Ring-style pointer (motor position arrowed)
As indicated earlier, irrespective of the type of mo-
tor, the single step angles are still too coarse for
smooth needle movement and much finer changes of
angle are required in practice; these finer steps are
termed ‘micro-steps’. They are achieved by opera-
ting both electromagnets at the same time. By
altering the strength of each of the magnetic fields it
is possible to produce a combined field that is mid
way between a full step; for example the Siemens
SM2 motor uses 128 micro steps between each full
step. The introduction of the micro-steps is not just
for improved accuracy but to produce smooth mo-
vement of the needle throughout its range.
The needle is generally mounted directly on the out-
put shaft of the gearbox. The needle is a tight press-
fit onto the spindle with a slight taper within the
needle’s hole which provides grip. Some have a he-
xagonal-shaped hole to provide extra grip.
Larger needles may have balance weights fitted to
the base of the needle (Fig 7). This has the effect of
also reducing the rotational torque during
deceleration. Such counterweights were commonly
used for mechanically-driven instruments but are less
common for stepper motor instruments.
Figure 7 A pointer with a counter weight fitted (arrowed)
Although normally a secure fit, the possibility of
slippage of a needle on its spindle under impact
conditions must be considered. Owing to the zero
rotational force encountered under frontal impacts,
232
and low rotational force encountered under near
frontal impacts, slippage in these instances is highly
improbable. However, slippage under violent side
impact conditions is more likely. It cannot be
detected on eddy-current or cross-coil instruments,
as these incorporate a hair-spring and end-stop sys-
tem to return the needle to its rest position, but
slippage can be detected on stepper motor instru-
ments. To do so requires the instrument cluster to be
powered up, which results in the needle motor
moving from its end stop or rest position to the zero
position.
This can be seen on many vehicles when the ignition
is turned on and the speedometer needle can be seen
to move from its rest position to the zero.. If the
needle had slipped on the shaft it would no longer
align with the zero position when powered.
Obviously this would alter the post-collision
condition of the instrument and so the benefit of
doing this should be balanced against the need to
preserve evidence.
Sources of electrical power
Electrical power to an instrument cluster is normally
provided by the alternator or battery. Power is
required even in the event of momentary loss, for
example in the event of a poor contact that is
disturbed when the vehicle receives a severe jolt. In
order to allow for this type of power loss, the instru-
ment panel PCB (Printed Circuit Board) has a num-
ber of capacitors fitted. These devices are typically
of the order of 500 µF and are intended to provide
power for no more than 20 – 30 milliseconds2. A
typical crash impulse during an NCAP test is
between 40 – 70 ms 6. It is not known where during a
crash event the power is lost. This is dependent on
amongst other factors the model of vehicle.
If there is a change of speed while power is lost
beyond the 20 – 30 milliseconds then the instrument
will not adjust for the change of speed during the
loss of power. Conversely if the input signal is lost
and power is maintained, then some instruments are
programmed to hold their position for a few seconds
and then return to zero.
The reliability of impact marks and ‘frozen’
reading of instruments
A technique that was used in the examination of
crashed aircraft for many years (and probably still is
for some small aircraft) was to examine the dial face
of the instrument for a ‘witness mark’. A similar
technique can be used for speedometers. Old-style
 A review of speedometers and the criteria to be considered before accepting ‘frozen’ readings and other marks
folded aluminium needles were flexible and could
easily strike the dial face, even in road traffic
crashes. However, moulded plastic needles used
nowadays are much more rigid and, in our experien-
ce, marks from these are only rarely encountered.
Figure 7 Witness mark made by the needle (arrowed)
For many years there has been the question of
whether ‘frozen’ readings are a reliable indication of
the pre-impact value. Some research on this matter
was carried out on eddy-current instruments by MI-
RA (Motor Industry Research Association) on behalf
of the Home Office Forensic Science Service in the
1980s; one of the authors of this paper was involved
in this research. Unfortunately this work was never
published but the findings supported the view of
many workers in this field i.e. speedometer readings
frequently could not to be relied upon. The problem
stemmed from distortion and contact between the
parts of the magnetic coupling, which could cause
the needle to be moved by direct rotational contact
between the parts of the magnetic coupling, rather
than only be eddy-currents. Equally damage,
distortion, or debris trapped elsewhere were able to
affect the reading significantly.
In recent years, instances of ‘frozen’ readings have
become much more frequent. Most of these are now
associated with instruments driven by stepper mo-
tors. It is known that the needle in this type of in-
strument should not move when all power is lost.
The forces encountered in a collision that has
resulted in the wiring loom or the batteries being
damaged are very severe and so could potentially
cause the needle to move.
In an attempt to put ‘frozen’ readings into a real-
world context, a study was undertaken by one of the
authors whilst employed by TRL. The study was
very limited and involved cars that had been crashed
at known speeds (as a result of other types of re-
search studies that had been carried out).
Most vehicles in this study did not have ‘frozen’
speedometer readings, either as a result of the power
not having been lost in the impact or no power
having been present during the test. None had their
engines running at the time of the crash and so did
not display any ‘frozen’ rev. counter reading.
However, seven cars were found to have residual
readings. Of these, two were off-scale but were of
the mechanically-driven (eddy-current) type. The
others appeared to be of the stepper-motor driven ty-
pe, but it was not permitted to completely dismantle
the units to investigate further. It was noted,
however, that the readings of all the electronic units
with ‘frozen’ readings were within the legally
permitted tolerance of the instrument (see later), ba-
sed on the known crash speed; this was reported via
a letter to ITAI ‘Contact’3. One further vehicle was
identified from subsequent crash tests and again its
reading was within the permitted tolerance.
In more recent times, other instances of ‘frozen’
readings have been encountered as part of ongoing
casework where validation of the speed was
available from other means (data recorders, tyre
marks, crush damage calculations, etc). Amongst
these, many instances have been encountered where
both the speedometer and the rev. counter have
‘frozen’ readings which enables self-validation to
occur. To do so, the make, specific model must be
identified, and the information about the gearing
must be obtained (from published vehicle tests, from
the manufacturer), or tests carried out on an identical
vehicle as part of the investigation. In most of those
instances, the ‘frozen’ speeds and rpm for the known
gear (or likely gear) have been corroborated. The
limitations are discussed below.
Factors affecting the reading
If the cover has been smashed, the needle(s) will
have been exposed and direct contact with the needle
is then possible, either in the crash or in subsequent
recovery, removal, or transit. The ease by which the
needle might have been moved should be considered
and evidence of rotational or other scuff marks on
the dial face should be looked for.
In the event of there being rotational scuff marks, or
scuff marks on the underside of the needle, the rea-
ding is not likely to be reliable.
233
 Christopher Goddard, David Price
Figure 8 Rotational scuffs (arrowed), made by the tip of
the needle
Figure 9 Worn tip of needle from Fig 8 (underside of
needle reflected using a scalpel blade inserted between the
dial face and needle
Similarly, if the needle or its spindle has damage, the
reading is not likely to be reliable. Such damage
could include displacement of the spindle and
distortion of the centre hole of the dial face.
The presence of pieces of debris trapped between the
dial face and the needle, or within the drive
mechanism, is likely to render the reading unreliable
because it may have either prevented movement of
the speedometer or as it became trapped it may have
caused movement of the needle.
Distortion of the magnetic coupling of an eddy-
current type of mechanically-driven speedometer or
rev. counter is likely to render the reading unreliable.
This is because there could have been alteration of
the eddy-currents or even direct mechanical coupling
of the two parts.
Where the instrument cluster is mounted
predominantly vertically, as in most cars or larger
vehicles, the force acting on the instrument in a fon-
tal collision is along the axis of the spindle, rather
than a lateral force that will try to deflect the needle
to one side or the other. If, however, the impact is to
the side of the vehicle, such as the vehicle sliding
broadside into a tree, there may be a force acting on
234
the needle that will try to change the reading. The
force will be greatest if the needle is vertical and
least if it is horizontal.
Impact tests using eddy-current instruments carried
out in the MIRA investigation, and separately, found
that there was almost no change in reading in a
violent frontal impact. Those instruments have a
very low static torque but normally have balanced
needle assemblies.
The un-geared stepper motor type of instrument has
such a low ‘static torque’ that the needle can move
even by tilting the panel, irrespective of it being
subjected to a violent impact with some lateral com-
ponent to the force. As identified by Anderson2 un-
geared stepper motors seem to be used more
commonly in USA than in Europe.
Without further work, it cannot be determined if the
needle of a stepper motor of the spur geared type is
likely to suffer movement as a result of a violent im-
pact. Based upon the cars involved in the TRL tests,
stepper motors with spur gears with a high gear-ratio
(e.g. around 40:1) are unlikely to be disturbed in a
longitudinal impact configuration, although no im-
pact tests have yet been carried out specifically to
verify this. In addition to the TRL tests, anecdotal
evidence from casework suggests that speedometer
readings can often be successfully corroborated with
rev counter readings. .
In addition to our investigations so far, it is hoped
that testing will be carried out using instruments of
this spur-gear type, mounted both laterally and
longitudinally, and subjected to a representative
collision envelope of force. It is envisaged that such
tests may resolve this issue and provide criteria for
investigators to accept or reject ‘frozen’ readings.
Stepper motors of the VDO Siemens worm-drive ty-
pe were tested by Kuranowski5. The VDO/Siemens
instruments have a high level of static torque as
identified previously. When mounted with the
needle aligned along the longitudinal and the lateral
directions relative to the impact, he found that the
needle did not move even when the vehicle was
subjected to high impulsive forces.
Speedometer accuracy
In the EU at least, speedometers have legal
tolerances defining their accuracy. Rev. counters,
and other instruments do not have such tolerance
requirements, other than that specified by the vehicle
manufacturer. The tolerance for speedometers is
specified in EEC Directive EC 75/443/EEC 75. The
permitted tolerance is not simple and is expressed as:
0≤V1-V2≤
V2 +4km/h
10
 A review of speedometers and the criteria to be considered before accepting ‘frozen’ readings and other marks
(where V1 is the indicated speed and V2 is the true
speed)
Expressed in a simpler form, the speedometer is not
permitted to under-read but may over-read by 10% +
4km/h (equivalent to 10% + 2.5mph).
The calibration of stepper motors is far more precise
than the mechanically driven types for which the
legislation was introduced. Most of the instruments
are quoted as achieving accuracy of 1.5% or better.
In addition to the inherent accuracy of the instru-
ment, the size of tyre and wheel fitted to the drive
axle can influence the accuracy, as can the degree of
wear of those tyres. Consequently, whenever
speedometer readings are being considered
evidentially, the size of tyre should be checked
against the manufacturer’s recommendation.
However, so long as the overall diameter of the tyre
is close to that of the manufacturer’s intended
fitment, a larger but squatter tyre may be acceptable.
As a rough guide, a car tyre worn to the legal limit
will typically affect the reading by about 2%.
We and the equipment manufacturers Visteon and
Continental consider that where both the
speedometer and rev. counter have ‘frozen’ readings,
and they can be matched to the known, or likely gear
ratio of the vehicle (using information obtained from
tests of identical model vehicles or obtained from
manufacturers), the readings can be considered to be
reliable. This is of course dependent on them being
of the same type of instrument, as is usually the case
nowadays.
Instruments that have a low static torque and are
fitted with a hair-spring to return the needle to its rest
position are likely to be of the eddy-current
mechanical type, a cross-coil type, or an un-geared
stepper motor type. For such instruments any
‘frozen’ reading is likely to have resulted from
damage to the instrument or from un-restrained late-
ral movement. For these instruments, the reading is
likely to be unreliable, although laboratory
examination can often help resolve this.
Un-geared stepper motors have a very low static
torque. In use the motor is driven and the ‘holding’
torque is considerably greater than the static torque
when no power is present (i.e. disruption of power
following severance of wiring or damage to the
battery). The reading can then change if the instru-
ment is tilted, hence it is likely to be unreliable.
These instruments can be identified by the spindle
being central and/or by the name NMB/Minebea on
the metal casing.
Results to determine the static torque of
various stepper motors
Testing of various motors has been carried out by the
authors to try to determine the static torque of geared
stepper motors. The method used was similar to that
used by Anderson 4. An instrument cluster was
positioned with a needle in a horizontal position, a
weight was added at a known distance from the cent-
re of the needle. The weight was gently increased
until it caused the needle to move. The torque was
then calculated from the final weight being applied at
a known distance with the following results.
Group A, currently only consists of the Continental
MW200 motor, which has a very high static torque.
Needle movement is actually the force required to
cause slippage of the needle on the shaft, rather than
the force required to rotate the motor.
Group B are motors that have a static torque of
around 1 mNm. Tests on those makes with gear
ratios of between 1:50 and 1:36 showed significant
static torque of around 1 mNm. In our experience,
those form a large percentage of the instruments
currently in use in the UK.
Group C consists of motors with a torque value be-
low 0.5 mNm. As these Instruments have lower
gearing ratios, they have lower static torque values
than those in the previous two groups, as might be
expected. Denso instrument fall into this category as
do the Switec units despite the latter having a high
gear ratio they exhibit very low static torque.
Group D motors have negligible resistance to mo-
vement of the needle. They were instruments
manufactured by Nippon Seiki and they showed little
resistance to movement. The low static torque
approached that of the un-geared instrument.
As can be seen, there is a huge variation in the static
torque between motor designs. This clearly helps
explain the inconsistencies in the confidence of
frozen speedometer readings and those highlighted
by Anderson as being inconsistent with the actual
crash speeds.
235
 Christopher Goddard, David Price

Motor Type Motor Gearbox ra- Needle step Typically Static Motor Ty-
Step angle tio angle found in Torque pe
mNm
Continental 90° 1:50 1.8° VW / Audi 22* A
MW2000
Sonceboz 6405 18° 1:36 0.5° Ford/SAAB 1.1 B
Fraen 18° 1:36 0.5° 1.1 B
FRSM 6405
Continental SM2 90° 1:43.2 2.08° VW 1.2 B
Sagem/Motometer 45º 1:75 0.6º Citroen/VW 1.1 B
Jaeger/Magneti 11.25º 1:16 0.7º Peugeot 1.44 B
Marelli 2.1*
Switec MS X15 / 60° 1:180 0.3° Jaguar, 0.37 C
JUKEN X27 Vauxhall
Denso 45° 1:24 1.8° Toyota 0.47 C
Nippon Seiki 180° 1:4 45° Land Rover 0.047 D
NMB PM20S-020 18º N/A 18º Fiat D

* Clockwise torque
The motors in the above table fell into four distinct groups. For reference, these groups have been designated with the let-
ters A to D

have be experienced in the collision. It is

Criteria for a ‘frozen’ reading to be
considered acceptable
Before a reading can be considered to be reliable the
conditions below should be considered as a precursor
to further investigation:
i. The instrument must be examined in detail
to determine the motor type. The motor type
should be either of type A or B stepper mo-
tors, as described above
ii. The vehicle should have experienced a
predominantly frontal impact
iii. There should be no evidence of multiple im-
pacts
iv. There should be no evidence of vehicle roll
over
v. There should be an obvious cause of power
loss identified and the loss of power should
have been instantaneous.
Our preliminary view is that:
i. ‘Frozen’ readings on the Continental Auto-
motive worm-drive instruments are likely to
be reliable, unless extremely high lateral
forces disturbing the needle on its spindle
may have been experienced in the collision.
ii. ‘Frozen’ readings on instruments with gear
ratios within the 1:50 to 1:36 range may well
be reliable unless high lateral forces may
hoped that impact testing will resolve this.
iii. ‘Frozen’ readings on instruments with gear
ratios below 1:24 may, or may not, be
reliable depending on the level of lateral for-
ce experienced in the collision. It is hoped
that impact testing will resolve this.
iv. ‘Frozen’ readings on un-geared stepper mo-
tors, or those of the low-ratio geared Nippon
Seiki type, are unlikely to be reliable if any
lateral forces have been experienced in the
collision.
Conclusions
Most speedometers up until the 1990s were
mechanically-driven, eddy-current instruments.
Previous investigations have shown that ‘frozen’
readings on this type of instrument are often
unreliable although laboratory examination can often
help resolve this.
Most instruments from around the 1990s have used
miniature stepper motors. Stepper motors are
constantly powered when in use and they
continuously correct the position of the needle.
When switched off, the needle is powered down to
zero. However, in the event of the wiring being
severed or the battery is smashed in a collision, there
is nothing to move the position of the needle other
than external forces.

236
 A review of speedometers and the criteria to be considered before accepting ‘frozen’ readings and other marks

In Europe, most, but not all, stepper motors used for
speedometer and rev. counters are fitted with
reduction gearing within the instrument body.
The gearing ratio and type of gearing used with step-
per motors varies widely between instrument
manufacturers. That influences considerably the
‘static torque’, which is the resistance to movement
when not powered. The needles of instruments
having worm-gearing are highly unlikely to move in
a collision.
The needles of instruments that have high gearing
ratios are less likely to move on impact and so any
‘frozen’ readings following a collision may well be
reliable, as demonstrated in limited research
undertaken so far. However, it is our intention that
laboratory-based realistic impact testing will be
carried out in order to help resolve this issue. The
needles of instruments that are either not geared or
have low ratio gearing can move easily and so any
‘frozen’ readings are likely to be unreliable in the
event of lateral forces having been experienced in a
collision.
When both the speedometer and rev. counter have
‘frozen’ readings following a collision, and are of the
same type of stepper motor, the speed can be
compared with expected readings for given revs in
different gears. That information may be obtained
either from published information or from
comparison tests using an identical vehicle.
Readings that correspond provide a considerable
degree of confidence in the speed reading.
In all cases, investigators should be wary of
accidents in which the vehicle has skidded or left the
road prior to collision, as the readings are likely to
change during these events. Additionally, the size of
tyres on the monitored axle should be checked to see
that they are those specified by the vehicle
manufacturer and have a rolling circumference
similar to that specified.
Peter Jennings (TCRI, Worcester)
Jon Neades (AiTS, South Cerney)
References
[1]
http://www.coregravity.com/assets/images/all_phase_step
per_motor.JPG
[2] ISO 7637-2: 2004(E) Road Vehicles – Electrical
disturbances from conduction and coupling.
[3] ITAI Contact, Edition 77, April 2004.
[4] R.D Anderson, Post Collision Speedometer Readings
and Vehicle Impact Speeds, Collision, Vol 5 Issue 2, p32-
41 2010.
[5] A Kuranowski, ‘Reliable determination of impact
velocity on the basis of indications of the speedometer
stopped after the collision’, Proceedings of the 6th
International Scientific Conference Transbaltica 2009.
[6] Wang, Q. and Gabler, H.C., “Accuracy of Vehicle
Frontal Stiffness Estimates for Crash Reconstuction”,
Proceedings of the Twentieth International Conference on
Enhanced Safety of Vehicles, Paper No. 07-0513, Lyons,
France (June 2007).
[7] Dhafer Marzougui, Development And Validation Of
An Ncap Simulation Using LS-DYNA3D,
FHWA/NHTSA National Crash Analysis Center, The
George Washington University, Virginia 22011
ntl.bts.gov/lib/6000/6800/6897/ncap.pdf.
Contact
Christopher Goddard
BSc IEng MIET
Collision Science
info@collisionscience.co.uk

Acknowledgements David Price

B.Tech MITAI
We would like to extend our sincere thanks for the Forensic Accident Investigation Services Ltd.
advice and knowledge provided by the following; info@forensic-ai.co.uk
Mike Lees, Visteon Ltd
Martin Ebenter, Continental, Switzerland
Dr Heinrich-Jochen Blume, Continental, Germany
Helmuth Mohr, Continental, Switzerland
Asger Jurgensen, Continental, Switzerland
Stefan Rupp, Continental, Switzerland
Betinna Zahnd-Sinzig, AXA Research Centre
Christian Planting, ACtronics Ltd
Bob Russell, HG Brunner Ltd

We would also like to extend our thanks for helpful

discussions held with:

237
 Christopher Goddard, David Price

238
© 2013 by EVU
EVU 2013 - 28

Muscular activity in normal driving and in rear end low

speed crash tests
Carlo Cialdai, Dario Vangi, Antonio Virga, Carlo Cialdai
University of Florence

Abstract
Surface electromyography (EMG) is a measurement technique that can been used to study muscular activity of human
subjects. In the work described in this paper electromyography was used to evaluate muscular activity during normal
driving and compare it to muscular activity prior low speed rear end crash tests, aimed at understanding if human
subjects’ condition is comparable to some real life driving phase, thus assessing crash tests usefulness in the study of
whiplash and in the determination of injury thresholds. The muscular activity of 9 human subjects sitting on the driver's
seat of a vehicle was monitored by means of surface electromyography, during a road test drive and during low speed
rear end collisions. Some muscles, chosen among those involved in whiplash mechanisms, were monitored, together
with forearm and shoulder muscles in order to evaluate bracing effort. Road tests (on a city route about 5.5 km long)
were carried out within normal, moderate traffic and also included some of the situations in which involvement in a rear-
end collision is most probable: keeping a constant speed, slowing down, braking suddenly, waiting at a traffic light, etc.
The same drivers were subjected to low speed rear-end vehicle to vehicle collisions; 36 crash tests of two types were
carried out (aware and unaware subject) with two nominal impact speeds (6 and 8 km/h). Volunteers were wearing a
lightweight plastic headband, carrying accelerometers and a video camera target. The tests were recorded by a high
speed video camera, synchronised with the accelerometers. The driving tests confirmed that different driving conditions
involve different muscular activation, braking violently usually producing the highest values. During the crash test little
difference was observed between aware and unaware subjects, unless a tense posture was maintained.

subjects monitored by accelerometers on a sled

Introduction
Whiplash in rear-end collisions is a relatively well-
known injury mechanism, and in past research
threshold injury levels have been defined in order to
use them for accident reconstruction purpose. It is a
common practice to use human subjects ([2], [7])
during low speed rear-end crash tests in order to
study whiplash related injuries. In fact, while in high
speed crash testing the use of surrogates such as
cadavers and dummies is the only possible solution,
in low speed tests human subjects can be
conveniently used, also considering the poor
performances that, in this kind of tests, the traditional
surrogates can show [10]. In the past surface
electromyography (EMG) has sometimes been used
to understand muscular activity during the tests, as in
[8], [5] or [10], where 10 tests were carried out on 5
subjects, either male and female, in order to compare
two different head restraints. Using EMG together
with other traditional techniques (such as
accelerometers and high speed video cameras) can
help understanding of many aspects of the problem.
Subjects' awareness can be a confounding factor in
such kind of experimentation: in [8] EMG was used
to study the effect of awareness; in [6] the role of
awareness was investigated by testing human
device; in [10] precautions were taken in order to
ensure that the subjects were unaware of the
impending collision. In [8] the effect of multiple
tests on the same subject was also investigated, by
carrying out a sequence of 11 tests on each subject.
Whiplash related collisions usually occur in well
characterised occasions, such as while decelerating,
braking or waiting at a traffic light; for instance in
[1] the effect of braking was investigated.
In the work described in this paper surface
electromyography was used to evaluate muscular
activity during normal driving and compare it to
muscular activity prior to and during crash tests,
aiming at understanding if human subjects’ condition
is comparable to some real life driving phase, thus
assessing crash tests usefulness in the study of
whiplash and in the determination of injury
thresholds. The effect of subjects’ awareness or
unawareness of the moment of collision was also
investigated. Then, having submitted each volunteer
to several tests, the effect of adaptation to repeated
consecutive collisions was evaluated.
Subjects
Nine subjects were enrolled for the experimentation
here described, 2 female and 7 male (see table 1). All

239
 Cialdai Carlo, Vangi Dario, Virga Antonio

of them declared to have never had injuries caused
by whiplash or presented illnesses in relation to the
muscular or skeletal apparatus. Subjects' age was
ranging from 26 to 68, their height was 165 to 183
cm, weight was 56 to 100 kg.
was recorded, aiming at correlating EMG signals
with subjects' movements. Besides, the vehicle was
equipped with a video camera to record the course
and a GPS acquisition device (Qstarz model BT-
1000EX) to record position and speed.

Tab. 1 - Characteristcs of the subjects participating to the Experimental set-up of crash tests
tests.
Subject id Gender Age Height Weight Two passenger cars were used, one as a target
(cm) (kg) vehicle (Fiat Punto 1st series, with a mass of 800 kg)
A F 26 168 56 and one as a bullet vehicle (Opel Corsa 2nd series,
B F 32 165 56 mass 950 kg). The tests took place at the crash
C M 56 183 92 testing facility managed by University of Florence.
D M 33 174 66 The target vehicle was equipped with an
E M 28 180 76
accelerometer (Dytran model 7521A6, full scale 100
g) installed near the centre of gravity. Impact
F M 29 180 72
velocity of the bullet vehicle was measured using a
G M 48 173 74 radar gun (Stalker Pro model Ats) and a
H M 68 169 76 photoelectric laser device (Age Scientific). The
I M 29 176 100 bullet vehicle was launched against the target vehicle
by means of two elastic cords joined in parallel array
(Sandow Technic model Avia) put in tension with an
Experimental set-up of driving tests electric winch (Ramsey model Rep 8000). Each test
was recorded by means of a high speed video camera
Road tests were carried out on a city course, (Fastec Inline, 250 fps, b/w), framing the target
characterised by moderate traffic; the test track vehicle and the subject; to facilitate the vision of
consisted in a ring with a length of about 5.5 km, subjects, the driver's front door of the car was
usually run in about 12 minutes. All subjects drove removed (fig. 1). To allow measurements on the
the same vehicle (Renault Clio II 1.2 16V) and faced video recordings some video targets were positioned:
the usual driving conditions that can be commonly one on the centre of the subject's head and four on
encountered and, among the others: the car at known distance, meant to be used as
- driving on straight road at constant speed reference. Besides EMG electrodes (with the same
(either 50 and 70 km/h); set-up as in driving tests), volunteers were measured
- driving along curves (right and left), usually at by means of two accelerometers, fixed on a
low speed; lightweight plastic headband: a triaxial one
- waiting at a traffic light; (Crossbow model CXL25GP3, full scale 25 g),
- slowing down and braking mildly. mounted at the top of the head, and a uniaxial one
At the end of each test the subjects were requested to (Dytran model 7521A6, full scale 100 g), mounted
brake violently, thus simulating an emergency on the forehead. Signals from the accelerometers
manoeuvre. were acquired by a laptop PC with a sampling rate of
Volunteers were monitored by means of surface 1024 Hz and synchronised with the high speed video
electromyography, aiming at evaluating their recording by means of a trigger signal.
muscular activity during normal driving. For this
reason the instrument BE Micro, designed and 36 low speed rear end collision tests were carried out
produced by EBNeuro (Florence, Italy) was used, (18 with bullet vehicle nominal velocity of 6 km/h
together with EBN disposable adhesive electrodes. and 18 at 8 km/h); the collision speed upper limit
More specifically surface electrodes were applied to was chosen to avoid any risks to the volunteers; in
5 couples of muscles chosen among those involved fact ([10], [7] and [3]) it is highly improbable that
in whiplash mechanisms, together with forearm and the subjects will suffer injuries when velocity change
shoulder muscles in order to evaluate bracing effort: -8 km/h, and velocity change is
sternocleidomastoid (shortly, in the following always less than impact velocity. Besides, such low
figures: SCM), splenius (SPL), flexor (FLX), velocity prevented the vehicles from being damaged.
trapezius (TRAP), deltoid (DLT), both right and left; The target vehicle was initially stopped and with the
they were acquired in differential mode (2 electrodes brakes released; the collision was always done with a
for each muscle) by a laptop PC at the sampling rate 100% overlap.
of 1024 Hz; a synchronised front video of the driver

240
 Muscular activity in normal driving and in rear end low speed crash tests

test27 F unaware 7.8 6.6 5.1

test28 F unaware 6.6 6.3 3.5
test29 D aware 6* 4.6 2.8
test30 D aware 8.0 n.a. n.a.
test31 D unaware 6.1 5.2 2.7
test32 D unaware 7.9 6.5 4.9
test33 I aware 5.9 4.7 2.6
test34 I aware 7.7 6.3 4.5
test35 I unaware 8.0 6.6 5.5
test36 I unaware 6.3 n.a. n.a.
(* nominal values)

The driver seat of the target vehicle was occupied by

the subject, wearing the safety belt; the seat was
positioned according to the subject's preference. The
bullet vehicle was empty. The tests were carried out
following two modes: with aware subject (A: the
volunteer could see and hear the bullet vehicle
approaching) and unaware subject (U: windshield
Fig. 1 - One of the subjects ready for a crash test. and rear mirrors obscured and earphones with music
at loud volume, aiming at avoiding the perception of
the beginning of the test). Besides, the waiting time
between the tests was randomly varied. Combining
Tab. 2 - Summary data of all the crash tests.
the two testing modes and the two impact velocities,
Test id Subje Test type Impact Peak
ct id vel. acc.
each volunteer was subjected to 4 tests, with some
(km/h) (km/h) (g) minutes pause (usually the testing sequence was A6,
A8, U6, U8, see table 2). Peak acceleration and
test01 A aware 6.6 3.5 1.8
test02 A aware 8.4 6.2 4.0 obtained from the accelerometer installed on the
test03 A unaware 6.0 4.5 2.2 target vehicle, while impact velocity was measured
test04 A unaware 7.7 6.1 4.3 with a radar gun or a photoelectric system. T-test
test05 B aware 6* 4.8 2.8 was used to decide if the two data sets were
test06 B aware 8.2 7.2 4.6 significantly different.
test07 B unaware 7.4 5.8 3.6
test08 B unaware 8.4 6.8 5.3 Results for driving tests
test09 H aware 8* 6.1 3.7
The electromyographic signals obtained during each
test10 H aware 6* 5.5 3.0
test were normalised by dividing by a mean value
test11 H unaware 8* 5.1 2.8 obtained by a reference signal (one for each subject
test12 H unaware 6* 6.3 4.5 and for each test) acquired during a relaxed
test13 G aware 5.7 4.9 2.6 condition, thus yielding adimensional numbers and
test14 G aware 7.5 6.2 3.8 allowing for the inevitable differences due to
test15 G unaware 5.8 4.8 2.4 physiologic variations, electrodes’ position and
test16 G unaware 7.4 6.2 3.9 coupling conditions. Thanks to GPS receiver and
test17 C aware 5.5 4.6 2.3 video camera recordings, the moments in which a
test18 C aware 7.2 6.0 3.8
particular driving condition occurred were identified;
in such moments the normalised signals were picked
test19 C unaware 7.3 5.9 4.1
and averaged over a time period of half a second. For
test20 C unaware 5.8 5.0 2.5 instance in figures 2 and 3 the values obtained on all
test21 E aware 6.1 5.8 2.9 the muscles for all the subjects, in different driving
test22 E aware 7.9 7.0 4.5 conditions are shown; each value is, in turn, the
test23 E unaware 6.1 5.0 3.0 average calculated for the same type of driving
test24 E unaware 7.9 6.5 4.4 conditions detected in each driving test. The high
test25 F aware 6.8 5.3 3.3 values shown by right flexor, deltoid and trapezius,
test26 F aware 7.8 6.4 4.3 in comparison to left ones, can be explained by the

241
 Cialdai Carlo, Vangi Dario, Virga Antonio

use of gear lever. In figure 4 the ratio of the different Results for crash tests
driving conditions in respect of normal driving on
straight road is shown, obtained by averaging for all For each subject, 4 low speed rear-end crash tests
muscles for each subject. As it can be seen, for all were carried out, usually in the sequence A6, A8,
subjects there is a marked difference in the muscles' U6, U8; electromyographic signals were acquired
activation in the specified driving conditions, waiting and processed in the same manner as for driving
at a traffic light usually giving the lowest values, tests. In figure 5, for instance, the results for all the
while during heavy braking (simulating an muscles monitored after averaging for all the
emergency braking) the highest values were usually subjects are shown.
recorded. Heavy braking was shown to have
statistically higher EMG values than all the other 10,0
conditions for 8 muscles out of 10; braking mildly 9,0 aware

normalised EMG mean

proved to be equivalent to normal driving (for all the 8,0
unaware
7,0
muscles), while waiting at a traffic light had lower
6,0
activation than normal driving for 4 muscles out of
5,0
10. 4,0
3,0
Mild braking 2,0
25,0
Heavy braking 1,0
normalised EMG mean

20,0 Traffic light

0,0
Straight normal driving
R-SCM L-SCM R-TRAP L-TRAP R-SPL L-SPL R-FLX L-FLX R-DLT L-DLT
15,0
muscles
10,0
Figure 5 - EMG normalised mean for the muscles,
5,0 averaged on all subjects.
0,0
R-SCM L-SCM R-TRAP L-TRAP R-SPL L-SPL
In figure 6 a comparison of the averaged normalised
muscles
amplitude of the muscles of the neck (i.e.
Figure 2 - EMG normalised mean for the muscles of the sternocleidomastoid, splenius and trapezius, right
neck, averaged on all subjects. and left) for all the subjects is shown; muscles'
activation is almost always higher for aware subjects.
The same can be said (figure 7) for the muscles of
25,0 Mild braking

Heavy braking
the shoulder (deltoid) and those of the forearm
(flexor). Considering the overall average, there is a
normalised EMG mean

20,0 Traffic light

Straight normal driving statistically significant difference, with a P-value

15,0
equal to 0.036. As regards the single muscles, only 2
10,0 out of 10 show a statistically significant difference
(left trapezius, P-value 0.040, and left
5,0
sternocleidomastoid, P-value 0.014). The other
0,0 muscles did not show a significant difference and, all
R-FLX L-FLX R-DLT L-DLT
in all, it can be said that no great difference is
muscles
detectable in muscles' activation. This is usually true
Figure 3 - EMG normalised mean for the muscles of for all cases but some in which the subjects chose to
shoulders and arms, averaged on all subjects. maintain a voluntarily tense (stiffened) position. One
Ratio to straight normal driving Mild braking
of the subjects (volunteer “I”, figure 8), was asked,
3,5 Heavy braking
Traffic light
while waitingfor the collision to stiffen their arms
normalised EMG mean

3,0
and legs and brace with force; such position caused a
2,5 great difference between aware and unaware
2,0 conditions, as can be seen in figures 6 and 7.
1,5 Comparing EMG values obtained during the driving
1,0 tests to those obtained by crash tests, it can be
0,5 pointed out that usually muscles' activation is higher
0,0 while driving (see figure 9, where EMG normalised
A B C D E F G H I
mean is shown for the different muscles, averaged on
subjects
all the subjects, in the different conditions); besides
Figure 4 - EMG ratio to straight normal driving for all the an activation level comparable to that obtained
subjects, averaged on all muscles. during emergency braking is obtained only with an

242
 Muscular activity in normal driving and in rear end low speed crash tests

aware and tense condition. In fact heavy braking 25,0 aware

unaware
showed a statistically higher EMG activation than

normalised EMG mean

20,0 mild braking
aware tests for 6 muscles out of 10, while normal heavy braking
driving was equivalent to aware tests for 5 muscles; 15,0
straight normal driving
traffic light
unaware tests yielded lower values than waiting at a
traffic light 5 times out of 10. Finally, normal driving 10,0

had higher activation than unaware tests 7 times out

5,0
of 10.
0,0

10,0 R-SCM L-SCM R-TRAP L-TRAP R-SPL L-SPL R-FLX L-FLX R-DLT L-DLT

9,0 muscles
aware
normalised EMG mean

8,0
unaware
Figure 9 - Comparison between EMG mean value for the
7,0
6,0
muscles obtained in crash tests (aware and unaware) and
5,0
in driving tests.
4,0
To focus on how muscles’ activation influenced
3,0
whiplash risks, longitudinal accelerations for head
2,0
1,0
and vehicle were analysed, and in particular an
0,0
adimensional coefficient k was used, defined as:
A B C D E F G H I (a x _ head ) max
subjects k
(a x _ car ) max
Figure 6 - EMG normalised mean for the subjects,
averaged on the muscles of the neck only.
meant to be used as an indicator for load
amplification on the head; the coefficient proved to
10,0 be statistically the same for aware and unaware
9,0 aware subjects (mean values were 1.85 and 1.90,
normalised EMG mean

8,0
unaware respectively, with a P-value equal to 0.4). This is to
7,0
6,0
say that the little difference observed in muscles’
5,0
activation, corresponded to a negligible difference in
4,0 k. Again, a difference was observed when a
3,0 deliberately tense position was maintained, as in the
2,0 case of subject “I”. In fact, as shown in figure 10, in
1,0
this condition about the same peak longitudinal
0,0
A B C D E F G H I
acceleration was reached for head and vehicle (k
subjects
close to 1), while with a normal aware condition
(figure 11), head acceleration was over two times
Figure 7 - EMG normalised mean for the subjects vehicle acceleration (k greater than 2). This means
averaged on the muscles of the shoulder and arm. that it is not enough to be aware of an impending
rear-end collision to minimise the risk of whiplash
injuries, if a highly defensive attitude is not
maintained.
7,5

vehicle
5,0 head
acceleration ax (g)

2,5

0,0
0,00 0,10 0,20 0,30 0,40

-2,5

-5,0
time (s)

Figure 10 - Longitudinal acceleration for head and

vehicle, aware and tense test for subject “I”.

Figure 8 - Subject “I”, waiting for the crash test in a tense

position, as recorded by high speed video camera.
243
 Cialdai Carlo, Vangi Dario, Virga Antonio
15,0
12,5
10,0
acceleration ax (g)
7,5
5,0
2,5
0,0
-2,5
0,00 0,10 0,20 0,30
-5,0
time (s)
Figure 11 - Longitudinal acceleration for head and
vehicle, aware test for subject “I”.
As said, all the volunteers were subjected to
consecutive crashes (usually 4, 2 aware and 2
unaware); separating these two types of tests to avoid
their confounding influence, the sample of first crash
was compared to that of second crash: in no case a
statistically significant difference was found, for
either the aware or unaware type of test, in respect of
EMG signals. In fact for the aware type of test a
minimum P-value of 0.21 was found, while for the
average of all the muscles it was 0.88; for the
unaware type of test minimum P-value was 0.11, and
0.43 for the average of all muscles. This indicates
that the first test did not influence the successive one.
Conclusions
Nine human subjects were enrolled for the research
project here described, with the purpose of studying
muscular activity during normal driving and before
the execution of low speed rear end crash tests.
Surface electromyography was employed as a
measuring technique for the evaluation of muscular
activation. The EMG instrumented human subjects
were asked to drive on a city route; afterwards they
were subjected to rear end collisions in two
conditions: aware and unaware of the impending
impact. EMG monitoring showed that a different
muscular activity can be recorded in different driving
conditions, braking violently usually causing the
highest activation, while waiting at a traffic light the
lowest. As regards the moments before crash tests,
aware condition usually produced a higher muscular
activation than unaware condition, limited to a little,
not always statistically significant, extent. Only
when a deliberately tense posture was maintained, a
relevant difference was observed. This was also
confirmed by the analysis of head to car acceleration
ratio. As regards the comparison to the muscular
activity during real life driving, the research
presented showed that usually during crash testing
muscles’ activation is lower; levels of activation
244
comparable to those typical of heavy braking are
recorded only if an aware and tense posture is held,
vehicle while in unaware conditions low activation levels are
head often recorded, even in comparison to waiting at a
traffic light.
0,40 References
[1] Anderson R. D., Welcher J. B., Szabo T. J., Eubanks J.
J., Haight W. R., Effect of Braking on Human Occupant
and Vehicle Kinematics in Low Speed Rear-End Collisions.
SAE Paper No. 980298 (1998).
[2] Braun T. A., Jhoun J. H., Braun M. J., Wong B. M.,
Boster T. A., Kobayashi T. M., Perez F. A., Hesler G. M.,
Rear-end impact testing with human test subjects. SAE
Paper No. 2001-01-0168 (2001).
[3] Kornhauser M., Delta-V Thresholds for Cervical Spine
Injury. SAE Paper No. 960093 (1996).
[4] Krafft M., Kullgren A., Tingvall C., Boström O.,
Fredriksson R., How crash severity in rear impacts
influences short- and long-term consequences to the neck.
Accident Analysis and Prevention 32 (2000) 187–195.
[5] Kumar S., Ferrari R., Narayan Y., Kinematic and
electromyographic response to whiplash loading in low-
velocity whiplash impacts—a review. Clinical
Biomechanics 20 (2005) 343–356.
[6] Kumar S., Narayan Y., Amell T., Role of awareness in
head-neck acceleration in low velocity rear-end impacts.
Accident Analysis and Prevention 32 (2000) 233–241.
[7] McConnell W. E., Howard R. P., Guzman H. M.,
Bomar J. B., Raddin J. H., Benedict J. V., Smith H. L.,
Hatsell C. P., Analysis of Human test subject kinematic
responses to low speed rear end impacts. SAE Paper No.
930889 (1993).
[8] Siegmund G. P., Sanderson D. J., Myers B. S., Inglis J.
T., Rapid neck muscle adaptation alters the head kinematics
of aware and unaware subjects undergoing multiple
whiplash-like perturbations. Journal of Biomechanics 36
(2003) 473–482.
[9] Szabo T. J., Voss D. P., Welcher J. B., Relationships
between impact pulse duration and occupant kinematics in
low speed rear impacts. SAE Paper No. 2002-01-029
(2002).
 Muscular activity in normal driving and in rear end low speed crash tests

[9] Szabo T.J., Welcher J. B., Human subject kinematics

and electromyography during low speed rear impacts. SAE
Paper No. 962432 (1996).

[10] Welcher J. B., Szabo T. J., Voss D. P., Human

occupant motion in rear-end impacts: effects of incremental
increases in velocity change. SAE Paper No. 2001-01-0899
(2001).

Contact
Dario Vangi, Professor, Antonio Virga, PhD, Carlo Cialdai,
PhD student
Department of Industrial Engineering
Via di S. Marta, 3.
50139, Florence, Italy
e-mails:
dario.vangi@unifi.it antonio.virga@unifi.it
carlo.cialdai@unifi.it
tel: ++ 390554796505

245
 Cialdai Carlo, Vangi Dario, Virga Antonio

246
© 2013 by EVU
EVU 2013 - 29

Impacts with roadside trees – just a German problem?

Alexander Berg, Jörg Ahlgrimm, Stephan Schlosser
DEKRA Automobil GmbH, Stuttgart, Germany
Prof. Dr. Hermann Steffan
Technical University of Graz, Graz, Austria

Abstract
Impacts with trees alongside the carriageway (so called “tree crashes”) may still be an area where the number of fatal accidents on our
roads could be significantly reduced. The official German statistics have included the attribute “impact with a roadside tree” since
1995. In that year, the proportion of road users killed in such crashes was 24% (2,284 out of 9,454 fatalities). Today, impacts with
trees are still one of the most important focal points of road deaths in Germany. For the year 2012, the latest figures show a proportion
of 19% (696 out of 3,600 fatalities). Most of these crashes occur on rural roads. In 2012, for crashes on roads outside urban areas
(excluding motorways) 2,151 fatalities were reported and 28% of them (595 fatalities) related to impacts with trees. It is estimated that
this may not be a problem only in Germany but also in other European countries. With the aim of launching international research into
this topic an additional study was conducted on tree-impacts in Austria and Germany. This paper reports on some of the analyses and
results. It seems that in Austria the problem of “tree impacts” is not as predominant as in Germany. Nevertheless, Austria also seems
to offer remarkable potential for a significant contribution to further reduction of road-user fatalities. The studies that are now being
conducted in a “2-country joint venture” could provide the basis for a more extensive accident research on “tree impacts” at a pan-
European level which could suggest ways of further reducing road fatalities by 50% which is in line with new strategic EU-27 target
by 2020.

1 Accident Statistics was not entirely met. Nevertheless, the reduction of

1.1 Vehicle and road safety - the long-
term trends
1.1.1 Road fatalities in the European Union
Over the past decades great strides have been made
in making European roads safer. This is impressively
demonstrated by corresponding accident statistics.
For example, the number of fatalities on the road per
year in the area covered by the European Union (EU-
27) during the period 1991 - 2001 fell by 28% to
54,302, Fig. 1, and in 2010 the number of fatalities
was 31,029. This represents a reduction of 43% on
2001. It means that the ambitious target of the 3rd
European Road Safety Action Program that ended in
2010 and set out to halve the number of annual road
fatalities (related to the number of fatalities in 2001)
43% is to be seen as a success. The new EU policy
orientations on road safety 2011-2020 aim at a
further reduction of the number of road fatalities by
50% [2]. Related to the area covered by the EU-27,
this means a reduction to 15,515 fatalities in 2020.
This goal would be achievable if the long-term
historical trends hold up and do not tail off over the
coming years.
In the long term, the strategic goal is set out in
"Vision Zero" [3]. This means a continuous
reduction of the number of road fatalities that
continues well beyond 2020. A period up until 2050
has been discussed, the number of road fatalities in
Europe falling to a sustainable figure of "virtually
zero" via the systematic exploitation of all safety
potential in the driver-vehicle-environment system.

247
 Berg Alexander, Ahlgrimm Jörg, Schlosser Stephan, Prof. Dr. Steffan Hermann

1991 2001 2010 2020 Evolution:

1991 - 2001: -21.124 (-28 %)
80.000 2001 - 2010: -23.273 (-43 %)
75.426

70.000 2010 - 2020: -15.515 (-50 %)

60.000 54.302
Killed road users

50.000

40.000

30.000 31.029
?
20.000
15.515
10.000

0
1992

1994

1996

1998

2000

2002

2004

2006

2008

2010

2012

2014

2016

2018

2020
Year

Figure 1 Historical evolution in the number of road fatalities in the European Union (EU-27)
from 1991 to 2012 and the new target for 2020
(Source: CARE European Road Accident Database, as of March 2013)

1.1.2 Road fatalities in Germany and out its intention to achieve a further 50% reduction
Austria in the number of road fatalities in its road safety
programme 2011-2020 [5]. This would still
Germany and Austria have both been able to register correspond to 276 fatalities in 2020.
considerable success in improving vehicle and road
safety, Fig. 2. The number of fatalities on Germany's The targets set out by Germany and Austria
roads during the period 1991 to 2001 fell by 38% respectively appear achievable if previous trends
from11,300 to 6,977, and during the period 2001 to continue in the coming years. Among other things,
2010 by 48% to 3,648. A halving of this figure by the continued improvement of active and passive
2020 in line with the EU targets would correspond to vehicle safety (see also [12]) has hitherto played a
1,824 road fatalities. However, officially, the fundamental role in these figures. As the current
German Road Safety Programme 2011 only national statistics show, these trends have so far
stipulates a reduction of 40% in the number of road remained on course. Although Germany recorded an
fatalities by 2020 [4]. increase in the road fatality figures in 2011 for the
first time in many years with 4,009 fatalities [6, 7],
Austria recorded 1,551 road fatalities in 1991. This the figure of 3,600 fatalities in 2012 once again
figure fell by 2001 to 958 (-38%). Related to the constituted an historic low [8]. 523 road fatalities
number of road fatalities in 2001, the fall to 552 road were recorded in Austria in 2011 [9]. According to
fatalities in 2010 corresponds to a relative decrease initial reports 522 persons were killed here on the
of 42%. Similar to the EU-27, Austria has also set roads in 2012 [10].

248
 Impacts with trees alongside the road - just a German problem?

1991 2001 2010 2020 Evolution in Germany:

1991 - 2001: -4.323 (-38 %)
12.000 11.300 2001 - 2010: -3.329 (-48 %)
11.000
2010 - 2020: -1.824 (-50 %)
10.000

9.000
Evolution in Austria:
Killed road users

8.000
6.977 1991 – 2001: - 593 (-38 %)
7.000
2001 – 2010: - 406 (-42 %)
6.000
2010 – 2020: - 276 (-50 %)
5.000
4.000

3.000 3.648

2.000
1.551
958 1.824
1.000 552
276
0
1992

1994

1996

1998

2000

2002

2004

2006

2008

2010

2012

2014

2016

2018

2020
Year

Figure 2 Historical evolutions in the numbers of road fatalities in Germany and Austria
from 1991 to 2012 and new targets for a further halving by 2020
(Sources: CARE European Road Accident Database, as of March 2013,
and national official statistics)

1.2 Accidents involving an impact with a Laender of the Federal Republic of Germany. In
tree along the road particular, in some Laender in eastern Germany the
dangers of accidents in tree lined roads (alleys)
1.2.1 Figures for Germany which are typical of these landscapes were discussed
Detailed figures on the accident breakdown in several times at road safety conferences and
Germany have been published for the year 2012 [6, corresponding preventative campaigns implemented
7]. Of the total of 3,600 road accident fatalities in over time (see for example [11]). This has been
this year in Germany 696 persons (19%) died as a reported about on several occasions (see, for
result of an impact with a tree lining the road, Fig. 3. example [12] and [13]).
In 1995, when the official statistics starts to include If the figures are considered on the basis of accident
the attribute "Impact with a roadside tree", the sites, in the year 2012 it can be seen that of the
corresponding share was still 24% (2,284 of a total 1,062 fatalities recorded within urban areas in
of 9,454 fatalities). While the number of all road Germany, 67 persons (6%) died as a result of an
fatalities fell from 9,454 by 62% to 3,600 during the impact with a tree, Fig. 4. Of the 2,151 fatalities
years 1995 to 2012, the fall in fatalities resulting recorded in accidents outside urban areas excluding
from an impact with a tree alongside the road was the Autobahn, 595 persons (28%) died as a result of
considerably greater at 70% (from 2,284 to 696). impacts with a tree. Of the 387 fatalities in accidents
on the Autobahn, 34 persons (9%) were killed by
Obviously, this reflects the success of targeted road collisions with a tree.
safety campaigns carried out in the individual

249
 Berg Alexander, Ahlgrimm Jörg, Schlosser Stephan, Prof. Dr. Steffan Hermann

10.000
9.454 (100 %)
9.000

8.000

7.000
-62 %
Killed road users

6.000

5.000

4.000
3.600 (100 %)
3.000
2.284 ( 24 %)
2.000
-70 %
1.000 696 ( 19 %)

0
1995
1996
1997
1998
1999

2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
Year

Figure 3 Historical evolutions in the number of all road fatalities and fatalities in accidents involving impacts with roadside
trees in Germany from 1995 to 2011 (Source: DESTATIS, [6])

Inside urban areas Impact with roadside Outside urban areas (without Autobahn)
1.062 fatalities obstacles: 2.151 fatalities

Tree

Pole 28%
6%
Abutment
5%
Barrier
78% 0%
Others 55%
1% 1%
10% no impact with obstacle 0%
8%
9%
44% 8%
0%

0%
Killed road users in the year 2012:
1.062 fatalities inside urban areas On the Autobahn
2.151 fatalities outside urban areas 387 fatalities
(without Autobahn)
387 fatalities on the Autobahn
38%
3.600 fatalities in total 9%

Figure 4: Percentage of fatalities resulting from collisions with obstacles next to the road broken down according to sites
of the accidents in Germany in 2012 (Source: DESTATIS, [6])

Impacts with a tree next to the road thus continues to even inside urban areas and on the Autobahn, should
be a strong focus in current accidents involving not be underestimated. Such observations could have
fatalities in Germany on national, regional and local been made years ago, when, although the absolute
roads outside urban areas, so-called “country roads” road fatality figures were greater, the percentages of
(Landstraßen). However, the role played by a fatalities in accidents involving a collision with a tree
collision with a tree in accidents involving fatalities next to the road were of a similar magnitude [12, 13].

250
 Impacts with trees alongside the road - just a German problem?

This area continues to offer great potential for further
reducing the number of road fatalities in Germany.
This potential in some respects varies greatly in the
individual German Laender. Mostly affected are the
larger Laender with an extensive network of country
roads and a large number of roads with trees next to
the carriageway, Fig. 5. Generally, the percentage of
road fatalities caused by accidents involving trees on
country roads is greater than the percentage of road
fatalities on all roads inside and outside urban areas
(including the Autobahn). Typical of these are,
firstly, accidents in alleys (roads with trees at a small
distance left and right alongside the carriageway).
But also individual trees located next to the road or
trees on the edge of woods, if they are located very
close to the road (which has also frequently
inadequately consolidated shoulders) with no
specific protective measures.

Germany
Year 2012
Hamburg
Berlin
Bremen on country roads
Rhineland-Palatinate in all sites

Baden-Württemberg
Bavaria
German Laender

Hesse
Saarland
Schleswig-Holstein
North Rhine-Westphalia
Thuringia
Saxony
Saxony-Anhalt
Mecklenburg-West Pomerania
Lower Saxony
Brandenburg

0 10 20 30 40 50 60% 70
Share of fatalities in roadside tree impacts

Figure 5 Percentages of fatalities in accidents involving impacts with roadside trees in the individual German Laender in
2012 in all sites and just on country roads (Source: DESTATIS, [7])

The greatest number of road fatalities resulting from
an impact with a roadside tree was recorded in
Lower Saxony. In 2012 there were 152 fatalities on
all roads, corresponding to a share of 32% of all
486 road fatalities in Lower Saxony. On country
roads alone, 146 fatalities were recorded for
accidents involving impacts with trees, which
represent 43% of the total 341 fatalities here.
The largest share of fatalities resulting from
accidents involving trees next to the road were in
Brandenburg. Here, there were 166 road fatalities on
all roads in 2012; more than one in three
(67 fatalities or 40%) resulted from
impacts with trees. 89 persons died in Brandenburg
on country roads. Of these, more than half (53
fatalities or 60%) resulted from impacts with a tree.
Nevertheless, for a long period of time Brandenburg
was one of the Laender that was able to record an
especially favourable relative trend in road fatalities
caused by tree accidents (see also [12]). Its figures
fell by 85% between 1995 (344 fatalities) and 2012
(53 fatalities).
Of the larger Laender, Rhineland-Palatinate
displayed the lowest percentage of tree accident
fatalities in 2012. There were 25 tree-accident
fatalities (12%) of the 209 fatalities on all roads and
20 (15%) of 133 fatalities on country roads.
251
 Berg Alexander, Ahlgrimm Jörg, Schlosser Stephan, Prof. Dr. Steffan Hermann
Fatal tree accidents are extremely rare in Berlin,
Bremen and Hamburg. Basically, this can be traced
to the fact that these so-called "city states" have very
few country roads. Although trees are indeed
frequently to be found next to the roads in urban
areas, the low speeds in urban areas mean that the
consequences of tree accidents are less serious.
1.2.2 Austria
The published official Austrian accident statistics do
not contain the attribute "Impact with a roadside
tree". As a result, it is not possible to carry out a
corresponding statistical evaluation as is the case for
Germany. On the basis of regionally delimited
analyses, however, some findings do exist for
accidents involving impacts with roadside trees in
Austria.
One study took a random sample in Lower Austria
and revealed that there were 39 accident fatalities in
the year 2002 resulting from impacts with a tree next
to the road [14]. The sample of the tree accidents
was taken from the accident reports of the safety
departments, a supplemental verification of data
from Statistik Austria being undertaken. The official
statistics show that a total of 261 road fatalities were
recorded in Lower Austria in 2002 [15]. Therefore,
as not all accidents were covered by the sample, the
percentage of road users killed by an impact with
trees next to the road in Lower Austria in 2002
totalled more than 15%. In Germany in 2002 a
corresponding figure of 23% (1,577 of
6,842 fatalities) was given.
As already known from earlier studies, a higher
severity of accident could be established for areas of
greater accident frequency in Lower Austria on
routes with trees as on treeless routes. In the
accidents analysed, single vehicle accidents on
straight roads predominated, a greater percentage
happening at night. Apart from leaving the road
surface (single vehicle accident) the accidents were
also caused by overtaking and evasive manoeuvres
occasioned by obstacles on the road.
A study conducted in 2007 covering accidents with
personal injury and impact with trees in Styria [15]
also cites another study in which the percentage of
persons killed in tree accidents in Lower Austria for
the years 2001 to 2004 is given as only around 8%
annually related to all persons killed in road
accidents in Lower Austria. The respective
percentages of the individual years being 7.4%
(2004), 9.7% (2003), 7.3% (2002) and 8% (2001).
These percentages are significantly lower than those
that can be derived from the results in [14].
252
According to the study on accidents involving
personal injury and impact with trees in Styria [16]
the number of fatal single vehicle accidents that
occurred in Austria from 1.1.2001 until 31.12.2004
totalled 1,309. 173 cases in which the accident
sketch suggested a tree accident were selected
individually from Statistik Austria. Of these,
19 accidents took place in Styria. These accidents
were supplemented with personal injury accidents in
Styria for the year 2004 and verified in more detail.
In the end, it was possible to subject 39 tree
accidents ending in personal injury (14 fatal crashes
from the years 2001 to 2004 as well as 25 non-fatal
crashes from the year 2004) to more extensive
analysis. This process also established that a large
proportion of the trees on the side where the vehicle
left the road were part of a forest. This tallies with
the fact that, as given in the study, 62% of the entire
surface area of the province of Styria is taken up by
forest. This is well above the Austrian average with
47% of the surface area of the state being covered by
forest. The number of tree-lined-road accidents was
very low with just one case (of the 29 investigated
cases). This obviously constitutes a key difference in
comparison with the known accidents in Germany.
The lack of typical tree-lined roads (alleys) could
thus be a reason why the tree accidents in Austria
represent a statistically less important accident focus
overall than in Germany.
Also, the tree impacts investigated in the Austrian
province of Styria frequently took place on narrow
roads with two lanes and narrow unconsolidated
shoulders on each side. These accident sites in the
lower echelon road network were only rarely
equipped with roadside barriers, curve warning signs
or specific road signs.
In five accidents the vehicles involved crashed into
individual, isolated trees. Four accidents involved an
impact with trees forming part of a row of trees. In
five accidents the impact took place with a tree that
was part of an existing area of greenery consisting of
irregularly distributed trees and undergrowth
confined to a limited area.
In order to supplement the information gleaned from
Austria and also as part of a master's thesis, 115 fatal
tree accidents were selected from the central
database for in-depth analysis of road accidents
(ZEDATU) and evaluated [17]. These accidents took
place across the entire area of the Republic of
Austria in the period between 25th January 2003 and
29th September 2008. The site of 111 of the accidents
could be determined: 80% took place outside urban
areas excluding motorways, 12% inside urban areas
and 8% on motorways. This roughly resembles the
 Impacts with trees alongside the road - just a German problem?
corresponding distribution in Germany, where, for
example, in 2005 of the total of 1,134 fatalities in
tree accidents, 950 (84%) occurred outside urban
areas excluding the Autobahn, 117 (10%) inside
urban areas and 67 (6%) on the Autobahn. As not all
the accidents in the ZEDATU have so far been put
into a format enabling more detailed analysis, the
data contained here does not enable an absolutely
reliable estimation of the tree accidents in the entire
period. For the year 2003 the data base contains
514 fatal accidents, of which 58 accidents involved a
vehicle impacting with a tree. This corresponds to an
11% share.
Overall, 768 road deaths were registered in statistics
in Austria in 2005. The 514 fatal accidents contained
in the ZEDATU thus reflect a large percentage of all
fatal accidents that occurred in Austria in 2005.
Therefore, the figure of 11% of fatalities for tree
accidents can also be seen as a relatively reliable
one. To what extent this percentage increases if only
the fatalities of accidents on country roads are taken
into consideration, is unknown and could be
investigated in a more detailed and lengthy study
(perhaps spanning several years). This likewise
applies to the regional differences in the individual
states of the Republic of Austria. Taking the relevant
findings and supplemental information on the road
network as a basis, it would be possible in the future
to draw up detailed comparisons with the accident
picture in individual German Laender.
2 In-Depth Case Study Analysis
2.1 Supplemental statistics
Apart from the master's thesis already mentioned
[17], a further study conducted at the Baden-
Crashes in Austria
December
November
October
September
August
Month
July
June
May
April
March
February
January
0 5
Relative frequency
Figure 6 Monthly distribution of tree accidents in Germany and Austria (Sources: [17, 18])
Württemberg Cooperative State University looked
into tree accidents in Germany [18]. It evaluated a
total of 47 tree crashes. Relevant information comes
from the accident analyses collated by DEKRA
experts. The accidents occurred in Germany in the
years 1995 to 2011. Of those persons involved in the
accidents, 34 were killed, 39 seriously injured and
6 slightly injured. 77% of the accidents occurred
outside urban areas (excluding the Autobahn), 21%
inside urban areas and 2% on the Autobahn.
Fig. 6 shows the accident distribution of both studies
across the months. These are more or less distributed
over the entire year; deviations from an equipartition
could be coincidental as a result of the low sample,
or also caused by seasonal influences (e.g. in
January).
Fig. 7 shows the distribution of accidents broken
down according to the time of day. In 88% of the
Austrian cases the light conditions could also be
determined at the time of the accident: 53% occurred
in daylight, 4% at twilight, 42% at night without
street lighting and 1% at night with street lighting.
In 107 of the Austrian cases the age of the driver at
the time of the accident was known: 42% were
between the ages of 15 and 25; 32% between the
ages of 26 and 40; 19% between the ages of 41 and
60, and 8% over 60.
Tree accidents are thus for the most part not the
typical night time "disco accidents" involving young
drivers (driving at excessive speeds and perhaps
under the influence of alcohol), but are frequently
accidents that take place during daytime, e.g. during
the journey to and from work, on holiday or for other
leisure pursuit purposes.
Crashes in Germany
N1 = 115 fatal tree crashes
in Austria
N2 = 47 tree crashes with
34 fatalities,
39 severely injured
and 6 slightly injured
in Germany
10 15 20 % 25
253
 Berg Alexander, Ahlgrimm Jörg, Schlosser Stephan, Prof. Dr. Steffan Hermann

Crashes in Germany Crashes in Austria

40
Ante meridiem Post meridiem
%

30
Relative frequency

20

10

0
00 to 6:00 6:01 to 12:00 12:01 to 18:00 18:01 to 00:00 no information
Time

Figure 7 Distribution of tree accidents in Germany and Austria broken down according to the time of day (Sources: [17,
18])

If the lower traffic density at night is also taken into The primary impact with the tree usually involves
account, it must nevertheless be stated that there is a one vehicle side, and the left-hand side (driver side)
great risk of a tree accident occurring at night. is more frequently, Fig. 9. In about one third of the
cases the impact was with the vehicle front. More
Involved in tree accidents were predominantly cars
rarely affected are the vehicle rear or the roof, after
(between 84 and 89%), but also on an individual
the vehicle had already tipped over on its side before
basis motorcycles and trucks as well as buses, Fig. 8.
the collision.

Crashes in Germany Crashs in Austria

100
%
80
Relative frequency

60

40

20

0
Motorcycle Car Truck under Truck more Bus
3.5 tons than 3.5
tonnes
Type of vehicle

Figure 8 Distributions of vehicle types in accidents involving impacts with a tree in Germany and Austria
(Sources: [17, 18])

254
 Impacts with trees alongside the road - just a German problem?

Crashes in Germany Crashes in Austria

50
%
40

Relative frequency
30

20

10

0
Front Side right Side left Rear end Roof
Main impact area on the vehicle

Figure 9 Distributions of main impact areas on the vehicle in accidents involving impacts with a tree in Germany and
Austria (Sources: [17, 18])

2.2 Examples of In-depth case injured passenger and fatally injured driver in light
studies of the enormous impact with the driver door.
An example of a single vehicle accident is given in Examination of the vehicle gave no cause to suggest
Fig. 10. The accident occurred in a rural area on a that technical faults could have contributed to the
Saturday shortly before midnight on a slightly accident. The reason why the vehicle left the road
downward sloping road that had several trees planted surface could not be clearly ascertained.
on one side. A car, a VW Golf IV, travelling through
It is highly likely that the presence of a roadside
a sweeping left-hand curve on a wet road surface
barrier could have prevented the departure from the
became unstable, partially left the road surface to the
road surface and the direct impact with the tree
right, spun in a clockwise direction about its vertical
followed by penetration of the trunk into the
axis and impacted at the area of the driver door
passenger compartment as far as the central tunnel.
against a tree located on the right-hand side of the
As for the vehicle, ESC (Electronic Stability
road. The maximum speed at the accident site is
Control) would have offered a certain potential
limited to 100 km/h. The speed of the vehicle, based
benefit in avoiding the accident or in reducing the
on its initial skid marks, was calculated at over 75 -
severity of the accident outcome. Without trees on
90 km/h, it being assumed that the speed in the curve
the side of the road the vehicle would probably have
was higher.
turned over in the adjacent field, which, in the case
The driver was killed and two passengers seriously of occupants wearing seat belts and if the vehicle had
injured. Both airbags in the vehicle were triggered. come naturally to rest would have generally resulted
Driver and the front passenger were clearly wearing in less serious injuries.
a safety belt. Even though the front airbags were
triggered and safety belts were worn, they proved
unable to offer sufficient protection to the seriously

255
 Berg Alexander, Ahlgrimm Jörg, Schlosser Stephan, Prof. Dr. Steffan Hermann
Figure 10 Example of a single vehicle accident with lateral impact of a car with a tree
Another example of a single vehicle accident is
shown in Fig. 11. The driver of an ambulance car
(based on the VW mini bus) was driving at around
7:35 a.m. in daylight following a slight right-hand
curve onto a straight, slightly upward sloping
country road. The 5.3 m wide road surface was dry
and there was no visual impairment. The permissible
maximum speed was limited to 80 km/h.
Starting on a straight and stable path the vehicle
gradually drifted to the left, departed from the road
surface, and impacted frontally against a tree located
at the beginning of a tree lined road. The collision
speed was determined as ranging between 75 and
88 km/h.
The driver was seriously injured in the accident. A
passenger sitting behind the driver in the second row
and a second passenger (patient) in a wheelchair
sitting likewise on the outside left were killed. All
occupants were wearing their seat belts and correctly
secured. The driver airbag had triggered during the
impact with the tree.
Examination of the vehicle gave no cause to suggest
that technical faults could have contributed to the
256
accident. The slow departure from the road surface
without any counter reaction like braking or steering
suggests that the driver was distracted or had fallen
asleep.
As far as the vehicle is concerned, a Lane Departure
Warning System (LDW) would have been one
measure that could have prevented this accident or
reduced the severity of the outcome. As the road
surface had an unbroken demarcation line marking
(Road Markings, Section 295 of the German Road
Traffic Regulations), LDW would have detected the
vehicle drift in good time and warned the driver
appropriately.
Roadside barriers would definitely have prevented
the vehicle from leaving the road surface and thus
from impacting with the tree. A broader consolidated
shoulder would likewise have contributed to
preventing the accident. Without densely clumped
trees at the side of the road the consequences of the
accident would have been significantly lower
assuming a similar course of events.
 Impacts with trees alongside the road - just a German problem?

Figure 11 Example of a single vehicle accident with frontal impact of an ambulance car with a tree

Fig. 12 shows an example of a fatal motorcycle an evasive manoeuvre, e.g. as in a non-contact

accident involving impact with a tree. The accident
occurred towards 7:50 p.m. in daylight. The accident
site is located outside urban area on a country road
with an area of forest adjacent on the right in the
driving direction of the motorcycle (Suzuki GN 72B)
and to the left isolated trees on the edge of a field.
The rider initially moved over onto the opposite lane
on a straight stretch of road, and then departed from
the road on the left. Motorcycle and rider then
collided whilst still upright and without applying
brakes frontally against a tree situated in an area of
grass on the edge of the road. The road slopes
slightly downwards around the accident area and has
a somewhat non-homogenous surface due to repair
work. The surface was also slightly wet at the time
of the accident. Small oil stains were found on the
road surface.
Examination of the vehicle gave no cause to suggest
that technical faults could have contributed to the
accident. The collision speed was determined as
being around 85 km/h. The reason why the
motorcycle departed from the road surface could not
be clearly ascertained. Possible are an overtaking or
accident with a wild animal. Also possible is a slight
instability of the path of the motorcycle perhaps
acerbated by passing over a patch of oil. This could
have caused the motorcyclist to cramp up, and in
turn prevented him from successfully controlling the
necessary evasive manoeuvre.
As there were no indications of a cyclist reaction
such as braking, for example, a motorcycle Anti
Lock Brake (ALB) would have had no potential
benefit. A motorcycle airbag would definitely have
been triggered and may have helped to lessen the
impact with the tree.
Roadside barriers would probably not have
prevented the motorcycle colliding with the tree;
however, they would have perhaps been able to
influence the path of the vehicle immediately before
the impact so that the motorcyclist would have been
propelled past the tree. If the tree had not existed at
the edge of the road the motorcycle would have
continued its path and gone into the adjacent field.

257
 Berg Alexander, Ahlgrimm Jörg, Schlosser Stephan, Prof. Dr. Steffan Hermann
Figure 12 Example of a single vehicle accident with frontal impact of a motorcycle with a tree
3 Outlook
The recent studies outlined here [17, 18] form part of
an entire corpus of studies on accidents involving an
impact with roadside trees that has undergone
already in Germany since several years [12, 13, 14,
16, 19, 20, 21, 22] now with an international focus.
In Germany, especially, there exist extensive
findings as well as effective proposals, the
implementation of which has already contributed to
obvious improvements in road safety, in particular in
the eastern part of the Federal Republic of Germany.
Despite this progress, however, there is also still
discernible potential for a further reduction in the
number of fatal tree accidents. This potential varies
widely according to region. For example, the
proportion of fatalities from tree accidents on
country roads in Brandenburg in the year 2012 was
still 60 %, despite a fall of 85% in all fatalities from
tree accidents in this area of Germany between 1995
and 2012.
The known findings so far suggest that the problem
of fatalities caused by tree accidents is not so
prominent in Austria, which recorded a share of
around 10%, as in Germany where the share in 2012
was 19%. That said, there still exist relatively little
258
information for the individual Austrian states. What
would be particularly important here would be
further differentiation according to site and type of
road, which would allow perhaps greater percentages
of tree accident fatalities (at a lower level of the
absolute numbers) to be detected. ZEDATU, if it
continues to be systematically expanded, would be a
suitable data source for this.
With road deaths in Austria at 10%, it can be
nevertheless assumed that there is indeed much
potential that the country can tap into make a
considerable contribution to a further reduction of
the number of road users killed. It is also interesting
to know why the problem of tree accidents in Austria
is not so acute as in Germany. Appropriate studies
containing comparison of different federal states and
regions could lead to further interesting findings.
Furthermore, more future research into the incidence
and structure of tree accidents on a pan-European
level should be encouraged. One would hope that the
findings could reveal whether and where a reduction
of tree accidents can support the new strategic goal
for the year 2020 of further reducing the number of
road deaths by 50%. Because the measures suitable
for this are already widely known, corresponding
 Impacts with trees alongside the road - just a German problem?

international studies could also be used to develop
"Best-Practice Guidelines".
And finally, with regard to Vision Zero, tree
accidents also form, over the longer term, part of
those interesting fields of accident research that can
be analysed to uncover residual potential for a
continuous and sustainable lessening of the number
of road deaths.

Sources [8] Statistisches Bundesamt (2013): Todesopfer im

Straßenverkehr: weniger denn je im Jahr 2012.
[1] COMMUNICATION FROM THE COMMISSION:
Pressemitteilung Nr. 068 vom 22.02.2013.
European Road Safety Action Programme. Halving the
Internetquelle (letzter Zugriff am 14.6.2013):
number of road accident victims in the European Union by
https://www.destatis.de/DE/PresseService/Presse/Pressemit
2010: A shared responsibility.
teilungen/2013/02/PD13_068_46241.html.
Online source: lats access: 2013-07-26): http://eur-
lex.europa.eu/LexUriServ/LexUriServ.do?uri=COM:2003:
0311:FIN:EN:PDF. [9] Statistik Austria (2012): Unfälle mit Personenschaden
2011.
Internetquelle (letzter Zugriff am 14. Juni 2013):
[2] COMMUNICATION FROM THE COMMISSION TO
https://www.statistik.at/web_de/statistiken/verkehr/strasse/u
THE EUROPEAN PARLIAMENT, THE COUNCIL, THE
nfaelle_mit_personenschaden/index.html.
EUROPEAN ECONOMIC AND SOCIAL COMMITTEE
AND THE COMMITTEE OF THE REGIONS: Towards a
European road safety area: policy orientations on road [10] Salzburger Nachrichten (2013): So wenig Verkehrstote
safety 2011-2020. wie noch nie in Österreich.
Online source (last access: 2013-07-26) http://eur- Internetquelle (letzter Zugriff am 14. Juni 2013):
lex.europa.eu/LexUriServ/LexUriServ.do?uri=COM:2010: http://www.salzburg.com/nachrichten/oesterreich/chronik/s
0389:FIN:EN:PDF. n/artikel/so-wenig-verkehrstote-wie-noch-nie-in-
oesterreich-41778/.
[3] DVR (2012): Vision Zero – Grundlagen & Strategien.
Schriftenreihe Verkehrssicherheit Nr. 16. Deutscher [11] Kompendium zur 3.Verkehrssicherheitskonferenz des
Verkehrssicherheitsrat DVR, Bonn, 2012. Landes Brandenburg. Ministerium für Stadtentwicklung,
Internetquelle (letzter zugriff am 18.06.2013): Wohnen und Verkehr des Landes Brandenburg, August
http://www.dvr.de/download2/p3046/3046_0.pdf. 1999.

[4] Bundesministerium für Verkehr, Bau und [12] Berg A., Ahlgrimm J. (2010): Baumunfälle – Nach wie
Stadtentwicklung (2011): vor Handlungs- und Forschungsbedarf. Verkehrsunfall und
Verkehrssicherheitsprogramm 2011. Fahrzeugtechnik 48 (2010) Heft 4, S. 118-126.
Internetquelle (letzter Zugriff am 14.6.2013):
http://www.bmvbs.de/cae/servlet/contentblob/74626/public
ationFile/76118/verkehrssicherheitsprogramm-2011.pdf. [13] Kühn M., Bakaba J E (2009): Bekämpfung von
Baumunfällen auf Landstraßen. Verkehrsunfall und
Fahrzeugtechnik 47 (2009) Heft 12, S. 374-380.
[5] Bundesministerium für Verkehr, Innovation und
Technologie (2011): Österreichisches
Verkehrssicherheitsprogramm 2011-2020. [14] Bigler, E. (2003): Baumunfälle – Beitrag zur aktuellen
Internetquelle (letzter Zugriff am 14.6.2013): Diskussion. Analyse von Geschwindigkeiten,
http://www.bmvit.gv.at/verkehr/gesamtverkehr/gvp/faktenb Unfallgeschehen, Unfallfolgen in Alleen. Diplomarbeit,
laetter/sicherheit/vsp2020.pdf. Technische Universität Wien, Forschungsbereich für
Verkehrsplanung und Verkehrstechnik.
Internetquelle (letzter Zugriff am 14.06.2013):
[6] DESTATIS (2013): Fachserie 8 Reihe 7 http://www.ivv.tuwien.ac.at/lehre/archiv-
Verkehrsunfälle Jahresbericht 2012 des Statistischen studentenarbeiten/diplomarbeiten/diplomarbeiten-
Bundesamtes, Wiesbaden, erschienen am 10. Juli 2013 2003/eric-bigler.html.
Internetquelle (letzter Zugriff am 11.Juli.2013:
https://www.destatis.de/DE/Publikationen/Thematisch/Tran
sportVerkehr/Verkehrsunfaelle/VerkehrsunfaelleJ20807001 [15] Statistik Austria, Zeitreihen: Das Unfallgeschehen
27004.pdf?__blob=publicationFile. 1992 bis 2011 nach Bundesländern.
Internetquelle (letzter Zugriff am 14.06.2013):
https://www.statistik.at/web_de/statistiken/verkehr/strasse/u
[7] DESTATIS (2013a): Verkehrsunfälle Zeitreihen nfaelle_mit_personenschaden/index.html.
Jahresbericht 2012 des Statistischen Bundesamtes,
Wiesbaden, erschienen am 10. Juli 2013
Internetquelle (letzter Zugriff am 11. Juli 2013): [16] Wruß C. (2007): Personenschadenunfälle mit Anprall
https://www.destatis.de/DE/Publikationen/Thematisch/Tran an Bäumen in der Steiermark. Diplomarbeit, Institut für
sportVerkehr/Verkehrsunfaelle/VerkehrsunfaelleZeitreihen Verkehrswesen, Department für Raum, Landschaft und
XLS_5462403.xls;jsessionid=BC9EF10AE893B37CE2144 Infrastruktur, Universität für Bodenkultur Wien, 2007.
C511C584BE5.cae2?__blob=publicationFile.
259
 Berg Alexander, Ahlgrimm Jörg, Schlosser Stephan, Prof. Dr. Steffan Hermann

[17] Wallner B. (2013): Untersuchung zum aktuellen

Unfallgeschehen mit Anprall an Bäumen neben der [22] Ministerium für Infrastruktur und Landwirtschaft
Fahrbahn. Master Thesis, TU Graz, Institut für (MIL) des Landes Brandenburg (2011): Baumunfälle in
Fahrzeugsicherheit, noch unveröffentlicht. Brandenburg. Analyse der Baumunfälle mit Getöteten 2008
bis 2010, Berlin, Oktober 2011.
[18] König M: Dokumentation von Verkehrsunfällen mit Internetquelle (letzter zugriff am 20.06.2013:
Anprall an Bäumen neben der Fahrbahn. Studienarbeit, http://www.mil.brandenburg.de/sixcms/detail.php/533484.
Duale Hochschule Baden-Württemberg, Mai 2013.

[19] Meewes V, (2001): Aufprallgeschwindigkeiten,

Unfälle und Unfallfolgen von Baumunfällen. Mitteilung
Nr. 39 des Instituts für Straßenverkehr (ISK) Köln des
GDV, Februar 2001.
Contacts
Alexander Berg, Jörg Ahlgrimm, Stephan Schosser,
DEKRA Automobil GmbH, Handwerkstraße 15, 70565
Stuttgart, GERMANY

[20] Eid D, Ellinghaus D, Funck P., Koch H., Manssen G.,

e-mails:
Meewes V., Neumann K., Peters J. (2005): Schutz von alexander.berg@dekra.com,
Mensch und Baum. Ergänzende Hinweise zu den ESAB. joerg.ahlgrimm@dekra.com,
GDV Gesamtverband der Deutschen stephan.schlosser@dekra.com
Versicherungswirtschaft, Berlin, September 2005
phone: +49 711 7861 0

[21] Mennicken C (2004): Schwere Unfälle an Landstraßen

mit Alleen in Brandenburg, Mecklenburg-Vorpommern Prof. Dr. Hermann Steffan, Technische Universität Graz,
und Niedersachsen. VSVI-Seminar „Entwurf und
Institut für Fahrzeugsicherheit, Inffeldgasse 23/1, 8019
Gestaltung von Straßenverkehrsanlagen“ am 19.02.2004 in
Hannover. Graz, AUSTRIA
Internetquelle (letzter Zugriff am 20.06.2013): e-mail:
http://www.ivh.unihannver.de/peb/de/Mitarbeiter/ehemalig h.steffan@tugraz.at
e/bachmann-
Dateien/veroeffentlichungen/schwere_Unfaelle_in_Alleen. phone: +43 316 873 30300
pdf.

260
© 2013 by EVU
EVU 2013 – 30

The importance of active control systems on accident

avoidance
Dr. Kakalis Leonidas
Dr.Sabbioni Edoardo, Prof. Cheli Federico

Abstract
When the adhesion limit is reached, the vehicle becomes difficult to control and drivers might not be able to avoid
dangerous situations since they are not experienced in dealing with these operating conditions.
To this end, vehicle safety systems have been significantly developed in the automotive industry in the past years. Most
of these systems regulate torque independently on the wheels to generate a yaw moment which is able to improve
vehicle stability and traction. ABS and ESP act only on braking torques, while active and semi-active differentials
regulate the driving torques.
In the present paper numerical results will be presented from ISO manoeuvres, demonstrating the benefits of the active
control systems such as active differential and ESP on critical situations over a passive vehicle. In addition, the use of a
sophisticated driver model can undoubtely provide some meaningful conclusions with regard to the driver’s actual inter-
action with the control system.

Introduction driving situations just before the accident occurs

In recent years TCS (Traction Control Systems) and
vehicle stability control systems like ESP, VDC,
have become more and more common in the market.
As these systems have limitations, a tendency to use
additional systems which help to overcome these li-
mitations and to add additional features can be
observed. Torque management devices (TMD) like
limited-slip differentials and on-demand couplings
offer the required improvements. In particular,
actively controllable TMD complements brake-based
traction control and stability control systems and
allow further improvements [1],[2].
showed that the steering, thus an evasive manoeuvre,
took place in approximately 55% of all accidents
[12], [13], [14], [15].
Other important statistics produced by the German
Association of Insurance Companies (GdV) revealed
that severe accidents in most of the cases involve
spinning cars (Fig. 1). These studies indicate that
most severe accidents involve vehicles spinning in
dry roads at speeds between 60km/h to 100km/h in
contradiction with the belief that the loss of control
is more likely to happen on slippery roads and at
high speeds [16].

In critical driving situations the tyre slip angles, and

consequently, the vehicle slip angle might increase
rapidly without a corresponding increase in lateral
force if a vehicle reaches its physical limit of
adhesion between tyres and the road. In the
aformentioned vehicle state, an average driver can
neither judge the friction coefficient of the road nor
the grip of the tyr remaining and therefore an attempt
to stabilise the vehicle and avoid a possible accident
can become cumbersome and in many cases is not
expected. If the adhesion limit is reached, in most
cases the driver panics and reacts in an irregular
way: excessive steering, excessive braking.

According to statistics, drivers are only responsible Fig 1: Severe accident caused by car spinning.
in 19% of accidents, vehicles in 31% and the envi-
ronment in 50% of all cases. Other more detailed in- TCS have been designed to stabilise the vehicle even
vestigation concerning the driver’s activity in critical in situations with panic reactions. These systems

261
 Dr. Kakalis Leonidas, Dr.Sabbioni Edoardo, Prof. Cheli Federico
have the goal of improving the traction and stability
of the vehicle by actively controlling the wheel slip
and consequently the vehicle yaw moment. The
control strategy behind these systems is referred to as
yaw moment control. Yaw moment can be generated
either by applying braking forces at the wheels or by
varying the torque split created by the differential
mechanism. In the present paper, a critical investiga-
tion of their function is going to be held presenting
their advantages and limitations [3], [4], [5].
TCS and torque vectoring systems
One of the most prevalent TCS is the electronic sta-
bility program, commonly known as ESP. Steering
of a vehicle results in a yaw moment which results in
a directional change. Excessive steering input during
a critical manoeuvre, can translate into a considera-
bly large vehicle side slip angle, and may lead to the
saturation of the tyre’s lateral and longitudinal force,
making the vehicle unstable and the chance of main-
taining control very small. Under these conditions,
the ESP system maintain the rotational stability of
the vehicle by applying independently controlled
braking forces on the individual wheels thus generat-
ing a compensatory artificial yaw moment (Fig. 2).
Additionally, the system extends its intervention by
controlling the engine torque whenever it recognises
that a more decisive correction must be done [6].
Fig 2: The impact of the drive torque distribution over the
understeering vehicle.
The desired yaw moment is based on a slip control-
ler, which needs a tiye model to calculate tyre slip
and contact force variation. Therefore, the actual and
optimal tyre slip (the longitudinal slip corresponding
to the maximum of the force – slip curve) must be
identified as they represent crucial factors in the con-
trol’s logic function (Fig. 3).
262
Driver Desire Car Motion
+ -
Car Motion
Controller
Brake & Traction
Slip Controller
Hydraulic Model
Pressure Controller
Hydraulic
Unit
Fig 3: ESP Control logic block diagram.
The value of the optimal longitudinal slip depends on
the tyre characteristics (size), the adherence condi-
tions, tyre wear and pressure. In order to test the ro-
bustness of the control logic empirical rules are in-
troduced and experimentally tuned.
An important attempt to bypass the speculative prin-
ciples of the ESP control logic and make its function
more efficient and accurate is the introduction of the
smart tyre. The aim of this system is to turn the tyre
into a sensor, able to provide direct information
about the contact dynamics and thus the tyre road
contact force variation. It is straightforward to under-
stand that tyre road contact interaction would allow a
significant enhancement of active systems since they
would be able to effectively respond and adjusts their
parameters before the driver is even aware of the ad-
herence change. It is clear that the combined on
board use of an ESP strategy and the smart tyre make
the vehicle extremely agile and consequently safer.
On the other hand, systems which have a great im-
pact on the yaw-side slip handling characteristics, are
the differential systems. The majority of the latest
systems employ a friction clutch (LSD) which pro-
vides a connection between the two driveshafts. In
contrast with the open differential, in non-uniform
condition of adherence of the wheels, these mecha-
nisms always transfer torque from the fastest wheel
to the slower one, imposing an asymmetric longitu-
dinal contact force distribution. Passive self-locking
differential are currently adopted in high perfor-
mance vehicles in order to improve traction and sta-
 The importance of active control systems on accident avoidance
bility in critical conditions like power on, power off
or μ – split road segments.
An important breakthrough is considered the use of
electronically controlled components where it is
possible to govern the yaw moment amplitude and
actuation time, transforming a passive device to a
semi-active one. in this case the kinematics of the the
driven wheels dictate the direction of the transferred
torque [7].
The key drawback in terms of controllability of
conventional differentials is that they will always
transfer torque to the slowest rotating wheel. For
yaw control applications, it is desirable to have
control over the direction as well as the magnitude of
the torque transfer. Active differentials make this
desirability a reality (Fig. 4). This way, the vehicle
can also be stabilised in the case of quick swerving
manoeuvres without having to brake. The steering
system responds more directly – with less effort and
fewer required corrections [8].
Fig 4: ZF’s active differential torque distribution.
Recently an attempt was made to combine the torque
vectoring interventions via combined brake and en-
gine interventions. Such a sophisticated system
promises to offer not only lateral stability but also
lateral and longitudinal promptness. The perfor-
mance of such a complex system will not be exam-
ined as its outputs exceed the purposes of this paper
[9].
The developed control logic which engages or disen-
gages the proper clutch is often based in a state
recognition algorithms which basically identify the
dynamic state of the vehicle and subsequently trig-
gers the appropriate state strategy (Fig. 5).
Steady-State, Power On
Step Steer
State Locking Torque
Recognition Target
Power Off Strategy
Fig 5: Control logic main structure.
Moreover the implemented control strategy for the
active differential has similar architecture to the ESP
(Fig. 3), taking into consideration the reference and
the actual motion.
Vehicle model and simulation tools
For the vehicle’s dynamic simulation there are
numerous models available, each one with different
characteristics: starting from simplified models like
the single-track models to more complicated models
with several degrees of freedom. The latest are the
optimum from computational speed point of view
and provide a user-friendly interface which in many
cases guarantees a simplified connection process
between the model and the control logic. It is
obvious that during their validation with experimen-
tal results, derived by the actual vehicle, deviations
may occur and therefore they should be interpreted
with extreme caution by the researcher.
A 14 degrees of freedom model (IPG CarMaker®) of
the examined vehicle has been used to test and
compare the performance of various control systems;
the rear driving vehicle model has been integrated
with the actuator (correspondant transfer function for
the hydraulic systems) and control logic model. The
control logic has been implemented in
MATHWORK SimulinkTM. In the present vehicle
model, the contact forces between the tyres and the
road are described through the MF-Tyre model. The
entire model has been validated by comparing its
outputs to experimental data relevant to the passive
vehicle equipped with a rear free differential. Fig. 7
and Fig. 8 depict the understeer curve and the
sideslip angle in relation with the lateral acceleration,
comparing the experimental with the numerical
results [10].
One of the most important features used in the pre-
sent study is the IPG – Driver. The IPG – Driver
helps to develop promising concepts by allowing
driving that will have very similar handling charac-
teristics when compared to a real driver. It offers
tools which enable the user to add the control actions
of a human driver to complete vehicle simulations.
263
 Dr. Kakalis Leonidas, Dr.Sabbioni Edoardo, Prof. Cheli Federico

These actions include the steering, braking, throttle
position, gear shifting and clutch operation. The vir-
tual driver results in highly accurate linear conditions
but is also proven to be adequate to reconstruct real-
istic behaviours on the limit area.
ATI 90-90 manoeuvre will be held in open loop and
a double lane change in closed loop.
In an attempt to demonstrate the function of the acti-
ve systems an ATI 90–90 manoeuvre carried out at
80km/h is used. During the whole manoeuvre the
steering wheel angle for all models remains the same
(Fig. 9). The adherence level was high (μ = 0.85) and
the maximum lateral acceleration of about 7m/s2 was
reached. In Fig. 10 the vehicle’s side slip angle is
reported for all three models while in Fig.11, the yaw
rate time history is compared for all three models
again.

100

80

60

40

20

SWA [deg]
0

-20

Fig 7: Steering pad constant radius: understeer curve. -40

-60

-80

-100
9 10 11 12 13 14 15 16
Time [sec]

Fig 9: ATI 90-90 manoeuvre, steering wheel angle time

history.

8
Passive
Active Diff
6
ESP

4
Sideslip Angle [deg]

-2
Fig 8: Steering pad constant radius: Side slip angle curve.
-4

The virtual driver actions are: -6


-8
Course options within the lane borders and 9 10 11 12 13
Time [sec]
14 15 16

choice of the driving speed according to the

Fig 10: ATI 90-90 manoeuvre, side slip angle time histo-
course and the vehicle behaviour;
ry.
 Automatic adaptation to the vehicle by iden- 30
Passive
tifying its current dynamic behaviour with Active Diff
the ability to store it, which gives the possi- 20
ESP

bility of use of this knowledge in later simu-

lations; 10
Yaw Rate [deg/s]

Performance evaluation
-10

In this section an attempt will be made to examine -20

the performance of the aforementioned systems from

a more quantitative perspective. Models with the -30
9 10 11 12 13 14 15 16
Time [sec]
ESP and the active differential are going to be
compared with the passive model. The results of an Fig 11: ATI 90-90 manoeuvre, yaw rate time history.

264
 The importance of active control systems on accident avoidance

It is obvious that two models equipped with the ac-
tive systems have a significant advantage over the
passive model as their activation permits to the vehi-
cle to remain within the margins of stability and not
to spin. Their crucial difference with the passive
model derives from their ability to moderate the re-
sultant longitudinal force (Fig. 12) and create an arti-
ficial yaw rate counterbalancing the tendency to
oversteer. In Fig. 12 the rear longitudinal forces are
presented for the three examined models. As can
been seen the active differential, by engaging the ap-
propriate clutch, moderates the driving torque to the
appropriate semi-axle creating the desired torque
split, while the function of the ESP brakes the appro-
priate wheel (Fig. 13) creating brief but again notice-
able corrections to the vehicle dynamic’s yaw mo-
ment.
be used to set a predictive model in real conditions.
This process can imitate actual driving conditions as
it includes the system “driver – vehicle – environ-
ment” and therefore can be considered as a useful
tool to evaluate the steer input, the control strategy
and to subjectively analyse the entire vehicle’s be-
haviour. In order to obtain a realistic behaviour from
the virtual driver, a prediction of the vehicle’s
movement is made. Therefore a correct reaction to
the prediction allows the driver to apprehend the dy-
namic behaviour of the vehicle. By using the IPG –
Driver tools, this is accomplished in general with a
learning procedure, where for example a yaw re-
sponse time is identified. With this information the
driver is able to adapt his control behaviour and cal-
culate the correct feedback signals in steering in each
situation. [17]

2500 In the next figures the results of a double lane change

2000 manoeuvre are presented. The vehicle holds a fixed
1500 velocity of 120km/h through the entire manoeuvre
1000
and the ground adherence is set to μ = 0.85. In Fig.
Longitudinal Force [N]

500
14, Fig. 15 and Fig.16 the steering wheel angle, the
sideslip angle and yaw rate are reported respectively.
0
As can be seen, all three vehicle models remain con-
-500
trollable and successfully remain within the accepta-
-1000 Passive L
Passive R
ble track margins.
-1500 Active Diff L
Active Diff R
-2000 80
ESP L Passive
ESP R Active Diff
-2500 60
9 10 11 12 13 14 15 16 ESP
Time [sec]
40
Fig 12: ATI 90-90 manoeuvre, longitudinal forces time
history. 20
SWA [deg]

0
70
Front Left
Rear Left -20
60 Front Right
Rear Right -40
50
-60
[Mpa]
Pressure [bar]

40
-80
0 50 100 150 200 250 300
Pressure

Distance [m]
30
Fig 14: Double lane change, speed 120km/h, μ = 0.85:
20 steering wheel angle distance history.

10

0
9 10 11 12 13 14 15 16
Time [sec]

Fig 13: ATI 90-90 manoeuvre, wheel brake pressure.

Similar results can be obtained by performing closed

loop tests. In particular a severe double lane-change
manoeuvre is herein presented. During this test the
driver must not only observe the given lanes but
mantain the vehicle path within the track limits. The
basic idea is to create a test method for a transient
road – holding ability characterisation which could
265
 Dr. Kakalis Leonidas, Dr.Sabbioni Edoardo, Prof. Cheli Federico

5
Passive
vehicles deviate from the target path substantially. In
4 Active Diff real conditions, such deviation may be prove fatal
ESP
3
and this should therefore be taken into consideration.
2
The different response between the two active sys-
tems can be adjusted by different tuning and the vir-
Sideslip Angle [deg]

1
tual driver’s correspondent adjustment during the
0
systems’ activation.
-1 6
Target Path
-2 Passive
5
Active Diff
-3 ESP
4
-4
3
-5
0 50 100 150 200 250 300

Y [m]
Distance [m] 2

Fig 15: Double lane change, speed 120km/h, μ = 0.85:

1
sideslip angle distance history.
0
20
Passive -1
Active Diff
15
ESP
-2
0 50 100 150 200 250 300
10 X [m]

Fig 17: Double lane change, speed 120km/h, μ = 0.85:

Yaw Rate [deg/s]

Trajectory.
0 4
Passive
-5 Active Diff
3
ESP

-10 2
Sideslip Angle [deg]

-15 1

-20 0
0 50 100 150 200 250 300
Distance [m]
-1
Fig 16: Double lane change, speed 120km/h, μ = 0.85:
yaw rate distance history. -2

-3
A look at the aforestated results, shows a noticeable
difference between the passive and the active models -4
0 50 100 150 200 250 300
in the sideslip angle and yaw rate peak values. These Distance [m]

considerable reductions greatly influence the driver’s Fig 18: Double lane change, speed 120km/h, μ = 0.5: side-
feeling relating to the vehicle response and provides slip angle distance history.
the driver with the opportunity to negotiate the lane 6
change with more confidence. Target Path
Passive
5
Active Diff
ESP
In Fig. 17 the trajectories of all models are plotted in 4
respect to the target path. In these conditions no no-
3
ticeable advantage in terms of lateral distance is no-
X [m]

ticed. 2

The importance of the presented active systems can 0

be further underlined by performing the same proce- -1

dure by setting the road adherence in lower levels (μ
= 0.50). Without any control systems the vehicle -2
0 50 100 150 200 250 300
Y [m]
spins immediately after the test is initiated. On the
other hand the other two vehicles manage to main- Fig 19: Double lane change, speed 120km/h, μ = 0.5: tra-
tain their steerability. Although the vehicles under jectory.
these controls succeed in completing the test, the
adopted trajectory (Fig. 18) is judged unsatisfactory.
In relation with the high adherence manoeuvre, the

266
 The importance of active control systems on accident avoidance
Conclusions
An attempt is made in this study to point out the
benefits of the active systems over the passive one.
The presented results have demonstrated clearly how
greatly their activation can influence the vehicle’s
overall dynamic performance. Under the considered
operating conditions, the studied mechanisms have
developed a great improvement in terms of yaw rate
and sideslip amplitude and peak and amplified the
stability margins. The driver can surely handle the
vehicle with minor effort (reduced steering wheel
corrections). On the other hand, under low adherence
the results have shown that although the active sys-
tems have positively altered the vehicle’s perfor-
mance, they still need significant improvement as
they required to constrain the vehicle into narrower
travelled path.
Obviously the problem of the adherence level identi-
fication remains. This difficult task can be handled
through the definition and the implementation of a
self-tuned recognising algorithm based on systems
like smart tyres. Particularly in the case of low ad-
herence a prompt recognition of the surplus grip can
definitely be used for further improvement of lateral
stability and active safety.
References
[1] Y. Hou., Z. Yu., The Evaluation of ABS Perfor-
mance, SAE Technical Paper 2003-01-0255.
[2] Y. Seto, T. Watanabe, T. Kuga, Y. Yamamura, K.
Watanabe, Research on a Braking System for Re-
ducing Collision Speed, SAE Technical paper
2003-01-0251.
[3] Y. A. Ghoneim, C. L. William, Y.K. Chin, D. M.
Sldlosky, Enchased Traction Stability Control
System, SAE Technical Paper 2005-01-1591.
[4] S. Anwar, A Predictive Control Algorithm for a
Yaw Stability Management System, SAE Tech-
nical Paper 2003-01-1284.
[5] S. Matsumoto, H. Yamaguchi, H. Inoue,
Y.Yasuno, Improvement of Vehicle Dy-
namics Through Braking Force Distribution, SAE
Technical Paper 1992.
[6] A. T. van Zanten, Bosch ESP Systems: 5 Years of
Experience, SAE Technical Paper 2000-01-1633.
[7] F. Cheli, M. Pedrinelli, F. Resta, A. Zorzutti, M.
Zanchetta, G.Travaglio, Development of a Control
Strategy for a Semi-Active Differential for a High
Performance Vehicle, SAE Technical Paper 2007-
01-0927.
[8] C.Granzow, W.D. Gruhle, M.Spiess, R.Denzier,
D. Baasch, R. Peter, Fahrprazision durch Torque
Vectoring – Das neue Achsgetriebe von ZF, Euro-
pean All – Wheel Drive Congress Graz, Proceed-
ings 2007.
[9] M.Knoop, A.Erban, F.Flehmig, Dynamic Wheel
Torque Control – DWT Agility enchancement by
networking of ESP® and Torque Vectoring, Euro-
pean All – Wheel Drive Congress graz, Proceed-
ings 2007.
[10] H.B Pacejka, Tire and Vehicle Dynamics, Second
Edition.
[11] E.Sabbioni, F.Cheli, S. Melzi, Development of an
ESP Control Logic Based on Force Measurements
Provided by Smart Tires, SAE Technical Paper
2013-01-0416.
[12] W.-D. Kappler, Beitrag zur Vorhersage von
Einschatzungen des Fahrverhaltens, VDI –
Fortschritt – Berichte, Reihe 12, Nr.198, 1993.
[13] G.W Brown, Analysis of 104 Eastern Iowa Motor
Vehicle Casualty Accidents, Proceedings of the
Third Triennial Congress on Medical and Related
Aspects of Motor Vehicle Accidents, Ann Arbor,
Michigan, Highway Safety Research Institute
1971, pp 216 – 218.
[14] K. Rompe, B. Heissing, Moglichkeiten zur
Bewertung der Fahreigenschaften, K. Rompe,
Bewertungsverfahren fur die Sichereit von
Personenwagen, Koln, Verlag TUV Rheinland
1984, pp 243 – 265
[15] M.L. Edwards, S. Malone, Driver Crash Avoid-
ance Behavior., in Driver Performance Data Book,
Washington DC, National High-way Traffic Safe-
ty Administration, Final Report DOT HS 807 121,
1987.
[16] K. Langwieder, Mit ESP schwere Unfalle
vermeiden oder mildern, ESP – Workshop, No-
vember 10, 1999, Boxberg, Germany.
[17] International Standard, ISO 3888 - 1, Passenger
cars – Test track for a severe lane – change ma-
noeuvre, Part 1: Double lane – change.
[18] Car Maker© Version 2.1 – IPG Driver user’s
guide, IPG.
Contacts
Leonidas Kakalis, PhD
EVU Member Hellas
I.Tselepi, 67
54352, Thessaloniki, Hellas
leonidas.kakalis@gmail.com
Edoardo Sabbioni, PhD
Politecnico di Milano
Via La Masa, 1
20156 Milano, Italy
edoardo.sabbioni@polimi.it
Federico Cheli, Full Professor
Politecnico di Milano
Via La Masa, 1
20156 Milano, Italy
federico.cheli@polimi.it
267
 Dr. Kakalis Leonidas, Dr.Sabbioni Edoardo, Prof. Cheli Federico

268
© 2013 by EVU
EVU 2013 - 31

Traffic accidents caused by electronic failure

Uzi Raz

Abstract
Failures of vehicle components that caused road accidents used to be categorised as: design failures, manufacturing fail-
ures, maintenance problems or fatigue etc. The usual solutions used by the manufacturers were either recall campaigns
to replace failed components with a similar type, or changes in maintenance procedure instructions.
With the development of many new electronic devices and computers in vehicles, these new elements became critical to
the vehicle's proper operation. The whole operation of the vehicle is controlled by computers. The computers sense the
environment, the vehicles status and the driver's instructions, relying on various sensors, all of which have electronic
outputs. According to these readings, the computers operate the vehicle elements. Faults in any of these elements can
and do cause accidents.
This paper will analyse an accident which occurred due to a failure of an electronic component. The vehicle involved
was a Mitsubishi Grandis. The accident caused the death of eight occupants, only one passenger survived this accident.
After the analysis of this accident, we searched for other cars, either the same type or with the same electronics, which
had a similar problem.
We found a few cars that were of the same make, but from a different manufacturing year, and some other cars with the
same electronics. All the examined cars demonstrated the same faulty electronic accelerator.

Introduction
The traffic accident statistics shows that about 10%
of accidents are caused by faulty vehicles. In Israel,
vehicles are randomly checked by the authorities on
the road. Naturally, the inspectors choose to check
the cars that look dilapidated.
The inspection yielded the results shown in fig.1.
About 60% of all vehicles inspected were found to Figure 1. Vehicle test on the road data.
be faulty and 25% of them were not permitted to
continue on the road due to their bad condition. The
conclusion of this is that there is a problem with ve-
hicle condition.
Up until recently, vehicle safety was divided into
two main features:
1. Accident prevention, called active safety
2. Prevention of injury in the car, called passive
safety.
In modern vehicles, both active and passive safety
rely on electronic equipment.
Vehicle malfunction was historically related to the
following categories:
 Vehicle design, see fig.2
 Vehicle manufacturing, see fig.3
Figure 2. Vehicle design failures – example. Component
 Modifications, see fig.4 too weak.
 Maintenance, see fig.5
 Component age, see fig.6

269
 Uzi Raz
Figure 3. Failures due to manufacturing. Faulty welding.
Figure 4. Failure due to modification. Engine change from
gasoline to LPG
Figure 5. Failure due to maintenance. Bad maintenance of
the exhaust system. Accident was caused by the exhaust
fumes.
All those malfunctions are mechanical, and can still
occur and cause accidents in a modern vehicle.
However, in a modern vehicle, most of the safety
features, such as steering, suspension, acceleration
and deceleration are electronically controlled.
270
Figure 6. Accident due to component age. Age-fatigue.
The electronic features improve vehicle efficiency
but introduce new additionaltypes of failures. The
electronic failure is of a different type altogether.
These failures are not easily detectable during nor-
mal maintenance procedures, or while driving.
Electronic components behave along the "bath tub"
curve, meaning that most failures occur during the
first or the last stage of operation, see fig.7. The
failures occur suddenly, without prior warnings.
Fig. 7. Electronic components failure according
“Bath Tube” distribution function.
The dangerous failures may occur due to mal-
functions of computers and sensors of systems such
as:
 Engine control
 ABS - ASR
 ESP
 Air bags
For example, the accelerator pedal of a modern vehi-
cle is a combination of a mechanical pedal, an elec-
tronic sensor and a connector, see fig.8. When a
driver presses the accelerator pedal, the movement of
the pedal is translated to an electronic signal, which
is transmitted to the computer controlling the engine.
The engine computer reacts in accordance with its
program and controls the amount of air and fuel mix-
ture into the engine via the throttle, see fig.9. The air
and fuel mixture supplied to the engine depends on
the status of the accelerator pedal.
 Traffic accidents caused by electronic failure
Figure 8. Elements of accelerator pedal
Figure 9. Throttle operation and sensor
Eventually, the power train consisting of the engine
and gear-box will achieve equilibrium and the speed
of the vehicle will be constant.
In a vehicle with an automatic gear-box, pressing the
accelerator pedal to the floor causes the gear-box
computer to down shift the gear-box achieving a
"kick-down", meaning a strong acceleration.
Methods
The analysis of failures in a modern vehicle after an
accident must consider the possibility of electronic
components failure and their effect on the perfor-
mance of the vehicle. When suspecting a certain
electronic failure, it is difficult, or even impossible,
to reconstruct the events that were caused by the
failure. However, sometimes, corrective action is
needed quickly, to prevent additional accidents of the
same type.
In aircraft accidents, for example, all planes of the
type involved in an accident can be grounded imme-
diately, pending the outcome of the inquiry. In traffic
accidents, this is not the case, since it is technically
impossible to stop all vehicles of the same type.
Failures resulting in accidents can be attributed to
many factors. The electronic failure can be attributed
to: oxidation, defective connections of electrical
components, temporary shorts or opens in the cir-
cuits, software bugs, temporary environmental fac-
tors, or it can be attributed to the driver's error.
In traffic accidents, isolating a device or electronic
component for failure analysis starts by forming a
possible cause theory. Given a specific failure mode,
there are a limited number of possible explanations;
by determining which of these is the most plausible,
an analyst may begin putting together a theory that
will help limit the number of possible failing compo-
nents. It may not always be possible to pinpoint the
cause of failure in a component, but recording all de-
tails and findings can help eliminate possible causes
and focus attention on those areas that may be the
most effective in correcting the problem.
Failure analysis conclusions should include objective
recommendations for corrective actions. This is a re-
quirement of the ISO9000 quality system, which
most companies adhere to.
Example:
A Mitsubishi Grandis with an automatic gearbox
(fig.10) was involved in a fatal accident in which
eight occupants had been killed and only one child
survived.
Figure 10. Mitsubishi Grandis (2009)
When the driver of the car approached a sharp curve
to the right, he lost control of the car. The car hit a
centre reserve barrier, was thrown over the barrier,
crossed the opposing traffic lanes, rolled downhill
off the side of the road into a ravine and burst into
flames. All this, while the driver was trying to get
some help from the police. The media reported that
"The driver called police from the vehicle prior to
the accident upon realising that the car's brakes had
ceased to function. According to police, the dis-
patcher who took the man's call heard the crash oc-
cur while he was on the line with the driver."
271
 Uzi Raz

According to the police, the driver phoned the police

more than four (4) minutes prior to the impact. He
reported that he could not brake the car and surmised
that he had lost his brakes. He reported that the car
was accelerating instead of braking, even though he
was not pressing the accelerator pedal and pressed
only the brake pedal.
An eye witness, standing at an intersection, reported
that the vehicle involved in the accident passed the
intersection and its wheel hubs were "flaming red
like being on fire".
Figure 12. Reconstruction of vehicle fall

Data collection

The following were found at the site of the accident:

 Marks of the vehicle hitting and going over
the central barrier, see fig.11.
 Signs of the vehicle falling off the road into
the ravine.
 Signs of the vehicle hitting the ground in the
ravine for the first time.
 Signs of a tree that was hit by the vehicle
during its fall see fig.13 point 1.
 Location of the final rest position of the ve-
hicle, see fig.12 & 13 point 3.
Figure 13. Pictures of points 1-3 in fig. 12
 The burnt vehicle.
The road leading to the accident site, from a distance
of eleven (11) km to the site itself, was carefully ex- Data interpretation
amined. The map in fig.14 shows this section of the
road. Marked on the map are road inclines in the The vehicle's velocity was deducted using the road
same direction as the vehicle involved was going. marks and throwing distance, as shown in fig.12. At
the moment in which the vehicle left the road and
fell down the ravine, it was travelling at a speed of
97 kph (27 m/s).
Calculating the energy loss when hitting the barrier
and travelling over it yields a velocity of 111.2 kph
when approaching the accident site.

Figure 11. Road marks

Figure 14. Map of accident’s scene

The vehicle's engine produces 120 kW when the ac-

celerator pedal is pressed down. This implies that
each wheel absorbs 30 kW.
272
 Traffic accidents caused by electronic failure
At the speed of 111.2 kph, during four (4) minutes,
the vehicle travels a distance of approximately 7.4
km.
As shown in fig.14, the distance between point 1 and
point 5 is 7.4 km. At point 5, the driver realised that
he had no brakes. On his way to point 4, he went
downhill. From point 4 to point 3 the distance is 2.5
km and the road is uphill at 1-2%, see fig.15. A vehi-
cle travelling this section in neutral gear, starting at
120 kph (at point 4), can stop before arriving at point
3.
Figure 15. Road from point 4 to point 3
There is an intersection near point 3, where the eye
witness stood. This means that while driving uphill,
the vehicle could not stop, even though the driver
pressed the braking pedal hard.
The road from point 3 to point 1 via point 2 is down-
hill, as shown in fig.16.
Analysis
The driver felt that he had lost his brakes when was
going downhill. However, when he had the chance to
let the vehicle reduce speed on the uphill section, he
did not succeed in doing so although he tried.
The excessive use of the brake can be surmised from
the eye witness testimony.
Figure 16. Road from point 2 to point 1
From the police report we know that the car was ac-
celerating going uphill, against the driver's actions to
brake the car.
All the above points to the possibility that the car's
engine computer was locked in an accelerating state.
Because the car underwent an excessive fire after the
fall, it could not be determined which element of the
electronic circuit was faulty, thus causing the acci-
dent.
In order to compare this phenomenon, we searched
for similar cases in Israel. We found two drivers who
experienced the same situation but managed to over-
come it unharmed. According to them, the problem
was in the accelerator pedal, since by playing with it
a few times, it was released and the acceleration
stopped.
We checked two more fatal accidents in Israel in
which the drivers survived. They claimed that a sud-
den acceleration occurred without them changing the
pressure on the accelerator pedal.
Note: This case is similar to cases in the U.S.A,
where a few Toyotas were involved in accidents be-
cause of sudden acceleration problems.
Conclusions
There are various possibilities for this kind of failure
to happen:
1. Mechanical failure of the accelerator pedal.
2. Failure of the accelerator pedal sensor.
3. Electrical connections failure between the
sensor and the engine computer.
4. Engine computer failure, both hardware and
software.
It is important to arrive at a final conclusion as to the
true cause of this malfunction, since the components
nowadays are manufactured for the car industry by
outside manufacturers and the same components are
used by several car manufacturers in several models.
Possible solution
A temporary solution to this problem is adding a
safety feature to the computer program which will
prevent such an occurrence. The safety feature rec-
ommended will cause the car engine to idle when the
engine computer receives conflicting signals of both
acceleration and braking.
References
[1] Google maps and pictures
273
 Uzi Raz

[2] Finn Jensen. Electronic component reliability. Wiley

1995

[3] Mattila T.T. Toward comprehensive reliability testing

of electronic components.11th International Conference
on Thermal, Mechanical & Multi-Physics Simulation, and
Experiments in Microelectronics and Microsystems.
EuroSimE 2010.
[4] Mitsubishi Grandis, data and specifications, Internet

[5] Reliability of electronic components. Murata Newslet-

ter. E-mail magazine

[6] Titu-Marius.; Băjenescu, I.; Marius, I. Bâzu. Compo-

nent reliability for electronic systems. Artech House 2010

Keywords
Failure, Accident analysis, Electronic Component,
Electronic sensors, Engine computer, Acceleration.

Contact
Prof. (PhD P.Eng.) Uzi Raz
The Israeli college for security and investigation
Hashaham,18
49170, Petach-Tikva, Israel
uziraz@gmail.com
tel: +972-3-9229991

274
22th EVU Annual Meeting
Firenze 2013

Proceedings