Hermes Final Report PDF

FEHRL Report 2006/01
Harmonization of European
Routine and research
Measuring Equipment
for Skid Resistance
Title: Harmonization of European Routine and research Measuring Equipment for

Skid Resistance
Authors: G. Descornet – BRRC, Belgium

B. Schmidt – DRI, Denmark
M. Boulet – LCPC, France
M. Gothié – LCPC, France
M-T. Do – LCPC, France
J. Fafié – DWW, Netherlands
M. Alonso – CEDEX, Spain
P. Roe – TRL, United Kingdom
R. Forest – TRL, United Kingdom
H. Viner – TRL, United Kingdom
Keywords: Roads, Skid Resistance, Measurements, Harmonization.
Although the FEHRL has done its best to ensure that any information given is accurate,
no liability or responsibility of any kind (including liability of negligence) is accepted in this
respect by FEHRL, its members of its agents.
© FEHRL 2006
ISSN 1362-6019
All rights reserved. No part of this publication may be reproduced, stored in a retrieval
system, or transmitted in any form or by any means, electronic, mechanical,
photocopying, recording or otherwise, without the prior written permission of FEHRL.
FEHRL Secretariat: S. Phillips

Boulevard de la Woluwe, 42
Bte 3, B-1200 Brussels
Belgium
Tel: +32 2 775 82 38

Fax: +32 2 775 82 29
E-mail: [email protected]
Web site: www.fehrl.org
HERMES Final Report
iii
Table of Contents
FOREWORD .................................................................................................................................................................... V
THE HERMES PROJECT ................................................................................................................................................. V
SPECIFIC OBJECTIVES..................................................................................................................................................... V
W ORK PROGRAMME SUMMARY...................................................................................................................................VI
FINANCING......................................................................................................................................................................VI
RESEARCH CONSORTIUM............................................................................................................................................ VII
STEERING COMMITTEE ................................................................................................................................................ VII
PARTICIPANTS IN THE FIELD TRIALS ........................................................................................................................ VIII
1 INTRODUCTION..................................................................................................................................................1
2 THE HERMES CALIBRATION EXERCISES ............................................................................................4
2.1 GENERAL DESCRIPTION OF TRIALS............................................................................................................... 4
2.1.1 Overview ....................................................................................................................................................4
2.1.2 Participating devices...............................................................................................................................5
2.1.3 Trial Planning...........................................................................................................................................8
2.1.4 The trials – general principles.............................................................................................................11
2.1.5 The trials – Round 1...............................................................................................................................15
2.1.6 The trials – Round 2...............................................................................................................................18
2.1.7 The trials – Round 3...............................................................................................................................20
2.1.8 The test surfacings..................................................................................................................................20
2.1.9 Discussion of practical technical issues.............................................................................................22
2.2 NON-TECHNICAL ISSUES .............................................................................................................................. 27
2.2.1 Use of questionnaires.............................................................................................................................27
2.2.2 Logistics...................................................................................................................................................29
2.2.3 Safety issues.............................................................................................................................................31
2.2.4 Financial aspects....................................................................................................................................32
2.3 SUMMARY OF THE HARMONISATION PROCESS ......................................................................................... 33
2.3.1 Calculation of the IFI using the PIARC-model.................................................................................33
2.3.2 From IFI to EFI......................................................................................................................................34
2.3.3 Determination of the EFI ......................................................................................................................34
2.3.4 The calibration procedure....................................................................................................................35
2.4 EXPERIMENTAL RESULTS............................................................................................................................. 37
2.4.1 Introduction.............................................................................................................................................37
2.4.2 Characteristics of the test surfaces......................................................................................................37
2.4.3 Texture measurements ...........................................................................................................................38
2.4.4 Friction device-specific parameters....................................................................................................41
2.4.5 The database collected..........................................................................................................................42
2.4.6 Repeatability of F ...................................................................................................................................42
2.4.7 Reproducibility of F ...............................................................................................................................44
2.4.8 Curve fitting on F(S)..............................................................................................................................47
2.4.9 Deriving S 0 from MPD ..........................................................................................................................48
2.4.10 Applying statistical tests to the EFI measurements.....................................................................48
2.4.11 Calibration results............................................................................................................................49
2.4.12 Discussion ..........................................................................................................................................52
3 REVISION OF THE FRICTION MODEL ..................................................................................................55
3.1 INTRODUCTION ............................................................................................................................................. 55
3.2 FURTHER IMPROVEMENTS TO THE MODEL................................................................................................ 55
3.2.1 Calculation of S 0 vs. MPDISO per device............................................................................................55
3.2.2 Quadratic regression analysis .............................................................................................................58
3.2.3 The LCPC-model....................................................................................................................................63
3.2.4 Investigation of alternative characteristics of surface texture depth............................................80
3.3 DISCUSSION ................................................................................................................................................... 85
4 ALTERNATIVE PROCESSING OF EXPERIMENTAL DATA ..........................................................88
FEHRL
HERMES Final Report
iv
4.1 INTRODUCTION ............................................................................................................................................. 88

4.2 IMPROVED MODEL FOR S0 VS. MPD........................................................................................................... 88
4.3 STATISTICAL TESTS...................................................................................................................................... 90
4.4 SCENARIO #2: APPLYING NEW STATISTICAL TESTS.................................................................................. 91
4.5 SCENARIO #3: APPLYING SIMPLER ST ATISTICAL TESTS........................................................................... 92
4.6 SCENARIO #4: REVERSING THE REGRESSION CALCULATION.................................................................. 97
4.7 SCENARIOS #5 TO #7: FORCING “A” TO ZERO.........................................................................................100
4.8 SCENARIO #8: TREATING BFC- AND SFC-DEVICES SEPARATELY .......................................................105
4.9 SCENARIOS #9 TO #11: IMPROVED MODEL FOR F(S).............................................................................106
4.10 A NALYSIS OF DEVIATIONS.........................................................................................................................110
4.11 SCENARIO #12: USING ACTUAL VALUES OF S0 INSTEAD OF PREDICTIONS..........................................111
4.12 REPEATABILITY OF EFI.............................................................................................................................112
4.13 SCENARIO #13. A VERAGING OUT THE REP EATABILITY DEVIATIONS OF EFI.....................................114
4.14 RECAPITULATION .......................................................................................................................................119
4.15 DISCUSSION .................................................................................................................................................120
5 ALTERNATIVE APPROACHES TO REFERENCE LEVELS ......................................................... 122
5.1 INTRODUCTION ...........................................................................................................................................122
5.2 A SINGLE REFERENCE DEVICE...................................................................................................................122
5.2.1 Purpose.................................................................................................................................................. 122
5.2.2 Approach............................................................................................................................................... 122
5.2.3 Historic and technical background .................................................................................................. 123
5.2.4 European needs.................................................................................................................................... 124
5.2.5 Proposed specification requirements............................................................................................... 128
5.3 REFERENCE SURFACES...............................................................................................................................131
5.3.1 The purpose of reference surfaces.................................................................................................... 131
5.3.2 Requirements for reference surfaces................................................................................................ 131
5.3.3 Past experience and existing practice.............................................................................................. 132
5.3.4 Possible solutions................................................................................................................................ 136
5.3.5 Outline specification for calibration reference surfaces .............................................................. 141
5.3.6 Further research.................................................................................................................................. 144
6 CONCLUSIONS ................................................................................................................................................ 146
7 RECOMMENDATIONS................................................................................................................................. 150
7.1 REVIEW OF THE CURRENT SITUATION......................................................................................................150
7.2 SPECIFIC RECOMMENDATIONS..................................................................................................................151
REFERENCES ............................................................................................................................................................. 152
ANNEX A: GLOSSARY OF TERMS .................................................ERROR! BOOKMARK NOT DEFINED.
ANNEX B: EXPERIMENTAL RESULTS-GRAPHS OF F(S) ...ERROR! BOOKMARK NOT DEFINED.
ANNEX C: EXPERIMENTAL RESULTS-GRAPHS OF S0 V MPD..........ERROR! BOOKMARK NOT
DEFINED.
: EXPERIMENTAL RESULTS-CALIBRATION GRAPHS <<EFI>> V EFI ...... ERROR! BOOKMARK
NOT DEFINED.
ANNEX E: LCPC-MODEL CONSTANTS........................................ERROR! BOOKMARK NOT DEFINED.
ANNEX F: ALTERNATIVE PROCESSING-GRAPHS OF S0 V MPD ........ERROR! BOOKMARK NOT
DEFINED.
ANNEX G: ALTERNATIVE PROCESSING-CALIBRATION GRAPHS <<EFI>> V EFI....... ERROR!
BOOKMARK NOT DEFINED.
ANNEX H: ALTERNATIVE PROCESSING-GRAPHS OF F V S ...............ERROR! BOOKMARK NOT
DEFINED.
ANNEX I: ALTERNATIVE PROCESSING-CALIBRATION GRAPHS <<EFI>> V <EFI> ... ERROR!
BOOKMARK NOT DEF INED.
FEHRL
HERMES Final Report
v
ANNEX J: REVISED DRAFT FOR ANNEX B OF PREN 13036-2 ...............ERROR! BOOKMARK NOT
DEFINED.
ANNEX K: REFERENCE DEVICE QUESTIONNAIRE ............ERROR! BOOKMARK NOT DEFINED.
ANNEX L: PROPOSED SPECIFICATION FOR A REFERENCE DEVICE..... ERROR! BOOKMARK
NOT DEFINED.
ANNEX M: SYMBOLS AND ABBREVIATIONS .........................ERROR! BOOKMARK NOT DEFINED.
Foreword
The HERMES Project
HERMES – “Harmonization of European Routine and research Measuring Equipment for

Skid Resistance” was a FEHRL-funded pre-normative project that aimed to lay the
foundations for consistent European standardisation of skid resistance measurement on
roads and runways. Underlying the work was the idea that the project should give the
initial impetus to the implementation of a common scale of friction, the so-called EFI
(European Friction Index), together with an associated harmonised calibration procedure,
in all European countries.
Specific objectives
The project had five specific objectives relating to theoretical, practical and longer-term
aspects of harmonising skid resistance measurements in Europe, namely:
1. To improve the model on which EFI is based, which relates the friction coefficient
to the measurement speed, slip ratio and road surface texture, by taking account
of recent developments in this field.
2. To demonstrate the reliability and feasibility of the procedure proposed by
CEN/TC227/WG5 for inter-laboratory calibrations of the various devices used in
Europe.
3. To produce a revised draft standard incorporating the findings from items 1 & 2.
4. To provide FEHRL with a practical scheme for setting up an organisation for the
calibration of skid resistance devices in Europe.
5. To submit draft specifications for a reference device and/or reference surfaces to
the member laboratories of FEHRL with a view to developing the next generation
of standards.
FEHRL
HERMES Final Report
vi
Work programme summary
To meet those five objectives, the following tasks were identified:
0 Scientific management.
1 Improvement of models.
2 Evaluation of proposed calibration procedures and their organisation.
2.1 Organisation of calibration exercises.
2.2 Analysis of the results.
2.3 Evaluation of non-technical aspects.
3 Definition of references for the future.
3.1 Specifications for a reference device.
3.2 Specification for reference surfaces.
Financing
The members of the HERMES Working Group thank the following institutions for their
contributions to the financing of this research:
Belgium: Ministère fédéral des Affaires Economiques - Division Compétitivité

NG III
Belgian Road Research Centre (BRRC)
Denmark: Danish Road Institute
Danish Road Directorate
France: Ministère de l’Equipement, du Logement et des Transports
Ministère de l’Education Nationale, de la Recherche et de la
Technologie
Netherlands: Directorate-general of Public Works and Water Management (RWS)
Spain: Centro de Estudios y Experimentación de Obras Públicas (CEDEX)
United Kingdom: Highways Agency
FEHRL
HERMES Final Report
vii
Research consortium
The following consortium of FEHRL laboratories was responsible for carrying out the
research. Individual laboratories took responsibility for leading specific tasks within the
programme and all contributed to other tasks.
Laboratory Role
BRRC (BE) Project co-ordinator, task 0
Belgian Road Research Centre Contributor task 2.1
Leader task 2.2
CEDEX (ES) Contributor task 2.1
Centro de Estudios y Experimentación de Obras Contributor task 3.1
Públicas Contributor task 3.2
DRI (DK) Leader task 2.3
Danish Road Institute Contributor task 3.1
Contributor task 3.2
DWW (NL) Contributor task 2.1
Dienst Weg- en Waterbouwkunde Leader task 1
LCPC (FR) Contributor task 1
Laboratoire Central des Ponts et Chaussées Contributor task 2.1
Leader task 3.1
Contributor task 3.2
TRL (GB) Contributor task 1
Transport Research Laboratory Leader task 2.1
Leader task 3.2
Steering committee
The project was carried out by a FEHRL Working Group, the membership of which is as
follows:
Guy DESCORNET (Convenor) BRRC Belgium

Bjarne SCHMIDT DRI Denmark
Michel BOULET LCPC France
Michel GOTHIÉ LCPC France
Minh-Tan DO LCPC France
Jeannot FAFIÉ DWW Netherlands
Marta ALONSO ANCHUELO CEDEX Spain
Peter ROE TRL United Kingdom
Rozenn FOREST TRL United Kingdom
Helen VINER TRL United Kingdom
FEHRL
HERMES Final Report
viii
Participants in the field trials
In addition to the work of the main research consortium, the following organizations sent
their equipment and teams to participate in calibration exercises several times, and in
different countries, during the course of the project. The Steering Group offer their
wholehearted thanks for these contributions, without which the project could not have
been carried out 1.
Organization Country Particular thanks to:

BRRC - Belgian Road Research Centre Belgium Michel GORSKI
MET - Ministère wallon de l’Equipement et
des Transports – Division de Structures Belgium Guy LEFÈBVRE
routières
DRI – Danish Road Institute Denmark B. Roland JENSEN
CETE - Centre d’Etudes Techniques de
France Michel GOTHIÉ
l’Equipement de Lyon
STBA – Service Technique des Bases Jean-Claude
France
Aériennes DEFFIEUX
United
TRL - Transport Research Laboratory Paul FÈVRE
Kingdom
DWW - Rijkswaterstaat - Dienst Weg- en
Netherlands Jeannot FAFIÉ
Waterbouwkunde
Staten Vegvesen Norway Bjorne Ove OFSTAD
NPRA – NRRL – Norwegian Road
Norway Per Harald HANSEN
Research Laboratory
IBDIM – Road and Bridge Research
Poland Bogumil SZWABIK 2
Institute
CEDEX – Centro de Estudios y Marta ALONSO
Spain
Experimentación de Obras Públicas ANCHUELO
1
The experimental data collected in this project will remain the property of FEHRL. They can be
made available on request to FEHRL-member laboratories as well as other non-FEHRL
participants in field trials.
2
The authors note with sadness that Mr Szwabik died before the project was completed. The
HERMES team would like to acknowledge his enthusiastic personal contribution to the field trials
and offer their condolences to his colleagues, family and friends.
FEHRL
HERMES Final Report
1
1 Introduction
The skid resistance of a road surface is affected by a great many factors, some to do with
the road and others to do with the measurement technique and conditions.
In Europe more than twelve different types of device are used to measure skid resistance.
Some make use of different principles and those that use similar principles may have
detail differences that affect their relative measurements. Devices may be operated at
different speeds and this, too affects the results. The effect that speed has on the
measurements depends upon a property of the road surface known as “macro texture” or
texture depth. This is a major factor that contributes to the differences between
measurements made by different devices.
Historically, each country has used one or two types of device and the requirements of
their standard tender specifications and maintenance policies have been set to match
them. The opening up of the single market, however, has created a need for
harmonisation, to encourage the development of homogeneous conditions of safety from
one country to the next. This harmonisation has been undertaken in CEN group
TC227/WG5. One of the objectives of that group is to develop a draft standard defining a
uniform procedure to determine skid resistance from a dynamic measurement.
It is unrealistic to expect such a procedure to rely on a single device if it is to be

acceptable to a majority of countries. That is why a philosophy has been adopted that
there should be a means of converting results delivered by different devices to values on
a common, harmonised, scale so that individual countries can continue to use their own
method – at least during a period of transition – but at the same time offering them the
possibility of relating their measurements to those used elsewhere.
However, because skid resistance is not constant and is affected by so many factors,
there is no absolute measure of friction that can be used to compare different skid-
resistance measurement devices. Therefore, a scale is needed that, in effect, can use the
“average” of all devices as a surrogate for the “correct” answer. If individual devices could
be linked to this scale in a way that takes account of their particular characteristics, then
the required harmonisation could be achieved.
Anticipating this need, in 1992, PIARC conducted an international experiment to compare

different devices and methods in use to measure friction and surface texture on road and
airfield pavements [1]. All the devices and methods used in Europe at that time were
represented. The experiment showed that virtually all of the participating devices could be
harmonised using a single equation to relate their outputs to a common scale, provided
that an allowance was made for a measurement of macro texture. The proposed scale
was called the International Friction Index (IFI). There were concerns, however, that this
scale was not sufficiently precise to be of effective practical value in Europe where so
many different devices were used.
A further analysis of the database from this experiment and additional tests were
performed in 1997-1998 under a contract with the Belgian Federal Government [2]. The
objective was to adapt the definition of IFI to relate more specifically to the sub-set of
devices used in Europe and to update it by taking into account new road surfacing
materials (for example porous asphalt, stone mastic asphalt and slurry seals) not covered
by the PIARC experiment.
FEHRL
HERMES Final Report
2
This resulted in a proposal for a European Friction Index (EFI) that would, potentially,
provide the harmonisation sought between the various types of device used in Europe.
The proposed EFI, building on the experience gained from the PIARC experiment and
Belgian research, uses the overall average of all devices used in Europe to define the
common scale and uses mathematical models that represent the effect of texture and
speed to adjust the measured values made at a particular test speed by a particular
device to their equivalent on this common scale at a reference speed. How this is done is
explained in more detail in Chapter 2.3 of this report.
It was also recognised that, for the system to operate effectively, a means of regularly
calibrating individual devices to the scale and for verifying that the scale itself remained
stable was also needed. This would also need to make provision for new devices to be
added to the system when necessary.
Based on this work, in 1998 a proposal was submitted to CEN/TC227/WG5 that defined
both the EFI 3 and a procedure for calibrating measuring devices based on that index. The
proposal was included in a draft standard in preparation, as an informative Annex [3].
However, after several examinations and subsequent revisions of the draft, the group
considered it necessary to test the proposed calibration procedure before formalising it in
a standard. The questions that remained unanswered related mainly to the reproducibility
of the EFI-values delivered by the various devices, the real magnitudes of the drifts of the
devices and, consequently, the required frequency of calibration needed to maintain the
desired accuracy. Non-technical aspects such as costs and any difficulties in practice also
required examination. Further, in the intervening period, new research results had become
available that might permit some improvements in the precision of EFI by allowing for an
additional parameter, namely, the slip ratio between the test tyre and the tested surface.
Meanwhile, the FEHRL working group on “European Harmonisation of Friction, Texture

and Evenness Measurements” had been mandated as early as in 1997 by the FEHRL
Board of Directors to work on the development of basic specifications for a European
reference device for use with the next generation of skid resistance standards. As there
were doubts about the real possibility of relying on EFI, that alternative option was to be
seriously considered despite its own inherent difficulties. One such difficulty is that such a
device has itself to be calibrated somehow, which requires some friction standard(s) to be
available. Therefore, the possibility of using reference surfaces for that purpose had to be
investigated as well.
The HERMES project was therefore conceived by the FEHRL working group to address
these issues, with specific objectives as set out in the Foreword to this report. The work
would be carried out by a core team of experts from laboratories represented in this
group, assisted by contributions from other FEHRL laboratories and organisations in
member countries.
3
The Draft prEN-13036-2 [3], proposes the term Skid Resistance Index (SRI) to describe the
harmonised scale, instead of European Friction Index (EFI). However, because EFI has been used
in several previous publications, and in order to avoid confusion in readers’ minds, EFI is also used
in this report. A change to using the term SRI may be eventually necessary if the prEN retains that
terminology and when it is used as a standard.
FEHRL
HERMES Final Report
3
This report presents the results of the project in five main Chapters.
Chapter 2 describes the experimental part of the project that aimed to test the feasibility of
the proposed calibration procedures under realistic conditions. Not only technical activities
(Section 2.1) but also practical aspects (Section 2.2) are evaluated. Section 2.3 explains
briefly the way in which the harmonisation process in the draft standard operates, and
how the associated calculations are made. Finally, Section 2.4 summarises the
measurements made and gives an initial analysis of the results obtained when the
procedures set out in the draft standard were followed.
Chapter 3 presents the results of various attempts to improve the mathematical models on
which the definition of EFI is based, with a view to possible improvements to the precision
and reliability of the index itself.
Chapter 4 presents a number of alternative ways of processing the results from the
calibration exercises (described in Section 2.3) that were considered with a view to
optimising the reproducibility of EFI as delivered by the different participating devices.
Chapter 5 looks to the future. Section 5.1 proposes specifications for a future single
standard skid resistance measurement device, based upon a consultation of users of
current equipment and detailed discussions within the project steering group. Section 5.2
discusses the background to past work in the field and the future possibilities for
designing, building and maintaining stable and reproducible reference surfaces for
calibrating friction measurement devices.
FEHRL
HERMES Final Report
4
2 The HERMES calibration exercises
2.1 General description of trials
2.1.1 Overview
As explained in Chapter 1 of this report, the EFI, in the absence of an absolute reference
value, is based upon the idea of relating different measurement devices to a common
scale that itself represents an overall “average” of all devices. However, because devices
could change over time, it is essential to ensure that the index itself does not change
significantly or gradually drift and that individual types of device are regularly calibrated to
the index.
To achieve this, the prEN [3] requires calibration exercises, in which different
combinations of a few devices would meet to re-establish the links with the main index
and keep the index itself consistent with the current status of the devices. This process
would lead to devices being established as “reference devices 4” that had been calibrated
to the index through one of these meetings, allowing other similar devices in national
fleets to be calibrated to the reference device and hence to the index. In due course, new
devices could join in a calibration trial and then themselves join the set of “reference
devices” for subsequent occasions.
The prEN [3] anticipated that such meetings would occur annually for reference devices.
The purposes of the HERMES calibration exercises (referred to subsequently in this

chapter and elsewhere by the shorthand “trials”) were therefore twofold:
To assess the practicality of following the procedure set out in the prEN.
To provide data that would allow the stability and precision of the EFI to be assessed
across a realistic range of European devices.
To consider the precision of the EFI, the programme therefore needed to include a range
of devices that would have to meet in different combinations at different times. The first
round of tests would allow the device-dependent coefficients to be recalculated (Section
2.3.12). Subsequent rounds of trials would provide for further refinement of the
coefficients and the index.
To assess the stability of the index over time, a number of trial cycles would be necessary.
To assess this within a practical overall project time, it was decided to accelerate the
programme of trials to mimic a three-year cycle, but within a one-year timetable. A series
of trials was therefore planned, with three “rounds”, each containing three individual
meetings.
In principle, each participating device would attend one trial in each round and, generally,
visit a different location and meet with a different group of devices on each occasion. This
4
The term “reference device” here is used specifically to mean devices that have been through this
process of calibration against the skid resistance index and on which the index is based. It does not
have the same meaning as an absolute reference against which other devices can be compared,
except as a means of relating other devices of the same type to the index in what is known as a
“Type-3 calibration”.
FEHRL
HERMES Final Report
5
would enable the variation in the device coefficients to be checked and gradually refined
and also assess any inherent variation in the EFI. More details of the way in which this
was achieved are given in Section 2.1.3
2.1.2 Participating devices
Altogether, fifteen skid resistance-measuring devices and eight texture-measuring devices

took part in the HERMES project. Some devices were initially considered to be “reference
devices” because they had taken part in the 1992 PIARC International Experiment. This
meant that initial values for the regression coefficients were available and their results
provided the overall average on which the EFI would be based.
Other devices were newcomers to the set, but once they had participated at one exercise
they joined the set of reference devices for the later meetings. The set of friction devices
originally invited to participate covered practically the full range of systems currently in use
on roads in Europe, including both longitudinal (BFC) and transverse (SFC)
measurements with slip ratios from 14% to 100% (locked wheel). Table 2.1 provides a
brief description of each skid resistance device participating. One device used on airfields
in France was added later in the programme.
In order to apply the mathematical models to harmonise the devices, it was necessary to
have measurements of texture depth (as Mean Profile Depth, MPD) for the test surfaces.
Measurements were made both by the organisers of each trial and by participating
devices where they were capable of measuring MPD. The various texture devices used
are listed in Table 2.2. Some were fitted to the friction device, others were separate, but
all of the texture devices complied with ISO 13473-1, 1996.
FEHRL
HERMES Final Report
6
Table 2.1 Skid resistance devices taking part in the HERMES trials
Device Operator Organisation Key Operating

Code Thumb-nail Photograph
Name (Country) Principles*
86% Fixed Slip.

PIARC Radial smooth tyre
DWW DWW Rijkswaterstaat
F01 at 200 kPa.
Trailer (NL)
0,5 mm water film
thickness.
Locked Wheel.
CETE de Lyon smooth PIARC tyre at
F02 ADHERA (FR) 180kPa. 1,0 mm water film
thickness.
Side Force 20º wheel

angle.
CEDEX Avon SCRIM smooth tyre
F03 SCRIM
(ES) at 350 kPa.
0,5 mm water film
thickness.

angle.
MET Avon SCRIM smooth tyre
F04 SCRIM
(BE) at 350 kPa.
0,5 mm water film
thickness.
15% fixed slip.

254 mm (10”) dia. smooth
MET
F05 GripTester tyre at 138 kPa.
(BE) 0,5 mm water film
thickness.
Variable slip device run at

20% fixed slip.
DRI
F06 ROAR ASTM 1551 smooth tyre.
(DK)
0,5 mm water film
thickness.

86% fixed slip.
DWW Rijkswaterstaat
F07 ROAR ASTM 1551 smooth tyre.
(NL) 0,5 mm water film
thickness.

angle.
PIARC ASTM E525 88
BRRC
F08 Odoliograph smooth tyre.
(BE)
0,5 mm water film
thickness spread by
preceding tanker.
FEHRL
HERMES Final Report
7

Locked wheel.
ASTM E524 smooth tyre
TRL
F09 PFT at 200 kPa.
(GB)
1,0 mm water film
thickness.

18% fixed slip.
(NPRA/NRRL) ASTM E524 smooth tyre
F10 OSCAR
(NO) at 207 kPa.
0,5 mm water film
thickness.

18% fixed slip.
Statens Vegvesen
F11 ROAR II ASTM E1551 smooth tyre.
(NO)
0,5 mm water film
thickness.
Locked wheel.
Commercial patterned tyre
IBDIM
F12 SRT-3 at 200 kPa.
(PL)
0,5 mm water film
thickness.

angle.
Avon SCRIM smooth tyre
TRL at 350kPa.
F13 SCRIM
(UK) Water flow 0.95l/s giving
0,5mm water film
thickness at 50km/h and
0,25mm at 90km/h.
angle.
PIARC ASTM E525 88
MET
F14 Odoliograph smooth tyre.
(BE) 0,5 mm water film
thickness spread by
preceding tanker.
15% Fixed slip.

STBA Ssmooth PIARC tyre.
F15 IMAG
(FR) 1,0 mm water film
thickness.
FEHRL
HERMES Final Report
8

Tyre pressures are given in SI units to the nearest kPa based upon information provided by the
Operating Organisation (some operators use Imperial (UK) or Metric units). The water film
thickness given here is the theoretical water film thickness at which the device normally operates
when wetting the road for a test, as advised by the Operating Organisation. (For some devices
this is controlled automatically by varying the flow rate according to the test speed, for others a
fixed flow rate may be used that results in a slight variation in water film thickness at different
operating speeds. In most cases in the HERMES trials, apart for the first machine on the first test
run, the road surface had already been wetted by preceding devices).
Table 2.2 Texture devices used in HERMES trials

Name of
Device Code Nationality Operator Organisation
Equipment
T1 GREENWOOD DK Danish Road Institute
T2 ARAN NL DWW Rijkswaterstaat
T3 Rugolaser FR CETE de Lyon
T4 GEOCISA ES CEDEX
T5 HARRIS GB TRL
T6 ROAR NO Statens Vegvesen
T7 SCRIMTEX GB TRL
T8 Stat TX meter BE BRRC
2.1.3 Trial Planning
So that the project would include a representative sample of European highway surfaces
and encompass the wide range of environmental conditions across Europe, the trials were
planned to take place in more than one country. Sites on test tracks and sites on public
roads would be included in order to assess the advantages and disadvantages of each
type of location. Holding the trials in more than one country would also allow a wider
range of important non-technical issues to be assessed than would be possible if the trials
were held in just one country.
It was arranged that five of the Steering Group members would host or arrange trial
exercises in their countries. These would be TRL (United Kingdom), CEDEX (Spain),
DWW (Netherlands), LCPC (France) and BRRC (Belgium), with overall co-ordination by
TRL, the Task Leader. As Task Leader, the TRL representative on the Steering Group
attended all of the trials apart from the last two, either as an independent observer or as a
device operator. This provided for continuity of observation at all of the different locations
(every member of the Steering Group attended at least one of the trials).
The trial schedule was constrained by the overall duration of the project and the need to
simulate the calibration periods specified in the prEN [3]. It was also necessary to provide
a realistic interval between the first and last set of trials so that any drift in the results of
FEHRL
HERMES Final Report
9
any one device could be identified and realistically incorporated in the data set, thus
ensuring that the derived procedures would be robust.
Overall, nine trial exercises would be held in three trial rounds, with three meetings in
each round. Each host organisation would arrange two trials, except CEDEX who would
only host one trial. The trial rounds were planned for autumn 2001, spring 2002 and
autumn 2002. The number of trials required was based on the number of participating
devices, allowing most devices to participate at least once in each trial round. Some
devices participated more than once in a trial round, usually when they were attached to
the host organisation for one of the trials.
The order of the trials was chosen to reduce the possibility of disruption due to bad
weather. For example, in the first round, which took place during autumn 2001, the United
Kingdom trial was held first, followed by the Netherlands and then Spain. Even so, heavy
rain affected several of the trials and snow on the roads in the preceding week almost
forced the postponement of the Spanish trial. In Belgium, BRRC worked through MET (the
Ministère Wallon de l'Equipement et des Transports) that carried out the detailed
organisation and ran the two trials there. Table 2.3 summarises the trial round
programme.
Table 2.3 Summary of trial rounds

Round, Host, No. of devices No. of
Dates Location and type of site test
Trial Country Friction Texture sections
1,1 TRL, GB 23-25/10/01 5 1 Crowthorne, TRL test track 8
Joure and Lelystad,
1,2 DWW, NL 29-31/10/01 4 2 6
Motorway + test track
Valencia area, Motorway,
1,3 CEDEX, ES 20-22/11/01 4 2 6
main road
Nivelles and Borders areas,
2,1 MET, BE 19-21/03/02 7 3 Motorway, main road, local 7
roads
2,2 LCPC, FR 26-28/03/02 5 1 Nantes, LCPC test track 6
Cuijk area, Motorway, main
2,3 DWW, NL 23-25/04/02 6 3 7
roads
3,1 TRL, GB 01-03/10/02 3 3 Crowthorne, TRL test track 8
Nivelles and Borders areas,
3,2 MET, BE 08-10/10/02 6 1 Motorway, main road, local 7
roads
3,3 LCPC, FR 15-17/10/02 7 2 Nantes, LCPC Test track 6
At the outset of the work it was recognised that there could be more devices available
than was necessary for the purposes of the project and that some might have constraints
on their availability that would rule them out. Therefore, an initial letter was sent to all of
the FEHRL Laboratories introducing the project and inviting them to consider taking part.
The letter included a general proposal regarding the trial schedule and the level of
commitment being requested. Each potential participant was asked to identify dates that
were most suitable for them. They were also asked to indicate whether they would be able
to attend any particular country. The final decision as to which devices would be invited to
join in the project and which trials they would attend rested with the project Steering
Group.
FEHRL
HERMES Final Report
10
Based on the response from this initial approach, a participation matrix was planned. In
principle, each device would attend three trials, in three different countries. In order to
reduce costs, one of these would be the trial of its host country where possible. In
addition, “host” devices would also attend subsequent trials in their own countries in
addition to the visiting devices.
The matrix was initially designed to ensure that the sets of devices that met at any one
trial were mixed in relation to measurement type and in such a way as to avoid repetitions
of meeting sets on subsequent occasions. In the event, equipment failures and other
operational commitments meant that some re-arrangement and substitution was
necessary and some devices experienced difficulties that prevented them from taking part
fully in all the trials that they attended.
Table 2.4 lists the devices attending each trial. Table 2.5 shows the full participation
matrix, indicating the trials where each device met the other devices.
Table 2.4 Devices attending each trial
Texture devices
Trial Reference Country Friction Devices attending
attending
1,1 UK F09, F10, F11, F12, F13 T5
1,2 Netherlands F01, F04, F07, F05, F06 T1, T2
1,3 Spain F01, F03, F02, F08 T3, T4
2,1 Belgium F01, F06, F09, F10, F05, F14 T6
2,2 France F02, F05, F08, F12, F15 T3
2,3 Netherlands F01, F03, F04, F11, F13, F15 T4, T6, T7
3,1 UK F01, F06, F13 T1, T5, T7
3,2 Belgium F04, F05, F13, F15 T6, T8
3,3 France F02, F03, F04, F05, F10, F11, T3, T4
F12
FEHRL
HERMES Final Report
11
Table 2.5 Participation matrix – trials attended by each device
Device F01 F02 F03 F04 F05 F06 F07 F08 F09 F10 F11 F12 F13 F14 F15
1,2 1,2
1,3 1,2 2,3
F01 1,3 2,1 2,1 1,2 1,3 2,1 2,1 2,3 2,1 2,3
2,3 2,1 3,1
2,3 3,1
1,3 2,2 1,3 2,2
F02 1,3 3,3 3,3 3,3 2,2
3,3 3,3 2,2 3,3
1,3 1,3 2,3 2,3
F03 3,3 1,3 3,3 3,3 2,3 2,3
2,3 3,3 3,3 3,3
1,2
1,2
2,3 2,1 1,2 2,1 2,3 2,3 2,1 2,3
F04 2,1 3,3 1,2 3,2 2,1 3,3
3,3 3,2 2,1 3,3 3,3 3,2 3,2 3,2
2,3
3,3
1,2
1,2 2,2 2,1 1,2 2,2 2,1 2,2 2,1 2,2
F05 3,3 1,2 2,1 3,3 3,2
2,1 3,3 3,2 2,1 3,2 3,3 3,3 3,2 3,2
3,3
1,2
1,2 1,2
F06 2,1 1,2 2,1 2,1 3,1 2,1
2,1 2,1
3,1
F07 1,2 1,2 1,2 1,2
1,3 2,2 2,2
F08 1,3 1,3 3,2 2,2 3,2 3,2
2,2 3,2 3,2
1,1
F09 2,1 2,1 2,1 2,1 1,1 1,1 1,1 2,1
2,1
2,1 2,1 1,1 1,1 1,1
F10 2,1 3,3 3,3 2,1 1,1 2,1
3,3 3,3 2,1 3,3 3,3
2,3 2,3 1,1 1,1 1,1
F11 2,3 3,3 3,3 1,1 2,3
3,3 3,3 3,3 3,3 2,3
2,2 2,2 1,1 1,1
F12 3,3 3,3 2,2 1,1 1,1 2,2
3,3 3,3 3,3 3,3
2,3 2,3 1,1 2,3
F13 2,3 3,2 3,1 3,2 1,1 1,1 1,1 3,2
3,1 3,2 2,3 3,2
2,1 2,1
F14 2,1 2,1 3,2 2,1 2,1 3,2 3,2
3,2 3,2
2,3 2,2 2,2 2,3
F15 2,3 2,2 2,3 2,3 2,2 3,2
3,2 3,2 3,2 3,2
2.1.4 The trials – general principles
The general trial procedure was standardised, although some local variation was
necessary in the light of experience or to take account of local circumstances or events as
they developed. Generally, each trial was planned to take place over a working week. In
principle, the Monday and Tuesday mornings were available for the participants to travel
to the location ready to attend a briefing and familiarisation session on the Tuesday
afternoon. The main test runs were carried out on the Wednesday and Thursday, followed
by a de-briefing session. Teams would then travel home on the Friday. The Friday was
also reserved for further testing in case bad weather or other unforeseen problems
interrupted the main programme. In principle, the preceding and succeeding weekends
would be available for travelling for devices with longer journeys.
FEHRL
HERMES Final Report
12
At each trial, a range of surfaces was to be tested by each device at three different
speeds: 30km/h, 90km/h and an intermediate speed that corresponded to the normal
survey speed for the specific device. A target of five replicate measurements on each
surface at each speed was set, with the intention of achieving at least three passes by
each device in the event of delays or disruptions. The order in which devices would test
the surfaces would generally be randomised between replicates. Figure 2.1 is an
example of a timetable chart from one of the trials, illustrating the randomisation process.
Each host organisation was responsible for measuring and reporting the texture depth of
the test sections but, in addition, an opportunity was provided to enable the visiting
devices that could measure texture to survey the test sections on the dry surface before
the commencement of the main wet-testing programme.
Each trial surface selected was to be of homogeneous friction and texture and not less
than 100 m long. The trial organisers were encouraged to select as wide a range of
surfaces as was practical, including asphalt and cement concrete, dense and porous. It
was intended that across all the trials the surfaces selected should vary in type, age and
condition and should include modern thin surfacings. The surfacings are discussed in
more detail in Section 2.1.8.
Device operators would process the data from their own machine and complete a
standardised data sheet with the results (i.e. the average skid resistance and actual test
speed recorded on each section for each run), either before leaving the trial location for
home or as soon as possible afterwards. Figure 2.2 shows an example of a completed
results sheet. This was part of a Microsoft Excel spreadsheet that contained a number of
similar worksheets to record the results from each device participating in the particular
trial, together with other information relating to the trial. Once completed for all the devices
in the trial, the data were sent to BRRC for further analysis.
In addition to recording their data, each device team was asked to complete a
questionnaire seeking feedback on various non-technical aspects of the trial. These were
subsequently used to review the major non-technical issues that are discussed in detail in
Section 2.2 below.
FEHRL
HERMES Final Report
13
APPENDIX 1 TRIAL MATRIX WEDNESDAY 2ND OCTOBER
Route 1 Practice Run Route 1 Texture Run

NO WATER NO WATER
Time No Speed Time No Speed
09:00 13 Any 09:15 13 50
9 Any 1 50
1 Any 6 60
6 Any
Route 1 Replicate 1 Replicate 2 Replicate 3 Replicate 4 Replicate 5
Time No Speed Time No Speed Time No Speed Time No Speed Time No Speed
9:30 1 30 10:10 6 30 10:50 13 30 11:30 13 30 12:10 9 30
6 30 13 30 1 30 9 30 6 30
9 30 1 30 9 30 6 30 13 30
13 30 9 30 6 30 1 30 1 30
9:40 1 50 10:20 6 60 11:00 13 50 11:40 13 50 12:20 9 60

6 60 13 50 1 50 9 60 6 60
9 60 1 50 9 60 6 60 13 50
13 50 9 60 6 60 1 50 1 50
9:50 1 90 10:30 6 90 11:10 13 90 11:50 13 90 12:30 9 90

6 90 13 90 1 90 9 90 6 90
9 90 1 90 9 90 6 90 13 90
13 90 9 90 6 90 1 90 1 90
LUNCH 12:45 to 13:30 TRANSPORT WILL BE PROVIDED TO ROOM B0151
Route 2 Practice Run Route 2 Texture Run

NO WATER NO WATER
Time No Speed Time No Speed
13:45 13 Any 14:00 13 50
9 Any 1 50
1 Any 6 60
6 Any
Route 2 Replicate 1 Replicate 2 Replicate 3 Replicate 4 Replicate 5
Time No Speed Time No Speed Time No Speed Time No Speed Time No Speed
14:20 1 30 15:00 6 30 15:40 13 30 16:20 13 30 17:00 9 30
6 30 13 30 1 30 9 30 6 30
9 30 1 30 9 30 6 30 13 30
13 30 9 30 6 30 1 30 1 30
14:30 1 50 15:10 6 60 15:50 13 50 16:30 13 50 17:10 9 60

6 60 13 50 1 50 9 60 6 60
9 60 1 50 9 60 6 60 13 50
13 50 9 60 6 60 1 50 1 50
Figure 2.1 Example of trial running timetable
FEHRL
HERMES Final Report
14
Harmonisation of European Routine and Research Measuring

Equipment for Skid Resistance of Roads and Runways
HER M ES
CALIBRATION TRIAL MEETING - RESULTS SPREADSHEET - FRICTION TEST RESULTS
This sheet summarises the results from Device 1(SCRIM from UK), at calibration meeting 1.1 held in United Kingdom on 23-
25 October 2001
Section Target speed 30km/h Intermediate speed Target speed 90km/h

Identifier
Section name
(eg UK1, actual friction actual friction actual friction
NL5) Run Run Run
speed value speed value speed value
1 29.8 0.550 1 50.4 0.523 1 89.3 0.453
2 30.2 0.555 2 50.1 0.529 2 88.8 0.450
UK1 Grooved concrete 3 30.7 0.550 3 50.0 0.530 3 89.1 0.455
4 31.2 0.550 4 50.0 0.531 4 89.1 0.448
5 31.0 0.556 5 49.6 0.531 5 89.5 0.452
1 29.2 0.546 1 50.1 0.479 1 90.3 0.349
2 30.1 0.560 2 49.8 0.493 2 90.6 0.342
UK2 Brushed concrete 3 30.4 0.548 3 50.0 0.502 3 89.3 0.358
4 30.5 0.556 4 49.8 0.495 4 90.5 0.339
5 30.1 0.555 5 50.1 0.496 5 89.9 0.340
1 31.0 0.687 1 49.8 0.652 1 89.3 0.611
2 31.0 0.718 2 50.0 0.691 2 90.1 0.610
UK6 "MARS" 14mm 3 31.0 0.718 3 50.2 0.708 3 91.0 0.605
4 30.0 0.737 4 49.7 0.688 4 88.1 0.602
5 30.3 0.739 5 50.1 0.700 5 92.9 0.584
1 30.0 0.704 1 50.0 0.699 1 89.1 0.603
2 31.5 0.686 2 49.6 0.709 2 90.4 0.570
UK5 "ENCAP" 3 30.0 0.753 3 50.0 0.691 3 91.0 0.573
4 30.2 0.715 4 50.0 0.681 4 90.8 0.550
5 31.0 0.718 5 50.0 0.689 5 90.8 0.552
1 30.4 0.653 1 50.2 0.647 1 87.5 0.595
2 30.5 0.642 2 49.8 0.684 2 89.5 0.606
UK8 Hot Rolled Asphalt 3 30.3 0.661 3 50.0 0.683 3 90.6 0.606
4 31.0 0.662 4 49.9 0.700 4 90.4 0.601
5 30.9 0.666 5 50.0 0.682 5 90.6 0.599
1 31.0 0.134 1 51.0 0.071 1 91.7 0.028
2 30.8 0.157 2 50.3 0.079 2 91.6 0.029
UK7 Epoxy 3 30.4 0.141 3 50.7 0.078 3 89.6 0.030
4 30.0 0.144 4 49.9 0.075 4 91.4 0.035
5 31.0 0.142 5 49.7 0.095 5 89.9 0.026
1 30.0 0.679 1 50.4 0.625 1 91.2 0.500
2 30.1 0.695 2 49.6 0.635 2 90.0 0.503
UK3 fine cold asphalt 3 30.1 0.686 3 50.0 0.642 3 89.6 0.529
4 30.0 0.688 4 49.7 0.633 4 90.2 0.511
5 30.0 0.684 5 50.0 0.636 5 90.1 0.484
1 30.0 0.529 1 49.3 0.454 1 89.8 0.354
2 30.0 0.533 2 50.0 0.456 2 89.3 0.349
UK4 polished concrete 3 30.0 0.525 3 49.9 0.466 3 89.8 0.361
4 30.5 0.527 4 50.0 0.460 4 89.8 0.357
5 30.0 0.525 5 49.6 0.458 5 90.1 0.348
1 1 1
2 2 2
3 3 3
4 4 4
5 5 5
1 1 1
2 2 2
3 3 3
4 4 4
5 5 5
Figure 2.2 Example of a completed data sheet for one device from a trial results
spreadsheet (in this case, Device F13 at Trial 1,1).
FEHRL
HERMES Final Report
15
2.1.5 The trials – Round 1
2.1.5.1 Trial 1,1 – UK

The first trial was based at the private research track of TRL Limited. This provided an
initial test for the operating procedures and a basic model for the subsequent trials in this
round in the Netherlands (1,2) and Spain (1,3).
Eight test sections were chosen. These included asphalt surfacings of a type widely used,
but specific to, the UK, brushed and grooved cement concrete and a smooth epoxy
surface laid for the purpose. The track layout meant that it was not possible to test all of
the sections in turn in a single pass, so the testing was arranged to cover them in three
“routes”. The tests were controlled from a central part of the test track (Figure 2.3).
Figure 2.3 Test vehicles waiting at the control area, trial 1,1
Initially, it had been planned to provide lines on the track as guides for the crews to follow
to when testing. Unfortunately, wet weather at the time meant that the marker tape used
would not adhere satisfactorily to the road surface. However, because the surfaces were
on a test track, and therefore not subject to the concentrated polishing that would occur on
an in-service road, this was not considered critical. To take account of the mixture of left-
side, right-side and centrally-mounted test wheels, crews were instructed to operate with
their test wheel in the centre of the lane. Marshals positioned close to each group of
sections checked that each device followed the correct line.
The original concept required organisers to allow for sufficient time to enable water from
one device to drain from the surface before the next device followed. However, this
proved to be unrealistic for sites of this type given the number of devices taking part and
the number of tests to be made. Therefore, a nominal two-minute interval between
devices was planned. The marshals ensured that there was sufficient time for any excess
water to drain from the test sections between each participating device.
Overall, in spite of delays due to heavy rain when the surfacings became waterlogged, the
full programme of tests was completed. Device F11 developed a fault just prior to the
main test runs. This prevented it recording its measurements at the planned fixed slip ratio
and so it ran with a lock-and-release cycle, interpreting a value equivalent to the standard
18% slip subsequently.
FEHRL
HERMES Final Report
16
2.1.5.2 Trial 1,2 - Netherlands

The second trial in the first round (1,2) was held in the Netherlands. The trial had been
brought forward by a day so that work could be completed before a public holiday that
affected some of the visiting crews. Therefore, there was little time between Trial 1,1
(which finished on the Thursday) and Trial 1,2 (with the briefing scheduled for the
following Monday afternoon) to enable lessons learned in the first trial to be applied.
Nevertheless, it was possible to take some points into account, which assisted the smooth
running of the second trial.
There were six test sections, which were divided between two locations: four on the A6
motorway near Joure and two on a privately operated test track at Lelystad. Five devices
took part in this second exercise. Crews were accommodated and the main briefing was
held in a hotel approximately mid-way between the two locations and about half and
hour’s drive from each of them. The tests on the motorway (Figure 2.4) were controlled
from a depot at Joure, close to the end of the motorway section. Devices were required to
return to the depot for despatch at the appropriate time and sequence.
Figure 2.4 Device F01 approaches the end of the test section on the A6 at Joure
An interval of five minutes was allowed between devices to give time for water drainage
and for vehicles that needed to stop to set up. In theory, this could have been shortened
but the lengthy run between motorway junctions to go from the depot to the site and return
to the depot meant that there was a natural interval of about four or five minutes between
the machines.
Some crews commented on the need to return to the depot between passes once the test
pattern was established because this required negotiating two roundabouts and an
awkward turn. However, this was necessary since the co-ordinator could not know what
was happening on the road and the crews would be unaware of any problems with other
devices or changes in running order that could affect them.
The tests at Lelystad (Figure 2.5) were more easily controlled, although there were
constraints resulting from shared use of the track. This prevented, for example, closer
observation of some of the sections on foot during the testing. Heavy-duty adhesive tape
markers were placed to identify the sections and indicate the test line. Unfortunately, as
was found in the UK, the tape did not adhere well in the wet conditions and the strips were
removed for safety reasons.
FEHRL
HERMES Final Report
17
Figure 2.5 Control area on the Lelystad test track
2.1.5.3 Trial 1,3 - Spain

The third trial (1,3) was held in Spain and organised by CEDEX. Test sites were identified
in the Valencia region and the base for the trials was the Road Maintenance Centre at El
Rebollar, alongside the motorway section of the A3 route from Valencia to Madrid. The
briefing meeting was held here, following a similar form to the previous occasions.
The test sections were on sites at two locations, both on in-service roads: part of the A3
motorway section on the Valencia side of El Rebollar (Figure 2.7) and four on the old non-
motorway N-III route about 30km west of this base. On the first day, tests were made on
the old road. Having been relieved by the motorway, this was now very lightly trafficked
and so traffic management was considered unnecessary. The test vehicles were able to
operate freely along the route (Figure 2.6). The tests were co-ordinated from an area of
ground near a service area, with devices operating generally at about four-minute
intervals.
On the second day, tests were made on the motorway sections. For this work, the test
lane and hard shoulder were closed to normal traffic by a line of cones. “Guards” from the
traffic management team prevented unauthorised traffic from entering the closed section
where it began.
The machines were all filled with water after the initial briefing, from a large tanker that
was filled from a well at the Maintenance Centre. This was towed to the more remote site
by an agricultural tractor to provide additional water supply at the roadside. Fewer re-fills
of water were needed during the tests on the motorway site because there were only two
test sections. Any re-fills that were necessary were made at the Maintenance Centre.
FEHRL
HERMES Final Report
18
Figure 2.6 Rural section on the old N-III in Spain
Figure 2.7 Device F02 on the Motorway site in Spain
As with the earlier trials, good planning ensured that the work went smoothly and most
tests were completed satisfactorily, in spite of the generally cold and wet conditions,
particularly on the first day.
2.1.6.1 Trial 2,1 – Belgium

The trials in the second round were held in February and March 2002. These built on the
experience gained in the first round.
Trial 2,1 was held in Belgium. The tests were organised through MET at their base in
Nivelles. The seven test sites were all on public roads and as in the Netherlands and
Spain, were located in two distinct areas. Five were on a stretch of the A25 main dual
carriageway east of Nivelles and two were on rural single carriageways near Borders.
The sites in the two areas were tested on the successive days.
FEHRL
HERMES Final Report
19
In addition to the main HERMES participants, additional Belgian devices also tested the
sites. For operational convenience and to avoid conflict with the main trial, these devices
operated in a separate group.
For this exercise, the tests on the main road route were organised slightly differently.
Meeting points were arranged in lay-by areas close to the turn-around point at one end of
the circuit. However, rather than requiring all vehicles to report back after each pass, as
had been the case at trials 1,2 and 1,3, crews were instructed to complete a set of
replicates. They would only report back if there were problems or at the end of a set of
measurements at one speed. Although the devices completed their replicates in a
sequence, the trial was organized so that the order of the devices was randomised for the
different speeds.
On the rural roads, both sites were controlled from parking areas that also served as turn-
around points. One of the sites needed informal traffic management to keep the section
clear for devices that had to travel on the wrong side of the road in order to follow the test
line.
As with the Round 1 trials, the weather was wet for part of the time, with heavy rain falling
for the final test site. This, however, was a well drained, coarse textured surface dressing
so build-up of water was not a problem. It was possible for all devices to complete at least
the minimum number of runs required.
2.1.6.2 Trial 2,2 - France

Trial 2,2 was held at the Nantes centre of LCPC in France and the test sections were all
on the LCPC test track. As at TRL and Lelystad, some test sections were on parallel
lanes, but at LCPC they were all on the same straight length (Figure 2.8). It was therefore
possible to test parallel sections in sequence, with one device following another on the
alternative line at a shorter interval than would have been possible with the normal interval
to allow the road to drain. This made testing very efficient and, combined with the fine
weather during the week, led to a very smooth trial.
Figure 2.8 Device F08, preceded by its tanker, on the track at LCPC
2.1.6.3 Trial 2,3 - Netherlands

Learning from their experience in the first round, the DWW team decided that Trial 2,3
should be held at a different location to Trial 1,2. This would simplify the operational
aspects and provide a wider range of surface characteristics than had been possible at
the earlier meeting. It is not a requirement of the Friction Index calibration concept that
FEHRL
HERMES Final Report
20
calibration meetings should always be in the same place and on the same surfaces, so
this provided a useful opportunity to test the procedures over a wider range of
circumstances.
The sections for this exercise were all on motorway standard public roads, near Cuijk. The
test sections were closed to normal traffic for the tests, but the traffic management
organisation limited the amount of time that the devices could have on site. Also, for one
site there was some traffic congestion approaching the lane closure that also affected the
test vehicles.
As a result there was insufficient time to complete a full set of five replicates at all speeds
but all devices achieved the minimum of three runs at each speed.
The third round of trials were all repeat exercises from the earlier rounds and all were held
in the same locations in their respective countries.
Both Trial 3,2 in Belgium and Trial 3,3 in France used the same sections and followed the
same patterns as they had in Round 2, but with different combinations of devices.
However, at TRL there had been some changes to the test track since Round 1 and
several of the surfacings or test lines were changed for Trial 3,1. The trial programme and
procedures were otherwise unaltered from Trial 1,1.
2.1.8 The test surfacings
A wide range of surfacings was eventually included in the HERMES exercises. These
included examples of purpose-made surfaces on test tracks and ordinary road surfaces.
The material types included smooth epoxy resins (on the test tracks at TRL and LCPC) as
well as cement concrete and different types of asphaltic material, including modern thin
layers and porous asphalt. On most of the in-service roads it proved difficult to find a wide
range of materials in a convenient location or a range of levels of skid resistance and
texture depth.
Nevertheless, sufficient surfacings were available to provide a robust test of the CEN
friction index and device calibration proposals. Initial measurements of friction and texture
depth were made by the host organisations to provide a guide to the surface
characteristics. Table 2.6 gives a summary description of the different test surfaces, as
provided by the organisers. Illustrative friction and MPD values have been included in the
table to show the range and combinations of levels at each trial; for convenience, the
sections are grouped by host country in this table.
FEHRL
HERMES Final Report
21
Table 2.6 List of the test surfacings used at the trials
Section Illustrative levels

Brief description of surfacing Trial
Code*
Friction (Device speed) † MPD‡
BE1/BE08 Brushed Cement Concrete 0,60 1,7
BE2/BE09 Bituminous Concrete 0,84 1,3
BE3/BE10 Epoxy-bound Surface Dressing 0,97 1,0 2,1
BE4/BE11 Asphaltic Concrete 0,55 (F0450) 0,9 3,2
BE5/BE12 Asphaltic Concrete 0,81 1,1
BE6/BE13 Asphaltic Concrete 0,67 0,8
BE7/BE14 Surface Dressing 0,66 2,4
ES1 Stone Mastic Asphalt 0,89 2,1
ES2 Stone Mastic Asphalt 0,87 2,0
ES3 Asphaltic Concrete 0,83 0,9
(F0360) 1,3
ES4 Asphaltic Concrete 0,89 0,8
ES5 Porous Asphalt 0,61 2,0
ES6 Porous Asphalt 0,63 2,2
FR1/FR07 Surface Dressing 0,52 0,6
FR2/FR08 Asphaltic Concrete 0,39 0,7
FR3/FR09 Very Thin Asphalt Layer 0,69 0,9
2,2
FR4/FR10 Smooth Epoxy 0,04 (F0260) 0,1 3,3
FR5 Sand-Asphalt 0,41 0,3
FR6/FR11 Porous Asphalt 0,47 1,5
FR12 Very Thin Layer Asphalt, Polished 0,62 0,5
GB1 Grooved Cement Concrete 0,53 1,1
GB2 Brushed Cement Concrete 0,49 0,6
GB3 Fine Cold Asphalt 0,63 0,5
GB4 Polished Cement Concrete 0,46 0,5
(F1350) 1,1
GB5 Stone Mastic Asphalt “ENCAP” 0,69 1,5
GB6 Thin Layer “MARS” 14mm 0,69 2,2
GB7 Smooth Epoxy 0,08 0,2
GB8 Hot Rolled Asphalt 0,68 1,7
GB9 GB1 in opposite direction 0,66 1,1
GB10 GB2 in opposite direction and shorter 0,63 0,6
GB11 As GB5 0,81 1,9
GB12 As GB6 but different line 0,79 2,4 3,1
(F1350)
GB13 As GB7 0,13 0,3
GB14 As GB8 0,83 1,7
GB15 Epoxy Bonded Pea Shingle 0,63 1,4
GB16 White aggregate on dense asphalt 0,80 1,1
NL1 Stone Mastic Asphalt 0/1 New 0,67 0,9
NL2 Dense Asphaltic Concrete 0/16 0,64 0,5
NL3 Emulsified Asphaltic Concrete 0,47 0,4
(F0150) 1,2
NL4 Stone Mastic Asphalt 0/11 Old 0,56 0,6
NL5 Cement Concrete 0,56 0,6
NL6 Dense Asphaltic Concrete 0/16 0,62 0,9
NL7 Dense Asphaltic Concrete 0,64 0,7
NL8 Dense Asphaltic Concrete 0,54 0,6
NL 9 Porous Asphalt 0,62 1,5
NL10 Porous Asphalt 0.69 (F0150) 1,9 2,3
NL11 Brushed Cement Concrete 0.56 0,6
NL12 Brushed Cement Concrete 0.49 0,5
NL13 Dense Asphaltic Concrete 0.55 0,7
* Several of the sections were used at at more than one trial: where this occured, separate numbers are used later
in the report to allow for the possible change in friction or texture over time between the two meetings.
†
The illustrative friction values are averages for the “home” device operating at its normal domestic test speed at
one of the trials at which the section was used.
‡
Illustrative MPD values are shown rounded to 1 decimal place.More precise measurements are given later in
Section 2.3.
FEHRL
HERMES Final Report
22
2.1.9 Discussion of practical technical issues
Overall, the trials process ran very successfully. A high level of co-operation between the
various teams, both organisers and participants, ensured that the required data were
gathered. As the organisers and teams gained experience, so the processes became
more efficient.
Nevertheless, as would be expected, a number of technical issues emerged and lessons

were learnt that have implications if the CEN procedure is to be revised and implemented
in practice.
2.1.9.1 Reliability and number of devices at a meeting

Most participating devices successfully attended and completed a trial in each round.
However, there were some mechanical or electrical problems with the equipment that did
affect some devices.
It might be expected that before attending a trial of this type, devices would be thoroughly
checked and arrive in a fully working condition. There is no reason to think that this was
not the case and it appeared that most of the faults actually occurred by chance at the trial
meetings. The problems are summarised below here and have been taken into account in
the data analysis.
• At trial 1,1, after its initial warm-up, F11 could not measure in the standard fixed-slip
mode it would normally use. Although measurements were made using a wheel-lock
cycle and values for 18% slip were interpolated from the friction/slip curve, these
data were less reliable and not necessarily the same as true fixed-slip values.
• F07 developed a wiring problem during Trial 1,2 that put it out of action for the rest
of the project.
• F06 had problems during two of the trials but it managed to complete three trials. At
Trial 1,2 the device had to operate with its alternative test wheel.
• F05 had limited water flow settings and could not test at speeds over 30km/h at trial
1,2, but this was resolved for later trials. (A similar device owned by a Spanish
operator was also to take part but at Trial 1,3 it failed early the testing and took no
further part in that exercise or subsequent trials).
• F09 successfully completed Trials 1,1 and 2,1 but developed a serious electrical
fault during the first test pass of 3,1 and was unable to gather any data on that
occasion (the problem was later traced to the knock-on effect of a failed water valve
causing a circuit overload).
This experience would suggest that there is always a risk that equipment will develop
problems during the trials that can cause delays to the procedures or result in devices
having to withdraw. Therefore, it is suggested that although the CEN procedure requires
just three devices for a calibration meeting, it is probably better to have more in case a
failure invalidates the entire trial. However, with three or four reference devices needed for
a valid calibration meeting, and the possibility of other devices attending for secondary
calibration, the overall number of devices to be included in any future meetings also needs
to be considered.
FEHRL
HERMES Final Report
23
Almost all of the HERMES trial meetings ran with more than the minimum of three devices
attending. This was mostly due to the addition of local devices at individual trials.
Practically, it was found that up to eight devices could be accommodated depending on
the site.
However, the presence of more machines introduced delays in completing the required
number of passes. This was most significant on the public road sites because of the time
taken for each device to complete a circuit. It was also relevant on the test tracks where
conflicting paths could occur, which meant that all devices would have to complete a
pattern of tests before beginning the next. It is therefore recommended that for any future
exercises the number of devices in any single calibration exercise should be limited to six.
2.1.9.2 Choice of trial sites and test surfaces

As anticipated the outset of the project, this proved to be one of the most difficult aspects
of the whole process. The host organisations who had access to their own test tracks
found it easier to plan and control the exercises, but they were constrained by the layout
and availability of surfacings on their particular tracks and the sections could be shorter
than ideal.
Hosts who had to use public roads had problems in finding a sufficient range of material
types and skid resistance or texture levels within a practical distance of their chosen
operating base. This was particularly apparent in the first trial in the Netherlands, where
the range of surfacings proved to be more limited than had been first thought, and hence
the need to change location for the second trial in that country. This was also a major
issue in Spain where the sections proved to be very similar to one another in character.
The range of surfaces in the trial overall was good, but was limited in some individual
meetings. This is likely to be the most difficult part of organising such exercises in the
longer term.
2.1.9.3 Test line

It was recognised that it would be important to identify the test line clearly. Various
methods were tried to define the test line. On the test tracks, adhesive tape placed at
intervals on the track to mark the width of the test path was proposed. This was tried
initially at TRL and at Lelystad but failed in wet conditions. The technique was more
successful at LCPC, having been laid in warmer and dry conditions. Special reflective
studs (as used for temporary lane markings during road works) were later used
successfully at TRL.
On in-service roads it was not possible to put marks on the roads, so the organisers had
to depend on natural features such as the normal wheel path or edge marking lines to act
as guides to the drivers.
However, whichever method is used, the range of approaches to the position of the test
wheel relative to the vehicle can cause difficulties or create potential safety hazards.
Some devices use trailers with one, two or three wheels and the test wheel can be in a
vehicle wheel path (some right, some left), offset or central. Others are mounted on the
back of a test truck, right or left. Yet others have an inboard test wheel that can be in line
with either the left or right wheel path of the vehicle. Test or towing vehicles can range
from saloon cars to large trucks, which means a variety of vehicle widths. This variety
caused some practical problems at different meetings, for example:
FEHRL
HERMES Final Report
24
• At the UK trials, the epoxy section was deliberately laid with a narrow width to
enable test devices to keep two wheels on the normal track surface. However,
establishing a safe test line was problem for a device with a centrally towed test
wheel. In this situation it was difficult for the driver to keep two vehicle wheels on the
higher-friction surface and the test wheel on in the centre of the section. (Figure
2.9).
• In the Netherlands, the public road sites were arranged so that vehicles with a left-
mounted wheel could use the hard shoulder if necessary to gain the correct line, but
this created a hazard with other maintenance vehicles stopped on the shoulder,
even in the coned area. Even so, some vehicles had initial difficulty in following the
wheel path.
• In Belgium one of the test sections was on the hard shoulder itself, which was
relatively narrow. That made it rather difficult for drivers of the left-wheeled
machines both to find the correct track and to stay on the road, especially at the
higher speeds.
Figure2.9 Device F12 on the narrow epoxy section at TRL
2.1.9.4 Other practical issues

It became apparent at the first trial that there was a clear need for a dummy run to
familiarise operators with the section layout in addition to any preliminary visits to see the
test sections. This important principle was applied in all subsequent trials.
Another point that emerged in the first two trials was that certain devices needed to stop
and set up before a test run. This had not been fully appreciated beforehand and this
caused some initial concerns for following devices. This problem was most noticeable on
the main road sites where devices were running at timed intervals. In this situation, a
device about to make a slow run at the end of a group could be stopping to set up on site
immediately ahead of a following device leading the next group at a higher speed. Once
aware of this problem it could be managed, and this information was specifically sought
ahead of the later trials.
A further operational issue came to light in the first trial that affected devices that use
servo systems to develop a fixed slip ratio. Problems can occur when changing from high-
or medium-friction surfacings to low- friction surfacings during a test run. The servo
system applies a certain force to the brakes to control the slip on the higher-friction
surface. However, that is then too great for the low-friction surface and so the control
FEHRL
HERMES Final Report
25
system enters a lock-and-release cycle to re-adjust the brakes to find the correct slip. As a
result, stable conditions may not be established in time to gather meaningful data.
This has possible implications for minimum test section length. The limited length of
some sections also meant that locked wheel devices could perhaps only conduct one or
two measurements on these sections particularly at the higher test speeds required.
2.1.9.5 Weather conditions and water build up

A concern at the beginning of the project was that significant build up of water on the
surface from repeated running could mean that the effective water depth might be rather
greater than the theoretical depth for each device.
The original plan for allowing a fixed interval between devices proved impractical,
particularly where larger numbers of devices were involved. Careful observation of the
sites suggested that there was no significant water build-up that would be likely to
significantly affect the results with normal testing. Even where a longer interval was
possible, such as in the Netherlands, it was found that where the test section road was not
open to normal traffic, the road did not drain or dry significantly more than with the shorter
test intervals. There was some evidence of slight rutting on some of the test sections, but
this was not sufficient to cause significant water build-up.
Although water build-up was not a general problem when the weather was dry, this was
not always the case during continuous heavy rain and at almost every meeting this proved
to be a problem for some of the time. One of the sections in Spain was on a slight
downhill stretch that tended to drain along the line of the road, so water from successive
test vehicles tended to build up towards the end of the section. There were some
concerns that the water depths were getting too great on this length, particularly when it
was raining.
It was found that, generally, the issue of water build-up can be managed by suspending
testing during the worst weather and by making regular close observations of the surface
condition.
As well as rain, the weather conditions during the testing on the tracks were windy. A line
of trees screening the side of the track meant that one of the sections on the Lelystad
track suffered from noticeable leaf fall in the wind, with leaves sticking to the wet road
surface (Figure 2.10). However, the frequent passage of the test devices (particularly
SCRIM, with its relatively high water flow rate and heavy test wheel) kept the test line
clear and the results are unlikely to have been affected.
FEHRL
HERMES Final Report
26
Figure 2.10 Leaf fall on the track at Lelystadt
It is likely that calibration meetings would always need to be held in the autumn or spring,
to avoid the main testing seasons for most devices. The issue of adverse weather is
therefore inevitable. Experience in these trials has shown that this can be managed, but
some flexibility in planning may be necessary, particularly in periods with prolonged heavy
rain.
2.1.9.6 Water supply

There were some concerns over water supply. This had been anticipated and every effort
was made by the local trial organisers to ensure that water was available. However, in
spite of attempts to resolve these in advance, there were inevitable problems with fittings
not matching. In the more remote locations, large tankers were used to supply water to
the devices.
In Spain, the tanker (figure 2.11) was fed from a well. It was noticed that it had picked up
some leaves and other detritus in the water, some of which was transferred to the device
tanks. There were concerns that this material could impede the water flow on some
machines, although there was no evidence of this occurring during the tests. It was
difficult to estimate precise water requirements and the final filling on the first day drained
the tank completely.
In the second Netherlands trial, the DWW ROAR tanker was used as a remote source.
However, with three SCRIMs joining in this exercise, the demand for water and the time
taken to transfer it to the test devices were rather greater than elsewhere and therefore
supplementary sources had to be found, including drawing water from a nearby canal.
Clearly, the issue of water supply is one that is important to consider in planning any trial
location. It may be worthwhile compiling a set of connectors that could be used to link any
of the devices to local hydrants or other water sources, but this is probably impractical. It
is also important to take account of local difficulties regarding water supply in countries or
areas where water is limited, as illustrated by the need to use a tanker fed from a well in
Spain.
Figure 2.11 Tanker used for remote water supply in Trial 1,3
FEHRL
HERMES Final Report
27
2.1.9.7 Device selection

The selection of devices for the HERMES experiment was simplified by limiting the
number that were involved and because the procedures were being followed for a known
period.
In future, calibrations being carried out on a regular basis, this aspect of planning will
require careful attention. In particular, it will be important to avoid developing sub-sets of
devices that meet together frequently. This could happen because operators may be
reluctant to travel too far for calibration exercises and therefore geographical groups could
emerge. Provision will also need to be made for the inclusion of devices for secondary
calibration. It is recommended that for the full operation of the CEN procedures, the co-
ordination of trial meetings and the allocation of devices to them should be the
responsibility of a single overseeing body.
2.2 Non-technical issues
2.2.1 Use of questionnaires
Task 2.2 was included in the project specifically to address those aspects of organising
and carrying out the trials that do not directly affect the technical or practical running of the
exercises but are nevertheless important in the overall scheme of the harmonised
calibration procedure. These would include issues such as the logistics of getting the
devices together, traffic management and other safety aspects and the costs of the
exercises.
At each trial, a member of the Steering Group (or a representative) was present to
observe the operation of the trial and to take note of any non-technical issues that arose.
The non-technical assessor was provided with a checklist to assist in this. In addition,
questionnaires were used to gauge the opinions of the operating crews and operating
organisations regarding their experience in attending the trials, and of the trial hosts
regarding the organising of their particular exercises.
The information gathered in this way from the early trials was used to advise the
organisers on ways of improving the later trials in the programme. Some of the important
issues that emerged are discussed in sections 2.2.2, 2.2.3 and 2.2.4. The commitment
and co-operation of the device crews were of considerable value in the success of the
experimental programme. The experience that they gained contributed particularly to the
smooth running of the later trials as they become more familiar with the processes
involved, even though they may have been visiting new sites. This experience could also
be of some value in any future calibration exercises where these teams are involved.
Overall, the information from the questionnaires and the experience gained were used to
prepare the advice regarding organising the trials to be included in the separate
Guidelines developed as part of this task as a stand-alone document.
2.2.1.1 Questionnaire for participants

This questionnaire was designed to enable the crew and operating organisation of each
device to record their impressions of each trial that they attended. Apart from standard
information to identify the particular trial, the device and contact details, the participants
were asked to provide some outline information on the costs to them of joining the
FEHRL
HERMES Final Report
28
exercise and to comment on various aspects of the trial, either by ranking them on a scale
or by adding their own remarks.
The main questions asked (shown here as bullet points for brevity) were as follows:
• What were your approximate costs in attending the calibration exercise? (These
were broken down by staff time, travel and accommodation costs).
• Did the participation in the calibration exercise have any influence on your routine
test programme in your country?
• What time period of the year is the most suitable for your organisation to participate
in this type exercise, should future exercises be conducted?
• Did you have sufficient time to prepare your participation in the exercise?
• Did you have to make any special arrangement or changes to your device in order
to be able to travel through foreign countries to reach the place for the exercise?
• What did you think of the traffic management on the sites?
• How was the access to the site?
• How did you find the safety of following the required test lines?
• How did you find the safety of test speeds at the test sites?
• Were you asked to carry any action you consider unsafe while testing?
• How do you rate the organisation of the test runs?
• How do you rate the clarity of instructions for the tests?
• How do you rate the completeness of the instructions?
• Did you meet any language problems while testing?
• How do you rate the marshalling throughout the tests?
• How do you rate the signing on the test sites?
• How do you rate the communication during the tests?
• How do you rate the assistance from the organisers?
• How was the assistance from other teams?
• How do you rate the quality of the initial instructions?
• How was the information for travel between the sites?
• How was the timing of the exercise?
2.2.1.2 Checklist for the non-technical assessor

The checklist that was provided for the non-technical assessor stressed that the assessor
should “observe” and not interfere with the running of the trial (the Task Leader or his
representative liaised with the organisers if it was felt that something should be changed).
The questions addressed all the issues (except costs) covered in the participants’
questionnaire but also included more specific comments or observations on aspects that
individual crews might not be in a position to notice, such as:
• The test section surfacing, etc.
• Whether there were any conflicts between testing vehicles.
• Accommodation.
• Whether test lines were followed properly.
• Any build-up of water on the test sections.
• Whether texture measurements were completed (on dry road).
2.2.1.3 Questionnaire for the host organisers

This much shorter form addressed issues specific to the organising of the local trial. As
with the participants, the organisers were asked for an estimate of their costs, in terms of
cash and staff time, in hosting the calibration exercise. They were also asked to comment
specifically on:
FEHRL
HERMES Final Report
29
• The site selection process.

• The planning of the trial runs.
• Whether the sites meet the requirements of the prEN and if not, whether non-
technical problems were the reason.
• Any problems relating to use of foreign devices.
• Their perception of the performance of the teams.
The organisers were also asked whether they had been able to perform MPD texture
measurements according to the ISO standard, and to describe any logistics (such as
arranging hotels and transportation) that they had carried out.
2.2.2 Logistics
2.2.2.1 Local co-ordination

The individual trials were co-ordinated by a host country representative from the Steering
Group. In the UK, the trials were hosted and organised by a team from TRL; in the
Netherlands by DWW, in France by LCPC and in Spain by CEDEX. In Belgium, the local
organisation was carried out by MET working with advice from BRRC.
The difficulties with the organisation depended largely on the type of site that the host
intended to use. TRL and LCPC, for example, were able to use their own test tracks and
this greatly simplified the arrangements because all the work took place on one site that
was under the host’s control. A disadvantage of this approach was that the operations
were constrained by the layout of the particular test sections on the tracks and in some
cases it was difficult to meet all the requirements of the standard. However, these
problems were offset by the ease with which the test runs could be controlled.
The tests on public roads introduced additional problems, mostly relating to the time taken
to carry out the tests as a result of the greater distances involved travelling along a site
and then returning to the start for a replicate run. In some cases, where teams had to
return to a central control point between passes, there was a time penalty as a result of
having to leave the main road. Conversely, it was more difficult to maintain contact with
the teams, especially those who were having problems, where they were not returning to
a central point between passes. As experience developed over the groups of trials, this
process improved so that marshals were stationed at contact points that were directly on,
or very close to, the route to be followed.
The use of a private test track for some sections for the first trial in the Netherlands was of
mixed success in that it provided a degree of control and some test efficiency, but
constraints placed upon the operation by the track authorities caused some potential
difficulties, such as water supply and ease of movement on and off the track.
Another aspect of local co-ordination that involved considerable work on the part of the
hosts was the arrangement of accommodation for the teams. In some places a centrally
located (compared with the test sites) hotel was used and in others the teams were
dispersed among a number of local hotels. This then involved provision of overnight
parking for the larger test vehicles and local transport to get teams to and from their
hotels. This will always be an issue for organisers and can only be dealt with in a
localised way. Overall, however, there were no significant problems for the test
programmes from this.
FEHRL
HERMES Final Report
30
The timing of the tests and the provision of meal breaks and rest facilities was a problem
on some sites, particularly those using public roads where the working sites were some
distance from the rest area. In some cases, the time taken to complete a pattern of tests
proved longer than anticipated and this had a knock-on effect on the timing, resulting in
fewer tests than planned being completed.
2.2.2.2 Travel to and from the trial Locations

Generally, there were no major problems in getting the devices to or from the main trial
locations. However, for those teams that needed to use ferries (to/from the UK for
instance, or from Norway/Denmark) there were additional time constraints. This could
mean that participant crews had significant downtime before or after the trial where ferries
were infrequent. Conversely, some teams were short of time because they found that in
the event the test site was actually some distance from the publicised “location” and they
were much further from their embarkation point than had been anticipated.
This indicates that as much advance information as possible should be provided.

However, in some cases the forward plans had to be made before the location of sites
was finalised. Although the participants knew in advance that they might be needed for
an extra “contingency” day, those that needed to book ferries usually, and
understandably, chose the earliest possible.
In some cases some effort was required by both the participants and the local co-ordinator
in order to gain permissions from the appropriate authorities for some of the devices to
operate outside their own country’s borders. Some of the potential problems of getting to
what might have been perceived as the more difficult locations were avoided simply by
participants declining to attend trials in those countries.
Another issue affecting longer-distance travel that had not been considered in advance
but that affected some teams, was the constraint in some countries that prevents heavy
goods vehicles travelling on Sundays.
All of these points emphasise that any future calibration system will need to have a
number of centres, positioned such that devices can reach them comparatively easily.
This does, however, have implications for the EFI which depends on the calibration
meetings being managed to involve different combinations of devices wherever possible:
there might be a trend for certain devices to “favour” particular locations and so create
local sub-sets of devices.
2.2.2.3 Travel between test sites

In the countries where some or all of the test sites were located on public roads, the test
sites were usually in groups at two or three different locations, sometimes many
kilometres apart and not always close to the central hotel where the teams were staying.
This also imposed a time penalty in collecting the devices together and then driving in
convoy to the new location.
Experience showed that keeping to a convoy over long distances (especially where the
driver of the lead vehicle did not appreciate that some of the test devices could not or
were not permitted to run at car speeds) was difficult and placed added strain on the
crews. Provision of good maps and clear instructions will be important if this situation is
likely to occur in future.
FEHRL
HERMES Final Report
31
2.2.3 Safety issues
Some specific safety issues have already been discussed in Section 2.1.9.3. This Section
discusses some more general matters that emerged during the trials.
2.2.3.1 Traffic management

Traffic control on the tests tracks owned by the hosts was generally straightforward to
arrange, with test vehicles generally circulating in one direction. The problem of some
devices needing to stop and set up (Section 2.1.9.4) was catered for in the later trials by
specifically asking participants to tell the organisers in advance that this was necessary.
The management of the public-road sites varied from full formal lane closures on
motorways and main roads through local organiser’s staff directing traffic or no control at
all on quiet rural routes.
Some other potential safety risks were identified that should be taken into account in
future planning, for example:
• Limitations of higher-speed testing in a closed lane alongside slower-moving traffic.
• Risk of live traffic following test vehicles into the closed lane.
• Risk of traffic management or organisers’ vehicles being parked in or very close to
the test lane.
• Lightweight cones being flicked by passing traffic into the test lane.
• Difficulty with test teams immediately communicating with marshals to resolve these
kinds of issues.
Another consequence of the need for traffic management was that of serious delay to the
programme when the management was not in place in time for the scheduled start of the
test programme. In one case, there were delays to the test teams because the devices
had to join the traffic queue created by the closure in order to get to it or to get back to the
control point. These effects combined to reduce the test programme to three replicates
only in one of the exercises.
2.2.3.2 Positioning of test sections

Although the test tracks were generally easier to manage, there were safety issues,
particularly relating to the higher-speed tests, as a result of the positioning of test sections,
either near the edge of the track (affecting right- or left-side test wheels) or close to run-
out points. The organisers generally had little control over where the sections were
(because in most cases they were already in existence).
Allowance had been made for the potential risks by the organisers, but in the event, there
were some concerns expressed by team members that needed attention for subsequent
visits, for example, Trial 1,1 at TRL, where it was necessary to brake and steer into a
curve immediately after a test section. At Lelystad the test runs were interspersed with
other users of adjacent lanes of the track.
The positioning of test sections on the carriageway or hard shoulder on the public roads
has already been mentioned (Section 2.1.9.3). There was a further safety issue relating
to traffic management at one location, where the test vehicles entered a closure protecting
the hard shoulder, but then had to leave it and rejoin traffic and cross straight over an exit
road. There was a risk of potential conflicts here particularly in the higher-speed test runs
where there was no opportunity to reduce speed after the test section before crossing the
live lanes and then to regain speed for the subsequent section.
FEHRL
HERMES Final Report
32
2.2.4 Financial aspects
Clearly, if the proposed calibration system for EFI is to be taken up in the longer term,
then the relative costs of the exercise should be considered. This section summarises in
broad terms the costs indicated by the various hosts and the participants.
2.2.4.1 Costs of organising individual trials

The costs of organising the trials varied considerably from country to country and the
make-up of the costs also varied, depending upon whether public roads, private test
tracks or host’s test tracks were to be used.
On test tracks, there was no need to fund traffic management on the highway, but usually
there was a charge for the use of the facility of the test track that had to be borne by the
project. The staff involvement also varied, with the number of marshals needed being
different in different locations.
Broadly, arranging and running a calibration meeting involved general costs in the range 3
000 to 5 000 euros and around 150-200 hours of staff time.
2.2.4.2 Costs to participants

The costs to the participants (which were not part of the main project budget and were
funded separately by the various FEHRL members involved) comprised two main
components – travel to and from the trial location and time spent at the trial itself. As with
the organisation of the trials, there was a wide range. Clearly, costs were lowest for
participants from the host countries because they were usually based at or relatively near
the test location, although in some cases they also travelled and stayed with the other
teams during the trial. As expected, costs were usually highest for those with some
distance to travel, especially where sea crossings were involved.
Typically, the costs of attending a calibration meeting were in the range 1 500 to 2 000
euros for devices with short distances to travel, such as those based in the host country or
close to the borders of a neighbouring country, or 2 000 to 5 000 euros for devices with
significant distances to travel or if ferries were involved.
The staff time could vary from 50 to 150 hours depending upon the travelling time and
how many people were in the crew (typically two). These hours could also include some
time for preparation, post-processing of data, booking ferries and other administrative
tasks associated with attending the trial.
There were also additional, perhaps hidden, costs (such as hire charges for the test
devices or loss of revenue-earning work) that varied widely and cannot be included easily
in general estimates such as these.
FEHRL
HERMES Final Report
33
2.3 Summary of the harmonisation process
2.3.1 Calculation of the IFI using the PIARC-model
As explained earlier, the first attempt to harmonise skid resistance measurements was the
IFI (International Friction Index), developed as part of the 1992 PIARC experiment. The
IFI uses two parameters: F60 (the common friction value at a slip speed of 60 km/h) and
Sp (the speed constant, which is based on a texture measurement). Using these two
parameters, a friction value can be predicted for any device at any test speed using a
process described here as the “PIARC-model”.
Equations 2.1 – 2.4 show how the calculation is performed. Given a skid resistance
measurement F, made with a certain device at a slip speed S and a texture measurement
Tx , the following steps have to be carried out to calculate the IFI:
• transform the measurement result at a slip speed S to a friction value at a slip speed
of 60 km/h;
• apply a correction for the specific device, to give a device-independent friction value
at a slip speed of 60 km/h.
The transformation to a slip speed of 60 km/h, is done using equation 2.1.
S −60
F ( 60) = F ( S ) ⋅ e S0
(2.1)
in which
% slip
S= ⋅V (2.2)
100
where V = the speed of the device and
S 0 = a + b ⋅ Tx (2.3)
From equation 2.3 it can be seen that the texture measurement is used for the speed
correction. The parameters “a” and “b” in this equation are specific to the device used for
the texture measurements.
The correction from the calculated value for the specific device (F(60)) at the reference
speed of 60 km/h to a device-independent general value at the reference speed (F60) is
achieved by linear regression (equation 2.4):
F 60 = A + B ⋅ F (60) + C ⋅ Tx (2.4)
where “A”, “B” and “C” are device specific parameters for the friction measuring device.
Parameter “C” is set to zero for devices using a smooth (blank) tyre.
In principle, when the IFI parameters F60 and S0 are known, the friction value of a road
surface at any other slip speed can be predicted. The intention of the IFI was to allow an
easy comparison of measurements made with different skid testers through their
conversion to a common scale.
FEHRL
HERMES Final Report
34
2.3.2 From IFI to EFI
The Belgian research project used the PIARC experiment database together with
additional data to optimise the parameters in the IFI definition and to focus more on the
results of the devices operated in Europe. This included recalculating texture depth values
in accordance with the new ISO-standard for the calculation of the Mean Profile Depth
(MPD) [4]. (Throughout this report, the term MPD refers to values calculated following this
standard). A second objective of the Belgian work was to extend the validation of the IFI to
road surfaces that were either not included or insufficiently represented in the PIARC
experiment and to this end additional test measurements were made on these surface
types with devices used in Belgium. The final goal was to make a proposal for calibration
of skid resistance and texture devices based on the revised index that would prevent
devices drifting away from the common standard (which might occur over time).
This work led to the definition of what had become known as the European Friction Index.
An important difference between the IFI and the EFI is that EFI, as well as being
calibrated to devices used in Europe, uses a reference slip speed of 30 km/h rather than
80km/h. In that respect, it had been shown that a reference speed of 30 km/h instead of
60 km/h was better adapted to the fleet of European devices. The results of the Belgian
work were incorporated into an informative annex to the draft CEN standard for the
determination of skid resistance of a pavement surface (prEN 13036-2) [3]. It is the
calibration procedures in this document that are being assessed by the HERMES project5.
2.3.3 Determination of the EFI
EFI is defined as:
EFI = A + B ⋅ F30 (2.5)
in which
F30 = F ⋅ e ( S −30) / S0 (2.6)
giving the general equation
s − 30
EFI = A + B ⋅ F ⋅ e s 0 (2.7)
Here, F30 is the measured friction coefficient, F, brought to the 30km/h reference slip
speed using the predicted value of S0 given by equation (2.8).
S 0 = a + b ⋅ MPD (2.8)
5
As explained previously, the Draft prEN-13036-2 [3], proposes the term Skid Resistance Index
(SRI) to describe the harmonised scale, instead of European Friction Index (EFI). However,
because EFI has been used in several previous publications, and in order to avoid confusion in
readers’ minds, EFI is also used in this report. A change to using the term SRI may be eventually
necessary if the prEN retains that terminology and when it is used as a standard.
FEHRL
HERMES Final Report
35
In practice, “A” and “B” are parameters specific to the friction measuring device used. The
coefficients “a” and “b” of the original PIARC-model have been replaced by fixed values,
either for a laser texture measurement (MPD) or a patch test (MTD), as in equations (2.8)
and (2.9):
S 0 = 57 + 56 ⋅ MPD (2.9)
or
S 0 = 43 + 70 ⋅ MTD (2.10)
It should be noted that, with these formulae, S0 is only an estimate of the real speed
constant for different sections. Further, as will be seen later, by using these equations, an
initial source of inaccuracy might be introduced into the calculations, depending on how
well the real speed constant is estimated by the general formula.
2.3.4 The calibration procedure
In the draft CEN-standard, initial “A” and “B” values are given for the so-called “reference
devices” that were in the original PIARC experiment. It is necessary to check that these
“A” and “B” values for the reference devices remain stable over time. This is achieved by a
periodic calibration of subsets of reference devices. In this “Type-1 calibration procedure”,
as it is known, new values for “A” and “B” are calculated for the reference devices.
The calibration procedure also provides for the inclusion of “new” devices. This is
achieved by means of a “Type-2 calibration” in which a new device runs alongside existing
reference devices and initial “A” and “B” values are calculated for the new device. Once
this has happened for a device, it joins the set of reference devices and will be included in
a Type-1 calibration in the next round.
The calibration friction tests in any particular meeting are carried out on a range of
surfaces (usually at least six) for which MPD has been measured. Friction tests are
performed at least three times at each of three operating speeds. The speeds must
include the speed that is standard for the device when operating in its own country and
two others so that the range from 30 to 90 km/h is covered.
If a device needs to apply corrections to the results (for temperature for example), these
should be made according to the standard procedure for that device and only the
corrected results should be reported and used in the subsequent calculations. Speed
corrections, of course, should not be made. The data from these measurements are then
used in a series of calculations as follows.
Initially, value of EFI is calculated for each individual measurement using equations (2.7)
and either (2.9) or (2.10), depending on the texture measurement method. Through most
of the analyses in this report, equation (2.9) is used since the texture measurements in the
various exercises determined MPD. The values for A and B used are those established for
the device from the previous calibration exercise. In the case of the first trial where a
“reference” device takes part, these would be the appropriate value provided in the draft
standard.
FEHRL
HERMES Final Report
36
Having calculated the individual EFI values, the next step is to apply tests to the data to
check for outlying tests and devices. For each device (i) and for each surface (j), calculate
the average of the r measured values (normally, r=9 for three runs each three speeds) of
the EFI, which yields EFIij (i=1,…,N; j=1,…,n). N is the total number of friction devices
taking part in the calibration exercise and n is the total number of surfaces used.
The standard deviation σij of each measurement series is calculated by means of the
following equation:
( EFIij − EFImij ) 2
σ ij = ∑m r −1
(2.11)
where m is the run number (m=1,…,r).
If σij > 0,04, the measurement which has the largest deviation from the mean is discarded
and the average is calculated again to give EFIij and σij with m=1,…,r-1. If, again, σij >
0,04, the whole measurement series of device (i) on surface (j) should be discarded.
For each surface, the “Grand Average” EFIj of the NR average values of EFI reported by
the “reference devices” only should be calculated. (In later sections of the report the form
“<<EFI>>” is also used as shorthand to represent the grand average of all the reference
device EFI values.)
For each device, the linear regression of EFIj versus EFI mij is calculated by the least
squares method, which gives
EFI j = α i + β i ⋅ EFI mij , with i = 1,…,N (2.12)
The residual standard deviations with respect to the regression lines, σi, are then
calculated using the following formula:
( EFI j − α i − β i ⋅ EFI mij ) 2

σi = ∑∑m j ( r ⋅ n − 2)
(2.13)
If σi > 0,07, device (i) should repeat the whole series of measurements after having taken
steps to fix the problem with the device.
Having established that the devices have provided valid results following the calibration
exercise, new values for “A” and “B” are determined. To do this for each device (i), the
value for EFI mij in equation (2.12) is replaced by its expression using the old values, i.e.:
EFI j = α i + β i ⋅ ( Ai, old + Bi, old . ⋅ F30 ) with i = 1,…,N

(2.14)
and hence,
Ai ,new = α i + βi ⋅ Ai, old (2.15)
FEHRL
HERMES Final Report
37
Bi ,new = β i ⋅ Bi ,old (2.16)

It would normally be expected that, assuming that the equipment is unchanged, changes
in these parameters would be small. Clearly, if large differences appear between the old
and the new values, the device should be checked to verify that it is producing proper
friction values. It is also important that a device is maintained and calibrated according to
the instructions of the manufacturer and should be in proper working order before it is sent
to a calibration exercise. When a device that is not working properly is sent to a calibration
exercise, it could introduce a problem for other devices because the EFI is based on the
grand average of all the measurements in a calibration exercise.
2.4 Experimental results
2.4.1 Introduction
This section presents the results of the measurements carried out in the nine calibration
trials. It also includes an initial analysis of the data following the process set out in the
CEN draft standard and explained in 2.3.4 above.
2.4.2 Characteristics of the test surfaces
As Table 2.7 shows, the test sections covered a wide range of types of surfaces including
dense materials, porous surface courses and open-textured thin layers, as well as
different binders (such as asphalt, cement or epoxy resin).
Table 2.7 Tested surface types and Mean Profile Depths in mm.
Code Type MPD Code Type MPD

BE01 Brushed Cement Concrete 1,69 GB01 Grooved Cement Concrete 1,09
BE02 Asphalt Concrete 1,31 GB02 Brushed Cement Concrete 0,57
BE03 Epoxy-Bound Surface Dressing 1,03 GB03 Fine Cold Asphalt 0,48
BE04 Asphalt Concrete 0,89 GB04 Polished Cement Concrete 0,54
BE05 Asphalt Concrete 1,13 GB05 Stone Mastic Asphalt "Encap" 1,54
BE06 Asphalt Concrete 0,79 GB06 Thin Layer "Mars" 14 Mm 2,23
BE07 Surface Dressing 2,42 GB07 Epoxy 0,19
BE08 Brushed Cement Concrete 1,71 GB08 Hot Rolled Asphalt 1,66
BE09 Asphalt Concrete 1,60 GB09 Grooved Cement Concrete 1,06
BE10 Epoxy-Bound Surface Dressing 0,98 GB10 Brushed Cement Concrete 0,62
BE11 Asphalt Concrete 0,90 GB11 Stone Mastic Asphalt "Encap" 1,85
BE12 Asphalt Concrete 1,00 GB12 As GB06 But Different Line 2,41
BE13 Asphalt Concrete 0,94 GB13 Epoxy 0,25
BE14 Surface Dressing 3,03 GB14 Hot Rolled Asphalt 1,73
ES01 Stone Mastic Asphalt 2,13 GB15 Epoxy Bonded Pea Shingle 1,43
White Aggregate on Dense
ES02 Stone Mastic Asphalt 1,99 GB16 0,86
Asphalt Concrete
ES03 Asphalt Concrete 0,92 NL01 Stone Mastic Asphalt 0/11 (new) 0,86
ES04 Asphalt Concrete 0,79 NL02 Dense Asphalt Concrete 0/16 0,52
FEHRL
HERMES Final Report
38
Code Type MPD Code Type MPD

ES05 Porous Asphalt 2,02 NL03 Emulsified Asphalt Concrete 0,36
ES06 Porous Asphalt 2,23 NL04 Stone Mastic Asphalt 0/11 (old) 0,56
FR01 Surface Dressing 0,57 NL05 Cement Concrete 0,62
FR02 Asphalt Concrete 0,73 NL06 Dense Asphalt Concrete 0/16 0,91
FR03 Very Thin Asphalt Concrete 0,90 NL07 Dense Asphalt Concrete 0,66
FR04 Epoxy 0,09 NL08 Dense Asphalt Concrete 0,62
FR05 Sand Asphalt 0,34 NL09 Porous Asphalt 1,76
FR06 Porous Asphalt 1,54 NL10 Porous Asphalt 1,87
FR07 Epoxy-bound Surface Dressing 0,76 NL11 Brushed Cement Concrete 0,55
FR08 Asphalt Concrete 0,84 NL12 Brushed Cement Concrete 0,47
FR09 Very Thin Asphalt Layer 1,13 NL13 Dense Asphalt Concrete 0,70
FR10 Epoxy 0,16
FR11 Porous Asphalt 1,78
FR12 Very Thin Asphalt Layer. Polished 0,54
The texture depths (measured as MPD) ranged from 0,09 to 3,03 mm. Two sections
(FR04/FR10 and GB07/GB13) were made from epoxy resin, both having an extremely
smooth surface. When more than one texture measuring device was used on a section
(see Section 2.3.3), the average of the reported values has been given in Table 2.7.
2.4.3 Texture measurements
Texture measurements with the participating devices were carried out on each test
surface at the start of the first day of friction testing. The device crews were asked to
report the Mean Profile Depth averaged over the whole length of the section. At some
trials, more than one texture profiler was used and this has provided an opportunity to
evaluate the reproducibility of the measurement of MPD by different devices claimed to
comply with ISO 13473-1 [4]. Table 2.8 lists the values reported by each individual device
on each section.
Table 2.9 provides some statistics derived from the data in Table 2.8. The standard
deviation (σ R ) of reproducibility between pairs of devices (1 & 2) has been calculated
using the following formula:
∑ (MPD
1
σR = 1i − MPD2i ) 2 (2.17)
2n i
where subscript i denotes the tested surface. Its overall value is 0,11 mm.
The linear regressions between pairs of devices, illustrated in Figures 2.12 to 2.20, have
been calculated with the intercept forced to zero. The reason for this is that when doing
so, the intercepts are generally of the order of the deviations from the regression line,
which means that they are not significant. The resulting slope ( a ) is generally close to
unity as expected. Its estimated relative standard deviation ( σ a ) is 4,7% on an average.
FEHRL
HERMES Final Report
39
In the subsequent analyses, the average reported values of MPD (as in Table 2.7) have
been used.
Table 2.8 Comparison of MPD(mm) values reported by different texture devices on a

given section.
Section Texture device

Code Type T1 T2 T3 T4 T5 T6 T7 T8
BE01 Brushed cement concrete 1,62 1,76
BE02 Dense asphalt concrete 1,18 1,44
BE03 Epoxy-bound surface dressing 0,98 1,08
BE07 Surface dressing 2,41 2,43
ES01 SMA 1,86 2,39
ES02 SMA 1,75 2,23
ES03 Dense asphalt concrete 0,76 1,07
ES04 Dense asphalt concrete 0,69 0,89
ES05 Porous asphalt 2,18 1,86
ES06 Porous asphalt 2,27 2,19
FR07 Surface dressing 0,58 0,93
FR08 Dense asphalt concrete 0,82 0,85
FR09 Very thin asphalt layer 0,91 1,34
FR10 Epoxy 0,11 0,20
FR11 Porous asphalt 1,52 2,03
FR12 Very thin asphalt layer. Polished 0,47 0,60
NL01 SMA 1 year old 0,85 0,86
NL02 Dense asphalt concrete 0/16 0,52 0,51
NL03 EAB Emulsified A Concrete 0,42 0,30
NL04 SMA 0/11 Old 0,60 0,52
NL07 Dense asphalt concrete 0,69 0,59 0,70
NL09 Porous Asphalt 1,78 1,71 1,80
NL10 Porous Asphalt 2,21 1,69 1,70
NL11 Brushed cement concrete 0,59 0,52 0,53
NL12 Brushed cement concrete 0,47 0,43 0,51
GB09 Grooved concrete 1,02 1,09 1,08
GB10 Brushed concrete 0,54 0,62 0,70
GB11 SMA "ENCAP" 1,50 2,05 2,00
GB12 Thin layer "MARS" 14 mm 2,40 2,55 2,29
GB13 Epoxy 0,24 0,26
GB14 HRA 1,68 1,71 1,82
GB15 Bonded Pea Shingle 1,27 1,47 1,53
GB16 White Stone Asphalt 0,80 0,91
FEHRL
HERMES Final Report
40
Table 2.9 Statistical comparison between MPD values reported by different devices (n:
number of measurements; ? R: standard deviation of reproducibility; a: slope
of the linear regression of y vs. x; ? a: estimated relative standard deviation of
the slope)
Trial x y n a sa sR
1,2 T1 T2 4 0,940 5,7%
0,05
1,2 T2 T1 4 1,053 6,4%
1,3 & 3,3 T3 T4 12 1,126 6,5%
0,24
1,3 & 3,3 T4 T3 12 0,857 4,9%
2,1 T6 T8 7 1,029 3,7%
0,09
2,1 T8 T6 7 0,964 3,5%
2,3 T4 T6 7 0,851 3,6%
0,15
2,3 T6 T4 7 1,162 4,9%
2,3 T6 T7 7 1,052 2,8%
0,06
2,3 T7 T6 7 0,947 2,5%
2,3 T4 T7 7 0,893 5,1%
0,14
2,3 T7 T4 7 1,098 6,3%
3,1 T1 T5 8 1,117 4,3%
0,16
3,1 T5 T1 8 0,886 3,4%
3,1 T5 T7 6 0,974 3,0%
0,09
3,1 T7 T5 6 1,021 3,2%
3,1 T1 T7 6 1,087 6,3%
0,18
3,1 T7 T1 6 0,905 5,2%
Average : 4,7% 0,11
HERMES TRIAL 3,1 UK HERMES TRIAL 3,1 UK
3,0 3,0
2,5 2,5
MPD (T5)
MPD (T7)
2,0 2,0
1,5 1,5
1,0 y = 1,1168x 1,0 y = 1,0869x
0,5 0,5
0,0 0,0
0,0 1,0 2,0 3,0 0,0 1,0 2,0 3,0
MPD (T1) MPD (T1)
HERMES TRIAL 3,1 UK HERMES TRIAL 2,3 Netherlands
3 3
MPD (T7)
MPD (T7)
2 2
1 y = 1,052x
1
y = 0,9744x
0
0
0 1 2 3
0 1 2 3
MPD (T6)
MPD (T5)
FEHRL
HERMES Final Report
41
HERMES TRIAL 2,3 Netherlands

HERMES TRIAL 2,3 Netherlands
3
3
MPD (T6)
MPD (T7)
2
2
1
1 y = 0,8511x
y = 0.8933x
0
0
0 1 1 2 2 3
0 1 1 2 2 3
MPD (T4)
MPD (T4)
HERMES TRIAL 2,1 Belgium HERMES TRIAL 1,3 Spain
3 3
MPD (T4)
MPD (T8)
2 2
1 1 y = 1,1264x
y = 1,0289x
0 0
0 1 2 3 0 1 2 3
MPD (T6) MPD (T3)
HERMES TRIAL 1,2 Netherlands Figures 2.12 to 2.20

Comparisons between
3 texture devices. Only y
MPD (T2)
2 vs. x is shown. The

regression parameters
1
y = 0,9398x for both y vs. x and x vs.
0 y are given in Table 2.9.
0 1 2 3
MPD (T1)
2.4.4 Friction device-specific parameters
Table 2.10 lists the participating devices, together with the initial values ( A = A0 and
B = B0 ) of their specific parameters to be used in the calculation of EFI (Section 2.4.10).
The values are derived from the PIARC International Experiment 1992 database but
restricted to the devices in use in Europe [2]. The shading denotes devices that did not
participate in the PIARC experiment and so were not initially regarded as reference
devices. Their initial parameters were set to A0 = 0 and B0 = 1 .
Device F09 did not participate in the experiment itself, but it is identical to one that did and
so has been treated as a reference device at this stage. Although device F08 did actually
participate in the PIARC experiment, it is not regarded as a reference device here
because the test wheel yaw angle was changed from 15° to 20° in the period between the
PIARC experiment and the HERMES trials.
FEHRL
HERMES Final Report
42
Table 2.10 List of friction devices along with their initial specific parameters
Friction device Lab, Measurement Slip

Tyre A0 B0
Code Name Country Principle Ratio
F01 DWW Trailer DWW, NL PIARC Blank BFC 0,86 0,100 0,751
F02 ADHERA LCPC, FR PIARC Blank BFC 1,00 0,203 0,700
F03 SCRIM CEDEX, ES AVON Blank SFC 0,34 0,115 0,815
F04 SCRIM MET, BE AVON Blank SFC 0,34 0,006 0,992
F05 GripTester MET, BE Findley-Irvine Blank BFC 0,15 0,190 0,779
F06 ROAR DRI, DK ASTM E1551 Blank BFC 0,20 0,000 1,000
F07 ROAR DWW, NL ASTM E1551 Blank BFC 0,86 0,000 1,000
F08 Odoliograph BRRC, BE PIARC Blank SFC 0,34 0,000 1,000
F09 PFT TRL, GB ASTM E524 Blank BFC 1,00 0,264 0,574
F10 OSCAR NRRL, NO ASTM E524 Blank BFC 0,18 0,000 1,000
F11 ROAR Mk2 NRRL, NO ASTM E1551 Blank BFC 0,18 0,000 1,000
F12 SRT-3 IBDIM, PL Patterned BFC 1,00 0,104 0,886
F13 SCRIM TRL, GB AVON Blank SFC 0,34 0,006 0,992
F14 Odoliograph MET, BE PIARC Blank SFC 0,34 0,291 0,514
F15 IMAG STBA, FR PIARC Blank BFC 0,15 0,000 1,000
2.4.5 The database collected
In total, 4 231 measurements were made. The following data were discarded:
• Device F05 in Trial 1,2 because the device made measurements at only one speed.
• Device F11 in Trial 1,1 because the device did not operate at a fixed slip ratio.
This left 4 084 measurements to take forward to the analysis. Table 2.11 gives the
number of measurements, broken down by device and by trial.
2.4.6 Repeatability of F
The repeatability standard deviation of the friction coefficient, F, reported from repeated
runs by a given device at a given speed on a given surface was calculated. No correction
was made for actual speeds being slightly different between repeated runs as those
deviations from the specified speed must be considered as one of the factors affecting
repeatability of the measurement in this context.
It was found that, in the event, the speed variations were generally rather small, as can be
seen in Annex B. Table 2.12 gives the average values of repeatability of F for each device
at each trial along with the averages per trial over all devices and the averages per device
over all trials and finally the grand average over the whole measurement campaign. i.e.
? r(F) = 0,025.
FEHRL
HERMES Final Report
43
Table 2.11 Number of measurements carried out per device and per trial
Trial →
1,1 1,2 1,3 2,1 2,2 2,3 3,1 3,2 3,3 Total
Device ↓
F01 90 90 81 75 120 456

F02 90 90 90 270
F03 90 75 90 255
F04 90 87 75 57 90 399
F05 80 90 87 90 347
F06 66 76 120 262
F07 70 70
F08 90 90 85 265
F09 103 78 181
F10 115 77 90 282
F11 73 89 162
F12 120 90 90 300
F13 120 75 120 84 399
F14 87 86 173
F15 90 75 98 263
Total 458 316 360 566 450 448 360 497 629 4084
Table 2.12 Repeatability of F for each device at each trial. Averages are weighted
according to the number of measurements from Table 2.11.
Trial→
1,1 1,2 1,3 2,1 2,2 2,3 3,1 3,2 3,3 Average
Device↓
F01 0,011 0,015 0,017 0,018 0,018 0,016

F02 0,018 0,020 0,017 0,018
F03 0,030 0,032 0,023 0,028
F04 0,061 0,022 0,029 0,013 0,020 0,035
F05 0,025 0,035 0,033 0,025 0,030
F06 0,025 0,026 0,029 0,027
F07 0,017 0,017
F08 0,027 0,027 0,022 0,026
F09 0,029 0,029 0,029
F10 0,023 0,036 0,023 0,027
F11 0,024 0,018 0,021
F12 0,015 0,016 0,016 0,016
F13 0,012 0,038 0,023 0,037 0,028
F14 0,026 0,017 0,022
F15 0,017 0,017 0,013 0,016
Average 0,020 0,036 0,023 0,026 0,024 0,027 0,024 0,025 0,021 0,025
FEHRL
HERMES Final Report
44
2.4.7 Reproducibility of F
In some trials, similar devices were participating and this permits an assessment of the
reproducibility of their measurement of F. This was possible with the two Odoliographs,
devices F08 and F14, and with the three SCRIM’s, devices F03, F04 and. The
reproducibility for the means of repeated runs is calculated by the following formula
comparing two devices:
∑∑ (F
1
σR = 1ij − F2ij ) 2 (2.17)
2n i j
in which the sums are over the speeds and the surfaces.
Equation 2.17 includes both the systematic and the random deviations; for this reason the
values for reproducibility given by equation 2.17 cannot be directly compared to the
reproducibility of a single measurement. The results of this analysis are given in Table
2.13, with graphical comparisons in Figure 2.21 for the Odoliographs and Figures 2.22,
2.23 and 2.24 for the SCRIMs.
Incidentally, it can be seen in Table 2.13 that the relationship between pairs of similar
devices varies from one round of calibration exercise to the next. This suggests that some
characteristics of the devices can vary significantly in a six-month period. This is not
surprising since, in spite of their apparent similarities, there are known differences
between the machines. Regarding the SCRIMs, for example, F13 had a left-side test
wheel whereas the other two machines had right-side test wheels. This could lead to
differences in test line, especially on sites where there was limited room for manoeuvre.
Further, it is known that, without careful control, differences can develop between
machines of the same type and none of these machines had been cross-checked with
one another in the way in which machines in a common fleet might be, such as occurs
annually with the UK SCRIM fleet.
Figures 2.25 & 2.26 compare the locked-wheel devices. In this case, it must be noted that
they differ in respect to the type of tyre used: F02 and F09 use a blank tyre while F12
uses a patterned tyre.
Table 2.13 Reproducibility standard deviation for the means of repeated runs
Devices Reproducibility Number of

Device type Trial
compared standard deviation measurements
Odoliograph F08-F14 3,2 0,041 75
F03-F13 2,3 0,056 75
3,3 0,051 85
F03-F04
2,3 0,120 63
SCRIM
2,3 0,171 84
F04-F13
3,2 0,059 76
Overall 0,103 383
FEHRL
HERMES Final Report
45
F14 vs F08 all speeds
1,0
0,8
0,6 Equality
Trial 3,2
0,4 Linear (Equality)
0,2
0,0
0,0 0,2 0,4 0,6 0,8 1,0
F08
Figure 2.21 Comparison between the friction values reported by the two Odoliographs in
Trial 3,2
1,2
1,0
0,8
Equality
0,6 Trial 2,3
Trial 3,3
0,2
0,0
0,0 0,2 0,4 0,6 0,8 1,0 1,2
F03
Figure 2.22 Comparison between the friction values reported by two SCRIMs in Trials 2,3
and 3,3
1,2
1,0
0,8
Equality
F13
0,6 Trial 2,3

Linear (Equality)
0,4
0,2
0,0
0,0 0,2 0,4 0,6 0,8 1,0 1,2
F03
Figure 2.23 Comparison between the friction values reported by two SCRIMs in Trials 2,3
FEHRL
HERMES Final Report
46
1,2
1,0
0,8 Equality
Trial 2,3
0,6
Trial 3,2
0,2
0,0
0,0 0,2 0,4 0,6 0,8 1,0 1,2
F04
Figure 2.24 Comparison of friction values reported by two SCRIM’s in Trials 2,3 and 3,2
1,0
0,8
Equalit y
0,6
2,2
3,3
0,4
Linear (Equalit y)
0,2
0,0
0,0 0,2 0,4 0,6 0,8 1,0
F02
Figure 2.25 Comparison of friction values reported by two locked-wheel devices in Trials
2,2 and 3,3. As both trials took place on the same site (LCPC, in Nantes),
several surfaces were the same. Note that they use different types of tyre.
1,2
1,0
0,8
Equality
0,6 1,1
Linear (Equ ality)
0,4
0,2
0,0
0,0 0,2 0,4 0,6 0,8 1,0 1,2
F09
Figure 2.26 Comparison between the friction values reported by two locked-wheel devices
in Trial 1,1 (note that they use different types of tyre).
FEHRL
HERMES Final Report
47
2.4.8 Curve fitting on F(S)
The first step of the analysis was to determine the parameters for the speed correction for
each device. Initially, values of S0 for each device were determined from the test data by
fitting the exponential curve
F = F0e −S / S 0 (2.19)
to the reported values of the friction coefficient (F) versus the slip speed (S), the latter
being calculated as the reported operating speed (V) multiplied by the slip ratio listed in
Table 2.10. A least-squares linear regression was calculated on the logarithmic form of
the equation:
ln F = ln F0 − S / S 0 (2.20)
A few measurement series where S0 was negative (that is, where friction increased with
increasing speed) were discarded from the analysis: an example of this is shown in Figure
2.27.
Device F05
1,0
0,9
0,8
0,7
0,6 Trial 2,1
F 0,5 Trial 3,3
0,4
0,3
0,2
0,1
0,0
0 2 4 6 8 10 12
S (km/h)
Figure 2.27 Examples of exponential curve fitting on F(S) yielding negative S0 values
In addition, some zero - or close to zero - values reported for F on the smooth epoxy
surfaces were discarded because they did not fit in the calculation of the exponential
regression. In total, less than 0,9% of the measurements were discarded for either of
these reasons (Table 2.14). Annex B presents all the graphs of F versus S.
Table 2.14 Number of measurements discarded per trial.
Trial→ 1,1 1,2 1,3 2,1 2,2 2,3 3,1 3,2 3,3 Total
F < 0,01 0 0 0 0 9 0 0 0 3 12
S0 < 0 0 0 0 9 0 0 0 0 15 24
Total 0 0 0 9 9 0 0 0 18 36
FEHRL
HERMES Final Report
48
2.4.9 Deriving S0 from MPD
According to the “PIARC-model” specified in the CEN standard [3], S0 is predicted from
the texture depth using the linear equation (2.8). When, as here, texture has been
measured using MPD method, the version of this formula in equation (2.9) is used, with
the constants set so as to give S0 in km/h from MPD in mm.
When plotting all the values of S0 versus MPD from the nine trials, as shown in Figure
2.28, it was found that the formula fitted the data reasonably well. However, the scatter
was rather wide. Annex C presents all the graphs of S0 versus MPD per device and per
trial.
700
y = 66 x + 32
S0 (km/h)
600
y = 56 x + 57
500
400
300
200
100
0
0,0 0,5 1,0 1,5 2,0 2,5 3,0 3,5
MPD (mm)
Figure 2.28 General correlation between S0 and MPD, all measurement results plotted
together. The solid line is the actual regression; the dotted line is the equation proposed in
the CEN standard [3].
2.4.10 Applying statistical tests to the EFI measurements
Values of EFI were calculated as explained in Section 2.3.3, together with the grand
average <<EFI>> for all the reference devices (see Table 2.10). As explained in 2.3.4, two
statistical tests are used to identify outlying measurements.
The first test is applied to the values of EFI obtained from repeated runs at the three
speeds by a given device on a given surface. When the repeatability standard deviation
( σ r ) of the series is larger than 0,04, the value that has the largest deviation from the
mean is discarded. If the test fails again, the whole series is discarded. When this test
was applied to the HERMES data it was found that, in all cases, eliminating the most
deviating value did not bring the standard deviation for the device and surface below the
threshold and the whole series had to be discarded. Table 2.15 shows how many
measurements had to be discarded per trial.
The second test (see Equation 2.10) is applied to the reproducibility of EFI. When the
residual standard deviation of the linear regression between <<EFI>> and all the
individual EFI-values reported by a given device in a given trial (Equation 2.11) is larger
than 0,07, the whole measurement series for that device is discarded and the calibration
of that device is declared invalid in that trial. When applied to the HERMES results after
FEHRL
HERMES Final Report
49
the repeatability test had been made, the reproducibility test never failed. However, for
devices F06 and F07 in Trial 1,2, the repeatability test left results from less than two
surfaces, in which case the regression (Equation 2.11) could not be calculated.
Table 2.15 Number of measurement data discarded after the repeatability test.
Number of discarded Percentage of discarded

Trial
measurements measurements
1,1 199 43%
1,2 229 72%
1,3 112 31%
2,1 156 28%
2,2 208 46%
2,3 234 52%
3,1 0 0%
3,2 41 8%
3,3 96 15%
Total 1275 31%
In view of the unacceptable severity of the repeatability test, the data were processed
again, but this time applying only the reproducibility test. In this case, only one
measurement series was flagged, namely from device F10 in Trial 1,1, and those
particular results were not used in the analysis that followed. Device F10, which was not
initially in the reference group, gained its first valid calibration parameters values (A, B)
after its participation in the second round, at Trial 2,1.
This second stage of processing will be referred to as “Scenario #1” because, later in this
report, different variants for the data processing are considered with a view to finding
ways that might improve the reliability and precision of the calibration procedure. In that
context, the strict application of the standard that was the first to be considered will be
referred to as “Scenario #0”.
2.4.11 Calibration results
New values of “A” and “B” were determined as explained in Section 2.3.4, using Equations
2.13, 2.14 and 2.15. This was repeated for each trial meeting, with those devices that at
first had not been defined as “reference devices” gaining their initial values of “A” and “B”
after their first trial and then being included as reference devices in subsequent trials.
The way in which “A” and “B” for each device evolved from trial to trial using the “Scenario
#1” analysis is shown in Tables 2.16 and 2.17 and is represented graphically n Figures
2.29 and 2.30. In these graphs, dotted lines are used to represent the devices that initially
were not “reference devices”. Table 2.18 and Figure 2.31 show the evolution of the
residual standard deviation ( σ EFI ) of EFI. The overall average value of σ EFI is 0,032. Its
average value over the last trial only is 0,024.
FEHRL
HERMES Final Report
50
Table 2.16 Evolution of calibration parameter “A” for each device according to Scenario #1
Device→
F01 F02 F03 F04 F05 F06 F07 F08 F09 F10 F11 F12 F13 F14 F15
Trial↓
Last-5 0,100 0,006
Last-4 0,535 0,421 0,190 0,006
Last-3 0,402 0,203 0,115 0,374 0,451 0,000 0,000 0,104 0,186 0,000
Last-2 0,396 0,366 0,334 0,506 0,405 0,444 0,454 0,264 0,000 0,000 0,185 0,403 0,291 0,432
Last-1 0,384 0,421 0,461 0,469 0,560 0,345 0,000 0,445 0,195 0,431 0,514 0,407 0,386 0,332 0,544
Last 0,372 0,504 0,460 0,464 0,513 0,413 0,383 0,437 0,401 0,488 0,489 0,482 0,455 0,475 0,541
Table 2.17 Evolution of calibration parameter “B” for each device according to Scenario #1
Device→
Trial↓
Last-5 0,751 0,992
Last-4 0,066 0,320 0,779 0,992
Last-3 0,354 0,700 0,815 0,399 0,319 1,000 1,000 0,886 0,818 1,000
Last-2 0,317 0,449 0,417 0,252 0,437 0,256 0,392 0,574 1,000 1,000 0,589 0,333 0,514 0,401
Last-1 0,368 0,338 0,265 0,275 0,159 0,468 1,000 0,405 0,738 0,358 0,198 0,281 0,355 0,454 0,169
Last 0,369 0,182 0,256 0,276 0,248 0,364 0,287 0,350 0,381 0,287 0,258 0,186 0,276 0,268 0,171
Table 2.18 Evolution of s EFI for each device according to Scenario #1
Device→
Trial↓
Last-5
Last-4 0,056 0,034
Last-3 0,025 0,031 0,016 0,048
Last-2 0,026 0,025 0,015 0,020 0,036 0,049 0,026 0,063 0,028 0,053
Last-1 0,023 0,042 0,021 0,012 0,016 0,020 0,044 0,066 0,040 0,023 0,053 0,015 0,014 0,024
Last 0,020 0,024 0,010 0,011 0,032 0,021 0,050 0,012 0,039 0,015 0,017 0,026 0,015 0,011 0,020
FEHRL
HERMES Final Report
51
0,6
F1
F2
0,5 F3
F4
F5
0,4 F6
F7
F8
A 0,3 F9
F10
F11
0,2
F12
F13
0,1 F14
F15
0,0
L-5 L-4 L-3 L-2 L-1 Last
Trial
Figure 2.29 Evolution of calibration parameter “A” for each device according to Scenario
#1. Devices that were “non-reference” initially are shown by dotted lines.
(Last) is the last trial in which the device took part.
1,0
F1
0,9 F2
F3
0,8 F4
F5
0,7
F6
0,6 F7
F8
B 0,5 F9
F10
0,4
F11
0,3 F12
F13
0,2 F14
F15
0,1
0,0
L-5 L-4 L-3 L-2 L-1 Last
Trial
Figure 2.30 Evolution of calibration parameter “B” for each device according to Scenario
#1. Devices that were “non-reference” initially are shown by dotted lines.
(Last) is the last trial in which the device took part.
0,07
F1
0,06 F2
F3
F4
0,05
F5
F6
σ (EFI)
0,04 F7
F8
F9
0,03
F10
F11
0,02 F12
F13
F14
0,01
F15
0,00
L-4 L-3 L-2 L-1 Last
Trial
Figure 2.31 Evolution of s EFI for each device according to Scenario #1. Dev ices that were
“non-reference” initially are shown by dotted lines. (Last) is the last trial in
which the device took part.
FEHRL
HERMES Final Report
52
2.4.12 Discussion
The analysis of the results of the test programme has revealed a number of drawbacks in
the draft CEN standard [3]. This section discusses the main areas where improvement is
needed.
The statistical tests are obviously inappropriate. An alternative approach would be to

apply a standard procedure, considering the different devices as different laboratories and
the measured object as the <<EFI>> value of a given surface. Then, consistency tests
could be applied to the repeatability and reproducibility standard deviations following an
established standard for assessing the precision of test methods such as ISO 5725-2 [5].
The prediction of S0 from MPD exhibits considerable scatter that can affect the precision
of EFI to a large extent, particularly when the measurement speed and equipment
principles are such that the slip speed is far from the reference speed of 30 km/h. The
more the predicted S0 departs from its actual value, the more EFI depends on speed as
the extrapolations of F(S) to F30 from different speeds diverge, as illustrated for one device
and trial in Figure 2.32.
In view of Annex D, it seems that some improvement could be gained by considering the
relationship S0(MPD) as device-dependent. Then, not only “A” and “B” but also “a” and “b”
would be device-specific parameters requiring calibration.
Device F12 in Trial 1,1
0,7
0,6
0,5
<<EFI>>
0,4 V=30 km/h

V=60 km/h
0,3 V=90 km/h
0,2
0,1
0,0
0 0,2 0,4 0,6 0,8 1 1,2
EFI
Figure 2.32 Example of operating speed (V) influence on EFI
In eight cases, occurring in Trials 1,1, 2,2 and 3,3, the predicted S0 has a small, negative
value for some low-MPD surfaces (see Annex D). This is not acceptable because it leads
to absurd values for F30. To avoid this, a non-linear model for S0(MPD) should be used,
which would also better fit most data sets.
Moreover, in view of some of the graphs in Annex D, in several cases, some S0 values
depart markedly from the general trend. In order to prevent such outlying values to have
too much influence on the curve fitting, it would be advisable to use a weighting that gives
less weight to S0 values derived from poorly-correlated F(S) data.
FEHRL
HERMES Final Report
53
From another viewpoint, even if S0 were accurately predicted, measurement series not
fitting the exponential model could also affect the precision of EFI to some extent. That is
what happens in number of instances in the HERMES tests, of which Figure 2.33 shows
an example. Therefore, a better model could be needed for F(S).
0,9
0,8
0,7
0,6
2,2/F08/FR1
3,2/F15/BE9
F 0,5
3,2/F05/BE12
0,4
0,3
0,2
0,1
0
0 5 10 15 20 25 30
S (km/h)
Figure 2.33 In these examples, the data exhibit a negative curvature, which is opposite to
the exponential model
As a given device undergoes successive calibrations, there is a general tendency for its
“A” parameter to increase and its “B” parameter to decrease, which has two undesirable
consequences. Firstly, “A” is effectively the EFI-value that would be obtained on a zero-
friction surface. That being so, values higher than 0,5 (see Table 2.16) are hardly
acceptable values for a very slippery surface, such an icy one, for example. Secondly, “B”
reducing to a range of 0,18–0,38 means that if the difference in friction coefficient between
two surfaces at a given speed is ? F, their difference in terms of EFI will decrease to 18-
38% of ? F, which would be a dramatic loss of sensitivity of EFI in comparison with the
skid resistance of the surface.
Moreover, the apparently good precision of the calibration, namely s EFI = 0,024 in the last
round of tests, is an artefact of the low “B” values. Indeed, for it to be compared to the
precision of EFI, the precision of actual measurements of F must be multiplied by B, which
is generally much lower than one.
The development of “A” and “B” over successive trials is considered to be mainly due to a
bias arising from the least squares linear regression method being used in a situation
where both the independent variable and the dependent variable have measurement
errors. This tends to pull the slope towards low values all the more because the scatter is
large. The effect is amplified as the procedure is repeated over successive trials, as
Figures 2.29 & 2.30 show.
A possible way to prevent that behaviour could be to calculate the regression of EFI
against <<EFI>> and then to reverse the equation to get back <<EFI>> against EFI. From
a statistical viewpoint, this is a better approach then the original regression. However, the
FEHRL
HERMES Final Report
54
project was already at an advanced stage when this approach was first introduced. An
initial attempt to analyse the data in this way was made (see 4.6) but this was inconclusive
and there was insufficient time to fully re-work the data. Another possibility might be to
force “A” to zero. Then, by a new definition, EFI would just become proportional to F30:
EFI = B ⋅ F30 (2.21)
This is a logical and practical approach, since EFI would then vanish on a zero-friction
surface. Moreover, for the data from the LCPC and TRL test tracks that included low-
friction epoxy surfaces, the regression lines (Equation 2.12) pass close to the origin (Trials
1,1; 2,2; 3,1; 3,3 in Annex D). This supports the idea that model (2.21) should be applied
in all cases. However, there might be individual situations where the form is not linear
through the whole friction range (compare, for example, the first two graphs in Annex G).
FEHRL
HERMES Final Report
55
3 Revision of the friction model
3.1 Introduction
In the Belgian study [2] that followed up the 1992 PIARC experiment, the reproducibility
and the repeatability of the EFI were calculated. The standard deviation describing the
repeatability over all the devices and sections was 0,08 and the standard deviation for the
reproducibility of the EFI by different devices was 0,14.
This means that, in the situation where one device is used to predict the measurement
results of another device, the confidence interval is very large. Therefore, the use of the
EFI in its present state is very limited and improvement of the model is desirable.
In this Chapter, the work carried out in the HERMES project to improve the model will be
presented. This work was carried out mainly by DWW and LCPC, with a contribution from
TRL.
3.2 Further improvements to the model
In the process of investigating ways of improving the EFI, three databases were used: the
HERMES database; a small database containing the results of an experiment carried out
by BASt, DRI and DWW in 1996, and the original database from the PIARC experiment in
1992. From the PIARC database, only data from friction devices with blank tyres were
included. For the texture devices, only the Belgian data processed according to the ISO
standard were used. Three possible improvements were tested:
• a specific relation between the speed parameter S0 and MPD for every device,
instead of the general relation according to Equation 2.9;
• a non linear relation between S0 and MPD with specific coefficients for every device;
• a model based on the lubrication theory developed in tribology (the Stribeck
formula).
3.2.1 Calculation of S0 vs. MPD ISO per device
In Section 2.3.3 the comment was made that inaccuracies might occur in calculating the
EFI because the speed constant is estimated from a general formula, independent of the
friction device. It was found that the “real” speed constant, calculated from the
measurement results, was not independent of the friction device. Therefore, an attempt
was made to derive a device-dependent relation to calculate S0 from a texture
measurement.
As a first step, a linear regression line was fitted for each device and site combination,
with
x = S − 30 (3.1)
and
FEHRL
HERMES Final Report
56
y = ln( F ) (3.2)
The results of this analysis (the regression coefficients and their standard deviations) were
saved and from this dataset, the following parameters were calculated:
S0, from the slope of the regression line as:
−1
S0 = .
slope
Note that this is the “real” S0 for the different devices and not an estimate as calculated
with equation 2.9;
the standard error of S0 as
se( slope )
se( S 0 ) = ;
slope 2
a weighting factor as
w(S 0 ) =
1
2
.
se( S 0 )
The next step was to determine the relationship between the S0 and the texture parameter
MPD. To find this relationship a weighted linear regression was fitted (for each device)
with:
x = MPD (3.3)
y = S0 (3.4)
with the weight w(S0).
With the introduction of a weighting factor, it is possible correct the influence that the
accuracy of the parameter S0 has on the regression. Values of S0 with small standard
errors will have a greater weight in the regression analyses than S0 values with a large
standard error. Together with this weighted analyses, the Cook’s distance was used to
check for outliers. The Cook’s distance is a method to indicate whether deletion of a
particular value will have a large influence on the regression coefficients.
For each device, the following graphs were drawn in order to judge the regression
analysis:
• the standardised residuals versus the fitted values;
• the Cook’s distances versus the fitted values;
• standardised partial residual plots (plots in which the effects of the other dependent
variables are eliminated).
When performing a linear regression with y = S0 and x = MPD and weight = w(S0), then
what actually is done, is to fit a linear regression without an intercept and with:
FEHRL
HERMES Final Report
57
x1 = w(S 0 ) (3.5)
x 2 = w(S 0 ) ⋅ MPD (3.6)
y = w(S 0 ) ⋅ S 0 (3.7)
In the situation where a weighted linear regression is used, no direct estimate is made of
the intercept of the regression line. The regression coefficient belonging to x 1 represents
the intercept, and the regression coefficient belonging to x 2 stands for the effect of the
MPD. In subsequent analyses (maximum 5 repetitions) the measurements with the largest
Cook’s distances were deleted from the regression analysis.
The analysis resulted in a specific relation between S0 and MPD (S0 = a + b·MPD) for
every friction device instead of the general relation S0 = 57 + 56·MPD derived from the
data from the PIARC experiment. The coefficients for the different devices are listed in
Table 3.1.
Table 3.1 Coefficients by device (Device reference numbers refer to the PIARC
experiment)
Device a b Device a b Device a b Device a b

B1 LKD 81 43 B5 LKD 94 34 C5 49 93 D3 47 32
B1 SLP 33 6 B5 SLP 28 22 C6E 35 39 D4 106 9
B2 LKD 115 53 C1 83 55 C9 -9 46 D5 52 26
B2 SLP 34 17 C10 -8 55 D1E 34 32 D6 3 93
B3 35 19 C3B 61 32 D2 56 6 D8 -4 53
With a specific relation established for each device, the original model was followed to
calculate values for the EFI per site and per device. This calculation showed that the
accuracy of the results was worse than the results from the original model. To
demonstrate this, the standard deviations for the reproducibility and the repeatability of the
new model and the original model are listed in Table 3.2. In fact, this is a very strange
result: by improving the model for the relationship between slip speed and texture, the
overall result worsens. The only explanation for this is that MPD alone is not the right
explanatory factor for the speed dependency of friction.
Table 3.2 Comparison of the standard deviation for the repeatability and the
reproducibility of both models
Standard deviation for
Repeatability Reproducibility
Original model 0,08 0,14
New model 0,13 0,13
With the dataset available after this analysis it was possible to check the findings
demonstrated at the SURF 2000 symposium [6] concerning a linear relation between the
FEHRL
HERMES Final Report
58
slope in the relation S0 vs. MPD and the slip percentage of the devices. Again a weighted
linear regression was used, with a weighting factor
1
w= .
se ( slope ) 2
Using this relation, an estimate was made for the coefficients “a” and “b”. In Table 3.3
these estimates are listed as (a* and b* ), together with the values that were calculated
from the weighted linear regression. Although most of the estimated “a” and “b” values
were very close to the calculated values, the results of the analysis were not very
satisfying. The percentage variance accounted for was 31%. The slope of the regression,
however, was very significantly different from zero (Probability = 0,994).
Table 3.3 Coefficients estimated from slip percentage
Estimated from Estimated from

% Calculated % Calculated
Device slip percentage Device slip percentage
slip slip
a b a* b* a b a* b*
B 1 LKD 100 81 43 68 56 C5 86 49 93 79 53
B 1 SLP 14 33 6 25 14 C 6E 34 35 39 40 33
B 2 LKD 100 115 53 113 56 C9 26 -9 46 2 27
B 2 SLP 20 34 17 30 21 D 1E 34 34 32 33 33
B3 20 35 19 33 21 D2 34 56 6 32 33
B 5 LKD 100 94 34 75 56 D3 34 47 32 47 33
B 5 SLP 20 28 22 29 21 D4 34 106 9 78 33
C1 100 83 55 82 56 D5 34 52 26 46 33
C 10 34 -8 55 6 33 D6 100 3 93 28 56
C 3B 34 61 32 61 33 D8 18 -4 53 17 19
When the values of the EFI, that were calculated with the new model were analysed, it
was found that these values were still dependent from the slip speed. From the “real”
relation between friction and slip speed, it was seen that some non-linearity exists in the
relation “ln (friction)” versus slip speed. As a possible measure to eliminate this
dependency, it was decided to perform a non-linear regression analysis for the relation
between “y = ln (F)” and “x = S-30”.
3.2.2 Quadratic regression analysis
As stated above, the new model with specific “a” and “b” values for every device, resulted
in EFI-values that still depended on the slip speed. Therefore the possibilities of finding a
better model to define the relation between “ln(friction)” and the slip speed were
discussed, the idea being to remove, or at least reduce, the influence of the slip speed on
the EFI.
As a result of this discussion, and based on the graphs from earlier calculations, a
quadratic model was chosen to describe the relation between “ln(friction)” and the slip
speed. It should be emphasized that a quadratic model was decided upon arbitrarily,
based on the shape of the graphs of the real relation; there was no physical reason for the
FEHRL
HERMES Final Report
59
choice. The quadratic model was chosen merely to demonstrate that a non-linear model
for the relation between “ln(friction)” and slip speed would improve the standard deviation
for the repeatability and reproducibility of the EFI. It was decided to determine a specific
model for each device again. The following relation was used for the analysis:
ln( F ) = b0 + b1 ⋅ ( S − 30) + b2 ⋅ ( S − 30) 2 (3.8)
As a first step, this model was fitted for the devices that were used in the experiment with
BASt, DRI and DWW, because this was a limited dataset. One of the results of the
quadratic model is illustrated in Figure 3.1.
In this experiment, six sections, with the following surfacings, were tested:
• S1 Sheep foot rolled;
• S2 Cement concrete steel brushed;
• S3 Rolled mastic;
• S81, S82 Bituminous concrete;
• S4 Porous asphalt;
• S71-S76 Porous asphalt < 6 months.
The MPD-values measured on these sections are shown in Figure 3.2.
DEVICE: C5 SECTION: S4 SUBSEC: B

-0.54
ln(Friction)
-0.55
-0.56
-0.57
-0.58
-0.59
-0.60
-0.61
0 10 20 30 40
S-30
Figure 3.1 A quadratic function in the relation of ln(friction) and slip speed
FEHRL
HERMES Final Report
60
MPDISO
1.50
S
1.00
A
0.50
S1 S2 S3 S4 S71 S72 S73 S74 S75 S76 S81 S82
Section
Figure 3.2 MPD-values of the test sections in the experiment of BASt, DRI and DWW
Since the analysis of the limited database showed good results, the exercise was
repeated for the PIARC database. When the coefficients b0, b1 and b2 were calculated, S0
could be calculated. With this parameter, the F30 was calculated for each device, site and
run. Then, the mean value over the runs per device was calculated for each site and the
EFI was calculated as the mean value of F30 per site. With this data a linear regression
was fitted per device with y=EFI and x= F30. From this linear regression the “A” and “B”
values were calculated.
The next step was to use these “A” and “B” values to calculate the EFI per site for every
device, from EFI = A + B F30. A result of this calculation is shown in Figure 3.3, in which
the relation between EFI and measured friction is broken down by site. Separate colours
for every device identify the different devices. It can be seen that the speed dependency
no longer exists. However, different devices still report different levels for the EFI on the
same section.
FEHRL
HERMES Final Report
61
TIFF
are
QuickTime™
TIFF
are
needed
QuickTime™
TIFF
are
QuickTime™
(Uncompressed)
needed
needed
(Uncompressed)
(Uncompressed)
toand
see
totoand
aand
see
this
seeadecompressor
a
this
picture.
thisdecompressor
picture.
decompressor
picture.
Figure 3.3 Relation between slip speed and EFI with a quadratic model and calculated
S0 and S1
3.2.2.1 Quadratic model with estimation of S0 and S1

So far the coefficients S0 and S1 used in the calculations were the “real” S0 and S1 as
calculated from the measurements at different vehicle speeds. In normal service practice
it will not be possible to use calculated values for S0 and S1, because measurement
passes will usually only be made once and at only one test vehicle speed. Therefore, in
order to calculate values for EFI routinely, the coefficients S0 and S1 have to be estimated,
preferably from the texture of the road surface.
To investigate the relation between S0, S1 and MPD, a regression was performed for the
relations S0 vs. MPD and S1 vs. MPD. It was found from these regression lines that both
S0 and S1 were almost independent of the texture. However, the estimates of S0 and S1
were implemented in the quadratic model and again EFI-values were calculated. From
Figure 3.4 it can be seen that the results are very disappointing, because of the fact that
the EFI-value is again dependent on the original friction value of the device (and with the
vehicle speed of the device). This is further evidence that MPD alone is not the parameter
that explains the speed dependency of friction. Some other influences are also playing a
part in the tyre-road interaction.
FEHRL
HERMES Final Report
62
Site 10 A
device
C10 D3
1.0 C3B D4
C5 D5
C9 D6
D
D2 D8
0.8
D
cdevice
EFI
J AHH A F
C10 D3
A
E B
0.6 J E
I
C3B G D4
E G
A I C C5 H
J HF G C GF II D5
JH A C
BJ J B
B
F G C F I
D
H B BB
I
C9 I D6
G C
EG DD E D2 J D8
0.4 A
H
D
D
0.2
0 25 50 75 100
Slip speed (s)
Figure 3.4 Relation between EFI and slip speed for the quadratic model with estimated
S0 and S1
3.2.2.2 Conclusions regarding the quadratic model

From this analysis it can be concluded that a non-linear (quadratic) model in the relation
“ln(friction)” – slip speed can result in EFI-values that are independent of the vehicle
speed as long as the “real” values for S0 an S1 are used. The model using this relation
shows an improvement compared with the previous models, as can be seen from the
standard deviations for the reproducibility and the repeatability listed in Table 3.4, which
makes a comparison with the models described earlier.
Table 3.4 Standard deviations for the reproducibility and the repeatability
Standard deviation for

Repeatability Reproducibility
Original PIARC-model 0,08 0,14
New linear model with “real” coefficients 0,13 0,13
New linear model with estimated coefficients 0,11 0,11
Quadratic model with “real” coefficients 0,01 0,06
Quadratic model with estimated coefficients 0,08 0,08
From these standard deviations, both the repeatability and the reproducibility of the EFI
using the model with estimated coefficients were calculated to be 0,23. Although this is an
improvement compared with the original model, the accuracy of the predicted EFI is poor
and a device and speed dependency remains which has to be dealt with before accurate
EFI-values can be calculated. Likely candidates for factors involved in these
dependencies are the slip percentage, the measurement principle or influences like the
contact area or the contact pressure.
FEHRL
HERMES Final Report
63
In practice, the results of the latest analysis show that it is not possible to calculate a good
value for the EFI that is independent of the vehicle speed and the device. This also means
that the measurement values obtained with a particular device cannot be used to predict
accurately the measurement values of another device.
More research is needed to describe the parameters that are responsible for the
interaction between devices and sites and, with this, those leading to the speed
dependency and the different levels of the EFI found on any one site.
3.2.3 The LCPC-model
In parallel with the work at DWW, research into the improvement of the original PIARC-
model was carried out at LCPC. The LCPC approach was based on the lubrication
theories developed in tribology (friction-related research). The idea is supported by the
fact that many similarities exist between the lubrication mechanisms investigated in
tribology and the nature of the contact system that occurs between the tyre and road in
the presence of water.
Before a model based on tribology was developed, the limitations of the PIARC model in
the description of the friction versus slip-speed curve were investigated. The second task
was to propose a more general alternative model. The validation of the new model was
based on two criteria:
• How the model fitted the experimental data;
• The correlation between the model constants and measurable parameters such as
the road and vehicle characteristics.
3.2.3.1 Statistical analyses based on the original PIARC model

As with the work of DWW, the LCPC analysis started with some statistical analyses based
on the original PIARC-model. For this work, the data from the HERMES calibration
meetings were used. This database contained the following information:
• Device-related characteristics (slip ratio, tyre type);
• Surface characteristics (MPD);
• Friction values at various slip speeds.
From the friction values at various slip speeds, it is possible to derive S0 values for each
measurement series of slip ratio and MPD. Each meeting can then be treated as a
factorial design experiment in which the MPD and the slip ratio are "the factors" and S0 is
"the response". Analyses of variance were then performed to establish how S0 varied with
MPD for different slip ratios or, conversely, how S0 varies with slip ratio for different MPD.
Strictly speaking, each meeting was a factorial design experiment with unbalanced data,
since the number of repetitions was not the same for all measurement series. In the
analyses this was not taken into account. All data were treated as though they were sets
of balanced data. Therefore, inaccuracies might occur in the results.
For the calculation of S0, a regression was performed for each device and for each
surface. Firstly, to obtain S0, the exponential function y = a·ebx , where y represents the
friction and x the corresponding slip speed, was fitted to the data. Then, S0 was calculated
as -1/b. As an example of the results from this analysis, two graphs using data from
meeting 1,1 are shown in Figure 3.5.
FEHRL
HERMES Final Report
64
Interaction Plot Interaction Plot

500 Slip 500 MPD
0.18 0.19
400 0.34 400 0.48
1 0.54
300 300 0.57
So
So
1.09
200 200 1.54
1.66
100 100 2.23
0 0
0.19 0.48 0.54 0.57 1.09 1.54 1.66 2.23 0.18 0.34 1
MPD Slip
Figure 3.5 Example of the variation of S0 with MPD and slip ratio
The graphs show that there is a general tendency for S0 to increase for increasing values
of both MPD and slip ratio. However, two unexpected points can be seen on the graphs:
those for device F10 on section GB6 and for F13 on GB8.
Examples of the Friction versus Slip Speed graphs for these cases are shown in Figure
3.6. For device F10, friction is constant with slip speed, leading to a very high value of S0,
while for F13, friction is steady for the first two speeds, decreasing for the highest speed.
80
70
60
friction
F10
50
F13
40
30
20
0 10 20 30 40
S (km/h)
Figure 3.6 Friction – slip speed curves for F10 on GB6 and F13 on GB8
Overall, the analysis of variance generally showed a non-linear increase of friction for
increasing values of both MPD and slip ratio. However there were twenty-eight
measurement combinations of device and surface that did not respect this trend
(representing about 9% of the whole HERMES calibration campaign). Since no apparent
measurement error could be found, except that it was known that device F05 had reported
the wrong results, any explanation must rely upon physical phenomena that are not
properly described by the model used for the analyses. It was found that, typically, either:
• The friction did not vary with slip speed (referred to subsequently as “type I”
behaviour, similar to F10 above), or;
• The friction remained stable for the first two speeds then decreased for the highest
speed (referred to subsequently as “type II” behaviour, similar to F13 above).
FEHRL
HERMES Final Report
65
For both type of behaviour, the exponential model cannot be fitted to the experimental
data and the speed parameter concept does not have any physical meaning. Device
behaviour is generally of “type I” on porous asphalt surfaces; that is why this type of
surface is generally considered to be a "bad" surface in the context of model validation.
The 28 measurement series that, unexpectedly, did not follow the general tendency are
listed in Table 3.5.
Table 3.5 Summary of unexpected measurement series
Meeting 1,1 1,2 1,3 2,1 2,2 2,3 3,1 3,2 3,3
F01/ES5 F01/BE7 F05/FR1 F01/GB15 F05/BE9 F05/FR7
Measurement F10/GB6 F02/ES5 F05/BE4 F05/FR2 F11/NL9 F06/GB15 F05/BE11 F05/FR8
series F13/GB8 F03/ES5 F09/BE4 F05/FR3 F11/NL10 F13/GB9 F08/BE9 F05/FR9
F08/ES5 F10/BE7 F05/FR6 F13/GB10 B13/BE12 F05/FR12
It should be noted that a high proportion (11 out of 28, roughly 40 %) of the “unexpected”
measurement series related to device F05. However, this may have been caused by the
“wrong” results for this device having been used in the analyses. (Correct data were
eventually provided, but too late to be taken into account in this analysis.)
Another important comment that can be made is that, even though the different MPD
devices quite well with one another in general (see Section 2.4.3), there was nevertheless
considerable scatter between MPD values between different texture measuring devices.
Differences of up to 30 % were observed between T3 and T4, for example (Table 2.9).
This scatter could have a significant influence on the correlation between S0 and the
coupling of MPD and the slip ratio.
(1) Regression analysis to derive the S0 function

In the general case, the interaction graphs show almost no interaction between MPD and
slip ratio. Therefore, it was decided to derive S0 from both MPD and slip ratio using a
power function (this was an arbitrary choice; other non-linear functions should also work).
The following expression was used:
S 0 = a1 ⋅ MPD a2 ⋅ ( slipratio ) a3 (3.9)
This formula can be transformed using the logarithm function:
log( S 0 ) = log( a1 ) + a 2 ⋅ log( MPD) + a3 ⋅ log( slipratio ) (3.10)
Where log(a1), a2 and a3 are constants to be determined from multiple linear regression.
(2) Regression with all data

Multiple-linear regression was performed on the whole database (nine calibration
meetings) to derive values of the constants. The comparison between measured and
calculated (estimated) values of S0 is shown in Figure 3.7. From this graph it can be seen
that there is a large scatter around the line of equality, possibly due to the “unexpected”
values described earlier.
FEHRL
HERMES Final Report
66
Figure 3.7 Validation of the regression from formula 3.10
(3) Regression with all data except “unexpected” measurement series

In order to reduce the effects of the scatter, the analysis was repeated, omitting the
“unexpected” values listed in Table 3.5, to derive modified values for the constants.
Comparison between measured and calculated values of S0 is shown in Figure 3.8.
Figure 3.8 Validation of the regression from formula 3.10 excluding “unexpected”
measurement series
The suppression of the “unexpected” measurement series improves the regression

coefficient (R² increased from 0,63 to 0,69), confirming what can be seen in the graphs.
However, it can also be seen that there is still a large scatter around the line of equality in
Figure 3.8. Further, some curvature can be seen in the data points. Since it was shown
that the “unexpected” measurements are due to specific behaviour for which the
exponential model cannot be applied, it was decided to use the constants obtained after
discarding those outlying measurements series for the ongoing analysis. The final formula
relating the speed constant to the slip ratio and MPD is thus:
log( S 0 ) = 2,22 + 0,73 ⋅ log( MPD) + 0,87 ⋅ ( slipratio ) (3.11)
FEHRL
HERMES Final Report
67
Formula 3.11 can be rewritten as:
S 0 = 167 ⋅ MPD 0, 73 ⋅ ( slipratio ) 0, 87 (3.12)
Based on the results from the HERMES calibration meetings, the relation in formula 3.12
explains almost 70 % of the variation of the speed parameter using the surface MPD and
the vehicle slip ratio.
Additionally, a regression per device was investigated. From the individual regression
plots it could be seen that there was both large scatter and a curvature in the graphs.
Values of regression constants “a” and “b” for the device-specific relations are given in
Table 3.6. The relation between the constant “b” (device dependent) and the slip ratio is
shown in Figure 3.9.
Table 3.6 Constants for the regression log(S0) = a·log(MPD) + b for each device
F01 F02 F03 F04 F05

a 0,575 0,703 0,729 0,953 0,564
b 2,169 2,129 1,899 1,849 1,387
F06 F07 F08 F09 F10
a 0,90 0,89 0,983 0,23 1,201
b 1,610 2,319 1,795 1,965 1,645
F11 F12 F13 F14 F15
a 0,847 0,36 0,771 0,746 0,857
b 1,526 2,260 1,880 1,801 1,457
2.5
2.0
constant (b)
1.5
1.0
0.5
0.0
0 0.2 0.4 0.6 0.8 1.0 1.2
slip ratio
Figure 3.9 Relationship between constant “b” and slip ratio
It is of interest to note the close relationship between “b” and the slip ratio. Two types of
curve were then investigated to fit these points: polynomial and logarithm. The
regressions obtained were as follows:
b = −2,09 ⋅ ( slipratio ) 2 + 3,17 ⋅ ( slipratio ) + 1,05 (3.13)
FEHRL
HERMES Final Report
68
b = 2,21 + 0,84 ⋅ log( slipratio ) (3.14)
The first regression (3.13) has an R2 of 0,93 and, for the second (3.14), R2 was 0,88. The
second regression is interesting since it is similar to the sum of the first and third terms of
equation (3.11). Besides, the values are almost the same: 2,22 and 0,87 compared with
2,21 and 0,84 for equations (3.11) and (3.14) respectively.
To summarise, individual regressions lead almost to the same formula as does general
regression. The only difference is the value of the MPD exponent (constant “a” in Table
3.6), which is now device-related.
Individual regressions were also performed using an exponential model, that is to say:
S0 = a ⋅ e b.MPD (3.15)
or
ln( S 0 ) = ln( a) + b ⋅ MPD (3.16)
However, the regression coefficients were no better than those obtained with the power
model. Since individual regressions do not significantly improve the correlation between
S0, MPD and slip ratio, the general regression has been applied. The exponential model
can be now written as:
 −S 
F ( S ) = F0 ⋅ exp  0 ,87 
(3.17)
167 ⋅ MPD ⋅ (slipratio ) 
0 ,73
The F30 values are calculated for each surface, each device and each run using the
following formula:
 S − 30 
F30 = F ( S ) ⋅ exp  0 , 87 
(3.18)
167 ⋅ MPD ⋅ (slipratio ) 
0 , 73
For many of the measurement series for individual device and surface combinations, it
was observed that F30 (the friction measurement adjusted to the 30km/h reference speed)
was speed-dependent. An example of this is shown in Figure 3.10 for the measurement
series of device F02 on section FR1. In this graph, the data have been plotted in test run
order, with five repeated runs at each of three speeds, with higher speeds for the later
groups of runs and corresponding reductions in F30.
FEHRL
HERMES Final Report
69
100,0
80,0
60,0
F30
F2
40,0
20,0
0,0
0 5 10 15
run
Figure 3.10 Dependence of F30 on slip speed (F02 on FR1)
In such cases, σij will be high and the standard deviations for the repeatability and the
reproducibility will be high also. The new formula for S0, therefore, is still not satisfactory.
At this stage, it is worthwhile looking for other approaches that might provide a better
description of the friction versus slip-speed curve.
(4) Validation limits of the exponential model

The exponential model was proposed almost 30 years ago to describe the decrease of
friction for increasing slip-speed. The HERMES trial results have shown that this
description is not complete. Depending on the device operating conditions and the
surface, one of what has been described as the “Type 1” or “Type 2” patterns of behaviour
can be observed. These different kinds of behaviour are frequent for porous asphalt
surfaces and devices operating at low slip speeds.
A literature research showed that such behaviours correspond to actual physical

phenomena or, more exactly, to the lubricated contact between the tyre and the road.
Stribeck investigated a complete description of the friction versus slip-speed curve in
1902. The commonly observed friction – speed curve for tyre/road interaction can be
regarded simply as a part of the general Stribeck curve. The next section of the report
considers a new model based on this idea.
3.2.3.2 Investigation of a new model

In a Stribeck curve (Figure 3.11), friction between two surfaces separated by a lubricant is
presented as a function of the parameter:
η ⋅S
H = (3.19)
p
Where:
S is speed, which can be relative sliding speed depending on the authors;
η is the lubricant viscosity;
p is the apparent normal pressure.
FEHRL
HERMES Final Report
70
friction
zone 1 zone 2 zone 3
ln(ηS/p)
Figure 3.11 Example of the Stribeck curve
In Figure 3.11, three zones can be seen:

Zone 1: the lubrication is called "boundary" because lubricant film is present at the
interface, but its thickness is such that direct contact between the surfaces is
possible without adhesion;
Zone 2: the lubrication is called "mixed" because the lubricant supports part of the
contact area;
Zone 3: the lubrication is called "(elasto) hydrodynamic" because the lubrication film
separates the surfaces.
In 1996, a Dutch working group proposed a modified H parameter for tire/road contact [7]:
η a ⋅ Sb
H = (3.20)
p (h, V ) c ⋅ htd
Where:
p(h, V) is the effective contact pressure;
h is water film thickness;
V is vehicle speed;
ht is the texture depth;
a, b, c, d are positive constants.
The main question is how to determine the shape of the effective-pressure function.
Emmens [8] studied the mixed lubrication between two rough steel sheets, where friction
is generated on isolated spots and reduced by the lubricant pressure at the sheet
interface (Figure 3.12).
FN
F T, S
A
αA
h
plub, η
Figure 3.12 Pressure exerted by the lubricant at the interface between two sliding
surfaces
FEHRL
HERMES Final Report
71
According to Emmens, the lubricant pressure is given by:
η⋅S
p lub = λ ⋅ (3.21)
h2
Where:
plub is the lubricant pressure;
η is the lubricant viscosity;
S is the relative sliding speed between the sheets;
h is the average gap height created by surface roughness;
λ is a constant.
Due to surface roughness, the real contact area is a fraction αA of the total area “A”
(Figure 3.12). The normal load is divided into two components:
FN = Fm + Flub (3.22)
Where:
FN is the normal load;
Fm is a metal bearing component;
Flub is a lubricant bearing component.
Since Flub = plub· (1-α)·A, the metal part is given by Fm = FN – plub· (1-α)·A. The friction force
is given by:
FT = Fm ⋅ µ m + τ 0 ⋅ (1 − α ) A (3.23)
Where:
FT is the friction force;
µm is the coefficient of friction at boundary lubrication ;
τ0 shear stress in the lubricant.
Neglecting the lubricant shear stress in the case of water, the coefficient of friction is given
by:
 η ⋅S A
= µ m ⋅ m = µ m 1 − λ ⋅ 2 ⋅ (1 − α ) ⋅
FT F
µ=
FN 
(3.24)
FN FN  h
FN η ⋅S
Defining p = (normal pressure) and H = , the coefficient of friction is then a
A P ⋅ h2
linear function of H. Emmens found further that the gap height is best described by the
Rpm parameter, which is the mean distance between the roughness profile mean line and
highest peak.
It is of interest to note that the H parameter employed by Emmens is quite similar to that
proposed by the Dutch working group, and that the standard definition of Rpm is similar to
that of the MPD. This leads to the idea that, on the one hand, the effective pressure can
FEHRL
HERMES Final Report
72
be approximated by the normal pressure and, on the other hand, the “h” parameter can be
represented by the MPD. The H parameter (Equation 3.20) can then be written as
ηa ⋅ S b
H= ,
p c ⋅ MPDd
where p is the normal pressure. Before being able to applying this to the HERMES data,
the major difficulty was to gather normal pressure values for the devices participating in
the HERMES meetings. These values had been recorded in 1992 for the devices that took
part in the PIARC experiment [1]. However, since H is quite sensitive to variations of p,
current values of p would be preferable.
(1) Similarity between Stribeck and Tyre/road friction curves

A representation of friction as a function of slip speed for some of the devices participating
in the HERMES exercises is provided in Figure 3.13.
For devices F05 and F08, it can be seen that the experimental points start from zone 1 of
the Stribeck curve – where friction is almost constant – and go to zone 2, the intermediate
speed being the transition point. For other devices, a linear tendency corresponding to
zone 2 is observed. Actually, the “type I” and the “type II” behaviours mentioned in the
statistical analyses correspond to the boundary lubrication and to the mixed lubrication
conditions respectively.
100
80
F2
60 F5
F8
40 F12
F15
20
0
1 10 100
S (km/h)
Figure 3.13 Example of friction – slip speed curves (Trial 2,2 – Surface FR1)
In the opinion of the authors of this report, the problems in relating the speed parameter to
the MPD and the slip ratio that are caused by the measurement series for some device
and surface combinations are not due to measurement errors. These have simply
demonstrated that the exponential model is not always suitable for representing the
variation of friction with speed and/or MPD is not sufficient to describe the road surface
texture.
FEHRL
HERMES Final Report
73
(2) Application of the Stribeck presentation to tyre/road contact

The question now is how to represent the inverse S-shape of the Stribeck curve. In the
European collaborative project VERT (Vehicle Road Tyre interaction), the following
function was proposed [10]:
 b1 
µ = µ ref ⋅  b0 +  (3.25)
 1 + b2 ⋅ e (b3 + b4 ⋅V ) 
Where:µ is the friction;

µref is "reference" friction;
V is test speed;
bi are constants (i=1, ...., 4).
Emmens [8] proposed an arctangent function, which reproduces the first two zones well,
even if zone 3, where friction starts to increase, cannot be represented. Van den Bol [9]
proposed another logistic function called "the Gompertz function", which can be written in
the following form:
−β ⋅( S −µ )
F = α + γ ⋅ e−e +ε (3.26)
Where:F is friction;
S is slip speed in km/h;
α, β, γ, µ are constants;
ε is an error term.
Values of the constant β must be strictly positive to obtain the required inverse S-shape
variation. Since there are many constants to be estimated in this function, alternative
functions were investigated. Studies related to failure phenomena such as fatigue usually
employ the Weibull reliability function, which is written in the following form [11]:
α
R(t ) = 1 − e − β ⋅t (3.27)
Where: R is the reliability function;

t is time;
α, β are constants.
Physically, the reliability function expresses the probability that an item will function up to
a time (t) and will fail only after time (t). Example of variation of R(t) with the time (t) is
shown in the Figure 3.14 (the simulation represents α = 3, β = 0,01; time is expressed in
arbitrary units). Since it has an S-shape, the appropriate function to give the inverted S-
shape like a Stribeck curve is the complement, 1 – R(t).
FEHRL
HERMES Final Report
74
reliability function R(t)

0,8
0,6
0,4
0,2
0
0 5 10 15 20 25 30
time (t)
Figure 3.14 Example of the Weibull reliability curve
Among the various functions mentioned above, the complement of the reliability function
proposed by Weibull, seems to be the most promising since it represents the inverse S
shape, it is simple and its form is quite similar to the PIARC-model. For convenience, this
will now be referred to as "the Stribeck function" (although it is only an approximation of
the Stribeck Function).
Actually, equation (3.27) can be adapted and rewritten as :
α
 S  
F = F0 ⋅ exp  −    (3.28)
  S 0  
With
1 α1
S0 = ( ) .
β
In this form, formula 3.28 represents a generalization of the PIARC-model, which

corresponds to the case where α = 1. The S0 term still preserves its meaning and can be
still called "the speed constant".
Thus, it was decided to adopt Equation 3.28 as a new friction model. It should be noted at
this stage that no model can be used that is able to describe the small increase of friction
for high slip speeds, because it is not possible to make an estimate of the coefficients of
such a model (like the Gompertz function). At this stage, it was decided to use only the
exponential term in order to simplify the calculations.
3.2.3.3 Fitting the new model to the HERMES data

Having arrived at a new model based on the Stribeck function, the next step was to fit this
to data from the HERMES calibration trials. To simplify the calculations, the following
equation was used to fit the data:
α
F = F0 ⋅ e − β ⋅S (3.29)
The following tasks were then performed:
FEHRL
HERMES Final Report
75
• Fitting equation (3.41) to each measurement series of device and surface, giving the
values of the constants F0, α and β;
• Transformation of β into S0;
• For the whole HERMES database, correlation between S0 and physical parameters
such as surface MPD or device slip ratio.
(1) Fitting constraints

In theory, for each measurement series of device and surface, there are between 9 and
15 values (3 slip speeds repeated 3 to 5 times) to fit a 3-constant function. In practice,
however, the values are grouped around 3 slip-speed levels, which are very close to each
other for low slip ratio devices. The major difficulty in fitting, then, is how to generate a
realistic inverse S-shape curve from only 3 values.
The first tentative attempts at fitting showed that F0 could be very high. Since this constant
represents the friction at very low slip speeds and friction is actually measured on wet
roads, it was thought that F0 could be reasonably imposed to be less than 1.
For the constant β, in order to reproduce the decrease in friction, values were required to
be positive.
It was noted that the relative widths of the zones 1 and 2 of the Stribeck curve depend on
both α and β. Therefore, the value of α was adjusted by trial and error to obtain the
observed zone 1 width from the HERMES data. It was found that α could be kept constant
and that the best fit was obtained for α = 3.
In summary, the following constraints were imposed when fitting the new model to each
measurement series for a surface and device combination, in order to reproduce an
estimate of the Stribeck curve: F0 < 1; β > 0; α = 3. It should be mentioned that these
constraints are quite arbitrary and further investigation is needed, especially to derive
eventually a physical meaning for the exponent α.
(2) Results of Fitting

Values of the constants F0, β and S0 (recalculated from β) are reported in Annex E for
Trials 1,1 to 3,3 respectively. Examples of the fitted curve are shown in Figure 3.15. On
the left-hand graph, even though the fitted curve appears to be a good representation,
uncertainties still exist, mainly in the right-hand part of the curve (corresponding to high
slip speeds) due to the absence of experimental data. The graph on the right shows that
the new model does not work as well as the exponential model in the case of locked-
wheel devices. This is caused by the constraints that are imposed in the curve-fitting; the
experimental data showed a clear S-shape.
PFT(F9)
SCRIM(F13)
1
1
0.8
0.8
0.6 mesures
0.6 mesures
F
F
0.4 modèle
0.4 modèle
0.2 0.2
0 0
0 20 40 60 80 100 0 20 40 60 80 100
S (Km/h) S (Km/h)
Figure 3.15 Examples of model fitting
FEHRL
HERMES Final Report
76
It could be said at this stage that, with the exception of locked-wheel devices the new
model fits the experimental data well. However, to be of practical use, the model
constants must be related to physical parameters. Investigation was therefore focused on
the relationship between the speed parameter S0, the surface MPD and the device slip
ratio. The constant F0 is ignored at first, since it is not required for the calculation of the
EFI (F0 disappears when expressing F(30) as a function of F(S)).
3.2.3.4 Relationship between S0, surface MPD and device slip ratio
Two steps are required to establish a formula relating S0 to MPD and slip ratio:
• Perform the analysis of variance for each HERMES meeting to derive the variation
of S0 with the MPD for different values of the slip ratio (and vice versa).
• Carry out the regression between S0, MPD and slip ratio.
(1) Analysis of variances

An example of graphs of variation of S0 is shown in Figure 3.16 (for Trial 2,2). It can be
said that S0 varies non-linearly with the slip ratio and, to a lesser degree, with the MPD.
The “unexpected” measurement series of device and surface are summarised in the Table
3.7.
Interaction Plot Interaction Plot

180 Slip 180 MPD
0.14 0.09
150 0.15 150 0.34
120 1 120 0.57
0.73
So
So
90 90 0.9
1.54
60 60
30 30
0 0
0.09 0.34 0.57 0.73 0.9 1.54 0.14 0.15 1
MPD Slip
Figure 3.16 Example of the variation of S0 with MPD and slip ratio
Table 3.7 Summary of unexpected measurement series
Meeting
1,1 1,2 1,3 2,1 2,2 2,3 3,1 3,2 3,3
F05/BE10
Measurement F05/BE3 F05/FR12
F05/BE11
series F05/BE13
No explanation has been found for these special cases. It can be noted that there are
fewer unexpected measurement series with the new model (5 cases) than with the original
exponential model (28 cases) and that some are different. This is further evidence of the
fact that the unexpected behaviour is caused by inaccurate models and not by any
physical problem. The most significant fact is that, this time, all the “unexpected”
FEHRL
HERMES Final Report
77
measurement series were related to the F05 device, which might have been caused by
the use of partially-incorrect data for this device.
(2) Regression
Next, it was proposed that S0 should be related to the MPD and the slip ratio using
Equation 3.30:
S 0 = a1 ⋅ MPD a2 ⋅ ( slipratio ) a3 (3.30)
Multiple-linear regression was performed on the logarithmic transformation of Equation

3.30 using the whole HERMES database except the special cases listed in Table 3.7. It
was found that the regression coefficient was high (R² = 0,91), demonstrating the
improved relevance of the model (a coefficient of 0,69 was obtained with the PIARC-
exponential model).
The comparison between measured and calculated values of log10(S0) is shown in Figure
3.17. Again, a curvature and a large scatter around the line of equality can be seen in this
graph.
The fitted relationship between S0, MPD and the slip ratio is then:
log( S 0 ) = 0,27 ⋅ log( MPD) + 0,92 ⋅ log( slipratio ) + 2,09 (3.31)
This regression analysis led to a subsidiary question: would it be possible to further

improve the prediction of S0 when device-specific relationships between S0 and MPD are
developed?
Figure 3.17 Validation of the regression log(S0) = log(a1) + a2·log(MPD) + a3·log(slip ratio)
(3) Derive device-dependent constants by individual regression between S0 and MPD

Data used to derive the relation (3.31) were divided into 15 sub-databases: each sub-
database included all S0 values obtained by one device from different HERMES meetings
as well as the corresponding values of MPD. An example of the graphs describing the
relation between S0 and MPD is shown in Figure 3.18. The colours identify the different
trials. It can be seen that the non-linear variation of S0 with MPD is more or less
pronounced, depending on the device.
FEHRL
HERMES Final Report
78
200,0 200,0
So (km/h)
150,0 150,0
So (km/h)
100,0 100,0
50,0 50,0
F4 F5
0,0 0,0
0,00 1,00 2,00 3,00 4,00 0,00 1,00 2,00 3,00 4,00
MPD (mm) MPD (mm)
Figure 3.18 Example of S0 – MPD graphs from different comparison meetings put
together.
Linear regression was performed on log10(S0) (dependent variable) and log 10(MPD)
(independent variable) to give:
log( S 0 ) = a ⋅ log( MPD) + b (3.32)
Values of the regression constants for each device are given in Table 3.8.
Table 3.8 Constants for the regression log1 0(S0) = alog10(MPD) + b for each Device
F01 F02 F03 F04 F05

a 0,158 0,191 0,265 0,361 0,191
b 2,032 2,029 1,684 1,665 1,269
F06 F07 F08 F09 F10
a 0,321 0,233 0,434 0,095 0,488
b 1,428 2,054 1,603 2,016 1,438
F11 F12 F13 F14 F15
a 0,428 0,095 0,249 0,291 0,291
b 1,402 2,086 1,688 1,547 1,307
It is interesting to notice that again there was a very close relationship between the
device-dependent constant “b” and the slip ratio (Figure 3.19).
2,50
2,00
constant b
1,50
1,00
0,50
0,00
0 0,2 0,4 0,6 0,8 1 1,2
slip ratio
Figure 3.19 Relationship between device-specific constant (b) and slip ratio
FEHRL
HERMES Final Report
79
Logarithmic fitting was performed, giving the following formula:
b = 0,90 ⋅ log( slipratio ) + 2,07 (3.33)
Formula 3.33 can now be written as:
log( S 0 ) = a ⋅ log( MPD) + 0,90 ⋅ log( slipratio ) + 2,07 (3.34)
Where: a is a device-specific constant.
The formulae 3.31 and 3.34 are quite similar. It can thence be concluded that individual
regression leads to the same formula as that given by the general regression. The only
difference is that the value of the MPD exponent is now device- dependent.
The values of S0 that emerged from fitting the physical data, called "observed values",
were then compared to those calculated from equations 3.43 and 3.46. For each device,
the relative difference between the observed and calculated values - the ratio ((S0
observed – S0 calculated)/S0 observed) - was determined (each ratio was related to a
specific road surface on which the friction measurements had been carried out) and
averaged. The mean ratios related to the general regression were then compared with
those related to device-specific regressions (Figure 3.20). The prediction was good when
the ratio was low. It can be seen that the prediction of the speed constant is slightly
improved when device-specific constants, mainly the MPD exponent, are used. From the
graph it can be seen that, for some devices this is more than a “improvement”. On the
other hand, the method chosen to compare both methods is not capable of determining
systematic effects.
0,35
calculated)/observed
0,30
0,25
(observed -
0,20
0,15
0,10
0,05
0,00
device
device-specific regression general regression
Figure 3.20 Improvement of the prediction of S0 using device-specific parameters
3.2.3.5 Resulting friction model based on tribology approach

The following formula is proposed for the calculation of the friction at a slip speed S,
based on the Stribeck function:
S 3
−( )
117⋅( MPD) a ⋅ ( slipratio)0 , 9
F ( S ) = F0 ⋅ e (3.35)
FEHRL
HERMES Final Report
80
Where:F0 is the “theoretical” friction at 0 km/h slip speed;

S0 is the speed constant, in km/h;
a is a device-specific constant.
Since the Stricbeck curve shows a stable-friction zone for low slip speeds, the constant F0
can be estimated from friction measured at low speeds; less than 20 km/h, say. Values of
the constant “a” for the devices participating in the HERMES calibration trials are
summarized in Table 3.9.
Table 3.9 Constant (a) of LCPC model for each HERMES Device
F01 F02 F03 F04 F05

a 0,158 0,191 0,265 0,361 0,191
F06 F07 F08 F09 F10
a 0,321 0,233 0,434 0,095 0,488
F11 F12 F13 F14 F15
a 0,428 0,095 0,249 0,291 0,291
3.2.4 Investigation of alternative characteristics of surface texture depth
In earlier Sections, it has been shown that MPD alone is not necessarily the correct
explanatory factor to use to predict the speed dependency of the friction devices. This
observation prompted an additional analysis at TRL to attempt to determine whether other
characteristics of the surface texture might be more successful.
For sites where the texture profile data were available, a number of additional
characteristics of the texture depth were calculated. The texture profile data were filtered
to pass wavelengths within octave bands centred at 2, 4, 8, 16, 31,5, 63, and 125 mm, as
well as the 2 mm to 100 mm wavelength range used for the MPD calculation. The filtered
profiles were then characterised by:
• the root mean square (RMS);
• ten percentile and ninety percentile measures of the filtered profile heights;
• the skewness of the distribution of profile heights.
A method similar to the MPD calculation was also applied at each of the different
wavelength ranges. In this method, for each 100 mm length of the filtered profile, the
highest profile point in each half of the 100 mm length was identified and the average of
these two values was calculated. This method is referred to below by the shorthand
“peak-pick” (i.e., choosing the highest point) to distinguish it from MPD, which requires a
specified wavelength range. The MPD values from the HERMES database and the
Sensor Measured Texture Depth (SMTD) method, a root mean square texture depth
method used in the UK since the late 1970s, were also included.
For analysis, the friction devices were grouped into categories by their slip ratio, taking
note of the observation that the speed dependency was influenced by the slip ratio. The
devices grouped in each category are listed in Table 3.10.
FEHRL
HERMES Final Report
81
For each category, the correlation coefficient for the linear regression between the “real”
S0 values and the values of each texture characteristic was calculated. The R2 values for
the most highly correlated texture characteristics are listed in Table 3.11.
Table 3.10 Device grouped by slip ratio
Grouping criterion Friction device

F05 Grip Tester
F06 ROAR
Low slip ratio (<0,2) F10 OSCAR
F11 ROAR
F15 IMAG
F03 SCRIM
F04 SCRIM
Intermediate slip ratio (0,34) F08 Odoliograph
F13 SCRIM
F14 Odoliograph
F01 DWW trailer
F02 ADHERA
High slip ratio (0,86 – 1,0) F07 ROAR
F09 PFT
F12 SRT3
Table 3.11 Texture characteristics most highly correlated with S0 values
High slip ratio Intermediate slip ratio Low slip ratio

2 2 2
Rank Characteristic* R Rank Characteristic R Rank Characteristic R
8 mm; 2-100 mm; 2-100 mm;

1 0,07 1 0,29 1 0,23
10 percentile peak-pick 90 percentile
4 mm; 2-100 mm; 2-100 mm;
2 0,07 2 0,25 2 0,21
10 percentile 90 percentile RMS
4 mm; 63 mm; 2-100 mm;
3 0,07 3 0,22 3 0,21
90 percentile peak-pick peak-pick
8 mm; 125 mm;
4 0,06 4 0,22 4 UK SMTD 0,21
90 percentile peak-pick
8 mm; 63 mm; 16 mm;
5 0,05 5 0,22 5 0,21
peak-pick 90 percentile 10 percentile
MPD 63 mm; 63 mm;
7 0,05 7 12 0,20
(HERMES) RMS 0,21 RMS
2-100 mm; 16 mm
13 0,02 14 UK SMTD 0,18 13 0,19
peak-pick RMS
MPD MPD
22 UK SMTD 0,01 19 0,14 20 0,17
(HERMES) (HERMES)
* Characteristic texture parameter listed as (wavelength range; method)
FEHRL
HERMES Final Report
82
In all cases, the low overall amount of variation explained by any individual texture
characteristic was disappointing. Furthermore, the performance of the 2 to 100 mm peak-
pick calculation performed for this analysis and the MPD values from the HERMES
dataset (both highlighted in Table 3.11) were different. Since these methods are
ostensibly the same, this again highlights problems with the definition of the MPD method.
As judged by the R2 values, the 2 to 100mm peak-pick method is the best single texture
parameter, or close to the best method, for both the intermediate and low slip ratio
categories. The ninety percentile method using the same wavelength range ranks highly,
as does the RMS-method for the low slip ratio devices. The ninety percentile method
would be worth considering as an alternative to MPD because it uses at least 10 texture
profile values to determine the texture.
For the intermediate and low slip ratio categories, the long wavelength features seem to
be the most important in determining the speed dependency. In contrast, for the high slip
ratio devices, the short wavelength texture characteristics are the most successful.
However, the correlation, while significant at the 90% level, explains only a very small
amount of the overall variation in S0 values for the high slip ratio devices.
Although it would be desirable to study the individual friction devices separately, the large
amount of variability in the measurements for individual devices in this dataset made it
difficult to determine trends for the individual friction devices with confidence. (This was
partly because texture profile data were not available for all the HERMES test sites).
Therefore, two larger UK databases were investigated to see if the larger amount of data
would make it possible to establish more robust relationships for the two UK friction
devices.
The first database consisted of friction measurements at 20, 50, 80 and 100 km/h for
device F09 (using the locked wheel method). These measurements were obtained from
nearly 150 UK sites covering a wide range of surface types. These friction values were
used to determine “real” S0 values, as described above. However, the values obtained
may not correspond exactly to those measured in the HERMES experiment because the
speed range is different. Texture measurements were available as Sensor Measured
Texture Depth (SMTD) values from the UK SCRIMtex, which, as well as participating in
HERMES, also participated in the PIARC experiment. MPD data were not available for
these sites, but MPD values are typically a factor of between 1 and 1,4 times larger than
SMTD values, depending on the surface type.
Figures 3.21 & 3.22 show that, for these data, the S0 values are highly correlated with the
texture depth, through a linear or power relationship, but it can also be seen that a
substantial scatter about the overall trend remains, particularly when porous surfacing
materials are included. The overall standard deviation of the residuals, excluding porous
materials, is 16,7, although this may be dependent on the SMTD value.
FEHRL
HERMES Final Report
83
300
y = 75,3x + 38,3
2
R = 0,79
250
200
S0
150
100 0,54
y = 113,4x
2
R = 0,78
50
0
0,0 0,5 1,0 1,5 2,0 2,5
SMTD (mm)
Figure 3.21 Relation between S0 and texture depth for device F09 (UK dataset, excluding
porous materials). Solid line: linear regression. Axis line: exponential
regression.
300
250
y = 77,8x + 43,8
2
R = 0,52
200
S0
150
100 0,57
y = 120,5x
2
R = 0,63
50
0
0 0,5 1 1,5 2 2,5
SMTD (mm)
Figure 3.22 Relation between S0 and texture depth for device F09 (UK dataset including
porous materials). Solid line: linear regression. Axis line: exponential
regression.
The choice of test tyre for device F09 (smooth or ribbed) also has a substantial influence
on the value of S0, as indicated in Figure 3.23, which shows the distribution of values
obtained on the same sites with different tyres. This is expected, because it is known that
FEHRL
HERMES Final Report
84
measurements using ribbed tyres are less influenced by speed than measurements using
smooth tyres. However, this result suggests that the different characteristics of the test
tyres could be one reason for the need for S0 values to be specific to the friction device.
40
30
Smooth tyre
N 20 Ribbed tyre
10
0
0-25 25-50 50-75 75- 100- 125- 150- 175- 200- 225- 250-
100 125 150 175 200 225 250 300
S0
Figure 3.23 Comparison of S0 values measured using smooth and ribbed tyres
The second dataset comprised measurements on over 45km of in-service roads made at
two test speeds by device F13 (the UK SCRIM). For each 100 m section, the “real” S0
value was calculated from these two measurements. Determining the S0 value in this way
(i.e., from only two points) will be less precise than in the earlier analysis but this is offset
in the analysis by the large amount of data available. The relationship between S0 and
SMTD is shown in Figure 3.24, by surface type. A small number of extreme S0 values
(those below zero or above 300) have been excluded from this analysis.
300
200
S0
100
0
0,3 0,8 1,3 1,8
SMTD (mm)
Cement concrete Hot rolled Porous asphalt Surface dressing Unknown
asphalt
Figure 3.24 Relation between S0 and texture depth for device F13 (UK dataset)
FEHRL
HERMES Final Report
85
In contrast with device F09, where there was a clear trend with the texture depth, for
device F13, the trend with texture depth is not significant at the 90% level. However, the
data indicate that the surface type will be significant in determining the S0 value for this
device.
This additional analysis indicates that different features of the surface type and surface
texture influence the speed dependency of the different friction measurement devices.
The variability of the measurements exhibited by the two UK devices when examined on a
larger dataset suggest that it may be necessary to determine the characteristics of the
individual devices on a much wider scale than was possible within the HERMES
experiment. Furthermore, should some devices be found to be more sensitive to effects
that cannot be adequately measured using current technology then it may be necessary to
consider whether they can be fully included in the harmonisation procedure.
3.3 Discussion
From the analyses in this chapter it can be concluded that, although the models can be
improved, the accuracy of the EFI is still poor. Where individual devices show values for
repeatability and reproducibility of about 0,03 and 0,04, the EFI has a repeatability of 0,23
and a reproducibility of 0,23 – 0,40 depending on the model that is used for the
calculation. This means that the EFI (and a comparison between devices) is not
sufficiently accurate for the purposes for which skid resistance measurements are
normally used (assessment of skid resistance after accidents, acceptance of new works,
and planning of maintenance works).
The inaccuracy can be understood by looking at the procedures for the calculation of the
EFI. In Figure 3.25 a theoretical example of a friction curve is shown. In this graph,
measurements are indicated for three devices with different slip percentages at three
different speeds.
Friction value
1,2
1
0,8
0,6
0,4
0,2
0
0 50 100 150
Slip percentage
Figure 3.25 Theoretical friction curves at a site X for three different devices.
With this information, the figure can be transformed to a graph showing the relation
between friction value and slip speed. This graph is shown in Figure 3.26.
FEHRL
HERMES Final Report
86
1,2
1
Friction value
0,8
device A
0,6 device B
0,4 device C
0,2
0
0 20 40 60 80 100
Slip speed (km/h)
Figure 3.26 Relation between the friction value and slip speed for site X
From Figure 3.26, the reason why a reference slip speed of 60 km/h gives such poor
results becomese apparent. Transforming a measurement value at a given slip speed to a
slip speed of 60 km/h means an extrapolation for all devices with the exception of those
with high slip percentages and devices with locked wheel. A reference speed of 30 km/h
means less extrapolation, demonstrating that the decision to use 30 km/h as reference
slip speed for EFI that was made in the Belgian study [2] was sound.
However, extrapolation is still necessary for devices with a low slip percentage and with
this extrapolation a source of error is introduced. This can be shown in Figure 3.27, in
which linear and quadratic regression lines are drawn for the three devices.
1,2
1
device A
Friction value [-]
0,8 device B
device C
0,6 Quadratic (device A)

Linear (device A)
0,4 Linear (device C)
Linear (device B)
0,2 Quadratic (device B)
Quadratic (device C)
0
0 20 40 60 80 100
Slip speed (km/h)
Figure 3.27 Example of different methods for the regression analysis
Devices with high slip percentages (and therefore a larger difference in slip speed with the
three vehicle speeds used in the experiments) show only small influences of the model
that is used. On the other hand, devices with low slip percentages (most of the devices in
the experiments) show a large influence of the model used for the calculation of F30. A
way to overcome this problem would be to operate devices with low slip percentages at
FEHRL
HERMES Final Report
87
other (higher) speeds than the devices with high slip percentages. This, however,
introduces new problems such as traffic safety during calibration meetings and the water
flow needed when measuring at higher speeds.
Since the different devices all measure on a different part of the friction/slip curve (before,
around or after the peak of the curve), it is difficult to find a general model to describe the
speed dependency of the friction measurement. New research into the interaction
between devices and sites is necessary to develop better models for the calculation of a
common scale for friction measurements. When new parameters are found or derived
from existing data it might even be possible to use non-linear models, from which it has
been shown that an improvement can be achieved. The practical problem at present that
no, or at least insufficient, physical parameters are available to determine the constants of
the more complex models.
Unfortunately, for the next few years, the normative Annex A [3] will be the best available
for a standard regarding road surface friction. A lot of extra research into the interaction
between devices and sites is necessary before an accurate harmonized friction value will
be available.
For the short term, it might be a solution to use different relationships for devices
operating at almost the same slip ratio. This, again, is far from a harmonised friction value.
Another possibility is to define pairs of devices of which the measurement results at the
normal operating speed of these devices show a good correlation against each other. The
data from the PIARC experiment and the HERMES calibration exercises should give
enough information to determine these relationships between devices.
The most important question to answer, however, is what level of accuracy is needed,
either using models or by direct comparison of devices.
FEHRL
HERMES Final Report
88
4 Alternative processing of experimental

data
4.1 Introduction
Section 2.4 presented an initial anlysis that followed the calculation procedures and used
the models specified in the draft CEN standard. Chapter 3 has explored possible ways of
improving the models. A further important part of the HERMES project, partly carried out
in parallel with the model development work, was to use the data from the calibration
meetings to explore ways of improving the reliability and precision of the calibration
procedure.
This chapter presents successive attempts to improve the procedures by applying

different treatment scenarios to the experimental data, including utilising the more
promising of the ideas discussed in Chapter 3.
The two scenarios already considered in Section 2.4 and the twelve more considered in
this chapter are brought together and summarised in a table at the end of this part of the
report, in Section 4.14.
4.2 Improved model for S 0 vs. MPD
In Section 2.4.9, it was pointed out that when plotting S0 vs. MPD the scatter is rather
wide. Looking at the graphs presented in Annex C, it appears that the relationship
between S0 and MPD depends on the friction device. This means that “a” and “b” in
equation (2.8) should be considered as device-specific. But then it is clear that a linear
equation is no longer adequate, for two reasons.
Firstly, as illustrated by the example in Figure 4.1, back-calculating S0 for small values of
MPD can yield small, negative values of S0, which is not acceptable because it can lead to
enormous, absurd, values of F30 and EFI. This happened eight times in the HERMES
data, in Trials 1,1, 2,2 and 3,3 where very smooth surfaces were tested (Annex C).
FEHRL
HERMES Final Report
89
Trial 2,2 - Device F02
800
700
600
500
S0 (km/h)
400
300
200
100
0
-100
-200
0 0,2 0,4 0,6 0,8 1 1,2 1,4 1,6 1,8
MPD (mm)
Figure 4.1 Example illustrating the need for a non-linear relationship between S0 and
MPD. The arrow points to a case where the predicted value for S0 is negative.
Secondly, a non-linear curve provides a better fit to the data. It has been found (Section
3.4.3) that, in general, a power law offers the closest fit, i.e.:
S 0 = a ⋅ MPD b (4.1)
where “a” and “b” are device-specific, as are “A” and “B”.
Values were determined for “a” and “b” using a least squares regression on the
logarithmic form of (4.1), i.e.:
ln S 0 = ln a + b ⋅ ln MPD (4.2)
As there were still many cases where the scatter was rather wide, a weighting was applied
in the calculation in order to minimize the influence of less significant S0 values. To that
end, the weighting (w) was set to be inversely proportional to the squared relative
standard deviation s So of S0 with respect to the regression line in equation (2.20). Then,
the regression equation becomes:
w ⋅ ln S 0 = w ⋅ ln a + w ⋅ b ⋅ ln MPD (4.3)
with
w = (S0 / σ So )2 (4.4)
A good example of the effect of weighting is shown in Figure 4.2. Annex F presents all the
graphs of S0 vs. MPD with the fitted weighted and unweighted curves.
FEHRL
HERMES Final Report
90
250
S0 (km/h) 200
150
100
50
0
0.0 0.5 1.0 1.5 2.0
MPD (mm)
Figure 4.2. This is an example of the effect of weighting in the regression of S0 vs. MPD.
The solid and dotted lines represent the weighted and unweighted
regressions respectively
In all subsequent processing of the experimental data, the new, weighted, model has
been applied, except in Scenario #12 (Section 4.11).
4.3 Statistical tests
As suggested in Section 2.4.12, an attempt to improve the procedure by using different

statistical tests has been made by applying the consistency tests described in ISO 5725-2
[5] while considering the different friction devices as different “laboratories”. The object of
the measurement is the “true” EFI-value of a given surface. We call the 5 repeated runs at
the 3 different speeds “a measurement series”. The “k-test” detects the measurement
series made by a given device on a given surface that exhibits a repeatability standard
deviation (s r) excessively large in comparison with the average repeatability of all the
devices on the surface being considered. The “h-test” detects those measurement series
made by a given device on a given surface that exhibit an average EFI-value (<EFI>)
excessively deviating from the grand average (<<EFI>>) of EFI-values reported by all the
devices on that surface. The latter test has, of course, to be applied after calibration, i.e.
on EFI-values re-calculated using Anew and Bnew .
When these tests were applied to the HERMES data, a probability level of 0,5% was
used. The “k-test” was applied first and the measurement series that did not pass that test
were discarded. The “h-test” was subsequently applied to calibrated EFI-values and those
measurement series that failed were discarded. Then, <<EFI>> had to be re-calculated,
together with “A” and “B”, using the remaining data.
In addition to those standard tests, a test has been applied on the correlation coefficient
(RF2) of the regression F against S. The measurement series with RF2 < 0,5 where
discarded from the calculations.
A similar test was made on the correlation coefficient (REFI 2) of the calibration regression
equation: << EFI >>= α + β ⋅ EFI . Cases where REFI 2<0,5 were flagged but not
discarded.
FEHRL
HERMES Final Report
91
4.4 Scenario #2: applying new statistical tests
In Scenarios #0 & #1 used in Section 2.4, the statistical tests proposed in the draft
standard [3] were applied. As an alternative, in Scenario #2, the four statistical tests
described in Section 4.3 were applied. Table 4.1 reports the percentage of measurements
eliminated by the different tests.
Table 4.1 Percentage of measurements eliminated by the different tests
Test Proportion eliminated at each trial

Total
applied
1,1 1,2 1,3 2,1 2,2 2,3 3,1 3,2 3,3
RF2 < 0,5 3,3% 0,0% 0,0% 6,7% 8,0% 2,5% 4,2% 5,2% 4,8% 4,2%
k>k0,5% 14,0% 20,9% 12,5% 16,6% 16,7% 16,7% 25,0% 14,5% 14,3% 16,4%
h>h0,5% 16,2% 13,3% 12,5% 5,8% 3,3% 11,4% 20,8% 2,4% 0,0% 8,5%
Total 33,4% 34,2% 25,0% 29,2% 28,0% 30,6% 50,0% 22,1% 19,1% 29,1%
The test on RF2 appears to be both effective and useful since it eliminates really poor
measurement series, as can be assessed by looking at the specific cases in Annex B. The
cases of interest in this context are listed in Table 4.2. In all subsequent Scenarios, that
test will be applied.
It appears that the consistency tests are again unacceptably severe. In some cases, the
application of the “k-test” did not allow the analysis to move on to the calibration
calculation because only the results from one surface were left. This happened for device
F05 in Trial 2,1 and device F11 in Trial 2,3. Moreover, despite having discarded failing
measurement series, in 33% participations (14 out of 42), devices were flagged because
REFI 2 remained lower than 0,5.
It was concluded that the “h- and k-tests” are helpful, but they cannot be used in practice
with the linear models because of the inaccuracy of these models.
FEHRL
HERMES Final Report
92
Table 4.2 Measurement series exhibiting a correlation coefficient of F(S) lower than 0,5
Trial Device Surface

1,1 F10 GB06
2,1 F05 BE03
2,1 F05 BE07
2,1 F09 BE04
2,1 F10 BE07
2,2 F08 FR04
2,2 F08 FR06
2,2 F15 FR06
2,3 F11 NL10
3,1 F06 GB05
3,2 F08 BE09
3,2 F13 BE12
3,3 F10 FR09
3,3 F10 FR11
4.5 Scenario #3: applying simpler statistical tests
Here, only the test on REFI 2 is applied. But, in addition, a test is made on whether the
coefficient of variation (CVEFI ) of the range of EFI values reported by the device being
considered is larger than 10% (Scenario #3a) or 5% (Scenario #3b). This supplementary
test is designed to prevent calibration regressions based on a very narrow range of EFI
values, even with a high correlation coefficient, to be taken into account and spoil the
whole procedure. The worst of such cases is shown in Figure 4.3.
With this scenario, the devices that did not pass either the test on REFI 2 or on CVEFI in a
given trial were not considered to be calibrated. In that event, for this analysis, the “old”
values for “A” and “B” were kept until they next participated in a trial. If any reference
devices were in this category, they were treated as though they were not reference
devices in that trial and the calculations were repeated without including their EFI-values
in the grand average <<EFI>>. (In practice, a device that was not calibrated would not be
able to be used to provide EFI values until it had successfully participated in a future trial).
The outcome of this approach is presented in Tables 4.3 and 4.4.
FEHRL
HERMES Final Report
93
0,8
0,7
0,6
0,5
0,4
0,3
0,2
y = 3,564x - 0,990
0,1 R2 = 0,593
0
0 0,05 0,1 0,15 0,2 0,25 0,3 0,35 0,4 0,45 0,5
EFI
Figure 4.3 An example of a calibration regression based on a very narrow range of EFI
values, while still exhibiting an acceptable correlation coefficient.
Table 4.3 The outcome of Scenario #3a. Superscript R means that the calculations have
been repeated after the reference devices failing either test were discarded.
2
Trial Device Reference REFI CV EFI
F09 YES 0,914 0,22
F10 NO 0,638 0,68
1,1
F12 YES 0,882 0,29
F13 YES 0,941 0,41
F01 YES 0,020 0,06
F04 YES 0,777 0,29
1,2
F06 NO 0,798 0,29
F07 NO 0,171 0,07
F04 YES 0,826 0,29
1,2 R F06 NO 0,701 0,29
F07 NO 0,078 0,07
F01 YES 0,850 0,12
F02 YES 0,792 0,10
1,3
F03 YES 0,905 0,12
F08 NO 0,285 0,24
F01 YES 0,709 0,14
F04 YES 0,798 0,15
F05 YES 0,945 0,19
2,1 F06 YES 0,887 0,16
F09 YES 0,333 0,14
F10 YES 0,387 0,19
F14 YES 0,788 0,09
F01 YES 0,812 0,14
F04 YES 0,884 0,15
2,1 R
F05 YES 0,868 0,19
F06 YES 0,741 0,16
F02 YES 0,602 0,21
F05 YES 0,596 0,05
2,2 F08 NO 0,303 0,41
F12 YES 0,940 0,27
F15 NO 0,161 0,76
2,2 R F02 YES 0,594 0,21
F08 NO 0,260 0,41
F12 YES 0,952 0,27
FEHRL
HERMES Final Report
94
2
F15 NO 0,132 0,76
F01 YES 0,444 0,07
F03 YES 0,943 0,13
F04 YES 0,947 0,24
2,3
F11 NO 0,851 0,64
F13 YES 0,612 0,07
F15 NO 0,842 0,97
F03 YES 0,944 0,10
F04 YES 0,946 0,10
2,3 R F11 YES 0,973 0,09
F13 NO 0,537 0,09
F15 YES 0,974 0,09
F03 YES 0,960 0,10
2,3 RR F04 YES 0,962 0,10
F13 NO 0,578 0,09
F01 YES 0,892 0,21
3,1 F06 YES 0,913 0,31
F13 YES 0,956 0,26
F04 YES 0,877 0,07
F05 YES 0,791 0,14
F08 NO 0,692 0,12
3,2
F13 YES 0,279 0,08
F14 YES 0,671 0,07
F15 YES 0,667 0,07
F05 YES 0,850 0,06
3,2 R
F08 YES 0,702 0,06
F02 YES 0,775 0,22
F03 YES 0,977 0,18
F04 YES 0,978 0,16
3,3 F05 YES 0,632 0,08
F10 NO 0,798 0,47
F11 YES 0,785 0,08
F12 YES 0,910 0,20
Table 4.4 Outcome of Scenario #3b. Superscript R means that the calculations have
been repeated after the reference devices failing either test are discarded.
2
F09 YES 0,914 0,22
F10 NO 0,638 0,68
1,1
F12 YES 0,882 0,29
F13 YES 0,941 0,41
F01 YES 0,020 0,06
F04 YES 0,777 0,29
1,2
F06 NO 0,798 0,29
F07 NO 0,171 0,07
F04 YES 0,826 0,29
1,2 R F06 NO 0,701 0,29
F07 NO 0,078 0,07
F01 YES 0,850 0,12
F02 YES 0,792 0,10
1,3
F03 YES 0,905 0,12
F08 NO 0,285 0,24
2,1 F01 YES 0,709 0,14
F04 YES 0,798 0,15
F05 YES 0,945 0,19
F06 YES 0,887 0,16
F09 YES 0,333 0,14
FEHRL
HERMES Final Report
95
2
F10 YES 0,387 0,19
F14 YES 0,788 0,09
F01 YES 0,833 0,14
F04 YES 0,878 0,15
2,1 R F05 YES 0,851 0,19
F06 YES 0,767 0,16
F14 YES 0,821 0,09
F02 YES 0,601 0,21
F05 YES 0,593 0,04
2,2 F08 NO 0,300 0,41
F12 YES 0,941 0,27
F15 NO 0,158 0,76
F02 YES 0,594 0,21
F08 NO 0,260 0,41
2,2 R
F12 YES 0,952 0,27
F15 NO 0,132 0,76
F01 YES 0,442 0,06
F03 YES 0,943 0,13
F04 YES 0,947 0,20
2,3
F11 NO 0,852 0,64
F13 YES 0,613 0,07
F15 NO 0,843 0,97
F03 YES 0,954 0,13
F04 YES 0,959 0,20
2,3 R F11 YES 0,900 0,64
F13 NO 0,607 0,07
F15 YES 0,894 0,97
F01 YES 0,892 0,19
3,1 F06 YES 0,910 0,28
F13 YES 0,959 0,35
F04 YES 0,891 0,08
F05 YES 0,754 0,13
F08 NO 0,693 0,12
3,2
F13 YES 0,289 0,09
F14 YES 0,697 0,10
F15 YES 0,628 0,09
F04 YES 0,845 0,08
F05 YES 0,837 0,13
3,2 R F08 NO 0,579 0,12
F14 YES 0,590 0,10
F15 YES 0,747 0,09
F02 YES 0,770 0,22
F03 YES 0,970 0,21
F04 YES 0,981 0,17
3,3 F05 YES 0,678 0,11
F10 YES 0,845 0,09
F11 YES 0,828 0,10
F12 YES 0,880 0,20
With a 10% threshold on CVEFI , the number of rejected measurement series is excessively
large: in Trials 2,2 and 2,3, only two devices are kept and the whole of Trial 3,2 is
cancelled. With a 5% level, only the worst case (the one shown on Figure 4.3) is detected.
In either Scenario, the problem with “A” and “B” pointed out in Section 2.4.12 is not
solved, as can be seen in Figures 4.4 to 4.7: “A” is still increasing and “B” decreasing
unacceptably.
FEHRL
HERMES Final Report
96
F1
0,6 F2
F3
0,5 F4
F5
F6
0,4
F7
F8
A 0,3 F9
F10
0,2 F11
F12
0,1 F13
F14
F15
0,0
L-5 L-4 L-3 L-2 L-1 Last
Trial
Figure 4.4 Evolution of “A” in Scenario #3a.
F1
F2
3,0 F3
F4
2,5 F5
F6
2,0 F7
F8
B 1,5 F9
F10
1,0 F11
F12
0,5 F13
F14
0,0 F15
L-5 L-4 L-3 L-2 L-1 Last
Trial
Figure 4.5 Evolution of “B” in Scenario #3a
F1
F2
0,6 F3
F4
0,5 F5
F6
0,4 F7
F8
A 0,3 F9
F10
0,2 F11
F12
0,1 F13
F14
0,0
F15
L-5 L-4 L-3 L-2 L-1 Last
Trial
FEHRL
HERMES Final Report
97
Figure 4.6 Evolution of “A” in Scenario #3b
F1
F2
3,0 F3
F4
2,5 F5
F6
2,0 F7
F8
B 1,5 F9
F10
1,0 F11
F12
0,5 F13
F14
0,0
F15
L-5 L-4 L-3 L-2 L-1 Last
Trial
Figure 4.7 Evolution of “B” in Scenario #3b
4.6 Scenario #4: reversing the regression calculation
As suggested in Section 2.4.12, in Scenario #4, the calibration procedure was reversed by
calculating first the regression of EFI (individual values) versus <<EFI>> (grand average):
EFI = α '+ β '⋅ << EFI >> (4.5)
and then calculating back <<EFI>> against EFI:
α' 1
<< EFI >>= α + β ⋅ EFI with α = − and β = (4.6)
β' β'
Taking this approach, it appeared that the calibration process diverged very quickly, for
example, leaving many negative values for “A”. This was caused by poor correlations in
the calibration regressions. To avoid this kind of problem, the test on REFI 2 had to be
applied to CVEFI (with a 10% threshold) as well. The outcome is presented in Table 4.5
and Figures 4.8 to 4.11.
FEHRL
HERMES Final Report
98
Table 4.5 Outcome of Scenario #4. Superscript R means that the calculations have
been repeated after the reference devices failing either test are discarded.
2
F09 Y 0,914 0,22
F10 N 0,613 0,70
1,1
F12 Y 0,882 0,29
F13 Y 0,941 0,41
F01 Y 0,020 0,06
F04 Y 0,777 0,29
1,2
F06 N 0,798 0,29
F07 N 0,171 0,07
F04 Y 0,826 0,29
1,2 R F06 N 0,701 0,29
F07 N 0,078 0,07
F01 Y 0,850 0,12
F02 Y 0,792 0,10
1,3
F03 Y 0,905 0,12
F08 N 0,285 0,24
F01 Y 0,722 0,16
F04 Y 0,871 0,17
F05 Y 0,941 0,16
2,1 F06 Y 0,835 0,21
F09 Y 0,339 0,14
F10 Y 0,369 0,25
F14 Y 0,795 0,09
F01 Y 0,754 0,16
F04 Y 0,797 0,17
2,1 R
F05 Y 0,949 0,16
F06 Y 0,829 0,21
F02 Y 0,595 0,27
F05 Y 0,869 0,08
2,2 F08 N 0,690 0,33
F12 Y 0,675 0,31
F15 N 0,749 0,82
F02 Y 0,597 0,27
F08 N 0,073 0,33
2,2 R
F12 Y 0,948 0,31
F15 N 0,407 0,82
F01 Y 0,454 0,10
F03 Y 0,941 0,15
F04 Y 0,944 0,41
2,3
F11 N 0,708 0,61
F13 Y 0,609 0,07
F15 N 0,834 0,97
F03 Y 0,960 0,15
F04 Y 0,963 0,41
2,3 R
F11 N 0,861 0,61
F15 N 0,922 0,97
F01 Y 0,891 0,34
3,1 F06 Y 0,924 0,59
F13 Y 0,934 0,27
F04 Y 0,791 0,15
F05 Y 0,886 0,24
F08 N 0,545 0,11
3,2
F13 Y 0,224 0,13
F14 Y 0,561 0,07
F15 Y 0,786 0,19
3,2 R F04 Y 0,675 0,15
F05 Y 0,934 0,24
FEHRL
HERMES Final Report
99
2
F08 N 0,342 0,11
F15 Y 0,920 0,19
F02 Y 0,736 0,45
F03 Y 0,986 0,48
F04 Y 0,981 0,43
3,3 F05 Y 0,986 0,08
F10 Y 0,963 0,37
F11 Y 0,756 0,22
F12 Y 0,920 0,25
F02 Y 0,776 0,45
F03 Y 0,984 0,48
F04 Y 0,977 0,43
3,3 R
F10 Y 0,945 0,37
F11 Y 0,715 0,22
F12 Y 0,912 0,25
F1
F2
0,6
F3
0,5 F4
F5
0,4
F6
0,3 F7
F8
A 0,2
F9
0,1 F10
0,0 F11
L-5 L-4 L-3 L-2 L-1 Last F12

-0,1 F13
-0,2 F14
Trial F15
Figure 4.8 Evolution of “A” in Scenario #4
F1
F2
3,0
F3
F4
2,5
F5
F6
2,0
F7
F8
B 1,5 F9
F10
1,0 F11
F12
0,5 F13
F14
0,0 F15
L-5 L-4 L-3 L-2 L-1 Last
Trial
Figure 4.9 - Evolution of “B” in Scenario #4
FEHRL
HERMES Final Report
100
F1
F2
1,0 F3
0,9 F4
0,8 F5
0,7 F6
F7
EFI (0,5)
0,6
0,5 F8
F9
0,4
F10
0,3
F11
0,2
F12
0,1
F13
0,0
F14
L-5 L-4 L-3 L-2 L-1 Last F15
Trial
Figure 4.10 - Evolution of EFI(0,5) = A + 0,5 B in Scenario #4
F1
F2
0,5 F3
0,5 F4
0,4 F5
0,4 F6
0,3 F7
σ(EFI)
F8
0,3
F9
0,2
F10
0,2
F11
0,1
F12
0,1
F13
0,0
F14
L-5 L-4 L-3 L-2 L-1 Last F15
Trial
Figure 4.11 - Evolution of s EFI in Scenario #4
In this scenario, “A” and “B” no longer exhibit any undesirable general trend. However,
now, they evolve in a rather chaotic way. As the intercept “A” is generally far away from
the cluster of data, a more realistic approach to evaluate the stability of the calibration
parameters would be to consider the evolution that the value of EFI would take at a
medium speed on a surface in the middle of the practical range of friction value, for
example at F30=0,5. Then, EFI(0,5)=A+0,5B. That is what Figure 4.10 shows. It does not
seem that EFI(0,5) has a tendency to stabilize. Regarding the precision of the calibration,
σEFI takes on values ranging from 0,023 to 0,119 with an average value of 0,062 in the
3rd round (last three trials).
4.7 Scenarios #5 to #7: forcing “A” to zero
As the first few attempts to prevent the undesirable trends of “A” and “B” have failed, the
next step suggested in Section 2.4.12 is to force “A” to zero, which makes sense from a
physical point of view since, ideally, EFI should vanish on a zero-friction surface (see
2.4.12). That option has been implemented in Scenario #5. Before following that process,
FEHRL
HERMES Final Report
101
the initial B-values of the reference devices had to be re-calculated from the database6 of
the PIARC International Experiment 1992 [1] with EFI re-defined as:
EFI = B ⋅ F30 (4.7)
Table 4.6 gives the initial values (B0') to be used when A0 = 0.
Table 4.6 Initial B-values (B0') when “A” is forced to zero.

Tyre B0'
Code Name country principle ratio
F01 DWW Trailer DWW, NL PIARC Blank BFC 0,86 0,908

F02 ADHERA LCPC, FR PIARC Blank BFC 1,00 1,089
F03 SCRIM CEDEX, ES AVON Blank SFC 0,34 1,045
F04 SCRIM MET, BE AVON Blank SFC 0,34 1,002
F05 GripTester MET, BE Findley-Irvine Blank BFC 0,15 1,183
F06 ROAR DRI, DK ASTM E1551 Blank BFC 0,20 1,000
F07 ROAR DWW, NL ASTM E1551 Blank BFC 0,86 1,000
F08 Odoliograph BRRC, BE PIARC Blank SFC 0,34 1,000
F09 PFT TRL, GB ASTM E524 Blank BFC 1,00 1,049
F10 OSCAR NRRL, NO ASTM E524 Blank BFC 0,18 1,000
F11 ROAR Mk2 NRRL, NO ASTM E1551 Blank BFC 0,18 1,000
F12 SRT-3 IBDIM, PL Patterned BFC 1,00 1,079
F13 SCRIM TRL, GB AVON Blank SFC 0,34 1,002
F14 Odoliograph MET, BE PIARC Blank SFC 0,34 1,065
F15 IMAG STBA, FR PIARC Blank BFC 0,15 1,000
Figure 4.12 shows the evolution of “B” in Scenario #5.
F1
F2
12 F3
F4
10 F5
F6
8 F7
F8
B 6 F9
F10
4 F11
F12
2 F13
F14
0 F15
L-5 L-4 L-3 L-2 L-1 Last
Trial
Figure 4.12 Evolution of “B” in Scenario #5 (A forced to zero)
The behaviour of device F05 is an obvious problem, introduced by the exponetial model
for friction against speed. The extremely high values of “B” that it reaches in some trials
6
Restricted to the devices in use in Europe.
FEHRL
HERMES Final Report
102
affect the calibration of the other devices. Therefore, it was discarded and the calculations
were repeated, to give Scenario #6. This time, F15 exhibited atypical behaviour (for the
same reason) regarding both the evolution of “B” (Figure 1.13) of σEFI , the residual
standard deviation of EFI (Figure 4.14).
F1
3,0 F2
F3
2,5 F4
F6
2,0 F7
F8
B 1,5 F9
F10
1,0 F11
F12
0,5 F13
F14
0,0 F15
L-5 L-4 L-3 L-2 L-1 Last
Trial
Figure 4.13 Evolution of “B” in Scenario #6 (Device F05 excluded)
0,50
0,45 F1
0,40 F2
F3
0,35
F4
0,30 F6
σ (EFI )
0,25 F7
F8
0,20
F9
0,15 F10
0,10 F11
F12
0,05
F13
0,00 F14
L-4 L-3 L-2 L-1 Last F15
Trial
Figure 4.14 Evolution of σEFI in Scenario #6 (Device F05 excluded)
For the same reason as for F05, F15 was discarded in the next round of calculations,
which became Scenario #7, the results of which are presented in Figures 4.15 and 4.16.
For completeness, all of the calibration graphs have been included in Annex G.
FEHRL
HERMES Final Report
103
F1
3,0 F2
F3
2,5 F4
F6
2,0 F7
F8
B 1,5 F9
F10
1,0
F11
F12
0,5
F13
F14
0,0
L-5 L-4 L-3 L-2 L-1 Last
Trial
Figure 4.15 Evolution of “B” in Scenario #7 (Devices F05 and F15 excluded).
0,50
0,45 F1
F2
0,40 F3
0,35 F4
F6
0,30
σ (EFI )
F7
0,25 F8
F9
0,20
F10
0,15 F11
0,10 F12
F13
0,05 F14
0,00
Trial
Figure 4.16 Evolution of σEFI in Scenario #7 (Devices F05 and F15 excluded).
It can be concluded from Figure 4.15 that forcing “A” to zero solves the problem: “B” no
longer decays. Instead, it shows a clear tendency to stabilise after one or two trials, which
is the expected behaviour. This demonstrates that convening limited subsets of devices is
sufficient to stabilise and maintain the EFI scale.
However, the benefit has a price: σEFI in the last trial in which any given device took part,
which is a measure of the precision of the calibration, ranges from 0,027 to 0,135 with an
average of 0,081 (Figure 4.16). While the lowest figure is close to the repeatability of F
(Section 2.4.6), which would be very satisfactory, the highest figure can hardly be
considered acceptable. Table 4.7 ranks the devices according to their final value of σEFI .
FEHRL
HERMES Final Report
104
Table 4.7 Ranking of the devices according to the last value of s EFI in Scenario #7 (F05
and F15 excluded)

s EFI
F05 GripTester MET, BE BFC 0,15 *

F15 IMAG STBA, FR BFC 0,15 *
F04 SCRIM MET, BE SFC 0,34 0,027
F03 SCRIM CEDEX, ES SFC 0,34 0,040
F13 SCRIM TRL, GB SFC 0,34 0,045
F14 Odoliograph MET, BE SFC 0,34 0,047
F07 ROAR DWW, NL BFC 0,86 0,051
F08 Odoliograph BRRC, BE SFC 0,34 0,051
F01 DWW Trailer DWW, NL BFC 0,86 0,069
F10 OSCAR NRRL, NO BFC 0,18 0,070
F06 ROAR DRI, DK BFC 0,20 0,090
F12 SRT-3 IBDIM, PL BFC 1,00 0,092
F09 PFT TRL, GB BFC 1,00 0,103
F02 ADHERA LCPC, FR BFC 1,00 0,133
F11 ROAR Mk2 NRRL, NO BFC 0,18 0,135
Figure 4.17 shows that, regarding ther ? EFI values, the devices fall into two populations
representing the SFC and BFC measuring principles, regardless of the slip ratio (at least
for BFC – all SFC devices in the trials had the same wheel angle). It could be argued that
there is actually a third population, represented by F05 and F15. These two devices have
a similar behaviour to one another that can be related to their rather low slip ratio, as
shown in Chapter 3. As can be seen in Annex B, their F-values decrease with speed more
rapidly than other devices and, because of a low slip ratio, their maximum slip speed is
some way from the 30 km/h reference level. This generally results in extrapolated F30
values that are much lower than for other devices, which explains the low values of EFI
with respect to <<EFI>> and the high values of “B” and σEFI .
0,20
0,15
σEFI
SFC
0,10
BFC
0,05
0
0% 50% 100%
Slip ratio
Figure 4.17 Values of σEFI versus slip ratio and measuring principle (Scenario #7)
FEHRL
HERMES Final Report
105
4.8 Scenario #8: treating BFC- and SFC-devices separately
In this scenario, the BFC devices and SFC devices, were processed as separate groups,
as Scenario #8a or #8b respectively. The outcomes of the separate treatments are shown
in Figures 4.18 and 4.19 for BFC devices and Figures 4.20 and 4.21 for SFC-devices.
From the σEFI values in the last trial of each device, it can be seen that considering BFC
and SFC devices separately does not bring any improvement in either case. In terms of
repeatability of EFI, the BFC group remains bad with 0,041< σEFI <0,144, whereas the SFC
group remains rather good with 0,024< σEFI <0,054. The latter class includes the two
Odoliographs and the three SCRIMs that participated in the trials. They all have the same
slip ratio of 34% so they might be expected to perform better in terms of σEFI .
3,0
2,5
F1
F2
2,0
F6
F7
B 1,5 F9
F10
1,0 F11
F12
0,5
0,0
L-5 L-4 L-3 L-2 L-1 Last
Trial
Figure 4.18 Evolution of "B” in Scenario #8a (BFC-devices only)
0,50
0,45
0,40 F1
0,35 F2
0,30 F6
σ (EFI)
F7
0,25
F9
0,20 F10
0,15 F11
0,10 F12
0,05
0,00
Trial
Figure 4.19 Evolution of σEFI in Scenario #8a (BFC-devices only)
FEHRL
HERMES Final Report
106
3,0
2,5
2,0 F3
F4
B 1,5 F8
F13
1,0 F14
0,5
0,0
L-5 L-4 L-3 L-2 L-1 Last
Trial
Figure 4.20 Evolution of “B” in Scenario #8b ( SFC-devices only)
0,50
0,45
0,40
0,35
F3
0,30
σ (EFI )
F4
0,25 F8
0,20 F13
F14
0,15
0,10
0,05
0,00
Trial
Figure 4.21 Evolution of σEFI in Scenario #8b (SFC-devices only)
4.9 Scenarios #9 to #11: improved model for F(S)
In Chapter 3 it was shown that F(S) could be better modelled by the following formula:
α
F = F0 ⋅ e − ( S / S0 ) (4.8)
With α > 1, measurement series exhibiting a negative curvature can be curve-fitted more
closely than with the simple exponential model (the so-called PIARC model).
The model has now three undetermined parameters: F0, S0 and α. Consequently, it can
no longer be linearized. It can, however, still be treated in this way with respect to F and
Sα by using the logarithm form:
−α
ln F = ln F0 − S 0 ⋅ Sα (4.9)
FEHRL
HERMES Final Report
107
In order to determine the best curve fit to this formula, the linear regression of lnF versus
Sα was calculated by the least squares method for each value of α from 0,05 to 9,95 in
steps of 0,05. The value of α giving the lowest residual deviation was then selected. The
different values obtained from the different measurement series made by a given device in
a given trial were then averaged and that average was used in the subsequent
processing. Figure 4.22 shows the average values of α obtained per device and per trial.
5,0 F01
4,5 F02
4,0 F03
F04
3,5
F05
3,0
F06
α 2,5 F07
2,0 F08
1,5 F09
F10
1,0
F11
0,5 F12
0,0 F13
1,1 1,2 1,3 2,1 2,2 2,3 3,1 3,2 3,3 F14
Trial F15
Figure 4.22 Values of α per device and per trial
It appears that, in general, α for a given device does not vary much from trial to trial,
except for devices F05 and F15. Its average value per device, shown in Figure 4.23,
seems to be related to the slip ratio: the lower the percent slip, the higher the value of α.
4,5
4,0
-0,60
y = 0,61x
3,5 2
R = 0,50
3,0
2,5
<α >
2,0
1,5
1,0
0,5
0,0
0% 20% 40% 60% 80% 100% 120%
Slip ratio
Figure 4.23 Values of α per device, averaged over the different trials
All of the regression curves using the new model (4.8) with the device-specific α values
are shown in Annex H. In this new model, the the power law and the weighting as
discussed in Section 4.2 continued to be used to relate S0 to MPD.
Reprocessing the data using the new model is Scenario #9. It appeared that device F05
would have to be discarded again. Although the new model provides a better curve fitting
for F(S), as can be seen in Annex H, the extrapolation to the reference speed leads to still
FEHRL
HERMES Final Report
108
lower values of F30 than were obtained with the original model. Figure 4.24 shows a
typical example of the effect of that on the calibration regression.
Trial 2,1
0,8
0,7
0,6
0,5
<<EFI>>
F05
0,4
0,3
0,2
y = 335120x
0,1
0,0
0,0000000 0,0000005 0,0000010 0,0000015 0,0000020
EFI
Figure 4.24 Example of impossible calibration of device F05.
The processing was then repeated without F05 (Scenario #10). As had occurred before,
device F15 exhibited a behaviour resembling the case of F05 as Annex H (Trials 2,2 – 2,3
– 3,2) and Figure 4.25 show.
Trial 3,2
1,2
1,0
0,8
<<EFI>>
0,6 F15
0,4
0,2 y = 1,3774x
0,0
0,000 0,100 0,200 0,300 0,400 0,500 0,600 0,700 0,800
EFI
Figure 4.25 Example of unacceptable calibration result of device F15.
Eventually, a further set of calculations, Scenario #11, was made that excluded both F05
and F15. The results in the form of the evolution of “B” and σEFI , are presented in Figures
4.26 and 4.27 respectively.
FEHRL
HERMES Final Report
109
F1
3,0 F2
F3
2,5 F4
F6
2,0 F7
F8
B 1,5 F9
F10
1,0 F11
F12
0,5 F13
F14
0,0
L-5 L-4 L-3 L-2 L-1 Last
Trial
Figure 4.26 Evolution of “B” in Scenario #11 (F05 and F15 discarded).
0,50
F1
0,45
F2
0,40 F3
0,35 F4
F6
0,30
σ (EFI )
F7
0,25 F8
F9
0,20
F10
0,15 F11
0,10 F12
F13
0,05
F14
0,00
Trial
Figure 4.27 Evolution of σEFI in Scenario #11 (F05 and F15 discarded).
In order to evaluate the possible benefit brought about by the improved model for F(S),
Figures 4.26 and 4.27 (Scenario #11) should be compared with Figures 4.15 and 4.16
(Scenario #7). It can be seen that the results are very similar.
The ultimate values of σEFI for Scenario #11 (given in Table 4.8) range from 0,026 to
0,155 with an average of 0,085, which is slightly worse than with Scenario #7. Therefore,
despite the fact that that the improved model better fits certain measurement series, it
does not seem to improve the precision of EFI.
FEHRL
HERMES Final Report
110
Table 4.8 Final calibration precision in Scenario #11

s EFI
F04 SCRIM MET, BE SFC 0,34 0,079

F03 SCRIM CEDEX, ES SFC 0,34 0,132
F13 SCRIM TRL, GB SFC 0,34 0,039
F14 Odoliograph MET, BE SFC 0,34 0,026
F07 ROAR DWW, NL BFC 0,86 0,100
F08 Odoliograph BRRC, BE SFC 0,34 0,049
F01 DWW Trailer DWW, NL BFC 0,86 0,049
F10 OSCAR NRRL, NO BFC 0,18 0,108
F06 ROAR DRI, DK BFC 0,20 0,066
F12 SRT-3 IBDIM, PL BFC 1,00 0,155
F09 PFT TRL, GB BFC 1,00 0,090
F02 ADHERA LCPC, FR BFC 1,00 0,050
F11 ROAR Mk2 NRRL, NO BFC 0,18 0,050
BFC only 0,097
SFC only 0,043
All 0,085
4.10 Analysis of deviations
Essentially, there are four potential sources of errors that could explain the values of σEFI :
1. Insufficient accuracy in the model for F(S).

2. The repeatability deviations of F.
3. The deviation of the predicted S0 with respect to the actual S0, due to the lack of
accuracy of the model for S0 (MPD).
4. The inaccuracy of the EFI concept itself due to some possible deficiency of the
friction model.
In Section 2.4.6, the repeatability of F was evaluated. In Section 4.9 in this chapter it has
been shown that improving the model for F(S) does not bring any improvement. The
following Sections of the report examine:
• The influence of the errors that occur when a predicted S0 is used (Scenario #12).
• The repeatability of the EFI.
• The residual standard deviation, ? EFI , of the calibration regression when the
deviations due to errors of type 1, 2 and 3 above have been averaged out (Scenario
#14).
FEHRL
HERMES Final Report
111
4.11 Scenario #12: using actual values of S0 instead of

predictions
The purpose of Scenario #12 is to evaluate the influence of the errors that arise from
using a prediction of S0 to determine F30 by using actual values of S0 instead. The
processing was otherwise identical to Scenario #7, i.e. using the original model for F(S)
with (? = 1). Although it is possible to make the calculations here, using the HERMES
data, this approach cannot be considered as a practical solution to the problem. This is
because, in a real situation, measurements are not repeated at different speeds and S0
therefore cannot be determined. In normal circumstances it will only be possible to use a
predicted value of S0 based on a texture measurement.
The results of this analysis are presented in Table 4.9 and Figure 4.28. The range of final
values of σEFI is 0,032 to 0,133 with an average of 0,072, which is not significantly better
than was obtained using the approach in Scenario #7. In some cases, ? EFI increases in
Scenario #12. This can happen because successive calibrations are influenced by the
results of those preceding. So, σEFI in the last trial does not depend only on the
improvement of the extrapolation to obtain F3 0.
Table 4.9 Comparison of the final calibration precision in Scenarios #7 and #12
Friction device Lab, Measurement Slip s EFI
Code Name country principle ratio Sc.#7 Sc.#12

F01 DWW Trailer DWW, NL BFC 0,86 0,069 0,055
F02 ADHERA LCPC, FR BFC 1,00 0,133 0,080
F03 SCRIM CEDEX, ES SFC 0,34 0,040 0,034
F04 SCRIM MET, BE SFC 0,34 0,027 0,032
F06 ROAR DRI, DK BFC 0,20 0,090 0,087
F07 ROAR DWW, NL BFC 0,86 0,051 0,059
F08 Odoliograph BRRC, BE SFC 0,34 0,051 0,055
F09 PFT TRL, GB BFC 1,00 0,103 0,089
F10 OSCAR NRRL, NO BFC 0,18 0,070 0,073
F11 ROAR Mk2 NRRL, NO BFC 0,18 0,135 0,133
F12 SRT-3 IBDIM, PL BFC 1,00 0,092 0,079
F13 SCRIM TRL, GB SFC 0,34 0,045 0,045
F14 Odoliograph MET, BE SFC 0,34 0,047 0,041
BFC only 0,093 0,082
SFC only 0,042 0,041
All 0,081 0,072
FEHRL
HERMES Final Report
112
0,50
F1
0,45
F2
0,40 F3
0,35 F4
σ (EFI )
0,30 F6
0,25 F7
0,20 F8
0,15 F9
0,10 F10
0,05 F11
0,00 F12
L-4 L-3 L-2 L-1 Last F13
F14
Trial
Figure 4.28 Evolution of σEFI in Scenario #12
4.12 Repeatability of EFI
The repeatability of EFI, is represented by the standard deviation of the EFI values
reported by a given device on a given surface from repeated measurements at different
speeds. The overall repeatability standard deviation of EFI essentially contains two
components:
• The repeatability errors in the friction measurements themselves.
• The errors that arise as a result of the inaccuracy of the models that relate friction to
speed, F(S), and the speed constant to texture, S0(MPD) which, in turn, are
reflected in the extrapolated values of F30 used to convert F into EFI.
Table 4.10 provides a comparison of the repeatability of EFI with the repeatabilty of F,
using the process described in Scenario #7. The table gives the average repeatability
standard deviation (σr ) of EFI for each device at each trial it attended, together with the
overall average and the average for the last trial round. These values are compared with
the average repeatability of the original friction measurements, σr(F), both overall and for
the last trial round. The averages incorporate the values of the repeatability of EFI,
calculated from each measurement series for each device and surface, weighted by the
number of measurements retained after the test on RF 2 (Table 4.11). Devices F05 and
F15 are excluded.
Clearly, an increase in σr would be expected when changing from F to EFI and this is
generally reflected in the summary columns of Table 4.10 (those showing the “average”
and “final” values). However, comparing the “final” columns, which represent the values
obtained when all the devices had been calibrated at least once, the increase found when
moving from σr (F) to σr (EFI) is moderate, especially when “all” devices are considered.
This obs ervation suggests that little further improvement can be expected from the
quest for better models.
FEHRL
HERMES Final Report
113
Table 4.10 Repeatability of EFI in Scenario #7 compared to the repeatability of F (Section

2.3.6, Table 2.12).”Final” means the three trials in the third round
Summary values
σr (EFI) at individual trials
Device σr (EFI) σr (F)
1,1 1,2 1,3 2,1 2,2 2,3 3,1 3,2 3,3 Average Final Average Final
F01 0,024 0,030 0,019 0,036 0,025 0,027 0,025 0,016 0,018
F02 0,047 0,142 0,073 0,096 0,073 0,018 0,017
F03 0,034 0,022 0,028 0,029 0,028 0,028 0,023
F04 0,059 0,023 0,028 0,033 0,022 0,036 0,022 0,035 0,020
F06 0,014 0,031 0,022 0,024 0,022 0,027 0,029
F07 0,031 0,031 0,031 0,017 0,017
F08 0,027 0,050 0,036 0,037 0,036 0,026 0,022
F09 0,055 0,055 0,055 0,055 0,029 0,029
F10 0,013 0,045 0,039 0,032 0,039 0,027 0,023
F11 0,015 0,024 0,021 0,024 0,021 0,018
F12 0,041 0,025 0,029 0,033 0,029 0,016 0,016
F13 0,022 0,043 0,030 0,025 0,030 0,025 0,028 0,037
F14 0,028 0,030 0,029 0,030 0,022 0,017
BFC 0,035 0,023 0,037 0,038 0,084 0,026 0,024 - 0,041 0,044 0,040 0,025 0,021
SFC 0,022 0,059 0,030 0,025 0,050 0,031 0,030 0,031 0,025 0,033 0,028 0,025 0,025
All 0,036 0,038 0,035 0,035 0,092 0,030 0,026 0,031 0,040 0,042 0,036 0,025 0,023
Table 4.11 Number of measurements retained after the test on R F2.
n 1,1 1,2 1,3 2,1 2,2 2,3 3,1 3,2 3,3 Total
F01 90 90 81 75 120 456
F02 90 90 90 270
F03 90 75 90 255
F04 90 87 75 57 90 399
F06 66 76 105 247
F07 70 70
F08 90 60 73 223
F09 103 69 172
F10 100 62 57 219
F11 62 89 151
F12 120 90 90 300
F13 120 75 120 69 384
F14 85 86 171
Total 443 316 360 460 240 362 345 285 506 3317
FEHRL
HERMES Final Report
114
4.13 Scenario #13. Averaging out the repeatability deviations of

EFI
In Scenario #13, the repeatability deviations of EFI have been averaged out by correlating
the grand average <<EFI>> to the average values <EFI> over each measurement series
instead of the individual values of EFI. By doing this, not only are the repeatability
deviations of F eliminated but the errors due to the imperfection of the models for F(S) and
S0(MPD) are removed as well. Apart from this, the processing is exactly as in Scenario
#7. Figures 4.29 and 4.30 show the evolution of “B” and σEFI , respectively. The values of
“B” are quite close to those in Scenario #7, as expected.
F1
3,0 F2
F3
2,5 F4
F6
2,0 F7
F8
B 1,5 F9
F10
1,0 F11
F12
0,5 F13
F14
0,0
L-5 L-4 L-3 L-2 L-1 Last
Trial
Figure 4.29 Evolution of “B” in Scenario #13
F1
0,50 F2
0,45 F3
0,40 F4
F6
0,35
F7
0,30
σ (EFI )
F8
0,25 F9
0,20 F10
0,15 F11
F12
0,10
F13
0,05 F14
0,00
Trial
Figure 4.30 Evolution of σEFI in Scenario #13
Table 4.12 provides a comparison between ? EFI in Scenarios #7 and #13. The range of
final values of σEFI is now 0,019 to 0,135 with an average of 0,073. It shows that
repeatability deviations do not contribute very much to σEFI .
FEHRL
HERMES Final Report
115
Table 4.12 Comparison between final calibration precision of EFI (Scenario #13) and final
repeatability of EFI (Scenario #7). The devices are ranked according to σEFI .
Friction device Lab, Measurement Slip s EFI
Code Name country principle ratio Sc. #13 Sc. #7

F04 SCRIM MET, BE SFC 0,34 0,019 0,027
F03 SCRIM CEDEX, ES SFC 0,34 0,031 0,040
F13 SCRIM TRL, GB SFC 0,34 0,036 0,045
F14 Odoliograph MET, BE SFC 0,34 0,036 0,047
F08 Odoliograph BRRC, BE SFC 0,34 0,041 0,051
F07 ROAR DWW, NL BFC 0,86 0,045 0,051
F10 OSCAR NRRL, NO BFC 0,18 0,062 0,070
F01 DWW Trailer DWW, NL BFC 0,86 0,065 0,069
F09 PFT TRL, GB BFC 1,00 0,085 0,103
F06 ROAR DRI, DK BFC 0,20 0,088 0,090
F12 SRT-3 IBDIM, PL BFC 1,00 0,089 0,092
F02 ADHERA LCPC, FR BFC 1,00 0,113 0,133
F11 ROAR Mk2 NRRL, NO BFC 0,18 0,135 0,135
BFC only 0,085 0,097
SFC only 0,033 0,043
All 0,073 0,080
Sets of three graphs showing the calibration lines for the trials in each round are shown in
Figures 4.31 to 4.33 for Rounds 1 to 3 respectively (See also Annex I).
FEHRL
HERMES Final Report
116
Trial 1,1
1,0
0,9
0,8
0,7
F09
0,6
<EFI>
F10
0,5
F12
0,4
F13
0,3
0,2
0,1
0,0
0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8
<<EFI>>
Trial 1,2
0,8
0,7
0,6
0,5 F01
<EFI>
F04
0,4
F06
0,3 F07
0,2
0,1
0,0
0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7
<<EFI>>
Trial 1,3
0,9
0,8
0,7
0,6
F01
<EFI>
0,5 F02
0,4 F03
F08
0,3
0,2
0,1
0,0
0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8
<<EFI>>
Figure 4.31 Comparison of EFI-values delivered by the devices participating in each trial
in Round 1 (Scenario #13). Solid lines represent reference devices taking part
in a Type-1 calibration in that trial. Dotted lines represent non-reference
devices taking part in a Type-2 calibration.
FEHRL
HERMES Final Report
117
Trial 2,1
1,2
1,0
F01
0,8 F04
<EFI>
F06
0,6
F09
F10
0,4
F14
0,2
0,0
0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9 1,0
<<EFI>>
Trial 2,2
1,0
0,9
0,8
0,7
0,6 F02
<EFI>
0,5 F08
0,4 F12
0,3
0,2
0,1
0,0
0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9
<<EFI>>
Trial 2,3
1,0
0,9
0,8
0,7 F01
0,6 F03
<EFI>
0,5 F04
0,4 F11
0,3 F13
0,2
0,1
0,0
0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8
<<EFI>>
Figure 4.32 Comparison of EFI-values delivered by the devices participating in the trials in
Round 2 (Scenario #13). Solid lines represent reference devices taking part in
a Type-1 calibration in that trial. Dotted lines represent non-reference devices
taking part in a Type-2 calibration
FEHRL
HERMES Final Report
118
Trial 3,1
0,9
0,8
0,7
0,6
F01
<EFI>
0,5
F06
0,4
F13
0,3
0,2
0,1
0,0
0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8
<<EFI>>
Trial 3,2
1,2
1,0
0,8 F04
<EFI>
F08
0,6
F13
0,4 F14
0,2
0,0
0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9
<<EFI>>
Trial 3,3
1,0
0,9
0,8
F02
0,7
F03
0,6
<EFI>
F04
0,5
F10
0,4
F11
0,3
F12
0,2
0,1
0,0
0,0 0,1 0,2 0,3 0,4 0,5 0,6 0,7 0,8 0,9
<<EFI>>
Figure 4.33 Comparison of EFI-values delivered by the devices participating in the trials in
Round 3 (Scenario #13). Solid lines represent reference devices taking part in
a Type-1 calibration in that trial. Dotted lines represent non-reference devices
taking part in a Type-2 calibration
FEHRL
HERMES Final Report
119
4.14 Recapitulation
Table 4.13 summarises the various processes used in the analyses described in the
earlier parts of this Chapter. There is a column for each of the fourteen data processing
scenarios described. An “X” in a box indicates which variant was used for each
component of the analysis in that scenario.
Table 4.13 Recapitulation of the different data processing scenarios applied in Section
2.3. and Chapter 4.
Scenario => #0 #1 #2 #3a #3b #4 #5 #6 #7 #8a #8b #9 #10 #11 #12 #13
F = F0 ⋅ e − S / S0 X X X X X X X X X X X X X
F-model
α
F = F0 ⋅ e − ( S / S0 ) X X X
S 0 = 57 + 56 ⋅ MPD X X
S0-model
S 0 = a ⋅ MPD b (1) X X X X X X X X X X X X X
Actual S0-value from F(S) X
EFI = A + B ⋅ F30 X X X X X X
EFI-model
EFI = B ⋅ F30 X X X X X X X X X X
<< EFI >>= α + β ⋅ EFI X X X X X
Calibration EFI = α '+ β '⋅ << EFI >> X

method
<< EFI >>= β ⋅ EFI X X X X X X X X X
<< EFI >>= β ⋅ < EFI > X
F > 0.01 X X X X X X X X X X X X X X X X
S0 > 0 X X X X X X X X X X X X X X X X
σ r ( F ) < 0.04 X X
σ EFI > 0.07 X

Statistical
tests RF 2 > 0.5 X X X X X X X X X X X X
REFI 2 > 0.5 X X X X
CVEFI > 10% 5% 10%

“k-test” (0,5%) X
“h-test” (0,5%) X
F05 F05 BFC F05 F05 F05 F05

Discarded devices F05 F05
F15 F15 (2) F15 F15 F15 F15
SFC
(1) With weighting . (2) F05 and F15 were also discarded here since they are of BFC-type
FEHRL
HERMES Final Report
120
4.15 Discussion
In Figures 4.31 to 4.33, the EFI-values delivered by the different devices in the first round
of trials were compared. Some of the devices (represented by solid lines in the graphs)
were regarded as reference devices from the start of the exercise because they had
participated in the 1992 PIARC International Experiment [1]. Their initial B-values (B0)
were derived from the database of that experiment.
Using those values should have enabled them to deliver equal EFI values. Generally,
however, this was not the case. In Trial 1,1, devices F09, F12 and F13 provided widely
differing EFI values even though they were initially reference devices. In Trial 1,2, devices
F01, F04 and F07 were in the reference group. In this trial, devices F01 and F07 agreed
with each other, the only case where this occurred, while F04 disagreed with them both. In
Trial 1,3, F02 and F03 were close but not equal 7. This demonstrates the necessity of
calibrating EFI.
Figures 4.37 to 4.39 compare the EFI-values delivered by the different devices in the last
round of trials. At this stage of the exercise, having participated in at least one earlier
calibration, all devices were now reference devices and so they should have delivered
similar EFI-values. This appears to have occurred quite well in Trial 3,1, albeit with some
scatter, and again, but not quite so well, in Trial 3.3. However, the improvement was not
so clear in Trial 3,2, in which two pairs of devices agreed well with each other, but not with
the other pair.
Now, it is possible that some devices might change some of their characteristics in the
interval between two rounds 6-months apart, and this could explain the discrepancy
between the two sets of devices in Trial 3,2. This point has already been made in relation
to the direct comparison between devices F04 and F13 (Figure 2.24). Taking that
possibility into account (which, incidentally illustrates why calibrations are required), it
appears that the concept of convening small sets of devices in separate trials and on
various test sites instead of bringing all the devices together actually works. This is further
illustrated by the fact that the B-value of each device more or less stabilises at the end of
the series of trials (Figure 4.15).
Table 4.14 givese the results of the calculations of the repeatability and reproducibility
standard deviations of the calibrated EFI after each trial in Scenario #7, in accordance
with ISO 5725-2 [5]. The overall values were obtained by weighting the variances with
respect to the number of measurements.
The rounded overall values were 0,04 for the repeatability and 0,10 for the reproducibility.
These values should be compared to the overall repeatability of F, which ranges from
0,016 to 0,035 (Table 2.12) and to the reproducibility of F by similar devices, which ranges
from 0,041 to 0,171 (Table 2.13).
7
In that trial, F01 was no longer using B0’ since it had been re-calibrated in trial 1,1.
FEHRL
HERMES Final Report
121
Table 4.14 Repeatability and reproducibility standard deviations of the calibrated EFI after
each trial in Scenario #7, according to ISO 5725-2 [5]
Trial s r (EFI) s R (EFI) n
1,1 0,037 0,112 443

1,2 0,040 0,094 316
1,3 0,038 0,072 360
2,1 0,034 0,101 460
2,2 0,079 0,137 240
2,3 0,030 0,122 362
3,1 0,025 0,082 345
3,2 0,031 0,045 285
3,3 0,039 0,099 506
Overall 0,040 0,099 3317
The increase of repeatability standard deviation in moving from F to EFI is due to the extra
deviation introduced into F30 by using the value S0 “predicted” from MPD instead of the
actual value derived from the exponential regression for F(S). That extra deviation
appears as a speed influence as explained in Section 2.4.13. As was apparent in Table
4.9, which showed the effect of using the predicted S0 on the calibration precision of EFI,
the extra repeatability deviations due to the imperfection of the model for S0(MPD) were
rather moderate. Little could be gained by further improving the model.
In summary, a general conclusion to this Chapter could be that the EFI concept neither
improves nor worsens the reproducibility of friction measurements by different devices
but, at least, it does seem to provide them with a stable common scale.
FEHRL
HERMES Final Report
122
5 Alternative approaches to reference levels
5.1 Introduction
It was recognised at the outset of the HERMES project that there could be problems in using the
EFI concept and associated the calibration procedure to provide a reliable common scale for the
comparison of measurements of skid resistance made by different devices in Europe.
Anticipating this, the project included two supplementary tasks that had the objective of
investigating alternative solutions to the problem of providing references against which devices
could be calibrated and these are discussed in this Chapter.
The first of these, described in Section 5.2, was to develop proposals for a set of specifications for
a single reference device that, conceptually at least, could be used either as a means of calibrating
existing devices or, in the longer term become the standard measurement method.
It was also recognised that any real device, even if it is based on a standard specification, will need
to be calibrated or checked somehow and to do this would require stable and reproducible
reference surfaces that could provide and maintain known levels of friction. The second approach,
covered in Section 5.3, discusses the topic of reference surfaces and presents proposals for a set
of requirements for such surfaces.
5.2 A single reference device
5.2.1 Purpose
As outlined in the introduction to this Chapter, one objective of the HERMES project was
to develop a specification for a single friction and texture measurement device. This
device would, at some time in the future, take over the role of the calibration process in
the draft standard prEN [3] by providing a direct reference, rather than the current
proposal in which the “grand average” of all devices provides a floating reference and
groups of different combinations of devices meet at different places at regular intervals for
calibration purposes.
In the long term, a fleet of devices of this type could possibly replace the many different
devices now in used in Europe.
5.2.2 Approach
In approaching its objective, the HERMES working group proposed that the “reference
device” should be a Friction Testing Device to which any device currently in use could be
compared. It was recognised that, for the forseeable future, the friction measurement
would have to rely on the basic principle of a test tyre running on a wetted road surface
and the likely requirements were developed on that basis.
It was also considered important that the design should not necessarily be a perpetuation
of present practice. Although it would be wise to draw on current experience, the
FEHRL
HERMES Final Report
123
reference device should, in effect, represent a “new generation” of devices and that
therefore the proposed specifications would not correspond to any existing device.
A lesson learnt from current experience in some countries that have operated large fleets
of similar devices over many years was that careful and consistent control of equipment
was essential to maintain consistent results between ostensibly similar machines.
(Evidence of this kind of problem was to emerge from the HERMES exercises, as noted in
Section 2.4.7 of this report.) Therefore, it was also considered particularly important that
the specifications for the new device should be designed to provide the best level of
reproducibility that would be practicable.
The HERMES team, therefore took into account these various aspects when considering
the requirements for a reference device. This included a questionnaire survey of FEHRL
member laboratories to canvass their views. Having established what the basic principles
should be, a draft specification was incorporated into a draft standard for making friction
measurements using the reference device, prepared in a format suitable for proposal to
CEN, which is included as Appendix K. The draft proposal for a test method would, of
course, be subject to revision and review through the normal channels before formal
adoption.
The following sections discuss the background and underlying ideas behind the proposed
specification.
5.2.3 Historic and technical background
Each European country uses at least one device to measure friction and texture.
Historically, some countries have developed their own devices; others have used devices
developed elsewhere. The earliest developments were made before the Second World
War, continuing in the 1950s. In the UK, for example, devices such as the Pendulum
tester and early side-force skid cars and brake-force devices were developed in this
period, but most devices in current use were developed in the late 1960s and 1970s.
An important driving force behind the development of friction measuring devices was the
need to explain why it appeared that, on certain sections of the road network, there would
be an accumulation of accidents under certain conditions. At the same time,
developments were made in order to satisfy the needs of the different countries for
research and, in due course, to provide for acceptance tests on new works, and network
surveys of various kinds. In parallel, there was an increasing desire to be able to measure
friction on airfield runways.
Many new types of equipment were devised and developed in different European
countries, for example (in no particular order): Adhera, µMeter, Odoliograph, SCRIM,
Skidometer, Stradograph, Stuttgarterreibungsmesser (SRM), and others in their turn.
Some of these are still in use while others have been replaced by more recent
developments. At the start of the HERMES project it was estimated that there were
around 18 different devices in use in Europe alone.
Over the last ten years or so it seems to have become increasingly important for the
European national Road Authorities to be able to know the level of the surface
characteristics of their own networks. This knowledge is an important aid to the optimising
of budgets used to maintain a good service level for road users. Skidding resistance and
FEHRL
HERMES Final Report
124
texture also have an important role in relation to safety, especially when road surfaces are
wet.
Apart from the UK, which had developed its side-force motorcycle combination in the
1920s, early knowledge of the skid resistance characteristics of a road surface in most
European countries was very often obtained by static devices or methods. What became
known as micro texture was characterised with a pendulum measurement and macro
texture was characterised by a volumetric measurement (the “Sand Patch” technique).
The introduction of dynamic methods for skid resistance measurement enabled the
influence of another important parameter, the test speed, to be explored. This had first
been observed in the 1930s by the UK team working with their motorcycle equipment.
This led on to the understanding of the relatively greater influence of micro texture at low
test speeds and the important influence of macro texture as test speed increased.
In the last past 20 years laser technology has been used to evaluate the macro texture.
Although the dynamic measurements of this parameter can now be made at normal traffic
speeds (on a dry surface), this is still not possible for micro texture. Lasers and digital
imaging technology can be used to assess micro texture but currently this is only possible
with with static measurements on a laboratory scale and the translation of the technology
to a vehicle moving a traffic speeds for routine use will be some years in the future. At
present, therefore, there is still a need for a direct friction measurement to evaluate the
influence of micro texture.
Many experts consider that, today, the best way to evaluate the likely friction
characteristics of a road surfacing is to associate a skid resistance measurement made at
a very low speed (allowing micro texture evaluation) with a laser-based profile
measurement (allowing macro texture evaluation). However, direct measurements at low
speeds may be difficult to obtain on a trafficked road.
It would appear, therefore, that the best practical approach to the problem would be to use
a system that can associate a laser macro texture measurement with a friction
measurement made at a low slip speed for the test wheel to assess micro texture. The
advantage of this approach is that, although the test vehicle may be running at a normal
traffic speed (60 km/h or 80 km/h), the test wheel slip speed can be very much lower (for
example 30 km/h). These lower slip speeds can be obtained with a test wheel set at an
angle (in which case slip speed is equal to the test speed multiplied by the sine of the
cornering angle) or with a slip ratio imposed by a mechanical or hydraulic system while
the test is being made.
In practice, all existing devices for dynamic skid resistance measurement operate on one
of these principles, which have been referred to elsewhere in this report as the SFC
(sideway-force coefficient, ie angled wheel) or BFC (brake-force coefficient, ie in-line
braking of some kind) principles. The BFC principle encompasses the full range of slip
ratio, up to the locked-wheel condition (100% slip ratio), in which the slip speed is the
same as the test speed.
5.2.4 European needs
All of the available techniques are used across Europe, with different countries having
adopted different methods to meet their own particular requirements. In different
countries, the measurements may be made for various purposes:
FEHRL
HERMES Final Report
125
• On airfield surfaces in real conditions, to give the information about runway condition
to the pilots before landing.
• On new works to control the quality level of the work.
• In network-scale surveys of skid resistance characteristics of in-service roads to assist
in identifying areas requiring treatment and to prioritise and optimise maintenance
work.
• For accident investigation.
• For research (for example, developing new friction models, assessing the behaviour of
new types of surfacings, testing the influence of different parameters on skid
resistance).
Although initally the principal purpose of the “reference device” would be to calibrate
existing devices, there was also the idea that the device might become a Europe-wide
standard device in the longer term. With this in mind, it was decided that, before drawing
up a specification, it would help if there was a proper understanding of the needs within
Europe and to see whether it was possible to conduct all these tests with a single device.
To this end, a questionnaire (Annex K) was sent to all FEHRL members. The
questionnaire identified the main potential areas of use for the device (excluding accident
investigation, which in this context was regarded as a special case of the research
category) and asked respondents to indicate which principles and key parameter ranges
they thought would best meet their needs. In the latter case, the respondents were asked
to “agree” or “disagree” with the proposal that the device should use the feature
concerned. A question was included as to whether simultaneous measurements of macro-
or megatexture would be required.
Twelve replies were received and syntheses of the results of this enquiry are shown in
Tables 5.1 & 5.2. Table 5.1 summarises the majority view regarding the key operating
principles broken down by the various types of use and Table 5.2 gives the percentage of
responses regarding detailed features. Figure 5.1 illustrates graphically the responses to
the 21 questions covered by Table 5.2 regarding network surveys.
Not surprisingly, the responses reflected to some extent the experience of the particular
organisation replying, the devices already used in their country and purposes for which
they were currently used. It was also apparent that the main common needs related to
conducting acceptance tests of new works and for routine network surveys.
Table 5.1 Main results of the responses to the reference device questionnaire.
Preferred option for each proposed use of device

Operational Feature
Airfields New works Research Network survey
Type of measurement Continuous Continuous ? Continuous

BFC fixed ratio
Principle BFC fixed ratio BFC fixed ratio ?
SFC
Test speed Up to 140 km/h 50 – 80 km/h All speeds 50 – 80 km/h
Test tyre Smooth Smooth Smooth Smooth
Water depth 0,5 mm 0,5 mm 0,5 mm to 3 mm 0,5 mm
Operating range Up to 5 km 10 – 100 km 10 – 100 km Up to 100 km
FEHRL
HERMES Final Report
126
Load on test wheel 200 daN 200 daN Variable 200 daN
The more loaded The more loaded
Wheel path Variable Variable
wheel path wheel path
Macro texture Yes Yes Yes Yes
Mega texture Yes Yes Yes Yes
FEHRL
HERMES Final Report
127
Table 5.2 Main results of the enquiry
Question asked Agree: Disagree:

1) Continuous type of measurement 75% 25%
2) SFC principle 33% 0%
3) BFC slip ratio 58% 0%
4) BFC fixed ratio > 80% 25% 0%
5) BFC variable ratio 17% 0%
6) BFC locked wheel 17% 0%
7) Blank tyre 67% 8%
8) PIARC tyre 58% 0%
9) Test speed low 30 – 50 km/h 0% 0%
10) Test speed medium 50 – 80 km/h 83% 0%
11) Test speed high > 80 km/h 33% 0%
12) water depth = 0,5 mm 100% 0%
13) Test wheel load = 200 daN 58% 42%
14) Wheel path - the most trafficked 92% 8%
15) Operating range < 10 km/h 0% 0%
16) Operating range 10 – 100 km/h 42% 0%
17) Operating range > 100 km/h 58% 0%
18) No texture measurements 17% 0%
19) Macro texture 33% 0%
20) Mega texture 0% 0%
21) Both Macro and Mega texture 50% 0%
Figure 5.1 Results of the enquiry for surveys reference device.
FEHRL
HERMES Final Report
128
5.2.5 Proposed specification requirements
Initially, there might be only one device built but eventually more than one may be
needed, either to increase flexibility for calibration of existing devices in different countries
or to replace those devices in the longer term. Experience over many years has shown
that as soon as more than one device is made, even following an identical specification
and detailed design, differences can occur between them. Therefore, the specifications
for the reference device have been chosen bearing in mind the need to keep the precision
of the measurements (i.e., the repeatablity and reproducibilty) as small as possible. As
well as the issue of precision, the specification requirements reflect the initial decision by
the HERMES group not to replicate existing devices.
A draft formal specification, in a format suitable for proposal to CEN has been prepared
and is included as Annex L. In the following paragraphs, the main aspects of the proposed
specification and the reasoning behind them are discussed briefly.
(1) Continuous measurements

It should be borne in mind that the primary purpose of the reference device is to provide a
means of calibrating other devices, with the secondary potential to become a routine tool.
In that regard, it would need to be capable of continuous surveys. In either case, the
device should be able to carry out repeated tests on long sections of road without serious
short-term deterioration of the test tyre.
(2) Measurement principle: Brake Force Coefficient (BFC) using controlled slip.
Because this was to be a reference device, in order to minimise concerns over possible
adverse effects such as transverse deformation of the tyre during a test and the uneven
tyre wear that can occur using an angled wheel, it was decided that in-line braking should
be used. To meet the continuous measurement requirement, a locked-wheel system
would not be suitable and therefore the basic operating principle of the device should use
in-line controlled slip to give BFC.
(3) Normal operating speeds of 40, 60 and 80 km/h

Three normal operating speeds are proposed: 40, 60 and 80 km/h. This range of speeds
will assist safe operation on most types of road, for example, 40 km/h for urban roads, 60
km/h on main roads and 80 km/h on motorways. This also provides a range of speeds for
comparative tests with other devices for calibration purposes. Clearly, for research use,
other speeds may also need to be used.
(4) Use of a variable slip ratio to provide a constant, low, slip speed
For calibrating other devices with EFI the device should, ideally, record its measurements
directly at the reference slip speed of 30 km/h. However, since the device will need to
operate at different speeds in practice, this imposes a special requirement on its
operation.
In order to achieve a constant slip speed, the actual speed of the test wheel will need to
be measured and compared with the vehicle speed so that the slip ratio can be varied
depending upon the vehicle speed:
• A test speed of 40 km/h requires a slip ratio of 75%.
• A test speed of 60 km/h requires a slip ratio of 50%.
• A test speed of 80 km/h requires a slip ratio of 37.5%.
FEHRL
HERMES Final Report
129
This proposed approach to using a variable slip ratio to maintain a constant slip speed, so
far as the Authors are aware, is new and has not been applied on any friction
measurement devices in current use. Some tests have been made on three of the
surfaces used in the HERMES trials in France (FR1, FR2 and FR3) using a device that
was able to measure the all points on the friction/slip curve (BFC = f(slip ratio)) at different
test speeds. These suggest that by using a different slip ratio in order to obtain the same
slip speed for different test speeds leads to a reduction in the variations in the measured
braking force coefficient as speed increases (Figure 5.2).
The BFC friction measurements obtained on the same surface using each of the three
proposed test speeds will not necessarily be equivalent, even though the slip speed
should be the same. However, the apparatus, after calibration and calculation of its “A”
and “B” values, will be able to calculate an EFI value for each of these test speeds and
these EFIs should be equivalent. The slip speed of 30 km/h has been chosen so that it
matches the reference speed for the EFI and therefore can be expected to minimise the
repeatability of the EFI value.
Figure 5.2 Influence of slip ratio on the BFC obtained at different test speeds
(5) Static Vertical Load of 3kN

The majority of answers to this question in the enquiry favoured a 2kg (200daN) vertical
load on the test wheel but it was clear that this had been influenced by the fact that this is
the standard load for a SCRIM. For the reference device, a value of 3kN (300daN) was
chosen, not just to be different from existing devices, but because it was more
representative of the average load of a normal quarter passenger car.
(6) Measured dynamic vertical load

Even with critical damping on the test wheel suspension, roughness or unevenness of the
road can induce variations in the vertical load applied to the test wheel. Additionally, the
response of the vehicle suspension relative to the test wheel (when entering, leaving or
negotiating a curve, for example) may also have an effect on the dynamic load. For this
FEHRL
HERMES Final Report
130
reason, the reference device should continuously measure the vertical load and use this
value when converting the frictional force to BFC.
(7) “Theoretical” water film thickness of 0.5mm, independent of speed.

As with all skid resistance measurement devices, the reference device will make
measurements on a wet road and it should carry its own water supply to wet the road just
in front of the test wheel.
The theoretical water film thickness can be defined as the water depth obtained on a
perfectly flat, smooth and waterproof surface. The actual water film thickness will depend
on the surface to which it is applied and will be affected by the texture and porosity of the
surface as well as the way in which it is applied. Other practical issues such as air
currents (both from side winds and the vehicle slipstream) may also influence the actual
application of water.
The constant theoretical thickness will be achieved by regulating the flow rate to match
the area over which the water is applied and the test speed. Careful design of the
application nozzle will also be required to maintain an even distribution of water across the
width of the wetted area.
The chosen value of 0,5 mm is already widely used by current European devices. It is also
is a value that should give a reasonable operating range for the reference device.
(8) Test tyre

It was obvious that test tyre used by the reference device would also have to be
standardised and that therefore a special reference tyre would be needed. It was decided
that the device should use a tyre that was similar to a normal passenger car tyre under
worst conditions, i.e., with a smooth tread (smooth) so that the device would test the road
surface characteristic and not the wear level of the tyre.
Proposals for such a tyre were already being developed through PIARC Committee C1 in
parallel with this work, so it was decided that the HERMES reference device should use
the tyre defined in that proposal [19] which, in addition to the smooth tread, covers other
characterisics such as compound, size, inflation pressure, hardness/resilience and
storage conditions.
(9) Other parameters to be measured or controlled

As well as the matters discussed above, a number of other parameters will need to be
either measured or controlled in some way.
Clearly, the horizontal force must be measured during the test, as must the distance
travelled. The vehicle speed should be measured and recorded (the speed is also needed
in order to determine and control the slip ratio).
In order to ensure that the measurements at different speeds reflect any the variations
along the road, the sampling interval (i.e. the period over which instantaneous values of
the forces are aggregated in order to provide a sample) should be based on distance
travelled, not on elapsed time. It was decided that a sample value should be obtained at
least every 100 mm (ie a sampling frequency of up to 220 Hz at 80km/h).
Inevitably there will be significant “noise” in the data as a result of vibrations in the
dynamic system and local irregularities in the surface being tested. To damp out these
FEHRL
HERMES Final Report
131
effects, the horizontal and vertical force samples will be averaged over 10 m before
making the calculation of the BFC.
5.3 Reference surfaces
The objective of this task within the HERMES project was to evaluate the feasibility of
designing stable reference surfaces for calibrating friction-testing devices. It was
recognised that it was unlikely that a full specification could be produced at this stage.
Rather, the task would review key aspects of the topic and make proposals that could be
developed in the future.
The work carried out by the project partners in this process included:
• A literature review on the general topic.
• Contacts with operators of test tracks in the motor industry and with other contacts
in the field in different countries.
• General discussion and pooling of expertise within the core group.
5.3.1 The purpose of reference surfaces
The reason that a friction index such as EFI is being considered is that there is no
absolute measure of skid resistance that all devices will record. Differences in
measurement method and operational parameters all influence the values recorded. The
characteristics of most conventional road surfaces also vary widely and change with time,
ambient conditions and the influence of traffic.
Section 5.2 discussed options for a reference device against which others could be
compared or which, in the long term, could become universally used. However, such a
device must itself be calibrated and in an ideal world it would also be checked against
known levels of friction.
Therefore, even though a harmonising index may be necessary to enable measurements

from different devices to be compared, it would be better to be able to calibrate devices
against standardised surfaces rather than against combinations of each other.
The purpose of a reference surface, therefore, is to provide a stable, known, level of skid
resistance that can be used for calibration purposes.
5.3.2 Requirements for reference surfaces
Ideally, a reference surface should have the following general characteristics:

• It should have a known, possibly predictable, level of skid resistance.
• The skid resistance should be stable over time (i.e., it does not change with use).
• The surface should have reproducible characteristics (so that more than one can be
made or a replacement can be produced).
If reference surfaces are to be used for calibration purposes, then it must be possible to
test any device over its practical range of operation. Therefore, several reference levels
are required. Also, for as long as different devices are to be used together with a
FEHRL
HERMES Final Report
132
harmonised scale, the practical range of operation will have to include a range of speeds
as well as of friction levels.
The two main surface characteristics that contribute to skid resistance are micro texture
and texture depth (macro texture), governing the underlying friction level and the change
in skid resistance with speed. Therefore, any set of reference surfaces should include
combinations of these two parameters.
Clearly, it would not be realistic to attempt to produce examples of all possible

combinations. Neither is it necessary, for calibration purposes, for surfaces to be
specifically representative of any particular type of road surfacing.
It is considered that a practical number of reference surfaces, therefore, would be four.

The different combinations of texture that could be encountered should be covered, in
broad terms, as suggested in Table 5.3.
Table 5.3 Proposed combinations of texture parameters
Micro texture Macro texture
Surface HH High High

Surface HL High Low
Surface LH Low High
Surface LL Low Low
At this stage, it is not clear how these parameters should be specified, but before any
device calibration testing were undertaken on such surfaces, the skid resistance and
texture would need to be validated in some way.
Ultimately, the reference device could be used but, until its development, the texture
depth, at least, could be measured using a standard procedure: either the volumetric
texture depth test (MTD) using glass beads or the mean profile depth (MPD). Friction
levels would probably need to be checked in the first instance using existing devices
representing the three main principles (in-line low slip ratio, side force intermediate slip
ratio and in-line locked-wheel or high slip ratio).
However, it is likely that the reference surfaces would need to be made to quite close
tolerances in order to ensure that the surfaces and the skid resistance values obtained are
reproducible. Current test methods may not be sufficient for this purpose, and this is a
point to bear in mind when considering any future specification.
5.3.3 Past experience and existing practice
5.3.3.1 Some historic background

Dynamic measurements of road skid resistance have been made in some countries for
decades. In the UK, for example, the first side force device was developed in the late
1920s, and locked-wheel devices were used from the 1950s. Accurate measurement of
skid resistance on roads became of increasing importance in the 1970s and since that
time there have been many attempts to establish appropriate reference standards. While
it has always been possible to carry out static calibration checks on force transducers,
FEHRL
HERMES Final Report
133
there has always been a desire to establish surfaces on which the whole friction
measurement system could be checked in a dynamic test.
In a laboratory exercise in 1974, Britton et al [13] investigated the criteria needed for the
design of primary standard reference pavement surfaces. Model surfaces were created
using particles of known and easily-controlled geometry on a flat substrate, such as
spheres set in epoxy resin. Adhering artificial or natural fines controlled the micro texture.
A wide range of macro texture was tested and several materials were used. The different
particles had the same shape factor but represented different chemical structures. The
friction measurements were made with a British Portable Skid Resistance Tester (the
“pendulum tester”) and the friction values reported as the BPN (British Pendulum
Number). Over 600 samples were made for the experiment.
No evident difference was observed between the samples made of different materials and
of the same particles size, within the limit of the sensitivity of the experiments, but the
effects of macro texture, size and shape were found to be more significant. Their work
indicated that pavement tyre/friction was influenced by both the macro texture (size of
aggregate, spacing and shape) and the micro texture (size of the fines, spacing and
shape), which, of course, had already been observed in practical work on roads. The use
of synthetic aggregates was investigated and the more promising type appeared to be
synthetic burnt clay aggregates.
Viewed from the present time, however, a limitation of this exercise was the use of the
pendulum test, which is not able to discriminate reliably between the relative effects of
micro texture and macro texture. (The test was originally designed to indicate the level of
friction when a patterned tyre (of the 1960s) skids at 50km/h on a medium-textured road
surface).
In 1983, Dunlop Limited (UK) investigated the creation of reference surfaces and
proposed a standard reference surface to the relevant International Organisation for
Standardisation (ISO) committee [14]. This proposal involved replica surfacings to which
quartz sand was applied to simulate micro texture. The polished stone value (PSV) was
measured for the different replicas, which were found to behave in a similar way to many
granite natural aggregates. The replicas reproduced micro texture to a high degree of
accuracy but, in use, the micro texture was removed rapidly by the tyre in a similar way to
that expected from traffic action had a natural aggregate been used.
Therefore, this could not be considered as a standard reference surface specification but
was a starting point that was taken into account in a further review in 1986, when ISO
published a technical report detailing the process for creating a standardised test surface
for high friction tests [15]. The work carried out to investigate this type of surface indicated
that the best results were achieved with a surface dressing of fine (passing1.2mm
retained on 0.6mm sieves) silica sand (natural Leighton Buzzard sand) spread, without
rolling, on to a bitumen-expanded epoxy binder. With this surface, the high friction
depended almost entirely on the micro texture produced by this aggregate. Silica sand
was selected because it represented the most wear-resistant material known.
In the mid-1970s, three field test centres were set up in the USA under the auspices of the
Federal Highway Administration (FHWA) in order to improve and standardise the
measurement of skid resistance [16]. At these centres, which were in separated
geographical locations, various “primary reference surfaces” (PRS) were constructed. The
surfaces were replicated in each location using the same contractor and similar selected
naturally occurring materials, including silica sand and river gravel, all in an epoxy seal
FEHRL
HERMES Final Report
134
coat. Initially, each centre had five primary reference surfaces 4.6m wide by 158m long.
Friction measuring devices (complying with the ASTM standard) from the various state
authorities were correlated individually against the PRS. A standard vehicle based at each
centre was used to provide a reference skid measurement system and provide a
correction to take account of variations in the PRS over time. Although three centres had
been set up initially, it soon became evident that only two were needed to service the
population of skid testers and one station was closed after only one year of operation.
Although five primary reference surfaces were construc ted at each site, there were
difficulties in achieving the required target levels of skid resistance. Initially, all the
surfaces had higher levels than anticipated. As a result the roughest surface, which was
abrading the test tyres of the candidate equipment, was abandoned and further primary
surfaces were constructed later to provide low skid resistance. In addition, some of the
other primary surfaces at one of the centres were affected by surface distress probably
brought about by defects in the original binder and by construction joints propagating from
the underlying base material.
Even though the primary reference surfaces were only trafficked during the testing
process, it was found that all exhibited significant variations in measured friction during the
year. It was also found that there could be significant variation over time: on one surface,
the Skid Number (the value recorded by ASTM devices) reduced from 63 to 46 over the
nine year period of the operation of one of the centres [17].
Thus, although so-called durable reference surfaces were made, the experience shows
the difficulty in defining and achieving skid resistance using natural materials and in
making materials that maintain a consistent value over time. In this situation a reference
device (but of the same type as the devices being calibrated) had to be used as well in
order to provide a correction to take account of the variations in the surfaces, but there
was no real standard against which that device could be compared.
5.3.3.2 Comparisons of devices on airfields

For some years, regular comparisons of friction devices (mostly for airfields) have been
made in the USA at the NASA airfield test site at Wallops Island [18]. Many different
surfaces have been laid on runways and taxiways to provide a range of friction and
texture levels and studied on regular basis since 1993. Although some 35 different
surfaces or variations have been included or added over the years, the eight surfaces that
have been included throughout and therefore on which most attention has been focussed
are described as:
• Concrete surface.
• Concrete surface with 1x1/4x1/4 inch grooves and canvas belt finishing.
• Surface with 1x1/4x1/4 inch grooves and burlap drag finishing.
• Aggregate asphalt.
• Aggregate asphalt with 2x1/4x1/4 inch groove.
• Aggregate asphalt with 1x1/4x1/4 inch groove.
• Aluminium plate.
• Driveway seal coated asphalt.
Measurements have included sand patch, MPD, BPN, and outflow meter for texture.
Various friction devices have made measurements, including some from Europe. A
review of data provided to the HERMES team by the organisers of the trial showed that
considerable variability has been observed both between devices (as might be expected)
FEHRL
HERMES Final Report
135
and over time. Each friction device made measurements at a range of test speeds, with
the expected decrease in measured friction with increased speed.
However, it is the stability of the friction and texture measurements over time, which is the
major factor of interest here. Some variation in texture depth was observed (the cement
concrete surfaces appear to have been the most consistent). Measurements of friction at
one test speed were compared over time for individual devices on those sections for
which most data were available for the period. Considerable variation could be observed
from year to year, the extent of which varied depending on the device making the
measurements. Although this exercise has provided many useful comparisons, the
analysis is still dependent upon using the average of all devices as a basis for them.
Because of the instability observed over time, it is considered that none of these surfaces
has the potential to become a long-term reference surface, although some of the ideas
could probably be developed.
Attempts to harmonise airfield friction devices have been made in recent years in Norway
on a calibration test track at Oslo Airport. These trials utilised a purpose built Calibration
Test Track. The test surfaces were of different asphalt mixes but here, as at Wallops, the
average of devices was used as a reference level for harmonisation in a similar way to
that proposed by the International Friction Index. Comparisons from trials from 1998 to
2000 using tests with examples of GripTesters and the Swedish BV11 devices showed
variation in friction level from year to year.
5.3.3.3 Test tracks operated by road research organisations

FEHRL members with research test tracks, such as LCPC and TRL have built test
surfaces over the years, some of which were used in the HERMES calibration meetings.
They, too, have found that it is possible to obtain different friction levels using different
combinations of materials. In fact, surface GB15 was prepared, partly for use in the
HERMES work, with the specific intention of providing an intermediate level of skid
resistance. It was based upon small particles (nominally 3mm) of gravel (with smooth
surfaces and hence no micro texture) fixed to the underlying asphalt surface with epoxy
resin (Figure 5.3).
GB15 – SFC 50 ~ 0.45 – 0.55
Figure 5.3 The "pea gravel" surface GB15
FEHRL
HERMES Final Report
136
While this technique successfully produced an intermediate friction level, it was not
possible to predict what that would be or to guarantee that it would remain stable over
time.
These various test track surfaces are all used for regular comparison of devices or for
correlation of devices of the same type, such as the annual SCRIM correlation trials held
at TRL each April. However, the correlation process relies on the comparative
measurements of the devices rather than a stable reference level from the surfaces
themselves.
5.3.3.4 The Motor Industry

The motor industry in different parts of the world uses proving grounds to evaluate the
performance of tyres and braking systems. However, although they may have surfaces
designed to give different friction levels and some materials are commonly used (such as
polished basalt tiles or ceramic to provide low friction), they do not have standardised
friction characteristics. Rather, they provide a means for comparative testing and friction is
deduced from the braking performance of the vehicle or tyre.
In Spain, for example, IDIADA - Instituto De Investigación Aplicada Del Automóvil (Applied
Automotive Research Institute) built a test pavement on their research track. The
pavement friction was tested at the end of the construction work, using the British
Pendulum tester. However, no dynamic skid resistance measurements with standardised
devices have been carried out. Instead, friction values are estimated from the braking
distance of the different commercial vehicles visiting the site. This process of calculation
was considered sufficient to characterise the surfaces since the vehicles had already been
officially approved by some standard.
In this context, the friction can be even more variable than with standard friction test
devices given the range of systems, tyre compounds and tread patterns likely to be used.
Some organisations carry out regular friction tests on their test surfaces using standard
methods (TRC, in the US, for example monitors its surfaces every two weeks using the
ASTM Skid trailer (similar to device F09 in the HERMES project) but because the sites are
out-doors, the surfaces are still subject to the variations that are associated with changing
seasonal conditions.
It is clear that the problem of specifying and producing surfaces with predictable
characteristics and stable friction levels that would be suitable for reference calibration
purposes has yet to be solved. In the remainder of this Section the various issues that
need to be addressed are considered and proposals for further work are made.
5.3.4 Possible solutions
In considering possible solutions to the issue of reference surfaces in the context of

calibrating devices for comparison with a European skid resistance index, several aspects
must be considered including:
• Where the surfaces are to be based.
• Where and how they are to be used.
• How they are to be made.
FEHRL
HERMES Final Report
137
These considerations, in turn, lead to further questions regarding the detail of the way in
which calibration reference surfaces might be used in Europe. This Section discusses the
general issues and Section 5.2.5 summarises the ideas in the form of some initial
suggestions for an outline specification.
5.3.4.1 Where should the reference surfaces be based?

Having established that the reference surfaces would need to be specially laid or
manufactured for this purpose, it follows that a dedicated site would be required. That
would need certain specific features, discussed later, but an important issue is where it
should be.
In an ideal world there would be one site, with one set of reference surfaces, reasonably
centrally located, which all devices that were to be included in the skid resistance index
would have to visit for regular calibration checks. This would avoid problems of
reproducibility between replicate surfaces. However, Europe covers a very large area, so
it would be unrealistic to consider just one location. Devices from as far apart as Norway,
and Spain took part in the HERMES exercises but while this was reasonably practical (as
discussed in Section 2.1.3), some devices did place constraints upon where they would
be prepared to go, given the travelling distances involved.
In principle, any country that can identify a suitable site and organisation to take
responsibility could build and maintain a set of reference surfaces. However this would be
a significant investment and might not be practical for some European countries. It would
be better, perhaps, to establish a limited number of separate test locations so that there is
one within a reasonable distance of most countries. It is recommended that at least three
sites should be established in Europe: one in the north and west, one in a central area
and one in the south or eastern region.
Conceptually, each site could have its own set of reference surfaces. Alternatively, a
common set of surfaces could be built with a modular structure, stored in one location and
then transported to a particular test site (designed to accommodate them) when required
for a calibration exercise.
5.3.4.2 What should a test site be like?

The HERMES project partners have carefully considered what would be required from a
test site in order to provide an effective location for carrying out calibrations using
reference surfaces.
Clearly, any test surfacing must be built on a structure or foundation that is capable of
carrying repeated passes of the weight of the test vehicles. In many cases the test wheel
is in line with the vehicle road wheels, either on the same chassis or a trailer. Because
they carry large water tanks, some are necessarily large goods vehicles with axle loads of
up to eight tonnes. (The Dutch ROAR in the HERMES exercises and a SCRIM that has
recently entered service in Belgium, for example, are both built on three-axle truck
chassis). Other devices have the test wheel on a trailer offset from the wheel path and so
the vehicle must pass with its wheels slightly to one side. Whatever form of construction is
chosen, it should be able to accommodate vehicle of these sizes.
As well as being able to carry the weight of the vehicles repeatedly, the test road structure
will also need to have adequate drainage to remove the water deposited by the test
devices. Ideally, a system that could positively remove excess water between passes
would be an advantage.
FEHRL
HERMES Final Report
138
Each test section also needs to be long enough to accommodate the necessary test
passes, including an allowance for accelerating to the required test speed and
decelerating afterwards. Although the length of surface on which friction will be measured
may only be 100 m or so long, some devices, usually those using a variable slip control
system, need a certain length in which to stabilise the required slip ratio at the particular
speed and friction level. This could require a test surfacing some 300 m long, together
with an approach and exit lane to allow for acceleration and braking when higher test
speeds are needed.
A crucial aspect of reference surfaces is that their characteristics should remain stable
over time. This means that ideally, they should be kept clean and should not be exposed
for long periods to extremes of weather, particularly frost, rain and strong sunlight.
Therefore, a test facility should include some means of protecting the test surfaces from
the elements when they are not in use. Consideration should also be given to how any
build-up of tyre deposits on the surfacings as a result of repeated testing can be removed
without adversely affecting the friction characteristics.
While it will be important that the characteristics should be stable and durable, the working
life of a reference surfacing need not be very long, provided that it can be reproduced and
replaced easily and reliably.
There are essentially two choices for the method of construction: the creation of four
permanent surfaces, or the creation of modular surface sections. Permanent surfaces
would probably be quicker to test, especially if laid in-line with one another, and would be
relatively easy to control. However, a large test site would be required and it would be
potentially rather difficult to keep clean and protect from the weather.
The general requirements suggest, therefore, that a modular form of construction that
allows surfaces to be removed and stored when not required could be the preferred
approach. An arrangement that enables the sections to be laid adjacent to one another
with approach and exit areas common to all sections would reduce the length of road
required compared with a linear structure where sections were laid one after the other.
This would allow the test surfaces to be relatively narrow (say 1m wide), with neutral
areas the width of a normal traffic lane to each side that would potentially allow either left-
or right-handed machines to test them without unnecessary wear and tear on the test
sections.
Ideally, the reference component of the test surfaces (i.e. the part where the calibration
friction levels will actually be measured rather than the approach length) should be under
cover, perhaps in a large hangar that is also equipped with water supplies and other
ancillary facilities. The building could include space for cleaning and storing modular
surface sections when they are not being used. Blank modules or slats could be provided
for the test area when it is not in use. This concept is illustrated schematically in Figure
5.4. In principle, an alternative might be to use a single line layout, with modules removed
and replaced with different surfacings during a test session. However, this is only likely to
be worthwhile if the modules can be changed quickly and this may not be a practicable
option. A separate storage building, possibly with a protective tunnel to covering the test
area might also be considered.
FEHRL
HERMES Final Report
139
~200m acceleration length 100m ref. length Deceleration length
1m
3m
Drainage
Observation/Service building/hangar
• water supply
• surface cleaning facilities etc
Figure 5.4 Concept of using parallel modular reference sections
Another advantage of the modular approach is that the sections could be transported
relatively easily from one facility to another if this were required. Is would also be much
easier to provide identical, interchangeable and replaceable surfaces. Spare modules
would need to be built for the different surfaces to allow for possible damage during
handling and use. If a typical module was in the form of a 1m x 5m slab, around 70 slabs
(60 + 10 repairs) would be needed for each surface, assuming a section is 300m long.
The essential feature of a reference surface is that its characteristics should be clearly
defined, easily reproduced and consistent through its working life. An individual module
need not last for a very long time if it can quickly be replaced by another with the same
characteristics.
5.3.4.3 What materials might be used?

The calibration surface itself could be made from three basic types of material:
• Natural aggregate with either a bitumen binder or in a Portland cement concrete.
• An artificial aggregate (such as a ceramic) fixed to a substrate with a resin-based
binder.
• A completely fabricated surface using man-made materials.
(1) Natural aggregates

Although crushed rock aggregates are easy and relatively cheap to obtain, their natural
characteristics are likely to be very variable in the context of creating a reference surface.
Natural aggregates are already known to change their characteristics with time due to
weathering and wear, particularly polishing by traffic.
The use of bitumen as a binder is also a marked disadvantage because there is a strong
possibility of initial contamination of the aggregate surface. This might be avoided with a
surface dressing technique, but this is unlikely to be a successful way of producing a
FEHRL
HERMES Final Report
140
material that will retain its texture depth with the repeated passage of test vehicles.
Similarly, it would be difficult to prepare a surface using cement as a binder because of
the risk of contamination.
Conventional asphalt mixes would be inappropriate for reference surfaces, not only
because of the contamination risk but also because where the asphalt matrix forms part of
the surface, this can be expected to change over time as the bitumen weathers. Cement
concrete mixtures are also likely to gradually wear.
In theory it might be possible to make a surface using conventional materials and then to
condition it in some way before use as a reference. However, the difficulty with this
approach is to know when the correct condition has been reached and what the friction
level would be.
Some gravel aggregates, particularly flint, may have naturally smooth, hard surfaces and
so it might be possible to use such materials to provide a combination of low micro texture
and intermediate or higher macro texture.
However, experience on test tracks using natural aggregate surfaces where skid
resistance devices are regularly compared has shown that their characteristics can
change as a result of repeated testing.
For all these reasons, it is not recommended that natural aggregates or normal asphalt or
concrete mixes should be used
(2) An artificial aggregate and a resin binder

There has been a great deal of research over the years into the production of artificial (or
synthetic) aggregates. Artificial aggregates can be produced with characteristics that can
be controlled and remain relatively stable. Most are derived from naturally-occurring
minerals, which are then treated in some way, for example by calcination (heating to high
temperatures).
Synthetic aggregate particles can be expected to be identical and regular in shape and
these therefore, fixed to a substrate with a suitable binder (such as epoxy resin) could be
a possibility for the creation of reference surfaces. The different levels of the two
components of texture could be achieved by varying the final particle sizes and the
asperities in the surface.
Examples of artificial aggregates that might be explored are listed in Table 5.4.
Table 5.4 Artificial aggregates that might be used in a reference surfacing
low friction high friction
Ceramics Calcined bauxite

Calcined Flint Burned clay
(3) Man-made materials

Using man-made materials means that pre-determined shapes and profiles can be
manufactured and replicated. The techniques might utilise moulded shapes using, for
example, fibreglass and resin, as was tried for special external drum surfaces by Dunlop
and TRL during the 1990s. The shapes could be basic geometric shapes such as
FEHRL
HERMES Final Report
141
hemispheres, cuboids tetrahedra or cylinders (Figure 5.5) or castings taken from actual
road surfaces.
An alternative to moulding surface profiles would be to machine or press them from a

metallic plate, although this technique might not allow some of the more complex patterns
to be easily reproduced over large areas.
The advantage of this type of technique is that it would allow very repeatable surfaces to
be made and would permit different forms and scales of macro texture to be produced.
Using casts from real roads, although theoretically possible, might not be appropriate for
developing reference surfaces. Apart from deciding what general type of road surface
should be used, this approach would make it difficult to ensure homogeneity, both along
each modular section and along the length of the assembled test surface.
A major limitation of this type of approach, however, is that the materials would be unlikely
to have a “natural” micro texture and so this would have to be added somehow, which has
proved a problem in the past. On the other hand, this approach could be useful to make a
low micro-, high macro texture surface (type LH in Table 5.3).
Figure 5.5 Specially formed surfaces using geometric shapes (l-r): hemispheres, cubes,
tetrahedra, cylinders.
5.3.5 Outline specification for calibration reference surfaces
The discussion of the various aspects of reference surfaces in Section 5.2.4 leads to the
view that, at the present time, it is not possible to develop a definitive specification for
reference surfaces suitable for use for the routine calibration of friction measuring devices
against a harmonised index. Further research is needed to take this idea forward (Section
FEHRL
HERMES Final Report
142
5.2.6), but an outline specification can be suggested here to provide a framework for
future investigations.
Table 5.5 gives this suggested outline specification, based on the assumption that a
modular form of construction would be used, allowing surfaces to be set up at a number of
sites or to be moved between suitable locations as required. Some of the details such
materials to be used, the layout of the site or facilities needed that have been discussed
above have not been included in this table since they would need to be finalised in the
light of experience from any future research.
FEHRL
HERMES Final Report
143
Table 5.5 Outline requirements for Calibration Reference surfaces
General Property General Requirements Other Comments

These requirements would apply to
the track on which vehicles run as
well as the test surface. With no
Alignment Straight and level, no cross-fall cross-fall, suitable drainage or other
mechanisims will be needed to rapidly
clear excess water from the test
surfaces.
This is the minimum test length. A
longer length (up to 300m) may be
required to accommodate some
Length 100 metres minimum
devices that use a fixed slip ratio
controlled by a servo system that
responds to changing surface friction.
There will also be a need for space
either side of the test surface when
installed to allow for the passage of
Width 1 metre wide test surface
the test vehicle, depending upon the
relative alignment of the test wheel
and the main vehicle’s tyres.
The load bearing capacity for the
installed module and associated
Similar interlocking modules
roadway will need to be able to
Construction Module size to be chosen to suit
support normal lorry axle-weights in
ease of construction and handling.
order to accommodate the larger test
vehicles.
Each module should be simlar, with
Texture (micro and macro) of the
no significant boundary edges in the
surfacing should be homogeneous
surface where the modules join.
along its length and across its width.
General Joints will beed to be secure and not
The texture should not be so
collect dirt or allow passage of water
aggressive that it causes excessive
(unless suitable sub-surface drainage
test tyre wear.
is provided).
Surface BFC 20 = 0.75-0.85 These are tentative suggestions of
Surface HH MPD = 1.5-2.0 mm target ranges for the key texture
characteristics parameters. Eventually, a more
Surface BFC 20 = 0.75-0.85 precise specification will be needed.
HL MPD = 0.2-0.4 mm
BFC (locked-wheel) values at 20km/h
Surface BFC 20 = 0.25-0.35 have been used here as a suggested
LH MPD = 1.5-2.0 mm indicator of the micro texture level
needed: in practice, other low slip-
Surface BFC 20 = 0.25-0.35 speed measurements could be used.
LL MPD = 0.2-0.4 mm
Should not be bitumen-based or use

Surfacing materials To be determined.
untreated crushed-rock aggregate.
Should maintain defined skid
resistance and texture depth over a
To assist this, it may be necessary to
practical temperature range for at
Stability keep and use the surfaces in a
least 3 years. Friction should not controlled environment.
change over a short period of
repeated testing.
The suggested number of passes is
Should maintain defined skid
sufficient to check up to 5 devices
resistance and texture depth for up
Durability (making 5 test passes at 3 speeds
to 1000 repeated test passes with
plus some additional passes) 3 times
up to 500kg test wheel load.
a year for 3 years.
FEHRL
HERMES Final Report
144
5.3.6 Further research
It is clear that further research is necessary to develop practical reference surfaces. It is

not practical to develop a full research proposal here, but some suggestions of what might
be included can be made.
5.3.6.1 Suggestions for materials to include in the investigation

As discussed in Section 5.2.4.3, the choice of suitable materials to achieve predictable
and stable performance will be difficult, so work to identify materials that could be reliably
specified for use for the reference surfaces will be a fundamental aspect of the research.
The challenge is to find suitable combinations of a regular and repeatable form and level
of macro texture with appropriate treatments or additives to provide predictable, controlled
and durable micro texture.
As a starting point, some or all of the following could be investigated:

• Castings of geometric shapes using resin/fibreglass to represent controlled macro
texture forms (such as in Figure 5.5).
• Cut or pressed shapes or patterns in metal.
• Proprietary anti-slip or high friction materials/coatings.
• Paints with suitable additives to provide micro texture (variations on some road
marking materials might prove suitable).
• Conventional materials, either to prove that, as expected, they would not be suitable
or to assess whether in certain conditions they might be used.
These suggestions assume an approach in which the required surfacing is applied as

some form of overlay to a suitable substrate. An alternative might be to use a material
that can be “refreshed” in some way to a standardised level.
Another aspect that could be considered would be the inclusion of porous or permeable
materials, since these are used on the network and a calibration check on the devices
with this type of surfacing might be appropriate.
5.3.6.2 Module structure

Although the main emphasis on the research would be to establish suitable test surfaces,
consideration should also be given to the structure of the modules on which they would be
used and how they should be supported at road level on the test site.
The research on this would initially be a desk study to assess how the modules might be
engineered, followed by practical tests on the most promising designs.
Possible forms of construction for the bearing surface might include:

• A timber bed.
• Concrete slabs.
• Steel/alloy plates.
All would need a suitable support framework, together with a carefully designed
mechanism to ensure that they would interlock reliably. Some might be ruled out as
impractical in order to achieve a workable size and weight for individual modules.
Tests using the surface materials applied to fixed substrate might also be considered to
allow for the possibility of a permanent installation.
FEHRL
HERMES Final Report
145
5.3.6.3 Outline research programme

In outline, the research might comprise a three-stage process involving the following
ideas:
(1) Investigation on a laboratory or pilot scale of suitable materials

• Tests to explore suitable material combinations to achieve the required levels of
friction and texture depth.
• Accelerated wear testing of the best of the possible materials.
• Comparisons of indoor and outdoor tests.
(2) Design and testing of a suitable form of construction

• Desk study to explore suitable forms of construction for the modules that would
carry the test surfaces.
• Practical durability tests on pilot-scale modules.
(3) Full-scale trial exercise

• Assess both weathering and wear-and-tear under repeated testing over a 3-year
period.
Testing of the reference surface materials would need to include measurements using
friction devices covering the three main principles, plus measurements of physical
characteristics using all suitable techniques available to assess gradual changes.
For accelerated wear tests on the surfaces, consideration should be given to using
laboratory scale indoor test machines or larger-scale outdoor machines such as the LCPC
carousel at Nantes.
For tests on the durability of the module construction, facilities such as the Pavement Test
Facility at TRL, which allows repeated passing of a lorry-sized wheel under controlled
load, might be used.
Clearly, this is a large programme of work that, as with the HERMES project, would
require co-operation between several organisations to achieve. It might be worthwhile
therefore, to limit the work initially to a feasibility study that concentrated on (1), the desk
component of (2) and limited tests for (3).
FEHRL
HERMES Final Report
146
6 Conclusions
The HERMES project was developed through FEHRL, and was carried out under the
guidance of a steering group comprising experts from six FEHRL member laboratories.
The project had the primary objective of testing the procedures for establishing and
maintaining a harmonised skid resistance index proposed by CEN in prEN13036-2 [3]. In
addition to testing the practical aspects, the project also was to consider possible
improvements to the models used to calculate the index and to explore alternative
approaches for possible future consideration. The findings of the the project in relation to
these main tasks are summarised below.
(1) Experimental procedures

The procedures described in Annex B of the prEN [3] have been tested in as realistic
conditions as possible. Nine calibration trials involved fifteen friction devices from eight
countries, covering virtually all the measurement principles normally used to measure skid
resistance at traffic speeds. Seven different devices (but all complying with the relevant
ISO-CEN standards [4]) were used to measure texture depth.
The trials took place in five different countries, either on public roads or on test tracks. The
test surfaces covered a wide range of materials and textures, including porous surfaces.
The trials were grouped in “rounds”, with three trials in each round being carried out in the
space of about one month. Further rounds were carried out at six-month intervals, thus
giving some opportunity for the characteristics of the devices to undergo changes and for
seasonal or weather influences to have an effect, as would be expected to happen in
practice.
The project has shown that the proposed methodology for bringing different groups of
devices together for calibration trials is practical to achieve and can be carried out
successfully. However, the process can be expensive, with costs broadly in the range
3,000 - 5000 euros and 150 - 200 hours of staff time required to organise a trial and from
1500 to 5000 euros and 50 - 100 hours of time for a device to attend a trial, depending
upon how far it has to travel to the test location. Lessons learnt progressively from the
running of the trials have been used to develop a Guideline that can be used as the basis
for organising such exercises in future.
(2) Analysis methods and improved m odels

The application of the analysis method specified in the prEN [3] revealed a number of
drawbacks:
1. The statistical tests proposed in the prEN for discarding outlying results are too
severe and not in line with those commonly used in international standards such
as ISO 5725 [5].
2. The exponential model for the variation of the friction coefficient (F) with slip speed
(S) did not fit the experimental data in a significant number of cases.
3. The prediction of the speed parameter (S0) from texture depth (MPD) is much too
imprecise, and this leads to a significant residual operating speed influence on
EFI.
FEHRL
HERMES Final Report
147
4. As the calibration is repeated trial after trial, the calibration parameters (A and B)
of any given device evolved in such a way that the sensitivity of EFI to the actual
friction characteristics of the surfaces decreased dramatically, leading for example
to unacceptably high values on very low friction surfaces.
In order to remedy those defects, a number of possible improvements have been

considered:
(i) Applying the statistical tests specified in ISO 5725-2 [5] ("k-test" and "h-test") to
remove outlying combinations of devices and sites.
(ii) Applying, as alternative statistical tests, the rule: R2>0,5 to the regressions.
(iii) Applying an additional condition on the coefficient of variation of EFI (CVEFI >5% or
10%) in order to discard measurement series bearing on too narrow a range of
friction levels.
(iv) Using alternative forms of regressions to F against S, namely polynomial and

Stribeck's curve in order to provide a better fit to the data.
(v) Considering the relationship S0 versus MPD as device-specific in order to reduce

the scatter of data with respect to the "universal" law specified in the prEN [3].
(vi) Using a power law for S0 versus MPD instead of a linear law to the fit the data
better.
(vii) Applying a weighting on the regression of S0 against MPD to reduce the influence
of outlying values.
(viii) Trying to relate S0 to texture parameters other than MPD to further improve the
prediction of S0.
(ix) Forcing calibration parameter “A” to zero with a view to preventing the loss of
sensitivity of EFI referred to in point 4 above.
(x) Calculating first the regression of EFI against its grand average <<EFI>> and then
back calculating <<EFI>> against EFI by reversing the equation.
Fourteen different attempts have been made to improve the consistency and precision of
the calibration procedure by resorting to different combinations of the alternative
treatments (i) – (x) above. It was found that better, although still not ideal, results were
obtained by adopting the following options:
Retaining the original exponential model proposed in the prEN [3] for the relationship
between friction and slip speed, i.e.
F = F0 e− S / S 0 (6.1)
Use a new model for the relationship between texture depth (expressed as MPD) and the
speed parameter S0, with device-specific coefficients “a” and “b”, i.e.
FEHRL
HERMES Final Report
148
S 0 = a * MPD b (6.2)
Apply the following weighting in the regression calculation to determine “a” and “b”:
w = ( S0 / σ So ) 2 (6.3)
where s So is the residual standard deviation of S0 with respect the exponential

regression of F against S.
In the definition of EFI, “A” is forced to zero to give:
EFI = B ∗ F30 . (6.4)
Both in the regression F(S) and in the regression yielding the value of “B”, the correlation
coefficient R2 must be higher than 0,5 otherwise, the corresponding measurement
series is discarded.
In the model chosen, both in the prEN and in the further analyses, a reference slip speed
of 30km/h was selected because this was close to that used by many devices in practice.
However, it was found that the GripTester and IMAG devices did not fit the chosen model
well. A possible explanation for this is that these devices have only a narrow range of slip
speeds over the practical range of operating speeds: this leads to unavoidable
extrapolation when relating these devices to the reference speed. It is important to stress
that this is a function of the models, not a criticism of the devices themselves. Including
the results of the GripTester and IMAG unacceptably disturbed the whole calibration
process of the other devices; therefore, they were discarded from the data set.
Further, it was found that two families could be distinguished within the remaining thirteen
devices, namely those using the BFC and the SFC measuring principles. The latter group
comprised the SCRIMs and the Odoliographs, both types types of device having the same
slip ratio of 34%. Although it could be assumed that the participating devices were in
good condition, direct comparison of devices revealed that significant changes may have
occurred to some of them from one trial to the next. However, it is not known to what
extent this could have influenced the outcome of the whole experiment.
(3) General conclusions from the experimental and analytical studies

In the light of the conditions applied in the analysis summarised above, the main results of
the experimental part of the project can be summarised as follows:
1. Regarding consistency, it was concluded that the calibration method works

satisfactorily: the procedure leads to a stable EFI-scale even though limited
subsets of devices operating on different sets of test surfaces are compared in
each calibration exercise.
2. Regarding precision, it was concluded that the reproducibility of the EFI value
delivered by different devices was acceptable for SFC devices with the same slip
ratio but not for BFC devices, which use a wider variety of measurement
principles.
FEHRL
HERMES Final Report
149
3. Analysis of the sources of deviations has shown that further improvements to the
models used for F(S) and S0(MPD) are unlikely to improve the reproducibility of
EFI.
An important conclusion from the project is that it appears that, based on the data from
this study and using the models developed so far, it is not yet practical to harmonise
satisfactorily all the device principles currently used in Europe by using the EFI approach.
(4) Alternative approaches

An alternative approach to the common scale and calibration process associated with the
EFI concept has been considered, in the form of a single reference device. Such a device
could, in the longer term, replace existing devices and itself become a Europe-wide
standard. A set of proposed specifications for a single friction and texture measurement
method based on this device has been prepared in a form that can be proposed to CEN.
The specifications take into account the views of FEHRL laboratories based on their
responses to an enquiry.
Although learning from experience from develping and using present equipment, the
proposed device is intended to be completely new and, so far as the authors are aware, it
uses a different approach to the measurement of skid resistance to that followed by
existing devices. It was decided that the reference device should have the following key
features:
• Simultaneous measurement of friction and macro texture on the same test line to
enable immediate evaluation of EFI at any test speed.
• Friction measurement on the BFC controlled-slip principle, but with a variable slip
ratio, controlled to provide a constant slip speed of 30 km/h.
• Three standard operating speeds (40, 60 and 80km/h).
• Simultaneous
• A standardised reference tyre based on a technical specification currently being
proposed by PIARC [19].
A mega texture indicator will be also given by this device on this same measurement line.
Any real device, even if it is based on a standard specification, will need to be calibrated
or checked somehow and to do this would require stable and reproducible reference
surfaces that could provide and maintain known levels of friction. A set of requirements
for such surfaces has been developed.
(5) Additional findings

Finally, it is noted that two interesting by-products emerged from the work carried out
during the course of the project:
• The quest for a better model for predicting S0 yielded a new relationship that
included not only the macro texture of the surface but also the slip ratio of the
device.
• As a result of the participation of seven texture profilers, the reproducibility of the
Mean Profile Depth (MPD) as defined by ISO 13473-1:1997 [4] has been
determined. It is characterized by an average relative standard deviation of 5%.
FEHRL
HERMES Final Report
150
7 Recommendations
In the sections below, the current situation in the light of the work in the HERMES project
is reviewed and specific recommendations are made as to how to move forward.
7.1 Review of the current situation
The underlying problem in harmonising skid resistance measurements is the absence of a

fixed reference friction level. The principle of EFI attempts to overcome this problem by
using the average of all machines to define a “correct” level. An alternative would be to
use a reference device for direct comparison. However, this still needs to be calibrated to
validate that it is stable. Another alternative is to use reference surfaces to provide stable
(over time), known levels of friction with which all devices can be compared and hence
linked to some form of common scale. However, the reference surfaces must also be
calibrated to verify that they continue to provide the expected level of friction. This would
need a “reference device”. Neither the reference device or reference surfaces remove the
need for an “EFI” kind of scale where different devices using different principles are used.
This gives rise to four possible “scenarios”, reflecting progress from the current position to
a fully-standardised situation:
1. No one type of device gives the same results as another (the current situation).
2. Devices are calibrated against one another on any kind of surface, using the EFI
approach to harmonise the results.
3. During a transition period, different measuring principles will be in use as countries

decide to change (or not) their current practices and devices. These devices must
be calibrated against a “reference device” using a set of representative “reference
surfaces” and correcting for the influence of various factors such as texture etc.
using the EFI approach.
4. In a “standardised” world, all devices would use the same principle – the
“reference device” would become, in effect, a “standard device”. The reference
surfaces would then be needed to verify that each individual “standard device” was
actually measuring the standard values.
In brief, the HERMES project found that:

• The procedure specified in the draft CEN standard [3] works (in the sense that it
allows a common, stable scale of friction to be kept between the various devices
that participated in the calibration exercises) in fully realistic conditions.
• The reproducibility of EFI has proved to be rather poor in comparison with the
reproducibility between devices of a similar type, which makes the procedure as yet
insuitable for a mandatory application.
• The project has paved the way to alternative solutions by drawing up specifications
for a proposed reference device and by setting up general specifications and
possible designs for reference surfaces.
FEHRL
HERMES Final Report
151
7.2 Specific recommendations
In view of the need to improve the present situation, the possible scenarios above and the
results of HERMES, it is recommended that:
(a) For the short term (less than 4 years):
o CEN/TC227/WG5 should be advised to set up a provisional testing and

calibration method based on the experiences gained in the experiments
carried out in HERMES. The purpose of this recommendation is to make the
results provided by various devices comparable through a common scale,
even though it is not perfect.
o The calibration routine initiated by HERMES should be kept running using
the harmonized procedure on a voluntary basis. It is recommended that for
the full operation of the CEN procedures, the co-ordination of trial meetings
and the allocation of devices to them should be the responsibility of a single
overseeing body
o Further analysis of the data collected by HERMES and additional equipment
comparisons should be carried out, with a view to improving the precision of
EFI.
o The prototype of a reference device should be developed.
o A feasibility study into the design of reference surfaces is begun.
(b) For the medium term (4-8 years):
o A second generation of the standard for dynamic measurement of skid

resistance should be prepared, based on a reference device associated to
reference surfaces for calibrating the existing systems, possibly using an
improved version of EFI.
(c) For the long term (more than 8 years):
o The reference device should be progressively substituted for existing

measuring devices.
Finally, it is recommended that the FEHRL Working Group on “Harmonization of friction,

texture and evenness measurement methods” that designed and launched the FILTER
and HERMES projects should be re-convened, in order for that extended group to
evaluate the situation and develop action plans.
Anticipating these future developments, the HERMES Group has prepared three
documents in addition to the general report:
• A proposal for the revision of the prEN 13036-2-Annex B (drafted as Annex J in this
report).
• Guidelines for organizing the calibration of skid resistance testing devices (drafted
as a stand-alone document accompanying this report).
• A proposal of a standard reference device for dynamic skid resistance testing of
pavements (drafted as Annex L of this report).
FEHRL
HERMES Final Report
152
References
[1] WAMBOLD J.C., ANTLE C.E., HENRY J.J., RADO Z., DESCORNET G., SANDBERG U.,
GOTHIE M., HUSCHEK S., "International PIARC Experiment to Compare and Harmonize Skid
Resistance and Texture Measurements", PIARC Publication n°01.04.T, Paris, 1995.
[2] DESCORNET G., “Proposal for a European standard in relation with the skid resistance of
road surfacings”, Final report, Research contract SSTC NO/C3/004, Belgian Road Research
Centre, Brussels, 1998.
[3] CEN/TC 227/WG5 5, prEN 13036-2:2003 “Surface characteristics of road and airfield
pavements - Procedure for determination of skid resistance of a pavement surface”, CEN.
[4] ISO 13473-1:1997, “Characterization of pavement texture by use of surface profiles: Part
1: Determination of Mean Profile Depth”, ISO.
[5] ISO 5725-2:1994, “Accuracy (trueness and precision) of measurement methods and
results – Part 2: Basic method for the determination of repeatability and reproducibility of a
standard measurement method”, ISO.
[6] VAN DEN BOL M.E., BENNIS T.A., “Measuring Systems for the Evaluation of Skid
Resistance and Texture – Relation between the friction Measurements from Different Devices and
the International Friction Index”, Proceeding of the 4th International Symposium on Pavement
Surface Characteristics of Roads and Airfields, pp. 137-182, Nantes, 2000.
[7] JACOBS M.M.J., GERRITSEN W., WENNINK M.P., VAN GORKUM F., “Optimization of
Skid Resistance Characteristics with Respect to Surface Materials and Road Function”,
Proceedings of the 3rd International Symposium on Pavement Surface Characteristics, pp. 283-
298, 1996.
[8] EMMENS W.C., “The Influence of Surface Roughness on Friction”, Proceedings of the
15th Vienna Congress of IDDRG, pp. 63-70, 1988.
[9] VAN DEN BOL M.E., Personal communication, March 2003.
[10] MANCOSU F., PARRY A., LA TORRE F., “Friction Variation Due to Speed and Water
Depth”, Proceedings of the 4th International Symposium on Pavement Surface Characteristics of
Roads and Airfields, pp. 249-258, Nantes, 2000.
[11] KENNEDY J.B., NEVILLE A.M., “Basic Statistical Methods for Engineers and Scientists”,
Thomas Y. Crowell Company, Inc. 2nd edition, pp. 335-244, 1976.
[12] GOTHIE, M.G. “Proposal for Specification for a Standard Test Tyre for Friction Coefficient
Measurement of a Pavement Surface: Smooth Test Tyre (1)”, PIARC C1 (Committee on Surface
Characteristics), Paris, 2003.
[13] S.C. BRITTON, W.B LEDBETTER, AND B.M.GALLAWAY. "Estimation of Skid Numbers
from Surface Texture Parameters in the Rational Design of Standard Reference Pavements for
Test Equipment Calibration". Journal of testing and Evaluation, JTEVA, Vol 2 , Nº2 p73-83, 1974.
[14] ISO/TC22/SC 9. "Vehicle dynamics and road-holding ability", ISO.
[15] ISO/TR 8350-1986 (E). "Road vehicles - High Friction test track surface- Specification",
ISO.
FEHRL
HERMES Final Report
153
[16] HUCKINS, H.C. “FHWA Skid Measurement Test Centres”. Public Roads. Federal Highway
Administration. Washington D.C. September, 1977.
[17] ELDRIDGE A. WHITEHURST. E. A AND NEUHARDT. J. B. “Time-History Performance of

Reference Surfaces. Page 61-71. The Tire Pavement Interface”. Pottinger/Yager edition. ASTM.
Baltimore, 1986.
[18] J WAMBOLD, J HENRY, "NASA Wallops Tire/Runway Friction Workshops". NASA report.
NASA Langley Research Centre Hampton, VA 23681-2199, 2002.
[19] PIARC STRATEGIC THEME 1, COMMITTEE C1, SURFACES CHARACTERISTICS,

Technical document: “Specification for a standard test tyre for friction coefficient measurement of a
pavement surface: Smooth test tyre”, PIARC.
FEHRL

Hermes Final Report PDF

Uploaded by

Copyright:

Available Formats

Hermes Final Report PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Hermes Final Report PDF

Uploaded by

Copyright:

Available Formats

FEHRL Report 2006/01

Title: Harmonization of European Routine and research Measuring Equipment for

Authors: G. Descornet – BRRC, Belgium

Keywords: Roads, Skid Resistance, Measurements, Harmonization.

FEHRL Secretariat: S. Phillips

Tel: +32 2 775 82 38

4.1 INTRODUCTION ............................................................................................................................................. 88

The HERMES Project

HERMES – “Harmonization of European Routine and research Measuring Equipment for

Work programme summary

To meet those five objectives, the following tasks were identified:

Belgium: Ministère fédéral des Affaires Economiques - Division Compétitivité

Guy DESCORNET (Convenor) BRRC Belgium

Participants in the field trials

Organization Country Particular thanks to:

It is unrealistic to expect such a procedure to rely on a single device if it is to be

Anticipating this need, in 1992, PIARC conducted an international experiment to compare

Meanwhile, the FEHRL working group on “European Harmonisation of Friction, Texture

2 The HERMES calibration exercises

2.1 General description of trials

The purposes of the HERMES calibration exercises (referred to subsequently in this

2.1.2 Participating devices

Altogether, fifteen skid resistance-measuring devices and eight texture-measuring devices

Device Operator Organisation Key Operating

86% Fixed Slip.

Side Force 20º wheel

Side Force 20º wheel

15% fixed slip.

Variable slip device run at

Variable slip device run at

Side Force 20º wheel

Device Operator Organisation Key Operating

Variable slip device run at

Variable slip device run at

Side Force 20º wheel

15% Fixed slip.

Device Operator Organisation Key Operating

Table 2.2 Texture devices used in HERMES trials

2.1.3 Trial Planning

Table 2.3 Summary of trial rounds

Table 2.4 Devices attending each trial

Table 2.5 Participation matrix – trials attended by each device

2.1.4 The trials – general principles

APPENDIX 1 TRIAL MATRIX WEDNESDAY 2ND OCTOBER

Route 1 Practice Run Route 1 Texture Run

Route 1 Replicate 1 Replicate 2 Replicate 3 Replicate 4 Replicate 5

9:40 1 50 10:20 6 60 11:00 13 50 11:40 13 50 12:20 9 60

9:50 1 90 10:30 6 90 11:10 13 90 11:50 13 90 12:30 9 90

LUNCH 12:45 to 13:30 TRANSPORT WILL BE PROVIDED TO ROOM B0151

Route 2 Practice Run Route 2 Texture Run

Route 2 Replicate 1 Replicate 2 Replicate 3 Replicate 4 Replicate 5

14:30 1 50 15:10 6 60 15:50 13 50 16:30 13 50 17:10 9 60

Figure 2.1 Example of trial running timetable

Harmonisation of European Routine and Research Measuring

Section Target speed 30km/h Intermediate speed Target speed 90km/h

2.1.5 The trials – Round 1

2.1.5.1 Trial 1,1 – UK

2.1.5.2 Trial 1,2 - Netherlands

Figure 2.5 Control area on the Lelystad test track