Richard Leach (Auth.), Richard Leach (Eds.) - Characterisation of Areal Surface Texture-Springer-Verlag Berlin Heidelberg (2013) PDF

Richard Leach Editor
Characterisation
of Areal Surface
Texture
Characterisation of Areal Surface Texture
Richard Leach
Editor
Characterisation of Areal
Surface Texture
123
Editor
Richard Leach
Engineering Measurement Division
National Physical Laboratory
Teddington
UK
ISBN 978-3-642-36457-0 ISBN 978-3-642-36458-7 (eBook)

DOI 10.1007/978-3-642-36458-7
Springer Heidelberg New York Dordrecht London
Library of Congress Control Number: 2013934384
Ó Springer-Verlag Berlin Heidelberg 2013

This work is subject to copyright. All rights are reserved by the Publisher, whether the whole or part of
the material is concerned, specifically the rights of translation, reprinting, reuse of illustrations,
recitation, broadcasting, reproduction on microfilms or in any other physical way, and transmission or
information storage and retrieval, electronic adaptation, computer software, or by similar or dissimilar
methodology now known or hereafter developed. Exempted from this legal reservation are brief
excerpts in connection with reviews or scholarly analysis or material supplied specifically for the
purpose of being entered and executed on a computer system, for exclusive use by the purchaser of the
work. Duplication of this publication or parts thereof is permitted only under the provisions of
the Copyright Law of the Publisher’s location, in its current version, and permission for use must always
be obtained from Springer. Permissions for use may be obtained through RightsLink at the Copyright
Clearance Center. Violations are liable to prosecution under the respective Copyright Law.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this
publication does not imply, even in the absence of a specific statement, that such names are exempt
from the relevant protective laws and regulations and therefore free for general use.
While the advice and information in this book are believed to be true and accurate at the date of
publication, neither the authors nor the editors nor the publisher can accept any legal responsibility for
any errors or omissions that may be made. The publisher makes no warranty, express or implied, with
respect to the material contained herein.
Printed on acid-free paper
Springer is part of Springer Science+Business Media (www.springer.com)

This book is dedicated to my beautiful mother
Preface
The function of a component part can be profoundly affected by its surface

topography. There are many examples in nature, of surfaces that have a well-
controlled topography to affect their function. Examples include the hydrophobic
effect of the lotus leaf, the reduction of fluid drag due to the riblet structure of
shark skin, the directional adhesion of the gecko foot and the angular sensitivity of
the multi-faceted fly eye. Surface structuring is also being used extensively in
modern manufacturing industries. Many properties can be altered, for example
optical, tribological, biological and fluidic. Previously, single line (profile) mea-
surements were adequate to control manufacture of surfaces, but as the need to
control the functionality of surfaces increases, there is a growing need for three-
dimensional (areal) measurement and characterisation techniques. For this reason
there has been considerable research, development and standardisation of areal
techniques. This book will present the areal framework being adopted by the
international community. Whereas previous books have concentrated on the
measurement aspects, this book will concentrate on the characterisation tech-
niques, i.e. how to interpret the measurement data to give the appropriate (func-
tional) information for a given task. The first part of the book presents the
characterisation methods and the second part presents case studies that highlight
the use of areal methods in a broad range of subject areas—from automobile
manufacture to the surfaces of roads.
vii
Acknowledgments
First and foremost I would like to thank all the chapter authors for their hard work
and dedication to this book. The following friends and colleagues have also
contributed to my understanding of the subject: Mr. Claudiu Giusca and Dr. Peter
Harris (NPL), Prof. Jane Jiang, Prof. Liam Blunt and Prof. Paul Scott (University
of Huddersfield). I also wish to thank my beautiful wife for allowing me to spend
hours writing and days travelling in order to become an expert in such an inter-
national field—thanks Sharmin. Last, but not least, my parents, sisters, son and
stepson also need to be thanked for their unwavering support.
ix
Contents
1 Introduction to Surface Topography . . . . . . . . . . . . . . . . . . . . . . 1

Richard Leach
2 The Areal Field Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

François Blateyron
3 The Areal Feature Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . 45

François Blateyron
4 Areal Filtering Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

Jörg Seewig
5 Areal Form Removal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

Alistair B. Forbes
6 Areal Fractal Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

Christopher A. Brown
7 Choosing the Appropriate Parameter . . . . . . . . . . . . . . . . . . . . . 155

Franz Helmli, Kerstin Pötsch and Claudia Repitsch
8 Characterisation of Individual Areal Features . . . . . . . . . . . . . . . 179

Nicola Senin and Liam Blunt
9 Multi-Scale Signature of Surface Topography . . . . . . . . . . . . . . . 217

H. Zahouani and M. El Mansori
10 Correlation of Areal Surface Texture Parameters

to Solar Cell Efficiency . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 269
Erik Novak and Nelson Blewett
xi
xii Contents
11 Characterisation of Cylinder Liner Honing Textures

for Production Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 281
Zlate Dimkovski, Cecilia Anderberg, Robert Ohlsson
and B.-G. Rosén
12 Characterisation of the Mechanical Bond Strength

for Copper on Glass Plating Applications . . . . . . . . . . . . . . . . . . 303
Baofeng He, Jon Petzing, Paul Webb, Paul Conway
and Richard Leach
13 Inspection of Laser Structured Cams and Conrods . . . . . . . . . . . 321

Franz Helmli and Stefan Lehmann
14 Road Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337

Alan Dunford
Index . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349
Chapter 1
Introduction to Surface Topography
Richard Leach
Abstract This chapter introduces the field of surface topography measurement.

The basic concepts of surface topography measurement and characterisation are
discussed, including a short historical overview. The differences between surface
profile and areal surface characterisation are presented along with the concepts of
software measurement standards. Finally, the current international standards
infrastructure and potential future direction for standards are highlighted.
1.1 Surface Topography Measurement

and Characterisation
Most manufactured parts rely on some form of control of their surface charac-
teristics. The surface is usually defined as the feature on a component or device
that interacts with the environment in which the component is housed or in which
the device operates, or with another surface. The surface topography, and of course
the material characteristics, of a part can affect how two bearing parts slide
together, how fluids interacts with the part, or how the part looks and feels. The
need to control, and hence, measure surface features becomes increasingly
important as we move into a miniaturised world (Leach et al. 2011). Surface
features can become the dominant functional features of a part and may become
large in comparison to the overall size of an object.
The list of terms and definitions in the field of surface texture is extensive.
Unless otherwise stated, this book will be consistent with ISO specification
standards (published and draft, Sect. 1.5) and the NPL good practice guides (Leach
R. Leach (&)
Engineering Measurement Division, National Physical Laboratory, Hampton Road,
Teddington, Middlesex, TW11 0LW, UK
e-mail: richard.leach@npl.co.uk
R. Leach (ed.), Characterisation of Areal Surface Texture, 1

DOI: 10.1007/978-3-642-36458-7_1, Springer-Verlag Berlin Heidelberg 2013
2 R. Leach
2001; Petzing et al. 2010). Here, surface topography is defined as the overall
surface structure of a part (i.e. all the surface features treated as a continuum of
spatial wavelengths), surface form as the underlying shape of a part (for example, a
cylinder liner has cylindrical form) and surface texture as the features that remain
once the form has been removed (for example, machining marks on the cylinder
liner). The manner in which a surface governs the functionality of a part is also
affected by the material characteristics and sub-surface physics, or surface integ-
rity (see Griffiths 2001), but this topic is not covered in this book.
The focus of this book is the characterisation of surface texture, that is to say
the handling of surface texture data to give meaningful information once a mea-
surement has been made. Only surface texture will be considered; the measure-
ment of surface form is covered in many other text books and references; see for
example, Malacara (2007). Research into surface texture measurement and char-
acterisation has been carried out for over a century and is still very active, espe-
cially as the new areal surface texture specification standards have now begun to
be introduced. The reader is referred elsewhere for more in-depth treatments of
surface measurement (Leach 2009, 2011; Mainsah et al. 2010; Whitehouse 2010;
Blunt and Jiang 2003).
The characterisation of surface topography is a complex branch of metrology
with a large range of parameters. The proliferation of surface texture character-
isation parameters has been referred to as ‘‘parameter rash’’ (Whitehouse 1982)—
at any one time there can be over one hundred parameters to choose from.
However, due to recent activities, there will soon be a coherent international
standards infrastructure to support surface texture characterisation. Profile char-
acterisation (see Sect. 1.2) has been standardised for some time now. Areal
characterisation (see Sect. 1.3) is still under development but the standards are
progressively being published (see Sect. 1.5).
The first important work on areal surface texture characterisation was carried
out by a consortium as part of a European project led by Ken Stout from the
University of Birmingham (Thomas 2008). This project ended with the publication
of the ‘‘Blue Book’’ (Stout et al. 1993) and the definition of the so-called ‘‘Bir-
mingham-14’’ parameters. Following this project, ISO initiated standardisation
work on areal surface texture. However, ISO experts rapidly realised that further
research work was needed to determine the stability of areal parameters and their
correlation with the functional criteria used by industry. A further project
(‘‘SURFSTAND’’) was carried out between 1998 and 2001, by a consortium of
universities and industrial partners, led by Liam Blunt of the University of
Huddersfield. SURFSTAND ended with the publication of the ‘‘Green Book’’
(Blunt and Jiang 2003) and generated the basic documents for forthcoming
specification standards.
Areal surface texture characterisation methods that are now either standardised
or are at the draft stage are covered in this book in depth. The book will also
address characterisation techniques that are not necessarily internationally
accepted, but are likely to form the basis for future standardisation or to be used for
specialised applications. There are many other parameters (and filtering methods)
1 Introduction to Surface Topography 3
that can be found on older/extant instrumentation and are used in many industries,
but it would never be possible to cover the entire field in one book. Further
methods for surface characterisation, including those from the fields of roundness
measurement, and frequency and waveform analysis can be found elsewhere
(Whitehouse 2010; Muralikrishnan and Raja 2008).
Parameters for areal surface texture have only been introduced recently and
there has been limited research on their use. For this reason this book is split into
two main parts. The first part describes the parameters and how to calculate them
(including filtering methods) and the second part covers industrial case studies.
In the author’s opinion, it is the latter part that will motivate industry to start
embracing areal methods. It is also expected that most users of surface texture
parameters will have access to software packages that can be used to calculate
parameters and will not attempt to code the parameters from scratch. However, it
is vital that software packages be checked for correctness, where possible, using
software measurement standards (see Sect. 1.4).
1.2 Surface Profile
Surface profile measurement is achieved by measuring a line across the surface

and representing that line mathematically as a height function with lateral
displacement, z(x). When measuring and characterising surface texture, use is
made of the rectangular coordinate of a right-handed Cartesian set, in which the
x axis provides the direction of the line, the y axis lies nominally on the real
surface, and the z axis is the outward direction from the material to the surrounding
medium (see Fig. 1.1).
With a stylus (or optical scanning instrument), profile measurement is carried
out by traversing the stylus across a line on the surface. With an areal optical
instrument, a profile is usually extracted in software after an areal measurement
has been performed. Figure 1.2 shows the result of a profile measurement
extracted from an areal measurement.
When making measurements with a stylus instrument, the traversing direction
for assessment purposes is defined in ISO 4287 (2000) as perpendicular to the
direction of the lay, unless otherwise indicated. The lay is the direction of the
predominant surface pattern. Lay usually derives from the actual production
process used to manufacture the surface and results in directional striations across
the surface. The appearance of the profile being assessed is affected by the
direction of the view relative to the direction of the lay and it is important to take
this into account when interpreting surface texture parameters. Surface profile
measurement and characterisation are covered in detail elsewhere (Leach 2001; BS
1134 2010).
4 R. Leach
Fig. 1.1 Coordinate system

for profile measurement
Fig. 1.2 Example of the result of a profile measurement
1.3 Areal Surface Topography
The measurement and characterisation of surfaces using the profile method

(see Sect. 1.2) has been carried out in manufacturing industry for over a century (see
Leach 2009 for a short historical overview and Hume 1980 for a more thorough
treatment). However, whereas the profile method may be useful for showing man-
ufacturing process change, much more functional information about the surface can
be gained from an analysis of the areal surface topography. Also, over the last few
decades there has been a change in the types of surfaces being used in manufac-
turing. Previously, stochastic and random surfaces, or the machining marks left by
the manufacturing process, were most often used to impart functionality into the
surface. More recently, deterministic patterning is being used to critically control the
function of a surface (Evans and Bryan 1999; De Chiffre et al. 2003; Jiang et al.
2007; Bruzzone et al. 2008).
To a large extent, the use of deterministic patterning to control function is
duplicating the way that surfaces have evolved in the natural world. For example,
the riblet micro-structures on a shark’s skin allows it to glide more easily through
water (Bechert et al. 1999) and the complex, multi-scale surface structures on the
skin of a snake allow it to have unique tribological and thermal properties (Abdel-
Aal et al. 2011). Modern manufacturing industry is now using a large range of
structuring techniques to affect the function of component parts. Examples include
the following:
• surface structuring to encourage the binding of biological implants, for example
to promote bone integration and healing (Shalabi 2006) or cell adhesion (Bächle
and Kohal 2004);
• micro-optical arrays for displays, lightings, safety signage, backlighters and
photo-voltaics (see Chap. 10);
• nanostructured surfaces that affect plasmonic interactions for anti-reflection
coatings, waveguides and colour control (Wang et al. 2007)—recent researchers
have attempted to mimic the multi-scale surfaces found in, for example moth-
eyes (Huang et al. 2008; Kettle et al. 2008), see Fig. 1.3;
Fig. 1.3 SEM image of FIB fabricated 2 9 2 array of moth-eye lenses, (10 9 10 9 2) lm (from
Kettle et al. 2008). The insert SEM zoom-in image of the patterned bottom of the micro-lenses
with nano-lenses, Ø150 9 50 nm, in hexagonal arrangement
6 R. Leach
• surfaces of microfluidic channels for flow control, mixing, lab-on-a-chip and

biological filtering; and
• deterministic patterning to control tribological characteristics such as friction,
rheology and wear, for example laser texturing of automotive components
(Etsion 2005; Nosonovsky and Bhushan 2007).
There are a number of significant differences between profile and areal analysis.
Firstly, most of the structures in the list above require areal characterisation to
predict or control their function. Whereas it may be possible to use the profile
method to control quality once a machining process has been shown to be suffi-
ciently stable, for problem diagnostics and function prediction, an areal mea-
surement is often required.
Also, with profile measurement and characterisation it is often difficult to deter-
mine the exact nature of a topographic feature. Figure 1.4 shows a profile and an areal
surface map of the same component covering the same measurement area. With the
profile alone a discrete pit is measured on the surface. However, when the areal
surface map is examined, it can be seen that the assumed pit is actually a valley and
may have far more bearing on the function of the surface than a discrete pit.
Lastly, an areal measurement will have more statistical significance than an
equivalent profile measurement, simply because there are more data points and an
areal map is a closer representation of the ‘‘real surface’’.
To conclude this section, the profile method has been used for over a century, is
relatively simple to apply, is well-established and is still the most utilised method for
surface characterisation in manufacturing industry, especially for process and quality
control purposes. But, as manufacturing industry is increasingly using deterministic
surface structuring methods to significantly enhance the functionality, efficiency
and usefulness of components, areal methods of analysis are becoming more
Fig. 1.4 A profile taken from a 3D measurement shows the possible ambiguity of 2D
measurement and characterisation (from Leach 2009)
commonplace. However, the complexity of areal analysis and the fact that an areal
measurement can take significantly longer than a profile measurement, means that if
profile methods can be used, they should be. This book is designed to help with the
choice of whether to adopt areal methods or not.
1.4 Software Measurement Standards
Surface texture characterisation involves a large array of filtering methods and

parameter calculations. The software packages that are supplied with surface
texture measuring instruments, and some stand-alone software applications, usu-
ally offer a bewildering range of options for characterisation. Where possible,
these software applications should be verified by comparing them to reference
software. ISO 5436 part 2 (2000) presents two types of software measurement
standard for profile measurement and ISO 25178 part 7 (2012) presents the two
areal counterparts. Only the profile software measurement standards will be dis-
cussed in detail here, but the general principles also apply in the areal case.
The two types of software measurement standards are:
• Type F1 (S1 for areal)—reference data files. These are digital representations of
a profile that are used as input to the software under test. The results from the
software under test are compared with the certified results provided with the
type F1 (S1) software measurement standard. Type F1 (S1) software measure-
ment standards are often referred to as softgauges.
• Type F2 (S2 for areal)—reference software. Reference software consists of
traceable computer software against which software in a measuring instrument
(or stand-alone package) can be compared. Type F2 (S2) software measurement
standards are used to test software by inputting a common data set into both the
software under test and the reference software and comparing the results.
Of course, the types F1 and F2 (S1 and S2) software measurement standards are
related. Types F1 (S1) standards can be generated as mathematically known
functions such as sinusoids, for which parameters can be calculated analytically
and independently. These functions can be input to candidate software, and if this
software passes the acceptance test for many different type F1 (S1) software
measurement standards, it can be considered as type F2 (S2) software.
Software measurement standards for profile characterisation are available from
some national measurement institute web sites, see for example Jung et al. (2004);
Bui and Vorburger (2004); Blunt et al. (2008). The user can either download type
F1 standards or upload data files for type F2 analyses.
At the time of writing, software measurement standards for areal character-
isation are only available on the web site of the National Physical Laboratory
(NPL) (Harris et al. 2012a) and for comparison purposes (that is to say, not as
reference software) on the web site of the National Institute of Standards and
Technology (NIST) (Bui and Vorburger 2006). A recent comparison of the NPL
8 R. Leach
software measurement standards with those from NIST, and a number of

commercial software packages, has shown good agreement, at least for the simpler
field parameters (Harris et al. 2012b)..
1.5 Current International Standards
Surface texture documentary standards are part of the scope of the International
Organization for Standardization (ISO) Technical Committee 213 (TC 213),
dealing with Dimensional and Geometrical Product Specifications and Verification
as well as many national committees. ISO TC 213 has developed a wide range of
standards for surface texture measurement for both profiling and areal methods and
has an ambitious agenda for future standards.
1.5.1 Profile Standards
There are nine ISO specification standards relating to the measurement and
characterisation of surface profile. These standards only cover measurements made
with stylus instruments. The details of most of the standards are presented in Leach
(2001) and summarised in Leach (2009), and their content is not discussed in detail
in this book. It should be noted that the current ISO plan for surface texture standards
is that the profile standards will become a sub-set of the areal standards. Whilst the
basic standards and details will probably not change significantly, the reader should
keep abreast of the latest developments in standards. The following is a list of the
profile specification standards as they stand at the time of writing of this book:
• Nominal characteristics of contact (stylus) instruments (ISO 3274 1996).
• Rules and procedures for the assessment of surface texture (ISO 4288 1996).
• Metrological characteristics of phase correct filters (ISO 11562 1996)
• Motif parameters (ISO 12085 1996).
• Surfaces having stratified functional properties—Filtering and general mea-
surement conditions (ISO 13565 part 1 1996).
• Surfaces having stratified functional properties—Height characterisation using
material ratio curve (ISO 13565 part 2 1998).
• Terms, definitions and surface texture parameters (ISO 4287 2000).
• Measurement standards—Material measures (ISO 5436 part 1 2000).
• Software measurement standards (ISO 5436 part 2 2000).
• Calibration of contact (stylus) instruments (ISO 12179 2000).
• Surfaces having stratified functional properties—Height characterisation using
material probability curve (ISO 13565 part 3 2000).
• Indication of surface texture in technical product documentation (ISO 1302
2002).
1.5.2 Areal Standards
In 2002, ISO technical committee 213 formed working group (WG) 16 to address
standardisation of areal surface texture measurement methods. WG 16 is devel-
oping a number of draft standards encompassing definitions of terms and param-
eters, calibration methods, file formats and characteristics of instruments. Several
of these standards have been published and a number are at various stages in the
review and approval process. The plan is that the profile standards will be a sub-set
of the areal standards (with appropriate re-numbering). Hence, the profile stan-
dards will be re-published after the areal standards (with some omissions, ambi-
guities and errors corrected) under a new numbering scheme that is consistent with
that of the areal standards. All the areal standards are part of ISO 25178, which
will consist of at least the parts shown in Table 1.1 (correct at the time of pub-
lication of this book), under the general title Geometrical product specification
(GPS)—Surface texture: Areal.
Part 1 presents the rules for the indication of surface texture in technical product
documentation such as drawings, specifications, contracts and reports. Part 2
presents the definitions of the surface texture parameters (i.e. the field and feature
parameters—see Chaps. 2 and 3) and how to calculate the parameters, including a
limited number of case studies. Part 3 describes the various default values, and
basic rules and procedures for areal surface topography measurement and char-
acterisation. Whereas the profile analysis standards include a standard on how to
filter surface texture data (ISO 11562 1996), there are so many filter types
available for areal analysis that a new suite of standards is being developed. The
areal filtering standards are all parts of ISO 16610 and are described in detail in
Chap. 4.
Part 4 on comparison rules and Part 5 on verification operators are currently not
considered mature enough to be able to produce definitive standards. The part
numbers have been reserved for future use when the appropriate research and
practical testing on these topics has been established.
Part 6 lists, and briefly describes, the various types of instrument for measuring
surface topography. Note that stylus and some of the optical measurement methods
listed in part 6 are further described by dedicated parts (the ‘‘60X series’’).
However, some techniques do not have a 60X equivalent part (for example,
scanning probe or electron beam techniques), but it is expected that these parts will
be developed in future standards.
Part 70 describes the artefacts that are used to calibrate areal surface topography
measuring instruments and includes the profile calibration artefacts from ISO 5436
part 1 (2000), but with new names. Part 71 describes the concepts and use of
software measurement standards (see Sect. 1.4) and part 72 an XML file format for
the standard data file types described in part 71.
There are two parts 60X standards that have been published: part 601 (stylus
instruments) and part 602 (confocal chromatic probes). At the time of writing,
parts 603 and 604 are at the FDIS stage, and are expected to be published in 2013,
10 R. Leach
Table 1.1 Current status of ISO 25178 areal specification standards. Note that these standards
are fully cited in the References section. Key: WD—working draft, CD—committee draft,
NS—not started, DIS—draft international standard, FDIS—final draft international standard,
PS—published standards
No. Title Status Date
1 Areal surface texture drawing indications CD 2009
2 Terms, definitions and surface texture parameters PS 2012
3 Specification operators PS 2012
4 Comparison rules NS –
5 Verification operators NS –
6 Classification of methods for measuring surface texture PS 2010
70 Measurement standards for areal surface texture measurement instruments DIS 2012
71 Software measurement standards PS 2012
72 Software measurement standards—XML file format CD 2012
600 Nominal characteristics of areal surface topography instruments WD 2012
601 Nominal characteristics of contact (stylus) instruments PS 2010
602 Nominal characteristics of non-contact (confocal chromatic probe) PS 2010
instruments
603 Nominal characteristics of non-contact (phase shifting interferometric FDIS 2012
microscopy) instruments
604 Nominal characteristics of non-contact (coherence scanning interferometry) FDIS 2012
instruments
605 Nominal characteristics of non-contact (point autofocus) instruments FDIS 2012
606 Nominal characteristics of non-contact (variable focus) instruments CD 2012
607 Nominal characteristics of non-contact (imaging confocal) instruments WD 2012
700 Calibration of areal surface measuring instruments WD 2011
701 Calibration and measurement standards for contact (stylus) instruments PS 2010
and parts 605–607 are working drafts. The 60X standards currently contain
common terminology, metrological characteristics and a list of parameters that can
influence the uncertainties when using the instrument. There are also technical
annexes that discuss the theory and operation of the instruments. However, as the
60X series developed, it was realised that there are a large number of sections in
the 60X parts that are common to all instruments based on a microscope objective.
For example, research into the metrological characteristics has shown that a
common set can be found that does not differ for each instrument type (see Giusca
et al. 2012a, b, 2013). Therefore, a new standard is under development (part 600),
which will cover all the common aspects. Once part 600 is published, the 60X
series will be withdrawn and re-issued with the common sections removed.
Part 701 is concerned with the calibration of stylus instruments. Part 700, which
is still under development, will cover the calibration of instruments and is expected
to be common across all instruments. Once part 700 is published, part 701 will be
withdrawn.
The American National Standards Institute has also published a comprehensive
documentary specification standard, ANSI/ASME B46.1 (2010) that includes
some areal analyses (mainly fractal-based).
References
Abdel-Aal HA, Vargiolu R, Zahouani H, El Mansori M (2011) A study on the frictional response
of reptilian shed skin. In: Proceeding 13th International Conference Metrology and Properties
of Engineering Surfaces, Twickenham pp 66–70
ANSI/ASME B46.1 (2010) Surface texture, surface roughness, waviness and lay. American
National Standards Institute, Washington
Bächle M, Kohal RJ (2004) A systematic review of the influence of different titanium surfaces on
proliferation, differentiation and protein synthesis of osteoblast-like MG63 cells. Clin Oral
Implant Res 15:683–693
Bechert DW, Bruse M, Hage W (1999) Experiments with three-dimensional riblets as an
idealised model of shark skin. Exp Fluids 28:403–412
Blunt LA, Jiang X (2003) Advanced techniques for assessment surface topography. Kogan Page
Science, London
Blunt L, Jiang X, Leach RK, Harris PM, Scott P (2008) The development of user-friendly
software measurement standards for surface topography software assessment. Wear
264:389–393
Bruzzone AAG, Costa HL, Lonardo PM, Lucca DA (2008) Advances in engineering surfaces for
functional performance. Ann CIRP 57:750–769
BS 1134 (2010) Assessment of surface texture: guidance and general information. British
Standards Institute, London
Bui S, Vorburger TV (2006) Surface metrology algorithm testing system. Precis Eng 31:218–225
Bui SH, Renegar TB, Vorburger TV, Raja J, Malburg MC (2004) Internet-based surface
metrology algorithm testing system. Wear 257:1213–1218
De Chiffre L, Kunzmann H, Peggs GN, Lucca DA (2003) Surfaces in precision engineering,
micro engineering and nanotechnology. Ann CIRP 52:561–577
Etsion I (2005) State of the art in laser surface texturing. J Tribol 127:248–253
Evans C, Bryan J (1999) ‘‘Structured’’, ‘‘textured’’, or ‘‘engineered’’ surfaces. Ann CIRP
48:451–456
Giusca CL, Leach RK, Henning A, Coupland JM (2013) Calibration of the scales of areal surface
topography measuring instruments: part 3—resolution. Meas Sci Technol to be published
Giusca CL, Leach RK, Helery F, Gutauskas T, Nimishakavi L (2012a) Calibration of the scales of
areal surface topography measuring instruments: part 1—measurement noise and residual
flatness. Meas Sci Technol 23:035008
Giusca CL, Leach RK, Helery F (2012b) Calibration of the scales of areal surface topography
measuring instruments: Part 2—Amplification coefficient, linearity and squareness. Meas Sci
Technol 23:065005
Griffiths B (2001) Manufacturing surface technology. Penton Press, London
Harris PM, Smith IM, Leach RK, Giusca CL, Jiang X, Scott PM (2012a) Software measurement
standards for areal surface texture parameters: part 1—algorithms. Meas Sci Technol
23:105008
Harris PM, Smith IM, Wang C, Giusca CL, Leach RK (2012b) Software measurement standards
for areal surface texture parameters: part 2—comparison of software. Meas Sci Technol
23:105009
Huang J, Wang X, Wang ZL (2008) Bio-inspired fabrication of antireflection nanostructures by
replicating fly eyes. Nanotechnology 19:025602
Hume KJ (1980) A history of engineering metrology. Mechanical Engineering Publications Ltd,
London
ISO 11562 (1996) Geometrical product specification (GPS)—surface texture: profile method—
metrological characteristics of phase correct filters. International Organization of
Standardization
ISO 12085 (1996) Geometrical product specifications (GPS)—surface texture: profile method—
motif parameters. International Organization for Standardization
12 R. Leach
Calibration of contact (stylus) instruments. International Organization for Standardization
ISO 1302 (2002) Geometrical product specification (GPS): indication of surface texture in
technical product documentation. International Organization of Standardization
ISO 13565 part 1 (1996) Geometrical product specification (GPS)—surface texture: profile
method—surfaces having stratified functional properties—filtering and general measurement
conditions. International Organization for Standardization
ISO 13565 part 2 (1998) Geometrical product specification (GPS—surface texture: profile
method—surfaces having stratified functional properties—height characterization using
material ratio curve. International Organization for Standardization
material probability curve. International Organization for Standardization
nominal characteristics of contact (stylus) instruments. International Organization of
Standardization
rules and procedures for the assessment of surface texture. International Organization of
Standardization
terms, definitions and surface texture parameters. International Organization of
Standardization
ISO 25178 part 6 (2010) Geometrical product specification (GPS)—surface texture: areal—part
6: classification of methods for measuring surface texture. International Organization for
Standardization
71: software measurement standards. International Organization for Standardization
ISO 25178 part 601 (2010) Geometrical product specification (GPS)—surface texture: areal—
part 601: nominal characteristics of contact (stylus) instruments. International Organization
for Standardization
ISO 25178 part 602 (2010) geometrical product specification (GPS)—surface texture: areal—part
602: nominal characteristics of non-contact (confocal chromatic probe) instruments.
International Organization for Standardization
700: calibration of non-contact instruments. International Organization for Standardization
part 701: calibration and measurement standards for contact (stylus) instruments. International
Organization for Standardization
method—measurement standards—material measures. International Organization of
Standardization
method—software measurement standards. International Organization of Standardization
ISO/CD 25178 part 1 (2011) Geometrical product specification (GPS)—surface texture: areal—
part 1: indication of surface texture. International Organization for Standardization
ISO/CD 25178 part 72 (2012) Geometrical product specification (GPS)—surface texture:
areal—part 72: XML soft gauge file format. International Organization for Standardization
ISO/DIS 25178 part 3 (2011) Geometrical product specification (GPS)—surface texture: areal—
part 3: specification operators. International Organization for Standardization
ISO/DIS 25178 part 70 (2012) Geometrical product specification (GPS)—surface texture: areal—
part 70: material measures. International Organization for Standardization
ISO/DIS 25178 part 603 (2011) Geometrical product specification (GPS)—surface texture:
areal—part 603: nominal characteristics of non-contact (phase shifting interferometric
microscopy) instruments. International Organization for Standardization
ISO/FDIS 25178 part 2 (2011) Geometrical product specification (GPS)—surface texture: areal—
part 2: terms, definitions and surface texture parameters. International Organization for
Standardization
ISO/FDIS 25178 part 604 (2012) Geometrical product specification (GPS)—surface texture:
areal—part 604: nominal characteristics of non-contact (coherence scanning interferometry)
instruments. International Organization for Standardization
ISO/FDIS 25178 part 605 (2012) Geometrical product specification (GPS)—surface texture:
areal—part 605: nominal characteristics of non-contact (point autofocusing) instruments.
ISO/WD 25178 part 606 (2011) Geometrical product specification (GPS)—surface texture:
areal—part 606: nominal characteristics of non-contact (point auto focusing) instruments.
areal—part 607: nominal characteristics of non-contact (focus variation) instruments.
areal—part 608: nominal characteristics of non-contact (imaging confocal microscope)
instruments. International Organization for Standardization
Jiang X, Scott PS, Whitehouse DJ, Blunt LA (2007) Paradigm shifts in surface metrology. Part II.
The current shift. Proc R Soc A 463:2071–2099
Jung L, Spranger B, Krüger-Sehm R, Krystek M (2004) Reference software for roughness
analysis: features and results. Proc XI Int Colloq Surfaces, Chemnitz, pp 164–170
Kettle J, Hoyle RT, Perks RM, Dimov S (2008) Overcoming material challenges for replication
of ‘‘Motheye lenses’’ using step and flash imprint lithography (S-FIL) for opto-electronic
applications. J Vac Sci Technol B 26:1794–1799
Leach RK (2009) Fundamental principles of engineering nanometrology. Elsevier, Amsterdam
Leach RK (2011) Optical measurement of surface topography. Springer, Berlin
Leach RK (2001) The measurement of surface texture using stylus instruments. NPL good
practice guide No 37. National Physical Laboratory
Leach RK, Boyd R, Burke T, Danzebrink H-U, Dirscherl K, Dziomba T, Gee M, Koenders L,
Morazzani V, Pidduck A, Roy D, Unger WES, Yacoot A (2011) The European
nanometrology landscape. Nanotechnology 22:062001
Mainsah E, Greenwood JA, Chetwynd DG (2010) Metrology and properties of engineering
surfaces. Kluwer Academic Publishers, Boston
Malacara D (2007) Optical shop testing, 3rd edn. Wiley Series in Pure and Applied Optics
Muralikrishnan B, Raja J (2008) Computational surface and roundness metrology. Springer,
London
Nosonovsky M, Bhushan B (2007) Multiscale friction mechanics and hierarchical surfaces in
nano- and bio-tribology. Mat Sci Eng 58:162–193
Petzing J, Coupland JM, Leach RK (2010) The measurement of rough surface topography using
coherence scanning interferometry. NPL good practice guide No 116. National Physical
Laboratory
Shalabi MM, Gortemaker A, Van’t Hof MA, Jansen JA, Creuger NHJ (2006) Implant surface
roughness and bone healing: a systematic review. J Dental Res 85:496–500
Stout KJ, Sullivan PJ, Dong WP, Mainsah E, Luo N, Mathia T, Zahouani H (1993) The
development of methods for the characterisation of roughness in three dimensions. Brussels,
Commission of the European Communities
Thomas TR (2008) Kenneth J Stout 1941–2006: a memorial. Wear 266:490–497
Wang H, Brandl DW, Nordlander P, Halas NJ (2007) Plasmonic nanostructures: artificial
molecules. Acc Chem Res 40:53–62
Whitehouse DJ (1982) The parameter rash—is there a cure? Wear 83:75–78
Whitehouse DJ (2010) Handbook of surface and nanometrology. CRC Press
Chapter 2
The Areal Field Parameters
François Blateyron
Abstract The vast majority of surface texture parameters are the field parameters.
The term field refers to the use of every data point measured in the evaluation area,
as opposed to feature parameters that only take into account specific points, lines
or areas. Field parameters allow the characterisation of surface heights, slopes,
complexity, wavelength content, etc. They are defined in the specification standard
ISO 25178 part 2. In this chapter the ISO areal field parameters will be presented
along with limited guidance on their use.
2.1 A Short History of Areal Parameters
The first areal surface texture measuring instruments were made available around
1987. Instrument manufacturers such as Zygo, Wyko and others started to provide
parameters calculated on the data. Early areal parameters were often simple
extrapolations of profile parameters (see Chap. 1 and Leach 2009 for a description
of the profile parameters) and were either named simply Ra or sRa, and sometimes
calculated using proprietary algorithms, leading to different parameter values on
different instruments.
The initial work by Stout et al. (1993a, b) leading to the ‘‘Blue Book’’ was
covered in Chap. 1. During the same period, ISO technical committee TC 57
introduced a new concept, called Geometrical Product Specification and Verifi-
cation (GPS for short) in order to unify specification standards dealing with
dimensional analysis and surface texture. Then, in 1996 a new committee was
created, TC 213, to develop GPS specification standards. One of the first actions of
TC 213 was to entrust a group of researchers with the aim of developing the basis
F. Blateyron (&)
Digital Surf sarl, 16 rue Lavoisier, 25000 Besançon, France
e-mail: fblateyron@digitalsurf.fr

16 F. Blateyron
of areal surface texture parameters. The historical perspective of the SURF-

STAND, which led to the so-called ‘‘Green Book’’ (Blunt and Jiang 2003) was
covered in Chap. 1.
The SURFSTAND results were presented to ISO in January 2002 and officially
transferred to TC 213, in order to start the standardisation process. In June 2002,
TC 213 voted for the creation of a new working group, which was assigned the
task of developing future international standards for areal surface texture. This
working group (known as WG 16) met for the first time in January 2003. At the
end of 2005, the ISO secretary allocated the number (ISO) 25178 to all areal
surface texture standards, thereby giving the standards their official birth. These
new standards were motivated by a shift that occurred in surface metrology: the
shift of surface measurement from profile to areal and contact to non-contact
(Jiang et al. 2007).
The work allocated to WG 16, which would be implemented in the ISO 25178
standards, contained two parts:
• to define the content of the areal surface texture standards, for specification and
verification; and
• to revise the existing profile standards to bring them into line with the new areal
standards.
The first part of ISO 25178, part 6 on classification of surface texture measuring
instruments, was published in January 2010 (ISO 25178 part 6 2010). This was
followed in June by a group of documents on instrument techniques: part 601 (ISO
25178 part 601 2010) and part 701 (ISO 25178 part 701 2010) on stylus profi-
lometer and part 602 (ISO 25178 part 602 2010) on confocal chromatic instruments.
The main document defining areal parameters, ISO 25178 part 2, was published
in September 2012 (ISO 25178 part 2 2012). As highlighted in Chap. 1, more
documents will follow in the forthcoming years.
2.2 Naming, Filtering and Calculation Conventions
2.2.1 Naming
ISO 25178 part 2 defines symbols for surface texture parameters that have a prefix
that is the capital letters S or V followed by one or several small letters that form
the suffix. This suffix should not be written as a subscript but on the same line as
the prefix. However, the final version of ISO 25178 part 2 uses subscripts in
parameter names to comply with ISO, which states that symbols should only
contain one letter and optional subscripts.
The prefix S is used for the majority of parameters (for example, Sq, Sdr, Smr),
the alternative being volume parameters that start with the letter V (for example,
Vmp, Vvc).
2 The Areal Field Parameters 17
2.2.2 Filtering Conditions
Profile parameters are named after the type of surface profile from which they are
calculated, for example R-parameters (Ra, Rsk, etc.) are calculated on the
roughness profile, W-parameters (Wa, Wsk, etc.) are calculated on the waviness
profile, while P-parameters are calculated on the primary profile (see ISO 4287
2000; Leach 2009). The type of surface is not taken into account with areal
parameters. The Sa parameter can be calculated on a primary surface or a filtered
surface—it will be called Sa in all cases. Therefore, it is important to provide the
filtering conditions together with the parameter value.
2.2.3 Sampling Area
Profile parameters are defined based either on a sampling length or the evaluation
length. If a parameter is defined on a sampling length, it is (by default) calculated
on each sampling length (ISO 4288 1996) and a mean value calculated (the default
number of sampling lengths is five). With surfaces and areal parameters, the
concepts of sampling and evaluation areas are still defined but the default is one
sampling area per evaluation area. This simply means that parameters are calcu-
lated on the measured surface without segmenting the surface into small sub-areas
that depend on the sampling length.
If the user wants to use more sampling areas in the evaluation of a surface, several
surfaces can be measured, either contiguous or separated, and a statistical evaluation
of the parameters calculated on each surface (mean, standard deviation, etc.).
2.2.4 Centred Heights
In surface texture parameter equations, the height function, z(x,y) must be centred.
This means that the mean height calculated on the definition area is already
subtracted from the heights. This leads to a simplified version of the parameter
equations as it is possible to express the equation as, for example
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ZZ
1
Sq ¼ z2 ðx; yÞdxdy; ð2:1Þ
A A
rather than
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ZZ
1
Sq ¼ ½zðx; yÞ z2 dxdy ð2:2Þ
A A
where z is the mean height of the surface on the definition area, A.

18 F. Blateyron
In practice, the above simplification can lead to rounding errors because heights
are usually coded with integer numbers and the average height is, therefore,
rounded to the closest integer. Each value in the sum is slightly biased, leading to
rounding errors in the result, especially when high order powers are involved in the
calculation, as is the case with Ssk (see Sect. 2.4.2.1) or Sku (see Sect. 2.4.2.2). A
correct digitisation of heights with a small quantisation step makes it possible to
reduce the uncertainty due to rounding errors. In the rest of this chapter, heights
z(x,y) are understood to be already centred.
2.3 Continuous Against Discrete Definitions
In modern specification standards, parameter definitions are always given for the
continuous case, i.e. expressed with integrals, although in practice measured
profiles and surfaces are always sampled and digitised. The use of continuous
definitions ensures the correctness of the definition and does not imply any
numerical approximation. National Metrology Institutes (NMIs) can implement
the most accurate versions of the parameter definitions without taking into account
the need for calculation speed (Harris et al. 2012a). However, software engineers
providing commercial products will implement an algorithm that has appropriate
speed and accuracy.
A simple discrete implementation uses summations. For example, in the case of
the parameter Sa (see Sect. 2.4.1.2), on a surface sampled with ny lines of nx
points, the equation will be approximated by
f 1 Xny 1 Xnx 1
Sa ¼ jzðx; yÞj: ð2:3Þ
n x ny y¼0 x¼0
This simple implementation provides sound results providing that the data density
is sufficiently high.
Figure 2.1 shows a classical representation of a profile where data points are
joined by line segments. In 3D, data points are connected through a triangular
facet. However, when parameters are calculated, the approximation method
replaces integrals by summations and, therefore, corresponds more to the repre-
sentation shown in Fig. 2.2.
Fig. 2.1 Sampled data points

represented as a series of line
segments
Fig. 2.2 Integration using a

simple summation of heights
Fig. 2.3 Integration using

linear interpolation
Fig. 2.4 Integration using a

spline interpolation
This simple summation of heights approximation can be used when the density
of points is high enough, that is to say, for most amplitude parameters where zero
crossing is not involved.
Linear interpolation gives a lower value for the Sa parameter as shown in
Fig. 2.3, since the area enclosed between the profile and the horizontal axis is
smaller around the zero crossings compared to the simple summation (Fig. 2.2).
Figure 2.4 shows a spline interpolation (Unser 1999), which is usually closer to
the continuous profile (Harris et al. 2012a). In this case, the value of the Sa
parameter will be more accurate and will lie somewhere in between the values
calculated in the cases shown in Figs. 2.2 and 2.3. The drawback of spline
interpolation is that it generates overshoots around peaks and tends to give
excessive values for peak-to-valley parameters. However, certain varieties of
splines do not have overshoot compared to simple cardinal splines (Catmull and
Rom 1974).
Software implementations of surface texture parameters can be tested inde-
pendently from instrument contributions by testing the calculation algorithms
against areal software measurement standards (Chap. 1; ISO 25178 part 71 2012;
ISO/CD 25178-72 2012; Harris et al. 2012b).
20 F. Blateyron
2.4 Height Parameters
The definitions of the height parameters are given in the following sections.
2.4.1 Mean Height of the Surface
2.4.1.1 Root Mean Square Height, Sq
The root mean square height or Sq parameter is defined as the root mean square
value of the surface departures, z(x,y), within the sampling area, A.
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ZZ
1
Sq ¼ zðx; yÞdxdy ð2:4Þ
A A
2.4.1.2 Arithmetic Mean Height, Sa
The arithmetic mean height or Sa parameter is defined as the arithmetic mean of

the absolute value of the height within a sampling area,
ZZ
1
Sa ¼ jzðx; yÞjdxdy: ð2:5Þ
A A
The Sa and Sq parameters are strongly correlated to each other (Blunt and Jiang
2003). The Sq parameter has more statistical significance (it is the standard devia-
tion) and often has a more physical grounding than Sa, for example, Sq is directly
related to surface energy and the way light is scattered from a surface (Leach 2009).
2.4.2 Skewness and Kurtosis
2.4.2.1 Skewness, Ssk
Skewness is the ratio of the mean of the height values cubed and the cube of Sq
within a sampling area,
ZZ
1 1
Ssk ¼ 3 z3 ðx; yÞdxdy: ð2:6Þ
Sq A A
This parameter can be positive, negative or zero, and is unit-less since it is nor-
malised by Sq. The Ssk parameter describes the shape of the topography height
distribution. For a surface with a random (or Gaussian) height distribution that has
symmetrical topography, the skewness is zero. The skewness is derived from the
amplitude distribution curve; it is the measure of the profile symmetry about the
mean line. This parameter cannot distinguish if the profile spikes are evenly dis-
tributed above or below the mean plane and is strongly influenced by isolated
peaks or isolated valleys. Skewness represents the degree of bias, either in the
upward or downward direction of an amplitude distribution curve. A symmetrical
profile gives an amplitude distribution curve that is symmetrical about the centre
line and an unsymmetrical profile results in a skewed curve. The direction of the
skew is dependent on whether the bulk of the material is above the mean line
(negative skew) or below the mean line (positive skew).
As an example, a porous, sintered or cast iron surface will have a large value of
skewness. A characteristic of a good bearing surface is that it should have a
negative skew, indicating the presence of comparatively few peaks that could wear
away quickly and relatively deep valleys to retain lubricant traces. A surface with a
positive skew is likely to have poor lubricant retention because of the lack of deep
valleys in which to retain lubricant traces. Surfaces with a positive skewness, such
as turned surfaces, have high spikes that protrude above the mean line. The Ssk
parameter correlates well with load carrying ability and porosity.
2.4.2.2 Kurtosis, Sku
The Sku parameter is a measure of the sharpness of the surface height distribution
and is the ratio of the mean of the fourth power of the height values and the fourth
power of Sq within the sampling area,
ZZ
1 1
Sku ¼ 4 z4 ðx; yÞdxdy: ð2:7Þ
Sq A A
Kurtosis is strictly positive and unit-less, and characterises the spread of the height
distribution. A surface with a Gaussian height distribution has a kurtosis value of
three. Unlike Ssk, use of this parameter not only detects whether the profile spikes
are evenly distributed but also provides a measure of the spikiness of the area. A
spiky surface will have a high kurtosis value and a bumpy surface will have a low
kurtosis value.
The Ssk and Sku parameters can be less mathematically stable than other
parameters since they use high order powers in their equations, leading to faster
error propagation.
2.4.3 Maximum Height of the Surface
The Sp parameter represents the maximum peak height, that is to say the height of
the highest point of the surface. The Sv parameter represents the maximum pit
height, i.e. the height of the lowest point of the surface. As heights are counted from
the mean plane and are signed, Sp is always positive and Sv is always negative.
22 F. Blateyron
The Sz parameter is the maximum height of the surface, i.e. is sum of the
absolute values of Sp and Sv,
Sz ¼ Sp þ jSvj ¼ Sp Sv: ð2:8Þ
The maximum height parameters are to be used with caution as they are sen-
sitive to isolated peaks and pits which may not be significant. However, Sz can be
pertinent on surfaces that have been filtered with a low-pass filter (S–F surfaces
with a large S nesting index or in other words, waviness surfaces—see Chap. 4) to
characterise the amplitude of waviness on the workpiece. Also, maximum height
parameters will succeed in finding unusual conditions such as a sharp spike or burr
on the surface that may be indicative of poor material or poor processing.
Alternative parameters that could be used as more robust versions of a maxi-
mum height parameter are the feature parameter S10z (see Chap. 3) and Sdc (see
Sect. 2.5.2.4).
2.5 Function Related Parameters
The definitions of the function related parameters are given in the following sections.
2.5.1 Height Distribution and Material Ratio Curve
The height distribution can be represented as a histogram of the surface heights that
quantifies the number of points on the surface that lie at a given height. The
material ratio curve is the cumulative curve of the distribution. The material ratio
curve is counted from the highest point on the surface (where the curve equals 0 %)
to its lowest point (where the curve reaches 100 %) (Fig. 2.5).
In the case of profiles, the material ratio is calculated using a cutting depth
c which is counted from the highest peak (ISO 4287 2000). This is not the most
0 10 20 30 40 50 60 70 80 90 100 %
5 5
4 4
3 3
2 2
1 1
0 0
-1 -1
-2 -2
-3 -3
-4 -4
-5 -5
0 0.5 1 1.5 2 2.5 %
µm µm
Fig. 2.5 The height distribution (left) and material ratio curve (right)
robust solution as it may be affected by outliers. In areal analysis, the value c is

counted on a surface from the mean plane, and this reference provides a more
robust definition for material ratio parameters.
2.5.2 Material Ratio Parameters
2.5.2.1 Areal Material Ratio, Smr
The areal material ratio is the ratio of the material at a specified height c to the
evaluation area expressed as a percentage (see Fig. 2.6). The heights are taken
from the reference plane. The Smr(c) function gives the material ratio p corre-
sponding to a cutting height c given as a parameter.
2.5.2.2 Inverse Areal Material Ratio, Smc
The Smc(p) function evaluates the height value c corresponding to a material ratio
p given as a parameter (see Fig. 2.7).
2.5.2.3 Peak Extreme Height, Sxp
The Sxp parameter (see Fig. 2.8) is aimed at characterising the upper part of the
surface, from the mean plane to the highest peak without taking into account a
small percentage of the highest peaks that may not be significant,
Sxp ¼ Smcð2:5%Þ Smcð50 %Þ: ð2:9Þ
Fig. 2.6 Smr(c) is the

material ratio 0 10 20 30 40 50 60 70 80 90 100 %
p corresponding to a section 5
height c 4
3
2
1
0
-1
-2
-3
-4
-5
µm
24 F. Blateyron
Fig. 2.7 Smc(p) is the height

section c corresponding to a 0 10 20 30 40 50 60 70 80 90 100 %
material ratio p 5
4
3
2
1
-1
-2
-3
-4
-5
µm
Fig. 2.8 Peak extreme 0 10 20 30 40 50 60 70 80 90 100 %

5
height Sxp defined as the
height difference between 4
two inverse material ratios at
3
2.5 and 50 %
2
-1
-2
-3
-4
-5
µm
The values 2.5 and 50 % are defined as the default values (ISO 25178-3 2011) and
can be set to other values depending on the application (they should be close to
these default values as this parameter is specifically defined for peak character-
isation). For more general height differences, the Sdc parameter should be used
(see Sect. 2.5.2.4).
2.5.2.4 Surface Section Difference, Sdc
This parameter is not strictly defined in ISO 25178 part 2 (2012) but it is a simple
extension of the Rdc parameter that was part of ISO 4287 (2000) (and see Leach
2009)
Sdc ¼ Smcð pÞ SmcðqÞ ð2:10Þ
where p and q are two material ratios that can be chosen freely depending on the
application.
The Sdc parameter can be used to give the maximum height of the surface when
the extreme peaks and valleys are removed or a threshold is applied, for example,
with p = 2 % and q = 98 %.
2.5.3 Characterisation of Stratified Surfaces
ISO 13565 part 2 (1998) is based on the German standard DIN 4777 (1990) that
was the first to introduce functional parameters based on a graphical construction
on the Abbott-Firestone curve (Whitehouse 2011). These parameters, Rk, Rvk,
Rpk, Mr1 and Mr2, are extracted from a filtered surface using a robust filter
specially designed for stratified surfaces. The parameters Sk, Spk, Svk, Sr1 and Sr2
are the areal equivalent of the parameters defined in ISO 13565 part 2 when the
Abbott-Firestone curve is built from an areal surface (see Fig. 2.9). The surface
may be filtered prior to the calculation of these parameters, preferably using a
robust Gaussian filter (ISO/CD 16610 part 71 2011; Muralikrishnan and Raja
2009, and see Chap. 4).
ISO 13565 part 3 (2000) defines three further parameters that are extracted:
Spq, Smq and Svq. These parameters are calculated on the probability curve and
are specifically designed for the evaluation of plateau honed surfaces (Malburg and
Raja 1993).
Fig. 2.9 Graphical

construction of Sk parameters
Spk
Sk
Svk
0 10 20 30 40 50 60 70 80 90 100 %
Smr1 Smr2
26 F. Blateyron
Fig. 2.10 Void volume Vv mr

below a section height 0 10 20 30 40 50 60 70 80 90 100 %
defined by a material ratio mr
Vv (mr)
2.5.4 Volume Parameters
2.5.4.1 Void Volume, Vv
The void volume or Vv(mr) parameter is the void volume calculated for a material
ratio mr. This parameter is calculated by integrating the volume enclosed above
the surface and below a horizontal cutting plane set at a height h = Smc(mr). This
can be expressed by the following
Z 100%
VvðmrÞ ¼ k ½Smcðmr Þ SmcðqÞdq ð2:11Þ
mr
where k is a factor to convert the volume into the required unit, either [lm3], [lm3/
mm2] or [ml/m2]. Void volume can be represented on the Abbott-Firestone curve
as shown in Fig. 2.10.
For mr = 100 %, the void volume is zero. For mr = 0 %, the void volume is a
maximum (the cutting plane below the lowest point).
Void volume calculations are often useful to evaluate the surface texture of
mechanical components that are used in contact with other surfaces.
2.5.4.2 Material Volume, Vm
The material volume or Vm(mr) parameter is the material volume calculated for a
material ratio mr. The parameter is calculated by integrating the volume enclosed
Fig. 2.11 Material volume mr

above the section height 0 10 20 30 40 50 60 70 80 90 100 %
defined by a material ratio mr
Vm (mr)
below the surface and above a horizontal cutting plane set at a height h = Sdc(mr).
This parameter can be expressed by the following
Z mr
Vmðmr Þ ¼ k ½SmcðqÞ SmcðmrÞdq: ð2:12Þ
0%
The material volume can be represented on the Abbott-Firestone curve as shown in

Fig. 2.11.
For mr = 100 %, the void volume is a maximum. For mr = 0 %, the void
volume is zero (the cutting plane above the highest point).
2.5.4.3 Peak Material Volume, Vmp
The peak material volume or Vmp parameter is the material volume calculated at a
fixed material ratio mr,
Vmp ¼ Vmðmr1Þ ð2:13Þ
where mr1 = 10 % by default. The ratio mr1 may be changed for specific
applications and will always be specified together with the value of Vmp.
The Vmp parameter can be used for the same purpose as the Spk parameter, i.e.
to characterise the volume of material which is likely to be removed during run-
ning-in of a component.
28 F. Blateyron
2.5.4.4 Core Material Volume, Vmc
The core material volume or Vmc parameter is the difference between two material
volume values calculated at different heights
Vmc ¼ Vmðmr2Þ Vmðmr1Þ ð2:14Þ
where mr2 = 80 % and mr1 = 10 % by default.

The Vmc parameter represents the part of the surface material which does not
interact with another surface in contact, and which does not play any role in
lubrication.
2.5.4.5 Core Void Volume, Vvc
The core void volume (the difference in void volume between the mr1 and mr2
material ratios) is given by
Vvc ¼ Vvðmr1Þ Vvðmr2Þ ð2:15Þ
where mr2 = 80 % and mr1 = 10 % by default.
2.5.4.6 Dales Void Volume, Vvv
The dale volume at mr2 material ratio is given by

Vvv ¼ Vvðmr2Þ ð2:16Þ
where mr2 = 80 % by default.
2.5.4.7 Examples of Volume Parameters
Figure 2.12 shows the four volume parameters Vmp, Vmc, Vvc and Vvv calculated
from two bearing ratio levels mr1 and mr2.
Volume parameters have shown good correlation with functional requirements in
several applications; see Waterworth (2006) for a thorough treatment. Volume
parameters have replaced the functional indices Sbi, Sci and Svi (Stout et al. 1993a, b)
as they have proved to be more stable while providing the same type of information
(see Sect. 2.9.1 and Jiang et al. 2000).
The Vvv parameter characterises the volume of fluid retention in the deepest
valleys of the surface. This parameter is not affected by wear processes applied on
the surface.
The Vmp parameter characterises the volume of material located on the highest
peaks of the surface which is removed during a wear process. On a used
mechanical component, after several hours of function, the highest peaks are cut
Fig. 2.12 Definition of mr1 mr2

volume parameters on the 0 10 20 30 40 50 60 70 80 90 100 %
bearing areal ratio curve
Vmp
Vvc
Vmc
Vvv
out or plastically deformed, and the corresponding particles of material are cap-
tured by the deepest valleys, so that the behaviour of the surface is more likely
described by Vmc and Vvc.
2.6 Hybrid Parameters
The hybrid parameters are defined in the following sections.
2.6.1 Root Mean Square Gradient, Sdq
The gradient of a surface point is defined for each axis x and y by oz=ox and oz=oy:
The implementation of these gradients on a sampled surface is given elsewhere
(Whitehouse 2011). The root mean square gradient is then calculated on the whole
surface with
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi
ZZ 2
1 oz oz 2
Sdq ¼ þ dxdy ð2:17Þ
A ox oy
where A is the projected area of the surface. The Sdq parameter has a unit-less
positive value. Optionally, it can be expressed in [lm/lm] or [lm/mm], or even as
an angle by calculating the arctangent of Eq. (2.17). The Sdq parameter is useful
30 F. Blateyron
Fig. 2.13 Orientation b and α

inclination a of a surface
facet
for assessing surfaces in sealing applications and for controlling surface cosmetic
appearance.
The Sdq parameter is also associated with two plots that represent the distri-
bution of the horizontal and vertical angles a and b (and see Fig. 2.13)
0sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi1
oz 2 oz 2 A
a ¼ tan1 @ þ ; ð2:18Þ
ox oy
oz2
!
1 oy
b ¼ tan : ð2:19Þ
oz2
ox
The angle a characterises the steepest gradient in the vertical plane, and is given
as an angle between 0 and 90, 0 being a horizontal facet and 90 a vertical facet.
The angle b, when calculated on the whole surface, characterises the mean
orientation of the surface facets and is an evaluation of the texture direction. It is
given as an angle between 0 and 360, with 0 in the direction of the x axis,
counter clockwise.
2.6.2 Developed Interfacial Area Ratio, Sdr
The developed interfacial area of a surface is calculated by summing the local area
when following the surface curvature. It can be approximated by the mean area of
two triangles formed between four adjacent points.
Referring to Fig. 2.14, the area of a triangle is half the cross product of two
vectors,
P11: z(x+1,y+1) P11: z(x+1,y+1)
P01 : z(x,y+1) P01: z(x,y+1)
P00: z(x,y)
P10: z(x+1,y) P00: z(x,y) P10: z(x+1,y)
Fig. 2.14 Area between four adjacent points calculated by the average of two triangulations
2 ! ! ! ! 3

1
P00 P01 P00 P10 þ P11 P01 P11 P10
162 7
A00 ¼ 4 þ 5: ð2:20Þ
2 1 ! ! ! !
2 P10 P00 P10 P11 þ P01 P00 P01 P11
As most surfaces are globally flat (the topography is seen only by expanding the
z axis), the developed area is usually slightly larger than the projected area—this is
why the Sdr parameter is expressed as the excess value above 100 %, thus
PP
Aij A
Sdr ¼ ; ð2:21Þ
A
where Aij is the mean area calculated at a point, and A is the projected area
calculated by the product of the lengths in x and y. Note that ISO 25178 part 2
(2012) defines the Sdr parameter for the continuous case, i.e. with integrals instead
of summations.
The Sdr parameter can be given as a unit-less positive number or as a per-
centage. It will usually produce a value of several percent (typically between 0 and
10 %). A perfectly flat and smooth surface would have Sdr = 0 %.
The Sdr parameter is used as a measure of the surface complexity, especially in
comparisons between several stages of processing on a surface, and it can provide
useful correlations in adhesion applications (Löberg et al. 2010; Barányi et al.
2011; Reizer and Pawlus 2011). The Sdr parameter is greatly influenced by the
sampling scheme (number of points and spacing in the x and y axes).
2.7 Spatial Parameters
The spatial parameters are defined in the following sections.

32 F. Blateyron
2.7.1 Autocorrelation Function
The autocorrelation function (ACF) evaluates the correlation of a part of an image

with respect to the whole image. The ACF is defined as a convolution of the
surface with itself, shifted by (sx, sy)
RR
zðx; yÞz x sx ; y sy dxdy

ACFðsx ; sy Þ ¼ RR : ð2:22Þ
zðx; yÞ2 dxdy
Fig. 2.15 Surface with PCB vias (left) and its autocorrelation plot (right)
Fig. 2.16 Abrasive surface with quartz grains (left) and its autocorrelation plot (right)
Fig. 2.17 Surface of a DVD stamper (left) and its autocorrelation plot (right)
Fig. 2.18 Using autocorrelation for detecting surface patterns. When the image of the circuit
(left) is correlated with this pattern, it creates an image with correlation peaks at positions where
the pattern is found in the image (right)
The ACF corresponds to the autocovariance normalised by Sq2 [the denominator

in Eq. (2.22)]. The ACF produces a value between -1 and +1 for each point on the
surface. An ACF of +1 means a perfect correlation and zero means no correlation.
34 F. Blateyron
The maximum of the ACF is always at the centre (for a zero shift). Figures 2.15,
2.16, 2.17 show several examples of surface textures and their autocorrelation
plots.
The ACF is used to study periodicities on a surface, i.e. when a texture motif is
reproduced several times on the surface (see Fig. 2.18), or is used to assess the
isotropy of a surface (see also Sect. 2.7.3).
2.7.2 Autocorrelation Length, Sal
The autocorrelation length, Sal, is defined as the horizontal distance of the ACF(tx, ty)
which has the fastest decay to a specified value s, with 0 B s \ 1. The Sal parameter
is given by
Fig. 2.19 Autocorrelation peak with an applied threshold of 0.2 (white part above the threshold)
Fig. 2.20 Shortest radius

measured from the centre to
the contour of the thresholded
lobe on the autocorrelation
plot
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Sal ¼ min tx2 þ ty2 : ð2:23Þ
Figure 2.19 shows the autocorrelation of a textured surface. The white part on the
central lobe is above the threshold s. A radius is calculated from the centre to the
perimeter of the lobe and the shortest radius is kept for Sal (see Fig. 2.20).
For all practical applications involving relatively smooth surfaces, the value for
s can be taken as 0.2 (ISO 25178 part 3 2012), although other values can be used
and will be subject to forthcoming areal specification standards. For an anisotropic
surface, Sal is in the direction perpendicular to the surface lay. A large value of Sal
denotes that that surface is dominated by low spatial frequency components, while
a small value for Sal denotes the opposite case.
The Sal parameter is a quantitative measure of the distance along the surface by
which a texture that is statistically different from that at the original location would
be found.
The contour of the central lobe is measured from the centre and the shortest
radius is identified. This radius gives the value of the Sal parameter.
2.7.3 Texture Aspect Ratio, Str
The texture aspect ratio parameter, Str is one of the most important parameters
when characterising a surface in an areal manner as it characterises the isotropy of
the surface.
The Sal parameter is calculated from the minimum radius on the central lobe of
the ACF. The Str parameter is calculated from the minimum, rmin, and maximum
radii, rmax (see Fig. 2.21), found under the same conditions, on the autocorrelation
plot after applying a threshold of 0.2
Fig. 2.21 Minimum and

maximum radii measured
on the central lobe of the
autocorrelation plot r max
r min
36 F. Blateyron
rmin
Str ¼ ð2:24Þ
rmax
The Str parameter is unit-less and its values lies between 0 and 1. It can also be
expressed as a percentage between 0 and 100 %.
The Str parameter is an evaluation of the surface texture isotropy. If Str is close
to unity, then the surface is isotropic, i.e. it has the same properties regardless of
the direction. On an isotropic surface, it is possible to assess the surface texture
using a 2D (profile) surface texture measuring instrument. If Str is close to 0, then
the surface is anisotropic, i.e. it has a dominant texture direction. In this case, the
parameter Std will give the direction angle of the texture.
Fig. 2.22 Surface of artificial leather (left) and its Fourier spectrum (right)
Fig. 2.23 Radial integration

of frequency amplitudes of
the Fourier spectrum
Fig. 2.24 Polar spectrum graph representing the texture directions
2.7.4 Texture Direction, Std
The texture direction parameter, Std, is assessed from the Fourier spectrum of the
surface. The Fourier spectrum gives the energy content of each spatial frequency
on the surface and is usually represented as a plot where amplitudes are coded with
a colour or grey level (see Fig. 2.22).
When moving from the centre to an edge of the spectrum in a given direction,
the spatial frequencies go from the lowest to the highest value. The frequency at
the centre corresponds to the continuous value in z (frequency of zero or infinite
wavelength). When the surface is centred, this offset is zero. Frequency amplitudes
along the radius at a given direction h can be integrated between two selected
spatial frequencies, fmin and fmax, in order to calculate a value A(h) that represents
the spatial frequency content in that direction (see Fig. 2.23).
By repeating this integration for all angles between 0 and 180, a polar spectrum
is obtained that can be represented with a semi-circular graph (see Fig. 2.24).
The maximum value of the graph shown by Fig. 2.24 is called the main texture
direction, or Std. The Std parameter is given in degrees between 0 and 180, and
should be considered as insignificant if the isotropy factor Str is below 0.6 and 0.8
(depending on the application). The definition of Std in ISO 25178 part 2 specifies
that the angle can be given from a reference angle s.
The Std parameter is a convenient parameter on surfaces showing scratches and
oriented texture (Schulz et al. 2010; McGarigal et al. 2009).
38 F. Blateyron
2.8 Areal Parameters from ASME B46.1
The US specification standard ASME B46.1 introduced areal parameters in its

1995 edition. The latest 2009 edition contains a set of areal parameters that is
similar to that in ISO 25178 part 2 (2012). ASME B46.1 (2010) defines the
following parameters identically to ISO 25178 part 2: Sa, Sq, Sp, Sv, Ssk, Sku, Sdq,
Str and Std. However, one parameter in ASME B46.1 is named differently: St
corresponding to Sz in the ISO standard.
Two other parameters, not defined in the ISO standard, are specific to the
ASME B46.1 standard:
SWt: peak to valley height of the waviness surface. This parameter can be emu-
lated with an ISO parameter by calculating Sz on an S-L surface with the L nesting
index set at the same value as kc.
Sdq(h): directional root mean square slope. This parameter corresponds to Pdq (as
per ISO 4287 2000 and see Leach 2009) calculated on a profile extracted along the
h direction.
The 2002 edition of ASME B46.1 also defined SDa(h), which it is not in the
2009 edition.
2.9 Areal Parameters from Earlier Reference Documents
2.9.1 European Project Report EUR 15178 EN (1993)
The European project report EUR 15178 EN (Stout et al. 1993a, b) established a
list of areal parameters grouped into several families: amplitude parameters,
spatial parameters, hybrid parameters and functional parameters (see Table 2.1).
The parameters in Table 2.1 are still widely used in some industries and in
scientific publications, although updated and improved parameters have been
available for a long time (De Chiffre et al. 2000).
2.9.2 Basis for 3D Surface Texture Standards

‘‘SURFSTAND’’
The aim of the European project SURFSTAND was to improve areal parameters
defined in EUR 15178 EN (Stout et al. 1993a, b), investigate their correlation with
surface function, and to prepare the basis of an international specification standard
(now ISO 25178 part 2 2012). The parameter set developed in SURFSTAND is,
therefore, based on the ISO parameter set but with several modifications given
below.
Table 2.1 Parameters from EUR 15178 EN (Stout et al. 1993a, b)
Symbol Parameter name Compatibility Family B’14*
Sa Arithmetic mean deviation Equivalent (see note 1) Amplitude
Sq Root-mean-square Equivalent (see note 1) Amplitude *
deviation
Sz Ten point height Different (see note 2) Amplitude *
Ssk Skewness of topography Equivalent (see note 1) Amplitude *
height distribution
Sku Kurtosis of topography Equivalent (see note 1) Amplitude *
height distribution
2 The Areal Field Parameters
Sds Density of summits Different (see note 3) Spatial *

Str Texture aspect ratio Equivalent Spatial *
Std Texture direction Equivalent (see note 4) Spatial *
Sal Fastest decay Equivalent Spatial *
autocorrelation length
SDq Root-mean-square slope Equivalent (see note 1) Hybrid *
Ssc Arithmetic mean summit Different (see note 5) Hybrid *
curvature
Sdr Developed interfacial ratio Equivalent (see note 1) Hybrid *
Stp Surface bearing ratio Renamed (see note 6) Area & Volume
Smr Material volume ratio Renamed (see note 7) Area & Volume
Svr Void volume ratio Renamed (see note 7) Area & Volume
Sbi Surface bearing index Specific Functional *
(continued)
39
40
Table 2.1 (continued)

Symbol Parameter name Compatibility Family B’14*
Sci Core fluid retention index Specific Functional *
Svi Valley fluid retention index Specific Functional *
Sk … Functional parameters from Equivalent (see note 8) Functional
DIN 4777
Sk, Spk, Svk, Sr1, Sr2
Note 1 Parameter equations are all given for the discrete case, but their definitions are compatible with those of ISO 25178 part 2 (2012)
Note 2 This parameter is defined here from the five highest peaks and the five deepest valleys
P5 P5
zpi þ i¼1 jzvi j
Sz ¼ i¼1 : ð2:25Þ
5
In order to discriminate significant peaks and valleys, only one peak and one valley should be found per autocorrelation area, with side length equal to twice
the fastest autocorrelation decay Sal
Note 3 Sds corresponds to Spd in ISO 25178 part 2 but the discrimination method is different
Note 4 Std is defined in EUR 15178EN with the origin 0 on the y axis while in ISO 25178 part 2 it is defined with an origin s that can be set at any angle
Note 5 Ssc corresponds to Spc in ISO 25178 part 2 but the discrimination method is different
Note 6 Stp corresponds to Smr in ISO 25178 part 2
Note 7 Smr corresponds to Vm in ISO 25178 part 2 (should not be confused with the bearing ratio parameter that has the same name); Svr corresponds to Vv
in ISO 25178 part 2
Note 8 These parameters are extensions in 3D of the parameters defined in ISO 13565 part 2 (1996). The Sr1 and Sr2 parameters have been renamed Smr1
and Smr2 in ISO 25178 part 2. The standard also includes Spq, Svq and Smq from ISO 13565 part 3 (1996)
* These parameters are part of the so-called ‘‘Birmingham 14 parameters’’
F. Blateyron
• Sp and Sv are now introduced as maximum surface peak height and maximum
surface valley depth.
• Sz is defined as the maximum height i.e. the sum of the absolute values of Sp and
Sv. The old ten-point height parameter Sz of the previous report is here renamed
as S10z.
• Ssc, Sds and S5z, which are parameters related to peaks, are now calculated from
peaks detected after a segmentation and Wolf pruning of 5 % of Sz (see Chap. 3).
The SURFSTAND report (published as a book, Blunt and Jiang 2003) also
introduces several new parameters given below.
• Sfd is the fractal dimension calculated from the volume-scale plot where the
volume is calculated between two morphological envelopes (see Chap. 6).
• Vmp and Vmc are introduced as material volume (see ISO 25178 part 2 2012),
respectively peak material volume and core material volume.
• Vvc and Vvv are introduced as void volume (see ISO 25178 part 2 2012),
respectively core void volume and valley void volume.
It is interesting to note that in the SURFSTAND report, the Sa parameter was
removed from the parameter list although it was part of the earlier work. The
authors of the study wanted to avoid encouraging people to use Sa in the same way
they use Ra on profiles, without really knowing if the parameter is the best cor-
related parameter for their needs. However, during the preparation of the ISO
25178 standard, some experts of WG 16 lobbied strongly in order to reintegrate Sa,
and it was finally added to the draft.
References
ANSI/ASME B46.1 (2010) Surface texture, surface roughness, waviness and lay. American
National Standards Institute
Barányi I, Czifra A, Kalácska G (2011) Height-independent topographic parameters of worn
surfaces. Sustain Constr Des 2(1): 35–40
Blunt LA, Jiang X (2003) Advanced techniques for assessment surface topography. Kogan Page
Science, London
Catmull E, Rom R (1974) A class of local interpolating splines. Compt Aided Geom Des
317–326 (Academic Press)
De Chiffre L, Lonardo P, Trumpold H, Lucca DA, Goch G, Brown CA, Raja J, Hansen HN
(2000) Quantitative characterisation of surface texture. Ann CIRP 49:635–652
DIN 4777 (1990) Metrology of surfaces; Profile filters for electrical contact stylus instruments;
Phase-corrected filters
Harris PM, Smith IM, Leach RK, Giusca CL, Jiang X, Scott PM (2012a) Software measurement
standards for areal surface texture parameters: Part 1—Algorithms. Meas Sci Technol
23:105008
Harris PM, Smith IM, Wang C, Giusca CL, Leach RK (2012b) Software measurement standards
for areal surface texture parameters: Part 2—Comparison of software. Meas Sci Technol
23:105009
42 F. Blateyron
ISO 13565 part 2 (1998) Geometrical product specification (GPS)—surface texture: Profile
material ratio curve. International Organization for Standardization
ISO 13565 part 3 (2000) Geometrical product specification (GPS)—surface texture: Profile
material probability curve. International Organization for Standardization
part 2: Terms, definitions and surface texture parameters. International Organization for
Standardization
part 3: Specification operators. International Organization for Standardization
part 6: Classification of methods for measuring surface texture. International Organization for
Standardization
part 601: Nominal characteristics of contact (stylus) instruments. International Organization
for Standardization
ISO 25178 part 602 (2010) Geometrical product specification (GPS)—Surface texture: areal—
part 602: nominal characteristics of non-contact (confocal chromatic probe) instruments.
part 701: calibration and measurement standards for contact (stylus) instruments. International
Organization for Standardization
part 71: Software measurement standards. International Organization for Standardization
ISO 4287 (2000) Geometrical product specification (GPS)—Surface texture: profile method—
terms, definitions and surface texture parameters. International Organization of
Standardization
ISO 4288 (1996) Geometrical product specifications (GPS)—surface texture: Profile method—
rules and procedures for the assessment of surface texture. International Organization of
Standardization
ISO/CD 16610 part 71 (2011) Geometrical product specifications (GPS)—robust areal filters:
Gaussian regression filters. International Organisation for Standardization
ISO/CD 25178-72 (2012) Geometrical product specification (GPS)—surface texture: areal—part
72: XML file format. International Organization for Standardization
Jiang XQ, Blunt L, Stout KJ (2000) Comparison study of areal functional parameters for rough
surfaces. Proc ASPE, Nashville
Jiang X, Scott PJ, Whitehouse DJ, Blunt L (2007) Paradigm shifts in surface metrology. Proc R
Soc A 463:2049–2099
Löberg J, Mattisson I, Hansson S, Ahlberg E (2010) Characterisation of titanium dental implants
I: critical assessment of surface roughness parameters. Open Biomater J 2:18–35
Malburg MC, Raja J (1993) Characterization of surface texture generated by plateau honing
process. Ann CIRP 42:637–640
McGarigal K, Tagil S, Cushman A (2009) Surface metrics: an alternative to patch metrics for the
quantification of landscape structure. Landscape Ecol 24:433–450
Muralikrishnan B, Raja J (2009) Computational surface and roundness metrology. Springer,
Berlin
Reizer R, Pawlus P (2011) 3D surface topography of cylinder liner forecasting during plateau
honing process. In: Proceedings 13th international conference metrology and properties of
engineering surfaces, Twickenham, April, pp 29–34
Schulz E, Calandra I, Kaiser TM (2010) Applying tribology to teeth of hoofed mammals.
Scanning 32:162–182
Stout KJ, Sullivan PJ, Dong WP, Mainsah E, Luo N, Mathia T, Zahouani H (1993a) The
development of methods for the characterization of roughness in three dimensions. Paton
Press, Chico
Stout KJ, Sullivan PJ, Dong WP, Mainsah E, Luo N, Mathia T, Zahouani H (1993b) The
development of methods for the characterisation of roughness in three dimensions.
Commission of the European Communities, Brussels
Unser M (1999) Splines: a perfect fit for signal and image processing. IEEE Signal Proc Mag
16:22–38
Waterworth A (2006) Quantitative characterization of surface finishes on stainless steel sheet
using 3D surface topography analysis. PhD Thesis, University of Huddersfield
Whitehouse DJ (2011) Handbook of surface and nanometrology, 2nd edn. CRC Press, Boca
Raton
Chapter 3
The Areal Feature Parameters
François Blateyron
Abstract Unlike field parameters, where every point on the surface is taken into
account in the calculation, the category of feature parameters takes into account only
identified features on the surface. Features are considered because they play a sig-
nificant role in a particular function (for example, large peaks and hills play a
functional role as contact zones in mechanical engineering applications; while
points located in valleys do not). Significant features are identified by segmentation
of the surface and selected by a discrimination method known as pruning. Param-
eters are then calculated to quantify the characteristics of the selected features.
3.1 Definitions of Topological Features
3.1.1 Introduction
Topological features have been studied for a long time in the field of geography
and cartography (Geographic Information Systems). One of the first authors to
describe topological features using mathematical definitions was Maxwell (1870)
who extended earlier work of Cayley (1859). A long list of contributors followed
Maxwell and in the 1980s the mathematical methods that had been developed were
adapted for use with computers.
The development of algorithms for the detection of topological features goes
back to the 1970s. Algorithms were published by Peucker and Douglas (1975) and
Warntz (1975); algorithms for building relation graphs connecting the topological
features were proposed by Pfaltz (1976) and further developed by Wolf (1984,
1993). A comprehensive history and summary of the techniques used to analyse
F. Blateyron (&)
Digital Surf sarl, 16 rue Lavoisier, 25000 Besançon, France
e-mail: fblateyron@digitalsurf.fr

DOI: 10.1007/978-3-642-36458-7_3, Ó Springer-Verlag Berlin Heidelberg 2013
46 F. Blateyron
surface networks and extract critical points can be found elsewhere (Rana and
Morley 2002).
In the 1990s, efficient algorithms for watershed segmentation were developed
by Beucher (1990), Vincent (1990), Vincent and Soille (1991). Scott formalised
the mathematical concepts for use in the field of surface analysis and described the
relationship that exists between segmentation and morphological filtering (Scott
1992).
Shortly afterwards, the first applications with industrial surfaces were devel-
oped and tested (Zahouani 1993, 1998; Barré 1997, 2001; Scott 1998, 2004).
Finally, watershed segmentation and a set of feature parameters were included in
the SURFSTAND project (Blunt and Jiang 2003, Chap. 1) and later in the inter-
national standard ISO 25178 part 2 (2012). Segmentation is now considered to be a
filter and is included in a separate standard, ISO/DIS 16610 part 85 (2012),
whereas feature parameters remain in ISO 25178 part 2.
The key concepts used to describe feature parameters based upon watershed
segmentation of a surface are: contour line, peak, pit, hill, dale, course line, ridge
line and saddle point, which are all described below.
3.1.2 Contour Lines
Points on digital elevation maps used in geophysics, as well as points on micro-

scale surface maps, represent heights measured from a reference surface. By
convention, the reference surface is represented by a horizontal plane which
defines the zero level (for example in geophysics, the reference surface for digital
elevation models is a geoid or an ellipsoid).
Fig. 3.1 Contour lines from the intersection of the surface with planes at different heights
3 The Areal Feature Parameters 47
When the horizontal plane is shifted to a given level above or below the zero
level (see Fig. 3.1), it intersects the surface in a series of closed lines that represent
the sets of points that have that given level. Shifting the plane to evenly spaced
height levels generates a series of contour lines (see Fig. 3.2).
Contour lines are a convenient way to represent lines of equal height on a black
and white map of a topographic surface. They are used to define critical points and
other concepts.
Fig. 3.2 Surface topography with ten contour lines (left) and twenty contour lines (right)
3.1.3 Hills
A surface point higher than its surrounding area is called a peak. The peak
neighbourhood is called a hill (see Fig. 3.3). From any point on a hill there is an
upward path that ends at a unique peak. If the upward path from a point ends at
another peak, then this point belongs to another hill. All points belonging to a hill
are enclosed by a course line.
Fig. 3.3 Representation of a

hill (B), its highest point, the
peak (A) and its surrounding
course line (C)
48 F. Blateyron
3.1.4 Dales
A surface point that is lower than its surrounding area is called a pit. The pit
neighbourhood is called a dale (see Fig. 3.4). From any point on a dale there is
a downward path that ends at a unique pit and the dale defines a catchment
basin: a drop of water starting from any point in the basin (dale) will run down
to the pit. If the downward path from a point ends at another pit, it means that
this point belongs to another dale. All points belonging to a dale are enclosed by
a ridge line.
Course lines are lines representing the natural downward flow of water dropped
on a side of a hill (see Fig. 3.5). Water flows down to the minimum point––the pit.
Fig. 3.4 Representation of a

dale (B), its lowest point, the
pit (A) and its surrounding
ridge line (C)
Fig. 3.5 A beck winding

across a meadow in the Alps
(courtesy of the author). This
is a natural version of a
course line
3.1.5 Saddle Points
Saddle points are at the intersection point of ridge lines and course lines. They
correspond to a maximum on a course line, and a minimum on a ridge line (see
Fig. 3.6).
There are several saddle points around a dale or around a ridge. The lowest
saddle of a dale has special significance: it corresponds to the point where the
water flows into an adjacent dale. This property creates a special connectivity with
the adjacent dale, which itself may have another adjacent dale where the water will
flow, and so on.
Fig. 3.6 Left representation of a saddle point (C) which is at the intersection of a ridge line
(A) and a course line (B). Right saddle in the Alps (courtesy of the author)
3.1.6 Motifs
The term ‘‘motif’’ was forged during the 1980s in the French automotive industry
when the Comité de Normalisation des Moyens de Production (CNOMO)
commission started to design a new method for characterising mechanical com-
ponents depending on their function (for example, sliding, sealing and friction).
This motif method was based on a graphical segmentation of the profile with
respect to motif shapes (see Fig. 3.7) that were defined as peak-valley-peak shapes.
The motif method was used in France for two decades (CNOMO E00.14.015.N
1983; Boulanger 1992), and was later adopted as an international standard under
the reference ISO 12085 (1996).
The term ‘‘motif’’ has been used to designate a texture cell on a surface, and can
be used to replace either a ‘‘dale’’ or a ‘‘hill’’. Strictly speaking, at least in
accordance with the original definition, a motif is closer to a dale since the peak-
valley-peak in ISO 12085 is replaced by a pit (valley) surrounded by a ridge line.
50 F. Blateyron
µm
1
0.75
0.5
0.25
0
-0.25
-0.5
-0.75
-1
-1.25
-1.5
-1.75
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 mm
Fig. 3.7 Roughness motifs according to ISO 12085 (1996)
Initially created for images and digital elevation models (see Fig. 3.8), then
adapted to areal surface topography, the segmentation method can also be applied
to profiles to identify motifs and calculate feature parameters. Feature parameters
calculated in this way are more stable and robust than the ISO 12085 parameters,
hence 2D watershed segmentation is a good candidate for replacing the methods in
ISO 12085 (Blateyron and Adam 2004).
Fig. 3.8 Motifs detected on

a surface that is an imprint of
human skin. Course lines are
shown in white and peaks
with black ‘‘plus’’ signs
3.2 Watershed Segmentation
The process of watershed segmentation starts from the lowest point of the surface.
Virtual water is poured over the whole surface from above, so that the dale around
the point is flooded progressively. At any time during the pouring process the
water occupies an area surrounded by a contour line. As the water level increases it
moves upwards to higher contour lines. Similarly, dales all over the surface are
flooded when their contour lines are at or below the flooding level.
When two adjacent dales are flooded to a level at which their contour lines
come into contact, a virtual wall is recorded at the contact points. The virtual walls
that are recorded this way are the ridge lines surrounding dales. When the flooding
level reaches the highest point on the surface, the watershed segmentation is
complete and all dales are segmented by ridge lines.
Without further processing, segmentation can lead to thousands of small and
insignificant dales; this is called over-segmentation. In this case, an additional step
is needed in the algorithm in order to merge small dales into larger ones and only
retain significant dales at the end (see Sect.. 3.3.2). Hills on a surface are seg-
mented by inverting the surface and applying the dales watershed segmentation
algorithm.
3.3 The Change Tree and Pruning
3.3.1 Definition of a Motif Height
Motif height plays a very important role in the process of selecting significant motifs
(see Sect. 3.3.2). The motif height for a dale is defined as the height difference
between the pit and the lowest saddle point on the surrounding ridge line (see
Fig. 3.9). Symmetrically, the motif height for a hill is defined as the height difference
between the highest saddle point on the surrounding course line and the peak.
A dale usually has several saddle points along its surrounding ridge line (see
Fig. 3.10). There is a special relationship between the pit and the lowest of these
saddle points (and symmetrically, for a hill, between its peak and the highest
saddle point on its course line). When virtual water overflows out of the dale, the
water will escape into an adjacent dale at the lowest saddle.
Fig. 3.9 The motif height hm is defined between the pit (1) and the lowest saddle point (2) which
corresponds to the point where the dale overflows when filled up with virtual water (3)
52 F. Blateyron
Fig. 3.10 Dale with three

saddle points on its
surrounding ridge line
3.3.2 Change Tree and Wolf Pruning
Peaks, saddles and pits are critical points connected by ridge lines and course lines.
Connections between them can be represented in a relation graph (see Fig. 3.11).
Over the past years several authors have proposed different data structures in
order to represent the topological relationships between the critical points (Morse
1969; Pfaltz 1976; Wolf 1984; Kweon and Kanade 1994; Takahashi et al. 1995;
Scott 1998). One of the most interesting of these data structures is the change tree
which makes it possible to reduce the number of motifs resulting from over-
segmentation. Figure 3.12 shows the profile of a topographic surface in combi-
nation with the respective change tree.
Fig. 3.11 Relationships

between peaks (P1–P5),
saddles (S1–S9) and pits (V1–
V7)
Fig. 3.12 Profile of a

topographic surface with its
respective change tree
Each peak is connected to its closest (highest adjacent) saddle and characterised
by the hill height (hm). A change tree represents all motifs on a surface (see
Fig. 3.13). By applying a threshold to motif heights, it is possible to merge small
insignificant motifs with larger adjacent ones and only retain significant motifs at
the end of the process. This action of merging small motifs into larger ones is
called pruning the change tree. Scott introduced the name Wolf pruning (after Gert
Wolf’s work) to designate merging with respect to a height threshold, which is
usually specified as a percentage of total height (Sz, see Chap. 2). Other thres-
holding criteria can be used depending on the application, such as minimum area,
perimeter, roundness, etc., or used together in combination.
Fig. 3.13 A complete change tree with peaks (P), saddles (S) and pits (V)
54 F. Blateyron
Figures 3.14, 3.15 and 3.16 show the effect of pruning on a surface with
several pruning criterion values.
Fig. 3.14 Representation of surface texture with contour lines (left). Segmentation without any
pruning resulting in 1638 hills (right)
Fig. 3.15 Segmentation with a Wolf pruning of 1 % of Sz resulting in 344 hills (left);
segmentation with a Wolf pruning of 5 % of Sz resulting in 134 hills (right)
Fig. 3.16 Segmentation with a Wolf pruning of 10 % of Sz resulting in sixty hills (left);
segmentation with a Wolf pruning of 20 % of Sz resulting in twenty-seven hills (right)
3.4 Feature Parameters
3.4.1 Peak Density, Spd
The peak density is calculated by dividing the number of peaks by the unit area. It
is given in units of [1/mm2] or [peaks/mm2]. This parameter evolved from the
density of summits, Sds, one of the Birmingham 14 parameters (see Chap. 2). The
main difference between the old and new parameters is that with Spd, only sig-
nificant peaks are taken into account, due to the watershed segmentation of the
surface and the pruning of the change tree by a specified pruning factor. The Spd
parameter can be seen as the areal equivalent of the peak count parameter RPc
defined in ISO 4287 Amendment 1 (2009).
The Spd parameter can be used in applications where contact is involved,
usually in combination with other parameters such as minimum or maximum
height parameters.
3.4.2 Peak Curvature, Spc
The Spc parameter is the arithmetic mean curvature of significant peaks. ISO
25178 part 3 (2012) specifies that Spc is given in [1/mm] by default. This
parameter is equivalent to the mean curvature of summits, Ssc, which was part of
the Birmingham 14 parameters (see Chap. 2). As with Spd, only significant peaks
are taken into account for Spc (see Sect. 3.4.1).
56 F. Blateyron
The curvature is not uniquely defined on a sampled surface (extracted surface).

A formula was proposed in the SURFSTAND report (Blunt and Jiang 2003) based
upon a least-squares polynomial estimated in several directions.
The curvature is the inverse of the radius of curvature; i.e. a large value of Spc
means a small radius of curvature. In this case, the peak is sharp and is likely to be
worn away during a sliding contact or be plastically deformed during a static contact.
A small value of Spc means a large radius of curvature. In this case, the peak is
wide, will exhibit good contact properties and may be subject to elastic defor-
mation in case of static contact.
3.4.3 Motif Slope
In the original French standard (CNOMO E00.14.015.N 1983), a slope parameter

was defined from the mean height and mean width of motifs on profiles
AR
KR ¼ ð3:1Þ
2R
where AR is the mean motif’s width and R is the mean motif’s height. The idea was
to evaluate the motif slope from the centre of the motif (AR/2). Low values are
observed on globally flat profiles, and large values of KR suggest that texture cells
(motifs) are deep and periodical.
A similar parameter can be devised for surfaces. An equivalent of the motif width
can be found with MED in Sect. 3.4.6 and the motif height hm is defined in Sect. 3.3.1.
It is proposed that the slope on each motif should be calculated and then averaged
with respect to the number of motifs (N). In accordance with this proposal, the
parameter (which is not standardised) would be named SKS, mean motifs slope
1 X MED
SKS ¼ : ð3:2Þ
N 2hm
where MED is the equivalent diameter of the motif (see Sect. 3.4.6) and hm is the
motif height (see Sect. 3.3.1). The ratio MED=2hm evaluates the relative flatness of
the motif. The SKS parameter could be given in [mm/lm] or [mm/mm], be unit-
less, or even be converted in angle.
3.4.4 Significant Heights
Quantifying surface texture heights has always been a challenge from the begin-
ning of surface measurement. With profile measurements, multiple parameters
have been designed to capture what could be called the roughness height [the
profile parameters––see Leach (2009)], with the idea of averaging local maximum
0 1 2 3 4 5 6 mm µm
0.0
300
0.5
1.0 250
ISO 25178
1.5
Height Parameters
2.0 200
Sp 87.45 µm Maximum peak height
2.5 Sv 236.0 µm Maximum pit height
150
3.0 Sz 323.5 µm Maximum height
3.5 100 Feature Parameters

4.0 S5p 53.84 µm Five point peak height
50 S5v 150.8 µm Five point pit height

4.5
S10z 204.6 µm Ten point height
5.0 0
mm
Fig. 3.17 Comparison of height parameters (Sp, Sv and Sz) with feature height parameters (S5p,
S5v and S10z)
heights on small segments (sampling lengths). However, with areal surface mea-
surements, the Sz parameter (maximum height of the surface) acts more like the Rt
parameter (the profile height parameter that is calculated over typically five
sampling lengths) because it does not average multiple values. Hence, Sz, as
defined in ISO 25178 part 2 (2012), is highly sensitive to outliers and cannot be
used as a robust evaluation of the areal surface texture height.
The definition of significant peaks and significant pits, based upon pruning the
change tree, can be used to establish robust parameters that estimate the surface
texture height: S5p, S5v and S10z, defined as follows.
S5p is the arithmetic mean height of the five highest significant peaks.
S5v is the arithmetic mean height of the five lowest significant pits. Note that
height is always positive and the ‘‘height’’ of a pit is in fact its depth.
S10z is simply the sum of S5p and S5v:
S10z ¼ S5p þ S5v: ð3:3Þ
These parameters are given in [lm] or more generally in the unit of the z axis used
in the surface representation.
Figure 3.17 shows height parameters calculated on a leather surface and
compares them with feature parameters.
3.4.5 Area and Volume
The dale area, Sda and hill area, Sha are calculated as the mean area of all
individual motifs. The area considered here is the horizontal area of the motif
projected onto the horizontal plane. The Sda and Sha parameters are given in
millimetres squared.
58 F. Blateyron
The dale volume, Sdv and hill volume, Shv are calculated as the mean of all
individual motifs. The volume is delimited horizontally by its ridge (course line),
and delimited vertically along the z axis by the surface on one hand and by a
horizontal plane placed at its lowest (highest) saddle point on the other. The Sdv
and Shv parameters are given in [mm3].
Area and volume parameters are usually calculated on closed motifs, i.e. on
motifs that are not cut by the edge of the evaluation area. However, in order to be
flexible, the specification should clearly indicate whether they are calculated on
closed, open or on all motifs.
3.4.6 Morphological Parameters
Besides the characterisation of the height or volume of a motif, there are cases in
which the horizontal shape needs to be characterised. A motif can be seen from
above as a flat shape that can be characterised only in the xy plane (see Fig. 3.18).
Russ (1995) summarised a set of morphological parameters that can be used to
estimate the shape of objects in an image. The following parameters can be retained
for applications such as grain and particle analysis. For simplicity, a symbol is
suggested for these parameters, based on the capital M (as in morphological) and a
suffix designing the parameter name. These symbols are not standardised.
The form factor, MFF evaluates the compacity of the shape (the filled volume
fraction). An elongated object will give a value close to zero while a compact
object will give a value close to unity. The MFF parameter is given by
4pA
MFF ¼ ð3:4Þ
P2
with A is the motif area and P its perimeter. The MFF parameter is unit-less.
The aspect ratio, MAR parameter discriminates between compact and oblong
motifs. For example, in the case of a disc, MAR = 1. In the case of an oblong motif,
its value will be greater than 1,
Fig. 3.18 Assessment of the

radius calculated from the
barycentre point to the
perimeter point
Dmax
MAR ¼ ð3:5Þ
Dmin
where Dmin and Dmax are the minimum and maximum diameters. The MAR
parameter is unit-less.
The roundness parameter, MRN is similar to the form factor. A round object will
give a value of unity, and an oblong object will give a value less than unity. The
MRN parameter is given by
4A
MRN ¼ ; ð3:6Þ
pD2max
and is unit-less.
The equivalent diameter, MED is the diameter of the circle of same area as the
motif, given by
rffiffiffiffiffiffi
4A
MED ¼ ð3:7Þ
p
where A the motif area. The MED parameter is given in [mm].
These morphological parameters can be calculated for each individual motif, or
given as an average for all motifs (all, closed or open). For convenience, symbols
are proposed (by the author) in Table 3.1, so that they can be used in
specifications.
Table 3.1 Proposed symbols for morphological parameters on motifs (hills or dales)
Form factor Aspect ratio Roundness Equivalent diameter
Single motif MFF MAR MRN MED
Average (all motifs) SMFF SMAR SMRN SMED
3.5 Automatic Partitioning and Levelling
Watershed segmentation, when associated with a pre-filtering that emphasises

edges of geometrical features, provides a powerful method for partitioning sur-
faces. This is particularly interesting for levelling a surface with respect to one or
more parts, or layers of a component, for example, contact zones on MEMS (see
Fig. 3.19) or PCBs.
Figure 3.20 shows how auto-partitioning is carried out. The surface shape is
segmented automatically and an image of the motifs is displayed (upper right). The
user can click on one or several motifs to define a reference area which is shown in
yellow. This reference area is used to level the whole surface (lower left) and is
extracted (partitioned) for further analysis, such as flatness measurement of the
contact zone (lower right). This solution is very straightforward when levelling a
structured surface, which would be difficult using a least-squares plane.
60 F. Blateyron
Fig. 3.19 3D view of a surface measured on a micromechanical sensor (left) and its automatic
partitioning into homogeneous parts (right)
Fig. 3.20 Auto-partitioning features in a commercial package (MountainsMapÒ)

3.6 Verification of Nominal Geometry
A structured surface contains geometrical patterns with elements that are repeated
periodically over the surface. The patterns implement a particular function that is
related to shape, form factor or developed surface, such as hydrophobia, absorp-
tion, photoelectric conversion, thermic transfer, etc. Therefore, it is important to
verify the geometry and tolerances of the surface with respect to geometrical
specifications.
Watershed segmentation can provide a first level of pattern recognition by
isolating each element in a pattern as a motif. Geometrical form can then be fitted
to the motifs in order to calculate the dimensions of pattern elements. Figure 3.21
shows a micro-sphere surface where each sphere cell is segmented by watersheds.
This application shows that verification of nominal geometry can be automated,
not only for spheres but also on other types of geometry.
A least-squares sphere is fitted to each motif. In addition to the radius, this
provides a sub-pixel evaluation of the centre position, making it possible to cal-
culate pitch between adjacent spheres (see Fig. 3.22).
Fig. 3.21 Micro-spheres

automatically segmented by
watersheds
62 F. Blateyron
Fig. 3.22 Fitting a least-

squares sphere (2) to the
points of a motif (1) allows
the calculation of the true
centre (3) and the true radius
(4). The sphere is usually
larger than the visible circle
(5) which is the intersection
of the sphere and the surface
3.7 Specification of Feature Parameters
ISO/CD 25178 part 1 (2009) describes the way surface specifications should be
written on engineering drawings. For example, consider the following specifica-
tion line:
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
S L 0; 005 0; 8=Salð0; 3Þ 45
This specification requires that the Sal parameter be calculated with a threshold of
0.3 (instead of the default 0.2), shall not be greater than 45 lm (the U symbol for
‘‘upper’’ is a default and, therefore, can be omitted from the specification). The
parameter is calculated on an S-L surface with nesting indexes of 5 lm and
0.8 mm (see Chap. 4). (Note that in ISO specification standards it is common to
use a comma to designate a decimal point, but this is not used throughout the rest
of this book.)
However, in the case of feature parameters, there is an additional possibility
where the parameter name can be described by a feature specification line:
FC; \feature[; \pruning[; \target[; \attribute[; \statistics[
where FC refers to feature characteristics. The following indication, feature,
defines the set of segmented features which is the subject of the specification:
Symbol Definition Type

H Hills Areal
D Dales
C Course line Line
R Ridge line
P Peak Point
V Pit
S Saddle
The following indication, pruning, specifies the pruning conditions:

Indication Definition
Wolf pruning x % Pruning if motif height \ x % of Sz
Area x % Pruning if motif area \ x % of definition area
volS x Pruning if motif volume \ x of specified volume
More criteria can be used such as morphological parameters (for example, form
factor). The following indication, target, selects features to be taken into account:
Symbol Definition
Closed x % Motif does not intersect the edge of the surface when thresholded at x %
of bearing ratio
Open x % Motif intersects the edges of the surface when thresholded at x % of bearing ratio
Top N One of the N highest peaks
Bottom N One of the N lowest pits
All All motifs are taken into account
The following indication, attribute, defines which operand is calculated on the

selected features:
Type Symbol Definition

Areal lpvh Local motif height
volS Motif volume
Area Motif area
Line Length Length (perimeter) of a line
Point pvh Absolute height
Curvature Curvature at the point
All Count Counts the number of features
The following indication, statistics, specifies how parameter values should be

used:
Indication Definition
Mean Average of all values
Max/min Maximum/minimum of all values
RMS Standard deviation of all values
Percent x % Percentage of values above the specified value
Sum Sum of all values
Density Sum of all values divided by the definition area
64 F. Blateyron
For example, the peak density parameter, Spd can be described by the following
specification line:
FC; H; Wolfprune 5 %; all; count; density.
The height parameter, S5v can be described by the following specification line:
FC; V; Wolfprune 5 %; bottom 5; lpvh; mean.
The percentage of peaks above a specified height, for example, is given by:
FC; H; Wolfprune 5 %; top; pvh; percent 50 lm.
Acknowledgments All figures showing surface topography, motifs and feature parameters were
produced using MountainsMap 6.2, the surface analysis software package from Digital Surf. The
author would like to thank Mr Antony Caulcutt (Digital Surf) and Dr. Gert W. Wolf (University
of Klagenfurt) for reviewing the manuscript and for suggesting useful corrections.
References
Barré F (1997) Contribution de l’analyse d’images à la caractérisation morphologique des

surfaces industrielles. PhD Thesis: Saint-Etienne, France
Barré F (2001) Watershed lines and catchment basins: a new 3D motif method. Int J Mach Tools
Manuf 40:1171–1184
Beucher S (1990) Segmentation d’images et morphologie mathématique. PhD Thesis: ENSMP
Blateyron F, Adam M (2004) Application of image segmentation to motifs evaluation of 2D
profiles. Proc XI Colloq Surfaces, Chemnitz, Germany, Feb 56–64
Blunt L, Jiang X (2003) Advanced techniques for assessment surface topography. Kogan page
science, London
Boulanger J (1992) The motifs method, an interesting complement to ISO parameters for some
functional problems. Int J Mach Tools Manuf 32:203–209
Cayley A (1859) On contour and slope lines. Phil Mag 4:264–268
CNOMO E00.14.015.N (1983) Etats géométriques de surface, calcul de paramètres de profil
ISO 12085 (1996) Geometrical product specifications (GPS)—surface texture: profile method––
Motif parameters. International Organization for Standardization
ISO 25178 part 2 (2012) Geometrical product specification (GPS)—surface texture: areal––part
2: terms, definitions and surface texture parameters. International Organization for
Standardization
ISO 25178 part 3 (2012) Geometrical product specification (GPS)—surface texture: areal––part
3: specification operators. International Organization for Standardization
ISO 4287 Ammendment 1 (2009) Geometrical product specifications (GPS)—surface texture:
profile method—terms, definitions and surface texture parameters; Amendment 1: peak count
number. International Organization for Standardization
ISO/CD 25178 part 1 (2009) Geometrical product specifications (GPS)—surface texture: areal—
part 1: drawing indications. International Organization for Standardization
ISO/DIS 16610 part 85 (2012) Geometrical product specifications (GPS)—filtration—part 85:
areal morphological: segmentation. International Organization for Standardization
Kweon IS, Kanade T (1994) Extracting topographic terrain features from elevation maps. Image
Underst 59(2):171–182
Maxwell JC (1870) On hills and dales. Phil Mag 30:233–240

Morse SP (1969) Concept of use in contour map. Commun in the ACM 12.3:147–152
MountainsMapÒ surface analysis package. www.digitalsurf.com
Peucker TK, Douglas DH (1975) Detection of surface-specific points by local parallel processing
of discrete terrain elevation data. Comp Graphics Image Proc 4:375–387
Pfaltz JL (1976) Surface networks. Geogr Anal 8:77–93
Rana S, Morley J (2002) Surface networks. Working papers series: Centre for Advanced Spatial
Analysis, UK
Russ JC (1995) The image processing handbook. CRC, USA
Scott PJ (1992) The mathematics of motif combination and their use for simulation. Int J Mach
Tools Manuf 32:69–73
Scott PJ (1998) Foundations of topological characterization of surface texture. Int J Mach Tools
Manuf 38:559–566
Scott PJ (2004) Pattern analysis and metrology: the extraction of stable features from observable
measurements. Proc R Soc A460:2845–2864
Takahashi S, Ikeda Y, Shinagawa Y, Kunii TL, Ueda M (1995) Algorithm for extracting correct
critical points and constructing topological graphs from discrete geographical elevation data.
Int J Eurographics Assos 14:181–192
Vincent L (1990) Algorithmes morphologiques à base de files d’attente et de lacets, extension aux
graphes. PhD Thesis: Ecole des Mines, Paris
Vincent L, Soille P (1991) Watersheds in digital spaces: an efficient algorithm based on
immersion simulations. IEEE Trans Pattern Analysis Mach Int 13:583–598
Warntz W (1975) Stream ordering and contour mapping. J Hydrology 25:209–227
Wolf GW (1984) A mathematical model of cartographic generalization. Geo-processing
2:271–286
Wolf GW (1993) Data structures for the topological characterization of topographic surfaces.
In: 6th Colloque Européen de Géographie Théorique et Quantitative, Montpellier, pp 24–34
Zahouani H (1993) Analysis of 3D pattern topology of manufactured surfaces in regards to
tribological situations. In: Proc 6th Int Congress Tribology, Budapest, Aug. pp 284–289
Zahouani H (1998) Spectral and 3D motifs identification of anisotropic topographical
components. Int J Mach Tools Manuf 38:615–623
Chapter 4
Areal Filtering Methods
Jörg Seewig
Abstract Filtering is essential for surface texture characterisation. Filtration

separates the measured data into different scales of interest. The extracted scales
can be characterised by parameters given in ISO 25178 part 2. In this chapter the
two draft areal filtering standards, ISO 16610 part 61 (areal Gaussian filter) and
ISO 16610 part 71 (robust areal Gaussian regression filter), are discussed. Spline,
morphological and wavelet filters are also briefly introduced.
4.1 Basic Concepts: The Scale Limited Surface
Any surface topography can be represented as a superposition of different geo-

metrical structures with different scales (from large scales down to very small
scales). Form, or form deviation, has a large scale whilst the fine structure on the
surface, roughness, has a small scale. Waviness is a periodic component and can
have large and small scales respectively. A typical surface with these different
scale features is shown in Fig. 4.1. The aim of filtration is to separate surface
measurement data into the large—and small-scale components. Filtration is
essential for further investigation of the data, because each scale component will
be the result of the manufacturing process, and each component will have an effect
on the functionality of the surface.
International standard ISO 25178 part 2 (2012) defines two surfaces: the S-F
surface and the S-L surface. The designation ‘‘S’’ indicates that the short-wave
surface deviations are eliminated by a low-pass filter. ks filtering for profile
evaluation according to ISO 3274 (1996) is the equivalent to S filtering (see Leach
(2009) for a discussion on profile filtering). The designation ‘‘F’’ indicates that the
J. Seewig (&)
University of Kaiserslautern, Gottlieb-Daimler-Straße, 67663 Kaiserslautern, Germany
e-mail: seewig@mv.uni-kl.de

68 J. Seewig
Fig. 4.1 Topography of a honed surface with form, waviness and roughness
Fig. 4.2 Definition of the S-F surface and the S-L surface
nominal form is removed from the surface dataset by using the so-called
F-operation. As the default, an F-operation is defined that performs a total least-
squares fit of the a priori nominal form. In individual cases, such as with freeform
surfaces or unknown nominal forms, a form filter can be used, for example a
second order regression filter according to ISO 16610 part 71 (2011), to eliminate
the nominal form. The designation ‘‘L’’ indicates high-pass filtering of the surface.
This means that long-wave surface deviations, such as components of the form
deviation or waviness, can be removed. The signal chain described in ISO 25178
part 2 can be seen in Fig. 4.2.
Two steps are necessary to apply a filter operation. The first step is to choose a
suitable filter method for a given problem, and the second step is to choose the
appropriate filter parameter to divide the measurement data into the large—and
small-scale components of interest. Both steps should be carried out with care,
because an inappropriate choice of either the filter method or the filter parameter
can lead to erroneous results. Users without experience in filtration should always
use the default values defined in ISO 25178 part 3 (2012) for areal surfaces and
ISO 4288 (1996) for profiles.
4 Areal Filtering Methods 69
4.2 Linear and Robust Filters
A linear or robust filter is an operation that can be applied to extract or suppress

certain scale ranges of the surface, zðx; yÞ. The filter operation to extract the long-
wave (large-scale) components can be described with the mathematical
formulation
filter
zðx; yÞ ! wðx; yÞ: ð4:1Þ
Equation (4.1) corresponds to low-pass filtering of the surface. The short-wave
(small-scale) components r ðx; yÞ are the result of subtracting the surface and the
filtered surface
r ðx; yÞ ¼ zðx; yÞ wðx; yÞ: ð4:2Þ
This is referred to as high-pass filtering of the surface.
Linear Filters
A filter is considered to be linear when it meets the following requirement
linear filter
a z1 ðx; yÞ þ b z2 ðx; yÞ ! a w1 ðx; yÞ þ b w2 ðx; yÞ: ð4:3Þ
Equation (4.3) states the following: when two surface datasets are additively
superimposed and a linear filter is then applied, the result is identical to the
additive superposition of the filtered surfaces az1 ðx; yÞ and bz2 ðx; yÞ:
A linear filter determines a moving average. However, the values are weighted
differently in the averaging process. Thus the respective function is called the
weighting function sðx; yÞ of the filter, which also determines the properties of the
filter. Determination of a moving average can be expressed with a double integral
for areal filtering
ZZ
wðx; yÞ ¼ sðx l; y mÞ zðl; mÞdl dm ð4:4Þ
where l and m are the variables of the integration in x and y directions respectively.
The value wðx; yÞ of the filtered surface is calculated by moving the origin of the
weighting function sð0; 0Þ to the position ðx; yÞ and by determining the weighted
average in the range of influence of the weighting function. This operation is also
referred to as convolution, since the values of the weighting function are reflected
over the ordinate before integration (averaging) occurs. To prevent an ideally
plane surface with the constant topography height zðx; yÞ ¼ z0 from being affected
by filtering, the following normalisation condition applies
ZZ
sðx l; y vÞdl dm ¼ 1: ð4:5Þ
With some filters, the resulting weighting function sðx; yÞ can be expressed as
multiplication of two single weighting functions, such that sðx; yÞ ¼ sð xÞ sð yÞ
70 J. Seewig
applies. In this case, the weighting function is called separable and the double
integral can be expressed as two single integrals
Z Z
wðx; yÞ ¼ sðy mÞ sðx lÞ zðl; mÞdl dm: ð4:6Þ
Graphically, surface filtering can be reduced to profile filtering, first in x

direction and then in y direction. This property results in numerically efficient and
fast filtering.
The filter effect of a linear filter is defined with the aid of sinusoidal—and
cosinusoidal-shaped surface deviations. For simplicity the topography is specified
as a wavefront with a complex-valued exponential function with the real amplitude
a0 that approximates the sine and the cosine terms

i 2p x
kx þ kyy
zðx; yÞ ¼ a0 e : ð4:7Þ
The wavefront, shown in Fig. 4.3, extends in the x direction with the wave-
length kx and in the y direction with the wavelength ky .
Figure 4.3 also shows the correlation k ¼ kx cosu and k ¼ ky sinu. Adding the
complex-valued exponential function to Eq. (4.4) leads to the filter transfer
function
ZZ
i 2p kxx þkyy
S kx ; ky ¼ sðl; mÞe dl dm: ð4:8Þ
Equation (4.8) corresponds to the Fourier transform of the weighting function. The
effect of the filter on the wavefront can then be expressed as
a1 a1
wðx; yÞ ¼ zðx; yÞ; ¼ S kx ; ky ð4:9Þ
a0 a0
where a1 is the amplitude of the filtered exponential function. The following
applies to the short-wave component r ðx; yÞ
Fig. 4.3 Wavefront to

calculate the transmission
characteristic for sinusoidal
topographies
a2 a2 a1
r ðx; yÞ ¼ zðx; yÞ; ¼1 : ð4:10Þ
a0 a0 a0
It is essential for dimensional metrology that there is no phase displacement
between the topography and the filter plane, i.e. the filter must be phase correct and
its transfer function must be positive real.
It is also necessary that the direction of filtering is irrelevant, thus the filter
transfer function is not a factor of the direction of propagation of the wavefront.
Filtering independent of the direction satisfies the following

k k
S ; ¼ constant; u 2 R: ð4:11Þ
cosu sinu
The cut-off wavelength is defined as the filter parameter for a linear filter. If the
filter works independently of the direction, filtering causes a sinusoidal surface
with a wavelength that is equal to the cut-off wavelength to be damped to 50 % of
its initial amplitude.
Figure 4.4 illustrates that any wavefront with amplitude a0 and wavelength k,
which is the same as the cut-off wavelength kc used in profile filtering, is always
damped to 50 % of the initial amplitude when filtering is independent of direction.
Robust Filters
A filter is called robust, if the output data are insensitive to specific phenomena in the
input data. Such phenomena can be outliers, which are local portions in a dataset that
are not representative or not typical for the partitioned integral feature, and are
characterised by magnitude and scale. Other phenomena are structures on the surface
such as isolated grooves. Such grooves serve as oil reservoirs on cylinder liners in
combustion engines and can significantly improve the tribological behavior of the
sliding surfaces (see Chap. 11). Because a linear filter is a moving average, it is
influenced by such isolated structures. Figure 4.5 shows the difference between a
Fig. 4.4 Definition of the cut-off wavelength

72 J. Seewig
Fig. 4.5 Response of the linear and robust filter in case of a surface with an outlier
linear filter and a robust filter when filtering a surface that has an outlier. The linear
filter follows the outlier while the robust filter is not influenced by the outlier.
The inadequate operation of a linear filter when outliers are present can be
explained as follows. The linear filter process is a weighted least-squares problem
expressed by
ZZ
2
min wðx;yÞ ðzðl; mÞ wðx; yÞÞ sðx l; y mÞdl dm : ð4:12Þ
As the residuals are squared, any outlier will influence the filter point. To avoid this
effect, a robust filter mitigates the weight of the outliers using a loss function q. In the
case of the weighted moving average, this leads to the minimisation problem
ZZ
min wðx;yÞ qðzðl; mÞ wðx; yÞÞ sðx l; y mÞdl dm : ð4:13Þ
A robust filter is non-linear and needs iterative methods to calculate the filtered
surface. Because of the non-linearity, robust filters have no transfer functions.
For a robust filter, a type of cut-off wavelength is also defined as the filter
parameter; however, the cut off wavelength follows the occurring widths of the
measured geometrical structures. Different robust filters have different transfer
properties, so the definition of the cut-off wavelength is always a function of the
respective filter.
4.2.1 Planar and Cylindrical Filters
In planar filtering the measurement dataset is only one section of the total surface
and, for example, for rectangular sections, is limited by four edges. Such a
Fig. 4.6 Planar surface as part of the topography of a cylinder liner
measurement dataset is referred to as open. The measured section consists of a

superposition of the form, form deviation, waviness and texture of the component
surface. An example of an open surface can be seen in Fig. 4.6, which shows a
result of a topography measurement of a cylinder liner surface. Separation of the
different surface features is achieved with planar filtering which is independent of
the direction.
If the areal surface texture is measured around the entire circumference of a
cylindrical component, the result is a closed surface in the circumferential direc-
tion of the cylinder. If the measured data are projected onto a plane, the measuring
area is also closed in the circumferential direction. In the axial direction, however,
there are two edges. Figure 4.7 illustrates a measurement of a cylindrical surface.
Such surfaces require cylindrical filtering, which has a different mathematical
description to that of planar filtering. The filter settings are made for cylindrical
filtering: in circumferential direction, damping is stated as waves on the circum-
ference fc and in the axial direction as the cut-off wavelength kcz . With this
specification, filtering is no longer necessarily independent of direction.
4.2.2 The Linear Gaussian Filter
If a Gaussian function is used to determine a weighted moving average, the filter is

known as a Gaussian filter. The Gaussian filter has become an integral part of
surface metrology. A distinctive feature of a Gaussian filter is that its transfer
function is again Gaussian.
74 J. Seewig
Fig. 4.7 Measurement of a cylindrical surface
4.2.2.1 The Linear Gaussian Planar Filter
The weighting function of the planar filter with the cut-off wavelength kc is then
1 pða xk Þ2 1 pða yk Þ2
sðx; yÞ ¼ e c e c : ð4:14Þ
akc a kc
|fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl} |fflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflffl}
sðxÞ sðyÞ
pffiffiffiffiffiffiffiffiffiffiffiffi
The filter constant a ¼ ln2=p ffi 0:4697 specifies amplitude damping of
50 %. The weighting function is axially symmetric and, according to Eq. (4.14),
separable. Figure 4.8 shows the shape of the weighting function.
The Gaussian weighting function has no compact support. No compact support
means that the weighting function is unbounded in the x and y directions. For a
practical determination of the moving average, the weighting function outside of a
pffiffiffiffiffiffiffiffiffiffiffiffiffiffi
defined radius must be set to zero, meaning sðx; yÞ ¼ 0 for x2 þ y2 [ Lc kc . The
surface area that can be evaluated is reduced around each edge by Lc kc . For
implementation in reference software, a value of Lc 0:6 is recommended
(Krystek 2005). To prevent the measuring area from being reduced too much, in
practice a value of Lc 0:5 is selected without significantly distorting the filter
transfer function.
Applying Eq. (4.8) and taking into consideration the wavefront from Fig. 4.3,
results in the following filter transfer function
ðkkc Þ2
a1 kc 2 1
Sðk; uÞ ¼ ¼ epða k Þ ¼ : ð4:15Þ
a0 2
Figure 4.9 illustrates the filter transfer function for the short-wave and long-wave
components depending on the quotient kc =k.
Fig. 4.8 Weighting function of the Gaussian planar filter
Fig. 4.9 Filter transfer function of the Gaussian planar filter for long- and short-wave
components
4.2.2.2 The Linear Gaussian Cylindrical Filter
The form of the weighting function of a cylindrical filter is determined by the cut-
off wavelength kcz in the axial direction and the cut-off wavelength L=fc in cir-
cumferential direction, where L is the circumference of the cylinder and fc is the
number of waves around the circumference. When the coordinates z for the axial
direction and t for the circumferential direction are added, the following applies to
the weighting function
fc 2 tfc 2
epðakcz Þ epðaLÞ :
z
sðz; tÞ ¼ 2
ð4:16Þ
a kcz L
76 J. Seewig
Fig. 4.10 Filter transfer function for the long-wave components of the two filter transfer
functions Sðkz Þ and Sð f Þ of the Gaussian cylindrical filter
The two possible filter settings (axial and circumferential) generally lead to a
weighting function that is not axially symmetric. Correspondingly, the weighting
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
function is set to zero outside the range of z2 =k2cz þ t2 fc2 =L2 [ Lc in order to be
able to determine the average. The values for Lc are identical to those for planar
filtering. However, the edges of the measurement section are reduced only in the
axial direction by Lc kcz each. The measurement field is closed in the circum-
ferential direction and filtering can occur without loss of valuable measurement
points.
Since the weighting function is separable, the filter transfer function Sðkz ; f Þ can
also be separated into the axial and circumferential directions. Thus,
ðkkcz Þ2 ðff Þ2
a1 1 z 1 c
Sðkz ; f Þ ¼ ¼ Sðkz ÞSð f Þ ¼ : ð4:17Þ
a0 2 2
Figure 4.10 shows the filter transfer function Sðkz Þ as a factor of the quotient kcz =k
and Sð f Þ as a factor of the quotient f =fc .
When a wavefront with the direction of propagation u and wavelength k as in
Fig. 4.11, is considered, kz ¼ k sin1 u applies to the axial component and L=k ¼
k cos1 u to the circumferential direction.
Added to Eq. (4.17), the resulting filter transfer function is
ð1k kcz sinðuÞÞ2 ð1k fL cosðuÞÞ2
1 1 c
Sð kz ; f Þ ¼ : ð4:18Þ
2 2
Equation (4.18) clearly indicates that the effect of filtering for any cut-off wave-
lengths, kcz in the axial direction and L=fc in the circumferential direction, depends
on the direction of the wavefront. Filtering independent of direction can be
achieved only for the case that kcz ¼ L=fc . This leads to a transfer function
given by
Fig. 4.11 Relationship

between the axial
wavelength, the
circumferential wavelength
and the resulting wavefront
ðkkcz Þ2 ð1k fL Þ2
1 1 c
Sð kz ; f Þ ¼ ¼ : ð4:19Þ
2 2
4.2.2.3 Digital Implementation of the Linear Gaussian Filter
Thus far in this chapter the spatial continuous representation of filtering has been
considered. Digital implementation of spatial continuous filters on a computer
generally occurs with equidistant sampling of the surface and of the weighting
function in increments of Dx and Dy. An integral can also be approximated by
summation for small increments of Dx and Dy. Figure 4.12 shows this approxi-
mation for the one-dimensional case.
If a finite number of samples Nx or Ny is examined, the convolution integral
from the Eq. (4.4) can be conveyed to a convolution sum
XNy 1 XNx 1
wðnDx; mDyÞ ¼ l¼0 k¼0
sððn kÞDx; ðm lÞDyÞzðkDx; lDyÞDxDy:
ð4:20Þ
For the weighting function to be completely superimposed on the topography,

meaning that it is spatially invariant, valid values for the filter plane can be
calculated only within the ranges Lc kc n Dx ltx Lc kc and
Lc kc m Dy lty Lc kc . The remaining area is referred to as the running-in and
running-out lengths of the filter. As discussed previously, 0:5 Lc 1:0 should be
selected. This reduces the measuring area that can be evaluated, the extent of
which is a factor of the cut-off wavelength. Section 4.2.2.5 describes a method that
can be applied to reduce the running-in and running-out lengths of a filter.
It is also essential that the increments Dx and Dy are small enough that sampling
does not significantly change the transfer properties of the linear weighting
function and the filter transfer function respectively. An adequate approximation
78 J. Seewig
Fig. 4.12 Approximation of an integral by a summation
Fig. 4.13 Deviation in percentage between the continuous amplitude transmission and the
digital implementation of the truncated Gaussian filter
of the transfer behaviour can be obtained for the increments Dx kc =20 and
Dy kc =20 (Krystek 2005). Figure 4.13 shows the deviations in percentage from
the ideal filter transfer function for sampling increments kc =10 und kc =50; as well
as for the limiting factors Lc ¼ f0:5; 0:6; 1:0g.
4.2.2.4 Limits of the Linear Gaussian Filter
The Gaussian filter has become an integral factor globally in the field of surface
metrology. Fast filtering on the basis of profile filters can be facilitated particularly
with separation of the weighting function. The Gaussian filter is not as suitable
when parts of the component shape are to be compiled and eliminated by filtering.
Depending on how pronounced the part of the shape is, there will typically be an
offset between the measured topography and the filter plane. Before applying a
Gaussian filter, it is always better to remove the form by applying an F-operation
(see Chap. 5). If the topography is merely tilted, the Gaussian filter can be used
without the F-operation.
4.2.2.5 Handling End Effects of the Linear Gaussian Filter
A distinction is made between two types of end effects when filtering: the loss of
valuable measuring area due to the running-in and running-out length of the
weighting function and the distorted transfer behaviour of the filter around the
edges of the measured data. First of all, it is possible to adapt the filter operation
such that filtering can be performed into the edge area. One way to do this is to
increase the measuring area by reflecting the measured data in the edge area. Then
measured data can be filtered all the way into the edge area. Figure 4.14 illustrates
reflection of the measured data.
A better way to eliminate end effects is to asymmetrically truncate the
weighting function around the edges and to correct the volume with position-
dependent rescaling to the value 1, in accordance with Eq. (4.5). This procedure
can be expressed with the convolution integral according to Eq. (4.4)
R lty R ltx
sðx l; y mÞzðl; mÞdl dm
wðx; yÞ ¼ 0 R lt0y R ltx : ð4:21Þ
0 0 sðx l; y mÞdl dm
Filtering is separable in this case too, and can be performed numerically effi-
ciently. First profile filtering in the x direction is performed
R ltx1 pða kc Þ
xl 2
0 akc e zðl; mÞdl

wðx; mÞ ¼ R ltx xl 2
; ð4:22Þ
1 pð Þ dl
0 a kc e
akc
followed by profile filtering in y direction

R lty ym
1 pð akc Þ
2
0 a kc e wðx; mÞdm
wðx; yÞ ¼ R lty ym 2
: ð4:23Þ
1 pð Þ dm
0 a kc e
akc
Figure 4.15 shows the re-scaled Gaussian weighting function. The correction
occurs in a corner of the measurement section. The weighting function is truncated
Fig. 4.14 Left the measured dataset is reduced by kc =2 at the edges of the surface. Right the
dataset is mirrored at the edges of the surface by kc =2
80 J. Seewig
Fig. 4.15 The re-scaling of the Gaussian weighting function is one method to avoid end effects
Fig. 4.16 Left S-F surface with roughness and waviness. Right S-L surface created with
rescaling of the Gaussian weighting function. The whole surface is still valid
and its amplitude increased in relation to the central position. This type of cor-
rection is described for profile filters in ISO 16610 part 28 (2010).
Figure 4.16 shows an example of a smoothed surface with some waviness. The
measuring area is 2 mm by 2 mm. Filtering is performed with the aid of the
weighting function correction described above. A cut-off wavelength of
kc = 0.25 mm was chosen. The right hand image shows the S-L surface. It is clear
that the waviness was eliminated without loss of valuable measuring area around
the edges of the topography.
4.2.3 The Linear and Robust Gaussian Regression Filter
As described in Sect. 4.2.2, the areal Gaussian filter according to ISO 16610 part
61 (2012) has three essential but limiting features:
• without additional corrections to the weighting function, running-in and
running-out lengths lead to a reduction of the valuable measuring area;
• it must be ensured that distinct components of the form are eliminated by an
F-operation before filtering; and
• the linear Gaussian filter is not resistant (robust) to outliers.
The effect of these factors is reduced or eliminated by the concept of regression
filtering according to ISO 16610 part 71 (2011). The regression filtering method
was introduced by Seewig (1999) and a brief overview of the concept is given in
elsewhere (Seewig 2005). The regression filtering method is best explained with a
simple example: a point ðx; yÞ within the topography zðx; yÞ is examined. At
exactly this position, a filter value wðx; yÞ is to be determined such that the least-
squares sum between the filter value and the topography assumes a minimum
across the entire measuring area (i.e. within the range 0 x ltx and 0 y lty ).
The squares are also weighted with the weighting function sðl; vÞ, the origin of
which is at the point ðx; yÞ. Mathematically, this can be expressed as
Z lty Z ltx
J¼ ðzðl; vÞ wðx; yÞÞ2 sðx l; y vÞdl dv: ð4:24Þ
0 0
Figure 4.17 shows the least-squares sums J, with and without weighting
function, for three different filter values wðx; yÞ. The optimisation problem
minwðx;yÞ ðJ Þ is solved by setting the first derivative with respect to the filter value to
zero. Thus
Z lty Z ltx
oJ
¼ 2ðzðl; vÞ wðx; yÞÞsðx l; y vÞdl dv ¼ 0: ð4:25Þ
owðx; yÞ 0 0
If Eq. (4.25) is solved for the filter value wðx; yÞ, the result is the correction
according to Eq. (4.21). If an indefinitely expanded topography is considered, the
result is the convolution integral according to Eq. (4.4).
The example illustrates the simplest case of regression filtering. ISO 16610 part
71, on the other hand, specifies a generalised approach
Z lty Z ltx X2 X2
minbh; jh ðx;yÞ qðzðl; vÞ j¼0
b
h¼0 h; jh
ðx; yÞðx lÞh ðy vÞjh
0 0
sðx l; y vÞdl dvÞ:
ð4:26Þ
The deviations between the topography and the unknown filter plane are initially
weighted with a loss function q: The filter value wðx; yÞ is expanded in the default
82 J. Seewig
Fig. 4.17 Three possible values for wðx; yÞ and the corresponding least square sum J. The
optimum for J with the lowest value is shown in the centre of the figure
case by a spatial variant second degree polynomial surface. For the filter value
ðx; yÞ ¼ b0 0 ðx; yÞ:The Gaussian function is used as the weighting function. This
results in the following filter properties.
• A square loss function for q leads to a linear filter.
• If for the loss function q the so-called bi-weight function is selected, the filter is
resistant to outliers.
• The polynomial surface replicates a Taylor series expansion of the topography
in the point ðx; yÞ. Therefore, form components can be approximated success-
fully with this filter.
• The filter has transfer properties comparable to those of the Gaussian filter, since
the Gaussian function is used as the weighting function.
• The filter works without running-in and running-out lengths, so there is no loss
of valuable measured values.
4.2.3.1 The Linear Gaussian Regression Planar Filter
The regression filter works as a linear filter when a square function is used for the
loss function q. Equation (4.26) can be ascribed to a convolution integral in this
case. For the weighting function of the second degree regression filter, the result
for planar filtering is
!!
2 2
x2 þ y2 1 p xc2þyk2
sðx; yÞ ¼ 2 p e c : ð4:27Þ
c2 k2c c2 k2c
kc is the cut-off wavelength at which amplitude damping of 50 % is reached. The

following applies to the filter parameter c ffi 0:7309 (refer to ISO 16610 part 71).
The axially symmetric weighting function is shown in Fig. 4.18.
Fig. 4.18 Weighting function of the linear Gaussian regression planar filter of second degree
A sectional plane in the direction of the x coordinate is plotted at y = 0 and the

course of the function is shown separately to clearly indicate the weighting
function. Compared to the Gaussian filter’s weighting function, negative function
values also occur. However, the function shape is dominated by the Gaussian
function. Unlike the Gaussian filter, the weighting function is not separable; so
regression filtering is numerically more complex. The axially symmetric weighting
function, independent of direction, ultimately results in the filter transfer function

k2 1 p c2 k2c2
SðkÞ ¼ 1 þ p c2 c2 e k : ð4:28Þ
k c2 k2c
The shape of the filter transfer function is shown in Fig. 4.19. For comparison
purposes, the graph also shows the transfer shape of the Gaussian filter as
described in Sect. 4.2.2.1. The regression filter slope is steeper than the Gaussian
filter slope in the transition zone, enabling ‘‘sharper’’ separation of the different
scale ranges.
Figure 4.20 shows an example of a surface with significant form components.
The linear Gaussian regression filter with a cut-off wavelength of kc = 0.8 mm was
used for filtering. The filter plane is valid over the entire topography and follows the
form component well (left image). The right image shows the difference between
the topography and the filter plane. The form is completely eliminated.
84 J. Seewig
Fig. 4.19 Filter transfer function of the Gaussian regression planar filter of second degree for
long- and short-wave components (the transfer function of the Gaussian filter for comparison is
shown in light gray)
Fig. 4.20 Filtration of a planar topography with the linear Gaussian regression filter. Left
measured topography with the filter plane. Right topography subtracted by the filter plane
4.2.3.2 The Linear Gaussian Regression Cylindrical Filter
The weighting function of the regression filter must be modified, as described in

Sect. 4.2.2.2, to be able to evaluate surface data of cylindrical components. With
the ordinate z for the axial direction and t for the circumferential direction, the
following applies to the weighting function
!! 2 2
2 t f
p z2 t2 fc2 fc p2 z2 þ 2c
sðz; tÞ ¼ 2 2 2 þ 2 e c kcz L
ð4:29Þ
c kcz L c2 kcz L
where kcz is the cut-off wavelength in the axial direction and L=fc the corre-
sponding cut-off wavelength in the circumferential direction, L is the circumfer-
ence of the cylinder and fc is the number of waves around the circumference. The
filter transfer function of the regression filter for measured data from cylindrical
components is
!! 2
k2cz
kcz f 2
2 f2 fc pc2
k2
þ2
Sð kz ; f Þ ¼ 1 þ pc 2
þ 2 2
e z fc ð4:30Þ
kz fc c kcz L
where kz is the wavelength in axial direction and f is the number of waves around
the circumference.
4.2.3.3 The Robust Gaussian Regression Filter
The robust Gaussian regression filter was introduced to be able to better filter
particularly plateau-type surfaces with valleys and peaks relevant to the function.
This includes, for example honed surfaces, laser-textured surfaces or metal matrix
composite (MMC) surface. With robust filters, the filter plane mostly follows the
plateau, thus providing an unequivocal basis for the evaluation of surface struc-
tures relevant to the function.
Figure 4.21 illustrates the behaviour of regression filters according to Eq. (4.26)
for various loss functions q. The shape of the function qLSQ , drawn in gray in
Fig. 4.21, increases exponentially (square) and is used with linear regression fil-
tering. The greater the deviation between topography and filter value, the more it
dominates the optimisation problem according to Eq. (4.26). The consequence is
that outliers (for example, pronounced valleys or peaks) significantly influence the
course of the filter plane. In this section the filter plane follows the ‘‘outliers’’ to
keep deviation between the topography and filter value to a minimum. In ISO
16610 part 71, the loss function qB according to Beaton (refer to Huber 2004) is
applied, such that deviations between topography and filter value above the
defined threshold c are only weighted constantly, while the function for lesser
deviations again progresses almost exponentially square. A careful choice of the
threshold c can nearly negate the effect of outliers on calculation of the filter plane
according to Eq. (4.26).
In ISO 16610 part 71 the threshold c is determined using
3
c ¼ pffiffiffi 1 DMAD : ð4:31Þ
2 erf 0; 5
Fig. 4.21 Robust and non-robust loss functions q

86 J. Seewig
The value DMAD is the median of the absolute values between the topography and
the unknown filter area and erf 1 is the inverse error function (Huber 2004). The
result of a Gaussian distributed deviation with the standard deviation r is
pffiffiffi
DMAD ¼ 2 r erf 1 0:5. Therefore, the threshold c is three times the standard
deviation.
Between the two loss functions, qLSQ and qB , there is now a substantial dif-
ference in regard to the numerical solvability of the system of equations according
to the Eq. (4.26). While qLSQ is a closed form expression, the loss function qB leads
to a non-linear system of equations that has to be solved iteratively. By taking into
consideration an additional weighting function wB the solution can be reached with
a re-weighted least-squares problem
Z lty Z ltx X2 X2
minbh;jh ðx;yÞ zðl; vÞ j¼0
b
h¼0 h;jh
ðx; yÞðx lÞh
0 0
2 w ðzðx; yÞ wðx; yÞÞ
ðy vÞjh B
sðx l; y vÞdl dv :
zðx; yÞ wðx; yÞ
ð4:32Þ
The function wB is referred to as an influence function, because it evaluates

topography values according to their amplitudes (Huber 2004). The function wB is
the result of the first derivative of the loss function and is defined according to ISO
16610 part 71 by
( u2 2
1 oqB ðuÞ u 1 for juj c :
wB ðu; cÞ ¼ ¼ c ð4:33Þ
2 ou 0 for juj [ c
Figure 4.22 shows the shape of the influence function. Figure 4.22 clearly
demonstrates that the influence function wB weights with zero amplitude values u
that are greater or less than the threshold c. Such values occur for example, in the
form of pronounced valleys or peaks and are thus not taken into consideration for
optimisation. In contrast, the influence function wLSQ weights topography values
linearly according to their amplitudes. A high amplitude leads to a respectively
high weight when optimising. Thus the square shape of the loss function wLSQ is
Fig. 4.22 Robust and non-robust influence functions wB and wLSQ respectively
Fig. 4.23 Left filtering of a structured surface with the linear Gaussian regression filter. Right the
same surface filtered with the robust Gaussian regression filter
not robust, because single extreme outliers strongly affect optimisation and the
position of the filter plane.
This non-robust behaviour is demonstrated with the simulated topography
shown in Fig. 4.23. The topography consists of a form component with an over-
lying peak in the centre. In the left hand image, the topography was filtered with a
linear Gaussian regression filter. The limiting scale is kc = 0.8 mm. The peak
significantly influences the position of the filter plane. The filter plane attempts to
follow the peak. When the difference between the topography and the filter
plane—or the S-F surface—is calculated, the peak is depicted as distorted. In
contrast, the robust Gaussian regression filter was used in the right hand image. In
this case the structure does not affect the position of the filter plane. The filter
plane conforms solely to the form component. The structure in the S-F surface
remains undistorted.
Figure 4.24 shows a metal matrix composite surface as a typical application for
the robust filter. A form component is overlaid. The intent is to extract only the
protruding structures to then characterise them with the aid of the parameters
specified in ISO 25178 part 2 (see Chaps. 2 and 3). A limiting scale of
kc = 0.25 mm was selected. The left hand image shows the topography with the
filter plane and the right hand image shows the extracted structures of the S-F
surface.
88 J. Seewig
Fig. 4.24 Robust filtering of a metal matrix composite surface
Fig. 4.25 Left cylindrical topography with form deviation and the groove. Right filter plane of
the robust Gaussian regression filter
The robust filter can be used for planar and cylindrical filtering. The left hand
image in Fig. 4.25 shows the topography of a shaft with a groove as an example of
the use of cylindrical filtering. The intention here is to approximate the long-wave
form deviation of the wave by the filter surface. The groove should not have any
effect on the position of the filter plane. As the limiting scale, kcz = 0.8 mm was
selected in the axial direction and fc = 15 upr (undulations per revolution) in the
circumferential direction. The right hand image shows the filter plane after use of
the robust Gaussian regression filter.
4.2.3.4 Filter Transfer Function and Recommendation for the Limiting

Scale of the Robust Gaussian Regression Filter
The robust Gaussian regression filter is non-linear and thus cannot be described
with a filter transfer function. The filter parameters kc , kcz and fc are referred to as
limiting scale and the filter parameter settings are a factor of the surface structure.
As a guideline, the limiting scale should be three times the greatest width of the
measured geometrical structure.
4.2.3.5 Digital Implementation of the Linear and Robust Gaussian

Regression Filter
To keep the use of the Gaussian regression filter simple, ISO 16610 part 71
specifies a definition only for the discrete case. Therefore, the user can convert the
standardised filter to source code with minimal effort.
4.2.4 Spline Filter
In shipbuilding, elastic strips were used in former times to approximate smooth

shapes of functions. The elastic strips, also called splines, were secured, for
example with nails. Mechanically, the resulting shape of the function wð xÞ can be
described as a bending beam that runs through defined points and whose bending
energy assumes a minimum. Mathematically, a function wð xÞ must be found for n-
fixation points ðxk ; zk Þ with k ¼ 1; . . .; n as well as for values xk that increase in a
monotonic manner. The function must solve the following optimisation problem:
Z xn 2 2 !
d w ð xÞ
minwðxÞ dx ; wðxk Þ ¼ zk ; k ¼ 1; ::; n: ð4:34Þ
x1 dx2
Solving the optimisation problem leads to cubic polynomial segments that run
continuously through the second derivative and pass directly through the knots
ðxk ; zk Þ. As a consequence of the mechanical beam bending model, the second
derivative of wð xÞ is zero at the edges x ¼ x1 and x ¼ xn , which is also referred to
as a natural boundary condition. Figure 4.26 illustrates the shape, for example for
n ¼ 9 fixation points.
Fig. 4.26 Interpolating cubic spline for n ¼ 9 fixation points

90 J. Seewig
Fig. 4.27 Smoothing spline for three different degrees of spring stiffness c
A smoothing spline is created when the elastic strip is not secured with nails but
instead the beam is suspended at the knots with elastic springs with stiffness c.
Depending on the spring stiffness, the strip progresses as an interpolating spline
ðc ! 1Þ, a smoothing function ð0\c\1Þ or, when all stiffness is removed, as a
straight line ðc ¼ 0Þ. Figure 4.27 shows the three smoothing splines discussed here.
The mechanical model leads to a mathematical variation problem, whereby the
spring flexibility c occurs as the Lagrange parameter k
Z xn 2 2 !
d wð xÞ Xn 2
minwðxÞ dx þ k ðz wðxk ÞÞ :
k¼1 k
ð4:35Þ
x1 dx2
Here too, the solution is piecewise cubic polynomials that again rise steadily
through the second derivative. Krystek (1996) applied the basic concept of the
smoothing spline as a way to adjust pairs of variates (de Boor 1978) to filtering of
measured surface data. This formed the foundation for the international standards
ISO 16610 part 22 (2012) for linear spline filtering and ISO 16610 part 32 (2009)
for robust spline filtering. Areal spline filters for evaluating measured topography
data are not yet available (but see Zhang et al. 2012). Spline filters provide filter
lines over the entire profile course and, in the linear case, lead to a roughness
profile without mean values. For spline filters, there are numerically-efficient
algorithms used in image processing, particularly to approximate the Gaussian
filter (D’Haeyer 1989). The variation problem described by Eq. (4.35) is expanded
such that the mechanical tension (first derivative) of the spline can be controlled by
an additional parameter b0 . The variation problem (4.35) is amended accordingly:
Z xn 2 2 Z xn !
d w ð xÞ dwð xÞ 2 Xn
0 0 2
minwðxÞ a dx þ b dx þ ðz wðxk ÞÞ :
i¼1 k
x1 dx2 x1 dx
ð4:36Þ
As shown in Sect. 4.2.4.1, solution of the variation problem leads to linear
filtering with characteristic spatially-dependent weighting functions. The tension
parameter b can be applied to decisively influence the course of the weighting

function. With a spline filter in accordance with Eq. (4.35), negative values occur
in the weighting function, as with the second degree Gaussian regression filter.
These lead to overshooting with discontinuous profile shapes and can be disrup-
tive. This effect can be diminished by applying the tension parameter b.
Equation (4.36) forms the foundation for the internationally standardised spline
filter described in ISO 16610 part 22 and part 32.
4.2.4.1 The Linear Spline Filter
Application of the variation problem given by Eq. (4.36) to equidistant sampled

profiles, results in the linear spline filter described in ISO 16610 part 22. This is
done by replacing the continuous derivatives with differential quotients. For the
natural boundary values (second derivative of wð xÞ at the edges x ¼ x1 and x ¼ xn
equal to zero), the result is

J ¼ b0 ðw2 w1 Þ2 þðwn wn1 Þ2 þ ðz1 w1 Þ2 þðzn wn Þ2
Xn1 ð4:37Þ
þ k¼2
a0 ðwkþ1 2wk þ wk1 Þ2 þb0 ðwk wk1 Þ2 þðzk wk Þ2 :
To obtain the minimum for the functional J, the first derivative is formed
according to the unknown values wk of the filter line and set to zero. A compact
representation of the resulting system of equations is obtained by writing the val-
ues wk of the filter line and the profile coordinates zk as column matrices w and z
0

1 þ b P þ a0 Q w ¼ z ð4:38Þ
with the matrices

0 1
1 2 1
0 1
1 1 B 2 5 4 1 C
B C
B 1 2 1 C B 1 4 6 4 1 C
B .. .. .. CC B .. .. .. .. .. C
P¼B
B . . .C ; Q ¼ B
B . . . .. C:
C
@ 1 2 1 A B 1 4 6 4 1 C
B C
1 1 @ 1 4 5 2 A
1 2 1
ð4:39Þ
Equation (4.37) demonstrates the filter behaviour for open profiles. The spline
filter for closed (periodic) profiles is obtained by continuing the first and last line in
P and continuing the first two and last two lines in Q
92 J. Seewig
0 1
6 4 1 1 4
0 1
2 1 1 B 4 6 4 1 1 C
B C
B 1 2 1 C B 1 4 6 4 1 C
B .. .. .. C ~ B
C .. .. .. .. .. C
~¼B
P . . .C ; Q ¼ B . . . .
. C:
B B C
@ 1 2 1 A B 1 4 6 4 1 C
B C
1 1 2 @ 1 1 4 6 4 A
4 1 1 4 6
ð4:40Þ
Specifically for 2\k\n 1, the result is a linear differential equation,
ð1 þ 2b0 þ 6a0 Þwk ðb0 þ 4a0 Þðwkþ1 þ wk1 Þ þ a0 ðwkþ2 þ wk2 Þ ¼ zk : ð4:41Þ
Taking into consideration the previously assumed equidistant profile points, how
the spatially discrete spline filter responds to sinusoidal profile shapes can now be
calculated, thus
2p yields 2p
zk ¼ a0 ei kx kDx ! wk ¼ a1 ei kx kDx : ð4:42Þ
Inserting the values for zk and wk in Eq. (4.40) provides the general filter transfer
function
a1 1
¼ : ð4:43Þ
a0 1 þ 4b0 sin2 p Dx þ 16a0 sin4 p Dx
kx kx
pffiffiffiffi
As shown in D’Haeyer (1989), b0 ¼ a0 results in a reasonable approximation
of the Gaussian filter. A similar approach is followed in ISO 16610 part 22. The
two parameters a0 and b0 are selected such that, for the cut-off wavelength kx ¼ kc ,
amplitude damping of 50 % results and the effect of the tension parameter b0 on
the filter transfer capacity is controllable. So, with a0 ¼ ð1 bÞa4 , b0 ¼ ba2 and
0 b 1, the discrete filter transfer function occurs, described in ISO 16610 part
22, for the long-wave profile component
a1 1 1
¼ ; a ¼ : ð4:44Þ
a0 1 þ 4ba2 sin2 p Dx þ 16ð1 bÞa4 sin4 p Dx 2 sin kpc Dx
kx kx
It is readily apparent that, for wavelengths kx ¼ kc , amplitude damping of 50 %

results, regardless of selection of the parameter b. Thus, a2 =a0 ¼ 1 a1 =a0
applies for the short-wave roughness profile. With this definition for the param-
eters a0 and b0 , and taking into consideration the Eq. (4.37), this is how the filter
line is calculated
1
w ¼ S z; S ¼ 1 þ ba2 P þ ð1 bÞa4 Q : ð4:45Þ
Fig. 4.28 Comparison of the Gaussian weighting function (grey solid line) and the weighting
function of the spline filter (dashed line) for b ¼ 0, b ¼ 0:625242 and b ¼ 1. The weighting
function for the middle of the profile was examined
It is evident that the filter line w of the spline filter is the result of multiplication
between the matrix S and the column vector z of the profile coordinates. Each line
of the matrix S can be construed as a spatially-dependent characteristic weighting
function of the spline filter. Figure 4.28 shows an example of the weighting
function of the spline filter for b ¼ 0 (no mechanical tension), b ¼ 0:625242
(‘‘magic number’’ described in ISO 16610 part 22) and b ¼ 1 (maximum
mechanical tension) around the centre of the profile. The Gaussian weighting
function is shown as a gray solid line for comparison purposes.
If the so-called ‘‘magic number’’ is selected for b, the spline filter comes close
to the filter behaviour of the Gaussian filter. The D’Haeyer approach results in a
pffiffiffi
very similar value: b ¼ 5 1 =2 ffi 0:618034. Figure 4.29 shows the devia-
tions of the filter transfer function of the Gaussian filter for the three previously
selected values of b. The amplitude is less than 5 % around the ‘‘magic number.’’
Fig. 4.29 Deviations of the filter transfer function of the spline filter from the filter transfer
function of the Gaussian filter. The course for b ¼ 0, b ¼ 0:625242 (‘‘magic number’’) and b ¼ 1
are shown
94 J. Seewig
4.2.4.2 The Robust Spline Filter
A simple modification to the functional J according to Eq. (4.36), results in the

robust spline filter, which reacts less sensitively to outliers in the surface profile.
Instead of the least-square sums between the profile coordinates zk and the
unknown filter values wk , the absolute values are now examined. By taking into
consideration a rescaling constant c, the dimension of the differential equation
remains unchanged. Thus

J ¼ b0 ðw2 w1 Þ2 þðwn wn1 Þ2 þ cjz1 w1 j þ cjzn wn j
Xn1 ð4:46Þ
0 2 0 2
þ k¼2
a ð wkþ1 2wk þ wk1 Þ þb ð w k wk1 Þ þc j z k wk j :
Here, too, the functional is derived according to the unknown values wk of the
filter line and set to zero. Following the notation described in Eq. (4.37), the
following applies to the robust spline filter

0 0 1 if u 0
ðb P þ a QÞw ¼ c sgnðz wÞ; sgnðuÞ ¼ ð4:47Þ
1 if u\0
or with a0 ¼ ð1 bÞa4 , b0 ¼ b a2 and 0 b 1

1 1
w ¼ b a2 P þ ð1 bÞa4 Q c sgnðz wÞ; a ¼ : ð4:48Þ
p
2sin kc Dx
The matrices P and Q should be selected for an open profile described by Eq.
(4.38) and for a closed profile described by Eq. (4.39). Finally, it should be noted
that the system of Eq. (4.47) is non-linear and must be solved iteratively. The
constant c should be selected such that it reacts proportionally to the amplitude of
the filter line. This is the only way to ensure that the filter transfer capacity of the
robust spline filter is not dependent on rescaling of the profiles. For a sinusoidal
profile with the wavelength kx ¼ kc , for example, selecting
p 1 Xn
c¼ jz wk j;
k¼1 k
ð4:49Þ
2n
leads to amplitude damping of 50 %, regardless of the amplitude of the profile
coordinates zk . However, because it is non-linear, no generally applicable filter
transfer function can be stated for the robust spline filter.
4.3 Morphological Filters
Morphological filtering (Srinivasan 1998, Soille 2003) is based on a geometric

element, referred to as a structuring element (SE), which is moved tangentially
along the surface. The centre form of the structuring element forms the filter plane.
A morphological operation is called dilation when the material-free side of the

topography is contacted. Contacting the topography on the side filled with material
is called erosion. Two structuring elements are defined in the ISO 16610 series: (1)
the spherical ball and (2) the plane element for surface filtering. However, mor-
phological operators are currently available only for profiles in ISO 16610 part 41
(2012) and ISO 16610 part 49 (2012). Discs and line segments are defined as
structuring elements in this chapter. Figure 4.30 shows the fundamental operations
of dilation and erosion on a profile section of the topography and on a disc as the
structuring element. The top of Fig. 4.30 shows dilation and the bottom erosion.
The radius of the disc is specified as the parameter of the operator.
If dilation is performed and followed by erosion, the process is referred to as a
morphological closing filter. A morphological opening filter is erosion followed by
dilation. The results of a closing filter and an opening filter are illustrated in
Fig. 4.31. The closing filter leads to an envelope above the topography, while the
opening filter leads to an envelope below the topography. The closing filter is used
for example, for tactile measurement of topographies. Each tip performs a dilation
based on the radius of its tip. The measurement dataset can then be ‘‘corrected’’ by
an erosion process.
The filtering effects of closing and opening are demonstrated with a simple
sample profile. A line element is used as the structuring element. A profile with
elevations is shown in Fig. 4.32. Two structuring elements of different lengths,
Fig. 4.30 Morphological operation with a disc as the structuring element (dilation above and
erosion below)
96 J. Seewig
Fig. 4.31 Morphological closing (above) and opening (below) with a disc as the structuring
element
SE1 and SE2, are used for the closing and opening filters. SE1 is somewhat shorter
than SE2, which is longer than the width of the elevation.
The closing filter is used on the left hand side of Fig. 4.32 and the opening filter
on the right hand side. It can be seen that the profile structure of (a), (b) and (c)
remains unchanged after filtering. Only in the case of (d) is the step completely
eliminated by SE2. Something different happens when a depth of the same width
Fig. 4.32 Morphological closing and opening of a height step with structuring elements of
different size
Fig. 4.33 Morphological closing and opening of a depth with structuring elements of different
size
W is filtered with the morphological filter (Fig. 4.33). The cases (a), (c) and (d)
remain unchanged, while the depth is completely filtered out in (b).
The closing operator, or opening filter, acts as a sieve. During closing, all of the
elevations with structure widths greater than the length of the structuring element
remain in the sieve. During opening, depths with widths greater than the length of
the structuring element remain in the sieve. Using a combination, for example, a
closing filter followed by an opening filter, elevations and depths that exceed a
certain width can be removed. This process is referred to as an M sieve. When this
type of filtering is performed sequentially with different rescaled structuring ele-
ments, it is referred to as an alternating sequence filter. This type of profile filtering
is described in ISO 16610 part 49, with the title ‘‘scale space techniques’’. This
enables extraction of structures of a certain size and for example, statement of their
relative spatial frequency. Decomposition of a profile into certain scale spaces is
shown in Fig. 4.34. Decomposition begins with the original profile S0 . A struc-
turing element with the smallest scale ‘‘1’’ is selected, then sieving is performed
with M1 . The result is the signal S1 , which contains only structures with scale
ranges greater than the structures of scale ‘‘1.’’ The ‘‘filtered out’’ structures,
designated as signal d1 , can be determined by calculating the difference between
S0 and S1 . The next scale, ‘‘2,’’ is sieved with M2 and produces the two signals S2
and d2 . Sieving is continued up to a largest scale.
4.4 Multi-Resolution Analysis by Wavelets
The concept of multi-resolution provides a tool to break down signals into different
wavebands, to analyse the wavebands and to reconstruct the signal by an inverse
transform. Wavelet transformation is one possibility of multi-resolution. ISO
98 J. Seewig
Fig. 4.34 Scale space decomposition of a profile
16610 part 29 (2006) defines interpolating spline wavelets for the evaluation of
topography data. To keep the mathematical description simple, the wavelet
transformation for profile sections is described in this chapter. Expanding to
include areal topography datasets is simple (refer to Fernandez et al. (1996); Jiang
et al. (2000)).
4.4.1 Fourier Transform
With the aid of the Fourier transform, a profile zðxÞ is approximated by super-
imposing cosine and sine waves of different amplitudes Zðf Þ and frequencies f (or
wavelengths). Mathematically, this can be ascribed to the following expression
Z 1 X1
zð x Þ ¼ Z ð f Þei 2p fx df ffi k¼1
Z ðk Df Þei 2p k Dfx Df : ð4:50Þ
1
The cosine and sine portions are expressed as a complex exponential function, i.e.
ei 2p fx ¼ cosð2p fxÞ þ isinð2p fxÞ. The frequency-dependent amplitudes Zðf Þ, also
referred to as the spectrum, are calculated with the Fourier transform
Z 1
Zð f Þ ¼ zð xÞei 2p fx dx: ð4:51Þ
1
Equation (4.50) is a so-called integral transform and effectively calculates how

similar the profile shape and a cosine or sine wave of a certain frequency are to one
another. If the match is good, the amplitude for Zðf Þ is particularly high. This
means that the Fourier transform is a good way to detect periodic surface devia-
tions. However, the Fourier transform is not so useful if a single topographic
feature such as a groove or a peak is to be detected in the profile. In this case,
surface deviations can be approximated only by superimposing many cosine and
sine components. The result is a wide range of amplitudes. Therefore, single
topographic features are difficult to characterise with a Fourier transform.
Figure 4.35 illustrates the use of the Fourier transform. The upper graph shows a
purely sinusoidal profile, while the lower shows a profile with a groove. The
corresponding spectrums (amplitudes) are shown on the right hand side.
Another restriction of the Fourier transform is that the spectrum makes it dif-
ficult to localise a single incidence of a topography dataset or profile dataset, i.e.,
effectively the Fourier transform does not have a spatial resolution. If, for example,
the groove in Fig. 4.35 is in a different place, the amplitudes in the Fourier
spectrum remain unchanged. Only the phase gives (poor) information about the
position. Therefore, the short-time Fourier transform (STFT) is applied to achieve
a spatial resolution. This is done by expanding Eq. (4.50) in such a way that
the profile is only examined locally using a spatially sliding window function
wðx x0 Þ (refer to Fig. 4.36). Then for each position x0 in the window, a spectrum
is given by
Z 1
Z ðf ; x 0 Þ ¼ zð xÞwðx x0 Þei 2p fx dx: ð4:52Þ
1
Fig. 4.35 Fourier transform of a sinusoidal signal (above) and a groove (below)
100 J. Seewig
Fig. 4.36 Sliding Gaussian window function to calculate the short-time Fourier transform
Under certain conditions, it can be shown that a Gaussian window function

offers the best resolution of attributes, or of the position and the frequency. The
STFT with a Gaussian window function is referred to as the Gabor transform, and
is closely related to the wavelet transform.
4.4.2 Continuous Wavelet Transform
As with the Fourier transform, the continuous wavelet transform (CWT) is an

integral transform and is defined by the following expression
Z 1
1 xb
Ww ða; bÞ ¼ zð xÞ pffiffiffi w dx: ð4:53Þ
1 a a
The function wðxÞ is called a wavelet. Reconstruction of the profile zðxÞ from the
wavelet transform Ww ða; bÞ is achieved by applying the inverse wavelet transform.
If Eq. (4.52) is interpreted as a type of convolution (refer to Eq. (4.4)), a wavelet
behaves as a band-pass filter that suppresses low and high frequencies in the
profile. The parameter a enables adjustment of the pass-band range of the band-
pass. There are many different wavelets with which a transformation can be per-
formed properly. Figure 4.37 shows an example of the well-known Mexican hat
wavelet.
The wavelet is similar to a window function and has significant function values
not equal to zero only within a certain range. Applying Eq. (4.52), the profile is
examined for similarity to the wavelet used. The wavelet can be positioned with
Fig. 4.37 The Mexican hat wavelet

Fig. 4.38 Mexican hat wavelet with different translation and dilation parameters
the translation parameter b on the abscissa and the width of the wavelet can be
pffiffiffi
influenced with the dilation parameter a. The pre-factor 1= a ensures that the
square area below the wavelet remains constant. Figure
4.38 shows the Mexican
hat wavelet for the parameter pair ða; bÞ ¼ ð1; 0Þ; 12 ; 2 ; ð1=4; 1Þ.
What is decisive now is to be able to adapt the width and position of the wavelet
to certain attributes of the profile by varying the parameters a and b. If profile
attributes resemble the wavelet, a largevalue results
for the amount of the wavelet
transform for the parameter pair ða; bÞ Ww ða; bÞ. If there is no similarity between

wavelet
and
profile attribute, the result is a correspondingly small value
Ww ða; bÞ. So the wavelet transform Ww ða; bÞ indicates how closely the profile
and the scaled and shifted wavelet correlate. The dilation parameter a can be
considered a type of frequency parameter. The smaller the value for a, the nar-
rower the wavelet, and high-frequency profile components can be better approx-
imated. Thus wavelets can be used to locally identify profile attributes with
varying frequency components. The STFT has this attribute, because the window
width remains constant and localisation of very short-wave or long-wave profile
attributes is not possible.
Figure 4.39 shows the CWT of a simulated signal. In this special case, two
Mexican hats were used as the signal to achieve optimum correlation with the
CWT. The first function was positioned at x = 1 mm with rescaling of a ¼ 0:5:
The point of symmetry of the second function is at x = -3 mm with rescaling of
a ¼ 0:1: The CWT is typically shown as a colour-coded image. In the example, the
grey values indicate the amplitude of the CWT depending on the translation and
dilation parameter. It is clear that optimum correlation (white area) is achieved
when the transform wavelet corresponds exactly to the wavelet in the signal. The
position and rescaling can be found in Fig. 4.39. If the transform is applied to a
real measured topography, a wavelet would be chosen that, for example, can
approximate function-critical attributes in the signal. With the CWT, these attri-
butes can be detected by locating the white areas and characterising them by their
shape. For dimensional metrology, ISO specifies interpolating spline wavelets as
the default case; many different manufacturing processes can be characterised with
these wavelets. The interpolating spline wavelets described in ISO 16610 part 29
102 J. Seewig
Fig. 4.39 CWT of a signal with a Mexican hat wavelet. The signal is a superimposition of two
translated and scaled Mexican hats
were first introduced by Fernandez and Sweldens in their software package

LIFTPACK (Fernandez et al. 1996) and later, Jiang et al. introduced the inter-
polating spline wavelets into surface measurement (see Jiang et al. 2000).
4.4.3 Discrete Wavelet Transform and Multi-Resolution
Section 4.4.2 introduced the continuous wavelet transform (CWT) with the con-
tinuous parameters a and b. If the profile is scanned with the increment Dx, it
makes sense to choose the translation parameter b such that the wavelet is always
shifted by a multiple of the scan increments. The same procedure should be chosen
for the dilation parameter a. This parameter is usually increased by a factor of two.
This definition leads to the spatially discrete wavelet transform for the scanned
profiles
X pffiffiffiffiffiffiffi
Ww ðj; kÞ ¼ n
zðnDxÞ 2j w n2j k Dx : ð4:54Þ
Equation (4.53) is a discretisation of the CWT. The numerically efficient imple-

mentation that is often used technically is the discrete wavelet transform (DWT). It
has been shown (refer to Mallat 1999) that translation and dilation of the wavelets
can be performed with filter banks. Each filter bank consists of a linear high-pass
(HP) and a low-pass filter (TP). The required filter coefficients are a factor of the
selected wavelet. Figure 4.40 is intended to illustrate wavelet analysis with the aid
of a filter bank. The analysis begins with the scanned profile zðxÞ, designated as
Fig. 4.40 Filter bank to calculate the DWT and IDWT of a signal
Fig. 4.41 DWT in the frequency domain
signal S0 . The signal S0 passes through the first filter bank. The HP filtered signal
d1 corresponds to the wavelet coefficients of the first dilation, referred to as scale
space j ¼ 1. The wavelets are narrowest here. Every coefficient in d1 specifies the
amplitude and position of the wavelet that is compared to the profile. The TP
filtered signal S1 then progresses to the next filter stage. The result is the wavelet
coefficients d2 of the second scale space j ¼ 2. The corresponding wavelet that is
compared to the signal is now twice as wide. Decomposition continues only until
wavelets are examined that are completely above the profile, making further
widening pointless. With the aid of the reconstructing filter banks, the profile can
finally be completely reconstructed from the coefficients. DWT complementary to
the HP and TP filters are needed. This transform is called the inverse discrete
wavelet transform (IDWT). If, for example, the profile was decomposed into n
scale spaces, the signals Sn ; dn ; d n1 ; . . .; d 2 ; d1 are needed for reconstruction. An
efficient way to calculate the DWT is the so-called lifting scheme, which is also
used in ISO 16610 part 29. The lifting scheme is characterised particularly by the
fact that no additional memory is needed to calculate the coefficient.
The wavelet coefficients d j of the scale spaces j [ 1 correspond to band-pass
filtering; they are calculated by low-pass filtering followed by high-pass filtering.
104 J. Seewig
Fig. 4.42 Relationship between the wavelet coefficients, and the roughness and waviness
Fig. 4.43 Decomposition of a turned profile in waviness ðj ¼ 3. . .8Þ and roughness ðj ¼ 8. . .11Þ
The band-pass filters have a characteristic transfer function according to their scale
space. Figure 4.41 qualitatively shows decomposition of the profile by the DWT as
the filter transfer function jHðf Þj. Decomposition to different wavebands is referred
to as multi-resolution.
The decomposition enables function-critical attributes in the profile shape to be
selectively reconstructed. The respective scale spaces are selected either from a
priori knowledge or by analysing the amplitudes of the wavelet coefficients, then
reconstruction is as shown in Fig. 4.40. It is also conceivable to decompose the

deviations in a surface to roughness and waviness as suggested in ISO 4287 (2000)
by specifying wavelength bands. The roughness is extracted by a band-pass with
the short-wave cut-off wavelength ks and the long-wave cut-off wavelength kc . For
waviness, a band-pass with the short-wave cut-off wavelength kc and the long-
wave cut-off wavelength kf is chosen. Figure 4.42 illustrates implementation with
the aid of the wavelet transform. The respective cut-off wavelengths can be freely
defined by selecting the assigned scale spaces.
Figure 4.43 shows an example of the analysis of a profile of a turned surface
(top image). The central image takes into consideration the scale spaces 11–8 for
With a scan increment of Dx ¼ 0:7 lm; the cut-off wavelengths are kf ¼
211 2Dx ¼ 2:8 mm and kc ¼ 28 2Dx ¼ 0:36 mm. The short-wave cut-off lengths
ks ¼ 23 2Dx ¼ 11 lm apply to extraction of the roughness.
4.5 Summary
This chapter has introduced the areal Gaussian filter presented in ISO 16610 part
61 and the robust areal Gaussian regression filter described in ISO 16610 part 71.
The filter properties were explained mathematically and demonstrated using
examples. Appropriate modification of the weighting function of the areal
Gaussian filter results in filtering without loss of valuable topography points. In
contract, the robust areal Gaussian regression filter described in ISO 16610 part 71
always supplies a filter plane for the entire topography. Finally, an outlook was
presented for spline filters (ISO 16610 part 22 and part 32), morphological filters
(ISO 16610 part 41 and part 49) and wavelet filters (ISO 16610 part 29), available
only for profiles so far.
Filters enable separation of surface deviations. Sensible characterisation of the
topography is strongly dependent on the right selection of the filter type and on the
nesting index. The suitability of the selected filter for the intended evaluation
strategy must always be thoroughly and critically examined.
References
D’Haeyer JPF (1989) Gaussian filtering of images: a regularization approach. Sig Proc
18:169–181
Fernàndez G, Periaswamy S, Sweldens W (1996) LIFTPACK: a software package for wavelet
transforms using lifting. Proc SPIE 2825:396–408
Huber PJ (2004) Robust statistics. Wiley, New York
nominal characteristics of contact (stylus) instruments. International organization of
standardization
106 J. Seewig
rules and procedures for the assessment of surface texture. International organization of
standardization
terms, definitions and surface texture parameters. International organization of standardization
Standardization
3: specification operators.International Organization for Standardization
ISO/CD 16610 part 61 (2012) Geometrical product specifications (GPS)—linear areal filters:
Gaussian filters.International Organization for Standardization
ISO/CD 16610 part 71 (2011) Geometrical product specifications (GPS)—robust areal filters:
Gaussian regression filters. International Organization for Standardization
ISO/DIS 16610 part 41 (2012) Geometrical product specifications (GPS)—morphological profile
filters: disk and horizontal line-segment filters. International Organization for Standardization
ISO/DIS 16610 part 49 (2012) Geometrical product specifications (GPS)—morphological profile
filters: scale space techniques. International Organization for Standardization
ISO/TS 16610 part 22 (2012) Geometrical product specifications (GPS)—linear profile filters:
spline filters. International Organization for Standardization
ISO/TS 16610 part 28 (2010) Geometrical product specifications (GPS)—profile filters: end
effects. International Organization for Standardization
ISO/TS 16610 part 29 (2006) Geometrical product specifications (GPS)—linear profile filters:
spline wavelets. International Organization for Standardization
ISO/TS 16610 part 32 (2009) Geometrical product specifications (GPS)—robust profile filters:
spline filters. International Organization for Standardization
Jiang XQ, Blunt LA, Stout KJ (2000) Development of a lifting wavelet representation for surface
characterization. Proc R Soc Lond A 456:2283–2313
Krystek M (1996) Form filtering by splines. Measurement 18:9–15
Krystek M (2005) The digital implementation of the Gaussian profile filter according to ISO
11562. Beuth-Verlag, Berlin
Mallat S (1999) A wavelet tour of signal processing. Academic Press, New York
Seewig J (1999) Praxisgerechte Signalverarbeitung zur Trennung der Gestaltabweichungen
technischer Oberflächen. Shaker Verlag GmbH
Seewig J (2005) Linear and robust Gaussian regression filters. J Phys: Conf Series 13:254–257
Soille P (2003) Morphological image analysis: principles and applications. Springer, Berlin
Srinivasan V (1998) Discrete morphological filters for metrology. Proceedings of 6th ISMQC
IMEKO, Wien, Austria, Sep 1998, pp 623–628
Zhang H, Yuan Y, Piao W (2012) The spline filter: a regularization approach for the Gaussian
filter. Precis Engineering 36:586–592
Chapter 5
Areal Form Removal
Alistair B. Forbes
Abstract This chapter deals with areal form removal, one of the basic operations
in areal surface texture analysis, and relies on determining an associated feature
from a set of data points, the primary extracted surface. The three main compo-
nents for form removal are discussed: measures of distance from a point to a
surface (or ideal associated feature), fitting criteria to determine the best-fit sur-
face, and the form removal operator. Special issues associated with areal form
removal, with detailed description of plane and spherical form removal are
reviewed. Approaches to removing more general forms are also discussed.
5.1 Introduction
Form removal is one of the operations to be performed in areal surface texture

assessment ISO 25178 part 2 (2012), ISO 25178 part 3 (2012). ISO 25178 part 2
(2012) defines the F-operator as the ‘‘operation which removes form from the
primary surface’’. Thus, the removal of form can be regarded as an operator
x ¼ F ðxÞ, F : R3 7! R3 , acting on points in 3D to allow areal parameters to be

evaluated on the modified points in order to minimise the influence of form on
these parameters. In practice, the F-operator is defined in terms of measured point
coordinates. The input to determining the form removal operation is the primary
extracted surface, a finite set of points sampled from the primary surface. If the
primary surface is associated with a particular geometric form, such as a sphere,
the F-operation removes the form from the primary surface, so that the resulting
S-F-surface is planar in the sense that repeating the form removal process, in
theory at least, would leave the S-F-surface unchanged.
A. B. Forbes (&)
National Physical Laboratory, Hampton Road,
e-mail: Alistair.Forbes@npl.co.uk

108 A. B. Forbes
Form removal has three main components. The first is the choice of model
surface, for example, geometric element, polynomial surface, tensor product
spline, to represent the form. The second is the criterion used to fit the form to the
measured data (the primary extracted surface), i.e., used to determine the associ-
ated feature. This component generally has two sub-components, the first relating
to how the distance from a measured point to the model surface is defined, the
second relating to how the distances for all the measured data points are aggre-
gated. The third component relates to how the fitted model surface is subtracted
from the data so that subsequently areal surface parameters can be evaluated.
Often the third component is a byproduct of the fitting process but this is not
necessarily so.
Section 5.2 describes ways of specifying the form, while Sect. 5.3 discusses
measures of distance from a point to a surface. Fitting criteria are discussed in
Sect. 5.4 while special features associated with areal form removal are discussed
in Sect. 5.5. Summary and concluding remarks are given in Sect. 5.6.
5.2 Form Model Definition
5.2.1 Functionally Defined Surfaces
A functionally defined surface describes the height of the surface, the z-compo-
nent, as a function of x and y:
z ¼ f ðx; y; aÞ; ð5:1Þ
for example, a plane is defined as
z ¼ f ðx; y; aÞ ¼ ax þ by þ c; a ¼ ða; b; cÞT : ð5:2Þ
This functional form of the definition of a surface is particularly useful as it

enables a simple form removal process.
5.2.2 Algebraically Defined Surfaces
An algebraically defined surface is the locus of points x such that
Fðx; aÞ ¼ 0: ð5:3Þ
For example a sphere of radius r0 centred at x0 ¼ ðx0 ; y0 ; z0 ÞT is defined alge-

braically with
Fðx; aÞ ¼ ðx x0 Þ2 þ ðy y0 Þ2 þ ðz z0 Þ2 r02 ¼ 0; a ¼ ðx0 ; y0 ; z0 ; r0 ÞT :

5 Areal Form Removal 109
The sphere is a special case of a quadric surface, the general case given by an
equation of the form
a1 x2 þ a2 y2 þ a3 z2 þ a4 xy þ a5 xz þ a6 yz þ a7 x þ a8 y þ a9 z þ a10 ¼ 0:
A single constraint is required on the faj g to define a parametrisation of the

surface. Often a10 is set to 1 but other constraints may be appropriate. For a patch
of a quadric surface that approximates the plane z ¼ 0, it is appropriate to set
a9 ¼ 1. This allows, with some algebraic manipulation, the surface to defined
functionally as z ¼ f ðx; y; aÞ. An example of this approach is given in Sect. 5.5.2.
For a non-degenerate quadric surface such as an ellipsoid, the nine parameters
define the centre location (three parameters), orientation (three parameters) and
shape (three semi-axis lengths).
5.2.3 Parametrically Defined Surfaces
For a parametrically defined surface, the coordinates ðf ; g; hÞ of a point lying on

the surface are defined in terms of two patch or footpoint parameters, u and v:
u 7! f ðu; aÞ ¼ ðf ðu; aÞ; gðu; aÞ; hðu; aÞÞT ; u ¼ ðu; vÞT :
For example, a cylinder whose axis is aligned with the z-axis is given paramet-
rically as
2 3
r0 cos u
6 7
u 7! f 0 ðu; r0 Þ ¼ 4 r0 sin u 5:
v
A cylinder in an arbitrary position can be specified as

2 3
x0
6 7
u 7! f ðu; aÞ ¼ x0 þ RT0 RTy ðbÞRTx ðaÞ f 0 ðu; r0 Þ; x0 ¼ RT0 4 y0 5;
0
involving five parameters a ¼ ðx0 ; y0 ; a; b; r0 Þ. Here Rx ðaÞ is a rotation through

angle a about the x-axis, Ry ðbÞ similarly represents a rotation about the y-axis,
2 3 2 3
1 0 0 cos b 0 sin b
6 7 6 7
Rx ðaÞ ¼ 4 0 cos a sin a 5; Ry ðbÞ ¼ 4 0 1 0 5 ð5:4Þ
0 sin a cos a sin b 0 cos b
and R0 is a fixed rotation matrix.

110 A. B. Forbes
The cylinder is an example of one type of surface of revolution that has the
general form
2 3
rðv; a0 Þ cos u
6 7
u 7! f 0 ðu; a0 Þ ¼ 4 rðv; a0 Þ sin u 5 ð5:5Þ
v
in standard position, with generator v 7! rðv; a0 Þ depending on parameters a0 .

Here rðv; a0 Þ is the radius of the surface at height v. Such a surface in an arbitrary
position can be described as
2 3
x0
T T T T6 7
u 7! f ðu; aÞ ¼ x0 þ R0 Ry ðbÞRx ðaÞ f 0 ðu; a0 Þ; x0 ¼ R0 4 y0 5: ð5:6Þ
z0
A second type of surface of revolution in standard position is given by
u 7! f 0 ðu; a0 Þ ¼ ðv cos u; v sin u; zðv; a0 ÞÞT ð5:7Þ
with generator v 7! zðv; a0 Þ where zðv; a0 Þ is the height at radius v. Aspheric

surfaces are often specified in this form.
An elliptic hyperboloid is standard position is given by
2 3
a0 cos u cosh v
6 7
u 7! f 0 ðu; a0 Þ ¼ 4 b0 sin u sinh v 5
c0 sinh v
is a simple surface that generally has no translational or rotational symmetry.

Many freeform surfaces are defined in terms of nonuniform rational B-splines
(NURBS); see, for example, Piegl and Tiller (1996).
A functionally defined surface (5.1) can also be described parametrically:
u 7! ðu; v; f ðu; v; aÞÞT :
Conversely, a near planar patch of a parametric surface that approximates the

plane z ¼ 0 can be described functionally z ¼ f ðx; y; aÞ, by solving x ¼ f ðu; v; aÞ
and y ¼ gðu; v; aÞ to describe u ¼ uðx; yÞ and v ¼ vðx; yÞ as functions of x and y.
5.3 Measures of Distance from a Point to a Surface
Almost all form removal algorithms involve fitting a surface to coordinate data,
minimising an aggregate measure of the closeness of the data points to the fitted
surface. This aggregate measure usually is built up from some measure of the
distance of a point to the surface. There are a number of approaches that can be
adopted.
5.3.1 Difference in Height
For functional surfaces defined as z ¼ f ðx; y; aÞ, the signed distance dðxi ; aÞ of a
data point xi ¼ ðxi ; yi ; zi ÞT to the surface is often measured as
dðxi ; aÞ ¼ zi f ðxi ; yi ; aÞ;
so in the case of a plane (5.2), for example,
dðxi ; aÞ ¼ zi axi byi c; ð5:8Þ
a linear function of the parameters a. Thus, the difference is measured along a

particular direction, the z-axis, associated with the coordinate system. For this
reason, this measure is not invariant with respect to changes in the coordinate
frame of reference.
5.3.2 Algebraic Distance
For surfaces defined algebraically, Sect. 5.2.2, the algebraic (signed) distance
dA ðxi ; aÞ from xi to the surface Fðx; aÞ ¼ 0 is measured by dA ðxi ; aÞ ¼ Fðxi ; aÞ.
Thus for a sphere
Fðxi ; aÞ ¼ ðxi x0 Þ2 þ ðyi y0 Þ2 þ ðzi z0 Þ2 r02 ;

¼ ri2 r02 ¼ ðri r0 Þðri þ r0 Þ;
where
ri ¼ ½ðxi x0 Þ2 þ ðyi y0 Þ2 þ ðzi z0 Þ2 1=2 ; ð5:9Þ
is the distance from the point xi to the sphere centre.
5.3.3 Distance Related to a Surface Feature
For a surface of revolution as in (5.6) it may be convenient to measure distance

orthogonal to the axis of revolution. If the surface is in standard position as in
(5.5), then the distance dðxi ; a0 Þ is given by
112 A. B. Forbes
dðxi ; a0 Þ ¼ ri rðzi ; a0 Þ; ri ¼ ðx2i þ y2i Þ1=2 :
Similarly for a surface of revolution as in (5.7) in standard position,
dðx; a0 Þ ¼ zi zðri ; a0 Þ; ri ¼ ðx2i þ y2i Þ1=2 :
If an axis is specified as x0 þ tn, where x0 is a point on the axis and n is a unit

vector pointing along the axis, then
r ¼ jjðx x0 Þ njj
is the distance from x to the axis and
z ¼ ðx x0 ÞT n
is the (signed) height of x above the plane passing through x0 with normal
vector n.
5.3.4 Orthogonal Distance from a Point to a Surface
Suppose x is close to a parametrically defined surface u 7! f ðu; bÞ, and let u

solve the footpoint problem
minðx f ðu; aÞÞT ðx f ðu; aÞÞ; ð5:10Þ

u
so that u ¼ u ðaÞ specifies the point f ¼ f ðu ; aÞ on f ðu; aÞ closest to x. Let

n ¼ nðaÞ be the normal to the surface at f , likewise a function of a, and set
dðx; aÞ ¼ ðx f ÞT n ¼ ðx f ðu ðaÞ; aÞÞT nðaÞ: ð5:11Þ
Then dðx; aÞ is the signed orthogonal distance of x from f ðu; aÞ, where the sign is
consistent with the convention for choosing the surface normal. Furthermore,
T
@d @f
¼ n; k ¼ 1; . . .; n; ð5:12Þ
@aj @aj
where all terms on the lefthand side are evaluated at u . For standard geometric
elements, the distance function dðx; aÞ can be defined as an explicit function of the
parameters, as in the case of a sphere above, but for free form surfaces, the optimal
footpoint parameters u have to be determined using numerical techniques
(Anderson et al. 1998; Bartholomew-Biggs et al. 2000; Forbes 2006; Forbes and
Minh 2011; Helfrich and Zwick 1995). For example, Forbes (2006) gives a
compact algorithm for determining the optimal footpoint parameters using a
simple sequential quadratic programming algorithm.
For a surface described functionally as in (5.1), the distance dðx; aÞ from a point
x ¼ ðx; y; zÞT to the surface is approximated by
z f ðx; y; aÞ @f @f
dðx; aÞ 1=2 ; fx ¼ ðx; y; aÞ; fy ¼ ðx; y; aÞ;
@x @y
1 þ fx2 þ fy2
with the approximation becoming better the closer x is to the surface. Similarly,
for a surface defined algebraically as in (5.3),
Fðx; y; aÞ @F @F
dðx; aÞ 1=2 ; Fx ¼ ðx; y; aÞ; Fy ¼ ðx; y; aÞ:
@x @y
Fx2 þ Fy2
5.4 Fitting Criteria
Suppose that a measure of the distance dðxi ; aÞ from a data point xi , i ¼ 1; . . .; m,

to a surface defined by parameters a has been defined. The best-fit surface to the
data is defined by minimising some aggregate measure EðaÞ of the goodness of fit
of the surface to the data involving the distances dðxi ; aÞ. For example, the
aggregate measure Ep ðaÞ can be defined in terms of p-norms:
" #1=p
X
m
Ep ðaÞ ¼ dip ðxi ; aÞ :
i¼1
By far the most common choices for p are p ¼ 2, least-squares or L2 approximation:

X
m
min d 2 ðxi ; aÞ;
a
i¼1
p ¼ 1, L1 approximation:
X
m
min jdðxi ; aÞj; ð5:13Þ
a
i¼1
and p ¼ 1, Chebyshev approximation:
min max jdðxi ; aÞj: ð5:14Þ

a i
These three choices of fitting criteria can be related to discretised versions of the
areal surface parameters Sq (p ¼ 2), Sa (p ¼ 1) and Sz (p ¼ 1). Thus, form
removal according to a least-squares criterion will have the effect of minimising
the subsequent evaluated Sq parameter value. In other words, the least-squares
114 A. B. Forbes
criterion maximises the removal of that component of Sq that can be attributed to

form. It is possible to devise criteria related to other areal surface parameters.
5.4.1 Least-Squares Model Fitting
The most straightforward fitting criterion to implement is the least-squares crite-

rion. The first case to consider is when the distance function is linear in the surface
parameters a ¼ ða1 ; . . .; an ÞT as in (5.8), for example. If di ¼ zi þ cTi a and C is the
m n matrix whose ith row is cTi , then the optimal a is given by
1
a ¼ C T C C T z; z ¼ ðzi ; . . .; zm ÞT :
If C has QR factorisation, Golub and Van Loan (1996),

R1
C ¼ QR ¼ ½ Q1 Q2 ¼ Q 1 R1 ;
0
where Q is an m m orthogonal matrix and R1 is an upper-triangular matrix, then
a ¼ R1 T
1 Q1 z.
If dðx; aÞ is a nonlinear function of a, an iterative approach, for example the
Gauss-Newton algorithm, is required, where at each iteration a linear least-squares
problem is solved. If J is the Jacobian matrix of partial derivatives
@d
Jij ¼ ðxi ; aÞ
@aj
evaluated at estimate a, then an updated estimate is given by

1
a :¼ a þ p; p ¼ J T J J T d; ð5:15Þ
where d ¼ dðaÞ ¼ ðd1 ; . . .; dm ÞT and di ¼ dðxi ; aÞ. Using a first order Taylor
expansion,
dða þ pÞ dðaÞ þ Jp;
p is determined so that, to first order, dða þ pÞ is minimised in the least-squares

sense. The update step p in (5.15) can be determined using the QR factorisation of
the Jacobian matrix J.
For parametrically defined surfaces u 7! f ðu; aÞ, the best-fit parameters a can
also be determined by minimising
X
m
min ðxi f ðui ; aÞÞT ðxi f ðui ; aÞÞ
a;fui g
i¼1
involving not only a but also the ui , i ¼ 1; . . .; m. This means that the Jacobian
matrix of partial derivatives is a 3m ð2m þ nÞ matrix involving partial deriva-
tives with respect to ui as well as those with respect to a, so that for large data sets
the computation is potentially time consuming. However, the fact that ui is
involved with only one data point means that the Jacobian matrix has a sparsity
structure that can be exploited to make the computation very efficient (Cox 1981;
Forbes 2006).
5.4.1.1 Uncertainties Associated with the Fitted Parameters
Suppose
xi ¼ f ðui ; aÞ þ i ; i 2 Nð0; r2 IÞ; ð5:16Þ
modelling the fact that the measured coordinates xi represents a point on f ðu; aÞ
perturbed by random effects drawn from the multivariate Gaussian distribution.
Using the law of propagation of uncertainty (BIPM et al. 2008; Cox and Harris
2010), the variance matrix Va associated with the fitted parameters is given by
1
Va ¼ r2 J T J : ð5:17Þ
1
If J has QR factorisation, J ¼ Q1 R1 , then Va ¼ r2 RT1 R1 . The standard uncer-
tainties uðaj Þ associated the fitted parameters aj are given by the square roots of the
diagonal elements of Va . If r is not known then a posterior estimate r^ of r is given by
1 X m
^2 ¼
r d 2 ðxi ; aÞ:
m n i¼1
This estimate can be used in place of r in (5.17).

The interpretation of the variance matrix Va depends on the interpretation of the
model in (5.16). The model can be interpreted as saying that the artefact has
perfect geometry, in the sense that there is an a that describes the geometry
exactly, and that the deviation from perfect geometry is due to random effects
associated with the measurement system. In this interpretation, the variance matrix
represents the variation that is expected if the geometry is re-measured many times
using the same measurement strategy, i.e., quantifies the propagation of the ran-
dom effects associated with the measurement system through to the fitted
parameters.
A second interpretation of the model (5.16) is that i represents the departure
from perfect geometry that is measured by a perfectly accurate measuring system.
In this case the variance matrix can be interpreted, perhaps, as the variation that
would be expected if the surface was measured using a different but similar
measurement strategy or measuring a different surface produced on the same
system. In this interpretation, the model assumes that the form error is random so
116 A. B. Forbes
that knowing the form error at one location provides no information about the form
error at a nearby location. This interpretation might not be particularly realistic.
A more general model accounts both for form error and random effects asso-
ciated with the measurement system. It has the form
xi ¼ f ðui ; aÞ þ ei ni þ i ; i 2 Nð0; r2 IÞ; ð5:18Þ
where ei is the (signed) size of the form error in the direction of the normal vector
ni at the point f ðui ; aÞ on the surface. For this model, the variance matrix in (5.17)
represents the influence only of the random effects associated measurement system
on the fitted parameters.
The model above assumes that the random effects associated with the mea-
surement system are uncorrelated and have equal variance. A more general model
is that the 3m vector of random effects is drawn from a multivariate distribution
with variance
V ¼ D þ GGT
where D is a 3m 3m diagonal matrix associated with random effects and G is a

3m p matrix that represents the uncertainty contribution of p random effects
associated with the measurement system, for example, scale and squareness effects
associated with the axis system. If N is the m 3m block diagonal matrix storing
nTi in the 1 3 diagonal blocks, then the uncertainty associated with the xi is
propagated through to the uncertainty associated with a according to
Va ¼ R1 T T T 1
1 Q1 NðD þ GG ÞN Q1 R1 :
If V ¼ r2 I then this expression is the same as that in (5.17).
5.4.1.2 Uncertainty Associated with the Residual Distances
The residual distances dðxi ; aÞ are determined from the data points xi and the fitted
parameters a. In practice xi and hence a are subject to random effects associated
with the measurement system. If J has QR factorisation

R1
J ¼ QR ¼ ½ Q1 Q2 ¼ Q1 R1 ;
0
then the variance matrix Vd associated with the vector of residual distances is
given by r2 Q2 QT2 , in the case V ¼ r2 I. For the more general case,
Vd ¼ Q2 QT2 NðD þ GGT ÞN T Q2 QT2 :
The matrix Vd is generally a full matrix so that the variance associated with di is
correlated with that associated with dj . This is because the evaluated residual
distances all depend on the common set of parameters a. For m data points and n
parameters, the variance matrix Vd has rank m n, since n degrees of freedom are
used to define the surface fit. Since areal surface parameters are usually determined
on the basis of the residual distances, random effects associated the measurement
systems propagate through to the residual distances and then through to the areal
parameters.
5.4.2 L1 Model Fitting
The L1 model fitting problem (5.13) can be reformulated as follows,

X
m
min si
a;fsi g
i¼1
subject to the constraints
si dðxi ; aÞ si :
If dðxi ; aÞ is linear in a, then the problem is a linear programming (LP) problem.

Note that the introduction of m parameters si means that a standard implementation
of an LP algorithm could be computationally expensive. Special implementations
of the LP problem such as algorithm ACM TOMS 552 Barrodale and Roberts
(1978, 1980) that take account of the particular structure of the problem have
proved effective. If the distance function dðx; aÞ is nonlinear in the parameters a,
then more advanced computational machinery such as the package IPOPT
Wächter and Biegler (2006) can be applied. IPOPT uses sparse matrix techniques
to improve computational efficiency. Both TOMS 552 and IPOPT are available in
the public domain.
Uncertainty evaluation associated with fitted parameters according to an L1
fitting criterion is not straightforward and is not considered further here. Monte
Carlo methods can be used if necessary.
5.4.3 Chebyshev Model Fitting
The Chebyshev model fitting problem (5.13) can be reformulated as
min s
s;a
subject to
s dðxi ; aÞ s; i ¼ 1; . . .; m;
involving one additional parameter s. If dðx; aÞ is linear in a, then the problem can
be solved efficiently using an LP algorithm, for example Algorithm ACM TOMS
118 A. B. Forbes
495 Barrodale and Phillips (1975). For nonlinear dðx; aÞ the IPOPT package can
be used effectively Wächter and Biegler (2006).
As far as uncertainty evaluation is concerned, remarks that apply to L1 fitting
also apply to Chebyshev model fitting.
5.4.4 Discussion on the Fitting Criteria
Least-squares fitting criteria have the advantages that they lead to straightforward
computational approaches and it possible to evaluate uncertainties associated with
the fitted parameters and derived quantities. In general, there are two variants,
ordinary least-squares (OLS) and total least-squares (TLS), the latter also known
as orthogonal distance regression.
The OLS approach generally applies using a measure of distance that depends
on the difference in heights, Sect. 5.3.1, while the TLS approach is based on using
the orthogonal distance. The OLS approach has the advantage of being more
straight forward, computationally. For a large class of models of form, the height
distance function is linear in the form parameters and the fitted surface can be
determined using direct numerical linear algebra techniques. The OLS approach
has the disadvantage that the fitted surface is not invariant with respect to rotations
of the coordinate system. The TLS method does have such invariance but at the
expense of a more complex computations. The fitting problem is usually nonlinear
requiring an iterative approach and for freeform geometries, the orthogonal dis-
tances themselves require iterative schemes to evaluate them.
For areal form removal, the surface patch is generally close to planar which
means that, following a re-orientation if necessary, the difference in height and the
orthogonal distance has provide essentially the same measure of distance, and
therefore the OLS and TLS approach give very similar fits that lead to equivalent
F-operations. For this case, the algebraic distance is often also related accurately to
the orthogonal distance and can also be used for form removal. Using an algebraic
measure of distance usually means that the optimisation problem can be solved
with the same computational tools as for OLS.
The main advantage of an L1 fitting approach is that the fitted surface is not
overly influenced by outliers, relative to least-squares methods for example. If the
data is likely to have isolated spikes, etc., then an L1 approach may be more
appropriate. In determining an estimate of the quantity from a number of repeat
measured values the LS fit defines the mean of the values while the L1 fit
determines the median of the values. The L1 need not be unique, for instance the
median could be defined by two values (or any value in between). For fitting
models to large data sets, non-uniqueness is unlikely to arise in practice. Uncer-
tainty evaluation for L1 fits is not straightforward as mentioned above.
Chebyshev fits are appropriate when it is required to minimise the maximum
error of approximation, for example, in approximating a function by simpler one.
Because of the computational complexity associated with L1 and Chebyshev

approximations and the difficulty in assigning uncertainties to the fitted parame-
ters, the rest of this chapter concentrates on mainly on least-squares fitting
criterion.
5.5 Areal Form Removal
A major feature for areal form removal is that, for the most part, the surface
analysis is performed on a near planar patch of a surface, irrespective of the shape
of the overall geometry, which may be a sphere, cylinder, etc. This means that
standard geometric fitting algorithms need to take into account the fact that only a
small section of the surface is available.
5.5.1 Plane
Ordinary least-squares approach. In this approach, the plane is defined as a

functional surface as in (5.2) involving parameters a ¼ ða; b; cÞT , the distance from
a point to a surface is measured in terms of a difference in height (Sect. 5.3.1), and
a least-squares fitting criterion is used, Sect. 5.4.1. Given data X ¼ fxi gm i¼1 , let C
be the m 3 matrix whose ith row is cTi ¼ ðxi ; yi ; 1Þ, and z ¼ ðz1 ; . . .; zm ÞT . Then
the best-fit parameters are given by
a ¼ ðCT CÞ1 CT z:
Given a ¼ ða; b; cÞT , and a point x ¼ ðx; y; zÞT , the corresponding F-operator is
x ¼ F ðx; aÞ where
defined by
2 3 2 3
x x
6 7 6 7
¼ 4 y 5 ¼ 4
x y 5: ð5:19Þ
z z ax by c
This is the simplest type of form error removal.

Total least-squares approach. The parametrisation of the plane used in (5.2) is
convenient for the ordinary least-squares approach. The total least-squares approach
uses orthogonal distance (Sect. 5.3.4) as a measure of the departure of a point from a
surface. In evaluating the orthogonal distance, it is generally more convenient to
work with another parametrisation. The following parametrisation is appropriate for
planes that are approximately horizontal. Let Rx ðaÞ and Ry ðbÞ be rotation matrices
defined in (5.4) representing rotation through an angle of a about the x-axis and an
angle of b around the y-axis, respectively. Then, for a and b small,
120 A. B. Forbes
2 3 2 3
0 sin b
T T 6 7 6 7 b
n ¼ nða; bÞ ¼ Rx ðaÞRy ðbÞ4 0 5 ¼ 4 sin a cos b 5 ð5:20Þ
a
1 cos a cos b
represents a parametrisation of the normal vector to the plane. The equation of the
plane is then
xT nða; bÞ ¼ d0 :
where d0 is the distance from the plane to the origin, and involves three parameters
a ¼ ða; b; d0 ÞT .
The orthogonal distance di from a point xi ¼ ðxi ; yi ; zi ÞT to the plane is
dðxi ; aÞ ¼ xTi n d0 :
The total least-squares best-fit plane (least-squares orthogonal distance regression

best-fit plane) is found by minimising
X
m
d2 ðxi ; aÞ;
i¼1
and a standard iterative algorithm can be applied. However, an alternative

approach using a singular value decomposition (SVD, Golub and Van Loan
(1996)) can be applied. The best-fit plane must pass through the centroid x ¼
ðx; y; zÞT of the data, where
1X m
x¼
xi :
m i¼1
be the m 3 matrix with xi

Let X has SVD
x in the ith row, and suppose X
¼ USV T ;
X
where U is an m n orthogonal matrix, V ¼ ½v1 ; v2 ; v3 is a 3 3 orthogonal

matrix, and S is an m 3 diagonal matrix
2 3
s1 0 0
60 s 07
6 2 7
6 7
60 0 s3 7
6 7
S¼6
07
;
60 0 7
6 . .. 7
.. 7
6 .
4 . . . 5
0 0 0
with diagonal elements s1 s2 s3 0. Then, the normal vector n to the best-fit

plane is given by v3 ¼ ðv13 ; v23 ; v33 ÞT , the right singular vector corresponding to
the smallest singular value, and the solution d0 is given by d0 ¼ xT n. The SVD is
related to the eigenvector decomposition in that the eigenvector decomposition of
TX
X ¼ VS2 V T . Software for evaluating the SVD of a matrix exists in the public
domain, for example, in LAPACK (1999).
In view of (5.20), the rotation angles a and b are given by

1 1 v23
b ¼ sin ðv13 Þ; a ¼ sin :
cos b
The corresponding F-operator is given by

02 3 2 3 1 2 3
x x x
B6 7 6 7 C 6 7
F ðx; aÞ ¼ Ry ðbÞRx ðaÞ@4 y 5 4 y 5A þ 4 y 5:
z z 0
The effect of the operator is to rotate the data points about the centroid so that the
best-fit plane to the data points is horizontal and then translate the data points
along the z-axis so that the centroid lies in the xy-plane. Depending on the tilt of
the best-fit plane, the grid spacing along the x- and y-axis may be slightly different
following the application of this F-operator whereas the application of the
F-operator in (5.19) leaves the grid spacing unchanged. If appropriate the
F-operator in (5.19) can be applied using the parameters determined from the TLS
associated feature.
L1 and Chebyshev. L1 and Chebyshev fits using a difference in height as a
measure of distance can be implemented using algorithms such as TOMS 552 and
TOMS 495, respectively, as mentioned in Sect. 5.4.
5.5.2 Sphere
While sphere fitting to general coordinate data has been considered elsewhere
Forbes (1989, 1990), here issues associated with areal applications are considered.
It is assumed that a plane form has already been removed so that the sphere section
lies approximately on the plane z ¼ 0.
Usually, a sphere is parametrised by its centre x0 ¼ ðx0 ; y0 ; z0 ÞT and radius r0 .
The orthogonal distance from a point xi to such a sphere is given by
dðxi ; aÞ ¼ ri r0
where ri is defined in (5.9). The TLS best-fit sphere can be determined using the
standard Gauss-Newton algorithm, Sect. 5.4.1. For a near planar section of a
122 A. B. Forbes
sphere, the sphere centre is far from the data and the radius is poorly defined. This
leads to ill-conditioning in the problem and to numerical difficulties when the
numerical scheme is implemented.
Alternative parametrisations come from an algebraic definition of a sphere
Aðx2 þ y2 þ z2 Þ þ Bx þ Cy þ Dz þ E ¼ 0:
This expression involves five coefficients, A–E. An additional constraint on these

coefficients leads to a well-defined parametrisation. For example setting, E ¼ 1,
spheres approximately centred at the origin can be parametrised. An alternative
parametrisation for sections of sphere surfaces approximately parallel to the
xy-plane derived from this algebraic expression is
Aðx2 þ y2 þ z2 Þ þ Bx þ Cy þ z þ E ¼ 0; ð5:21Þ
parametrising the sphere in terms of a ¼ ðA; B; C; EÞT . If A ¼ 0, the equation

defines a plane and for a near-planar section of a sphere, A is small relative to 1.
For A 6¼ 0, the sphere centre x0 and radius r0 are given by
2 3
B
1 6 7 1 þ B2 þ C 2 4AE
x0 ¼ 4 C 5; r0 ¼ :
2A 2jAj
1
The algebraic distance from a point to the sphere surface is given by
dA ðx; aÞ ¼ Aðx2 þ y2 þ z2 Þ þ Bx þ Cy þ z þ E;
The height of the surface specified by a as function of x and y is given by

2Wðx; yÞ
f ðx; y; aÞ ¼ ; Wðx; yÞ ¼ Aðx2 þ y2 Þ þ Bx þ Cy þ E;
1 þ ð1 4AWðx; yÞÞ1=2
ð5:22Þ
while the orthogonal distance is given by
2dA ðx; aÞ
dðx; aÞ ¼ ;
UðxÞ þ V
where
UðxÞ ¼ ðð2Ax þ BÞ2 þ ð2Ay þ CÞ2 þ ð2Az þ 1Þ2 Þ1=2 ; V ¼ ð1 þ B2 þ C 2 4AEÞ1=2 :
Note that if the sphere surface section is near the plane z ¼ 0, then A 0, B 0,
C 0 and E 0, U 1 and V 1, so that both dA ðx; aÞ and dðx; aÞ are
approximately the same as the difference in heights.
Fig. 5.1 Random data on a 0.5

section on a sphere 0
−0.5
−1
−1.5
−2
−2.5
10
5 10
0 5
0
−5 −5
−10 −10
The advantage of this parametrisation is that all fitting calculations, using any
measure of distance (orthogonal, height, algebraic) or using any fitting criteria
(least-squares, L1, Chebyshev), can be performed in a numerically stable way.
The simplest form removal operation is based on a difference in heights
approach:
F ðx; aÞ ¼ ðx; y; z f ðx; y; aÞÞT ; ð5:23Þ
where f ðx; y; aÞ is defined in (5.22). Applying these operations to data for example,
the sequence of operations is (1) fit a plane to the data, (2) apply the plane form
removal operator to the data, (3) determine the best-fit sphere to the transformed
data, and (4) apply the sphere form removal operator. Figures 5.1 and 5.2 represent
the results of this form removal process applied to random data.
L1 and Chebyshev. L1 and Chebyshev fits using the difference in height or
algebraic distance as measures of distance can be implemented using algorithms
such as TOMS 552 and TOMS 495, respectively, as mentioned in Sect. 5.4.
Fig. 5.2 Data in Fig. 5.1

with sphere form removed 0.06
0.04
0.02
0
−0.02
−0.04
−0.06
10
5 10
0 5
0
−5 −5
−10 −10
124 A. B. Forbes
5.5.3 Other Geometric Elements
The parametrisation of the sphere given by (5.21) successfully avoids any

numerical difficulties associated with fitting a sphere to data representing a small
section of a sphere surface. For other geometric elements, for example, a cylinder,
no such parametrisation is available and form removal algorithms have to cope
with the potential for rank deficiency. One approach is to use additional prior
information. For example, the radius of the cylinder may be known along with the
approximate orientation of the axis. This enables some of the ill-conditioning to be
removed.
5.5.4 Empirical Surfaces with Spatial Correlation Structure
An alternative to fitting standard geometric elements to surface patches and

thereby potentially encountering ill conditioning is to regard the surface patch as
an empirical surface that could be represented quite accurately by tensor product
Chebyshev polynomials or tensor product splines for example, or other empirical
models. By increasing the number of basis functions, very accurate representations
of geometric elements, aspheric surfaces, etc., can be determined (with no
numerical difficulties). For example, a tensor product spline with 23 23 ¼ 529
basis functions can accurately represent the cylindrical patch in Fig. 5.3 to a
relative accuracy of better than 1 part in 107 . However, the increased flexibility
means that the model surface can also represent surfaces features that would be
characterised as waviness, rather than form, an extremely undesirable feature for
areal form removal. Figure 5.4 shows the result of fitting a tensor production spline
with 529 basis functions to the cylindrical patch with Gaussian noise added. The
graph shows the tensor product surface with the cylindrical form removed. Clearly,
the spline surface has features that are derived purely from the noise.
Fig. 5.3 Patch of a

cylindrical surface
1
0.5
0
120
100 120
100
80
80
60
60
40 40
20 20
0 0
Fig. 5.4 Tensor product

spline fit to a simulated rough
cylindrical patch with 0.015
cylinder form removed
0.01
0.005
0
−0.005
−0.01
−0.015
150
100 150
100
50
50
0 0
One approach to balance the approximating power of these empirical models

f ðx; y; aÞ without overfitting to local surface features is to impose a spatial cor-
relation structure on the models, Rasmussen and Williams (2006). The correlation
structure is assigned as
covðf ; f 0 Þ ¼ kðx; y; x0 ; y0 jrÞ;
where k is a correlation kernel, for example,

1
kðx; y; x0 ; y0 jrÞ ¼ r2 exp 2 ððx x0 Þ2 þ ðy y0 Þ2 Þ þ r20 d; r ¼ ðr; r0 ; kÞT ;
2k
ð5:24Þ
where dðxÞ ¼ 1 if x ¼ x0 and y ¼ y0 and is zero otherwise. Let V ¼ VðrÞ be the

variance matrix determined by the correlation kernel k above evaluated at ðxi ; yi Þ
and C the coefficient matrix associated with basis functions evaluated at these
points. The fitted parameters a are required to be such that the evaluated
f ¼ ðf1 ; . . .; fm ÞT , fi ¼ f ðxi ; yi ; aÞ looks like a sample from the distribution Nð0; VÞ.
The likelihood pðyÞ of observing y 2 Nð0; VÞ is such that

1 T 1
pðyÞ / exp y V y :
2
so that, applying Bayes theorem, Gelman et al. (2004),

1
pðajf ¼ CaÞ / exp aT CT V 1 Ca
2
from which it is inferred that

1
a Nð0; Va Þ; Va ¼ C T V 1 C :
126 A. B. Forbes
Fig. 5.5 Spatially correlated

tensor product spline fit to a
simulated rough cylindrical 0.06
patch with cylinder form
removed 0.04
0.02
−0.02
−0.04
150
100 150
100
50
50
0 0
If Va has Cholesky factorisation Va ¼ La LTa , then the fitted parameters a are

determined by solving an augmented least-squares system,

C z
1
a
La 0
incorporating the smoothing term derived from La .

Figure 5.5 shows the result of fitting a spatially correlated tensor product spline
surface to the cylindrical patch with Gaussian noise added. The graph shows the
tensor product surface with the cylindrical form removed. Compared to Fig. 5.4,
the spline surface does not follow the random noise in the data.
The spatial correlation model in (5.24) does not reflect the fact that the nominal
surface is a cylinder. It would be possible to change the correlation lengths along a
generator and orthogonal to a generator. It is also possible to combine a spatially
correlated empirical model with a nominal form, with suitable prior information.
The parameter k can be chosen to define the spatial correlation scale, just as its
counterpart in Gaussian filtering.
Once an empirical surface has been fitted to the data determining the surface
z ¼ f ðx; y; aÞ, the most straightforward form removal operator to apply is that in
(5.23). A plane form should be removed first from the surface before the empirical
model is fitted.
5.5.5 Form Removal Using Filters
Form removal using empirical functions, particularly spatially correlated empirical

models discussed above, is not dissimilar from form removal using filters. In ISO
25178 part 3 (2012), form removal using the nominal form is the default procedure
but a filtration method can also be used.
Given a set of data points xi ¼ ðxi ; yi ; zi ÞT , the filter produces a new set of data
points ðxi ; yi ; ^zi ÞT where the new ^zi are the filtered heights. The corresponding
F-operator is
F ðxi Þ ¼ ðxi ; yi ; zi zî ÞT :
This F-operator is similar to other F-operators, only that the operator does not
depend on parameters a defining the associated feature, but on the outputs of the
filter acting on the data set.
Filters are discussed in Chap. 4.
5.6 Summary and Concluding Remarks
A general procedure for areal form removal based on measured data X0 is as

follows:
1. Determine the TLS best-fit plane to the data, determining plane parameters a0 .
2. Apply the plane form removal operator F ðxi ; a0 Þ to the data, determining a new
data set X1 .
3. Express the nominal form to be removed as a functional surface z ¼ f ðx; y; aÞ.
4. Determine the best-fit surface to the data set X1 , specified by best-fit parameters
a1 .
5. Apply the form removal operator to X1 ,
F ðx; aÞ ¼ ðx; y; z f ðx; y; a1 ÞÞT ;
to determined data set XF , the data set with the nominal form removed.
The following comments can be made:
a. The combined form removal operations is F ðxÞ ¼ F ðF ðx; a0 Þ; a1 Þ, the com-
position of the two operations.
b. The best-fit surface in 4, above, can be determined using an OLS or TLS
approach. Irrespective of which one is used the combined operation does not
depend on the coordinate system since the plane form removal is invariant.
c. The combined operator depends of the fitted parameters a0 and a1 . Uncer-
tainties associated with fitted parameters can be propagated through to those
associated with the modified data points.
d. If the data X0 is subject to outliers or spikes, a L1 best-fit plane can be
determined after step 1 (to maintain invariance with respect to the coordinate
system). The fitted parameters ~ a0 can be used to perform an additional plane
form removal prior to step 3.
128 A. B. Forbes
Acknowledgments This work was funded by the UK’s National Measurement Office pro-
gramme for Mathematics and Modelling.
References
Anderson IJ, Cox MG, Forbes AB, Mason JC, Turner DA (1998) An efficient and robust
algorithm for solving the footpoint problem. In: Daehlen M, Lyche T, Schumaker LL (eds)
Mathematical methods for curves and surfaces II. Vanderbilt University Press, Nashville,
pp 9–16
Barrodale I, Phillips C (1975) Algorithm 495: solution of an overdetermined system of linear
equations in the Chebyshev norm. ACM Trans Math Softw 1(3):264–270
Barrodale I, Roberts FDK (1978) An efficient algorithm for discrete l1 linear approximation with
linear constraints. SIAM J Numer Anal 15:603–611
Barrodale I, Roberts FDK (1980) Solution of the constrained ‘1 linear approximation problem.
ACM Trans Math Soft 6:231–235
Bartholomew-Biggs M, Butler BP, Forbes AB (2000) Optimisation algorithms for generalised
regression on metrology. In: Ciarlini P, Forbes AB, Pavese F, Richter D (eds) Advanced
mathematical and computational tools in metrology IV. World Scientific, Singapore,
pp 21–31
BIPM, Iec, IFCC, Ilac, ISO, Iupac, IUPAP, OIM (2008) Guide to the expression of uncertainty in
measurement. International Organization for Standardization, Geneva
Cox MG (1981) The least-squares solution of overdetermined linear equations having band or
augmented band structure. IMA J Numer Anal 1:3–22
Cox MG, Harris PM (2010) SS fM best practice guide no. 6, uncertainty evaluation. NPL Report.
MS 6, National Physical Laboratory, Teddington, UK
Forbes AB (1989) Least-squares best-fit geometric elements. NPL Report DITC 140/89, National
Physical Laboratory, Teddington
Forbes AB (1990) Least-squares best-fit geometric elements. In: Mason JC, Cox MG (eds)
Algorithms for approximation II. Chapman and Hall, London, pp 311–319
Forbes AB (2006) Structured nonlinear Gauss-Markov problems. In: Iske A, Levesley J (eds)
Algorithms for approximation V. Springer, Berlin, pp 167–186
Forbes AB, Minh HD (2011) Form assessment in coordinate metrology. In: Georgoulis EH, Iske
A, Levesley J (eds) Approximation Algorithms for complex systems. Springer proceedings in
mathematics, vol 3. Springer, Heidelberg, pp 69–90
Gelman A, Carlin JB, Stern HS, Rubin DB (2004) Bayesian data analysis, 2nd edn. Chapman and
Hall/CRC, Boca Raton
Golub GH, Van Loan CF (1996) Matrix computations, 3rd edn. John Hopkins University Press,
Baltimore
Helfrich HP, Zwick D (1995) Trust region algorithms for the nonlinear distance problem. Numer
Alg 9:171–179
ISO 25178 part 2 (2012) Geometric product specifications (GPS)—surface texture: areal, part 2:
terms, definitions and surface texture parameters
ISO 25178 part 3 (2012) Geometric product specifications (GPS)—surface texture: areal, part 3:
specification operators
Piegl L, Tiller W (1996) The NURBS book, 2nd edn. Springer, New York
Rasmussen CE, Williams CKI (2006) Gaussian processes for machine learning. MIT Press,
Cambridge
The LAPACK User’s Guide (1999) The LAPACK User’s Guide, 3rd edn. SIAM, Philadelphia
Wächter A, Biegler LT (2006) On the implementation of a primal-dual interior point filter line
search algorithm for large-scale nonlinear programming. Mathematical programming
106:25–57
Chapter 6
Areal Fractal Methods
Christopher A. Brown
Abstract In this chapter the basic nature of fractals is discussed and practical
applications of fractal-based methods for useful characterisation of surface
topography are examined. Length-scale and area-scale fractal analysis methods are
examined in detail because these methods have industrial and scientific value, and
are now part of ASME and ISO specification standards. The context and rationale
for using fractal analysis for describing surface topography are examined first,
followed by discussion of the chaotic nature of measured surface topographies.
Length-scale and area-scale analysis methods, which exploit the fractal nature of
topographies, are then reviewed, with an emphasis on areal methods. In addition,
examples are presented of the discoveries of functional correlations, and the
abilities to discriminate that are made possible by multi-scale fractal analysis.
6.1 Introduction to Areal Fractal Methods
In this chapter the basic nature of fractals is discussed and practical applications of
fractal-based methods for useful characterisation of surface topography are
examined. Fractal-based characterisation has potential for providing orderly
descriptions of chaos (Mandelbrot 1977). In this chapter the development of fractal
analysis as applied to areal topography is also discussed, along with the inclusion
of fractal-based methods in national and international standards.
This chapter is not intended to be a review of the large variety of fractal analysis
methods that have been applied to surface measurements. Such a review is pre-
sented elsewhere (De Chiffre et al. 2000). Rather, length-scale and area-scale
fractal analysis methods are discussed, because they appear to have industrial and
C. A. Brown (&)
Worcester Polytechnic Institute, Worcester, MA, USA
e-mail: brown@WPI.EDU

130 C. A. Brown
scientific value, and because they now included in international specification

standards.
The term fractal was coined by Mandelbrot in the late 1970s. In geometry, the
term fractal has been used to describe topographies that have certain properties.
One property is that they are continuous, but nowhere differentiable. Another
property is that fractals exhibit self-similarity, or self-affinity, relative to scale
(Mandelbrot 1977).
Geometrically, fractals have forms, or features, that repeat at different sizes
over ranges of scales. These features can repeat exactly, such as the triangles that
repeat with scale on a Koch snowflake or Minkowski sausage (Fig. 6.1). Or, the
features might repeat statistically, as on ground or abraded surfaces, where these
repeating features create patterns of little scratches on larger scratches over a range
of scales. Turned surfaces also show these characteristics inside the feed marks of
the surface (see Fig. 6.2).
Mathematical fractal shapes can be generated by recursive algorithms, or rec-
ipes, such as the repeating triangles in the Koch snowflake (Fig. 6.1). Conven-
tional geometric shapes are generated by equations. This makes fractals
particularly well suited for computers. Recursive algorithms can be similar to the
phenomena that generate topographies.
The fractal dimension can be used to characterise the intricacy or complexity of
geometries. The fractal dimension can be fractional. In Fig. 6.1, the Minkowski
sausage has a dimension of 1.5 and is more complex than the snowflake, with a
dimension of 1.26. For a profile z = z(x), the fractal dimension is greater than or
equal to one and less than or equal to two. For a measured surface z = z(x, y), the
fractal dimension is greater than or equal to two, and less than three.
Fig. 6.1 The Koch snowflake shows a triangular generator or feature and the Minkowski
sausage a jog that repeat exactly over a range of scales to form fractal patterns
6 Areal Fractal Methods 131
Fig. 6.2 Smooth approximation and chaotic elements. A turned surface on 6061 T6 aluminium
alloy measured with a scanning laser confocal microscope. At the larger scales the turned surface
shows the cusps from the smooth approximation. At the finer scales it shows scratches that repeat
statistically over a range of scales
Self-affine surfaces demonstrate self-similarity only with an appropriate hori-

zontal to vertical scaling ratio (Mandelbrot 1985). It can be argued that engi-
neering surfaces are by nature self-affine, because they tend to be smooth at large
scales of observation and rough at some sufficiently fine scale. Therefore, if it is
required to maintain apparent self-similarity through such a range of scales, then
the scaling ratios must change.
The scaling ratio, in addition to the fractal dimension, can be used in the
characterisation of self-affine topographies. In practice, it is found that the fractal
dimension and the scaling ratio can both change with respect to scale. All surfaces,
including self-affine surfaces, can be considered as surfaces that are multi-fractal
with respect to scale in that the fractal dimension, and its relation to scale change
with respect to scale. Multi-scale, or scale-based, fractal analysis is the principal
approach to characterising topographies that will be examined in more detail in
this chapter.
Profiles, where the heights z are given as a function of position x, are often
called 2D in the field of surface metrology. Surfaces, where the heights are given
as a function of position in x and y, are called 3D. The designation of profiles and
surfaces by their conventional dimensionality should not be confused with their
characterisation by the fractal dimension.
Fractal analysis of profiles is discussed here in the context of characterising
areal topographic measurements. It makes sense to look at length-scale analysis as
well. The logical development is from length-scale profile to area-scale areal
analyses, which provides some continuity of understanding. More importantly,
length-scale analysis of profiles in multiple directions on areal measurements can
be used to characterise anisotropy, which is an important form of areal analysis.
132 C. A. Brown
The context and rationale for using fractal analysis for describing topographies
are examined first in this chapter, followed by discussion of the chaotic nature of
measured surface topographies. Length-scale and area-scale analysis methods,
which are based on fractal theories, are then reviewed, with an emphasis on areal
methods. In addition, examples are presented of the discoveries of functional
correlations, and the ability to discriminate that is made possible by multi-scale
fractal analysis.
6.1.1 Rationale: The Value in Characterising Topographies
The context and rationale for characterisation of surface topographies can be based
on utility, which is to say that it has value. To be valuable, characterisations of
topographies must go beyond simply providing another means of describing
topographies. Many kinds of fractal analyses merely provide the basis for claiming
that surface topographies are fractal in nature (De Chiffre et al. 2000). To have value,
particularly for applications in engineering, technology, and science, the descrip-
tions of topographies should advance the understanding of topographic interactions,
i.e. advance the ability to solve problems related to surface metrology.
There is value to industry in supporting quality control, as well as product and
process design. This is accomplished by providing the means for discriminating
surfaces, and by providing the means for correlating topographies with topo-
graphically-dependent phenomena.
There are two fundamental kinds of topographic interactions. The first kind is
manufacturing, creating or modifying topographies. Studies of this kind of inter-
action seek to understand how topographies are influenced by processing condi-
tions during their creation or modification. The second kind of interaction concerns
how topographies influence surface behaviour. Studies of this kind of interaction
seek to understand the relationships between the topographies and performance.
These two kinds of interactions are not mutually exclusive. Some kinds of surface
interactions can both modify the behaviour and be modified by it simultaneously.
The understanding of topographic interactions can help advance science and can
solve problems in engineering and technology.
Interactions of both kinds can be described by functional correlations. Func-
tional correlations describe the relationship between particular characterisations of
the topography and of the interaction, for example, processing or performance
characterisations.
Understanding the first kind of topographic interaction has value in that it
provides a basis for creating a desired topography through manufacturing process
design. The ability to optimally design manufacturing processes can depend on
knowledge of the functional correlations between the processes and the topogra-
phies. The first kind of interaction is also useful for determining how surface
topographies are created or modified, which has value in forensics, anthropology
and archaeology.
Understanding the second kind of topographic interaction has value in

designing products. The ability to specify optimal surface topographies, those that
will have optimal behaviours for topographic-dependent interactions, requires
knowledge of the functional correlation between the topography and the
behaviour.
Both of these kinds of interactions, with valuable applications in product and
process design, depend on being able to find an appropriate characterisation of the
topography; one that is capable of elucidating these correlations.
The ability to discriminate surfaces also has value. This is the essential element
in most quality control systems, discriminating good product from bad. Demon-
strating the ability to discriminate surfaces based on their topographies can also be
useful in developing technology and advancing science.
6.1.2 The Chaotic Nature of Roughness and Topographies
Roughness can be thought of as a geometric irregularity. Fractal objects are

irregular in shape. Some kinds of fractals resemble the popular conception of what
roughness on surfaces should look like. Many natural surfaces have some kind of
fractal appearance over a wide range of scales, clouds (Schertzer and Lovejoy
1988) and mountains, for example. Many manufactured objects, however, only
appear to be rough and have this kind of fractal appearance when observed at
sufficiently fine scales.
The forms of physical objects are often designed at the largest scales by using
the inherently smooth shapes of classical geometry, which are at least piecewise
differentiable. The roughness that appears at the finest scales in manufactured
objects might only be there because it is too difficult to eliminate. Perhaps it is felt
that there is not sufficient value added to make it worth precisely controlling and
creating the topography during manufacturing at these fine scales. Also, little may
be known about how to design surface topographies for desired functions at
extremely fine scales.
At the finest scales the work piece microstructures and the manufacturing tools
can have a kind of chaotic heterogeneity that interacts to create chaotic topogra-
phies. Some of these chaotic topographies can be engineered to be useful. How-
ever, to do this engineering effectively and repeatedly, design and manufacturing
engineers need to be able to characterise these chaotic geometries in useful ways.
Because it is adept at describing chaos, fractal analysis can be valuable for
facilitating the design and manufacture of rough surfaces.
The smooth shapes of classical geometry are orderly and can be described with
equations. Fractal shapes can be rough, or chaotic. Fractal geometry can be
described with recipes or recursive algorithms (Mandelbrot 1977).
The difference between classical and fractal geometry can be described in terms
of entropy or information. Statistically, entropy is given by
134 C. A. Brown
S ¼ k ln w ð6:1Þ
where w is the number of possible states and k is Boltzman’s constant. Fractal

surfaces have higher entropies and are more complex than the smooth surfaces
described by classical geometry. Complexity, or information, can be defined as

I ¼ ln 1=p ð6:2Þ
where p is the probability of success in accomplishing some function (Suh 1990),

in this case describing the surface. The probability of describing the surface could
be interpreted as the likelihood of accurately determining the height at a location
intermediate to locations with known heights on a surface. In the case of smooth
surfaces the probability is high. In the case of fractal surfaces the probability is low
and they, therefore, have higher information content and greater complexity than
the smooth surfaces described with classical geometry.
In classical, or smooth, geometries, the location of every point in relation to
every other point on a surface can be known exactly. The order is perfect, and the
entropy is zero. The probability of describing the height of every point on the
surface is one; therefore, the information content would be zero, indicating that the
description of the surface geometry is perfect, and there is no complexity in
describing the geometry.
6.1.3 The Inherent Chaotic Nature of Measured Surface

Topographies
Surface measurements contain heights z as a function of position in x and y, such

that z = z(x, y). Real surface measurements do not produce data sets that are like
the surfaces that are defined in smooth, classical geometry. There are two reasons
for this.
Firstly, the height of a mathematical point cannot be measured on a real surface.
This is because it is infinitesimally small. Sensors that make height measurements
must sample some kind of interaction with the surface. This interaction must be
sampled over a finite sampling zone. That sampling of the surface is then used to
determine a height at that position. The measured height in a topographic data set
must represent some function of the heights on the surface in that sampling zone.
The actual function depends on the individual sensor and how it interacts with the
surface. The function might be something like an arithmetic average, a weighted
average, or the maximum. The function is seldom known, if ever, for an individual
sensor and surface combination (Fig. 6.3).
Secondly, regardless of how many positions have heights measured on a real
surface, the exact heights at other positions that have not been measured can only
be predicted within some amount of uncertainty. This is because measured surface
Fig. 6.3 Sampling zones and

sampling intervals. The
measurement in the
background represents the
range of heights that exist
within a sampling zone
topographies tend to have chaotic components. The nature of this chaotic com-
ponent could be related to the technology used to make the measurement. Because
of this chaotic nature, statistical descriptions are used for characterising real sur-
face topography measurements.
Chaos by its nature is not random. If two heights are known on a profile, then it
can be illuminating to ask how well an intermediate height might be known. If the
profile is smooth, as in classical geometry, then the location of the intermediate
height can be known exactly. If the profile is random, then the intermediate height
cannot be known to any degree of certainty, i.e. it could be anywhere. In this sense,
a ‘‘random profile’’ is an oxymoron because it would not be continuous, or
coherent, which it must be in order to be a profile.
In order to be a real topographic profile, there must be some short-range cor-
relation between the proximate measured heights in order for a profile to represent
a real, continuous surface. If the profile is chaotic, then the height can only be
known within some bounds of probability. The distribution of this probability
could be characterised by the variance of the measured heights in the profile. This
variance can depend on the scale on which the profile has been measured, i.e. the
resolution. The resolution is limited in part by the lateral distance between the
measured heights, which is the sampling interval. And, the resolution can be
limited by the lateral and vertical resolution of the sensor, the former being
dependent upon the size and properties of the sensor’s sampling zone. If a mea-
sured profile appears to be perfectly smooth, i.e. with no chaotic elements, then the
resolution is not good enough to detect the chaotic nature of the surface.
One way of creating or modelling fractal profiles is by the midpoint displace-
ment method. In this method, intermediate heights are successively estimated from
some probability distribution. The result is used to create a model, or fractal
forgery, of a profile that can look convincingly like a real profile.
It should be noted that a fractal-based characterisation of the surface can be
valuable, even if the geometry of the physical object is not chaotic through all
ranges of scale. At the largest scales, most manufactured objects are best char-
acterised as smooth, which usually would be according to the intent of the design.
136 C. A. Brown
At and below atomic scales, there are issues with the physics of defining the
boundaries of the surface that are not addressed here. Nonetheless, many, if not all,
surfaces, both natural and manufactured, appear to have some geometric properties
that can be adeptly characterised using fractal-based methods, at least through
some ranges of scales.
Fractal geometry can provide valuable ways of characterising many kinds of
topographies through certain ranges of scales that can be of interest in design and
manufacturing. Value in this context comes from the ability to discriminate or to
correlate. Discrimination of surfaces that are known to be different, based on
processing or performance, is valuable in quality control, for example. The ability
to correlate topographies, at least through some range of scales, with their per-
formance characteristics or processing variables has value in the design of prod-
ucts and processes.
6.2 Diverse Fractal Characterisation Methods
Many different kinds of fractal-based characterisation methods of topographies can

be found in the literature. Generally, these methods have been used to make
statements about the mathematical nature of surfaces, in that they are related to
some particular aspect of fractal geometries, i.e. have some sort of fractal prop-
erties. Fractal characterisation methods have also been used to create mathematical
models that represent surfaces for modelling some kinds of surface interactions.
These efforts are summarised elsewhere (De Chiffre et al. 2000).
The fractal dimension has been used to characterise the complexity or intricacy
of surface topographies in a variety of cases. The early efforts in fractal analysis of
topographies largely concentrated on the determination of the fractal dimension.
Several different methods have been developed to determine some kind of fractal
dimension of topographies. There are extensive examples in the literature (Thomas
1988; Russ 1994; Kaye 1994; Whitehouse 2010; De Chiffre et al. 2000).
Fractal models have been used for modelling chaotic surfaces, in addition to
characterising measured surfaces. These models are then used to model surface
interactions. One such method is based on the Weierstrass-Mandelbrot function
(Zahouani et al. 1998; Whitehouse 2010; Majumdar and Bhushan 1991). This
function provides a model surface with adjustable parameters to provide a desired
complexity in applications such as modelling the size and number of multi-scale
contacts (Zahouani et al. 1998), which can be applied to conductance (Majumdar
and Tien 1990). The Weierstrass-Mandelbrot function method has shown that
electrical contacts follow power laws.
In the context of this chapter, it must be noted that the fractal dimension by itself
has not proven to be an especially valuable tool in surface metrology. The purported
value has been largely to show that the surfaces can be modelled as fractal.
The fractal dimension might be used to discriminate surfaces. However, this
ability is not emphasised in the literature. The fractal dimension by itself has an
important shortcoming for characterising topographies and that is its scale inde-
pendence. Mandelbrot pointed out that the fractal dimension alone could not
distinguish the topography of a runway and the Himalayas, because it could be the
same. However, the smooth-rough crossover scale (SRC) would be different for
pavement and mountains by several orders of magnitude. The SRC is the scale
above which a surface is apparently smooth and below which it is apparently
rough. The fractal dimension in combination with the SRC would be a better
discriminator than either alone.
At scales above the SRC the topography would best be described by classical
geometry, at scales below the SRC the topography might be described statistically
and by fractal methods.
The fractal dimension, or any measures of complexity that the fractal dimension
might quantify, do not generally appear in fundamental equations that describe the
interactions with topographies. One exception to this is found in electrochemical
impendence spectra (EIS). A model (McRae et al. 2002) notes that the fractal
dimension could describe the crack patterns, which would define a resistance and
capacitor network that would correlate with EIS. McRae’s results support this
model for EIS.
For a variety of reasons, the diverse methods of determining the fractal
dimension do not always produce similar numerical results for the fractal
dimension. This diversity of results provokes interesting discussions about how to
determine a true fractal dimension. In this chapter, a discussion of the appropriate
mathematics for fractal characterisation is avoided. The methods that have been
successful at enabling discrimination and discovering correlations, methods that
have appeared in specification standards, will be discussed.
6.2.1 Fractals and Basic Geometric Properties of Surfaces
Fractal analysis can be used to characterise basic geometric properties of chaotic

surfaces. It is important to understand that basic geometric properties can vary
with the scale of observation, or of calculation, on chaotic surfaces. Basic geo-
metric properties include the area of surfaces, the lengths of profiles on the sur-
faces, the curvature of surfaces, slopes or inclinations on surfaces and volumes in
surfaces. These properties are important because they are probably many of the
basic geometric properties that would be used in modelling interactions that create
or modify topographies, or properties that influence topographic-dependent
behaviours.
These geometric properties depend on the specific positioning of the heights on
the surface. Multi-scale characterisations of these geometric properties inherently
provide more knowledge about the surface than conventional analyses. Consider,
for example, the conventional characterisation parameters that describe the sta-
tistics of the distribution of the heights on the surfaces, such as the three statistical
moments, standard deviation, or RMS roughness Sq, skewness Ssk, and kurtosis
138 C. A. Brown
Sku (see Chap. 2). These conventional parameters do not take into account the
relative position of the heights on the surface. In the absence of levelling and
filtering, any ordering of the same collection of measured heights on the surface
would give the same statistical results.
6.2.2 Notes on the Use of the Term ‘‘Scale’’
In this chapter the term ‘‘scale of calculation’’ is used to refer to the physical size
of the virtual object used to calculate length or area. The scale of calculation in this
chapter can have units of length or area. It is noted that basic geometric properties,
such as length, slope and area, can vary with the size of the physical element used
to assess them on a particular object.
The ‘‘scale of observation,’’ a term that is often found in the literature on
fractals, can be interpreted as the scale of measurement or as the scale of calcu-
lation. A topographic measurement, however, contains a range of scales, from the
overall dimensions of the measured region to the sampling interval between the
measured heights. In the analyses discussed here, the values of the geometric
properties of length, slope and area, are based on calculations that are performed at
different scales on the same topographic measurement. These calculations can be
performed over the range of scales contained in the measured surface.
6.3 Length-Scale Analysis
A discussion of length-scale analysis, which is essentially a profile analysis

method, is appropriate in a book on areal methods, for two reasons. One reason is
that length-scale analysis was developed and published first and formed the basis
for developing areal fractal analyses. Therefore length-scale analysis contains the
essential elements of areal, scale-dependent topographic analysis in a simpler
form. Another reason is that length-scale analysis can be used to study anisotropy
on a measured surface by applying it in different directions and comparing the
results as a function of direction. Multi-directional length-scale analysis is a
method for characterise anisotropy that requires areal measurements.
The length of coastlines was investigated by Richardson, as cited by Man-
delbrot (1977). A familiar question is: ‘‘how long is the coast of Britain?’’ The key
observation was that the length of coastlines appears to depend on the scale of
calculation. A coastline can be evaluated at different scales; for example, it can be
measured on maps with different levels of detail. At the finer scales, more bays and
peninsulas are evident. This tends to increase the apparent length of the coastline
beyond calculations based on less detailed maps. The analysis has been called the
coastline, Richardson, and the compass method. The latter comes about because
the length of a coastline can be determined at different scales on one map by
stepping around the coastline with a set of dividers, also called a compass. The
scale in this case is the distance separating the two points defined by the compass
or dividers. The inference is that the length of a coastline-like profile, i.e. a profile
with chaotic geometric components, can appear as if it could become infinitely
long when the size of the measuring step tends to zero. There is an intriguing
inference: an infinitely long coastline can encircle an island that is of finite pro-
jected area. The reason for the use of the qualifier projected, will be evident in the
discussion of area-scale analysis.
Applying the compass method to profiles (see Fig. 6.4) could suggest that, if
there were no limit to the fineness of the scale, then there should be no limit to the
length. Practically, there are limits to the ability of instruments to resolve the
topographic details. Also, there are scale limits in the physical nature of surfaces.
Clearly, the mathematical models of topographies that have infinite fineness of
scales, and never ending intricacy of details, will no longer be applicable at the
scale of atoms.
Figure 6.4 shows the compass method applied to a cartoon profile according to
ASME B46.1 (2009). The calculation algorithm steps along the profile like a
compass. This stepping exercise is equivalent to tiling the profile with line seg-
ments. The scale is the separation of the legs of the compass, or the step length,
Fig. 6.4 Three tiling exercises for length-scale fractal analysis on a simulated profile. The scale
is represented by the length of the step and decreases with each tiling exercise. The relative length
is the ratio of the measured to the nominal length. The nominal lengths are indicated in each case
140 C. A. Brown
which is represented by the length of the line segments in Fig. 6.4. The calculated
length is the step length multiplied by the number of steps. The step length is
constant, regardless of inclination in each tiling exercise. The stepping, or tiling,
exercise is repeated over a range of step lengths (Brown and Savary 1991).
The height samples that compose the profile are represented by points spaced at
the sampling interval. The ending point of the tiling exercise depends on the
starting point and the scale. When the step lands in between two points, which it
generally does, then a linear interpolation between the two adjacent measured
points is used to locate the start of the next step.
The relative length, at a particular scale, is the calculated length at that scale,
divided by the projected, or nominal, length of the portion of the profile that was
measured in the stepping exercise at that scale. The projection is made onto the
datum, which serves as a nominal, horizontal, straight reference. The relative
length is a dimensionless, normalised length and must be greater than or equal to
one. The relative length varies with scale, generally increasing as the scale
decreases, and is given by

calculated length ðiÞ scale ðiÞ
Rel Li ¼ ¼ Ni ð6:3Þ
nominal length ðiÞ nominal length ðiÞ
where i is the scale and N is the number of lengths.

Using the relative length rather than the calculated length accomplishes two
things. Firstly, the relative length accounts for the different end points encountered
at different scales in the tiling algorithm. It achieves this by normalising by the
projected length. Secondly, the relative length provides a dimensionless charac-
terisation parameter for the extra length of the profile, i.e. the profile tortuosity, at a
particular scale.
The relative length at a particular scale is also equal to a weighted average of
unity divided by the average cosine of the inclinations havri on the surface at that
scale
X
Rel Li ¼ 1=cos havri ¼ j
1=cos hj pij =Pi ð6:4Þ
where i indicates the scale, Pi is the length along the datum of the profile used for
normalising at the scale i, pij is the length of the jth line segment projected onto the
datum, and hj is the inclination of the jth element. The weighting factor is pij /Pi.
This weighting factor tends towards zero as 1/coshj tends to infinity, which keeps
the function well behaved. In any event, the inclination between two measured
heights on a measured profile can never reach 90°, regardless of the difference in
heights, because there is always a finite sampling interval between the two heights.
The relative lengths can be plotted against scale as a log–log, length-scale plot
(ASME B46.1 2009), as shown in Fig. 6.5. At larger scales, the relative areas tend
to unity for a profile measured on a surface that is flat and smooth at such scales.
At some finer scale, the relative lengths just start to deviate importantly from unity.
Fig. 6.5 Length-scale plots, parallel and perpendicular to the turning axis. The approximate
SRCs are indicated. The SRC for the plot relative lengths perpendicular to the turning axis is at a
scale markedly finer than parallel. The relative minima on the plot of relative lengths parallel to
the turning axis are even are at multiples of the feed
This is when the line segments become short enough, compared to the topographic
features, so that they begin to tilt significantly when the ends land down in the
valleys and up on the peaks of the profile. This finer scale is the SRC, the scale
below which the surface is rough and above which it is smooth (ASME B46.1
2009). For profiles with chaotic elements at the finest scales, the relative lengths
continue the tendency to increase as the scale decreases.
When the profile is self-similar over some range of scales, the logarithm of the
relative area will increase linearly, with a decrease in the logarithm of the scale.
The length-scale fractal dimension is the difference between unity and the slope of
the length-scale plot (ASME B46.1 2009)
Dls ¼ 1 slope: ð6:5Þ
Because the slope is always zero or negative when considered over significant
scale ranges, the fractal dimension of a profile will always be greater than or equal
to unity. When the profile is self-affine, the slope of the length-scale plot and,
therefore, the fractal dimension, will change steadily with respect to scale. The
qualifier ‘‘length-scale’’ on the fractal dimension is used to distinguish it from the
fractal dimension calculated by other means.
142 C. A. Brown
The area-scale complexity (Asfc) is defined as -1,000 times the slope of the
length-scale plot (ASME B46.1 2009). Steeper negative slopes indicate greater
complexity.
At the finest scales, when the sampling interval is reached, the complexity goes
to zero. That is because there is no more topographic detail recorded in the
measurement at scales finer than the sampling interval. Therefore, tiling algorithms
do not need to extend to scales below the sampling interval.
If there are periodic elements in the profile, aliasing can occur in the stepping
exercise. This is apparent because the relative length decreases steadily with a
decrease in scale over a small range of scales (see Fig. 6.5). There will be a series
of relative minima occurring when the scale is equal to even multiples of the
spatial wavelength of the periodicity (Brown et al. 1996). This property can be
useful for finding periodicity in profiles.
In implementing the length-scale tiling method, the selection of the starting
point is arbitrary and generally the start of the measured profile. The exact result of
the tiling, i.e. the relative length at that scale, depends to some degree on the
selection of the starting point. Certain starting points on profiles with periodic
elements can result in aliasing at half the wavelength of the period. These starting
points will be indicated by relative minima in relative lengths, at scales of half the
wavelength as well. This would be the case with a sine wave, when the half wave-
length step begins at zero or some integer multiple of p.
The differences in the length-scale plots parallel and perpendicular to the
direction of tool motion in Fig. 6.5 suggest a means for characterising anisotropy
in areal measurements. The relative lengths can be compared in multiple directions
to characterise anisotropy. Such a method has been developed by anthropologists
to study diets by analysing dental microwear (Scott et al. 2005, 2006).
A measure of anisotropy called exact proportion length-scale anisotropy of
relief (epLsari) was developed by Scott et al. (2005, 2006) to quantify the degree
of directionality in a surface at a scale of i. At a given scale, epLsar is the
fractional part of the relative length in a certain direction, normalised by the sum
of the fractional parts’ relative lengths in all the other directions at 5° intervals,
ðRel Lai 1Þ
epLsari ¼ P : ð6:6Þ
a ð1 Rel Lai Þ
A rosette can be plotted consisting of normalised relative length vectors in their

respective directions. The epLsar parameter utilises the variation of anisotropy
with scale to characterise an important component of the geometric character of
the surface.
The scale dependence of anisotropy means that characterisation algorithms
should account for different sampling intervals in the different directions on
measured surfaces. The structure of the height data in the measurement restricts
the assessment algorithms for anisotropy.
Measured surface textures generally consist of heights measured at regular
intervals in two orthogonal directions. Consider the case where the sampling
intervals in the x and y directions are equal. Then at 45° to the x and y axes, the
sampling interval would be the square root of two (H2) multiplied by the sampling
interval in the x and y directions. After the x and y directions, 45 and 135° are
directions of the shortest sampling intervals. These are directions where the
locations of measured heights align in the structure of the measured data set. Other
directions will have longer sampling intervals. The method used by Scott and co-
workers only uses directions where the sampling intervals are within an order of
magnitude of the sampling interval in the x and y directions. The sampling interval
limits the finest scales that can be used to determine the anisotropy.
This scale-based characterisation for assessing anisotropy using epLsar has
provided value by discriminating microwear on teeth that are associated with dif-
ferent diets (Scott et al. 2005, 2006). Scott et al. (2006) also demonstrate, through a
cross-correlation matrix, that the epLsar measure of anisotropy is independent of the
other parameters that they selected for describing topographies of teeth.
6.4 Area-Scale Analysis
Just as fractal analysis shows that the length of a coastline can vary with the scale
of observation, so can the area of a rough surface vary with scale. Just as Rich-
ardson inquired about the length of the coast of Britain, one might just as well
inquire about the area of Switzerland. Area-scale analysis calculates the areas of
surfaces as a function of scale (ASME B46.1 2009; ISO 25178-2 2012). The areas
of surfaces with chaotic components vary as a function of scale, in a similar
fashion to those of the lengths of profiles.
Area has a clear physical interpretation for understanding performance, in that
many of the interactions that control the performance depend on surface area. This
observation supports the potential value of area-scale analysis for discriminating
surfaces that behave differently and for correlating with performance and behaviour.
There are many topographically-dependent functions that depend on the area of
the surface at some scale. However, the best scale for demonstrating this depen-
dence may not be known. The intensity of interactions with and transport across
surfaces are typically modelled as dependent on the area available for interaction
and transport.
Newton’s equation for heat transfer across a boundary, for example, can be
written to show that the heat transfer rate is proportional to the temperature dif-
ferential times the area of the boundary. Generally, area of the boundary is cal-
culated at a macro-scale. The influence of the fine scale topography of the surface
area is then included in the heat transfer coefficient, which accounts for the
intrinsic heat transfer properties of the boundary. However, the topographic
components of this kind of heat transfer coefficient could be extracted from the
topography directly. This would require some appropriate analysis of the topog-
raphy to extract the effective area at the appropriate scale. This kind of charac-
terisation could then be used to advantage in finding correlations with many kinds
144 C. A. Brown
of performance observations and the topography. This additional knowledge about

the topography should be valuable and should support the design of better surfaces
and the processes to manufacture them. This additional knowledge should also
help to advance the scientific understanding of topographically related phenomena.
The developed interfacial area ratio of the scale limited surface Sdr (ISO
25178-2 2012; Chap. 2) can be used to calculate the developed area of a surface.
This method, however, calculates the area of the surface at only one scale, the
scale of the sampling interval used in the measurement. Most often, the selection
of the sampling interval is limited by the instrument and the accompanying soft-
ware. Such selection as there is generally would be based on measurement and
data storage considerations. For this reason, the developed area as a character-
isation of topographies has been of less value than if the scale at which it was
calculated were somehow adjustable. It would make more sense to base the
selection of scale on the abilities to discriminate surfaces or to correlate the sur-
faces with behavior. Both of these abilities would have value.
The approach used in area-scale analysis is intended to provide the ability to
calculate the apparent areas across the entire range of scales inherent in an areal
topographic measurement. The corresponding range of calculated areas, each
associated with a corresponding scale, can then be tested, scale by scale, for the
ability to discriminate surfaces suspected of being different, or to correlate with
behaviour that is thought to be topographically-dependent.
6.4.1 Triangular Tiling for Calculating Area
A schematic of a graphical representation of the tiling algorithm, which is discussed

below, is shown in Fig. 6.6 (Brown et al. 1993, 1994). In 2002, area-scale was
included in the US standards for characterising surface textures (ASME B46.1
2009). In 2012 area-scale analysis was also included in ISO 25178-2 (2012).
Unfortunately, there are some important, and probably unintended, differences
between the version in the current ISO standard and the ASME standard, so caution
is advised. It is recommended by the chapter author to follow the ASME version.
The area as a function of scale is calculated using a virtual tiling algorithm similar
to that used in the length-scale analysis. The length-scale algorithm analysis tiles
with line segments. The area-scale analysis tiles with triangles. The area of the
triangle represents an areal scale of the calculation. In each tiling exercise, the virtual
triangles used for tiling all have the same area in 3D. The area of the triangles
projected onto the datum, or nominal xy plane, will vary with their inclination.
The tiling algorithm used in these examples places the vertices on the tiling
triangles in line with one of two active rows or columns of heights. These active
rows or columns are separated in the x or y direction by a distance which is the
square root of twice the area of the tiling triangle. The tiling algorithm starts with
heights located along each of these rows or columns. The algorithm then looks for
the two adjacent heights along one or the other of the active rows or columns, such
Fig. 6.6 Triangular tiling for area-scale fractal analysis tiling exercise diagram schematic. The
dots represent measured heights. The inclination of the triangle is shown as hij. The projected area
of one triangle (ij) is shown as aij. The total projected area covered by the triangular tiling is Ai
that the areas of the triangles formed by these heights bracket the scale for the
current tiling exercise. A linear interpolation is performed, to find the location
along the active row or column that will result in a triangle of the appropriate area
in 3D for that tiling exercise. That point is used as one of the two starting points for
the next triangle. At the beginning of each row or column, the first two heights that
are used for two vertices of the first triangle in that row are measured. After the
first two vertices in a set of rows or columns, each of the other vertices of the
subsequent triangles would most likely be located on a height that is interpolated.
At one scale, the tiling exercise can start at any of the four corners and can proceed
in either rows or columns. The results can then be averaged. At the largest scales,
this procedure results in using more of the measured heights, and perhaps a better
representation of the area at that scale.
At each scale, a relative area is calculated. The relative area at a particular scale
i is the calculated area divided by the nominal area. The calculated area is the
number of triangles used in the tiling Ni multiplied by the scale, divided by the
nominal area of the surface covered in the tiling, thus

calculated area ðiÞ scale ðiÞ
Rel Ai ¼ ¼ Ni ð6:7Þ
nominal area ðiÞ nominal area ðiÞ
The relative area at a particular scale Rel Ai can also be related to the mean
inclinations on a surface. In this case, the angle havri represents the angle between a
normal to the triangular tiling element and normal to the nominal surface.
X
Rel Ai ¼ 1= cos havri ¼ j
1= cos hj aij =Ai ð6:8Þ
where aij represents the projected area of tiling triangle j at scale i, i.e. projected
onto the nominal xy plane. The area Ai represents the total projected area tiled at
the scale i.
Figure 6.7 shows a series of three virtual tiling exercises on the measurement
of the turned surface shown in Fig. 6.2. The corresponding area-scale plot is shown
in Fig. 6.8. Note that the horizontal scale is in areal units. This is according to
146 C. A. Brown
Fig. 6.7 Three tiling exercises at different scales performed on the measurement shown in
Fig. 6.2. The scales, and resulting relative areas, are indicated on the figure. The scale is the area
of the triangular tiles, which all have the same actual area and different projected areas,
depending on the inclination. Maximum z height is approximately 7.995 lm
ASME B46.1 (2009). If it were according to ISO 25178 part 2 (2012), the plot would
have linear units. The discussion and presentation that follows is all according to the
current ASME standard. To the author’s knowledge, all publications use the defi-
nitions in the ASME standard and none use the definitions that are particular to the
current ISO standard.
Fig. 6.8 Area-scale plot showing the relative areas as a function of scale calculated from the
kinds of tiling exercises shown in Fig. 6.7
At large scales, the relative areas are close to unity. At finer scales, the values
differ significantly from unity. The area-scale fractal complexity (Asfc) is the slope
of the area-scale plot over a one-decade-wide window in areal scale. A plot of the
area-scale complexity against scale is shown in Fig. 6.9. This is all similar to
length-scale analysis. The complexity is essentially the scale-based derivative of
the area-scale relationship.
6.4.2 Value Adding Applications of Area-Scale Analyses
Area-scale analysis is a form of multi-scale analysis, which characterises area, an

important geometric property. The relative area and the complexity (Asfc), at a
certain scale, can be used as characterisation parameters. The algorithm discussed
above can be applied meaningfully to a measurement, which covers a range of
scales. This range is limited by the size of the measurement region at the large end
and the sampling interval at the fine end.
148 C. A. Brown
Fig. 6.9 Complexity-scale plot, showing the scale derivative of the relative area versus scale.
This is related to the fractal dimension. This is the slope of the area-scale plot multiplied by
1,000. The fractal dimension is two minus twice the slope of the area-scale plot. This plot is based
on the area-scale plot shown in Fig. 6.8
It is important to note that many of the differences in surfaces or correlations

with behaviour can only be found over certain, limited scale ranges. Conventional
height parameters with conventional filtering will not be able to identify many of
these differences. This is because conventional height parameters tend only to be
sensitive to surface features at relatively large scales. Conventional height
parameters miss the differences that might be found between surfaces at some of
the finer scales.
Indications of differences in measurement and filtering can be shown using
area-scale analysis as a function of scale (Bergstrom et al. 2004). Area-scale
analysis can be used to observe the influences of filtering and measurement con-
ditions. The relative influences of high-pass and low-pass filters are clear in area-
scale plots (Cohen et al. 1997). These indications of the influence of filtering and
measurement conditions, which are evident on area-scale plots, can be useful for
assessing filtering and for selection of measurement systems and parameters.
The influence of topographies on the strength of interactions, such as adhesion,
with surfaces can sometimes be better understood, if the interaction can be
discretised with respect to the local area. In these cases, the strength of macro-
scopic interactions can be derived from discrete, microscopic interactions. In these
situations, where the interactions can be modelled as discrete, the macroscopic
behaviour, such as adhesive strength, can be modelled as the sum of contributions
of some finite number of discrete microscopic bonds at some fundamental scale of
interaction (Brown and Siegmann 2001). The units at the macroscopic scale would
be strength of the interaction per nominal area, i.e. macroscopic or global area. The
units for the discrete interactions would also be the strength per area. However, in
the discrete case, the area would be determined on the local microscopic scale, or
perhaps nanoscopic scale, appropriate to the fundamental interaction. For exper-
imental verification of this kind of model, the topography needs to be measured
and analysed at the appropriate scale.
In the case of adhesion of a coating on a substrate, the units to describe the
interaction are force per unit area. The force would be that which is necessary to
cause adhesive failure of the coating, i.e. to exceed the bond strength. The overall
bond strength is the result of a finite number of discrete, fundamental microscopic
or nanoscopic bonds. The size and nature of the bonds depends on the adhesive
system.
Each of the fundamental or discrete bonds would be modelled as having a
certain adhesive strength, i.e. force for de-bonding. Each of the bonds would
occupy a certain area, which would be defined locally on the surface at the scale of
the bond. This area could be the smallest area, or scale, at which the adhesive bond
would be able to exist between the coating material and the substrate. To a first
approximation, at least, the theoretical estimation of macroscopic adhesive
strength would, therefore, depend on the number of bonds that the topography
allows per nominal area on the surface.
The macroscopic interaction strength (Smacro) is the macroscopic interaction
strength (Fmacro) divided by the nominal area (Anom). The macroscopic interaction
strength is equal to the number of discrete interactions (N) times the strength of a
fundamental interaction Ffun
Fmacro Ffun
Smacro ¼ ¼N : ð6:9Þ
Anom Anom
The number of discrete interactions is equal to the total area of the surface,
calculated at the scale of a fundamental interaction ATfun divided by the area of a
fundamental interaction Afun. The expectation is that N would be a large number,
because the area for a discrete fundamental interaction should be small,
ATfun
N¼ : ð6:10Þ
Afun
Substituting and rearranging gives

ATfun Ffun
Smacro ¼ ¼ ðRel As ÞðSfun Þ: ð6:11Þ
Anom Afun
150 C. A. Brown
The ratio of the total surface area at the scale of a fundamental interaction to the
nominal area is the relative area at the fundamental scale (Rel As). The ratio of the
strength of the fundamental interaction to the area of the fundamental interaction
characterises the fundamental interaction in the same units as the macroscopic
interaction. The relative area at the fundamental scale is a dimensionless parameter
that characterises the topography appropriately for this interaction. In this way, it
can be seen that relative area has a clear physical interpretation as an appropriate
characterisation parameter in the context of some kinds of topographically-
dependent surface interactions that can be modelled as discrete.
In the case of adhesion, this kind of area-scale characterisation has been applied
to thermal spray coatings on grit-blasted substrates. Correlation coefficients (R2)
approaching 0.9 were found between the relative area and adhesion over a small
range of scales (Brown and Siegmann 2001). In the case of mass transfer, area-
scale analysis has been applied to carburisation (Karabelchtchikova et al. 2007)
and to oil adsorption in fried foods (Moreno et al. 2010). In both cases, strong
statistical correlations are found between the characterisation of the transport and
the relative areas in a certain scale range.
These examples of adhesion and of mass transfer are cases of correlations of the
second kind, correlations between a topographic characterisation parameter and
characterisation of the surface performance or behaviour. In these experiments, a
variety of surface roughnesses are created. Over a range of scales, the strength of
the interaction is regressed against the relative areas. The scale of the strongest
regression coefficient is sought. The corresponding scale could be said to be the
scale of the fundamental interaction.
In another case of an interaction of the second kind, the friction between sheet
metal and a machined die surface, strong correlations are found between friction
coefficients and the relative areas over a narrow range of scales (Berglund et al.
2010a). In that study, even stronger regression coefficients were found when
regressing friction coefficients with the area-scale fractal complexity. With the
same experimental results and conventional height parameters, strong correlations
were also found, but only when using a narrow band-pass filter (Berglund et al.
2010b). This created a topographic data set with narrow ranges of scale for the
measured textures. The values of the conventional height parameters were cal-
culated in a series of narrow bands covering the total scale ranges, and the scale of
the narrow band with the strongest correlation was sought (Berglund et al. 2010b).
Berglund et al. concluded that the scales used in finding the correlations are of
primary importance.
It is reasonable to suggest that the slopes on the harder die surface at a par-
ticular scale are an important geometric characteristic for the surface to use in
determining the friction with the sheet metal. It could be that conventional height
parameters are indicative of slopes on the surfaces, and surface areas, at a par-
ticular scale. When the scale over which they are calculated is restricted, by using
a band-pass filter, then the slopes would apply to this scale range. This seems
likely and should be tested. If it is true, then it would also be true that the height
parameters would relate to the apparent surface area at a particular scale. In that
case, the height parameters calculated over narrow scale ranges could be said to be
another kind of multi-scale fractal property. Correlations of the first kind, for
forming topographies have been found with the fracture of chocolate. The frac-
tographic properties of chocolate depend on the temperature of the chocolate when
it is fractured. Chocolate is an interesting composite material that bears some
similarity to tool steel, in that it is a mixture of ductile and brittle phases. The
fracture surface of chocolate can bear a strong resemblance to fracture surfaces on
tool steels. Particularly strong correlations are found between the relative area of
the fracture surface over a particularly narrow scale and the temperature of
chocolate when fractured (Cantor and Brown 2009).
Area-scale fractal analysis can also be used to discriminate surfaces made with
different processes. In the case of ground polyethylene ski bases, relative areas at a
particular scale have been shown to discriminate surfaces ground with slightly
different conditions. To demonstrate the capability, an F-test is used, scale by
scale, over a range of scales. The F-test is used to determine, as a function of scale,
the confidence level for discriminating measurements on surfaces created with two
different grinds. Area-scale analysis was successful in discriminating surfaces of
ground polyethylene ski bases that could not be distinguished by conventional
parameters (Jordan and Brown 2006). In the case of pharmaceutical compacts,
relative areas are also shown to be capable of discriminating surfaces where other
texture characterisation methods have failed (Narayan et al. 2006).
6.5 Concluding Remarks
Many surface topographies have chaotic components and exhibit fractal properties
at some sufficiently fine scale. These surfaces might be best described over these
scales through some kind of fractal analysis.
From the theory associated with fractal geometries it is known that essential
geometric properties, such as, lengths, areas and slopes, change with respect to the
scale of calculation (i.e. observation or measurement) on chaotic surfaces.
There may be narrow scale ranges where some geometric properties will cor-
relate with processing or performance, or the use of such properties will permit
discriminations on surfaces that were processed differently or on surfaces that
perform differently.
For topographic analysis to facilitate the discovery of functional correlations
and the ability to discriminate surfaces, both the appropriate aspect of geometry
(for example, surface area) and the appropriate scale could be considered. Scale-
sensitive fractal analysis has the ability to provide this kind of facilitation.
Acknowledgments The author of this chapter would like to gratefully acknowledge the
excellent editing done by Shelia McAvey, the work on the figures done by Steven Kordell and
Adam Lemoine both WPI’16, the generous support of Olympus for use of the LEXT OLS 4000
scanning laser microscope and Digital Surf for use of Mountains analysis software. Surfract,
152 C. A. Brown
which supplies the length-scale and area-scale analysis software, is owned by the author of this
chapter, Christopher A. Brown.
References
ASME B46.1 (2009) Surface texture (surface roughness, waviness, and lay). Am Soc Mech Eng
Berglund J, Brown CA, Rosen BG, Bay N (2010a) Milled die steel surface roughness correlation
with steel sheet friction. Ann CIRP 59:577–580
Berglund J, Agunwamba C, Powers B, Brown CA, Rosén BG (2010b) On discovering relevant
scales in Surface roughness measurement—an evaluation of a band-pass method. Scanning
32:244–249
Bergstrom TS, Hamel RS, Kummailil J, Gray AR, Brown CA (2004) Comparison of surface
texture measurement systems. In: XI Int Colloq Surfaces, Feb, Chemnitz, Germany, pp 13–21
Brown CA, Savary G (1991) Describing ground surface texture using contact profilometry and
fractal analysis. Wear 141:211–226
Brown CA, Siegmann S (2001) Fundamental scales of adhesion and area-scale fractal analysis.
Int J Mach Tools Manuf 41:1927–1933
Brown CA, Charles PD, Johnsen WA, Chesters S (1993) Fractal analysis of topographic data by
the Patchwork method. Wear 161:61–67
Brown CA, William A, Johnsen P, Charles D (1994) Method of quantifying the topographic
structure of a surface. U.S. Patent 5,307,292
Brown CA, Johnsen WA, Butland RM (1996) Scale-sensitive fractal analysis of turned surfaces.
Ann CIRP 45:515–518
Cantor GJ, Brown CA (2009) Scale-based correlations of relative areas with fracture of
chocolate. Wear 266:609–612
Cohen DK, Brown CA, Johnse WA, Hoch P (1997) An investigation of filtering on 3D surface
texture measurements using scale-sensitive fractal analysis and PSD advances in surface
metrology. In: Proceedings of ASPE, Raleigh, USA, Nov, pp 30–35
De Chiffre L, Lonardo P, Trumphold H, Lucca DA, Goch G, Brown CA, Raja J, Hansen HN
(2000) Quantitative characterisation of surface texture. Ann CIRP 49:635–652
ISO 25178-2 (2012) Geometrical product specifications (GPS)—Surface texture: Areal—Part 2:
Terms, definitions and surface texture parameters, TC 213 - Dimensional and geometrical
product specifications and verification
Jordan SE, Brown CA (2006) Comparing texture characterisation parameters on their ability to
differentiate ground polyethylene ski bases. Wear 261:398–409
Karabelchtchikova O, Brown CA, Sisson RD Jr (2007) Effect of surface roughness on kinetics of
mass transfer during gas carburizing. Int Heat Treat Surf Eng 1:164–170
Kaye B (1994) A random walk through fractal dimension. Wiley-VCH, Berlin
Majumdar A, Bhushan B (1991) Fractal model of elastic-plastic contact between rough surfaces.
J Tribology Trans ASME 113:1–11
Majumdar A, Tien CL (1990) Fractal characterisation and simulation of rough surfaces. Wear
136:313–327
Mandelbrot BB (1977) Fractals form chance and dimension. WH Freeman, San Francisco
Mandelbrot BB (1985) Self-affine fractals and fractal dimension. Phys Scr 32:257–260
McRae GA, Maguire MA, Jeffrey CA, Guzonas DA, Brown CA (2002) Atomic force microscopy
of fractal anodic oxides on Zr-2.5Nb. J Appl Surf Sci 191/1-4:94–105
Moreno MC, Bouchon P, Brown CA (2010) Evaluating the ability of different characterisation
parameters to describe the surface of fried foods. Scanning 32:212–218
Narayan P, Hancock B, Hamel R, Bergstrom TS, Brown CA (2006) Using fractal analysis to
differentiate the surface topography of various pharmaceutical excipient compacts. Mater Sci
Eng A: Struct Mater: Prop Microstruct Process 430:79–89
Russ JC (1994) Fractal surfaces. Plenum Press, New York

Schertzer D, Lovejoy S (1988) Multifractal simulations and analysis of clouds by multiplicative
processes. Atmos Res 21:337–361
Scott RS, Ungar PS, Bergstrom TS, Brown CA, Grine FE, Teaford MF, Walker A (2005) Dental
microwear texture analysis within-species diet variability in fossil hominins. Nature
436:693–695
Scott RS, Ungar PS, Bergstrom TS, Brown CA, Childs B, Teaford MF, Walker A (2006) Dental
microwear texture analysis. J Hum Evol 51:339–349
Suh NP (1990) The principles of design. Oxford University Press, New York
Thomas TR (1988) Rough surfaces. Imperial College Press, London
Whitehouse DJ (2010) Handbook of surface and nanometrology. CRC Press, Boca Raton
Zahouani H, Vargiolu R, Loubet J-L (1998) Fractal models of surface topography and contact
mechanics. Math Comput Model 28:517–534
Chapter 7
Choosing the Appropriate Parameter
Franz Helmli, Kerstin Pötsch and Claudia Repitsch
Abstract Modern production has to fulfill the needs of smarter surfaces and
cost-effectiveness. In both cases the measurement and analysis of the surface
texture of functional surfaces is a valuable process. Surface texture analysis has
developed a great deal during the last decade. National and international standards
have been defined. One problem in modern surface metrology is the presence of
too many available parameters. It is difficult for a typical operator to decide which
surface texture parameters are most appropriate for monitoring whether the desired
surface properties are met. In this chapter several approaches for how to identify
the most suitable parameters for an application are presented. The approaches are
based on classification algorithms which are widely used in the field of machine
learning. After presenting the necessary mathematical background, the application
of the classification algorithms for surface texture analysis are discussed. Finally, a
practical example on the discrimination of new and used abrasive paper is given in
which, twenty-one surface texture parameters were measured. Classification is
used to identify those parameters that are most suitable for the discrimination
between new and used abrasive paper.
7.1 Problem Description
Before going into detail about specific solutions to the problem of choosing an
appropriate parameter, the basic definition of the problem and some definitions are
required. It is assumed that a set of surface parts g1… gn of class G, and a set of
surface parts b1… bm of class B, for example a set of good parts and a set of bad
parts respectively, is present. Each of these parts can be described by a set of
F. Helmli (&) K. Pötsch C. Repitsch

Alicona Imaging GmbH, Dr.-Auner-Straße 21a, 8074 Raaba/Graz, Austria
e-mail: Franz.Helmli@alicona.com

156 F. Helmli et al.
surface texture parameters. Let u be a new surface part. Is it possible to decide

whether u belongs to class G or to class B based on the set of surface texture
parameters? Which of the surface texture parameters can be used for such a
decision?
The above problem is known as classification. In this chapter the classification
of surface parts based on surface texture parameters is presented.
The following section is an introduction to the subject, including the basic
principles of classification and evaluation, the state of the art and the mathematical
background. Next a simple method known as ‘‘thresholding’’ is introduced for
choosing the appropriate surface texture parameter for classification. In the third
section, additional classification methods, such as Naïve Bayes and Boosting are
described. Finally, a practical example on abrasive paper and a conclusion are
presented.
7.2 Introduction
Classification is a very broad research field. This section discusses the basic
principles and mathematical background of classification, and state of the art
methods.
7.2.1 Classification
Classification can be described as the process of assigning class labels to samples

(Duda et al. 2001; Sonka et al. 2007). Classification can be grouped roughly into
binary classification and multi-class classification. In binary classification, only
two classes are involved, whereas in multi-class classification, the differentiation is
made on three or more classes. A special case of classification is one-class clas-
sification, which is discussed in more detail later in this chapter.
The process of classification can be split into four steps: (1) build the training
data by measuring samples and their features, (2) learn a classifier by setting up a
decision rule, (3) classify new unknown samples, and (4) evaluation. Classification
with labeled training data is called supervised classification. If no labels are
available, the classification is called unsupervised classification. For binary and
supervised classification with class labels, the four steps are given in more detail
below.
Step 1: Building the training data

The first step involves the measurement of surface samples p1 . . . pnþm (of class
G and class B) and the calculation of the surface texture parameters. From each
part pi a series of surface texture parameter si ¼ si1 . . . sik can be calculated, which
7 Choosing the Appropriate Parameter 157
build feature vectors for each part. For each sample, and consequently for each
surface texture parameter, it can easily be determined whether it belongs to class
G or to class B. Therefore, the training data {(s1,y1),…, (si, yi),…,(sn+m, yn+m)} can
be built; where yi[{0,1} defines the label to which class the parameter belongs.
Step 2: Learn a classifier

Given the trainings data, the next step is to set up a classifier C, which computes a
class label for each new input sample. There are several methods to learn such a
classifier, for example Naïve Bayes, Boosting, Support Vector Machines, etc. (see
Sect. 7.3), and some methods are discussed later in this chapter for the application
of surface texture classification. The central idea behind the learning process is to
find features (surface texture parameters), which are discriminative for one class,
but not for the other.
Step 3: Classify unknown samples

In the third step, the learned classifier can be used to classify an unknown surface
part u. For this, the part is measured to determine the surface texture parameters
s1… sk. These surface texture parameters are the input of the classifier C,
c ¼ Cðs1 . . .sk Þ with c 2 f0; 1g:
Given a set of surface texture parameters, the optimum case would be to have
one single parameter for the classification. In this case a threshold can be defined
to distinguish between class G and class B. A simple method to find such a surface
texture parameter and its discrimination threshold is described in the following
section. If no such single parameter is available, a more sophisticated learning
procedure can be used to setup a classifier, for example a combination of surface
texture parameters (see Sects. 7.3 and 7.4).
Step 4: Evaluation
After a classifier has been learned, its quality needs to be measured. A classifier
assigns a class label to an unknown sample. One method for evaluation is the
receiver operating characteristic (ROC) analysis (see Metz 1978; Fawcett 2006).
In binary classification, there are four possible outputs:
• true positive (TP): classifier output is G and expected value is G;
• true negative (TN): classifier output is B and expected value is B;
• false positive (FP): type I error—classifier output is G but expected output is B;
and
• false negative (FN): type II error—classifier output is B, but expected value is G.
One possibility to specify the accuracy of a classifier is to calculate the sen-
sitivity [also called the true positive rate (TPR)] and the specificity [also called the
true negative rate (TNR)] or similarly, the false positive rate (FPR) and the false
negative rate (FNR). These terms are summarised below:
#TP #TP
sensitivity ¼ TPR ¼ ¼ ;
#P #TP þ #FN
#TN #TN
specificity ¼ TNR ¼ ¼ ;
#N #TN þ #FP
#FN
FNR ¼ 1 sensitivitiy ¼ ; and
#P
#FP
FPR ¼ 1 specificity ¼
#N
where P is the number of positive training samples and N is the number of negative
training samples, and # represents ‘‘the number of’’.
The accuracy is then given by
#TP þ #TN
accuracy ¼ :
#P þ #N
The ROC (Metz 1978; Fawcett 2006) curve is a graphical representation of the
FPR against the TP rate. In binary classification this curve is generated by vari-
ation of the discrimination threshold s. The point (0, 1) represents a perfect
classification and values that are above the line of no-discrimination are better than
those which would be obtained by random estimation. Figure 7.2 shows an ROC
curve for the distributions shown in Fig. 7.1. On the basis of the ROC curve, the
best operating point can be chosen depending of the cost of the TP, TN, FP and FN
decisions. This best operating point is the tangent to a line with defined slope.
Normally, this point is that which is the closest to the perfect classification point
lying on a 45 line. This means equal TP and FN costs.
Fig. 7.1 Illustration of TNR, FPR, TPR and FNR and variation of threshold s for an ROC curve
(adapted from Metz 1978)
Fig. 7.2 ROC curve showing the best operating point
7.2.2 Multi-Class Classification
In multi-class classification,
more than two classes are involved. Given K classes
and training data ðs1 ; y1 Þ; . . .; ðsi ; yi Þ; . . .; ðsnþm ; ynþm Þ with yi 2 f1; . . .; K g;
the aim is to learn a multi-class classifier C which computes a class label for a new
unknown sample. Some binary classification methods can be extended easily to the
multi-class case, such as Support Vector Machines or Naïve Bayes (see Sect. 7.4).
Other classification methods have to be adapted for multi-class classification. An
overview of multi-class classification methods can be found in Aly (2005).
7.2.3 Mathematical and Statistical Background
In this section some basic definitions used later in the chapter are presented. In
probability theory, two events A and B are independent, if and only if the prob-
ability of their intersection is equal to the product of their probabilities. Thus, the
following equation is satisfied, P(A \ B) = P(A)*P(B). Intuitively, probabilistic
independency means that the occurrence of one event has no influence on the
occurrence of the second event. For example, rolling a six with a die has no
influence on the second roll.
With P(A|B), the conditional probability of event A under the assumption that
event B has occurred is denoted.
Two events A and B are conditionally independent with respect to event C if
and only if the conditional probabilities fulfill the following equation:
PðA \ BjCÞ ¼ PðAjCÞ PðBjC Þ:
Bayes’ Theorem states that the conditional probability of an event A, knowing

the occurrence of a further event B is given by
PðBjAÞ Pð AÞ
PðAjBÞ ¼ :
PðBÞ
Bayes’ Theorem is often used to compute posterior probabilities given obser-
vation. In this context the following wording is common:
• P(A): prior probability, i.e., the initial degree of belief in A, having made no
observation so far;
• P(A|B): a posterior probability, i.e., the degree of belief in A, after having
observed B; and
• PðBjA ÞPð AÞ
PðBÞ : is called the support that B provides for A.
Applying Bayes’ Theorem in the field of classification, event A represents a

class, and event B describes a sample, i.e., it is the feature-vector s = (s1…sk) of a
sample. For each class, the posterior probability P(A|B) is calculated, and the
sample is assigned to the class with the highest posterior probability. In Sect. 7.4.1,
the Naïve Bayesian classifier is described in more detail.
The probability density function (PDF) of a normal distribution (sometimes
referred to as a Gaussian distribution) is defined by (Ross 2005)
1 1 xl 2
fnð xÞ ¼ pffiffiffiffiffiffi e2ð r Þ ð1\x\1Þ
r 2p
where l is the expectation value and r is the standard deviation. A normal dis-
tribution with l = 0 and r2 = 1, is called a standard normal distribution. The
notation of a normally distributed variable X is X * N(l, r2).
The normal PDF is a bell-shaped symmetric curve (see Fig. 7.3). More accu-
rately, the curve is symmetric about l and the height and the width depend on the
parameter r. For different k, the following probabilities are known:
• k = 1: 68 % is covered between l - r and l ? r;
• k = 2: 95 % is covered between l - 2r and l ? 2r;
• k = 3: 99.7 % is covered between l - 3r and l ? 3r;
• and so on.
7.3 Thresholding
After surface texture parameters have been calculated, the question often arises:
which one is significant? There are several methods to calculate the significance of
parameters. In the following, the correlation coefficient method and the more
robust average and standard deviation method are described.
Fig. 7.3 Gaussian bell curve—approximate areas under a normal curve for k = 1, 2, 3
The easiest method to choose a significant parameter is to use a single

parameter for discrimination between two classes, G and B. In most situations this
method will be successful if the significant spatial bandwidth of the surface texture
can be found. As this method is simple, but not very powerful, a significant
parameter has to be found by the classifier. A brute force method is to use many
possible surface texture parameters at different spatial bandwidths. A given dataset
containing areal information between a given S-filter and L-filter, can be divided
into smaller spatial bandwidths in the following manner:
S … L1
L1 … L2
L2 … L3
Ln … L
Each interval gives a set of surface texture parameters. The advantage is that this
can help in selecting the L-filter that gives the most significant parameters. For
such an investigation, the measurement should also have an appropriate spatial
bandwidth.
7.3.1 Correlation Coefficient
In the case of a classification problem with only two classes, G and B (for example,
good and bad parts), a very simple way of finding the appropriate parameter is to
use the cross correlation coefficient. Using the data from the training set, for each
surface texture parameter Sv, the cross correlation coefficient rGB between the two
classes is calculated according to
Pn G;i G B;i B
i¼1 ðsv l Þðsv l Þ
rGB ðSv Þ ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Pn P ffi:
ðs G;i
l G Þ2 n
ðs B;i
l B Þ2
i¼1 v i¼1 v
B;i
Here lG (lB) denotes the mean of the samples in class G (class B) and sG;i
v sv is
the texture parameter Sv of the ith sample in class G (class B). The parameter
n denotes the samples in class G (class B).
The parameter with the highest absolute cross correlation is the most significant
parameter. Comparing the calculated cross correlation coefficients rGB ðSv Þ; for
v = 1,…, k, results in one of the cases described below.
1. The maximum absolute cross correlation is close to one. In this case the cor-
responding parameter can be used as a classifier.
2. The maximum absolute cross correlation is close to zero. This result indicates
that most likely a single parameter cannot be used as a classifier. In this case, it
would be advisable to apply a more sophisticated approach. A possible alter-
native is Boosting, which is described in more detail in Sect. 7.4.2.
7.3.2 Average and Standard Deviation Method
The average and standard deviation method analys.es the variation of each
parameter on the basis of the standard deviation and confidence intervals. The
method can be summarised in the following steps:

1. For each parameter s0 i ¼ s0 1i . . . s0 ni of class G and s00 i ¼ s00 1i . . . s00 ni of class
B, the average l and the standard deviation r is calculated
Pn Pn
l0 i ¼ 1n k¼1 s0 ki l00 i ¼ 1n k¼1 s00 ki ;
pffiffiffiffiffiffiffi 0 pffiffiffiffiffiffiffi 00
r0 i ¼ varðs i Þ r00 i ¼ varðs i Þ:
2. For each parameter, an interval for good parts and for bad parts is calculated
with the coverage factor k,
I 0 i ¼ l0 i k r 0 i ;
I 00 i ¼ l00 i k r00 i :
3. If the intervals I 0 i and Ii00 for a parameter si are disjunct, this parameter can be
used for thresholding and the significance Si of this parameter can be computed.
The parameter with the highest significance value is that which can be used for
classification. To find the most significant surface texture parameter, the sig-
nificance values must be comparable. This is achieved by normalising them
with the average values. The significance Si is computed on the basis of the
intervals and the means
dðI 0 i ; I 00 i Þ
Si ¼ 1 0 00
:
2 ðl i þ l i Þ
4. One method for computing the discrimination threshold s is to compute the

intersection point of the densities by solving the quadratic equation for s,
ðs l00 i Þ2 ðs l0 i Þ2 r0 i
00
0
¼ 00 :
2r i 2r i r i
Another method for the computation of the discrimination threshold s is based
on the ROC curve, by choosing the best operating point. If the intervals are not
disjunct, then the classification errors can be computed by the FPR and the FNR.
7.3.3 Example of Computation of the Significance Value Si
Using the average and standard deviation method for parameter s0i of class G and
s00i of class B, the computation of the significance value Si is demonstrated. On the
basis of the values (see Table 7.1) and Fig. 7.4, it can be seen that the two classes
are well separable on the basis of the surface texture parameter si (see also
Figs. 7.4 and 7.5). This surface texture parameter is an appropriate choice for
classification.
7.3.4 Example of Overlapping Intervals
There is not always a surface texture parameter with disjunct intervals. Never-
theless, it is possible to define a threshold for this sort of parameter, but with lower
accuracy. Table 7.2 shows the results of the average and standard deviation
method. Figure 7.6 shows overlapping intervals for k = 2. By computing the best
operating point, a threshold s with TPR \ 1 can be defined (see Fig. 7.7).
7.4 Classification Methods
Using the average and standard deviation method, the significance of each
parameter Si is known. By using the parameter with the highest significance, the
Table 7.1 Results of the average and standard deviation method

Class G Class B
s0 i s00 i
1 11.59702129 17.902779
2 8.084521223 21.87106
3 9.078186256 20.4325753
4 9.71231168 21.6981515
5 11.23946288 18.7105335
6 9.901993117 16.7368411
7 9.668845718 20.4125894
8 11.24789048 18.0754563
9 11.08010977 19.5692714
10 9.179953711 20.1539732
11 10.31249917 18.2433351
12 10.41928543 20.4759712
l0 10.12684006 l00 19.5235447
r0 1.056766798 r00 1.59409733
k 3 k 3
Imin 6.956539667 Imin 14.7412527
Imax 13.29714045 Imax 24.3058367
Si 0.097409343
s 13.947
TPR 1
Fig. 7.4 Gaussian bell curve for the example data in Table 7.1. The surface texture parameter
used is useful even with k = 3
Fig. 7.5 ROC curve for data in Table 7.1. The best operating point is in corner (0, 1) which
means perfect classification
Table 7.2 Results of the average and standard deviation method

Class G Class B
s0 i s00 i
1 0.79 0.67
2 0.82 0.43
3 0.62 0.52
4 0.68 0.53
5 0.71 0.68
6 0.74 0.56
7 0.88 0.48
8 0.63 0.60
9 0.77 0.47
10 0.76 0.60
l0 0.74 l00 0.55
r0 0.08219 r00 0.08383
k 2 k 2
Imin 0.58 Imin 0.39
Imax 0.90 Imax 0.72
Si -0.22 Negative
s 0.646653 Significant
TPR 0.87197
classification can be solved if the intervals are disjunct. But, what happens if there
is no surface texture parameter with disjunct confidence intervals? If there are
several parameters with high significance, would it be a good idea to use only one
of them for classification?
Fig. 7.6 Illustration of Gaussian bell curve for the example data in Table 7.2
Fig. 7.7 ROC curve of for the example data in Table 7.2
In the following section, two well-known classification methods, Naïve Bayes

and Boosting, are described.
7.4.1 Naïve Bayesian Classification
Naïve Bayesian classification is a statistical classification method which is based

on Bayes’ Theorem (see Sect. 7.2.3). Each new sample is assigned to the class with
the maximum posterior probability. The classification method works as follows:
1. A training set of n samples, {(p1, y1), …, (pi, yi), …, (pn, yn)}, each sample pi
with the correct class label yi [{{0, 1, … m} defining the class to which the
i
sample belongs is given. Thus, there are i m iclassesi in total. Each sample p is
represented by a k-dimensional vector s1 ; s2 ; . . .; sk depicting the k measured
values of the attributes S1…Sk, respectively.
2. Given a new sample pl with its attribute vector sl1 ; sl2 ; . . .; slk ; the Naïve
Bayesian classifier assigns sample pl to class C with the highest posterior
probability P Cj sl1 ; sl2 ; . . .; slk ; i.e., the sample is assigned to class Cj if

P Cj j sl1 ; sl2 ; . . .; slk [ P Cv j sl1 ; sl2 ; . . .; slk for all v 2 f0; 1; . . . mg:

The posterior probabilities P Cj j sl1 ; sl2 ; . . .; slk ; j ¼ 1; . . .; m are calculated by
exploiting Bayes’ formula. Additionally, in order to reduce the computational
costs, the assumption of conditional independence of the attributes S1…Sk within
each class is made. Thus, the following equations apply

P Cj j sl1 ; sl2 ; . . .; slk ¼

P sl1 ; sl2 ; . . .; slk jCj P Cj =P sl1 ; sl2 ; . . .; slk ¼

P sl1 jCj P sl2 jCj . . . P slk jCj P Cj =P sl1 ; sl2 ; . . .; slk :

Note, that the denominator P sl1 ; sl2 ; . . .; slk is equal for all classes and thus, only

the numerator P sl1 jCj P sl2 jCj . . . P slk jCj P Cj needs to be calculated
in order to identify class Ci with the maximum posteriori probability.
Here, estimates for the a priori probabilities P(Cj) and for the attributes prob-
ability distributions P slr jCj can be determined from the training data. These
estimates may be approximated with the relative frequencies given by the training
set.
The a priori probabilities P(Cj) can be estimated by the number of samples in the
training set that belong to class Cj, i.e. P(Cj) = (number of samples in class Cj)/N.
One further common approach is to assume a probability distribution (for
example,
a Gaussian distribution) for the attribute probability distributions
P slr jCj : Then, using maximum likelihood estimation, the underlying parameters
(for example, the mean and standard deviation of the Gaussian distribution) are
calculated from the training data.
Due to the conditional independence assumption, if one single attribute dis-
tribution is equal to zero, then the posterior probability is also zero. In order to
avoid this undesired behaviour, a small correction term may be added to all
probabilities. In the case of frequency-based estimates, a simple maneuver is
proposed in the literature namely, unity is added to all the numerators and the
denominators are increased by the number of added ones (Cestnik 1990).
The main advantage of the Naïve Bayes classifier is its short computation time
for training. Naïve Bayesian classifiers have been employed in a wide range of
applications (for example, text classification, data mining, medical domain), and
have exhibited high accuracy and speed when applied to experimental data (for
example, Cestnik et al. 1987).
However, Naïve Bayesian classification is based on the conditional indepen-
dence assumption, which may be unrealistic in certain application fields, for
example, medical diagnosis (correlation between age and certain symptoms) and,
therefore, it is usually less accurate than more sophisticated approaches. Several
modifications of the basic model have been proposed to increase its performance
(Friedman et al. 1997; Kononenko 1991).
7.4.2 Boosting
Boosting (Schapire 2003) is a machine learning algorithm which combines

approximate classification rules into one single higher accuracy prediction rule—a
so-called strong classifier. Such approximate classification rules are called weak
classifiers and often are simply based on one feature of the feature space, for
example one surface texture parameter. However, a single parameter may not be
sufficient to distinguish between a good and a bad part. To improve the accuracy, a
Boosting algorithm learns m weak classifiers ci (each with decisions better than
random estimation) to one strong classifier C.
There are a high number of different boosting algorithms, for example Ada-
boost, GentleBoost, LPBoost (Demiriz et al. 2002). One of the most popular
Boosting algorithms is AdaBoost (short for adaptive boosting). First, AdaBoost
was proposed by Freund and Schapire (1999). AdaBoost iteratively learns the
strong classifier as a weighted combination of weak classifiers, thus
X
T
C ð xÞ ¼ sgn at ct ðpÞ
t¼1
where ct are weak classifiers and at are the corresponding weights. Let ci be a
classifier that uses one single surface texture parameter si for classification. The
decision whether a part is good or bad is made on the basis of a threshold li,

þ1 if si li
ci ðsÞ ¼ ci ðs1 ; s2 ; . . .sm Þ ¼ :
1 if si \li
For classification of good or bad parts using surface texture parameters, the
AdaBoost method takes as input a training set of parameter vectors with known
labels (+1, -1). At the start of the process, the weights D1j of the parameters are set to
be equal. In each round of the training, the set of parameters is adapted by computing
new weights. Thus, weights of incorrectly classified parameter vectors are increased
and weights of correctly classified parameter vectors are decreased. The higher the
value of a weight; the more important is the parameter vector for the training of the
next classifier. The AdaBoost algorithm can be summarised as follows:
1. Input: n samples with parameter vectors pj with known status yj = (+1, -1).
2. Initialise each parameter vector pj equally with D1j ¼ 1=n; j ¼ 1; . . .; n:
3. For t = 1, …, T (T = number of classifier = number of parameters = m):
a. train weak classifier ct
b. obtain weak classifier ct with minimum error with respect to distribution Dtj
X
et ¼ Dtj
8i : ct ðp j Þ
is wrongly
classified
c. if et ¼ 0 or et [ 1=2 ) stop
d. calculate weight
1 1 et
at ¼ log
2 et
e. update
j i j
Dtþ1 ¼ Dtj eat y ct ðp Þ
4. output of final hypothesis C

X
T
C ð xÞ ¼ sgn at c t ð pÞ
t¼1
7.4.3 Other Classification Algorithms
The development of classification algorithms has been an active field of research.

Various approaches are proposed in the literature: logic based algorithms (decision
trees (Breiman et al. 1984) and rule-based classifiers), algorithms based on neural
networks (Bishop 1995), statistical learning techniques, support vector machines
(Vapnik 1995), cluster analysis, etc. A thorough review of supervised classification
(see Sect. 7.2.1 Classification) can be found in Kotsiantis (2007).
7.5 Practical Solutions
7.5.1 Dimensioning of the Task
For all the methods for choosing parameters discussed in this chapter, variation
across the parts being measured is needed in order to setup a classifier. Therefore,
several repeat measurements have to be made on different samples or different

positions over one sample. The number of repeat measurements is dependent on
the distribution of the heights on the surface.
7.5.2 Classification with No Bad Parts
Nowadays, industry spends a lot of money in machines for mass production in the
electronics and automobile industries. The actual accuracy and repeatability of
such machines is relatively high in relation to the available surface topography
measuring instruments. Due to the high accuracy of manufacturing machines, there
are often only samples of one class available (good parts) but no (or less) samples
of class B (samples with defects, outliers, etc.). However, to have just one dam-
aged part in production would be a big problem for industrial sectors such as
automobile or aerospace.
If just samples of one class are available, a binary or multi-class classification is
not applicable. A special case is the one-class classification (Tax 2001) (also
known as outlier detection or novelty detection), which assumes that the training
set contains only samples of one class G. Here, there are three special cases: the
training set contains only positive samples of class G, it contains positive samples
and negative samples, or it contains positive samples of class G and unlabeled
samples (Liu et al. 2003). Known methods for one-class classification are
boundary methods, density estimation or reconstruction methods (Tax 2001).
7.6 Practical Example on Abrasive Paper
One main application for surface texture classification is the discrimination

between good and bad parts (wear, defects, etc.). In this practical example the aim
is the discrimination of new (class G) and used (class B) abrasive papers and the
associated choice of appropriate surface texture parameters.
The training data consists of new abrasive paper of type Bosch with graining of
180 (class G) and of used abrasive paper of type Bosch with graining of 180
(class B).
These datasets were selected from measurements taken with a focus variation
microscope (Leach 2011) instrument (the InfiniteFocus from Alicona) with a 209
objective, a lateral resolution of 2 lm and a vertical resolution of 100 nm.
An example surface topography of each class is shown in Figs. 7.8 and 7.9. A
detailed cut-out in pseudo-colours is shown in Figs. 7.10 and 7.11. There is not
much visual difference between class G and class B (the high peaks are slightly
reduced in class B).
Twenty-one surface texture parameters for seven class G and eight class B sam-
ples were measured. The measured values are summarised in Tables 7.3 and 7.4.
Fig. 7.8 Bosch 180 new: 3D surface texture view
Fig. 7.9 Bosch 180 used: 3D surface texture view
The results of the average and standard deviation method are summarised in
Table 7.5. A value of k = 2 for the computation of the lower and upper bounds has
been used.
Fig. 7.10 Bosch 180 new: cut-out of 3D surface texture view in pseudo-colours
Fig. 7.11 Bosch 180 used: cut-out of 3D surface texture view in pseudo-colours
Table 7.5 shows that four parameters, Sp, Sz, FLTt and Vmp, have disjunct
intervals (Chap. 2 describes these parameters and FLTt is flatness). The signifi-
cance of these parameters can be calculated. The parameter with the highest
significance is Sp. However, Sp defines the maximum peak height of the selected
area. This parameter is not really significant for the discrimination between new
and used abrasive paper because it only describes one specific surface feature.
Table 7.3 Measured surface texture parameter values for Bosch 180 new
Bosch 180 new
Sample 1 2 3 4 5 6 7
Sa 22.48 24.91 24.42 23.36 23.39 23.07 23.09 lm
Sq 27.73 30.03 29.66 28.58 28.96 28.13 28.29 lm
Sp 114.36 115.71 112.13 134.20 123.37 115.60 124.00 lm
Sv 66.96 69.15 71.36 67.54 65.90 68.50 74.55 lm
Sz 181.32 184.86 183.49 201.74 189.27 184.09 198.55 lm
S10z 170.91 169.25 171.16 197.60 175.38 173.28 186.72 lm
Ssk 0.79 0.62 0.68 0.71 0.74 0.63 0.76
Sku 3.28 2.72 2.84 3.19 3.18 2.91 3.16
Sdq 1.32 1.35 1.36 1.38 1.38 1.27 1.32
Sdr 63.82 65.33 66.97 67.14 67.14 59.70 63.48 %
FLTt 181.32 184.86 183.49 201.74 189.27 184.09 198.55 lm
Sk 60.45 72.02 70.41 69.62 66.77 67.20 61.25 lm
Spk 40.43 35.34 38.19 36.44 40.03 36.75 40.46 lm
Svk 10.54 9.92 9.03 10.16 10.57 10.36 9.90 lm
Smr1 19.35 18.33 18.00 16.47 17.82 17.38 20.57 %
Smr2 95.42 96.08 96.51 96.79 95.38 96.59 96.12 %
Vmp 1.66 1.55 1.59 1.61 1.74 1.59 1.66 ml/m2
Vmc 23.63 27.36 26.87 25.75 25.23 25.42 24.56 ml/m2
Vvc 40.19 42.29 42.68 40.30 41.37 40.21 40.95 ml/m2
Vvv 1.66 1.75 1.68 1.66 1.79 1.61 1.52 ml/m2
Vvc/Vmc 1.70 1.55 1.59 1.56 1.64 1.58 1.67
Lc 800.00 800.00 800.00 800.00 800.00 800.00 800.00 lm
Table 7.4 Measured surface texture parameter values for Bosch 180 used
Bosch 180 used
Sample 1 2 3 4 5 6 7 8
Sa 21.79 22.49 21.87 22.69 22.04 23.01 23.24 21.53 lm
Sq 26.49 27.05 26.40 27.74 26.61 27.53 28.06 25.85 lm
Sp 86.60 88.04 92.18 92.62 91.68 92.33 88.31 92.68 lm
Sv 59.47 62.67 67.63 61.23 63.96 63.43 70.00 67.40 lm
Sz 146.07 150.71 159.81 153.85 155.64 155.77 158.31 160.08 lm
S10z 138.98 139.65 142.01 148.56 149.26 151.23 154.57 148.68 lm
Ssk 0.67 0.52 0.53 0.68 0.56 0.48 0.60 0.47
Sku 2.78 2.56 2.62 2.76 2.64 2.43 2.62 2.47
Sdq 1.23 1.25 1.23 1.35 1.33 1.36 1.38 1.19
Sdr 57.41 58.25 57.29 65.55 62.79 66.65 67.91 53.14 %
FLTt 146.07 150.71 159.81 153.85 155.64 155.77 158.31 160.08 lm
Sk 61.47 70.40 64.39 65.54 65.14 72.42 65.82 68.02 lm
Spk 35.76 30.12 31.93 37.55 32.04 28.11 36.05 27.08 lm
Svk 6.98 8.38 9.93 7.98 8.57 8.29 9.31 9.34 lm
Smr1 18.65 14.92 17.46 18.09 17.36 15.43 18.86 14.43 %
Smr2 96.46 96.84 96.40 96.87 96.41 97.23 96.75 95.75 %
Vmp 1.42 1.36 1.30 1.42 1.35 1.19 1.36 1.16 ml/m2
Vmc 23.88 25.29 24.30 25.12 24.25 26.37 25.74 24.33 ml/m2
Vvc 38.94 37.97 38.08 41.37 38.13 39.30 41.44 36.56 ml/m2
Vvv 1.43 1.64 1.56 1.55 1.61 1.62 1.50 1.71 ml/m2
Vvc/Vmc 1.63 1.50 1.57 1.65 1.57 1.49 1.61 1.50
Lc 800.00 800.00 800.00 800.00 800.00 800.00 800.00 800.00 lm
Table 7.5 Results of average and standard deviation method

Average and standard deviation method
l0 r0 l00 r00 I0 min l0 max I00 min l00 max Si
Sa 23.53 0.84189 22.33 0.56818 21.85 25.22 21.20 23.47
Sq 28.77 0.83401 26.97 0.66032 27.10 30.44 25.65 28.29
Sp 119.91 7.75082 90.56 2.50610 104.41 135.41 85.54 95.57 0.08
Sv 69.14 2.95568 64.47 3.63878 63.23 75.05 57.20 71.75
Sz 189.05 8.00165 155.03 4.67653 173.04 205.05 145.68 164.38 0.05
S10z 177.76 10.50829 146.62 6.09333 156.74 198.77 134.43 158.80
Ssk 0.70 0.06451 0.56 0.07631 0.58 0.83 0.41 0.72
Sku 3.04 0.21346 2.61 0.11874 2.61 3.47 2.37 2.85
Sdq 1.34 0.03958 1.29 0.06528 1.26 1.42 1.16 1.42
Sdr 64.80 2.72806 61.12 4.59353 59.34 70.25 51.94 70.31
FLTt 189.05 8.00165 155.03 4.67653 173.04 205.05 145.68 164.38 0.05
Sk 66.82 4.46312 66.65 3.72443 57.89 75.74 59.20 74.10
Spk 38.23 2.11349 32.33 3.46051 34.01 42.46 25.41 39.25
Svk 10.07 0.53168 8.60 0.94381 9.01 11.13 6.71 10.49
Smr1 18.27 1.33961 16.90 1.53053 15.60 20.95 13.84 19.96
Smr2 96.13 0.55689 96.59 0.30635 95.01 97.24 95.98 97.20
Vmp 1.63 0.06309 1.32 0.07931 1.50 1.75 1.16 1.48 0.02
Vmc 25.55 1.28090 24.91 0.89718 22.98 28.11 23.12 26.70
Vvc 41.14 1.02112 38.97 1.50653 39.10 43.18 35.96 41.99
Vvv 1.67 0.08864 1.58 0.07426 1.49 1.84 1.43 1.73
Vvc/Vmc 1.61 0.05765 1.57 0.06161 1.50 1.73 1.44 1.69
Fig. 7.12 Vmp values for

Bosch 180 new and Bosch
180 used. The values are
clearly separable at a
threshold of approximately
1.5
Here a manual pre-selection of parameters is useful. Even if the parameter Vmp

has a low significance, it should be chosen because it fits to the known sample
difference. The Vmp parameter is the amount of material present in the peaks
(above the core height).
Fig. 7.13 Gaussian curve for parameter Vmp with l0 = 1.63, l00 = 1.32, r0 = 0.06309,
r00 = 0.07931 and s = 1.492
Fig. 7.14 ROC curve and best operating point for parameter Vmp
Figure 7.12 shows the Vmp values for Bosch 180 new and Bosch 180 used. It
can be seen that the values are clearly separable so that a threshold can be defined
(see also Fig. 7.13). In contrast, a plot of the overlapping values for parameter Sa
is shown in Figs. 7.15 and 7.14 shows the ROC curve for the parameter Vmp. For
the chosen best operating point, the TPR = 0.98564.
Fig. 7.15 Overlapping

values for parameter Sa
7.7 Conclusions
In this chapter, methods for searching and finding the most appropriate surface
texture parameters have been described. Firstly, the theoretical part shows simple
methods, such as the average and standard deviation method, and more advanced
methods, such as the boosting algorithm. Secondly, a practical example on abra-
sive paper was shown.
References
Aly M (2005) Survey on multi-class classification methods. Technical Report, Caltech, USA
Bishop CM (1995) Neural networks for pattern recognition. Oxford University Press, Oxford
Breiman L, Friedman JH, Olshen RA, Stone CJ (1984) Classification and regression trees.
Wadsworth & Brooks, Pacific Grove
Cestnik B, Kononenko I, Bratko I (1987) Assistant 86: a knowledge elicitation tool for
sophisticated users. In: Proceedings of second European working session on learning,
Wilmslow, pp 31–45
Cestnik B (1990) Estimating probabilities: a crucial task in machine learning. In: Proceedings of
9th European conference on artificial intelligence (ECAI-90), Stockholm, pp 147–150
Demiriz A, Bennett KP, Shawe-Taylor J (2002) Linear programming boosting via column
generation. Mach Learn 46:225–254
Duda RO, Hart PE, Stork DG (2001) Pattern classification, 2nd edn. Wiley, New York
Fawcett T (2006) An introduction to ROC analysis. Pattern Recogn Lett 27:861–874
Freund Y, Schapire RE (1999) A short introduction to boosting. Jap Soc Artif Intell 14:771–780
Friedman N, Geiger D, Goldszmidt M (1997) Bayesian network classifiers. Mach Learn
29:131–163
Liu B, Dai Y, Li X, Lee WS, Yu PS (2003) Building text classifiers using positive and unlabeled
examples. In: Proceedings of Third IEEE international conference on data mining, Florida,
pp 179–186
Kononenko I (1991) Semi-naive Bayesian classifier. In: Proceedings of 6th European working
session on learning, Porto, pp 206–219
Kotsiantis SB (2007) Supervised machine learning: a review of classification techniques.
Informatica (Slovenia) 31:249–268
Metz CE (1978) Basic principles of ROC analysis. Semin Nucl Med 8:283–298
Ross SM (2005) Introductory statistics, 2nd edn. Academic, Elsevier
Schapire RE (2003) The boosting approach to machine learning: an overview. In: Lecture Notes
in Statistics, vol 171. Springer, New York, pp 149–171
Sonka M, Hlavac V, Boyle R (2007) Image processing, analysis, and machine vision, 3rd edn.
Thomson Learning, Toronto
Tax DMJ (2001) One-class classification. Thesis, Delft University of Technology, Delft
Vapnik V (1995) The nature of statistical learning theory. Springer, New York
Chapter 8
Characterisation of Individual
Areal Features
Nicola Senin and Liam Blunt
Abstract Along with the characterisation of areal surface topography through tex-
ture parameters, surface metrology is increasingly facing challenges related to the
dimensional and geometric characterisation of individual surface features. Typical
scenarios range from the inspection of individual elements of micro-parts and
devices, to the characterisation of pattern units in structured surfaces, and to the
analysis of scratches, pores, bumps and other singularities either generated by
the manufacturing process, or originated during the operational life of the surface.
The characterisation of individual surface features opens up a wide array of new
application scenarios and creates novel challenges for surface metrology. Early
approaches are not as consolidated as what is available for the characterisation of
surface texture and see the convergence of mathematical models, methods and
algorithmic solutions coming from heterogeneous disciplines such as image pro-
cessing, computer vision, coordinate metrology, reverse engineering, and computer-
aided design. In this chapter an overview of the tools available in current surface
metrology software is provided first. Then, the main challenges and open issues of
achieving full metrological characterisation of individual surface features are intro-
duced and discussed, as well as the current research approaches addressing them.
8.1 Surface Metrology and the Characterisation

of Individual Areal Features
Surface metrology is conventionally concerned with the characterisation of surface

finish. Once form has been removed from an areal topography dataset, the
remaining information is surface texture, which is described over the entire
N. Senin (&)
University of Perugia, Via G. Duranti 67, 06125 Perugia (PG), Italy
e-mail: nicola.senin@gmail.com
L. Blunt
University of Huddersfield, Huddersfield, UK

180 N. Senin and L. Blunt
Fig. 8.1 Example typical areal features which need to be characterised individually; a geometric
elements of a MEMS (courtesy of D. Gohil, National Physical Laboratory), b groove from scratch
testing (courtesy of G. Campatelli, University of Firenze)
scale-limited surface (see Chap. 2). For this reason, the recent ISO 25178 part 2
(2012) standard stresses the concept of field parameters, i.e. parameters that define
surface texture over the entire scale-limited surface (Chap. 2, Blateyron 2006;
Jiang et al. 2007). ISO 25178 part 2 also refers to feature parameters, which are
computed by aggregating attributes (area, perimeter, depth, etc.) originally
extracted from individual topographic features (Maxwellian hills and dales) (Chap.
3, Scott 2004). However, feature parameters are still representative of topographic
properties which pertain to the entire surface.
In spite of the conventional approach, there is a strong and growing interest in
the characterisation of individual areal surface features considered as standalone
entities, i.e. without the need to revert the results back into an indication of the
overall surface state. For the purposes of this chapter, an individual areal feature
can be simply considered as a portion/region of the scale-limited surface, whose
topography is of interest in some application.
The example MEMS illustrated in Fig. 8.1a is a typical micro-fabricated product
that can undergo dimensional/geometric verification through the measurement of
its surfaces. A surface metrology instrument can be used to acquire its areal surface
topography. However, a simple characterisation of overall surface texture would
not solve the verification problem. The groove shown in Fig. 8.1b is a scratch mark
generated by a scratch test. As with the MEMS example, a surface metrology
instrument can be used to acquire the topography where the groove is located, but
field parameters may not provide meaningful insight on groove geometry.
The surfaces shown in Fig. 8.1 are of different type. The MEMS is a structured
surface, whose topography has been explicitly manufactured to match a design
specification. The scratch mark is the result of a material testing procedure; its
geometry does not match a design, but is still important for material character-
isation. Albeit different, both scenarios suggest a similar inspection procedure
based on extracting the individual feature and computing its relevant dimensions/
shape attributes.
8 Characterisation of Individual Areal Features 181
Similar needs for characterising individual features apply to many structured

surfaces, where multiple instances of an individual surface feature are often spread
into patterns by design, in order to achieve a specific function. In more conven-
tional surfaces, fractures, pores, bumps, pile-up material and other localised sin-
gularities generated either by functional interaction, or as a by-product of
manufacturing processes, are additional examples of individual surface features
which may need to be characterised.
Contact and optical areal surface topography measuring instruments are not the
only means available for acquiring information about individual surface features at
micro- and sub-micro-scales. Scanning electron microscopy (SEM) imaging is ideal
for gathering qualitative knowledge about shape (the examples shown in Fig. 8.1 are
SEM images). However, current SEM technology lacks the capability of returning
reliable, quantitative dimensional information. Moreover, SEM images are intrin-
sically two-dimensional and only recently attempts have been made to give me-
trological significance to stereo-SEM approaches, which can be used to reconstruct
3D information (Bariani et al. 2005). The other notable alternative to conventional
measurement techniques is the micro-CMM (see Leach 2011 for an overview) which
has the theoretical advantage of allowing the application of verification procedures
developed for coordinate metrology of standard-sized parts. However, micro-CMMs
are still at their infancy, require skilled operators, and—most importantly—are
intrinsically slow given that points are acquired sequentially.
8.2 Current Practice and State of the Art
Although areal surface topography measuring instruments have been around for a
few decades, only recently has a strong need surfaced for their full metrological
characterisation. Even calibration, originally aimed at making sure the instruments
would return consistent roughness parameters, is now more aggressively focusing
on the problem of returning more accurate surface topography information as a
whole in order to favour dimensional metrology applications (Leach et al. 2012).
Despite the immaturity of calibration, many manufacturing companies operat-
ing in the semiconductor, precision optics, microfluidics, micromachining, and
other precision industries, are already using areal surface topography data to obtain
quantitative insight on the dimensions/shape attributes of individual areal features.
8.2.1 Commercial Software
A simple procedure often applied with the tools available in current commercial
surface metrology software is shown in Fig. 8.2. The procedure consists in mea-
suring the distance between markers manually placed onto pixels. This is typically
done either using a top-view or on cross-section images.
Fig. 8.2 Manual placement of markers and computation of Euclidean distances between pixels
in current commercial surface metrology software; a distance on top-view; b distance on cross-
section profile (part of a MEMS device)
The procedure illustrated in Fig. 8.2 is fairly straightforward; however, it is

clearly not robust from a metrological standpoint. The selected pixels may not be
the best representatives of the edges; they may be misaligned with respect to the
intended measurement direction; or the reference plane selected for the mea-
surement procedure may be misplaced. Moreover, while sharp edges make
boundary visual identification simpler (for example, step-like geometries), gradual
transitions may make it more difficult (for example, the groove shown in
Fig. 8.1b). Ultimately, the manual process is not traceable, repeatable and/or
reproducible, and the uncertainty associated with the results is difficult to quantify.
Some commercial surface metrology software products are starting to include
algorithmic functionality to trace contours (on planes parallel to the xy plane) or
profiles (on planes passing through the z axis). Some packages also provide tools to
fit simple 2D geometric elements such as lines, circles, etc. to contours/profiles.
This allows a more robust computation of linear dimensions, distances and angles
(for example, see Fig. 8.3). Some software packages are also beginning to provide
functions compliant with ISO specification standards for profile assessment, for
Fig. 8.3 Computer-assisted evaluation of linear distances, dimensions and angles in current
commercial surface metrology software; a original topography; b extraction of contours on the xy
plane, fitting to 2D geometry (lines) and computation of linear distances; c extraction of z profile
and computation of step height according to the ISO 5436 part 1 (2000); d extraction of z profile,
fitting to 2D geometry (lines) and computation of a draft angle
example ISO 12181 part 2 (2011) for roundness assessment and ISO 5436 part 1
(2000) for step height evaluation.
Despite the availability of tracing and fitting algorithms within commercial
software, there are still significant issues with accuracy, reproducibility and
traceability; mostly due to the influence of human judgement in the process. This
principally includes the choices made in the initial filtering and levelling of the
surface topography, in the extraction of z profiles and xy contours, and in the
identification of the contour/profile point subsets to be used for each fitting.
Finally, since only contours and profiles are targeted, only 2D characterisation
is generally possible. For more complex types of geometric analysis, areal
topography data is usually exported to coordinate metrology/reverse engineering
software.
8.2.2 Literature
While the measurement of individual features has been widely explored in

CMM-based metrology (Hocken and Pereira 2012), there has been little research
into the characterisation of individual features in surface metrology. The authors
have developed analysis procedures targeting specific individual features in
selected application domains (Senin et al. 2012a), and have later proposed a more
generalised analysis procedure—albeit limited to step-like features—in line with
ISO 17450 part 1 (2011), the specification standard devoted to geometric verifi-
cation (Senin et al. 2012b). Additional insight for developing novel characterisa-
tion methods can be gained once the relationship between areal surface topography
data and conventional 2D intensity images is fully acknowledged. In fact, many
algorithmic procedures can be developed using previous findings in image pro-
cessing, computer vision and photogrammetry, at least as starting points.
8.3 Definitions and Data Structures for Areal Features
ISO 25178 part 2 (2012) defines point, line and areal features, but only in the
context of an areal surface topography imagined as a collection of Maxwellian
hills and dales (see Chap. 3). From this perspective, point features are peak, pit and
saddle points; line features are course and ridge lines; and areal features are the
hills and dales themselves (for further details, see Chap. 3 and also Scott 2004).
However, in this chapter the concept of areal feature is taken into a broader
context: an areal feature is any region of interest located within the areal surface
topography, therefore, not necessarily a hill or dale.
Both the definition in ISO 25178 part 2, and the one introduced in this chapter,
apply to continuous geometry, while the output of a surface topography measuring
instrument is intrinsically discrete. It is, therefore, necessary to first illustrate an
areal feature in terms of raw data. The following sections make use of an example
L-shaped feature with vertical side walls (a step-like areal feature, see Fig. 8.4)
and assume an optical instrument has been used for the measurement. However,
the following considerations are applicable to any other feature shape measured
with any surface topography measuring instrument, so long as the instrument
returns data on a uniform grid, as illustrated in the following section.
8.3.1 The Acquisition Process as Gridded Sampling

of Height Information
Areal surface topography measuring instruments usually sample height informa-

tion at specified (x,y) positions along the rows and columns of a regular grid. In
this chapter this mode of acquisition is referred to as gridded sampling of height
information. In Fig. 8.4a, the typical coverage obtained via gridded sampling is
shown for an example step-like, L-shaped feature (for better clarity, grid quadrant
sizes are exaggerated). For an optical instrument, the gridded layout on the
measurand surface replicates the organisation of the pixels on the CCD sensor; the
actual surface area covered by each grid quadrant depends on both the CCD
resolution and the optical magnification. The output of each CCD pixel is a single
height value, somewhat representing the entire surface region enclosed within the
Fig. 8.4 Data structures for individual areal features; a characteristic pattern of gridded
sampling over a step-like surface feature; b acquisition result as a range image; c as an intensity
image; d as a point cloud; e as a triangulated surface
corresponding quadrant (for more details of how height is computed in optical

instruments refer to Leach 2011).
The gridded sampling model applies to all optical instruments; different probing
technologies (for example, focus variation, coherence scanning interferometry,
etc.) determine how the pixel height value is reconstructed from the measured
data. An analogous model also applies to stylus instruments, as long as they
operate in a raster scanning mode; the value encoded within each quadrant results
from the interaction of the stylus tip and the surface in that specific region.
8.3.2 The Acquired Dataset as a Range/Intensity Image
Under the assumption of constant quadrant width/spacing in the x and y axes, the
position of each quadrant of the sampling grid can be encoded in terms of (row,
col) indices of a matrix. The entire result of the acquisition process can be handled
as a digital image, i.e. a matrix of scalar values, where each value represents height
information (z). This type of image is commonly referred to as a range image
(Fig. 8.4b), since z is a distance to a reference zero set by the probe. The dataset
can also be equivalently handled as a conventional 2D intensity image (Fig. 8.4c),
where each z value can be seen as a simple scalar value (i.e. an intensity value),
with no geometric interpretation attached.
The greatest advantage of being able to handle areal data as a conventional 2D
range/intensity image is the possibility of applying algorithms from the field of
digital image processing. The use of the range/intensity image has proved vital for
most procedures developed for filtering, manipulation, and many other types of
data processing in surface metrology, and is of paramount importance for many
procedures described in this chapter. The use of the range/intensity image is also
the reason why areal surface topography measuring instruments are commonly
referred to as imaging devices. The representation as an intensity image is also the
main representation adopted by surface metrology software.
In general, any individual areal feature can be seen as occupying a region of the
image, i.e. a specific subset of pixels; therefore, in terms of raw data, an areal
feature is a subset of pixels of the range/intensity image.
8.3.3 The Acquired Dataset as a Point Cloud
Sometimes it is useful to convert the image into a point cloud (Fig. 8.4d). This is
done by turning the (row, col) indices of each pixel of the range/intensity image
into (x,y) coordinates (using pixel spacing information) and by using pixel values
as the z coordinates. The resulting point cloud is often referred to as an unstruc-
tured dataset, since topological information (i.e. point neighbouring information)
is generally not retained in the conversion process. An areal feature, originally a
subset of pixels, becomes a subset of points within the cloud. The point cloud
representation is currently not used much in conventional surface metrology.
However, it will be shown later in this chapter that the use of a point cloud is a
necessary precondition to fitting, one of the key steps for the characterisation of
individual features. The point cloud representation is also potentially useful
because it represents the starting point for many characterisation algorithms which
have been originally developed for CMM metrology and reverse engineering (see
Hocken and Periera 2012), which could be used for surface metrology.
8.3.4 The Acquired Dataset as a Tessellated Surface
A tessellated (usually triangulated) surface (Fig. 8.4e) can be constructed from any
subset of the point cloud by reconstructing the local topological information (i.e.
the point neighbouring information). This reconstruction process can be carried out
by projecting the points onto the xy plane and then applying a 2D Delaunay
triangulation algorithm; the connectivity defined by the triangle edges essentially
replicates the point neighbouring information (De Berg et al. 2008). However, it is
often easier to reconstruct the triangles directly from contiguous (row, col) indices
of the range image, which results in a triangulation also consistent with the
Delaunay properties. Triangulated representations are useful because they are the
simplest form of continuous geometry that can be reconstructed from the dataset.
However, the triangulation may introduce errors or false expectations about the
underlying topography, since continuity is achieved by interpolation, and inter-
polation introduces information about spatial frequencies which were not mea-
sured. For this reason, tessellated surfaces are mostly used for visualisation
purposes, again at risk of being visually misleading (for example, compare
Fig. 8.4a with Fig. 8.4e). In the characterisation of individual features, tessellated
representations are nevertheless useful for reconstructing the geometric deviations
of a feature with respect to its nominal counterpart, as discussed in Sect. 8.6.6.
When tessellated representations are adopted for individual features, the
applicable triangulation methods are the same as described above for entire sur-
faces; additional care must be taken when creating triangles for the feature
boundary points, since the boundary will typically be non-rectangular and may
lead to degenerate facets.
8.4 Issues in Imaging of Areal Features
Most of the major issues related to acquiring areal features using imaging devices
are intrinsically evident in the example feature shown in Fig. 8.4. A few issues
must be discussed in more detail as being more relevant to dimensional metrology.
Some issues are scale-independent, and strictly related to the intrinsic properties of
acquisition using imaging devices, while others are more specific to imaging at the
micrometre and sub-micrometre scales.
8.4.1 Gridded Sampling
Gridded sampling implies uniform point (x, y) density. This may be ideal for
computing field parameters, because it fixes the spatial frequencies under obser-
vation, but may be sub-optimal for the dimensional and geometric characterisation
of individual features. Uniform density makes it impossible to accumulate sample
points in regions where they may be needed the most (for example, regions with
high local curvature), which on the contrary is common practice in CMM
metrology. Furthermore, low-density sampling grids, applied to surface features
with particularly small footprints, will introduce significant aliasing effects (for
example, see Fig. 8.5). In Fig. 8.5, it is important to distinguish between the
detectability of the features, which obeys the Nyquist sampling criterion, and the
actual measurement of their dimensions, which requires higher x and y axis res-
olutions to localise the feature edges.
Fig. 8.5 Aliasing effects on step-like features (simulation); a low resolution: the two features are
not entirely distinguishable from each other (poor detectability of each individual feature);
b intermediate-resolution: the features are distinguishable (improved detectability) but the
contours of each are not clearly defined (poor measurability); b higher resolution: the features are
well distinguished (good detectability) and the contours can be identified with increased accuracy
(improved measurability)
Lateral resolution is a major issue for optical surface topography measuring

instruments. While increasing the instrument detector resolution is generally fairly
straightforward and reasonably affordable, the real limitation is in the minimum
resolvable distance provided by the optics, which is linked to wavelength and
numerical aperture (Leach 2011). Lateral resolution is particularly poor for optical
instruments if compared to the actual lateral resolutions that some micro-
manufacturing processes are currently capable of achieving; which makes aliasing
a significant problem.
8.4.2 Unidirectional Observation and Maximum

Detectable Slope
Since imaging takes place along the direction of observation (z axis), vertical
surfaces, undercuts and regions shadowed by other regions will not be measured
(for example, see Fig. 8.6a). Several micro-fabrication processes are capable of
producing complex, fully 3D geometries, which cannot be adequately captured by
unidirectional observation. Areal surface topography measuring instruments have
also a limitation on the maximum local slope of the surfaces they can measure
(Leach 2011). As the simulated hemispherical protrusion in Fig. 8.6b clearly
shows, limitations on maximum detectable slope may lead to problems in
reconstructing the exact boundaries of a feature.
One way to overcome instrument slope limitations is to use stitching algorithms
to combine images taken from multiple viewpoints; however, stitching is still a
subject of considerable research as it introduces additional errors, which are
Fig. 8.6 Unidirectional observation and maximum detectable slope reduce coverage of
individual features (simulated gridded sampling on simplified geometries); a vertical walls,
undercuts and shadowed regions on step-like geometry; b high-sloped regions on hemispheric
protrusion
detrimental to metrological characterisation (Wyant and Schmit 1998; Marinello

et al. 2007). A detailed illustration of the issues involved in stitching is beyond the
scope of this chapter and only single images will be assumed available for the
characterisation process.
8.4.3 Limitations on Aspect Ratio and Assessment

of Metrological Performance
Since conventional surface metrology has always been primarily focused on

assessing surface finish, most instruments have been designed to provide high
accuracy in determining the smallest height changes while covering the widest
possible areas in the xy plane. This has resulted in instruments featuring a very
high vertical z resolution over a limited range, and a much lower lateral resolution
with wider lateral range. This dynamic range may be contrasted with areal features
characterised by much higher aspect ratios; for example MEMS manufactured
with LIGA or deep reactive ion etching processes.
In recent years there has been open debate on how to assess the metrological
performance of areal surface topography measuring instruments, and in particular
on how to model the measurement uncertainty for these devices (Giusca et al.
2012a, b). These metrological considerations also apply to instruments for con-
ventional imaging, for example in photogrammetry and computer vision appli-
cations (Yi et al. 1994; Zhou and Gu 1997; Weißensee et al. 2010). For areal
surface topography measuring instruments, the metrological issues are even more
difficult because pixel values have higher metrological significance. In computer
vision, grey levels are often only used as a contrast mechanism to detect edges; in
surface metrology pixels contain actual z measurements, which are used in the
dimensional/geometric characterisation of the measurand surface.
(a) (b) (c)
Fig. 8.7 Association of pixels to feature surfaces; a typical imaging scenario: the acquisition
area covers the feature (pyramid) and part of the background (flat surface), no explicit association
exists between each pixel and the actual measurand surface it was sampled from; b result of
feature identification: the dataset is partitioned so that the points associated to the background and
to the feature are well discriminated; c a further partitioning associates feature points to specific
surfaces of the pyramid
8.4.4 Association of Pixels/Points to Feature Surfaces
In CMM measurement it is possible to move the probe so that it acquires points

that belong to specific parts of the workpiece surface. In imaging, the whole
exposed workpiece area is sampled at the same time with no possibility to dis-
criminate the pixels based on what part of the surface they are referring to. The
areal topography data in its native form does not contain any information allowing
each pixel/point to be traced back to the specific measured region it was sampled
from. While this is generally not a problem in the assessment of surface finish
using field parameters, it becomes a primary issue for dimensional metrology of
individual features. Pixels/points belonging to the measurand feature must be
differentiated from those belonging to the background. Pixels belonging to the
feature may need to be further differentiated between themselves if the feature is
characterised by multiple surfaces (for example, see Fig. 8.7).
The association of pixels/points to specific regions of the measurand is probably
the most important challenge which must be faced in the characterisation of
individual features, and is the primary objective of feature identification; one of the
key steps of the entire feature characterisation procedure, as illustrated in detail in
Sect. 8.6.2.
8.5 Characterisation Scenarios and Approaches
A general procedure can be envisioned for the characterisation of individual

features, consisting of the following main steps:
• identify the areal feature within the areal surface topography;

• extract the feature and evaluate its relevant geometric attributes; and
• take the necessary actions based on the characterisation requirements.
8.5.1 Feature Inspection and Feature Verification
In this chapter, the term feature inspection is used to refer to the general procedure
of identifying and extracting an areal feature, and assessing its relevant geometric
attributes. The term feature verification is used consistently with ISO 17450 part 1
(2011) to refer to an extension of the inspection process where a nominal model is
assumed available (for example, a CAD model) and feature attribute values are
checked for conformance to specifications (for example, dimensional and geo-
metric tolerances).
In the characterisation of individual areal features defined at micrometre and
sub-micrometre scales, the difference between inspection and verification is often
quite subtle. As a matter of fact, even in those scenarios where a nominal model is
not available, reference values for comparing feature attribute values are often
assumed. For example, in the assessment of wear/damage of a surface, the depths/
widths of individual cracks/pits may be compared against reference values to
assess the gravity of the situation. Furthermore, in the realm of micro-fabrication,
it is still quite common for manufacturers not to use a CAD model of the surface
features they produce, since the control on the manufacturing process is generally
not as good as required to fully replicate it. Even less common is the use of
tolerances at small scales, since they often end up being the same order of mag-
nitude as the nominal dimensions (Hsu 2008), which also blurs the line demar-
cating inspection from verification.
8.5.2 Individual Features and Feature Patterns
For many structured surfaces, function is not achieved by individual surface fea-
tures, but by patterns of a repeated feature. In this case, the inspection/verification
processes are generally aimed at identifying and extracting the individual features,
computing their relevant geometric attributes, and then aggregating the results into
synthetic descriptors that summarise the texture properties of the entire surface
(see Fig. 8.8).
For feature patterns, the procedures illustrated in this chapter follow a con-
ceptual approach similar to feature parameters: partition the surface into features,
compute feature attributes and aggregate feature attributes into descriptors of the
whole surface topography. The only difference is the added freedom of computing
feature attributes that do not necessarily come from a topography representation
exclusively formed from hills and dales.
(a) (b) (c)
Fig. 8.8 Feature pattern analysis through aggregation of individual feature attributes; a structured
surface with regular dimple layout; b identification and extraction of individual dimples (areal
features) and evaluation of their diameter (feature attribute); d aggregation into synthetic
descriptor (for example, dimple mean diameter)
8.5.3 Pathways to Feature Characterisation: Coordinate

Decoupling and 2.5D/3D Characterisation
As illustrated in Sect. 8.4.2, the main limitation of unidirectional observation is

that the full 3D nature of a feature’s geometry cannot be captured. Instead, the
resulting range image contains 2.5D geometry, which favours some feature sur-
faces over others depending on orientation with respect to the direction of
observation. This limitation forces the inspection/verification process to exclu-
sively focus on the surfaces which are most visible in the areal topography data,
and on their related attributes.
The 2.5D nature of the areal topography data has also promoted the adoption of
a simplified approach to verification, here referred to as coordinate decoupling.
The underlying assumption of coordinate decoupling is that, if compatible with the
application requirements, an areal feature can be satisfactorily characterised by
two independent sets of specifications defined in separate 2D planes. For example,
a step-like feature belonging to a MEMS (Fig. 8.9a) can be characterised by a
single measure of step height, computed in a cross-sectioning plane containing the
z axis (Fig. 8.9b), and in terms of its xy footprint (i.e. boundary contour—
Fig. 8.9c). The advantage of coordinate decoupling is that the disjointed charac-
terisation problems can usually be approached with simplified 2D algorithms. For
example, the identification of the feature xy contour can be turned into a 2D edge
detection problem, followed by a 2D fitting (lines, circles, etc.). Similarly, step
height computation may be implemented by applying the ISO 5436 part 1 (2000)
procedure on the cross-sectional profile. Most surface metrology commercial
software products now support this type of disjointed analysis.
In a semi-decoupled approach, a feature is still described in parallel, dimen-
sionally simplified spaces, but the analyses are dependent on each other. For
example, the same step-like feature (Fig. 8.9a) could be characterised in terms of
(b) (f)
(a) (e)
(c) (g)
(h)
(d)
(i)
Fig. 8.9 Approaches to the characterisation of areal surface features; a example step-like
feature; b feature z profile analysis (for example, step-height evaluation); c feature xy contour
analysis; d flatness assessment of the feature top surface; e example hemispheric protrusion;
f feature z profile analysis; g feature xy contour analysis; h analysis of the topography of the
hemispheric surface; i full 3D characterisation: 3D sphere fitting and 3D plane fitting; feature
boundary retrieved from the intersection of the fitted geometries
its 2D contour on the xy plane (Fig. 8.9c), and in terms of the flatness of its top
surface (Fig. 8.9d). The assessment of flatness requires knowledge of the feature xy
boundary in order to extract the correct subset of points that are then fitted to the
reference planes. Analogous considerations apply to other areal features—the
hemispheric protrusion shown in Fig. 8.9e can be characterised by a fully
decoupled approach (see Fig. 8.9f, g), and/or by a semi-decoupled approach (see
Fig. 8.9g, h). In addition, some applications may require a full 3D approach
(Fig. 8.9i), where the example hemispheric protrusion is fitted to a sphere and the
background is fitted to a plane (3D fitting of 3D nominal geometries), and the
feature boundary is extracted from the intersection of the sphere with the plane
itself.
The choice of a decoupled, semi-decoupled or full 3D approach is generally
application dependent, and essentially related to the specifications associated with
the feature. The existence of multiple pathways to the characterisation of indi-
vidual features, in particular balancing computational complexity (by operating on
reduced dimensionality) and completeness of the characterisation results (obvi-
ously reduced when operating in coordinate decoupling modes), is a fundamental
premise to the characterisation procedures which will be illustrated in the
following sections.
8.6 Steps of an Inspection Process Targeting Individual

Features
The fundamental steps that must be implemented by an inspection process aimed

at the characterisation of individual areal features are discussed in the following
sections. The result of inspection is a set of dimensions and/or shape attributes.
A verification process would be comprised of the same steps, with an added final
checking for conformance to specifications.
It is assumed as a premise that the entire inspection process relies on a single
acquired image containing the target feature, which means that issues related to
incomplete feature coverage and stitching are not considered.
8.6.1 Areal Topography Data Pre-processing
Albeit in this chapter it is assumed that the measurement instrument axes have
been properly calibrated and any aberration has been corrected, a few specific
types of error remain that must be typically dealt with. For optical instruments this
refers, in particular, to missing measurement values and measurement artefacts.
8.6.1.1 Handling of Missing Measurement Values
Missing measurement values (here also referred to as voids) are generated when
the probe does not receive sufficient information to compute height for a specific
pixel (Fig. 8.10).
Voids are easily represented within point cloud data structures, where no points
are generated for missing z coordinates. Similarly, in triangulated geometries,
triangles are not generated for missing vertices. However, voids are not as easily
handled in range/intensity images, where a validity mask is required, i.e. a binary
image maintained in parallel to the original image to keep track of invalid pixels.
Masked images (range/intensity images and associated validity masks) are dif-
ficult to handle because most image processing algorithms (for example, convo-
lution, fast Fourier transform, wavelet decomposition, etc.) are not easily adapted to
deal with such data. Even common surface metrology tasks, such as field parameter
computation, require significant rethinking in the presence of masks. In order to
avoid such problems, most surface metrology software does not support masking,
and adopts padding, i.e. replacement of voids with a reference value (for example,
simply a zero, or some value obtained by extrapolation of valid neighbours).
Evidently, padding introduces a source of error. Such error is usually small when
the analysis is carried out over the entire image, because the number of voids is
small compared to the image size. This is why padding is generally well accepted
when computing field parameters. On the contrary, voids (and padding error) are
Fig. 8.10 Detail of areal

topography data featuring
multiple clustered voids. The
dataset comes from a real
measurement with a
coherence scanning
interferometer on a MEMS
surface; heights at the
boundaries of the void region
are truncated to reduce the
batwing effect (see Sect.
8.6.1.2) and produce a better
visualisation
likely to have more relevance in the analysis of individual features. This is because
the feature is generally smaller than the entire image, and because voids may be
placed in critical regions of the feature itself. Voids may make it harder to deter-
mine a feature boundary, and/or compute a proper geometric fitting. Analogously,
padding error may alter a feature boundary, and/or lead to significant fitting error.
Smart padding strategies are, therefore, often essential to make sure the char-
acterisation process of an individual feature can be accomplished. For this purpose,
the authors have investigated void replacement by linear and non-linear interpo-
lation of valid neighbours (Senin et al. 2012a, b). For step-like geometries, the
most promising solution appears to be interpolation via a median operator com-
puted on a m 9 n moving window centred on the void pixel to be padded (window
size being application dependent). The median operator is known from image
processing literature to better preserve sharp transitions, such as step-like features
(Gonzalez and Woods 2008). The procedure must be repeated multiple times in
order to handle multiple clustered voids.
However, the effects of different padding strategies on the dimensional char-
acterisation of individual features are still largely unexplored. More advanced
smart padding strategies may imply some knowledge of the underlying topography
and the adoption of statistical modelling techniques in order to reproduce the
missing regions more accurately.
8.6.1.2 Handling of Measurement Artefacts
Not to be confused with physical artefacts used for calibration, measurement

artefacts are false features appearing on the measured areal topography as a
consequence of probe-surface interactions. Batwing artefacts (Gao et al. 2008) are
typical of optical instruments such as coherence scanning interferometers and
usually appear in correspondence to sharp topographic transitions on the surface
Fig. 8.11 Detail of areal

topography data showing
batwing artefacts. The dataset
is from a real measurement
obtained with a coherence
scanning interferometer on a
MEMS surface; positive and
negative batwings can be
observed consistently
spreading along the top and
bottom edges of the step
(Fig. 8.11). Different artefacts characterise other optical instruments. Unlike voids,
measurement artefacts must be identified before they can be treated.
Measurement artefacts, such as batwings, are a significant threat to the char-
acterisation of individual areal features because they may alter the shape and/or
hide the true boundaries of the feature. The presence of batwing artefacts may also
significantly alter the outcome of a levelling operation (more on this in Sect.
8.6.1.3).
A comprehensive solution for measurement artefact identification on areal
surface topography data would include mathematical modelling of the measure-
ment process so that the expected topography of the artefacts can be accurately
predicted (see, for example Mandal et al. 2012). However, in this chapter a simpler
technique, previously developed by the authors, is illustrated based on post-pro-
cessing the resulting areal surface topography data (Senin et al. 2012a, b). The
procedure replicates a common solution for the identification of sharp noise
artefacts in conventional digital images (Gonzalez and Woods 2008). Batwings are
recognised as outliers in the residual image obtained by subtracting a smoothed
version of the image from the original. Smoothing is achieved by filtering, for
example, median on a m 9 n window. Once identified, artefacts can either be
turned into null values (voiding), or replaced by interpolation of valid neighbours
(padding) depending on characterisation needs. Notice that for step-like features,
voiding multiple artefacts located on the feature boundary may lead to the
impossibility of determining the exact location of the boundary itself, which is
why padding is often preferred, at least for analyses on the xy plane, when
coordinate decoupling is possible.
8.6.1.3 Areal Topography Levelling
Levelling by removal of least-squares mean reference polynomial surfaces is a

widely accepted procedure in surface metrology for removing form and specimen
(a) (b)
Fig. 8.12 Levelling by subtraction against levelling by rotation on example cross-section profile;
a levelling by subtraction: dimensions and shapes are altered but uniform xy point spacing is
retained (the result is still an image); b levelling by rotation: dimensions and shapes are retained
but uniform point spacing is lost (the result is not an image anymore)
placement error (see Chap. 5). For nominally planar surfaces, the least-squares
mean (LSM) plane is typically used for removing the influence of specimen tilt
error from the topography height values, ahead of computing field parameters.
However, when dealing with the characterisation of individual features, two main
issues should be considered, as discussed in the following.
Levelling involves only the z coordinates

The least-squares formulation generally adopted in surface metrology considers
only the z coordinates (i.e. z least-squares as opposed to total least-squares).
Moreover, the mean plane is removed by simple z subtraction (referred to as
levelling by subtraction—Fig. 8.12a), as opposed to finding the rigid rotation
transform that registers the mean plane to the xy plane (referred to as levelling by
rotation—Fig. 8.12b). Levelling by subtraction is a non-rigid transform since it
introduces distortion (Fig. 8.12a). Distortion is generally negligible when com-
puting field parameters, but may become relevant when doing dimensional veri-
fication of individual features. Previous findings by the authors, albeit on a limited
set of test cases consisting of step-like geometries with comparatively low aspect-
ratios, have found such distortions to be at least one order of magnitude smaller
than the lateral pixel spacing in the original dataset (Senin et al. 2012b). However,
this may not be the case in application scenarios involving many MEMS fabricated
with higher aspect-ratios. On the other hand, levelling by rotation preserves
dimensions and shapes (Fig. 8.12b), but disrupts uniform point xy spacing, thus the
levelled result cannot be handled as an image unless resampling is applied.
Resampling is itself potentially dangerous because it is generally based on inter-
polation, and thus introduces additional sources of error (see Chap. 5).
Fig. 8.13 Global and selective levelling of a MEMS, the LSM plane is shown in semi-
transparent colour, levelling is implemented by subtraction; a global levelling: the LSM plane is
affected by the tall protrusions on the right and levelling leads to a tilted topography; b feature-
based levelling: the LSM plane is only affected by the background region, as a consequence
levelling results in a topography aligned to the background
The suggested procedure for handling the levelling problem (Senin et al. 2012a)
consists of initially adopting levelling by subtraction in order to retain the image
data structure as long as possible, and then compensating for distortions at the final
stage by reconstructing the correct position of the points as if they were processed
by rotation. This process should to be carried out before the fitting stage, as
discussed in Sect. 8.6.5.
Levelling makes use of all the available data points

In general all the available data points are used to compute the LSM plane
(referred to as global levelling). While again this is usually acceptable for com-
puting field parameters on surfaces with approximately uniform surface topogra-
phy, it may not always be advisable when targeting individual areal features, or for
topographies with large, asymmetrically located, protruding or recessed forma-
tions (Fig. 8.13a). While several surface metrology software products now offer
solutions based on manual selection of specific points to identify custom reference
planes to level against (for example, three-point levelling), a more robust and
repeatable analysis procedure should make use of algorithmic procedures to
automatically identify and extract regions that are better suited to form reference
planes (i.e. selective levelling). For analysis processes targeting individual fea-
tures, it is often useful to identify and use the background regions (i.e. regions
surrounding the features of interest) as references to build the LSM plane (Senin
et al. 2012a) (for example, Fig. 8.13b). Notice that the background regions are
simply another type of areal feature, therefore, the same algorithms described in
Sect. 8.6.2 for the identification of the features of interest can be applied at this
stage for identifying the background. This is why, in the remaining parts of this
chapter, selective levelling is often referred to as feature-based levelling.
Notice that the differences introduced by global compared to selective levelling
are generally relevant only to those scenarios where a significant part of the
characterisation process is carried out through image processing techniques
(including levelling by subtraction). If a full 3D characterisation approach is
pursued exclusively based on point clouds or tessellated geometries, the angular
position of the dataset is not relevant.
8.6.2 Feature Identification and Extraction
As illustrated in Sect. 8.4.4, one of the most peculiar aspects of areal feature
imaging is the inability to obtain information linking each measured point/pixel to
the surface it refers to. Consequently, there is no initial way of knowing whether
each pixel/point either belongs to the background or to any of the relevant surfaces
of the feature under investigation (refer to Fig. 8.7). There are many ways to tackle
this problem, and countless examples can be found both in the literature of image
processing/computer vision, when considering images (see for example Gonzalez
and Woods 2008) and in the literature dedicated to geometric reconstruction/
reverse engineering, when considering point clouds (see for example, Bi and Wang
2010). In this chapter, only the methods operating on images are discussed,
classified as either based on segmentation or on template matching.
8.6.2.1 Segmentation and Heuristic Region Selection
Segmentation is a term coming from the image processing literature. It refers to

partitioning an image into regions/segments so that all pixels within a region are
characterised by similar properties.
A feature identification procedure based on segmentation for areal surface
topography data may consist of the following procedure (Senin et al. 2007, 2012a):
• identify a set of pixel properties that are expected to be similar for all the pixels
of the target feature;
• use such properties to run a segmentation of the areal data; and
• scan through the resulting segments applying predefined heuristics in order to
identify the segment/region (or a combination of them) that actually contain the
feature.
The above general procedure can be implemented in multiple variants, as
described in the following sections. Importantly, while all the illustrated solutions
are adaptations of techniques already popular in digital image processing, only
some are currently applied in surface metrology.
Thresholding on height
Thresholding on height is based on classifying pixels based onto whether their
height is above or below a user-defined threshold value. This is a binary classi-
fication process that will produce multiple regions depending on the topological
connectivity of the pixels either passing or failing the threshold rule.
Figure 8.14 shows a thresholding operation applied to an example topography
containing a groove (target feature). Under the assumption that the topography has
been appropriately levelled, a specific z value can be identified that results in a
partitioning of all the dale-like features (including the groove) from the remaining
topography (the background regions). In Fig. 8.14a, a tentative z threshold value
translates into a truncation plane. The truncation plane gives origin to the binary
classification image of Fig. 8.14b, which contains multiple, topologically dis-
connected, 2D regions (i.e. segments). First and second order moments are com-
puted on each segment to encode their shapes through equivalent ellipses. The
region containing the groove is identified through simple heuristics applied to
ellipse properties; for example, by setting thresholds on ellipse area and elongation
(the groove is a large, elongated region). In Fig. 8.14c, the selected 2D segment is
shown mapped to the original 2.5D topography data, highlighting the region to be
extracted.
Segmentation by thresholding on height is a very common approach and widely
available in commercial surface metrology software. The main criticality is the
identification of appropriate threshold values to single out a specific target feature.
This is generally difficult, and often impossible depending on the surface topog-
raphy. Specific approaches that can be adopted for the identification of threshold
values range from the analysis of the shape of the histogram of heights (similar to
conventional image processing), to the use of specific percentages of suitable field
parameters (for example, Sz, Spk, Sk and Svk). However, as stated above, for some
topographies, none of the above threshold identification techniques may work
appropriately. Even in the example shown in Fig. 8.14, it can be seen that the
Fig. 8.14 Groove identification via thresholding on height; a identification of an appropriate

threshold value and generation of a truncation plane (blue); b analysis of the binary segmentation
image through equivalent ellipses (the image is shown in transparent colour, overlaid over the
original topography); c the segment identified as the groove through thresholds on ellipse
properties (area and aspect ratio) is mapped back onto the original areal surface topography
Fig. 8.15 Multiple thresholding for the identification of step-like features in MEMS; colours
indicate regions identified by different threshold values
groove boundaries have not been captured accurately, which is ultimately down to
a poor selection of threshold value.
Thresholding techniques may work better for step-like features, where the
introduction of multiple threshold values can also help discriminate features with
different heights (Fig. 8.15). However, once again, care must be taken to ensure
the topography is suitable to be partitioned by simply defining threshold height
values. Levelling error may also influence the result, making feature-based lev-
elling (see Sect. 8.6.1.3) an essential preliminary step. Finally, as Fig. 8.15 clearly
shows, additional post-processing is needed to discriminate between features
whose top surfaces are at the same height. As for the example previously reported
in Fig. 8.14, the typical approach consists in identifying topologically discon-
nected regions belonging to each thresholded class.
Morphological segmentation (hill-dale segmentation)

Morphological segmentation is the only segmentation approach currently featured
in ISO 25178 part 2 (2012). As stated earlier, and described in detail in Chap. 3,
morphological segmentation involves the partitioning of a topography into a set of
contiguous Maxwellian hills (regions around a local maximum—peak—such that
all maximal upwards paths end at the peak), and dales (regions around a local
minimum—pit—such that all maximal downwards paths end at the pit). Mor-
phological segmentation is referred to as hill-dale segmentation in this chapter.
When applied to typical areal surface topography data, morphologic segmen-

tation generally results in large numbers of tiny partitions (over-segmentation),
because local roughness tends to give origin to a large amount of peaks/pits.
Combination techniques are then usually adopted to simplify the partitioning by
merging smaller, less significant formations into larger, more relevant conglom-
erates. Significance is determined by applying thresholds to the area, perimeter,
volume, height/depth, etc. of each hill or dale (Scott 2004, Chap. 3).
While further details on morphological segmentation and its role in computing
feature parameters can be found in Chap. 3, this chapter focuses on how mor-
phological segmentation can be used to identify and extract individual areal fea-
tures. The idea is that an appropriate combination rule can lead to an aggregation
of hills or dales that in the end is coincident to the feature of interest (Senin et al.
2012b). This approach works well in some cases. For example, a groove can be
imagined as a large, elongated dale, and thus a combination rule can be sought for
that merges smaller, individual dales (originated from the irregularities in the
internal regions of the groove) into a larger one, representing the groove itself.
Figure 8.16 shows this type of approach. The challenge is the identification of
appropriate combination techniques and threshold values, which currently requires
trial and error. Unfortunately, even when an acceptable combination has been
found, additional post-processing of the segmentation map is often needed to
obtain the target feature. For example, in Fig. 8.16c a merging of the two largest
segments is needed in order to reconstruct the full groove topography. Moreover,
errors at the feature boundaries are still possible due to morphological segmen-
tation strictly following the contours of the original hills or dales (for example, see
the final result in Fig. 8.16d).
For step-like features, and for any other topography where feature edges are
well demarcated, a segmentation approach which works well consists of pre-
processing the original topography data (for example, the MEMS shown in
Fig. 8.17a) with an edge detection filter, for example, by taking the absolute value
of a 3 9 3 Sobel operator (Blunt and Xiao 2011; Senin et al. 2012a). The result is
a virtual, intermediate topography, where planar regions in the original image
appear as dales, and height transitions are turned into crests surrounding the dales
(Fig. 8.17b). A morphological (dale) segmentation applied onto such intermediate
topography results in a partitioning (Fig. 8.17c) that can be mapped back onto the
original topography (Fig. 8.17d), where it correctly discriminates the features from
the background.
The accuracy of this segmentation process is entirely dependent on the edge
detection phase; which is why it tends to work better for step-like, regular surfaces.
Once again, the result of the segmentation must be post-processed in order to
identify what segments/regions actually correspond to the feature(s) of interest. It
has been suggested elsewhere (Senin et al. 2012b) that the segmentation result (2D
image) can be aligned to a template drawing (2D vector geometry) describing the
nominal layout of the features, then each segment can be correctly matched to
either a feature of interest or to the background, depending on its localisation
within the aligned template drawing.
Fig. 8.16 Groove identification via morphologic (dale) segmentation; a initial dale segmenta-
tion; b dale map simplification through the application of a combination rule based on pruning the
change tree (as defined in ISO 25178 part 2; Wolf pruning: 5 % of Sz, area pruning: 0.01 %);
c post-processing of the final segments to obtain the groove region (merging of the two largest
segments, identified via thresholding on area); d identified groove mapped back into the original
topography
Texture-based segmentation
Texture-based segmentation is used in digital image processing. The idea is to
partition a digital image into regions characterised by uniform texture properties.
Albeit the concept of texture in conventional digital imaging refers to the presence
of a visible pattern of some kind, it can be safely translated and generalised for
areal surface topography data, where it matches the definition of texture provided
earlier, i.e. any topographic formation present on the surface once the form has
been removed. Thus, texture-based segmentation is any segmentation process
where areal topography data is partitioned into regions characterised by uniform
texture (in the more general sense).
One of the principal challenges of texture-based segmentation is how to
describe and compute the texture properties associated with each pixel from areal
topography data. One way to do it is to consider as texture properties those which
can be computed from a surrounding region of the pixel itself. Then, any linear or
non-linear function designed to provide a synthetic description of the topographic
Fig. 8.17 Segmentation of step-like features (MEMS) via combined edge detection and dale
segmentation; a original topography (detail); b virtual, intermediate topography obtained through
an edge detection filter; c dale segmentation on virtual topography; d segmentation result mapped
back onto the original topography
properties of a region can be applied onto a moving window that scans through the
pixels of the original topography (Senin et al. 2007). Descriptors of local topog-
raphy include local slope, local curvature or local height. As a matter of fact, any
field parameter could be adapted to work as a synthetic descriptor of region
properties, as long as it is understood that the moving window imposes constraints
on the actual range of spatial wavelengths which will be captured by the
descriptor. In Fig. 8.18, two examples of texture-based segmentation are shown;
both rely on the fact that the machined feature (a groove in Fig. 8.18a and a laser
dimple in Fig. 8.18b) have different bottom roughness than the surroundings,
therefore, the Sq parameter is adopted as the local topography descriptor. The final
segmentation results shown in Fig. 8.18 are obtained by a two-steps process: pixels
with similar descriptor values are first grouped into classes by means of statistical
clustering (specifically k-means clustering, see MacQueen 1967); then, clusters are
further split into subsets (the actual segments) made by topologically connected
pixels (clustering alone cannot guarantee the topological connectedness of the
pixels belonging to the same class).
Fig. 8.18 Examples of texture-based segmentation applied to areal topography data; a groove
segmentation; Sq parameter computed on 19 9 19 window, k-means clustering with k = 4: the
groove is singled out as being smoother than the rest; b laser dimple segmentation; Sq parameter
computed on 9 9 9 window, k-means clustering with k = 4; the region surrounding the dimple is
singled out as being smoother
Alternatives to clustering include region growing and split and merge tech-
niques, also derived from the literature of segmentation for conventional digital
images (Gonzalez and Woods 2008). As with thresholding and morphological
segmentation, texture-based segmentation requires a final step of heuristic region
selection, where segments are either mapped to the areal features of interest or to
the background. Again, approaches such as computing properties on equivalent
ellipses (image moments) are possible.
8.6.2.2 Template Matching Methods
Template matching methods constitute the most notable alternative to segmenta-

tion and heuristic region selection. Template matching methods are based on
creating a template shape representing the feature being searched for, and on
scanning the surface to look for the closest local match to the template. Template
matching methods imply the adoption of a shape coding solution (i.e. a way to
represent a shape/pattern as numbers) and a shape similarity metric (i.e. a way to
obtain a quantitative measure of similarity/difference between shapes). Template
matching methods are very common in conventional imaging and in reverse
engineering problems based on point clouds, but are rarely applied to surface
metrology applications, with a few exceptions (Jiang et al. 2010; Senin et al.
2010).
In Fig. 8.19, an example combination of shape coding and similarity metric is
illustrated applied to the comparison of a template step-like feature with a simu-
lated feature, having alterations in orientation and shape (adapted from Senin et al.
2010). The shape descriptor in this case is not a single number, but a curve
originated from a modification of the ring projection transform originally
(a)
(b)
Fig. 8.19 Template matching applied to an example step-like feature; a computation of the
modified ring projection transform for a 2.5D shape; b application to the comparison of a
measured feature (simulated) and a template feature
conceived for intensity (greyscale) images (Lin and Chen 2008). The value of the
transform curve at r (Fig. 8.19a) corresponds to the integration of the topography
height values located at radial distance r from the feature centroid. The descriptor
is invariant to feature rotation (but not to scale) and robust to small shape varia-
tions. The similarity metric is based on computing a quantitative indicator of curve
similarity (for example, the normalised, cumulative, point-to-point difference
between the two curves), as shown in Fig. 8.19b.
The main challenge of shape matching is robustness to feature variability. For
example, if the areal feature is a groove (for example, a groove from scratch
testing, or a groove due to damage/wear), it is very difficult to identify an
appropriate mathematical shape representation (for the template and associated
similarity metric), capable of capturing the salient traits of any groove geometry,
while at the same time being robust to the variations that may characterise each
individual groove in terms of width, depth, orientation, etc. Even if the feature
under investigation is manufactured by design (for example, part of a MEMS), the
variability due to poor control of the manufacturing process may be a factor to be
taken into account. Template matching for feature identification in surface
metrology still has many unsolved challenges.
8.6.3 Sub-feature Partitioning
Once the target feature has been identified, the characterisation task may require it
to be further partitioned. This is often carried out to extract individual surfaces
from the feature itself.
The same segmentation and template matching techniques, illustrated in Sect.
8.6.2, can be applied for sub-feature partitioning, as long as they are adapted to
operate on masked image data (Senin et al. 2007). This is because the previously
identified individual feature will likely occupy a non-rectangular region of the
original topography and, therefore, it will be associated with a mask. It should be
noted that this is the same problem that feature identification algorithms face if
they are designed to operate in the presence of void data points.
Finally, it should be noted that in some very specific cases it may still be
possible to extract individual feature surfaces without resorting to masking,
because feature and sub-feature partitioning can take place at the same time, over
the whole surface. An example is shown in Fig. 8.20.
A special case of sub-feature partitioning is the extraction of a feature
boundary. This problem is also referred to as identification of a feature xy contour,
since it is a simple 2D problem that typically applies to those scenarios where
characterisation is done in coordinate decoupling mode. Once the region repre-
senting the areal feature has been identified, boundary/contour extraction is typi-
cally solved by applying a common sequence of two morphological operators
defined for conventional binary images (Gonzalez and Woods 2008), i.e. erosion
and subtraction. The operators are applied to the binary image containing the
region, as shown in the example in Fig. 8.21.
In approaches based on coordinate decoupling, the extraction of a feature
boundary is the last critical step in the characterisation of a feature property in the
Fig. 8.20 Example application where feature and sub-feature partitioning are done at the same
time (bike reflector surface—simulated): individual features (pyramids) are obtained by
recombining flat facets (sub-features) obtained directly from partitioning the original topography.
The facets are obtained through a texture segmentation process (colours identify segments): local
topography descriptors: components of the local normal vector; multi-attribute k-means
clustering with k = 7 (three clusters for capturing the three main facet orientations, the
remaining four for capturing the intermediate orientations at the edges)
Fig. 8.21 Example extraction of an areal feature boundary (laser dimple topography); a original
segment/region identified as the target feature; b feature boundary extraction through the
application of morphologic operators to the identified region
xy plane. From the accuracy of such boundary derives the accuracy of the fitting
results (see Sect. 8.6.4), and consequently the accuracy of any measurement of size,
localisation or tolerances concerning the feature. It is clear that boundary extraction
is actually a very simple step, and the real performance of the entire procedure is
mostly dependent on the accuracy of the previous feature identification step.
Often, and especially for step-like features, the feature boundary forms a
complex contour which cannot be captured by fitting a single 2D geometric entity.
In such a case there is the additional problem of further partitioning the boundary
into simpler subsets. Current commercial surface metrology software typically
leaves this problem to the user, who is in charge of selecting the subset of
boundary pixels which must be fitted to a specific line, circle, etc. The automation
of such a process implies an additional segmentation operation carried out on
boundary pixels. For step-like features characterised by piecewise-linear bound-
aries, segmentation based on local edge direction can be applied (for example, see
Fig. 8.22).
8.6.4 Vectorisation and Error Compensation
Feature identification and sub-feature partitioning (for example, boundary extrac-

tion) always end up identifying a set of image pixels. Before such data can be
processed any further (that is, before they can be processed by fitting), they must be
converted into points (i.e. vector geometries) in a R2 or R3 Cartesian space. This is
known as vectorisation (Gonzalez and Woods 2008). For a fully calibrated set of
instrument axes, vectorisation is straightforward, as illustrated previously when
discussing about transforming range images into point clouds (see Sect. 8.3.3). The
(row, col) coordinates of each pixel are turned into (x, y) coordinates by means of
Fig. 8.22 Segmentation of feature boundaries based on local edge orientation; a example step-
like features with straight boundaries; b segmentation result (colours indicate different edge
orientations)
pixel width/spacing information (for example, see Fig. 8.23a, b). For conversions
into the R3 Cartesian space, pixel values are directly converted into z coordinates.
Vectorisation marks the transition from image to unstructured point set,
implying that no image processing algorithms can be applied from this point on.
For this reason, it is recommended that most error compensation algorithms, which
may result in alteration of point xy spacing, are applied directly after vectorisation
(for example, see Fig. 8.23b, c). Since error compensation typically results in point
lateral displacement, an earlier application would either result in abandoning the
image structure sooner, or in the need to resample the point set into a regular grid
in order to recreate a suitable image (and paying the price of adding interpolation
error).
An example error compensation procedure that can be applied at this stage
(Senin et al. 2012a) addresses the aberrations introduced by levelling by sub-
traction described in Sect. 8.6.1.3. The procedure consists of first removing the
effects of levelling by subtraction by adding the z component of the LSM plane
back into the z value of each point (for example, see Fig. 8.24a), and then in
applying the rigid rotation transform to align the plane to the reference z = 0 plane
(Fig. 8.24b).
(a) (b) (c)
Fig. 8.23 Conversion of pixels to points (vectorisation) and error compensation in the xy plane;
a original pixels; b converted points; c application of displacement vectors for error
compensation
(a) (b)
Fig. 8.24 Compensation for the error introduced by levelling by subtraction; a reversal of
levelling by subtraction; b levelling by rotation
8.6.5 Fitting
Fitting refers to the process of adapting nominal geometric entities to the target
feature under scrutiny. Fitting may involve the entire feature (for example, fitting
an entire groove topography to its nominal counterpart, such as a cylindrical
surface) or just parts of it (for example, fitting a portion of a feature boundary to a
straight line, or to a circle). Fitting is generally approached as a minimisation
problem, where the quantity being minimised is representative of the cumulative
difference (error) between the actual geometric entity and the nominal reference.
Such a difference is generally assessed as an aggregation of distance-related
quantities; for example, as a sum of squared distances. The unknowns in a fitting
problem are typically the position and orientation of the nominal reference, but
may include parameters defining the size and shape of the reference itself (for
example, the diameter of a circle or the coefficients of a polynomial curve). In the
latter case, fitting helps not only in determining the deviation with respect to
the nominal, but also helps in gaining additional insight on the size/shape of the
feature itself. In Fig. 8.25a, the boundary of the laser hole (yellow) is fitted to
Fig. 8.25 2D fitting examples; a fitting to a 2D circle to compute the hole diameter; b fitting to
multiple 2D lines to determine the geometry and critical dimensions of step-like features on a
MEMS
a 2D circle of unknown radius with a least-squares minimisation approach; the

best-fit circle (red) provides an indication of the hole localisation and approximate
diameter. In Fig. 8.25b, the boundaries (yellow) of multiple, step-like features of a
MEMS device are fitted to 2D lines, which are then automatically trimmed into
segments (red).
The examples in Fig. 8.25 illustrate least-squares fitting. However, depending
on the application needs, and in particular when tolerances must be determined,
other non-least-squares fittings are possible. Examples include least-sum-of-dis-
tances fitting, two and one-sided minmax fitting, smallest circumscribed and
largest inscribed fitting (Feng and Hopp 1991).
Moreover, the examples illustrated in Fig. 8.25 refer to 2D fitting of 2D geo-
metric entities, and thus are typical of coordinate decoupling approaches. More
complex fitting problems may be faced when dealing with 2.5D or even full 3D
characterisation approaches. For example, in Fig. 8.26a, the top surface of a step-
like feature is fitted to a plane (2.5D fitting involving least-squares minimisation
over the z coordinates only). In Fig. 8.26b a cylinder is fitted directly to the 3D
point cloud resulting from the extraction of the groove region (full 3D fitting
involving total least-squares).
The accuracy of fitting results such as those shown in Figs. 8.25 and 8.26, is firstly
related to the accuracy in identifying the pixels that define the regions to be fitted,
and then to the process of turning such pixels into vector geometry, introducing point
displacement vectors for error compensation. Also, regardless of the segmentation
technique, a few outlier pixels will almost invariably be present, therefore, robust
fitting strategies should be preferentially adopted whenever possible.
Finally, in some applications the nominal model may have a complex shape,
which could be difficult to capture in an analytical form suitable for fitting. In such
circumstances, assuming the nominal model is available as CAD data, it is still
possible to reach an alignment between nominal and measured shape by applying
Fig. 8.26 2.5D and 3D least-squares fitting examples; a top surface of step-like feature fitted to a
plane (2.5D fitting with z least-squares); b cylinder fitted to the groove region (3D fitting with
total least-squares)
3D registration algorithms such as the iterative closest points (ICP) algorithm and
its variants (Rusinkiewicz and Levoy 2001). ICP is popular in coordinate
metrology and reverse engineering, but still seldom applied to surface metrology.
In Fig. 8.27, an example application of an ICP variant (Senin et al. 2010, 2012c) is
shown where a CAD surface model, representing the nominal shape of a step-like
feature of a MEMS, is registered to its measured counterpart. The application of
ICP leads to a final registration solution, which is consistent with a full 3D
minimisation of cumulative error in the least-squares sense, and that can be used as
a starting point to evaluate deviations in feature verification applications.
Fig. 8.27 Registration of a

nominal model (CAD data) of
a MEMS feature to a portion
of measured topography
through the application of an
ICP algorithm
Fig. 8.28 Example approaches to analyse geometric deviations; a selective magnification and
height-based coloring to highlight the deviations of the top surface of a MEMS with respect to the
mean plane; the magnification is applied to a cropped region of the top surface to avoid
amplification of the residual batwing artifacts; b volumetric error between the measured feature
and its nominal counterpart (Fig. 8.27) rendered as voxelised data: blue identifies void regions
(lack of material with respect to the nominal), red identifies regions with excess material
8.6.6 Computing Critical Dimensions and Solving

Verification Problems
Once the fitted geometry has been obtained, computing critical dimensions and
tolerance intervals essentially becomes a task of solving geometric problems in a
R2 or R3 Cartesian space involving the fitted geometric entities. Verification
becomes a matter of comparing computed values against references defined by the
specifications.
At this stage, the scenario and related research challenges are largely scale-
independent, and countless examples can be found in the literature dedicated to
coordinate metrology of standard-sized parts, therefore, there is no need to discuss
them further in this chapter. However, additional peculiar types of analysis are
available for dimensional verification starting from areal data, in particular due to
the 2.5D nature of the areal topography datasets. The combined use of selective
magnification and height-based colouring to investigate the deviations of the
example MEMS top surface is shown in Fig. 8.28a. Deviations are measured along
z with respect to the local LSM plane. In Fig. 8.28b, the volumetric error between
the nominal and measured features registered as previously shown in Fig. 8.27 is
computed at each pixel xy location and rendered as voxelised data.
8.7 Summary and Outlook
An increasing amount of research and industrial scenarios involving the charac-

terisation of engineered surfaces, resort in the use of surface metrology tools to
investigate and verify individual areal surface features. While some of these
scenarios are only concerned with checking the presence/absence of a feature (for
example, defect identification), in other circumstances (the main subject of this
chapter), full dimensional and/or geometrical characterisation of the feature may
be required. This is the case for the dimensional and geometric verification of
MEMS and other elements of structured surfaces, but also for the dimensional
quantification of unstructured features (for example, the depth and diameter of a
series of microscopic holes/pores, lengths, widths and orientation of scratches and
other marks).
Since current surface metrology instruments and data analysis procedures have
been developed so far for the most part to accomplish the primary task of assessing
surface finish over the whole surface (for example, by computing areal field
parameters), in this chapter a whole range of issues and non-standard aspects have
been discussed, which must be addressed in order to accommodate what is
available when individual features are of concern.
In this chapter, the problem has been approached from the perspective of data
analysis; that is, how data obtained from an areal surface topography measurement
instrument could be processed and analysed in order to solve tasks which are
essentially from the field of dimensional metrology, and thus would likely benefit
from data collection solutions more suitable to dimensional and tolerance verifi-
cation (for example, verification by means of a micro-CMM).
In the chapter, the main issues related to the identification, extraction and
measurement of individual areal features from areal topography data have been
discussed. The tools available in commercial software and the most recent ongoing
research approaches have been discussed, with the support of simulations and
actual examples pertaining to real-life industrial applications.
Amongst the main open challenges, a predominant role is taken by the need to
assess the metrological quality of the dimensional characterisation results that can
be obtained by starting from areal topography data. In particular, the problem of
determining the measurement uncertainty associated with each dimensional
characterisation result is still far from solved, and it is likely to become the subject
of considerable debate in the research community in the near future.
Acknowledgements The authors gratefully acknowledge the UK’s Engineering and Physical
Sciences Research Council (EPSRC) funding of the EPSRC Centre for Innovative Manufacturing
in Advanced Metrology (Grant Ref: EP/I033424/1).
References
Bariani P, De Chiffre L, Hansen HN, Horsewell A (2005) Investigation on the traceability of three
dimensional scanning electron microscope measurements based on the stereo-pair technique.
Precis Eng 29:219–228
Bi ZM, Wang L (2010) Advances in 3D data acquisition and processing for industrial
applications. Robot Comp Integr Manuf 26:403–4013
Blateyron F (2006) 3D parameters and new filtration techniques. Proc JSPE (March). Tokyo,
Japan, pp 21–27
Blunt L, Xiao S (2011) The use of surface segmentation methods to characterise laser zone
surface structure on hard disc drives. Wear 271:604–609
De Berg M, Cheong O, Van Kreveld M, Overmars M (2008) Computational geometry:
algorithms and applications. Springer, Berlin
Feng SC, Hopp T (1991) A review of current geometric tolerancing theories and CMM inspection
data analysis algorithms. NIST interagency/internal report (NISTIR), p 4509
Gao F, Leach RK, Petzing J, Coupland JM (2008) Surface measurement errors using commercial
scanning white light interferometer. Meas Sci Technol 19:015303
Giusca CL, Leach RK, Helery F, Gutauskas T, Nimishakavi L (2012a) Calibration of the scales of
areal surface topography measuring instruments: part 1–measurement noise and residual
flatness. Meas Sci Technol 23:035008
Giusca CL, Leach RK, Helery F (2012b) Calibration of the scales of areal surface topography
measuring instruments: part 2—amplification coefficient, linearity and squareness. Meas Sci
Technol 23:065005
Gonzalez RC, Woods RE (2008) Digital image processing. Pearson Education, NJ
Hocken RJ, Periera PH (2012) Coordinate measuring machines and systems, 2nd edn. CRC Press,
Bosa Roca
Hsu T-R (2008) MEMS and microsystems: design, manufacture, and nanoscale engineering.
Wiley, Hoboken
method—measurement standards—material measures. International Organization of
Standardization
ISO 12181 part 2 (2011) Geometrical product specification (GPS)—roundness–part 2:
specification operators. International Organization of Standardization
ISO 17450 part 1 (2011) Geometrical product specification (GPS)—general concepts–part 1:
model for geometrical specification and characterisation. International Organization of
Standardization
ISO 25178 part 2 (2012) Geometrical product specification (GPS)—surface texture: areal—part 2:
terms, definitions and surface texture parameters. International Organization for Standardization
Jiang X, Scott PJ, Whitehouse DJ, Blunt L (2007) Paradigm shifts in surface metrology. Part II:
the current shift. Proc R Soc Lond A463:2071–2099
Jiang X, Zhang X, Scott PJ (2010) Template matching of freeform surfaces based on orthogonal
distance fitting for precision metrology. Meas Sci Technol 21:1–10
Leach RK, Giusca CL, Coupland JM (2012) Advances in calibration methods for micro- to
nanoscale surface measurements. Proc SPIE 8430:84300H1–84300H9
Lin YH, Chen CH (2008) Template matching using the parametric template vector with
translation, rotation and scale invariance. Pattern Recogn 41:2413–2421
MacQueen JB (1967) Some methods for classification and analysis of multivariate observations.
In: Proceedings of 5th Berkeley symposium on mathematical statistics and probability,
University of California Berkeley, pp 281–297
Mandal R, Palodhi K, Coupland JM, Leach RK, Mansfield D (2012) Application of linear systems
theory to characterize coherence scanning interferometry. Proc SPIE 8430:84300T1-10
Marinello F, Bariani P, De Chiffre L, Hansen HN (2007) Development and analysis of a software

tool for stitching three dimensional surface topography. Meas Sci Technol 18:1404–1412
Rusinkiewicz S, Levoy M (2001) Efficient variants of the ICP algorithm. In: Proceedings of
3DIM, Québec City, Canada, 1–8 May
Scott PJ (2004) Pattern analysis and metrology: the extraction of stable features from observable
measurements. Proc R Soc Lond A 460:2845–2864
Senin N, Ziliotti M, Groppetti R (2007) Three-dimensional surface topography segmentation
through clustering. Wear 262:395–410
Senin N, Pini S, Groppetti R (2010) Identification of micro topographic surface features and form
error assessment. In: Colosimo BM, Senin N (eds) Geometric tolerances: impact on product
design, quality inspection and statistical process monitoring. Springer, London
Senin N, Blunt LA, Tolley M (2012a) The use of areal surface topography analysis for the
inspection of micro-fabricated thin foil laser targets for ion acceleration. Meas Sci Technol
23:105004
Senin N, Blunt LA, Tolley M (2012b) Dimensional metrology of micro parts by optical three-
dimensional profilometry and areal surface topography analysis. J Eng Manuf 226:1819–1832
Senin N, Colosimo BM, Pacella M (2012c) Point set augmentation through fitting for enhanced
ICP registration of point clouds in multisensor coordinate metrology. Robot Comput Integr
Manuf 29:39–52
Weißensee K, Kühn O, Linß G (2010) Knowledge-based uncertainty estimation of dimensional
measurements using visual sensors. J Phys Conf Ser 238:1–8
Wyant JC, Schmit J (1998) Large field of view, high spatial resolution, surface measurements.
J Mach Tools Manuf 38:691–698
Yi S, Haralick RM, Shapiro LG (1994) Error propagation in machine vision. Mach Vis Appl
7:93–114
Zhou L-X, Gu W-K (1997) Evaluation, evolution, and optimal manipulation of uncertainties in
computer vision. In: Proceedings of IEEE ICIPS (October), Beijing, China, pp 1473–1477
Chapter 9
Multi-Scale Signature of Surface
Topography
H. Zahouani and M. El Mansori
Abstract In this chapter the relationship between surface topography and its
functional performance is partially investigated using the statistical technique of
the roughness amplitude distribution. The three dimensional morphological
approach is used to better understand the connection between surface topography
and functional performance.
9.1 Introduction
The relationship between surface topography and its functional performance has
been partially investigated using the statistical technique of the roughness
amplitude distribution. But this limited analysis technique is not yet fully under-
stood, and does not allow the correlation of the nature of surface roughness and its
tribological function. However, it is becoming increasingly obvious that a full
understanding of the connection between surface topography and functional per-
formance can only be achieved if a three dimensional (3D) morphological
approach is used. From a tribological point of view, the fluid flow phenomena
between contacting rough surfaces is interesting for investigating lubrication
mechanisms and sealing problems. For example, metal-to-metal sealing parts are
sometimes used as separable fluid connectors. It is well known that the surface
H. Zahouani (&)
Laboratoire de Tribologie et Dynamique des Systèmes, UMR CNRS 5513,
ENISE, 58, Rue Jean Parot, 42023 Saint-Etienne, France
e-mail: Hassan.Zahouani@ec-lyon.fr
M. El Mansori
Laboratoire de Mécanique et Procédés de Fabrication (LMPF-EA4106),
Arts et Métiers ParisTech, Rue Saint Dominique, BP 508, 51006
Châlons en Champagne, France
e-mail: mohamed.elmansori@ensam.eu

218 H. Zahouani and M. El Mansori
morphology of contacting parts governs the pressure distribution and the leakage
across metal seals; essentially the local morphology plays an important role in
contact mechanics and leakage rate. Recent research has shown the importance of
the surface waviness related to the summits envelope in Coulombic friction and
adherence, the pressure supported by the summits during an elastic contact and the
size of the motifs in mixed or hydrodynamic lubrication.
For the clarity, the first part of the chapter is focused on the characterisation of
surface summits and the modelling of contact mechanics and friction. The second
part of the chapter concerns the development of a new approach for surface
morphology assessment with regard to the tribological performance. This approach
consists of a morphological tree transform of the surface motifs. Compared to the
Fourier transform, the motif transform is based on the detection of surface motifs,
which can be defined as a combination of the summits and pits for a linear
morphology, and as an association of the summit to the valley of the surface in the
case of a mixed morphology.
Three parameters are necessary to define a 3D motif: r, l and q. The depth r is
computed as the maximum difference between the summits and pits, the width l is
defined as the Cartesian distance between the summits and q is the orientation of
the motif. The 3D morphological tree can be represented by the density of the
motifs oriented in each q direction at different scales of r and wavelength l. This
multi-scale representation is useful in tribology; the summit of the morphological
tree permits the characterisation of the waviness of the summits envelope—each
summit can be characterised by the height and the radius of curvature in the x and
y directions necessary for computing the Hertzian pressure. The local slopes of the
summits envelope play an important role in Coulombic friction, adherence and
leakage. The fluid reservoir can be characterised by the multi-scale aspect of the
motifs size and their orientation.
9.2 Importance of the Summit Geometry in Contact

Between Non-conforming Surfaces
It is well known that the contact between rough surfaces occurs at asperities and
results in high pressure in the contacting asperities, and a small real contact area,
which is only a fraction of the nominal contact area. Relationships between the real
contact area and the load, and between separation and load, are two basic char-
acteristics necessary for understanding many tribological situations, such as fric-
tion, adhesion, wear, thermal and electrical conductance, and sealing. Analysis of
the contact process shows that it is very difficult to predict the nature of contact
owing to the random diversity of the shape and height of contacting asperities, the
elastoplastic type of strain, material hardening and asperity deformation. However,
several attempts have been made in this field. According to the deformation laws
9 Multi-Scale Signature of Surface Topography 219
which have been applied, the models could in most cases be easily divided into
two opposite types: purely plastic and purely elastic models.
One of the first plastic models was introduced by Abbot and Firestone (1933).
According to this model, the real contact area is the area of geometrical inter-
section between a rough surface and a plane at the intersection level. Greenwood
and Williamson (GW) (1966) introduced an asperity based elastic model. The
Hertz solution has been used as the deformation law for an individual asperity,
thus limiting the model’s applicability to elastic contact. The GW model has been
successively extended to include such aspects as anisotropic surfaces, non-uniform
radius of curvature of asperity summits and non-Gaussian empirical distributions
of asperity heights (Greenwood and Williamson 1966; Pullen and Williamson
1972; Tsukioze 1968). A further surface parameter of interest in contact problems
is the radius of curvature of each contacting summit, which controls the load
supported by each summit, the individual area of contact, the local mean pressure
and the components of friction force.
The geometrical parameters of the relief which will be identified as descriptors
of the multi-scale signature are shown in Fig. 9.1. The fundamental parameter of
the process signature is the basic motif, which can be defined as a combination of
two summits and a pit on the surface, with a depth q and width k. The ratio q/k is
the scale factor of the surface topography at a given scale. From the point of view
of metrology, this parameter represents the ratio of the vertical resolution and the
lateral resolution of the measuring device. The parameters of the upper envelope
will be defined by the local descriptors: the radius of curvature R of the summits,
the summit’s angle 2a, and the attack angle b of each asperity. The degree of
openness of the local pattern is defined by the angle 2c and the direction of the
grooves machining angle h.
Applications that will be discussed to illustrate the major role of the morpho-
logical approach include: contact mechanics, adhesion, friction, lubrication, tactile
perception and the surface morphology of human skin during aging.
Fig. 9.1 Fundamental descriptors of roughness and the basic motif of the process signature
9.2.1 Geomorphological Identification of Local Summits
9.2.1.1 Summit Identification
The conceptual method created to characterise surface morphology is based on the

geographical approach developed initially by Peucker and Douglas (1975), in
order to classify the geomorphologic features of the Earth’s relief as a family of
summits, crests, passes, valleys, ridges, slopes, etc. This approach has been
modified and applied to real engineering surfaces (Zahouani et al. 1994; Zahouani
1995a). The geomorphologic analysis supposes that every point on a surface
can be identified and classified by analysis of its neighbors. Consider a point X
whose elevation with reference to a mean plane is ZX . This elevation can
be subtracted from those of its neighbours in either a clockwise or counter
clockwise sequence (Table 9.1). The result is an array of positive or negative
numbers difference Zij ZX ¼ Dþ or D allowing the characteristic features to
be recognised. The following quantities are usually introduced: n = number of
neighbouring points, Di = height difference between a feature and one of its
neighbours (i = 1, n), Dþ = sum of all positive differences, D = sum of all
Table 9.1 Algorithm approach for summit identification
Fig. 9.2 Identification of summits

negative differences, NX = number of sign changes and LX = number of points

between two sign changes. The summit topological identification is shown in
Table 9.1 where ts is the threshold for summits: 2 % ts 10 % of total roughness.
The example in Fig. 9.2 shows the localisation of summits of different mor-
phologies. Each family of a particular topology is statistically analysed by
determining the 3D Cartesian co-ordinates (x,y,z) of each detected element, the
height distribution of the summits and statistical parameters.
9.2.1.2 Principal Radius of Curvature of Summits
3D surface topology detection was applied to estimate the position (x,y,z) of the
summit, the radius of curvature in x and y directions, the root mean square of the
summits height and summit height distribution. The summit radius of curvature of
each asperity in x and y directions relative to the mean square plane of the surface
are defined as Fig. 9.3 (Zahouani et al. 1997).
The radius of curvature of summits is given by
180ðDxÞ2
Rxs ¼ ;
ð2zi3; j 27zi2; j þ 270zi1; j 490zi; j þ 270ziþ 1;j 27ziþ2; j þ 2ziþ3; j Þ
ð9:1Þ
180ðDyÞ2
Rys ¼ :
ð2zi;j3 27zi;j2 þ 270zi; j1 490zi; j þ 270zi; jþ1 27zi; jþ2 þ 2zi; jþ3 Þ
where Dx and Dy are the sampling steps in the x and y direction, and zij is the
height of summit neighbours in the x and y directions. The mean radius of summit
curvature is defined as
1=2
Rs ¼ Rxs Rys : ð9:2Þ
Fig. 9.3 Radius of curvature

of summit in x and
y directions
9.3 Involvement of Radii of Curvature in Elastic Models

of Multiple Asperity Contacts
9.3.1 Dry Contact with Spherical Geometry
The treatment of a contact problem between two surfaces is simple: the total force
is the sum of all of the equal ‘‘summit’’ forces, which can be calculated using
Hertzian contact theory (Hertz 1881). The individual ‘‘micro’’ contact areas and,
therefore, the total contact area are, in this case, A F 2=3 : This expression con-
tradicts not only direct experiments, but also Amontons’ law of friction, according
to which the frictional force is approximately proportional to the normal force.
Therefore, a somewhat linear increase in the contact area with respect to normal
force is expected. The situation changes if a real surface which is stochastically
rough is taken into account.
The simplest method for modelling an irregular surface was proposed by
Greenwood and Williamson (1966). In this model the authors assumed that all
roughness peaks (asperities) have the same radius of curvature and that the heights
of the peaks are stochastically distributed around a mean value, Fig. 9.1. If the
contacting peaks are far enough away from each other, then their deformations can
be considered independent of each other. Thus, the positions of the peaks and,
therefore, the exact configuration of the surface, matters little in the contact
problem (under the given assumptions). Only the height distribution of the peaks is
of importance. The probability density of an asperity can be described as having
the maximum height z as U(z). This means that the probability that an asperity has
its maximum height in the interval [z, z ? dz] is equal to U(z)dz. If the total
number of asperities is N0, then the number of asperities in the interval [z, z ? dz]
is equal to N0 U(z)dz. For many technical and natural surfaces, it can be assumed
that the heights of the peaks are normally distributed
!1=2 !
1 z2
UðzÞ ¼ exp ð9:3Þ
2pr2p 2r2p
where the value rp is the root mean square of the height distribution
pffiffiffiffiffiffiffiffi
rp ¼ hz2 i: ð9:4Þ
Consider a contact between an elastic body with the described statistical
roughness and a rigid plane at a distance h0 from the centre level, at which the
value of zero for the z axis is assumed (see Fig. 9.4). Under the assumption that the
elastic interactions between the asperities can be neglected, all of the asperities
with a height z [ h0 are in contact with the rigid plane. The ‘‘penetration depth’’ of
an asperity with height z is d = z - h0. For a single contact, a2 = Rd, according to
Hertzian theory. Therefore, the contact area of a single asperity is
Fig. 9.4 Stochastic model according to GW
Ai ¼ pa2i ¼ pRdi ¼ pRðzi h0 Þ; ð9:5Þ
the single force is

4 3=2 4
Fni ¼ E di R1=2 ¼ E R1=2 ðzi h0 Þ3=2 ð9:6Þ
3 3
with
1 1 m2i 1 m2s
¼ þ : ð9:7Þ
E Ei Es
E* is the reduced elastic modulus of the indenter-sample system, Ei and vi are the
Young’s modulus and Poisson’s ratio of the indenter respectively, and Es and vs
the Young’s modulus and Poisson’s ratio of the sample respectively. The total
number of contacts, the total contact area and the total normal force Fn are found
through integration over all of the asperities in contact. This means that the
integration must be performed over all heights from z = h0 to infinity (Greenwood
and Williamson 1966)
Z 1
N¼ N0 UðzÞdz: ð9:8Þ
h0
The total area of contact and normal load are respectively given as
Z 1
Ar ¼ N0 pRðzs h0 ÞUðzs Þdzs ð9:9Þ
h0
and
Z 1
4
Fn ¼ N0 E R1=2 ðzs h0 Þ3=2 Uðzs Þdzs : ð9:10Þ
3 h0
The average contact pressure is given by Abbot and Firestone (1933), Tabor
(1951)
Fig. 9.5 Contact area and pressure of spherical morphology in contact with a rigid plane
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 ffi
4E d 1=2 4E z h0 1=2 4E z
r ¼ ¼ at h0 ¼ 0: ð9:11Þ
3p R 3p R 3p R
The radius of curvature for an asperity is calculated using 1=R ¼ z00 . Therefore,
the mean pressure is given by Popov (2010)
qffiffiffiffiffiffiffiffiffi
4E pffiffiffiffiffiffiffiffiffiffiffiffiffi00 4E
hri ¼ z z ¼ hz0 2 i:
3p 3p ð9:12Þ
For Eq. (9.12), it is taken into account that the average z z00 is defined as the
RL
integral ð1=LÞ 0 zð xÞ z00 ð xÞdz over a sufficiently large distance L. Partial inte-
RL
gration yields ð1=LÞ 0 z0 ð xÞ z0 ð xÞdx and, thus z02 . Equation (9.12) can be
02
summarised by using the symbol rh ¼ z for the root mean square of the slope
of the surface profile.
Figure 9.5 shows an example of contact between a rigid plane and a surface
modelled by a spherical roughness with different radii of curvature (Zahouani et al.
2011).
9.3.2 Involvement of Radii of Curvature in the Elastic–

Plastic Model of Multiple Asperity Contacts
For the sake of simplicity the discussion is restricted to the contact between a
rough surface and a perfectly smooth plane. Assume a surface roughness with
spherical asperities of identical radius and neglect the elastic interaction between
the asperities. With these assumptions, the roughness will be described by the
distribution of the asperity heights. Denoting by zi the summit height of a given
asperity and by d the position of the smooth plane with regard to the mean surface
plane, the local deformation of asperity i (i = 1, …, N) will be
di ¼ zi d: ð9:13Þ
The local elastic–plastic behaviour of this asperity will be described by the
relationship between di and the corresponding transmitted normal force Fi. The
related significant variables will be: Ai contact area, ai contact radius and pi = Fi/
Ai the mean contact pressure. The asperity behaviour will then be described by an
asperity model defined by the three following functions
Fi ðdi Þ; Ai ðdi Þ; pi ðdi Þ: ð9:14Þ
This model depends on the mechanical properties of the material and on the
geometry of the asperity. In the simple case discussed here, the model will be
identical for all asperities. Starting from a given asperity height distribution and a
prescribed value of d, the local deformation is computed for all asperities and, if
positive, used in the asperity model. Summation of all these contributions will
finally provide the global contact behaviour: contact force, contact area and
pressure.
9.3.2.1 An Elastic–Plastic Asperity Model
The material is considered as elastic and ideally plastic with reduced Young’s
modulus E* = E/(1 - m2) and tensile yield stress Y related to the hardness
H (Tabor 1951; Johnson 1985). The elastic model is obtained from Hertz theory
(Tsukioze 1968) as: the local deformation:
di ¼ zi d; ð9:15Þ
Herzian force:
4 1=2 3=2
Fei ¼ E Ri di ; ð9:16Þ
3
the local elastic contact radius:
ai ¼ ½ðzs dÞRi 1=2 ; ð9:17Þ
the local elastic contact area:

Aei ¼ pdi Ri ; ð9:18Þ
the local elastic contact pressure:

4E di 1=2
pi ¼ : ð9:19Þ
3p Ri
The plastic deformation will occur for p = 0.39H (H is the the hardness of
material) and the local transition deformation from the elastic to the plastic regime
is given by the critical penetration depth

H 2
di ¼ 0:92 Ri : ð9:20Þ
E
In the rigid plastic limit, the pressure p reaches the hardness, while the contact
force increase results from the increasing contact area A, which in turn is usually
assumed to be given by the geometrical condition:
the plastic contact area:
Ap ¼ pa2 ¼ 2pdR; ð9:21Þ
so that, the plastic force:

Fp ¼ 2pdRH: ð9:22Þ
An asymptotic regime, which, according to Zhao et al. (2000), is obtained when

the contact force at fully plastic deformation, F2 is approximately equal to 400
times that at initial yielding F1. Hence, the relationship between the critical
deformations is d2 = 54d1. Between these two limiting values, an elastic–plastic
transition is obtained. The elasto–plastic pressure pep is expressed as a logarithmic
function of deformation d, as proposed in the literature (Tabor 1951)

d
pep ¼ k1 þ k2 ln ; ð9:23Þ
a
while the elastic–plastic contact area is given by

Aep ¼ padR; 1\a\2: ð9:24Þ
The pressure continuity requires that pep = p(d1) in the elastic regime for
d = d1. It also requires the pressure pep(d2) = H for d = d2. By simultaneously
solving the equations related to these two conditions, k1 and k2 can be determined
(Zhao et al. 2000). Therefore, the relationship pep(d) is given by
lndc2 lnd
pep ¼ H Hð1 0:6Þ : ð9:35Þ
lndc2 lndc1
Analysis of the boundary conditions leads also to the following expression for the
elasto–plastic area (Zhao et al. 2000)
" #
d dc1 3 d dc1 2
Aep ¼ 2pRd 2 þ3 : ð9:26Þ
dc2 dc1 dc2 dc1
The function, which links the area Aep to the d1 \ d \ d2 interval, increases in a
monotonous manner. Therefore, the contact force is expressed by the following
relationship
Fig. 9.6 Prediction of the contact areas of spherical asperities at different stages of deformation
Fep ¼ pmep Aep

" #
lndc2 lnd d dc1 3 d dc1 2
¼ pRd H Hð1 kÞ 12 þ3 :
lndc2 lndc1 dc2 dc1 dc2 dc1
ð9:27Þ
Figure 9.6 shows the various local contact situations of a rough 52100 steel
surface (H = 8 GPa), in contact with a smooth rigid plane at different stages of
deformation.
9.3.2.2 Adhesion Effect
During a compressive elastic contact, the energy of the spherical asperity is

composed of an elastic contribution and an adhesive contribution. This total force
is the sum of elastic (Hertzian force) and adhesive force
F ¼ Fe þ Fa ð9:28Þ
with
4 a3 4 1=2 3=2
Fe ¼ E ¼ E R d
3 R 3 ð9:29Þ
1=2
Fa ¼ 8ca E pa3 :
The maximum negative value of this force is reached when the contact radius
a reaches the critical value (Popov 2010)
1=3
9ca pR2
a ¼ acrit ¼ ð9:30Þ
8E
and is equal to
3
Famax ¼ ca pR: ð9:31Þ
2
The absolute value of this force is called the adhesive force. The penetration
depth in the critical state is given by
2 1=3
3p ca R
dcrit ¼ : ð9:32Þ
64E2
9.3.3 Elastic Contact Model with Conical Geometry

of Asperities: Effect of Slopes
9.3.3.1 Dry Contact
The proposed analysis of a single conical asperity was developed by Sirghi and
Rossi (2006), in nanoscale indentation experiments (using tips of AFM probes).
This approach introduces the effect of indenter-sample adhesion force in nanoscale
indentation experiments. Assessment of elasticity at the nanoscale is usually based
on Sneddon’s solutions of elastic indentation of half space samples by rigid ind-
enters of arbitrary axis-symmetric profiles (see Fig. 9.7). The stiffness of the axis-
symmetric elastic contact, Kz ¼ dFn =dd, is
Kz ¼ 2E ac ð9:33Þ
where Fn is the loading force, d the indentation depth, ac the radius of the indenter-
sample contact. The pure elastic force for the conical tip-sample contact is given as
Sirghi and Rossi (2006)
Fig. 9.7 A schematic representation of multi-asperity contact with a conical morphology

indenting an elastic body
Fig. 9.8 Contact area and pressure supported by anisotropic morphology

2 2E
Fe ð dÞ ¼ d tan a: ð9:34Þ
p
Figure 9.8 shows an example of the elastic model of the contact area and
pressure, supported by anisotropic morphology, during a contact with a spherical
indenter (Zahouani et al. 2011).
9.3.3.2 Role of Geometrical Parameters on Abrasion Process
The assessment of the properties of coated abrasives for use in belt finishing is a
complex problem due to variation of grit morphology from particle to particle. A
3D assessment method providing an accurate description of the belt surface is
necessary. Morphological image analysis of the belt surface topography consists of
the application of local morphological analysis of motifs. This analysis allows the
detection of the cutting edges of abrasive grits in the belt surface; it takes into
account the grain morphology by considering that an abrasive grain usually
Fig. 9.9 Statistical analysis of the half apex angle of the abrasive grain diameter of 9 lm (a) and
30 lm in diameter (b)
presents multiple summits. Then the average attack angle b; or the average of the
summit’s angle a of the abrasive cutting edge area, is considered as a major
signature for the characterisation of the wear evolution of the abrasive belt.
Figure 9.9 gives an example of the statistical analysis of the half apex angle of the
abrasive grain diameter of 9 lm (a) and 30 lm (b).
The correlation between the results shown in Fig. 9.7, and the increase of the
removed material when the abrasive grit size increases, shows the fundamental
role played by the angularity of the abrasive grains in the transition from cutting to
plowing and sliding mechanisms. Indeed, the greater the attack angle hs of the
abrasive cutting edges, the more the edge favours the cutting phenomena and,
therefore, material removal. Figure 9.10 shows an example of the attack angle
distribution of a successive finishing with grains of diameter 9 and 30 lm on a
surface initially machined with hard turning process.
9.3.3.3 Effect of Abrasive Finishing on the Contact Pressure

and Bearing Area
To predict the behaviour of dry contact surfaces from belt grind finishing, a model
of contact with a rigid plane of surfaces allows the evolution of the contact
pressure and the bearing area to be studied after different finishing stages (see
Fig. 9.11) (Jourani et al. 2005b).
Fig. 9.10 Attack angle distribution of a finishing surface with abrasive grains of diameter 9 and
30 lm (the surface was initially machined with hard turning process)
9.3.3.4 Effect of Adhesion
To take into account the adhesion effect (Sirghi and Rossi 2006), consider the
external loading force applied to the indenter during the unloading process, Fn, is
the sum of the elastic force of the sample, Fe, and adhesive force Fa
Fn ¼ Fe þ Fa : ð9:35Þ
The external force of the contact is given by Johnson (1985)

2E tan a 2 ca 8 tan a
Fn ðdÞ ¼ d d ¼ Fe ðdÞ þ Fa ðdÞ: ð9:36Þ
p p cos a
To extend Eq. (9.36) to the asperities in elastic contact with a smooth plane (see
Fig. 9.3), Eq. (9.32) introduces the average value of asperity cone angles, to
determine the total average force in the presence of adhesion
2E d2 8c d D tan a E
h Fn i ¼ htan ai a : ð9:37Þ
p p cos a
9.3.3.5 Application of Multi-Scale Roughness to Fractal Models
The analysis so far has focused on rough surfaces, whose texture often exhibits
fractal properties (see Chap. 6). Several machining processes lead to surfaces
Fig. 9.11 Modelling of the evolution of contact pressure and the bearing area after different
finishing stages
having fractal properties. In particular Majumdar and Bhushan (1991) and Bhu-
shan and Majumdar (1992) show that processes producing deterministic texture do
not yield self-affine fractal surfaces, whereas those producing random textures do.
The basic properties of self-affinity are presented in more details in Feder (1988)
and can be defined as follows.
A profile Z(x) is self-affine if it is (statistically) invariant under the affine
transformation

x ! kx
: ð9:38Þ
ZðxÞ ! lZðxÞ
The group properties imply that l should be an homogeneous function of k. The

homogeneity index H is such that
l ¼ kH ð9:39Þ
where H is the roughness Hurst exponent or Hölder condition, with 0 \ H 1.

Note that for a self-similar invariance, H is unity. In this case, both scaling factors
are equal.
The details of Z(x) depend on the length scale; each realisation of Z(x) is
assumed to be a continuous, but non differentiable function. This means that the
presence of any small roughness elements may prevent a satisfactory limit from
being reached (Zahouani et al. 2011), thus
ðZðx þ DÞ ZðxÞÞ
as D ! 0: ð9:40Þ
D
A simple way to obtain this behaviour for a function Z(x) is to assume that the
increment of Z(x) is related to D by the self-affinity relationship
jZðx þ DÞ ZðxÞj / DH ; 0 \ H \ 1 D ! 0 ð9:41Þ

This relationship can be generalised to dimensions E [ 1 as follows
D E
ðZðx þ DÞ ZðxÞÞ2 / kDk2H ; 0 \ H \ 1; D ! 0; ð9:42Þ
where x represents a point in E-dimensional Euclidean space and || || stands for the
usual norm in this space. This is not a new condition; it is verified for all kDk
values by the fractional Brownian motion.
9.3.3.6 Modeling Roughness with a Large-Scale of Slopes

by the Random Midpoint Displacement Model
For random rough surfaces, the random displacement method used for a fractal
profile provides one of the simplest algorithms to generate random fractal surfaces.
For analysis proposes, it is particularly useful to have a numerical tool that creates
the same kind of morphological data arrays as most scanning acquisition methods
for real range images, namely an x, y array of Z values: Z(x,y). The algorithm of the
random displacement method for rough surfaces is used to generate fractal sur-
faces in a wide range of scales. Figure 9.12 shows a fractal surface generated with
Fig. 9.12 Fractal surfaces generated with the variation of local slopes
this approach; the size of each generated surface is 512 9 512 points with the
steps in the x and y directions of 1 lm.
9.3.3.7 Effect of Roughness Scale on Contact Stiffness
To study the effect of roughness scale on contact stiffness without adhesion

(ca = 0) (Jourani et al. 2005a; Vallet et al. 2009a, b; Zahouani et al. 2009), a
numerical indentation can be performed with a 3D elastic model on a poly-
dimethylsiloxane (PDMS) surface (Young’s modulus of 2.2 MPa) (Zahouani et al.
2011). The texture of the steel indenter is generated with a fractal morphology.
The procedure of the 3D elastic simulation consists of bringing two surfaces
numerically close to each other, and solving the equation for each level of pen-
etration d. For each normal position d of the rigid plane, contact pressures p(x,y) is
determined by using the direct and inverse Fourier transforms of the contact
equations. The step of penetration is chosen as d = 0.01 lm. The sum of the local
efforts calculated for each penetration d is given as
XX
FðdÞ ¼ pij ðx; yÞDxDy : ð9:43Þ
i; j
In the range of penetration, d is chosen between 0.01 and 0.5 lm. The fractal
dimensions used for this simulation are D = 1.15, D = 1.3 and D = 1.5. Fig-
ure 9.13 shows the fractal scale effect on the contact stiffness in the absence of an
adhesive force.
9.3.3.8 Effect of Roughness Scale and Adhesive Force on Contact

Stiffness
To take into account the adhesion effect on the elastic contact, the adhesive term of
the relationship defined in Eq. (9.31) is introduced. The total normal load of elastic
adhesive contact can be written as
Fig. 9.13 Effect of fractal

dimension on contact
stiffness
Fig. 9.14 Effect of adhesion and fractal dimension on contact stiffness
Fn ¼ Fe þ Fa : ð9:44Þ
The total normal load using the 3D elastic model and the adhesion effect can be
written, for each penetration of a rigid rough plane, as a combined elastic and
adhesive force (Zahouani et al. 2011)
XX 8c dk D tan a E
Fn ðdk Þ ¼ pij ðx; yÞDxDy a : ð9:45Þ
i; j
p cos a
In the adhesive component, note the effect of the local roughness scale in terms of
local slopes, where the mean slope value a is computed in the contact region at any
depth of penetration dk (k the index of the penetration depth, the step of penetration
is chosen as d = 0.01 lm and the maximum of penetration depth is 0.5 lm). The
load–displacement represented in Fig. 9.14 clearly shows the combined effect of
adhesion and fractal dimension on contact stiffness.
9.4 Application to Tire-Road Adherence
Road transport is the main means of transport in most developed countries.

Nevertheless, it is a very dangerous form of transport. The World Health Orga-
nisation predicts that by 2030, road accidents will become the fifth leading cause
of death worldwide. This is why governments are trying to combat the phenom-
enon. Although the statistics on road safety cannot precisely quantify the pro-
portion of accidents caused by a loss of adherence on wet road, the studies agree
on the fact that rain is an important risk factor during accidents.
Fig. 9.15 Multi-scale texture

of a road
9.4.1 Scales of Road Texture
The combination of aggregate, sand and binder give the road pavement a multi-
scale texture. Usually two roughness scales are distinguished: macrotexture and
microtexture (see Fig. 9.15 and Chap. 14). Adhesive friction is usually connected
to microtexture, whilst friction hysteresis is mainly influenced by macrotexture. In
the case of wet road surfaces, adhesive friction drops dramatically with increasing
speed, so that the friction of hysteresis is responsible for 90 % of the adhesion
between the tire and the road.
9.4.1.1 Macrotexture
Macrotexture encompasses wavelengths between 0.5 and 50 mm, with a vertical

amplitude between 0.1 and 20 mm, as shown in Fig. 9.15. Macrotexture is related
to the particle size distribution of the road and its implementation. This is the
major contribution to the generation of hysteresis. In the case of rain, macrotexture
affects the drainage capacity (training natural water in the soil and down the side)
and discharge (by the tire) of water on the floor.
9.4.1.2 Microtexture
Microtexture encompasses wavelengths between 1 lm and 0.5 mm, with an

amplitude between 1 lm and 0.2 mm. This scale of texture is due to small
asperities on the surface of the aggregates (see Fig. 9.16), fine sand and embedded
particles.
To evaluate the effect of the scale of the texture on contact pressure, the scale
parameter is defined as the ratio of the height and width. This parameter defines the
deformation transmitted by the texture to the tire (see Fig. 9.17), and can be
defined as
z
p ¼ E ; ð9:46Þ
k
Fig. 9.16 3D morphology of the road texture
Fig. 9.17 The scale effect on

local contact pressure
with kz ¼ e; the deformation defined by indentation theory.

The study presented here focuses only on low water depths (less than 1 mm).
The phenomena involved are very different from those that lead to hydroplaning,
see Fig. 9.18. This loss of adherence on thick layer water has been widely studied
in the literature.
Two mechanisms are responsible for the generation of frictional forces: the
deformation of the tire rubber on the asperities (called hysteresis) and adhesion,
which is the result of molecular interactions between the road surface and the tire.
If it is assumed that there is negligible interaction between the adhesion and
deformation processes during sliding, they can be summed, and the total intrinsic
frictional force Fi equals the force needed to shear adhered junctions Fa, and the
force needed to supply the energy of deformation Fd. Therefore, the friction
coefficient can be expressed as li = la ? ld.
Fig. 9.18 Tribological

situation
9.4.1.3 Deformation Component (Hysteresis)
The tire is a visco-elastic material. This property creates an asymmetry in the

deformation of the tire rubber asperities on the road surface, giving rise to a
resultant force opposed to sliding. As shown in Fig. 9.19, this component of
friction is the most important friction on wet road surfaces.
For a spherical asperity of radius R in contact with a softer body, the expression
for ld is
rffiffiffiffiffi
4 a 4 rs
ld ¼ or ; ð9:47Þ
3p R 3p R
with a being a contact radius, rs the root mean square height of summits, and R is
the mean radius of summits.
9.4.1.4 Adhesion Component
The adhesive friction is the result of molecular interactions between the tire and
the road. It requires direct contact between the rubber and the road surface. The
molecular bonds are formed, and then stretch break during sliding of the rubber on
the surface, as shown in Fig. 9.19. This load dissipates energy by friction of the
molecular chains together. This phenomenon could no longer occur if the presence
Fig. 9.19 Mechanisms of adhesive friction

of a contaminant prevented direct contact between the rubber and the road surface,
which explains the loss of adherence between tire and wet road.
From the classical theory of adhesion and to a first approximation, the friction
force Fa is defined as follows. For a dry contact
F a ¼ s a Ar : ð9:48Þ
For a contact with a partial film

Fa ¼ Ar ½asa þ ð1 aÞsl ð9:49Þ
and sl ¼ ghl V , where sa and sl are the average shear strengths of the dry contact and
of the lubricant film, respectively; a is the fraction of unlubricated area, gl is the
dynamic viscosity of the lubricant; V is the relative sliding velocity; and h is the
liquid film thickness.
For dry contact, the coefficient of adhesive friction is
s a Ar
la ¼ ð9:50Þ
pr
where pr is the mean real pressure. For elastic contacts
3:2sa
la rs 1=2 ð9:51Þ
E R
where rs is the root mean square height of summits.

After several experimental tests, the friction measurements with a pendulum
showed excellent correlation between the deformation component and the mea-
sured friction coefficient. The linear trend between the coefficient of friction and
the angle of attack of the asperities, shows the role of microtexture in adhesion,
and the importance of the attack angle of the microtexture, see Fig. 9.20.
Fig. 9.20 Correlation

between the friction
coefficient and the mean
attack angle of texture
9.5 Application to Human Touch: Effect of Roughness

Geometry on Vibration of the Human Finger During
a Friction Test
To characterise vibrations transmitted to the finger during a test of tactile per-

ception, a tribo-haptic system was developed, see Fig. 9.21. The human finger
plays the role of a resonator and the roughness of the tested material as a source of
excitation (Zahouani et al. 2013). When the finger is rubbed on the test surface,
vibrations are generated and excite different tactile receptors in a wide range of
frequencies depending on the roughness and stiffness of the surface tested. An
accelerometer, which detects the vibrations, is glued to the human finger as shown
in Fig. 9.21, and has the following characteristics: mass of 0.14 g, diameter of
3.58 mm, sensor gain of 1 Vm-1s-2 and vibration sensor sensitivity of
0.5 pCm-1s-2. The normal force is measured by a force sensor placed just below
the support of the surface to be tested. The maximum effort can reach 0.8 N with a
resolution of 1 mN.
The displacement of the test surface is provided by an ultra-compact linear
actuator and linear motor. The monolithic construction leads to a high mechanical
stiffness. This linear displacement system operates on the same principle as torque
motors, but is open and placed flat. The electromagnetic interaction between a coil
assembly (primary assembly) and a path of permanent magnets (all secondary),
converts electrical energy into mechanical energy with high efficiency.
9.5.1 Correlation Between Roughness of Random Texture

and Vibration Level
A test touch on different abrasive papers was made at a speed of 10 mm s-1.

Figure 9.22 shows the morphology of abrasive paper and the vibration signal of
Fig. 9.21 View of the

experimental ‘‘Haptic
tribometer’’: (1)
accelerometer, (2) normal
force sensor, (3) displacement
system (ultra-compact linear
axis)
Fig. 9.22 Areal topography of abrasive paper and vibration signal transmitted to the finger:
a = 600, b = 1200, c = 2400, d = 4000
the human finger (Zahouani et al. 2013). These results show an increase in the
average level of vibration depending on the scale of roughness via the mean of
attack angle, see Fig. 9.23.
9.5.2 Contact Model: Multi-Stimuli and the Human

Finger—The Transfer of Roughness to the Fingerprint
To identify the role of the digital print of the human finger on the contact, an
elastic model was adopted allowing for interactions between all contact points and
to predict the contact geometries of the finger topography under load with a
spherical stimulus (Zahouani et al. 2013). The effect of multiple stimuli was
investigated by texturing the surface of the spherical indenter with a periodic
Fig. 9.23 Correlation between mean level of vibration (Lv [dB]) and the mean attack angle of
texture
roughness (Ra = 40 lm and periodicity of 35 lm). The study of contact shown in

Fig. 9.24 shows how the stimuli print their forms on the surface of the finger, and
how the contact area and the distribution of the pressure field are modified by the
morphology of the texture of the sphere. This morphology transmits a discrete
deformation field, which is consistent with the roughness of the sphere, contrary to
the contact with a smooth sphere which prints its shape only. This result is fun-
damental in understanding the interaction between the morphology of the finger-
print with textures at different scales. The impression of the fingerprint by texture
gives a new vision in the interpretation of cutaneous mechanoreceptors excitement
and vibrations generated.
If the real contact area A is taken into account to determine the local contact
frequency by setting the frequency for a given contact area, f ¼ pvffiffiAffi, the fre-
quencies of vibration generated by the smooth and the rough sphere can be
compared. The calculation shows that for the rough sphere, the root mean square
Fig. 9.24 Elastic modeling of human finger deformation

Fig. 9.25 Frequencies

generated by a smooth and
rough ball depending on the
scanning speed
of the real contact area is equal to the wavelength of the fingerprint. For the rough
sphere, the frequencies generated are much higher than for the rough sphere (see
Fig. 9.25). This simple calculation shows that, in general, the friction of the finger
on a smooth surface generates vibrations whose wavelength is that of the finger-
print. For rough surfaces, with wavelengths less than the wavelength of the fin-
gerprint, the vibration frequencies generated are the result of the interaction
between the roughened surface and the fingerprint.
9.6 3D Motifs Topology
Fluid flow phenomena between contacting rough surfaces are interesting for the
analysis of lubrication mechanisms and sealing problems. Recent investigations
involving wear experiments show that the asperities can be described as a con-
tinuous area with heights exceeding the estimated film thickness. The lifetime of
lubricated systems can be estimated relative to the number of the summits pene-
trating the lubricant film. In other words, permanent lubrication is conditioned by
the capacity of the surface to retain a necessary quantity of lubricant in the valley
of the surface (refer back to Fig. 9.1). The fluid flow phenomena can be influenced
by the directional properties of the local motif and the multi-scale aspects of the
surface morphology.
Consider the motifs of the surface as keys elements in a tribological system.
The summits and their wave envelope constitute the boundary surface of contact
area friction and wear; the form, scale and orientation of the motifs play an
important role in lubrication and leakage problems. The 2D motif is defined as the
part of the profile which associates two peaks separated by a deep valley (pit)
(Zahouani et al. 2001). The extension of the 2D motif definition to 3D morphology
needs the distinction between two categories of motifs families:
A: motifs with linear topology of the grooves which are generally oriented
depending on the manufacturing process (grinding, turning, milling, honing,
etc.).
B: motifs presented as a mixed distribution of holes and bumps at different
scales.
9.6.1 3D Motif with Linear Topology
In this morphology, a 3D motif is defined as the association of two summits

separated with a pit. The topological definition of summit and tip is shown in
Table 9.2 and the schematic diagram of a motif’s definition is given in Fig. 9.26.
Three parameters q; k; h, are necessary to define a 3D linear motif: the depth q
computed as the maximum difference between the summit and the pit

qij ¼ sup Zsummit Zpit ; ð9:52Þ
and the width k defined as the Cartesian distance between two summits in the
direction of the motif
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
k ¼ ðx2 x 1 Þ2 þ ðy 2 y 1 Þ2 : ð9:53Þ
The direction h of the motif is the orientation which coincides with the general
direction of a principal manufacturing scratch or valley.
9.6.2 Morphological Rose of Motifs Direction
Motif direction is achieved by the identification of the maximum local gradient

variation. Once the position (i, j) of a pit has been identified, the two summits that
correspond to the maximum local variation in relief must be found by working
around the valley bottom in all directions. Two factors are used in this
identification:
Table 9.2 3D topological definition of surface tip

Fig. 9.26 Schematic diagram of a motif’s definition
• The direction must coincide with the normal area of the plane determined by the
three elements of the motif illustrated in Fig. 9.27, if the first summit is point A,
point C is the pit, and the second summit is point B. The direction of the motif is
collinear within the limits of the normal N ~ which is given by the vectorial
product
~ ¼ !
N
!
CB CA ð9:54Þ
• For motif direction to coincide with that of a principal valley composed of a
succession of motifs, it must be determined whether the pit point C(i, j) is
related to the orientation of neighbouring pits; this is obtained by determining
the intersection between a half straight from B(i, j) and the nearest pits. If there
is an even number of points, pit C(i, j) is not within the orientation of the groove.
Otherwise it is definitively within the orientation of the principal valley.
The morphological rose is generated by computing the density of motifs ori-
ented in a direction h, identified with respect to the x axis in the original sample.
The geometrical form of the morphological rose represents the local and global
motif anisotropy and allows quantification of the phase of the surface topography.
9.6.3 Anisotropy Index
A perfectly isotropic surface leads to a circular direction rose. For an anisotropic

morphology of motifs, the morphological rose presents the orientation of the
Fig. 9.27 Local detection of the anisotropy of motifs
motifs in privileged directions. So if N is the number of angular sectors, inspected

between 0 and p, an anisotropy index (AI) can be defined as
PN1 S

1 i¼0 i R =N
AI ¼ 100% ð9:55Þ
2 S S=N
where Ri is the rose value corresponding to angular sector i and

XN1
S¼ i¼0
Ri : ð9:56Þ
S/N should be the Ri value for all i in the case of a perfectly isotropic surface
and the factor 1/2 comes from the fact that Ri values greater than S/N must be
exactly compensated by lower values. The example in Fig. 9.27 shows the mor-
phological rose of random and oriented texture.
9.7 Application to the Honing Process
The texture of cylinder bores mainly consists of two sets of straight, approximately
parallel grooves placed stochastically and appearing at different angles to the
cylinder axis (Chap. 11, Tsukioze 1968; Greenwood and Williamson 1966).
Figure 9.28 shows a typical honed surface texture from an engine cylinder.
The texture features of a honed surface can be split into two principal com-
ponents: plateaus and valleys. The plateau component is related to friction and
wear. It has been demonstrated that the coefficient of friction in the piston ring-
liner interface increases with the average roughness of surface liner (Abbot and
Firestone 1933). The valleys component is intuitively associated with lubricant
circulation and reservoirs. The prominence of the honing grooves suggests that
they play an important role in the effect of surface texture on ring-pack perfor-
mance. Friction is also an important factor to consider when working with sliding
surfaces with high contact pressures. A major proportion of friction losses are
determined by the piston/piston ring/running surface module. A reduction in the
friction losses increases the engine output and lowers the fuel consumption of the
engine. Despite a high number of experimental studies of groove texture on
lubrication performance (Pullen and Williamson 1972), its effect on friction is not
well understood.
In this section the effect of lateral groove pattern characteristics on friction
occurring in the interfacial contact between piston ring and cylinder liner surfaces
will be discussed. To this aim, a numerical model, which takes into account the
real topography of the cylinder liner, is developed. The model serves to predict
qualitatively, the friction of the ring-liner-piston contact.
Fig. 9.28 Texture of cylinder liner

9.7.1 Effect of Honing Angle
Honing angle is determined by the vertical and rotational movement of the honing
head

h Va
tan ¼ ð9:57Þ
2 Vr
where Va is the axial speed and Vr is the rotation speed.

Figure 9.29a and b show an example of the identification of the morphology of
the machining grooves for three types of honing by means of the opening angle of
the valleys, the autocorrelation function, the correlation lengths in both directions
(x and y) (see Fig. 9.29a) and the anisotropy quantified by the rose of anisotropy of
manufacturing grooves and the distribution histogram of grooves (see Fig. 9.29b).
These examples clearly show the signature of honing and how the morphological
diversity of the local geometry can alter the tribological behaviour of surfaces
obtained by honing.
It is widely accepted that the honing angle is directly related to oil consumption
and noxious emissions (Sabri et al. 2011). Scientific studies and practical appli-
cations have shown that oil consumption can be reduced by increasing honing
angle. However, the effect of this geometric property on friction has not been
studied previously because groove patterns are always associated with lubricant
performance. Therefore, honed surface topographies with different honing angles
were simulated to study the effect of this parameter on the friction performance of
the cylinder ring contact (see Fig. 9.30).
The coefficient of friction between the piston ring and the cylinder liner is
predicted with the model developed by Mezghani et al. (2012). The model reflects
the energy consumption of the oscillating components, indicating the probability
of wear of the sliding surfaces, and the lubrication conditions at the respective
positions of the piston.
Figure 9.31 shows the effect of honing angle on the coefficient of friction. The
two honing angle ranges are commonly used in two kinds of honing: plateau
honing and helical slide honing. The minimum observed in the second range was
determined in experimental test results given in Zahouani et al. (1997). It has been
shown that the 140 honed liners show the same performance in wear and friction
properties for the top dead centre area compared to a 40 honing angle.
Knowing that oil consumption is reduced with increasing honing angle
(Greenwood and Williamson 1966; Zahouani et al. 1997), the slide honing (115–
130) could be the optimised honing process. It could provide the best global
compromise for engine performance. To study the effect of the other groove
features, the honing angle is fixed to 50.
Fig. 9.29 a Morphology of the machining grooves (angle of valleys and correlation lengths).
b Anisotropy of the the machining grooves
Fig. 9.30 Simulated honed surface with different honing angle (a) 20, (b) 50, (c) 120, and
(d) 160
Fig. 9.31 The influence of

honing angle on the
coefficient of friction. Two
minima of friction are
observed respectively for
honing angle range of 40 to
55 and 115 to 130
9.7.2 Effect of Groove Width
Before investigating the effect of groove width on friction, it is important to note

that the honed surfaces are obtained by a succession of three honing stages.
The first stage, often categorised as rough honing, establishes the form of the bore.
The second so-called ‘‘finish honing’’ stage creates the basic surface texture of the
hole. This enables the third honing stage—plateau stage—to remove the surface
peaks, increasing the micro-relief quality without surface damage (Popov 2010;
Tabor 1951). Nevertheless, experimental studies have shown that the plateau
honing stage allows only a reduction of oil consumption and running-in time. Sabri
et al. (2011) showed, moreover, using a multi-scale characterisation based on
continuous wavelet transform decomposition, that plateau honing only affects
wavelengths lower than 20 lm. These plateau honed surfaces usually have
grooves of 15–35 lm wide in industrial production. Figure 9.32 shows the pre-
dicted result of the friction of the ring cylinder contact for various groove width
variants from 8 to 64 lm. Figure 9.32 shows that the coefficient of friction is
reduced for grooves with widths greater than 20 lm (Mezghani et al. 2012). This
result confirms that the plateau honing stage benefit is limited to the reduction of
oil consumption. Furthermore, careful control of the finish honing operation is
central to optimising friction, since the groove texture is provided by this honing
stage. In the following section, the groove width is fixed at 32 lm.
9.7.3 Effect of Grooves
The groove density parameter Dg was determined for various plateau honed
surfaces derived from a honing production series. The conformity of these surfaces
Fig. 9.32 Friction coefficient

against groove width for a
contact speed of 20 ms-1
Fig. 9.33 Friction coefficient

against groove density
is assured by visual inspection by an expert. It is seen that groove density varies in

the range of 20–40 grooves mm-2. Especially, it was noted that increasing groove
density leads to higher oil consumption (Zahouani et al. 1997). Different groove
textures were simulated with groove densities from 16 to 42 grooves mm-2. The
influence of groove density on friction performance is shown in Fig. 9.33. It is
clear from this figure that optimal friction performance is obtained in the range
20–30 grooves mm-2 (Mezghani et al. 2012).
9.8 3D Multi-Scale Morphological Tree Decomposition

of Motifs
The overall analysis of the 3D morphology of the patterns is illustrated on the

example of an anisotropic surface (see Fig. 9.34). Mapping the height of the
patterns is shown in Fig. 9.34b, rose anisotropy in Fig. 9.34c and the distribution
of heights and widths of the motifs in Figs. 9.34d and e respectively. This first
global analysis is used to store the three parameters of patterns, q; k and h, nec-
essary to determine the 3D morphological spectrum (Zahouani et al. 2001).
9.8.1 3D Multi-Scale Morphological Tree Decomposition
When the local height of each motif q is known, the orientation h against the
height q of the motif can be represented by
h ¼ f ðDqÞ; ð9:58Þ
Fig. 9.34 Morphological decomposition
and each motif height width can be quantified by

k ¼ gðDqÞ: ð9:59Þ
This morphological transformation is used to study the multi-scale property of the

topography and determine the morphological spectrum of a 3D surface in the form
of a multi-scale morphological tree (see Fig. 9.35).
Fig. 9.35 3D morphological tree concept
The example in Fig. 9.36 shows the morphological decomposition of a milled

surface. The morphological tree shows the hierarchy of scales and the 3D
anisotropy of morphology (Zahouani et al. 2001).
9.8.2 3D Motifs with Mixed Surface Topology
The methodology described above can be extended to the mixed motifs mor-
phology. In this case the 3D motif can be defined as the association of the summit
and crest, to the valleys and pits. The summary of the topological definition of
each motif element is described in Table 9.3. The crests and valleys family can be
defined by the thresholds of each topoligical point: tc, tv.
The example, Fig. 9.37 shows the morphological tree of a porous and sand-
blasted surface. The morphology of the tree is consistent with the local mor-
phology of the surface in a hierarchical form. Comparison with the Abbot and
Firestone curve shows good similarity in evolution, with the advantage of the
multi-scale property of the morphological tree.
Fig. 9.36 3D morphological tree of motifs of an anisotropic surface
Table 9.3 Algorithm approach for crests and valleys
9.9 3D Morphology of Biological Surfaces: Human Skin

Topography During Ageing
The surface of skin shows a specific topography depending on anatomical site, age
and sex. In general, skin morphology presents a 3D network of lines, which
expresses all the multi-directional tensions of elastic fibres and the collagen beams.
Micro-lines, primary lines, fine wrinkles and wrinkles represent the special
organisation of collagen bundles and elastic fibres in the superficial dermis, and
there is a relationship between the morphology of the skin lines and the elastic
Fig. 9.37 Morphological tree of a porous and sandblasted surface
network. Different functions can be attributed to the lines network. The first
function is the retention and drainage canals of the sebum and sweat. The canals
preferentially collect and retain the substances applied to the skin for a long time;
they are thus preferential sites for percutaneous absorption. This reservoir function
allows the applied topical products to be stored on the skin surface and then
eventually to diffuse into its different layers. The second function is mechanical—
during ageing, the depth, width, density and orientation of the skin lines changes.
Some lines become more marked; they evolve progressively in marked anisotropy
connected to the decrease of the elasticity of the collagen fibres.
9.9.1 Skin Tension and Anisotropy of Skin Lines Network
When analysing the mechanics of skin in vivo, a significant property is its natural
tension. Discovered by Dupuytren, and mapped by Langer (1978), non-uniform
skin tension lines exist. Langer identified these lines by puncturing the skin with a
circular device (see Fig. 9.38). The wounds then assume an elliptical shape and by
joining the major axes of the ellipses, a system of lines can be drawn.
Skin resistance to traction determines the directions of the Langer’s lines and
varies with body site. On all body sites, the skin tension is greater in the direction
of Langer’s lines (see Fig. 9.39).
Fig. 9.38 Direction of

Langer’s lines: evolution of
circular incisions on the face
This phenomenon is the source of Young’s modulus anisotropy (Greenwood and

Williamson 1966), whose distribution angle shows a maximum in the Langer’s lines
axis. This result favours a similar orientation of the elastic fibres involved in the skin.
Assuming that the fibres are independent, it has been calculated that, on the calf,
76 % of the fibres were in the direction of the Langer lines and 5.1 % were per-
pendicular to them. Of course, the distribution concerns only the elastic fibres, which
are parallel to the skin surface. Observation of the dermis with scanning electron
microscopy confirms this data (Zahouani et al. 1994). In retracted skin, the collagen
bundles appear tortuous, with no special direction, and sinuous elastic fibres are
fixed to them in several places, especially in their concave part. In non-retracted
skin, the thinnest collagen bundles and the elastic fibres are straightened in the
direction of the Langer’s lines, and almost parallel; the thickest bundles remain
tortuous and oriented in all directions, but their shape seems to be modified by the
traction from the oriented bundles and fibres. Contrary to common past belief,
Langer’s lines do not reflect anisotropy of the collagen density, but anisotropy of the
reticular dermis collagen bundles’ direction and elastic fibres’ tension.
9.9.2 Identification of Langer’s Lines
The morphology of lines network reflects the intrinsic tension of the skin, which is
distinct from the additional tension induced by increases in the volume of the
Fig. 9.39 Schematic representation of Langer’s lines on different areas of the body
underlying tissues, such as muscle contraction, edema, or in particular, posture

which stretches the skin. The identification of Langer’s lines must, therefore, be
made on relaxed skin.
9.9.2.1 Skin Micro-Topography Method
With the exception of palms and soles, the skin miro-relief is made of plateaus
separated by valleys. The valleys are roughly parallel and oriented in different
directions, and this layout is characteristic of each body area. The direction of the
deepest valleys matches Langer’s lines. There may be one or two other preferential
directions, indicating an ordered, non-orthogonal mechanical anisotropy. This
analysis has an advantage over the others as it is insensitive to extrinsic skin
tensions (Zahouani et al. 1998b). Its physiological interpretation is simple. The
cutis is normally retracted (skin tension) and extensible, whereas the epidermis has
none of these properties. Therefore, the epidermal creasing responsible for the
Fig. 9.40 Anisotropic signature of human skin topography
micro-relief appears to be a transformation of tension, allowing the creases to be

flattened by stretching them. The superficial dermis, an intermediary zone between
epidermis and cutis, is precisely the place where the skin relief begins. This
mechanical transduction is one of its functions.
It is this approach that has been developed to study the chronological aging of
human skin. The method uses the wealth of 3D imaging of the skin and the
possibility of linking a signature printed on the surface of the stratum corneum
with the organisation in volume of the different skin layers. Figure 9.40 shows the
anisotropy signature of different sites of the body.
9.9.3 Morphological Evolution of Skin Lines Network

of Caucasian French Women During Aging
Several studies have demonstrated changes in the network of lines with age, leading
to deepening of certain lines and the disappearance of others (Zahouani and Vargiolu
2000). However, quantitative and detailed descriptions of the modifications of skin
lines with age are rarely reported in the literature, since most of the published results
are based on standard parameters, which give an overall description of the topog-
raphy of any surface, without specificity for the skin morphology. For this reason, a
3D confocal microscope working with a high vertical and lateral resolution has been
used, which enables accurate characterisation of the skin lines network. Negative
skin replicas were taken with silicone rubber (Silfo, Flexico Ltd, UK) from 120
Caucasian French women equally divided into six age groups (20–29 years,
30–39 years, 40–49 years, 50–59 years, 60–69 years and 70–80 years). Replicas
were taken from the women’s left volar forearm at the same pre-determined area,
Fig. 9.41 Morphology of skin relief of Caucasian women during aging (volar forearm site)
after a 30 min rest period in an environmentally controlled room (temperature

21 ± 2 8C and relative humidity 50 ± 5 %). The vertical resolution of the confocal
microscope is approximately 0.01 lm with a vertical range of 1,000 lm. The lateral
resolution depends on the quality of displacement in the plane x and y axes. In the
case of skin aging, the lateral resolution was 1 lm. Figure 9.41 shows the evolution
of skin volar forearm morphology of Caucasian women aged 20–80 years (Zahouani
and Vargiolu 2000).
The images in Fig. 9.41 reflect the 3D character of the skin lines network. The
scale height is expressed here by a colour scale. The deepest lines are expressed by
the colour blue, heights that are at the top of plates expressed by the colours red to
black. The full scale (peak-to-valley) can reach 500 lm.
To study the transformation of the 3D skin line network during aging, it is
necessary to identify all its local motifs. A motif of skin line is defined by the
association of two peaks separated by the hollow of a valley, and the height is
determined by the difference between the highest peak and the hollow of the valley
(Zahouani and Vargiolu 2000) (see Fig. 9.42).
The width of the motif k is given by the distance between both peaks. The
direction of the motif, which coincides with the main direction of the line, is
defined in the orthogonal direction at the maximal variation of the local gradient.
This quantitative approach enables the evolution of the different families of
lines in relation to age to be followed. The depth Z depends on the skin site and
Fig. 9.42 Definition of tension lines patterns
two classes of depth Z1 B 60 lm and Z2 [ 60 lm were adopted for volar forearm

aging. This choice was fixed after sampling every 10 lm as the class depth. The
results showed a marked decrease in the density of the family lines of depth less
than 60 lm and a net increase in the density of the family of lines at depths
exceeding 60 lm:
• Z1 B 60 lm, related to the tension effect of elastic fibers network
• Z2 [ 60 lm, related to Langer’s lines
Analysis of the morphology of skin tension lines of Caucasian women aged
between 20 and 80 years, shows a significant decrease of the density of lines of
depth less than 60 lm and an augmentation of deep lines beyond 60 lm. This
important result, presented in Fig. 9.43, shows the mechanical role of elastic fibres
in maintaining skin tension and the firmness of young skin. This elastic relaxation
mechanism, of the reduction of elastic fibres according to age, is the basis of the
phenomenon of the appearance of wrinkles.
Fig. 9.43 Evolution of the scale lines of tension during aging

Fig. 9.44 Aging effect on

the dynamic rotation of skin
line network of 120
Caucasian women
9.9.4 Dynamic Rotation of Skin Line Network During Aging
To monitor the rotational dynamics of skin lines during aging, the overall results
for the densities of line orientation from 20 to 80 years were collected for a
comprehensive representation of the dynamic change of direction in function of
age. This change is represented for those aged under 60 and over 60 years. Fig-
ure 9.44 shows the rotational dynamics and the significant decrease in tension lines
in the directions between 90 and 180, and the establishment of a marked
anisotropy between the directions 20 and 60. This result demonstrates the
relationship between the voltage loss of elastic fibres, and lower voltage lines
printed on the plates of the relief, and depths less than 60 lm.
The increase of the density of lines deeper than 60 lm, and the diminution of
the density of lines \60 lm, results in the increase of the anisotropy index sig-
nificantly during aging (see Fig. 9.45).
Fig. 9.45 Aging effect on

the skin line anisotropy index
of 120 Caucasian women
9.9.4.1 3D Reconstruction of Skin Line Network: Volumetric

Anisotropy and Tree of Skin Tension Network
The dermis is the layer of living skin. It is a supporting connective tissue rich in
fibre which gives the skin elasticity and strength. The dermis contains the
appendices of the skin. Histologically, the dermis can be divided into two layers:
the papillary and reticular layer. The dermis and epidermis are closely meshed into
each other through many outgrowths of the surface ripples of the dermis called
papillae. The papillary dermis contains many nerve endings (thermo-receptors,
tactile receptors). The reticular dermis consists of a network of collagen bundles
(thick, wavy, perpendicular to the basal membrane) more visible because they are
more dense within the reticular dermis. The network of elastic fibres, which
underlies the undulations of collagen fibre bundles, is anchored to their concavity.
The reticular dermis is the strongest part of the dermis. Its mobility results from the
unfolding of the undulating collagen fibre bundles, thus allowing their extension
and their return to their original position by the action of elastic fibres.
This structuring of the dermis with a network of collagen and elastic fibres gives a
3D structure to the skin network with directions perpendicular and parallel to the
skin surface. This 3D morphology prints on the surface of the stratum corneum a 3D
network of lines, which express the mechanical tension of the skin at rest.
The approach developed specifically for skin morphology allows the identifi-
cation of the lines network anisotropy at different scales of depth and orientation.
For each plane at a certain depth of the skin surface, three parameters of the point
belonging to the line of tension are determined: the density of depth z, the width of
the line and the rose of directions between 0 and 180 (Zahouani and Vargiolu
1998a, 2000; Zahouani 2006, 2007). Figure 9.46a represents the identification of
the network of skin tension lines in different directions for a 25 year old
(approximate age). The depth of the skin lines is illustrated by the range of colours
from blue to red. The family of skin lines printed on the plates can be identified:
secondary lines (colours of green, yellow and red correspond to a variation of
depths between 17 and 50 lm). The family of skin tension lines in the main colour
blue is in a scale between 17 and 84 lm. The identification of the orientation of
skin lines in different directions is illustrated in Fig. 9.46b, c, d and e. Figure 9.47
illustrates the anisotropy distribution against the depth of skin network families.
9.9.5 Morphological Tree of Skin Line Network
Appropriate and quantitative representation of the skin line network has been
developed. It allows all network lines to be reconstructed as a morphological tree.
Each trunk of the tree represents the density of lines in a given direction and for a
given depth (see Fig. 9.48).
One family of lines is described as a branch of the tree, in relation to its depth
and direction. With this method, it is possible for the first time to quantify the
Fig. 9.46 Network of skin tension lines
Fig. 9.47 Reconstruction volumetric skin tension lines anisotropy

Fig. 9.48 3D reconstruction: morphological tree of skin tension lines network
different scales of skin line accurately and to follow the morphological changes of
the surface in relation to age. This multi-morphological decomposition of line
network can be used to assess mechanical tension of elastic fibres and collagen
bundles during ageing. Also, this approach can be used as a preventive test for
certain diseases of the elasticity of the skin. The examples in Fig. 9.49 show the
use of this approach in the analysis of the transformation of the 3D line network
after a cosmetic application.
Fig. 9.49 Cosmetic effect on the tree of skin tension lines network
References
Abbot EJ, Firestone FA (1933) Specifying surface quantity-a method based on accurate
measurements and comparison. Mech Eng 55:569–572
Bhushan B, Majumdar A (1992) Elastic-plastic contact model of bifractal surfaces. Wear
153:53–64
Feder J (1988) Fractals. Plenum Press, New York
Greenwood JA, Williamson JBP (1966) Contact of nominally flat surfaces. Proc R Soc Lond A
295:249–267
Hertz H (1881) On the elastic contact of elastic solids. J Reine Angew Math 92:156–171
Johnson KL (1985) Contact mechanics. Cambridge University Press, Cambridge, London, New
York, New Rochelle
Jourani A, Delalleau A, Dursapt M, Sidoroff F, Zahouani H (2005a) Effect of local slopes of
roughness during contact between solids. Revue Européenne des éléments finis 14:271–286
Jourani A, Dursapt M, Hamdi H, Rech J, Zahouani H (2005b) Effect of the belt grinding on the
surface texture: modeling of the contact and abrasive wear. Wear 259:1137–1143
Langer K (1978) On the anatomy and physiology of the skin. 1: the cleavability of the cutis
(trans: Gibson T). Br J Plast Sur 31:3–8
Majumdar A, Bhushan B (1991) Fractal model of elastic-plastic contact between rough surfaces.
ASME J Trib 113:1–11
Mezghani S, Demirci I, Zahouani H, El Mansori M (2012) The effect of groove texture patterns
on piston-ring pack friction. Precis Eng 36:210–217
Peucker K, Douglas D (1975) Detection of surface specific points by local parallel processing of
discrete terrain elevation data. Computer Graphics and Image Processing 4:375–387
Popov L (2010) Contact mechanics and friction. Springer
Pullen J, Williamson JBP (1972) On the plastic contact of rough surfaces. Proc R Soc Lond A
327:159–173
Sabri L, Mezghani S, El Mansori M, Zahouani H (2011) Multiscale study of finish-honing process
in mass production of cylinder liner. Wear 271:509–515
Sirghi L, Rossi F (2006) Adhesion and elasticity in nanoscale indentation. Appl Phys Lett
89:2431181–2431183
Tabor D (1951) The hardness of metals. Oxford University Press, Oxford
Tsukioze T (1968) On the mechanism of contact between metal surfaces. ASME J Lub Tech
F90:81–89
Vallet C, Lasseux D, Sainsot P, Zahouani H (2009a) Real versus synthesized fractal surfaces:
contact mechanics and transport properties. Tribol Int 42:250–259
Vallet C, Lasseux D, Zahouani H, Sainsot P (2009b) Sampling effect on contact and transport
properties between fractal surfaces. Tribol Int 42:1132–1145
Zahouani H (1995a) Filtrage tridimensionnel des surfaces rugueuses, Bulletin de la Société des
Sciences et des lettres de LODZ. Recherche sur les déformations 20:131–163
Zahouani H (2006) Skin tension lines network during ageing. Serup J, Jemec GBE, Grove GL
(eds) Handbook of non-invasive methods and the sin, 2nd edn. CRC Taylor & Francis
Zahouani H (2007) Tension des fibres élastiques et lignes cutanées: application au vieillissement
et à la cicatrisation. Lévy JL (ed) Les secrets de l’anti-âge. Groupe Liaisons SA
Zahouani H, Vargiolu R (1998a) 3D morphological tree representation of the skin relief. A new
approach of skin imaging characterization. 20th congress international federation of the
societies of cosmetic chemists, Cannes, France, Sep, pp 69–80
Zahouani H, Vargiolu R (2000) Mesures du relief cutané et des rides. Collection explorations
fonctionnelles humaines. Physiologie de la Peau et Explorations Fonctionnelles Cutanées.
Sous la direction du Professeur Pierre Agache
Zahouani H, Mathia T, Rousseau J (1994) Morphology of engineering surfaces in deformation
mechanism. In: Proceedings of 6th Nordic Symposium on Tribology, pp 369–379
Zahouani H, Vargiolu R, Loubet JL (1998b) Fractal models of surface topography and contact
mechanics. Math Comput Model, Contact Mech 28:517–534
Zahouani H, Mezghani S, Pailler-mattei C, El mansori M (2009) Effect of roughness scale on
contact stiffness between solids. Wear 266:589–591
Zahouani H, Assoul M, Vargiolu R, Mathia T (2001) The morphological tree transform of surface
motifs. Int J Mach Tools Manufac 41:1961–1979
Zahouani H, Ben Tkaya M, Mezghani S, Pailler-Mattéi C (2011) Adhesive contact in the context
of multi-asperity interaction. Comptes Rendus Mécanique 339:502–517
Zahouani H, Kapsa P, Vargiolu R, Loubet J L, Mathia T (1997) Incidence de la morphologie
locale et effet d’échelle, pp 57–69
Zahouani H, Mezghani S, Vargiolu R, Hoc T, EL Mansori M (2013) Effect of roughness on
vibration of human finger during a friction test. In Press Wear, Available online 5 December
2012
Zhao Y, Maietta DM, Chang L (2000) An asperity microcontact model incorporating the
transition from elastic deformation to fully plastic flow. J Tribol ASME 122:86–93
Chapter 10
Correlation of Areal Surface Texture
Parameters to Solar Cell Efficiency
Erik Novak and Nelson Blewett
Abstract With increasing competition in the photovoltaic industry, quality

control to maximise efficiency is critical. Surface texture is broadly known to
qualitatively affect solar cell efficiency. However, quantitative correlation between
specific measures of surface texture and efficiency has only recently been dem-
onstrated, through use of large-area, three dimensional measurements and asso-
ciated areal field parameters. This chapter discusses how certain surface texture
parameters have been shown to linearly correlate with solar cell efficiency in
different manufacturing processes. A three-dimensional, non-contact microscope,
which uses specialised objectives to obtain areal surface information over a large
field of view in seconds, was used to measure the surface texture of a variety of
solar cells from different processes. Areal surface texture parameters, defined in
ISO 25178 part 2, were compared with the relative efficiencies of various cells.
The key parameters that linearly correlate to efficiency were identified. The ability
to rapidly quantify surface texture in a way that correlates to cell efficiency allows
process optimisation and control, lowering the cost per watt of photovoltaic cells.
10.1 Background
With increasing competition in the photovoltaic industry, quality control to

maximise efficiency is critical. Certain surface texture parameters have been
shown to linearly correlate with solar cell efficiency in different manufacturing
processes. Thus, monitoring surface texture to maintain process parameters
increases the quality of the solar cells produced. This decreases the price per watt,
because fewer bad cells are created, preventing wasted time and materials.
E. Novak (&) N. Blewett

Bruker Nano Surfaces Division, 3400 E. Britannia Rd., Suite 150,
Tucson, AZ 85706, USA
e-mail: Erik.Novak@bruker-nano.com

270 E. Novak and N. Blewett
Coherence scanning interferometers (CSIs) (Leach 2011) acquire three

dimensional (3D) information over a relatively large field of view in seconds
without contacting or otherwise damaging the sample. Although the data is taken
quickly, CSIs can have nanometre-level accuracy and repeatability, making them
ideal for production monitoring.
Lowering the surface reflectance of silicon wafers by etching or some form of
deterministic texturing is critical to maximise the efficiency of silicon solar cells.
There are different theories on which type of texturing helps to achieve a higher
level of efficiency and how best to monitor texture (Lee et al. 2009; Montesdeoca-
Santana et al. 2008; Green et al. 2007; Saito and Kosuge 2007). Some organisa-
tions simply visually examine cells for reflectance, or use a basic reflectance
monitor. Other organisations use manual peak counting of key features, such as
etched pits or pyramids. Still others use the average roughness, or Ra parameter,
from profile stylus instrumentation, the traditional method of quantifying texture
(Leach 2009; Whitehouse 2010). However, such methods have their limitations,
and even when properly executed, have been found to only weakly correlate with
solar cell efficiency. More comprehensive, areal surface texture parameters have
been found to directly correlate surface texture with efficiency; these parameters
are essentially areal extensions of profile roughness parameters, for example Ra
and Rq, and have recently been standardised (ISO 25178 part 2 2012).
10.2 Description of Samples
The first measurements performed as part of this study on solar cells involved
measuring several cells whose efficiency was quoted by the manufacturer. The
goal was to determine whether surface texture parameters could be used to dif-
ferentiate between the good and bad cells. The cells were monocrystalline solar
cells on wafers of 100 mm diameter. After the success of this first experiment,
more comprehensive studies were performed on a broader set of samples.
Due to the proprietary nature of many of the solar cell manufacturing processes,
and reluctance to publicly share results, subsequent trials were first conducted on
several sets of commercially procured samples. Two sets of monocrystalline and
one set of polycrystalline solar cell samples were procured from different vendors
with no knowledge of the underlying manufacturing steps. These cells were not
measured prior to being procured, and any correlation of surface texture param-
eters to efficiency was not known in advance.
Once the results shown in this chapter were established, multiple studies were
conducted directly with various manufacturers of mono- and polycrystalline solar
cells. While permission to share details of those studies could not be obtained for
this chapter, in each case at least one areal parameter was found to correlate to the
solar cell efficiency to an 85 % or higher R2 level. The surface texture was found to
be a more dominant indicator of solar cell efficiency than the trace properties, or
other electrical properties of the cells.
10 Correlation of Areal Surface Texture Parameters to Solar Cell Efficiency 271
10.3 Description of Experimental Setup
In order to examine the correlation of solar cell efficiency to surface texture, an

easy, rapid method of determining the efficiency of each cell was required. Bruker
wanted a simple, low-cost, repeatable setup for examining the relative efficiency of
one cell to another. This setup would allow the experimentation to be recreated and
verified in any of our worldwide applications laboratories, or customer sites, to
recreate the results with minimal effort on new, unknown samples.
The basic setup is shown in Fig. 10.1. A halogen lamp was firmly mounted to
an optical table, facing downward, 300 mm above the table surface. The solar cells
were placed on two raised bars (anodised aluminum bars with Teflon tape used to
prevent accidental electrical contact) to provide easy electrical access to the
conductive undersides for those cells (where this was necessary). Simple screws
were used as mounting points, capable of repeatably mounting both square and
circular cells. A Fluke multimeter with alligator clips for attachment to the solar
cells was used to measure the potential difference across the devices to four
significant digits.
While this setup did not use a calibrated source and, therefore, could not pro-
vide absolute solar cell efficiency, the goal was to examine relative efficiency
among a family of cells, to evaluate which parameters affected efficiency within
that process. The setup was improved several times in terms of ease of loading and
stability of results until a trial of ten remove/replace operations for several cells
Fig. 10.1 Setup for

measuring the relative
efficiency of the solar cells.
Key components are the light
source, hard stops for locating
the solar cells, the digital
multimeter and electrical
leads
yielded variations of less than 1 % peak to valley on the measured potential

differences.
For the surface texture measurements, a Bruker ContourGT-K CSI instrument
was used. An example of such a CSI is shown in Fig. 10.2. The CSI has vertical
resolution independent of the microscope objective employed; at any field of view
the vertical resolution remains unchanged. For these tests, the vertical scanning
interferometry (VSI) measurement mode was used, as it provides capability for
surfaces with differing reflectance and roughness values. Measurement speed was
set to its lowest level (approximately 8 lm s-1) to maximise the vertical resolu-
tion. The CSI scans the objective through focus to build up a vertical height map of
the sample; the scan length used was 20 lm.
The CSI employs various objectives and can, therefore, have both variable field
of view and lateral resolution. It is generally desirable to maximise areal coverage
of a sample to increase statistical significance. Thus, measurements were first
performed with a 5X magnification objective, which provided a field of view of
2.3 mm by 1.7 mm. This relatively large field of view enabled many of the local
features to be imaged simultaneously and was ultimately found to produce highly
correlated results. In addition, a 50X objective with 0.23 mm by 0.17 mm field of
view was employed to test whether there was correlation in roughness parameters
at a more local scale; results for this magnification indicated no reasonable cor-
relation to photovoltaic efficiency.
While achieving high quality surface topography data is a relatively simple task
with the CSI, the critical aspect of the exercise was to discover any relationships
between surface structure and solar cell efficiency. The system can automatically
store to a database thousands of different parameters, but the focus was on methods
already in use in industry as well as the ISO S-parameters, which have been
recently adopted (see Chap. 2). Thus, Ra and Sa of the centre trace were stored.
Fig. 10.2 A benchtop

Bruker ContourGT CSI.
A turret holds multiple
microscope objectives.
Samples are placed on the
150 mm automated stage for
measurement
Also, for those processes producing a non-random surface (such as periodic

dishing or pyramids), the number of such features was stored; since this count is
used in some production lines. In addition, the remaining areal S-parameters were
also stored, to determine which, if any, would correlate to solar cell performance.
10.4 Measurements and Analysis
The measurements consisted of two parts: the determination of cell output and the
surface characterisation. The solar cell output was logged under the controlled
conditions described previously. Even though the result was not an absolute effi-
ciency, these results were normalised against an area, so were reported in units of
micro-watts per square millimetre, to allow for potential comparison between cells
of different sizes under the same lighting conditions.
Since a direct correlation between a given surface texture parameter and solar
cell performance had not been published prior to Bruker’s initial studies, a large
number of parameters, including all of the areal S-parameters, were recorded; as
analysis time is minimal, the goal was to capture every probable metric and then
determine which correlated to performance. With the computing and storage
available to most metrology instruments, such a technique is increasingly common
in industry; log everything and then examine correlation of any or all of the logged
parameters to a wanted or unwanted trait in the components being produced.
As discussed above, the first experiment involved several known samples from
a solar cell manufacturer who wished to determine differentiating parameters.
Cells could be visually differentiated but the Ra value of both cells was equivalent
at approximately 14 nm. The manufacturer wished to determine if there was a
quantifiable difference between the cells that could be used to control the pro-
duction process. Ten locations were measured across each cell using a low mag-
nification (5X) objective, which provided approximately a 2 mm by 2 mm field of
view (Fig. 10.3 and Table 10.1).
Fig. 10.3 Areal surface maps of two solar cells. The cell on the left is 13 % more efficient than
the cell on the right
Table. 10.1 Output power and some areal surface texture parameters.
Output Power Avg Std Avg Std Avg Std
(W cm2) Sdr (%) Sdr (%) Sbi (%) Sbi (%) Sdq (°) Sdq (°)
Cell 1 10.78 0.07 0.01 0.92 0.08 2.13 0.15
Cell 2 9.39 0.82 0.72 1.33 0.67 7.10 4.08
Ratio 87 % 12 72 1.4 8.4 3.3 27
Avg = average, Std = standard deviation, Sdr is the developed interfacial area ratio (Sect. 2.6.2),
Sbi is a surface bearing index (Sect. 2.9.1, an older parameter that does not feature in ISO 25178
part 2) and Sdq is the root mean square gradient (Sect. 2.6.1)
The manufacturer wished to determine whether certain parameters showed

large differences between the cells, although the efficiencies were only 13 %
different. Parameters that show a larger ratio between the average, are more
effective at differentiating between the cells. Also, the standard deviations of the
parameters were studied, and it was seen that the more efficient cell also had
greater stability of these parameters across the ten measurement locations. Ulti-
mately the Sdr parameter was determined to best differentiate between good and
bad cells; the average Sdr value differs most between the good and bad cells, and
Sdr value of the less efficient cell varied greatly across the cell. This was verified
through measurement of a large number of cells with varying efficiencies.
After the initial success, the next solar cells studied were monocrystalline with
surface morphologies characterised by large dished regions distributed across the
cell surface. Four of the ten cells examined are shown in Fig. 10.4 in a false height
map; blue represents low regions, red high regions, and green is the mean of the
surface. The vertical scale (red to blue) represents 3 lm as seen on the accom-
panying scale. Each of the images in Fig. 10.4 came from a different cell from the
same manufacturer using the same basic process. A significant difference in the
size and distribution of the regions can be seen. The cells in Fig. 10.4 are arranged
in order of decreasing efficiency.
Neither counting the number of features nor measuring Ra of the centerline
showed any correlation to cell efficiency—both produced fits with R2 values of less
than 0.3. Upon further examination it was seen that Ra could vary by more than a
factor of three depending on where exactly the 10 mm trace was located across the
sample.
The Sa values also showed poor correlation to cell efficiency. Although the R2
value was 0.64, which is generally indicative of something more than random
scatter, this is mainly due to two of the ten points being large outliers from the bulk
of the group, therefore producing a somewhat linear correlation due to the large
deviations. The results are shown in Fig. 10.5.
Examining the other S-parameters, Ssk was found to correlate very well with the
photovoltaic efficiency of the cells, while no other parameter exhibited such high
correlations. The Ssk parameter is a measure of surface skewness, and involves
examining the histogram of the heights of all measured points, and calculating the
symmetry and deviation from an ideal normal distribution (see Sect. 2.4.2).
Fig. 10.4 Surface texture across four of the monocrystalline cells initially studied
Fig. 10.5 Efficiency against

Sa value, showing poor
correlation
Figure 10.6 is a graph of efficiency against Ssk value for the first set of
monocrystalline cells. The graph is highly linear with points evenly spaced about
the curve. The efficiency measurements were both repeated, looking at a different
subsection of the surface on the second trial, and correlation was equally strong,
confirming the first set of results.
The correlation of Ssk value to cell efficiency is not particularly surprising. A
positive Ssk value indicates a predominance of peaks on a surface, while a negative
Ssk value indicates a majority of valleys. A good solar cell is one that traps
incoming light, and thus it would be expected that if a surface is comprised of
more valleys, there is more probability that a photon will enter the cell and be
reflected from both sides of the valley and become absorbed, leading to higher
Fig. 10.6 Efficiency against

Ssk value for an
approximately 2 mm by
2 mm field of view
efficiency. This is borne out by the data, where the greater the relative number of
valleys, the more efficient the solar cell.
Following the initial positive results, the cells were again measured, but this
time with a 50X magnification objective, such as might be used in a confocal
microscope to achieve acceptable vertical resolution for surface texture mea-
surement. The field of view in this case was approximately 200 lm on a side, and
a typical surface image is shown in Fig. 10.7.
Using the 50X objective, only a handful of features are seen in a given area.
When examining the correlation between surface texture parameters and cell
efficiency, no correlation was seen either via linear or lower order polynomial fits.
Figure 10.8 shows the general randomness of the results.
Thus, it is important to understand the spatial scale of the dominant surface
structures and to ensure appropriate areal surface coverage to obtain statistically
Fig. 10.7 Typical high-magnification measurement of solar cell from Fig. 10.4 showing only a
few of the depressions typical of the surfaces
Fig. 10.8 Cell efficiency

against Ssk value when using
a 50X field of view
significant results. There is no hard and fast rule for what coverage is required. One
good measure is the consistency of the results when moving to various positions on
a surface. Statistical significance can be assessed mathematically in a variety of
ways, but generally results should be stable to within a few percent to be able to
draw meaningful conclusions.
Different etching processes produce significantly different surface morpholo-
gies. Thus, the correlation discussed above would not necessarily apply to a dif-
ferent solar cell manufacturing process. As such, a different set of monocrystalline
cells, with very different surface structures, was procured. The surface texture of
the new cells was much more randomly distributed, with no large-scale features as
observed in the previous set of cells (see Fig. 10.9).
Surface texture measurements were again taken with a large field of view and
all S-parameters were logged to a database. The results are shown in Fig. 10.10.
Fig. 10.9 Monocrystalline cell with a random distribution of surface heights

Fig. 10.10 Efficiency

against Ssk for a second set of
monocrystalline solar cells
Again, a good linear correlation between the cell performance and Ssk is seen
across a broad range of cell performance.
The next cells examined were polycrystalline cells. Polycrystalline cells exhibit
a variety of morphologies across the cell, depending on the local crystal orienta-
tions. Figure 10.11 shows three approximately 1 mm square regions on the cells
with the same false colour height scale range of ±8 lm. Each region has a sig-
nificantly different range of heights and localised surface structure. As such, to
study this cell, ten measurements were taken on each cell at random locations, with
all results averaged, in order to get a better representation of the overall cell
structure. As each single measurement took only 5 s, the total measurement time
per cell was still only about 1 min, so throughput is still sufficient for most
Fig. 10.11 Three areas from a single polycrystalline solar cell used to examine the correlation of
surface texture parameters to cell performance
Fig. 10.12 Ssk and Sbi for the polycrystalline solar cells studied. The Ssk parameter shows no
correlation, while Sbi shows strong correlation to cell performance
statistical process control encountered in the solar industry. The averaged surface
parameters were then again studied for correlation with cell efficiency.
Figure 10.12 shows the results for this set of solar cells. The Ssk parameter was
found to have essentially no correlation to solar cell performance, even when
considering the average across various locations. However, Sbi, the surface
bearing index (see Sect. 2.9.1), did have a fairly strong correlation to performance,
with an R2 value greater than 0.90. The Sbi parameter is a measure, relative to Sq,
of the surface height at the 5 % bearing area ratio. The Sbi parameter typically
ranges from 0 to 3. A greater Sbi value indicates a greater relative load bearing
area.
10.5 Conclusions
Maximise solar cell output and maintaining a consistent process are critical to the
highly competitive solar industry. Gaining quantitative measures that enable high
output and consistent results leads directly to increased yield, and margin capa-
bility on products. Traditional methods of controlling surface texture via two-
dimensional stylus traces, or visual inspection methods, are qualitative and do not
reliably predict cell performance, potentially leading to acceptance of poor quality

products, and rejection of high quality products. Areal surface topography mea-
suring instruments, such as the coherence scanning interferometer used in these
studies, provide rapid, accurate surface measurements over relatively large areas of
solar cells.
Areal parameters were found to correlate strongly to solar cell performance. For
two sets of monocrystalline solar cells, Ssk, the surface skewness, correlated very
well to cell efficiency. An additional study was made on polycrystalline cells,
where the surface statistics for many areas were averaged to gain representative
numbers for each cell. In this case, the Ssk parameter was not found to correlate to
cell performance, but the Sbi parameter was found to substantially correlate to
relative cell efficiency.
After these initial promising results, various photovoltaic manufacturers were
approached for further validation. Over the course of approximately one year,
multiple manufacturers worked with Bruker to validate areal parameter correlation
to cell efficiency on a wide variety of silicon solar cells. In all cases at least one S-
parameter was found to strongly correlate to cell performance. Certain manufac-
turers found even stronger correlation when using a combination of multiple S-
parameters as the figure of merit. In all cases, the photovoltaic manufacturers had
not previously had a quantified way to relate the surface texturing process with
performance, but rather only had broad checks to determine parts which were
wildly out of normal tolerances. The additional information provided by large field
of view measurements from an areal surface topography measuring instrument was
enough to reliably and quantitatively relate surface texture to performance.
References
Green MA (2007) Thin-film solar cells: review of materials, technologies and commercial status.
Mater Sci: Mater Electron 18:S15–S19
Standardization
Lee J, Lakshminarayan N, Dhungel SK, Kim K, Yi J (2009) Optimization of fabrication process
of high-efficiency and low-cost crystalline silicon solar cell for industrial applications. Sol
Energy Mater Sol Cells 93:256–261
Montesdeoca-Santana A, González-Díaz B, Jiménez-Rodríguez E, Ziegler J, Velázquez JJ,
Hohage S, Borchert D, Guerrero-Lemus R (2008) Influence of stain etching on low minority
carrier lifetime areas of multicrystalline silicon for solar cells. Mater Sci Eng B 176:541–1545
Saito Y, Kosuge T (2007) Honeycomb-textured structures on crystalline silicon surfaces for solar
cells by spontaneous dry etching with chlorine trifluoride gas. Sol Energy Mater Sol Cells
91:1800–1804
Whitehouse DJ (2010) Handbook of surface and nanometrology. CRC Press, Abingdon
Chapter 11
Characterisation of Cylinder Liner
Honing Textures for Production Control
Zlate Dimkovski, Cecilia Anderberg, Robert Ohlsson

and B.-G. Rosén
Abstract It is of common interest to reduce oil consumption and frictional losses

in internal combustion engines, which are heavily influenced by the quality of the
cylinder liner surface. The plateau cross-hatch topography of a cylinder liner
consists of a system of grooves of different density, width and depth, some parts
covered by folded metal, and some parts totally interrupted and unbalanced as a
result of imperfection in the honing process. These grooves are critical for good
liner function, and need to be quickly and objectively quantified for an efficient
surface finish development. A suitable way to do this is to use 3D interference
measurements and to combine profile and image analysis. Thus, the features/
parameters, such as honing angle, balance of honing texture, groove interrupts,
width, height, and distance between grooves, are successively quantified. Here,
these parameters, along with areal surface texture parameters in the published ISO
specification standard were used in two case studies. The first case study is on the
effect of the folded metal on the surfaces of run truck liners and the second is an
evaluation of the improvements of the surface quality introduced by the diamond
honing in production of car liners. In addition, based on the significant parameters
of the surface, a general characterisation tool for qualifying the surface quality and
determination of the required number of measurements is presented.
Z. Dimkovski C. Anderberg B.-G. Rosén (&)

Functional Surfaces Research Group, School of Business and Engineering Halmstad
University, P.O. Box 823 SE-301 18 Halmstad, Sweden
e-mail: bg.rosen@set.hh.se
C. Anderberg
Volvo Cars Corp. Base Engine Department Gothenburg, Halmstad, Sweden
R. Ohlsson
Volvo Power Train Corp, Volvo Group Gothenburg, Halmstad, Sweden

282 Z. Dimkovski et al.
List of symbols
a groove width
Bal groove balance
Blech Blechmantel parameter
c groove height
C matrix of groove widths and heights
d distance between grooves
Gcov groove coverage parameter
Holes holes parameter
Intrp interrupt parameter
MVI mean volume of islands
n number of grooves
Stray percentage of stray grooves
a groove angle
b main angle
k reciprocal mean groove separation
l mean
m groove order
h angle between normal and x-axis
q distance to origin
r root-mean-square groove angle difference
11.1 Introduction
Cylinder liners are an important class of technical components, produced in large

numbers for internal combustion engines of all sizes. The specification and control
of their surface topography is, therefore, an important manufacturing requirement
(Pawlus et al. 2009). In the past, liner surfaces were produced in a single
machining process, and the resulting high peaks had to be slowly and carefully
worn down in service, by the process known as ‘‘running-in’’, in order to produce
an acceptable load-bearing surface (Keller et al. 2009). Nowadays run-in liner
surfaces are ready-created in the factory by a three-stage honing process (Sabri
et al. 2011). The first stage sets the form of the bore, while the second stage, often
called coarse honing (Grabon et al. 2010), produces a rough surface with coarse
peaks and valleys as a result of the abrasive honing grits ploughing the liner
material (Michalski and Wos 2011). Finally the third step, plateau honing with
abrasive grits of smaller size, replaces the coarse honing peaks with a much finer
texture but leaves the valleys unaltered.
Since the control of the cylinder bore surface topography needs to be improved,
there is a need for improvement and facilitation of its characterisation (Anderberg
et al. 2009; Kanthababu et al. 2009). To date, the most common practice used, to
11 Characterisation of Cylinder Liner Honing Textures 283
fully inspect the quality of the bore topography, is to measure surface profiles, to
check whether the profile surface texture parameters are within tolerance, and
scanning electron microscope (SEM) images are used to gain lateral information.
Groove width, height, distance between grooves, material ratio at a certain depth
and macro-waviness, evaluated from profiles, and honing angle, orientation of
grooves, plateau formation and groove appearance (i.e. groove interrupt and
smeared metal-‘‘Blechmantel’’—see Sect. 11.2.1), evaluated from SEM images,
have been suggested as process control parameters (Goetze 1988) and used by
experts for many years. Based on these parameters, rating factors are determined,
weighted and an overall grade is obtained for the surface. The most inferior
measured grade dictates the quality of the surface. As the parameters are evaluated
manually by an expert, grade determination is a subjective, tedious and time
consuming process. In order to make grade determination more objective, Beyerer
et al. (2001) used image analysis (fast Fourier and Radon transforms with SEM
images) to extract the groove widths, heights, distance between grooves, and
groove angles and unify them in a ‘‘Balance of the honing texture’’ parameter.
Further improvements were carried out by Anderberg et al. (2006) by using two
different techniques. In one technique, the SEM images were used for extraction of
Blechmantel (see Sect. 11.2.1), irregularities, holes, balance, interrupt, orientation,
stray grooves and honing angle parameters by employing edge detection, fast
Fourier and Hough transform techniques. In the other technique, surface profiles
were used for automatic extraction of the groove width, height, distance between
grooves, and other standard profile parameters.
As SEM measurements can be time consuming and are not feasible for an
automatic inspection, there is a need of extraction and computation of the nec-
essary parameters from areal measurements. The fact that the surfaces are 3D in
nature and the potential high-speed data acquisition of optical measuring instru-
ments, which enables automated inspection, has led to a trend for areal cylinder
bore characterisation. A further characterisation issue is to determine the number
of measurements required, which depends on the variation of the roughness over
the measured surface and the differences between liner samples.
The aim of the work reported in this chapter, is to automate the inspection of the
liner surface topography using areal interference measurements by complementing
the standard areal parameters (i.e. those described in Chap. 2) with honing angle,
balance of honing texture, groove interrupts, width, height, and distance between
grooves. Two case studies are presented to demonstrate the application of the
characterisation methods. The first is a case study on the effect of the folded metal-
Blechmantel partly covering the deep honing grooves on the surfaces of run truck
liners. The second case study is an evaluation of the improvements of the surface
quality introduced by diamond honing in production of car liners. In addition, the
second case study introduces a rating method applicable, not only for liner sur-
faces, but for any type of surface.
11.2 Methods
In order to improve the characterisation, two methods are suggested: a method for
extraction of the groove parameters specific for the honing texture of liner surfaces
and a method for rating of the surfaces, based on the selected parameters important
for good liner function.
11.2.1 Method for Extraction of the Groove Parameters
The groove parameters: honing angle, groove width, height, distance between
grooves, groove balance, Blechmantel, groove interrupt and stray grooves are
shown schematically in Fig. 11.1. If the uncut or folded material inside the honing
groove partly covers or blocks it, it is referred to as Blechmantel, while the
interrupt parameter describes a total interruption in the course of the groove. The
balance parameter reflects how much the honing texture is balanced/homogeneous
and is a combination of the groove width, height, distance between grooves and
honing angle, which are shown in Fig. 11.1.
As the honing grooves have an almost linear form, first, the lines need to be
detected. For that, the Hough transform (Hough 1962) has been proven to be
efficient in image analysis. The Hough transform, is a point-to-curve transfor-
mation utilising the normal representation of a line
q ¼ x cos h þ y sin h ð11:1Þ
where q is the perpendicular distance of the line to the origin and h is the angle
between a normal to the line and the positive x axis. To use the Hough transform in
this form, each point (xi,yi) is mapped to all points in a q - h parameter space that
Fig. 11.1 Overview of the groove parameters

Fig. 11.2 Left: three co-linear points in x–y space; right: their intersection in q - h space
specifies a possible line through the point. This set will form a sinusoid, and points
on a particular line (see the line—l Fig. 11.2 left) will all map to sinusoids that
intersect at a common point—l, (Fig. 11.2 right), whose coordinates give the
parameters of the line.
The transform is implemented by quantising the Hough q - h parameter space
into finite intervals or accumulator cells. As the algorithm runs, each (xi, yi) is
transformed into a discretised (q,h) curve and the accumulator cells which lie
along this curve are incremented as in a voting procedure. Resulting peaks in the
accumulator array represent strong evidence that a corresponding straight line
exists in the image.
11.2.1.1 Blechmantel
To be able to isolate and detect lines from areal measurements, the image analysis
needs to be combined with the profile analysis. It is also convenient to follow the
flowchart (Fig. 11.5) of the algorithm used for extracting the Blechmantel parameter-
Blech. Firstly, each axial profile is extracted from a roughness surface (Fig. 11.3
top) and the valleys reaching and crossing a certain ‘‘lower’’ threshold (the lower
solid line at -1 lm in Fig. 11.3 bottom) are identified. From these valleys, their
bottom points are found as minima, while the left and right points are found by
tracking left and right respectively from the bottom points (Fig. 11.3 bottom) to the
first point over an ‘‘upper’’ threshold (the upper solid line at 0 lm in Fig. 11.3
bottom). This implies that the ‘‘upper’’ and ‘‘lower’’ thresholds, which depend on the
depth of the grooves/scratches in question, need to be found in advance by experi-
menting with the surfaces. The bottom points are mapped and a Hough transform is
used to find the central groove lines (Fig. 11.4 top). These bottom points are asso-
ciated with their respective grooves, and their left and right points for each profile.
Fig. 11.3 Top: a roughness surface from an areal interference measurement of a truck liner
sample; bottom: identification of the grooves’ left, right and bottom points from the last axial
profile of the above surface. Reprinted from Dimkovski et al. (2009), Copyright (2009), with
permission from Elsevier
The left and right groove lines, which determine the axial groove width, are identified
as the most frequent distances from the groove’s central line.
A correction is applied where two or three parallel lines are located very close
so that they are joined in one large groove. Referring to Fig. 11.4, the parallel
grooves very close to each other: 1–2, 3–4–5, 6–7, 8–9, 10–11 and 12–13, are
joined in six large grooves. After this, the surface is thresholded at the same depth
threshold and converted to binary form by setting ones for the ‘‘on’’ pixels above
and zeros for the ‘‘off’’ pixels below the threshold. The Blechmantel pixels are
found by tracking and checking if the count of the ‘‘on’’ pixels is less than the total
number of pixels within the left and right groove lines for each profile, and each
groove. The detected Blechmantel pixels can be seen as a transparent white colour
in Fig. 11.4 bottom. Finally, the Blechmantel coverage is quantified as a ratio
Fig. 11.4 Top: detection of the central groove lines based on the identified bottom points for
each profile. The parallel grooves very close to each other: 1–2, 3–4–5, 6–7, 8–9, 10–11 and
12–13 are joined in six large grooves; bottom: identification and quantification of Blechmantel,
marked with a transparent white colour. Reprinted from Dimkovski et al. (2009), Copyright
(2009), with permission from Elsevier
between the number of Blechmantel pixels and the total number of image pixels
multiplied by one hundred to express it as a percentage.
The relative amount of Blechmantel has been defined as a mean of the ratios
between the Blechmantel pixels and the number of pixels occupying the respective
grooves. This parameter definition may be useful when comparing measurements
of different sizes or measurements with large differences in the number of grooves.
11.2.1.2 Other Parameters
Interrupt
If the number of ‘‘on’’ pixels (see Fig. 11.5) is equal to the total number of
pixels within the left and right groove lines, those pixels will belong to a groove
interrupt. The count of all those pixels divided by the total number of image pixels,
and multiplied by one hundred defines the Interrupt parameter Intrp. A relative
amount of Interrupt can be defined in the same way as for the Blechmantel
parameter.
Groove Coverage
The number of pixels that do not belong to Blechmantel or Interrupt, divided by
the total image pixels, and multiplied by one hundred defines the Groove Coverage
parameter Gcov.
Holes
The number of ‘‘off’’ pixels that do not belong to the grooves divided by the
total number of image pixels, and multiplied by one hundred defines the Holes
parameter Holes.
Number of Grooves
The Number of Grooves NG is simply the counted number of grooves.
Honing Angle
The two main texture directions of the groove sets are found from the analysis
of the Hough transform. The grooves with these main directions are associated and
separated into two main groove sets: one groove set for grooves with angles
greater than 90, and one for grooves with angles less than 90. Then, the Honing
Angle is determined from the geometry of the lines of these two sets.
Stray Grooves
The grooves that do not belong to the groove sets of the two main directions are
counted, divided by the total number of grooves, and multiplied by one hundred.
This yields the percentage of Stray Grooves Stray.
Groove Width
The axial groove width is found by subtracting the x coordinates (axial-sliding
direction) of the left from the right groove lines for each groove (see Figs. 11.4 and
11.5). As the angle of each groove is known, the normal groove widths can be
calculated and averaged to give the Groove Width parameter value a.
Distance between Grooves
The axial distances between the grooves is found by subtracting the x coordi-
nates of the preceding grooves from the following central groove lines that belong
Fig. 11.5 Flowchart of the algorithm for calculation of the Blechmantel parameter Blech.
Reprinted from Dimkovski et al. (2009), Copyright (2009), with permission from Elsevier
to the respective groove set of the two main texture directions (see Figs. 11.4 and
11.5). As the angles of the groove sets of the two main texture directions are
known, the normal distances between the grooves can be calculated for the
respective groove set, and averaged to give the Distance between Grooves
parameter value d.
Groove Height
The Groove Height c is determined by finding the minimum of each profile
within the groove, and averaging these minima values later on.
Groove Balance
According to Beyerer et al. (2001), the Groove Balance Bal can be calculated
using

lC1 lC2 T 1 lC1 lC2
Bal ¼ 0:5 C1 þ C21
la1 la2 la1 la2

1 k1 k2
þ 0:5 trace C1 C2 þ C21 C1 2 þ þ 2 þ 0:5ðra1 þ ra2 Þ
k2 k1
ð11:2Þ
where
1X ni
1X ni
1
lCi ¼ civ ; lai ¼ aiv ; ki ¼ nP
;
ni v¼1 ni v¼1 i 1
1
ni 1 div
v¼1
ni T
1X ni
1 X civ lci civ lci
r2ai ¼ ðaiv bi Þ2 ; and Ci ¼
ni v¼1 ni v¼1
aiv lai aiv lai
whereby the heights civ , widths aiv , distances div , angles aiv , number of grooves ni
and main angles bi are calculated for each groove v in the respective groove set
i. This parameter has values greater than or equal to zero (Bal C 0), meaning that a
value approximately equal to zero indicates a very well balanced honing texture.
11.2.2 Method for Rating of the Surfaces
As part of the process to quickly and objectively characterise the different liner
samples measured, a rating method effective when the control limits are unknown
(estimations can be made from the results suggested by simulations that are
experimentally unconfirmed at the early stage of the given project) and known
(results experimentally confirmed) is proposed. This method was applied on the
areal parameters: Sq, Sk, Spk, Svk, Sds, Ssc, Sdq, Angle, a, c, d (width, height and
distance of the grooves or axial wear scratches), Blech, Intrp, Holes (pixels below
the threshold but not counted as Blechmantel), and number of grooves NG chosen
in the second case study, but the method can be applied for any parameters and any
surfaces. This method is an analogy of the method described in the Goetze Honing
Guide (1988).
Firstly, from the values of the chosen parameters of all the measurements of all
the liner samples to be rated, the probability density estimates are computed using
Fig. 11.6 Finding the unknown control limits and grading factors by specifying percentages of
acceptability
the kernel smoothing Matlab1 function ksdensity. For example, to estimate the Sk
density, forty-four Sk values were input into ksdensity for the rating of three
different surface types. Types 1 and 2 were examples of liners currently in use,
while type 3 was an example of the latest development of the improved slide
honing process. Sixteen measurements were taken on both type 1 and 3 and twelve
measurements on type 2 (forty-four in total). Then, this process was repeated for
the other parameters (Sq, Svk, Spk, etc.) selected in the rating. The construction of
the probability densities is not only restricted to the parameter values from the
sample measurements to be rated, but can also be constructed from reference
parameter values of larger numbers of data previously collected from high quality
surfaces (it is statistically preferable to have as much data as possible). In this case,
the comparison would be with respect to the high quality surfaces. Generally, there
are three areas under the parameters’ densities of interest when determining the
quality: lower area (the lower value the better quality, such as Sk, Spk, Ssc, Sdq,
Angle, Blech, and Intrp parameter, see Fig. 11.6 left), central area (Sq, Svk, a, d, c
and Holes, see Fig. 11.6 middle) and upper area (the higher value the better
quality, such as Sds and NG, see Fig. 11.6 right).
The valley depth and distance between grooves are preferred to be medium
because it is known that the debris are collected and oil is retained in the deep
grooves. These ‘‘deep’’ grooves are the only visible honing marks after the wear
process, which sooner or later will take place. However, it is not clear what
the optimal values of these parameters are. When the control limits are unknown,
the rating can be made by specifying the acceptable percentage of the area under the
parameter density of interest, for instance 70 % acceptable and 30 % unacceptable
(see Fig. 11.6). For the central area of interest, the percent of the greater unac-
ceptable area (70 % in Fig. 11.6 middle, that is 0.7 9 30 % on the right part) needs
to be specified additionally. When the control limits are known, the rating is made
1
MathWorks Inc., USA, www.mathworks.com.
by specifying: lower parameter value for parameters with a lower area of interest,
upper parameter value for parameters with an upper area of interest, and lower and
upper parameter values for parameters with a central area of interest. Once the
control limits are determined, the grading factors can be found by dividing the
enclosed areas under the parameter density within these limits in an equal number
of parts defined by the number of grades.
Here, six grades were adopted, from which 1, 2 and 3 are unacceptable and 4, 5
and 6 are acceptable, meaning that 1 is the worst and 6 is the best. For six grades
the area under the densities of the parameters with lower area of interest (see
Fig. 11.6 left) is divided into six parts. The first three parts form the acceptable
area with three acceptable grading factors (6, 5 and 4) and the second three parts
form the unacceptable area with three unacceptable factors (3, 2 and 1). Finding
the control limits for the parameters with a central area of interest is a little
different. First, the position of the maximum of the density is examined. If it is on
the lower part side, then the greater part of the unacceptable area (defined by the
greater percentage of unacceptable area additionally specified) is on the opposite
side and vice versa. If the position of the maximum of the density is in the middle,
the unacceptable area is divided into two equal parts, lower and upper part. The
lower control limit is then found by dividing the lower unacceptable area in three
parts giving three unacceptable factors (1, 2 and 3). The upper control limit is
found by dividing the acceptable area (defined by the percentage of acceptable
area specified) in five equal parts giving factors 4, 5, 6, 5 and 4 accordingly. The
grading factors for a parameter with an upper area of interest are found in an
opposite way to those with a lower area of interest. This rating method was
incorporated in an areal characterisation tool called ‘‘LinerSurf3D’’ (see
Fig. 11.7).
The rest of the rating method is just to follow the Goetze’s procedure as follows.
• For a parameter value of a measurement, a parameter grading factor is found
from the corresponding parameter density (Fig. 11.6).
• These parameter grading factors are multiplied by their ‘‘Weights’’ respectively,
summed and rounded up or down resulting in a ‘‘Measurement Grade’’ (see the
list box labelled as ‘‘Evaluation Table’’ in the right top corner of Fig. 11.7).
Here the ‘‘Weights’’ express the importance of the respective parameter.
• The final ‘‘Sample Grade’’ is its lowest ‘‘Measurement Grade’’ (see the list box
labelled as: ‘‘3D-Parameters’’ for the ‘‘Measurement Grades’’ and the list box
labelled as ‘‘Sample Parameters’’ for the final ‘‘Sample Grades’’ in Fig. 11.7).
In the same tool, the required number of measurements can be easily deter-
mined by using the t-distribution and selecting the confidence interval and toler-
ance limits (denoted as ‘‘Dispersion’’ in Fig. 11.7).
Fig. 11.7 Overview of the areal characterisation tool ‘‘LinerSurf3D’’ where the rating method is
incorporated. The list box labeled as ‘‘Evaluation Table’’ in the right top corner presents the
determination of the grade of the measurement: ‘‘R9_110_2.mnt’’-highlighted in the list box
labelled as ‘‘3D-Parameters’’ and plotted in the left top corner of the figure. From the middle list
box all the sixteen measurements (parameters’ values and grades) of the first liner sample can
also be seen. In the list box labelled as ‘‘Sample Parameters’’ the parameter values are averaged
per liner samples and the samples’ final grades are presented. The bottom list box shows the
needed number of measurements for the chosen parameters of the third sample
11.3 Case Study 1: Blechmantel Effect on the Wear

of Truck Liners
In this case study (Dimkovski et al. 2009, 2011a, b), areal measurements of liner
samples were taken from the top (top ring reversal location), middle and bottom
(unworn surface used as a reference for comparison) regions, representing regions
with different wear regimes. The size of the axial scratches caused by abrasive
wear was measured by a 509 objective (0.16 9 0.12 mm measurement area),
while for the estimation of Blechmantel and distribution of wear scratches a
109 objective (0.8 9 0.6 mm measurement area) on a coherence scanning
interferometer (CSI) was used. The Blechmantel parameter was evaluated by using
the method described in Sect. 11.2.1, and for each measurement an image was
generated to check the result. Such images from top, middle and bottom regions
Fig. 11.8 Identification and quantification of Blechmantel of a top region surface. Reprinted
from Dimkovski et al. (2009), Copyright (2009), with permission from Elsevier
Fig. 11.9 Identification and quantification of Blechmantel of a middle region surface. Reprinted
from Dimkovski et al. (2009), Copyright (2009), with permission from Elsevier
Fig. 11.10 Identification and quantification of Blechmantel of a bottom region surface.

are shown in Figs. 11.8, 11.9, and 11.10 respectively. All the parameters were
averaged per region since nine measurements were taken from each region. The
mean values of the parameters are given in Table 11.1, where the liner’s regions
are denoted as: ‘B’—Bottom region, ‘M’—Middle region and ‘T’—Top region.
For a qualitative assessment of the distribution of the wear scratches and surface
modification in general, four stitched 109 objective measurements from the top,
middle and bottom region of liner #6 are shown in Fig. 11.11.
By analysing the values of the parameters from Table 11.1, a general trend of
smoothing of the surfaces is seen in decreasing the values of Blech, Sa, Ssc and
Spk moving from the bottom to the middle to the top region almost for each liner.
This can also be noted by looking at Figs. 11.8, 11.9, 11.10, and 11.11. It is in the
top region (Fig. 11.8) that the Blechmantel is mostly removed as very few
(somewhere none) grooves can survive the smoothing which reflects the smaller
values of the computed Blech parameter compared to the other two regions. It
should be pointed out here that the removal of Blechmantel is not the only con-
tributor to surface smoothing but also an accumulation of deposits occurs in this
region at the same time (Dimkovski et al. 2011b). In the middle region, axial traces
of wear are clearly visible, but the Blechmantel is almost the same as in the bottom
region. By a careful inspection of Fig. 11.11 it can be seen that the folds are
rounded off rather than detached.
Table 11.1 Areal standard parameters, Blech, H and W of axial scratches. ‘B’, ‘M’, and ‘T’ in
the first column stand for: bottom, middle and top region of the liner’s sample respectively
Liner Blech Sa Sz Sds Sdq Ssc Sk Spk Svk Sr1 Sr2 H w
% lm lm pks/ lm/ 1/lm lm lm lm % % lm lm
mm2 lm
IB 4.31 0.37 7.01 3,743 0.12 0.0314 0.70 0.25 1.50 8.14 78.89
1M 5.94 0.37 6.50 3,939 0.14 0.0326 0.52 0.22 1.82 8.20 77.55 0.34 4.13
IT 0.30 0.06 4.68 7,638 0.04 0.0157 0.17 0.09 0.16 8.80 87.35 0.3 13
2B 5.77 0.67 12.60 3,317 0.17 0.0310 0.68 0.29 3.10 5.86 71.06
2M 4.90 0.63 12.15 3,444 0.20 0.0360 0.69 0.18 3.13 5.83 73.73 0.402 4.12
2T 4.13 0.31 7.60 3,554 0.12 0.0255 0.54 0.16 1.55 7.23 81.18 0.3185 3.61
3B 3.16 0.37 10.20 3,668 0.11 0.0306 0.68 0.23 1.64 7.86 80.31 0.39 5.16
3M 5.30 0.39 8.07 3,550 0.12 0.0253 0.63 0.15 1.65 4.50 76.20 0.52 5.16
3T 1.66 0.34 5.59 4,078 0.14 0.0335 0.80 0.20 0.98 5.70 79.15 0.53 4.13
4B 3.96 0.28 7.86 3,759 0.11 0.0310 0.57 0.27 1.30 9.62 83.37
4M 7.02 0.37 8.20 3,502 0.13 0.0271 0.33 0.23 1.61 7.43 68.13 0.435 4.13
4T 2.36 0.18 5.25 3,703 0.09 0.0197 0.40 0.18 0.70 9.22 83.47
5B 4.14 0.33 9.79 3,522 0.12 0.0317 0.40 0.32 2.04 14.60 81.60
5M 6.08 0.61 11.94 3,296 0.17 0.0329 0.44 0.29 2.59 7.60 65.15 0.27 4.13
5T 3.87 0.36 8.41 3,271 0.14 0.0291 0.69 0.17 1.51 4.03 80.18
6B 5.38 0.51 9.91 3,419 0.15 0.0343 0.75 0.23 2.17 6.32 74.92
6M 5.44 0.52 11.73 3,534 0.16 0.0312 0.68 0.18 2.18 4.97 72.02 1 5.675
6T 4.98 0.41 10.14 3,847 0.17 0.0323 0.60 0.16 2.10 5.62 77.87
7B 6.12 0.37 7.55 3,693 0.12 0.0325 0.70 0.25 1.48 8.02 79.02
7M 6.33 0.43 5.93 3,511 0.13 0.0277 0.60 0.16 1.79 5.94 72.79 0.21 4.13
7T 3.87 0.25 7.07 3,597 0.10 0.0237 0.57 0.22 1.04 9.50 84.50 0.42 4.13
8B 4.21 0.25 7.80 3,743 0.10 0.0315 0.42 0.36 1.24 14.83 84.18
8M 6.03 0.42 6.66 3,375 0.14 0.0311 0.54 0.23 1.70 7.75 71.60 0.16 3.1
8T 2.97 0.19 8.54 3,822 0.10 0.0223 0.40 0.18 0.77 5.80 81.25 0.37 4.13
9B 4.23 0.23 6.36 3,763 0.09 0.0298 0.40 0.34 1.09 14.59 84.24
9M 5.36 0.37 8.30 3,554 0.14 0.0318 0.45 0.36 1.51 8.96 72.36
9T 3.25 0.18 6.43 3,447 0.09 0.0215 0.33 0.25 0.70 10.95 81.15 0.56 6.19
10B 3.99 0.24 11.26 3,633 0.10 0.0291 0.36 0.35 1.30 16.40 84.40
10M 5.46 0.39 6.11 3,447 0.12 0.0270 0.34 0.21 1.63 7.25 66.90 0.29 3.1
10T 3.33 0.22 7.62 3,673 0.11 0.0237 0.28 0.19 1.07 11.62 76.37 0.43 6.19
Average 0.41 4.95
Comparing the parameters, in some liner samples, the values of the parameter
Blech (also Sa, Ssc and Spk) from the middle region are higher than those from the
bottom region. This could be attributed to the very mild wear in the middle region
and its higher initial roughness before the running. A precise relocation before and
after the tests is necessary to overcome this problem.
The presence of an axial scratch in the bottom region (Figs. 11.10 and 11.11) is
a result of the interaction between the piston skirt, some abrasive particle and the
Fig. 11.11 Four (of nine) 109 objective stitched measurements from the top middle and bottom
region of the tested liner #6 for a qualitative assessment of the distribution of the wear scratches
and surface modifications in general. Reprinted from Dimkovski et al. (2009), Copyright (2009),
with permission from Elsevier
liner surface. Furthermore, this scratch has a very common height and width
(H 9 W = 0.39 9 5.16 lm for the sample 3B from Table 11.1) and also looks
very similar to the larger scratches from the other regions. These ‘‘larger’’ scrat-
ches have an average (the last row of Table 11.1) of 0.41 lm in height and
4.95 lm in width. Observing the wear scratches in all three regions in Fig. 11.11,
it can be seen that they are finest and densest in the top region, less denser and
largest in the middle region and almost absent in the bottom region.
11.4 Case Study 2: Assessment of Surface Finish

Improvement of Car Liners
The areal parameters: honing Angle, groove width a, height c, distance between
grooves d, Blechmantel Blech, groove interruption Intrp, Holes, number of
grooves NG and groove balance Bal were developed in this case study for a more
Fig. 11.12 Identification of Blechmantel, groove interruption and holes marked in transparent
white, yellow and green colour respectively for the three surface types (type 1—top, type 2—left
and type 3—right)
complete characterisation of liner surfaces (Dimkovski et al. 2008). They were

employed to evaluate the extent of improvement of the surface finish of car liners.
Three surface types (see Fig. 11.12) were monitored: type 1, 2 and 3; all honed
with diamond tools. Sixteen 109 objective measurements were randomly taken
from types 1 and 3, and twelve from the type 2.
The algorithms described in Sect. 11.2.1 were employed to calculate the new
parameters and, together with the standard areal parameters, were averaged per
surface type. The mean parameter values are presented in the upper part of
Table 11.2, while the required number of measurements calculated for ±20 %
tolerance limits and 95 % confidence interval are presented in the lower part of
Table 11.2.
The parameters: Sq, Sk, Spk, Svk, Sds, Ssc, Sdq, Angle, a, c, d, Blech, Intrp,
Holes and NG were chosen for rating of the surface types by using the method for
unknown control limits (see Sect. 11.2.2, Fig. 11.7). Six grades were adopted: 1, 2
and 3 are unacceptable and 4, 5 and 6 are acceptable (i.e. 1-worst, 6-best). Sk, Spk,
Ssc, Sdq, Blech, Intrp and Angle were selected to have lower values of interest (the
lower value, the better quality) with 70 % acceptable area under their probability
densities; Sq, Svk, a, c, d and Holes with central values of interest, 70 % acceptable
area and 75 % unacceptable area of the greater part; and Sds and NG with higher
values of interest and 70 % acceptable area. This selection (especially the
parameters Sq, Svk, a, c, d, Holes and NG) should be treated with caution since the
overall effect of some features is not clear. Having lower values of Sk, Spk, Ssc and
Sdq indicates smoother plateaus, better sealing and in turn a lower oil consump-
tion. These parameters were found to be negatively correlated with Sds so that a
higher value of Sds means smoother plateaus as well. Since a large part of the
interrupts may be a folded material (i.e., Blechmantel), high amounts of Blech-
mantel (Blech parameter) and interrupt (Intrp parameter) would be undesirable
from a wear perspective. Based on the simulation results of Jocsak et al. (2006), a
lower honing Angle was chosen to be preferable. The Sq parameter was selected
for three reasons: it is a very stable parameter; it is an input parameter in the ring-
liner simulations; and it gives information for the surface roughness in general.
Even though Bal is an important parameter it was excluded from the rating
because of its instability (see the high number of measurements needed in
Table 11.2).
The results of the ratings were that the type 1 and type 3 surfaces were both
graded as acceptable Grade 4, while the type 2 surface was graded as unacceptable
Grade 3. On inspection of Fig. 11.12 and the values of the roughness parameters in
Table 11.2, it can be seen that the surface type 2 is significantly rough (highest Sq
value), with large grooves (highest groove width a and height c), bad groove
balance (highest Bal), high Blechmantel coverage (highest Blech), with most holes
(highest percentage of Holes) and with large islands (highest mean volume of
islands MVI) compared to the other two types. It should be noted that this liner
type is the oldest one manufactured, implying that the honing at that time was not
developed as well as now. Furthermore, a comparison between type 1 and 3 yields:
type 3 has narrower, denser and more grooves (lower a and d values, and higher
Table 11.2 Mean values of the roughness parameters (table’s upper part) and required number of measurements (lower part) within ±20 % using the t-
300
distribution with 95 % confidence interval for the three surface types

Mean values of the roughness parameters
Standard parameters
Sulface Sq Sz Sa Sdq Sds Ssc SK Spk Svk Sr1 Sr2 MVl
lm lm lm (-) 1/lm2 1/lm lm lm lm % % lm3
Type 1 0.43 8.30 0.27 0.14 15,594 0.090 0.55 0.12 0.93 4.93 78.00 60.01
Type 2 0.82 10.33 0.53 0.22 18,338 0.165 1.14 0.35 1.71 7.37 78.36 147.55
Type 3 0.48 11.77 0.29 0.26 36,143 0.272 0.61 0.33 1.09 10.04 82.90 24.35
Complemented groove parameters
Surface a c d Blech BlechREL Intrp IntrpREL Balance Angle Stray Holes NG
lm lm lm % % % % (-) deg % % (-)
Type 1 22.16 0.53 115.24 14.47 43.18 8.08 26.76 8.12 49.5 11.16 1.23 14
Type 2 26.54 1.02 68.32 23.79 42.52 9.20 17.40 18.05 45.7 6.45 2.14 19
Type 3 16.09 0.64 79.87 16.25 40.72 11.94 31.12 13.47 45.1 2.14 1.20 20
Needed number of measurements
Standard parameters
Sulface Sq Sz Sa Sdq Sds Ssc SK Spk SvK Sr1 Sr2 MVI
Type 1 13 20 14 10 5 16 17 22 20 12 2 27
Type 2 11 9 10 4 3 8 24 24 26 17 6 35
Type 3 8 15 8 3 3 4 5 6 13 4 2 35
Complemented groove parameters
Surface a c d Blech BlechREL Intrp IntrpREL Balance Angle Stray Holes NG
Type 1 8 17 13 13 4 19 20 58 15 56 23 12
Type 2 11 22 10 7 8 22 19 64 4 101 28 8
Type 3 4 11 11 8 3 10 10 38 1 129 21 7
Z. Dimkovski et al.
NG), with a lower honing Angle and smaller islands (smallest MVI) than type 1,
while the type 1 has better groove balance (lowest Bal value), less Blechmantel
(lowest Blech) and less interruption (lowest Intrp) than type 3.
Regarding the required number of measurements, Sds shows the best stability
while Stray grooves and Bal show the worst. Sk, Spk and Svk exceed the made
number of measurements and caution is required when interpreting their values in
the rating of the liners.
11.5 Conclusion
Two methods: a method for extraction of the groove parameters characteristic for
honed liners, and a method for rating of the surface quality, were developed and
applied in two case studies. In the first case, the effect of the folded metal-
Blechmantel parameter on the wear of truck liners was investigated. It has been
revealed that most of the Blechmantel remained on the surfaces after running in
the engines, which performed acceptably well. In the second case, the groove
parameters, together with the standard areal parameters, were used to assess the
improvement of surface finish of car liners. The rating method was applied to rank
the surfaces, suggesting that the newer surface finish is better than the older ones,
but still some improvements regarding the groove balance, interrupt and Blech-
mantel should be made. In addition, the parameters, Sds and Blech, were found to
be suitably stable (because of the low number of measurements needed) while the
parameters, groove balance and stray grooves, were very unstable and, therefore,
not used in the rating.
Acknowledgments The authors wish to thank the Swedish Knowledge Foundation for funding,
Volvo Cars and Volvo Trucks for their kind contribution with money, man hours and liners, and
Digital Surf for providing a free version of MountainsMap software.
References
Anderberg C, Cabanettes F, Dimkovski Z, Ohlsson R, Rosén B-G (2006) The cylinder liners and
consequences of improved honing. In: Proceedings of the 12th Nordic symposium in
tribology, Helsingor, Denmark, Jun, Paper NT2006-11-71
Anderberg C, Pawlus P, Rosén B-G, Thomas TR (2009) Alternative descriptions of roughness for
cylinder liner production. J Mater Proc Technol 209:1936–1942
Beyerer J, Krahe D, Puente FP (2001) Characterisation of cylinder bores. In: Mainsah E,
Greenwood JA, Chetwynd DG (eds) Metrology and properties of engineered surfaces.
Dordrecht, Kluwer, pp 243–281
Dimkovski Z, Anderberg C, Ohlsson R, Rosén B-G (2008) Complementing 3D roughness
parameters for monitoring of improved honing of cylinder bores, 2nd Swedish production
symposium, Göteborg, Sweden, November 2008
Dimkovski Z, Anderberg C, Ohlsson R, Rosén B-G (2011a) Characterisation of worn cylinder
liner surfaces by segmentation of honing and wear scratches. Wear 271:548–552
Dimkovski Z, Bååth L, Rosén S, Ohlsson R, Rosén B-G (2011b) Interference measurements of

deposits on cylinder liner surfaces. Wear 270:247–251
Dimkovski Z, Anderberg C, Rosén B-G, Ohlsson R, Thomas TR (2009) Quantification of the cold
worked material inside the deep honing grooves on cylinder liner surfaces and its effect on
wear. Wear 267:2235–2242
Goetze AG Honing Guide (1988) Rating criteria for the honing of cylinder running surfaces,
Burscheid, Hauptverwaltung, D-5093
Grabon W, Pawlus P, Sep J (2010) Tribological characteristics of one-process and two-process
cylinder liner honed surfaces under reciprocating sliding conditions. Tribol Int 43:1882–1892
Hough P V C (1962) Methods and means for recognizing complex patterns. US patent 3069654
Jocsak J, Li Y, Tian T, Wong VW (2006) Modeling and optimizing honing texture for reduced
friction in internal combustion engines. SAE Technical paper series 2006-01-0647
Johansson S, Nilsson PH, Ohlsson R, Anderberg C, Rosén B-G (2008) New cylinder liner
surfaces for low oil consumption. Tribol Int 41:854–859
Kanthababu M, Shunmugam MS, Singaperumal M (2009) Identification of significant parameters
and appropriate levels in honing of cylinder liners. Int J Mach Mach Mater 5:80–96
Keller J, Fridrici V, Kapsa Ph, Vidaller S, Huard JF (2009) Influence of material nature and
surface texturing on wear of heavy-duty diesel engine cylinder liners. Tribol Trans
52:121–126
Michalski J, Wos P (2011) The effect of cylinder liner surface topography on abrasive wear of
piston-cylinder assembly in combustion engine. Wear 271(12):582–589
Pawlus P, Cieslak T, Mathia T (2009) The study of cylinder liner plateau honing process. J Mater
Proc Technol 209:6078–6086
Sabri L, Mezghani S, El Mansori M, Zahouani H (2011) Multiscale study of finish-honing process
in mass production of cylinder liner. Wear 271:509–513
Chapter 12
Characterisation of the Mechanical
Bond Strength for Copper on Glass
Plating Applications
Baofeng He, Jon Petzing, Paul Webb, Paul Conway

and Richard Leach
Abstract This chapter describes research into the role that surface topography
plays in influencing the mechanical bond strength of the electroless copper plating
of novel glass substrates. The work considers bespoke laser machining of glass
substrates, electroless plating chemistry, areal surface topography analysis using
non-contact optical techniques, parametrisation of the surfaces using areal
parameters described in ISO 25178 part 2, and scratch testing of plated copper to
measure the adhesive bond strength. By correlating bond strength to appropriate
areal parameters, it is anticipated that better mechanical adhesive potential of
machined glass surfaces can be achieved.
12.1 Introduction and Background
The research presented in this chapter starts to evaluate the role that surface
topography may have on influencing the mechanical bond strength of the elec-
troless copper plating process. The central aims of the research are to:
• improve mechanical adhesion of copper on glass, thus improving electronic
functionality and life cycle; and
• correlate adhesion quality with areal surface texture parameters to identify
parameters that can reduce the design/testing cycle of glass surface preparation
and acceptance.
B. He J. Petzing P. Webb P. Conway

Wolfson School for Mechanical and Manufacturing Engineering, Loughborough University,
Loughborough, LE11 3TU, UK
R. Leach (&)
Engineering Measurement Division, National Physical Laboratory, Hampton Road,
e-mail: richard.leach@npl.co.uk

304 B. He et al.
The drive towards increasing densities of components and integrated circuit

inputs/outputs in electronics is pushing the capabilities of conventional printed
circuit board (PCB) manufacture to its limits. Whilst sub-100 lm metal features
can be produced, the dimensional instability of organic material-based boards in
response to changes in temperature or humidity means such features cannot be
reliably aligned. Glass is attractive as an alternative substrate material due to its
relatively low cost and high thermal stability, with a coefficient of thermal
expansion similar to that of silicon. Glass has the added benefit that it is trans-
parent, which could simplify assembly and inspection of components with area
array or hidden interconnects, and facilitate developments in optoelectronic cir-
cuitry with optical interconnects.
The concept of using glass for high density interconnects has been explored by
several research groups (Schröder et al. 2003). Previous work at Loughborough
University (LU) has examined the feasibility of the major process steps required to
form a high density, multilayer glass circuit board; glass layer lamination, drilling
of microvias, and electrical interconnect pattern formation (Cui et al. 2008a, b).
The method chosen for the latter was electroless plating of copper due to its
process advantages over other alternatives, such as reduced energy consumption
and cost, and enhanced deposition rates. It was possible to plate layers up to
150 nm in thickness with good adhesion when using a silane adhesion promotion
coating, however, thicker layers tended to peel off.
A further measure to improve plating adhesion would be to texture the glass
surface. In the previous work, randomly rough surfaces showed improved plating
retention compared to smooth untreated glass, although a systematic understanding
of the influence of surface texture was not previously a research priority. However,
it seems likely that controlling the surface texture would improve adhesion still
further, potentially allowing mechanical enhancement to the bond strength
between plated metals (such as copper) and glass substrates, in addition to the
traditionally considered electrochemical bonding characteristics.
Whilst previous researchers have reported that micro-columnar array structures
could enhance adhesive bonding strength for metals and alloys (Baburaj et al.
2007; Zhang et al. 1997; Starikov et al. 2004), there has been little progress in
generating micro-scale patterned structures on glass. Techniques have, therefore,
been developed at LU to produce a variety of textures and structures on glass
substrates, allowing a systematic study of the effect of surface texture on elec-
troless plated copper adhesion. This has primarily involved the use of excimer
laser machining technology to selectively ablate small areas of glass.
The complex relationship between surface texture and adhesion has interested
scientists and engineers for more than fifty years. Examples of research output are
identified elsewhere (Takadoum et al. 1997; Hallab et al. 2001; Shahid and Hashim
2002; Chong et al. 2003; Garbacz et al. 2006; Bénard et al. 2006; Menezes et al.
2006; Jiang et al. 2007; Zappone et al. 2007), where authors identify that varying
degrees and types of surface texture appear to have beneficial effects on adhesion.
Analysis of these examples indicates that surface topography is typically consid-
ered in a superficial manner. Furthermore, although surface profile parameters may
12 Characterisation of the Mechanical Bond Strength 305
potentially be restrictive and misleading, very few adhesion researchers have

considered areal surface texture parameters to characterise surface texture over the
last ten years, a period of time within which equipment, data processing software
and published texts have provided access to the use of areal parameters. Whilst an
example of the use of the Sa parameter can be cited in the context of adhesion
(Bénard et al. 2006), little attempt has been made to consider the breadth of
parameters (and consequently surface description) available. It can also be iden-
tified that no evidence can be found of publications concerning bespoke texturing
of glass for electroless copper plating.
In this chapter the development of the experimental techniques used is
described, initial results obtained and analysis are presented and discussed, with
reference to chosen areal parameters. The novelty of the work reported here is in
identifying and producing the optimal surface topography for adhesion of plated
metal layers on glass, and linking this to areal parameters that describe the sur-
faces. It should also be noted that the manufacturing techniques used within this
research are not proposed as production methods for future glass based PCBs, but
have been chosen for their versatility and controllability within the context of
experimental investigation.
12.2 Aims and Objectives
The research presented here starts to question the role that surface topography may
have on influencing the mechanical bond strength of electroless copper plating.
The aims of the research are to:
• improve mechanical copper adhesion on glass, thus improving electronic
functionality and life cycle; and
• correlate adhesion quality with areal parameters to identify parameters that can
reduce the design/testing cycle of glass surface preparation/acceptance.
The following objectives are identified as key measures of project progress.
• Investigate manufacturing processes for the machining of micrometre scale
features into glass substrates.
• Investigate manufacturing processes for producing (controlled) randomly rough
surfaces on glass substrates.
• Develop contact/non-contact metrology techniques to routinely measure
micrometre scale features on glass substrates.
• Assimilate and develop copper plating chemistry for textured glass surfaces.
• Develop and apply quantifiable scratch testing techniques for assessing copper/
glass bond strength (critical load) using equipment based at the National
Physical Laboratory (NPL).
• Produce a number of glass batches with sequentially refined surfaces, and test at
NPL.
306 B. He et al.
• Produce statistics that demonstrate the quality of correlation between critical

load values (adhesive bond strength) and a range of ISO 25178 areal field
parameters.
• Identify key ISO 25178 areal field parameters that provide routinely robust
characterisation of glass textured surfaces that are identified as being suitable for
electroless copper plating and enhanced cooper bond strength.
12.3 Processing and Measurement of Glass
A range of methods and techniques for the production of textured glass surfaces
has been considered during this research, and can be categorised into the pro-
duction of random rough surfaces and the production of structured surfaces.
• The production of random rough surfaces has been investigated and the fol-
lowing techniques considered: acid etch, bead blasting, micro-sand blasting and
plasma etching, all of which produce randomly rough surfaces, but with limited
control over volume of material removal per unit area, and subsequent rough-
ness values.
• The preparation of structured surfaces has been investigated and the following
techniques considered: mechanical machining and laser machining. Laser
machining (or ablation) provides finesse of material removal, which mechanical
machining struggles to match, unless diamond turning techniques are used.
The research needed to initially address the ability to produce glass surfaces
with the maximum amount of control, repeatability and material removal finesse.
These criteria maximise the opportunity to study cause and effect, in terms of glass
surface mechanical structure (characterised via areal parameters) and copper
adhesion quality. The requirement to eventually consider volume or large scale
manufacturing can then be subsequently addressed in terms of process change to
optimise lead time and cost reduction. To this end, the majority of glass surface
preparation has used excimer laser machining (ablation).
12.3.1 Laser Materials Processing
Cerium doped glass (CMG specification) supplied by Qioptiq Ltd was chosen as a
substrate material due to the beneficial absorption characteristics at the excimer
laser wavelength available (248 nm) thus maximising glass removal efficiency,
and the similarity of thermal expansion coefficient to silicon. The glass sheets were
supplied as square samples with lateral dimensions of 40 mm by 40 mm, and
thicknesses of 100 lm and 500 lm.
A krypton fluoride excimer laser (model EMG 203, Lambda Physik) operating
at 248 nm was used for machining the CMG glass. The laser system parameters,
including output pulse energy (200 mJ to 250 mJ), repetition rate (5 Hz to 25 Hz),
and shots per area (5–200), were varied and optimised according to the surface
topography and adhesion performance of the samples. Various metal masks with
square, circular and triangular apertures were placed in the laser path to tailor the
size and the shape of laser beam projected onto the glass work substrates. The
excimer laser was typically operated at energy density of 26 mJmm-2 and a pulse
frequency of 10 Hz.
Excimer lasers can generate micro-scale pattern array structures on glass by
using a mask projection and dragging process (Tseng 2007). Complex micro-scale
structures were typically produced in two steps. Firstly, the work piece (glass) was
mounted on the computer numerically controlled (CNC) work stage to form a
groove contour along one direction, with step-wise repetition of the process
forming overlapping grooves. Secondly, and where necessary, grooves were
machined orthogonally to the initially machined grooves, creating segmented
periodic structures.
A range of novel and complex micro-scale patterned structures of various
depths were generated on CMG glass by changing the mask geometry, mask
dimensions and laser operation parameters. Laser variables were investigated
systematically to minimise the micro-roughness effect during production of the
structured surfaces; micro-roughness being an inherent function of the fluctuating
spatial energy density of the laser beam profile.
12.4 Basic Methodology
The basic methodology and techniques used to correlate surface topography with
adhesion performance are shown in Fig. 12.1, and can be summarised as:
• excimer laser machining with control of machining variables;
• application of contact/non-contact surface metrology to characterise the
machined surfaces, prior to coating, via areal parameters;
• electroless copper plating with control of plating variables;
• scratch testing of copper plating to identify critical load of failure (adhesive
bond failure); and
• correlation of failure critical loads against areal parameter values.
The surface topographies were measured using non-contact areal surface
topography measurement techniques such as coherence scanning interferometry
(CSI) and focus variation (FV) (see Leach 2011 for an overview of optical mea-
surement methods), coupled with the new ISO 25178 part 2 (2012) areal surface
texture parameters. Typically a 109 lens was used on each instrument, providing a
lateral field of view of approximately 700 lm by 500 lm.
308 B. He et al.
Fig. 12.1 Basic methodologies and techniques
A Zygo NewView 5000 CSI system was predominantly employed to measure

the majority of the surface topographies of each machined micro-pattern. Typical
areal surface topography images machined by different masks are shown in
Figs. 12.2 and 12.3. These images show novel grid pattern micro-structures on the
CMG glass surfaces produced by using circular and square masks. Micro-ramp and
pyramid pattern structures can be produced by using triangular masks. When using
fixed mask and laser parameter settings, the topography of the micro-pattern
structure may be modified, by varying the extent of overlaps and substrate rotation.
Thermal shock from the laser photo-thermal process causes micro-cracking of
the glass substrate, but these cracks are typically sub-surface and do not contribute
to the surface topography. The micro-cracks will act as fatigue crack initiation
sites if the glass substrate is subject to cyclic stress reversals, causing premature
failure of the substrate.
Fig. 12.2 Areal view of micro pattern structures machined using a square mask
Fig. 12.3 Areal view of micro pattern structures machined using a triangular mask
12.5 Electroless Copper Plating of Glass
Electroless plating (auto-catalytic plating) of the glass is necessary because the

glass substrate is non-conductive and, therefore, not suitable for traditional elec-
troplating techniques. Typical electroless copper plating procedures include sub-
strate cleaning, catalyst dipping and copper bath dipping. Electroless deposition is
a self-accelerating process in which metal ions are chemically reduced at catalytic
surface substrates. The specific processes used for this research are illustrated in
Fig. 12.4 and can be summarised as follows:
• substrate cleaning with Decon 90 surface active cleaning agent at 60 C, for ten
minutes followed by water wash;
• pretreatment dip using (3-aminopropyl) trimethoxysilane (APTS) to form a self-
assembled monolayer (SAM) on the glass substrate to change the chemical
functionality of the surface;
• CircupositTM catalyst 3344/4444 to prepare the pretreated glass surface;
• final dip of the treated glass substrates in agitated CircupositTM 3350-1 at ele-
vated temperature (typically 45 C); and
• final wash and dry.
Fig. 12.4 Electroless copper plating procedure for CMG glass

310 B. He et al.
Fig. 12.5 Electroless copper plating quality as a function of dip time
Table 12.1 Experimental procedure for electroless copper plating on CMG glass
Step Process Solution Time Temperature
1 Cleaning Decon 90 5 min 60 C
2 Rinse Deionised water 3–5 min Room temperature
3 SAM deposition APTS (5 9 10-3 mol l-1) 1h Room temperature
Solvent: methanol (95 %)
5 Catalyst Circuposit 3344 5 min 36 C
7 Electroless copper Circuposit 3350-1 15 min 46 C
8 Rinse and dry Deionised water 10 min Room temperature
Process variables investigated included wash times, wash conditions, APTS and
catalyst concentrations, solution temperatures, rate of agitation and dip times.
Examples of efficacy and thickness of coating are illustrated in Fig. 12.5, showing
increasing plating coverage and thickness as a function of time. Final plating
experimental procedure was established and is identified in Table 12.1. Typical
copper plating thickness on all glass samples was approximately 0.5 lm. Exam-
ples of copper plating on laser machined glass patches are shown in Fig. 12.6, with
copper firmly adhered on the machined substrate regions and copper typically not
adhering to the non-machined substrate.
Initial experimentation resulted in machined areas of 40 mm by 10 mm (first
two samples), but this was later reduced to machined areas of 10 mm by 10 mm
(third sample). Furthermore, electroless chemical process improvement resulted in
samples with minimal copper plating of smooth glass, as evidenced by the three
samples shown in Fig. 12.6.
Fig. 12.6 Examples of

electroless copper plating on
excimer laser machined CMG
glass
12.6 Scratch Testing of Glass
Scratch testing was chosen as the most effective quantitative assessment of

adhesion strength between copper coating and the CMG glass substrates. The
technique involves a diamond stylus being loaded against the sample and then
drawn across the surface to cause a scratch. The load applied to the stylus is either
held constant or linearly increased. Scratches can be single or multi-pass. Coating
312 B. He et al.
Fig. 12.7 Different types of damage that may be observed in scratch testing (Bull and
Berasetegui 2006)
failure is defined at the point where surface defects are visually identified, although
failure point identification can be subjective and prone to uncertainty.
Existing researchers have demonstrated correlation of coating adhesion mea-
surement (by scratch testing) with adhesive bond strength, through a number of
proposed failure models (Chalker et al. 1991; Maxwell 2001; Bull and Berasetegui
2006). Figure 12.7 illustrates the typical failure elements found during coating
delamination, cracking, spalling, coating detachment and coating penetration/
delamination.
This research used the ST 3001 test system at NPL, which incorporates a
200 lm Rockwell C diamond stylus. Ramp loading was increased from a mini-
mum load of 1 N to a maximum load 15 N incrementing in thirty steps. As a
consequence of the scratch test result, knowledge of the sample stress state leading
to delamination failure is available through direct measurement of the distance
from the loading start point to the coating penetration point and by theoretical
calculation. Examples of test scratches on copper coated glass samples can be seen
Fig. 12.8 Optical CMM images for scratches of copper coating on CMG glass
Fig. 12.9 SEM images of scratches on copper coating on CMG glass a overall scratches
b spalling position c delamination position
in Fig. 12.8 (optical coordinate measuring machine (CMM) images) and Fig. 12.9
(scanning electron microscope (SEM) images).
Failure points along the scratch are typically distinguished by backlighting
penetrating through the copper layer and are easier to recognise using white light
techniques compared to SEM images. Recognition of different failure modes may
change depending on eventual copper/glass use and manufacturer. Failure was
initially identified as the point of first observable failure (any mode as defined in
Fig. 12.7).
Damage patterns on the surface are correlated with acoustic emission and
friction forces from the ST 3001 test system, to derive the critical loads at which
various failure mechanisms occur. Graphical representation of the experimental
data provides an alternative direct method of identifying the critical load of coating
delamination, as shown in Fig. 12.10. Point (a) on the graph identifies the probable
point of copper delamination.
Other graphical scratch testing results do not provide definitive changes and
require visual analysis to identify the points of failure. Furthermore, the graphical
analysis in Fig. 12.10 does not allow for the identification of the three failure
criterion as defined in Fig. 12.7.
The testing of adhesive bond strength is, however, susceptible to a number of
interfering factors that should be controlled as much as possible, for instance:
314 B. He et al.
Fig. 12.10 Scratch test result

of CMG glass with structured
surface
• copper coating quality (variation of thickness) affects the scratch testing result;
• coating failure occurs easily on weak and thin coated areas;
• preferential failure occurs at sharp discontinuities of the glass surface (stress
concentrations), that may not be representative of the bulk adhesion;
• identification of spalling and delamination is subjective and prone to human
error; and
• graphical analysis (example in Fig. 12.10) does not necessarily allow for dif-
ferent failure modes to be identified.
12.7 Correlation of Areal Parameters with Critical Load
As identified in Sect. 12.3, a Zygo NewView 5000 CSI system was used to measure
the surface topography of each machined micro-pattern, with consistent lens and
magnification settings (109) and field of view (700 lm by 500 lm). Surface
topography and areal parameters were generated and analysed using Talymap v5.1
(DigitalSurf) surface texture processing software that is designed to be compliant
with ISO 25178 part 2 (2012). Filtering of the data initially considered a consistent
and fixed set of wavelength cut-off values (typically 80 lm). However, the use of
standard filter parameters for all glass samples did not allow appropriate differen-
tiation between micro-roughness components and structural components of the
glass, leading to subsequently limited correlation with critical load.
Re-evaluation of the laser machined surfaces identified three scenarios that
exist, which required clarification and quantification:
• the correlation between the micro-roughness component of the glass surface and
the critical load;
• the correlation between the structural component of the glass surface and the
critical load; and
• the correlation between the micro-roughness and structural components of the
glass surface and the critical load.
Micro-roughness on the glass surface is an inherent function of the stochastic

nature of the excimer laser wavefront spatial characteristics (Sect. 12.3), and has
been evaluated in its own right with glass samples featuring micro-roughness, but
no structure.
Furthermore, process consideration of this data (and other data sets) confirmed
the requirement to investigate a number of potential failure modes for data pro-
cessing investigation, because the failure mode analysis is prone to premature and/
or random copper bond failure, which is illustrated in the two sets of tracks in
Fig. 12.8 (black points along the scratches indicate possible localised failure
points). This is typified by a scratch testing stylus impacting on a structural dis-
continuity, causing a localised stress concentration resulting in localised failure
that is not indicative of bulk bond strength and is illustrated in Fig. 12.11.
Alignment of the scratch testing stylus with the sample is a manual process and,
therefore, it is difficult to guarantee alignment with the preferential direction of the
structured surface. Figure 12.11 shows slight misalignment of the scratch track
causing a prolonged impact on a structure edge along each track (half way along
each track—seen as a black line across the scratch). Once the stylus leaves the
feature edge and continues along the structure surface, plating failure is no longer
observed, until the stylus load reaches the critical value at the end of each scratch.
Failure at this point is again indicated by long dark elements on the scratch.
Consequently, analysis of the data has led to the development of three failure
modes as being more representative of bulk copper adhesion (and potential user
application scenarios):
• Mode 1—simple—the point where the first failure occurs on the plated surface
(as illustrated in Fig. 12.11);
• Mode 2—consecutive—the point where three consecutive observable individual
failure points are identified within a predefined length of 0.5 mm (typically a
higher critical load compared to the first definition); and
Fig. 12.11 Premature

plating failure as a function of
stylus/discontinuity impact
316 B. He et al.
• Mode 3—continuous—the point where continuous delamination occurs for a

minimum length of 0.2 mm (typically the highest critical load compared to the
first two definitions).
Areal parameters (including field and feature parameters—see Chaps. 2 and 3)
for samples typical of those shown in Figs. 12.2 and 12.3 were calculated
according to ISO 25178 part 2 (2012). Two correlation coefficients have been
employed to quantify the strength of relationship between the areal parameter
values and the mean critical load value.
• Pearson product moment correlation coefficient (q). This coefficient provides a
measure of the strength of the linear dependence between two variables, giving a
value between +1.0 and -1.0.
• Spearman’s rank correlation coefficient (r). This coefficient provides a non-
parametric measure of the statistical dependence between two variables, varying
from -1.0 to +1.0, and does not require a linear dependence between the two
variables.
In this manner, it has been possible to categorise each areal parameter in terms
of the strength of correlation with the critical load values, being mindful of the
definition and relevance of each parameter. In each analysis case, the quality of
correlation has increased as the failure mode has been changed from simple, to
consecutive, to continuous. Table 12.2 shows correlation values for four param-
eters that have demonstrated the strongest behavioural relationships with the
copper/glass critical load values.
The parameters in Table 12.2 have been identified as having potential for
appropriately describing the glass surfaces in the context of bonding, on the basis
of the correlation results within the data sets, but also with reference to their
descriptions and mathematical functions as identified in ISO 25178 part 2 (2012)
and Chap. 2.
Individual graphs are presented in Figs. 12.12, 12.13, 12.14 and 12.15 and
illustrate the initial relationships between critical load from the scratch testing
results and the specific areal parameters. These data are based on the failure modes
identified above, with each mode plotted for each areal parameter.
In all four cases, strong trends have been identified between critical load and the
respective areal parameter, although the Vvc and Sq parameters show more scatter
as the areal value increases. In comparison, the Spc and Sdq parameters are
showing very little scatter with the exception of the 5 N critical load value data
Table 12.2 Areal parameters and associated correlation coefficient values

Areal parameter Spearman coefficient value (r) Pearson coefficient value (q)
Spc -0.86 -0.81
Sq -0.81 -0.76
Vvc -0.80 -0.75
Sdq -0.77 -0.73
Fig. 12.12 The Spc parameter value as a function of critical load and failure mode (r = -0.86,
q = -0.81)
Fig. 12.13 The Sq parameter value as a function of critical load and failure mode (r = -0.81,
q = -0.76)
point in each case. It should be noted that the maximum load applied by the scratch
testing equipment was 15 N; this force being determined by the failure of the glass
substrate at greater loads. If no copper plating failure is observed during a scratch
test, then the default critical load value will be 15 N. Consequently, it is possible
that some of the copper/glass samples may have higher bond strengths than 15 N,
but it has not been possible to quantify this strength. However, it is recognised that
there is a need to reinforce these graphical and numerical comparisons with further
data points from additional copper plated glass samples.
318 B. He et al.
Fig. 12.14 The Vvc parameter value as a function of critical load and failure mode (r = -0.80,
q = -0.75)
Fig. 12.15 The Sdq parameter value as a function of critical load and failure mode (r = -0.77,
q = -0.73)
Analysis of failure modes 1 and 2 shows that there are no significant correla-
tions between areal parameter value and critical load. This is not surprising
because the definitions of these two failure modes are susceptible to spurious, non-
representative plating failure as a function of edge effect stress concentrations in
the case of mode 1, and localised random defects in the case of mode 2. This fact
reinforces the decision to regard failure mode 3 as the most rigorous and consistent
test for copper plating failure (and, therefore, adhesive bond strength).
12.8 Conclusions and Further Work
The research presented here identifies the role that surface topography may have
on influencing the mechanical bond strength of the electroless copper plating on
glass substrates. This process is being achieved via controlled excimer based laser
based machining of glass, areal parametrisation of the surfaces, electroless copper
plating, critical failure mode identification using scratch testing, and finally cor-
relation/statistical analysis of critical load to areal parameter. Current achieve-
ments and conclusions of this work are listed below.
• Novel and bespoke structured geometries have been machined onto CMG glass
substrates using controlled excimer laser ablation, with a variety of mask
geometries, mask dimensions and laser parameter settings.
• Geometry structure has ranged laterally from 20 lm to 100 lm, with typical
feature depths of between 5 lm and 10 lm.
• Areal data sets have been produced, typically using CSI techniques, with ISO
25178 part 2 (2012) areal parameters being identified and produced for different
glass structures.
• Copper has been deposited onto machined CMG glass surfaces to form inter-
connection tracks by electroless copper plating. Optimal copper plating
parameters such as substrate cleaning, process temperatures, solution concen-
tration, agitation characteristics and dipping time have been identified.
• Scratch testing techniques have been implemented to quantify adhesive bond
strength of the copper plating, leading to the identification of critical loads.
• Visual comparisons and numerical correlation comparisons have been com-
pleted on current data, initially identifying four ISO 25178 part 2 (2012) areal
parameters of interest (Sdq, Spc, Sq and Vvc). These areal parameters have
shown strong correlations (in the range of 73–86 %) between parameter values
and critical load values (adhesive bond strength), causing further consideration
with respect to failure mode type and data filtering strategies.
Acknowledgments This research is a collaboration between Loughborough University (UK)

and the UK National Physical Laboratory. The research has been part funded by the EPSRC
3D-Mintegration Grand Challenge project, the Wolfson School of Mechanical and Manufacturing
Engineering, and the 2008–2011 NMO Engineering and Flow Metrology Programme. The
authors would also like to thank Qioptiq Ltd for the supply of CMG glass.
References
Baburaj EG, Starikov D, Evans J, Shafeev GA, Bensaoula A (2007) Enhancement of adhesive
joint strength by laser surface modification. Int J Adhes Adhes 27:268–276
Bénard Q, Fois M, Grisel M (2006) Surface treatment of carbon/epoxy and glass/epoxy
composites with an excimer laser beam International. J Adhes Adhes 26:543–549
320 B. He et al.
Bull SJ, Berasetegui EG (2006) An overview of the potential of quantitative coating adhesion
measurement by scratch testing. Tribo Int 39:99–114
Chalker PR, Bull SJ, Rickerby DS (1991) A review of the methods for the evaluation of coating-
substrate adhesion. In: Proceedings of the 2nd international conference on plasma surface
engineering (September), Garmisch-Partenkirchen, pp 583–592
Chong EK, Stevens MG, Nissen KE (2003) Effect of surface roughness on the adhesion of
electrolessly plated platinum to poly (ethylene terephthalate) films. J Adhes 79:667–681
Cui X, Bhatt D, Khoshnaw FM, Hutt DA, Conway PP (2008) Glass as a substrate for high density
electrical interconnect. In: Proceedings of the 10th international IEEE electronics packaging
technology conference (December), Singapore, 12–17
Cui X, Hutt DA, Conway PP (2008) An investigation of electroless copper films deposited on
glass. In: Proceedings of the 2nd IEEE electronics system-integration technology conference
(September), London, pp 105–110
Garbacz A, Courard L, Kostana K (2006) Characterization of concrete surface roughness and its
relation to adhesion in repair systems. Mat Charac 56:281–289
Hallab NJ, Bundy KJ, O’Connor K, Moses RL, Jacobs JJ (2001) Evaluation of metallic and
polymeric biomaterial surface energy and surface roughness characteristics for directed cell
adhesion. Tissue Eng 7:55–71
ISO 25178 part 2 (2012) Geometrical product specifications (GPS)—Surface texture: areal—part
2: terms, definitions and surface texture parameters. ISO
Jiang ZX, Huang YD, Liu L, Long J (2007) Effects of roughness on interfacial performances of
silica glass and non-polar polyarylacetylene resin composites. Appl Surf Sci 253:9357–9364
Maxwell AS (2001) Review of test methods for coating adhesion. NPL Report MATC (A) 49
Menezes PL, Kishore A, Kailas SV (2006) Studies on friction and transfer layer: role of surface
texture. Tribo Lett 24:265–273
Schröder H, Arndt-Staufenbiel N, Cygon M, Scheel W (2003) Planar glass waveguides for high
performance Electrical-Optical-Circuit Boards (EOCB)—The glass layer concept. In:
Proceedings of the 53rd electronic components and technology conference (May), New
Orleans, pp 1053–1059
Shahid M, Hashim SA (2002) Effect of surface roughness on the strength of cleavage joints. Int J
Adhes Adhes 22:235–244
Starikov D, Boney C, Pillai R, Bensaoula A, Shafeev GA, Simakin AV (2004) Spectral and
surface analysis of heated micro-column arrays fabricated by laser-assisted surface
modification Infrared. Phys Technol 45:159–167
Takadoum J, Houmid Bennani H (1997) Influence of substrate roughness and coating thickness
on adhesion, friction and wear of TiN films. Surf Coat Technol 96:272–282
Tseng AA (2007) Recent developments in micromachining of fused silica and quartz using
excimer lasers. Physica Status Solidi A204:709–729
Zappone B, Rosenberg KJ, Israelachvili J (2007) Role of nanometer roughness on the adhesion
and friction of a rough polymer surface and a molecularly smooth mica surface. Tribo Lett
26:191–201
Zhang XM, Yue TM, Man HC (1997) Enhancement of ceramic-to-metal adhesive bonding by
excimer laser surface treatment. Materials Lett 30:327–332
Chapter 13
Inspection of Laser Structured Cams
and Conrods
Franz Helmli and Stefan Lehmann
Abstract The inspection of laser structured cams and conrods has become an
essential factor in the manufacturing processes of the automotive industry.
Important measures of quality are the surface characteristics of these components,
especially the roughness. Efficient and accurate measurements of these charac-
teristics become essential in the mass production of automobiles. In this chapter, a
fast and fully automatic method is presented. The proposed method takes advan-
tage of the capabilities of focus variation technology to accomplish fast areal
measurements and to measure both topography and surface roughness in one step.
The basic focus variation measurement system is enhanced with a unique sample
holder for cams and conrods. A special mirror enables the system to measure the
inner surface of these components. Automatic form removal and parameter cal-
culation deliver the means to efficiently report the quality of the manufactured
parts. This method reduces the inspection time significantly. Where conventional
approaches, such as tactile measurements, can take hours to accomplish one
inspection involving manual analysis, the proposed method completes an inspec-
tion, including automated analysis, within a few minutes.
13.1 Introduction
13.1.1 Problem Description
Cams and connecting rods (or conrods) are essential parts of every automobile
engine (see Fig. 13.1). Their quality is, amongst other factors, responsible for the
functionality, reliability and durability of an engine.
F. Helmli (&) S. Lehmann

Alicona Imaging GmbH, Dr.-Auner-Straße 21a, 8074 Raaba/Graz, Austria
e-mail: Franz.Helmli@alicona.com

322 F. Helmli and S. Lehmann
Fig. 13.1 Design of an

engine. 1 Cam of exhaust
vent. 2 Water jacket for
coolant flow. 3 Spark plug.
4 Cam of intake vent.
5 Exhaust and intake valves.
6 Piston. 7 Conrod.
8 Crankshaft
Fig. 13.2 Connecting rod

(conrod)
A conrod (see Fig. 13.2) serves as a connector between the oscillating move-
ment of the piston and the rotation of the crankshaft. Together, the conrod and
crankshaft translate linear motion into rotary motion, which is transmitted to the
gearbox. In the gearbox, the motion is transferred to the drive wheels of the
automobile.
Part of the rotation of the crankshaft is reused to regulate the rotation of
the camshaft. The cams (see Fig. 13.3), which are part of the camshaft, retranslate
the rotary motion of the crankshaft into the oscillating motion used to operate the
intake and exhaust valves of the cylinders. These valves control the intake of
the air/fuel mixture and the ejection of exhaust gases from the cylinder.
Fig. 13.3 Cam

13 Inspection of Laser Structured Cams and Conrods 323
Fig. 13.4 Thermal joining of

cam and camshaft
Recent years have seen some changes in production of these parts, for example
instead of casting a camshaft in one piece, it is assembled from several parts.
These changes allow reductions in cost and weight as well as higher flexibility in
production. However, such innovations also demand more sophisticated approa-
ches in manufacturing and inspection of those parts.
While cams are mounted onto the camshaft, conrods are fixed to the crankshaft.
For both constructions, it is essential that there is adequate linkage between the
matching surfaces to avoid unwanted shifts, especially when the engine runs at
high speed. Even the smallest displacement between the surfaces may compromise
the operation of the engine.
Therefore, thermal joining (see Fig. 13.4) is applied, which results in a very
stable assembly. The quality of the micro-structures of the matching surfaces is the
key to make these constructions feasible.
One of the latest emerging technologies, the laser texturing technique, is
increasingly being applied to produce appropriate micro-structures. Laser texturing
enables the manufacturing processes to meet the requirements raised by the
described assembly routines.
Accordingly, the inspection of the surface characteristics of cams and conrods
is essential in the manufacturing processes of the automotive industry. At the same
time, the manufacturing processes face new challenges which must be overcome in
order to provide an efficient and feasible inspection solution.
13.1.2 Subject of Measurement
The subjects of the inspection described in this chapter are the laser textured
micro-structures of the inner surface of cams and conrods. These surfaces are
decisive in allowing an adequate seal for the thermal joining of cams and camshaft,
or conrods and crankshaft respectively.
Parallel grooves are induced by the laser texturing in four areas around the inner
surface. Each area is split into six separated regions (see Fig. 13.5). As a by-product
Fig. 13.5 A laser textured

region of a cam. Each region
is split into six separate areas
Fig. 13.6 Measurement

showing the laser grooves
and the ridges (coloured)
of the laser texturing, ridges arise at the borders of the grooves (see Fig. 13.6).
During the thermal joining of the cam and camshaft, these ridges penetrate the
relatively soft surface of the camshaft and create an additional anchorage. The
surface topography of these ridges and the shape of the inner surface fundamentally
affect the stability of the joining. The same effect applies to the joining of conrods
and crankshaft.
Several parameters have been determined to describe the characteristics of the
grooves and ridges in a suitable manner. On the one hand, the width and distance
of the grooves are measured and on the other hand, ridge height and width are the
subject of inspection. Additionally, the waviness of the turning grooves, which
result from the boring, is measured. All the measured parameters describe
important characteristics of the ridges of the laser textured grooves. These
parameters have to be measured on the four laser textured regions, located at
ninety degree intervals around the inner surface.
13.1.3 Problems with Conventional Inspection Techniques
While conventional inspection has been successfully implemented, improved

manufacturing and the accompanying application of laser texturing demands
refined inspection methods (Zhengde 2010).
Conventional inspection approaches involve tactile measurements, which can

have several drawbacks. Firstly, the parts that are to be measured have to be cut up
and subsequently cannot be used. Secondly, tactile measurements take a large
amount of time, even to measure a small number of profiles. Therefore, inspection
becomes inefficiently slow. Thirdly, the inspection results are, due to the small
number of measured profiles, unreliable. Also, unrepresentative outlier measure-
ments may result in an averaged value that lies outside the tolerances, even when
the measured part fulfills the requirements.
13.1.4 Proposed Solution
In order to overcome the drawbacks of conventional methods, a new measurement

method is proposed based on focus variation (FV) technology (see Leach 2011,
ISO/WD 25178-6 2010). FV has the advantage of a reduced inspection time along
with a large number of measureable profiles that can be acquired within a single
measurement.
Another preferable aspect of FV is that the measurement of the produced parts
is non-destructive. While this is not trivial in this particular application, a solution
has been found by implementing a dedicated hardware setup, including a mirror,
for measurement.
The sample holder is able to hold cams and conrods at a defined location. The
inner surface of these parts is orientated perpendicular to the scanning axis of the
FV system. Inspection is realised using a mirror that enables the system to measure
the inner surface.
13.2 Hardware Setup
FV combines the small depth of focus of an optical system with vertical scanning
to provide topographical and colour information from the variation of focus. One
of the main benefits of FV technology is its ability to carry out fast, areal and non-
contact measurements of a wide range of specimens.
For the inspection of cams and conrods, an objective with 509 magnification is
appropriate, which allows vertical resolutions down to 20 nm and sampling dis-
tances of approximately 180 nm. An integrated and changeable mirror allows the
measurement of the inner side of the cams and conrods without the need to cut
them. The measurement setup is shown in Fig. 13.7. The measurement system is
enabled to focus on the sample via the mirror, which redirects the system’s
viewing direction by 90°.
By means of automatically rotating the mirror, any region on the inner surface
of a cam or conrod can be measured without adjusting the sample manually. Thus,
the samples can be measured automatically.
Fig. 13.7 Cam placed on

sample holder with mirror
13.2.1 Considerations for the Optical Setup
The use of a mirror to allow the measurement of the inner surfaces of cams and
conrods poses some challenges for the optical setup that have to be considered.
The constructed system restricts the accessible measurement positions on the inner
surface of a sample. Limitations are caused by the numerical aperture of the
objective, its working distance and the geometry of the sample.
At a particular position, the edge of the sample will shadow part of the objec-
tive, as illustrated in Fig. 13.8. The closer a measurement position on the sample
approaches the bottom of the sample, the larger the overlap on the objective
becomes. This overlap can only be tolerated to a certain extent, since it will
decrease the quality of the measurement.
Fig. 13.8 Optical setup

showing the shadowing
of the objective
With reference to Fig. 13.8, the half aperture angle a of the objective can be
obtained from the numerical aperture NA , since.
NA ¼ sin a:
Together with the working distance WD; the numerical aperture describes the
visibility cone of the objective. The base radius r of the cone is given by
r ¼ WD tan a:
The focus position is limited by the working distance, meaning that y ¼ a þ b
may not exceed WD: If the mirror is higher than the measured sample, the focus
position will be limited by the distance of the objective to the mirror.
Additionally, the closer a measurement position is to the bottom of the sample,
meaning a decrease of a and a corresponding increase of b, the larger the clipping
of the edge u1 of the sample, since
u1 ¼ x a
with x ¼ y tan a being constant for lateral movement of the FV system. Along with
an increase of u1 ; an increased overlap u2 is created on the objective, since
r1 a
u2 ¼ r r1 and ¼ :
r x
Appropriate experiments help to determine which values can be allowed for the
overlap u2 to obtain suitable measurements.
The measurement setup utilises an objective with a numerical aperture of 0.55
and a working distance of 13 mm, allowing an overlap u2 of 1.22 mm on the
objective.
13.3 Cam and Conrod Measurement
Each measurement of a cam or conrod starts with the positioning of the sample on
the sample holder. Samples with different diameters can be measured. For each
possible diameter, there is a specific mirror that is attached in a fixed position. The
system is approximately focused to the first desired measurement position and the
measurement begins. A measurement consists of the data acquisition, the pro-
cessing of data, which includes a form removal step to prepare the data for analysis
and, finally, the calculation of the parameters which determine whether the mea-
sured specimen satisfies the specification. As soon as a sample is positioned, data
acquisition, processing and analysis are conducted automatically, providing the
operator with a finished assessment of the measured sample.
Fig. 13.9 Measurement

areas in a laser textured
region on a cam
Fig. 13.10 Measured dataset

of the laser textured structure
on the inner surface of a cam
13.3.1 Automatic 3D Data Acquisition
Each of the four laser textured fields of interest has to be measured for inspection.
A measurement of the whole area would be inefficient and unnecessary. Instead,
certain areas are defined for each laser textured field, which are representative of
the quality of the whole field. These fields are located in different regions of each
laser textured field (see Fig. 13.9).
At each measurement position an image-field measurement is accomplished,
meaning that overlapping measurements are taken and combined into a single
dataset which effectively extends the field of view of the system. These image-
fields typically contain three laser textured grooves for this application. These
measurements are repeated for each of the four laser fields. The result of one
measurement is shown in Fig. 13.10.
13.3.2 Form Removal
The system measures both the macro-scale form of a sample and the micro-scale
structure of its surface. The parameters of interest, as part of the micro-structure,
can only be evaluated if the macro-scale form can be removed from the acquired
datasets.
Fig. 13.11 Dataset with

cylindrical shape and turning
grooves
The form of the sample is affected by two different influences: the concavity of
the cylindrical shape and the turning grooves that remain from the boring
(Fig. 13.11). Both influences need to be removed.
The form induced by the cylindrical shape and the turning grooves can be
described by a specific mathematical model. This model includes a polynomial fit,
which describes the cylindrical shape of the measured surfaces and a Fourier series
(Schwarz 1997) that describes the periodic shape of the turning grooves.
Since the cylindrical shape does not match a perfect cylinder, a polynomial
surface of second order is approximated to describe the curvature of the inner
surface. A polynomial surface is a multi-dimensional polynomial with more than
one argument.
A Fourier series is generally used to decompose periodic functions, such as the
form of the turning grooves, into the sum of a set of sine and cosine functions.
With a Fourier series, a periodic function f with period T [ 0 can be approximated
by a finite set of sine and cosine functions whose frequencies are integer multiples
of the base frequency x ¼ 2p=T: The approximated function fn is a trigonometric
polynomial of the form
a0 X n
f n ðt Þ ¼ þ ðak cosðkxtÞ þ bk sinðkxtÞÞ:
2 k¼1
A least-squares solution for the coefficients ak and bk provides a close approxi-

mation for the function f : The higher the number of summands n; the closer the
approximation will be.
The final approximation of the macro-scale form of the surface, including both
the curvature of the inner surface and the form of the turning grooves, can then be
removed. The result is a leveled, form-reduced dataset that contains only the
micro-structure of the measured data, as shown in Fig. 13.12.
13.3.3 Surface Measurement
Once the form is removed from the dataset, the characteristics of the surface can
be measured to investigate its suitability for the thermal joining process. In the
Fig. 13.12 Left form dataset. Right form-reduced dataset
following sections, two possibilities for the measurement are proposed. The first
possibility relies on the extraction of profiles from the surface, and the second
possibility involves the determination of areal surface texture parameters.
13.3.4 Profile Extraction
There are two different types of profile that are extracted from the surface. The first
type is used to characterise the ridges. These profiles are positioned orthogonal to
the laser textured grooves and on top of the turning grooves, since only the ridges
on top of the turning grooves influence the stability of the thermal joining (see
Fig. 13.13). The characteristics of the ridges at the bottom and the slope of
the turning grooves are not relevant. The correct positioning of the profiles in the
form-reduced dataset is determined in the form dataset, where the mean of the
Fourier series polynomials is used to determine the peaks of the turning grooves.
From all profiles that are extracted from one measurement, a mean and a maximum
profile is calculated and used for parameter evaluation.
Fig. 13.13 Left profile on top of a turning groove orthogonal to the laser structure. Right profile
in between the laser textured grooves and orthogonal to the turning grooves
The second type of profile is used to characterise the turning grooves. These
profiles are positioned in between and parallel to the laser textured grooves
(Fig. 13.13). The width of the profiles has to be less than the distance between the
laser textured grooves. Over the width of the profiles, a large number of adjoining
profiles is summarised to a mean profile to ensure a robust parameter calculation.
13.3.5 Parameter Calculation from Profiles
Once the profiles have been extracted from the surface, parameter calculations can
take place. The calculated parameters (see Table 13.1) can be divided into three
groups.
The first group of parameters describes the character of the ridges of the laser
textured grooves: the maximal ridge height GRs on the peaks of the turning
grooves, the mean maximal ridge height GRsm; which is the mean of the maximal
ridge height of each ridge in the measured field, and the mean ridge height in the
measured field GRm: These parameters are defined thus,
1X n
1X n
GRm ¼ pi ; GRsm ¼ pmax i and GRs ¼ Maxðpmax i Þ; i ¼ 1. . . n:
n i¼1 n i¼1
These three parameters are analysed to evaluate whether the characteristics of the
ridges are suited for the purpose of strengthening the thermal joining of cam and
camshaft, and likewise, conrod and crankshaft. These parameters are measured
from the slim profiles on top of the turning grooves, which are orthogonal to the
laser textured grooves (Fig. 13.14).
The second group of parameters characterises the laser textured grooves—the
mean distance LZa and the mean width of the laser grooves LZb (see Fig. 13.14).
The parameters are defined thus
1X n
1X n
Wsm ¼ Wsi and Wz ¼ Wzi :
n i¼1 n i¼1
Table 13.1 Short description Parameter Description

of the measured parameters
GRs Maximum ridge height
GRsm Mean maximum ridge height
GRm Mean ridge height
LZa Mean laser line distance
LZb Mean laser line width
Wz Mean peak to valley turning groove height
Wsm Mean spacing of turning groove peaks
Fig. 13.14 Profiles

orthogonal to the laser
textured grooves
These two parameters are of interest since they describe the distances between
the ridges on the borders of the laser textured grooves. An appropriate density of
ridges is necessary to make the joining robust against production tolerances of the
form of the manufactured parts. A lower density of the ridges may result in an
unstable joining in areas where the cam and camshaft, or the conrod and crankshaft
respectively, do not closely fit. These parameters are measured from the same
profiles as the parameters from the first group for the same reason: only the ridge
characteristics on top of the turning grooves are relevant for the stability of the
thermal joining.
The final group of parameters characterises the turning grooves with the
parameters Wz; which describes the mean peak to valley height of the turning
grooves, and Wsm; which describes the mean spacing of turning groove peaks (see
Fig. 13.15). Since the laser has a fixed focal width, it cuts nearly consistently deep,
regardless of the waviness of the turning grooves. Therefore, a high waviness will
result in irregular laser textured groove ridges since more material will be accu-
mulated on some positions and less on others. In short, by measuring the waviness
of the turning grooves the uniformity of the ridges of the laser grooves can be
evaluated. As mentioned before, the parameters of the third group are measured on
the profiles in between and parallel to the laser textured grooves.
Finally, the evaluated parameters can then be used to classify a sample as
adequate or as insufficient. The appropriate tolerances have to be defined by the
user of the inspection solution.
Fig. 13.15 Profile parallel to

the laser textured grooves in a
region that has not been laser
textured
Fig. 13.16 Bearing area

curve with parameters Sk,
Spk, SMr1 and SMr2
13.3.6 Areal Surface Texture Parameter Calculation
The evaluation of the laser textured areal surface characteristics relies on the
determination of the parameter Spk, as described in Chap. 2 (ISO 25178 part 2
2012). The tolerance limits have to be defined by the user of the inspection solution.
The reduced peak height Spk serves as a measure of the peak height above the
core roughness of the surface. The core roughness is described by the core
roughness depth Sk which is the peak-to-valley height with the predominant peaks
and valleys removed (see Chap. 2). Both Sk and Spk are derived from the bearing
area curve of the form-reduced dataset, as shown in Fig. 13.16. Referring to
Fig. 13.16, SMr1 gives the percentage of the peak structures that are not associated
with Sk. Accordingly, the value SMr2 gives the percentage of the valley structures
not associated with Sk.
The parameter Sk is determined by fitting a linear curve to the bearing area
curve into the central 40 % of the measurement points. The central area lies where
the slope of the secant over DSMr = 40 % is a minimum. The Sk parameter is the
height difference between the cutting points of the fitted line with the ordinate at
SMr1 = 0 % and SMr2 = 100 %.
Figure 13.17 shows a pseudo-coloured surface dataset, where only the height
values above the range of Sk are shown (meaning the areas that are regarded as
peaks). Figure 13.17 shows that the peak areas coincide well with the ridges that
are of interest here.
The parameter Spk is determined on the whole form-reduced surface dataset.
The resulting value, including the irrelevant regions at the bottom of the turning
Fig. 13.17 Pseudo-colour

dataset reduced to the height
values below SMr1
grooves, appropriately corresponds to the desired characteristic of strong thermal

joining ability. Such areal surface measurements could be advantageously applied
for measurements of cams or conrods that have curved laser textured structures.
13.4 Assessment of the Solution
A solution as proposed in this chapter can only be of practical use if it provides

accurate as well as repeatable results. The accuracy of the results can be dem-
onstrated by means of calibration. A variety of traceable calibration artefacts have
been used to calibrate the measurement system and, thereby, verify the accuracy of
its measurements. It is also essential to gain comparable results repeatedly. Rep-
resentative for the cam and conrod measurements, a repeatability study has been
carried out on a cam to verify the stability and consistency of the measurements.
The repeatability study was designed to allow an assessment in accordance with
the Measurement System Analysis Method (MSA 2002). The MSA is a widely
used method to assess the quality of measurement systems. It consists of five
methods, of which the first is shown here. This first method focuses on the
capability of a measurement process with respect to the variation and position of a
measured value in the tolerance zone of the measured feature. The capability index
Cg can be determined to evaluate the achieved repeatability of the measurement
system. The capability index is calculated as
0:2T
Cg ¼
4r
where T is the allowed tolerance and r is the standard deviation of the measured
values. A measurement process can be considered capable with respect to its
repeatability if the condition Cg 1:33 is fulfilled.
To assess the repeatability of the proposed measurement setup, twenty-five
measurements of a reference cam have been taken by a single user under constant
conditions. The results are shown in Table 13.2.
The resulting Cg values prove the capability of the proposed measurement setup
to accomplish measurements of cams and conrods with the required repeatability.
Table 13.2 Assessment of the repeatability of the proposed measurement setup

Parameter xg (lm) T (lm) r (lm) Cg
GRs 8.37 5.0 0.120 2.09
GRsm 3.23 3.5 0.011 15.73
GRm 1.25 3.8 0.005 38.94
LZa 93.58 10.0 0.047 10.66
LZb 48.14 20.0 0.081 12.33
Wz 1.80 3.0 0.025 6.11
Wsm 108.97 60.0 0.153 19.67
13.5 Conclusion
A solution for the inspection of cams and conrods in the automotive industry has
been proposed. This solution meets new requirements, which arose from refined
manufacturing processes employing previously-unused technologies, such as laser
texturing. The new measurement system succeeds in solving an inspection task
which exceeds the capabilities of conventional solutions. Key to the improvement
is the use of focus variation technology. The advantages of FV and the automation
of measurements has helped to reduce the inspection time and cost.
A repeatability study has demonstrated the capability of the inspection solution
to acquire measurements that fulfill the demands of an industrial inspection
solution. For an even wider capability assessment, a full Gauge Repeatability and
Reproducibility study would need to be carried out.
References
ISO 25178 part 2 (2012) Geometrical product specifications (GPS)—surface texture: areal—part
1: terms, definitions and surface texture parameters. International Organization of
Standardization
ISO 25178-6 (2010) Geometrical product specifications (GPS)—Surface texture: areal—part 6:
classification of methods for measuring surface texture. International Organization of
Standardization
AIAG - Chrysler Corp, Ford Motor Co, General Motors Corp (2002) Measurement Systems
Analysis, Reference Manual, 3rd edn. Michigan, USA
Schwarz H R (1997) Numerische Mathematik. B G Teubner, Stuttgart
Zhengde Z (2010) Application analysis of 2D measurement and 3D measurement for surface
microstructure. In: J 14th national automobile test technology annual conference, Huangshuan
Mountain, China, 71–79 June
Chapter 14
Road Surfaces
Alan Dunford
Abstract Adequate road surface friction is essential for the prevention of skidding
accidents. In order to maintain a safe and serviceable road network, many coun-
tries routinely monitor friction by direct measurement with a sliding tyre. Better
methods for characterising the features of the road surface texture that are related
to friction could lead, through more efficient monitoring and specification, to
improved road safety. A robust methodology has been developed for inspection of
the changing surface texture of road stones as they undergo simulated traffic
polishing. Areal measurement and characterisation of surface roughness are likely
to be essential for understanding complicated road/tyre interactions.
14.1 Background
The geometric profile of a road in the vertical plane has traditionally been divided
into different scales depending on the dynamic response of interest to vehicles or
vehicle passengers, for example, ride comfort, safety, rolling noise and vehicle
operating costs. After the 1987 congress, PIARC (World Road Association)
specified these divisions, in descending order of wavelength, as unevenness,
megatexture, macrotexture and microtexture (PIARC 1987). In general, for good
road-holding, large amplitudes are desirable when wavelengths are below 10 mm
and small amplitudes are desirable for wavelengths above 10 mm (i.e. low
unevenness and megatexture give a more comfortable ride, whereas high macro-
texture and high microtexture contribute to safer road-holding).
Microtexture, with wavelengths less than 0.5 mm, is present on the surfaces of
coarse aggregate particles or the texture of bituminous mortar and fine material in
A. Dunford (&)
TRL, Crowthorne House, Nine Mile Ride, Berkshire, RG40 3GA, UK
e-mail: adunford@trl.co.uk

338 A. Dunford
asphalt surfaces, and is associated with good skidding resistance on wet roads
(PIARC 1987). The macrotexture (typically due to the gaps in between aggregate
particles in the surface and measured in millimetres) can affect the rate at which
wet skid resistance reduces with increasing speed. It is generally agreed that, under
wet conditions, macrotexture is required so that water can be removed as quickly
as possible from the path of vehicle tyres, while the microtexture is required to
break through the surface of any remaining film of water so that contact with the
tyre can be retained (Moore 1975). It is expected that polishing by traffic pre-
dominantly affects the microtexture of the surface.
The mechanism by which friction is generated at the tyre/road interface is
complicated and its dependence on the roughness of the surface (at any wave-
length) is not well understood. Several attempts have been made to characterise
surface microtexture in such a way that it can be directly related to friction (Chen
and Wang 2011; Do et al. 2009). Due to the complexity of the interaction and the
variability of the surface (aggregates typically having random, naturally formed
surface texture), conclusive results have not been demonstrated.
Do et al. (2009) describe an experiment where two types of aggregate were
polished in the laboratory and surface profile measurements were taken at three
stages. Characterisation of surface roughness was undertaken using Rq and a
proprietary, two dimensional, calculation of the sharpness of surface asperities.
A weak relationship between roughness and friction was developed for the three
stages of polishing and an interpretation of the mechanism by which the aggregate
polished was suggested, which is dependent on the mineral composition of the
aggregate.
Chen and Wang (2011) also made measurements on aggregate specimens
polished in the laboratory. For three different aggregates, profile measurements
were made at seven stages during the polishing cycle. The measurements were
analysed using fractal and spectral approaches (see Chap. 6). It was concluded that
the evolution of friction is governed by a polishing mechanism whereby global
surface height reduction is accompanied by differential wear of individual minerals
depending on their hardness.
The experiment described in the case study in this chapter proposes an
improved methodology for examination of the surface texture of aggregate par-
ticles, in three dimensions, as they are polished in the laboratory. A Wehner-
Schulze machine (see Sect. 14.2.2) was used to abrade the surface of a sample of
aggregate particles in a progressive fashion. At five stages of polishing, the
aggregate particle surfaces were examined using an optical, areal surface topog-
raphy measurement system. The results allow intuitive and detailed exploration of
the link between abrasion and surface texture, and have the potential to provide a
clearer understanding of the link between surface texture and friction. Use of areal
measurements of surface topography (1) increases confidence that the same area is
inspected during surface evolution; (2) allows a more intuitive understanding of
the changes that can be observed; and (3) potentially improves the statistical
significance of surface characterisation since more data is analysed than with
profile measurements.
14 Road Surfaces 339
14.2 Methodology
14.2.1 Aggregate Selection
The aggregate used in the surface course of road pavements is polished by the
action of traffic, and different aggregates polish by different amounts depending on
their geology and the effect of passing traffic. At locations where the wearing effect
of traffic is greatest, such as bends and approaches to traffic lights, a more polish-
resistant aggregate is required so that skid resistance is retained at a higher level
(Hosking and Woodford 1976).
The specification of polish resistance of aggregates is achieved using a labo-
ratory-based test which measures the skid resistance provided by an aggregate
surface after it has undergone a standardised amount of polishing, designed to
simulate the polishing that would be experienced on the road. In the UK, the
polished stone value (PSV) test is used (BS EN 1097 part 8 2009); aggregates with
a higher PSV are likely to provide higher long-term skid resistance when the
aggregate is used in the surface course of a road. An alternative, but less well-
established method for simulating the polishing action of traffic in the laboratory,
is the Wehner-Schulze procedure (Woodbridge et al. 2006).
As in the work of Do et al. (2009), limestone was chosen as the subject
aggregate. Limestone is not used as road surface course aggregate in the UK
because it has a low resistance to polishing. It is this low resistance to polishing
(and, therefore, the wide range of friction levels observed when it is polished in the
laboratory) that makes limestone a good initial candidate for development of the
methodology.
Limestone is typically very homogenous, predominantly composed of a single
mineral (calcite), with only small amounts of secondary minerals such as quartz or
feldspars. A petrographic analysis of limestone taken from the same quarry as the
limestone used in this experiment states that the ‘‘limestone is composed of robust
fragments that resist breakage; pits and cavities are uncommon’’. Accordingly, the
surface texture of limestone should be amongst the simplest to interpret of any
aggregate type.
A test plate was prepared for use in the Wehner-Schulze machine (see Sect.
14.2.2). The test plate consists of limestone aggregate particles arranged in a
circular mosaic so their flat surfaces are uppermost. Aggregate particles were
sieved so that they passed through a 10 mm square sieve, but were retained on a
flake sorting sieve with a slot width of 7.2 mm and a slot length of 40 mm. This
ensures the particles have approximately square faces and that they are relatively
uniform in size—nominally 10 mm. The aggregate particles are held rigidly in a
matrix of epoxy resin and sand, as shown in Fig. 14.1.
340 A. Dunford
Fig. 14.1 Limestone

aggregate test plate
14.2.2 Wehner-Schulze Machine
The Wehner-Schulze (W-S) test equipment was developed during the 1960s in
Germany, at the Technical University of Berlin, as an alternative laboratory test
procedure for assessing the polishing of aggregates in road surfaces. The procedure
is designed to simulate accelerated polishing on road surfacing materials, and
measure the friction associated with the sliding of rubber blocks across the speci-
men before and after that polishing. Testing is carried out using a purpose-designed
machine that is available commercially (Fig. 14.2). The machine meets the spec-
ification outlined in a draft European Standard (BS EN 12697 part 49 2011).
Polishing is achieved in the W-S machine by lowering a polishing head con-
sisting of three independently suspended conical rubber rollers so that they are
forced into contact with the test surface. The polishing head is rotated at a speed of
500 rpm in a clockwise direction, while a suspension of silicon dioxide in water is
pumped onto the surface to act as a polishing medium. The polishing head has a
mass of 40 kg and, according to calculations made elsewhere (Huschek 2004), this
results in each roller being independently loaded onto the test surface with an
average contact pressure of approximately 0.4 N mm-2, equivalent to 4 bar
(58 psi), typical of the tyre pressures of a commercial vehicle. The same research
has suggested that, although the rollers are free to rotate, there is some drag due to
the mounting bearings, giving a slight slip of 0.5–1.0 %.
Friction is calculated from measurements of the torque imparted to the test sur-
face when the test head, comprising three separate rubber sliders, and rotating at a
pre-defined speed, is dropped onto the surface and allowed to slide to a halt under its
own mass. Just before the head is dropped onto the test surface, water is sprayed on to
Fig. 14.2 A Wehner-Schulze machine
the test surface at a rate of twenty litres per minute to attain a theoretical water film
thickness of 0.5 mm until the test has finished. The test head has a mass of 25 kg and
it has been calculated that each rubber slider imparts a static pressure of approxi-
mately 0.2 N mm-2 equivalent to 2 bar (29 psi), in tyre pressure (Huschek 2004).
14.2.3 Progressive Polishing Regime
The aggregate test plate was polished through a series of stages, allowing mea-
surement of surface texture and friction at five different ‘stress levels’. Initial
surface texture was measured before the first measurement of friction. In the
standard W-S procedure, the friction test head is accelerated to a tangential speed
of 100 km per hour before it is dropped onto the test surface. During the exper-
iment described here, an initial speed of 50 km per hour was used so that any
polishing associated with abrasion of the aggregate during the friction test itself
was minimised. For this experiment the friction reported, lPWS20, was measured
when the friction test head had slowed to a tangential speed of 20 km per hour.
342 A. Dunford
Fig. 14.3 Friction measured

on limestone test specimen as
a function of polishing in the
W-S machine
Following the initial measurements, the aggregate specimen was polished only
a small amount before the surface texture was measured again. Friction was also
measured for a second time before a further period of polishing in the W-S
machine. This process was repeated until surface texture and friction measure-
ments had been made at five distinct stress levels. The amount of polishing applied
at each stage was calculated to achieve a similar change in friction between the
stress levels. Experience has shown that the evolution of friction for aggregate
surfaces when polished in the laboratory (and indeed when polished by traffic on
the road) is approximately logarithmic. Five sets of surface measurements were
taken after 0, 90, 900, 9,000 and 90,000 polishing rollers had passed over the
specimen. Note that 90,000 passes is the standard amount of polishing applied
using the W-S machine. Figure 14.3 shows the friction measured at each stress
level and the number of roller passes is indicated on a logarithmic scale at the top
of the graph (for display purposes stress level zero has been shown with one
polishing roller pass instead of zero passes).
14.3 Measurement of Surface Texture
Surface texture measurements were made after each stage of polishing using an
Alicona InfiniteFocus focus variation microscope (Leach 2011) at the National
Physical Laboratory, UK. The microscope combines the small depth of field of an
optical system with vertical scanning to provide topographical information derived
from the variation in focus. Using a beam-splitting mirror, light emerging from a
white light source enters into the optical path of the system and is focused onto the
specimen via an objective lens. Light rays reflecting from the specimen and
entering the objective lens are gathered by a light sensitive sensor behind the
beam-splitting mirror. Due to the small depth of field of the optics, only small
regions of the specimen are sharply imaged when the lens is at any given height.
The lens is moved vertically along the optical axis while continuously capturing
data from the surface until each region of the specimen has been imaged in sharp
focus. Algorithms are used to analyse the variation of focus along the vertical axis
and convert the acquired data into 3D topographical information.
The measurement process is fast compared to line scanning profilometry, but
still not fast enough to practically capture high resolution information for the entire
specimen surface. Therefore, two stones in the test specimen were chosen at
random and these were scanned at each stress level. Surface texture on each whole
stone was measured first, at a relatively low (stated) resolution of 7 lm laterally
and 1 lm vertically, before three areas on the highest parts of each stone (i.e.
regions where contact with the polishing rollers would be expected first) were
chosen and characterised at three magnification levels. According to the operating
software supplied with the microscope, using the 59 objective lens achieves a
Fig. 14.4 Topographical map of Stone 1 and three areas within it selected for closer inspection,
stress level 0 and at 59 magnification
344 A. Dunford
resolution of 1.75 lm laterally and 0.5 lm vertically, while the 109 objective lens
achieves 881 nm laterally and 124 nm vertically, and the 209 objective lens
achieves 438 nm laterally and 59 nm vertically. The number of measured points is
the same at each magnification (1,628 points by 1,236 points) and the actual size of
the area measured is, therefore, reduced accordingly.
Figure 14.4 shows the height variation across the surface of Stone 1 and the
three areas selected for closer inspection are indicated. These images, all further
topographical images and the analysis reported in this chapter were produced using
MountainsMap Universal software from Digital Surf.
In order to measure the same areas at each stress level, the specimen was placed
under the microscope with approximately the same orientation, and the coordi-
nates of corners and edges for each stone were used for navigation to the same
position. The photographic images recorded for the previous stress level were then
used to locate the same area. Measurement in 3D, and use of photographic images
at each stage of polishing in this way, greatly aids the process of aligning the
specimen surface. Careful alignment of the surface allows direct inspection of the
effect of abrasion in the laboratory. Correspondingly, quantitative analysis of
surface texture is more intuitive when the whole surface is considered.
14.4 Analysis
14.4.1 Qualitative Changes in Surface Texture
Figure 14.5 shows the surface of Stone 1 before and after polishing (i.e. Stress 0
and 4). There are high planes and ridges fanning out from the top right of the stone
surface before polishing; after polishing, these areas are less well defined and, in
places, distinctly broader.
Figure 14.6 shows the difference, at each point on the surface of Stone 1,
between the heights at Stress 0 and at Stress 4. The differences were calculated by
the microscope’s operating software after the two sets of measurements had been
automatically aligned using a least-squares best fit. The largest changes in height
indicated in Fig. 14.6 correspond to the highest points on the stone (shown in
Fig. 14.5).
14.4.2 Characterisation of Surface Texture
On each of the two stones, three areas at the highest parts were chosen for closer
inspection at 5, 10 and 209 magnification. Areal surface texture parameters were
calculated for each of the six areas at each magnification (ISO 25178 part 2 2012).
In addition, calculations were made after application of a Gaussian filter with
various nesting indices: 0.025, 0.08, 0.25, 0.8 and 2.5 mm, limited by the size of
Fig. 14.5 Topographical map of Stone 1 before and after polishing
Fig. 14.6 Map showing the

change in height in Stone 1
before and after polishing
the measured area. After application of the filter, the resultant surface retains only
those height variations with wavelengths less than the nesting index.
If the average roughness at each stress level is compared to the corresponding
friction measured at each stress level, some strong relationships emerge: various
parameters decrease with decreasing friction (or incrementally increasing stress
level). Figure 14.7 shows the average value of Sa calculated over all six areas,
without application of any surface filtering, for measurements made using the 209
objective lens, against friction measured. The range of roughness on the six areas
346 A. Dunford

between average roughness
(Sa) and test plate friction.
Sa calculated from
measurements made using
209 objective lens with no
filtering applied
is indicated with ‘+’ and ‘-’ symbols indicating the highest and lowest mea-
surements respectively. The relationship in Fig. 14.8 is derived from the same
measurements after they have been filtered using a 0.25 mm Gaussian filter.
Application of the Gaussian filter removes some of the largest height variation
on the surface, which results in lower values of Sa. The filter also apparently
emphasises the relative change in Sa with changing friction (the slope of the
relationship is steeper) and reduces the range of measurements from the six sep-
arate areas.
Similar relationships can be found at all magnifications, and after application of
most of the filters. However, the strongest can be found after application of a
0.25 mm filter to measurements made using the 209 objective lens for parameters
including height parameters Sa, as above, and Sq, and volumetric functional
parameters such as the core material and void volumes, Vvc and Vmc respectively.
An inverse relationship (increasing parameter with decreasing friction) can also
be shown for areal material ratio (the relative amount of area under the highest
peak), Smr.
These relationships are all consistent with observations of the surface and with
the interpretation that less deviation from a mean (or bulk) level results in lower
friction, perhaps because there is less physical interaction with the rubber sliders
used to measure friction.

between average roughness
(Sa) and test plate friction.
Sa calculated from
measurements made using
209 objective lens with
0.25 mm Gaussian filter
applied
14.5 Conclusions
The work described here was primarily focussed on development of a robust

methodology to investigate changes in the surface texture of road stones as they
undergo controlled polishing in the laboratory. Areal measurement of surface
texture is essential for confidence in the location accuracy of measurements when
attempting to observe changes at higher magnification (where edges of aggregate
particles cannot be used as datum points). Areal measurement also gives an
indication of the extent of physical changes that cannot be achieved using profile
measurement. The large amount of data collected is likely to improve the statis-
tical significance of calculations of surface roughness.
It has been shown that changes in roughness calculated using standard height
and volume parameters reflect observed physical changes in surface texture, and
that it is possible to relate these changes to expected friction behaviour. Filtering of
the data improves the relationships between roughness and friction and may permit
further investigation of the effect on friction of texture of different wavelengths, as
has been attempted by other researchers (Chen and Wang 2011; Do et al. 2009;
Yandell 1970).
It is expected that further analysis of the measurements made on the limestone
specimen will enable refinement of the methodology, which will be applied to
alternative, more geologically complicated, aggregates.
348 A. Dunford
Acknowledgments The work described in this chapter was supported by TRL Academy.
Thanks are due to Helen Viner of TRL, Tony Parry and Phil Shipway at the University of
Nottingham, Louise Caudwell and Ramesh Sinhal of the Highways Agency, and Richard Leach
and Claudiu Giusca of the National Physical Laboratory.
References
BS EN 1097 part 8 (2009) Tests for mechanical and physical properties of aggregates—Part 8:
determination of the polished stone value. British Standards Institute
BS EN 12697 part 49 (2011) Bituminous mixtures. Test methods for hot mix asphalt. Part 49:
determination of friction after polishing. draft. British Standards Institute
Chen XH, Wang DW (2011) Fractal and spectral analysis of aggregate surface profile in polishing
process. Wear 271:2746–2750
Do M-T, Tang Z, Kane M, de Larrard F (2009) Evolution of road-surface skid-resistance and
texture due to polishing. Wear 266:574–577
Hosking JR, Woodford GC (1976) LR738 measurement of skidding resistance. Part II. Factors
affecting the slipperiness of a road surface. TRL Ltd, Crowthorne
Huschek S (2004) Experience with skid resistance prediction based on traffic simulation. In: 5th
Symposium of pavement surface Characteristics, Toronto, Canada
ISO 25178 part 2 (2012) Geometrical product specifications (GPS)—Surface texture: areal.
Part 2: Terms, definitions and surface texture parameters. International Organization for
Standardization
Moore D (1975) The friction of pneumatic tyres. Elsevier, Amsterdam
PIARC (1987) Report of the committee on surface characteristics. XVIII World Road Congress,
Brussels
Woodbridge ME, Dunford A, Roe PG (2006) PPR144, Wehner-Schulze machine: first UK
experiences with a new test for polishing resistance in aggregates.TRL, Crowthorne
Yandell WO (1970) The measurement of surface texture of road stones with particular regard to
the effect on the frictional properties of road surfaces. University of New South Wales
Index
A B
Abbott-Firestone curve, 25 Bayes Theorem, 160
Abrasive paper, 170 Binary classification, 156
Accuracy, 158 Birmingham 14 parameters, 2
ACF. See autocorrelation function, 32 Bi-weight function, 82
AdaBoost. See adaptive boosting, 168 Blechmantel, 284
Adaptive boosting, 168 Bond strength, 304
Adhesion, 303 Boosting, 168
Aggregate particle, 338
Algebraic distance, 111
Algebraically defined surface, 108 C
Aliasing, 142, 187 Cam rod, 321
Alternating sequence filter, 97 Capability index, 334
American National Standards Institute, 10 Car liner, 299
Amontons’ law, 222 Cerium doped glass, 306
Anisotropy, 131, 142 Change tree, 51
ANSI. See American National Standards Chaos, 133
Institute, 10 Chebyshev approximation, 113
Areal feature, 184 Class label, 156
Areal material ratio, 23 Classification, 156
Areal scale fractal complexity, 147 Closed surface, 73
Areal surface topography, 4 CMM. See coordinate measuring machine, 313
Area-scale analysis, 143 Coherence scanning interferometer, 270, 293,
Area-scale complexity, 142 307
Arithmetic mean height of the five highest Compacity, 58
significant peaks, 57 Conditional independence, 167
Arithmetic mean height of the five lowest Conditional probability, 159
significant pits, 57 Connecting rod, 321
Arithmetic mean height, 20 Conrod. See connecting rod, 321
Aspect ratio, 58 Contact mechanics, 218
Auto-catalytic plating. See electroless plating, Continuous wavelet transform, 100, 251
309 Contour line, 47, 69
Autocorrelation function, 32 Coordinate decoupling, 192
Autocorrelation length, 34 Coordinate measuring machine, 313
Autocovariance, 33 Core fluid retention index, 40
Auto-partitioning, 59 Core material volume, 28

DOI: 10.1007/978-3-642-36458-7, Ó Springer-Verlag Berlin Heidelberg 2013
350 Index
Core void volume, 28 Filter bank, 102

Coulombic friction, 218 Filter, 126
Course line, 47 Filtration, 67
Crankshaft, 322 Finish honing, 251
Cross correlation coefficient, 161 Fitting criteria, 113
CSI. See coherence scanning interferometer, Fitting, 210
270, 293 FNR. See false negative rate, 157
Cut-off wavelength, 71 Focus variation, 307, 325, 342
Cutting height, 23 F-operation, 68
CWT. See continuous wavelet transform, 100 F-operator, 107
Cylinder liner, 247, 282 Form error, 116
Cylindrical filter, 73 Form removal, 107
Fourier series, 329
Fourier spectrum, 37
D Fourier transform, 70, 98, 218
Dale area, 57 FPR. See false positive rate, 157
Dale volume, 28, 58 Fractal analysis, 129
Dale, 48 Fractal dimension, 41, 130
Definition area, 17 Fractal, 129
Delaunay triangulation, 186 Freeform, 68
Density of motifs, 245 Function related parameter, 22
Density of summits, 55 Functionally defined surface, 108
Deterministic patterning, 5 FV. See focus variation, 325
Developed interfacial area, 30
Dilation, 95
Directional root mean square slope, 38 G
Discrete wavelet transform, 102 Gabor transform, 100
Discrimination threshold, 158 Gaussian cylindrical filter, 75
Distance between grooves, 284 Gaussian distribution. See normal
Distance measure, 111 distribution, 160
Documentary standard, 8 Gaussian filter, 73
Dynamic range, 189 Gaussian function, 73
Gaussian planar filter, 74, 75
Gaussian weighting function, 74
E Gauss-Network algorithm, 121
Elasto–plastic asperity model, 225 Geometric element, 108
Electroless copper plating, 303 Geometrical Product Specification and Verifi-
Empirical surface, 124 cation, 15
End effect, 79 Geomorphologic feature, 220
Entropy, 133 Global leveling, 198
Equivalent diameter, 59 GPS. See Geometrical Product Specification
Erosion, 95 and Verification, 15
Evaluation area, 17 Gridded sampling, 184
Exact proportion length-scale anisotropy, 142 Groove balance, 284
Excimer laser machining, 304 Groove interrupt, 284
Groove parameter, 284
Groove width, 284
F
False negative rate, 157
False positive rate, 157 H
Feature identification, 190 Height difference, 111
Feature inspection, 191 Height distribution, 22
Feature parameter, 15 Height function, 17
Feature verification, 191 Height parameter, 20
Index 351
Helical slide honing, 248 Least-squares mean plane, 197

Hertzian contact theory, 222 Least-squares, 113
Hertzian pressure, 218 Lngth-scale analysis, 138
High-pass filter, 69 Length-scale fractal dimension, 141
Hill area, 57 Lifting scheme, 103
Hill volume, 58 Linear filter, 69
Hill, 47 Linear Gaussian regression
Hill-dale segmentation. See morphological cylindrical filter, 84
segmentation, 201 Linear Gaussian regression
Hölder condition, 233 planar filter, 82
Honing angle, 248, 284 Linear interpolation, 19
Honing, 282 Linear spline filter, 90, 91
Hough transform, 284 Loss function, 72
Hurst exponent, 233 Low-pass filter, 69
Hybrid parameter, 29 LSM. See least-squares mean plane, 197
I M
IIP. See iterative closest points algorithm, 212 Macrotexture, 236, 337
IDWT. See inverse discrete wavelet transform, Magic number, 93
103 Main texture direction, 37
Individual areal feature, 180 Material ratio curve, 22
Influence function, 86 Material volume, 26
Intensity image, 185 Maximum height, 21
Internal combustion engine, 282 Mean motifs slope, 56
International Organization for Standardization, Measurement System Analysis Method, 334
8 Median operator, 195
Interpolating spline wavelet, 98 Megatexture, 337
Interpolating spline, 90 Mexican hat wavelet, 100
Inverse areal material ratio, 23 Micro-CMM, 181
Inverse discrete wavelet transform, 103 Microtexture, 236, 337
Inverse error function, 86 Midpoint displacement method, 135
Inverse wavelet transform, 100 Model surface, 108
ISO. See International Organization for Stan- Morphologic segmentation, 202
dardization, 8 Morphological closing filter, 95
Isotropy, 36 Morphological filter, 94
Iterative closest points algorithm, 212 Morphological opening filter, 95
Morphological parameter, 58
Morphological rose, 245
J Morphological tree, 218, 253
Jacobian matrix, 114 Motif height, 51
Motif transform, 218
Motif, 49
K Moving average, 69
Knot, 89 MSA. See Measurement System Analysis
Kurtosis, 21 Method, 334
Multi-class classification, 156
Multi-scale fractal analysis, 131
L
L1 approximation, 113
Langer’s lines, 256 N
Laser machining, 306 Naïve Bayesian classification, 166
Laser texturing, 323 National Institute of Standards
Lay, 3 and Technology, 7
352 Index
National Metrology Institute, 18 Q

National Physical Laboratory, 7 QR factorisation, 114
NIST. See National Institute of Standards Quantisation step, 18
and Technology, 8
NMI. See National Metrology Institute, 18
Normal distribution, 160 R
NPL. See National Physical Laboratory, 7 Random rough surface, 306
Numerical aperture, 327 Range image, 185
Nyquist sampling criterion, 187 Receiver operating characteristic analysis, 157
Recursive algorithm, 130
Reference software, 7
O Regression filtering, 81
One-class classification, 170 Relative length, 140
Open surface, 73 Riblet, 5
Optoelectronic circuitry, 304 Ridge line, 48
Ordinarly least squares, 118 Ring projection transform, 205
Orthogonal distance regression, 118 Robust filter, 69
Orthogonal distance, 112 Robust Gaussian regression filter, 85
Outlier, 71 Robust spline filter, 90
Over-segmentation, 51 ROC. See receiver operating characteristic
analysis, 157
Root mean square gradient, 29
P Root mean square height, 20
Padding, 194 Rotation matrix, 109
Parametrically defined surface, 109 Roughness amplitude distribution, 217
PCB. See printed circuit board, 304 Rounding error, 18
PDF. See probability density function, 160 Roundness parameter, 59
Peak curvature, 55 Running-in length, 77
Peak density, 55 Running-out length, 77
Peak extreme height, 23
Peak material volume, 27
Peak to valley height of the waviness, 38 S
Peak, 47 Saddle point, 49
Pearson product moment correlation Sampling area, 17
coefficient, 316 Scale-based, fractal analysis, 131
Percutaneous absorption, 256 Scaling ratio, 131
Phase correct, 71 Scanning electron microscope, 181, 283, 313
Photovoltaic, 269 Scratch testing, 311
Pit, 48 Segmentation, 50, 199
Planar filter, 72 Selective leveling, 199
Planar form removal, 119 Self-affinity, 130
Plastic model, 219 Self-similarity, 130
Plateau honing, 248, 282 SEM. See scanning electron
Point cloud, 186 microscope, 181, 283
Polished stone value test, 339 Sensitivity, 157
Posterior probability, 160 S-F surface, 67, 107
Printed circuit board, 304 Short-time Fourier transform, 99
Prior probability, 160 Sieving, 97
Probability curve, 25 Silicon wafer, 270
Probability density function, 160 Skewness, 20
Pruning, 53 Skidding resistance, 338
PSV. See polished stone value test, 339 Skin, 255
Index 353
S-L surface, 67 Texture-based segmentation, 203

Smoothing spline, 90 Thermal joining, 323
Smooth-rough crossover, 137 Threshold, 286
Sneddon’s solutions, 228 Thresholding, 162, 200
Sobel operator, 202 Tiling algorithm, 144
Softgauge, 7 TNR. See specificity, 157
Software measurement standard, 7 Topological feature, 45
Solar cell, 270 Tortuosity, 140
Spatial correlation, 125 Total least squares, 118
Spatial parameter, 31 Total least-squares, 68
Spearman’s rank correlation coefficient, 316 TPR. See sensitivity, 157
Specification standard, 8 Transfer function, 70
Specificity, 157 Triangulation, 187
Spherical form removal, 121 Tribology, 218
Spline filter, 89 True negative rate. See specificity, 157
Spline interpolation, 19 True positive rate. See sensitivity, 157
SRC. See smooth-rough crossover, 137
Standards, 8
Statistical clustering, 204 U
STFT. See the short-time Fourier transform, 99 Uncertainties associated with fitted
Stratified surface, 25 parameters, 115
Stray grooves, 284 Unevenness, 337
Structured surface, 306 Unsupervised classification, 156
Structuring element, 94
Stylus instrument, 3
Sub-feature partitioning, 207 V
Supervised classification, 156 Valley fluid retention index, 40
Surface bearing index, 39 Variance matrix, 115
Surface bearing ratio, 39 Vectorisation, 208
Surface form, 2 Void volume, 26
Surface integrity, 2 Void, 194
Surface profile, 3 Volume of islands, 299
Surface section difference, 24 Volume parameter, 28
Surface texture parameter, 3
Surface texture, 2
Surface topography, 1 W
Surface, 1 Watershed segmentation, 46, 50
Systematic effects associated with a Wavelet transform, 100
measurement system, 116 Wavelet, 98
Wehner-Schulze machine, 338
Weierstrass-Mandelbrot function, 136
T Weighting function, 69
Template matching, 205 Wolf pruning, 53
Tessellated surface, 186 Working distance, 327
Texture aspect ratio, 35
Texture direction, 37

Richard Leach (Auth.), Richard Leach (Eds.) - Characterisation of Areal Surface Texture-Springer-Verlag Berlin Heidelberg (2013) 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

Richard Leach (Auth.), Richard Leach (Eds.) - Characterisation of Areal Surface Texture-Springer-Verlag Berlin Heidelberg (2013) PDF

Uploaded by

Copyright:

Available Formats

Richard Leach Editor

ISBN 978-3-642-36457-0 ISBN 978-3-642-36458-7 (eBook)

Library of Congress Control Number: 2013934384

Ó Springer-Verlag Berlin Heidelberg 2013

Printed on acid-free paper

Springer is part of Springer Science+Business Media (www.springer.com)

The function of a component part can be profoundly affected by its surface

1 Introduction to Surface Topography . . . . . . . . . . . . . . . . . . . . . . 1

2 The Areal Field Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15

3 The Areal Feature Parameters . . . . . . . . . . . . . . . . . . . . . . . . . . 45

4 Areal Filtering Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 67

5 Areal Form Removal . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107

6 Areal Fractal Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 129

7 Choosing the Appropriate Parameter . . . . . . . . . . . . . . . . . . . . . 155

8 Characterisation of Individual Areal Features . . . . . . . . . . . . . . . 179

9 Multi-Scale Signature of Surface Topography . . . . . . . . . . . . . . . 217

10 Correlation of Areal Surface Texture Parameters

11 Characterisation of Cylinder Liner Honing Textures

12 Characterisation of the Mechanical Bond Strength

13 Inspection of Laser Structured Cams and Conrods . . . . . . . . . . . 321

14 Road Surfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337

Abstract This chapter introduces the field of surface topography measurement.

1.1 Surface Topography Measurement

R. Leach (ed.), Characterisation of Areal Surface Texture, 1

1.2 Surface Profile

Surface profile measurement is achieved by measuring a line across the surface

Fig. 1.1 Coordinate system

Fig. 1.2 Example of the result of a profile measurement

1.3 Areal Surface Topography

The measurement and characterisation of surfaces using the profile method

• surfaces of microfluidic channels for flow control, mixing, lab-on-a-chip and

1.4 Software Measurement Standards

Surface texture characterisation involves a large array of filtering methods and

software measurement standards with those from NIST, and a number of

1.5 Current International Standards

1.5.1 Profile Standards

1.5.2 Areal Standards

2.1 A Short History of Areal Parameters

R. Leach (ed.), Characterisation of Areal Surface Texture, 15

of areal surface texture parameters. The historical perspective of the SURF-

2.2 Naming, Filtering and Calculation Conventions

2.2.2 Filtering Conditions

2.2.3 Sampling Area

2.2.4 Centred Heights

where z is the mean height of the surface on the definition area, A.

2.3 Continuous Against Discrete Definitions

Fig. 2.1 Sampled data points

Fig. 2.2 Integration using a

Fig. 2.3 Integration using

Fig. 2.4 Integration using a

2.4 Height Parameters

2.4.1 Mean Height of the Surface

2.4.1.1 Root Mean Square Height, Sq

2.4.1.2 Arithmetic Mean Height, Sa

The arithmetic mean height or Sa parameter is defined as the arithmetic mean of

2.4.2 Skewness and Kurtosis

2.4.2.1 Skewness, Ssk

2.4.2.2 Kurtosis, Sku

2.4.3 Maximum Height of the Surface

2.5 Function Related Parameters

2.5.1 Height Distribution and Material Ratio Curve

robust solution as it may be affected by outliers. In areal analysis, the value c is

2.5.2 Material Ratio Parameters

2 ! ! ! ! 3