Metrology

Precision Manufacturing
Series Editor: Liangchi Zhang
Wei Gao Editor
Metrology
Precision Manufacturing
Series Editor
Liangchi Zhang
The University of New South Wales
New South Wales, Australia
This series of handbooks covers a comprehensive range of scientific and technolog-
ical matters in precision manufacturing. The proposed handbook series aims to
bridge the gaps by a systematically designed strategy to cover the required range
of knowledge and essential understanding, and hence provide researchers and
engineers a vehicle for achieving the optimization of the intelligent manufacturing
chain. The readers will understand their role and position in precision manufacturing
chain and hence understand how they could progress more efficiently and effec-
tively. Junior researchers and engineers could seek their starting points of career
development more easily and grab essential knowledge more systematically with a
clear direction.
More information about this series at http://www.springer.com/series/15575

Wei Gao
Editor
Metrology
With 476 Figures and 36 Tables

Editor
Wei Gao
Tohoku University Department of Finemechanics
Sendai, Miyagi, Japan
ISSN 2522-5464 ISSN 2522-5472 (electronic)

ISBN 978-981-10-4937-8 ISBN 978-981-10-4938-5 (eBook)
ISBN 978-981-10-4939-2 (print and electronic bundle)
https://doi.org/10.1007/978-981-10-4938-5
© Springer Nature Singapore Pte Ltd. 2019
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.
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.
The publisher, the authors, and the editors are safe to assume that the advice and information in this book
are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or the
editors give a warranty, express or implied, with respect to the material contained herein or for any errors
or omissions that may have been made. The publisher remains neutral with regard to jurisdictional claims
in published maps and institutional affiliations.
This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd.
The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721,
Singapore
Series Preface
Manufacturing has always been a major wealth-creating sector in developed econ-

omies and will remain the cornerstone of long-term economic growth. It is
manufacturing that underpins the modern scientific and technological development,
such as the advances in energy-, bio-, micro-, and nano-technologies. Most of
today’s complex and important technological problems are inseparably connected
to manufacturing issues. Successful innovative solutions in almost all disciplines
rely on manufacturing, because deep insights can only be obtained with the aid of
instruments properly manufactured. Since the new century, the advances in elec-
tronics, optics, telecommunication, biology, medical surgery, energy generation,
resource exploration, and environment protection have brought about further chal-
lenges and have produced veritable onslaught of fundamentals and technologies that
require the capacity of precision manufacturing.
The production of precision components and systems is sensitive to many
complex sets of conditions in which they are manufactured. Over the past decades,
the design and manufacture of high-integrity systems have been improved but are
limited by the lack of understanding of the production processes as an integrated
whole. The discipline advancement of precision manufacturing requires that
researchers and engineers master the fundamentals of materials to use, the pro-
cessing technologies for transforming such materials with minimized defects to
functional products, the strategies and technologies to make machines that are
accurate enough to realize the precision production, and the solutions to environ-
mental impact issues. Each aspect of the above plays a critical role in the precision
manufacturing chain. The chain can be damaged if any of the aspects suffer from
imperfections. If researchers and engineers do not have a comprehensive vision and
understanding of the chain, optimization of the precision manufacturing is difficult
and forms a major hurdle to realizing intelligent/smart manufacturing for new
industrial revolution. This has imposed severe restrictions on our ability to analyze
the complex processes of precision manufacturing. It is because so much is now
demanded of high-integrity systems that the slightest imperfection in their manufac-
ture has become a serious matter.
The objective of this book series is to redress the shortcomings in the isolated
single aspect studies in the chain of precision manufacturing. The books will cover a
comprehensive range of scientific and technological matters to bridge the gaps by a
v
vi Series Preface
systematically designed strategy to reinforce the required knowledge and essential

understanding, and hence provide the reader with a vehicle for achieving the
optimization of the intelligent manufacturing chain. Specific emphasis of the book
series will be on the fundamentals of materials and mechanics for precision
manufacturing, minimal damage and damage-free design in precision manufactur-
ing, precision machines and their control, numerical simulation for precision
manufacturing, precision forming, precision optics, precision additive manufactur-
ing, precision biomedical manufacturing, precision sensing and measurement, non-
traditional precision manufacturing, eco-technologies, and remanufacture of
precision elements.
The books are suitable to senior undergraduate students, postgraduate students,
junior researchers, and engineers who are interested in or working in the field of
precision manufacturing.
Volume Preface
Precision manufacturing is a manufacturing technology for mass production of

precision parts with high accuracies and low tolerances, in which precision machines
are employed to produce designed parts from raw materials. Precision manufacturing
is often composed of complicated processes involved with different machines and
influenced by many factors; the actual manufactured part will deviate from the
designed values. Sensing and measurement are essential for process and quality
control in precision manufacturing, which can not only distinguish whether the
manufactured part meets the assigned tolerances through inspection but also, in
many cases, reduce the deviation of the manufactured part from the designed values
through improvement of the process or compensation manufacturing based on the
sensing and measurement results.
This book describes a variety of sensing technologies and measurement
systems that are vital for precision manufacturing. The topics and chapters, which
have been contributed by a group of leading researchers in this field, include:
Dimensional Metrology Using Mode-Locked Lasers, Nanopositioning and
Nanomeasuring Machines, Confocal Microscopy for Surface Profilometry, Optical
Sensors for Machine Tool Metrology, Optical Micro-CMM, Machine Tool Calibra-
tion, Accuracy and Performance Analysis of Machine Tools, Machine Tool Progno-
sis for Precision Manufacturing, Cylindrical Gear Metrology, High-Speed
Measurement of Complex Shaped Parts by Laser Triangulation for In-Line Inspec-
tion, Micro-dimensional Measurement by a Micro-probing System, Measurements
in Additive Manufacturing, In-Line Measurement Technology and Quality Control,
In-Process Measurement of Subwavelength Structures, Optical Scatterometry for
Nanostructure Metrology, Contact-Type Micro Thermal Sensor for Surface Defect
Detection, X-Ray Computed Tomography for Dimensional Metrology, Uncertainty
Estimation in Computational Tools in Metrology, Molecular Dynamics Characteri-
zation of a Force Sensor Integrated Fast Tool Servo for On-Machine Surface
Metrology.
As the editor, I believe that this book presents the newest information on precision
manufacturing metrology and it will be beneficial to postgraduate students
and engineers in the fields of precision manufacturing, quality control, and precision
measurement. I would like to thank all the authors for their great contributions.
vii
viii Volume Preface
I would also like to thank Professor Liangchi Zhang, the Series Editor, for involving
me in this book project. The publishing staff of Springer are highly appreciated for
their dedicated efforts in making this book possible.
July 2019 Wei Gao

Contents
1 Dimensional Metrology Using Mode-Locked Lasers . . . . . . . . . . . 1

Seung-Woo Kim, Yoon-Soo Jang, Jiyong Park, and Wooram Kim
2 Nanopositioning and Nanomeasuring Machines .............. 35

Eberhard Manske
3 Confocal Microscopy for Surface Profilometry ............... 59

Liang-Chia Chen
4 Optical Sensors for Machine Tool Metrology . . . . . . . . . . . . . . . . . 93

Zhi-Feng Lou and Kuang-Chao Fan
5 Optical Micro-CMM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

Yasuhiro Takaya and Masaki Michihata
6 Machine Tool Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189

J. R. R. Mayer
7 Accuracy and Performance Analysis of Machine Tools . . . . . . . . . 215

Andreas Archenti and Theodoros Laspas
8 Machine Tool Prognosis for Precision Manufacturing . . . . . . . . . . 245

Robert X. Gao, Peng Wang, and Ruqiang Yan
9 Cylindrical Gear Metrology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277

Kang Ni, Yue Peng, Dirk Stöbener, and Gert Goch
10 High-Speed Measurement of Complex Shaped Parts by Laser

Triangulation for In-Line Inspection . . . . . . . . . . . . . . . . . . . . . . . 307
Alexander Schöch and Enrico Savio
11 Micro-dimensional Measurement by a Micro-probing System . . . . 329

So Ito
12 Measurements in Additive Manufacturing . . . . . . . . . . . . . . . . . . . 369

David Bue Pedersen, Sebastian Aagaard Andersen, and Hans
Nørgaard Hansen
ix
x Contents
13 In-Line Measurement Technology and Quality Control . . . . . . . . . 399

Gisela Lanza, Benjamin Haefner, Leonard Schild, Dietrich Berger,
Niclas Eschner, Raphael Wagner, and Marielouise Zaiß
14 In-Process Measurement of Subwavelength Structures . . . . . . . . . 435
Satoru Takahashi
15 Optical Scatterometry for Nanostructure Metrology . . . . . . . . . . . 477
Xiuguo Chen and Shiyuan Liu
16 Contact-Type Micro Thermal Sensor for Surface Defect
Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 515
Yuki Shimizu
17 X-Ray Computed Tomography for Dimensional Metrology ..... 537
Filippo Zanini and Simone Carmignato
18 Uncertainty Estimation in Computational Tools in Metrology ... 585
Jean-Marc Linares
19 Molecular Dynamics Characterization of a Force Sensor
Integrated Fast Tool Servo for On-Machine Surface
Metrology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 601
Yindi Cai
About the Series Editor
Liangchi Zhang is Scientia Professor, Professor of

Mechanical Engineering, and Head of the Laboratory
for Precision and Nano Processing Technologies at the
University of New South Wales, Australia. He is also the
Fellow of the Australian Academy of Technological
Science and Engineering (FTSE). Zhang obtained his
B.Sc. and M.Eng. from Zhejiang University and his
Ph.D. from Peking University, China. Prior to joining
the University of Sydney, Australia, Zhang has worked
at the University of Cambridge, UK, and the National
Mechanical Engineering Laboratory, MITI, Japan. In
2009, he was invited to take up the Scientia Professor-
ship at the University of New South Wales. Zhang carries
out research on both the fundamentals and industrial
applications in the cross-disciplinary field of advanced
manufacturing, advanced materials, tribology, nanotech-
nology, solid mechanics, and biomechanics. He has
published extensively in his research areas, with some in
multiple languages. His research outcomes have led to
significant economic benefits for manufacturing industry.
Zhang has been granted many awards and honors. In
2005, he was awarded a higher doctoral degree – Doctor
of Engineering – by the University of Sydney for his
contributions to precision manufacturing. Home page:
http://www.precision-manufacturing.unsw.edu.au/.
xi
About the Editor
Wei Gao received his Bachelor of Precision Instrumen-

tation from Shanghai Jiao Tong University, China, in
1986, followed by M.Sc. and Ph.D. from Tohoku Uni-
versity, Japan, in 1991 and 1994, respectively. He is
currently a Professor in the Department of
Finemechanics at Tohoku University. His research inter-
ests lie primarily in the field of precision engineering,
specialized in precision metrology and micro-/
nanometrology. He and his group have developed a
number of surface metrology systems as well as a couple
of optical sensor technologies for precision measure-
ment and nanometrology. He is a Fellow of the Interna-
tional Academy for Production Engineering (CIRP), the
International Society for Nanomanufacturing (ISNM),
and the Japan Society for Precision Engineering (JSPE).
He served as the Chairman of the Scientific Technical
Committee Precision Engineering and Metrology of
CIRP from 2015 to 2018 and as Vice President of
JSPE in 2015. He is the author of the book Precision
Nanometrology (Springer). He and his group have won
seven paper awards from JSPE. He was also awarded
the Prize for Science and Technology from the Govern-
ment of Japan in 2019.
xiii
Contributors
Sebastian Aagaard Andersen Technical University of Denmark, Lyngby,

Denmark
Andreas Archenti KTH Royal Institute of Technology, Stockholm, Sweden
Dietrich Berger wbk Institute of Production Science, Karlsruhe Institute of Tech-
nology (KIT), Karlsruhe, Germany
Yindi Cai Dalian University of Technology, School of Mechanical Engineering,
Dalian, Liaoning, China
Simone Carmignato Department of Management and Engineering (DTG), Uni-
versity of Padova, Vicenza, Italy
Liang-Chia Chen Department of Mechanical Engineering, National Taiwan Uni-
versity, Taipei, Taiwan
Xiuguo Chen School of Mechanical Science and Engineering, Huazhong Univer-
sity of Science and Technology, Wuhan, China
Niclas Eschner wbk Institute of Production Science, Karlsruhe Institute of Tech-
Kuang-Chao Fan Key Laboratory for Precision and Non-traditional Machining of
Ministry of Education, School of Mechanical Engineering, Dalian University of
Technology, Dalian, China
Department of Mechanical Engineering, National Taiwan University, Taipei, Taiwan
Robert X. Gao Department of Mechanical and Aerospace Engineering, Case
Western Reserve University, Cleveland, OH, USA
Gert Goch Center for Precision Metrology, University of North Carolina at Char-
lotte, Charlotte, NC, USA
Benjamin Haefner wbk Institute of Production Science, Karlsruhe Institute of
Technology (KIT), Karlsruhe, Germany
Hans Nørgaard Hansen Technical University of Denmark, Lyngby, Denmark
xv
xvi Contributors
So Ito Department of Intelligent Robotics, Toyama Prefectural University, Imizu,

Japan
Yoon-Soo Jang Department of Mechanical Engineering, Korea Advanced Institute
of Science and Technology (KAIST), Daejeon, South Korea
Seung-Woo Kim Department of Mechanical Engineering, Korea Advanced Insti-
tute of Science and Technology (KAIST), Daejeon, South Korea
Wooram Kim Department of Mechanical Engineering, Korea Advanced Institute
of Science and Technology (KAIST), Daejeon, South Korea
Gisela Lanza wbk Institute of Production Science, Karlsruhe Institute of Technol-
ogy (KIT), Karlsruhe, Germany
Theodoros Laspas KTH Royal Institute of Technology, Stockholm, Sweden
Jean-Marc Linares Aix Marseille University, Marseille, France
Shiyuan Liu School of Mechanical Science and Engineering, Huazhong Univer-
sity of Science and Technology, Wuhan, China
Zhi-Feng Lou Key Laboratory for Precision and Non-traditional Machining of
Ministry of Education, School of Mechanical Engineering, Dalian University of
Technology, Dalian, China
Eberhard Manske Department of Mechanical Engineering, Technische
Universität Ilmenau, Ilmenau, Germany
J. R. R. Mayer Department of Mechanical Engineering, Polytechnique Montreal,
Montreal, QC, Canada
Masaki Michihata The University of Tokyo, Tokyo, Japan
Kang Ni Center for Precision Metrology, University of North Carolina at Charlotte,
Charlotte, NC, USA
Research and Development, The Timken Company, North Canton, OH, USA
Jiyong Park Department of Mechanical Engineering, Korea Advanced Institute of
Science and Technology (KAIST), Daejeon, South Korea
David Bue Pedersen Technical University of Denmark, Lyngby, Denmark
Yue Peng Center for Precision Metrology, University of North Carolina at Char-
lotte, Charlotte, NC, USA
Enrico Savio Department of Industrial Engineering, University of Padova, Padova,
Italy
Leonard Schild wbk Institute of Production Science, Karlsruhe Institute of Tech-
Contributors xvii
Yuki Shimizu Department of Finemechanics, School of Engineering, Tohoku

University, Sendai, Japan
Alexander Schöch Institute for Production Metrology, Materials and Optics, NTB
University for Technology, Buchs, Switzerland
Dirk Stöbener Bremen Institute for Measurement, Automation and Quality Sci-
ence (BIMAQ), University of Bremen, Bremen, Germany
Satoru Takahashi Research Center for Advanced Science and Technology
(RCAST), The University of Tokyo, Tokyo, Japan
Yasuhiro Takaya Osaka University, Osaka, Japan
Raphael Wagner wbk Institute of Production Science, Karlsruhe Institute of Tech-
Peng Wang Department of Mechanical and Aerospace Engineering, Case Western
Reserve University, Cleveland, OH, USA
Ruqiang Yan School of Mechanical Engineering, Xi’an Jiaotong University,
Xi’an, China
Marielouise Zaiß wbk Institute of Production Science, Karlsruhe Institute of
Technology (KIT), Karlsruhe, Germany
Filippo Zanini Department of Management and Engineering (DTG), University of
Padova, Vicenza, Italy
Dimensional Metrology Using
Mode-Locked Lasers 1
Seung-Woo Kim, Yoon-Soo Jang, Jiyong Park, and Wooram Kim
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2
Mode-Locked Lasers for Dimensional Metrology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
Multiwavelength Interferometer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
Time-of-Flight Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3-D Surface Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
Summary and Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
Abstract
Laser-based optical interferometry has long been playing a central role in the
progress of dimensional metrology for precision manufacturing. Meanwhile,
mode-locked lasers are nowadays available to facilitate the progress by responding
to ever-growing industrial demands on the measurement precision and functionality
beyond the capabilities of conventional lasers. The optical spectrum of mode-
locked lasers, referred to as the frequency comb, acts as the ruler enabling ultra-
stable wavelengths to be produced for precision interferometry with traceability to
the atomic clock. In addition, mode-locked lasers are employed directly as the light
source offering ultrashort pulses, of which the time-of-flight can be detected with
unprecedented precision in implementing long distance measurement. Further,
mode-locked lasers are able to provide well-controlled temporal coherence in
combination with high spatial coherence, being suited to overcome the technical
barriers long standing in 3-D profiling of rough surfaces. In summary, mode-locked
S.-W. Kim (*) · Y.-S. Jang · J. Park · W. Kim

Department of Mechanical Engineering, Korea Advanced Institute of Science and Technology
(KAIST), Daejeon, South Korea
e-mail: swk@kaist.ac.kr
© Springer Nature Singapore Pte Ltd. 2019 1

W. Gao (ed.), Metrology, Precision Manufacturing,
https://doi.org/10.1007/978-981-10-4938-5_1
2 S.-W. Kim et al.
lasers are now ready to lead the advance of dimensional metrology by providing
unique temporal and spectral benefits over conventional continuous wave lasers or
broad-spectrum light sources.
Keywords
Dinemsional metrology · Length measurement · Surface profile measurement ·
Laser-based optical interferometry · Mode-locked lasers · Ultrashort lasers
Introduction
The role of dimensional metrology in manufacturing is to quantify the physical size

of a machined part or distance of a moving object within a reference frame (Berkovic
and Shafir 2012; Gao et al. 2015). The task requires the use of mechanical and
optoelectronic instruments in precise calibration with reference to the basic unit of
length – the meter – currently being defined as the length of the path travelled by
light in vacuum during a time interval of 1/299,792,458 s in the International System
of Units (SI) (Giacomo 1984). The well-known Michelson interferometer was first
proposed in 1887, which instigated the approach of using the light wavelength as a
basic ruler of length measurement. Nowadays, diverse lasers are available with
wavelengths calibrated with respect to several radiation sources recommended as
the optical wavelength standards with well-certified uncertainties with reference to
the SI definition of the meter (Quinn 2003; Felder 2005). Optical interferometers
available today for dimensional metrology mostly rely on a single-wavelength laser
as the light source (Bobroff 1993; Demarest 1998). The single-wavelength interfer-
ometry is inherently subject to a short non-ambiguity range constrained to just half of
the used wavelength. The measurement of a length or distance has to be made by
continuously accumulating the displacement induced by moving the target mirror.
This incremental feature of single-wavelength interferometry restricts its usage in
practical implementation of dimensional metrology, particularly in the situation of
dealing with large size objects or measuring long distances.
There are in fact some different optical schemes available to extend the non-
ambiguity range. They are collectively categorized as absolute distance measure-
ments (ADMs) as they are all intended to measure a given distance in an absolute
way with the target mirror being fixed at the distance to be measured. ADM
examples include time-of-flight measurement using light pulses (Bender et al.
1973; Degnan 1985), laser intensity modulation (Fujima et al. 1998), wavelength
scanning using a frequency-tunable laser (Uttam and Culshaw 1985; Xiaoli and
Katuo 1998), and incorporation of multiple lasers of different wavelengths
(Dandliker et al. 1988; Hartmann et al. 2008). These variant ADM schemes permit
long distance measurements to be implemented with an extensive or limitless non-
ambiguity range. Nonetheless, the uncertainty of absolute measurements is not yet
comparable to that of single-wavelength interferometry, mainly due to the wave-
length accuracy achievable from conventional lasers. In the meanwhile, mode-
locked lasers became available recently as the light source to bring breakthroughs
1 Dimensional Metrology Using Mode-Locked Lasers 3
for dimensional metrology in the measuring range and resolution (Kim 2009). As a
result, many noble ADM principles have been demonstrated successfully to imple-
ment dimensional metrology with superior performance over conventional light
sources in terms of the measurement resolution and update rate. The examples
include synthetic wavelength interferometry (Minosima and Matsumoto 2000),
multiwavelength interferometry (Schuhler et al. 2006; Salvadé et al. 2008), disper-
sive interferometry (Joo and Kim 2006; van den Berg et al. 2012), dual-comb
time-of-flight interferometry (Coddington et al. 2009), and optical cross-correlation
time-of-flight measurement (Lee et al. 2010; Han et al. 2015).
In this chapter, a comprehensive review is made on how mode-locked lasers can
be utilized as the light source for dimensional metrology. Specifically, section
“Mode-Locked Lasers for Dimensional Metrology” describes the temporal as well
as spectral characteristics provided by mode-locked lasers uniquely in contrast to
conventional lasers and broad spectral light sources. Section “Multiwavelength
Interferometer” focuses on the spectral feature represented by the frequency comb,
which allows multiple wavelengths to be generated simultaneously so as to measure
long distances without losing the sub-wavelength precision of laser-based interfer-
ometry. Section “Time-of-Flight Measurement” explains the temporal advantage of
utilizing ultrashort pulses to improve the time-of-flight measurement beyond the
performance limit of using conventional light pulses. Section “3-D Surface Mea-
surements” deals with 3-D surface profile measurements of very rough surfaces, an
extended field-of-view achieved by combining the low temporal but high spatial
coherence of mode-locked lasers. Finally, section “Summary and Outlook” gives a
summary and outlook on how mode-locked lasers will make contributions to the
progress of dimensional metrology in the near future.
Mode-Locked Lasers for Dimensional Metrology
Mode-locked lasers were made available first in the 1970s to produce short pulses by
means of mode locking, i.e., synchronizing together a large number of discrete
modes radiated from a broad spectral gain medium by stimulated emission within
a resonance cavity (Keller 2003). In the beginning period, dye lasers were used as the
gain medium and later solid-state bulk materials such as ion-doped crystals or glass
began to be devised to generate ultrashort pulses of sub-picosecond or few femto-
second duration (https://www.rp-photonics.com/mode_locked_lasers.html). Nowa-
days, a variety of mode-locked lasers has been developed with distinct gain materials
and mode-locking mechanisms, including fiber-type oscillators emitting infrared
pulses of tens of femtoseconds with improved immunity to environmental distur-
bance (Kim and Song 2016). This diversified mode-locked laser technology benefits
many fields of science and technology, with no exception on optical metrology. The
most remarkable breakthrough was made at the turn of this century by making the
success of stabilizing the optical modes of a whole frequency comb collectively to
the radio-frequency atomic clock. This enabled precision calibration of optical laser
wavelengths directly with reference to the microwave atomic clock of time/
4 S.-W. Kim et al.
frequency standard (Jones et al. 2000; Diddams et al. 2000). Furthermore, being
stabilized to optical clocks, the frequency comb is now able to disseminate very
stable optical frequencies for diverse applications for the progress of fundamental
sciences and technologies (https://www.rp-photonics.com/optical_clocks.html).
From the viewpoint of optical interferometry, a mode-locked laser can be distin-
guished from conventional lasers in its distinct temporal and spectral characteristics.
In the time domain, as depicted in Fig. 1, a mode-locked laser appears as a train of
short pulses repeated at a radio-frequency rate. The pulse duration falls into the sub-
picosecond to few femtosecond range. When the pulse train is Fourier-transformed
to the frequency domain, as in Fig. 2, the resulting optical spectrum is dubbed the
Fig. 1 Temporal characteristic of a mode-locked laser as a train of ultrashort pulses
Fig. 2 Spectral characteristic of a mode-locked laser as the frequency comb

frequency comb, of which the spectral shape comprises numerous discrete frequency
modes equally spaced with a spacing equal to the pulse repetition rate ( fr). The
spectral bandwidth of the frequency comb is related to the pulse duration; shorter
pulses lead to broader bandwidths. Each optical mode within the frequency comb
has a narrow linewidth with its center frequency expressed as fi = i fr + fo with the
integer i denoting the individual mode number and fo being the carrier-envelope
offset frequency. The repetition rate fr is directly observed using a photo detector,
while the carrier-envelope offset frequency fo is identified using a self-referencing
f-2f interferometer (Kim et al. 2009a). Both fr and fo are in the radio-frequency
regime, so they can be phase-locked to the microwave atomic clock (e.g., rubidium
clock or cesium clock). This stabilizes the comb modes all together concurrently. For
more precise stabilization, the frequency comb can be locked to an optical clock
through a high-finesse cavity (https://www.rp-photonics.com/finesse.html).
Once the frequency comb is stabilized, it can be utilized as the ruler to generate
stable monochromatic wavelengths to be used for optical interferometry. This
process is called the optical frequency generation (OFG) and can be performed
using two methods as illustrated in Fig. 3. One method is to choose a comb mode
being at the frequency position of interest, and a working laser is brought in to lock it
to the comb mode. The phase-locked loop (PLL) control technique is utilized to
stabilize the beat frequency of the working laser with the particular comb mode to the
atomic clock. The working laser needs to offer an appropriate control bandwidth in
Fig. 3 Optical frequency generation from the frequency comb of a stabilized mode-locked laser
6 S.-W. Kim et al.
tuning its output frequency; thus an extra-cavity laser diode or distributed feedback
laser is preferable (Chun et al. 2013). This method is able to stabilize the working
laser with its linewidth usually falling into the range of several MHz.
The other method of optical frequency generation, illustrated in Fig. 3, is intended
to transfer the narrow linewidth of the comb mode to the working laser without
broadening. This task is conducted by extracting the selected comb mode by optical
band-pass filtering with a very narrow transmission window. The extracted comb
mode is then made incident to the working laser directly by means of injection
locking (Kim et al. 2009b). This process permits amplifying the optical power of the
extracted comb mode, which is usually a few nW to a mW level to be used as the
light source of optical interferometry. In addition, the process of injection locking
leads the working laser to be in perfect synchronization in its output frequency with
the comb mode. More importantly, the output wavelength generated from the
working laser yields a very narrow linewidth of the same order as the original
comb mode without significant degradation. This feature permits achieving an
extremely narrow linewidth of a few Hz in the output wavelength of the working
laser (Chun et al. 2016).
Multiwavelength Interferometer
Most of laser interferometers used today for dimensional metrology rely on single-
wavelength lasers such as HeNe gas laser or solid-state semiconductor diode laser.
These incremental-type laser interferometers operate on homodyne or heterodyne
phase-measuring techniques (Bobroff 1993; Demarest 1998) and determine a length
by translating the target mirror from the reference datum position. On the other hand,
the concept of absolute distance measurement attempts to measure the target distance
without the incremental translation of the target mirror as depicted in Fig. 4.
A widely used method of absolute distance measurement is multiwavelength inter-
ferometry (MWI), which can be implemented by generating multiple wavelengths
from the frequency comb (Jin et al. 2006). The target distance L is given as Lk =
(λk/2nk) (mk + ek) with λk being the vacuum wavelength of the k-th laser, nk the air
refractive index for λk, mk the integer multiple, and ek the excess fraction value
obtained from the measured interferometric phase for λk. The target distance L is
determined by numerical iteration as the mean of Lk of all λk (Jin et al. 2007).
Figure 5 illustrates the light source of a multiwavelength interferometer developed
at KAIST to calibrate gauge blocks by absolute distance measurement (Jin et al.
2006, 2007; Hyun et al. 2009, 2010). The frequency comb as the wavelength ruler is
constructed from a Ti/sapphire oscillator stabilized to the rubidium (Rb) atomic
clock. The frequency comb has a 60 nm spectral bandwidth and a 780 nm center
wavelength. The repletion rate ( fr) is 81 MHz and the carrier-envelop offset fre-
quency ( fo) is 15 MHz. An external-cavity laser diode (ECLD) is used as the
working laser with a 12 mW output power. The output frequency of the working
laser is precisely phase-locked to four comb modes with a beat frequency
( fb, 10 MHz), i.e., f = i fr + fo + fb. Then, the interferometric phase to calculate
Fig. 4 Concept of multiwavelength interferometry for absolute distance measurement
the excess fraction e is measured by adding a heterodyne beat of 40 MHz to the

measurement beam using a pair of acousto-optic modulator (AOM). Then the
working laser is tuned to other comb modes, one by one in sequence, to obtain
different wavelengths: 770.204349 nm, 779.953524 nm, 780.203668 nm, and
780.206961 nm. This scheme of multiwavelength generation permits calibrating
gauge blocks with a fractional uncertainty of one part in billion with traceability to
the erbium atomic clock.
Figure 6 illustrates a multiwavelength interferometer system configured at
KAIST to perform absolute distance measurement using the four wavelengths
generated in Fig. 5. The interferometer optics constructed for the purpose is a
modified Michelson type. The interferometric phase of each wavelength quantified
simultaneously by applying the heterodyne phase-measuring technique. The mea-
surement beam is separated from the reference beam through a set of four non-
polarizing splitters configured in a box configuration. This is intended to perform the
heterodyne phase measurement without nonlinear cyclic errors caused by unwanted
polarization mixing and leaking. The heterodyne beat frequency is generated
through a pair of two acousto-optic modulators (AOM1 and AOM2) installed in
tandem. Two avalanche photodetectors (APD) are used to monitor the resulting
8 S.-W. Kim et al.
Fig. 5 Generation of multiple wavelengths using the frequency comb as the wavelength ruler.
(Adapted from Hyun et al. 2010 with permission)
Fig. 6 Multiwavelength interferometer for length measurement. (Adapted from Hyun et al. 2009
with permission)
interference signals; one is the reference electric signal of A cos [2π

( fAOM1 – fAOM2)t + φref], and the other is the measurement electric signal of
B cos [2π( fAOM1 – fAOM2)t + φmea], where A and B denote amplitudes of each
interference signal. The phase difference Δφ = φmea – φref, given by Δφ = 4π L/λ, is
determined using a precision phasemeter with a 0.1 resolution. For real-time
compensation of the air refractive index using Edlen’s formula (Birch and Downs
1994), the ambient air parameters of temperature, pressure, relative humidity, and
CO2 concentration are continuously monitored. The measurement path of the inter-
ferometer is installed on a Zerodur plate of low thermal expansion under 0.005 K
temperature control.
Figure 7 explains how the interferometer system of Fig. 6 is operated to measure a
length of ~1.2 m. The measurement procedure consists of two steps: sweeping the
working wavelength from λ1 to λ2 and hopping the wavelength to λ3 and λ4. The
wavelength sweeping is intended to estimate the length L based on the method of
frequency sweeping interferometry, i.e., L = Λ(Δm + Δe) where Λ = λ1 λ2/
jλ1 λ2j. The next wavelength hopping is to determine L precisely by multi-
wavelength interferometry using the interferometric phases of the four wavelengths.
The instantaneous wavelengths are compensated continuously by monitoring the
refractive index of the air. To be more specific, during the wavelength sweeping from
Fig. 7 Measurement procedure of multiwavelength interferometry. (Adapted from Hyun et al.

2009 with permission)
10 S.-W. Kim et al.
λ1 to λ2, Δm and Δe are measured to be 13 and 0.06940, respectively. This result

leads to an estimation of L as 1195.205502 mm with Λ being 184.853 mm. The true
value of L is finally determined to be 1195.287863 mm, which is the absolute length
permitting the excess fractions for all the wavelengths falling in a standard deviation
of 0.001.
Figure 8 shows the linearity test result obtained while translating the target mirror
over a travel of 800 mm. For comparative evaluation, the same displacement of the
target mirror was also measured using a commercial HeNe laser interferometer. The
comparative result shows an extremely good linearity between the two measure-
ments with a residual difference of 200 nm in peak-to-valley. The difference is
reckoned attributable mainly to the effect of inhomogeneous environmental distur-
bance that cannot be taken into compensation by detection of the ambient temper-
ature at a fixed point. Further, Edlen’s formula used for compensation of the
refractive index of the ambient air provides an uncertainty usually in the range of
107–108 due to the errors encountered in sensing the ambient temperature,
pressure, relative humidity, and carbon dioxide concentration. In addition, the
long-term frequency stability of the HeNe laser source of the commercial interfer-
ometer is considered to contribute to the linearity error. This comparative test
concludes that the linearity of the absolute distance measurement is within a possible
maximum fluctuation of 63.9 nm in terms of standard deviation.
Fig. 8 Linearity test result. (Adapted from Hyun et al. 2010 with permission)
The sequential procedure of multiwavelength interferometry given in Fig. 7

consumes a total time of 17 s by using a single external-cavity laser diode for the
wavelength sweeping from λ1 to λ2 plus hopping from λ3 to λ4. The measurement
time has to be shortened if the distance measurement has to be performed in real
time. This task requires generating four wavelengths simultaneously in parallel
(Wang et al. 2015; Jang et al. 2016). As illustrated in Fig. 9, an Er-doped fiber
oscillator is used to produce the frequency comb as the wavelength ruler with
stabilization to the Rb atomic clock. The repetition rate ( fr) is 100 MHz and the
carrier-envelop offset frequency ( fo) is set to 30 MHz. The working laser source is
comprised of four distributed feedback (DFB) laser diodes of 10 mW optical power.
Each diode is phase-locked to its designated comb mode within the frequency comb
by phase-locked loop (PLL) control. The output beams of the DFB laser diodes are
combined into a single-mode fiber and a heterodyne shift of 40 kHz is applied to the
four wavelengths individually using an acousto-optic modulator (AOM). The inter-
ferometric phases of the four wavelengths are then detected simultaneously using a
multichannel phasemeter. The distance to be measured is determined by applying the
algorithm of multiwavelength interferometry previously explained in Fig. 4. For
comparative evaluation, an incremental-type HeNe laser displacement (IDM) inter-
ferometer is installed on the same measurement optical path. The target mirror is a
corner cube-type retroreflector and installed on a granite air-bearing stage with 3-m-
long travel range. The whole interferometer hardware system is enclosed by a
Fig. 9 Real-time multiwavelength interferometer

12 S.-W. Kim et al.
10-mm-thick acrylic shield chamber to minimize the environmental disturbance of

air turbulence, temperature change, and acoustic noise. In addition, the ambient
parameters of air temperature, relative humidity, air pressure, and CO2 concentration
are monitored for compensation of the air refractive index using Ciddor’s equation
(Ciddor 1996; Jang and Kim 2017).
As shown in Fig. 10, the optical spectrum of the light source reveals that the four
wavelengths λ1, λ2, λ3, and λ4 are produced with a high signal-to-noise ratio of 50 dB.
In terms of the optical frequency, each wavelength is given as fi = i fr + fo + fb as
explained before. Then, the first-order frequency variation of Δfi in its fractional term
is expressed as Δfi/fi = [(ΔfRb /fRb)2 + (Δfr /fr)2 + (Δfo/f )2 + (Δfb /f )2]1/2 with fRb being
the frequency of the Rb atomic clock used as the frequency reference in stabilizing the
frequency comb in this study. In terms of the Allan deviation in the averaging time
from 1 to 500 s, the most dominant factor arises from the Rb atomic clock as
2.57 1011 at 1 s averaging. All other terms are estimated to be much smaller as
shown. Since Δf/f = Δλ/λ, the fractional frequency stability is directly transferred to
the fractional stability of wavelength. In consequence, the four wavelengths generated
are estimated to have a linewidth less than 1.0 106 nm, being specified as
1530.279693 nm, 1531.040888 nm, 1554.179409 nm, and 1554.937151 nm, respec-
tively, with ten significant digits.
Figure 11 shows a measurement result; the acronym ADM indicates the multi-
wavelength interferometer presented in Fig. 9, and the IDM is the commercial
incremental-type HeNe laser incorporated for comparative evaluation. This mea-
surement was taken over a period of 400 s while the target distance was fixed to
~3.8 m. The result indicates that both the ADM and IDM produce almost identical
readings in the time domain; the fluctuation is about 20 nm in standard deviation for
a 100 Hz update rate, and it reduces to a few nm when the update rate increases to
1 Hz. In the frequency domain in which the measured data was Fourier-transformed
up to 50 Hz, spectral power concentration is seen in the range of 20–45 Hz due to
mechanical vibration disturbance. In addition, the Allan deviations of the ADM and
IDM readings are taken for more detailed analysis of the measurement stability,
which reveals a 109 level for the ADM for an averaging time of 0.1–0.5 s. The
measurement stability for longer averaging time becomes deteriorated due to the
slow-varying environmental drift of temperature and pressure. All these observa-
tions confirm that the ADM interferometer is able to provide almost the same or even
better performance of the incremental IDM interferometer in terms of the measure-
ment resolution, speed, and stability even with its remarkable capability of absolute
distance measurement.
Next, the real-time measurement capability of the multiwavelength interferometer
is demonstrated in Fig. 12. In this test, a mechanical chopper is installed to block the
ADM beam path on and off periodically at a 2 Hz switching rate. Meanwhile, the
IDM interferometer is continuously operated without any interruption. The resulting
ADM (blue color) and IDM (pink color) readings are overlapped to reveal no notable
difference between the two results. The high-frequency fluctuation of a ~50 nm
Fig. 10 Four-wavelength light source. (Adapted from Jang et al. 2016 with permission)
14 S.-W. Kim et al.
Fig. 11 Performance evaluation result. (Adapted from Wang et al. 2015 with permission)
amplitude is attributable to the air turbulence and vibrational disturbance induced by

the mechanical chopper. The comparison results clearly show that the ADM inter-
ferometer catches up with that of the IDM results every time when the ADM
measurement beam is recovered from blockage. This verifies that the ADM inter-
ferometer is capable of updating the distance information instantly without any
accumulation of prior information.
Figure 13 shows the measurement linearity of the multiwavelength interferometer
with reference to a conventional HeNe incremental laser interferometer. The abso-
lute distance determined by λ1 was evaluated up to 3.0 m with every 150 mm step by
averaging 300 measurements. The test results reveal a discrepancy of 35.3 nm in
peak-to-valley, corresponding to a 4.6 109 linearity error over the entire range
of 3.0 m. For more strict evaluation, the distance determined by individual wave-
lengths (λ1, λ2, λ3, and λ4) is compared with each other. The absolute distance values
from the individual wavelengths are denoted as L1, L2, L3, and L4, respectively. With
distance differences of L1-L2, L1-L3, and L1-L4 lying within a 20 nm peak-to-valley
bound, the maximum linearity error is worked out to be 2.6 109 between the
measured distances from individual wavelengths. The inter-wavelength differences
are reckoned to arise from the imperfect compensation of the refractive index of the
ambient air due to the 108 level uncertainty of Ciddor’s equation formula (Ciddor
1996; Jang and Kim 2017).
Figure 14 shows additional test results for a fixed target distance of ~3.8 m, which
was monitored over 12 h at a 1 Hz update rate. After compensating the air refractive
index in the overall optical path, the absolute distances denoted by L1, L2, L3, and L4
follow a common long-term drift of gradual decrease by ~5 μm. These distance
decreases appear to be affected by a thermal expansion of the stage monitored during
Fig. 12 Interrupted operation of multiwavelength interferometry. (Adapted from Wang et al. 2015
with permission)
the entire test period. During the measurement, the stage temperature was linearly
decreased by an amount of 0.27 C, which corresponds to 5.7 μm when the thermal
coefficient of the granite stage is considered to be 7 106. Despite the thermal
expansion of the stage, the individual distances of L1, L2, L3, and L4 show no
significant differences. The inter-wavelength distance differences of L3–L2 and
L4–L1 remain almost constant without any long-term drift. The differences of
L3–L2 and L4–L1 are found to be 12.8 nm and 7.1 nm, respectively; they are not
exactly identical since the air refractive index compensation can be made with an
uncertainty of 108 level. When the stability of the differences are evaluated in terms
of the Allan deviation, the stability of L3–L2 and L4–L1 is estimated to be 3.4 nm at
1 s averaging with a fractional stability of 8.9 1010, which further improves to
0.57 nm at 100 s averaging, corresponding to 1.5 1010 in the fractional stability.
This implies that the frequency-comb-referenced multiwavelength interferometer is
16 S.-W. Kim et al.
Fig. 13 Linearity test result. (Adapted from Jang et al. 2016 with permission)
able to achieve a 1010 level of measurement stability by adjusting the averaging

time, even though Ciddor’s formula for the air refractive index compensation is
limited to a 108 level of uncertainty. However, when the averaging time is taken
longer than 100 s, the measurement stability begins to weaken due to the long-term
drift of the phase detection electronics and the environment sensors to monitor the air
temperature, pressure, humidity, and CO2 concentration.
Table 1 summarizes the measurement uncertainty evaluated for the multi-
wavelength interferometer. The uncertainty u(L ) of the measured distance is con-
tributed by three major terms relating to the optical frequency, the refractive
index compensation, and the phase detection, i.e., u(L ) = [{u(e) (λ/2)}2 +
{u(n) L}2 + {u( f ) L}2]1/2. The first term u(e) represents the phase measurement
error. Using a function generator calibrated by the Rb atomic clock, the multichannel
phase meter used in the phase detection is evaluated to have a maximum systematic
phase offset of 8.33 104 over the entire 360 range and a random repeatability of
2.94 105, bringing about a 0.64 nm distance error. This uncertainty for the phase
detection is independent of the target distance L. The next term u(n) represents the
refractive index compensation error. This term is affected by two error sources: the
inaccuracy arising in environment monitoring and the uncertainty of the empirical
equation of Ciddor’s formula itself. The latter is known to be of 108 order. As for
the former error source, the state-of-the-art precision sensors permit observing the
ambient temperature within an error of 5 mK, the air pressure within 2.5 Pa, the air
humidity within 1%, and the carbon dioxide concentration within 41 ppm. The total
uncertainty of the refractive index compensation is estimated to be 1.62 108 L
with L being the target distance.
The last term arises from the frequency uncertainty of the light source denoted by
u( f ). The random frequency error is estimated to be ~2.57 1011 at 1 s averaging
Fig. 14 Performance test result. (Adapted from Jang et al. 2016 with permission)
18 S.-W. Kim et al.
Table 1 Uncertainty evaluation

Uncertainty sources Uncertainty
Uncertainty for phase detection 8.34 104 λ/2 (0.64 nm)
Phase offset (systematic) 8.33 104 λ/2
Repeatability (random) 2.94 105 λ/2
Uncertainty for refractive index 1.62108 L
Temperature of air (ΔT = 5mK) 4.56 109 L
Pressure of air (ΔP = 2.5 pa) 6.52 109 L
Humidity of air (ΔH = 1%) 8.79 109 L
CO2 concentration (ΔPCO2 = 41 ppm) 4.67 109 L
Ciddor’s formula 1.00 108 L
Uncertainty for frequency 5.63 1011 L
Frequency offset of fRb (systematic) 5.00 1011 L
Stability of fRb (random) 2.57 1011 L
Stability of fr (random) 3.08 1012 L
Stability of fceo(random) 1.16 1014 L
Stability of fb(random) 8.49 1014 L
Combined standard uncertainty (k = 1) [(0.64 nm)2 + (1.62 108 L )2]1/2
including the contribution from the Rb atomic clock. In addition, the systematic
frequency offset of the Rb atomic clock is estimated to be 5.00 1011. The total
uncertainty for the frequency is calculated to be 5.63 1011. In consequence, the
combined uncertainty (k = 1) of the distance measurement is worked out to be
u(L ) = [(0.64 nm)2 + (1.62 108 L)2]1/2 corresponding to a total error of 62 nm
for a target distance of 3.8 m. The combined uncertainty is dominated by the
inaccuracy of refractive index compensation. If the measurement is enacted in
vacuum such as outer space environment, the combined uncertainty improves to
u(L ) = [(0.64 nm)2 + (5.63 1011 L )2]1/2. Further, when the frequency comb as
the wavelength is stabilized to an advanced reference such as the Hg+ optical clock,
the combined uncertainty can reach to a sub-nanometer level over an extremely long
distance up to tens of hundred kilometers.
Time-of-Flight Measurement
The principle of time-of-flight (TOF) measurement of light pulses allows absolute

measurement of a distance to be performed strictly in accordance with the SI
definition based on the constancy and finiteness of the speed of light. This principle
is relatively simple to implement and widely used to determine long distances by
detecting the time duration between the start and stop pulses. The achievable
measurement accuracy is affected by not only the light pulse duration but also the
electronics adopted for pulse timing as depicted in Fig. 15. Particularly, photodetec-
tors available today offer a detection bandwidth of the order of a few picoseconds at
best, converted to a millimeter level in distance. In consequence, the employment of
Fig. 15 Balanced cross-correlation method for time-of-flight measurement using femtosecond

laser pulses
mode-locked ultrashort pulse lasers, replacing conventional long pulses, makes no

improvement straight to the measurement precision due to the slow photodetection
bandwidth. Nonetheless, the cross-correlation detection of ultrashort pulses using
nonlinear crystals permits overcoming the long-standing constraint related to slow
photodetectors, thus advancing the TOF measurement beyond the conventional
limit, even to the sub-micrometer regime, as will be described hereafter in this
section.
Figure 16 illustrates in detail how the optical cross-correlation technique can be
utilized in detecting the time-of-flight of femtosecond laser pulses (Lee et al. 2010,
2012). The measurement system is based on a Michelson interferometer, which
splits the source beam of femtosecond pulse train into two arms; one is the reference
arm equipped with a stationary mirror, and the other is the measurement arm with a
target mirror in motion. The measurement pulse train is recombined with the
reference pulse train, in which the former is adjusted to have an orthogonal polar-
ization to the latter. Then the recombined beam is passed through a nonlinear PPKTP
(periodically poled KTiOPO4) crystal that creates a sub-pulse by means of second
harmonic generation. The optical intensity of the sub-pulse becomes proportional to
the temporal overlap between two pulses. Another sub-pulse is generated by revers-
ing the recombined beam to pass through the PPKTP crystal one more time. Then,
the two sub-pulses are detected in a balanced scheme, of which the output electrical
signal is proportional to the temporal offset between the measurement and reference
pulses. This method, named the balanced cross-correlation (BXCOR), enables
ultrashort pulses to be detected with rejection of common opto-electrical noise.
The BXCOR output electrical signal yields an S-shaped response to the temporal
offset between the two sub-pulses as shown in Fig. 16. The zero-crossing point of the
BXCOR signal indicates the very moment when the measurement pulse is in
complete overlap with the reference pulse. Near the zero-crossing point, a highly
linear span exits with a steep signal sensitivity of mV/fs.
For the sake of TOF measurement, the pulse repetition rate ( fr) of the femtosec-
ond laser source is controlled so that the BXCOR signal is precisely locked to its
zero-crossing point for a given target distance L. This makes the TOF between the
reference and measurement pulses become an integer multiple of the pulse-to-pulse
20 S.-W. Kim et al.
Fig. 16 Overall configuration of balanced cross-correlation for time-of-flight measurement using

femtosecond laser pulses
spacing, so the target distance is determined as L = mc/(2frng) with m being the

integer, c the speed of light in vacuum, and ng the group refractive index of air. In
order to determine the integer value of m, the pulse repetition rate ( fr) is changed to
yield another locking, i.e., L = (m + 1)c/(2( fr + Δfr)ng), which leads to m = fr /Δfr
with Δfr being the increment of fr. The optical hardware system of Fig. 16 relies on
an Er-doped fiber laser as the light source, emitting 220 fs pulses with optical power
of 240 mW. The BXCOR signal has a high sensitivity of ~3 mV/fs with background
noise of 27.2 μV/Hz1/2. The instant pulse repetition rate is precisely determined with
a high-speed frequency counter which is referenced to the Rb atomic clock.
Figure 17 illustrates an exemplary result of long distance ranging. The TOF
interferometer system was installed on the top floor of a ME building at the
KAIST campus. The target mirror made of a 25 mm corner with retroreflector was
located on a remote building 700 m away with a 1.4 km return path. The target
distance was measured at an update rate of 200 Hz to have a mean value of
695.798133809 m. The measured data was subject to a low-frequency fluctuation
of a ~20 μm amplitude, which was mainly induced by the refractive index of air. The
measurement resolution was verified by applying a 150 nm amplitude sinusoidal
modulation to the target mirror by activating a piezoelectric actuator at a 10 Hz
frequency. The measured distance clearly shows the given modulation with a
Fig. 17 Time-of-flight measurement using femtosecond laser pulses for a long distance of 700 m.
(Adapted from Lee et al. 2010 with permission)
nanometer resolution. Furthermore, the Fourier-transformed data clearly indicates a

10 Hz peak that corresponds to the modulation frequency on the target mirror. The
environmental disturbance is centered near 2 Hz in the frequency domain.
The measurement precision is estimated by analyzing the BXCOR signal
obtained during measurement as shown in Fig. 18. The jitter in the BXCOR signal
is converted to the time-dependent fluctuation of the measured distance. For a long
22 S.-W. Kim et al.
Fig. 18 Measurement stability in terms of the Allan deviation. (Adapted from Lee et al. 2010 with
permission)
distance of ~0.7 km, the fluctuation is influenced mainly by the air refractive index
that changes severely with air turbulence arising along the measurement optical path.
For a very short distance of 1.5 m, on the other hand, the measured fluctuation is
induced dominantly by the intrinsic optical and electrical instability of the measure-
ment system itself since the effect of the air refractive index is not relatively
significant. Thus, the short-term fluctuation represents the ultimate precision of the
measurement system, which can be achieved without air particularly for outer space
missions. The ultimate precision is estimated to be 8.7 nm at 10 ms averaging, which
improves further to 1.1 nm at 1 s averaging.
Figure 19 shows another TOF measurement system configured to measure
multiple targets simultaneously. This system incorporates the dual-comb method in
detecting ultrashort pulses together with the optical cross-correlation technique. The
light source is made of two lasers; one is referred to as the signal laser and the other
as the local laser. In fact, both are Er-doped fiber femtosecond lasers emitting 90 fs
pulses with 10 mW optical power. The signal laser is used as the measurement beam
with a repetition rate of 100 MHz. The local laser is set to operate at a slightly
Fig. 19 Measurement system for multiple-target absolute ranging. C circulator, DOE diffractive
optical element, PBS polarizing beam splitter, L lens, M target mirror, PD photodetector, DM
dichroic mirror, and PPKTP periodically poled KTP crystal
different repetition rate of 99.998 Mz, i.e., with a difference of 2.0 kHz from that of
the signal laser. A diffractive optical element (DOE) is used to split the signal laser
beam toward four different targets. The reference point Mref of distance measure-
ment is established on the front surface of the DOE with a coating of partial
reflection. Every pulse of the signal laser is sent to all the targets and reflected
backward. So the pulse from Mref arrives first, followed by other pulses reflected
from M1, M2, etc.
Figure 20 illustrates how the principle of balanced cross-correlation (BCC) is
implemented by adopting dual-comb interferometry. The distance from Mref to each
target is determined as d = cΔt/(2 N ) with c being the speed of light in vacuum, N the
group refractive index of air, and Δt the time-of-flight of the pulse between Mref and
the target. The BCC method by dual-comb interferometry operates with the signal
laser of a repetition rate fr and the local laser with an offset Δfr from fr. Then, the
BCC signal is generated with a period of 1/Δfr, which permits the time interval
between the reference pulse signal and target pulse signals in the down-converted
time scale by a factor of fr /Δfr. The experimental result shows an actual S-shaped
BCC signal sampled using a 14-bit digitizer at a sampling rate of 200 MHz. The
effective sampling rate of the BCC signal displayed in the “effective time” scale is
calculated to be 10.0 THz. The effective sampling rate referred to as the up-
conversion sampling rate is defined as fs0 = fs ( fr/Δfr), which implies that the
timing resolution in the effective time domain is equivalent to 0.1 ps. The time-of-
flight Δt is measured by locating the peak point of the cross-correlation signal for
each pulse. This dual-comb sampling method brings a drastic improvement in the
resolution of the pulse timing for determining Δt.
Figure 21 presents test results obtained to evaluate the performance in terms of the
measurement repeatability, linearity, and speed. Firstly, the repeatability is given in
terms of the Allan deviation for two distances, d1 = 0.015 m and d2 = 1.6 m. The
repeatability turns out 0.936 μm for d1 at 0.5 ms averaging, which is equivalent to
0.01 ps in standard deviation. The repeatability improves to 17 nm for 0.5 s sampling
time. The distance d1 is short enough to neglect the environmental disturbance due to
24 S.-W. Kim et al.
Fig. 20 Cross-correlation processing. (a) An exemplary BCC signal sampled from three targets.
(b) Reference BCC signal pattern for cross-correlation. (c) Cross-correlation between the
reference and the measurement BCC signal. (d) Enlarged views of cross-correlation curves for
peak detection. Yellow lines indicate original sampled BCC signals, and red curves are calculated
cross-correlation data
air temperature fluctuation, mechanical vibration, and air turbulence. Thus, the
repeatability for d1 can be considered as the intrinsic noise level of the measured
time-of-flight Δt. The repeatability for distance d2 is worse due to the influence of
external disturbance. The linearity test was conducted in comparison to an incre-
mental HeNe laser interferometer. The target mirror M4 was moved on an aerostatic
stage over a 1.0 m travel in 100 mm steps. The relative linearity is 79 nm in standard
deviation or 279 nm in peak-to-valley. No noticeable cyclic error is observed. Next
the measurement speed was verified by applying a ~10 μm modulation through
piezoelectric actuators (PZTs) placed underneath the targets. to the targets. The
modulation frequency was 10, 20, and 30 Hz for M1, M2, and M3, respectively.
The result shows that the distance variation due to the sinusoidal modulation of each
target is precisely observed, confirming that the multiple absolute distances with
sinusoidal modulations are clearly resolved in the frequency domain with a high
signal-to-noise ratio by means of Fourier frequency analysis.
Fig. 21 Absolute multi-target ranging performance. (a) Repeatability in Allan deviation at

two distances. (b) Linearity test result compared with an incremental HeNe laser interferometer.
(c) Speed of absolute ranging of multiple target objects. Three targets were modulated at 10, 20, and
30 Hz, respectively, using PZTs. During all the tests, the environmental disturbance was controlled
as specified in Table 1 in terms of temperature, pressure, humidity, and CO2 concentration.
(Adapted from Han et al. 2015 with permission)
26 S.-W. Kim et al.
Figure 22 shows the multi-degree-of-freedom (multi-DOF) sensing ability of the

measurement system. The target object is configured as a rigid body to simulate the
satellite motion, from which multiple target points are distance-measured. Four
target points are selected with 90 apart from each other and illuminated simulta-
neously along the 1st-order diffraction lines of the DOE. The target distances are
given 1.0 mm initial offsets from each other in order to distinguish them in the
BCC signal sampled in the effective time domain. The local Cartesian coordinates
Fig. 22 Distance and angle measurement for multi-DOF sensing of a rigid body motion.
(a) Configuration of the experiment and (b) measurement geometry
are set on the rigid body so that the x-axis extends from M1 to M3 while the y-axis
from M2 to M3. The normal vector set on the origin of the x-y plane indicates the
z-axis direction. The orientation of the normal vector is described by the yaw angle
θx and the pitch angle θy defined with respect to the 0-th order diffraction line of
the DOE. Having measured the distances d1, d2, d3, and d4 to the four target mirrors,
the nominal distance d to the rigid body is calculated by the arithmetic mean as
d = (d1 + d2 + d3 + d4)/4. At the same time, the yaw and pitch angles representing the
attitude of rigid body are calculated instantaneously: θx = sin1[(d1 d3)/A] and
θy = sin1[(d2 d4)/A].
Figure 23 shows the measurement result obtained from a nominal distance of
~3.7 m. The repeatability of angular measurement turns out 5.289 arcsec for a
single measurement of 0.5 ms, which decreases to 0.073 arcsec, increasing the
averaging time to 0.5 s. Next, when the target object is given a continuous tilt
motion with a 1 Hz modulation frequency, the four distances are measured with
difference amplitudes. Fourier-transform frequency analysis reveals that the sinu-
soidal modulation is clearly resolved not only from the measured target distances but
also the calculated yaw and pitch angles with a high signal-to-noise ratio. It is also noted
that second and third harmonic peaks caused by imperfect modulation are also observed
at 2 Hz and 3 Hz.
a b
Allan deviation (arcsec)
3 d1
d2
-3681.7500 mm
Distance (mm)
100 d3
2
d4
Angle (arcsec)
-6300
10-1
1
-6250
0 1 2 3 4 5
Time (s)
10-2
10-3 10-2 10-1 100 0 2 4 6 8 10
Averaging time (s) Time (s)
c d
-7000 qx 300 qx
Amplitude (a.u.)
-6800 200
Angle (arcsec)
100
-6600 0
-5200 qy 300 qy
200
-5000
100
-4800 0
0 2 4 6 8 10 1 2 3 4 5 6 7 8
Time (s) Frequency (Hz)
Fig. 23 Angle measurement test result. (a) Repeatability in terms of Allan deviation. (b) Distance
(d1, d2, d3, and d4) to four targets under 1 Hz excitation. (c) Reconstruction angular motions.
(d) Fourier transformation spectra. (Adapted from Han et al. 2015 with permission)
28 S.-W. Kim et al.
3-D Surface Measurements
Conventional optical interferometric methods for 3-D surface profile measure-

ments rely on either continuous wave lasers or white-light lamps as the light
source. Continuous wave lasers such as HeNe gas lasers or solid-state lasers permit
profiling fine surfaces with high precision and alignment flexibility between the
reference and target surfaces due to their high spatial and temporal coherence.
Rough surfaces are preferably dealt with by low-coherence interferometry using a
white-light source such as a tungsten-halogen lamp offering low temporal coher-
ence. The use of a white-light source requires an equal-path interferometer con-
figuration in which the optical path difference between the reference and target
surfaces needs to be near zero. The spatial coherence of a white-light source is also
very low, which confines the lateral field-of-view of interferometric measurement
to be a narrow area.
The advent of ultrashort mode-locked lasers began to bring forth new possibilities
of advancing the extent of low-coherence interferometry beyond the conventional
limitation in several respects. Firstly, the spatial coherence can be increased to allow
for large field-of-view profilometry by configuring nonsymmetric schemes to deal
with large surfaces effectively with a smaller reference mirror. Secondly, the tem-
poral coherence can be customized to a given surface simply by adjusting the pulse
width Δt. The spectral bandwidth Δv that relates to the temporal coherence is given
by the relation of ΔtΔv K with K being a factor determined by the pulse shape.
Thirdly, the vertical measuring range along the surface height direction can be
extended significantly by configuring unequal interferometer schemes capable of
providing a long delay line to allow two distinct pulses generated at different times
from the source to interfere. Fourthly, the scanning mechanism needed for low-
coherence interferometry can be eliminated by control of the pulse repetition rate of a
mode-locked laser, which permits high-speed, long-range measurements to be real-
ized without increasing the hardware complexity in configuring the needed interfer-
ometer system.
Figure 24 shows two nonsymmetric schemes of low-coherence interferometry
demonstrated using a mode-locked laser as the light source (Oh and Kim 2005). One
is an unequal-arm Fizeau interferometer, in which a large test surface can be
positioned far from the reference surface with alignment flexibility. The other is a
nonsymmetric Twyman-Green interferometer configured to test a large surface with
a small reference surface. With a Ti/sapphire femtosecond laser as the light source, a
temporally resolved interferogram is sampled at a single CCD pixel while scanning
the reference mirror. The carrier signal is clearly seen together with a narrow
envelope peak, being suitable for high precision low-coherence scanning measure-
ment. The spatial interferogram captured over the entire CCD pixels within a single
frame also demonstrates an excellent fringe contrast without notable parasite fringe
noise, being superior to other spatial interferograms obtained from a HeNe laser and
a halogen lamp with filtering.
Figure 25 shows an example of using a fiber-type mode-locked laser, of which
the pulse repetition rate ( fr) is varied by extending the oscillator cavity length (Lc) as
Fig. 24 Low-coherence scanning interferometry using a mode-locked laser. (a) Unequal-path

Fizeau interferometer system. (b) Unequal-path, nonsymmetric Twyman-Green interferometer
system. (c) Typical temporal low-coherence interferogram. (d) Comparison of spatial interfero-
grams from a Ti/sapphire mode-locked laser, a HeNe gas laser, and a halogen lamp. MMF
multimode fiber of a 200 m core diameter. SMF single-mode fiber of a 3 μm core diameter.
(Adapted from Oh and Kim 2005 with permission)
fr = c/Lc with c being the speed of the light (Joo et al. 2013). The interferometer
system is configured as an unequal-path type with a fiber spool inserted in the
measurement arm to provide an optical path difference of mLc between the
measurement and reference arms with m being a large integer. In consequence, the
interference on the CCD camera is made between the 0-th reference pulse and the m-
th measurement pulse as illustrated in the figure. Then the optical scanning induced
by shifting the pulse repetition rate of the fiber oscillator is augmented by a factor of
m. This unequal-path low-coherence interferometer using a fiber-type mode-locked
laser is well suited for the industrial inspection of microelectronics products whose
feature heights are much larger, up to a few millimeters, than the light wavelength.
Furthermore, the high spatial coherence of the mode-locked laser allows a large
field-of-view up to a few tens of millimeters.
The fiber laser used as the light source is an Er-doped fiber oscillator emitting
infrared 100 fs pulses of a center wavelength of 1560 nm. The infrared pulses are
converted to the visible range by means of second harmonic generation through a
PPLN (periodically poled lithium niobate) crystal so that an ordinary CCD camera
30 S.-W. Kim et al.
Fig. 25 3-D profilometry of unequal-path low-coherence scanning interferometry using an

Er-doped fiber oscillator. (a) Interferometer system layout. (b) Interference during fr-scanning.
(c) Scanning range variations with various fr-tuning methods by a PZT (piezoelectric transducer),
an EOM (electro-optic modulator), and a motorized mechanical stage. Other abbreviations are SMF
single-mode fiber, DCF dispersion compensating fiber, DM dichroic mirror, PPLN periodically
poled lithium niobate, and Lc cavity length of the fiber oscillator. (Adapted from Joo et al. 2013 with
permission)
can be used for monitoring the resulting interferogram. Further, the fr-tuning range
varies with different devices used to extend the oscillator cavity lengths, such as a
piezoelectric actuator (PZT), electro-optic modulator (EOM), and motorized
mechanical stage. The EOM permits a fast fr-tuning speed but its total range is
very limited to a few tens of Hz. On the other hand, when a motorized mechanical
stage is used, the tuning speed is slow, but the range can be made extensive. As
illustrated in the figure, when the fr-tuning range of a mechanical stage is combined
with a long unequal-arm length of ~300 m, the low-coherence scanning interferom-
etry can be performed up to a 1 m range. Finally, Fig. 26 shows several measurement
examples obtained using an EOM and a PZT to demonstrate the measuring precision
and field-of-view in the vertical and lateral directions.
Fig. 26 Measurement examples. (a) Interferograms during fast fr-scanning using an intracavity
electro-optic modulator (EOM). (b) Reconstructed 3-D height map from interferograms of (a).
(c) Interferograms from a 69.624 μm step-height specimen by fr-scanning with a piezoelectric
actuator (PZT). (d) Reconstructed 3-D surface profile from interferograms of (c). (e) Schematic for
step-height measurement over a large field-of-view. (f) 3-D profile over 21 16 mm (or 14.5 mm)
field-of-view. (Adapted from Joo et al. 2013 with permission)
Summary and Outlook
The research work pioneered at KAIST in recent years is demonstrating that mode-
locked lasers provide the potential of making a significant contribution to the
progress of dimensional metrology made during the last several years by means of
laser-based optical interferometry. In comparison to conventional single-wavelength
lasers or broad spectral light sources, mode-locked lasers are capable of enhancing
the performance of optical interferometry to meet the ever-growing demand on the
measurement uncertainty and functionality. This optimistic prediction is based on
several spectral and temporal characteristics uniquely provided mode-locked lasers.
Firstly, the frequency comb of a mode-locked laser offers a large number of discrete
32 S.-W. Kim et al.
optical frequencies evenly distributed over a wide spectral bandwidth. This permits
generating multiple wavelengths with precise traceability to the atomic clock, so long
distances up to a few meters can be measured in an absolute way without losing the
measurement uncertainty of laser-based interferometry. Secondly, ultrashort pulses
emitted from a mode-locked laser in the time domain can be utilized to implement
time-of-flight measurement with a timing resolution at a sub-femtosecond level,
allowing the measurement precision in the nanometer regime. Thirdly, a mode-locked
laser is able to extend the 3-D surface profiling capability of optical interferometry as a
new source providing well-controlled temporal coherence together with high spatial
coherence. This unique characteristic permits a rough surface to be profiled with a
large field-of-view beyond the capability of white-light sources long used to conduct
low-coherence optical interferometry.
Despite the many advantages discussed so far here, not many mode-locked lasers
have yet been put to actual industrial applications of dimensional metrology. The
main reason is due to the cost and complexity of mode-lacked lasers available today.
Much effort is being exerted worldwide to make the use of mode-locked lasers be
more robust to handle at a reasonable price. No doubt soon in the very near future,
mode-locked lasers will be a general tool of dimensional metrology, which will
benefit manufacturing industries by effectively responding to the ever-growing
demand on the measurement precision and functionality.
Acknowledgment This work was supported by the National Research Foundation of the Republic
of Korea (NRF-2012R1A3A1050386).
References
Bender PL, Currie DG, Poultney SK, Alley CO, Dicke RH, Wilkinson DT, Eckhardt DH, Faller JE,
Kaula WM, Mulholland JD, Plotkin HH, Silverberg EC, Williams JG (1973) The lunar laser
ranging experiment. Science 182(4109):229–238
Berkovic G, Shafir E (2012) Optical methods for distance and displacement measurements. Adv
Opt Photon 4(4):441–471
Birch KP, Downs MJ (1994) Correction to the updated Edlen equation for the refractive index of air.
Metrologia 31(4):315–316
Bobroff N (1993) Recent advances in displacement measuring interferometry. Meas Sci Technol
4(9):907–926
Chun BJ, Hyun S, Kim S, Kim S-W, Kim Y-J (2013) Frequency-comb-referenced multi-channel
fiber laser for DWDM communication. Opt Express 21(24):29179–29185
Chun BJ, Kim Y-J, Kim S-W (2016) Inter-comb synchronization by mode-to-mode locking. Laser
Phys Lett 13(8):085301
Ciddor PE (1996) Refractive index of air: new equations for the visible and near infrared. Appl Opt
35(9):1566–1573
Coddington I, Swann WC, Nenadovic L, Newbury NR (2009) Rapid and precise absolute distance
measurements at long range. Nat Photonics 3(6):351–356
Dandliker R, Thalmann R, Prongue D (1988) Two-wavelength laser interferometry using super-
heterodyne detection. Opt Lett 13(5):339–341
Degnan JJ (1985) Satellite laser ranging: current status and future prospects. IEEE Trans Geosci
Remote Sens GE-23(4):398–413
Demarest FC (1998) High-resolution, high-speed, low data age uncertainty, heterodyne displace-
ment measuring interferometer electronics. Meas Sci Technol 9(7):1024–1030
Diddams SA, Jones DJ, Ye J, Cundiff ST, Hall JL, Ranka JK, Windeler RS, Holzwarth R, Udem T,
Hansch T-W (2000) Direct link between microwave and optical frequencies with a 300 THz
femtosecond laser comb. Phys Rev Lett 84(22):5102–5105
Felder R (2005) Practical realization of the definition of the metre, including recommended
radiations of other optical frequency standards (2003). Metrologia 42(4):323–325
Fujima I, Iwasaki S, Seta K (1998) High-resolution distance meter using optical intensity modula-
tion at 28 GHz. Meas Sci Technol 9(7):1049–1052
Gao W, Kim S-W, Bosse H, Haitjema H, Chen YL, Lu XD, Knapp W, Weckenmann A, Estler WT,
Kunzmann H (2015) Measurement technologies for precision positioning. CIRP Ann Manuf
Technol 64(2):773–796
Giacomo P (1984) News from the BIPM. Metrologia 20(1):25–30
Han S, Kim Y-J, Kim S-W (2015) Parallel determination of absolute distances to multiple targets
by time-of-flight measurement using femtosecond light pulses. Opt Express 23(20):
25874–25882
Hartmann L, Meiners-Hagen K, Abou-Zeid A (2008) An absolute distance interferometer with two
external cavity diode lasers. Meas Sci Technol 19(4):045307
Hyun S, Kim Y-J, Kim Y, Jin J, Kim S-W (2009) Absolute length measurement with the frequency
comb of a femtosecond laser. Meas Sci Technol 20(9):095302
Hyun S, Kim Y-J, Kim S-W (2010) Absolute distance measurement using the frequency comb of a
femtosecond laser. CIRP Ann Manuf Technol 59(1):555–558
Jang Y-S, Kim S-W (2017) Compensation of the refractive index of air in laser interferometer for
distance measurement: a review. Int J Precis Eng Manuf 18(12):1881–1890
Jang Y-S, Wang G, Hyun S, Kang HJ, Chun BJ, Kim Y-J, Kim S-W (2016) Comb-referenced laser
distance interferometer for industrial nanotechnology. Sci Rep 6:31770
Jin J, Kim Y-J, Kim Y, Kim S-W, Kang C-S (2006) Absolute length calibration of gauge blocks
using optical comb of a femtosecond pulse laser. Opt Express 14(13):5986–5974
Jin J, Kim Y-J, Kim Y, Kim S-W (2007) Absolute distance measurements using the optical comb of
a femtosecond pulse laser. Int J Precis Eng Manuf 8(4):22–26
Jones DJ, Diddams SA, Ranka JK, Stentz A, Windeler RS, Hall JL, Cundiff ST (2000) Carrier-
envelope phase control of femtosecond mode-locked lasers and direct optical frequency syn-
thesis. Science 288:635–639
Joo K-N, Kim S-W (2006) Absolute distance measurement by dispersive interferometry using a
femtosecond pulse laser. Opt Express 14(13):5954–5960
Joo W-D, Kim S, Park J, Lee K, Lee J, Kim S, Kim Y-J, Kim S-W (2013) Femtosecond laser pulses
for fast 3-D surface profilometry of microelectronic step-structures. Opt Express
21(13):15323–15334
Keller U (2003) Recent developments in compact ultrafast lasers. Nature 424(6950):831–838
Kim S-W (2009) Metrology: combs rule. Nat Photon 3:313–341
Kim J, Song Y (2016) Ultralow-noise mode-locked fiber lasers and frequency combs: principles,
status, and applications. Adv Opt Photon 8:465–540
Kim Y, Kim S, Kim Y-J, Hussein H, Kim S-W (2009a) Er-doped fiber frequency comb with mHz
relative linewidth. Opt Express 17(14):11972–11977
Kim Y-J, Kim Y, Chun BJ, Hyun S, Kim S-W (2009b) All-fiber-based optical frequency generation
from an Er-doped fiber femtosecond laser. Opt Express 17(13):10939–10945
Lee J, Kim Y-J, Lee K, Lee S, Kim S-W (2010) Time-of-flight measurement using femtosecond
light pulses. Nat Photonics 4(10):716–720
Lee J, Lee K, Lee S, Kim S-W, Kim Y-J (2012) High precision laser ranging by time-of-flight
measurement of femtosecond pulses. Meas Sci Technol 23(6):065203
Minosima K, Matsumoto H (2000) High-accuracy measurement of 240-m distance in an optical
tunnel by use of a compact femtosecond laser. Appl Opt 39(30):5512–5517
Oh J-S, Kim S-W (2005) Femtosecond laser pulses for surface-profile metrology. Opt Lett
30(19):2650–2652
34 S.-W. Kim et al.
Quinn TJ (2003) Practical realization of the definition of the metre, including recommended
radiations of other optical frequency standards (2001). Metrologia 40(2):103–133
Salvadé Y, Schuhler N, Lévêque S, Floch SL (2008) High-accuracy absolute distance measurement
using frequency comb referenced multiwavelength source. Appl Opt 47(14):2715–2720
Schuhler N, Salvade Y, Leveque S, Dandliker R, Holzwarth R (2006) Frequency-comb-referenced
two-wavelength source for absolute distance measurement. Opt Lett 31(21):3101–3103
Uttam D, Culshaw B (1985) Precision time domain reflectometry in optical fiber systems using a
frequency modulated continuous wave ranging technique. J Lightwave Technol 3(5):971–977
van den Berg SA, Persijn ST, Kok GJP, Zeitouny MG, Bhattacharya N (2012) Many-wavelength
interferometry with thousands of lasers for absolute distance measurement. Phys Rev Lett
108(18):183901
Wang G, Jang Y-S, Hyun S, Chun BJ, Kang HJ, Yan S, Kim S-W, Kim Y-J (2015) Absolute
positioning by multi-wavelength interferometry referenced to the frequency comb of a femto-
second laser. Opt Express 23(7):9121–9129
Xiaoli D, Katuo S (1998) High-accuracy absolute distance measurement by means of wavelength
scanning heterodyne interferometry. Meas Sci Technol 9(7):1031–1035
Nanopositioning and Nanomeasuring
Machines 2
Eberhard Manske
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
Basic Measurement Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
Machine Concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
Length Measurement Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
Laser Interferometer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
Stabilized He-Ne Laser . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
Nanoprobes and Multi-Probe Concept . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
Measurement Uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 48
Metrological Properties and Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
Summary and Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 57
Abstract
Continuing engineering progress in precision fabrication technologies, especially
in the semiconductor industry, precision optics fabrication, and the diversified
micro- and nanotechnologies, stimulates the advance in precision metrology.
Fabricated structures reach atomic dimensions in ever-larger areas, thus becom-
ing more and more complex, also in three dimensions. Consequently, measure-
ments are made – to an increasing extent – of larger surface regions and sidewalls
with higher aspect ratios as well as fully 3D micro- and nanostructures. Advanced
high precision measurement technology is more and more an enabling technology
for nanotechnologies. Today, nanopositioning and nanomeasuring machines pro-
vide high-precision measurements and the positioning of objects across different
scales, from subnanometers up to several centimeters. This chapter deals with the
E. Manske (*)
Department of Mechanical Engineering, Technische Universität Ilmenau, Ilmenau, Germany
e-mail: eberhard.manske@tu-ilmenau.de

https://doi.org/10.1007/978-981-10-4938-5_2
36 E. Manske
requirements for highest measurement performance at the limits of physics and

technology, resulting from the progress and the goals of modern high-tech
fabrication technologies. The fundamentals of the nanopositioning and nano-
measuring machine, developed at the Institute of Process Measurement and
Sensor Technology of the Ilmenau University of Technology and manufactured
by the SIOS Meßtechnik GmbH Ilmenau, are described, and the measurement
capabilities, potential applications, progress in research, and prospects of the
device for the near future are pointed out.
Keywords
Nanopositioning and nanomeasuring machines · Abbe comparator principle ·
Laser interferometer systems · Nano probes
Introduction
The enormous growth-stimulating potential of future-oriented technologies such

as nanotechnology or advanced optical technologies is one of the supporting
pillars for meeting the economic and social challenges of the twenty-first century.
Here, the speed at which this development process takes place as well as its
efficiency depends, to an ever greater extent, on a powerful device-engineering
infrastructure. Therefore, the manufacture and in particular the analysis of ever
larger objects with nanoscale structures and features – from semiconductor tech-
nology, microsystems technology via precision optics up to bioanalytics – require
new concepts in the three-dimensional multi-scale nanopositioning and nano-
measurement technology. Whereas up to now a 2D positioning and a 2.5D
measuring machine have been sufficient in the semiconductor industry, the addi-
tional, metrologically traceable positioning and nanoprecision measurement in the
third dimension is becoming more and more important due to ever more complex
spatial structures in microsystems technology and in precision optics, but also in
semiconductor metrology.
In order to meet those high requirements, nanopositioning and nanomeasuring
machines (NPM-machines) are supposed to enable the positioning, sampling, mea-
surement, analysis, modification, and manipulation of three-dimensional objects in
large spatial ranges with nanometer precision (Manske et al. 2012). (NPM-machines)
are supposed to enable the positioning, sampling, measurement, analysis, modifica-
tion, and manipulation of three-dimensional objects in large spatial ranges with
nanometer precision. Thus, their functionality largely exceeds that of micro- and
nano-coordinate measuring machines. They present not only a highly developed
nearly error-free measuring concept and an adequate machine concept, but also the
most modern high-precision multicoordinate measuring systems in combination with
variable nanoscanning systems.
A multisensor concept allows a variety of metrological applications. Some first
approaches concerning the application of NPM-machines for nanofabrication are
presented.
2 Nanopositioning and Nanomeasuring Machines 37
Basic Measurement Approach
In 1890, Ernst Abbe formulated the Abbe-comparator principle, as follows: “The

measuring apparatus is to be arranged in such a way that the distance to be measured
is a straight-line extension of the graduation used as a scale” (Abbe 1890).
In the case of a real length measurement where tilts are inevitable, such an
arrangement allows errors of first order to be avoided. The development of
3D coordinate measuring machines represented enormous technical progress.
High-precision air bearing guides permit only very small tilts of the 3D scanning
system. However, due to the scanning system, which can be moved in three axes, the
distance to be measured and the scale are arranged mostly in the x- and y-axis in
a nonaligned way, thus limiting the achievable measurement uncertainty.
In order to realize highest precision, it is necessary to apply the Abbe comparator
principle in all three coordinate axes. The Abbe error in three dimensions can be
described as follows:
Δl i ¼ l offi sin αi 8i x,y,z (1)
This means that the Abbe offset loff should be close to zero in every axis.
l offi ! 0 8i x,y,z (2)
The inevitable consequence is that the probe system must be immobile and,
thus, must act as zero indicator (in the case of 3D sensors, in all three coordi-
nates), which allows the three orthogonal measuring axes to meet in the common
probing point (of the fixed sensor) at any point in time. For making a measure-
ment, the measuring object is moved in three dimensions such that it is scanned
sequentially by the zero indicator. In doing so, the displacement is measured
three-dimensionally by laser interferometers. Here, it is necessary for the mea-
suring object to be positioned on an orthogonal mirror corner, which is scanned
three-dimensionally by the three laser interferometers (Fig. 1). Thus, the position
of the measuring object can be detected in three dimensions synchronously to
probing with the nanosensor. It must be assumed that – after minimizing the
Abbe offset – a residual deviation of Δloff of the order of magnitude of
100–500 μm will remain (Fig. 2).
Thus, this deviation results in a residual Abbe error Δl.
Δl i ¼ Δl offi sin αi 8i x,y,z (3)
For good linear and planar guidance systems, angular deviations of the order
of magnitude of 3000 must be expected. This leads to residual errors (cf. Eq. 3)
of 15–73 nm. Even high-precision air bearings present guidance deviations of 200 and
more, which means residual deviations of 1–5 nm. In order to achieve smaller
guidance deviations, an additional angular measurement as well as a permanent
compensation for the tilts are necessary. In this way, the Abbe error can be further
minimized.
38 E. Manske
Fig. 1 3D Abbe comparator principle of the NPM-machine
Fig. 2 Occurrence of a residual Abbe error
αi ! 0 8i x,y,z (4)
An angular measurement and regulation can be carried out in all three coordinate
axes by employing additional autocollimator sensors (Fig. 1). In this way, a conven-
tional PID-control enables angular deviations <0.0500 to be achieved (Schmidt et al.
2007). Thus, the Abbe-error can be reduced significantly to 0.02–0.12 nm. On the
basis of this metrological concept of the consistent application of a measuring
principle presenting minimal errors in all six dimensional degrees of freedom
(3 lengths, 3 angles), a nanopositioning and nanomeasuring machine can be set up
which is able to achieve nanometer to subnanometer precision.
Machine Concept
Hence, a corresponding machine concept can be derived from this basic metrological
approach. The concept is metrologically based on three laser interferometers, two
autocollimator sensors, and some corresponding nanosensor systems. They must be
arranged to each other with highest accuracy and long-term stability. This is realized by
a mechanically stable metrological frame with slightest thermal expansion (cf. Fig. 3).
The nanosensor systems only serve as zero indicators, thus presenting nearly no
measuring movement at all. The measurement table has the shape of a Cartesian mirror
corner, also made of Zerodur, and must be able to house the device under test (DUT).
Furthermore, it must be moved three-dimensionally in the space. This is realized via
three orthogonal rolling element linear guidance systems, which can be moved by linear
actuators both over large travel ranges and in a nanometer-sensitive range.
As the mirror corner must always be adjusted in vertical direction at a well-
defined height, it is advantageous to employ an electromechanical weight force
compensation. Thus, the driving energy of the linear actuators for height movement
and angular control can be minimized.
Various temperature sensors, which are shown in Fig. 3, are used for measuring
and correcting the influence of the refractive index on the laser wavelength of the
interferometer systems.
A stable Zerodur cover plate (top plate mount Fig. 3) closes the metrological
frame over the nanoprobe systems (cf. Fig. 4). Here, the nanoprobe systems are
Fig. 3 Basic design of a nanopositioning and nanomeasuring machine (without probing systems)
40 E. Manske
Fig. 4 Mounted laser focus probe
arranged in such a way that the probe tip (or also the laser focus of an optical sensor)
lies exactly at the intersection of the three interferometer measuring axes.
A universal probe mount is well able to hold different nanoprobe systems such as
laser focus probe, according to Fig. 4.
Length Measurement Systems
Laser Interferometer
In order to carry out a three-dimensional interferometric length measurement versus

three orthogonal mirror surfaces, plane mirror interferometers must be provided.
Here, it is necessary to ensure that the invariance of the interferometer is maintained
in the case of the tilting of the measurement mirror. In general, this can be realized
through a special interferometer configuration using triple prisms. However, this
implies that two laser beams hit the plane mirror at a more or less defined distance.
Therefore, for providing a common intersection of three interferometer axes, it is
considerably easier and more appropriate to employ single-beam laser interferome-
ters. Here, the tilting invariance is achieved through a special optical diaphragm
configuration (Büchner and Jäger 2006). These interferometers constitute a decisive
prerequisite for minimizing both the Abbe offset loff in all three axes and the
interferometric length measurement error (Fig. 5).
Fig. 5 Plane mirror

interferometer used in
NPMM. PSB – polarizing
beam splitter; BS –
non-polarizing beam
splitter; R, M – plane
reference and measuring
mirrors; D – diaphragm; 1, 2,
3 – quarter-wave plates; PD1,
PD2 – photodetectors; and
4, 5 – polarizers (Büchner and
Jäger 2006)
A suitable signal processing and demodulation of the photoelectric signals

make it possible – on the basis of the laser wavelength used (632.8 nm) – to obtain
a smallest measurable increment of 0.02 nm (Hausotte et al. 2012). The measuring
values of the three interferometer axes are recorded at a synchronicity of 0.2 ns.
Assuming a travel speed of 10 mm/s, the synchronization error is <2 pm.
The wavelength correction of the laser is carried out via the measurement of the
temperature, pressure, and humidity by applying the Edlen formula.
In order to minimize the temperature dependence and the dead path error of the
interferometer in der NPM-machine, the measuring and the reference beam have
been adapted in such a way that a good symmetry at the stage’s center position is
achieved (cf. Fig. 6).
In addition low expansion material (Invar) was used for the reference mirror
holder. The reference mirror mounting was implemented in a virtually self-contained
massive body providing higher mechanical stability, lower oscillation affinity, better
thermal conduction, and thus more homogeneous temperature profiles within the
reference beam path.
On the basis of this interferometer arrangement, a high stability of the length
measurement in the machine of 25 pm standard deviation was achieved (Fig. 7).
Stabilized He-Ne Laser
Due to the fact that the metrological requirements to be met are extremely high, only
stabilized He-Ne lasers can be considered as laser source for the plane mirror
interferometers. Both the large size and the high thermal load of these lasers are
very disadvantageously for a high precision machine. Therefore, the laser light is
feed into the interferometers via optical fibers. Thus, the thermal influences are
greatly reduced, which permits a very compact mechanical design to be realized
42 E. Manske
Fig. 6 Symmetric plane mirror interferometer
Fig. 7 Stability of laser interferometer measurement
(cf. Fig. 8). The He-Ne lasers thermally stabilized according to the two-mode
comparative method reach a relative frequency stability of < 2.109 (http://
www.sios-de.com/products/stabilized-hene-lasers).
Nanoprobes and Multi-Probe Concept
Now, several probes of most different types can be integrated into the NPM-machine.
They operate as zero point indicators. This implies the implementation of a multiple
sensor arrangement. Here, again, it is advantageous to use a modular sensor concept
allowing the NPM-machine to act as an excellent metrological basis with subnanometer
resolution, a large measurement range, and high nanometer precision.
Fig. 8 Stabilized OEM He-Ne Laser with fiber coupling. (Courtesy SIOS Meßtechnik GmbH)
The key probe system of the NPM-machine is a laser focus probe (Mastylo et al.
2005) (cf. Fig. 9). The central component part of the sensor is a so-called hologram
laser unit (Yoshida et al. 1990). Normally, this unit is used in DVD players. It houses
a semiconductor laser diode at 650 nm, photodiodes, and their preamplifiers. A small
hologram positioned directly in front of the laser unit performs the different func-
tions of beam splitting and deflection to generate the playback signals and the signals
for focusing and tracking error detection. The generation of the focal error signal is
based on the knife-edge method.
In order to enable this hologram laser unit to operate as a zero indicator, only
the focus error signal of the device is used. Additionally, the sum signal of the
focus error photodiodes is used for an automatic intensity scaling of the sensor
signal. Therefore, it is possible to measure on object surfaces with variable
reflectivity.
The basic structure of the entire laser focus sensor is shown in Fig. 9. Because it
works as a point sensor, it is necessary to trace the laser spot of the optical scanning
on the sample surface. Therefore, the probe has been combined with a CCD camera
microscope. This allows a very comfortable discovery and also the later retrieval of
interesting regions on the sample surface. On the other hand, the camera is used in
the modular setup as detection unit for white-light interference, as described later. In
order to use the full vertical measuring range of the NPM-machine, a long working
44 E. Manske
Fig. 9 Principle of laser

focus probe
distance lens is used. On this basis, for example, step heights up to the measurement
range of the NPM-machine of 5 mm can be measured with nanometer resolution
(Fig. 10).
This optical probe shows the typical properties of optical sensors. The optical
nontactile measurement prevents a mechanical elastic or even plastic deformation
of the surface of the specimen. This laser focus probe is a point sensor with a laser
spot size <1 μm. The DUT has to be scanned in a defined manner. The data
collection is sequential. On the other side, the noncontact method allows a high
scanning speed up to 1 mm/s (and more) with nanometer precision. A basic property
of optical probes is, for example, the appearance of diffraction effects at step flanks
or sharp edges. Therefore, if these effects influence the measurement significantly,
it is desirable to apply tactile stylus probes. Here, the tip diameters are in the order of
4 μm. Batwing effects do not exist but an adulteration of the measurement result
regarding the tip geometry is stated. Again, this influence can be reduced by using
sharp tips in the order of some nanometers with the aid of atomic force microscopy.
In consequence, there is no optimal nanoprobe for all characteristics of specimen
or measurement requirements in the nanometer range. To realize a variety of
measurement tasks, a combination of different probes is useful. Therefore, a multi-
probe concept is proposed (Fig. 11).
A probe changer in the form of a microscope turret can turn the desired probe in
front of the sensor head. The probes a and c work together with the hologram laser
Fig. 10 Step height (top) and roughness (bottom) measurement with laser focus probe and NPM-
machine
46 E. Manske
Fig. 11 Multi-probe concept (a-focus probe, b-stylus probe, c-AFM probe, d-white light interfer-
ence probe)
unit. The focus error signal is detected as zero signal. In the stylus probe (b) a small
stylus with a diamond tip is guided by a parallel membrane spring (Hofmann et al.
2015). The laser spot is focused on the rear side of the shaft. In the same way, the
laser beam is focused on the cantilever of an AFM probe. Here, a variety of tip
geometries is commercially available (from 500 nm up to 2 nm, with high aspect
ratio or with special tip design).
The white-light interference microscope probe (Fig. 11, probe d) works together
with the CCD camera of the sensor head. The parallel data acquisition (of each pixel)
is performed by a z-scan of the NPM-machine stage (Kapusi et al. 2008). If the
collection of all white-light interferograms (for all pixels) is complete, the surface
geometry of the full field can be derived with subnanometer precision. All these
probes are arranged in a microscope turret in the NPM-machine (cf. Fig. 12).
The probes shown above are so-called surface probes (sometimes called 2.5D
probes). They can only measure from top. They cannot measure sidewalls, cavities,
holes, or undercuts. Even the measurement of steep flanks is restricted to a certain extent.
Additionally, not yet integrated into the probe changer, tactile 3D micro probes
can be applied in the NPM-machine (cf. Fig. 13). Different 3D microprobes have
been implemented, tested, and compared (Hofmann et al. 2009). Most microprobes
produced use ruby balls with a diameter of 300 μm. However, also diameters of
120 μm or 75 μm are increasingly used. A big problem is the slightly high
uncertainty in the estimation of diameter and roundness deviation of these spheres.
The three-sphere calibration method first shown at METAS (Küng et al. 2007) at
Fig. 12 Multi-probe turret in

the NPM-machine
Fig. 13 3D microprobe
integrated into the NPM-
machine
1 mm-spheres with a repeatability of 5 nm can be considered as a standard method,

nevertheless at high expense. Up to now, it has not been trivial to achieve an absolute
full-sphere characterization with an uncertainty of less than 80 nm up to 40 nm as
presented at the Gannen XP (XPRESS Precision Engineering 2017).
Besides the uncertainty of the microsphere contact forces, isotropy of the forces in
all directions, stiffness probing mass and uncertainty, the measuring range, and the
linearity of the probe measuring system are important parameters. The possibility of
48 E. Manske
exchanging the stylus not only in the case of damage but also for calibration
purposes is advantageous (Küng et al. 2007; Balzer et al. 2011).
The depth-from-focus method also has been integrated in the NPM-machine as
measurement technique working in parallel (Machleidt et al. 2012). By means of the
CCD camera microscope, a z-image stack of the structure to be measured is
captured, with the positions of highest image definition being determined for each
single pixel. As opposed to white-light microscopy, the achievable z-resolution is
considerably lower. However, its advantage is that measurements can be made also
on steep surfaces. A basic prerequisite for this is that the measurement surface
presents sufficient roughness so as to allow enough scattered light to enter the
measurement lens.
Measurement Uncertainty
The excellent performance of the machine is based on four basic characteristics: the
Abbe-free arrangement, the application of the interferometric measurement princi-
ple, of an advanced closed-loop control of angular deviations, and of nanoprobes as
zero-point indicators (Jäger et al. 2009). To analyze the performance of the
NPM-machine, an extended uncertainty budget of the full metrological chain of
the machine has to be established. Because of the fact that it is a 3D measuring
system, a mathematical 3D model for the uncertainty budgeting is needed. Vectorial
modeling of the system parts is an effective approach, because the full metrological
chain can be assembled by a sum of vectorial submodels, which cover different
relevant influence quantities (Füßl et al. 2006; Füßl et al. 2011) even in the presence
of nonlinearities (Manske et al. 2015). So, the modeling can be traced to the 3D
vectorial analysis of two states – starting point A and end point B – of a measurement
process (cf. Fig. 14). In this case, the chain is divided into seven subvectors. Each of
them can be expressed by a mathematical context and contain several physical or
⇀
geometrical influence quantities. Again, subvectors can be subdivided (e.g., rFRa if
necessary. The vectorial sum of the measurement chain is zero. The vector regarding
measurement point B is derived in the same manner. The length of a measured length
⇀
rM can be determined by:
! ! ! !
rM ¼ r Bb r Aa r ab (5)
!
The vector r ab considers the possible shift of the distance vector itself between
time a and time b.
!
The combined uncertainty of the distance vector r M can be calculated according
to the GUM (Büchner and Jäger 2006):
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u
uX @f 2 X X @f @f
¼t
! ! 2 ! ! !
uc rM ! u r i þ 2 ! ! u r i , r j (6)
i @ri i j @ r i@ r j
Fig. 14 Vectorial model in starting point A

!
The combined uncertainty of the norm of the distance vector uc r M is in
accordance with the GUM provided by
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi

! ðxM uðxM ÞÞ2 þ ðyM uðyM ÞÞ2 þ ðzM uðzM ÞÞ2
uc r M ¼ (7)
x2M þ y2M þ z2M
in which the uncertainties of the measured

coordinates
! !
uðxM Þ,uðyM Þ,uðzM Þ are derived from the uncertainty vector u c r M (Eq. 6).
The error models of the different probe systems (shown in the previous chapter)
can be specified and are easily exchangeable, while the other subvectors remain
unchanged.
On this basis, error influences can be determined very effectively. So, conclusions
for the design of the machine or components as well as for operating the machine or
the measuring conditions can be derived (Schmidt et al. 2007). Under well-defined
measuring conditions in a thermally stabilized (<0.02 K) and acoustically insulated
enclosure with vibration insulation it is possible to achieve an expanded combined
3D uncertainty in the order of 5 nm of the full measuring range of 25 mm of the
NPM-machine.
Metrological Properties and Applications
To verify the metrological during the execution of consecutive 1 nm-steps in all three
axes of motion, a standard deviation of <0.3 nm on every step plateau was found
(Fig. 15).
50 E. Manske
Fig. 15 1 nm steps in all axes with the NPM-machine
Fig. 16 Comparison of step height measurements with different probes
Various step height normals have been used for comparing different tactile
sensors with each other. Figure 16 shows the results of the step height measurement
of a 70 nm step height normal for four different sensors in the NPM-machine. The
results obtained for all sensors lie nearly completely in the admissible uncertainty
interval of the calibration of 1.2 nm.
The reproducibility of lateral measurements can be checked, for example, with

pitch measurements of test grids. The measurement uncertainty for determining the
grid constant by measuring 200 consecutive grid periods of a 3 μm grid is better 30
pm (Dorozhovets et al. 2007). The reproducibility of the edge probing is <1 nm both
for the integrated AFM and for the laser focus sensor.
The measurement tasks to be solved are highly varied and place quite different
demands on the size of the measuring field, the lateral resolution, the point density,
whether an optical or tactile method should be applied, and many more. So
a combination of various sensor technologies seems to be desirable. With the aid
of the NPM-machine, measurements of most various kinds using different probe
systems can be carried out. So it is possible to make also several measurements with
various sensors on a DUT (device under test) within shortest time on the basis of the
sensor changer in the entire measuring range of 25 mm 25 mm (Fig. 17).
Here, it is extremely important to have knowledge of the different properties
of the sensors. By using the focus sensor, surfaces of 25 mm 25 mm can be
scanned within few minutes due to the high scanning speeds, but with larger line
distances (Fig. 18a). In the same way, it is possible to make scans at highest optical
resolution (<1 μm scan line distance) (Fig. 18b).
Furthermore, scans can be made in the same way with AFM sensors. As the
scanning speed is usually about 40 μm/s, the measurement time often turns out to
be very long. Here, the good long-term stability of the NPM-machine is very
Fig. 17 Grid measurements (a) optical, (b) tactile, and (c) AFM
52 E. Manske
Fig. 18 Scan with laser focus probe (a) lens 22 22 mm2 (b) lens with structure 400 400 μm2
advantageous. In the past, measurements were already made over up to 80 h without

any noteworthy drift effects. However, large amounts of data result which must
appropriately be processed and saved.
By employing CCD camera-based sensors, one million or even more measuring
data – depending on the number of pixels – can be recorded in parallel. Nevertheless,
Fig. 19 Zooming into specific regions of measured data
the surface which can be measured with a z-scan (e.g., 1 1 mm2) is very small
compared with the measuring range of the NPM-machine (25 25 mm2). Therefore,
a stitching method has been developed which allows very many single images to be
assembled with the high precision of the NPM-machine.
To prove this, a CCD camera with 782 582 pixels and a maximum image
frequency of 64 Hz was used to capture a 25 mm wafer1. Approximately 2.2 h were
required to fully capture the entire 25 mm 25 mm field, generating about 7828
individual images (CCD camera microscope: currently 0.5 million pixels, resolution:
0.417 μm) to be assembled automatically. This process creates a 3.6 GB file
containing 3.6 billion pixels (Fig. 19).
At present, there is no suitable commercially available measurement data pro-
cessing software which can handle those huge amounts of data. For this reason, a
suitable software has been developed for capturing, stitching, and visualizing (Birli
et al. 2013).
A special tile server for three dimensions has been developed and implemented in
a file format converter. The data of the entire measurement range are stored in cubes
with a defined edge length (known as tiles). Thus, it is well possible to zoom quickly
from a larger image field with lower resolution into a smaller image field with higher
resolution (Fig. 19).
When working with complex structures of several Gigabytes, it is
extremely difficult, for example, to manually find flaws in these images. For
this reason, methods for the automatic identification of areas of interest are
necessary. First, methods were investigated and implemented to identify those
areas with defects automatically and to add these areas of interest to the batch list
(Birli et al. 2013). Here, considering the enormous amounts of data, a lot of
challenges and opportunities for developing new complex measurement strate-
gies on the basis of the high-precision NPM-machine with its large measurement
volume are offered.
Most optical sensors allow measurements to be made only up to a limited slope of
the surface due to the limited aperture of the lenses. If scattered light is sufficient, the
depth-from-focus method enables also measurements on very steep edges. In this
1
Courtesy of X-Fab Semiconductor Foundries AG
54 E. Manske
way, it was possible to completely measure a microstructure standard presenting

very steep edges (Fig. 20) (Machleidt et al. 2012).
The measurement was carried out over a surface of 25 mm 360 μm and consists
of 90 stitched areas. For each measuring area, z-image stacks consisting of
600 images were generated. The measurement duration was around 2.5 h, with the
standard deviation achieved lying in the order of 50 nm.
With the NPM-machine, truly long-range and long-term AFM measurements are
possible (Vorbringer-Dorozhovets et al. 2016). Thus, measurements have already
been made on nanoimprint-structures over a measurement area of 1.5 μm 5 mm at
a line distance of 5 nm (Fig. 21). This corresponds to a measuring path of approx-
imately 1.5 m, with the measurement duration lying at around 28 h and a storage
capacity of 40 GB being required.
Fig. 20 Complete measurement of a micro contour standard with steep edges
Fig. 21 Three sections from the scanned structure (5 μm 1.5 μm)

Thus, the file contains 1000 of the sections shown in Fig. 21 including a total of
about 30,000 nanodots. The exact evaluation of the measurement data, however,
represents a big challenge.
Due to the tip wear, the achievable scanning length is restricted. Therefore, an
additional electromagnetic tip changer was developed. The tip changer comprises
three SPM probes in a special rig outside the measuring range (Vorbringer-
Dorozhovets et al. 2014). The AFM-cantilevers can now be purposefully ejected
electromagnetically by the AFM in the NPM-machine. On the other hand, a new
cantilever can be picked up directly by the AFM. In order to retrieve the previous tip
positions, additional fiducial marks have been developed. The repeatability of
relocation is less than 10 nm. The automatic tip changer and fiducial marks are
integrated into a sample holder. The tip changer in combination with fiducial marks
allows scanning distances three times longer (with the same type of SPM probes)
(Fig. 22).
The automatic tip changer is also advantageous for multifunctional nanoanalytics.
For example, for Kelvin probe force microscopy (KPFM), the measurement of
topography and surface potential with different SPM tips is necessary. Sample
KPFM measurements made on a special material standard (BAM-L200) are shown
(cf. Fig. 23) (Vorbringer-Dorozhovets et al. 2014).
Here, the real KPFM measurements (c, curve 3) have been deconvoluted (d, curve
2) regarding the topography data (b, curve 1). As a result, KPFM data with a
considerably higher lateral resolution are obtained. For this application, the use of
the tip changer can also be advantageous as the different cantilevers can be captured
consecutively.
Fig. 22 Metrological SPM and tip changer integrated into the NPM-machine (Vorbringer-
Dorozhovets et al. 2014)
56 E. Manske
Fig. 23 Topography and KPFM measurements at an BAM-L200 standard: (a) illustration of the
BAM-L200 standard, (b) topography measurement, (c) KPFM measurement, (d) deconvolution of the
KPFM data, (e) profile of the results ((1) topography, (2) deconvoluted KPFM data, (3) KPFM data)
Summary and Outlook
The continuing rapid progress in some key fabrication technologies has led
to ambitious challenges for precision measurement technology. Today’s nano-
positioning and nanomeasuring technology allows most versatile and complex
measurement and positioning of objects with subnanometer resolution across
multiple scales up to several centimeters. The traceable 3D metrology combined
with nanometer precision over a range of several tens of millimeters requires the
measurement and control in 6 DOF by applying the most advanced fiber-coupled
laser interferometers with a resolution of <20 pm and highly stabilized He-Ne lasers.
Several optical, tactile and AFM probes with reproducibilities less than 1 nm have
shown very good comparability when used for step height measurements. To solve
most measuring tasks, a multi-probe approach has been realized in the Nanopositioning
and Nanomeasuring Machine. Furthermore, also 3D-micro probes can be employed.
At present, there is an ever-stronger trend toward multisensor technology and
multiparameter characterization. At the same time, new requirements have to be met
by data fusion as well as by the processing and visualization of huge amounts of data
records. For increasing the potential offered by the NPM-technology, the positioning
range has already been extended to 200 mm 200 mm 25 mm (Jäger et al. 2016).
In the near future, the outstanding metrological properties of nanopositioning and
nanomeasuring machines will allow also a number of new alternative approaches of
micro- and nanofabrication down to the sub-10 nm range in vast working ranges.
References
Abbe E (1890) Meßapparate für Physiker. Z Instrumentenkd 10:446–448
Balzer FG, Hausotte T, Dorozhovets N, Manske E, Jäger G (2011) Tactile 3D microprobe system
with exchangeable styli. Meas Sci Technol 22(9): 094018 (7pp), IOP Publishing, Bristol.
ISSN 13616501
Birli O, Franke K-H, Linß G, Machleidt T, Manske E, Schale F, Schwannecke H-C, Sparrer E, Weiß
M (2013) Measurement, visualization and analysis of extremely large data sets with a nano-
positioning and nanomeasuring machine. In: Conference: optical measurement systems for
industrial inspection VIII. In: Proceedings of SPIE: Munich, Germany, 13–16 May 2013.vol
8788, pp 87880X (7 pp)
Büchner H-J, Jäger G (2006) A novel plane mirror interferometer without using corner cube
reflectors. Meas Sci Technol 17:746–752
Dorozhovets N, Hausotte T, Jäger G, Manske E (2007)Application of the metrological scanning
probe microscope for high-precision, long-range, traceable measurements. In: Proc. of the SPIE
6616 (2007): Munich, Germany, 661624 (7 pp)
Füßl R, Jäger G, Grünwald R, Schmidt I (2006) A new vectorial model for the estimation
of uncertainty in nano coordinate measuring machines. In: XVIII IMEKO world congress.
Metrology for a sustainable development, September 17–22, 2006, Rio de Janeiro, Brazil
Füßl R, Manske E, Kreutzer Ph. Modeling of 3D-measuremnt chains in nanopositioning and
nanomeasuring machines. In: Joint international IMEKO TC1+ TC7+ TC13 symposium,
August 31st–September 2nd, 2011, Jena. urn:nbn:de:gbv:ilm1-2011imeko-38.7
Hausotte T, Percle B, Gerhardt U, Dontsov D, Manske E, Jäger G (2012) Interference signal demod-
ulation for nanopositioning and nanomeasuring machines. Meas Sci Technol 23:074004. (8pp)
Hofmann N, Tibrewala A, Balzer FG, Hausotte T, Manske E, Jäger G, Büttgenbach S (2009)
Characterisation of the metrological properties of 3-D microprobes. In: Laser metrology and
machine performance IX/International conference and exhibition on laser metrology, machine
tool, CMM & robotic performance, 9 (Uxbridge). Euspen, Cranfield, pp 243–252
Hofmann N, Mastylo R, Manske E et al (2015) A compact tactile surface profiler for multi-sensor
systems in nano measuring machines. TM-Technisches Messen 82(7–8):377–383, Published:
Jul 2015
Jäger G, Manske E, Hausotte T, Büchner H-J (2009) The metrological basis and operation
of nanopositioning and nanomeasuring machine NMM-1. TM-Technisches Messen
5:227–234
Jäger G, Manske E, Hausotte T, Müller A, Balzer F (2016) Nanopositioning and nanomeasuring
machine NPMM-200 – a new powerful tool for large-range micro- and nanotechnology. Surf
Topogr: Metrol Prop 4:034004
Kapusi D, Machleidt T, Manske E, Franke K-H, Jahn R (2008) White light interferometry, a new
application for the nanopositioning and nanomeasuring machine. In: Proceedings of ICPM
Ilmenau, Verl. ISLE, pp 37–38. ISBN 978-3-938843-38-3
Küng A, Meli F, Thalmann R (2007) Ultraprecision micro-CMM using a low force 3D touch probe.
Meas Sci Technol 18:319–327. https://doi.org/10.1088/0957-0233/18/2/S01
Machleidt T, Sparrer E, Manske E, Kapusi D, Franke K-H (2012) Area-based optical 2.5D
sensors of a nanopositioning and nanomeasuring machine. Meas Sci Techno 23: 074010 (6 pp)
Manske E, Jäger G, Hausotte T, Füßl R (2012) Recent developments and challenges of nano-
positioning and nanomeasuring technology. Meas Sci Techno 23: 074001 (10pp)
Manske E, Fröhlich T, Füßl R (2015) Measurement uncertainty consideration in the case of
nonlinear models for precision length measurement. J Phys Conf Ser 588(1). https://doi.org/
10.1088/1742-6596/588/1/012033
Mastylo R, Dontsov D, Manske E, Jäger G (2005) A focus sensor for an application in
a nanopositioning and nanomeasuring machine. In: SPIE, München, vol 5856, pp 238–244
Schmidt I, Hausotte T, Gerhardt U, Manske E, Jäger G (2007) Investigations and calculations into
decreasing the uncertainty of a nanopositioning and nanomeasuring machine. Meas Sci Technol
18:482–486
58 E. Manske
SIOS Stabilized HeNe Lasers. http://www.sios-de.com/products/stabilized-hene-lasers

Vorbringer-Dorozhovets N, Manske E (2016) Scanning strategies for long term and long distance
AFM measurements. In: Proceedings of the 16th international conference of the European
Society for Precision Engineering and Nanotechnology: May 30th–3rd June 2016, Nottingham,
UK. Euspen, Bedford, P1.60, 2 pp
Vorbringer-Dorozhovets N, Goj B, Machleidt T, Franke K-H, Hoffmann M, Manske E (2014)
Multifunctional nanoanalytics and long-range scanning probe microscope using a nano-
positioning and nanomeasuring machine. Meas Sci Technol 25:044006 (7pp), https://doi.org/
10.1088/0957-0233/25/4/044006
XPRESS Precision Engineering: Gannen series – 3D tactile scanning probes with nanometer
uncertainty. http://www.xpresspe.com/download/Xpress-Gannen-series.pdf. 29 Nov 2017)
Yoshida Y, Miyake T, Kurata Y, Ishikawa T (1990) Three beam CD optical pickup using
a holographic optical element. In: Proceedings of SPIE, vol 1401. Optical data storage
technologies, Singapore, pp 58–65
Confocal Microscopy for Surface
Profilometry 3
Liang-Chia Chen
Contents
Basic Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
Virtual Confocal Microscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
Optical Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
Measuring Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 66
Measuring Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 68
Lateral Scanning Confocal Microscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
Optical Configuration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
Measurement Principle and Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71
Chromatic Confocal Microscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
Basic Optical Configuration and Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
Line-Scanned Chromatic Confocal Microscopy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 81
Analysis of Sensor Cross Talks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84
Measurement Examples and Analyses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85
Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
Abstract
The application of automated optical inspection (AOI) to advanced manufactur-
ing processes with tight takt time and specifications is critical in winning today’s
global competition. In the past decades, great effort had been devoted to devel-
oping novel solutions for in-line optical inspection of surfaces and the dynamic
characteristics of tested components or devices. Conventional approaches to
microscale 3D surface profilometry have adopted novel optics or concepts in
confocal microscopy for measuring 3D surface characteristics with high speed
and precision. One-shot measurement capability is demanded to minimize
measured uncertainty from environmental vibration or system instability.
L.-C. Chen (*)

e-mail: lchen@ntu.edu.tw

https://doi.org/10.1007/978-981-10-4938-5_3
60 L.-C. Chen
Nevertheless, extremely high-speed microscopic 3D confocal profilometric meth-

odologies for 100% full-field inspection are yet to be developed. This chapter
intends to provide the measurement principle and development of confocal
optical profilometry in overcoming bottlenecks and developing feasible solutions
for novel manufacturing technologies, such as roll-to-roll nanoimprinting or
semiconductor processes. It contains the basic optics, the measuring algorithms
and industrial applications in optical confocal microscopy required for surface
profilometry.
Keywords
Microscopy · Confocal microscopy · Optical metrology · Automated optical
inspection (AOI) · Microscopic confocal profilometry · 3D measurement ·
Optical profilometry · In situ inspection · Precision manufacturing · One-shot
imaging
Basic Theory
Confocal microscopy was first proposed by Minsky in 1961 for its main application
in biological investigation by using fluorescence and in-focus contrast (Minsky
1961). For industrial inspection, the main difference is the use of illumination light
source and emitted light reflection or backscattering for nonfluorophores and exci-
tation. In confocal microscopy, axially segmented images of the tested sample are
generally achieved by confining the illuminating area by the use of an illumination
aperture and detecting the reflected light using a pinhole, slit, or other structured
patterns for clearly distinguishing in-focus light from out-of-focus one. A confocal
microscope works by geometrically matching two conjugate focal points in image
space. As the sample is scanned through the focal point, peak intensity can be
detected only when the focal point lies directly on the surface of the sample.
Vertically scanning along the optical axis of the objective or tested sample is
essential to generate a series of optically sectioned image, in which a depth response
curve can be produced to detect the best focus position by the means of finding the
peak intensity and the surface height of the tested sample can be determined.
Scanning confocal microscopy is a well-known and useful measurement tech-
nique for surface topography measurement due to its superior lateral and vertical
resolution over conventional microscopy in in situ automatic optical inspection
(AOI) for microstructures (Chen et al. 2016). Confocal microscopy can achieve
high resolution, depth discrimination, and excellent sectioning capability in 3D
profile measurement (Chen et al. 2012).
In a point-scanned confocal microscope, illumination light source lightens the
tested object using an objective and the reflected or backscattered light is received
coaxially by an optical detector for sizing intensity of the reflected light. The
detection is achieved using a pinhole or slit in front of the detector. A basic
configuration of traditional confocal microscope is illustrated in Fig. 1. Only the
light that is reflected from the focal plane of the system can precisely pass through
3 Confocal Microscopy for Surface Profilometry 61
Photo detector
Pinhole
Illuminating aperture
Point light source

Beam splitter
Objective
Z-distance
Sample
Intensity
Fig. 1 Structure of traditional confocal microscope
the pinhole or slit and is detected by the detector. This optical conjugate configura-
tion generates a strong optical sectioning property for vertically scanning 3D profiles
of the tested object. A vertical scanning along the optical axis of the microscope can
be used to generate a depth-response curve (DRC), in which the depth associated
with the maximum light intensity of the curve can represent the depth of the scanned
surface. A complete volumetric structure or surface topography of the tested object
can be scanned by a spatially lateral scanning (X, Y direction) in cooperating with a
vertical scanning along the optical axis for each scanned object’s point. Traditional
confocal microscopes generally use a linear translation stage for lateral scanning and
piezoelectric transducer (PZT) for vertical scanning.
A schematic illustration of a basic optical confocal configuration is shown in Fig. 2.
As the configuration illustrates, a coherent or incoherent illumination light intensity, S
(v, w), is projected through an illumination aperture and a microscope’s objective
having a pupil function, P(ξ, η), and then focused onto a tested sample having a
62 L.-C. Chen
Detector D(v,w)
Imaging pinhole
Illumination aperture
Illumination light source

S(v,w)
Objective
P(x,h)
r
Sample to be tested
Fig. 2 Basic optical confocal microscopic configuration
reflectance distribution defined as r before it is reflected or backscattered to the

objective and then focused onto an imaging pinhole or a slit by following an optical
detector having sensitivity, defined as D(v, w) for photon detection using a photon
multiplexer transducer (PMT) or a charge couple device (CCD) (Artigas 2011).
When the incident light is projected through an illumination aperture with out-of-
focus by a quantity, defined as u along the optical axis, i(u, v, w) represents the
illumination light pattern being out-of-focus by a quantity, u along the optical axis on
the object plane, given by
O
iðu,v,wÞ ¼ S ðv,wÞ jhðu,v,wÞj2 (1)
where h(u, v, w) is the transfer function in representation of the Fourier transform of

the pupil distribution to be expressed as follows:
ð ð1
jk
hðx2 , y2 Þ ¼ Pðx1 , y1 Þe f fx1 x2 þy1 y2 g dx1 dy1 (2)
1
where x1 and y1 are the spatial coordinates on the first propagation plane; x2 and y2
are the spatial coordinates on the second propagation plane.
In general, the parameters x and y are expressed in the normalized optical
coordinates; the above transfer function can be rewritten in a more expressive format
as follows:
ð
1
hðu,v,wÞ ¼ Pðξ,ηÞeiðξvþμwÞ eiΔW ðu,v,wÞ dξdη (3)

1
where ξ and η are the pupil coordinates, which are perpendicular to the aperture
radius of the pupil; sinα is the objective’s numerical aperture; ΔW(u, v, w) is the
wavefront consisting of two terms in focusing and aberration; the parameters v and
w are defined as the normalized optical coordinates, given by
2π
v¼ x sin α
λ (4)
2π
w ¼ y sin α
λ
For a circular pupil, the pupil coordinates, ξ and η are defined as:
x
ξ¼ (5)
a
y
η¼ (6)
a
ξ ¼ η ¼ ρ sin θ (7)
Using Eq. (7), Eq. (3) can be further simplified as:
ð1
hðu,vÞ ¼ PðρÞJ 0 ðvρÞeiΔW ðu,ρÞ ρdρ (8)
0
where J0(x) is a first-order Bessel function of the first type.

The defocus term for the wavefront in the pupil region is expressed as
1
ΔW ðu,ρÞ ¼ uρ2 (9)
2
where
8π
u¼ z sin2 ðα=2Þ (10)
λ
Eq. (8) can then be simplified to
ð1
hðu,vÞ ¼ PðρÞJ 0 ðvρÞei2uρ ρdρ
1 2
(11)
0
64 L.-C. Chen
Thus, when i(u, v, w) is reflected or backscattered to the objective and focused

through an imaging pinhole then onto an optical detector, the detected light intensity
can be given as follows:
I ðu,v, vd Þ ¼
1 2 v 1 2
ð ðd ð

2 PðρÞexp 1 juρ2 J 0 ðρvÞρdρ 2 PðρÞexp 1 juρ2 J 0 ðρvÞρdρ vdv (12)

2 2
0 0 0
where vd is the normalized aperture size of the pinhole, vd ¼ 2πλ r sin α, in which r is
the radius of pinhole and sinα is the numerical aperture of the objective.
Equation (12) can be also simplified and expressed in a convolution form as
follows:
O O
I ðu,v,wÞ ¼ S ðv,wÞ jhðu,v,wÞj2 Dðv,wÞ jhð0,v,wÞj2 (13)
When vd = 0, the optical configuration can be called an ideal confocal microscope

with a pinhole having an infinite tiny point size. The light intensity being detected
can be simplified and expressed as
1 4
ð
1
I ðu,v,0Þ ¼ 2 PðρÞexp juρ J 0 ðρvÞρdρ
2
(14)
2
0
In contrast, when vd ! 1, the optical configuration can be called a conventional

optical microscope. Figure 3 illustrates the normalized light intensity responding to
various normalized pinhole sizes, in which the pinhole size significantly affects the
light intensity distribution.
Virtual Confocal Microscopy
Optical Configuration
The greatest difference of a virtual confocal microscope from a conventional one lies
in the fact that no pinhole or slit is being used at the image focus plane. The object
focus plane is searched by measuring the focus degree of a projected structured
pattern and is detected when the image focus reaches to its maximum. This makes
the technique being called virtual or pseudo confocal measurement. As the depth
information is measured by the focus degree in the image plane, people also call it
shape from focus (SFF).
The schematic diagram of a typical virtual confocal microscope is illustrated in
Fig. 4 (Chen et al. 2014). When the tested sample is with a kind of textureless
Intensity response for different sizes of pinhole

1
Vd=0
0.9 Vd=1
Vd=2
0.8
Vd=3
Vd=4
Normalized Intensity I(u)
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
-10 -8 -6 -4 -2 0 2 4 6 8 10
u
Fig. 3 Schematic diagram of the normalized light intensity responding to various normalized
pinhole sizes
surface, a light source with a designed wavelength with selected bandwidth is

employed as an active illumination and is projected through a structured pattern as
the illumination aperture. When the tested sample is with a kind of surface texture
possessing certain roughness, the active illumination can be replaced by using a
normal optical aperture when the light is incident into the microscope. A set of
focus lenses is usually used to collimate and focus the structured light into a
microscope objective. A vertical linear scanner using either a piezoelectric trans-
ducer (PZT) or precise translation stage is integrated with the objective for vertical
scanning. The reflected or backscattered light is collected by the objective and
passes through the beam splitter and then focuses on an area-scan image sensor at
its image plane.
Considering the reflectance characteristics of the tested sample, the design
of light transmittance and grating size of the pattern are generally optimized
to achieve best lateral measurement resolution. In general, the smaller the
structured pattern is, the higher measuring resolution can be achieved.
However, this will be restricted by the image detection limit. The structured
pattern can be made by chrome coating, holography, or other printing techniques.
A different size and pitch of structured pattern can be designed to achieve
required lateral resolution. For example, a glass grating with checkerboard
patterns having various pitch sizes can be employed in structured projection.
66 L.-C. Chen
IR CCD
Focus lens
Beam splitter
Fiber
IR light source
PZT
Optical grating
Objective
IC sample
Stage
Fig. 4 Schematic diagram of a typical virtual confocal microscope
When an optical aperture is used to measure a tested sample with surface texture,
the measuring spatial resolution can normally reach to the image detection limit.
In contrast, the active illumination method can only achieve the scanning up to
the pattern resolution. Its full-field image can be realized by shifting the illumi-
nation pattern for lateral scanning but this normally consumes more time in
inspection.
Measuring Principle
Figure 5 illustrates the brief operation principle of the virtual confocal microscopy.
Two conjugate points located in the object and image planes of the virtual confocal
microscope are matched by searching for its maximum degree of image focus. When
the object point scans at the focus plane, the image focus detected on its
corresponding conjugate point in the image plane can reach a maximum along its
scanning axis. Depth map can be reconstructed by defining a focus operator and
depth estimation algorithm.
The intensity of defocused image on the image plane, Id(x, y), is given by
convolution of the focused image, If(x, y), with the point spread function (PSF) of
the image system, h(x, y), as follows:
Fig. 5 Confocal vertical scanning principle
I d ðx,yÞ ¼ hðx,yÞ I f ðx,yÞ (15)

2 2
1 x þy2
hðx,yÞ ¼ e 2σ h
(16)
2πσ h 2
where σ h is the spread parameter, which is proportional to the radius of the circular
patch on the object surfaces.
To precisely measure the focus degree of a detecting image pixel, a high-pass
focus measure operator o(x, y) is used to evaluate the image consisting of the pixel
with its neighboring pixels as a subimage (Nayar and Nakagawa 1990). A sum-
modified-Laplacian (SOL) operator, used as the focus index, can quantify the degree
of the focus on the detecting image, given as

MLðx,yÞ ¼ 2I ðx,yÞ I x step, y
I xþ step, y (17)
þ 2I ðx,yÞ I x, y step I x, y þ step

X
iþN X
jþN
F ði,jÞ ¼ MLðx,yÞ for MLðx,yÞ T (18)
x¼iN y¼jN
68 L.-C. Chen
Fig. 6 Flow chart diagram of virtual confocal measuring method
where the parameter N is the window size and T is a threshold set to filter image
noise.
To make the measurement principle clear, a measurement flow chart, shown in
Fig. 6, is illustrated and described. A system calibration is performed to determine
best measuring parameters, such as the pitch and size of illumination pattern,
sensor exposure time, and vertical scanning pitch to satisfy required measurement
resolution and precision. During the vertical scanning operation, a series of
sectioned images are acquired by actuating the linear motion stage. The signal-
to-noise ratio of the scanned image is enhanced by adequate image preprocessing
algorithms. The focus index is determined using the SML operator for each tested
pixel. With focus index along the depth axis, a depth response curve (DRC) is
established for peak detection on the maximum focus index. An effective peak
search algorithm is needed to determine the depth of each measured point. A full-
field surface contour of the tested surface can be reconstructed for quality insur-
ance purposes.
Measuring Examples
A precalibrated standard gauge block with a step height of 32.54 μm was employed
to verify measurement accuracy of the developed system. A structured grating with
black and white checkboard pattern of a pitch size of 60*60 μm2 was used for active
light illumination. By performing the measurement procedure described above, the
3D profile of the sample was reconstructed. Figure 7 shows the reconstructed 3D
3D Profile
Z axis (μm)
40
20
0
600
500
400 500
300 400
200 300
X axis (μm)
200
100 100 Y axis (μm)
Fig. 7 Reconstructed 3D contour for a precalibrated precision gauge block
contour of the tested object. The repeatability of the measurement can reach less than
0.2 μm at a confidence level of three standard deviations.
Using the method for IC dicing inspection, Fig. 8 shows a typical example of
chipping defect which appeared on the IC dicing process. By using the developed
measurement method with IR light source of 1100 nm, the 3D profile of the IC chip
with a defect can be measured.
Lateral Scanning Confocal Microscopy
Optical Configuration
To expedite in situ three-dimensional shape measurement, lateral confocal scan-

ning (LCS) microscopy provides an effective method, since it can skip vertical
scanning operation normally required by traditional confocal microscopy in
contour scanning. Optically, the LCS is different from the traditional confocal
microscopy, in which one angle between the lateral scanning plane and the
focus plane is configured, so that the optical inspecting axis is tilting with the
sample, shown in Fig. 9. Fantastically, with such configuration, one detected
image can include all focus information acquired from different vertical scan-
ning depths. Essentially, the imaging sensor observes one inspecting point by
using a line array of optical pixels, so all the focus information along the tilting
axis, which locate in different depth sections, can be detected by the sensor
simultaneously. Consequently, when the microscope scans along the tested
sample laterally, all the focus information existing in three-dimensional space
can be detected and the vertical scanning operation can be totally skipped.
Obviously, the penalty of the LSC will be considerable increase of the sensing
elements, so the imaging detection rate can be less efficient. Moreover, the
70 L.-C. Chen
3D Profile
Z axis(μm)
20 600
0
-20 500
400
100 300
200
300 200
400 100 X axis(μm)
500
Y axis(μm)
Fig. 8 A typical example of IC chipping defects: (a) 2D image of chipping defect captured with IR
camera and (b) the 3D profile
lateral scanning is suggested to be performed on a precise scanning stage, so the

line scanning does not introduce substantial positioning errors into the measure-
ment result.
To inspect a sample with spectacular light property or textureless condition, the
LSC system generally employs an illumination structured pattern which goes
through an optical filter and an optical grating for generation of high-quality
structured grid patterns. To suit various kinds of the object’s surface characteristics,
the light intensity needs to be controlled to produce good image contrast of the
projected pattern. With this design, it is capable of profiling microstructures having
low reflectivity and high slope surfaces.
Illuminated in Fig. 10, for the purpose of the lateral scan measurement, the optical
objective is orientated two angles, α and β, with the scanning axis Y and the vertical
Z axis, respectively (Chen et al. 2010). The first angle α is designed to set the image
focus plane to be parallel with the tested microstructure surface, in which the
projected structure pattern is focused on the tested surface, shown in Fig. 10a. As
CCD sensor
Optical grating
Light source
Non-polarizing beam splitter
Green filter
Horizontal positional stage
Fig. 9 Schematic diagram of lateral confocal scanning microscopy
illustrated in Fig. 10b, β is designed to generate a focusing-and-defocusing process

along the Y-axis during the lateral scanning operation, in which the structured light
being projected on the tested surface has a focus variance trend along the scanning
direction. In the method, LCS is performed by synchronizing a linear moving stage
with the full-field image acquisition, in which the maximum resolution of the
scanning step is defined as the finest resolution of the stage. A depth response
curve (DRC) of each detected pixel can be obtained along its scanning direction. It
is clear to see that a modulated focusing function is obtained along the scanning
direction, in which a focus measure signal is peaked at the corresponding confocal
scanning depth of the detected surface point.
Measurement Principle and Algorithm
In the LSC, the optical system is inclined by an adequate angle, β, with the vertical
axis, thus orientating its focus plane to be tilted one angle with its lateral scanning
plane. Figure 11 shows the image formation geometry associated with the LCS being
developed here.
72 L.-C. Chen
CCD sensor
a
Optical grating
Light source
Green filter
Z
a
The direction of scanning X Z
Focus plane X
Optical grating
Light source
Green filter
b
Y
The direction of scanning

Z
Fig. 10 Optical configuration of the developed LCS system: (a) tilting angle of α with respect to
the scanning direction, Y-axis; (b) yaw angle of β with respect to the vertical direction, Z-axis
The optical relationship between the object distance, o, the image distance, i, and
the effective focal length, f, are given by Gaussian lens law as follows:
Sensor plane
O
ƒ
Optical axis
P'
Object reference plane
Fig. 11 Image formation geometry of the LCS optical configuration
1 1 1
þ ¼ (19)
o i f
When the sample surface and the focus plane are not parallel, only some portion
of the tested surface is located within the focus range of the image plane and can be
captured by an imaging sensor. The focus range, Fr, is given by:
DOF
Fr ¼ (20)
tan β
In Fig. 12, Point o(xr, yr) is a cross position between the focus plane and the tested
surface. A maximum light intensity can be only obtained on this position while the
intensity of other positions is attenuated. Illustrated in Fig. 13, the light intensity of
the detected modulated DRC is similar to a sin-modulated Gaussian function along
the lateral scanning direction. Thus, the intensity, I(i, j), of the DRC can be modeled
as follows:
!
ð1M Þ2
2π
I ði,jÞ ¼ sin i a e 2ð1=βÞ
2
(21)
f
where

sin 2π
f i is the structured light distribution
f is the period of the structured light pattern
i is the pixel position of the image detection
a is the modulation amplitude
74 L.-C. Chen
Peak
Intensity
Focus position
Cross profile
Position Y
Cross profile
X 2D image
X (640, 480)
O (xr,yr)
Light source
CCD sensor
Optical grating
Z
Objective
X b
Focus plane
d
X
Lateral scanning moving stage
Fig. 12 Relationship between the focus range and the DOF in the developed optical configuration
T chart
40
20
Intensity
-20
-40
0 19 46 73 106 145 184 223 261 300 339 378 417 455 494 533 572 610
Picture number
Fig. 13 Light intensity of the detected DRC
M is the peak position of the Gaussian curve

β is the yaw angle with respect to Z-axis
To determine the peak accurately, a peal detection algorithm of a DRC is

described. A block diagram of the algorithm is shown in Fig. 14. Low
frequency components, such as the DC term of the DRC are first removed
200
Intensity
150
To acquire DRC
100
019 4674 107 146 185224 263302 341 380 419 458 497536 575614
Image frame index
40
20
Intensity
0
To perform DC terms
-20
subtraction
-40
019 4673 106 145 184223 261300 339 378 417 455 494533 572610
Image frame index
2000
Intensity
To perform square-law 1,500

rectifier and low pass 1,000
filter 500
0
020 4876 110 150 190230 270310 351 391 431 471 511551 591631
Image frame index
2000 Peak position

Intensity
To determine the peak 1,500

using gravity center 1,000
detection 500
0
020 4876 110 150 190230 270310 351 391 431 471 511551 591631
Image frame index
Fig. 14 Peak detection algorithm for DRC
from the DRC by taking the first derivative operation, and the signal is then
rectified by square-law detection and low-pass filtering. As a result, the peak of
the low-pass filter output is located and the vertical position that corresponds to
the peak is detected.
Meanwhile, in the surface measurement process, the horizontal interval (Sp) of
scanning is critical to the detection resolution. It depends on the angle β and the
scanning resolution (k) and can be defined as follows:
S p ¼ k= cos β (22)
Space resolution k is generally determined by the CCD pixel size. Figure 15

shows that Sp varies against β ranging from 0 to 60 . The system’s depth resolution
Dr can be controlled by adequately controlling α and β. The depth resolution Dr can
be described as follows:
Dr ¼ k ð sin α þ cos α sin βÞ (23)

76 L.-C. Chen
Fig. 15 Relationship 0.4

between the tilted angle β and
scan pitch Sp
0.35
Scan pitch (μm)

0.3
0.25
0.2
0 10 20 30 40 50 60
Tilt angle(degree)
Fig. 16 LCS process of a calibrated standard step height
Measurement Examples
A measurement on a precalibrated step height surface is performed for measurement
accuracy evaluation. Its lateral scanning process is shown in Fig. 16, in which its step
height of 5.14 μm is evaluated. By performing the measurement procedure as
described, Fig. 17a, b demonstrates the 3D map and cross-section profile being
reconstructed from the step-height surface, respectively. From the analysis of the
measured result, the average height of the step height was 5.14 μm and the standard
deviation was 0.033 μm. Table 1 illustrates the measurement accuracy and repeat-
ability by a 30-time repeatability test.
3D Profile
Z axis(μm)
15
10
80
600
X 60 500
ax 400
is( 40
μm 300 )
) 20 200
xis (μm
100
Ya
profile
20
18
16
14
Z (μm)
12
10
4
0 100 200 300 400 500 600 700
X (pixel)
Fig. 17 Measurement results of calibrated step height: (a) 3D map and (b) cross-section profile
Meanwhile, an industrial sample having 30 V-groove microstructures, which

was made by roll-to-roll nanoimprinting process, was measured by the LSC method.
As shown in Fig. 18, to measure both the slit sides of the V-groove simultaneously,
the measuring optical design employs two microscopic probes to project structured
light patterns and two imaging sensors to acquire reflected light from the detected
78 L.-C. Chen
Table 1 Accuracy and repeatability obtained from 30-time repeatability test of the calibrated step
height
Measurement results
Number of scanned images Standard height Measured result Standard deviation
1300 5.14 μm 5.13 μm 0.033 μm
CCD sensor
CCD Sensor

Optical grating
Optical grating
Green filter
Green filter
Light source
Z
Light source
Y
X
The direction of scanning
Fig. 18 Schematic diagram of optical configuration of LSC system developed for measuring
microscopic V grooves
object’s surface from two adequate tilting angles, in which the probe is aligned with
the detected surface to obtain maximum reflected light (Fig. 19). An optical objective
with magnification of x50, NA of 0.95, and an illumination grating pattern of 50 μm
were used for the task. By using the LSC method, the 3D map was reconstructed and
the cross section was evaluated by comparison with the reference values, in terms of
the height, angle, and width of the V-groove, shown in Fig. 20. From a 30-time
repeatability test, Table 2 illustrates the measurement repeatability being obtained
from a 30-time repeatability test
Chromatic Confocal Microscopy
Basic Optical Configuration and Theory
Traditional confocal microscopy is by means of vertical scanning to establish a depth

response curve of the light intensity, which primarily depends on the light wave-
length and the numerical aperture of the objective. To avoid vertical scanning and
increase measurement efficiency, a spectrum scanning confocal measurement
Fig. 19 LCS process of standard angle target: (a), (b), and (c) are obtained by left side CCD; (d),
(e), and (f) are obtained by right side CCD
a b
8
6 X:24.97 X:74.39
3D profile 4 Y:16.06 Y:16.6
2 Width angle
0
20
8
Z axis(μm)
15
6 X:50.1
10
4 Y:2.642
5 20
40 2
0 Height
60 0
120 100
80 80
60 2
40 20 100 0 10 20 30 40 50 60 70 80 90 100
Y axis(μm) X axis(μm) μm
Fig. 20 Measurement result of the reference target: (a) top view; (b) 3-D map; and (c) cross profile
method employs a chromatic objective to disperse the light vertically according to

light wavelength and focus on various depth positions. This kind of microscope
generally equips with a predefined broadband light source in combination with a
diffractive lens as well as wavelength-to-depth coding for instantaneous point-
scanned or line-scanned profile measurement of a three-dimensional surface. This
technology takes the advantage of chromatic light dispersion to completely skip
vertical scanning operation, which is required for establishment of the depth
response curve. Figure 21 illustrates the schematic diagram of a chromatic confocal
microscope utilizing a chromatic objective for multi-wavelength light dispersion and
a pinhole to precisely filter a light wavelength corresponding to the object’s test
height (Chen et al. 2011a, b).
80 L.-C. Chen
Table 2 Measurement repeatability being obtained from a 30-time repeatability test of the 30 V-
groove microstructures
Measurement results
Average value Standard deviation
Height 14.65 μm 0.23 μm
Width 48.79 μm 0.54 μm
Angle 119.55 0.78
Fig. 21 Schematic diagram

of chromatic confocal
microscopy
In Fig. 21, when one chromatic objective is used as both the illuminating and
imaging lens, the light intensity function of acquired signals is given as:
1 4
ð
ð juρ =2Þ

I ðu,vÞ ¼ 2 PðρÞe
2
J 0 ðρvÞρdρ (24)

0
where P(ρ) denotes the pupil function of objective; J0 is the zero-order Bessel
function; u and v are the normalized optical radii.
2π 8π α
0
v r sin α0 u z sin2 (25)
λ λ 2
Moreover, when using a finite pinhole with a fixed size, the axial response
function of the system is given
Fig. 22 Axial light intensity Depth response curve

distribution for various 1
pinhole sizes vD=3
0.8 vD=6
Normalized Intensity
0.6
0.4
0.2
0
–15 –10 –5 0 5 10 15
u(displacement)
1 4
vðD ð

I ðu,vÞ ¼ 2 PðρÞeðjuρ =2Þ J 0 ðρvÞρdρ vdv
2
(26)

0 0
where vD is the normalized aperture size of a pinhole.

Using Eq. (26), the axial light intensity for various pinhole sizes is illustrated as
Fig. 22.
Line-Scanned Chromatic Confocal Microscopy
Line-scanned chromatic confocal microscopy is generally more efficient as one

sectioned-line profile can be measured by one image exposure (or one shot imaging).
However, the vertical measuring resolution and precision is usually decreased due to
the signal cross talk between adjacent detection points. The diffraction coupling
problem between two incident light beams along the slit direction has been regarded
as one of significant problems in restraining slit-scan chromatic confocal measure-
ment from possessing high measurement accuracy. The signal cross talk can poten-
tially increase the full width at half-maximum (FWHM) of the depth response signal
and potentially decrease the depth detection resolution for more than 10%. To
address this, a common strategy is to increase the scanning pitch between two
neighboring points, so the cross-talk phenomenon is minimized. However, by
doing so, the lateral measuring resolution is decreased as the measuring points are
sparser. In general, an adequate selection is made to make the best trade-off between
measurement efficiency and precision according to various measurement require-
ments in AOI.
Shown in Fig. 23, a fiber-based line-scanned chromatic confocal microscope is
introduced. It uses a chromatic dispersive objective for vertical light dispersion in a
82 L.-C. Chen
Fig. 23 Schematic diagram Spectrometer

of line-scanned chromatic a
confocal microscope: (a) Optical fiber
optical configuration and (b) Conjugate point
prototype system
Slit
Optical fiber Beam splitter
Lens
Light
source Chromatic
objective
designed depth range. A pair of linear coherent image fiber arrays are employed and
integrated in a multi-wavelength line-slit optical configuration in consideration of
minimizing the signal cross-talk effect, illustrated in Fig. 23a. Using the chromatic
objective, the incident light is dispersed according to a predefined depth ranging
from a few micrometers to even several millimeters. Each fiber is arranged in a one-
to-one conjugate relationship between each incident light and its corresponding
detected object point; the cross-talk effect is minimized, in which unfocused light
spatially and other possible stray lights are filtered away from the corresponding
spectrometer.
A hardware setup of the microscope is shown in Fig. 23b. The developed

optical system setup employs a pair of linear coherent fiber arrays at the incident
and imaging conjugate focal plane positions, respectively. The light source used is
a broadband light source generated by either a xenon arc lamp, halogen light, or
light-emitting diodes (LED). The light first focuses and passes through a slit with
aid of a set of optical lenses. The axial dispersion of light is generated by a
chromatic confocal objective in which light having a specific wavelength can be
focused on a corresponding depth position at the tested sample’s surface. In the
above configuration, the dispersive objective is designed to modulate an incident
broadband light for generation of a predefined axial chromatic light dispersion
along the object-focusing zone. Each image fiber is spatially configured and
employed to match a detecting incident light beam from the light source with its
corresponding reflected light from an object underlying inspection, in which a
spatially conjugate one-to-one relation between each incident light and its
matching detected object point can be formed accurately. Meanwhile, another
conjugate one-to-one optical relation between each detected object point and its
matching image pixel is also established through the above image-fiber mapping.
Each fiber is designed to spatially remove the unfocused light and other possible
stray lights, thereby minimizing the potential lateral cross talks between the
detected image sensors. Then, the one-to-one conjugate relationship between the
incident fiber point and its corresponding detected fiber point can be established.
Meanwhile, the chromatic optical objective with various optical magnification can
be used to disperse the incident white light with a spectrum range onto a series of
focal distances with respect to the corresponding wavelength. Figure 24a illus-
trates an example of vertical light dispersion between 400 and 700 nm by a
chromatic optical objective, while its chromatic spectral light distribution and
calibrated mapping curve between a light wavelength range of 300 nm and a
detected depth of 90 μm are shown in Fig. 24b, c, respectively. The linearity of
the curve represents the degree of a uniform vertical resolution of the depth
measurement. More linear the curve is, the uniform the vertical resolution is.
The design of the chromatic objective can significantly influence the uniformness
of the vertical resolution along the measurement depth range.
To obtain a reasonably accurate model of the above calibration-mapping curve, a
least-squares second-order polynomial approximation can be used to perform the
best fit and the model is given as:
zðλÞ ¼ aλ2 þ bλ þ c (27)
A longitudinal chromatic aberration relationship between wavelength and depth

can be further simplified and expressed as follows:

2
sin ðu=2Þ
I ðλÞ ¼ C (28)
ðu=2Þ

where u ¼ 2πλ NA2 aλ2 þ bλ þ c .
84 L.-C. Chen
Fig. 24 Dispersion of a
chromatic light: (a)
illustration of axial chromatic
light dispersion generated
by the chromatic objective;
(b) chromatic spectral B G R
light distribution; and

(c) calibration-mapping
curve between light
wavelength and measuring
depth
b
l
λ
c 620
Focal_point
Least-squares Fit
600
580
Wavelength(nm)
560
540
520
500
480
460
0 10 20 30 40 50 60 70 80 90
Depth position (mm)
Analysis of Sensor Cross Talks
In general, the traditional confocal system usually employs a micro pinhole or slit to
minimize the potential light cross talk between the neighboring detecting sensors.
The cross-talk problem is mainly generated by the general optical point spread
phenomenon, in which it potentially reduces the image quality. The FWHM of the
depth response curve for three different measurement methods, in terms of point, slit,
and area types, are shown in Fig. 25. It is clear to see that the FWHM is significantly
increased when the pinhole or slit is removed from its conjugate position, in which a
detecting pixel can cross talk (or convolute in Mathematics) with many of its
a c
CCD pixel
Point
Slit
Area
Reflection light focus point
Fig. 25 Schematic diagram of cross-talk phenomenon: (a) two reflection light beam focusing on
two neighboring pixels; (b) light cross-talk phenomenon; and (c) the depth response curve for three
different confocal measurement methods, in terms of pinhole, slit, and area types
neighboring pixels. Without employment of a pinhole or slit, the resultant light

intensity of a single detecting sensor from its cross talk neighboring region can be
expressed by the following equation:
1 X
X 1
I ði,jÞ ¼ 2 C iþp,jþq I i,j ði þ p, j þ qÞ (29)
p¼0 q¼0
where Ci + p, j + q is the surface reflection coefficient at Pixel (i + p, j + q).

To minimize the above cross-talk effect, a set of spatially matching image fiber
pairs can be used to establish one-to-one conjugate relationship between the incident
fiber point and its corresponding detected pixel. Figure 26 elucidates an optical
conjugate design using a fiber core diameter and a fiber pitch for accurate
corresponding matching between the object and the image planes. The fiber is
designed to spatially filter unfocused light and other possible stray lights from the
corresponding detection sensors, in which an adequate pitch of the fiber array is
designed in consideration of the point spread function (PSF) of the optical focusing
lens used in the optical fiber array.
Measurement Examples and Analyses
The flow chart diagram of the chromatic confocal measuring method is illustrated in
Fig. 27. A system calibration procedure including light intensity optimization,
86 L.-C. Chen
Receive optical fiber
Projection optical fiber
Beam splitter
Fiber core diameter f
Reflection light focus point
Fig. 26 Schematic diagram of the spatially matching image fiber pairs: (a) optical configuration of
fiber pairs and (b) optimal design of the fiber core diameter and fiber pitch
positioning verification of the translation stage, and optimal control of light exposure
time for best image contrast is performed to ensure an optimal spectrum imaging
condition for chromatic scanning. Following this, implementation of a vertical
scanning calibration using a precalibrated target surface and precise linear moving
stage is performed to establish accurate mapping between profile depth and light
wavelength. With the calibrated function, the tested sample can be scanned to
produce line-sectioned profiles and further reconstruct 3D map by performing lateral
scanning.
Fig. 27 Flow chart of the chromatic confocal measuring method
To evaluate the effect of the fiber size on the FWHM of the depth response curve,
a standard calibrated flat mirror was mounted on a precalibrated PZT linear stage for
performing vertical scanning. In the test, four setup using a continuous slit with an
opening width of 60 μm, a single fiber with a diameter of 62.5 μm for simulating a
pinhole, a linear fiber array with a diameter 9 μm having a pitch of 125 μm and a
linear fiber array with a diameter of 62.5 μm, and a pitch of 125 μm were employed
to perform a 30-time repeatability test on the same depth position of the tested
target. The experimental results are shown in Fig. 28, in which the fiber array with
a diameter of 9 μm and a pitch of 125 μm has a comparable result in comparison
with the single pinhole with a diameter of 62.5 μm. It is also clear to see that the
linear fiber array with a diameter 62.5 μm and a pitch of 125 μm has a narrower
spectral response curve (SRC) than the continuous slit with an opening width of
60 μm. Figure 29 illustrates the 30-time repeatability measurement results on the
FWHM of the SRC obtained from the four experimental setups. It is clear to
observe that the linear fiber array employed in the tested system has significantly
reduced the cross-talk effects. As seen, the reliability of the measurement has been
significantly improved when the result using the linear fiber array is compared with
the continuous slit having a similar opening width. By detecting the peak of the
SRC, the reliability of the depth measurement on the same calibrated target surface
can be shown in Table 3. The listed results have reasonably confirmed with the
above outcome. As seen, it is observed that the result of the slit with a 60 μm
opening width has much higher standard deviation on FWHM than the other three
tests. This clearly indicates that the measuring performance of this setup is not as
precise as the others.
88 L.-C. Chen
Fig. 28 Spectral response curve (SRC) obtained by using various confocal filters: (a) continuous
slit having an opening width of 60 μm; (b) a linear fiber array with a diameter 9 μm and a pitch of
125 μm; (c) a linear fiber array with a diameter 62.5 μm and a pitch of 125 μm; and (d) a single fiber
with a diameter of 62.5 μm
40
35
FWHM (nm)
Pinhole(60mm)
30
Fiber(9mm)
25
Fiber(62.5mm)
20 Slit(60mm)
1 3 5 7 9 11 13 15 17 19 21 23 25 27 29
Measurement sequence
Fig. 29 FWHM of the spectral response curve (SRC) using various confocal filters in a 30-time
repeatability test
Table 3 30-time repeatability test results of SRC using various confocal filters
Slit Fiber Fiber Pinhole
(60 μm) (62.5 μm) (9 μm) (60 μm)
Standard deviation of peak position (nm) 0.8045 0.0595 0.0305 0.0144
Mean FWHM (nm) 30.200 31.853 25.570 21.140
Standard deviation of FWHM (nm) 3.214 0.836 0.693 0.667
Meanwhile, to verify the measurement accuracy of the chromatic confocal mea-

suring method, a 25.4 μm standard step-height target was employed to perform step-
height profile measurement. Figure 30a exhibits the step measurement result using
the continuous slit having an opening width of 60 μm, in which the average height
and standard deviation were 24.98 μm and 0.066 μm, respectively, in a 30-time
repeatability test. By using the linear coherent fiber array with a diameter 62.5 μm
and a pitch of 125 μm, the standard deviation was slightly decreased to 0.061 μm, as
seen in Fig. 30b. Significantly noticed in Fig. 30c, the standard deviation was
considerably decreased to 0.009 μm when employing the linear fiber array with a
diameter 9.0 μm and a pitch of 125 μm. Moreover, it is also importantly noted that
the measurement of this kind of linear fiber array has been very close to the one to be
achieved by the single pinhole. Table 4 lists the measurement results achieved by the
four tested different filters. From the experimental results, it confirms that the method
using linear fiber arrays is capable of minimizing the cross-talk errors and enhancing
the measurement accuracy for chromatic confocal measurement by more than
16-folds. The measurement result by using the developed system was comparably
close to the one achieved by the single-pinhole confocal measurement method, while
the measurement efficiency can be enhanced due to its multiple-point measuring
configuration. Unquestionably, the lateral measuring resolution of the fiber array is
inferior than the one achieved by the slit type. This is always a trade-off decision to
be made by the technology users in consideration of various in situ measurement
requirements.
An industrial sample of a printed circuit board (PCB) with micro round bumps
having a nominal height of 25 μm was tested to measure profile of the bumps. A line-
scan chromatic measuring method inspection was performed to measure the bumps.
Figure 31a, b present 3D reconstructed map and cross-section contours of the
measured bumps, correspondingly. The measurements result indicate that the
method is capable to measure industrial microstructures with highly varying surface
reflectance on the tested sample. The measuring repeatability from a 30-times run
was verified to be within a submicrometer scale.
Conclusions
A confocal microscopic profilometer is extremely important and useful to achieve in

situ surface profilometry and automated optical inspection (AOI) of microstructures
in modern advanced manufacturing. Based on confocal microscopy, various kinds of
90 L.-C. Chen
Fig. 30 Measurement repeatability results on a 25.4 μm standard step-height target: (a) continuous
slit having an opening width of 60 μm; (b) a linear fiber array with a diameter 62.5 μm and a pitch of
125 μm; (c) a linear fiber array with a diameter 9.0 μm and a pitch of 125 μm; and (d) a single fiber
with a diameter of 62.5 μm
Table 4 30-time repeatability test results on measuring a 25.4 μm step-height target using various
confocal filters
60 μm slit 62.5 μm fiber 9 μm fiber 60 μm pinhole
Average height (μm) 24.98 25.56 25.41 25.406
Standard deviation (μm) 0.066 0.061 0.009 0.0076
3D Profile
Z axis(µm)
150
60
40
20 100
0 Y axis(µm)
50 50
100
150
X axis(µm) 200
250 0
300
Cross-section profile
60
50
X:71.12 X:103.9
Z axis(µm)
Y:49.7 Y:50.9
40 X:135.9
Y:36.71
X:12.6
Y:25.97 X:163.5
30 Y:26.19
20
0 20 40 60 80 100 120 140 160 180
Y axis(µm)
Fig. 31 Measurements of a micro round bump in a printed circuit board (PCB): (a) reconstructed
3D map and (b) cross-sectional profile
92 L.-C. Chen
optical configuration and techniques have been quickly developed to satisfy

demanding measurement requirements in tight tolerance and high precision, as
well as efficiency. The developed measurement method and system can be widely
employed to in situ microstructure profile measurement. It is worth noting that, using
the confocal method, the detection resolution of surface profilometry can reach
below 0.03% of the overall detection range. Nonetheless, there is still huge potential
in the field to be further developed to satisfy demanding needs in the fast developing
world. Novel optics and modern computer technologies such as artificial technology
can be utilized to break through the current measuring limits, especially in greatly
shortening of takt time consumed by the confocal method. The developed system
can be further developed with advancement of the hardware performance such as the
imaging sensor for its optimal performance in in situ AOI.
References
Artigas R (2011) Imaging confocal microscopy. In: Leach R (ed) Optical measurement of surface
topography. Springer, Heidelberg, pp 237–286
Chen LC, Chang YW, Chen SH, Li ZK, Kuo SH, Lai HW (2010) In-situ microscopic surface
profilometry by lateral confocal scanning (LCS) for roll-to-roll thin-film fabrication.
Paper presented at the 10th international symposium on measurement and quality, Osaka, 5–9
September 2010
Chen LC, Chang YW, Wu YL (2011a) In-situ chromatic confocal surface profilometry employing
image fiber correspondence for resolving lateral cross talk problems. Paper presented at the 10th
international symposium on measurement technology and intelligent instruments, Korea, 2011
Chen LC, Li HW, Chang YW (2011b) Full-field chromatic confocal surface profilometry
employing DMD correspondence for minimizing lateral cross talks. Paper presented at the
10th international symposium on measurement technology and intelligent instruments,
Korea, 2011
Chen LC, Chang YW, Li HW (2012) Full-field chromatic confocal surface profilometry employing
DMD correspondence for minimizing lateral cross talks. Opt Eng 51(8). https://doi.org/10.1117/
1.OE.51.8.081507
Chen LC, Le MT, Phuc DC, Lin ST (2014) In-situ volumetric topography of IC chips for defect
detection using infrared confocal measurement with active structured light. Meas Sci Technol
25(9):094013
Chen LC, Nguyen DT, Chang YW (2016) Precise optical surface profilometry using innovative
chromatic differential confocal microscopy. Opt Lett 41(24):5660–5663
Minsky M (1961) Microscopy apparatus. US patent 3,013,467, 19 Dec 1961
Nayar SK, Nakagawa Y (1990) Shape from focus: an effective approach for rough surfaces.
In: International conference on robotics and automation, vol 1, pp 218–225, May 1990
Optical Sensors for Machine Tool Metrology
4
Zhi-Feng Lou and Kuang-Chao Fan
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
6-DOF Geometric Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6-DOF Geometric Errors in Linear Stage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
6-DOF Geometric Errors in Rotary Stage . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 97
Machine Tool Accuracy Caused by Abbe Principle and Bryan Principle . . . . . . . . . . . . . . . . . . . . . 97
Low-Cost Optical Sensors for Geometric Error Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
Straightness Error Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
Pitch and Yaw Angle Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
Roll Angle Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
5-DOF Measuring System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 107
5-DOF Sensors for Abbe Error Compensation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 109
Parallelism Error Measurement of a Pair of Guideways . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 110
Accuracy Improvement of Linear Stages in Assembly Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 120
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
Abstract
The development of smart machine tools will be the trend toward the worldwide
need of intelligent manufacturing technology nowadays, which is the goal of
industry 4.0 as well as cyber-physical system and China made 2025. Current
Z.-F. Lou
Key Laboratory for Precision and Non-traditional Machining of Ministry of Education, School of
Mechanical Engineering, Dalian University of Technology, Dalian, China
K.-C. Fan (*)
Key Laboratory for Precision and Non-traditional Machining of Ministry of Education, School of
Mechanical Engineering, Dalian University of Technology, Dalian, China
e-mail: fan@dlut.edu.cn

https://doi.org/10.1007/978-981-10-4938-5_4
94 Z.-F. Lou and K.-C. Fan
techniques for machine tool metrology are implemented by measuring instru-

ments mainly suggested by the ISO 230 series of international standard. Although
these instruments are used to measure geometric errors of each axis of machine
tool, they cannot be used as sensors for real-time detection since they are
expensive and are in bulky sizes. In order to understand the functions of the
optical sensors for machine tool metrology, this chapter firstly addresses the
importance of geometric errors to the accuracy of machine tools. Abbe principle
and Bryan principle are two important guidelines in machine design and mea-
surement technology. The volumetric errors of machine tools are largely affected
by these two principles. This is the second topic highlighted in this chapter. Some
small and low-cost optical sensors for geometric error measurements are pre-
sented. These sensors can be portably mounted onto the machine frame for online
measurement. They can also be embedded in the machine structure as feedback
sensors for respective geometric errors. The integration of individual sensors into
a multi-degree-of-freedom measuring (MDFM) module for volumetric error
measurement and compensation is also introduced. Some applications of devel-
oped optical sensors to measure machine tool errors are described in the last part
of this chapter. Experimental validation shows the feasibility of developed optical
sensors for machine tool metrology and error compensation. This chapter neither
includes the measuring instruments for ISO 230 series as these are the subject of
other chapter nor those MDFM systems reported by many other researchers
because those are mainly at the laboratory level and not ready for practice on
machine tools.
Keywords
Geometric errors · Volumetric errors · Optical sensors · Machine tools · Abbe
principle · Bryan principle · 6-DOF · MDFM · Assembly process
Introduction
Precision stages are widely used in machine tools, photolithography steppers, and
measuring instrument. A normal serial type three-axis NC machine tool is
constructed by three linear stages to provide motions in X, Y, and Z axes, respec-
tively. For a four-axis or five-axis machine tool, it is to add one or two rotary stages
on top of the particular linear axis. However, it is difficult to achieve precise
positioning due to inherent geometric errors in each axis. As described in the
international standard (ISO230-1 2012), each linear or rotary stage has inevitable
six-degree-of-freedom (6-DOF) geometric errors, including three linear (or called
translational) errors and three angular (or called rotational) errors, caused by
manufacturing errors of the parts and assembly errors of the stages. Geometric errors
are the main sources that cause volumetric errors of the machine tool through
kinematic chains (Okafor and Ertekin 2000; Tlusty 1980), Abbe error (Abbe
1890), and Bryan error (Bryan 1979). Precision measurements of geometric errors
of each axis are the fundamentals of machine tool metrology (Gao et al. 2015).
4 Optical Sensors for Machine Tool Metrology 95
A complete list of measuring instruments of testing the accuracy of machine

tools, operating either under no-load or quasi-static conditions, has been documented
in the standard (ISO230-11.3 2011). Detailed setup and testing methods are specified
in various parts of ISO230 series (ISO230-2 2014; ISO230-4 2015; ISO230-6 2012;
ISO230-7 2015). Measuring equipment suggested by the ISO 230 standards includes
mechanical artifacts, displacement probes, digital scales, angle reference, and optical
instruments. Details can be found in relevant literature or the other chapter of this
book. In this chapter, the focus of presentation will be outside the contents of ISO
230 series. Some simple and low-cost optical sensors for measuring geometric errors
of machine tools will be introduced. Applications of these sensors to the assembly
process in order to reduce geometric errors will be presented. The methods of
converting measured error to another point in the working space will be described.
This is the basis of interpretation of volumetric error of machine tools based on Abbe
principle and Bryan principle, rather than the conventional Homogeneous Transfor-
mation Matrix (HTM) methods (Soons et al. 1992; Ferreira and Liu 1993).
The optical sensors defined in this chapter can be used not only to measure those
geometric errors of machine tools as other measuring instrument but also embedded
sensors to feedback geometric errors during motion. Some examples will be intro-
duced in the last part of this chapter. These include the integration of individual
sensor into a multi-degree-of-freedom measurement module, positioning error com-
pensation based on Abbe principle, parallelism error measurement of two guide-
ways, and accuracy improvement during machine assembly.
6-DOF Geometric Errors
6-DOF Geometric Errors in Linear Stage
Normal 6-DOF geometric errors of a moving stage consist of the positioning error
along the moving axis, two straightness errors in horizontal and vertical directions
perpendicular to the moving axis, and three angular errors called pitch, yaw, and roll
along three perpendicular axes, respectively, as shown in Fig. 1. These 6-DOF errors
can be categorized into two types. One type is the error in linear or translational
motion, including positioning and two straightness errors. The other type is the error
in angular or rotational motion, including pitch, yaw, and roll errors.
From the viewpoint of kinematic error chain of a linear stage, there are three
levels: (1) geometric errors, (2) joint kinematic errors, and (3) volumetric errors, as
shown in Fig. 2. The upper level error causes originate from the lower level errors
(Ekinci and Mayer 2007). Geometric errors are classified under two different sub-
categories. The first one is constituted of Guideway geometric errors which are the
surface straightness errors of the guideways upon which a machine axis carriage
moves as well as the relative location errors of the guideways of a particular axis.
The other subcategory is made of the link geometric errors including general shape
and assembly errors of the machine structures which determine the relative location
of the end functional point, i.e., the volumetric error indicated in Fig. 2. Therefore,
Yaw
Horizontal straightness
Roll
Pitch
Moving direction
Vertical straightness
Fig. 1 6-DOF geometric errors of a linear moving stage
Volumetric error
Linkage geometric error
Guideway geometric error Kinematic error
Ideal guideway
Fig. 2 Machine errors of a linear stage
not only the basic geometric errors are necessary to be measured, the spatial
positioning and trajectory of the functional point are also important to be discovered.
Figure 2, however, only illustrates the two-dimensional error chain originated
from an improper guideway. In practice, all moving stages are guided by a pair of
guideways. It is apparent that not only the guideway straightness will create the
stage’s straightness error and pitch error but also the parallelism of two guideways
will cause the stage’s yaw error and roll error. Figure 3 clearly shows the yaw error of
the stage is induced by nonparallel error in horizontal plane of two supporting
guideways. Likewise, the vertical nonparallel error of the guideways will cause the
roll error of the stage, as shown in Fig. 4. It can be clearly seen in Fig. 2, the pitch
error is due to the straightness error of the guideway in vertical direction. Since the
straightness error and the corresponding angular error have the same trend, one
curiosity would arise: can the angular error convert to straightness error and vice
versa? It has to be noted that although they can be convertible in surface plate
measurement by Moody method (Drescher 2003), up to the present there is no
positive answer in machine tool straightness error measurement (Ekinci and Mayer
Fig. 3 Yaw error due to

horizontally nonparallel
guideways
Fig. 4 Roll error due to

vertically nonparallel
guideways
2007; Majda 2012). Consequently, in order to reduce all angular errors of the stage,
the straightness and the parallelism of the guideways must be properly adjusted to
the best condition.
6-DOF Geometric Errors in Rotary Stage
Similar to the existing 6-DOF errors in the linear stage, there are also 6-DOF
geometric errors in the rotary table, including three linear errors and three angular
errors, as shown in Fig. 5 (ISO230-1 2012; ISO230-7 2015; Marsh 2010). Linear
errors can be separated into an axial slip error (δZ) and two radial motion errors (δX
and δY), which are also called run-out error of the spindle. Angular errors can be
separated into two tilt errors (εX and εY) and one angular positioning error (εZ),
which is also called the indexing error. The cause of these errors can be due to
imperfect bearing that constrains the shaft, as shown in Fig. 6 (Li and Fan 2017).
Machine Tool Accuracy Caused by Abbe Principle and Bryan

Principle
Ernst Abbe proposed the benefit of “Arranging the displacement measuring scale to
be in line with the distance to be measured” (Abbe 1890). Jim Bryan in 1979
extended this concept from displacement measurement to the straightness
Fig. 5 6-DOF geometric ZR

errors in rotary axis Axis of rotation
Reference axis
θZ (including εZ)
δX
δZ
δY
εX εY
Rotational body
XR YR
Fig. 6 6-DOF geometric

errors of rotary stage occurs
from imperfect bearing
measurement and proposed the Bryan principle as “The straightness measuring

system should be in line with the functional point whose straightness is to be
measured” (Bryan 1979). Bryan further interpreted the case that “If Abbe principle
is not possible in the machine design the consequence of angular induced error of the
stage should be considered.” This condition also applies to the Bryan principle. Both
principles are regarded as guidelines by worldwide engineers and researchers in the
design of linear motion machines, such as linear positioning stages, machine tools,
and coordinate measuring machines (CMMs). The difference between these two
principles is that Abbe principle only concerns the positioning error while Bryan
principle only applies to the straightness error of the linear stage. It is known that the
volumetric error at any spatial point of a machine tool contains three components,
Fig. 7 The linear stage with δ

Abbe error
Spindle
Cutter
Functional axis
Guideway
Lz
Ballscrew
Sensor Axis
Feedback point (Reference axis)
θ
namely one positioning error in the moving direction and two straightness errors
perpendicular to the moving direction. The positioning error can be derived with
Abbe principle and the straightness errors can be derived with Bryan principle.
In order to demonstrate that linear motion errors of a linear stage are affected by
effective angular motion errors, two examples are given. Figure 7 shows the first
example of the X-stage of a machine tool (Fan et al. 2012). The positioning reference
axis is the optical linear scale and the sensing point (feedback point) is the scale’s
pickup head. The functional point is the cutter location and the functional axis is the
line along the cutting direction, which has an Abbe offset in the Z direction (Lz) from
the reference axis. The pitch angular error (θ) of the moving table at the sensing point
will cause the positioning error (δ) of the functional point along the functional axis
by the amount of Lzθ. This is a typical Abbe error.
The second example, as shown in Fig. 8, depicts the conventional straightness
measurement of a linear stage using a straightedge, which is mounted on one side of
the slideway (functional axis). The distance between the functional axis and the
reference axis (straightedge) can be called the Bryan offset (Lyb). The roll angle of
the slide will cause the vertical straightness measurement of the dial gauge by the
amount of Lybεx. This is a typical Bryan error.
It has to be noted here that these two examples are only 2-D views. For the real
3-D view of a linear stage, more effective angular errors will generate more Abbe
error in positioning and Bryan error in straightness. In addition, the Bryan offset is
different from the Abbe offset and such a Bryan offset has rarely been noticed
before, especially in the volumetric error analysis of a NC machine tool. Figure 9
illustrates the case that the measured point of straightness error of an X-stage is at
the QPD point, which is away from the position sensing point at the read head of
the linear scale. If the linear scale has its own position error δxa(x), then, according
to the Abbe principle, the positioning error at the tool center point (TCP) of the
cutter will be
Ex ðxÞ ¼ δxa ðxÞ εz ðxÞLYa ðxÞ þ ϵy ðxÞLZa ðxÞ (1)

Fig. 8 Straightness Functional axis

measurement of a linear stage
Lyb Reference axis
with Bryan error
Slideway
Roll
Slide
Dial indicator
Z
X
εX
Y
εY
εZ
Fig. 9 Applying Abbe

principle and Bryan principle Z
LXb LXa Cutter
to a linear stage X
LYb Y
Laser beam LYa
LZb
LD
QPD
LZa
Linear scale Read head
where, LYa and LZa are Abbe offsets, εz(x) is the yaw error, and ϵy(x) is the pitch error
of the stage along X-motion. Meanwhile, if the straightness errors of the stage at the
sensing point (QPD) are δyb(x) and δzb(x), then, according to the Bryan principle, the
straightness error at the TCP of the cutter has two directions in y and z, respectively,
expressed by
Ey ðxÞ ¼ δyb ðxÞ þ εz ðxÞLXb ðxÞ εx ðxÞLZb ðxÞ (2)
and
Ez ðxÞ ¼ δzb ðxÞ εy ðxÞLXb ðxÞ þ εx ðxÞLYb ðxÞ (3)

where, LXb, LYb, and LZb are Bryan offsets, and εx(x) is the roll error of the stage
along X-motion. It is apparent that if geometric errors can be measured, the volu-
metric accuracy of machine tool can be estimated.
From the above explanation, it is essential to understand the effects of geometric
errors to the machine tool accuracy. More importantly, it is important to know the
meaning of Abbe principle and Bryan principle, and the influence on the volumetric
errors of machine tools. In order to enable constancy in monitoring the accuracy of
machine tools, it is necessary to install some appropriate sensors that can detect the
geometric errors of each axis of motion.
Low-Cost Optical Sensors for Geometric Error Measurement
A complete list of measuring instruments for machine tool metrology has been
recommended in ISO 230-11, among which the most frequently used optical instru-
ments are the laser interferometer for individual 5-DOF geometric error measure-
ment (except roll error) and the autocollimator for pitch and yaw angular error
measurements. These commercial optical instruments are bulky and expensive.
They are suitable for the accuracy calibration of machine tools but not for the role
of the sensors, which can be embedded into the machine structure for real-time
monitoring of the geometric errors of each axis. In order to become a smart machine
tool, it has to be mounted with some sensors so that the healthy condition can be
diagnosed through big data, model analysis, cloud computing, etc. to achieve the
goal of cyber-physical system (CPS) or industry 4.0. The cost of the sensor must be
low and its size must be small. Therefore, in order to become a smart machine tool
for accuracy monitoring and diagnosis, low-cost optical sensors for geometric error
measurement are required. The light source can be a He-Ne laser. However, consid-
ering the need for small and low-cost sensor, the laser diode (LD) is preferred. Some
examples will be given in the following.
Straightness Error Measurement
The System Configuration

The most commonly used sensor for straightness measurement is the laser straight-
ness measurement system consisting of a collimated laser diode and a four-quadrant
photodetector (QPD, with analog output) or a charge coupled device (CCD, with
digital output). Since the sensor must have fast response, QPD is preferred, as shown
in Fig. 10. The laser beam is treated as a reference line and the QPD is the target.
When the QPD is moved with the slide (or stage), any vertical or horizontal
straightness error of the slide will result in the laser spot movement on the QPD in
the opposite direction. Dual directional straightness errors can thus be easily
obtained. Considering that the electrical cables of QPD will be dragged or pushed
during motion that may cause signal noise, some users would like to replace the QPD
by a corner cube retroreflector (CCR) and move QPD to the laser side. The drawback
Z
Y x
QPD θp
Collimated LD θy
Slideway
Fig. 10 Straightness error measurement of a linear slide
Fig. 11 Laser and QPD with adjustable mount
is that doubling the optical path could reduce the light intensity and increase the spot
size on the QPD. Both effects will significantly change the sensitivity if the travel
length is long. Alternately, the analog output of QPD can be digitized through a
microprocessor and emitted wireless signals through Bluetooth so that cables can be
removed. For details of QPD refer Gao (2010). In practical applications of the laser
straightness measuring sensor, the laser beam needs to be adjusted in two angular
direction and the QPD can be mounted on the translational stage, as shown in
Fig. 11. Normally the QPD with the stable laser source can reach the resolution of
0.1–0.01 μm, measuring range of 100 μm, and accuracy less than 0.5 μm. Uncer-
tainty is less than 1 μm. These technical specifications are good enough for use in
machine tools.
Effect of Angular Drift of Laser Beam

All laser systems have the same physical property of angular drift, which would shift
the target spot. The cause of angular drift of laser beam could be due to the
temperature change of laser source and the optical mounts when the ambient
temperature changes. The effect of such an angular drift must be eliminated. There
are many solutions to this problem. One case is to use fiber-coupled laser diode
(LD) with collimating lens output so that the angular drift of the laser beam can be
minimized (Zhu et al. 2013; Fan and Zhao 2000). It is because the fiber will not
conduct the laser heat to the other end and the collimator is treated as a cold light
source, as shown in Fig. 12. Another solution is to use a beam expander to reduce the
beam drift (Ni et al. 1992). The drifted angle will be reduced in inverse
proportional

ratio. As given in Fig. 13, the output drifted angle is reduced to θ2 ¼ ff 12 θ1, where f
is the focal length.
The above-mentioned two methods can only reduce the drifted angle of the laser
beam; the effect still exists. The other method is to directly measure the drifted angle
using a dual-angle sensor, which will be addressed in the next section. As the drifted
angle can be detected continuously, the beam spot shift on the QPD target can be
estimated as a function of distance and, therefore, can be compensated through
calculation. Figure 14 shows the system where autocollimator 1 is used to detect
the angular drift of the emitted laser. The autocollimator 2 is used to measure the
angular change of mirror placed at a distance away. Since the mirror is fixed, it
should have no angular movement. However, due to the beam drift in space mainly
caused by the laser source, two collimators would detect the same angular error, as
shown in Fig. 15. After error compensation, the mirror actually moves with only a
Fig. 12 Fiber coupled laser straightness measuring system
Fig. 13 Beam expander can θθ22

reduce angular drift
θθ11
LD
ff11 ff22
Fig. 14 Beam drifted angle

detection in autocollimator
Mirror
Autocollimator 2 Autocollimator 1
LD
Fig. 15 Compensation of 8
beam drifted angle by 6
autocollimator
Beam drift (arc sec)
4
2 After compensation
0
0 50 100 150 200
-2
-4
Detected beam drift
-6
-8
Time (min)
very small angle. It has to be noted that the mirror was actually fixed to the base; the
detected angle of movement could be due to other error sources, such as the unstable
optics mounts. It can be a good research topic for interesting people.
Pitch and Yaw Angle Measurements
The optical structure of a dual-angle measurement system is shown in Fig. 16a. The
principle is based on the classic model of autocollimator. A divergent laser beam
outputs from a laser diode (LD) can be collimated to a parallel beam through a
collimating lens (CL). Reflected by a polarizing beam splitter (PBS) and passed
through a quarter waveplate (QWP), the beam will change to the circularly polariz-
ing state. The reflected beam from the plane mirror will change to linear polarization
when it passes through the QWP again and transmits through the PBS. The objective
lens (OBJ) will then focuses the beam to the surface of the quadrant photodetector
(QPD), which can detect two inclined angles from the normal based on the principle
of autocollimator. Figure 16b shows the compact size of the developed pitch and
yaw angular sensor. A proper signal processing circuit can easily reach 0.1 arc-sec
resolution with measuring range 100 arc-sec and uncertainty less than 0.5 arc-sec.
a Mirror
QPD
OBJ
QWP
PBS
CL
B A
Y -axis C D
X -axis
LD
b
Pitch
Flexure
mechanism
Mirror
QWP&PBS
Objective
Yaw Lens
QPD
Collimated LD
Fig. 16 Optical autocollimator: (a) principle, (b) compact size structure

a Moving mirror b
CCR
M2
M3
Q2 NPBS PD1
Q1 Q3
PD2
M1
PBS3
PBS2
PBS1
PD4
PD3
LD
Fig. 17 Optical angle interferometer: (a) principle, (b) compact size structure
Another type of pitch and yaw sensor is based on the Michelson interferom-
eter. The optical principle is shown in Fig. 17a (Fan et al. 2014). The polarized
beam from the laser diode is split by the polarization beam splitter PBS1. The
P-polarized beam passes through and the S-polarized beam is reflected to the left.
With careful rotation of the PBS1 these two beams will have equal intensity.
Then, the reflective mirrors M1, M2, and M3 guide these two beams to the object
mirror in parallel and equal path distance. When the object mirror has an angle
displacement, the change of the optical path difference will cause interference of
two returned beams after joining together, which can be converted into
corresponding angle value. After passing through the quarter waveplate Q1
twice, the left-arm beam will be converted into P-polarized beam and pass
through PBS1. The right-arm beam has similar feature. This design is to avoid
the beam returning back to the laser diode. After passing through Q3 the left-arm
beam and right-arm beam will be converted into right-circularly and left-
circularly polarized beams, respectively. The NPBS divides both beams into
two split beams of equal intensity. These four beams will be separated by
0–90–180–270 by PBS2 and PBS3 (set fast axis to 45 ) and interfere with
each other. Four photodetectors (PD) will convert the beam intensity to
corresponding current. A proper sinusoidal signal processing circuit can reach
0.1 arc-sec resolution. Figure 17b is the compact size design of the developed
yaw angle sensor with CCR. For the pitch sensor of this type, it is just to arrange
the optical plane in the vertical direction. Because of the interferometer-based
principle, the measuring range, accuracy, and uncertainty are all better than the
autocollimator.
Fig. 18 Roll angle

measurement by parallel laser QPD2
QPD2 QPD1
QPD1
beams
δz2
δz2 δz1
S
LD2
LS2 LS1
LD1
Roll Angle Measurement
There are a number of measuring principles for the roll angle. One concept is to
rotate a functional optics and detect its output change. One example is to rotate a
quarter-wave plate (QWP) in the path of a laser beam so that elliptical polarized light
is produced in heterodyne phase detection. The roll angle is sensitive to the polar-
izing angle (Jiang and Yin 2000). Another case is to rotate a rectangular prism on
which a beam splitter is attached to face the incoming beam. The roll motion of the
prism will produce differential spot shift on two OPDs (Zhai et al. 2012). This kind
of method is not good enough in the resolution and accuracy. A simple method is to
use two parallel straightness measurement kits. As shown in Fig. 18, two parallel
laser beams can be generated from two separate lasers or one laser with optics. The
roll motion of the stage causes the difference in Z-straightness of the two QPDs.
Dividing such difference by the distance of two laser beams it easily obtains the roll
error. It is noted that such a setup can also measure vertical and horizontal straight-
ness errors of the stage, being a 3-DOF sensor. This kind of differential QPDs can
reach the resolution to 0.1 arc-sec and accuracy about 1 arc-sec. Uncertainty is
around 1 arc-sec.
Applications
5-DOF Measuring System
A miniature 5-DOF sensor module has been developed for simultaneously measur-
ing three angular motions of pitch, yaw, and roll, and two straightness motions of the
machine tool, as shown the schematic diagram in Fig. 19. This is the integration of
two laser straightness measurement modules and one laser autocollimator module.
The optical module is fixed to the bed of the moving axis. The moving module is
mounted on the moving table of each axis. Pitch and yaw motions of the moving
table are transferred to the central reflective mirror and detected by the
Fig. 19 Design of a 5-DOF sensor module
autocollimator. Each side of the collimated LD and QPD pair detects the straightness
errors in two axes. The difference of vertical straightness errors of two QPDs reflects
the roll error. The physical size of the developed three-angle sensor is about
160 130 80 in the unit of mm. After calibration, the pitch and yaw sensors
can reach 0.3 arc-sec accuracy for a range of 100 arc-sec, and for roll angle it is
1 arc-sec accuracy for a range of 150 arc-sec. Each QPD has a resolution of
0.1 μm in a measuring range of 150 μm for straightness errors in both directions.
These performances are good enough for use in machine tools. Figure 20 shows the
photo of installing this 5-DOF senor in two axes of a machine tool.
It is known that any multi-degree-of-freedom measuring system has to confirm
the parallelism of all laser beams. It is actually not an easy task, especially if the
system has three emitted laser beams. If the beams are not parallel, they would be
gradually separated and out of measurement tolerance. Therefore, for the measure-
ment of a medium to large machine tool, the adjustment of laser beams to parallel
condition is a primary and necessary requirement. This part of technique has rarely
been disclosed in any literature. To this 5-DOF sensor module, one experience from
the author’s group is to adjust two-sided laser beams to be parallel to the axis of
motion and also conform to the actual roll errors of the linear stage. Figure 21 shows
the beam alignment principle that two parallel beams have to be adjusted to match
the actual roll errors at the beginning and ending positions of the linear stage. To
achieve this principle, a precision level has to be used to measure the actual roll
errors of the linear stage.
Fig. 20 Installing 5-DOF sensor in two axes of a machine tool
5-DOF Sensors for Abbe Error Compensation
From Eq. (1), it is clear that the positioning error of the TCP at any position within
the working space is induced by the angular errors of pitch and yaw associated with
corresponding Abbe offsets. A microprocessor that processes the angle signals and
calculates Eq. (1) is called the Abbe error compensator, which can dynamically
acquire the current three coordinate positions from the NC controller and, after
processing, send the compensated command into the controller. By this way, the
cutting point can be automatically adjusted in real time with the amount of Abbe
errors in space.
A test trial has been carried out on a small NC machine tool with linear scale
feedback in each axis. The experimental setup is shown in Fig. 22 for the X-axis
motion. A laser interferometer of HP5529 was mounted at different Z heights of the
spindle head as a calibration reference. Same procedure can also be conducted for the
Y-motion. Figure 23 shows the comparison of positioning errors with and without
the Abbe error compensation in the X-axis. The kinematic error of the table can be
regarded as a rigid body motion. It is clearly seen that the positioning errors can be
significantly reduced when the Abbe error compensation scheme is activated at
different heights. Figure 24 proves the effectiveness of compensation function in
Y-axis. In addition, from these two results, it can be clearly seen that the positioning
Precision Level
Moving QPDs
Fixed LDs
Fig. 21 Beam alignment for roll error assisted with a precision level
error is proportional to the Abbe offset and is mainly due to the pitch error of the
moving table. Analyzed from Eq. (1), it can be concluded that the linear scale is very
precise so that the first term in Eq. (1), i.e., the datum positioning error measured by
the sensor δx(x), is nearly zero. Also, the yaw-induced positioning error is very
small, which can be represented by the residual error in Fig. 23. In Fig. 24, the
yaw-induced positioning error is nearly zero. Figure 25 plots the positioning errors
together with the pitch error of another test. The error trends of the pitch and all
positing errors are almost the same. This can obviously prove the correctness of the
role of Abbe principle in positioning error analysis.
Parallelism Error Measurement of a Pair of Guideways
Any linear stage is composed of one moving carrier (or called table), two supporting
guideways, actuator, transmission mechanism, bearings, and displacement sensor.
As explained in section “6-DOF Geometric Errors in Linear Stage,” geometric errors
of the linear stage are caused by the improper assembly of two guideways that must
be ideally parallel and straight, as shown in Fig. 26. The primary slide and the
secondary slide must be parallel at any position along the travel. In fact, it is not
possible to achieve ideal parallel condition in machine assembly. Nonparallel
Fig. 22 Experimental setup for positioning error test
Z-offset before compensation after compensation

23cm
27cm 40
30
Positioning error (µm)
20
10
0
–50 –40 –30 –20 –10 –10 0
–20
–30
–40
X position (mm)
Fig. 23 Experimental results of X-positioning error calibration
condition can be illustrated by Fig. 27 in which the primary guideway is taken as the
reference and secondary guideway is not parallel to the reference guideway in XY
plane and XZ plane separately. The former one would cause yaw error and the latter
one would induce roll error of the stage. Therefore, measurement of parallelism of
two guideways is essential to the quality control. Conventional assembly technique
for guideway parallelism is measured by a standard parallel block and two dial
Z Offset before compensation after compensation

13cm
50 17cm
40
Positioning error (µm)
30
20
10
0
0 10 20 30 40 50
–10
Y position (mm)
Fig. 24 Experimental results of Y-positioning error calibration
Z offset 220 mm
Z Offset 235 mm
Z Offset 250 mm
50 50
Pitch angle
40 40
30 Pitch angle (arcsec)
30
Position error (mm)
20
20
10
10
0
-100 -80 -60 -40 -20 0
-10 0
-20 -10
X position (mm)
Fig. 25 Same trend of pitch and positioning errors
gauges as shown in Fig. 28. This is mostly used by machine tool builders and almost
all places in the world, also specified in ISO230-1 as one test method for machine
tool parallelism error. This method actually violates the Bryan principle two times as
the reference is not in line with the measured axis for both guideways. Each
guideway has inevitable roll and yaw errors before or after assembly. According to
the Bryan principle, the angular induced straightness error of each guideway would
combine to form the parallelism error of the pair of guideways. Hence, similar to the
Fig. 26 Ideal parallel

guideways of the linear stage Z O2
O2
Y
O2
X
O1 O2
O1
O1
Secondary
O1 guide
Primary
guide
Fig. 27 Situation of nonparallel guideways
Fig. 28 Parallelism
measurement of two
guideways with standard Secondary
parallel block guide
Dial gauge 2
Standard
parallel
block
Dial gauge 1
Primary
guide
straightness error measurement that uses a laser beam as the reference line, in
parallelism measurement two parallel laser beams with each arranged in line with
the corresponding measured guideway shall be provided. This is the required key.
a
Pentaprism
Guideway 2
b
Pentaprism
QPD Precision
Level
Guideway 1
Tilt adjuster
LD
Fig. 29 Two parallel reference beams for two guideways: (a) horizontal plane and (b) vertical
plane adjustment
In order to conform to Bryan principle, the parallelism of two guideways has to be

compared with two parallel reference lines and these two reference lines must be in
line with the two measured guideways, respectively. Figure 29a shows the concept of
generating two parallel laser beams by using a single laser source from the laser
straightness measurement system and a pentagon prism. The laser beam is first
deflected 90 through the pentagon prism to the direction of primary guideway
and the normal straightness error measurement is carried out. Having finished, just
moving the pentagon prism to the other side allows the same laser beam to deflect
90 again to the direction of the secondary guideway. This is to assure that during the
change of laser beam from the first guideway to the second guideway, the laser beam
must remain parallel to each other in both of horizontal and vertical directions.
However, by its nature, the pentagon prism can only assure the squareness in the
horizontal plane when moved from the front of first guideway to the second
guideway, i.e., the XY plane in Fig. 29. Therefore, for aligning the laser beam to
assure parallelism in the vertical XZ plane, a precision level has to be used to adjust
the pentagon prism to be equal in tilting angle at both places so that two laser beams
are at an equal angle in vertical direction, as shown in Fig. 29b. This is the second
key in parallelism measurement. Moreover, the zero setting of the QPD can only be
made to the first target point during the straightness measurement of the first
guideway. This is to keep only one reference point. If it is adjusted to zero again at
the first point of the second guideway, the initial reference datum will be lost.
Figure 30 shows the photo of setting up the LD, pentaprism, QPD, and level and
tilt adjusters for measuring the straightness error of each guideway and the parallel-
ism error of two guideways.
Accuracy Improvement of Linear Stages in Assembly Process
It has been detailed above that the accuracy of a precision machine is entirely
dependent on the geometric errors of each axis. Most researches in the past focused
QPD
LD
Dual-angle Pentaprism Level &

adjuster
Tilt adjuster
Fig. 30 Experimental setup for parallelism measurement of two guideways
on the measurement and compensation of existing errors of the assembled moving

stage. It is known that geometric errors, including three linear and three angular
terms, in each axis are entirely due to improper assembly of the supporting guide-
ways. If guideway errors could be reduced during the assembly process, final
volumetric errors of the machine would be largely reduced even without compen-
sation strategy. This is to point out the fact that tracing the error to the source where it
happens is more important than just measuring the error at the functional point of the
machine. It is because that if the source error could be suppressed the resulted
volumetric error would be reduced after assembly. Unfortunately, the task of guide-
way assembly in most factories is carried out by low-level technicians who know
only a little about the fundamentals of precision machinery. In fact, it needs skillful
and experienced engineers to implement the task, because many instant judgments
are required from the measured data during assembly. Following will address the
experience of proper guideway assembly from the author’s group.
Any linear guideway is purchased from the market with specified accuracy grade.
It means that even if with the highest quality, the guideway still has somewhat form
straightness error. Loosely speaking, the guideway can be regarded very straight.
Strictly speaking, the guideway can be regarded small bent. Without direct mea-
surement, the user would not know where and how much the local out-of-straight-
ness is along the length of a guideway. In other words, the exact freeform of the
guideway is unknown. To build up a linear stage, at least two parallel guideways are
needed to mount onto the machine bed of the stage, as shown in Fig. 31. The
freeform of the rail is unknown. The reference plane that contacts the rail is a ground
surface. There is also the form error of each contact plane and such error is unknown
Vertical
adjusting
screw b
Horizontal Vertical screw
screw
Horizontal
adjusting screw
Contact
planes
Pressed
rod
Fig. 31 Photo showing two guideways on the machine bed (a) showing detailed adjusting screws
of the rail (b)
as well. Therefore, it is impossible to keep contact of the rail along with the reference
plane without local gaps. In other words, in order to obtain a very straight guideway
after assembly, the screw at different place should be exerted with different tighten-
ing torque according to the actual shape of the rail and the reference surface at that
place. This requirement has been overlooked in most places where a constant torque
to set the adjusting screw is the standard of operation for each assembly technician.
Although fine adjustments of some particular screws will be carried out after the
straightness measurement with a straightedge, this kind of measurement brings in
another Bryan error term. Consequently, the assembled guideway could never be
assured the final straightness. No need to say that the final parallelism must be out of
tolerance. A proper assembly process of the stage is proposed as follows:
Step 1 It is to measure the free shape of the guideway using the laser straightness
measuring system. The measurement setup is referred from Fig. 29. Two end screws
are tightened up and all other screws are free. The laser beam should hit the QPD
center at these two end points, which is the process of leveling the guideway. The
straightness errors thus measured reflect the free-shape of the used guideway.
Figure 32 shows one example of this case. The initial measured free-shape straight-
ness errors in Y-axis and Z-axis are found very large both at the 600 mm point. The
free-shape in Y direction is bent-out before 400 mm and bent-in after 400 mm. The
free-shape in Z direction is bent-out after 400 mm.
Step 2 It is to decide the assembly sequence and procedure based on the measured
shape. There are two situations: one is away from the contact surface (bent-out
condition) and other one is closer to the contact surface (bent-in condition).
Therefore at the most bent-out (peak) point, the nearest screw should be twisted
down in order to reduce the gap. While at the most bent-in (valley) point, a piece of
a 40 b
20 60
Y-error (μm)
Z-error (μm)
- 20 0 40
200 400 600 800
-40 20
-60
0
-80
0 200 400 600 800 1000
- 100 -20
X-Displacement (mm) X-Displacement (mm)
Fig. 32 The initial free-shape straightness error of the guideway in (a) Y direction and (b) Z
direction
Fig. 33 Insert a thickness

pad to lift up the valley point
Fig. 34 Adjusted shape of

the first guideway from free- 40 After Before
shape condition 20
Y-error (μm)
0
-20 0 200 400 600 800 1000
-40
-60
-80
-100
-120
X-Displacement (mm)
thickness pad has to be insert into in order to lift the point up, as schematically
shown in Fig. 33. The thickness of the pad is selectable according to the real case.
The results of adjusting screws in Y-direction straightness error are shown in
Fig. 34. Compared to Fig. 32a, it can be seen that valley point at 600 mm has
been pushed up 40 μm by inserting a thickness pad and the original peak point has
been pressed down to nearly flat.
Step 3 It is to repeat the above two steps and gradually tighten the effective screws
from the shape previously measured. During the process, because of the high rigidity
of the steel guideway, it is not able to press the setting screw by force, which will be
in vain to get any movement because of very high constraint force from already fixed
screws of two sides. Therefore, it has to release the nearby screws before adjusting
on the middle screw. The amount of nearby screws and the degree of releasement are
dependent on experiences and skills. This is an important expert system that has to
be learned. Figure 35 shows the example continued from the last step. The straight-
ness error in Y direction is gradually reduced with repeated process of this step. The
final assembled guideway has been adjusted to the error from 100 μm to within
20 μm, being improved by around 80% of the accuracy. Figure 36 shows the results
of straightness error in Z direction after the above-mentioned three steps. The
maximum error, from initial 65 μm of the free-shape to 10 μm in the end, becomes
a great improvement too.
It has to be pointed out here that the guideway adjusting task is done on a very
poor accuracy sample. Even to the best efforts in the laboratory, the final straightness
error is still relatively large compared to normal machine tool guideways. This is a
laborious task but need experience to achieve the job. The on-site technician in the
machine tool assembly plant has to be trained very well. This, however, has been
ignored by traditional machine tool builders.
Fig. 35 Improvement in
Yaxis1
Step 2 Yaxis2
Final
Y-straightness accuracy after 40
assembly
20
0
Y-error (μm)
-20 0 200 400 600 800 1000
-40
-60
-80
-100
-120
X-Displacement (mm)
Fig. 36 Improvement in Original Step 2 Final

Z-straightness accuracy after 70
assembly 60
Z-error (μm)
50
40
30
20
10
0
-10 0 200 400 600 800 1000
-20
-30
X-Displacement (mm)
Step 4 It is to adjust the parallelism of the second guideway with respect to the first
guideway. The method of parallelism measurement of two guideways has been
described in section “Parallelism Error Measurement of a Pair of Guideways.”
This part of work is to adjust screws of the second guideway not only for the
straightness but also for the parallelism at the same time. The way to achieve parallel
laser beams can follow the procedure given in Fig. 30 by keeping the same tilted
angle in the vertical plane, or that given in Fig. 21 by keeping the same roll angle at
both ends of the travel. Figure 37 shows the parallelism of two practiced guideways
before and after adjustments. The final error has been suppressed to very small angle,
represented by the slope difference of two straightness error forms. The parallelism
error is about a few arc-sec.
Step 5 Fine tuning adjustment of angular errors. As given in the first part of this
chapter, machine tool accuracy is originated from 6-DOF geometric errors of each
axis. The volumetric error within the working space is particularly affected by
remaining angular errors multiplied by the corresponding Abbe offset (for position-
ing error) and Bryan offset (for two lateral straightness errors). Therefore, even if the
straightness error and parallelism error of the supporting guideways have been
adjusted to the best condition, it does not mean that the pitch, yaw, and roll errors
are small. By looking at the local difference of two adjusted guideways in Fig. 37,
the difference can be up to the order of about 30 μm at positions of 300 mm and
500 mm. The generated yaw angular error is still large. Therefore, the final fine
tuning process to reduce the angular errors is necessary. This procedure is not paid
attention by most machine tool builders.
Since a linear stage is supported by two guideways. Each guideway can also be
treated as a stage composed of a rail and a slide (carrier). The angular error of the
slide moving along the rail can be adjusted by tuning two consecutive screws in
different directions without changing other screws. This process is called fine tuning.
The experience from the author’s group indicates that the final angular error of the
moving table of the linear stage is the average of two angular errors of the slides at
the same position. Figure 38 shows the results of pitch error before and after
Fig. 37 Results of nd
parallelism adjustment of two
1st
參考 2nd before 2 after
待測軌調整後
100
guideways 50
0
Y-error (μm)
-50 0 200 400 600 800

-100
-150
-200
-250
-300
-350
-400
X-Displacement (mm)
Fig. 38 Comparison of pitch 250

error with and without
Pitch error (arc sec)

adjustment 200
Before
150
100
After
50
0
0 200 400 600 800
X (mm)
Fig. 39 Comparison of yaw 200

Yaw error (arc sec)
error with and without

adjustment 150
Before
100
50
After
0
-50 0
200 400 600 800
X (mm)
adjustment. A significant improvement can be achieved if the adjustment effort is

paid. The yaw and roll errors are also possible to be controlled to acceptable
conditions, as shown in Figs. 39 and 40, respectively.
From this finding, it can be concluded that each guideway has to be adjusted to
reduce not only the straightness error but also the angular error. The final 5-DOF
geometric errors, including two straightness and three angular errors, of the stage
will thus be largely reduced. The final volumetric error will accordingly be signif-
icantly suppressed.
The author would like to point out that the procedure of adjusting the screws to
reduce the geometric errors of the linear axis of a machine tool is not limited to the
given example. It is indeed a skillful technique that must be learned and trained in the
factory and laboratory. It could be an expert system assisted by optical sensors.
Conclusions
The development of smart machine tools will be the trend toward the worldwide need
of intelligent manufacturing technology nowadays, which is the goal of industry 4.0, as
well as cyber-physical system and China made 2025. In order to monitor the machine
tool accuracy under cutting condition, small and low-cost embedded sensors for real-
600
Roll error (arc sec)

500
Before
400
300
200
100 After
0
0 100 200 300 400 500 600 700 800
X (mm)
600
Roll error (arc sec)
500
Before
400
300
200
100 After
0
0 100 200 300 400 500 600 700 800
X (mm)
Fig. 40 Comparison of roll error with and without adjustment
time detecting geometric errors of each axis of machine tools are urgently needed. This
report highlights the importance of Abbe principle and Bryan principle to the volu-
metric accuracy of machine tools through respective angular errors and Abbe or Bryan
offsets. Some geometric error sensors are introduced in this report. The integration of
different sensors into a 5-DOF sensor module has demonstrated the feasibility of
volumetric error measurement and compensation in a NC machine tool. Proper use
of optical sensors in machine tool metrology could help improve the accuracy. Some
examples have been addressed from the past experiences of the author’s group. Optical
sensors capable of measuring geometric errors of the machine tool can be various. The
embedment of error compensator into the numerical controller is a key technology to
overcome. A variety of research topics can be generated for developing optical sensors
required for machine tool metrology.
References
Abbé E (1890) Meßapparate für physiker. Z Instrumentenkd 10:446–448
Bryan JB (1979) The Abbe principle revisit: an updated interpretation. Precis Eng 1:129–132
Drescher J (2003) Analytical estimation of measurement uncertainty in surface plate calibration by
the Moody method using differential levels. Precis Eng 27:323–332
Ekinci TO, Mayer JRP (2007) Relationships between straightness and angular kinematic errors in
machines. Int J Mach Tools Manuf 47:1997–2004
Fan KC, Zhao Y (2000) A laser straightness measurement system using optical fiber and modula-
tion technique. Int J Mach Tools Manuf 40:2073–2081
Fan KC, Wang TH, Wang CH, Chen HM (2012) Development of an Abbe error compensator for
NC machine tools. In: Proceedings of the 37th MATADOR conference, Manchester, 25–27 July,
pp 317–322
Fan KC, Wang HY, Yang HW, Chen LM (2014) Techniques of multi-degree-of-freedom measure-
ment on the linear motion error of precision machines. Adv Opt Technol 3(4):375–386
Ferreira PM, Liu CR (1993) A method for estimating and compensating quasi-static errors of
machine tools. Trans ASME J Eng Ind 115:149–159
Gao W (2010) Precision nanometrology. Springer, New York
Gao W, Wu SM, Boss H, Haitjema H et al (2015) Measurement technologies for precision
positioning. CIRP Ann Manuf Technol 64:773–796
ISO 230-1:2012. Test code for machine tools – Part 1: geometric accuracy of machines operating
under no-load or quasi-static load conditions
ISO 230-2:2014. Test code for machine tools – Part 2: determination of accuracy and repeatability
of positioning of numerically controlled axes
ISO 230-4:2015. Test code for machine tools – Part 4: circular tests for numerically controlled
machine tools
ISO 230-6:2012. Test code for machine tools – Part 6: determination of positioning accuracy on
body and face diagonals (Diagonal displacement tests)
ISO 230-7:2015. Test code for machine tools – Part 7: geometric accuracy of axes of rotation
ISO/WDTR 230-11.3:2011. Test code for machine tools – Part 11: measuring instruments and their
application to the machine tool geometry tests
Jiang H, Yin C (2000) Sensitivity enhanced roll angle measurement. Opt Eng 39(2):516–519
Li YT, Fan KC (2017) A novel method of angular positioning error analysis of rotary stages based
on the Abbe principle. Proc Inst Mech Eng B J Eng Manuf, Special issue article, 1–8
Majda P (2012) Relation between kinematic straightness errors and angular errors of machine tool.
Adv Manuf Sci Technol 36(1):47–53
Marsh E (2010) Precision spindle metrology, 2nd edn. DEStech Publications, Lancaster
Ni J, Huang PS, Wu SM (1992) A multi-degree-of-freedom measuring system for CMM geometric
errors. Trans ASME J Eng Ind 114:362–269
Okafor AC, Ertekin YM (2000) Derivation of machine tool error models and error compensation
procedure for three axes vertical machining center using rigid body kinematics. Int J Mach Tools
Manuf 40(8):1199–1213
Soons J, Theuws F, Schellekens P (1992) Modeling the errors of multi-axis machines: a general
methodology. Precis Eng 14(1):5–19
Tlusty J (1980) Testing of accuracy of machine tool. In: Technology of machine tools, supplement
1. National Livermore National Laboratory, Livermore
Zhai Y, Feng Q, Zhang B (2012) A simple roll measurement method based on rectangular-prism.
Opt Laser Technol 44:839–843
Zhu F, Tan JB, Cui JW (2013) Beam splitting target reflector based compensation for angular drift
of laser beam in laser autocollimation of measuring small angle deviations. Rev Sci Instrum
84:065116
Optical Micro-CMM
5
Yasuhiro Takaya and Masaki Michihata
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
Coordinate Metrology for Micro-products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124
Requirements for the Micro-CMM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
Problems in the Micro-probe System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130
Principle of the Laser Trapping Micro-probe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 137
Physical Basis of Optical Trap Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 139
Numerical Analysis of Dynamical Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145
Experimental Analysis of Position Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 156
Development of Micro-CMM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
The Structure of the Micro-CMM System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 167
Coordinate Stage System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
Micro-probe System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
Measurement Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
Experimental Analysis of Probing Property . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171
Detection of Position and Normal Direction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 177
Coordinate Measurement of a Micro-glass Sphere . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 182
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 184
Abstract
Rapid progress in microfabrication technology with great precision has been
made owing to the miniaturization of complex product. However, no concomitant
development has taken place in the evaluation technology for micro-components,
particularly in three-dimensional metrology, which is still in development. The
gap between the traditional CMM’s specification and the demand for geometric
Y. Takaya (*)
Osaka University, Osaka, Japan
e-mail: takaya@mech.eng.osaka-u.ac.jp
M. Michihata
The University of Tokyo, Tokyo, Japan
e-mail: michihata@nanolab.t.u-tokyo.ac.jp

https://doi.org/10.1007/978-981-10-4938-5_5
124 Y. Takaya and M. Michihata
dimensioning and tolerancing of a micro-component with the basic size of

less than several mm has been widening because a measuring accuracy of 100
to 10 nm is needed for the basic size of 10 to 1.0 mm. This situation has recently
led to increased focus on the development of technology for practical micro-
CMMs. The key – and hence the bottleneck – in realizing a micro-CMM system
is the micro-probe system. This is because for features smaller than several dozen
micrometers, the probing uncertainty is considerably exacerbated by scaling
effects. Accordingly, the development of the probing system is attempted with a
different approach from the traditional concept for the sake of scale effect that is
remarkable with the use of a touch trigger probe. Therefore, the basic concept of
the laser trapping micro-probe was proposed, whose principle is based on the
single-beam gradient force optical trap of a micro-probe sphere in air. This probe
can perform noncontact measurement of a position that is optically sensed based
on the displacement of the micro-probe sphere given by dynamical interaction
with the surface. Moreover, the improved circular motion probing technique
enables to measure a surface position and a normal direction simultaneously.
This probing system was integrated with a prototype micro-CMM that was
specially designed for evaluating the performance of the probe system. The
dynamic properties of the coordinate stage were evaluated by a heterodyne
interferometer. The test results showed that the stage had a positioning accuracy
of several dozen nanometers.
Keywords
Coordinate metrology · Coordinate measuring machine · Macro-machining ·
Micro-product · Nano-tolerance · Optical radiation pressure · Micro-probe ·
Laser trapping · Vibration probing mode · Circular motion probe
Introduction
Coordinate Metrology for Micro-products
In only a few decades, microfabrication technology has grown from nascency into
an established manufacturing system with submicrometer accuracy. However, no
concomitant development has taken place in the evaluation technology for micro-
components, particularly in three-dimensional (3D) metrology, which is still in
development. Currently, expansion in demand for micro-products in various indus-
trial fields of mechanical, electrical, optical, chemical, biomedical, etc. is acceler-
ated. Rapid progress in microfabrication technology with great precision has been
made owing to the miniaturization of complex product. Widely spreading miniatur-
ized products such as micro-machine, micro-electro-mechanical system (MEMS),
micro-optical-electro-mechanical system (MOEMS), micro-total analysis system
(μTAS), biomedical device, micro-parts and components, etc., are needed to fabri-
cate, namely micro-gear, micro-connector, micro-lens, and micro-forceps shown
in Fig. 1.
5 Optical Micro-CMM 125
Complexity
Simple Complex
1m
0.1 m
Turbine gear (MUSASHI Corporation 2018)
Dimension of structure
30 mm
0.01 m Cycloid gear (SYVEC Corporation 2018)
1 mm
1 mm
Micro-motor
Micro-lens (Impress
Corporation 2018)
Micro-gear (Carl Zeiss Inc. 2018) Micro-connector (Plunkett

0.1 mm Associates Ltd. 2018)
0.1 mm
Micro-forceps (Komatsuseiki Kosakusho.Co.,Ltd 2018 )
Fig. 1 Trends in micro-products with complexity
Not only improving conventional ultraprecision machining tool but also devel-
opment of the novel machining techniques capable of achieving the accuracy of
materials processing with the order of nanometers has been made, as predicted by
Taniguchi (1983). Micro-mechanical parts with microfine structure (Alting et al.
2003) as well as size smaller than several mm could be realized by using machining
techniques with an accuracy higher than submicrometer, such as mechanical micro-
machining (Dornfeld et al. 2006; Kuriyagawa 2009; Suzuki et al. 2006; Takeuchi
2008; Moriwaki and Shamoto 1995), micro-stereolithography (Kajihara et al. 2007;
Kawata et al. 2001; Maruo and Ikuta 2000; Nishino et al. 2003; Varadan et al. 2001),
lithography (Becker et al. 1986; Becker and Heim 2000) such as LIGA, micro-
electro discharged machining (EDM) (Masuzawa 2009; Kunieda et al. 2005; Mohri
2005), and plastic moulding (Heckele and Schomburg 2004).
It is successful to develop ultraprecision machining capable of achieving
the accuracy of materials processing with the order of nanometers. Introducing
the precision mechanical technologies such as air bearings and hydrodynamic
bearings, linear motors, static pressure guide structure to a mechanical micro-
machining tool, machining accuracy is drastically improved to nanometer level.
The machining accuracy performed by the ultraprecision machining tool for
industrial use, for example, ROBOnanoUi by FANUC corporation and Ultra
Nano100 by Sodic Co. Ltd., is achieved higher than 10 nm. Micro-stereo-
lithography is an emerging technology, which achieves machining accuracy of
submicrometer by employing near-field and two-photon absorption to overcome

diffraction limit. Micro-EDM with the electrode of 1 μm in diameter was devel-
oped. In addition, improvement in electrode material and working fluid with
graphite powder enables submicrometer material removal resolution. The micro-
gear of 0.15 mm in diameter can be formed by plastic injection molding using the
high precision mold with feedback-controlled injection.
As these various micro-products and parts are realized, it is necessary for quality
control to develop measurement and evaluation technologies. Especially, mechanical
parts and optical parts are needed to accredit accuracy of geometrical quantities (Pfeifer
et al. 2001); for instance, the form accuracy of pickup lenses for DVD is required in
nanometer order. In measuring of two-dimensional (2D) and two-and-a-half dimen-
sional (2.5D) profile for micro-parts like these, a light optical microscope (LOM) such as
fringe projection (Savio et al. 2007a), coherence scanning interferometry (CSI), confocal
microscope (Jordan et al. 1998), focus variation microscopy (Danzl et al. 2009), and
auto-focus microscopy as well as a conventional stylus profilometer is used (Lonardy et
al. 2003; Yoshizumi et al. 2002; Leach and Giusca 2008; Hidaka et al. 2003). In
addition, a critical dimension-scanning electron microscopy (CD-SEM) is developed
to measure a fine profile with an accuracy of nanometer order.
On the other hand, geometric dimensioning and tolerancing (GD&T) for
machined parts is based on coordinate metrology. The common coordinate measur-
ing machines (CMMs) are playing an important role to guarantee the precision
of products as a standard instrument for 3D shape measurement. However, it is
impossible to perform GD&T of a micro-machined part by employing a conventional
probe of 1 to 0.5 mm in diameter. As another approach to 3D-fine profile measure-
ment with nanometer accuracy, the improvement of a micro X-ray computed tomog-
raphy (Micro-CT) (Bartscher et al. 2007; Carmignato et al. 2009) and the scanning
probe microscopy (SPM) (Savio et al. 2007b; Yamamoto et al. 2000; Danzebrink
et al. 2005; Garnaes et al. 2006) have been attempted.
These measurement methods are not sufficient to satisfy required measurement
performances for the geometrical complexity realized by progressing micro-fabrica-
tion technologies. As shown in Fig. 2 (Hansen et al. 2006), increasing micro-
components with more complex 3D-structure, nevertheless, coordinates metrology
for micro-products which enables to complete 3D-measurement with high accuracy
at an early stage of development.
Requirements for the Micro-CMM
Geometrical quantities of precise machined parts, which compose a conventional

product, should be guaranteed relative accuracy of 104 to 106 in GD&T.
As shown, the typical high-precision CMM of UPMC1200 Carat (Carl Zeiss Inc.)
in Fig. 3, a CMM which achieves higher than 105 in relative accuracy, is generally
used for 3D shape measurement in industry. The performance of CMM is specified
as a standard measuring equipment in the current international standards of
ISO 10360-7. These CMMs are enabled to comprehensively cover the required
measuring performances for a machined part with the basic size of 0.1–1.0 m
Light optical
microscope (LOM)
Stylus instrument
Coordinate measuring
machine (CMM)
10–3
Structural dimensions [m]
microscope (SEM)
Scanning Electron
3D SEM
10–6
?
Scanning probe
microscope (SPM)
10–9
2D 21/2D 3D
Structural complexity
Fig. 2 Classification of measurement methods for micro-components (Hansen et al. 2006)
in the conventional manufacturing system so far. However, the gap between the
traditional CMM’s specification and the demand for GD&T of a micro-component
with the basic size of less than several mm has been widening because a measuring
accuracy of 100 to 10 nm is needed for the basic size of 10 to 1.0 mm. Moreover,
the traceability system including calibration techniques and artefacts has not been
improved yet. Therefore, the development of the CMM for micro-scale machined
parts, that is to say, the micro-CMM is attempted by the national laboratories
to establish the scale-interface of measurement techniques which enables to satisfy
measuring accuracy for basic size of meter to millimeter seamlessly.
K. Takamasu suggested that key factors are a scale, an actuator, a stage system,
a micro-probe, and material to establish the micro-CMM (Takamasu et al. 1996).
H. Schwenke placed emphasis on the probing system for a miniaturized high-
precision CMM also (Schwenke et al. 1999). In addition, the technological solution
is to prepare an artifact which is guaranteed nanometer order accuracy for calibrating
a micro-CMM. As those problems pointed out, there are many technological
difficulties to be overcome in establishment of a micro-CMM. Thus, the trend
in 3D shape measuring instruments such as an ultra-accurate and micro-CMM are
summarized briefly.
b
X-axis 1150 mm
Measuring range Y-axis 1500 mm
Z-axis 1000 mm
Accuracy 1.5 + L/300 µm
Resolution 0.08 µm
Measuring force 0.2 N
Fig. 3 Conventional CMM by Zeiss (Carl Zeiss Inc. 2009). (a) Photograph. (Courtesy of Carl
Zeiss Inc.). (b) Specification
From the latter half of the 1990s to the beginning of 2000, mainly European national
laboratories were concerned to develop the high-precision CMMs. From the beginning
of the 2000s, a variety of ultra-accurate and micro-CMMs have been developed and
realized commercialization by companies as well as universities, and European national
laboratories began to provide highly accurate calibration services using the developed
high-precision CMM. The typical Nano CMM project in Europe related to those trends
was initiated in 2007. The purpose of the project was to establish an ultra-accurate
CMM which could realize coordinate metrology with a volumetric uncertainty of less
than 100 nm and calibration services using the artefact. Ultra-accurate and micro-CMM
technology was accelerated to progress by vigorous research and development activ-
ities as shown in Table 1. Moreover, a lot of efforts have been made to improve
accuracy of CMM in outstanding research and developments of key factors, namely,
the stage system and the probe system (Peggs et al. 1999; Vít and Prokop 2007;
Vermeulen et al. 1998; Kirkland et al. 2004; Fan et al. 2006, 2007; Kim 2001; Kim
et al. 2003; Cao et al. 2002; Brand and Kirchhoff 2005; Takamasu et al. 1997;
Yoshizumi et al. 1987; Tsutsumi et al. 2005; Meli and Kung 2004; Kung et al. 2007;
Bos et al. 1998; Shiozawa et al. 1998; Stoup and Doiron 2003).
In particular, two representative cases of developing the ultra-accurate CMM in
England and Germany are introduced. National physical laboratory (NPL) started up
the research project concerning the need for CMM for micro-scale components at an
earlier stage. As a result, the small CMM (Leach et al. 2001, 2004) was promptly
developed as shown in Fig. 4. The basic structure is that the miniature probing
system is positioned at the center of three mirrors as indicated in the schematic
Table 1 The state of arts ultra-accurate coordinate measuring machine

Measuring Uncertainty
Name Institute Country Capability volume (mm) (nm)
μCMM PTB German 3D 25 40 25 100
Small CMM NPL UK 3D 50 50 50 100
Micro-CMM METAS Swiss Lands 3D 90 90 38 50
ISARA IBS The 3D 100 100 40 30
precision Netherlands
Nanocord Mitutoyo Japan 3D 300 200 100 (300 + 2L/
106)a
F25 Zeiss German 3D 100 100 50 250
Nano-CMM Eindhoven The 3D 53 50 4 25
University Netherlands
UA3P-L Panasonic Japan 2.5D 100 100 50 150
3D-CMM NIKON Japan 2.5D 400 400 100 20
Volumetric KAIST Korea 3D 300 300 300 30
interferometry
CMM
a
L is measuring length (mm)
diagram of Fig. 4a and a picture of Fig. 4b. The displacement of the mirror is
measured using the interferometer so that Abbe error is reduced effectively. The
interferometer is fixed to the metrology-frame made of low thermal expansion alloy
Inver™, on which a micro-component is also placed. The structural drawing of the
micro-probe is shown in Fig. 4c. The stem with 1 mm probe ball is sustained by three
tungsten carbide tubes jointed to BeCu strips with the target disc such that the
displacement of each target disc is measured using the electrostatic capacitance
sensor. The probing system with measuring force of less than 0.1 mN is realized.
It is introduced to ISARA system (IBS precision engineering Corp.).
In Physikalisch-technische bundesanstalt (PTB), the CMM for micro-scale
components, the so-called μCMM (Cao et al. 2002; Brand and Kirchhoff 2005), is
developed as shown in Fig. 5. Figure 5a illustrates the schematic diagram of the
μCMM together with the appearance in Fig. 5b. A micro-component is placed on the
metrology frame made of Inver™ on which three reference mirrors are attached. In
contrast to the fixed probe system with the reference cuboid made of Zerodur ®, the
metrology frame is precisely driven by the positioning system which consists of air
bearings and liner scales of 10 nm in resolution while measuring displacement and
angle between the reference mirrors and the reference cuboid using six inter-
ferometers. Both the opto-tactile micro-probe and the 3D-boss-micro-probe are
implemented. The structural drawing of the opto-tactile micro-probe is shown in
Fig. 5c. Illuminating the probe sphere through the optical fiber with which is
inconjunction, the scattered light is detected the CCD to sense x and y-position.
Z-position can be sensed by detecting an image of the z-target mark. This probing
system achieves the uncertainty of 0.7 μm, which is commercialized by Werth-
messtechnik as Werth 3D Fiber Probe (Werth-messtechnik 2018).
b
a
CMM z-axis
interferometer
3 mirrors on
reflector cube
frame
autocollimator
component miniature probing

system
metrology
frame
c
probe body tungsten carbide tube
fastening sensors
screws
sensor
target discs
tungsten
1 mm probe ball
Be-Cu strip carbide
on stem
tube
Fig. 4 Small CMM from NPL (Leach et al. 2001). (a) CMM system. (b) Picture of the system.
(c) Probe system
Problems in the Micro-probe System
A micro-CMM is required for coordinate metrology with nanometer order accuracy.

The key – and hence the bottleneck – in realizing such a micro-CMM is the micro-
probe system. This is because for features smaller than several dozen micrometers, the
probing uncertainty is considerably exacerbated by scaling effects. Indeed, there are
little feasible micro-probe systems for industrial use yet (Weckenmann et al. 2004;
Schwenke et al. 2002). In this section, trend of research and development in a micro-
probe for CMM is summarized; after that, technological problems are extracted.
From the beginning of the 1990s, a variety of micro-probes have been developed
as with ultra-accurate and micro-CMMs as illustrated in Fig. 6. Takamasu proposed
the novel principle of a pneumatic ball probe where the probe sphere displacement
is detected using laser light scattering (Fig. 6a). The micro-probing system of which
the basic structure is a kind of suspension was developed by NPL (Fig. 6b) and
Fig. 5 Micro-CMM from PTB (Cao et al. 2002). (a) CMM system. (b) Picture of the system.
(c) Probe system
Eindhoven University. The bending can be measured by high-resolution sensing

methods such as capacitive displacement sensors (Fig. 6c), an optical position
detection (Fig. 6d), and the integrated strain gauges (Fig. 6e). This type of probing
system achieves 2 nm repeatability over whole range in any direction and the 3D
probe uncertainty of 10 nm, which is commercialized by XPRESS precision engi-
neering as an ultra-precision 3D probe (XPRESS precision engineering 2018).
The basic design of a micro-probe with an elastic hinge structure is shown in Fig.
6f. Furutani proposed the optical method to detect the displacement of a probe sphere.
The laser light is focused to the quadrant photodiode (QPD) by using the ball lens
which is fixed to the other end of stylus with the probe sphere. National Institute of
Advanced Industrial Science and Technology (AIST) in Japan introduce almost the
same design as the micro-probing system (Fig. 6g) by Furutani to construct the micro-
factory. Swiss Federal Office of Metrology and Accreditation (METAS) developed the
Fig. 6 Micro-probe system structured based on hinge and so on. (a) Takamasu (Takamasu et al.
1998). (b) NPL probe (Leach et al. 2004). (c) Eindhoven probe (Haitjema et al. 2001) (capacitive
sensor). (d) Eindhoven probe (Haitjema et al. 2001) (optical sensor). (e) Eindhoven probe (Haitjema
et al. 2001). (f) Furutani probe (Furutani et al. 2001). (g) AIST probe (Ogura and Okazaki 2009).
(h) PTB 3D-boss probe (Cao et al. 2002)
ultraprecision micro-CMM using a low force probe of which design is the three-
dimensional complex elastic hinge structure with electromagnetic induction type
displacement sensors (Fig. 6h). Beside a ruby probe sphere, probes made of glass or
tungsten were tried to be miniaturized (Kung and Meli 2008). The 3D-boss-micro-
probe (Fig. 6i) where a silicon thin film is used as an elastic hinge structure was
developed by PTB. The piezoresistive Wheatstone bridge circuit is implemented on
the silicon thin film as the principle of position detection (Dai et al. 2009). All these
probes have a probe sphere of several hundreds of micrometers in diameter which is
too big to use for a micro-CMM. Therefore, to realize a probe sphere smaller than
several dozen micrometers is the critical issues for a micro-CMM technology.
Fig. 7 Micro-probe system based on optical fiber. (a) Ikeno grass (Takakai et al. 2006). (b) Oiwa
probe (Oiwa and Tanaka 2005). (c) Takaya probe (Eom et al. 2009). (d) Murakami probe
(Murakami et al. 2009). (e) Ji probe (Ji et al. 2009). (f) NIST probe (Muralikrishnan et al. 2006)
Start with the opto-tactile micro-probe by PTB, an optical fiber with a micrometer
size probe sphere (optical fiber type micro-probe) is proposed in several researches
as shown in Fig. 7. As the laser machining technique in early stage, sphering tip
of glass tube was investigated by Ikeno (Fig. 7a). Oiwa used bundled three optical
fibers with which a probe sphere is glued as an optical fiber type micro-probe
(Fig. 7b). Takaya proposed the novel optical fiber type micro-probe where a probe
sphere with the diameter of 8 μm is trapped by radiation pressure of laser light at the
tip of the optical fiber of 5 μm in diameter (Fig. 7c), which was only achieved in
water. The optical fiber type micro-probe with optical fiber of 50 μm and a probe
sphere of 30 μm in diameter is fabricated by Murakami. The deflection of the fiber at
contact is measured by the optical method as shown in Fig. 7d. This Murakami probe
is improved to the smallest stylus tip of 1.2 μm in diameter using CO2 laser
fabrication method (Murakami et al. 2017). To make a signal-detecting position
robust, the fiber Bragg grating was introduced by Ji (Fig. 7e). An optical fiber type
micro-probe was considered in National Institute of Standards and Technology
(NIST) also (Fig. 7f). This type of probe has technological problems such as the
stiffness of the optical fiber is low, the sensitivity in z-direction is not enough, and
3D-isotropic position detection is hardly obtained in principle.
As another type of a micro-probing system, the bottom-up approach based on
SPM (SPM type micro-probe) is attempted for the purpose to achieve nanometer
order accuracy as indicated in Fig. 8. The basic idea to use a cantilever of atomic
force microscopy (AFM) for the CMM probe was proposed in the 1990s (Chiffre
Fig. 8 Micro-probe system based on SPM cantilever. (a) Masuzawa probe (Masuzawa et al. 1997).
(b) Mitsui probe (Goda and Mitsui 2008). (c) Dai CMM probe (Dai et al. 2008)
et al. 1999). The vibroscanning probe consists of two cantilever styli with a sharpen
tip. As shown in Fig. 8a, using a probe with microvibration, the principle of
detection is based on sensing the switch of the electrical circuit of probes between
the probe and the object surface in nondestructive. Mitsui improved the SPM type
micro-probe to the scanning tunnel microscopy (STM) mode probe which is made
by positioning control with detecting the tunnel current between the probe and the
object surface (Fig. 8b). On the other hand, the novel atomic force probe with two
cantilevers in right angle relation is developed for sidewall scanning as shown in Fig.
8c. The principle of detection for the novel atomic force probe is based on the optical
lever. This sensing technique is applied to the CMM probing system as illustrated in
Fig. 8d. As different sensing method from the optical lever, the implementation of
Wheatstone bridge circuit on the cantilever, which is given a tilt, was made to realize
a compact monolithic probe unit (Fig. 8e).
The repeatability of a micro-probing system decreases significantly because
a probe sphere smaller than several dozen micrometers is adhesive to the surface.
To overcome this problem, the vibration system is introduced as shown in Fig. 9.
As a typical commercialized micro-probing system, a small-sized ultrasonic probe
is shown in Fig. 9a. The probe is driven by using a PZT (Lead Titanate Zirconate)
piezoelectric actuator to give axial vibration. Leach introduced the triskelion struc-
ture with six piezoelectric actuators (Fig. 9b) to give isotropic sensitivity in three
dimensions. The advanced vibration probe using carbon fiber of 5 mm in length
and 7 μm in diameter is developed by InsituTec Inc. in USA. This probe enables
to measure a high aspect structure.
The trend in an ultra-precision micro-probing system is advancing with the
development of a variety of novel principles such as the spherical probe based on
a tunneling current sensor (Weckenmann et al. 2008; Hoffmann et al. 2008) and the
CMM probe along with the terahertz-based imaging (Pfeifer 2009).
The principal specifications for leading micro-probing systems are listed in Table
2. The developed micro-probes are categorized into four types, that is, elastic hinge
type, SPM type, optical fiber type, and vibration type. The former two are charac-
terized by high resolution and the latter two by downsizing. To establish a micro-
probing system as the key device of a micro-CMM, there exist technological
problems that can be neglected in traditional probes. A short overview of specified
problems for a micro-probe is given here, that is, (a) probe size and sphericity,
(b) sticking, (c) probing force, and (d) isotropy in sensitivity.
(a) Probe Size and Sphericity

In the traditional CMM, considering that the probe diameter of 1 mm is adequate
to the measuring volume of 1 m, the probe size for a micro-CMM should be of
less than 10 μm in the measuring volume of 10 mm. The advantages of smaller
probe size are low uncertainty as well as accessibility. In order to decrease
measuring uncertainty results from the calibration of an effective diameter of
the probe sphere, it is important to use a smaller size probe and compensate
a probe vector more accurately as shown in Fig. 10. Namely, using a probe with
the diameter of r, if a probe vector is unknown with the difference in angle of
α (rad), the measurement error δe is given by
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
δe ¼ r 2ð1 cos αÞ (1)
In addition to these conditions, it is necessary to evaluate sphericity of a

micro-probe as precisely as possible. However, the problem is that the standard
measuring technique with an accuracy of 10 nm order is still in development.
(b) Sticking
It is a serious problem for features smaller than several dozen micrometers
that dominant force shifts from body force such as gravity and inertia to surface
force, namely, surface tension, friction, and adsorption, because of scale effect.
Therefore, even if the probe with size of several hundreds of micrometers can
achieve nanometer resolution, when the probe is downsized of less than several
dozen micrometers, the measurement resolution and the repeatability become
worse by unexpected behavior such as sticking, as shown in Fig. 11. This is
suggested by Table 2.
(c) Probing Force
The measurement uncertainty caused by the deformation under the probing force
at a contact can no longer be neglected specifically in micrometer regime. The
measured object is damaged by the plastic deformation; as a matter of course, the
elastic deformation is an error factor in position detection much greater than
nanometer order accuracy required for a micro-probe.
Fig. 9 Micro-probing system based on vibration system. (a) Mitutoyo probe (Hidaka and
Schellekens 2006). (b) NPL probe (Claverley and Leach 2009). (Courtesy of R.K. Leach).
(c) Insitu Tech (Bauza et al. 2005)
Fig. 10 Probe compensation Approaching
O´
r a δe
A
B
y
z x
Stick
Adhesive
Fig. 11 Sticking of probe to surface
The trade-off between high rigidity of the probe structure and small probing
force is a difficult problem to be solved because the higher rigid a probe is, the
higher the contact pressure becomes.
(d) Isotropy in Sensitivity
Three-dimensional anisotropic behavior in position sensing (Bos et al. 2007) is
the principal uncertainty factor that makes it difficult to achieve a high precision
micro-probe. For example, the three-dimensional complex elastic hinge struc-
ture of the micro-probe developed by METAS (Fig. 6h) is carefully designed
with considering isotropy in three dimensions.
Principle of the Laser Trapping Micro-probe
To meet the demand for coordinate metrology for a machined micro-scale component
with nano-tolerances, it is necessary to establish a micro-CMM which requires harsh
specifications for a micro-probe as listed in Table 3. Accordingly, the development of the
138
Table 2 Specifications of proposed micro-probe

Probe Measuring Measuring
Capability diameter Shaft length Resolution Repeatability Stiffness force range
K Takamasu 2D 1 mm <5 mm – 1.1 μm – 0.193 mN 0.4 mm
R. Furutani 3D – – 31 nm – – 1 mN –
NPL Hinge probe 3D 300 μm 15 mm 3 nm – 10 N/m 0.2 mN XY: 20 μm
Z: 20 μm
Εindhoven ΡΖT hinge 3D 500 μm 8 mm 1 nm – XY: 100 N/m <1 mN 25 μm
probe Z: 450 N/m
AIST 3D 400 μm 5 mm – 50 nm 200 N/m 200 μN –
METAS Rubby 3D 100 μm – – 5 nm 20 N/m 500 μN 25 μm
METAS Glass 3D 40 μm 250 μm – 5 nm 20 N/m 500 μN 25 μm
PTB Opt-tactile 2D 25 μm 0.1 μm 0.1 N/m a few μN 0.4 mm
PTB 3D boss 3D 300 μm 15 mm XY: 3 nm XY: <10 nm XY: 89 N/m XY: 0.5 μN –
Z: 5 nm Z: <20 nm Z: 55 kN/m Z: 0.3 mN
T. Oiwa 3D 1.2 mm 15 mm 0.5 μm 0.36 μm – 0.1 mN –
Y. Takaya Fiber probe 3D *in 8 μm – – 100 nm 24.2 μN/m – –
water
H. Murakami 2D 50 μm – 15 nm – – – 0.35 μm
H. Ji 3D 280 μm 5 mm 60 nm – – 0.1 μN –
PTB ACP 3D 50 μm 1 mm – 5 nm – – –
Mitutoyo 3D 30 μm 3 mm – 0.1 μm – 0.15 μN a few μm
NPL Triskelion 3D
Southern Taiwan University 3D 0.3 mm About 7 mm – 46 nm – 0.1 mN –
A. Weckenman 3D 0.3 mm – 1 nm 3 nm – – –
Y. Takaya and M. Michihata
Table 3 Required Probe diameter 10 μm

specifications for micro-
Shaft length 1 mm
probe
Resolution 1 nm
Measuring force <100 μΝ
Stiffness 10 N/m
Measuring range 10 μm
Fig. 12 Concept of laser

trapping probe Probe
Approaching
Micro parts
probing system is attempted with a different approach from the traditional concept for the
sake of scale effect that is remarkable with the use of a touch trigger probe. Therefore, the
basic concept of the laser trapping micro-probe was proposed (Takaya et al. 1999), whose
principle is based on the single-beam gradient-force optical trap (Ashkin et al. 1986) of a
micro-probe sphere in air. The laser trapping micro-probe can perform noncontact
measurement of a position that is optically sensed based on the displacement of the
micro-probe sphere given by dynamical interaction between the probe sphere and the
surface as indicated in (Fig. 12). The laser trapping micro-probe has two distinctive
features as following: one is the probe sphere size and the other is the high sensitivity. As
regards the probe sphere size, the optical trap technique has a high affinity with dynamical
controllability of a microsphere with the size (<25 μm) that is the most suitable for a
micro-CMM. On the other hand, the high sensitivity is the resultant from radiation force
with several hundreds of piconewtons (pN) that makes it possible to sense acutely the
dynamical interaction between a probe and an object’s surface.
Physical Basis of Optical Trap Technology
The light exerts a force on an object. This force is called optical radiation force
or light pressure. Its existence was predicted by Sir Isaac Newton (1642–1727) based
on the corpuscular theory more than 300 years ago. J. C. Maxwell (1831–1879) gave
it theoretical formula based on electromagnetic theory. The pressure exerted by
light was actually measured first in 1901 by P. N. Lebedev (1866–1912), and also
by Nichols and Hull. About 70 years after that, in 1970, the optical trap (optically
levitating small objects) was demonstrated first by A. Ashkin.
In this section, the dynamical theory about optical radiation force is presented as
the physical basis of the optical trap technique that is the fundamental to understand
the laser trapping micro-probe. After that, general characteristics in dynamical
properties of the laser trapping micro-probe are suggested.
Radiation Pressure Based on Electromagnetic Wave

The radiation pressure equals the energy density of the electromagnetic wave
as suggested by Maxwell. If the electric field E (the E-field), the magnetic field H
(the H-field), the electric flux density D, the magnetic flux density B, the energy
density per unit volume W is given by
ð ð
@D @B e μ
W ¼ E dtþ H dt ¼ E2 þ H2 (2)
@t @t 2 2
where e and μ are the permittivity and the electric permeability respectively. The first
item gives the energy density of the E-field and the second item the energy density of
the H-field. The energy flows of the light in the direction of the propagation is given
by the Poynting vector S.
S¼EH (3)
The Poynting vector is related to momentum of light. The energy conservation

law of electromagnetic field is given by
@W
þ∇S ¼ 0 (4)
@t
Therefore, the radiation pressure that equals W can be expressed in terms of
the magnitude of the Poynting vector. Inasmuch as the E- and H-fields are rapidly
varying, S is rapidly varying, so the light intensity per unit area I can be observed as
the average in an oscillating period hSi, which is expressed by
rffiffiffiffiffi
n e0
I ¼ hSi ¼ jE0 j2 (5)
2 μ0
where n is the refractive index and e0 and μ0 are the permittivity and the electric
permeability of free space, respectively. Thus, it is eminently practical to deal with
the average radiation pressure expressed by the light intensity. As noted above, the
light can give an object the dynamical action caused by radiation pressure. Namely,
in the case of an object such as a point-charge, in a dielectric body and a magnetic

body, the force can be derived directly. If the current density J, the electric polari-
zation density P, the magnetization density M, and the charge density ρ, ρare gicen,
the electromagnetic field exerts the force fe, fd and fm on a point-charge, a dielectric
body and a magnetic body respectively.
f e ¼ ρE þ J B (6a)
@P
f d ¼ P∇Eþ B (6b)
@t
f m ¼ μ0 M∇H þ μ0 Mð∇HÞ (6c)
The force fe is well known as Lorentz force. In these forces, especially the
force exerted on a dielectric body is strongly related to an optical trap technique.
So, dynamical properties that is implied in the relation of Eq. (6b) are considered in
detail. In Eq. (6b), the quantity @P/@t is the displacement current corresponding to
the current density in Eq. (6a), which is caused by migration of bound charge.
Therefore, from Eq. (6a),
@P
f d ¼ P∇Eþμ0 H (7)
@t
Using the relation between P and E given by P = e0 χ E,

@E
f d ¼ e0 χ E∇Eþμ0 H (8)
@t
where χ is electrical susceptibility. Transforming Eq. (8),

1 2 @
f d ¼ e0 χ ∇ E þ μ0 ðEHÞ (9)
2 @t
As shown in Eq. (9), the light exerts the force on a dielectric body, which can be
decomposed into two types of forces, namely gradient force and scattering force.
The former term, gradient force, comes from the gradient of E-field, that is, optical
potential. The latter term, scattering force, is caused by the change of the Poynting
vector in propagating direction, that is, momentum change of the light since momen-
tum of light is given by S/c2, where c is the speed of light.
Optical Force Based on Photon

The optical radiation force could be described also by employing the quantum
particle picture of light. Considering the light containing photons that each of
them carries energy, linear momentum, and angular momentum, a force per unit
area on an object results from the change in momentum of light. In the photon
picture, each constituent photon has an energy E given by the product of Planck’s
Constant h and the frequency υ of the radiation field:
E ¼ hυ (10)
Considering the relation between energy and momentum P, which is given by
hυ h
P¼ ¼ (11)
c λ
where λ is the wavelength. Its vector momentum P would be
P ¼ ℏk (12)
where k is the propagation vector and ℏ h/2π. The optical radiation force acts on
an object with changing momentum of a photon. So, as shown in Fig. 13, if a photon
travels through a dielectric sphere with a refractive index of n, the exerted force
vector f is given by
f ¼ nℏðk0 k0 Þ (13)
where k0 and k0 are the propagation vector of the photon into the dielectric sphere
and out from it, respectively. Further on, consider about the optical radiation force
if a laser beam with the light intensity per unit area I is incident to a dielectric sphere.
In this case, since the number of photons is I/hν, the incident momentum per second
of I in a medium of index of a refraction n is nI/c. So, introducing the dimensionless
factor Q from 0 to 2.0 of an effective index as a trapping efficiency, that is, a ratio of
the incident momentum per second of I and that is transferred to the force change of
the momentum (Ashkin 1992), the optical radiation force is given by
nI
f ¼Q (14)
c
Actually, the trapping efficiency Q is dependent of conditions of scattering and
reflection on an object’s surface. Then, as shown in Fig. 14, consider the forces if a
Fig. 13 Optical radiation

force driven by one photon
φ
Force, F
n
Fig. 14 Generation of optical Incident light Reflected light

radiation pressure. θ θ
(a) Reflection. (b) Refraction
n1
n2
Object A
(a)
FT
Incident light Reflected light
θ θ
n1
FR n2
ϕ
Object A
Refracted light
(b)
laser beam I is incident to a boundary of air (refractive index; n1) and dielectric
medium A (refractive index; n2). When the incident light with an incident angle of
θ is perfectly reflected at the boundary (Fig. 14a), that is, reflectance R = 1 and
transmittance T = 0, the force F is exerted to the dielectric object;
2n1 I
F5 cos θ (15)
c
In general (Fig. 14b), the optical radiation force F exerted to the dielectric object
is given by,
I
F ¼ ðn1 R cos θ þ n1 cos θ n1 T cos ϕÞ (16)
c
where of ϕ is refractive angle.
Principle of Laser Trap

The laser trap is a technique that enables to three-dimensionally sustain and manip-
ulate a micro-scale object by controlling optical radiation forces. The first optical
trap using the laser was realized by A. Ashkin in 1970 to demonstrate that laser
beams could controllably manipulate a particle of matter through the radiation force
(Ashkin 1970). In the past several decades, a wide variety of laser trap techniques
such as tweezers trap, levitation trap, standing wave trap, evanescent light trap,
surface plasmon trap, and optical fiber trap have been demonstrated or proposed as
shown in Fig. 15. Especially, the tweezers trap of micron-sized dielectric spheres
based on the principle of the single-beam gradient trap is useful to construct a micro-
probing system because of its simple optical system. As shown in Fig. 16, a micron-
sized dielectric sphere is optically trapped in air using a laser beam focused by an
objective. When the tightly focused laser light is incident to the sphere, optical
radiation forces are exerted on the boundary of air and dielectric medium. In this
case, a gradient force for each light ray acts on the sphere in the normal direction at
the boundary depending on the ratio of refractive index of air to dielectric medium.
Consequently, an axial trapping force, that is, a lifting force, is generated as the
resultant of these gradient forces. In free space, the micron-sized dielectric sphere is
suspended stably at a region near the incident laser focus by the dynamical balance
between the axial trapping force and the forces of gravity. So, it can be manipulated
to an arbitrary position in three-dimension by controlling the laser focus position.
Moreover, steep gradient of optical potential generated by the focused laser
light confines the sphere nearby the laser focus. It is deduced from these dynamic
properties that an optically trapped sphere can be modeled as a mass placed in a
three-dimensional harmonic potential. Therefore, the particle is accelerated in the
axial direction by the restoring force and shifts to the balance point along with the
optical axis.
Moreover, the dynamic property of an optically trapped sphere is also affected by
a polarization state of the focused laser light. This is because the optical radiation
Tweezers trap Levitation trap Standing wave trap
Evanescent light trap Surface plasmon trap Optical fiber trap
Fig. 15 Various types of the laser trapping technique

Fig. 16 Principle of laser trapping
force is decomposed into a gradient force exerted by the transmitted light power and
a scattering force by the reflected light power. A ratio of transmittance and reflec-
tance depends on the polarization state of the incident light, which is well known as
Fresnel reflection shown in Fig. 17. So, by changing the ratio of gradient forces and
scattering forces according to a polarization state of an incident light, the dynamic
balance in the laser trap of a micron-sized dielectric sphere can be controlled.
Namely, when a sphere is illuminated by laser light with linear polarization, both
p- and s-polarization are incident on the sphere surface as indicated in Fig. 18. In this
case, the dynamic balance that result from the ratio of gradient forces to scattering
forces changes with azimuthal angle. However, modulating a polarization state of
the laser light to radial polarization (Fig. 18), enhanced gradient forces brought by
isotropic p-polarization for azimuthal angle enable stiff and stable trap of a micron-
sized dielectric sphere.
Numerical Analysis of Dynamical Properties
The laser trapping micro-probe encounters the specific problems inherent in

an optically trapped microsphere, not to mention the unexpected and complicated
behaviors that are remarkable in micrometer regime because of scale effect, as given
the overview in the subsection of problems in the micro-probe system. Most of
those properties of the laser trapping micro-probe which could be revealed based on
specific dynamical properties in a probing mechanism results from optical radiation
force regime.
Fig. 17 Fresnel reflection (n = 1.44)
Linear pol. Radial pol.

Optical axis
Optical ray
Plane A
Plane A
Sphere
VIEW X
VIEW X
VIEW Y
VIEW Y
polarization direction
Fig. 18 Effect of polarization in laser trapping
In this section, the numerical simulations are employed based on the model of
a single-beam gradient force trapped microsphere to comprehensively understand
dynamical properties of the laser trapping micro-probe in probing such as
approaching an object and detecting position.
Ray Optics Model

The optical model for numerical analysis is performed based on the ray tracing
method proposed by A. Ashkin. In the ray optics regime, the total light beam is
decomposed into individual rays. Each ray has the characteristics of appropriate
intensity, direction, and state of polarization, which propagate in straight lines in
media of uniform refractive index and can change directions when it reflects,
refracts, and changes polarization at dielectric interfaces according to the usual
Fresnel formulas. This simulation method can also take into account the radial
polarization (Kawauchi et al. 2007) as well as polarization state of an incident
light such as linear polarization and circular polarization. In the ray tracing simula-
tions, ray optics is simply used to describe the scattering and optical momentum
transfer to a micro-scale sphere in the Mie size regime, where the diameter is large
compared with wavelength λ.
Consider first the force due to the scattering of a single incident ray by a dielectric
sphere. Figure 19 shows geometry of the ray tracing model for calculating the
scattering force and the gradient force. Taking initiate ray tracing as a single ray of
power P incident to a dielectric sphere at an angle of incidence θ with incident
momentum per second of n1P/c, where n1 is the refractive index of a sphere.
Assuming that the quantities R and T are the Fresnel reflectance and transmittance
of the surface at θ, the power of incident ray is shared with the reflected ray power
PR and infinite set of refracted rays power PT2Rn. Thus, the resultant force on the
sphere is summed over contributions due to the reflected ray of power PR and the
infinite number of emergent refracted rays of successively decreasing power PT2,
PT2R, . . . PT2Rn, . . .. The net force acting through the origin O can be broken into
Fig. 19 Optical path of PR

optical ray at arbitrary
incident angle (Ashkin 1992) q
P q
r a
PT2R2 α+β
q o
z β
PT2
n1
Y
n0
PT2R
n0: Refractive index of surrounding medium

n1: Refractive index of a sphere
components of FZ pointing in the direction of the ray and FY pointing in the direction
perpendicular to the ray, which are given by
" #
n1 P n1 PR X1
n1 P 2 n
FZ 5 cos ðπ 2θÞ þ T R cos ðα þ nβÞ (17a)
c c n¼0
c
" #
n1 PR X1
n1 P 2 n
F Y ¼ 0 sin ðπ2θÞ T R sin ðαþnβÞ (17b)
c n¼0
c
Considering the resultant force FT in the complex plane,
F T ¼ F Z þiF Y (18a)
n1 P n1 P n1 P 2 X1
FT ¼ ½1þR cos 2θ þ i R sin 2θ T Rn eiðαþnβÞ (18b)
c c c n¼0
Transforming Eq. (18b),

n1 P n1 P n1 P 2 iα 1
FT 5 ½1 þ R cos 2θþi R sin 2θ T e (19)
c c c 1 Reiβ
Using the geometric relations α = 2(θ r), β = π 2r, the real part and the
imaginary part are given by
F T ¼ F s þiF g (20a)

n1 P T 2 ½ cos 2ðθrÞ þ R cos 2θ
Fs ¼ 1þR cos 2θ (20b)
c 1 þ R2 þ2R cos 2r

n1 P T 2 ½ sin 2ðθ rÞþR sin 2θ
Fg ¼ R sin 2θ (20c)
c 1 þ R2 þ 2 R cos 2r
In Eq. (20b), the real part denotes the FZ component pointing in the direction
of the incident ray as the scattering force component Fs for a single ray. Similarly,
in Eq. (20c) the imaginary part denotes the FY component pointing in the direction
perpendicular to the ray as the gradient force component Fg for the ray. The trapping
forces on a sphere can be analyzed by calculating the vector sums of the scattering
and gradient force contributions of the individual rays of the beam.
(a) Axial Trapping Force

The simple ray optics model for analyzing the axial trapping forces on a
micro-scale sphere in the Mie size regime in single-beam gradient force trap
is illustrated in Fig. 20. Consider an incident beam is the highly convergent
Gaussian mode beam focused by using an objective lens with high numerical
aperture (NA), the intensity distribution is given by
Ray
Optical axis
w´ Optical axis Ray
rmax r
r rmax
β
n
θ
φ φmax φ
Fg
R dz
dz x w
O w
O Y φ
n1 Fs
n0
z Z
(a) (b)
n0 : Refractive index of surrounding medium

n1 : Refractive index of a sphere
Fig. 20 Ray optics model for analyzing the axial forces in single-beam gradient force trap. (a) Bird
view. (b) XZ plane
P ðrÞ ¼ P 0 e2r =rmax 2

2
(21)
When the input beam is divided into light rays of the number n and m in radial
and circumferential direction, respectively, the power of each ray which is
incident on the sphere at an angle ϕ with respect to the optical axis is given by
P 0 2r2 =rmax 2
e (22)
mn
In order to analyze dynamical properties when a sphere shifts along the

optical axis, the focus of the trapping beam is located along the optical axis at
a distance dz above the center of the sphere at O, as shown in Fig. 20a. A single
ray of the convergent beam striking the sphere at an angle θ on the XY plane is
depicted in Fig. 20b, where the direction of the scattering force component and
the gradient force component of the ray is expressed. Considering an axially-
symmetric plane-polarized input trapping beam, the axial force on the sphere can
be obtained by the integral along circumferential direction. The maximum
incident angle of the ray ϕmax is determined by
NA ¼ n0 sin ϕmax (23)
When the working distance of an objective lens is H, the rmax is given by

rmax ¼ H tan 1 ϕmax (24)
We consider the triangle connecting the three points of the laser spot,
the center of the sphere at O, and the incidence position of a ray on the sphere
surface, from the sine theorem,
R dz
5 (25)
sin ðπϕÞ sin θ
In Eq. (25) the incident angle θ of the ray hitting the sphere surface at an
angle ϕ. At this moment, the refracting angle r show in Fig. 19 can be obtained
using Snell’s law
sin r
5n1 (26)
sin θ
Here, the reflectance and the transmittance are changed by the polarization
and the incident angle. The quantities R and T are given by the Fresnel
reflectance and transmittance of the surface at θ. Then, consider the polariza-
tion of the ray of which a component of p- and s-polarization are fp and fs
respectively. Assuming linear polarization or radial polarization, fp and fs for
each are given by
fp ¼ sin 2 β (27a)
fs ¼ cos 2 β (27b)
fs ¼ 0 (28a)
fp ¼ 1 (28b)
Taking account of p- and s-polarization, the scattering force and the gradient
force for each ray are possible to be calculated. The force Fz pointing in the
direction of the optical axis for each ray consists of the scattering force and the
gradient force as follows:
F Z ¼ F s cos ϕ F p sin ϕ (29)
Therefore, the axial trapping force on the sphere is the sum of contributions
due to all rays.
(b) Transverse Trapping Force
We next examine the restoring forces for the case where a micro-scale sphere
is displaced in the direction of perpendicular to the optical axis. The ray optics
model for analyzing the transverse trapping forces on a micro-scale sphere in
the Mie size regime in single-beam gradient force trap is illustrated in Fig. 21.
The focus of the trapping beam is located transversely along the Y-axis of the
sphere at a distance dy from the center of the sphere at O, as shown in Fig. 21a.
Ray
Optical axis
W´
r rmax
β Ray
n
n
θ
θ
φ μ
γ R Fg R
dr O
Y
n1 n0 Y
dy O
Z Fs
(a) (b)
n 0 : Refractive index of surrounding medium

n 1 : Refractive index of a sphere
Fig. 21 Ray optics model for analyzing the transverse forces in single-beam gradient force trap.
(a) Bird view. (b) Ray incident plane
Since the model is not axial symmetry, consider an angle γ between an incident
ray and Y axis as indicated in Fig. 21b. Then, γ is geometrically given by
γ5 cos ðπϕÞ cos β (30)
From the sine theorem,
dy R
5 (31)
sin θ sin γ
The incident angle θ of the ray hitting the sphere surface can be obtained from
Eq. (31). The quantities R and T can be calculated using the Fresnel reflectance
and transmittance of the surface at θ. Here, we introduce the variable μ with
respect to the spherical coordinates to take account of the polarization of a ray.
The variable μ represents the angle between the incident plane and Z-Y plane as
indicated in Fig. 21a.
tan ðπϕÞ
μ¼ (32)
tan γ
The propagating direction of the light is given by the Poynting vector S.

Using the variable μ, we consider the polarization of the ray. The components
of p- and s-polarization are fp and fs, respectively. Assuming linear polarization
or radial polarization, fp and fs for each are given by
fp ¼ ð cos β sin μ sin β cos μÞ2 (33a)
fs ¼ ð cos β cos μ þ sin β sin μÞ2 (33b)
fp ¼ cos 2 μ (34a)
fs ¼ sin 2 μ (34b)
The forces FY pointing in the orthogonal direction to the optical axis for each
ray consists of the scattering force and the gradient force as follows:
F Y 5F s cos ϕþF g sin γ (35)
Therefore, the transverse trapping forces on the sphere is the sum of contri-
butions due to all rays.
Fundamental Properties
The dynamical properties of the laser trapping micro-probe such as the equilibrium
position, the axial and transverse trap strength, and the axial and transverse spring
constant are revealed to design and control it as a micro-probe for a micro-CMM.
An optically trapped silica particle (Refractive index; 1.44) with the diameter of
8 μm in air is employed as a probe sphere of the laser trapping micro-probe. Consider
first, the probe sphere which is suspended by means of the single-beam gradient trap at
the focal spot of the highly convergent Gaussian mode beam with linear polarization
focused by using an objective lens with NA of 0.95. The polarization direction is
orthogonal to Y-axis. In the ray optics model for the numerical analysis, the input
beam is divided into light rays of the number n = 300 and m = 400 in radial and
circumferential direction respectively. Both axial and transverse trapping efficiency
changing with a distance the focal spot and the center of the sphere along Z- and
Y-axis are calculated. Figure 22 depicts the trapping efficiency as a function of probe
sphere position. Our trapping efficiency curves were verified by the simulation results
of J. Bai (Bai et al. 1999) as shown in plots. As for an axial trapping force, steep
gradient of optical potential generated by the focused laser light confines a probe
sphere nearby the laser focus as indicated in Fig. 22a. The probe sphere is suspended
stably at a region near the incident laser focus by the dynamical balance between the
axial trapping force and the forces of gravity acting on the probe sphere of 5.25 pg in
mass. Whereas, a transverse trapping force has the stable equilibrium position on the
optical axis as plotted in Fig. 22b, at where the probe sphere is stably sustained. The
dynamic property of the probe sphere with shifting along Z- or Y-axis is based on these
force curves. In this dynamic situation, the probe sphere is accelerated toward the
Fig. 22 Trapping efficiency

as a function of probe sphere
position. (a) Axial trapping
efficiency. (b) Transverse
trapping efficiency
balance point by the restoring force. The axial trapping position of a probe sphere
depends on also an input laser power as shown in Fig. 23. Especially the laser power
exceeds 50 mW, the axial trapping position is drastically changed due to the decreas-
ing the gradient force. However, using the laser power higher than 200 mW, the
change in the axial trapping position is as small as 10 nm in 100 mW.
Measurement performance of a micro-CMM is dependent of probing character-
istics such as frequency response and sensitivity of detecting a position. When the
laser trapping micro-probe is approaching to a measured object for probing with the
speed of 1 mm/s, the probe sphere moves against Stokes’s dragging force of 1.4 pN.
Since the transverse trapping force is estimated 133 pN from the transverse trapping
efficiency Q = 0.2 with the laser power of 200 mW, the transverse trap strength is
high enough to be negligible to the effect of the viscous resistance which is only 1%
of the transverse trapping force.
Fig. 23 Equilibrium axial trapping position
The spring constant of a probe is the measure of sensitivity. The axial and
transverse spring constants of the laser trapping micro-probe with the laser power
of 200 mW are estimated as 63 μN/m and 64 μN/m, respectively. Since this is
far smaller than other probes and a cantilever for AFM (0.01–100 N/m), the laser
trapping micro-probe enables to achieve much higher sensitivity with responding to
the forces smaller than atomic force level. Besides, the resonance frequency of the
laser trapping micro-probe could be estimated from the spring constant. Namely, it
is estimated to be about 2700 Hz from the spring constant of 150 μN/m of the laser
trapping micro-probe with laser power of 500 mW. However, the fastest probing
speed of about 3 μm/s is the limit to attain a sensitivity of detecting a position of
1 nm. It is hard to avoid such a trade-off between high sensitivity and probing speed.
Trapping Beam Properties

The dynamical properties of the laser trapping micro-probe are principally domi-
nated by the optical characteristics of an input laser light for trapping a microsphere,
that is, trapping beam properties. We next examine those properties such as beam
condensing angle, beam profile, and beam polarization to envisage optimization for
the laser trapping micro-probe in the dynamic conditions.
(a) Beam Condensing Angle

When a laser light is highly focused by using an objective lens, the beam
condensing angle depends on the objective NA. The larger the NA is, the
more tightly the input beam is converged. To examine the influence of a trapping
condition by the objective NA, maximum trapping efficiency in axial and
transverse direction due to the change of NA is simulated as shown in Fig. 24.
Fig. 24 Maximum trapping efficiency changing with objective NA for trapping
It is confirmed that the NA of more than 0.75 is required to act an upward force
on a microsphere. When using the objective NA of 0.95, we can obtain the
maximum axial trapping efficiency up to 0.2. However, as the objective NA
increases, the axial trapping efficiency increases, on the contrary, the transverse
trapping efficiency decreases.
(b) Beam Profile
Since the balances of scattering force and gradient force depends on a light intensity
distribution on a microsphere, single beam gradient force trap condition is affected
by an input beam profile. In the case of the single-beam gradient trap with a high
objective NA as usual, an input Gaussian mode beam with very large spot size is
demanded to achieve a uniformly filled aperture in practice, which is wasteful of laser
power. That is to say, the beam profile of Gaussian mode has high power density
surrounding the optical axis, in which mainly scattering forces act on the microsphere
to push downward, so it is not effective for single beam gradient force trap.
We therefore consider the improvement of trapping efficiency for other cases of
Gaussian mode input beam profiles, as well as an annular beam of “do-nut” type
beam profiles which preferentially concentrate input light intensity distribution. For
the sake of simplicity, the cross-sectional intensity distribution is uniform in the
beam, and the input beam power is a constant. Figure 25 depicts the relation between
the ratio of an inner diameter to an outer diameter in an annular beam profile that is
defined as (Rmax Rin)/Rmax and maximum trapping efficiency in axial and trans-
verse direction. The effect of an annular beam is suggested that the axial trap
efficiency increases with the inner diameter of the annular beam, which reaches up
to two times at the maximum. The transverse trapping efficiency decreases by about
20% as the axial direction doubles.
Fig. 25 Influence of beam

intensity profile to trapping
efficiency
(c) Beam Polarization

In order to raise a trapping efficiency for the dynamically stable laser trapping
micro-probe, introducing radially polarized light as input beam is proposed. When
using an input beam with linear polarization, s- and p-polarized incident light
coexist on a microsphere. On the other hand, using an input beam with radial
polarization, only the p-polarized light is incident on a microsphere. In this case, an
axial trapping efficiency is increased due to the enhancement of the effect of
gradient force because the power of refracted p-polarized light is much higher
than s-polarized light. The simulation results of the comparison between radial
polarization and linear polarization in the trapping efficiency is shown in Fig. 26,
where the input beam profile is Gaussian mode. It is confirmed by Fig. 26a that the
maximum axial trapping efficiency is improved by 1.6 times by using the input
beam with radial polarization. Moreover, the axial spring constant also increases by
1.14 times. On the other hand, the maximum transverse trapping efficiency is
reduced 0.68 times as shown in Fig. 26b. This results from the conservation of
energy between axial and transverse trapping efficiency. In terms of the spring
constants, the difference between the axial spring constant and the transverse spring
constant becomes significantly small. This means that the extent of anisotropy in
the three-dimensional spring constants is estimated about 20% for linear polariza-
tion, which is improved 8% for radial polarization. The radially polarized trapping
beam has advantages for improving the laser trapping micro-probe, which can
achieve high trapping efficiency and three-dimensional isotropic sensitivity.
Experimental Analysis of Position Detection
In this section, we examine experimentally the feasibility of detecting the positions

of a micro-object using the laser trapping micro-probe with a vibration probing
mode. On the analogy of AFM with a contact mode and a tapping mode, the laser
trapping micro-probe has also a static probing mode and a vibration probing mode.
In the practical use, a static type probing mode has a lot of uncertainty factors
Fig. 26 Polarization effect to trapping efficiency. (a) Axial trapping efficiency. (b) Transverse
trapping efficiency
by unexpected behaviors remarkable in micrometer regime such as fragile, unstable,

adsorption, sticking, and low sensitivity. In order to achieve high sensitivity of
probing techniques for a micro-CMM, we developed optically vibrated micro-
probes, namely, an axial oscillation probe and a transverse oscillation probe.
Axial Oscillation Probe

At first, the method to drive an optically trapped microsphere with axial oscillation is
described here. A micro-probe sphere is forced to vibrate with several hundreds of
nanometer amplitude in the axial direction. The driving force of an optical radiation
force is generated by using a laser diode (LD), coaxially introduced with a trapping beam.
At the same time, the magnitude of optical radiation force is sinusoidally modulated by
changing input LD power to give the micro-probe sphere periodic displacement.
Next, we describe the method of measuring the displacement of a micro-probe
sphere which is oscillating in the axial direction. The principle of measurement is
based on detecting of backscattered light intensity from the micro-probe sphere,
which is sensed using a sensitive low-noise photodiode (PD) with fast response
frequency. Figure 27 depicts the configuration of the lens system for measuring axial
displacement of a micro-probe sphere. A tube lens is aligned so that an irradiation
area of backscattered light is larger than the sensitive area of a PD. The average light
intensity Pave of backscattered light detected by the PD is given by

Rdet 2
Pave ¼ Ptotal (36)
Rim
Fig. 27 Probe axial position Rim

measuring system Rdet Detector
Detector
plane
θ
L
ΔZ
Z Tube lens
Objective
z
Probe
Δz
where, Ptotal is the total light amount of the backscattered light, Rdet is the radius
of the sensitive area of the PD and Rim is the radius of the irradiation area of
backscattered light. Assuming that the divergence angle θ of backscattered light
does not change when the probe sphere moves in the axial direction,
Rim ¼ ðL Z Þ tan θ (37)
where, Z is the distance from the focal point of a tube lens to the light receiving
surface of the PD, and L is the distance from the tube lens to the image plane.
Further, if the probe sphere moves a distance Δz, and the image of the backscattered
light moves a distance ΔZ, the longitudinal and transverse magnifications of Ml
and Mt is given by
ΔZ ¼ M l Δz ¼ M t 2 Δz (38)
Therefore, using the above three formulas, the amplitude variations of power ΔPZ
due to the variations in the probe sphere position is given by
dP ave 2P ave M l
ΔP Z ¼ Δz ffi Δz (39)
dz LZ
Here, the amplitude of oscillating ΔPZ are proportional approximately to the
oscillating amplitude of the probe sphere. This method enables to measure an axial
displacement of probe sphere.
Schematic diagram of the optical system for the experimental analysis of position
detection using an axial oscillation probe is illustrated in Fig. 28. This optical system
is designed to compose the three elements: the laser trapping system, the probe
a
b
CCD Imaging optics
Eyepiece Detector
Dichroic Photodiode
minor
IRTF
LD driver Function 1
generator
Gaussian
Nomalized intensity
Illumination 0.8
LD (687nm) Nd:YAG(1064nm) 0.6

PBS2
0.4
PBSI
Obstruction
0.2 Annular
Objective
0
Piezo dirven 0 0.5 1 1.5 2 2.5 3 3.5 4
sample stage
Distance in cross-section [mm]
Fig. 28 Optical system for axially oscillated probe. (a) Optical system. (b) Intensity profile of
beam cross section
signal system for measuring axial displacement of a micro-probe sphere, and the
microscope for imaging it, as shown in Fig. 28a. A Nd: YAG laser (TEM00 mode,
wavelength: 1064 nm, polarization: linear) is employed to trap a silica particle
(refractive index: 1.44) with the diameter of 8 μm in air. In this experiment, the
Gaussian mode beam is obstructed at the center so as to form an annular beam to
improve the axial trap efficiency. As an obstacle filter, a glass plate on which the
center is deposited with a metal film disc is fabricated. The disc diameter of 1.6 mm
at the center of the input beam of 3.6 mm in diameter is obstructed. A micro-probe
sphere is optically trapped near the focus of an objective lens (Olympus, NA0.95,
focal length: 2.0 mm, working distance: 0.3 mm). The modulated LD (wavelength:
687 nm, power: 10 mW) illuminates the micro-probe sphere coaxially to give it axial
oscillation. The motion of the micro-probe sphere can be measured by detecting the
backscattered light of the trapping beam from the surface of it. An infrared trans-
mission filter is used to block the LD light scattered on the micro-probe sphere
surface. Backscattering light of the Nd: YAG laser from the micro-probe sphere
is detected by a PD after being reflected by a dichroic mirror. CCD camera takes the
image of the trapped probe sphere. The measured object is precisely positioned using
a Piezo-driven 3-axes stage.
To confirm an effect of annular beam, the axial trap efficiency is examined. Figure
28b shows the intensity distributions of both Gaussian mode and annular beam. The
power of the trapping beam is gradually decreased, and then the minimum laser
power is measured at the moment when the micro-probe sphere is released from trap.
The axial trap efficiency Qg is calculated using this measured minimum laser power
on the dynamical condition that the axial trapping force and the forces of gravity
acting on the probe sphere are equal as in Eq. (40), given by
4=3πr2 ðρm ρs Þg
Qg ¼ (40)
n1 P=c
where g is the gravitational acceleration, n1 is the refractive index of the surrounding

medium, r is the radius of the micro-probe sphere, and ρm and ρs are the densities of
the micro-probe sphere and the surrounding medium, respectively.
As a result, the minimum laser power required for trapping the micro-probe
sphere is 121 mW for the Gaussian mode beam and 64 mW for the annular beam.
The maximum axial trapping efficiencies are 0.013 and 0.025, respectively. It is
confirmed that the axial trap efficiency is improved by about two times by using the
annular beam.
The axial oscillating conditions are examined by measuring the variations in the
probe sphere position. Figure 29 depicts the measurement results of axial oscillation
of the micro-probe sphere when the power modulation of the LD is applied at 10 mW
and 86 Hz. The output signal of the PD in time-domain is low in the S/N ratio;
however, periodic variations are recognizable as indicated in Fig. 29a. In addition,
the oscillation frequency can be confirmed from the significant peak at 86 Hz in
frequency-domain signal by FFT analysis as shown in Fig. 29b. From these mea-
surement results, the fundamental properties of the axial oscillation probe could
Output [V] 0.6
0.5
0.4
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8
Time [Sec]
(a)
0.005
0.004
PSD [a..u.]
0.003
0.002
0.001
0
1 10 100 1000
Frequency [Hz]
(b)
Fig. 29 Measurement results of axial oscillation of the micro-probe sphere. (a) Time-domain.
(b) Frequency-domain
be verified by the oscillating probe driven by power modulation as well as the

quantitative sensing of axial oscillation. Since a peak magnitude of power spectrum
density (PSD) reflects the axial oscillation amplitude, it is usable to evaluate the axial
oscillation amplitude.
The position detection performance of the axial oscillation probe is evaluated.
Giving the laser trapping micro-probe vibration at the frequency of 500 Hz, the
changes in axial oscillation amplitude were measured when probing a standard glass
microsphere (Duke Scientific) which has the NIST traceable diameter of 168 μm
with a standard deviation of 8 μm. The measurement results of frequency-domain
signal in position sensing are shown in Fig. 30. Approaching the probe to the top
of the standard glass microsphere, an axial oscillation amplitude was evaluated by
the PSD peak with 25 nm interval. As shown in Fig. 30a, the value of PSD peak at
the frequency of 500 Hz attenuates in the vicinity of the surface of the standard glass
microsphere. When the probe moves 25 nm further from the traveling distance of
4150 nm, the PSD peak is completely buried in the noise, that is, the axial oscillation
amplitude is estimated approximately zero. In order to check the deterioration in
0.0025 0.0006 0.0006

PSD [a.u.]
PSD [a.u.]
PSD [a.u.]
0 0 0
400 500 600 400 500 600 400 500 600
Frequency [Hz] Frequency [Hz] Frequency [Hz]
Z=0 nm Z=4150 nm Z=4175 nm
0.0025 0.0006 0.0006

PSD [a.u.]
PSD [a.u.]
PSD [a.u.]
0 0 0
400 500 600 400 500 600 400 500 600
Frequency [Hz] Frequency [Hz] Frequency [Hz]
Z=0 nm Z=3150 nm Z=3175 nm
Fig. 30 Position sensing with axially oscillated probe. (a) On horizontal surface. (b) On 45 -tilted
surface
position detection sensitivity due to the tilted surface, the tangential plane with the
inclination angle 45 was measured. The measurement result is indicated in Fig. 30b,
where the PSD peak value drastically decreases when the probe is approaching in
the vicinity of the surface as same as the behavior shown in Fig. 30a. Consequently,
an axial oscillation probe was demonstrated to have a sensitivity of 25 nm or more in
position detection, and a vibration probing mode can achieve the improvement of
sensitivity.
Transverse Oscillation Probe

In the case of near-field scanning optical microscopy (NSOM), a photo-detection
probe is necessary to approach the surface in the vicinity of several hundred nm.
In order to find the surface, the photo-detection probe such as a fiber probe is given
transverse oscillation for sensing the share forces acting on the probe tip. On the
analogy of NSOM fiber probe, a transverse oscillation probe is one of feasible
methods to improve sensitivity of the laser trapping micro-probe.
At first, the method to drive an optically trapped microsphere with transverse
oscillation is described here. A micro-probe sphere is forced to vibrate with several
tens of nanometer amplitude in the transverse direction. Using the characteristic that
steep gradient of optical potential generated by the focused laser light confines
the sphere nearby the laser focus, an optically trapped microsphere is driven by
the lateral scanning of the focal position that is performed using an acousto-optic
deflector (AOD). The motion of an optically trapped micro-probe sphere in air is
governed by a dynamic mass–spring–damper system using the Langevin equation
(Takaya et al. 2004).
Next, we describe the method of measuring displacement of a micro-probe sphere
which is oscillating in the transverse direction. The principle of measurement is the
same as the axial oscillation probe, which is based on detection of backscattered light
intensity from the micro-probe sphere. As shown in Fig. 31, the configuration of the
lens system is the same as the axial oscillation probe; however, the optical alignment
is different from it. The horizontal adjustment of a PD is slightly shifted from the
optical axis in the oscillation direction so that half of the sensitive area of the PD
is illuminated by the image of the backscattered light based on the geometrical
arrangement of the sensitive area and the image of the backscattered light as
illustrated in Fig. 31a, b. The image of the backscattered light MtΔx shifts laterally
according to the lateral displacement of the micro-probe sphere Δx, where Mt is
microscope magnification. Assuming the sensitive area of 1.5 mm and the image of
2.5 mm in diameter, the light power on the sensitive area of PD as a function of the
lateral displacement of the micro-probe sphere is simulated based on the geometrical
relation as shown in Fig. 31c. The simulation results show that the detected light
power is proportional to the lateral displacement of the micro-probe sphere just
around the center of the sensitive area but out of linearity in the edge region. The
detected light power responds linearly in the range of actual oscillation amplitude
of 0.1 mm; so the measurement method is feasible enough to be applied to
a transverse oscillation probe.
X
Detector
Detector
plane
Detector
a x b
Amount of light received

1.5
1
Tube lens
0.5
0
–0.5 0 0.5
Position of backscattering light
from the probe sphere: x
Amount of light received
1.2
c
Objective
1.0
0.8
0.6
Trapped –0.2 –0.1 0 0.1 0.2
particle
Δx Position of backscattering light
from the probe sphere: x
Fig. 31 Probe transverse position measuring system
The optical system for the experimental analysis of position detection using
a transverse oscillation probe is revised on the basis of optical system illustrated
in Fig. 28. To optimize for a transverse oscillation probe, the optical systems for
trapping beam deflection and for sensing a position of an oscillating micro-probe
sphere are introduced as shown in the schematic diagram of Fig. 32. In order to
transversely vibrate the micro-probe sphere, the AOD designed for the wavelength
of the trapping beam of Nd: YAG laser light is used for changing incident angle to
the objective lens aperture. The relay lens system is incorporated into the trapping
beam optical system to make the optical conjugate arrangement of the deflection
Photo diode
Interference
filter
Eyepiece CCD
Half mirror
Imaging optics
Tube lens
Illumination
He-Ne (632.8nm)
YAG (1064nm)
Beam splitter f=256.4 f=62.0
PBS L2 L1
Two axis AOD
Objective
Piezo dirven
sample stage
Fig. 32 Optical system for transversely oscillated probe
center of the AOD and the back focal point of the objective lens for the precise
control of beam deflection. The AOD is designed to use Gaussian mode beam for the
highest efficiency, therefore the trapping beam is not the annular beam but a linear
polarized Gaussian mode beam in this experiment. As the light source for measuring
displacement of the oscillating micro-probe sphere, a He–Ne laser beam (wave-
length: 633 nm) is coaxially introduced with Nd: YAG laser of which scattered light
is cut using an interference filter to detect only the scattered He–Ne laser light from
the micro-probe sphere.
To examine transverse oscillation conditions, the variations in the probe sphere
position are measured. Figure 33 indicates the measurement results of transverse
oscillation of the micro-probe sphere when it is forced to vibrate with the frequency
of 50 Hz and the amplitude of 800 nm. The output signal of the PD in time-domain
is obtained at the sampling rate of 2.5 kHz as indicated in Fig. 33a. The regular
variations of the micro-probe sphere in position can be confirmed by an approximate
sinusoidal signal. In addition, the oscillation frequency is precisely controlled as
shown in the PSD plot of Fig. 33b in which the significant peak can be found at 50 Hz.
a
0.12
0.1
Output [V]
0.08
0.06
0.04
0.02
0
1.5 1.6 1.7 1.8 1.9 2
Time [Sec]
b 0.005
0.004
PSD [a.u.]
0.003
0.002
0.001
0
0 50 100 150 200 250
Frequency [Hz]
Fig. 33 Measurement results of transverse oscillation of the micro-probe sphere. (a) Time-domain
(b) Frequency-domain
From these measurement results, the fundamental properties of the transverse

oscillation probe could be verified by the oscillating probe driven by controlling the
trapping beam deflection as well as the quantitative sensing of transverse oscillation.
The amplitude of transverse oscillation can be evaluated by a peak magnitude of
power spectrum density (PSD).
In order to examine the feasibility of the transverse oscillation probe, the changes
in transverse oscillation amplitude were measured when probing a standard glass
microsphere (Duke Scientific) which has the NIST traceable diameter of 168 μm
with the standard deviation of 8 μm. The micro-probe sphere is forced to transversely
vibrate with the amplitude of 40 nm and the frequency of 100 Hz. Probing was
performed in the Z direction to the tangential plane with the inclination angle 60 on
the standard glass microsphere as depicted in Fig. 34a. Approaching the probe to the
surface of standard glass microsphere, a transverse oscillation amplitude was eval-
uated by the PSD peak with 25 nm interval. The measurement results of frequency-
domain signal in position sensing are shown in Fig. 34b. It was confirmed that the
PSD peak value gradually attenuated when the probe moves the traveling distance
from 4650 nm to 4900 nm. The PSD peak value reached the minimum in the region
Fig. 34 Position sensing with transversely oscillated probe. (a) CCD image of probing.
(b) Amplitude of probe oscillation
of 50 nm further from the traveling distance of 4850 nm, where the probe is
estimated to almost contact with the surface. In this way, it is demonstrated that a
transverse oscillation probe can detect position with a sensitivity of 25 nm or more
even for a steep angle surface and the sensitivity of a vibration probing mode was
confirmed to be high.
Development of Micro-CMM
The Structure of the Micro-CMM System
To realize micro-coordinate measurement using the laser trapping based micro-

probe, we specially designed and established the micro-CMM. The developed
micro-CMM consists of four elements – the probe system, the coordinate stage,
the base stage, and the control system. The illustration and photograph of the micro-
CMM is shown in Fig. 35. Figure 35a shows the structure of the micro-CMM.
The probe system is centered on the measuring system, which is attached on the
Z-axis stage. The probe tip, namely the optically trapped microsphere, is located at
the bottom of the probe system. The XYZ nano-coordinate stage is composed of the
three linear translation stages. The stage drive is a linear motor. Displacement of the
stages is monitored using the optical linear encoder with a resolution of 0.14 nm.
For travelling guide, the crossed roller bearing was adopted. The full stroke and the
maximum travelling speed of the stages are 20 mm and 10 mm/s, respectively.
These fundamental specifications of the coordinate stages are summarized in
Table 4. To support the coordinate stage and the probe system, the gate-type base
stage was designed. A sample to be measured is placed under the base stage, so
that the probe system is able to access the sample from the upper side (see Fig. 35a).
The base stage is made of carbon steel and its dimension is 210 mm (H) 1400 mm
(W) 500 mm (D), which is on the passive vibration isolation table.
Fig. 35 Developed micro-coordinate measuring machine equipped with the laser-trapping–based

micro-probe system. (a) Illustration (cross-sectional view). (b) Photograph
Table 4 Specifications of the micro-coordinate measuring machine

X-axis Y-axis Z-axis
Stroke 20 mm 20 mm 20 mm
Pitch 0.24 arcsec 0.28 arcsec 0.28 arcsec
Yaw 0.26 arcsec 0.22 arcsec 0.28 arcsec
Roll 0.18 arcsec 0.28 arcsec 0.28 arcsec
Straightness Horizontal 80 nm/40 mm 100 nm/40 mm 230 nm/40 mm
Vertical 150 nm/40 mm 130 nm/40 mm 180 nm/40 mm
Maximum traveling speed 10 mm/s 10 mm/s 10 mm/s
Resonant frequency 260 Hz 260 Hz 220 Hz
The control system includes the stage controller and the interpolators of the
optical linear encoders. The digitized signals of displacement of each translation
stage are sent to the control system for feedback control. The desktop computer also
receives these displacement signals of the coordinate stage as well as the probing
signal and the measured temperature. Whole signal flow and the structure of the
developed micro-CMM are depicted in Fig. 36. The desktop computer synchronizes
the probe system and the coordinate stage under LabVIEW software. The whole
measurement system, excluding the computer and the control system, is accommo-
dated in the temperature-controlled class 5000 clean booth. Environmental temper-
ature in the clean booth and several points on the micro-CMM is measured by the
calibrated thermistor thermometers with a resolution of 5 mK.
Coordinate Stage System
Positioning performances of the coordinate stage was verified using the calibrated
heterodyne interferometer with a resolution of 2.5 nm. The target retroreflector was
attached to the position equivalent to the probe tip. Each axis of the coordinate stage
was evaluated separately. To ensure stability of the interferometer and environmental
conditions, the measurements were conducted a few hours after adjusting the
Fig. 36 Schematic of data flow in the developed micro-coordinate measuring system
Table 5 Positioning performances of the coordinate stage

X-axis Y-axis Z-axis
Repeatability 3 nm 3 nm 9 nm
Accuracy 19 nm 28 nm 25 nm
Lost motion 16 nm 16 nm 13 nm
Thermal drift 1231 nm/ C 622 nm/ C 2087 nm/ C
alignment of the interferometer. The positioning performances of the coordinate

stage were evaluated in terms of positioning repeatability, accuracy, and lost motion.
Also thermal drift was measured. Positioning repeatability was defined as the
relative positioning error when the stage reaches the absolute origin while it shuttles
five times across a 10-mm distance. Positioning accuracy was evaluated as the error
between the actual coordinates measured by the interferometer and the commanded
coordinates when the stage moves five times in a step motion with a constant step
width. Lost motion was calculated as the positioning error between the forward and
backward motion. Thermal drift was measured from the gradient of the stage drift
when the environmental temperature changes after 6 h. The evaluated results are
summarized in Table 5. It can be observed that the positioning performances of the
coordinate stage were greater than 30 nm, which is accurate enough as a coordinate
stage of micro-CMMs. Since thermal drift was relatively high, the temperature
change during coordinate measurements should be carefully monitored.
Micro-probe System
The developed probe system is shown in Fig. 37. The optical system of the laser
trapping based micro-probe (Fig. 37a) was packaged compactly in the probe unit
to integrate the micro-CMM. Figure 37b shows the photograph of the probe unit.
Dimensions and mass of the probe unit are 230 mm (W) 450 mm (H) 140 mm (D)
and 13 kg, respectively. The optical system has mainly two functions: laser trapping and
detection of the trapped microsphere motion. A fiber laser was used as the light source
for laser trapping, whose wavelength is 1064 nm and maximum output power is 2 W.
This single continuous wave mode laser (TEM00) is collimated with the collimator lens
and then expanded such that it has the same diameter as the pupil of the objective on the
back surface by the relay lens system 1. After passing though the reflection prism, the
polarization conversion filter, and the harmonic separator, the trapping laser reaches the
objective (N.A. 0.95 and W.D. 0.3 mm). The polarization conversion filter converts
linear polarization of the beam into radial polarization in order to enhance the trapping
efficiency (Michihata et al. 2009). As a probe tip, a microsphere is trapped underneath
the objective. A silica sphere with a diameter of 8 μm was employed as the probe tip (see
Fig. 38), which is remarkably small. Diameters of conventional microbes are approxi-
mately 100 μm or larger (Weckenmann et al. 2004). A typical laser power for stable
trapping of the silica microsphere was 300 mW measured before the objective lens.
Horizontal position of the trapped microsphere can be controlled by deflection of the
focal spot of the trapping laser in the focal plane using the 2-axes acousto-optic deflector
Fig. 37 Laser-trapping–based micro-probe system. (a) Illustration of the optical system.

(b) Photograph
Fig. 38 Silica microspheres

used as the probe tip.
Its nominal diameter is
8 0.15 μm and sphericity is
better than 100 nm
(AOD). A designed resolution of position control of the microsphere is 15 nm. The laser
diode that illuminates the trapped microsphere coaxially with the trapping laser was used
as the second light source. The horizontal motion of the microsphere is measured from
the backscattered light of the laser diode (LD). Backscattered lights are collected by the
objective and then detected by the quadrant photodetector (QPD).
Measurement Properties
Experimental Analysis of Probing Property
The laser trapped microsphere was applied to the probe tip for the micro-CMM.
Advantages of the laser trapping based micro-probe are that a highly spherical, small
micro-glass particle is applicable and a measuring force is extremely small, which
means that the measured surface does not have damage due to probing. Therefore,
a stylus of the laser trapping based micro-probe is the glass microsphere that is
optically trapped in air by means of single beam gradient force optical trapping
(Fig. 39) (Ashkin et al. 1986). The probe tip is sinusoidally oscillating in the
horizontal plane (focal plane). When probing a surface, this oscillating microsphere
undergoes a dynamical interaction with the surface. The laser trapping based micro-
probe performs the position sensing of the surface by detecting the changes in the
oscillating motion. The laser trapped micro-probe is very soft, which means there is a
very low stiffness of the probe system, therefore high sensitivity can be realized.
First, the dynamics of the optically-trapped microsphere was theoretically ana-
lyzed. In single-beam gradient force trapping, the trapped microsphere can be dealt
with as a point mass. The trapped microsphere is maintained at the laser focal spot and
the force exerted on the microsphere acts as a spring force. When the trapped
microsphere is oscillating, the motion is damped by the viscosity resistance of the
surrounding air. Therefore, the transverse dynamic equation along a certain direction is
well described in terms of a mass–spring–damper system by the following equation:
Fig. 39 Laser-trapping–based micro-probe system. The white line arrow in (a) indicates the
position of the trapped microsphere where the scattered light of white light illumination can be
seen. (a) Optically trapped microsphere. (b) Trapped microsphere near surface
x þ Dx_ þ kðx A sin 2πftÞ ¼ F ðt Þ

m€ (41)
where m is the mass of the trapped microsphere, D is the viscous drag coefficient
of the surrounding medium, k is the spring constant, x is the lateral position of the
trapped microsphere, F(t) is the random force from air molecules, A is the oscillating
amplitude, and f is the oscillating frequency. The influence of the random force
F(t) on the microsphere is as small as negligible. A general solution to Eq. (41)
is assumed to have a harmonic form with an amplitude of X and a phase delay of φ,
given by
x ¼ X exp½ið2πft φÞ (42)

qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2
X ¼ Af n = f 2n f 2 þ ðDf =2πmÞ2
2
(43)
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
2 2
φ ¼ cos 1 f n f = f 2n f 2 þ ðDf =2πmÞ2
2
(44)
pffiffiffiffiffiffiffiffiffi
where f n ¼ 1=2π k=m is the resonant frequency of the probe system.

A position sensing of a sample surface with the laser trapping based micro-probe
is based on monitoring the probe’s oscillating conditions such as amplitude
and phase (Eqs. 42, 43, and 44). At far away from a surface, a fluid drag working
on the oscillating microsphere is considered as the Stokes’s law. Once the micro-
sphere approaches to be adjacent to the surface, the drag force working on the
microsphere becomes different from Stokes’s law. Now suppose that the trapped
microsphere is oscillating along with normal direction to surface. When the micro-
sphere is approaching to the surface, the air between the microsphere and the surface
is compressed, resulting in the resistance force which is also called the damping
force (Fig. 40a). When the microsphere is separating, the expanded air also generates
Fig. 40 Mechanism of damping force caused on the oscillating microsphere near surface.
Gray arrows indicate a direction of air flow. (a) Approaching. (b) Separating
the damping force (Fig. 40b). As a result, the oscillating microsphere near the surface
undergoes stronger resistance force than in free space, and then the amplitude of the
oscillating probe is decreased and the phase of the oscillation is delayed.
This damping phenomenon can be addressed analytically. The hydrodynamics
of a microsphere near a surface is well studied by J. Happel and H. Brenner.
They gave the analytical equations of a viscous drag coefficient in the case of
a microsphere oscillating either horizontally or vertically to a surface as shown in
following equations (Happel and Brenner 1983):
6πηr
Dk ¼
9 r
1 r
3 45 r
4 1 r
5
(45)
1 þ
16 h 8 h 256 h 16 h
" #
X
1
nðn þ 1Þ 2sinhð2n 1Þα þ ð2n 1Þsinh 2α
D⊥ ¼ 8πηr sinh α 1
n¼1
ð2n 1Þð2n þ 3Þ 4sinh2 ðn þ 1=2Þα ð2n þ 1Þ2 sinh2 α
(46)
where α = cosh1(h/r), h is the distance between the center of the microsphere

and the surface, η is the viscosity coefficient, and r is the radius of the microsphere.
D|| and D⊥ are the modified drag coefficient impinging on the microsphere oscillat-
ing parallel and perpendicular to the surface, respectively.
Figure 41a shows the calculated viscous drag coefficients using Eqs. (45)
and (46) as a function of the distance between the microsphere and the surface.
Figure 41b indicates the theoretical amplitude damping of oscillating microsphere
calculated using Eq. (43). From Fig. 41a, viscous drag coefficients drastically
increase when the microsphere comes a few micrometers closer to the surface.
Corresponding to these dragging force changes, the respond amplitudes of the
oscillating microsphere are decreased, particularly the stronger damping can be
seen in the case of oscillation perpendicular to the surface than parallel to.
Here only the amplitude change is shown but the phase of the oscillating micro-
sphere is also delayed near the surface with same tendency. These theoretical results
Fig. 41 Theoretical
calculations of the damping
behavior of oscillating
microsphere. (a) Viscos drag
coefficient. (b) Respond
amplitude
showed that the oscillating microsphere motion is damped due to the drag force near
the surface. In summary, the oscillating microsphere is damped near a surface,
resulting in the amplitude decrease and the phase delay. By setting an appropriate
threshold for the oscillating conditions of the trapped microsphere, the surface can be
detected high sensitively.
In order to confirm the surface detecting model, experiments were conducted.
Fundamental properties of the developed optical system were first checked. Figure
42a, b shows the signal detected from the QPD while the microsphere was oscillating
in either direction along X-axis or Y-axis with a frequency of 2000 Hz and an
amplitude of 300 nm. As shown in the figures, it was able to measure the sinusoidal
motion of the oscillating microsphere using the QPD although there was negligibly
small cross-talk. Then, the time domain signal was converted into the frequency
domain signal by means of the fast Fourier transform (FFT), as shown in Fig. 42c, d;
this proves that the motion of the microsphere can be controlled by the frequency.
Figure 42e, f shows the relation between the oscillating and the response amplitude
of the microsphere, which allows us to confirm the amplitude linearity of the probe
20 10
Signal [mV]
Signal [mV]
10 0
0
–10
–10
–20 –20
X-axis Y-axis X-axis Y-axis
–30 –30
0 2 4 6 8 10 0 2 4 6 8 10
Time [ms] Time [ms]
(a) (b)
1.E+03 1.E+03
PSD [a.u.]
PSD [a.u.]
1.E+01 1.E+01
1.E–01 1.E–01
1.E–03 1.E–03
1.E–05 1.E–05
0 1000 2000 3000 4000 5000 6000 7000 8000 0 1000 2000 3000 4000 5000 6000 7000 8000
Frequency [Hz] Frequency [Hz]
(c) (d)
40 45
Amp. output from QPD [mV]
Amp. output from QPD [mV]

35 40
X-axis
30 Y-axis 35 X-axis
Y-axis
30
25
25
20
20
15
15
10 10
5 5
0 0
0 100 200 300 400 500 600 700 800 0 100 200 300 400 500 600 700 800
Oscillating amplitude [mV] Oscillating amplitude [mV]
(e) (f)
Fig. 42 Fundamental performance of controlling and measuring the microsphere position.

(a) Microsphere displacement along with X-axis. (b) Microsphere displacement along with
Y-axis. (c) Frequency spectrum of (a). (d) Frequency spectrum of (b). (e) Amplitude linearity
along X-axis. (f) Amplitude linearity along Y-axis
system. The characteristics of the probe system are determined by the resonant
frequency of the laser trapped microsphere. The laser-trapping based probe has a
mass–spring–damper system, and the typical frequency responses for the amplitude
and the phase delay are shown in Fig. 43. The resonance frequency depends on the
laser trapping force, that is, the laser power. When the laser power was 450 mW, the
resonance frequency was approximately 2950 Hz that is corresponding to the spring
constant of 202 μN/m, which is super small if compared with conventional micro-
probe systems (Weckenmann et al. 2004). Typical stiffness of the micro-probe
system is of the order of 100 N/m. It is considered that the stiffness of the micro-
probe should be high enough because of detecting a contact with a surface sensi-
tively. However, the laser trapping based micro-probe detects a surface using near-
surface-air-damping, which requires a soft probe system. This pseudo contact probe
system has potential not only to realize high sensitivity, but also to measure soft
materials (Data were not shown in this article but in fact even the surface of liquid
such as oil can be detected).
Finally, it was examined for the surface sensing properties. Respond amplitude
and phase delay were measured during approach of the oscillating microsphere to the
Fig. 43 Frequency response of the oscillating microsphere. The black dot indicates the measured
value and red solid line indicates fitting curve using Eqs. (43) and (44). (a) Amplitude response.
(b) Phase delay
surface. The microsphere was oscillated in a direction of normal to the surface with
a frequency 1000 Hz and an amplitude of 200 nm. The resonant frequency was
1700 Hz. As-cleaved silicon wafer with a smooth surface (Ra < 1 nm) was used as
the sample surface. The trapped microsphere was approached step by step using the
coordinate stage until the microsphere had contact with the surface. The step width
was set to 200 nm initially and 50 nm near the surface. At each point, the amplitude
and the phase delay were measured five times. The measured results are shown in
Fig. 44 Properties of respond

amplitude and phase delay
during microsphere
approaching to surface.
(a) Amplitude response.
(b) Phase delay
Fig. 44. Both the respond amplitude and the phase delay were changing steeply near
the surface. Regarding the amplitude, the changing rate was 126 a.u./μm and
the variation of the measured value was 0.79 a.u., so that the sensing resolution of
6 nm (0.79 a.u./126 a.u./μm) could be estimated. About the phase delay, the sensing
resolution of 22 nm (0.55 /25 /μm) could be obtained by same procedures as the
case of the amplitude. These results insist that the laser trapping based micro-probe
shows sensitivity as high as approximately 10 nm.
Detection of Position and Normal Direction
The dragging force working on the oscillating microsphere changes remarkably due to
presence of a surface. As already described, changing tendency of the dragging force
is different with respect to the oscillating direction of the microsphere against the
surface. Careful analysis of the amplitude damping behaviors made it possible to not
only a position but also a normal direction of the surface, which is a very advantageous
function for the probe system of a CMM. Conventionally, the orientation of surface is
recognized from neighboring three-point detection. Here it was proposed that the
detection method of a position and a normal direction of a surface with single sensing.
First, we examined the dependence of the viscous drag forces on the oscillating
detection. One-dimensional oscillation either perpendicular or parallel to the surface
was given to the trapped microsphere. As-cleaved silicon wafer with a smooth
surface was used as the surface. At different distances from the surface, the viscous
drag coefficient was measured. The viscous drag coefficient was calculated by fitting
Eq. (43) into the experimentally obtained frequency-amplitude response as shown in
Fig. 43. The viscous drag coefficient as a function of the microsphere-surface
distance is shown in Fig. 45a. The obtained viscous drag coefficients were well
agreed with one from the theoretical analysis in Fig. 41a. In a few micrometers
before the microsphere contacts with the surface, the perpendicularly oscillating
microsphere was damped more than the parallel oscillating one. The amplitude of the
microsphere, while probing the silicon surface, is shown in Fig. 45b. As following
the viscous drag force increases, the amplitude of the perpendicularly oscillating
microsphere was damped more than the parallel-oscillating microsphere, as esti-
mated in theoretical model.
Fig. 45 Experimentally
measured oscillated probe
dynamics. (a) Viscos drag
coefficient. (b) Respond
amplitude
Using these anisotropic properties of damping of oscillating microsphere, surface

detecting probe that has a function of detecting both the position and the normal
direction of a surface was developed. The detection is executed based on monitoring
the orbit of circularly rotated microsphere motion, which is so called circular motion
probe. Figure 46 illustrates the concept used for detection using the circular motion
probe. Circular motion in the plane perpendicular to the optical axis (XY plane) is
provided to the trapped microsphere. When the circular motion probe approaches
a surface, the orbit of the microsphere motion is compressed perpendicular to
the surface, resulting in an elliptical orbit. Then, the major and minor axes of the
elliptical orbit are determined. The length of the minor axis, Amin, decreases with
the increasing distance between the microsphere and the surface due to compression
of the ellipse. The position of the surface is detected by monitoring the length of the
minor axis. The direction of the minor axis corresponds to the direction of the normal
direction to the surface. Therefore, the normal direction to the surface is detected by
tracking the rotated angle with respect to the minor axis.
Performances of the circular motion probe was experimentally examined. It
was estimated that the circular motion of the microsphere was compressed into
Probe tip behavior Probe signal from QPD

y x Ax
Microsphere Decomposition of the signal t

along X-axis and Y-axis
j=90 deg.
y AY
x
t
Approach
x Ax
t
Minor axis angle y j AY
y’ q
Normal direction x’ t
of surface
Rotating the coordinate system
x x 1 2Ax Ay cos
Minor axis =R = tan
length Amin y y 2 Ax2 A2y
R : Rotating matrix ,
Ax
y x’
t
y’ ,
Ay = Amin
z x t
Sample
Fig. 46 Concept of the circular motion probe system detecting both a position and a normal
direction of a surface
Fig. 47 CCD image of

sensing an inclined surface
Fig. 48 Measured orbit of the circular motion probe approaching the surface
then elliptical motion near the surface. The circularly oscillating microsphere was
approached toward the surface step by step. Figure 47 shows microscopic image
taken with CCD when sensing surface. Figure 48 shows that the motion trajectory of
the circular motion probe at the distance between the microsphere and the surface is
Fig. 49 Performance of circular motion probe. (a) Surface sensing. (b) Measurement of normal
direction of surface
4 μm (near contact), 5 μm, 7 μm, and far from the surface where a radius of the
microsphere was 4 μm. The microsphere was approaching in direction of minus
X-axis, and the sample surface was parallel to the Y-axis, that is, when the exper-
imental result shown in Fig. 49 was conducted, the surface inclination angle (α) was
zero. From the results, when the circular motion probe approaches the surface, the
orbit of the oscillating microsphere obviously becomes an ellipse compressed
perpendicular to the surface. It was also found that the direction of the minor axis
on the elliptical motion was perpendicular to the surface, as expected.
The circular motion probe detects both the position and the normal direction of
a surface based on the change in the orbit of the microsphere motion. The important
parameter for position detection was the minor-axis length of the elliptical orbit. To
investigate the resolution of the position sensing, experiments that the microsphere
approaches to the surface horizontally were carried out. An oscillating frequency
of 1800 Hz and an amplitude of 500 nm were used as the oscillation conditions. For
each plot, data were taken 50 times and the mean value and the standard deviation
were calculated. The plots were taken at intervals of 50 nm. As-cleaved silicon
wafer with a smooth surface was used as the surface. The sample surface was aligned
parallel to the Y-axis of the absolute coordinate of the system. And the microsphere
approached along with the X-axis. Figure 49a shows the minor-axis length as a
function of the distance between the microsphere and the surface. The minor-axis
length decreased as the microsphere approached the surface. It should be noted that
the decreasing change is approximately linear at a few micrometers (indicated by the
straight line in Fig. 49a) before the contact. The gradient indicates the sensitivity of
the probe system for position-detection. The minor-axis length decreased to 57 nm
with an approach distance of 1 μm. Thus, the circular motion probe performs with
a sensitivity of 0.018 μm/nm, and the standard deviation of the minor-axis length
was 1.1 nm. Therefore, a resolution of the circular motion probe was estimated to be
approximately 39 nm. When compared with the result in Fig. 44, the resolution was
six times larger. The reason of which was that the compressing pressure of the air
between the microsphere and the surface was weaker in circular motion probe.
Table 6 Measurement result of direction of surface using circular motion probe

No. 1 No. 2 No. 3 No. 4 No. 5
Inclined angle measured by CCD image 1 12 23 32 43
Measured angle 2 12 23 31 40
Difference 1 0 0 1 3
As explained, the circular motion probe can be used to measure the normal
direction of the surface. The measurement performance was evaluated by using the
surface inclined at various angles – 1 , 12 , 23 , 32 , and 43 – which are measured
by image processing using CCD images shown in Fig. 47 as references value. The
probing conditions of amplitude and frequency of the oscillation were 500 nm and
1800 Hz, respectively, which were used for both axes. Figure 47b shows the results
of tracking the minor-axis angle while the microsphere approached the inclined
surfaces. The horizontal axis was the distance between the microsphere surface and
the sample surface. The data were acquired at intervals of 50 nm. Table 6 summa-
rizes the measurement results. For each inclined surface, the differences from the
reference were within 3 . It is thus experimentally proved that the circular motion
probe performs measurements of the normal direction of the surface with an accu-
racy of 3 . In summary, the current system of the circular motion probe has the
ability to detect the surface with a resolution of 39 nm and to measure the normal
direction with an accuracy of 3 .
Coordinate Measurement of a Micro-glass Sphere
Coordinate measurement was conducted using the developed laser trapping

based micro-probe. Firstly, the diameter traceable standard sphere was measured to
evaluate performance of the three-dimensional coordinate measurement. Secondly,
an equatorial plane of another standard sphere was measured to test the functionality
of measuring both a position and a normal direction of surface.
As a reference sample, the NIST traceable standard sphere, which is made of soda
lime glass, was used. Its nominal diameter and standard deviation are 553 μm and
29 μm, respectively. Probe oscillating conditions were amplitude of 300 nm and
a frequency of 1500 Hz. The resonant frequency was 2300 Hz. The threshold value
for detection was set to 50% of the initial respond amplitude. The probing direction
was along with Z-axis with an approaching speed of 5 μm/s. Measurements were
conducted at different, 509, points. After measurement, the microsphere radius
of 4 μm was compensated. The measurement result is shown in Fig. 50a. Total
measuring time was 3 h. During the measurement, the temperature deviation was
less than 0.29 C. The measured diameter of the sphere was 565.47 μm by means of
least square method, which was well agreed with the certificated value of the sphere.
A certain YZ cross- sectional data is shown in Fig. 50b. The differences of the fitting
curve and measured data were around 5 μm. In experiments, it was found that dusts
on the measured sphere scatters the light of the LD. The scattered light comes to the
objective, which influences the probing signal. Since the QPD detects the
Fig. 50 Three-dimensional coordinate measurement of standard microsphere. (a) Whole measured

data. (b) One cross section
backscattered light intensity of the LD, the disturbed probing signal disrupted the
surface detection. Without these outliers, uncertainty analysis reveals that the
expanded measurement uncertainty of our coordinate measurement as better than
600 nm. Main factor of the uncertainty was thermal drift.
Using the circular motion probe, the position and the normal angle of surfaces were
measured. As a sample, the NIST traceable standard sphere with a diameter of
168 8.4 μm was used. Probe oscillating conditions were an amplitude of 512 nm
Fig. 51 Measurement of an equatorial plane of standard microsphere using circular motion probe.
(a) Position of surface. (b) Normal direction of surface
and a frequency of 2000 Hz. The resonant frequency was 2000 Hz. The threshold value
to detection was set to 64 nm for the respond amplitude. Probing was in direction of Y-
axis. An interval of measurement on X-axis was 10 μm. At each point, surface detection
was repeated 11 times. The measurement result is shown in Fig. 51. Figure 51a shows
the measured position of the surface and Fig. 51b shows the normal direction. The solid
black lines means the regression curve. The repeatability of the surface sensing was
better than 0.58 μm. This relatively worse repeatability was because the probing angle
was too shallow at 80 μm and 80 μm on X-axis. Except these points, differences from
the regression curve were within 1 μm for a position and 3 for a normal angle.
These results proofed feasibility of the circular motion probe. The circular motion probe
only measures two-dimensional normal direction of a surface. To extend it to the three-
dimensional coordinate measurement in future, three-dimensionally fluctuated
Brownian motion of the trapped microsphere is promising (Michihata et al. 2010).
References
Alting L, Kimura F, Hansen HN, Bissacco G (2003) Micro engineering. CIRP Ann Manuf Technol
52(2):635–657
Ashkin A (1970) Acceleration and trapping of particles by radiation pressure. Phys Rev Lett 24(4):
156–159
Ashkin A (1992) Forces of a single-beam gradient laser trap on a dielectric sphere in the ray optics
regime. Biophys J 61:596–582
Ashkin A, Dziedzic JM, Bjorkholm JE, Chu S (1986) Observation of a single-beam gradient force
optical trap for dielectric particles. Opt Lett 11:288–290
Bai J, Miyoshi T, Takaya T, Takahashi S (1999) Computer simulation for laser trapping on micro-
particles with arbitrary shape. Int J Jpn Soc Precis Eng 33:363–368
Bartscher M, Hilpert U, Goebbels J, Weidemann G (2007) Enhancement and proof of accuracy of
industrial computed tomography (CT) measurements. CIRP Ann Manuf Technol 56(1):495–498
Bauza MB, Hocken RJ, Smith ST, Woody SC (2005) Development of a virtual probe tip with an
application to high aspect ratio microscale features. Rev Sci Instrum 76:095112
Becker H, Heim U (2000) Hot embossing as a method for the fabrication of polymer high aspect
ratio structures. Sens Actuators A Phys 83:130–135
Becker EW, Ehrfeld W, Hagmann P, Maner A, Miinchmeyer D (1986) Fabrication of microstruc-
tures with high aspect ratios and great structural heights by synchrotron radiation lithography,
galvanoforming, and plastic moulding (LIGA process). Microelectron Eng 4:35–56
Bos EJC, Delbressine FLM, Haitjema H (1998) High-accuracy CMM metrology for micro systems.
In: Proceedings of the 8th IMEKO world congress, pp 8–15
Bos EJC, Heldens RWP, Delbressine FLM, Schellekens PHJ, Dietzel A (2007) Compensation of the
anisotropic behavior of single crystalline silicon in a 3D tactile sensor. Sens Actuators A Phys
134:374–381
Brand U, Kirchhoff J (2005) A micro-CMM with metrology frame for low uncertainty measure-
ments. Meas Sci Technol 16:2489–2497
Cao S, Brand U, Kleine-Besten T, Hoffmann W, Schwenke H, Butefisch S, Buttgenbach S (2002)
Recent developments in dimensional metrology for microsystem components. Microsyst
Technol 8:3–6
Carl Zeiss Inc. (2009) UPMC series. http://www.zeiss.com. Accessed 31 Aug 2009
Carl Zeiss Inc. (2018) A micro-component for medical instruments. http://www.zeiss.com.
Accessed 31 Mar 2018
Carmignato S, Dreossi D, Mancini L, Marinello F (2009) Testing of x-ray microtomography
systems using a traceable geometrical standard. Meas Sci Technol 20:084021
Claverley JD, Leach RK (2009) A vibrating micro-scale CMM probe for measuring high aspect
ratio structures. In: Proceedings of the HARMST
Dai G, Wolff H, Pohlenz F, Danzebrink HU, Wilkening G (2006) Atomic force probe for sidewall
scanning of nano- and microstructures. Appl Phys Lett 88:171908
Dai G, Pohlenz F, Butefisch S, Hartig F, Danzebrink HU (2008) Measurements of micro gear using
a nano CMM. In: Proceedings of the EUSPEN, vol 2, pp 366–370
Dai G, Butefisch S, Pohlenz F, Danzebrink HU (2009) A high precision micro/nano CMM using
piezoresistive tactile probes. Meas Sci Technol 20:084001
Danzebrink HU, Dai G, Pohlenz F, Wilkening G (2005) Overview of the metrological scanning
probe microscopes at PTB. Microsc Microanal 11:2–5
Danzl R, Helmli F, Scherer S (2009) Focus variation -a new technology for high resolution optical
3D surface metrology. In: The 10th international conference of the Slovenian Society for Non-
Destructive Testing
De Chiffre L, Hansen HN, Kofod N (1999) Surface topography characterization using an atomic
force microscope mounted on a coordinate measuring machine. CIRP Ann Manuf Technol
48(1):463–466
Dornfeld D, Min S, Takeuchi Y (2006) Recent advances in mechanical micromachining. CIRP Ann
Manuf Technol 55(2):245–768
Eom SI, Takaya Y, Hayashi T (2009) Novel contact probing method using single fiber optical
trapping probe. Precis Eng 33(2):235–242
Fan KC, Fei YT, Yu XF, Chen YJ, Wang WL, Chen F, Liu YS (2006) Development of a low-cost
micro-CMM for 3D micro/nano measurements. Meas Sci Technol 17:524–532
Fan KC, Fei Y, Yu X, Wang W, Chen Y (2007) Study of a noncontact type micro-CMM with arch-
bridge and nanopositioning stages. Robot Comput Integr Manuf 23:276–284
Furutani R, Takamoto T, Younf-Won C (2001) Development of high sensitive probe using optical
displacement detector (1st report) – estimation of capability of sensitivity of planar displace-
ment. J Jpn Soc Precis Eng 67(10):1670–1674
Garnaes J, Hansen PE, Agersnap N, Holm J, Borsetto F, Kühle A (2006) Profiles of a high-aspect-
ratio grating determined by spectroscopic scatterometry and atomic-force microscopy. Appl Opt
45:3201–3212
Goda J, Mitsui K (2008) Measuring method for shape and dimension of micro-components. J Jpn
Soc Precis Eng 74(3):226–229
Haitjema H, Pril WO, Schellekens PHJ (2001) Development of a silicon-based nanoprobe system
for 3-D measurements. CIRP Ann Manuf Technol 50(1):365–368
Hansen HN, Carneiro K, Haitjema H, De Chiffre L (2006) Dimensional micro and nano metrology.
CIRP Ann Manuf Technol 55(2):721–743
Happel J, Brenner H (1983) Low Reynolds number hydrodynamics: with special applications to
particulate media. Springer, Netherlands
Heckele M, Schomburg WK (2004) Review on micro molding of thermoplastic polymers.
J Micromech Microeng 14:R1–R14
Hidaka K, Schellekens PHJ (2006) Study of a small-sized ultrasonic probe. CIRP Ann Manuf
Technol 55(1):567–570
Hidaka K, Saito A, Koga S (2003) Study of a micro-roughness probe with ultrasonic sensor. CIRP
Ann Manuf Technol 57(1):489–492
Hoffmann J, Weckenmann A, Sun Z (2008) Electrical probing for dimensional micro metrology.
CIRP J Manuf Sci Technol 1:59–62
Impress corporation (2018) A micro-lens for DVD. https://av.watch.impress.co.jp/docs/20011011/
hitachi.htm. Accessed 31 Mar 2018
Ji H, Hsu HY, Knog LX, Wedding AB (2009) Development of a contact probe incorporating
a Bragg grating strain sensor for nano coordinate measuring machines. Meas Sci Technol
20:095304
Jordan HJ, Wegner M, Tiziani H (1998) Highly accurate non-contact characterization of engineer-
ing surfaces using confocal microscopy. Meas Sci Technol 9:1142–1151
Kajihara Y, Inazuki Y, Takahashi S, Takamasu K (2007) Study of nano-stereolithography using
evanescent light (2nd report): photofabrication of fine lattice structures using standing evanes-
cent light. J Jpn Soc Precis Eng 73(8):934–939
Kawata S, Sun HB, Tanaka T, Takada K (2001) Finer features for functional microdevice. Nature
412(6848):697–698
Kawauchi H, Yonezawa K, Kozawa Y, Sato S (2007) Calculation of optical trapping forces on a
dielectric sphere in the ray optics regime produced by a radially polarized laser beam. Opt Lett
32:1839–1841
Kim SW (2001) New design of precision CMM based upon volumetric phase-measuring interfer-
ometry. CIRP Ann Manuf Technol 50(1):357–360
Kim SW, Rhee HG, Chu JY (2003) Volumetric phase-measuring interferometer for three-dimen-
sional coordinate metrology. Precis Eng 27:205–215
Kirkland E, Kurfess TR, Liang SY (2004) A optical coordinate measuring machine for nanoscale
dimensional metrology. J Adv Comput Intell Intell Inform 8(1):39–44
Komatsuseiki Kosakusho.Co.,Ltd (2018) A micro-component for medical instruments. http://www.
komatsuseiki.co.jp. Accessed 31 Mar 2018
Kung A, Meli F (2008) Versatile probes for the METAS 3D micro-CMM. In: Proceedings of the
EUSPEN 2008, vol 2, pp 338–342
Kung A, Meli F, Thalmann R (2007) Ultraprecision micro-CMM using a low force 3D touch probe.
Meas Sci Technol 18:319–327
Kunieda M, Lauwers B, Rajurkar KP, Schumacher BM (2005) Advancing EDM through funda-
mental insight into the process. CIRP Ann Manuf Technol 54(2):64–87
Kuriyagawa T (2009) Future technologies of nano-precision micro/meso mechanical manufactur-
ing. J Jpn Soc Precis Eng 75(1):62–63
Leach RK, Giusca C (2008) Traceable measurement of areal surface texture. In: Proceedings of the
4M conference
Leach R, Haycocks J, Jackson K, Lewis A, Oldfield A, Yacoot A (2001) Advances in traceable
nanometrology at the National Physical Laboratory. Nanotechnology 12:R1–R6
Leach RK, Murphy J, Wilson A (2004) Design of a co-ordinate measuring probing system for
characterising three-dimensional micro-structures. NPL report, UK CBTLM 30
Lonardy PM, Lucca DA, De Chiffre L (2003) Emerging trends in surface metrology. CIRP Ann
Maruo S, Ikuta K (2000) Submicron manipulation tools driven by light in a liquid. Appl Phys Lett
76(19):2656–2658
Masuzawa T (2009) Electrical discharge machining. J Jpn Soc Precis Eng 75(1):68–69
Masuzawa T, Kim BJ, Bergaud C, Fujino M (1997) Twin-probe vibroscanning method for
dimensional measurement of microholes. CIRP Ann Manuf Technol 46(1):437–440
Meli F, Kung A (2004) Performance of a low force 3D touch probe on an ultra precision CMM for
small parts. In: Proceedings of the EUSPEN, vol 2004, pp 270–271
Michihata M, Hayashi H, Takaya Y (2009) Measurement of vertical and lateral trapping stiffness
of optical tweezers in air using radially polarized beam. Appl Opt 48(32):6143–6151
Michihata M, Takaya Y, Hayashi H (2010) Laser trapping based surface sensing probe using Brownian
motion for nano-coordinate measurement. In: Proceedings of the 14th international conference on
mechatronics technology, Osaka University, Japan, 24–26 November 2010, pp 313–316
Mohri N (2005) New development in electrical discharge machining technology (Special Issue
Novel development of electrical discharge machining). J Jpn Soc Precis Eng 71(5):537–540
Moriwaki T, Shamoto E (1995) Ultrasonic elliptical vibration cutting. CIRP Ann Manuf Technol
44(1):31–34
Murakami H, Katsuki A, Onikura H, Sajima T (2009) Development of a micro hole measuring
system using an optical fiber probe – evaluation of a fabricated apparatus and evaluative
experiment. In: Proceedings of 2009 JSPE general meeting in spring, pp 35–36
Murakami H, Katsuki A, Sajima T, Uchiyama K (2017) Fabrication of ultra-small diameter optical-
fiber probe using acid-etch technique and CO2 laser for 3D-micro metrology. Int J Autom
Technol 11(5):699–706
Muralikrishnan B, Stone JA, Stoup JR (2006) Fiber deflection probe for small hole metrology.
Precis Eng 30:154–164
MUSASHI Corporation (2018) An aluminum product by 5-axis machining. http://www.musashi-
co.co.jp/com.html. Accessed 31 Mar 2018
Nishino H, Miyoshi T, Takaya Y, Takahashi S, Hayashi T, Kimura K (2003) Study on nonlaminate
micro stereolithography using LCD mask (2nd report): thin layer laminating fabrication using
LCD live-motion mask. J Jpn Soc Precis Eng 69(10):1417–1422
Ogura I, Okazaki Y (2009) Development for micro measurement probe for micro factory.
In: Proceedings of 2009 JSPE general meeting in spring, pp 781–782
Oiwa T, Tanaka T (2005) Miniaturized three-dimensional touch trigger probe using optical fibre
bundle. Meas Sci Technol 16:1574–1581
Peggs GN, Lewis AJ, Oldfield S (1999) Design for a compact high-accuracy CMM. CIRP Ann
Manuf Technol 48:417–420
Peiner E, Balke M, Doering L, Brand U (2008) Tactile probes for dimensional metrology with
microcomponents at nanometre resolution. Meas Sci Technol 19:064001
Pfeifer T (2009) Terahertz based imaging for inspection and spectroscopic analysis. In: Plenary talk
in the 10th IMEKO world congress
Pfeifer T, Freudenberg R, Dussler G, Broccher B (2001) Quality control and process observation for
the micro assembly process. Measurement 30(1):1–18
Plunkett Associates Ltd (2018) High resolution stereolithography. https://www.plunkettassociates.
co.uk. Accessed 31 Mar 2018
Savio E, De Chiffre L, Schmitt R (2007a) Metrology of freeform shaped parts. CIRP Ann Manuf
Technol 57(2):810–835
Savio CD, Dejima S, Danzebrink HU, Gotszalk T (2007b) 3D metrology with a compact scanning
probe microscope based on self-sensing cantilever probes. Meas Sci Technol 18:328–333
Schwenke H, Härtig F, Wendt K, Wäldele F (1999) Future challenges in coordinate metrology addressing
metrological problems for very small and very large parts. In: IDW conference, Knoxville
Schwenke H, Neuschaefer-Rube U, Pfeifer T, Kunzmann H (2002) Optical methods for dimen-
sional metrology in production engineering. CIRP Ann Manuf Technol 51(2):685–699
Shiozawa H, Fukutomi Y, Ushioda U, Yoshimura S (1998) Development of ultra-precision 3D-
CMM based on 3-D metrology frame. In: Proceedings of the 13th ASPE, pp 15–18
Stoup J, Doiron T (2003) Measurements of large silicon spheres using the NIST M48 coordinate
measuring machine. Proc SPIE 5190:277
Suzuki H, Yamagata Y, Higuchi T (2006) Recent trend in ultra-precision machining system. J Jpn
Soc Precis Eng 72(4):417–421
SYVEC Corporation (2018) A high precision cycloid gear for automotive part formed by stamping.
http://www.syvec.co.jp. Accessed 31 Mar 2018
Takakai D, Ikeno J, Omori Y, Shibutani H, Suzuki A (2006) Fabrication of micro glass ball by laser
beam and its application. In: Proceedings of 2006 JSPE general meeting in spring, pp 333–334
Takamasu K, Ozawa S, Asano T, Suzuki A, Furutani R, Ozono S (1996) Basic concepts of nano-CMM.
In: The Japan – China bilateral symposium on advanced manufacturing engineering, pp 155158
Takamasu K, Furutani R, Ozono S (1997) Development of nano-CMM (coordinate measuring
machine with nanometer resolution). In: Proceedings of the 6th IMEKO world congress,
pp 34–39
Takamasu K, Chih-Che K, Suzuki A, Hiraki M, Furutani R, Ozono S (1998) Development of
pneumatic ball probe (1st report) – basic construction. J Jpn Soc Precis Eng 64(8):1153–1157
Takaya Y, Takahashi S, Miyoshi T, Saito K (1999) Development of the nano-CMM probe based on
laser trapping technology. CIRP Ann Manuf Technol 48(1):421–424
Takaya Y, Imai K, Ha T, Miyoshi T, Kinoshita N (2004) Vibrational probing technique for the nano-
CMM based on optical radiation pressure control. CIRP Ann Manuf Technol 53(1):421–424
Takeuchi Y (2008) Ultraprecision micro mechanical machining. The Nikkan Kogyo Shimbun,
Tokyo
Taniguchi N (1983) Current status in, and future trends of, ultraprecision machining and ultrafine
materials processing. CIRP Ann Manuf Technol 32(2):573–582
Tsutsumi H, Yoshizumi K, Takeuchi H (2005) Ultrahigh accurate 3-D profilometer. Proc SPIE
5638:387–394
Varadan VK, Jiang X, Varadan VV (2001) Microstereolithography and other fabrication techniques
for 3D MEMS. Wiley, New York
Vermeulen MMPA, Rosielle PCJN, Schellekens PHJ (1998) Design of a high-precision
3D-coordinate measuring machine. CIRP Ann Manuf Technol 47(1):447–450
Vít Z, Prokop Z (2007) Multi-sensor coordinate measuring machine and internet virtual multi-
sensor measuring machine in CMI Prague. In: Proceedings of the 9th ISMQC, pp 126–130
Weckenmann W, Estler T, Peggs G, McMurtry (2004) Probing systems in dimensional metrology.
Weckenmann A, Hoffmann J, Schuler A (2008) Development of a tunnelling current sensor for
a long-range nano-positioning device. Meas Sci Technol 19:064002
Werth-messtechnik (2018) Fiber probe WFP. https://www.werth.de/en/start/home.html. Accessed
31 May 2018
XPRESS precision engineering (2018) Gannen series XP, XM. http://www.xpresspe.com/gannen.
php. Accessed 31 May 2018
Yamamoto M, Takeuchi H, Aoki S (2000) Dimensional measurement of high aspect ratio micro
structures with a resonating micro cantilever probe. Microsyst Technol 6:179–183
Yoshizumi K, Murao T, Masui J, Imanaka R, Okino Y (1987) Ultrahigh accuracy 3-D profilometer.
Appl Opt 26(9):1647–1653
Yoshizumi K, Kubo K, Takeuchi H, Handa K, Kassai T (2002) Ultra accurate 3-D profilometer
using atomic force measure nanometer. J Jpn Soc Precis Eng 68(3):361–366
Machine Tool Calibration
6
J. R. R. Mayer
Contents
Machine Tool Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
Machine Tool Errors or Deviations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
Machine Tool Error Classifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 190
A Serial Kinematic View of Geometric Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
Machine Tool Interaxis Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 192
Machine Tool Intra-axis Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 199
Volumetric Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 200
Machine Tool Error Modelling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 203
Vectorial Approach . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 204
Homogenous Transformation Matrix (HTM) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 208
Forward Model with Machine Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
Machine Tool Metrology and Error Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
Direct Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 209
Indirect Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210
Machine Tool Compensation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
Lookup Table . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
Inverse Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 212
Abstract
Machine tools produce parts by moving a tool relative to a workpiece. Any
deviation from the command path may result in errors on the part thus degrading
its quality. Machine tool calibration aims to quantify and compensate the machine
errors in order to make better parts. This chapter reviews the definitions, nomen-
clature, and some principles associated with machine tool geometric errors.
Forward mathematical models are also presented that calculates the volumetric
errors at the tool tip as functions of the causal interaxis and intraaxis errors of the
J. R. R. Mayer (*)
Department of Mechanical Engineering, Polytechnique Montreal, Montreal, QC, Canada
e-mail: rene.mayer@polymtl.ca

https://doi.org/10.1007/978-981-10-4938-5_6
190 J. R. R. Mayer
machine with examples covering three- and five-axis machines. Finally, measure-
ment approaches and compensation schemes are briefly covered
Keywords
Machine tool · Calibration · Metrology · Geometric errors · Error modeling
Machine Tool Calibration
A machine tool effects a change in a workpiece by moving a tool relative to this

workpiece. The change is produced through a process which is often one of material
removal. However, more generally, the tool could also add material, through an
additive process, or simply induce physical changes to the workpiece.
The ability of the machine tool to position, and possibly orientate, its tool relative
to the workpiece affects the quality of the part such as its surface and subsurface
quality and the geometry and dimensions of its various features.
Calibration in the context of measurement is defined as (International vocabulary
2012) “operation that, under specified conditions, in a first step, establishes a relation
between the quantity values with measurement uncertainties provided by measure-
ment standards and corresponding indications with associated measurement uncer-
tainties and, in a second step, uses this information to establish a relation for
obtaining a measurement result from an indication.”
In the field of machine tools, calibration could be said to consist in “operation
that, under specified conditions, in a first step, establishes a relation between the
quantity values with measurement uncertainties provided by measurement standards
or instruments and corresponding workpiece physical realization or machine indi-
cations with associated uncertainties and, in a second step, uses this information to
establish a relation for obtaining from the machine tool a physical realization or
measurement result from its indications”. In simpler terms calibration consists in
gathering quantitative data on quantities associated with a machine tool’s ability to
position and orientate, or to know the position and orientation of, its tool with respect
to the workpiece and using such data to improve this ability. The improvement is
obtained through some form of physical correction or compensation. The compen-
sation involves updating single numerical parameter values in the machine controller
or multivalued tables. The controller then uses such values to make small changes to
the axes commands.
Machine Tool Errors or Deviations
Machine Tool Error Classifications
In the field of metrology, the term error is defined in (International vocabulary 2012)
as “measured quantity value minus a reference quantity value.” In the field of
machine tool, the term error refers to unwanted motion (ISO 230-1 2012), departure
6 Machine Tool Calibration 191
from target position, or simply the actual behavior minus the desired one. When
using the machine tool as a measuring machine, what matters is that the machine
knows where its tool is. However, when machining, just knowing its position is not
sufficient because the tool must also be where the program requires it to be since its
path will generate the part.
Errors are quantities (https://en.oxforddictionaries.com/definition/) and thus can
be expressed numerically. The term deviation may be more appropriate in the field of
machining as it simply means “the action of departing from an established course or
accepted standard” but “error” is widely adopted. Here we categorize errors
according to their causes. Errors can be viewed using the causality principle (Onat
Ekinci et al. 2009) whereby fundamental causes of errors are propagated to cause
errors of the tool position and orientation relative to the workpiece.
Quasi-static errors: in physics, “static” means “concerned with bodies at rest or
forces in equilibrium” (https://en.oxforddictionaries.com/definition/), whereas
“quasi” means “being partly or almost.” Quasi-static errors refer to the operating
conditions of the machine tool under which the errors are present and not to the
characteristics or physical causes of the errors. As a result, in the field of machine
tool errors, “quasi-static” refers to errors that are present when the machine is
moving at very low speed either in single axis or multi-axis mode and under no-
load conditions (self-weight only), i.e., with a workpiece of negligeable weight and
without cutting forces. Quasi-static errors include geometric errors and hysteresis
effects.
Geometric errors: geometric errors (for an unloaded (self-weight), “cold” (has
been on for a while) and stationary machine). Note that geometric errors, which
should primarily refer to errors (or deviations) of the machine parts and assembly
errors between parts, also include some amount of thermal effects since the machine
will not be at 20 C throughout. There will also be some elastic deformation due to
the machine’s own weight. Geometric errors are generally associated with quasi-
static conditions. However, changing thermal and dynamic conditions will affect the
geometry of the machine and induce non quasi-static geometric errors.
Dynamic errors: dynamics refers to the relationship between wrench (force and
torque) and motion twist (linear and angular displacements), so for machine tools,
dynamic errors broadly refer to errors that appear or are modified during movement
and changes in direction at various speeds such as those used during machining or
errors due to changes in loading conditions, whether the machine is moving or not.
The errors are due to elastic deformation and servo-errors. Elastic deformation
occurs due to inertial forces caused by accelerations and the self-mass and the
mass of the workpiece as well as changes in gravitational forces due to workpiece
weight. Servo-errors will change with axis speeds and acceleration and with the
weight of the workpiece. Dynamic effects will also appear during cutting which
produces cutting forces. Some dynamic errors are seen by the axis feedback systems,
i.e., the linear and rotary encoders, while others coming from the machine mechan-
ical structure are not. Thus, some dynamic errors can be subcategorized as structural
errors which are not seen by the controller and servo-errors.
Thermal errors: thermal errors are associated with thermal expansion due to the
temperature distribution of the various components of the machine and their
192 J. R. R. Mayer
coefficient of thermal expansion. The machine has internal heat sources such as
motors, drives, guiding systems, spindle, and cutting fluids, which produce or
conduct heat and consequently cause temperature changes. It also has external
heat sources as it operates in an environment which can also affect its temperature
through convection, conduction, and radiation. Since a machine is unlikely to
operate at 20 C, the ISO reference temperature for geometrical product specification
and verification, it generally exhibits thermal errors. Thermal errors could be defined
as changes to quasi-static geometric errors, known for a specified reference machine
thermal state, as the machine thermal state changes from that reference state.
Servo-errors: servo-errors are inherent to control systems and quantify the differ-
ence between the desired and the actual axis position at a given time. Servo-errors
are mainly present under dynamic conditions.
Numerical control errors: while executing a G-code program, which contains
motion instructions for the machine to accomplish, the machine controller’s algo-
rithms issue individual axis commands which may involve various approximations
such as converting curvilinear motion into a suite of linear segments thus producing
errors from the desired trajectory.
A Serial Kinematic View of Geometric Errors
In serial kinematics, a chain of link and joint pairs connect the workpiece to the tool.
Any error, as deviation from nominal, in this chain can result in a volumetric error
which is any deviation between the desired and actual location of the tool relative to
the workpiece. In this chapter, location refers to position and orientation. The
volumetric error results from (1) interaxis errors occurring at the link holding the
next axis, (2) the propagation of intra-axis errors occurring after joint displacement at
the joint (axis) producing the axis motion, (3) erroneous knowledge of the tool
geometry (tool setup), or (4) erroneous knowledge of the location of the workpiece
relative to the axis holding the workpiece (workpiece setup).
Machine Tool Interaxis Errors
The interaxis errors have also been called link errors, position-independent geomet-
ric errors, or axis location errors. Because interaxis errors are relative between an
axis and other axes, it is necessary to define the axes used as primary and secondary
references. For example, in Fig. 1 (left), the X-axis is taken as the primary axis so
that it cannot have any interaxis error. The Y-axis, which for this example should
move perpendicularly to the X-axis, may depart from this desired orientation and be
affected by an out-of-squareness error EC(0X)Y where E stands for “error,” C is the
axis around which the orientation error of the Y-axis is defined, and Y is the axis
affected by this error. The errors are associated with the quality of the motion of the
tool relative to the workpiece and not the underlying systems, such as guideways and
bearings, producing the motion.
Fig. 1 Interaxis error between two linear axes. The out-of-squareness error is associated with the
tool trajectory obtained when moving one axis and then the other axis. Also, note how the sign of
the error changes when the reference, or datum, axis changes
In practice, linear axes do not necessarily move on perfect straight lines, due to
the presence of intra-axis errors, and as a result, two additional notions are required
to have a realistic view of the interaxis error. The first notion is that of substitution
line and the other is that of locality.
Substitution Straight Line of Motion for Linear Axes and Substitution

Rotation Axis for Rotary
The actual trajectory of a linear axis will likely slightly depart from a straight line due
to the presence of intra-axis errors. As a result, in order to determine the interaxis
error of a linear axis, it becomes necessary to assign a substituted straight line of
motion to the actual axis. Possible substitution candidates are the straight line
minimizing the sum of the square of the residuals between points on the actual
line and the substituted straight line or more simply described as the ordinary least
square best fit line or Gaussian line. An alternative is to take the end points of the
actual line to define this straight line [ISO 230-1:2012]. A similar condition applies
to rotation axes in the presence of intra-axis errors. The actual rotation does not
necessarily take place around a unique axis, and so a substituted fixed axis is defined
called the axis average line.
Locality
The notion of locality refers to the likely change in the value of, for example, EC(0X)Y
depending on the range of travel exercised for the measurements. As a result,
interaxis errors should be accompanied by a statement of the range of axes motion
that was used for estimation.
194 J. R. R. Mayer
Fig. 2 Locality in interaxis definition. In the presence of intra-axis errors on the linear axes, the
estimated value for the interaxis error such as the out-of-squareness of the Y-axis with respect to the
X-axis may vary depending on the range of motion used
Figure 2 shows a further example of the difficulty in defining out-of-squareness

between linear axes in the presence of intra-axis error, here straightness errors, and
the effect of locality on the estimated value.
Complete Set of Interaxis Errors for a Linear Axis

A linear axis provides a linear motion in a direction and has up to two indepen-
dent interaxis errors. It is possible to assign a direction to a linear axis but not an
unambiguous zero position. The linear motion shows a relative displacement but
not an absolute position. This has been recognized in the field of mechanisms
(Everett and Hsu 1988; Everett and Suryohadiprojo 1988). As a result, up to two
angular interaxis errors can be defined. One linear axis is chosen as primary and
as such has no other axis to be compared with so that it has no interaxis error. A
second linear axis is chosen as secondary and can only have one angular error,
such as an out-of-squareness error, relative to the primary axis. The primary and
secondary axes together define a plane of motion so that a third linear axis, which
normally would be required, on most machines, to be nominally perpendicular to
that plane, will have two angular interaxis errors. This yields a total of three
angular errors, such as three out-of-squarenesses for a three linear axis machine
tool (Fig. 3).
The angular and linear error convention is as shown in Fig. 4.
Fig. 3 Interaxis errors of the Z-axis with respect to the X-axis and Y-axis. The positive sense of the
angular error is given by the right-hand rule
Interaxis Errors of a Five-Axis Machine Tool with Two Rotary Axes

The minimum number, N, of interaxis error for a complete definition of the interaxis
geometric errors of a serial machine, like serial kinematic machine tools, as opposed
to parallel kinematic machine tools, is given by the following equation (Everett and
Hsu 1988; Everett and Suryohadiprojo 1988):
N ¼ 2P þ 4R 6
where P is the number of linear (P for prismatic) axes, R is the number of rotary axes,
and 6 is to remove the geometric parameters associated with the position and
orientation (location) of the machine relative to some external reference coordinate
frame. A prismatic axis is fully defined by its direction which requires two angles. A
rotary axis is fully defined by its direction, so two angles, and its position which
requires a point on the axis. Because the point can be anywhere along the axis, it has
two radial translational constraints, and so two scalar quantities suffice. For a five-
axis machine, with two rotary and three linear axes, N = 2 3 + 4 26 = 8
(Abbaszadeh-Mir et al. 2002). However, the position and orientation of the spindle as
a rotary axis, on a milling machine for example, is also relevant and adds a further
four error parameters, N = 2 3 + 4 36 = 12. The minimum number of interaxis
196 J. R. R. Mayer
Fig. 4 (Left) The right-hand rule shows the positive sense of the linear and angular errors, both
inter- and intra-axis, relative to the axes of the machine coordinate system. (Right) The right-hand
rule also shows the positive sense for angular errors around each axis. Note how the thumb points in
the positive sense of the axis and how the fingers, wrapped around the axis, indicate the positive
sense of an angular error and the thumb the positive sense of a linear error
errors is independent of the particular topology of the machine tool. The topology is
the sequence of axes between the workpiece, w, and the tool, t. It is described by a
sequence of letters with a “w” at one end and a “t” at the other end with an “f” or a
“b” to separate the axes sequence of the workpiece and tool branches. “b” stands for
the machine base (ISO 230-1 2012), but to avoid a possible confusion with a B-axis,
the letter “f” is preferred here. The machine illustrated in Fig. 5 has the topology
wCBXfZY(C1)t with C1 representing the spindle axis. Its kinematic diagram is
shown in Fig. 6.
Figure 7 shows the interaxis errors for a machine with topology wCBXfZY(C1)t.
Because interaxis errors are deviations between axes, it becomes necessary to define
datums. Here, the X-axis was arbitrarily selected as primary to provide a datum, and
as a result, it is perfectly aligned with itself and has no interaxis error. The Z-axis is
arbitrarily selected as a secondary axis and as a result has only one angular interaxis
error around y, a B error named EB(0X)Z, defined with respect to the primary axis X,
although the error itself is a rotation around the y-axis of the reference frame of
the X-axis, ^j X , which is perpendicular to the X-axis and to the plane formed by the
actual substituted straight lines for the X- and Z-axes. Continuing along the tool
Fig. 5 Illustrative diagram of a wCBXfZY(C1)t machine tool
Fig. 6 Kinematic diagram of a nominal, without errors, machine tool with topology wCBXfZY
(C1)t also showing the tool and workpiece coordinate frames
198 J. R. R. Mayer
Fig. 7 Kinematic diagram of a machine tool with topology wCBXfZY(C1)t showing the interaxis
errors for the workpiece and tool branches as well as the linear components (vector) of the
volumetric error
branch, The Y-axis has two angular interaxis errors, one around x, an A error such as
EA(0Z)Y, and the other around z, a C error such as EC(0X)Y. Putting aside the spindle for
now, let us consider the rotary axes on the workpiece branch side. The B-axis has only
two angular interaxis errors and none of the potential two linear errors. This is because,
in kinematics, the X-, Y-, and Z-axes, as prismatic axes, do not have the capacity to
define translational references like rotary axes do. As a result, the B-axis is a transla-
tional datum in x and in z. The two angular interaxis errors are one error around x, an A
error named EA(0Z)B, and another around z, a C error named EC(0X)B. EA(0Z)B and
EC(0X)B are quantified relatively to the secondary axis Z and the primary axis X,
respectively. Next the C-axis has three interaxis errors out of the potential four. The
linear error in the y direction cannot be defined since there is yet no available
translational datum in y. The errors are two angular interaxis errors, EA(0B)C and
EB(0X)C, and one linear error, EX(0B)C. EA(0B)C is quantified relatively to the B-axis,
but the Z-axis could also be used. EB(0X)C is quantified relatively to the X-axis
although the Z-axis could also be used. EX(0B)C is quantified relatively to the B-axis,
and there is at this point no alternative since only the B-axis provides the necessary
translational datum. Finally, back to the tool branch, the spindle axis (C1) has all four
potential interaxis errors of a rotary axis, two angular and two linear which are EA(0Y)
C1, EB(0Z)C1, EX(0B)C1, and EY(0C)C1, respectively. Here too, alternative reference axes
for their quantification are possible. In serial kinematics it is usual to define axes
relative to the nearest neighboring axis toward the base capable of providing the
Table 1 A possible set of interaxis errors (ISO 230-1 2012; Abbaszadeh-Mir et al. 2002) for the
machine schematics shown in Fig. 7
Axis with the error
Type of error X Y Z B C C1
Translation in x – – – EX(0B)C EX(0B)C1
Translation in y – – – EY(0C)C1
Translation in z – – – –
A, rotation around x – EA(0Z)Y – EA(0Z)B EA(0B)C EA(0Y)C1
B, rotation around y – – EB(0X)Z EB(0X)C EB(0Z)C1
C, rotation around z – EC(0X)Y – EC(0X)B – –
Table 2 Descriptive names for the interaxis errors

Symbol Descriptive name
EB(0X)Z Out-of-squareness of the Z-axis relative to the X-axis
EA(0Z)Y Out-of-squareness of the Y-axis relative to the Z-axis
EC(0X)Y Out-of-squareness of the Y-axis relative to the X-axis
EA(0Z)B Out-of-squareness of the B-axis relative to the Z-axis
EC(0X)B Out-of-squareness of the B-axis relative to the X-axis
EX(0B)C Cross-axis offset of the C-axis relative to the B-axis
EA(0B)C Out-of-squareness of the C-axis relative to the B-axis
EB(0X)C Out-of-squareness of the C-axis relative to the X-axis
EX(0B)C1 Cross-axis offset of the spindle relative to the B-axis
EY(0C)C1 Offset in y of the spindle relative to the C-axis
EA(0Y)C1 Out-of-squareness of the spindle relative to the Y-axis
EB(0Z)C1 Out-of-parallelism around y of the spindle relative to the Z-axis
necessary constraints, and this is the approach that was used here. However, an equally
valid approach is to use the base frame defined from the primary and secondary axes as
the reference to quantify the axis location errors from nominal.
Table 1 summarizes the interaxis errors of the machine and Table 2 suggests
descriptive names.
Machine Tool Intra-axis Errors
Intra-axis errors describe any departure from the perfect, nominal, or desired motion
of a single axis such as the X-, Y-, Z-, A-, B-, C-, or C1-axis of a serial kinematic
machine tool. The action of an axis is to move a physical object such as a table or a
tool or another set of axes down the kinematic chain (toward the tool or the
workpiece). An object, in R3, requires three linear and three angular quantities to
fully describe its position and orientation. Similarly, these six quantities are required
to describe the departure of an object’s position and orientation from its desired one.
As a result, a 3 linear axis serial kinematic machine will have 18 intra-axis errors and
200 J. R. R. Mayer
3 interaxis errors for a grand total of 21 errors, whereas a serial kinematic five-axis
machine which includes 2 rotary axes will have 5 6 = 30 intra-axis errors and
8 interaxis errors plus another 4 interaxis errors if the spindle location is considered
for a grand total of 42 errors.
Various nomenclatures have been used in the scientific literature. Here, a power-
ful nomenclature clearly defined in (ISO 230-1 2012) and maybe originating in part
in (Schultschik 1977) is adopted. As an example, EYX is a linear error in the y
direction, indicated by the first subscript, afflicting the nominal, desired, or perfect
motion of the X-axis, as indicated by the second subscript. The intra-axis error of an
axis cannot be a single-value scalar; otherwise this would represent an interaxis error.
Minimally, the intra-axis error varies with the indicated position of the axis. On a
machine tool, this is usually the commanded position in quasi-static mode when the
servo-error is null. However, as the motion of an axis can affect the loading of
another, an axis intra-axis errors can also be a function of the indicated positions of
other axes. When considering thermal effects, the intra-axis errors can also be
affected by thermal expansion which depends on the ambient temperature and
temperatures of various parts of the machine.
Intra-axis Errors of a Linear Axis

Figure 8 illustrates the six intra-axis error of an X-axis. Table 3 lists the intra-axis
errors of the X-axis with their symbols and descriptions. An axis without errors is
such that any points of a body rigidly connected to the moving part of the axis, the
carriage, would follow the same rectilinear trajectory.
Various functions have been used to attempt to model quasi-static errors such as
ordinary polynomials (Slamani et al. 2011), Legendre polynomials (Kruth et al.
1994), and Fourier series (Bringmann and Knapp 2009). The machine, unfortu-
nately, does not have to conform to any such model. Correlations between the six
intra-axis errors were reported (Ferreira and Liu 1986; Florussen et al. 2001; Pahk
et al. 1997; Bryan 1979; Ekinci and Mayer 2007) and are likely due to the guideway
and drive systems which do not control the six degrees of freedom independently
from each other.
Intra-axis Errors of a Rotary Axis

Rotary axes also potentially exhibit six intra-axis errors as deviations from their
nominal (desired), or pure, rotation (Fig. 9) (Table 4). The nominal rotation is one
where all points of a rigid body, attached to the moving part of this axis, follow a
circular trajectory around that axis and in a plane perpendicular to that axis and with
the same angular motion around that axis.
Volumetric Errors
The volumetric error, as illustrated in Fig. 7, characterizes the deviation in position

and orientation of the tool with respect to its desired position an orientation relative
to the workpiece. Up to six error terms, three translational errors and three angular
Fig. 8 The intra-axis errors

of a linear axis, such as the
X-axis, are the six errors, three
linear and three angular,
expressing the departure from
the nominal (desired or
perfect) motion of the axis.
Errors are visually amplified
for explanation purposes
Table 3 Symbols and descriptions of intra-axis errors of the X-axis

Symbol Description
EXX Linear positioning error in x of axis X
EYX Straightness error in y of axis X
EZX Straightness error in z of axis X
EAX Angular error motion around x (roll) of axis X
EBX Angular error motion around y (pitch) of axis X
ECX Angular error motion around z (yaw) of axis X
errors, can be considered. Any unwanted motion of the tool or the workpiece may
result in a volumetric error.
Sensitive Direction or Relevant Volumetric Error

Not all volumetric errors degrade the precision of the part. This is because the
process may be insensitive to certain volumetric errors. Figure 10 illustrates a turning
process. Using geometric arguments, the following can be concluded:
– A radial volumetric error potentially affecting the roundness and the local
diameter.
– An axial volumetric error changes the axial position of a shoulder feature on the
part.
– A tangential volumetric error produces a negligible effect on the local part radius
and no effect in the axial direction. As is schematized in Fig. 11, the radial effect
202 J. R. R. Mayer
Fig. 9 The intra-axis errors

of a rotary axis, such as the
C-axis, are the six errors, three
linear and three angular,
expressing the deviation from
the nominal (desired or
perfect) motion of the axis.
Errors are visually amplified
for explanation purposes
Table 4 Symbols and descriptions of intra-axis errors of the C-axis

Symbol Description
EXC Radial error motion in x (after c rotation) of axis C
EYC Radial error motion in y (after c rotation) of axis C
EZC Axial error motion in z (after c rotation) of axis C
EAC Tilt error motion around x (after c rotation) of axis C
EBC Tilt error motion around y (after c rotation) of axis C
ECC Angular positioning error around z (after c rotation) of axis C
is a cosine error for small volumetric errors as a proportion of the desired radial
position, R, of the tool:

R 1 1 cos θ
ΔR ¼ R0 R ¼ R¼R 1 ¼R
cos θ 0 cos θ cos θ 1

θ2 θ4
B1 1 2 þ 24 . . . C
B C
¼ RB C (1)
@ θ2 θ4 A
1 þ ...
2 24
θ2 t2
R
2 2R
Fig. 10 Sensitivity of the part precision to the linear components of the volumetric error in a
turning process
Fig. 11 Effect of a tangential

volumetric error on the cut
radius which is negligible for
a small tangential volumetric
error, t
Similar analyses can be conducted for a milling process considering the interaction
between the tool and the workpiece in the case of a flat end mill, a ball end mill, or a
flank milling process.
Machine Tool Error Modelling
Machine tool error modelling has evolved in parallel with similar developments in
the field of robotics. Two approaches can be found in the literature. One is to use
exact formulation and then assume small errors to neglect products of errors. The
204 J. R. R. Mayer
other is to directly propagate the errors to the locations of interest, typically the tool
and its corresponding desired location in the workpiece frame.
Vectorial Approach
A quick reminder of coordinate transformations may be useful at this point. The

coordinate in reference frame A, A p 1 , of a point p1 , with its coordinates known in
frame B, B p 1 , can be calculated from the coordinates of the origin of frame B in
frame A, A p oB , and the projection matrix, or rotation matrix, made of the base
vectors of reference frame A projected along the directions of frame B, A B R as
follows:
2A 3 2B 3
p oB,x î A
A
p 1 ¼ A p oB þ A R B
p ¼ 4 A
p oB,y
5 þ 4 B^ 5B
jA p1 (2)
B 1
A B^
p oB,z kA
As will be shown later, A B R can also be obtained from knowing the three
successive rotations to align frame A with frame B using either the fixed frame
(the rotations are around the original frame position) or the moving frame (the
rotation is around the axis resulting from the previous rotations) conventions.
When using the fixed frame convention, the basic rotation matrices are concatenated
from right to left, whereas for the moving frame convention, they are concatenated
from left to right (Fig. 12).
The vectorial approach is used to calculate the vector between a point on the
workpiece (w) attached to the last workpiece branch axis frame and the tool tip (t)
known in the last tool branch axis frame as shown in Fig. 13.
The following development has many similarities with that in (Schultschik 1977).
In the case of a nominal (N) machine, i.e., without errors, the equation is
! ! ! ! ! !
¼ x î X þ y ^j Y þ z k^Z þ t w ¼ x þ y þ z þ t w
w N !
pt (3)
where, for example

2 3 2 3 2 3 2 3
x 0 0 tx
! ! ! !
x ¼ 4 0 5; y ¼ 4 y 5; z ¼ 4 0 5; t ¼ 4 t y 5: (4)
0 0 z tz
!
Generally on a machine tool, t ¼ ½0 0 tool lengthT .
If we assume that the tool is at its desired position relative to the workpiece, then
!
the nominal workpiece position is obtained by setting w ptN ¼ 0 in Eq. (3):
! ! ! ! ! !
0¼ x þ y þ z þ t w
!
(5)
! ! ! !
w ¼xþ yþ zþ t :
Fig. 12 Coordinate transformation
Fig. 13 Kinematic diagram of a nominal wfXYZt machine

206 J. R. R. Mayer
The contact point between the tool and the workpiece is defined as the functional
point in (ISO 230-1 2012). It is at this point that the volumetric error should be
!
relevantly estimated. It is the position given by w in the last workpiece branch frame
(Givi and Mayer 2015).
For a machine with errors, the nature of the error, linear or angular, and its
location of action must be defined. The 21 intra- and interaxis errors and their
location of actions are shown in Fig. 14.
The position of the tool center point and of the workpiece can be calculated
independently in the foundation frame and the volumetric error calculated as
follows:
Fig. 14 Kinematic diagram of a nominal wfXYZt machine with the action and location of the 21
geometric errors
!
! w !N F !V F !V F !N F !N
E TV ¼ w p V
t p t ¼ p t p w p t p w

! F !N ! F !N
¼ FpV t pt F pV w pw : (6)
where ETV is the vector of translational volumetric error and
w !V
pt is the true (V for verum which is true in Latin) tool position relative to the
workpiece at the functional point;
F !V
p t is the true tool position in the foundation frame;
F !V
p w is the true workpiece position in the foundation frame,
with
! h ! h ! !
ii
F !V ! ! X X0 ! Y Y0 !
pt ¼FpV
t ¼ x þ E TX þ X0 R Y0 R y þ E TY þ Y 0 R Z0 R z þ E TZ þ Z0 R t
Z
, (7)
where, as examples,
2 3
cos E Cð0XÞY sin E Cð0XÞY 0
X0
4 sin ECð0XÞY cos E Cð0XÞY 0 5,
Y0 R¼ (8)
0 0 0
2 32 32 3
cos E CZ sin ECZ 0 cos E BZ 0 sin EBZ 1 0 0
Z
Z0 R¼ 4 sin ECZ cos ECZ 0 54 0 1 0 54 0 cos E AZ sin E AZ 5
0 0 1 sin E BZ 0 cos E BZ 0 sin EAZ cos E AZ
(9)
(which uses successive basic rotations around z-, y-, and x-axes using the moving
frame convention),
2 3
!
EXZ
E TZ ¼ 4 EYZ 5, (10)
EZZ
and
F !V !
pw ¼w : (11)
It is customary at this point to make the hypothesis of small errors leading to scalar
equations which could then be represented in a single matrix form. The products are
executed; any terms with a product of two or more error terms are neglected, and the
following simplifications are applied:
sin θ θ and cos θ 1f or θ 1:

208 J. R. R. Mayer
The result can be presented in scalar form as follows:
E XV ¼ w pV
t,x E XX þ E XY þ E XZ ðE CX þ E Cð0XÞY Þðy þ t y Þ
þ ðE BY þ E Bð0XÞZ Þðz þ t z Þ þ EBX ðz þ t z Þ E CY t y E CZ t y þ EBZ t z

E YV ¼ w V
p t ,y EYX þ E YY þ E YZ ðE AY þ E Að0YÞZ Þðz þ t z Þ
(12)
E AX ðz þ t z Þ þ ðE CX þ E Cð0XÞY Þt x þ E CY t x þ ECZ t x E AZ t z
E ZV ¼ w pV
t,z E ZX þ E ZY þ E ZZ þ E AX ðy þ t y Þ E BX t x
ðE BY þ E Bð0XÞZ Þt x EBZ t x þ ðE AY þ EAð0YÞZ Þt y þ EAZ t y
Inspection of Eq. (12) reveals that the translational intra-axis errors contribute
unaltered to the translational components of the volumetric errors. However, the
angular inter- and intra-axis errors propagate proportionally to the relative coor-
dinates to the tool and so contribute to the translational volumetric errors through
Abbe and Bryan offset (Bryan 1979), thus causing the associated Abbe and
Bryan errors. What is not obvious from the above, and rather demanding analyt-
ical formulation, is that the same result can be obtained through the use of the
cross product to propagate the effect of small angular errors at a distance through
a rigid body. The use of cross products yields the estimated volumetric error
directly as
! ! ! ! h !
i h !
i
! ! !
E TV ¼ E TX þ E TY þ E TZ þ ERX y þ z þ t þ ERY z þ t
!
þ ERZ t (13)
where as an example,
ERX ¼ ½E AX E BX ðE CX þ E Cð0XÞY ÞT : (14)
Homogenous Transformation Matrix (HTM)
Homogenous transformation matrices have various properties which are not all
used in the context of kinematic modelling where they are basically a formalism
which simplifies the development of equations and the subsequent coding
although they may not be computationally efficient requiring numerous multipli-
cations by zero.
The HTM combines the effect of a 3D translation followed by three basic
rotations about the reference frame axes. The HTM executes the 3D translation of
a reference frame, associated with a body, followed by its rotation. For example,
!
referring to Fig. 12, the coordinates, A p 1 , of a known point, 1, in the moved frame
location, B, can then be calculated in the original frame location, A, as follows:
2 A
32 B 3
poB,x p 1 ,x
6 poB,y 7 6B 7
B R 33
A A
A!
p1 ¼A 6 76 p 1,y 7:
B T p1 ¼ 4
B
(15)
A 5
poB,z 4 B p 1,z 5
0 0 0 1 1
The mathematical formulation to calculate the position of the tool tip relative to a
workpiece feature frame for a machine tool made of two separate kinematic chains
such as the one shown in Fig. 6 is
feature !N
1
p tool ¼ ffeature T C1f T C1 ptool ¼ feature f T C1f T C1 ptool

1 f Z
¼ Xf TX B TC Tw Tfeature TÞ Z TY TC1 T ptool :
B C w Y C1
(16)
Forward Model with Machine Errors
Using HTMs the forward kinematic model of a serial kinematic machine tool
including inter- and intra-axis errors is obtained by enriching each mechanical axis
HTM by incorporating the errors at the appropriate positions in the sequence of
HTM by following the order of the error action locations along the serial chain as
shown in Fig. 14. The enrich HTM for a single axis becomes
X0 0 0
Y0 T¼ Y0 X TY0 0 Y0 TY Y0 TY0 Y T
¼Tnominal axis location Tinteraxis errors Tnominal axis motion Tintraaxis errors (17)
Machine Tool Metrology and Error Estimation
The purpose of machine tool metrology is to obtain values for quantities such as
volumetric errors and inter- and intra-axis errors. Whereas volumetric errors are measured
always directly, inter- and intra-axis errors are measured either directly or indirectly. The
indirect approach consists in using mathematical models of the relationship between
causal inter- and intra-axis errors and their effects on the volumetric error.
Direct Measurement
Direct measurement of an error refers to obtaining values for an error, such as

EXX(x), free from disturbance from other errors. It is usually accepted, for instance,
210 J. R. R. Mayer
that using a linear laser interferometer directly provides EXX(x) values. However,
as was demonstrated by Abbe and Bryan, angular errors affect the linear errors
such as EXX, EYX, EYY, etc. which means that the measured values may change,
in the presence of EA?, EB?, and EC? errors, depending on the position of the
measurement lines and other setup quantities. On the other hand, angular errors
can be uniquely measured as they do not suffer such coupling. Direct measure-
ment offers dense data sets but is often described as time consuming requiring
hours of downtime (the machine being unproductive) because of the need for
multiple setups (Okafor and Ertekin 2000a, b; Gao et al. 2006). However,
multiple error measuring units, described as 5D or even 6D, can acquire the
error values in a single setup (Qibo et al. 2013; Fan et al. 1998). Note that
attention must be given to the measurement uncertainty of the various
techniques.
Measurement of angular errors of a linear axis such as EBX and ECX can be
measured using an angular interferometer. Interferometers come in a variety of form
but fundamentally count and interpolate interference fringes. By cleverly using
various optics to produce secondary laser beams and recombining them for interfer-
ence purposes, up to five of the six intra-errors are measured for linear axes. Roll, or
rotation around the axis moving direction, such as EAX, EBY, and ECZ, is not directly
measurable by interferometry. Also, error of rotary axes will require a
counterrotating table with calibrated reference positions with the measurement
performed by the interferometer.
Autocollimators and calibrated polygon mirrors can measure the angular
errors of rotary axes, with a number of points limited by the number of mirror
facets on the polygon mirror. It can also measure two of the three angular
errors of linear axes with a single mirror. Another approach uses five capac-
itive sensors and two on axis balls for measuring five of the six intra-errors
(except the angular positioning error) of a rotary axis such as a spindle (Grejda
et al. 2005).
Indirect Measurement
Indirect measurement involves gathering readings of the volumetric errors which are
sensitive to at least one of the fundamental inter- and intra-axis errors. As a result,
posttreatment of the data is necessary to separate and quantify the fundamental
errors. Examples of such systems are the telescoping magnetic ball-bar (Bryan
1982a, b; Knapp 1983) with an interpolated circular trajectory, which measures the
distance between a ball on the workpiece table and another ball attached to the
toolholder effectively measuring the change in the distance between the two balls
due to the 3D volumetric error. This change can be approximated by the projection of
the volumetric error in the direction b^ defined by the two balls’ nominal center
coordinates (Kakino et al. 1987):
N ! N V N V
! ! ! ! ! !V !N !N
ρ ¼ w p t þ E TV w p t ¼ w p t w p t ¼ F p t F p w F p t F p w
! w !N !
p ^
E TV • w !Nt
E TV • b:
pt
(18)
In order to restrict the ball-bar length change and so obtain a small resolution at a
reasonable cost, the test is made of an interpolated circle so that a perfect machine
and setup will preserve the ball-bar length. The added ball-bar acts as an additional
passive axis with two ball joints which allow the test to be conducted using two or
three of a machine’s linear axes. Other test procedures have been proposed based on
the ball-bar to measure five of the six intra-axis errors (not the angular positioning
errors) of rotary axes (Zargarbashi and Mayer 2006; Lee et al. 2012) via the
measurement of volumetric errors.
Richer data sets can be gathered by using measuring heads equipped with
three or more proximity sensors (Weikert and Knapp 2004; Zargarbashi
and Mayer 2009; Bitar-Nehme and Mayer 2016; Bringmann and Knapp 2006)
to provide all three translational components of the volumetric error.
Touch probes have also been proposed to gather the necessary data (Lei and
Hsu 2002a). A richer data set is beneficial for separating causal error sources.
The disadvantage is that the machine must have redundant mechanical axes so
that the relative position between the ball and the sensing head is nominally
constant while the machine moves its axes. As a result, it cannot be used for two
or three linear axis measurement. Such approach is well suited to five-axis
machine tools.
In principle, any method that provides some metrological information
about the location of any part of the machine for known commands and/or
axis position readouts can be used for indirect measurement. For three-axis
machines, this usually takes the form of measuring artefacts such as arrays of
gauge blocks or step gauges. For five-axis machines, the concept of closed chain
is used. A ball-bar connects the tool location to the workpiece location; or
sensors such as an array of proximity sensors, a touch probe, or a noncontact
laser probe are mounted at the tool attachment location and used to gather data on
reference artefacts on the machine worktable (Abbaszadeh-Mir et al. 2002;
Bringmann and Knapp 2006; Lei and Hsu 2002b; Zargarbashi and Mayer
2009; Yang et al. 2015; Mayer 2012; Rahman and Mayer 2015; Ibaraki and
Knapp 2012; Schwenke et al. 2008). The separation of errors can be done in a
number of mathematical ways. Once a forward model containing the effect of
causal errors on the measurement is developed to calculate the predicted indica-
tions, the challenge then is to generate a sufficiently rich data set, the measured
indications, and to estimate the causal errors, the measurands, with sufficiently
small uncertainties as follows:
212 J. R. R. Mayer
X
arg min ðmeasured indication predicted indicationÞ2 : (19)
causal errors
The calculation can be done using a minimization algorithm such as the Levenberg-
Marquatz or solving the linearized equations generated by using small error approx-
imations. Understanding the particular reading patterns for specific causal errors may
also be used to quantify the causal errors.
Machine Tool Compensation
Lookup Table
Machine tool CNC provides numerous facilities to reduce systematic errors. Back-
lash, due to hysteresis, can be reduced by setting appropriate values in defined
registers, and numerous lookup tables are provided to reduce intra- and interaxis
errors. However, the correspondence between ISO-defined errors and CNC table
may not be straightforward. The simplest form of compensation is that to reduce the
axis positioning errors such as EXX, EYY, EZZ, EAA, EBB, and ECC which takes the
form of a table of incremental corrections. Straightness compensation may also be
achieved using slave axis compensation whereby, for example, small Z-axis correc-
tions are generated as a function of the X-axis to compensate an EZX error. 3D lattice
or grid may be defined within the X-, Y-, and Z-axis domain and 3D volumetric
errors measured at each node. The lattice can then be interpolated for any position
within the grid. More complex schemes involving mathematical functions operating
between tables are also available in modern CNCs.
Inverse Kinematics
Model-based compensation involves either using an error-enriched forward kine-

matic model to calculate the volumetric error followed by a linearized inverse model
to iteratively compute exact individual axis command compensations (Givi and
Mayer 2016) or using an error-enriched inverse kinematic model to directly calculate
commands (Yang et al. 2015). Such models require much computational power and
have been used as off-line procedures and not in real-time.
References
Abbaszadeh-Mir Y et al (2002) Theory and simulation for the identification of the link geometric
errors for a five-axis machine tool using a telescoping magnetic ball-bar. Int J Prod Res
40(18):4781–4797
Bitar-Nehme E, Mayer JRR (2016) Thermal volumetric effects under axes cycling using an invar
R-test device and reference length. Int J Mach Tools Manuf 105(Suppl C):14–22
Bringmann B, Knapp W (2006) Model-based ‘Chase-the-Ball’ calibration of a 5-axes machining

center. CIRP Ann Manuf Technol 55(1):531–534
Bringmann B, Knapp W (2009) Machine tool calibration: geometric test uncertainty depends on
machine tool performance. Precis Eng 33(4):524–529
Bryan JB (1979) The Abbé principle revisited: an updated interpretation. Precis Eng 1(3):129–132
Bryan JB (1982a) Simple method for testing measuring machine and machine tools Part 1:
principles and applications. Precis Eng 4(3):61–69
Bryan JB (1982b) Simple method for testing measuring machines and machine tools Part 2:
construction details. Precis Eng 4(3):125–138
Ekinci TO, Mayer JRR (2007) Relationships between straightness and angular kinematic errors in
machines. Int J Mach Tools Manuf 47(12):1997–2004
Everett LJ, Hsu T-W (1988) Theory of kinematic parameter identification for industrial robots.
J Dyn Syst Meas Control Trans ASME 110(1):96–100
Everett LJ, Suryohadiprojo AH (1988) A study of kinematic models for forward calibration of
manipulators. In: Robotics and automation, 1988. Proceedings of the 1988 IEEE international
conference on
Fan KC, Chen MJ, Huang WM (1998) A six-degree-of-freedom measurement system for the
motion accuracy of linear stages. Int J Mach Tools Manuf 38(3):155–164
Ferreira PM, Liu CR (1986) Contribution to the analysis and compensation of the geometric error of
a machining center. In: CIRP annals 1986: manufacturing technology, Annals of the Interna-
tional Institution for Production Engineering Research. Verlag Technische Rundschau,
Jerusalem
Florussen GHJ et al (2001) Assessing geometrical errors of multi-axis machines by three-dimen-
sional length measurements. Measurement 30(4):241–255
Gao W et al (2006) Measurement of multi-degree-of-freedom error motions of a precision linear air-
bearing stage. Precis Eng 30(1):96–103
Givi M, Mayer JRR (2015) Volumetric error formulation and mismatch test for five-axis CNC
machine compensation using differential kinematics and ephemeral G-code. Int J Adv Manuf
Technol 77(9):1645–1653
Givi M, Mayer JRR (2016) Optimized volumetric error compensation for five-axis machine tools
considering relevance and compensability. CIRP J Manuf Sci Technol 12(Suppl C):44–55
Grejda R, Marsh E, Vallance R (2005) Techniques for calibrating spindles with nanometer error
motion. Precis Eng 29(1):113–123
Ibaraki S, Knapp W (2012) Indirect measurement of volumetric accuracy for three-axis and five-
axis machine tools: a review. Int J Autom Technol 6(2):110–124
International vocabulary of metrology – basic and general concepts and associated terms (VIM 3rd
edition), JCGM 200:2012
ISO 230-1:2012(E). Test code for machine tools – Part 1: geometric accuracy of machines operating
under no-load or quasi-static conditions
Kakino Y et al (1987) The measurement of motion errors of NC machine tools and diagnosis of their
origins by using telescoping magnetic ball bar method. CIRP Ann 36(1):377–380
Knapp W (1983) Circular test for three-coordinate measuring machines and machine tools. Precis
Eng 5(3):115–124
Kruth JP, Vanherck P, De Jonge L (1994) Self-calibration method and software error correction for
three-dimensional coordinate measuring machines using artefact measurements. Measurement
14(2):157–167
Lee K-I, Lee D-M, Yang S-H (2012) Parametric modeling and estimation of geometric errors for a
rotary axis using double ball-bar. Int J Adv Manuf Technol 62(5):741–750
Lei WT, Hsu YY (2002a) Accuracy test of five-axis CNC machine tool with 3D probe-ball. Part I:
design and modeling. Int J Mach Tools Manuf 42(10):1153–1162
Lei WT, Hsu YY (2002b) Accuracy test of five-axis CNC machine tool with 3D probe-ball. Part II:
errors estimation. Int J Mach Tools Manuf 42(10):1163–1170
Mayer JRR (2012) Five-axis machine tool calibration by probing a scale enriched reconfigurable
uncalibrated master balls artefact. CIRP Ann Manuf Technol 61(1):515–518
214 J. R. R. Mayer
Okafor AC, Ertekin YM (2000a) Vertical machining center accuracy characterization using laser
interferometer Part 1. Linear positional errors. J Mater Process Technol 105
(Compendex):394–406
Okafor AC, Ertekin YM (2000b) Vertical machining center accuracy characterization using laser
interferometer Part 2. Angular errors. J Mater Process Technol 105(Compendex):407–420
Onat Ekinci T, Mayer JRR, Cloutier GM (2009) Investigation of accuracy of aerostatic guideways.
Int J Mach Tools Manuf 49(6):478–487
Pahk HJ, Kim YS, Moon JH (1997) New technique for volumetric error assessment of CNC
machine tools incorporating ball bar measurement and 3D volumetric error model. Int J Mach
Tool Manu 37(11):1583–1596
Qibo F et al (2013) Development of a simple system for simultaneously measuring 6DOF geometric
motion errors of a linear guide. Opt Express 21(22):25805–25819
Rahman MM, Mayer JRR (2015) Five axis machine tool volumetric error prediction through an
indirect estimation of intra- and inter-axis error parameters by probing facets on a scale enriched
uncalibrated indigenous artefact. Precis Eng 40(Suppl C):94–105
Schultschik R (1977) The components of volumetric accuracy. Ann CIRP 26:223–228
Schwenke H et al (2008) Geometric error measurement and compensation of machines–an update.
Slamani M, Mayer JRR, Cloutier GM (2011) Modeling and experimental validation of machine tool
motion errors using degree optimized polynomial including motion hysteresis. Exp Tech
35(1):37–44
Weikert S, Knapp W (2004) R-test, a new device for accuracy measurements on five axis machine
tools. CIRP Ann Manuf Technol 53(1):429–432
Yang J, Mayer JRR, Altintas Y (2015) A position independent geometric errors identification and
correction method for five-axis serial machines based on screw theory. Int J Mach Tools Manuf
95(Suppl C):52–66
Zargarbashi SHH, Mayer JRR (2006) Assessment of machine tool trunnion axis motion error, using
magnetic double ball bar. Int J Mach Tools Manuf 46(14):1823–1834
Zargarbashi SHH, Mayer JRR (2009) Single setup estimation of a five-axis machine tool eight link
errors by programmed end point constraint and on the fly measurement with Capball sensor. Int
J Mach Tools Manuf 49(Compendex/10):759–766
Accuracy and Performance Analysis of
Machine Tools 7
Andreas Archenti and Theodoros Laspas
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
Why Accuracy and Performance Analysis of Machine Tools? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 216
Basic Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 218
Error Sources in Machine Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
Introduction to Error Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 221
Geometric and Kinematic Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
Thermal-Mechanical Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222
Static (Quasi-static) Load-Induced Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223
Dynamic Load-Induced Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
Motion Control Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 224
Accuracy Evaluation of Machine Tools . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 225
Geometric and Kinematic Accuracy Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 227
Machining Test Parts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 238
Dynamic Accuracy: Dynamic Flexibility . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 240
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 243
Abstract
The key to solve manufacturing quality and productivity problems in the machin-
ing of parts is to understand the physical attributes’ geometric/kinematic, static,
dynamic, and thermal behavior of machine tools. In this chapter basic definitions,
error sources, and instruments and methodologies for the identification and
evaluation of machine tools’ physical attributes will be outlined.
A. Archenti (*) · T. Laspas

KTH Royal Institute of Technology, Stockholm, Sweden
e-mail: archenti@kth.se; laspas@kth.se

https://doi.org/10.1007/978-981-10-4938-5_7
216 A. Archenti and T. Laspas
The first section presents the background and answers “why” it is important to
measure and evaluate machine tools under no-load and loaded condition. Basic
concepts and definitions of metrological terms will be given. In the second part,
error sources in machine tools are introduced, and in the third part, instruments
and methodologies for the accuracy evaluation of machine tools will be given.
Keywords
Machine tool · Accuracy · Performance · Error sources · Analysis ·
Measurement · Instrumentation
Introduction
The importance of productivity increase and quality improvement in the manufactur-

ing of complex shape components with tighter tolerances and high surface accuracy
has led to the increased need for accuracy and performance evaluation of machine
tools.
Traditionally, the cutting process and the machine tool in a machining system
were designed and optimized to machine, with rather low cutting feeds and
cutting speeds, components made of conventional materials with relatively sim-
ple shapes. The geometric and dimensional accuracy of the components and
the surface finish were often in a feasible range without compromising produc-
tivity or product quality. One reason for this was that the machine tools often
used hade large masses and high damping because heavy construction elements
(columns, beams and castings, etc.) were used in their structures (Archenti 2011).
In recent years there has been an introduction of knowledge-based and smart
manufacturing in machining, where an increased understanding about workpiece
material properties and structural loading of the system and improved analysis
and design techniques have led to the development of new optimized machine
tool designs. Such designs in combination with high cutting feed and cutting
speeds have led to machining systems often excited by high-energy sources,
which can create intense loading conditions of the operational systems. One
example of this is vibration excitation problems as, for instance, modern machine
tool design results in systems with high stiffness and low mass (low inherent
damping).
Why Accuracy and Performance Analysis of Machine Tools?
To produce parts with required accuracy, the relationship between the machine tool
characteristics (physical attributes) and part accuracy/surface finish must be evalu-
ated in order to control deviations within required tolerances. Further, to uphold the
performance and functions of the machine tool, proper maintenance activities need
to be planned and executed. However, the machine tool has become a complex
mechatronic system consisting of more than hundreds of mechanical and electronic
7 Accuracy and Performance Analysis of Machine Tools 217
components, which are grouped into numerous subsystems. Components interact

and function together during the machining process thereby making the comprehen-
sion of the cause and effect of machine tool degradation at the component and
subsystem level complex. The importance of machine tool accuracy analysis and
performance evaluation becomes ever more important when considering the
very high precision requirements in machining and micromachining applications.
Technological advancements created a demand for parts and components of
increased precision that can achieve tighter fits and improved functionality. This
has shifted the notion of very high precision requirements away from miniscule and
microcomponents where tolerances are in a comparable order of magnitude to the
part dimensions, hence having large components having critical feature tolerances in
the microns or submicron level (Smith 2016). Achieving that level of accuracy
requires having deep knowledge of the machine performance in order to achieve
such precise and repeatable positioning.
One way to keep track of and to control the performance is to continuously
measure and quantify the ability of the machine tool physical attributes. The term
machine tool ability includes the individual and aggregated status of the machine
tool related to accuracy and the workspace availability through determining
functional ability of the machine tool structure (Sadasivam et al. 2018). The
functional ability is characterized by the variations in the performance as function
of the physical attributes’ geometric, kinematic, static, dynamic, and thermal
behavior of the machine tool. In other words, the machine tool ability is in second
step related to the workpiece characteristics through the machining system capa-
bility which is adding the effect of machining process. A machining system
described as a material processing unit where a machine tool structure, cutting
tools, the workpiece holding fixtures, and a workpiece interact with the cutting
process to produce a part.
Machining Systems, i.e., Interaction of Manufacturing Process

and Machine Tool
Machining system is a closed-loop system between the cutting process and the
machine tool elastic structure (Brecher et al. 2009). In Fig. 1 Fnom(t) is the cutting
force nominal value, Δd(t) is the total deviation of the relative displacement x(t), the
input of the machine tool elastic structure, F(t), is the instantaneous cutting force,
and the output x(t) is the relative displacement between the cutting tool and work-
piece. This relative movement affects chip thickness and depth of cut. Since the
cutting force is a function of these parameters, there is a feedback process in
operation.
Any change in force due to, e.g., a hard spot in the material, or material clearing
itself from a built-up edge on the cutting tool, will cause a change in the amount that
the machine deflects and hence in the depth of cut and chip thickness. This will lead
to a further variation in the cutting force and so on.
Dd(t) and D(t) are disturbances such as change in relative position between the
cutting tool and workpiece due to degradation and wear of guideways, tool wear,
thermal distortion of the elastic structure, variation of rigidity of the elastic structure
Dd(t) D(t)
Machine tool elastic
Fnom(t) F(t) structure x(t) Δd(t)
Including clamping device,
workpiece, tool, tool holder,
spindle system etc.
Variation in cutting
parameters
ΔF(t) Cutting process dynamics
Milling, turning, grinding etc.
Fig. 1 Closed-loop representation of machining system. The machining system represented as

the interdependence of two subsystems, the elastic structure of the machine tool and the cutting
process dynamics. The machine tool elastic structure forms the primary loop of the machining
system, while the cutting process dynamics is represented as a subsystem in the feedback loop
(Archenti 2011)
during a machining process, variation in the machine tool spindle ball bearing
characteristics, variation of cutting parameters, etc.
Machine Tool and Its Subsystems

As said in previous section, a machining system can be defined as a system
consisting of several physical modules mainly the machine tool and its associated
main modules (machine base/stand, spindle unit, ball screws, slides, measurement
system) and processes (e.g., milling and turning) linked to each other. Each module
of the system is essential in order to control the relative movement between cutting
tool and workpiece and to manufacture the desired shapes and features of the final
part. Every module may be comprised of other parts or subsystems, each having
certain attributes that determine and affect the outcome during their interaction.
Commonly, the main module of a machining system is the machine tool elastic
structure, which acts as the connecting link to the other modules.
As one of the fundaments of a machining system, machine tool capability has
a determinant role in the ability of the machining system to produce accurate parts
within the specified tolerances.
Basic Definitions
Before proceeding with describing machine tool errors, basic concepts and defini-
tions of metrological terms must be highlighted. The international vocabulary of
basic terms in metrology (JCGM 200:2012 2012) defines the terms accuracy,
precision, resolution, and uncertainty with respect to general metrological terms.
In the case of mechanisms and machines, with particular focus in machine tools, the
same terms are often attributed a differing meaning or cover a broader perspective,
relating to the machine performance and accuracy. In this context, the definitions
stated below are adapted to machine tools and mechanisms.
Accuracy, Repeatability, and Uncertainty
Accuracy
According to VIM, accuracy is the closeness of agreement between a measured
quantity and a true value of a measurand. In terms of machine tools, accuracy can be
defined as the maximum translational or rotational error between any two points in
the machine’s work volume (Slocum 1992) or, in the broader sense of a machining
operation, the closeness of the agreement between the actual value resulting from the
operation and a target value of the quantity. Accuracy is a qualitative description and
sometimes is referred to as “error.” In this sense, a machine tool can be said to be
more accurate when the deviation between nominal and the actual position in the
work volume is smaller.
Precision
Precision or repeatability, as it is oftentimes referred to, can be defined as the
closeness of agreement between successive measurements of the same quantity or
result of an operation (e.g., move of an axis) (Schellekens and Rosielle 1998) carried
out under the same conditions (repeatability conditions). When it comes to machine
tools, it can be understood as the error between a number of repeated attempts to
move the machine to the same position (Slocum 1992) under the same specified
conditions.
Resolution
Resolution refers to the smallest increment or meaningful change between
two indications of a quantity that can be detected or measured and causes a
perceptible change (JCGM 200:2012 2012). For machine tools, resolution usually
relates to the smallest mechanical step the machine can make during point-to-point
motion (Slocum 1992). As a general rule, resolution should be one order of magni-
tude better than the machine’s desired accuracy. Resolution is important because it
gives a lower bound on the repeatability that one could obtain if one really tried
(Slocum 1992).
Uncertainty
Uncertainty of measurement as a parameter characterizes the dispersion of the
resulting values of a measured quantity or an operation (e.g., motion to a specified
position) that can be attributed to this value with sufficient certainty. Some authors
refer to it as the maximum expectable error in a measurement with a probability
factor (Lopez de Lacalle and Lamikiz 2009). Uncertainty is a combination of many
components such as instrument uncertainty, environmental influences, uncertainty
due to the machine itself, as well as uncertainty due to incorrect usage or application
of the instrument or mechanism. It can be evaluated by statistical means, e.g.,
characterized by the standard deviation of the distribution of repeated measurements
or by assumed distributions based on other information, i.e., calibration certificates
or previous experience.
For machine tools, a typical example procedure of evaluating and reporting the
uncertainty for a linear axis is given in annex A of ISO 230-2 (2014), highlighting
the different contributing factors to the total uncertainty budget.
Functional Point
For testing of machine tool accuracy and specifically when referring to geometric
errors, it is important to state that these errors are defined and measured at the
position or in the trajectory of the functional point. The ISO 230-1:2012 (2012)
standard defines functional point as “cutting tool center point or point associated
with a component on the machine tool where cutting tool would contact the part for
the purposes of material removal.”
This is a single point that is attached to a machine tool component that can move
within the machine tool’s work envelope (see Fig. 2).
Although, measurement of axis geometric errors can be done anywhere in the
workspace, it is considered good practice the measurement point or trajectory to
coincide with the functional point, as it relates the measured error motion directly to
the geometric characteristics of the machine tool.
Additionally, offset between the functional point and the measurement point has a
magnifying effect to the uncertainty of the predicted accuracy, due to a lever effect
from the angular errors. Thus, minimizing the distance between the two points
minimizes the contribution to the uncertainty.
Applying measurement setups that represent the cutting tool’s trajectory allows
the measured errors to include the contribution of angular error (roll, pitch, and yaw)
of the axes.
b [µrad] ECX angular error motion (pitch)

10
5
0
-5
a +Y -10
Functional point 2 -15 µrad
-15
1000 mm X position
-20
Functional point 1 0 200 400 600 800 1000 1200 1400 1600 1800 2000 [mm]
+EXX [µm] EXX positioning error motion

+ECX
30
25
+X
20
+Z Functional point 2
15 15 µm
10
5 Functional point 1
0
-5
X position
-10
0 200 400 600 800 1000 1200 1400 1600 1800 2000 [mm]
Fig. 2 Functional points: (a) two functional points (point 1 and point 2) separated by the distance
of 1000 mm and (b) the same angular error yields different positing error motions (ISO 230-1:2012)
Error Sources in Machine Tools
Errors in machined parts are tightly associated with errors in the machine tool, and
usually they are identified as deviations of the actual cutting point from the nominal
or commanded position or trajectory in the workspace. These errors are related to the
design and physical attributes of the machine tool components, and their interaction
in multi-axis machines collectively determine the total part accuracy. Hence, regard-
less of how well a machine tool is designed and manufactured, these errors will
always impose a limit on the achievable machine tool performance.
Apart from the contribution of the machine tool components, errors in the
machine part can arise due to other causes such as fixture stiffness and clamping
force, the physical attributes of the part material and geometry, the cutting tool
geometry, wear and deflection, as well as the process itself. This allows categorizing
the errors in two broad types: (i) the errors attributed to sources in the machine tool
and its components and (ii) sources attributed to the cutting process. For the
evaluation and measurement of machine tool accuracy, it is deemed important to
determine and understand the major sources of error in a machine tool.
Introduction to Error Sources
Traditionally, machine tool accuracy is defined with respect to the accuracy and
repeatability in the movement of the machine axes that position the tooltip at the
specified workpiece position. However, errors induced due to machine tool stiffness,
thermal stability, and static and dynamic loads ultimately have an effect on the
machine performance; hence they are of interest to be measured and evaluated.
Machine tool errors can be classified into two categories: (i) the quasi-static type
which are slowly varying in time and are related to the machine tool structure
physical attributes and characteristics (geometry, thermal behavior and stiffness)
and (ii) dynamic errors which are dependent on operational conditions of the
machine tool (feed speeds, accelerations, jerk, spindle motion), process-related
vibrations and chatter, and dynamic stability. As such, the main error sources can
be categorized in five main sources (Schwenke et al. 2008):
• Kinematic/geometric errors
• Thermal errors
• Dynamic errors
• Static load effects
• Motion control and servo errors
In addition, a further distinction can be made into systematic (related to accuracy)

and random (related to precision) errors (Slocum 1992). Systematic errors are the
ones that can be measured and corrected by using sound engineering and metrology
methods and analysis. Random on the other hand are referring to the errors that do
not have a clear origin or are not repeatable at least with regard to how close one
looks to identify the origin of the error, the corresponding time frame, and the
utilized methods of measurement.
Geometric and Kinematic Errors
Geometric errors mainly relate to the design of the machine tool components
such as guideways, machine bed, bearings, carriages, lead screws, rams, etc. and
manufacturing imperfections in their geometry and form, as well as misalignment in
their assembly and wear in the couplings and joints (Schwenke et al. 2008; Sartori
and Zhang 1995). Due to their nature, geometric errors are affected by thermal
gradients and static loads (e.g., component’s own weight and workpiece weight).
They affect the relative position and orientation of the tool with respect to the
workpiece resulting in dimensional and geometric errors of the latter. They often
demonstrate a systematic behavior that varies slowly with time exhibiting good
repeatability, or they can have hysteretic and random behavior. In the context of
machine tools’ geometric accuracy, geometric errors are evaluated under quasi-static
or no-load conditions, meaning no operational errors (e.g., stick slip or other types of
vibrations) or other machine characteristics (speeds, feeds) are evaluated, and their
effect is kept as low as possible during measurement.
Kinematic errors are primarily concerned with errors in the trajectory of moving
points or average lines of motion of moving machine components (e.g., linear axes).
They are caused by misalignments of the machine components in the structure
assembly or due to their improper dimension and sizing. Typical misalignments
are squareness error between linear axis and parallelism error between the average
axis of rotation of rotary axes and the spindle. These errors are significant during
simultaneous movement of multiple axes performing linear or circular interpolation
motion, and they have a significant effect in the form accuracy of the machine parts.
Additionally, hysteresis and backlash errors in the axis motion also affect geometric
and kinematic accuracy.
Thermal-Mechanical Errors
Thermomechanical errors are one of the principal sources that limit machine tool
accuracy (Lopez de Lacalle and Lamikiz 2009). The term thermomechanical
relates to the variation of mechanical characteristics of a material with temper-
ature. For machine tools, this relates to geometric and dimensional changes of the
machine structure due to thermal expansion of the different components by
conduction or radiation of heat. Due to the transient thermal state of a machine
and the nonuniform expansion due to different thermal expansion coefficients of
structural materials, distortions and thermal stresses can result in location and
position errors. The generated thermal loads and temperature gradients can
originate from internal and external heat sources. Typical internal heat sources
relate to the operation of the machine tool and are identified to servo drives and
spindles, pumps, electrical systems, friction in ball screws and bearings, as well
as the process itself, with cutting chips carrying a significant amount of the
generated thermal loads and cooling fluids. With the exception of the process-
generated thermal variations, the rest of heat sources can be partially compen-
sated, maintaining the systematic nature of the geometric errors. Minimizing
thermal distortions in high precision machines can be achieved by, e.g., using
cooling channels for maintaining a stable thermal state of components (e.g., ball
screws) and by following design practices for isolating or constraining significant
heat or by designing structures with symmetry to reduce the effect of the thermal
expansions.
In addition to the internal heat sources, external sources relating to environmental
conditions, characterized by fluctuations of the air temperature from indoor climate
control (e.g., air-conditioning systems) and exposure to heat from surrounding
facilities or direct sunlight, can generate thermal gradients and stresses to
the machine structure, particularly to large machine tools that can be affected by
vertical temperature gradients (Weck and Brecher 2006). Environmental effects such
as thermal variations are difficult to be adequately measured, quantified, and com-
pensated; however for high precision machining applications, a controlled workshop
environment and floor planning can reduce or eliminate a significant amount of
these thermal errors.
Static (Quasi-static) Load-Induced Errors
Limited static stiffness of the machine structure can induce deflections of the
machine components causing errors in the relative position of the tool with respect
to the workpiece. Static and quasi-static loads and forces due to machine compo-
nents’ own weight, the workpiece mass, fixtures, and the cutting process create
deformations of the machine structure, which can be dependent on the machine axes
position influencing the machine geometry. As an example, consider the flexural
deformation of a linear axis guideways under the effect of the weight of the
workpiece and the corresponding slide. By moving the slide and workpiece, the
weight is repositioned thus changing the weight distribution on the machine struc-
ture. The effect of such static loads can be identified by measuring the geometric
errors of the machine tool. An important contribution to static load errors comes
from the joints and bearing interfaces in the structure, which can have a significant
effect on the machine tool rigidity particularly for machine of serial kinematic
structures. Static load errors exhibit a systematic nature and good repeatability
enabling their measurement.
Dynamic Load-Induced Errors
The structural loop of a machine tool is usually subject to dynamic effects that
influence its dynamic behavior, producing distortions in the tool path trajectory to be
realized. Such effects originate from varying forces such as rapidly changing cutting
forces during the machining process, inertial forces caused by acceleration or
deceleration of machine axes, as well as vibration-induced distortions, due to uneven
dynamic characteristics of the structural elements or tool wear. Since the dynamic
stiffness of the system’s structural loop changes under different machining condi-
tions due to the varying nature of these effects, deformations arising during machin-
ing are hard to predict and to compensate.
Self-excited vibration, or chatter as it is named when talking about metal cutting,
is induced by variations in the cutting forces (caused by changes in the cutting
velocity or chip cross section), built-up edge, metallurgical variations in the work-
piece material, stick-slip dry friction in sliding guideways, and regenerative effects.
This unwanted vibration has been found to be the main cause of poor quality surface
finish and reduced tool life, as well as poor machining accuracy. Chatter can be
considered an erratic phenomenon because it depends on the design and configura-
tion of both the machine and tooling structures, on workpiece and cutting tool
materials, as well as on machining regimes. The stiffness of the tool, spindle,
workpiece, and fixture are important factors for machining system static and
dynamic behavior. In this respect, cutting process through the cutting stiffness of
the workpiece material is also an important factor; for example, steels have a greater
tendency than aluminum to cause chatter. Cutting conditions, such as the depth of
cut, width of cut, and cutting speed, greatly affect the onset of chatter.
Motion Control Errors
The type and effectiveness of the control system has an influence on performance of
the machine tool. Motion control errors can affect the programmed axis position, the
achievable feed and speeds of the axis, and the simultaneous synchronous motion in
multi-axis machines.
Precise control of the servo drives used for positioning of the axes in the
programmed points in workspace can introduce errors in the contouring performance
of the machining system. Feedback from sensors and position encoders ensures that
the control system responds accurately and fast to required axis position changes by
adjusting the feed and is capable to achieve smooth accelerations avoiding over-
shoots in position, maintaining a dynamically stable motion. Stick-slip-induced
errors and servo response lag can be attributed to motion control errors.
In addition, interpretation of complex paths defined by the numerical control
(NC) commands and the accuracy in interpolation of each axis by precise adjustment
of the servo motors that provide motion to the axis has an effect in contouring
accuracy of the machine tool.
Measurement of motion control (or motion dynamic) errors is mostly done

with indirect measurement methods, and it is simpler to distinguish from
other types of errors.
Accuracy Evaluation of Machine Tools
Introduction
Machine tool accuracy analysis and performance evaluation is an essential task

during the entire life cycle of a machine tool: from its manufacture phase through
the purchase, installation and operational phase, to the end of life cycle. As discussed
previously, dimensional, form, and surface accuracy of a machined part is directly
affected by the machine tool performance and accuracy. Hence, one of the
key reasons for the performance and accuracy analysis of a machine tool is to verify
and assess the machine’s capability to produce parts according to the design toler-
ances and specifications. In addition, performance evaluation is vital for verifying
that the machine is capable to perform according to its production and quality
specifications as outlined or promised by the machine tool manufacturer. The latter,
is particularly important for machines utilized in a variety of part designs and
processes thus requiring of maintaining a stable overall accuracy and performance
over the entire work volume during their life. Moreover, an accurate machine effects
overall productivity, cutting tool, and machine components’ life and helps in reduc-
ing scrap as well as production and maintenance costs.
The general objective of accuracy and performance evaluation is to measure
and document all possible machine tool errors enabling the determination of their
sources, understanding of the behavior of physical components’ characteristics, and
quantify their influence in the overall machine performance. These components
range from guideways, bearings, spindles, frame, rotary axes, servo drives, etc.
Due to the inherent complexity of machine tools, evaluating and quantifying the
total accuracy and performance is deemed not a trivial task. Hence, no single
solution exists that can in one test assess the total machine accuracy.
A typical criterion for assessing machine tool accuracy is referenced to the
accuracy and repeatability in positioning of the tool center point (or cutting point)
with respect to the workpiece in the machine working volume. Based on this
characteristic, various national and international standards (i.e., ISO 230-1 (2012),
ANSI/ASME B5.54 (2005), etc.) have been developed providing guidelines on how
to evaluate positioning accuracy. However, the total accuracy of a machine tool
is dependent not only on the positioning capabilities of each axis (one-dimensional
accuracy) but also on the planar (two-dimensional) and volumetric accuracy (three-
dimensional), including the influence of thermal gradients that affect the machine
components, the stiffness of the different system components – individually as well
as collectively as elements of the machine tool structural chain – and the dynamic
stability under the effect of vibrations, e.g., from spindle rotation, cutting
process, or external sources. It is thus the volumetric accuracy inside the entire
workspace – under the effect of all the error sources –that will ultimately determine
the machine performance and capability.
To this extent, various measurement instruments, methods, and systems have
been developed to facilitate and enable measurement of machine tool accuracy
relating to measurement of deviations of single degree of freedom in an individual
axis or simultaneous measurement of the deviations under the aggregated effect of
multiple axis or machine moving components. In the following paragraphs, some of
the instruments and methods will be presented in relation to the metrological
reference each method is based upon and the type of errors that can be measured.
Direct and Indirect Measurement

The most suitable measurement and analysis method for evaluating machine tool
accuracy depend on the purpose and type of error to be measured and the respective
machine component and the configuration in the machine structural loop. Measure-
ment methods can thus be distinguished into direct and indirect, in relation to the
provided data of different aspects of the machine tool.
Direct measurement allows a single mechanical error of an individual machine
component to be measured excluding contributions of other components (e.g., axes).
A classification based on the metrological reference or the principle way of carrying
a measurement can be done, identifying three subgroups: material-based methods
that use artifacts (e.g., straightedges, precision mandrels, calibrated scale systems
and artifacts), laser-based methods that use the laser light wavelength and linear
propagation as a reference (e.g., laser interferometer, autocollimator), and gravity-
based methods (e.g., inclinometers, spirit levels, etc.) that use as reference the earth
gravity field. One of the most commonly used methods for geometric error mea-
surement is laser interferometry, which, combined with special optics, is suitable for
measuring almost every geometric error of a linear axis.
Indirect measurement methods on the other hand detect superposed errors asso-
ciated with simultaneous motion of two or more machine axes. They provide a quick
way for assessing the positioning capability and contour accuracy of a machine
partially or fully identifying all the geometric errors of the machine tool axis. As with
direct methods, a variety of measurement techniques are available including the use
of artifacts (calibrated or uncalibrated) and contour measurements (straight lines in
3D space or circular paths utilizing two or more axes) (Schwenke et al. 2008).
Classification of measurement methods into direct and indirect can also be done
according to the immediacy of the application of the measurement method to the
machine tool or component of interest. In this regard, “direct” methods account for
all error measurement methods applied directly on the machine tool and provide
measurement data involving one or more axes of motion. Direct methods can be
further classified into operational and off-operational depending on the number of
axes involved in the measurement process and their cumulative contribution. In this
regard, measurement of a single axis constitutes off-operational direct measurement
method, as it does not represent simultaneous motion of different machine tool axes.
“Indirect” methods, on the other hand, obtain error data by measuring the output of
the system on an interim object. Measurement results provide data for making
inferences about the state of the machine and the attributes or errors of interest.
The most common and typical example of an indirect measurement according to this
interpretation is machining of a test piece (e.g., according ISO 10791-7 (2014))
which links machined part dimensional and form errors to the machine tool errors
enabling the assessment of the machine tool performance.
Geometric and Kinematic Accuracy Evaluation
In machine tool performance evaluation, a significant portion of the achievable

accuracy can be attributed to geometric and kinematic errors. As discussed in section
“Error Sources in Machine Tools,” geometric errors are related to the dimensional
and geometric accuracy of the components comprising the machine tool structure
and to misalignments in their assembly. In addition, temperature gradients and
reduced stiffness of the machine components can affect their geometry, which
influence the machine tool positioning capability.
Test methods and instruments for geometric error measurement and evaluation
have been developed and established over several decades and are documented by
national and international standards. A notable example is the international standard
ISO 230-1 that provides general guidelines for the evaluation of geometric accuracy
of machines under no-load or quasi-static conditions and covers a wide range of
methods and techniques applicable for many different machine tool types. Several
different techniques and instrumentation exist, covering both direct where a single
error parameter of an axis is measured and indirect methods suitable for quick
performance evaluation and periodic health checks of a machine tool condition.
Laser Interferometry for Positioning and Geometric Accuracy

Geometric errors in machine tools can be identified as the linear or rotational
positioning accuracy of the axis of motion (for a linear or rotary axis, respectively),
straightness errors, angular or tilt errors, out-of-squareness between two axes (linear
or rotation axes), out-of-parallelism, and offset errors. In a three-axis machine tool,
for example, there are 21 geometric errors, 6 for each linear axis and 3 interaxis
errors or link errors, which are systematic and can be measured using different
techniques.
Laser-based measurement systems are the most typically used and well-
established instruments for geometric error measurement. Specifically laser interfer-
ometry offers a highly precise, reliable, and traceable measurement of the errors
relying on the frequency of a stable, monochromatic, coherent laser source as
reference. Combined with modules of special optics (reflectors, beam splitters,
etc.), laser interferometers are suitable for measuring almost every geometric error
of a linear axis of motion as well as some of the geometric errors of rotary axis.
A typical laser interferometer is used for the determination of a single error
parameter of a single degree of freedom direction (one linear axis of motion). The
Fig. 3 Illustration of (a) setup for linear positioning accuracy and repeatability measurement
(Image courtesy of Renishaw plc), (b) principle of measurement of linear positioning accuracy
(Image courtesy of Renishaw plc)
most common measurement corresponds to the evaluation of the positioning accu-

racy and repeatability of a linear axis of motion (see Fig. 3) as described in
ISO 230-2:2014. It is important to note here that even though positioning error is
part of the geometric accuracy of the machine tool, in reality it is dependent on the
geometry of the axis components, the drive system, and the numerical control and its
feedback loop.
A typical measurement sequence for a linear axis starts by defining the number
of measurement target positions, which usually are equally spaced and distributed
along the entire length of the linear axis. The optical reflector, the linear interferom-
eter, and the laser beam must be aligned to the machine axis line of motion.
Misalignment of the optical components with the laser beam will introduce errors
in the measurement and will increase the uncertainty of the result resulting in
erroneous evaluation of the axis accuracy. Five repetitions are preformed (according
to ISO 230-2) in both directions of motion (forward and reverse direction of
approach of each of the targets) in order to identify the effect of error such as
mechanical backlash of the axis. Statistical analysis of the measurement results
yields figures of bidirectional repeatability, reversal error, and bidirectional position-
ing error of the axis under test. The same statistics can be calculated for each target
position.
An important note to consider during evaluation of linear positioning errors of
an axis is the fact that the measured positioning errors are influenced and include
contributions from the angular errors of the axis of motion (i.e., in the case of moving
along X-axis, rotations around Y-axis (pitch), will have contributed in the position-
ing error of X-axis). In case the measurements are performed at a different trajectory
than the trajectory of the functional point, the effect of the angular errors has to be
taken into account in order to compensate for the induced deviations.
In a similar manner, using suitable optics and following a testing procedure as
described for positioning error measurement, most of the geometric errors of a linear
axis can be identified. For straightness error measurements (in the horizontal and
vertical plane with respect to the nominal line of motion of the axis) a straightness
Fig. 4 Illustration of (a) setup for straightness measurement and (b) principle of measurement of
straightness error (Image courtesy of Renishaw plc)
interferometer is used instead o f a beam splitter in combination with a straightness

reflector (see Fig. 4b).
Rotational positioning accuracy of axis of rotation (e.g., rotary tables)
can also be measured and evaluated using laser interferometer combined with a
precision indexing table and an angular reflector providing a level of accuracy of
1 arc second. The resulting positioning error of the rotary axis is calculated by
comparing the rotary axis position to the indexing table position and the laser
interferometer reading.
Despite their capability for highly precise and repeatable measurements, laser
systems are sensitive to changes in the environmental conditions. Hence, care must
be taken while performing interferometric measurements by monitoring the temper-
ature, humidity, and air pressure in order to account for disturbances. Many modern
laser systems have a built-in capability to automatically monitor and compensate for
environmental variations.
Once the positioning and geometric errors have been measured, corrective actions
can be taken in the form of compensation tables in the machine tool controller or by
mechanical adjustment of, e.g., the machine tool base (particularly true for medium
to large machines) and axis guide rails. In addition, fitting the measured geometric
and positioning error parameters to machine tool-specific kinematic model enables
the volumetric accuracy evaluation.
Lately several laser interferometer systems have been introduced which are
capable of measuring several error parameters of a linear axis simultaneously. The
advantages such systems offer are significant reduction in measurement time and
setup complexity, reduction in setup errors, and reduction of thermal drift influences
on the measured error values. Several commercial systems exist that are capable of
measuring five and even six degrees of freedom error parameters of a linear axis
(positioning error, horizontal and vertical straightness, pitch, yaw, and roll) simulta-
neously in one setup.
Circular Test
The basic idea of a circular test is to run a circle path, and as the test proceeds, all
deviations from the base circle are registered (Archenti et al. 2012a). An error-free
machine tool results in a perfect, circular path. Circular test methods measure
the changes in the distance between the spindle nose, or the end of the tool, and
the center of a circle on the table. The accuracy of the motion is evaluated from
the motion error traces, which can be diagnosed by analyzing the traces. The
measurement offers a high amount of detailed information due to the fact that the
most common errors distort the test path in a mathematically definable way (Knapp
1983a).
Many different test methods based on circular test principle are available. The
cross-grid encoder (Fig. 5a) is a device that makes it possible to perform all kinds
of free-form tests in a plane, including circular tests (Heidenhain GmbH). Then there
are laser-based measuring systems like laser tracking, multifunction laser interfer-
ometer (Schneider 2004), laser ball bar (LBB) (Ziegert and Mize 1994), and laser
circular test (Optodyne Inc. 2018). Another way of performing a circular test is by
the use of a scanning probe and a master artifact (ISO 10791-7:2014). Both test
pieces and ordinary workpieces can be evaluated by the use of coordinate measuring
machines (CMM) (Lopez de Lacalle and Lamikiz 2009).
The double ball bar (DBB), based on circular test (Archenti et al. 2012a), is
a device that can be used for quick tests of machine tools, developed in the early
1980s (Bryan 1982), and it is adopted by ISO 230-4 and ANSI B5.54-2005 as an
instrument for circular test (ANSI/ASME B5.54 2005; ISO 230-4:2005). This
ensures unified practice in ball bar measurements and makes it possible to reliably
find many machine tool deviation types out of this simple measurement. DBB is the
most common circular test method for machine tools and can be used to analyze
unloaded machine tool structures. The device is based on a very accurate linear scale,
which measures changes in radial direction during a circular motion (see Fig. 5b).
Fig. 5 Example of instruments for performing circular test: (a) cross-grid encoder (Image courtesy of
Dr. Johannes Heidenhain GmbH) and (b) telescoping double ball bar (Image courtesy of Renishaw plc)
The evaluation of the results is well studied, and many of the most significant
formulae can be found in the literature.
Artifact Measurement
Assessment and measurement of machine accuracy using calibrated master artifacts
and calibrated test pieces of known dimensions has been utilized for several years
prior to the development of laser measurement instruments. Examples such
as straightedges for the evaluation of straightness of a machine axis (see Fig. 6b),
reference precision squares and dial indicator for assessing of out-of-squareness of
linear axes (see Fig. 6a), test mandrels with dial gauges for spindle runout, and
parallelism evaluation have been standardized (e.g., ISO 230-1 and ISO 10791-2)
and used to directly measure machine tool errors.
Utilizing the on-machine probing ability of modern machine tools using touch-
trigger probes and noncontact or optical probing systems, more complex artifacts
have been possible to be used, whose calibrated feature’s coordinates can be
measured enabling the quick assessment of positioning performance and geometric
accuracy of machine axes and their calibration or compensation. As such, a wide
variety of master artifacts have been developed such as 1D ball arrays, 2D and 3D
ball plates (see Fig. 7), 3D artifacts with various features (e.g., holes, spheres, slots),
and special-purpose test pieces or calibrated production parts with features suitable
for identifying and determining the geometric accuracy of the machine tool. Usage
of artifacts provides a fast method for interim checks and periodical evaluation of
machine performance due to the ease of use and potential low machine downtime. In
addition, thermally stable calibrated test pieces are a suitable method for evaluating
thermal drifts and stability of machine tools.
Fig. 6 Illustration of (a) precision granite square for the evaluation of out-of-squareness (Image
courtesy of Microbas Precision AB) and (b) straightedge beam for the evaluation of straightness of a
linear axis of motion (Image courtesy of Microbas Precision AB)
Fig. 7 Example on artifacts for measurements: (a) A 2D calibrated ball artifact with precision spheres
(Image courtesy of IBS Precision Engineering) and (b) A 2D calibrated ball plate used in a hybrid 3D
ball measurement (Bringmann et al. 2005; Bringmann and Knapp 2009; Liebrich et al. 2009)
Proper usage of the calibrated artifacts requires the correct setup and alignment
on the machine table, probing of the control features (e.g., balls or holes), and data
analysis to identify the machine geometric errors, accuracy, and repeatability.
By adopting proper measurement strategies, some of the artifacts can extend
their potential usage. For example, the 2D ball plate shown in Fig. 7b contains
36 calibrated spheres placed in grid arrangement of mesh pitch size calibrated by
a CMM. By using spacers with precision kinematic couplings, the 2D ball plate can
be located in different planes of the machine work volume enabling the evaluation of
the volumetric accuracy of the machine tool and identification of geometric and
positioning errors of the axes including the rotary table.
The same strategy utilized with the 2D ball plate approach can be implemented
with a grid encoder using an optical probing system (see Fig. 5a). The grid plate can
be perceived as a calibrated artifact with the advantage that it is up to the user to
define the pitch length and the number of distinct locations to be measured. Of
course limitations due to the grid plate dimensions might limit the work area
available for evaluation but offer the flexibility to select a type of form or toolpath
to measure.
Test artifacts and calibrated test pieces require regular calibration (e.g., by
a CMM) to ensure their dimensional accuracy and repeatability and to maintain a
low measurement uncertainty. In addition, even though they are considered for rapid
evaluation, it can oftentimes take significant amount of time to set up and make a full
measurement which might make their use not desirable.
Volumetric Positional Accuracy

In ▶ Chap. 6, “Machine Tool Calibration,” volumetric errors have been defined as
the relative difference in position and orientation (jointly considered as pose)
between two coordinate systems. One coordinate system is attached to the tool
side, and one is attached to the workpiece side, thus describing the relative
difference in pose between the tool and workpiece. The volumetric accuracy of a
machine tool is the absolute minimum and maximum of the relative differences in
position and orientation in the concerned volume of the machine tool (ISO 230-1
2012). As the units for the translational and rotational components are non-
homogeneous, e.g., a positional volumetric error of 10 μm and a rotational
volumetric error of 10 mas cannot simply be added, one might want to talk
about volumetric accuracies, in plural. However, one usually talks about the
volumetric error of the TCP (as the TCP is a point, it has no orientation); it is
possible to transform the rotational volumetric error into an additional positional
volumetric error component, thus describing the volumetric error solely as 15 μm.
This way of expressing the volumetric error appears to be much closer to one’s
intuition. Thus, one possible way of expressing the volumetric accuracy is to say
that it varies in the interval of [5 μm, 15 μm] or it varies on the intervals of [3 μm,
10 μm] and [4 mas, 10 mas] (the values are arbitrarily selected, however in a
reasonable range).
Because there is no easy way to relate this figure to the ability to comply with
GDT and one would need to consider the importance of sensitive and insensitive
directions, the authors consider the informative value of the volumetric accuracy to
be debatable.
Now that we have established a common understanding of what the volumetric
accuracy of a machine tool is; the question on how we can measure and identify it
needs to be addressed. Please heed that there is no designated instrument for the
measurement of the volumetric error of a machine tool. That there is no such
equipment could be due to the varying importance of this information as well as
the varying complexity of its measurement. The volumetric error is the result of the
errors of the structural loop components of any machine tool. Therefore, it is
common to directly measure the individual errors (inter- and intra-axis errors,
▶ Chap. 6, “Machine Tool Calibration”) and model the machine tool’s kinematics
such that the volumetric error can be evaluated. Inter- and intra-axis errors are also
referred to as component and location errors (Smith 2016).
A non-comprehensive overview of metrological instruments, which can be used
for the identification of the inter- and intra-axis errors, thus the volumetric accuracy
of a machine tool can be found in Table 1.
The instruments are grouped according to their maximum measurement range,
accuracy, and positional as well as rotational identification capability of component
errors; one should notice that all systems employ laser-based measurement
principles.
There exists other measurement equipment to identify component and location
errors of machine tools such as the Leica AT-960 ® or the FARO Vantage Laser
Table 1 Measurement instruments for the identification of component and location errors of
machine tools
Name Measurement range [m] Accuracy [μm/m] DOF []
Renishaw XM-60 ® (2016) 10 1 6
Renishaw XL-80 ® (2018) 40 1 5
Etalon LaserTracer-NG® (2018) 20 0.2 6
API XD Laser ® (2018) 45 0.2 6
SIOS SP 15000 C Series ® (2018) 15 1 5
MEAX MT30 ® (2018) 3 1 5
Tracker ®. However, this measurement equipment is not specifically designed for this
purpose; thus they are not enlisted.
Other Performance Tests, Spindle Analysis, and Compliance

Several diagnostic instruments falling under indirect measurement methods that
utilize master artifacts, laser sources, or displacement sensors (contact and capaci-
tive) exist for evaluating machine tool performance in terms of static and quasi-static
accuracy and dynamic motion accuracy. Particularly the ones that cover aspects of
errors due to thermal effects, spindle performance, and rotary axis accuracy and
are considered important in enhancing the understanding of machine tool’s accuracy
are presented here, without meaning that several other existing methods are not
of similar importance or less capable for performance evaluation.
Rotary Axis Evaluation

Since five-axis machine tool utilization in manufacturing is becoming more and
more popular, the geometric and kinematic accuracy and performance of machines
that employ both linear and rotary axes has been in the focus. This increase in the
machine’s axes affects kinematic complexity, thus requiring enhanced knowledge of
the rotary axes positional and motion accuracy (rotation speed stability) as well as
identification of their alignment accuracy (location errors) with respect to linear
machine axes and the machine coordinate system since location errors are deemed as
fundamental factors in the five-axis kinematic accuracy. In addition, assessing the
ability of the machines to control the simultaneous interpolation of rotary and linear
axes is necessary in achieving the required path accuracy and precision.
The international standards ISO 230-1, ISO 230-7 (2006), and ISO 10791-6:2014
(2014) provide guidelines and define testing procedures for evaluating positional
accuracy of rotary axes and accuracy of circular trajectories by simultaneously
interpolating two linear and one rotary axes or all five axes (three linear and two
rotary axes), for machine types with axes of rotation on the tool or workpiece side.
For this purpose, a test method called R-test has been developed which can measure
the deviations from a programmed circular path and quantify a machine’s perfor-
mance by static and dynamic measurements. The system consists of a precision
master ball and a measurement head equipped with contact displacement sensor
(Fig. 8b) or capacitive sensors (Fig. 8a), configured in orthogonal or inclined
Fig. 8 (a) Implementation of R-test measurement system (Image courtesy of IBS Precision
Engineering). (b) Prototype R-test device with contact displacement sensors (Ibaraki et al. 2011)
position, which measure the deviation of the tool (in this the probe head) relative to
the workpiece (master ball) in three coordinate directions. With a static measurement
approach, the actual location (offsets in the two orthogonal directions) of the axis can
be determined in the machine coordinate system as well as the axis squareness with
respect to the normal coordinate plane, by moving in incremental steps of the rotary
axis through its entire range of motion, with the linear axis following the motion.
Similar evaluation can be performed with a dynamic measurement approach, where
the axes are moving at a constant feed rate. This way additional information about
the dynamic performance of the machine can be determined regarding its interpola-
tion accuracy and following errors of the axis of rotation. Dynamic testing method
can further evaluate the machine five-axis contouring accuracy and dynamic motion
characteristics by interpolating all machine axes synchronously at constant feed rate
and measure the deviations from the programmed tool point trajectory.
Due to the simultaneous motion of the linear and rotary axis during the R-test
procedure, it is essential to verify prior to testing that the linear axes influence to
the measured deviations are small and their contribution is minimal.
Spindle Analysis
The spindle is one of the most important and complex machine components. It is
the element that holds the cutting tool or workpiece (in case of turning) and links it to
the machine structure and the axes motions. Therefore, apart from its typical
performance characteristics of power and torque and their relation to rotational
speed, other characteristics like accuracy, reliability, stiffness and thermal stability,
and speed range capability (Stephenson and Agapiou 2016) are of importance for
ensuring machine performance and high quality of machined parts. Methods and
tests for the evaluation and identification of these properties are covered extensively
in ISO 230-7 standard that provides guidelines for the evaluation of the motion
errors and speed-induced shifts of axis of rotation and in ISO 230-3 for thermal
distortions and drifts from rotating spindles. Angular positioning accuracy of the
axis of rotation is covered by ISO 230-1 and has been briefly discussed in section
“Laser Interferometry for Positioning and Geometric Accuracy.”
A typical measurement method for spindle performance analysis utilizes a preci-
sion test artifact that can also incorporate one or two spherical features located at
known positions in combination with high precision capacitance sensors mounted on
a specially designed fixture (see Fig. 9b). The number of sensors used can vary from
three to five; however a minimum of three sensors is necessary for identifying radial
and axial errors of the spindle. Test is performed at different spindle speeds for
evaluating the radial and axial errors’ motions of the spindle. Additionally, it
is possible to identify and evaluate synchronous and asynchronous tilt error motions
of the spindle using two sensors for each direction (X and Y) and positioned at
known distance. Measurements are usually performed at three different spindle
speeds (10%, 50%, and 100% of maximum spindle speed). Moreover, the methods
enable the evaluation of thermal drifts of the spindle as part of the performance
evaluation. It is important to highlight that most of the identified errors from the
spindle analysis method can be directly associated to machine-related error sources
Fig. 9 (a) Spindle analysis with two spherical artifacts (Image courtesy of Lion Precision).
(b) Instrument for spindle error analysis with one sphere and three sensors (Image courtesy of
IBS Precision Engineering)
(e.g., control system and spindle bearings) or errors on the machined part, i.e.,
roundness error of bore, surface finish, cylindricity of deep holes, etc.
Machine Tool Compliance and Hysteresis Tests

Static compliance (inverse of static stiffness) of machine tools is a characteristic that
affect the machine behavior under the effect of the cutting forces and inertial loads
and can directly affect machined part dimensional and geometric accuracy. ISO
230-1 provides guidelines for testing and evaluating compliance and hysteresis of
a machine tool when a static load is applied between tool and workpiece. This testing
procedure provides a simplified measure of the machine’s static compliance and an
indication for measuring using more detailed methods. It is important to highlight
that compliance might vary as function of axis positions and force direction; hence
testing at different positions would provide a more thorough identification of the
static machine behavior. A typical example of the test can be seen in Fig. 10a where
a load cell and linear actuator are used in combination with a probe that measures
spindle deflections. Measurements are performed close to the center of the
workspace volume for each linear axis separately. For rotary axis, a similar principle
is followed; however, the force application point is offset radially from the axis
center of rotation (see Fig. 10b).
Following similar approach as the one described in ISO 230-1, researchers have
been evaluating machine tools’ static stiffness utilizing force actuators (hydraulic,
electromechanical), pressure and force sensors, dynamometer, and displacement
sensors for evaluating static compliance of the machine structure in three-
dimensional Cartesian space (e.g., see (Gao et al. 2016)).
A more elaborate testing device that combines the circular test ability (see section
“Circular Test”) for evaluating geometric, kinematic, and contouring accuracy of
Fig. 10 Setup for machine static compliance and hysteresis by externally applied force (a) using a
load cell (3), differential screw (2), and a probe (4) mounted on the table (5) measuring the
deflections at the spindle (1) and in (b) using load cell (2) displacement sensor with a special
fixture (4) mounted on the rotary table (5). Here R is the radial offset from the axis center of rotation,
and h is the offset of the point of deflection measurement from the point where the force is applied
machine tools with the capability to apply a variable load between machine tool table
and the toolholder called loaded double ball bar (LDBB) has been presented in
(Archenti et al. 2012b; Archenti and Nicolescu 2013). The systems enable the
evaluation of the machine tool equivalent stiffness in the circular trajectory by
capturing the aggregated deformation of the machine tool (due to the kinematic/
geometric errors and load-induced deviations).
Thermal Effects and Drift

Temperature variations and thermal loads originating from internal or external sources
affect all machine tool components influencing positioning and volumetric accuracy.
Standardized test procedures for evaluation of these effects on machine tool perfor-
mance are specified in ISO 230-3 (2007). The tests are designed to capture the effects
of the thermal expansions of the machine components due to thermal loads generated
by the moving machine axes (linear and rotary) and from the environment by measur-
ing the relative deviations between the tool and the workpiece, hence evaluating the
influence in the positioning accuracy of the machine. Suitable metrological instrumen-
tation can vary from laser interferometers to capacitive or contact displacement sensors
in conjunction with test mandrels, or other artifacts combined with temperature sensors
(e.g., thermocouple) can be used. In particular laser interferometers, spindle analyzer or
rotary axis analyzer following specifically designed measurement steps can be used for
evaluating thermal distortions and temperature effects from environmental sources,
linear axes, and rotating spindles.
Machining Test Parts
Machining of test parts can be the most representative method for machine
performance evaluation, since it will capture an operational view of the machine
under test including the effects due to machining forces even though the tests are
performed under finishing conditions. The objective of the test is to evaluate the overall
machine performance including geometric accuracy, thermal stability, static stiffness
and dynamic characteristics and the interaction of the machine tool and workpiece, by
using a CMM to measure the produced workpiece features’ accuracy and associate
them to machine capabilities. The method is simple and can be implemented in a
variety of machine tools; however good planning is required since the workpiece size
and material, cutting tools, process parameters, and machining location will all affect
the test outcome. In addition the level of precision when measuring the part is quite
high, and the one-to-one association of the measured features is not always easy since
confounding and correlations between the machine error parameters are existing.
Two well-known test pieces are the standard NAS 979 (see Fig. 11) and
the respective ISO 10791-7. The NAS 979 test piece was designed by the Aerospace
Industries Association (AIA) intended as a uniform cutting test initially for multi-axis
machine tools with indexable milling heads in mind. The test piece has various features
that each describes certain machine performance characteristics including the perfor-
mance of simultaneous five-axis machining. Some of the cutting tests in the NAS
7
Accuracy and Performance Analysis of Machine Tools
239
Fig. 11 Artifacts for testing of machine tools (a) NAS 979 and (b) specifications with tolerances of the test piece according to ISO 10791-7
979 standard are also included in ISO 10791-7, which provides guidelines and
tolerances of the finished test pieces. The standard includes tests for evaluating
machining performance, positioning, and contouring accuracy associated with the
geometric properties of interpolating one or two axes or the simultaneous five-axis
interpolation of machining centers.
Test Specimens for Deformation Testing

A conventional approach for testing machines under loaded conditions (see section
“Machining Test Parts”) is to produce some specially designed standard specimens,
which are designed to generate predefined forces during the machining operation. The
parts are then measured and evaluated, in order to draw conclusions of the behavior of
the machine tool. The disadvantages with this type of test methods are that expensive
tools and parts are needed and that the results depend on the condition of these. The
results from different tests are not comparable if not exactly the same test specimens,
tools, and process parameters are used.
An example of this type of test is the so-called BAS (Bofors AB, Alfa Laval, ASEA,
and Saab-Scania) machine tool test (Asea et al. 1970). The goal of the BAS machine test is
to evaluate machine tool stiffness under real cutting conditions without the need for a
complex cutting procedure in a simple and predefined way. However, assessment of the
result is relied on rough guidelines and experience of machining experts and testers. In
addition, as mentioned above, it requires the supply of the specific material, tool, and
expertise of given machine tool configuration and machining system behavior. However
unlike other methods, it has the advantage that it takes into account the response of the
entire machining system (including servo drives, linear guideways and guide rails, control
system response, etc.) when evaluating the stiffness behavior under different loads.
The BAS test can be done for different machine tool configurations, e.g.,
turning and milling machines. In case of a conventional milling machine, the
test workpiece is milled (climb milling) along the entire long side, the direction
of feed is reversed, and the cutter is allowed to run out, being withdrawn from the
surface halfway along the workpiece. The undercut in each direction is taken as a
measure of the deflection (measured in radial direction milling in the directions
of X, Y, and Z). The workpiece is mounted in that way, so the surface that will be
machined coincides at half of table radius (applicable for horizontal machining),
and in case of vertical machining, the workpiece should be placed in that way, so
the middle of it coincides at half of the table radius. In case of swiveling milling
head, deflection should be measured in both horizontal and vertical plane to get a
more complete view of the stability of the system. The test can be done for both
positive and negative feed directions. The results from the BAS test are used to
check if the machine is within the acceptable limits of deflection and so accurate
enough to be used for production (Fig. 12).
Dynamic Accuracy: Dynamic Flexibility
The dynamic accuracy of a machine tool structure can be quantified trough the
dynamic flexibility of the systems. As the dynamic stiffness is, generally, lower
Fig. 12 Illustration of BAS test procedure. In (a) down-milling configuration of the test piece is
performed in feed direction (1). Once the tool exits the material, next step is (b) reverse the feed direction
to direction (2). The cutter is allowed to run to the halfway of the test piece side (3) which exits the
material. The undercut Δt is measured indicating the machine stiffness in the direction normal to the
cutting feed
than static stiffness, it is desirable to take it into consideration when designing a

machine tool or when selecting cutting parameters (Österlind 2013). The dynamic
stiffness, Kdynamic, is characterized by the coefficient, Kdynamic = kdynamic/kstatic. It
is generally dependent on frequency and/or amplitude and Kdynamic < 1 for most
structures.
Dynamic stiffness is defined by the vibrating mass, static stiffness, and damping
of the system. Damping is the ability of transforming mechanical energy of
a vibratory motion into other forms of energy (generally heat). Damping in a
mechanical system decreases the intensity of vibratory motion, close to resonance,
and can therefore be used in machine tools to prevent/suppress vibrations. When
machining with a rotating tool, the cutter teeth are repeatedly entering and exiting the
workpiece. This creates impact forces (as the kinetic energy of the impacting mass is
transformed into potential energy in the form of elastic deformation of the tooth)
that, with insufficient damping, lead to tool vibration and bad surface quality of
the machined part.
Experimental modal analysis (EMA) by impact testing with accelerometers is
the most common way of estimating the dynamic properties of machine tool
structures. EMA through the so-called impact testing on the cutting tool – machine
tool and workpiece – is a rather fast method of estimating the dynamic response of
a machining system at its current state and at standstill, i.e., non-rotational cutting
tool. In general a few accelerometers are attached to the measurement object,
which is excited by an impact hammer instrumented with a force sensor, and the
frequency response function (FRF) of the structure is estimated; see Fig. 13 for a
typical setup for estimation of the dynamic properties of the cutting tool tip. For
larger structures, shakers, instead of impact hammers, can be used to excite the
structure.
b
970e-9 1.00
Synthesized FRF
F
Point1:+Y/Point2:+Y
F
Synthesized FRF
Point2:+Y/Point2:+Y
Synthesized FRF
F
Point3:+Y/Point2:+Y Amplitude
Amplitude
(m/N)
0.00 1332.58 2235.73 0.00

0.00 Hz 4000.00
Fig. 13 (a) Impact hammer and cutting tool (clamped in the toolholder) instrumented with
accelerometers in a typical setup for modal analysis of the tool. (b) FRF displaying identified
natural frequencies when tool–toolholder is attached in spindle. The figure shows the measured
response in three positions on the system tool–toolholder
References
ANSI/ASME B5.54:2005:R2015 Methods for performance evaluation of computer numerically
controlled machining centers, American Society of Mechanical Engineers
Archenti A (2011) A computational framework for control of machining system capability,
PhD thesis, Stockholm
Archenti A, Nicolescu M (2013) Accuracy analysis of machine tools using Elastically Linked
Systems. CIRP Ann 62(1):506–506
Archenti A, Österlind T, Nicolescu C (2012a) Evaluation and representation of machine tool
deformations. J Mach Eng 12(1):118–129
Archenti A, Nicolescu CM, Casterman G, Hjelm S (2012b) A new method for circular testing of
machine tools under loaded condition. Proc CIRP 1:575–580
Automated Precision Inc. XD Laser. [Online]. Available: https://apisensor.com/products/mth/xd-
laser/. Accessed 10 May 2018
Asea, Bofors, Scania Alfa Laval (1970) Bearbeitungstests zur Untersuchung des dynamischen
Maschinenverhaltens der Firmen AB Bofors, Alfa-Laval AB, ASEA and SAAB-Scania
(BAS-norm). Sweden
BIPM, IEC, IFCC, ILAC, IUPAC, IUPAP, ISO, OIML (2012) The international vocabulary of
metrology–basic and general concepts and associated terms (VIM), 3rd edn. JCGM 200:2012
Brecher C, Esser M, Witt S (2009) Interaction of manufacturing process and machine tool. CIRP
Ann 58(2):588–607
Bringmann B, Knapp W (2009) Machine tool calibration: geometric test uncertainty depends on
machine tool performance. Precis Eng 33(4):524–529
Bringmann B, Küng A, Knapp W (2005) A measuring artefact for true 3d machine testing and
calibration. CIRP Ann 54(1):471–474
Bryan JB (1982) A simple method for testing measuring machines and machine tools, Part 1 and
2. Precis Eng 4:61–69
Etalon AG. LaserTRACER-NG. [Online]. Available: http://www.etalon-ag.com/en/products/
lasertracer/. Accessed 10 May 2018
Gao X, Li B, Hong J, Guo J (2016) Stiffness modeling of machine tools based on machining space
analysis. Int J Adv Manuf Technol 86(5–8):2093–2106
Heidenhain GmbH. [Online]. Available: https://www.heidenhain.com/fileadmin/pdb/media/img/
208871-29_Measuring_Devices_For_Machine_Tool_Inspection_and_Acceptance_Testing_01.
pdf. Accessed 30 July 2018
Ibaraki S, Hong C, Oyama C (2011) Construction of an error map of rotary axes by static R-test.
In: Proceedings of the 6th international conference on leading edge manufacturing in 21st
century (LEM21), vol 2011.6
ISO 10791-6:2014(E). Test conditions for machining centres – accuracy of feeds, speeds and
interpolations. ISO, Geneva
ISO 10791-7:2014. Test conditions for machining centres – Part 7: accuracy of finished test pieces.
ISO, Geneva
ISO 230-1 (2012) Test code for machine tools – Part 1: geometric accuracy of machines operating
under no-load or quasi-static conditions. International Organization for Standardization, Geneva
ISO 230-2:2014(E). Test code for machine tools. Part 2. Determination of accuracy and repeatabil-
ity of positioning of numerically controlled axes. ISO, Geneva
ISO 230-3:2007. Test code for machine tools. Part 3. Determination of thermal effects. ISO, Geneva
ISO 230-4:2005. Test code for machine tools. Part 4. Circular tests for numerically controlled
machine tools. ISO, Geneva
ISO 230-7:2006(E). Test code for machine tools. Part 7. Geometric accuracy of axes of rotation.
ISO, Geneva
Knapp W (1983a) Circular test for three-coordinate measuring machines and machine tools. Precis
Eng 5(3):115–124
Liebrich T, Bringmann B, Knapp W (2009) Calibration of a 3D-ball plate. Precis Eng 33(1):1–6
Lopez de Lacalle L, Lamikiz A (2009) Machine tools for high performance machining. Springer,
London
MEAX. Machine Tool Calibration with STREVEL | MEAX MT30. [Online]. Available: http://
www.meax.com/en/products/meax-mt30/. Accessed 10 May 2018
Optodyne Inc. [Online]. Available: http://www.optodyne.com/opnew5/TecArt/tech11.pdf.
Accessed 10 July 2018
Österlind T (2013) An analysis of machining system capability and its link with machined
component quality. KTH Royal Institute of Technology, Stockholm
Rensihaw PLC (2016) Renishaw unveils the new XM-60 multi-axis calibrator. [Online]. Available:
http://www.renishaw.com/en/renishaw-unveils-the-new-xm-60-multi-axis-calibrator%2D%
2D39275. Accessed 09 May 2018
Rensihaw PLC. XL-80 laser system. [Online]. Available: http://www.renishaw.com/en/xl-80-laser-
system%2D%2D8268. Accessed 09 May 2018
Sadasivam L, Archenti A, Sandberg U (2018) Machine tool ability representation: a review. J Mach
Eng 18(2):5–16
Sartori S, Zhang G (1995) Geomteric error measurement and compensation of machines. Ann CIRP
44(2):599–609
Schellekens P, Rosielle N (1998) Design for precision: current status and trends. Ann CIRP
47(2):557–586
Schneider C (2004) LaserTracer – a new type of self tracking laser interferometer. In: IWAA 2004.
CERN, Geneva
Schwenke H, Knapp W, Haitjema H, Weckenmann A, Schmitt R, Delbressine F (2008) Geometric
error measurement and compensation of machines – an update. CIRP Ann 57(2):660–675
SIOS MEßTECHNIK GmbH. Calibration Interferometer. [Online]. Available: https://sios-de.com/
products/calibration-systems/calibration-interferometer/. Accessed 10 May 2018
Slocum A (1992) Precision machine design. Prentice Hall, Englewood Cliffs
Smith GT (2016) Machine tool metrology - an industrial handbook. Springer International Pub-
lishing, Switzerland
Stephenson DA, Agapiou JS (2016) Metal cutting theory and practice, 3rd edn. CRC Press Taylor,
Bosa Roca
Weck M, Brecher C (2006) Werkzeugmaschinen 5 Messtechnische Untersuchung und Beurteilung,
dynamische Stabilität. Springer, Berlin
Ziegert J, Mize C (1994) Laser ball bar: a new instrument for machine tool metrology. Precis Eng
16(4):259–267
Machine Tool Prognosis for Precision
Manufacturing 8
Robert X. Gao, Peng Wang, and Ruqiang Yan
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246
Particle Filter for Prognostic Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
Fundamentals of Particle Filter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 249
Adaptive Resampling-Based Local Search PF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 252
Simulation Study of Local Search PF . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 255
Tool Wear Modeling . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
Derivation of Tool Wear Evolution Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
Derivation of Measurement Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 263
Experimental Study . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
Conclusion and Future Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 272
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 274
Abstract
Increasing demand for precision-manufactured parts for high-tech applications
in aerospace, nuclear power, transportation, etc. continually drives the advance-
ment of precision manufacturing technologies. As the precision and quality
of manufactured parts are significantly affected by the performance of the
machine tools, accurate and reliable condition monitoring, performance predic-
tion, and maintenance of machine tools become one important part of precision
manufacturing. In this chapter, a stochastic modeling technique is presented for
R. X. Gao (*) · P. Wang

Department of Mechanical and Aerospace Engineering, Case Western Reserve University,
Cleveland, OH, USA
e-mail: Robert.Gao@case.edu
R. Yan
School of Mechanical Engineering, Xi’an Jiaotong University, Xi’an, China

https://doi.org/10.1007/978-981-10-4938-5_8
246 R. X. Gao et al.
prediction of machine tool performance degradation based on sensing data from

the tool wear. Specifically, to account for the nonlinear and non-Gaussian char-
acteristic of operating and environmental conditions on the wear propagation and
machine performance degradation, particle filter (PF) that approximates proba-
bility distributions through a set of weighted particles is investigated. To improve
the reliability of time-varying degradation tracking and prediction, an adaptive
resampling particle filter method is developed. Specifically, particles are recur-
sively updated and resampled from the neighborhoods that are determined by
particles’ estimation performance from the last iteration, to characterize the
temporal variation in the tool degradation rates. This leads to improved tracking
and prediction accuracy with progressively narrowed confidence interval. The
developed method has been experimentally evaluated using a set of benchmark
data that were measured on a high-speed CNC machine.
Keywords
Precision manufacturing · CNC machine tools · Condition monitoring ·
Prognosis · Stochastic modeling · Remaining life prediction · Particle filter ·
Maintenance
Introduction
Increasing demand for precision-manufactured parts that find applications in

aerospace, nuclear power, transportation, and other high-tech areas has continued
to drive the advancement of precision manufacturing technologies. Compared to
traditional technologies, precision manufacturing is characterized by significantly
improved product quality control through higher machine accuracy and reduced
scrap rate. The dimensional accuracy and precision of manufactured products are
affected by the machine tool performance, which in turn is affected by aging and
wear (Dornfeld et al. 2003). Therefore, machine condition monitoring, performance
tracking, prediction, and timely maintenance are of critical importance (Zhou and
Xue 2018).
Machine condition monitoring refers to the health assessment of the machine
or its constituent components through proper analytics of sensor data (Byrne and
O’Donnell 2007). Through condition monitoring, performance deviation can be
detected, and the type, severity, location, and root cause of the fault can be subse-
quently identified (Rehorn et al. 2005). This provides the basis for condition-based
maintenance (CBM) actions to minimize the effect of the fault (Mehnen et al. 2014).
With the advancement of predictive science, prognosis has been increasingly recog-
nized as a valuable complement to CBM in manufacturing. This has led to a more
efficient maintenance approach termed intelligent preventive maintenance (IPM).
Since failure in a machine typically progresses through several stages, from failure
initiation to functional failure, predictive techniques can help determine how long
it takes for a machine’s functional degradation to progress from its current state to
8 Machine Tool Prognosis for Precision Manufacturing 247
final failure, which is commonly defined as the remaining useful life (RUL) (Le et al.
2013). Based on this knowledge, actions can be taken to prevent catastrophic failures
from happening, and reliability of machine operations can be improved.
Over the past decades, research on prognostic technologies has continued to
evolve, providing the basis for CBM and IPM. Jardine et al. (2006) summarized
diagnostic and prognostic methods for CBM implementation. Peng et al. (2010)
and An et al. (2013) presented a comparative study of representative prognostic
methods in terms of their strengths and limitations. Gao et al. (2015) presented an
extensive review of prognostic techniques and their potential application in the
context of the emerging cloud computing environment. Si et al. (2011) discussed
relevant statistical methods for performing prognosis. Sikorska et al. (2011)
compared different modeling options for RUL estimation, from the perspective
of industry and business applications. Baraldi et al. (2013) investigated the
capabilities of prognostic approaches to deal with various sources of uncertainty
in the RUL prediction, focusing on particle filter (PF) and bootstrap-centered
techniques. Heng et al. (2009) and Sun et al. (2012) discussed the potential
benefits, challenges, and opportunities associated with prognosis of rotating
machinery.
Depending on the types of data and information needed to characterize
the systems of interest and predict its future behavior, prognostic techniques can
be classified into three categories, physics-based, data-driven, and model-based, as
shown in Fig. 1.
Using tool wear as an example, physics-based techniques investigate the physical
mechanism of the wear by establishing a relationship between tool wear propagation
(e.g., wear width, measured by microscopes) and parameters such as machine
settings and material prosperities. Based on a large amount of experimental data
Operation Fatigue Propagation

Physics-Based
Tool-life prediction
Kalman Filtering
Wear
Mechanism Model-Based Particle Filtering
Manufacturing 1 Neural Network

System Fuzzy Theory
0.5
Wiener Process
0 Data-Driven
0 20 40
Frequency (Hz)
60
Gamma Process
Maintenance Measurement/ Proportional Hazards Model
Features Hidden Markov Model
Fig. 1 Classification of prognostic methods (Gao et al. 2015)

(Li 2012), analytical expressions are established, and associated coefficients in the
expressions are empirically determined. As an example, a generalized form of
extended Taylor’s law is presented to predict the tool life as a function of cutting
parameters and workpiece hardness (Poulachon et al. 2001). Different mechanisms,
such as abrasion, adhesion, and diffusion, are investigated to describe the tool wear
rate under different machine settings and materials (Huang and Liang 2004).
Additional factors, such as the tool geometry (Karpuschewski et al. 2015)
and cutting temperature (Luo et al. 2005), have also been investigated. Since it is
generally not feasible to comprehensively cover the complex physical nature of tool
wear, labelled parameters based on experimental data would vary under different
operational conditions. Accordingly, parameters need to be iteratively calibrated,
which is time-consuming and not feasible for online applications. Data-driven
techniques, in comparison, use observable sensor measurements (e.g., current,
force, vibration, acoustic emission, etc.) to infer the state of tool wear. The limitation,
however, is that a large amount of data for model training is required. Furthermore,
it generally does not provide a quantification of uncertainties involved in the
machining and modeling processes.
Model-based prognostic techniques integrate the strength of the aforementioned
two methods, thereby improving the prediction accuracy and robustness. Based on
the relevant physical mechanisms, state evolution models that describe the progres-
sion of tool wear and measurement models that relate sensor data to the wear severity
are established. Subsequently, machine state is inferred based on new measurements,
by means of estimating the posterior probability density function (PDF) through
Bayesian inference (Gasperin et al. 2011; Caesarendra et al. 2011). For RUL
prediction, once a posterior PDF is determined, the RUL is defined as the conditional
expected time to failure, given the current state. Kalman filter (KF) and particle
filter (PF) are two commonly used techniques to implement Bayesian inference.
Compared with KF, PF has shown advantage in nonlinear and non-Gaussian system
modeling, as it approximates the posterior PDF via a set of weighted particles
(Doucet et al. 2000; Arulampalam et al. 2002).
One constraint of the classical PF techniques is sample impoverishment,
which leads to significant decrease in the particle diversity and global search
ability during the iteration process. This is due to the fact that the same set of
particles are being repeatedly resampled with replacement. This problem limits
PF in tracking a degradation process with time-varying rates. To solve the
problem, an adaptive resampling particle filter has been developed, by which
the particles in the resampling process are resampled in a neighborhood beyond
the initial prior distribution to maintain the particles’ diversity. The resampling
range is determined by the particle’s estimation performance from the last
iteration. As a result, the estimation process can progressively reduce the effect
of initial sampling from the prior distribution by keeping the amount of active
particles and consequently increasing the estimation accuracy and narrowing
down the confidence interval. This enables the derivation of the tool wear
propagation as well as the measurement model, which can be experimentally
evaluated by systematic machining tests.
Particle Filter for Prognostic Modeling
Fundamentals of Particle Filter
Particle filter (PF) uses a set of particles to approximate a probability distribution.

The particles are randomly sampled, following a Monte Carlo method. Figure 2
shows an example of PF for estimating a Gamma distribution (4, 1). First, a set of
particles are randomly sampled from the distribution. Then, the statistical histogram
of the sampled particles is drawn to represent the estimated distribution by PF. It is
found that the estimation accuracy increases with the increase of number of particles.
The effect of number of particles on performance degradation tracking and predic-
tion accuracy will be discussed in detail in the following chapters.
Prognostic modeling based on PF involves two models: state evolution model
xk = f(xk 1, θ, mk 1) and measurement model zk = h(xk, vk). The symbol x denotes
the system states (i.e., health parameters that can characterize system status) that
cannot be measured directly and thus need to be estimated, and z denotes available,
onboard measurements. The symbol θ denotes a set of unknown parameters that
define the system’s evolution (e.g., performance degradation). Estimation of the two
8 10 12 14 16 18 20
20
500 particles
15
Histogram
Distribution
Distribution
10
6
5
4
2
0
0
0.45
0.4
0.35
0.3
0.25
0.2
0.15
0.1
0.05
0
0 50 100 150 200 250 300 350 400 450 500

Particle Number
20
20
15
15
5,000 particles
Distribution
Histogram
Distribution
10
10
5
5
0
0
0.25
0.2
0.15
0.1
0.05
0 500 1000 150020002500 30003500 40004500 5000

Particle Number
Weight
Fig. 2 Particle filter for characterizing a probability distribution, using a set of randomly sampled
particles; estimation accuracy increases with increase of particle numbers
latent variables x and θ constitutes a joint state and parameter estimation. It should be
noted that parameter θ, defining the degradation rates, may or may not change over
time. The symbols f and h describe the state transition and measurement function,
where the former characterizes a nonlinear evolution pattern of the system state x and
the latter represents nonlinear relationships between the states and measurements.
Then the objective of system estimation based on PF is to estimate the posterior
probability density function (PDF) p(x1 : k, θ1 : k| z1 : k), based on which to predict the
state evolution in the future p(xk + 1, θk + 1| z1 : k).
In PF, the unknown system states andparameters are represented by a set of
N random particles and associated weights xik , θik , wik , i ¼ 1,2, . . . ,N . Each particle
is associated with three variables, with the first two xik and θik initially sampled
from predefined probability distributions (e.g., prior distribution as commonly used).
The weights wki (representing the importance of particles) are initially assigned with
equal values, giving all particles equal initial weighting. The estimation process
includes two steps: prediction and correction. Through the prediction step, the
posterior PDF can be approximated by summarizing the following weighted
particles (Doucet et al. 2000):
Ð
pðxk , θk j zk1 Þ ¼ pðxk , θk j xk1 , θk1 Þpðxk1 , θk1 j zk1 Þdxk1
X N
¼ wik1 δ ðxk1 , θk1 Þ xik1 , θik1 pðxk , θk j xk1 , θk1 Þ
i¼1 (1)
X
N
¼ wik1 p xk , θk j xik1 , θik1
i¼1
Equation (1) indicates that the estimation

of the posterior
PDF starts with
sampling from a given prior distribution p xik , θik j xik1 , θik1 (e.g., normal distribu-
tion, uniform distribution). Without loss of generality, the prior distribution is just
a special case of importance density function (IDF), from which particles are
sampled. It will be discussed in the following chapters that for standard PF
(known as the first-generation, sequential importance sampling (SIS) or second-
generation, sequential importance resampling (SIR)), the selection of the prior
distribution directly affects the final estimation accuracy. When a new observation
zk is available, the update step (Doucet et al. 2001) can be expressed as an update
process of particles’ weights, which literally equals update of the posterior PDF:

wik / wik1 p zk j xik , θik (2)
By following Eqs. (1) and (2), the latent variables x and θ can be recursively
updated. With the latest updated parameter θ, future system state and system
degradation path can be predicted, as system degradation is exclusively dependent
on θ. Equations (1) and (2) are known as the first generation of particle filter,
sequential importance sampling (SIS).
An inherent problem associated with the conventional (or sequential importance
sampling, SIS) particle filter is particle degeneracy. This refers to the phenomenon
where weights of most particles (e.g., those outside of the three-sigma range of the
posterior distribution) become negligible after several iterations. As a result, most of
the computational time is spent on updating the particles with negligible contribution
to the state or parameter update. This is caused by the mismatch between the
posterior PDF (distributions of system states and/or parameters in the given obser-
vation, denoted by the red line in Fig. 3) to be estimated and the prior PDF (assumed
distributions characterizing evolutions of states and parameters, denoted by the blue
line) from which particles are initially sampled. A general solution is resampling,
which is proposed by Gordon et al. (1993). Resampling removes particles with small
weights and retains particles with large weights (as quantified by comparing the
accumulating weight to a random number between 0 and 1). This is the basic concept
of sequential importance resampling (SIR) (Arulampalam et al. 2002).
However, SIR introduces the particle impoverishment problem, where one sam-
pling position contains repeated particles and the number of unique and active
particles decreases significantly, leading to diminished accuracy in the estimation
(Wang and Gao 2015). This is due to the fact that the resampling process in SIR does
not change the discrete approximation nature of the conventional SIS, where limited
numbers of particles are not able to cover the entire PDF to be estimated. The root
cause for the discrete approximation problem in SIS and SIR is that the positions of
particles that are initially sampled from the prior distribution do not vary with time
and the same set of particles is resampled with replacement iteratively, leading to
particles concentrating on few positions (as illustrated in Fig. 3).
Efforts to further address the sample degeneracy and particle impoverishment
can be classified into two categories: improvements through sampling selection of
importance density function and improvements through refining resampling strategy.
Constrained by the particle degeneracy and impoverishment problem, applying the
classical PF (i.e., SIS and SIR) techniques to practical problems of tracking and
Posterior pdf to be
Prior pdf for estimated
initial sampling
Weight
update Degeneracy
Standard sequential Sample

importance impoverishment
resampling
Fig. 3 Initial sampling, weight update, and sequential importance resampling (SIR)
predicting machine performance degradation still faces challenges: (1) time-varying

degradation rates (different coefficients in one degradation function) and modes
(different degradation functions), (2) transient changes in system states due to
abrupt fault occurrence, and (3) varying operating conditions that cause varying
degradation paths.
Adaptive Resampling-Based Local Search PF
To tackle particle degeneracy and sample impoverishment associated with SIS and
SIR, the resampling strategy needs to be changed from discrete approximation to
continuous approximation, while maintaining a balance between keeping particle
diversity (a degree to quantify unique and active particles) and ensuring particles’
tracking performance (diverse particles may increase the confidence interval of the
estimation, leading to reduced estimation accuracy). This can be done by dispersing
the particles with large weights in the resampling process from fixed positions to a
wider range (see Fig. 4) by adding a perturbation to each particle. The perturbation
for each particle is sampled from a normal distribution, which is determined by their
estimation accuracy in the last iteration step (Wang and Gao 2015):
Prior pdf for Posterior pdf to be

initial sampling estimated
Weight
update Degeneracy
Standard sequential Sample

(a) importance impoverishment
resampling
Local Search
(b) Importance
resampling
Fig. 4 Initial sampling, weight update, (a) standard sequential importance resampling (SIR), and
(b) local search importance resampling; degeneracy problem and sample impoverishment problem
in SIR (Wang and Gao 2016b)
qffiffiffiffiffi
rðl Þ r ðl Þ
p θkþ1 j θk
i
/ N θkþ1 j θk , h Pik
i
(3)
where P represents the variance of particles’ estimations

" ! !T #
rðl Þ rðl Þ
X
N
rðl Þ
X
N
Pk ¼E θk wik θik θk wik θik (4)
i¼1 i¼1
The expression
( )
X
l
r ðl Þ : wjk rand (5)
j¼1
indicates that the lth particle from iteration k is selected to enter into the
estimation process at iteration step k+1 as the ith particle. This particle is
added by a perturbation to explore a wider neighborhood before entering into
step k+1. This perturbation is generated from a normal distribution, the variance
of which is actually dominated by the difference between this particle’s estima-
P
N
tion result and the mean of all particles’ estimation results N1 wik θik (regarded as
i¼1
the optimal estimation at iteration k). The symbol h denotes the shrinkage
coefficient, to ensure particles gradually concentrate on an optimal exploration
range.
Among all particles that are selected to enter into the next iteration, a particle
that is closer to the optimal estimation will be assigned with a smaller exploration
range, as the normal distribution that is used to generate the perturbation has a
smaller variance. Otherwise, particles will be assigned with a larger exploration
range. Dispersing samples not only increase the number of individual particles but
also cause particles in the subsequent iterations to continuously move toward an
optimal solution. Hence, the effect of initial sampling can be minimized in the
final estimation result. The decreasing shrinkage coefficient ensures that samples
will gradually converge at the optimal location, to help narrow down the confidence
interval and provide accurate prediction. Iterative resampling performed as such
can be regarded as approximation of continuous sampling, to break through the
restriction on estimation accuracy due to limited and discrete particles (Wang and
Gao 2016a, 2017).
The shrinkage coefficient ensures that samples will gradually converge at the
optimal location, to help narrow down the confidence interval and provide accurate
prediction. In other words, the shrinkage coefficient enables the variance of particles
to gradually decrease.
From Eq. (3), the variation of the particles after resampling at time k can be
derived as
! ! " ! !T #
X
N X
N X
N X
N X
N
V θik ¼V θi
k1 þ 2
hE θi
k wik θik θi
k wik θik
i¼1 i¼1 i¼1 i¼1 i¼1
! " ! !T #
X
N XN X
N X
N
¼V θi
k1 þh E 2
θi
k wik θik θi
k wik θik
i¼1 i¼1 i¼1 i¼1
!
X
N
¼ 1þh V θi
2
k1
i¼1
(6)
where * in θik1 represents the particles being selected in the resampling process.
After resampling process, some particles with small weights are discarded, so that
the variation of particles’ weights (describing the particles’ variance) decreases. If
a symbol α is assigned to quantify the variance decrease due to discarding some
particles, then Eq. (6) can be expressed as
! !
X
N X
N
V θik ¼ ð1 αÞ 1 þ h V 2
θik1 (7)
i¼1 i¼1
To make sure the particles gradually converge over iterations, namely, variance of
particles gradually decreases, i.e.,
! !
X
N X
N
V θik θik1 (8)
i¼1 i¼1
the shrinkage coefficient h in Eq. (7) should satisfy:

rffiffiffiffiffiffiffiffiffiffiffi
α
h (9)
1α
Then the question becomes how to quantify α, the variance decrease, after
resampling. Assuming there is no big jump of the system between time k1 and k,
the weights’ variance change before and after resampling at time k1 can be used to
estimate α at time k, as
P
N P
N
^ ik1
w wik1
i¼1 i¼1
αk ¼ (10)
P
N
^ ik1
w
i¼1
^ ik1
w and ^ ik1
w
represent the particles’ weights before and after resampling at time
k1, respectively. Substituting Eq. (10) into Eq. (9) gives a range for h selection.
It should be mentioned that at time k = 1, as there is no available information for
h selection and the initial selection of h does not affect the subsequent estimation,
h can be selected as 1.
Below in Algorithm 1, a step-by-step description of the developed LSPF

algorithm is presented.
Algorithm 1 Local Search Particle Filter (LSPF)

Input: Raw sensor data z1:k collected at time k > 1
Output: Filtered system state x1:k
• Initial sampling
– for particles i = 1: N

Draw particle xi0 , θi0 from a prior distribution (e.g., even distribution)
Assign a weight 1/N to each particle
– end for
• Weight update (at time k = 1) n o
– Predict system states and parameters xî1 , θ^1 via the state evolution function
i

^ i1 ¼ p zk j xî1 , θ^1
i
– Update weights of particles by calculating likelihood of w
P
N
– Calculate total weight t ¼ w ^ i1 ¼ t 1 w
^ i1 . Normalize particle weights w ^ i1
i¼1
• Resampling (at time k = 1)
P
N
θ^1
i
– Calculate mean value of parameters N1
i¼1
– for i = 1: N
Select a random number and determine selected particle number to enter next
iteration, according to Eq. (5)
Calculate variances of parameters for the selected particle, according to
Eq. (4)
Assign new particle values xi1 , θi1 , according to Eq. (3) (using h = 1)
Calculate the particle weight wi1
– end for
• Repeat update and resampling processesnfor k > o1
i î
– Predict system states and parameters x^ , θ according to the state evolution
k k
function
^ ik
– Update weights of particles w
– Select shrinkage coefficient h, according to Eqs. (9) and (10)
– Perform resampling according to Eqs. (3, 4, and 5)
– Calculated weights of resampled particles wik
• End Algorithm 1
Simulation Study of Local Search PF
To evaluate the developed LSPF algorithm with adaptive resampling strategy,

a simulation study on tracking a nonlinear system is performed. The system state
evolution model and measurement model in the simulation are formulated as
8 x
>
< xk ¼ 0:5 xk1 þ 25 þ 8 cos ð1:2 ðk 1ÞÞ þ mk
1 þ x2
2 (11)
>
: z k ¼ xk þ vk
20
From Eq. (11), it can be noted that the system state evolution follows a strong
nonlinear model. The process noise mk and measurement noise nk are both sampled
from a [0 sqrt(10)] even distribution to simulate the system randomness.
The performance of LSPF on tracking the system is compared to standard
particle filter (i.e., SIR), extended Kalman filter (EKF), and unscented Kalman
filter (UKF). For the particle filters, 500 particles are used. For UKF, nine sigma
points are used. The performance is shown in Fig. 5. The tracking error is shown
in Fig. 6, and the average tracking error is shown in Table 1.
It can be seen from Fig. 6 and Table 1 that LSPF outperforms the other three
algorithms, while PF algorithms outperform KF variants in overall. It should be
mentioned that in Fig. 5, LSPF tracks the system variation more precisely than
SIR, especially at the inflection points where the trend of system variation
changes.
In the above simulation, the state transition is not dependent on parameters.
So one advantage of LSPF that, confidence intervals (or estimation variance) of
estimated parameters/states gradually shrink, is not reflected. Another simulation, as
illustrated below, involves a joint state and parameter estimation. The state evolution
Fig. 5 Tracking performance of LSPF, with comparison to EKF, UKF, and standard PF (SIR)
Fig. 6 Tracking error of EKF, UKF, standard PF (SIR) and LSPF, in terms of RMS
Table 1 Average tracking error of EKF, UKF, standard PF and LSPF, in terms of RMSE
EKF UKF PF (SIR) LSPF
RMSE 13.9 9.3 6.6 4.1
is conditional on two parameters (A = 0.0005 and B = 0.4), as shown in Eq. (12).

These two parameters determine the system variation (or degradation) rate:
B
xk ¼ xk1 A ek (12)
Initially, particles in both standard PF (SIR) and LSPF are sampled from even
distributions, [0 0.001] for parameter A and [0 1] for parameter B, assuming a prior
and rough knowledge of the system is known. The trajectories of parameter estima-
tion by LSPF and SIR are shown in Figs. 7 and 8.
It can be seen from Fig. 7 that the confidence intervals of parameter estimation
gradually decrease and finally converge at a small range, while the confidence
intervals finally become a single line in the SIR estimation shown in Fig. 8. This
does not mean that SIR achieves a better result than LSPF. In contrast, it means that
SIR suffers from severe particle impoverishment problem that particle diversity
greatly decreases over the iterations. This can be demonstrated by Fig. 9, histograms
of SIR and LSPF estimation on parameter A when time is 30. It can be noted that in
SIR case, there are only three active particles.
Fig. 7 Trajectories of
parameter estimation by LSPF
Fig. 8 Trajectories of parameter estimation by SIR

400
350
Number of particles 300
250
200
150
100
50
0
2 2.5 3 3.5 4 4.5 5 5.5 6
Value of parameter A × 10-4
140
120
Number of particles
100
80
60
40
20
0
3 3.5 4 4.5 5 5.5 6 6.5 7
Value of parameter A × 10-4
Fig. 9 Histograms of SIR and LSPF estimation on parameter A
Tool Wear Modeling
Derivation of Tool Wear Evolution Model
Friction between machining tool and material during the manufacturing process (see
Fig. 10) can impose a significant effect on product quality, tool life, and machining
cost. When tool wear reaches a certain degree, increasing cutting force, vibration and
cutting temperature can cause deteriorated surface integrity and dimensional error
Fig. 10 Mechanism for

friction and tool wear (Wang
and Gao 2016a)
greater than tolerance and lead to the end of tool life (Li 2012). Consequently, in situ
tool wear monitoring and tool life prognosis are important to ensuring machining
precision and cost-effective machining operations.
Traditionally, quantifying tool wear and predicting remaining tool life are
performed through physics enabled models. These models establish empirical rela-
tionships between tool wear/life and certain cutting parameters (e.g., cutting speed,
surface temperature, etc.), but do not exhaust the physical nature of tool wear.
Unknown parameters in these empirical models are trained with a large amount
of experiments under different operation conditions (Marksberry and Jawahir 2008).
Paris’ law is a common approach to quantify tool wear or crack growth (Paris 1961),
where the stress intensity factor at the tip of a crack and crack length are modeled
with respect to two material parameters. Aslantaş and Taşgetiren (2004) extended
Paris’ formula by mixing mode stress intensity factors to develop an analytical
model for two-dimensional rolling sliding contact situations. Among physical
models describing tool life, an important branch is around Taylor’s tool life equation
(Mills and Rdeford1983), which relates tool life to cutting speed in a reverse
exponential relationship (Karandikar et al. 2013). Hoffman (1984) introduced an
extended Taylor’s equation and discussed the effects of feed rate and cutting depth
on tool life besides cutting speed. However, this kind of approach requires detailed
and complete knowledge of system behavior, which is not available for most
manufacturing components and systems. Moreover, the majority of coefficients
involved in the physical models need to be determined experimentally, causing the
models to be application specific.
Contradictory to the above discussed direct measurement of tool wear,
another approach arises with the development of sensing techniques that use the
fused information of indirect measurements (e.g., force, vibration and acoustic
emission, etc.) generated in the machining process to infer the underlying tool
state (Teti et al. 2010).
Thus the problem is transferred to establishing a quantitative relationship linking
the measurements or extracted features to system states (i.e., tool wear) based on
various estimation methods. Once the tool wear model is determined, the remaining
Fig. 11 Architecture of LSPF-based tool wear and RUL prediction
useful life (RUL) of tools given a certain threshold of wear can be predicted
correspondingly. The process can be implemented by particle filter, to fully
take uncertainties into account. The architecture is shown in Fig. 11 (Wang and
Gao 2015).
The tool wear rate model can be seen as a particular type of crack growth model or
fatigue spall progression model. Generally, a crack growth model is characterized by
the stress intensity factor at the tip of a crack K = f (a, σ), with a being the half crack
size and σ being the nominal stress. Theoretically, the crack is assumed to not
propagate when K is smaller than a threshold value, after which, the crack growth
rate will be governed by a power law, such as Paris’ law (Paris 1961):
dx
¼ C ðΔK Þm (13)
dt
and
pffiffiffiffiffi
ΔK ¼ Δσ πx (14)
where t is the number of cutting cycles, dx/dt denotes the tool wear rate, parameters
C and m are related to material properties, and Δσ is the stress range, which is
assumed to be constant. Combining Eqs. (13) and (14), it becomes
dx pffiffiffiffiffim pffiffiffim m 0
¼ C Δσ πx ¼ C Δσ π x 2 ¼ C 0 xm (15)
dt
The above equation integrates all constants independent from tool wear size x into
a single variable C0 . Taking integration on both sides in Eq. (15), it can be rewritten
in the form that relates tool wear severity at the current time to the previous time
(replacing C0 and m0 with C and m):
h i1=ð1mÞ
ð1mÞ
xt ¼ xt1 þ C ð1 mÞ þ ut1 (16)
Thus, the current tool wear is fully described by the previous state and unknown
parameters C and m (determined by material properties). In other words, when
applying LSPF to estimate tool wear, the dimension of each particle is two
corresponding to parameters C and m.
However, Eq. (16) neglects the effect of other machine settings besides workpiece
material: cutting speed, feed rate, and cut depth. As an example shown in Fig. 12, the
relationship between tool wear rate and feed rate under certain cutting speed
8
v = 120m/min
6 v = 90m/min
Flank wear (µm)
v = 60m/min
0
0.1 0.2 0.3 0.4 0.5
Feed rate (mm/rev)
Fig. 12 Effect of feed rate on tool flank wear (Wang and Gao 2016a)
Fig. 13 Tool wear 0.6

propagation with respect to
machine settings
0.5
Wear width (mm)

0.4
0.3
f=0.5;d=1.5
0.2 f=0.5;d=0.75
f=0.25;d=0.75
f=0.25;d=1.5
0.1
0
0 20 40 60 80 100
Machining time (min)
(in symbol v) and depth is quadratic, which means an optimal feed rate exists for
longest tool life.
To include other machining settings, such as spindle speed, feed rate, and f cutting
depth d, Eq. (13) can be modified as (Astakhov 2007):
dx
¼ f A d B xC (17)
dt
Coefficients A and B represent the effects of f and d on wear progression.
Expression of wear x as a function of time can then be obtained as (Wang and
Gao 2016a)
1C
1
xk ¼ x1C
k1 þ f d ð1 C Þðt k t k1 Þ
A B
(18)
One example demonstrating the effect of feed rate and cutting depth is shown in
Fig. 13. It indicates the tool wear rate decreases with the decrease of the feed rate and
cutting depth and cutting depth influences more than feed rate. The influences of
machining setting factors would be reflected in estimated coefficients/parameters
A and B. It should be noted that these parameters are assumed to be constant under
the same machining setting.
Derivation of Measurement Model
Online measurements, such as force and vibration, are always with a high sampling
rate, and there is no need to estimate the tool wear when each new measurement is
available. Also, the measurement is usually noisy and cannot be employed directly.
Hence, feature extraction is needed to perform the estimation with the time unit
Table 2 List of extracted features for tool wear inference

Domain Features Expression
Statistical Mean P
N
z ¼ 1=N j zi j
i¼1
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi
RMS zRMS ¼ 1n z21 þ z22 þ . . . þ z2n
X
Variance zvar ¼ 1=N ðzi zÞ2
N
Kurtosis X zi μ4
zKURT ¼ 1n
n
σ
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
P Pk
Frequency Spectral mean
f ¼ k
i¼1 f i S ðf i Þ=
2
i¼1 S ðf i Þ
Spectral kurtosis Pk f i f4

f var ¼ i¼1 σ S ðf i Þ
PN
Time frequency Wavelet energy E WT ¼ i¼1 wtφ ðiÞ=N 2
Energy ratio Rwt ¼ E xWT =E zWT
being the cut number. Previous work (Wang et al. 2015) has investigated various
features in the time domain and frequency domain, as shown in Table 2. On the other
hand, by fully taking advantage of the fact that all measurements within one cut can
be seen as a distribution, statistical features, such as Kullback-Leibler (KL) diver-
gence, are investigated.
It is assumed that the distribution of measurement within one cut shifts when tool
wear deteriorates. Thus the distance between two distributions can be seen as an
indicator to reveal the wear. Let p1(x) and p2(x) be two distributions; the information
of KL divergence from p1 to p2 is defined as (Eguchi and Copas 2006)
ð
p1 ð x Þ
KLðp1 , p2 Þ ¼ p1 ðxÞlog dx (19)
p2 ð x Þ
Smaller values of the information quantity KL( p1, p2) mean that the distance
between two distributions is smaller. That is, the larger the distance between two
distributions, the larger the difference between two distributions. The distribution
obtained from the initial time is taken as the reference distribution, and the new
distribution is compared to the reference distribution to calculate the KL information,
which is subsequently applied to estimate the tool wear. To evaluate the quality of
extracted features, a correlation coefficient is applied:
X
ðxi xÞðzi zÞ
i
C c ¼ rX
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
X ffi (20)
ðxi xÞ2 ðzi zÞ2
i i
Due to quite the complex mechanism of both tool wear and force or vibration
measurement, it is difficult to establish a definite relationship between wear and
extracted KL information. To tackle this problem, in this research, both wear and
KL indicators are normalized to a common range. Considering that wear and KL
indicators show a similar development trend over time, the measurement model can
be rewritten as
KLk ¼ xk þ vk (21)
This is a general way to obtain a measurement model when a relationship based

on physics is unavailable.
Experimental Study
To experimentally evaluate the performance of proposed LSPF-based tool

wear prediction framework, two sets of experimental data are used. One set of
experimental data is measured from a high-speed CNC machine under a dry milling
operation (Li et al. 2009). In this dataset, experiments were conducted under the
same machine settings but with high sampling rate. The other set of experimental
data was measured under different machine settings (Agogino and Goebel 2007).
For the first dataset (Li et al. 2009), a three-flute ball nose tungsten carbide cutter
was tested to mill a workpiece (material: stainless steel, HRC52) in a down-milling
operation. The spindle was running at 10,400 rpm, while the feed rate was set at
1555 mm/min in the x direction. The depth of cut (radial) was 0.125 mm in the
y direction, while the depth of cut (axial) in the z direction was 0.2 mm. A Kistler
quartz three-component platform dynamometer was mounted between the work-
piece and machining table to measure the cutting forces. The diagram of experimen-
tal setup is shown in Fig. 14. Online measurements including force and vibration
were recorded at the sampling frequency 50 kHz. A total of around 300 cuts’ data
were collected during the tool life test.
CNC Milling Machine

Offline Measurement
Tool Wear
Microscope
Width
LEICA MZ12
Tool State
Cutting Tooth
PC
z y
Workpiece x Online Measurement
Accelerometer
Machining Data
Force Sensor Charge Force, Vibration
Table Acquisition
Amplifier Measurement
Card
Fig. 14 Schematic diagram of experimental setup (Wang et al. 2015)

1.2 Measurements
Measurement
Ground truth
True wear
Median of prediction
predictions Wear threshold
1 90% confidence
confidence limits
limits
Amplitude (Normalized)
Prediction Paths
paths
0.8
0.6
0.4
Estimation Prediction
0.2
0 50 100 150 200 250
Time
Fig. 15 200 steps ahead tool wear prediction (Wang and Gao 2015)
For feature extraction, previous work indicates that the ratio of wavelet energy in
the x direction to the z direction has the highest correlation coefficient of 0.985. The
KL divergence applied in this research has the correlation coefficient of 0.993 with
respect to true wear. The normalized actual tool wear and calculated KL divergence
are shown in Fig. 15, represented by a blue line and black star, respectively.
For this dataset, as only machine setting is employed, Eq. (16) is used as the
state evolution model. Initially, the unknown parameter pair m and C are modeled
as probability distributions following a certain distribution (e.g., uniform distri-
bution in this research). The initial distribution selection can determine the
performance of standard SIR, but not LSPF. In the learning stage, based on the
state model and measurement model, the unknown parameters combined with the
state transition probability p(xk| xk1) can be estimated recursively and obtained a
priori. Once the measurements stop, the posterior distribution function p(xk + l| zk)
can be calculated to predict a one-step ahead tool wear based on the latest
updated parameters. Also, multi-step ahead prediction can be achieved by
performing the prediction process in Bayesian inference. The multi-step ahead
prediction can be in turn used to calculate the remaining useful life when the
failure threshold is given.
Figure 15 shows an example of tool wear prediction based on LSPF, using the
information of the first 100 cuts as the prior knowledge. Five hundred particles are
applied, with gray lines as their prediction paths. The result indicates that median of
prediction (represented by red line) can generally track the tool wear. Figure 16
shows the evolution of parameter estimation, with the blue lines as median estima-
tions and red lines as 90% confidence limits of estimations. It can be noted that the
2.5
2
m Evolution
1.5
1 Median
90% confidence
0.5
-3
x 10
8
C Evolution
4
Prediction
2
0 50 100 150 200 250
Time (Cut)
Fig. 16 Evolution of distribution of parameters m and C for 200 steps ahead prediction (Wang and
Gao 2015)
estimated parameters’ distributions concentrate on smaller ranges (i.e., the width

between two red lines) continuously, which is the advantage of LSPF over standard
SIR, due to that the resampled particles keep searching the local area instead of
staying the same.
Figure 17 shows the prediction starting from the 200th cut, with the evolution of
parameter distribution. It is noted that longer step ahead prediction has a broader
distribution of prediction path or, in the terms of RUL prediction, has a larger
confidence interval, which is also proved in Fig. 17.
Given the wear prediction at different steps, RUL can be calculated with respect
to a predefined threshold. Figure 18 draws the RUL estimation result using a-λ
accuracy. a-λ accuracy determines whether at given point in time (specified by λ)
prediction accuracy is within desired accuracy levels (specified by a). Desired
accuracy levels (specified by β) are expressed as a percentage of true RUL. In this
research, a is set as 0.1 and β is set as 50%, which means the prediction is accepted
if more than half of the prediction distribution falls within 1 a of true RUL.
Figure 19 specifies the distributions of predicted RUL when λ is equal to
100 and 200 cuts, respectively. To demonstrate the effectiveness and robustness of
the proposed method, Monte Carlo simulations have been performed to process both
the data set used for the above analysis and a different data set measured from
a separate run-to-failure experiment. Each scenario has been run for 100 times. The
result showing the performance of LSPF compared to SIR is listed in Table 3, which
indicates the former can effectively reduce the prediction error by over 30%.
Measurement
1.1
True wear
Median of prediction Wear threshold
1
90% confidence limits
Amplitude (Normalized)
0.9 Prediction paths

m evolution
0.8
C evolution 2
0.01
0.7
1
0.005
0 50 100 150
0.6
0
0 50 100 150
0.5
0.4 Estimation Prediction
0 50 100 150 200 250

Time (Cut)
Fig. 17 100 steps ahead tool wear prediction (Wang and Gao 2015)
300
True RUL
(1±ɑ ) RUL
Interquartile Range
250
200
RUL (Cut)
150 True
True 96.0%
52.0% True True True
100 False True 60.0% 72.0% 58.2%
41.0% 83.0% True True
76.0% True 98.0%
86.0%
50
0
0 50 100 150 200 250 300
Time (Cut)
Fig. 18 RUL prediction with respect to a-λ accuracy (Wang and Gao 2015)
ɑ-bounds
0.3
RUL prediction at
Probability
100th cut
0.2
0.1
0
100 150 200 250 300 350 400
0.2
ɑ-bounds RUL prediction at
200th cut
Probability
0.1
0
70 80 90 100 110 120
RUL (Cut)
Fig. 19 Distributions of predicted RUL at 100th and 200th cut (Wang and Gao 2015)
Table 3 Comparison of SIR and LSPF on different steps ahead RUL prediction (Wang and
Gao 2015)
200 steps ahead prediction (%) 100 steps ahead prediction (%)
SIR 15.4 11.7
LSPF 9.6 3.5
For the second dataset, experiments were conducted on a Matsuura milling

machine MC-510V under different machining settings. The experimental setup
and data acquisition are shown in Fig. 20. Data sampled by five different sensors,
two acoustic emission sensors (one on table, one on the spindle), two vibration
sensors, and one current sensor are installed to determine the state of tool wear
(Agogino and Goebel 2007). The evolution of acoustic emission data with respect to
different wear severity is shown in Fig. 21 as an example. Two sets of data under four
different machining settings (as shown in Table 4, and their effects on tool wear
progression shown in Fig. 13) are investigated in this research, with one used for
training process to estimate the parameters in Eq. (18) and the other one to predict
tool wear and testify the effectiveness of the proposed method.
Initially, the unknown parameters A–C in Eq. (18) are modeled as probability
distributions following a certain distribution (e.g., uniform distribution). The initial
distribution selection can determine the performance of standard PF, but not
Fig. 20 Experimental setup Matsuura milling machine MC-510V

and data acquisition (Wang
and Gao 2016a) Motor
Current sensor
Vibration sensor Preamplifier

Spindle
AE s
Tool en sor Charger Amplifier
Workpiece
so r Charger Amplifier
sen
Table AE
Vibration sensor Preamplifier
Knee elevating
screw
0.12
Probability
0.08
0.04
0
0.08
0.13
0.20
0.31
0.35 0.5
Wear (mm) 0.4
0.40 0.3
0.2
0.49 0.1 Acoustic emission distribution
Fig. 21 Distributions of acoustic emission for different wear levels (Wang and Gao 2016a)
improved PF proposed in this research. In the learning stage, based on the state
model Eq. (18) and measurement model Eq. (21), the unknown parameters A–C can
be estimated recursively and obtained as a priori. The latest updated parameters A–C
Table 4 Design of experiments in Agogino and Goebel (2007)

Cutting speed (m/min) Cutting depth (mm) Feed rate (mm/rev)
Case 1 200 m/min 1.5 0.5
Case 2 0.75 0.5
Case 3 0.75 0.25
Case 4 1.5 0.25
0.6
0.5 Cutting speed: 200 m/min

Cutting depth: 1.5 mm
Wear (mm)
0.4
0.3
0.2
0.1
0
7 11 15 19 22 26 29 32 35 38 41 44
Time (min)
0.6
0.5 Cutting speed: 200 m/min

0.4 Cutting depth: 0.75 mm
Wear (mm)
0.3
0.2
0.1
0
9 12 15 2224 27 33 39 45 51 59 66 72
Time (min)
Fig. 22 Tool wear prediction for different cutting depths (Wang and Gao 2016a)
under different machining settings would be used for prediction. It should be noted
in this research that the machining settings are assumed to be known in advance and
the parameters representing the effect and machining factors on tool wear progres-
sion are assumed to maintain constant under the same machining settings. Figure 22
shows the predicted tool wear for different cutting depths. The black box in the figure
represents the interquartile range (50% confidence range) of the prediction.
Figure 23 shows the evolution of parameter estimation for Case 2, with the blue
lines as median estimations and red lines as 90% confidence limits of estimations.
Fig. 23 Evolution of 10
Parameter C Parameter B Parameter A

Median
estimated parameters in state 90% confidence
evolution model (Wang and 5
Gao 2016a)
0
10
0
1
0.5
0
-0.5
0 2 4 6 8 10 12
Estimation iteration
It can be noted that the estimated parameters’ distributions concentrate on smaller

ranges (i.e., the width between two red lines) continuously, which is the advantage
of LSPF over standard PF.
Since PF is an estimation method based on random sampling, the stability
and robustness of the algorithm should be demonstrated. Results of Monte Carlo
simulations are shown in Fig. 24. Each case has been run for 100 times. Both
the median (blue line) of the predictions and error bars are shown in the figure.
Both Figs. 22 and 24 demonstrate the tool wear predictions for different machining
settings based on improved PF are accurate and robust.
Estimated parameters for different machining settings based on the train data by
LSPF are shown in Table 5. When the planned machining settings are known, the
future tool wear can then be predicted. Figure 25 shows tool wear prediction for
varying machining settings, in which the tool undergoes different wear rates under
phase 1 and phase 2. The results demonstrate the proposed method, in which
machining setting factors are modeled as parameters in the state evolution model
and measurement model and then estimated by LSPF and used for prediction, is
effective.
Through Monte Carlo simulation tests, the developed LSPF has demonstrated
its ability on accurately determining the effects of machining settings on tool wear
propagation and precisely predicting the tool wear. The determination of effect of
machining settings can subsequently be used to finding the optimal machining
settings online.
Conclusion and Future Work
This chapter describes advanced prognostic modeling technique for machine tool
condition monitoring and performance prognosis, which is an enabling technique
that contributes to improving the science base for precision manufacturing.
0.6
0.5
Wear (mm)
0.4
0.3
0.2 Predicted wear

True wear
0.1
0 10 20 30 40 50 60 70 80
Time (min)
0.5
0.4
Wear (mm)
0.3
0.2
0.1
0
0 5 10 15 20 25 30 35 40
Time (min)
Fig. 24 Tool wear prediction for different feeding rates (Wang and Gao 2016a)
Table 5 Estimated parameters by LSPF

Case f d Parameter A Parameter B
1 0.5 1.5 4.8867 2.2227
2 0.5 0.75 4.7073 4.6192
3 0.25 0.75 1.7151 4.4621
4 0.25 1.5 1.6469 7.7653
A particle filter enabled joint-state-and-parameter estimation and prediction method

has been investigated. Specifically, a local search particle filter with an adaptive
resampling strategy has been developed to address the particle degeneracy problem
and improve the estimation accuracy by narrowing down the confidence interval of
estimation and prediction, especially when time-varying performance degradation is
encountered.
Case studies on prediction of the severity of tool wear under different machining
settings have been performed to evaluate the performance of the developed method.
For this purpose, machining settings have been taken as unknown parameters in the
state evolution model and measurement model. The estimated parameters and state
Fig. 25 Tool wear prediction

for varying machining settings
(Wang and Gao 2016a)
evolution model were then utilized to predict tool wear when the machining settings
are known. With the performance evaluated by a tool wear test on a CNC milling
machine, the developed method has shown to be able to accurately and reliably
determine the effect of machine settings and predict the tool wear growth. The
obtained knowledge can be further leveraged for optimized scheduling of tool
changes.
As for the measurement model, it is usually difficult to derive an explicit
relationship (nonlinear in most dynamical systems) between the system states and
sensor measurements, unless the system can be represented by physical parameters
that are fully derived from measurable quantities. Literature review has shown that
machine learning techniques may provide a promising choice for obtaining non-
linear measurement models. This will be explored in the future.
References
Agogino A, Goebel K (2007) Milling data set. NASA Ames Prognostics Data Repository,
NASA Ames Research Center
An D, Choi JH, Kim NH (2013) Options for prognostics methods: a review of data-driven
and physics-based prognostics. In: 54th AIAA/ASME/ASCE/AHS/ASC structures, structural
dynamics, and materials conference, Boston, p 1940
Arulampalam MS, Maskell S, Gordon N, Clapp T (2002) A tutorial on particle filters for online
nonlinear/non-Gaussian Bayesian tracking. IEEE Trans Signal Process 50(2):174–188
Aslantaş K, Taşgetiren S (2004) A study of spur gear pitting formation and life prediction. Wear
257(11):1167–1175
Astakhov VP (2007) Effects of the cutting feed, depth of cut, and workpiece (bore) diameter on the
tool wear rate. Int J Adv Manuf Technol 34(7–8):631–640
Baraldi P, Mangili F, Zio E (2013) Investigation of uncertainty treatment capability of model- based
and data-driven prognostic methods using simulated data. Reliab Eng Syst Saf 112:94–108
Byrne G, O’Donnell GE (2007) An integrated force sensor solution for process monitoring of
drilling operations. CIRP Ann Manuf Technol 56(1):89–92
Caesarendra W, Widodo A, Thom PH, Yang B, Setiawan JD (2011) Combined probability approach
and indirect data-driven method for bearing degradation prognostics. IEEE Trans Reliab
60(1):14–20
Dornfeld DA, Lee Y, Chang A (2003) Monitoring of ultraprecision machining processes. Int J Adv
Doucet A, Godsill S, Andrieu C (2000) On sequential Monte Carlo sampling methods for Bayesian
filtering. Stat Comput 10(3):197–208
Doucet A, Gordon NJ, Krishnamurthy V (2001) Particle filters for state estimation of jump Markov
linear systems. IEEE Trans Signal Process 49(3):613–624
Eguchi S, Copas J (2006) Interpreting Kullback–Leibler divergence with the Neyman–Pearson
lemma. J Multivar Anal 97(9):2034–2040
Gao R, Wang L, Teti R, Dornfeld D, Kumara S, Mori M, Helu M (2015) Cloud-enabled prognosis
for manufacturing. CIRP Ann Manuf Technol 64(2):749–772
Gašperin M, Juričić Ɖ, Boškoski P, Vižintin J (2011) Model-based prognostics of gear health using
stochastic dynamical models. Mech Syst Signal Process 25(2):537548
Gordon NJ, Salmond DJ, Smith AF (1993) Novel approach to nonlinear/non-Gaussian Bayesian
state estimation. IEE Proc F Radar Signal Process 140(2):107113
Heng A, Zhang S, Tan AC, Mathew J (2009) Rotating machinery prognostics: state of the art,
challenges and opportunities. Mech Syst Signal Process 23(3):724–739
Hoffman EG (1984) Fundamentals of tool design. Society of Manufacturing Engineers, Dearborn
Huang Y, Liang S (2004) Modeling of CBN tool flank wear progression in finish hard turning.
Trans ASME J Manuf Sci Eng 126:98–106
Jardine AK, Lin D, Banjevic D (2006) A review on machinery diagnostics and prognostics
implementing condition-based maintenance. Mech Syst Signal Process 20(7):1483–1510
Karandikar JM, Abbas AE, Schmitz TL (2013) Tool life prediction using random walk Bayesian
updating. Mach Sci Technol 17(3):410–442
Karpuschewski B, Schmidt K, Beno J, Mankova I, Frohumller R, Prilukova J (2015) An approach
to the microscopic study of wear mechanism during hard turning with coated ceramics. Wear
342–343:222–233
Le SK, Fouladirad M, Barros A, Levrat E, Lung B (2013) Remaining useful life estimation
based on stochastic deterioration models: a comparative study. Reliab Eng Syst Saf
112:165–175
Li B (2012) A review of tool wear estimation using theoretical analysis and numerical simulation
technologies. Int J Refract Met Hard Mater 35:143–151
Li X, Lim BS, Zhou JH, Huang S, Phua SJ, Shaw KC, Er MJ (2009) Fuzzy neural network
modelling for tool wear estimation in dry milling operation. In: Annual conference of the
prognostics and health management society, San Diego, pp 1–11
Luo X, Cheng K, Holt R, Liu X (2005) Modeling flank wear of carbide tool insert in metal cutting.
Wear 259:1235–1240
Marksberry PW, Jawahir IS (2008) A comprehensive tool-wear/tool-life performance model in the
evaluation of NDM (near dry machining) for sustainable manufacturing. Int J Mach Tools
Manuf 48(7–8):878–886
Mehnen J, Tinsley L, Roy R (2014) Automated in-service damage identification. CIRP Ann Manuf
Technol 63(1):33–36
Mills B, Rdeford A (1983) Machinability of engineering materials. Applied Science, London
Paris PC (1961) A rational analytic theory of fatigue. Trends Eng 13:9–14
Peng Y, Dong M, Zuo MJ (2010) Current status of machine prognostics in condition-based
maintenance: a review. Int J Adv Manuf Technol 50(1–4):297–313
Poulachon G, Moisan A, Jawhir IS (2001) Tool-wear mechanisms in hard turning with polycrys-
talline cubic boron nitride tools. Wear 250(1–12):576–586
Rehorn AG, Jiang J, Orban PE (2005) State-of-the-art methods and results in tool condition
monitoring: a review. Int J Adv Manuf Technol 26(7–8):693–710
Si X, Wang W, Hu C, Zhou D (2011) Remaining useful life estimation-a review on the statistical
data driven approaches. Eur J Oper Res 213(1):1–14
Sikorska JZ, Hodkiewicz M, Ma L (2011) Prognostic modelling options for remaining useful life
estimation by industry. Mech Syst Signal Process 25(5):1803–1836
Sun B, Zeng S, Kang R, Pecht MG (2012) Benefits and challenges of system prognostics. IEEE
Trans Reliab 61(2):323–335
Teti R, Jemielniak K, Donnell GO, Dornfeld D (2010) Advanced monitoring of machining
operations. CIRP Ann Manuf Technol 59(2):717–739
Wang P, Gao RX (2015) Adaptive resampling-based particle filtering for tool life prediction.
J Manuf Syst 37:528–534
Wang P, Gao RX (2016a) Stochastic tool wear prediction for sustainable manufacturing. Procedia
CIRP 48:236–241
Wang P, Gao RX (2016b) Markov nonlinear system estimation for engine performance tracking.
ASME J Eng Gas Turbines Power 138(9):091201
Wang P, Gao RX (2017) Automated performance tracking for heat exchangers in HVAC.
IEEE Trans Autom Sci Eng 14(2):634–645
Wang J, Wang P, Gao RX (2015) Enhanced particle filter for tool wear prediction. J Manuf Syst
36:35–45
Zhou Y, Xue W (2018) Review of tool condition monitoring methods in milling processes.
Int J Adv Manuf Technol 96:5–8
Cylindrical Gear Metrology
9
Kang Ni, Yue Peng, Dirk Stöbener, and Gert Goch
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 278
Geometry of Cylindrical Gears . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
Geometry of 2D Involute Curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 279
Geometry of 3D Involute Flank . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 280
Surface Coordinate System on an Involute Gear Flank . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 282
Flank Modifications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 283
Geometric Features on Gear Flanks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 285
Distance Definitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 288
Gear-Measuring Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
Standard Inspection Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 289
Hardware Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290
Characteristics and Evaluation Algorithms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 291
Calibration of Gear-Measuring Instruments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 292
Emerging Technologies for Gear Inspection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
Optical Gear Metrology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295
Areal-Oriented Measurement and Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 298
Traceability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300
Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303
K. Ni (*)
Center for Precision Metrology, University of North Carolina at Charlotte, Charlotte, NC, USA
Research and Development, The Timken Company, North Canton, OH, USA
e-mail: kang.ni@timken.com
Y. Peng · G. Goch
Center for Precision Metrology, University of North Carolina at Charlotte, Charlotte, NC, USA
e-mail: ypeng5@uncc.edu; fgoch@uncc.edu
D. Stöbener
Bremen Institute for Measurement, Automation and Quality Science (BIMAQ),
University of Bremen, Bremen, Germany
e-mail: d.stoebener@bimaq.de

https://doi.org/10.1007/978-981-10-4938-5_9
278 K. Ni et al.
Abstract
This chapter presents the basic principles and techniques for measuring the
geometrical features of cylindrical gears. The mathematical models for nominal
cylindrical gear geometry are given in a two-dimensional (2D) space and
extended to a three-dimensional (3D) space. The geometric parameters for
assessing the conformance of gear design and manufacturing are highlighted
based on the current international standards. Conventional gear inspections by
tactile measuring systems such as gear-measuring instruments (GMIs) and coor-
dinate measuring machines (CMMs) are discussed in detail, including measuring
strategies and evaluations of “raw” spatial data sets, methods of calibrating gear-
measuring systems, and the estimation of measurement uncertainty. Emerging
technologies including optical measuring systems and areal evaluation methods
are introduced as part of future cylindrical gear metrology.
Keywords
Involute geometry · Flank modifications · Measuring principles · Instrument
calibration · Artifacts · Uncertainty · Gear standards · Optical sensors · Areal
evaluation · Virtual CMM
Introduction
Gears are widely used in modern power train systems to transmit power, torque, and
angular velocity or position between rotational components. Gear drive applications are
highly visible in automotive, machine tool, automation, aviation, and energy handling
and conveyor systems, to name only the most important industrial areas. The perfor-
mance goals of the gears in those applications – such as lower noise emission, higher
power density, and longer lifetimes – have increased significantly in the past decades,
while simultaneously facing ever-growing demand for affordable gear components.
A multitude of manufacturing options include primary processing gear teeth cutting,
heat treatment, and precision flank finishing (Bouzakis et al. 2008; Karpuschewski et al.
2008). Modern production of high-performance gears combines a specific selection of
those options with a multi-step manufacturing process, aiming at targeted gear geometry
with tight tolerances and robust material properties. One of the design approaches to
improve gear performance is to apply complex geometric features to gear flanks. For
cylindrical gears, those features are known as “flank modifications,” defined as an
intentional alternation of the “pure” involute geometry of a tooth flank. Consequently,
measurements with high accuracy levels and short cycle times are indispensable to
enable stable gear production, offering reliable information about the manufacturing
process and for effective quality control (Goch et al. 2017a).
For a state-of-the-art overview of today’s gear inspection technologies, this chapter
starts with basic definitions of the key measurands of cylindrical gears then presents
the measuring principles based on tactile instruments including both hardware con-
figuration and software evaluation methods. The calibration of gear-measuring
9 Cylindrical Gear Metrology 279
instruments and the uncertainty in the evaluation of gear measurement results based on
physical artifacts and tactile probing technologies are reviewed. Emerging noncontact
gear inspection technologies for future gear metrology, such as optical gear-measuring
systems and areal gear evaluation algorithms, conclude this chapter.
Geometry of Cylindrical Gears
Geometry of 2D Involute Curve
An involute is a curve that is generated on geometries such as circles and ellipses.

The involute curve generated on a circle is most commonly used for gear profiles
because of a unique property that keeps a constant ratio of angular speeds between
two meshing gears with two different pitch diameters.
In a 2D Cartesian coordinate system, Eq. (1) defines the coordinates of a nominal
point Pnom(xnom, ynom) on an involute curve (see Fig. 1), starting from the base circle
with the radius rb:

xnom ¼ rb ½ cos ðξnom þ ΛÞ þ ξnom sin ðξnom þ ΛÞ
(1)
ynom ¼ rb ½ sin ðξnom þ ΛÞ ξnom cos ðξnom þ ΛÞ
where Λ denotes the angular position of the involute’s starting point at the base circle
and ξnom is the rolling angle assigned to Pnom.
In a cylindrical coordinate system, Eq. (2) expresses the position of the nominal
point Pnom as a function of the distance ρnom from the origin and the polar angle
ϕnom, starting from the X axis:
8 pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
< ρnom ¼ x2 þ y2 ¼ rb 1 þ ξ2
nom nom nom
y (2)
: ϕnom ¼ arctan nom
xnom
The relationship between an arbitrary rolling angle (ξ) and an arbitrary pressure
angle (α) is governed by the involute Eq. (3):
ξ ¼ α þ inv α ¼ tan α (3)
Figure 2 shows teeth that have been generated on a base circle in a 2D plane,
referred to as the transverse plane. The zoomed view illustrates the base space half-
angle ηb, which can be calculated by:
π 4 x tan αn
ηb ¼ inv αt (4)
2j z j
where x is the profile shift coefficient, αn is the normal pressure angle, |z| is the
absolute value of the number of teeth, and inv αt is the involute angle.
280 K. Ni et al.
Fig. 1 Involute curve in 2D

plane
Fig. 2 Involute gear tooth in a 2D plane
Geometry of 3D Involute Flank
Based on the 2D involute curve in the transverse plane, the 3D geometry

of a helical gear flank surface can be generated by a “threading” model
(Lotze and Haertig 2001). This leads to an areal mathematical description of a
helical gear flank in 3D Cartesian coordinates (Günther et al. 2009), given
by Eq. (5):
8
< xnom ¼ rb ½ cos ðξnom þ Λðznom ÞÞ þ ξnom sin ðξnom þ Λðznom ÞÞ
y ¼ rb ½ sin ðξnom þ Λðznom ÞÞ ξnom cos ðξnom þ Λðznom ÞÞ (5)
: nom
znom ¼ znom
Fig. 3 Helical gear in a 3D Cartesian coordinate system: (a) one gear flank generated by threading
model, (b) all gear teeth flanks
Based on this model, the origin (denoted by the yellow dot in Fig. 3a) of each
involute profile in a transverse plane is “threaded up” along the base helix line
(illustrated by the tilted black line with the base helix angle βb) from the bottom to
the top, covering the entire gear face width. The coordinate of a nominal point on the
pure involute surface is mathematically realized by describing the nominal initial
angle Λ(znom) of the 2D involute profile on each stacked transverse plane as a
function of the third dimension, as:
tan βb
Λðznom Þ ¼ ηb þ znom (6)
rb
Given the radius of the base circle, a gear’s dimensions are limited by setting the
start and end values of the roll angle ξ and the z- coordinate in the profile and axial
directions, respectively:

ξmin ξnom ξmax
(7)
0 znom b
where (ξnom, znom) defines a nominal point located on the reference cylinder
in the middle of the gear flank, as illustrated in Fig. 3a. The maximum
and minimum values of the roll angle are determined by the tip form
radius and root form radius, which are denoted, respectively, by rFa and rFf in
Fig. 2.
Table 1 lists the parameters for determining the nominal dimensions of involute
gear teeth. Figure 3b shows a 3D representation of simulated teeth with both left and
right flanks on the gear, generated by the parameters in Table 1. ISO 21771 (ISO
2007) gives detailed explanations and derivations of the analytic relationships
between those quantities.
282 K. Ni et al.
Table 1 Example of design parameters for an external helical gear

Name Symbol Value Unit
Number of teeth z 20 Dimensionless
Normal module mn 3 mm
Normal pressure angle αn 20 degree
Face width b 30 mm
Helix angle β 20 degree
Helical direction Right hand 1 Dimensionless
Profile shift coefficient x 0 Dimensionless
Surface Coordinate System on an Involute Gear Flank
Parametric representations of complex surfaces are extensively used in various

disciplines. To extend the analytic description of modifications on the flank of a
helical gear from a 2D line-oriented function to a 3D surface function, a pair of
curvilinear coordinates (u, v) is assigned to the nominal involute flank, as shown in
Fig. 4. The coordinates of the nominal point Pnom can be described by the surface
coordinates u and v, given by:

u ¼ rb ξnom
(8)
v ¼ znom
In this surface coordinate system, the U axis follows the generation

principle describing an involute curve, and the V axis is parallel to the direction
of the Z axis in the 3D Cartesian coordinate system shown in Fig. 3. The
combination of u and v axes also accommodates the typical computerized
numerical control (CNC) motion control system for generation cutting and
conventional gear inspection.
The nominal surface normal vector can be derived by the cross product of two
vectors formed by the partial derivatives along the u and v directions:
0 1
! ! ξnom sin ðξ þ Λðznom ÞÞ
! @Pnom @Pnom @
N¼ ¼ ξnom cos ðξ þ Λðznom ÞÞ A (9)
@u @v
ξnom tan βb
which leads to the normalized unit normal vector expressed as:

0 1 0 1
! nx sin ðξ þ Λðznom ÞÞ
! N 1
n ¼ ! ¼ @ ny A ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi @ cos ðξ þ Λðznom ÞÞ A (10)
N nz
2
1 þ ð tan βb Þ2 tan βb
Here, nx, ny, and nz are the components of the unit vectors in the X, Y, and Z axes,
respectively (Günther1996).
Fig. 4 Surface coordinate

system and surface normal
vector on an involute gear
flank
Flank Modifications
Over time, engineers have introduced flank modifications to improve the meshing
properties of gear pairs under various load conditions. Flank modifications on cylindrical
gears are desirable geometric alterations that have been intentionally superimposed on the
nominal involute flank. Figure 5 illustrates the affected zone on a gear flank realized by
symmetrical parabolically shaped modifications in both the profile and helix directions.
Typical flank modifications can be categorized into three general types:
• In the profile direction: slope, crowning, tip, and root relief

• In the helix direction: slope, crowning, top, and bottom end relief
• On the overall tooth surface: flank twist and triangular end relief
Figure 6 depicts six kinds of flank modifications on the tooth of a cylindrical gear
in 3D space.
An analytic equation for profile slope and crowning modifications is expressed by
surface coordinates u and v. First introduced in Günther (1996), curve arclength,
radial distance, and roll length are each applicable as the independent variable of a
modification function (Ni 2017) such as Eq. (11). In this chapter, roll length is used
in the analytic function.
4C α 2 4C α þ C Hα
AðuÞ ¼ u þ u (11)
L2α Lα
Here, CHα, Cα, and Lα denote the profile slope modification, the profile crowning
modification, and the range of modification defined in the profile direction, respec-
tively. Figure 7a illustrates the profile shape after profile slope and crowning
modifications.
Correspondingly, Eq. (12) describes the slope and crowning modifications in the
helix direction as a function of v:
284 K. Ni et al.
Fig. 5 Flank modifications in profile and helix directions
Fig. 6 Typical flank modifications for cylindrical gears in 3D space: (a) profile slope modification,
(b) profile crowning modification, (c) tip and root relief, (d) helix slope modifications, (e) helix
crowning modification, (f) end relief at top and bottom face. (After ISO 2007)
4C β 2 4C β þ C Hβ
BðvÞ ¼ v þ v (12)
L2β Lβ
where CHβ, Cβ, and Lβ denote the helix slope modification, the helix crowning
modification, and the range of modification defined in the V direction, respectively.
Figure 7b depicts the helix shape after helix slope and crowning modifications.
Fig. 7 Slope and crowning modifications: (a) profile modifications; (b) helix modifications
Depending on the gear size and specific applications, the magnitude of flank
modifications can vary from a few micrometers to several tens of micrometers.
Geometric Features on Gear Flanks
Based on the descriptions of the nominal gear geometry in the previous section,
various types of measurable gear features must be captured for quantitative assess-
ment. The current gear measurement standards require four evenly spaced teeth to be
measured on a gear, including both the left and right flanks (ISO 2013). On an
individual flank, one or more profiles and helix lines are probed, usually via
continuous scanning.
Between adjacent flanks, points are also captured by mechanical probing, such
that a calculated distance between each two points on adjacent flanks can be used
for assessing the uniformity of tooth-to-tooth geometry (known as pitch
deviation).
Today’s technology of CMMs and GMIs enables continuous scanning motions to
capture a gear flank’s raw data. Figure 8 shows two scanned paths (in red) on a gear
flank, required for the assessment of profile and helix features on that flank.
Profile
The profile feature is a 2D curve defined on a transverse plane. This plane is offset by
a specified distance from the datum face, usually half the face width. A profile
feature can be viewed as the curve generated by intersecting the transverse plane
with the gear body. The start and end of the measured profile is determined by the
corresponding diameters near the root and tip of a tooth.
Helix
Based on the helix angle of design (e.g., the helix angle on the base cylinder), the
distance from the gear center defines the radial location of a helix feature to be
measured. Its start and end positions are determined in the axial direction. The axial
range of this feature usually covers 90–95% of the face width of a gear. For a spur
gear, the helix feature is a 2D straight line, parallel to the gear axis. For a helical gear,
the helix feature is a 3D line, inclined to the gear axis by the corresponding helix
angle.
286 K. Ni et al.
Fig. 8 Typical features

measured on gear flanks
Specific procedures for obtaining deviation and modification characteristics such

as slope deviation, form deviation, and crowning will be discussed in the next
section for both profile and helix features.
Pitch
Due to the rotational symmetry of a gear, all teeth collectively form a circle on which
the relative position of two equal-handed flanks is separated by a constant angle.
This angle is known as angular pitch and is given by:
2π
τ¼ (13)
j zj
As shown in Fig. 9, various pitch quantities can be defined depending on the

diameter of interest. The most commonly used pitch feature is the transverse pitch,
which is specified on the reference circle and whose diameter is defined as reference
diameter (d). The pitch obtained on the transverse plane is denoted as pt, indicated by
the subscript “t.” For pitch investigations at flank locations other than the reference
diameter, the circle is defined as the “measurement” circle, the diameter of which is
denoted by dM (see Fig. 10).
To assess the tooth-to-tooth deviation, singular points on all teeth (left and right
flanks) must be captured. The location of those points is specified on a certain
diameter. For example, three probed points (yellow dots) on three consecutive left
flanks of a helical gear are shown Fig. 8.
In conventional pitch measurements, pitch quantities are expressed as arc length.
This means that the nominal pitch values and pitch deviations depend directly on the
probing diameter. Equation (14) gives the nominal value of a transverse pitch based
on nominal gear design parameters (see Table 1):
Fig. 9 Graphical definitions

of nominal pitch features on a
cylindrical gear. (After ISO
2007)
Fig. 10 Single pitch

deviation of an individual
tooth obtained on the
measurement diameter. (After
ISO 2013)
πmn
pt ¼ (14)
cos β
The relationship between transverse pitch (pt) and angular pitch (τ) on the
reference diameter is:
d
pt ¼ τ (15)
2
The relative position accuracy of a certain tooth (either right or left flank) with
respect to other teeth (corresponding right or left flank) is characterized by pitch
deviations. As shown in Fig. 10, the nominal position of a flank is given by the dash-
dotted arc and the nominal pitch (on the measurement circle dM) is represented by
the arrowed curve.
The single pitch deviation of that individual flank is denoted by fpi, the difference
between the measured pitch ptM and its nominal counterpart on the measurement
circle.
288 K. Ni et al.
Fig. 11 Definition of distances for evaluating gear deviation parameters: (a) Euclidean distance;
(b) projected distance; (c) plumb line distance
Distance Definitions
In coordinate metrology, three types of distances are commonly used to quantify the
deviation between a measured geometry and its nominal counterpart, the latter of
which might be formed by a set of discrete nominal points or an analytic representation
of the nominal geometry. These distances are the Euclidean distance, the projected
distance, and the plumb line distance, illustrated in Fig. 11a, b, and c, respectively.
Transferring these concepts to 3D Cartesian coordinate systems and given the
coordinate of a nominal point and a measured point in correspondence, the Euclid-
ean distance is defined by:
! qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
! ! 2 2 2 2
d Euc ¼ Pm Pnom ¼ ðxm xnom Þ þ ðym ynom Þ þ ðzm znom Þ (16)
where (xnom, ynom, znom) are the coordinates of the nominal point (represented by
!
vector Pnom ) and (xm, ym, zm) are the coordinates of the actual point (represented by
! !
vector Pm ). Similarly, the projected distance d proj (see Fig. 11b) results from the
! ! !
projection of d Euc on the unit normal vector n at Pnom .
!
! ! !
d proj ¼ Pm Pnom n
¼ ðxm xnom Þ nx þ ðym ynom Þ ny þ ðzm znom Þ nz (17)
where (nx, ny, nz) are the coordinates of the unit normal vector at the nominal point
!
Pnom .
Thus, the coordinates of nominal points must be used in correspondence with those
of the measured points in order to obtain the gear deviation distances based on
Eqs. (16) and (17) – an algorithm often used in conventional gear evaluation software.
!
However, the deviation between a given nominal point Pm is best characterized
!
by the plumb line distance d lot as illustrated in Fig. 11c. Its vectorial description is
given by Eqs. (18) and (19):
! ! !
d lot ¼ Pm Pnom (18)

! ! ! !
Pm ¼ Pnom þ d lot n (19)
The plumb line distance for calculating cylindrical gear deviation was first
derived by Günther (1996), who offered the analytic expression as Eq. (20):
(sffiffiffiffiffiffiffiffiffiffiffiffiffiffi sffiffiffiffiffiffiffiffiffiffiffiffiffiffi! )
! ρ2m ρ2m tan βb
rb
d lot ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1 arctan 1 ϕm þ η b þ z m
2 r2b r2b rb
1 þ ð tan βb Þ2
(20)
where (ρm, ϕm, zm) are the coordinates of the measured point in a cylindrical
coordinate system.
The three unique properties of the plumb line distance for cylindrical gears are:
• The plumb line distance is the shortest distance from a measured point to the
nominal geometry (i.e., pure involute flank)
• The plumb line distance can be directly evaluated by the coordinates of a
measured point without knowing the corresponding nominal coordinates
• Setting Equation (20) to zero leads to an implicit analytic function (ρnom, ϕnom,
znom) = 0, equivalent to the nominal involute gear geometry.
Gear-Measuring Process
Standard Inspection Procedure
To inspect the key geometric features of cylindrical gears, a typical measurement

procedure consists of:
• Measuring one profile and one helix on both flanks of three or four out of all the
gear teeth
• Measuring one pitch point on each flank
• Evaluating the profile, helix, and pitch deviations according to standards.
The requirements of the measurement and evaluation procedures are specified in

various international, national, and industrial standards. For example, ISO 1328-1
(ISO 2013) imposes requirements of minimum deviation parameters to be inspected
on a gear and be based on the reference diameter, module, and objective tolerance
class of the gear, as well as requiring a minimum of three teeth to be measured for
evaluation of the profile and helix parameters. It also includes methodology for
calculating the tolerance values for the parameters, based on the gear class.
VDI/VDE 2612 (VDI 2000) and 2613 (VDI 2003) provide introductions to
290 K. Ni et al.
measuring instruments and probing strategies. Both the ISO and VDI standards
contain definitions of deviation parameters, locations of the line features to be
measured, evaluation requirements, and guidance for measurement practices.
Hardware Systems
Tactile instruments have long been the dominant means of geometrically measuring
gears. In a tactile system, a probe physically contacts the gear flank surface, resulting
in a deflection of the sensing components to capture measuring points.
Conventional gear measurements are based on the generation principles of
involutes and helices. A perfect involute curve in a transverse plane is generated
by rotating the base circle while moving a point at a constant velocity along the line
of action (a fixed tangent to the base circle) in the same direction as the rotation.
When applied to gear-measuring instruments, this means the gear to be measured is
rotated while the probe follows the linear motion along the tangent to the base
cylinder in the transverse plane. This superposition of two movements, one rota-
tional and one linear, generates a nominal 2D involute curve representing the
reference for the profile measurement.
On an actual gear, which contains deviations from the nominal involute geometry,
the probe is deflected in the sensitive direction (along the tangent and perpendicular
to the involute curve), thus detecting a protrusion or lack of material on the flank
surface. Therefore, the output of the sensor shows the deviations to the nominal
involute directly. The probing can be carried out by single point measurements or
scanning on the profiles.
To measure a helix, a similar procedure is applied in which the probe moves along
the axial direction while the gear rotates with a constant angular speed, thus
generating a nominal helix as the reference. For this method, a rotary table and at
least one probe are necessary. More probes can be used for alignment or for better
access to the flanks. GMIs such as the one shown in Fig. 12a usually follow the
generation principle.
Procedures based on the generation principle, such as these, are the best in terms
of minimal measurement uncertainty. However, the deviations calculated from these
measurements assume that the contact point between the probing element and gear
flank stays in the same transverse plane and that the probe is moving linearly at an
even speed, which imposes requirements on the control of the system – otherwise,
uncertainty will be introduced.
A second strategy to inspect a gear is to obtain 3D coordinates of surface points
on the gear flanks and calculate the local deviations for the actual points. A rotary
table is not necessary for this procedure; general-purpose CMMs can be used in this
way to measure gears. The probe still touches individual points or scans on the same
designated profiles or helices required in the standards. In this case, a star probe
configuration such as the one in Fig. 12b must be used to access the flanks in
different directions. If the measuring device is equipped with a rotary table, it
positions the flank to be measured at an accessible position, such that all flanks
Fig. 12 Probe configurations for tactile gear measurements: (a) two probes used on a GMI, (b) star
probe on a CMM, (c) three probes used on a CMM with rotary table
can be measured with one probe (as illustrated in Fig. 12c). The deflections of the
probe are captured by the probing head sensors in three directions instead of a
specific sensitive direction. The deviation calculation and radius correction involve
the definitions of distances as explained in earlier sections.
The common advantages of current tactile gear-measuring systems, compared to
non-contact methods, are that the measurement uncertainty is relatively smaller and
the procedure for evaluation of the measurement uncertainty is well established. The
disadvantages of tactile systems are that the inspection areas are restricted on
individual points or lines and the inspection speed is limited due to the physical
nature of tactile probing systems.
In addition to GMIs and CMMs, an angular indexing device with a tactile probe
can be used for pitch inspection based on direct angular measurements. A tooth
space comparator can measure the relative distance between two flanks (Farago and
Curtis 1994; Goch 2003). Gear runout can be measured by either a single-probe or a
two-probe tester. Tooth thickness can be inspected by measurement over pins.
Characteristics and Evaluation Algorithms
Once the flank deviations have been captured by the measurement instruments, data
processing and evaluations are conducted to present the deviation parameters for
quality assessment. The “data” refers to the deviations directly measured by the
instruments or calculated from the coordinates of the surface points. For a profile, the
deviations are plotted versus the roll path length, whereas for a helix, the deviations
are plotted with respect to the probe motion in the axial direction. Usually, a trace
that is longer than the required evaluation range is measured, so limiting elements
such as the profile control diameter are used to mark the edge of the evaluation range.
The data outside this range is not included in the evaluation.
First, a low-pass filter (Gaussian 50% type per both ISO and VDI standards)
applied to the original deviation plot extracts only the long-wave components. Next,
a linear or second-order curve obtained by least-square approximation of the filtered
data enables an easy determination of the deviation parameters. First-order polyno-
mial models are used for unmodified flanks or modified flanks with designed slopes
292 K. Ni et al.
Fig. 13 Profile deviations on an unmodified involute: (a) total profile deviation, (b) profile form
deviation, (c) profile slope deviation. (Figures revised from ISO 1328-1)
(see Fig. 13b and c). Second-order models are applied for modified flanks with
crownings.
The most commonly used deviation parameters (symbols following ISO 1328-1)
are profile and helix slope deviations ( fHα and fHβ), profile and helix form deviations
( ffα and ffβ), profile and helix total deviations (Fα and Fβ), single pitch deviation ( fp),
total cumulative pitch deviation (Fp), and tooth thickness (s). Figure 13 illustrates
three of these parameters, measured on an unmodified involute profile.
Finally, the resulting deviation parameters are compared to designs and tolerances
to assess the gear quality by assigning the obtained deviation parameter values to one
of 11 accuracy grades. Parameter tolerances and their corresponding accuracy grades
are specified in ISO 1328-1 and 2 (ISO 2003, 1997).
Calibration of Gear-Measuring Instruments
The instruments used to measure gears should be calibrated regularly. ISO 10360
(ISO 2000–2016) specifies acceptance tests for CMMs. ISO 18653 (ISO 2003) and
ISO/TR 10064 (ISO 1992–1998) specifies the evaluation of measuring instruments
used for profile, helix, pitch, and runout inspection. The traceability of gear mea-
surements is established by physical gear artifacts, calibrated by the national metrol-
ogy institutes or accredited labs. The artifacts and the part to be inspected should be
similar in geometry and are both measured on the instrument to be calibrated. They
should be mounted at the same location within the measuring volume, under the
same environmental conditions, and measured following the same procedures.
Physical Artifacts
Physical artifacts are the basis for a reliable, traceable measurement of gear charac-
teristics. They are therefore indispensable for the precise measurement of gear
geometry. Normal-sized gears, such as those used in automobile drive train systems,
utilize two different types of gear artifacts:
(a) Non-workpiece-like representations of the gear flanks are generated by simple

geometric elements like spheres, planes, or cylinders. Since these geometric
elements can be manufactured with very low form deviations, observed mea-
surement deviations can be assigned to the calibrated measuring system. The
necessary geometrical elements for the alignment of the artifact on the calibrated
measuring device are incorporated into the artifact’s base or are derived from the
mounted elements (e.g., determination of x-y plane by approximation through
the measured center points of several spheres). Another non-workpiece-like
representation is to embody only one very precisely manufactured flank mounted
on a cylinder representing the gear axis (e.g., helix or involute normal).
(b) Workpiece-like artifacts are much more similar to regular gears and allow the use
of only slightly modified standard inspection procedures for performing calibra-
tion measurements. In addition, their similarity to the workpieces does not lead
to any significant deviations caused by, for example, different probing point
distributions or varying weights of the artifacts and the inspected gears. How-
ever, in the production of high-precision artifacts, often only a few flanks or just
one flank of the tooth can actually be produced in order to achieve a high level of
precision.
In addition to the gear flank line artifacts, pitch artifacts are also used to
calibrate the pitch measurements of a gear-measuring instrument (see Figs. 14c
and 15).
In 2009, PTB and NMIJ/AIST published a method for the high-precision pitch
calibration of gears and ball plates (Kniel et al. 2009), which is based on the rosette
method by Noch and Steiner (Noch and Steiner 1966). In regular pitch measure-
ments, the deviation components generated by the measuring instrument, rotary
table, and gear toothing are superimposed. The published method employs self-
calibrating fault separation, which does not require a substitution artifact. For
calibration, the pitch measurements are carried out for all teeth of the inspected
gear or for all spheres of a ball plate. The measured object is then rotated by one pitch
position and measured again completely. This process is repeated until the measured
object is again in its initial position. The deviation components caused by the
measuring instrument and the rotary table can then be separated mathematically
from the pitch data of the inspected gear.
Gear Measurement Uncertainty Evaluation

The major contributors to gear measurement uncertainty include:
• Artifact (uncertainty of the artifact calibration as well as geometric and material

differences between the artifact and the part to be inspected)
• Instrument (uncertainty regarding components of all axes, probe deflection,
filtering, and probing sphere radius correction)
• Measuring procedure (mounting variations, repeatability and reproducibility of
artifact, and part measurement)
• Part to be inspected (deviations in datum surfaces and surface finish)
294 K. Ni et al.
Fig. 14 Non-workpiece-like gear artifacts: (a) a profile artifact, (b) a lead artifact, (c) a pitch
artifact (Härtig 2005)
Fig. 15 A workpiece-like
gear artifact
• Environment (differences between the measurement conditions of the calibration

lab and the measuring lab, temperature gradients, and drifts that cause object and
instrument deformations as well as vibrations)
• Evaluation (establishment of datum features, calculation of deviations, filter, and
algorithms used to evaluate the deviation parameters)
• Operator conducting the measurements
An uncertainty estimation following the GUM and its supplements (BIPM 2008)
is usually applied in calibration labs. ISO 18653 (ISO 2003) describes the compar-
ator method, which uses an artifact with gear tooth geometry that has been widely
adopted in industry. Equation (21) shows the general form of the uncertainty
equation. U95 is the expanded measurement uncertainty with a 95% confidence
level. For the coverage factor, k = 2 is usually selected. um is the standard uncertainty
of the reproducibility tests, un is the standard uncertainty of the artifact calibration, ug
represents the geometrical dissimilarity between the artifact and measured part, uw
covers other dissimilarities between them, and |E| is the bias.
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
U 95 ¼ k u2m þ u2n þ u2g þ u2w þ jE j (21)
Emerging Technologies for Gear Inspection
Optical Gear Metrology
Currently, gear measurement techniques are still dominated by tactile measurement

methods. In recent years, however, various optical measurement approaches have
been presented, which are mainly based on distance measurement principles. These
methods have the potential to replace the tactile measurements. However, significant
deviations may still occur on nonmetallic gear surfaces if the surfaces exhibit
translucent behavior.
The methods presented here include measurement methods based on
non-coherent measuring principles and interferometry. These processes have been
further developed for use on gear-measuring devices, and some are now commer-
cially available from individual manufacturers.
Non-coherent Measuring Principles

Historically, several different optical measuring principles have been tested for
applicability to gear measurement. The laser triangulation and stripe pattern projec-
tion principles revealed great potential for such applications. Both of these optical
principles require a cooperative surface of the part under inspection, because the
surface must be able to reflect a part of the incident light back to the sensor.
In recent years, remarkable steps have been achieved in the development of the
illumination sources and the sensitivity of the detectors. Today’s laser triangulation
and stripe pattern projection sensors are thus able to acquire evaluable signals from
relatively smooth surfaces, even at large angles of incidence – corresponding to the
operating conditions under which tooth flanks are measured.
According to the manufacturers, commercially available optical gear-measuring
devices based on a laser line triangulation sensor – when applied to gear measure-
ment – reach gear deviation parameter uncertainties of only a few micrometers. A
multi-axis optical CMM with a laser triangulation sensor incorporated was used to
296 K. Ni et al.
Fig. 16 3D point clouds data

captured on a helical gear by
an optical CMM (Peng et al.
2017)
capture 3D point clouds from a helical gear, as shown in Fig. 16. The reported results
achieved agreement on profile and helix deviation parameters of several micrometers
in comparison to GMI- and CMM-based measurements on the same gear sample
(Goch et al. 2017; Peng et al. 2017).
A triangulation line sensor applied to the measurement of spur gears could not
reach these low values because of the harsher measurement conditions surrounding
the optical accessibility of the gear flanks. Its deviations from tactile measurements
amounted to about 20 μm due to systematic issues (reflections between the gear
flanks), statistical uncertainty contributions, and also the abovementioned differ-
ences in the acquired data sets (Auerswald et al. 2017).
As long ago as 2006, the fringe projection method was being tested with regard to
its suitability for gear measurement. Studies showed that measurements on tooth
flanks were possible only with precision-forged gears, since these had a surface
roughness high enough to reflect sufficient light to the detector for an evaluable
shape measurement (Meeß et al. 2006).
Taking into account the current developments in the field of sensor technology, it
is expected that today’s fringe projection systems can generate evaluable data sets
from smoother flank surfaces. However, the 2006 studies also showed that the
changes in the surface slope of the flank in the measuring zone led to significant
deviations of the profile deviation parameters from tactile measurement results. To
address this issue, algorithms were developed that reduce the deviations of the
measurements to values below 40 μm – much lower, but not low enough. With
new approaches like miniaturized fringe projection, deviations can probably be
further minimized in the future (Matthias et al. 2016).
Interferometric Measuring Principles

Interferometric measurement methods can be divided into two categories: a point-
wise scheme and an areal data acquisition scheme.
A point-wise distance sensor based on frequency shifting interferometry (Bosch

et al. 1992) has been described by Günther and Balzer et al. The measurement
principle operates with a frequency-shifted laser source and determines the beat
frequency between the reference laser beam and the beam reflected by the object.
The optical path difference between both laser beams results in a frequency shift,
which leads to the measurable beat frequency during beam superposition at the
detector.
The commercially available optical fiber-coupled sensor is used much like
the way the tactile probe of a CMM is used. The sensor offers an acceptance
angle of 30 , a measurement distance of ~10 mm, and a measurement rate of
1000 measurements per second, which allows faster scanning speeds than are
available with conventional tactile scanning probes. Due to its small diameter
(only a few millimeters), it can reach into the tooth gaps of automotive drive train
systems. A comparison of the optically determined gear deviation parameters of
a large gear artifact to tactile measurement results showed profile and helix
slope deviation differences of less than 0.5 μm (Günther et al. 2014; Balzer et al.
2015, 2017).
An areal interferometric measurement method has been presented by Fang et al.
and is based on the principle of phase-shifting interferometry (Fang et al. 2011,
2014; Wang et al. 2015). This method requires the reflection of the measurement
beam from the gear’s surface and delivers areal interferograms that can be converted
into flank deviation values via a correction algorithm based on simulated interfero-
grams. A comparison with tactile measurement results showed differences of less
than 1 μm in the determined flank geometry forms (A calculation of the gear
deviation parameters was not presented by the authors).
In contrast to non-contacting optical probes, tactile measurement systems with
optical position determination also exist. Härtig et al. evaluated a commercially
available, manually operated laser-tracker system for its usefulness in gear measure-
ment by comparing the measured gear deviation parameters with the calibrated
values of a large gear artifact. The evaluation of the measured data showed that
the measurement uncertainty of the manually applied probe was too large for precise
gear measurements (Härtig et al. 2013).
Another approach uses a double-beam laser interferometer to determine the
position of the tactile probe of a CMM with improved uncertainty through a
reduction in abbe errors. Two reflectors are mounted close to the probe sphere
position; for pitch and profile measurements, the measuring laser beams are oriented
in the circumferential direction of the gear tangential to the base circle, and for helix
measurements, they are oriented parallel to the gear’s axis. After mathematically
correcting the interferometric measurement results for bending deviations of the
tactile probe, an uncertainty of about 0.3 μm is obtained for the position measure-
ment (Hu et al. 2015).
In addition to distance measurements, white-light interferometry-based sensors
for the surface topography measurement of tooth flanks have also been developed.
However, due to the size of the measuring heads, it is currently not possible to
perform complete measurements in the involute area of the tooth flanks. Only those
298 K. Ni et al.
areas of the tooth flank that have almost vertical observation direction on the flank
surface, and are not covered by one of the adjacent teeth, can be measured (Günther
et al. 2014).
Areal-Oriented Measurement and Evaluation
Classical gear measurement and evaluation procedures are based on the inspection
of individual lines in a certain transverse plane (profile) or on a designated cylinder
(helix). They are therefore referred to as “line-oriented inspections.” Topograph-
ical measurements can also be conducted to show changes in the deviations on
entire gear flanks or the whole gear. However, two major drawbacks must be
considered:
• First, using a tactile system significantly increases the inspection time – an issue
that could be mitigated by introducing optical measurement technologies.
• Second, the deviation parameters explained previously are calculated on single
lines, even though it is possible to obtain a large data cloud distributed over the
entire gear flank.
Area-oriented (or “areal”) evaluation methods to establish quantitative character-

izations of gear flanks as 3D surfaces have been proposed (Pfeifer et al. 2001, 2002;
Stöbener et al. 2012; Jukl 2014; Goch et al. 2017; Ni 2017; Peng et al. 2017). Areal
evaluation uses a 3D gear model instead of a 2D profile and helix lines as the
reference geometry for gear inspection. The plumb line distance of each measured
point to the reference involute surface is calculated in a 3D coordinate system,
generating a deviation map of the measured flank.
To represent the most important features of the actual gear geometry across the
gear flank, areal deviation parameters can be created by extending the line-oriented
deviation parameters to the 3D space. Characterization algorithms such as least-
square approximation and decomposition with Chebyshev polynomials are applied
to the deviation map to evaluate the areal parameters (Ni et al. 2016). Figure 17
shows six possible areal parameters with their illustration on a gear tooth and
representation in the deviation map.
As an extension of the conventional line-oriented gear inspection, data clouds
measured on all teeth enable a holistic evaluation (Peng 2017), which means that one
set of global deviation parameters represents the common deviation patterns
observed on all teeth. Such a common deviation trend appearing similarly on all
teeth indicates systematic deviations in the manufacturing procedure – for example,
tool geometry deviations or motion control errors – and helps improve the assess-
ment of production processes and functionality predictions.
In addition to the common and single-flank deviations derived with the areal data
evaluation, point clouds enable us to determine the waviness of the flank topography,
which is essential information for the investigation of meshing noise generation.
Gravel et al. developed algorithms based on high-density areal data sets that can
9
Cylindrical Gear Metrology
Fig. 17 Areal deviation parameters

299
300 K. Ni et al.
determine periodic structures in the surface topography of the tooth flanks which
then can be correlated with the acoustic emissions of the gears (Gravel 2014; Gravel
and Kahnenbley 2017).
Traceability
In addition to the scientific and industrial advances in the field of measurement

technology, there have been some new developments in the field of gear artifacts.
Large Gear Artifacts

Until a few years ago, no traceable artifacts were available for large or micro-gears.
In 2010, the first almost workpiece-like artifact for large involute gears with a
diameter of 1 m was introduced. It contains three tooth gaps on a 60 segment of a
solid gearwheel body. The width of the gear is about 400 mm, and the gaps represent
three different gears with helix angles of 0 , 10 , and 20 in both helix directions.
The artifact is equipped with hemispherical stands to ensure a statically determined
positioning. The alignment of the artifact is carried out by the measurement of two
precisely manufactured testing waistbands with form deviations below 1 μm (Härtig
et al. 2010; Kniel and Härtig 2014).
In 2014, the world’s largest steel ring gear artifact was developed. Its dimensions
are outer diameter, 1980 mm; inner diameter, 1685 mm; width, 460 mm; and weight,
2.7 t. It embodies three different internal and three external gears in three equally
distributed positions on the ring circumference. All embodied gears are specified as
follows: normal module, 18 mm; pressure angle, 20 ; and helix angles, 0 , 10 right
and 20 left. During measurements, the ring rests on a base with six specially
developed bearing points that use gravitational action to align the ring in the middle
of the base to an accuracy of a few micrometers and ensure the lowest possible
deformation of the ring (Wiemann et al. 2017; Balzer et al. 2017) (Fig. 18).
This artifact was calibrated at PTB in Germany using the M3D3 calibration
method, which combines common tactile probing by a coordinate measuring device
with optical multilateration measurements based on at least four laser tracers (Wendt
et al. 2010, 2012; Härtig et al. 2011).
With this method, expanded measurement uncertainties smaller than 3.5 μm for
all gear parameters were achieved for the large ring gear artifact. As part of a German
national intercomparison, various project partners carried out measurements on the
artifact. This resulted in only minor deviations of less than 4 μm for all calibrated
profile and flank line deviations. Except for a few outliers, all results for all
determined gearing deviation parameters lay within the calibrated uncertainty ranges
(Wiemann et al. 2017; Balzer et al. 2017).
Micro-gear Artifacts
For micro-gears, a double-ball artifact (μ-DBA) has been developed, the basic
structure of which is similar to the already known macroscopic double-ball artifact
(Takatsuji et al. 2005; Kondo et al. 2008). It consists of three calibrated ruby spheres
Fig. 18 Large gear ring

artifact with three groups of
embodied gears on the inner
and outer surfaces
with diameters of 400 μm, 600 μm, and 700 μm. The smallest sphere in the center
represents the toothing axis, while the two larger spheres represent the involutes of
two gears with different base radii (1.8016 and 2.0386 mm) with a uniform module
of 0.169 mm. For the spheres, calibration uncertainties of less than 0.1 μm (shape
deviation) and 0.033 μm (distance deviation) have been achieved. Depending on the
measured quantity (Fα, Ffα, Fhα) and the involute embodiment used, these calibration
values resulted in estimated measurement uncertainties of 0.2–2.2 μm, obtained with
a fiber probe on a coordinate measuring device (Weckenmann et al. 2008; Lanza and
Viering 2011).
Another approach consists of five microcylinders on a cylindrical invar base
plate, which embody not only the profile lines of a gear segment but also the flank
lines and the pitch. The cylinders are fixed to the base plate by laser welding and are
commercially available in 0.001 mm increments. The authors do not indicate any
values for the measurement uncertainties achievable with this artifact in their
publication (Lanza and Viering 2011).
Workpiece-like artifacts have also been developed and tested. Wedmann et al.
presented two micro-gear artifacts, one made of titanium and one of carbide, which
were manufactured by wire erosion (Wedmann et al. 2014). They both embody four
outer straight gears with varying normal modules (1.0, 0.5, 0.2, and 0.1 mm) in four
sequences, which are equidistantly distributed along the reference circle. In order to
easily handle the artifacts, a cylindrical adapter 20 mm in height was glued to the
artifacts. The calibrated uncertainties of the slope and the form deviations are in the
range of 0.2–0.5 μm, values that are quite similar to the uncertainties of the double-
ball artifact.
Self-Calibrating Pitch Artifact

Gear artifacts for flank properties are not the only area of current research. Pitch
artifacts and their associated evaluation algorithms have been further developed as
302 K. Ni et al.
well. The calibration measurements for the rosette method mentioned in a previous
section are relatively slow due to the many measuring positions. An analysis of the
algorithm showed that the measurements of certain positions can be omitted if it can
reasonably be assumed that certain deviations with high spatial frequencies do not
occur at the rotary table. A revision of the algorithm under this assumption resulted
in a significant reduction of measurement times and showed only slight increases in
the determined uncertainties to values of approximately 0.5 μm for the pitch
deviations (Keller et al. 2017).
Uncertainty Evaluation with Virtual CMM

In addition to the very complex and time-consuming determination of the uncer-
tainty budget, according to the Guide to the Expression of Uncertainty in Measure-
ment (GUM), a Monte Carlo simulation investigating the superposition of the
individual uncertainty contributions for the respective measuring task can also be
used to assess the uncertainties of a gear measurement.
A corresponding software package (virtual CMM) was created for the determi-
nation of the uncertainties of coordinate measurements in 2000, in cooperation with
PTB, and is available to users via the manufacturers of coordinate measuring devices
(Schwenke et al. 2000; Wäldele and Schwenke 2001).
The application of the virtual CMM requires a one-time determination of the
major, individual sources of uncertainty and their contributions to the uncertainty of
coordinate measurement. In this context, both the position-dependent deviations of
the individual CMM axes, the directional deviations of the axes, and the temporary
deviations – as well as cross-sensitivities to environmental influences (such as the
mean measuring room temperature or local and temporary temperature gradients) –
must be determined with complex test measurements using reference devices (e.g.,
laser tracers) and artifacts (e.g., spherical rods). Further, there are direction-
dependent deviations in the probing process and uncertainties in probing ball and
probing head calibration that must be considered (Wäldele and Schwenke 2002).
For the uncertainty determination of a certain measuring task, normal measure-
ments are carried out with the CMM on the component to be tested. The virtual
CMM software uses this measurement data set as the basis for generating several
hundred virtual measurement data sets by adding systematic and random measure-
ment deviations to each set. The added measurement deviations result from the
superposition of randomly selected deviations in the previously determined sources
of uncertainty, whereby the selection of the deviation amounts corresponds to the
determined uncertainties of the sources. The evaluation of the virtual data sets
generated in this way leads to different results. Their statistical evaluation gives
the measurement uncertainty of the component characteristics to be determined
(Rost et al. 2016; Rost 2017).
However, the original virtual CMM version was not suitable for the determination
of gear measurement uncertainties. A corresponding project carried out at PTB
successfully extended the virtual CMM software package to measurements of
involute gears. The derived uncertainties are comparable to or lower than the
uncertainties derived from conventional uncertainty estimations (Rost et al. 2016).
An alternative approach is presented by Takeoka et al., who developed a virtual

gear checker. It is also based on the determination of the uncertainty sources of the
gear-measuring instrument and a subsequent Monte Carlo simulation. In this appli-
cation, simulated data instead of real gear geometry measurement data are used as
the measurement database (Takeoka et al. 2009).
Summary
As key components in transmission systems, the quality of gears must be controlled

to ensure their performance. Cylindrical gears and involute geometry have been
known for centuries. Classical line-oriented gear metrology has been well
established and standardized across the entire gear industry. But to fulfill the
increasing performance requirements of today’s transmission systems, classical
gear metrology had to be improved upon to meet the tight tolerances, complex
designs, and short cycle times. Consequently, developments in gear metrology over
the last two decades have focused on new sensing technologies, geometric confor-
mance evaluation techniques, and traceability establishment methods. The remark-
ably large range of gear sizes and weights has extended the limits of many gear
technologies. Advancements in optical metrology and computational power have
largely facilitated this progress.
References
Auerswald M, Freyberg A, Fischer A (2017) Optical sensor system for 3D measurements on large
gears. In: Proceedings of AMA conferences – SENSOR, pp 227–232
Balzer F, Schäfer M, Lindner I, Günther A, Stöbener D, Westerkamp J (2015) Recent advances in
optical gear measurements. In: Proceedings of international conference on gears, 11 pp
Balzer F, Steffens N, Stein M, Kniel K (2017) Traceable measurement of large gears with micron
accuracy. In: Proceedings of 59th Ilmenau Scientific Colloquium, urn: nbn:de: gbv: ilm1-
2017iwk-139:9, 16 pp
BIPM (2008) Evaluation of measurement data – guide to the expression of uncertainty in measure-
ment. International Organization for Standardization, Genève
Bosch T, Lescure M, Roviras D (1992) The physical principles of wavelength-shift interferometric
laser rangefinders (in French). J Opt 23(3):117–123
Bouzakis K, Lili E, Michailidis N, Friderikos O (2008) Manufacturing of cylindrical gears by
generating cutting processes: a critical synthesis of analysis methods. CIRP Ann Manuf Technol
57(2):676–696
Fang S, Wang L, Liu S, Komori M, Kubo A (2011) Positioning the actual interference fringe pattern
on the tooth flank in measuring gear tooth flanks by laser interferometry. Opt Eng 50(5):055601,
6 pp
Fang S, Zhu X, Yang P, Cai Q, Komori M, Kubo A (2014) Analysis and compensation method for
installation error in measuring gear tooth flank with laser interferometry. Opt Eng 53(8):084111,
9 pp
Farago F, Curtis M (1994) Handbook of dimensional measurement. Industrial Press Inc, New York
Goch G (2003) Gear metrology. CIRP Ann Manuf Technol 52(2):659–695
Goch G, Ni K, Peng Y, Guenther A (2017a) Future gear metrology based on areal measurements
and improved holistic evaluations. CIRP Ann Manuf Technol 66(1):469–474
304 K. Ni et al.
Goch G, Ni K, Peng Y, Guenther A (2017b) Paradigm change in cylindrical gear metrology using
areal measurement and evaluation. In: International conference on gears, Garching
Goch G, Ni K, Peng Y, Guenther A (2017c) Paradigm change in gear inspection based on a holistic
description, measurement and evaluation of gear flanks. In: 32nd ASPE annual meeting,
American Society for Precision Engineering, Charlotte
Goch G, Peng Y, Ni K, Guenther A (2017d) Optical and areal measurement and evaluations of
cylindrical gears. In: 17th international VDI congress, Bonn
Gravel G (2014) Simulation as support of the analysis of waviness. VDI reports 2236. VDI Verlag,
Düsseldorf, pp 69–80
Gravel G, Kahnenbley T (2017) New developments for the waviness analysis of acoustically
conspicuous gears. VDI reports 2316. VDI Verlag, Düsseldorf, pp 43–54
Günther A (1996) Flächenhafte Beschreibung und Ausrichtung von Zylinderrädern mit
Evolventenprofil, Diploma Thesis, Ulm University
Günther A, Peters J, Goch G (2009) Flächenhafte numerische Beschreibung, Ausrichtung und
Auswertung von Zylinderrädern (3D-Surface-like Numerical Description, Alignment, and
Evaluation of Involute Cylindrical Gears). tm – Technisches Messen Plattform für Methoden,
Systeme und Anwendungen der Messtechnik. 68(4/2001)
Günther A, Balzer F, Lindner I, Stöbener D, Westerkamp J, Goch G (2014) Application of
coordinate measuring devices for large gearings (in German). VDI reports 2243. VDI Verlag,
Düsseldorf, pp 139–154
Härtig F (2005) New developments on gear metrology. In: Third tri-national conference of the
North American Coordinate Metrology Association (presentation)
Härtig F, Rost K, Goch G (2010) Large gear material standard for the traceability of gears for
transmission manufacturing. VDI reports 2108. VDI Verlag, Düsseldorf, pp 991–1004
Härtig F, Franke M, Kniel K, Wendt K (2011) Coordinate measurement technique considering the
3D-Abbe principle. In: Proceedings of 10th IMEKO symposium on laser metrology for preci-
sion measurement and inspection in industry, Braunschweig, 12 pp
Härtig F, Lin H, Kniel K, Shi Z (2013) Laser tracker performance quantification for the measure-
ment of involute profile and helix measurements. Measurement 46:2837–2844
Hu L, Zi X, Yang G (2015) Correction of Abbe error in involute gear measurement using a laser
interferometric system. In: Proceedings of SPIE 9671, 96711N
ISO 10360 (2000 to 2016) Geometrical product specifications (GPS) – acceptance and reverification
tests for coordinate measuring machines (CMM). Part 1 to 10 and 12, 2000 to 2016
ISO 1328-1 (2013, 1997) Cylindrical gears – ISO system of accuracy. Part 1, 2013, Part 2, 1997
ISO 18653 (2003) Gears – evaluation of instruments for the measurement of individual gears
ISO 21771 (2007) Gears – cylindrical involute gears and gear pairs – concepts and geometry
lSO/TR 10064 (1992 to 1998) Cylindrical gears – code of inspection practice. Part 1 to 5, 1992-03
to 1998-05
Jukl P (2014) Point clouds of gearings (in German). VDI reports 2236. VDI Verlag, Düsseldorf, pp
105–121
Karpuschewski B, Knoche H, Hipke M (2008) Gear finishing by abrasive processes. CIRP Ann
Keller F, Stein M, Kniel K (2017) A shortened rosette method for the calibration of pitch deviations
(in German). VDI reports 2316. VDI Verlag, Düsseldorf, pp 65–76
Kniel K, Härtig F (2014) National and international intercomparisons of gear measurements
Kniel K, Härtig F, Osawa S, Sato O (2009) Two highly accurate methods for pitch calibration. Meas
Sci Technol 20(11):115110
Kondo Y, Sasajima K, Noguchi S, Kondo K, Osawa S, Naoi K, Takatsuji T (2008) Tooth form
evaluation using ball artifact – development of a measuring instrument of a ball center distance
traceable to national standard of length. Key Eng Mater 381–382:595–598
Lanza G, Viering B (2011) A novel standard for the experimental estimation of the uncertainty of
measurement for micro gear measurements. CIRP Ann Manuf Technol 60:543–546
Lotze W, Haertig F (2001) 3D gear measurement by CMM. Laser Met Mach Perform
34:333–344
Matthias S, Loderer A, Koch S, Gröne M, Kästner M, Hübner S, Krimm R, Reithmeier E,
Hausotte T, Behrens BA (2016) Metrological solutions for an adapted inspection of parts and
tools of a sheet-bulk metal forming process. Prod Eng Res Dev 10:51–61
Meeß K, Kästner M, Seewig J (2006) Reduction and evaluation of the uncertainty of measure-
ment of optical gear measurement using Fringe projection (in German). Tech Mess
73(11):603–610
Ni K (2017) Areal gear metrology with modified flanks. Dissertation, University of North Carolina
at Charlotte
Ni K, Peng Y, Goch G (2016) Characterization and evaluation of involute gear flank data using an
areal model. In: 31st ASPE annual meeting, American Society for Precision Engineering,
Portland, pp 184–189
Noch R, Steiner O (1966) Die Bestimmung von Kreisteilungsfehlern nach einem Rosetten-
verfahren. Zeitschrift für Instrumentenkunde, Braunschweig
Peng Y, Ni K, Goch G (2017) Areal evaluation of involute gear flanks with three-dimensional
surface data. AGMA, Fall technical meeting, October 22–24, 2017, Columbus, OH
Pfeifer T, Kurokawa S, Meyer S (2001) Derivation of parameters of global form deviations for
3-dimensional surfaces in actual manufacturing processes. Measurement 29:179–200
Pfeifer T, Napierala A, Mandt D (2002) Functional-orientated evaluation of modified tooth flanks.
VDI reports 1665. VDI Verlag, Düsseldorf, pp 769–783
Rost K (2017) Ermittlung einer aufgabensp ezifischen Messunsicherheit für Verzahnungs-
messungen. Dissertation, University of Bremen
Rost K, Wendt K, Härtig F (2016) Evaluating a task-specific measurement uncertainty for gear
measuring instruments via Monte Carlo simulation. Precis Eng 44:220–230
Schwenke H, Siebert B, Wäldele F, Kunzmann H (2000) Assessment of uncertainties in dimen-
sional metrology by Monte Carlo simulation: proposal of a modular and visual software. Ann
CIRP 49(I):395–398
Stöbener D, Freyberg A, Fuhrmann M, Goch G (2012) Areal parameters for the characterisation of
gear distortions. Mater Werkst 43(1–2):120–124
Takatsuji T, Kondo K, Kubo A, Härtig F, Osawa S, Naoi K, Kurosawa T, Komori M (2005)
Performance assessment of involute gear measurement by CMM using a double-ball artifact. In:
Proceedings of SPIE 5879 recent developments in traceable dimensional measurements III,
58790Q
Takeoka F, Komori M, Takahashi M, Kubo A, Takatsuji T, Osawa S, Sato O (2009) Gear checker
analysis and evaluation using a virtual gear checker. Meas Sci Technol 20:045104, 11 pp
VDI/VDE 2612 (2000) Profile and helix checking of involute cylindrical gears. Beuth, Berlin
VDI/VDE 2613 (2003) Pitch and runout testing on gearings – cylindrical gears, worm wheels, bevel
gears
Wäldele F, Schwenke H (2001) Determination of the measurement uncertainty by simu-
lation – the virtual CMM (in German). VDI reports 1618. VDI Verlag, Düsseldorf,
pp 103–114
Wäldele F, Schwenke H (2002) Automated calculation of measurement uncertainties on CMMs –
towards industrial application (in German). Tech Mess 69(12):550–557
Wang L, Fang S, Yang P, Meng L (2015) Comparison of three methods for identifying fringe
regions of interference fringe patterns in measuring gear tooth flanks by laser interferometry.
Optik 126(24):5668–5671
Weckenmann A, Wiedenhöfer T, Büttgenbach S, Krah T, Fleischer J, Buchholz I, Viering B,
Kranzmann A, Ritter M, Krüger-Sehm R, Bakucz P, Schmitt R, Körfer F (2008) Trends in
development of standards for micro- and nanometrology: chances and challenges (in German).
Tech Mess 75(5):288–297
Wedmann A, Kniel K, Dunovska V, Härtig F, Klemm M (2014) Tracing of micro gear measure-
ments (in German). VDI reports 2236. VDI Verlag, Düsseldorf, pp 199–212
306 K. Ni et al.
Wendt K, Franke M, Härtig F (2010) Mobile multi-lateration measuring system for high accurate
and traceable 3D measurement of large objects. In: Proceedings of the 10th ISMQC,
pp 224–227
Wendt K, Franke M, Härtig F (2012) Measuring large 3D structures using four portable tracking
laser interferometers. Measurement 45:2339–2345
Wiemann A, Stein M, Kniel K (2017) Traceability of gear measurements for large gear boxes
High-Speed Measurement of Complex
Shaped Parts by Laser Triangulation for 10
In-Line Inspection
Alexander Schöch and Enrico Savio
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 308
Overview of Measuring Techniques . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
Coordinate Measuring Machines . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 309
Photogrammetry . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 310
Fringe Projection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 310
Laser Tracking . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 311
Other Systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312
Optical Triangulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312
Case Study: Inspection of Complex Shaped Parts at Elevated Temperature . . . . . . . . . . . . . . . . . . 315
System Description . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 316
2D Scanner . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 317
Linear Motion System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320
Data Fusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320
Modelling and Correction of Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320
Temperature Effects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321
Laser Planes Alignment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 322
System Characterization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 324
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327
Abstract
Measurement of complex shaped parts is of interest in many applications.
Complex functional freeform surfaces may have a great influence on the
A. Schöch
Institute for Production Metrology, Materials and Optics, NTB University for Technology, Buchs,
Switzerland
e-mail: alexander.schoech@ntb.ch
E. Savio (*)
Department of Industrial Engineering, University of Padova, Padova, Italy
e-mail: enrico.savio@unipd.it

https://doi.org/10.1007/978-981-10-4938-5_10
308 A. Schöch and E. Savio
performances of a product. Geometrical deviations in manufacturing can cause

waste of large quantities of energy. Testing of parts having freeform surfaces is
a key activity during the development of products with better performances.
Depending on the workpiece shape, its manufacturing process, and tolerance
limits, it is required to measure freeforms densely, for sufficiently high proba-
bility to capture crucial regions. This often requires fast measurement processes
to keep measurement time feasible low. Fast measurements can be obtained by
increasing the speed of acquiring a single point and/or parallelizing this process
by acquiring multiple points at the same time. A brief overview of methods with
high measurement speed is given. The concept of optical triangulation is
discussed in detail, and case study is presented to illustrate an implementation
in industry of a dedicated measuring system for high-speed measurement of
complex shapes.
Keywords
Freeform · Complex shape · Optical measurement · Optical triangulation · Laser
triangulation · High-speed measurement · Multi-sensor measurement · In-line
inspection · Dense sampling · Industrial application
Introduction
The measurement of complex shaped parts is of interest in many applications. A

typical example is parts having freeform surfaces in industries such as automotive,
aerospace, household appliances, and others. Freeforms (also called sculptured
surfaces or curved surfaces (Savio et al. 2007)) are defined by having no invariance
degree. Similar to the degree of freedom in kinematics, the invariance degree is the
number of principle rotations (rotations around coordinate system axes) and trans-
lations for which an ideal feature is kept identical (Savio et al. 2007). Another
definition describes freeforms as arbitrary complex surfaces without rotational or
translational symmetries (Jiang et al. 2007).
Functional freeform surfaces may have a great influence on the performances of a
product; in the design of a turbomachine, for example, freeform geometry for both
static and rotating components is of paramount importance. Geometrical deviations
in manufacturing cause inefficiencies that can cause waste of large quantities of
energy. Design, engineering, and testing of parts having freeform surfaces are key
activities for the development of products with better performances (Savio et al.
2007). Depending on the workpiece shape, its manufacturing process, and tolerance
limits, it is required to measure freeforms densely, for sufficiently high probability to
capture crucial regions. Denser sampling requires either a long measurement time or
fast measurement process. The latter can be obtained by increasing the speed of
acquiring a single point and/or parallelizing this process by acquiring multiple points
at the same time.
For some demanding applications, it can be advantageous to measure the
workpiece during the production process; in-line metrology allows improvement
10 High-Speed Measurement of Complex Shaped Parts by Laser Triangulation. . . 309
of accuracy and a reduction of machining time (Nishiguchi et al. 1991) through the
elimination of repositioning and alignment operations. More in general, for both
higher efficiency and effectiveness, measurements should be performed as early as
possible in the production process (e.g., Keferstein 2015; Savio 2012;
Weckenmann 2012). If a defective part can be identified early, there is no reason
for further processing. Further, the impact of defective parts on overall production
costs rises as the manufacturing process progresses (Kunzmann et al. 2005;
Savio et al. 2016).
The scope of this chapter is to provide a short overview on the measuring
techniques that can be helpful for the measurement of complex shaped
parts, with focus on throughput (parts measured per unit time). There exists
a variety of methodologies to measure complex shaped parts, of which some
prominent ones are briefly reviewed here. For a more in-depth discussion
of measurement of complex shaped parts, the reader is referred to Savio
et al. (2007).
More information is provided for laser triangulation in a dedicated section, due to
its relevance for industrial applications. A case study is then presented, to illustrate in
detail an example of implementation in industry of a dedicated measuring system for
high-speed measurement of complex shapes.
Overview of Measuring Techniques
The measuring techniques of interest at those having the capability to acquire

multiple points on the part surface, in a short time. Depending on the application
requirements, the most suitable and cost effective measuring technique can be
selected among several available. The most important measuring systems are pre-
sented in the following, with focus on freeform measuring capabilities and
productivity.
Coordinate Measuring Machines
The most important general-purpose instruments for the inspection of complex

shaped parts in industry are Coordinate Measuring Machines (CMMs), which can
be equipped with both contact and noncontact probing systems. CMMs are well
accepted in industry because they are very flexible and allow the measurement of
points in space with high accuracy. They are today typically provided with numeric
control and scanning probes, both contact and noncontact; automatic measurements
of thousands of points are easily performed, even on complex surfaces. When the
geometrical accuracy of freeform surfaces is critical for the functional performance,
or for the appearance of a product, the inspection process is typically conducted on
CMMs. Contact measuring probing systems are discussed in Weckenmann et al.
(2004); they can measure up to 200 points/s at speeds as high as 150 mm/s. Faster
measurements are possible with noncontact probing systems based on optical
principles like autofocus, triangulation, and others. Their advantages are, in general,
the noncontact nature of the measurement principle, fast acquisition of a large
number of points, and high degree of flexibility in measurement settings. Details
of the optical measuring principles for dimensional metrology in production can be
found in Schwenke et al. (2002).
The mechanical linkage between a fixed reference and the probing system can
also be provided by a series of connected segments with encoders, called “articulated
arm CMM.” Measuring volume can reach few meters in radius. The device is
manually operated and is portable; both contact and noncontact probing systems
(e.g., a laser scanner) are available on commercial products. The joint angles
encoders positions and distance data from the laser scanner are thereby acquired
and fused into one dataset located in the machine coordinate-system. Integrating
multiple such datasets during the movement eventually yields a holistic representa-
tion of the parts surface. Being portable, it can be useful for the measurement of
complex parts with accessibility difficulties on-site, such as die cavities, heavy
equipment, or simply parts that cannot be moved. Manual operations are providing
high flexibility; however, the articulated arm CMM is not optimal in cases where a
number of similar parts have to be measured in short time, due to the manual
measurement procedure.
Photogrammetry
Photogrammetry is based on the projective transformation involved in photo-

graphing a part from different viewing angles. By taking at least two images with
one or multiple calibrated cameras, the 3D shape of the object can be described
qunatitatively (Weckenmann 2012). This triangulation-based principle requires to
solve the correspondence problem, i.e., ascertaining which regions in one image
correspond to which region in another image. A feasible solution implies that
individual regions can be sufficiently uniquely identified, i.e., exhibiting a non-
homogeneous surface topography in the spectrum of the camera. If the part to be
inspected does not have an adequate texture, physical markers can be attached to the
surface (Savio et al. 2007). This results in a significant loss of lateral resolution, since
only the markers 3D positions are recorded in contrast to many surface regions.
Photogrammetry can be used to implement measurement of freeform parts.
However, if markers become necessary due to lack of nonhomogeneous surface
topography, the degree of automation, measurement speed, and resolution are
drastically reduced.
Fringe Projection
Fringe projection is a technique in which the position of points of interest is also

calculated via optical triangulation; unlike photogrammetry, in which the measured
surface is usually provided with physical markers, fringe projection techniques
measure surfaces without physical markers. Optical triangulation is obtained, for

example, projecting a narrow stripe light to a complex shaped surface, producing an
illumination that appears distorted from other perspectives. Observing this deformed
stripe of light with a camera, multiple points can be acquired using the triangulation
principle. The projection of patterns consisting of multiple light stripes at once
allows the simultaneous acquisition of a high number of points over an area. The
use of this approach from different viewpoints allows the digitization of the full
geometry of a part.
A typical measurement volume of a fringe projection system is in the range of
side between 0.1 m and 1 m, with relative accuracy of up to 10 4. This level of
accuracy and the high speed of measurement are compatible with the requirements
of some industrial measurement tasks; therefore, the technique is often used in
industry.
Laser Tracking
The measurement of large parts requires techniques capable of recording point

coordinates with measuring ranges of several meters; the most common techniques
are laser tracking interferometry, photogrammetry, and laser radar. Measuring prin-
ciples, capabilities, limitations, and application examples can be found in Estler
et al. (2002).
The laser tracker measures the position of an independent portable target,
usually a cube corner retroreflector, which is manually positioned on the surface
of interest by an operator. Laser trackers emit laser radiation and adjust two
angular actors such that the laser ray follows a physical reflector. Contemporary,
the distance along the ray to the reflector is measured, e.g., by interferometric
methods. The need of an operator for moving the target is somewhat limiting the
number of probing positions, so the measuring process is intrinsically slow.
Measurements can be performed at rates of multiple thousand points per second
(Savio et al. 2007) at distances up to 100 m (Weckenmann 2012) while moving the
target; therefore, high density profiles can easily be scanned over large freeform
surfaces.
Laser trackers can be equipped with a 6 DOF-probe instead of a reflector. The
laser tracker determines not only one 3D position but also the orientation of the
probe. The probe simultaneously measures the distance(s) to the parts surface, e.g.,
implemented as a laser triangulation sensor (Hexagon Manufacturing Intelligence
2018). An example of laser tracker system with an integrated handheld laser scanner
is shown in Fig. 1. The measuring principle of the laser scanner is presented in the
next section.
The laser radar is other type of laser-based measuring system for large scale
metrology, using no targets, in which a coherent frequency modulated laser provides
a robust and eye-safe signal for a noncontact measurement capability up to 50 m
radius. The main advantage is the complete automation of the measurements process
(Nikon Metrology 2018).
Fig. 1 Example of laser tracker system with an integrated handheld laser scanner (Hexagon
Manufacturing Intelligence 2018)
Other Systems
Machine tools can also act as positioning systems for optical sensors.
Industrial robots are another possible fully automated solution for the positioning
of optical measuring systems, as an alternative to CMMs, for repetitive
inspection tasks on production lines where the need of accuracy is limited,
i.e., for sheet metal applications. Other systems are available on the market for
specific tasks.
Automated laser scanners with positioning and rotating stages are common
examples found in applications where high productivity is needed. When parts to
be measured are similar in shape and dimensions, dedicated measuring equipment
can be designed for application-specific solutions. An example of special-purpose
measuring device is described in the last section of this chapter.
Optical Triangulation
Optical triangulation methods such as photogrammetry, fringe projection, or

laser triangulation are among the most common approaches used in optical
distance measuring systems (Pfeifer 2002). The measurement principle of
triangulation relies on the geometrical arrangement of at least two sensors
(passive, i.e., only ambient light is used and no energy is emitted for sensing
purposes) or a combination of an emitter and a sensor (active, i.e., emitting of light

(pattern) on the workpiece for sensing purposes). The sensors are position-
sensitive, i.e., they can infer the position where the reflected light from the
workpiece hits the sensor.
In optical triangulation, the optical axes of both light emitter and photosensitive
sensor are adjusted towards a common point. The object to be measured is positioned
approximately at this point. By sufficiently diffuse reflection of the object, light from
the actor is reflected towards the sensor(s). By known distance and angle – termed
“baseline” and “triangulation angle,” respectively – between sensor and actor, a
distance to the object can be calculated by geometrical means (case 1 and case 2 in
Fig. 2). If a light emitter is used, its light is to be collimated to illuminate only a finite
region on the objects surface. The setup forms a triangle as shown in Fig. 2, thus the
name triangulation.
The illuminated region is to be imaged by the photosensitive sensors optics, e.g.,
a digital camera with lens. Since the distance between sensor and illuminated surface
varies during a measurement, optics with a sufficient depth of field are to be used or
the Scheimpflug principle is to be applied (Schwenke et al. 2002). The latter tilts the
camera sensor relative to the imaging lens as depicted in Fig. 3. Thus, the depth of
field is tilted relative to the optical axis. By tilting by the triangulation angle, each
position on the laser light path is in focus.
For active optical triangulation, a position sensitive diode (PSD) or a digital
camera (CCD, CMOS) can be utilized as sensors. The latter provides a better
accuracy due to image processing methods applied to accurately determine the
position on the sensor (Schwenke et al. 2002). A comparison of such methods is
given in Fisher and Naidu (1996).
Fig. 2 Sketch of a
triangulation measurement
with one active component,
i.e., a laser emitter, and a
passive digital camera. The
triangulation angle is
represented by α and the
baseline by b. A change in the
distance between camera and
workpiece changes the
position where the laser hits
the surface. This will be
evident in the cameras image
as indicated by case ① and
case ②
Fig. 3 Sketch of the

Scheimpflug principle. Note
that the angle between object
and lens plane is equal to the
angle between lens plane and
image plane
Mostly driven by the advances in the semiconductor domain, digital cameras

performance increased and prices dropped significantly over the past decade, what
made triangulation sensors very attractive for many industrial measurement tasks.
They parallelize the acquisition of 3D coordinates to a high degree by the use of
adequate light emitters, i.e., a laser that projects a line instead of a spot, generating
many distance measurements in one single image. As with other optical systems, a
high number of coordinates per time are acquired by such optical light-sectioning
systems (Hocken and Pereira 2012). In multisensor systems equipped with high-
speed sensors, millions of coordinates per second can be acquired (Schöch et al.
2015).
However, optical methods can be sensitive to ambient lighting which is often
inevitable in industrial environment. Passive triangulation methods such as photo-
grammetry require an adequate parts texture. Since no collimated light emitter is
involved, positions on an object must be inferred from its surface, i.e., texture,
illuminated by ambient light. The ambient light can be replaced by a noncollimated
artificial light source what leads to higher repeatability of such systems, minimizing
the influence on often not controllable ambient light sources. Enclosing of a mea-
surement system is, however, often not feasible in industry, e.g., for in situ
metrology.
Active triangulation methods project a structured pattern, e.g., a laser line, onto
the workpieces surface, thus resulting in more texture agnostic performance. Active
systems are also less sensitive to ambient light due to the possibilities of light
modulation and spectral filtering. The latter can be leveraged by laser triangulation
systems due to the monochromatic nature of laser light. By placing an appropriate
optical bandpass filter inside the optical path of the camera, the influence of typical
ambient light sources can be reduced significantly (Schöch et al. 2014).
The accuracy of passive and active methods depends to a high degree on the
surface of the workpiece (Schwenke et al. 2002). Translucency, nonuniform specu-
larity, and nonuniform reflectivity cause deviations in measurement results. The
perfect optically cooperative surface would have a completely diffuse uniform
reflection, which is rarely encountered in reality. To increase the performance for
noncooperative surfaces, a diffuse coating can be applied for the measurement
process.
Another source of uncertainty is the relative pose between sensor and workpiece
surface. For example, with partly specular surfaces, overexposure on the sensor can
appear for specific triangulation angles. This can lead to blooming, where intensity
drains into nearby pixels on the chip, potentially increasing the uncertainty of the
measurement. Methods to compensate for deviations resulting from pose differences
exist (Isheil et al. 2011; Xi et al. 2001).
Case Study: Inspection of Complex Shaped Parts at Elevated

Temperature
In the following, an industrial application of high-speed measurement of complex

parts by laser triangulation is described. A multisensor light sectioning coordinate
measuring systems (CMS) was developed for the inspection of forged turbine blades
at elevated temperature, right after the hot forging step (Ghiotti et al. 2015; Schöch
2016; Schöch et al. 2014, 2015).
Turbine blades are used in a variety of industrial products (Savio et al. 2007), e.g.,
cooling fans, turbo charger systems, and jet engines. They have both freeform
geometries and root geometries; typical tolerances are between 0.1 mm and
0.5 mm profile tolerance on the freeform aerofoil and between 0.02 mm and
0.05 mm on root geometries, where coupled to the turbine rotor (Fig. 4). The
manufacturing process chain includes multistep hot forging and machining opera-
tions as shown in Fig. 5.
Early measurements performed at intermediate manufacturing steps are of great
interest, because this may reduce the overall manufacturing costs (e.g., Keferstein
2015; Savio 2012; Weckenmann 2012). If a defective part can be identified early,
there is no reason for further processing. Further, the economic impact of defective
parts rises as the production process progresses, since the part exhibits higher
economic value.
Process-intermittent measurements at elevated temperature have several advan-
tages as compared to postprocess measurements. For example, if other process steps
follow, a statement about their feasibility can be given (Fig. 5, cooling prediction). In
addition, measurements directly after a production process step, i.e., process-inter-
mittent measurements (Hocken and Pereira 2012), when the process is “momentarily
halted,” allow for selective optimization of the individual step (Fig. 5, process
adjustment).
Measuring workpieces at elevated temperature is complex and subject of current
research. Most recently, optical technologies are utilized for dimensional
Fig. 4 Example of turbine blade, at the end of the manufacturing operations
Fig. 5 Schematic representation of the phases of the process-chain for production of turbine
blades, including process-intermittent measurements
measurements at elevated temperature. Therein, three different technologies are

mainly reported: time of flight (TOF), optical triangulation, and image processing.
An overview of scientific reports and commercial products is given in
Schöch (2016).
Recall that the systems purposes are process-intermittent measurements, directly
after the forging step. At that stage, forging tolerances to be verified are in the order
10 3 m. The target Maximum Permissible Error (MPE) for length measurements
was set at 0.2 mm level.
System Description
The prototype CMS is composed of two main subsystems: a 2D scanner and a

custom linear axis. The scanner acquires 2D cross-sections of the workpiece by
means of laser line triangulation, while the linear axis moves the piece through the
Fig. 6 CAD model of the CMS with machine coordinate system and sensor arrangement as
overlay. Top right: sketch of the metrology frame with measurement area of the 2D scanner inside
the yellow circle
2D scanner. From cross-sections, a 3D model is generated. Figure 6 shows a model

of the system. A summary of main technical data is available in Table 1.
2D Scanner
The 2D scanner has a modular design to mount up to eight light sectioning sensors.
Each sensor, consisting of a line laser, a Complementary Metal-Oxide-Semiconduc-
tor (CMOS) camera, and optical bandpass filter, is arranged on an octagonal
metrology frame (Fig. 6, top right and Fig. 7). Eight sensors are used to obtain
complete surface information, even for partly concave cross-sections. Using the
maximum number of sensors, the measurement area of the 2D scanner is a circle
800 mm diameter.
To keep sensors free from dust in the harsh shop floor environment, an integrated
blowing system is providing a constant flow of clean air in front of each of the sensor
windows. Temperature sensors (Pt100, class B) are installed on the top of the
octagonal frame, on the housing and inside the housing to monitor the temperature
Table 1 Summary of main technical data of the prototype CMS

Characteristic Value
Dimensions 2.6 2.5 1 m
(H W D)
Measurement speed 320 Hz, 1,000,000 points =s (typical)
Measurement time <10 s (for 800 mm long parts)
Measurement volume 860 mm disc with ø 800 mm (cylindrical)
Sensor camera type CMOS sensor, 2048 1088, global shutter, GigE interface,
5.5 5.5 μm pixel size
Sensor depth 0.03 mm (typical)
resolution
Sensor sampling 0.3 mm (typical)
distance
Laser emitter Laser diode with cylindrical lens
Laser wavelengths 405 nm, 450 nm, 640 nm and 670 nm
Optical bandpass FWHM = 20 nm, centered at the respective laser wavelength
filters
Internal air 20 1 C
temperature
Pyrometer type Optris CTlaser 3 M (GmbH 2018)
Pyrometer accuracy (0.3% of reading +2 C)
Fig. 7 3D sketch of the

metrology frame. Laser
wavelengths are in the
respective colors and sensors
are schematically shown with
their numbering
of the respective components. Special attention was given to the thermal insulation
of the metrology frame from the heat radiation caused by workpieces at elevated
temperature. In particular, the metal plates protecting the system from radiation are
water-cooled by cooling coils. The air temperature inside of the housing is kept
within 20 1 C. This is achieved by two conditioning systems: an air chiller and an
auxiliary water flow inside the metrology frame. The latter is enabled only if the
chiller is not sufficient to maintain the temperature within the specified range.
Each sensors laser plane is generated by a point laser emitter (laser diode) using
optics that contain a cylindrical lens. Laser emitters have four different wavelengths
(405 nm, 450 nm, 640 nm, and 670 nm) where opposite lasers are sharing the same
wavelength (Fig. 7). This setup was chosen to separate data between sensors, i.e.,
each camera acquires only light from its laser emitter. Opposite laser light is assumed
invisible to a sensor due to the presence of an opaque workpiece between them. An
optical bandpass filter is installed in front of each camera to minimize the interfer-
ence with other lasers, ambient light, and the infrared radiation originating from the
workpiece at elevated temperature.
The laser plane generates a line on the surface of a workpiece (Fig. 8). The CMOS
cameras record this laser line and extract its positions on the image sensor by means of a
Field Programmable Gate Array (FPGA) and appropriate algorithms. The baseline, i.e.,
the distance between camera and laser emitter, has a nominal length of 300 mm and the
angle a between optical axis of camera and laser emitter is nominally 30 (Fig. 7). Laser
line positions in pixel coordinates are then transformed into coordinates of the laser
plane to obtain a set of distance measurements. The datasets from all sensors are
appropriately combined to achieve a holistic measurement of a cross-section.
In addition to light sectioning sensors, up to three pyrometers can be installed on
top of the 2D scanner (Fig. 6). They measure the workpiece surface temperature at
the cross-section with an accuracy of (0.3% of reading +2 C), as stated by the
manufacturer.
Fig. 8 Photo of a hot billet at

1000 C inside the CMS.
Laser lines are diffusely
reflected on the workpiece and
visible in the picture
Linear Motion System
The linear motion system (Fig. 6) is providing z-position information by a magnetic

encoder system with a resolution of 2 μm. A crucial advantage of a magnetic encoder
system, in contrast to an optical encoder system, is its resistance to harsh industrial
environment conditions (e.g., dusty air). The magnetic scale is protected from heat
radiation in order to reduce distortions due to temperature gradients. To avoid
occlusion by the linear motion system, it was designed such that its two rails are
located outside the measurement area of the laser scanner.
The maximal travel distance is 860 mm what is also the measurement range of the
CMS along the Z-axis. Combining the circular measurement area of the laser scanner with
this measurement range yields a cylindrical measurement volume as specified in Table 1.
Data Fusion
All data of 2D sections, position, and pyrometer temperature are collected synchro-
nously using an electronic trigger signal. The section points from all cameras are
merged into complete 2D sections. Measurement points in 3D space are obtained by
augmenting the 2D point information of the sections with information from the
encoder system. By combining measurement points from each section, a 3D point-
cloud is constructed. A customized meshing algorithm (Schöch 2016) then generates
a Triangulated Irregular Network (TIN) model (i.e., a STL file) of the pointcloud as
shown for a turbine blade (Fig. 9) in Fig. 10. A comparison between the geometry at
850 C and at workshop temperature is given in Fig. 11.
Depending on the exposure time of the CMOS cameras and their AOI (Area of
Interest), the system can deliver complete section profiles at acquisition frequencies
of up to 200 Hz (for 2D measurements the frequency can be up to 500 Hz). For a
typical blade (e.g., Fig. 6), the number of points per cross-contour is about 3000,
resulting in 600,000 points/s. The motion platform can move at about 0.16 m/s. For a
700 mm long turbine blade, a full 3D measurement takes about 4.2 s, generates
840 cross-contours and approx. 2.5 million points.
The loading and unloading of a turbine blade on/from the prototype motion
platform takes about 10 s each. Forward and backward movement of the linear
axis takes about 12 s each, which results in a total measurement cycle time of about
42 s. In the built prototype, flexibility was preferred over measurement speed. In a
final realization in forging processes, the transportation system can be automated and
optimized to minimize cycle times. For endless products, the loading and unloading
processes are not necessary at all.
Modelling and Correction of Errors
Temperature effects and the alignment of laser planes (i.e., coplanarity of laser
curtains) were found to be the two most important error contributors. Other error
Fig. 9 Photograph of a turbine blade at high temperature during measurement. Laser lines are
diffusely reflected on the workpiece
Fig. 10 STL Model

reconstructed after
measurement of a turbine
blade measured at approx.
800 C
sources, e.g., vibration in forging plant, geometrical errors of the linear axis, were
estimated and considered to be negligible in relation to the target MPE.
Temperature Effects
The magnitude of temperature effects on the CMS was assessed by repeated

measurements on a calibrated artefact during 30 min of operation with the temper-
ature control disabled, showing errors up to 0.2 mm. By the use of two conditioning
systems – an air chiller and a water flow inside the octagonal frame – the internal
temperature is controlled at 20 1 C. In this way, temperature effects on the CMS
Fig. 11 Deviations map of the same blade at 850 C and at workshop temperatures
could be reduced to 50 μm for length measurements of a calibrated sample of 120 mm

nominal size. Thermal stability was also verified during 8 h of operation, while 10 hot
workpieces were measured (Schöch et al. 2014). The temperature influence on the
sensor components (CMOS chip and focal length elongation) was estimated according
to (Clarke et al. 1991) for an upper bound of ΔT = 5 C with a resulting error of less
than 10 μm. The heat radiation from the hot workpiece was found to have no direct
influence on the raw image data. Due to the installed optical bandpass filters, the
radiation effect on the image sensor lies below the threshold used for laser line
detection while attenuating the laser signal by 10%. The refraction of the laser light
in air due to temperature gradient was also considered. Using Ciddors equation
(Ciddor 1996) and a simplified model with a single refractive index change between
sensor (30 C) and workpiece (100 C) yields a maximum error of 30 μm for distance
measurements. Due to the relatively low influence of internal, external, and sensor
temperature residual deviations, these effects were not corrected.
Laser Planes Alignment
Bernstein (2011) investigated the influence of angular misalignments for 2D measure-

ments on constant cross-sectional profiles with the conclusion that the influence of a
0.6 misalignment has negligible influence on measurements of constant cross-sectional
profiles. Nevertheless, for nonconstant cross-sectional profile workpieces, this effect is
more severe since the slope of the workpiece in z-direction directly affects its sensitivity.
The maximum error due to planes misalignment was quantified in 0.5 mm for distance
measurements on steepest details on the turbine blade shown in Fig. 9.
An offset of the planes along the z-axis results in datasets acquired at different
positions, while an angular misalignment around the x- or y-axis will cause the
datasets to be distorted.
A method to estimate the sensor z-positions and x/y-orientations is described in
the following, based on the use of a calibrated conical sample. We propose the use
of a high-slope radially symmetric artefact, i.e., a conical sample. The sample
artifact is made of aluminum and white coated to obtain an optically cooperative
surface (Fig. 12). The nominal half aperture angle is 74 , this being the maximum
steepness on typical workpieces to be measured. Sample conicity, perpendicularity
of the cone axis to the base, and flatness of the base were identified as relevant
parameters for the successful implementation of the method and calibrated using a
tactile CMM.
Previously, both the internal (e.g., focal length, principle point) and external
(position and orientation w.r.t. laser plane) camera parameters are calculated using
available methods (Hartley and Zisserman 2003). The sample is fixed at four known
positions, where the laser plane light hits the cone surface (Fig. 13). All resulting
partial conics (i.e., partial ellipses) in the 2D laser plane space are then augmented to
3D and altogether rigidly transformed (z-position, x/y-orientation) to minimize their
squared distance to the cone surface at all positions. The resulting transformation
parameters describe the laser plane z-position and x/y-orientation relative to the cone
positions. By using this information, mutual misalignments can be minimized in two
ways: (i) mechanical adjustment of sensors alignment, with lower bound given by
the resolution of the adjustment procedure, and (ii) software adjustment of the
recorded points in a common coordinate system (Weckenmann et al. 2009), resulting
on more complex data generated by the CMS (i.e., 3D datasets) instead of simple 2D
Fig. 12 Left: sketch of the system with cone artifact positioned inside the measurement area
showing the light of a single sensor. Right: photo of the system with a zoom on the artifact with
enabled lasers
Fig. 13 Intersections (red curves) between a single laser plane (1 tilt around x-axis) and the cone
at four positions. Left: top view. Right: side view, z-axis magnified for better clarity
profiles. While the former procedure is preferable for constant cross-section parts,
the 3D nature of the latter is well suited for freeform parts and, moreover, faster in the
setup phase.
To demonstrate the accuracy improvement by the proposed procedure, several
experiments were performed on parts featuring different geometry. In case of limited
slope, the improvement is intrinsically small, while on steeper surfaces the advan-
tage is evident: e.g., for the steepest details on turbine blades, typical deviations
between measurements of two misaligned adjacent sensors were reduced using
software correction from 0.5 mm to 0.1 mm (Fig. 14) with 5-times improvement
by the presented method.
System Characterization
To test the general performance of the CMS under industrial conditions, hot geometries
were acquired: a series of forged blades at about 800 C (10) and billets at about
1000 C (8) at the forging plant of Pietro Rosa TBM s.r.l. (Pietro Rosa TBM srl 2018).
It was validated that the system is well insulated. In fact, the average air
temperature inside the housing remained between 19 C and 21 C when measuring
the hot parts for about 30 min. Due to the optical bandpass filters, the geometry
acquisition was not disturbed by the infrared light emitted by the glowing steel
(Zhang et al. 2014).
During repeated measurements of turbine blades at approx. 800 C, the temper-
atures of the whole system were monitored (Fig. 15). The temperature of the
protection shields increases over time because of the workpiece radiation. After
about 1800 s (30 min) of repeated measurements, the shields reached a steady state at
a maximum temperature of about 31 C. Due to the conditioning provided by the air
chillers, the temperature of the metrology frame remained at 20 0.3 and the
Fig. 14 Deviation map of data coming from two different sensors on a turbine blade foot. Left: without
plane correction. Right: with correction. Top: location of the measured area on the turbine blade foot
Fig. 15 System temperatures as measured by internal sensors during the first minutes of loading
and measuring of workpieces at elevated temperature
housing temperature was in the range of 20 1.1 C (Fig. 15). The water cooling
system, designed to condition the temperature of the frame, has never been activated
during the test phase. Thus, it could be stated that the system is able to handle even
hotter workpieces.
In order to test the metrological performance of the CMS, bidirectional length

measurements were performed on a series of calibrated parts.
Experiments were performed using commercially available 130 70 4 mm.
glass-ceramic plates (Schott Nextrema ® (Schott 2018)) with rounded surfaces in
the smallest dimension, having a CTE of (0.63 0.2) 10 6 K 1 in the range of
20 C to 300 C, as stated by the manufacturer. Calibration of the two main
dimensions was performed using a scanning tactile CMM (MPE of length mea-
surement = 2,7 + L(mm)/300 μm) with measurand definition, measuring strategy,
i.e., area of interest and point density, and data processing similar to those active
while measuring on the optical CMS, obtaining a calibration uncertainty lower
than 5 μm.
To reach environmental conditions as close as possible to those active while
measuring actual workpieces, a series of billets at elevated temperature with mass
similar to a typical turbine blade were individually loaded onto the CMS and moved
back and forth as in normal operation for 1 h. During this time, the average
temperature of the billet was starting at approx. 850 C and decreasing to approx.
400 C before loading the next one. Then, eight calibrated glass-ceramic plates were
loaded on a dedicated frame positioned above a billet at about 850 C (Fig. 16). The
plates were fixed in two different orientations, in order to measure the two distinct
calibrated lengths of 70 mm and 130 mm. Individual part alignment was on purpose
slightly different, resulting in an angle of few degrees between moving direction and
corresponding workpiece axis. The glass-ceramic plates and the billet at elevated
temperature were measured continuously for 40 min, moving backwards and for-
wards, resulting in 50 measurements with billet temperature decreasing from approx.
800 C to 300 C.
Testing results of length measurements at elevated temperature are given in
Fig. 17, showing deviations from the reference values well within the target MPE
of the prototype CMS (0.2 mm).
Fig. 16 Calibrated glass-ceramic plates with low CTE and alternating orientation on a dedicated
frame (top) and billet at elevated temperature (bottom) during testing. Glass-ceramic plate num-
bering as overlay
Fig. 17 Results of testing in hot conditions. Even and odd plate numbers refer to glass-ceramic
plates oriented to measure the 70 mm length and the 130 mm length, respectively. Error bars
represent the standard deviation of 50 repetitions
Acknowledgments The authors acknowledge the financial support in the HOTGAUGE project
(E!6692), funded in the framework of the Eurostars Programme, and the support and advice from
colleagues at authors’ institutions, as well as industrial partners Zumbach Electronic AG (Switzer-
land) and Pietro Rosa TBM s.r.l. (Italy).
References
Bernstein J (2011) Optisches Multi-Sensor-Messverfahren zur dimensionellen In-line-Messung von
Strangprofilen im Fertigungsprozess, volume Bd. 23 of Berichte aus dem Lehrstuhl Qualitäts-
management und Fertigungsmeßtechnik. Friedrich-Alexander-Universität Erlangen-Nürnberg,
Shaker, Aachen. ISBN 978-3844003451
Ciddor PE (1996) Refractive index of air: new equations for the visible and near infrared. Appl Opt
35(9):1566–1573
Clarke TA, Grattan KTV, Lindsey NE, Grover CP (1991) Laser-based triangulation techniques in
optical inspection of industrial structures. In: San Diego – DL Tentative, SPIE proceedings,
SPIE, pp 474–486
Estler WT, Edmundson KL, Peggs GN, Parker DH (2002) Large-scale metrology – an update. CIRP
Ann Manuf Technol 51(2):587–609
Fisher RB, Naidu DK (1996) A comparison of algorithms for subpixel peak detection. In: Sanz JLC
(ed) Image technology. Springer, Berlin/Heidelberg, pp 385–404. ISBN 978-3-642-63528-1
Ghiotti A, Schöch A, Salvadori A, Carmignato S, Savio E (2015) Enhancing the accuracy of
high-speed laser triangulation measurement of freeform parts at elevated temperature. CIRP Ann
Manuf Technol 64/1:499–502. https://doi.org/10.1016/j.cirp.2015.04.012
Hartley R, Zisserman A (2003) Multiple view geometry in computer vision. Cambridge University
Press, Cambridge/New York, p 2. ISBN 0521540518
Hexagon Manufacturing Intelligence (2018) http://www.hexagonmi.com
Hocken RJ, Pereira PH (2012) Coordinate measuring machines and systems, volume 76 of
Manufacturing engineering and materials processing, 2nd edn. CRC Press, Boca Raton. ISBN
978-1574446524
Isheil A, Gonnet J-P, Joannic D, Fontaine J-F (2011) Systematic error correction of a 3D laser
scanning measurement device. Opt Lasers Eng 49(1):16–24
Jiang X, Scott PJ, Whitehouse DJ, Blunt L (2007) Paradigm shifts in surface metrology. Part II. The
current shift. Proc R Soc A 463(2085):2071–2099
Keferstein CP (2015) Fertigungsmesstechnik: praxisorientierte Grundlagen, moderne
Messverfahren. Vieweg + Teubner, Wiesbaden, 8. auflage edition. ISBN 978-3-8348-2582-7
Kunzmann H, Pfeifer T, Schmitt R, Schwenke H, Weckenmann A (2005) Productive metrology –
adding value to manufacture. CIRP Ann Manuf Technol 54(2):155–168
Nikon metrology (2018) http://www.nikonmetrology.com
Nishiguchi T, Koizumi Y, Maeda Y, Masuda M, Nagayama K, Okamura K (1991) Improvement of
productivity in aspherical precision machining with insitu metrology. CIRP Ann Manuf Technol
40(1):367–370
Optris GmbH (2018) http://www.optris.de
Pfeifer T (2002) Production metrology. Oldenbourg Wissenschaftsverlag, Berlin/Boston. ISBN
978-3-486-81042-4
Pietro rosa TBM srl (2018) http://www.pietrorosatbm.it/
Savio E (2012) A methodology for the quantification of value-adding by manufacturing metrology.
Savio E, de Chiffre L, Schmitt R (2007) Metrology of freeform shaped parts. CIRP Ann Manuf
Technol 56(2):810–835
Savio E, de Chiffre L, Carmignato S, Meinertz J (2016) Economic benefits of metrology in
manufacturing. CIRP Ann Manuf Technol 65(1):495–498
Schöch A (2016) Quality control of freeform parts at elevated temperature. PhD thesis, University
of Padua. http://paduaresearch.cab.unipd.it/9192/
Schöch A, Salvadori A, Germann I, Balemi S, Bach C, Ghiotti A, Carmignato S, Maurizio AL,
Savio E (2014) Fast measurement of freeform parts at elevated temperature using laser-
triangulation principle. In: IMEKO LMPMI
Schöch A, Salvadori A, Germann I, Balemi S, Bach C, Ghiotti A, Carmignato S, Maurizio AL,
Savio E (2015) High-speed measurement of complex shaped parts at elevated temperature by
laser triangulation. Int J Autom Technol 9(9):558–566
Schott AG (2018) www.schott.com
Schwenke H, Neuschaefer-Rube U, Pfeifer T, Kunzmann H (2002) Optical methods for dimen-
sional metrology in production engineering. CIRP Ann Manuf Technol 51(2):685–699
Weckenmann A (2012) Koordinatenmesstechnik: flexible Strategien für funktions- und fertigungs-
gerechtes Prüfen. Hanser, München, 2., vollständig überarb. aufl edition. ISBN 978-3-446-
40739-8
Weckenmann A, Estler T, Peggs G, McMurtry D (2004) Probing Systems in Dimensional Metrol-
ogy. CIRP Ann Manuf Technol 53(2):657–684
Weckenmann A, Jiang X, Sommer K-D, Neuschaefer-Rube U, Seewig J, Shaw L, Estler T (2009)
Multisensor data fusion in dimensional metrology. CIRP Ann Manuf Technol 58(2):701–721
Xi F, Liu Y, Feng H-Y (2001) Error compensation for three-dimensional line laser scanning data. Int
J Adv Manuf Technol 18(3):211–216
Zhang Y, Han J, Fu X, Zhang F (2014) Measurement and control technology of the size for large hot
forgings. Measurement 49:52–59
Micro-dimensional Measurement by a
Micro-probing System 11
So Ito
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 330
Miniaturization of Styli . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 331
Sensor for Micro-probing System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 333
Performance of Shear-Mode Detection Micro-probe . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337
Precision Measurement System of the Micro-slit Gap Width . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 343
Online Qualification of Tip Ball Diameter . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 349
Uncertain Due to Alignment Errors of the Measurement System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352
Experimental Results of Online Qualification and Gap Width Measurements . . . . . . . . . . . . . . . . 358
Summary and Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365
Abstract
Micro-probing system has become a remarkable technique for the dimensional
measurement of complex micrometric features on the micro-parts and precision
tools. In especial, the tactile micro-probes are one of the most effective micro-
probing systems since the miniaturized micro-stylus and high-sensitive probing
sensor of the micro-probing systems allows both the capability of three-
dimensional accessibility and nanometric resolution for the complex micrometric
features. Therefore, there have been many efforts for the miniaturization of
the stylus size with high-accuracy tip shape. In addition, since high sensitivity
of the tactile micro-probing system is influenced by the external interaction force
which has been ignored in the previous probing systems, new principle of
the probing sensor and novel calibration method of the micro-probing system
have been required.
On the other hand, the probe tip ball of the micro-probing systems is composed
with the high-accuracy microsphere, so that their diameter is smaller than the
S. Ito (*)
Department of Intelligent Robotics, Toyama Prefectural University, Imizu, Japan
e-mail: so.ito@pu-toyama.ac.jp

https://doi.org/10.1007/978-981-10-4938-5_11
330 S. Ito
micrometric features of the measuring workpiece generally in order to realize

good accessibility for complex features. Therefore, precise qualification of the
probe tip dimension is also important issues for the calibration of the micro-
probing system because the uncertainty of the dimensional measurement is also
affected by the geometrical tolerance of the probe tip shape. The uncertainty of
the micro-dimensional measurement is dominated not only by the reliability of
the micro-probe and the probe positioning instruments as well as the conventional
probing system but also by the nanometer-scale deformation of the surface and
the nanometric dimension of probe tip. In this chapter, a high-sensitive micro-
probing system utilizing the local interaction force has been described. With
respect to the compensation of the geometrical tolerance of the tip of the micro-
probing system, an online qualification method has been introduced. Finally,
dimensional measurement of micrometric feature by a micro-probing system is
described, and its measurement results are investigated according to the uncer-
tainty analysis.
Keywords
Micro-dimensional measurement · Micro-probing system · Micro-stylus · Micro-
CMM · Shear-mode detection · Gap width measurement · Effective diameter ·
Online qualification · Gap width uniformity
Introduction
Precision parts which consisted of the micrometric complex structures have been
widely utilized in various fields such as micro-optics (Gao et al. 2003; Chen et al.
2014), micro-gear (Ferreira et al. 2014), fuel injections (Alting et al. 2003; Hansen
et al. 2006), micro-molding (Heckele and Schomburg 2004), nanoimprint (Chou
et al. 1996), roll-to-roll patterning (Krebs et al. 2009), precision cutting tools (Noh
et al. 2009; Gao et al. 2013), and so on. Since the performance and reliability of those
precision parts depend on the shape and dimension of the micrometric features,
precision measurement technologies with nanometric uncertainty have also played
an important role. Optical measurement techniques represented by an optical micros-
copy (Durakbasa et al. 2011) and an interferometer (Kim et al. 2013) have been
widely applied for the dimensional measurement, surface profile measurement, and
three-dimensional inspection of the workpiece. The optical measurement methods
can achieve both of short measuring time and noncontact condition, so that those
optical measurement techniques are indispensable in the industrial fields (Gao et al.
2015). However, the accessibility of the light irradiated on the object to be measured
is confined by the optical properties or complex surface shape, so that it is difficult to
apply for the measurement of the side walls and bottoms of the deep trench structures
(Bos 2011).
Since the advances of precision machining technologies allow the miniaturization
of the complex structures, the number of the micro-parts consisting of the high-
aspect ratio, which is defined as the ratio of the depth with respect to the
11 Micro-dimensional Measurement by a Micro-probing System 331
aperture size, has been increased. Consequently, the demands of the dimensional
measurement for the high-aspect ratio micrometric features have been increased
in the recent years. With respect to the dimensional measurement of the macroscopic
workpieces, the coordinate measuring machines (CMMs) with the tactile probing
systems have been utilized (Claverley and Leach 2015). The commercially available
tactile probes involving the touch-trigger probes and the scanning probes are consisted
of elongate stylus shaft and small probe tip ball, so that it has satisfactory accessibility
to the complex shapes on the measuring objects. Therefore, the use of the miniaturized
tactile probes will also be effective solution for the complex micrometric features on
the precision parts in the dimensional measurement (Thalmann et al. 2016). The
measurement uncertainty of conventional macroscopic CMMs, which is affected by
the measurement range, is ranging from 1 μm to 1 mm. On the other hand, the
measurement ranges for the micro precision parts are approximately 1 μm to 1 mm
in usual, so that the measurement uncertainty has been required from 10 nm to 100 nm.
Here, the dimensional measurement instrument which is consisted of a tactile micro-
probe is called as a micro-probing system.
Various kinds of the micro-probing systems have been proposed to realize three-
dimensional measurements of complex micro-features with nanometer-scale
measurement uncertainty (Weckenmann et al. 2006). The development of the
micro-probing system has been reported from the various viewpoints, so that the
major development challenges can be divided to three points. The first challenge is
the miniaturization of probing system to improve the accessibility of the probe tip for
the measuring points on complex geometrical features. The second challenge is high
sensitivity of the probe sensor to realize low measuring force and nanometer-scale
resolution simultaneously. The third challenge is high-precision calibration of
the micro-probe, especially the qualification of the tip ball diameter. In this chapter,
for the miniaturization of the probe, fabrication of the micro-stylus using the micro
glass capillary tube is described, firstly. Secondly, the principle of nanometric
resolution and high-accuracy probe contact sensing based on local interaction
force is described, and the configuration of the highly sensitive micro-probe is
explained. Thirdly, the qualification of the tip ball diameter of the micro-probing
system is described as a calibration of the developed measuring system with
the micro-probe. Finally, as an application of the developed micro-probing system,
a dimensional measurement of the slit gap width of the precision coating
tool is presented.
Miniaturization of Styli
The accessibility and measurable dimension of the probing system are confined
partly by the stylus tip size and the shape of the stylus. Since a stylus possessing
a spherical tip ball maintains the uniform shape with respect to all directions, it is
widely employed to three-dimensional measurement not only for the micrometer-
scale structures but also for the millimeter- or meter-scale workpieces. Therefore, the
fabrication of a miniaturized micro-stylus with a spherical ball tip is one of the
332 S. Ito
important challenges in the development of the micro-probing system. With regard

to the accessibility of the stylus tip for the surface of the complex geometrical micro-
features, the shape of the stylus is also one of the important factors. In order to access
the inner side wall or the bottom of the concave shape such as a slit or trench, the
shape of the micro-styli is required sufficient the effective length and high-aspect
ratio of the probe with respect to the measuring object. In this chapter, the aspect
ratio of the probe is defined as the ratio of the effective length of the probe with
respect to the effective diameter of the probe tip ball. There have been various reports
for the miniaturization of the stylus possessing a micrometer-scale spherical tip ball.
Micro-styli possessing a micro-spherical tip ball have been fabricated by the forming
from the single material or the assembling of the shaft and the tip ball. The micro-tip
ball forming from the single material has been performed by micromachining
processes such as a micro-electro discharge machining (EDM) (Sheu 2004). As for
the other method, a micro-ball fabrication using melting glass (Hidaka and
Schellekens 2006) has also been proposed since the spherical shape can be formed
spontaneously by the surface tension of the melted material. Although the micro-tip
ball forming from the single material can produce a micro-stylus with the seamless
between the stylus shaft and the tip ball, it is difficult to control the sphericity,
diameter, and surface roughness, precisely. The stylus fabrication by the assembling
of the stylus shaft and high-accuracy sphere is a typical method for the stylus used in
the probing system of the conventional CMMs. The miniaturized styli for the micro-
probing system are also fabricated by the assembling of microsphere and elongate
stylus shaft. The tip balls of the micro-styli have been made of synthetic ruby (Goj
et al. 2014), glass, and so on. In some of the studies for the fabrication of the micro-
styli, a glass microsphere with accurate diameter has been employed as a tip ball
because of the ease of available (Murakami et al. 2010; Ito et al. 2014). Since the
glass spheres are easy to miniaturize, the glass microspheres are commercially
utilized as the ball lens and spacer for the liquid crystal displays (LCD). Therefore,
the precision glass microspheres of various diameters can be selected as the tip ball
of the micro-probe. The shaft of the typical styli used for the conventional probing
system for the CMMs has been made of stainless steel, tungsten carbide, carbon
fiber, and so on. From the view point of the ease of the fabrication, glass capillary
tubes have been applied for the micro-styli shafts in the study introduced in this
chapter. Figure 1 shows the microphotographs of the micro-stylus consisted of the
glass capillary tube shaft and glass microsphere. A hollow glass tube with 1.0 mm
outer diameter and 0.6 mm inner diameter was formed into an elongated and
sharpened capillary glass tube by a commercially available Pt filament-type glass
pipette puller. The sharpened edge of the capillary glass tube has an aperture with
micrometer-scale diameter. The sharpened glass capillary tubes were employed as
the material of the stylus shaft. The diameter of the sharpened edge and the tapered
angle of the glass capillary tube can be adjusted by the heating temperature and the
pulling force, so the glass capillary tube styli can be applied for the micro-tip ball
with various diameters. With regard to the tip ball materials, the glass microspheres
which are utilized for the spacer of the LCD are employed as the tip ball of the micro-
styli because of the high accuracy of diameter. The diameter of the glass
Fig. 1 Microphotographs of micro-styli consisting of a capillary glass tube and a glass microsphere
microspheres are ranging from 2 μm to 50 μm. The glass microspheres were attached
on the sharpened edge of the capillary glass tubes by thermosetting resin. The
assembling of the micro-stylus has been carried out by using the micromanipulator
under the optical microscope observation with two CCD cameras (Gao 2010)
because the accurate positioning of the microsphere is required. The procedure for
the assembling of the micro-styli is as follows. Firstly, the edge of the sharpened
glass capillary glass tube is slightly contacted with the droplet of the thermosetting
resin. When the edge of the capillary glass tube is contacted with the droplet, the
inner side of the glass tube can be filled with the thermosetting resin owing to
the capillary phenomenon. Secondly, a glass microsphere is attached on the edge of
the capillary glass tube under the optical microscope observation. When the edge of
the capillary glass tube filled with the thermosetting resin contacted with the
microsphere, the microsphere is stuck at the edge by the adhesiveness of the resin.
Once the tip ball is stuck at the edge of the sharpened capillary tube, the microsphere
is aligned spontaneously at the center of the stylus edge by the surface tension of the
thermosetting resin. After the attachment of the microsphere, the stylus shaft
is heated to cure the thermosetting resin.
Sensor for Micro-probing System
Many of the probing systems with the tactile probes have done the probing by
detection of the contact between the probe tip and the surface of the workpiece
(Weckenmann et al. 2004). Touch-trigger probe systems and scanning-probe sys-
tems, which are utilized in general CMMs, can detect the contact between the probe
tip and the surface of the measuring object based on the displacement or deflection of
the stylus shaft owing to the displacement of the probe tip ball. On the other hand,
since the stiffness of the miniaturized styli becomes lower than that of the styli used
334 S. Ito
in the conventional CMMs due to its smaller diameters of the stylus shaft, lower
measuring force and high-sensitivity for the contact sensing have been required.
In addition, nanometer-scale resolution detection of the tip ball displacement
is also necessary to realize the nanometer-scale measurement uncertainty. The
micro-styli are suspended by the beams (Nesterov and Brand 2005), leaf springs
(Li 2016), or flexure hinge (Liebrich and Knapp 2010), which have low spring
constant, to realize the low measuring force. The probe tip ball can be moved in three
axial directions by the suspending stylus, so that they can be applicable for three-
dimensional measurement.
With respect to the sensor for the detection of the stylus movement, the optical
type displacement sensors (Murakami et al. 2014; Schwenke et al. 2001) and angle
sensors (Li et al. 2016; Liebrich and Knapp 2010; Fan et al. 2010) have been widely
applied for the detection of the stylus displacement and inclination because they can
realize the high sensitivity and the low measuring force owing to non-contact
between the sensors and the stylus. The capacitive-type displacement sensors are
also applied for the sensors of the micro-probes owing to the displacement measure-
ment in noncontact condition (Küng et al. 2007; Thalmann et al. 2016; He et al.
2013). However, the adoptions of those displacement sensors or the angle sensors
make the probing systems larger, so the inertia caused by the weight of the probing
system itself induces measurement errors (Weckenmann et al. 2004). Therefore,
there have been various efforts for the miniaturization of the micro-probe units by
utilizing the compact sensors or integrated sensors. Some of the sensors for the
micro-probes are composed with the strain gauges and piezoelectric materials which
are installed in the suspended parts of the micro-styli (Lei et al. 2014; Dai et al.
2009). Those sensors have been fabricated by the semiconductor manufacturing
processes and MEMS fabrication techniques. In the sensing of the probe tip ball
contact by the detection of the displacement or inclination of the stylus, the measur-
ing force of the probing is generally determined by the spring constant of the stylus
and the suspended parts and the displacement amount of the probe tip ball. In order
to realize higher sensitivity and lower measuring force, the probing principles based
on the variation of the dynamic behavior of the micro-stylus have been proposed.
The micro-probing systems using the vibrating micro-stylus have been proposed
because the measuring force can be reduced by the intermittent contact between the
probe tip and the surface of the workpiece. Bauza et al. have proposed a lateral
vibration probe for the measurement of the thru hole with micrometer-scale diameter
(Bauza et al. 2005, 2011). The stylus is made of glass fiber, and it doesn’t have a tip
ball at the edge of the stylus shank. The stylus is vibrated in parallel with long axis of
the stylus by a tuning fork quartz crystal resonator. As well as a dynamic-mode
atomic force microscope, the attenuation of the probe oscillation amplitude due to
the intermittent contact has been utilized for the recognition of the contact condition
between the probe shank and the surface. Since the detection axis of the lateral
vibration probe is limited to the stylus vibration direction, it is difficult to be applied
for dimensional measurement of 3D micrometric structures. Leach and Claverley
have proposed a vibrating microscale CMM probe for three-dimensional measure-
ment of complex micro-features (Claverley and Leach 2010, 2013). Hidaka has
proposed small-sized ultrasonic probe, which can detect the contact in three axial
directions by the uniaxial stylus vibration (Hidaka et al. 2008, 2010).
With respect to the methods of the detection of the contact between the probe tip
and the surface, various principles for the probing detection have been proposed
based on the local phenomenon around the probe tip. Masuzawa et al. have devel-
oped a vibroscanning method for the micro-probing system (Masuzawa et al. 1993;
Kim et al. 1999). Since the styli of the vibroscanning method is made of tungsten
carbide (WC) alloy, the styli are conductive. Although the vibroscanning probe also
utilizes a lateral vibrating micro-stylus, the sensing of the stylus contact on the
surface is carried out by the electrical contact between the stylus and the workpiece.
Weckenmann et al. also propose the 3D micro-CMM using the conductive styli,
which can detect a tunneling current (Weckenmann et al. 2008; Hoffmann et al.
2008). Detection of the tunneling current is utilized for the nanometer-scale surface
imaging by the scanning tunneling microscope (STM) because the tunneling current
sensor can detect the tip-to-surface distance with nanometric resolution in non-
contact condition. However, the detection of the local electrical interactions, such
as the electrical conductive or the tunneling current, cannot be applied to the
dimensional measurement of the structures consisted of the insulating materials.
Since the principle of the probing detection is applicable regardless of the conduc-
tivity of the material, the probing based on the detection of the local interaction force
due to the water layer, which exists on the surface of the workpiece in atmosphere,
has been proposed. Goj et al. have introduced triaxial probing systems by using the
vertical oscillating probe (Goj et al. 2015). Since the condition of the vibrating stylus
is affected by the capillary force due to thin water layer on the surface of the
workpiece, nanometric distance between the probe tip and workpiece surface can
be detected with high sensitivity. The water layer on the surface is derived from the
humidity of the atmohsphere, and its existence has been recognized by AFM (Santos
et al. 2011; Zitzler et al. 2002). Thickness of the water layer is ranging from several
nanometers to several tens nanometers (Christenson 1994; Maeda et al. 2003; Butt and
Kappl 2009), and it is varied by the humidity of the measurement environment and the
surface energy of the workpiece. Therefore, the thin water layer is present on the
surface of the workpiece without any special pretreatment. Goj et al. have developed
an electrostatic probe system and investigated the phase change of the vibration probe
by the interaction of the water layer (Goj et al. 2014). The water layer often induces the
adhesion of the probe tip ball on the surface or the probing errors, so that it will
become one of the error sources of the uncertainty for the dimensional measurement
(Claverley and Leach 2013; Bos 2011; Murakami et al. 2016). Ito et al. have proposed
a micro-probing system based on the detection of the local interaction of the water
layer (Ito et al. 2014). The proposed micro-probing system can detect the approach
between the probe tip and the surface of the workpiece with nanometer-scale resolu-
tion by detecting the interaction force by the water layer on the surface. The micro-
probing system employs a micro-stylus vibrated in parallel with the measuring surface
to detect the interaction force owing to the thin water layer, so it is named as a shear-
mode detection micro-probing system. Figure 2 shows a schematic of the principle of
the shear-mode detection micro-probe. The micro-stylus is vibrated in Y-direction on
336 S. Ito
a Micro-stylus b
Viscous resistance
Amplitude
Z Water layer
Surface of workpiece Δf
Frequency
X
Fig. 2 Principle of a shear-mode detection micro-probe. (a) Schematic of shear-mode detection.

(b) Variation of amplitude and frequency of the resonance frequency of micro-stylus vibration
its resonance frequency, and the probe tip ball is moved toward the surface of the
workpiece, as shown in Fig. 2a. When the vibrating micro-stylus is contacted with the
water layer on the surface from X-direction, the amplitude of the micro-stylus vibra-
tion is reduced by the viscous resistance of the water layer which exists on the surface
of the measuring object (Antognozzi et al. 2001). At that moment, the frequency of the
micro-stylus vibration is shifted due to the capillary force of the water layer (Eoma
et al. 2009), as shown in Fig. 2b. In the case of the probe approach from Z-direction,
similar behavior of the micro-stylus vibration has occurred because the tip of the stylus
vibrates horizontally with respect to the water layer on the surface of the workpiece.
Consequently, the shear-mode detection micro-probing system can detect the approach
from two axial directions by the uniaxial vibrating micro-stylus. As for a sensor for the
detection of the micro-stylus vibration, a tuning fork quartz crystal resonator
(TF-QCR) has been used as the sensor for the micro-stylus vibration. The TF-QCR
is widely used in the quartz clock since it is supplied at low cost by the mass
production. Figure 3 shows the photograph of the shear-mode detection micro-probe
which is made of the micro-stylus and TF-QCR. A micro-stylus, which is composed of
a capillary glass tube and a glass microsphere, is attached on one beam of the TF-QCR.
The length of the stylus shaft is set to be 1.5 mm. The TF-QCR with the resonance
frequency of 32.768 kHz, which is generally used for the quartz clock, has been
employed for the shear-mode detection micro-probing system. The micro-stylus is
vibrated in Y-direction by a PZT transducer together with the TF-QCR. The vibration
of the micro-stylus is detected by the piezoelectric effect of the quartz crystal. By
utilizing a TF-QCR as the vibration sensor, it is possible to make the micro-probe
compact in comparison with the use of the optical sensors. The current generated by
the piezoelectric effect of the TF-QCR is converted into the voltage signal by the
current-to-voltage converter (I/V converter), and the output signal of I/V converter is
used as a probe vibration signal. The use of the TF-QCR for the vibration sensor
enables high Q-factor of the stylus oscillation. The Q-factor of the TF-QCR is
approximately 10,000. Since the minimum detectable force by the oscillating probe
Fig. 3 Photograph of the

assembled shear-mode
detection micro-probe
is affected by the Q-factor, the micro-probe (Albrecht et al. 1991), high Q-factor can
realize low measuring force. Q-factor of the developed shear-mode detection probe
with TF-QCR is larger than 2000. Similarly, the dynamic-mode AFM employs a
vibrating cantilever to detect the intermittent contact force on the surface of the
specimen, and Q-factor of the AFM cantilever generally used is approximately less
than 1000 in the atmosphere. Therefore, the developed vibrating micro-probe can
realize high Q-factor and high sensitivity for the detection of the probing at the
measuring surface.
Performance of Shear-Mode Detection Micro-probe
In the shear-mode detection micro-probing system, the variation of the micro-stylus

vibration condition, such as the amplitude and frequency, is utilized as the trigger
signal for the probing. Two types of the experimental configurations have been
introduced as the detection methods of the variation of the micro-stylus vibration
condition. Figure 4a shows the configuration for the detection of the variation of the
amplitude of the micro-stylus vibration (Ito et al. 2014). The amplitude of the probe
vibration signal, which is detected and amplified by I/V converter, can be extracted
by using a lock-in amplifier (LIA). The detection of the contact point is carried out
based on the amplitude attenuation of the probe vibration signal, so it is named as
an amplitude modulation (AM) method in this study. On the other hand, the
configuration for the detection of the frequency change of the micro-stylus vibration
(Ito et al. 2016a) is shown in Fig. 4b. The micro-stylus is oscillated and maintained
on the resonance frequency by a self-excitation circuit. The frequency shift from the
resonance frequency is detected by using a commercially available phase-locked
loop (PLL) circuit that is designed for the frequency modulation (FM) AFM. In this
study, the detection of the contact point based on the detection of the frequency shift
of the prove vibration is named as a FM method.
338 S. Ito
a AM detection unit
Reference signal
FG
Vibration PZT transducer
direction I/V converter
LIA
Z TF-QCR
Amplitude signal
Micro-stylus
X
Y
FM detection unit
b
Self-excitation circuit
Vibration PZT transducer
direction I/V converter
Z Phase-locked loop circuit
TF-QCR
Frequency signal
Micro-stylus
X
Y
Fig. 4 Configuration of the shear-mode detection micro-probe. (a) Amplitude modulation (AM)
detection configuration (Ito et al. 2014). (b) Frequency modulation (FM) detection configuration
(Ito et al. 2016a)
In order to evaluate the characteristics of the shear-mode detection micro-

probe, a prototype micro-probing system, which is shown in Fig. 5, has been
constructed by utilizing the shear-mode detection micro-probe and the X-Y and
Z-directional linear motion positioning stages. A micro-probe using a micro-stylus
with a tip ball with 10 μm in diameter was used in this experiment. The moving
tables of the positioning stages can be driven by the piezoelectric actuators with
nanometer-scale resolution. The displacements of the positioning stages are mea-
sured by the capacitive-type displacement sensors with a resolution of 0.4 nm. The
outputs of the displacement sensors are also utilized for the closed-loop control of
the position of the moving table to compensate the drift and creep of the piezo-
electric actuators. Therefore, the positioning accuracy of the stage has been
achieved to be 1.0 nm. Three-axis coordinates of the contact point are obtained
based on the output of the displacement sensors, and three-axis micro-probe
position is recorded through the analog-to-digital (A/D) converter installed in
the personal computer (PC). The shear-mode detection micro-probe is mounted
PZT transducer
AM or FM detection unit
TF-QCR
Micro-stylus Probe vibration signal
PC
PZT driven Stage drivers D/A converter
X-Y-Z stages A/D converter
Z Positioning feedback signal
Output signal of stage displacement

X
Y
Fig. 5 Experimental setup of the prototype micro-probing system for the evaluation of the
characteristics of the shear-mode detection
on the moving table of X-Y positioning stages. The measurement axis of X-axial
displacement sensor is aligned coaxially with the tip ball of the micro-stylus in
order to reduce the measurement errors due to the Abbe offset. With respect to the
workpiece for the probing detection, the reference surface of the gauge block made
of a stainless steel has been employed as a test workpiece. A workpiece is put on
the Z-directional positioning stage. The Z-directional positioning stage is installed
independently of X-Y positioning stages. The tip of the micro-stylus is also aligned
coaxially with the measurement axis of Z-axial displacement sensor to decrease
the Abbe offset.
Figure 6 shows the behavior of the probe vibration signal when the tip of the
micro-probe is approached toward the surface of the test workpiece. The test
workpiece was moved along Z-direction in 5 nm step. Blue lines in Fig. 6 indicate
the displacement of the stage table in Z-direction. The amplitude attenuation of
the micro-probe vibration is shown in Fig. 6a. The amplitude of the probe vibration
signal is normalized by the amplitude before the approaching. When the tip of the
micro-probe was located in the vicinity of the surface of the test workpiece, the
amplitude of the probe vibration signal was decreased with the approaching of
the test workpiece. As shown in Fig. 6a, the amplitude attenuation of the probe
vibration signal showed the behavior corresponding to the stepwise nanometer-scale
displacement of the positioning stage. Detection of the nanometer-scale displace-
ment between the probe tip and the measuring surface based on the amplitude
detection of the probe oscillation has been widely utilized in the tapping-mode
AFMs. The AM method of the shear-mode detection micro-probing system can
realize the nanometer-scale resolution of the displacement detection. On the other
hand, Fig. 6b shows the relationship between Z-directional displacement of the test
workpiece and the frequency shift of the probe vibration. As shown in Fig. 6b, the
340 S. Ito
a Amplitude attenuation ratio (a.u.) 1.1
Stage displacement (1 nm/div.)

Amplitude attenuation ratio
Stage displacement
1
0.9
0.8
0.7
Time (1 s/div.)
b
Stage displacement (1 nm/div.)

1.6
Frequency shift amount (Hz)
Frequency shift amount

1.4
Stage displacement
1.2
1
0.8
0.6
0.4
0.2
0
-0.2
Time (1 s/div.)
Fig. 6 Behavior of the probe vibration signal during the probe approaching. (a) Amplitude
attenuation. (b) Frequency shift amount
frequency shift of the probe vibration was increased with the approaching of the test
workpiece, and the stepwise frequency shift could also be obtained. As a result, it
was confirmed that the shear-mode detection micro-probe could detect nanometer-
scale displacement between the micro-stylus tip and the measuring surface by
detecting the amplitude attenuation or the frequency shift of the vibrating micro-
stylus. In comparison with the AM method of the shear-mode detection probe, the
shear-mode detection micro-probe with the FM method was able to more clearly
recognize the stepwise change with respect to the stage displacement. Therefore,
the shear-mode detection micro-probe can realize the nanometer-scale resolution by
using the frequency shift of the micro-probe vibration as the trigger signal for the
probing.
According to the experimental results shown in Fig. 6, the shear-mode detection

micro-probing system with FM method can detect the nanometer-scale displacement
more clearly. Probing capability for three axes is necessary to achieve a three-
dimensional measurement of the micrometric complex features. The sensitivity
and linearity in three axes of the developed micro-probe were investigated by the
prototype micro-probing system. Figure 7 shows the frequency shifts of the shear-
mode detection micro-probe with FM method as the function of the displacement of
the test workpiece on each direction. The reference surface of the gauge block was
approached toward the stylus tip of the micro-probe from X-, Y-, and Z-directions,
respectively. The stage moving speed was set to be 170 nm/s, and that of the
approaching and the withdrawing were same. The blue lines and red lines in Fig. 7
indicate the behaviors of the frequency shift during the approaching and with-
drawing, respectively. The vertical axes indicate the shift amount from the initial
resonance frequency of the micro-probe vibration. The zero points on the vertical
axes are defined by the initial frequency at the position where the probe tips are
sufficiently separated from the surface. The horizontal axes indicate the distance
between the tip of the micro-probe and the surface of the workpiece. The zero points
on the horizontal axes indicate the points at which the frequency shift begins to
change, so that it is assumed that the probe tip has contacted with the water layer on
the surface. Figure 7a shows the frequency shift of the micro-probe vibration when
the workpiece is approached and withdrawn from the Z-direction. The sensitivity of
the micro-probe with respect to the displacement is calculated by the linearity
approximating the frequency shift amount during the approaching using the least
squares method. The sensitivity of the shear-mode detection micro-probe in
Z-direction was estimated to be 0.173 Hz/nm. Similarly, Fig. 7b shows the frequency
shift when the workpiece is approached and withdrawn from the X-direction. The
sensitivity in X-direction was calculated to be 0.371 Hz/nm. As shown in Fig. 7b,
there can be seen the disagreement of the frequency shift between the approaching
and the withdrawing. This disagreement might be caused by the adhesion due to the
water layer and the deflection of the stylus shaft. The adhesion due the water layer is
well-known in the AFM observation. The attractive force of the adhesion interfered
with the frequency shift during the withdrawing. In addition, the elongated stylus
shaft is susceptible to the bending load in the case of the approaching of the surface
from X-direction. On the other hand, since the stiffness of the bucking load of the
stylus shaft is larger than that of the bending force, the influence of deflection in
Z-direction becomes small in comparison with X-direction. Consequently, the sen-
sitivity in X-direction is higher than that of Z-direction; however, the nonlinearity of
the frequency shift becomes larger. As for the use of the touch-trigger probe, higher
sensitivity of the detection from one direction is important since the dimensional
measurement is carried out based on the discrete probing on the workpiece.
Figure 7c shows the frequency shift of the micro-probe vibration when the work-
piece is approached and withdrawn from Y-direction. The sensitivity in Y-direction
was calculated to be 0.02 Hz/nm. In Y-direction, the stylus tip is contacted with the
water layer intermittently because of the probe vibration direction. Therefore, the
interaction force applied to the stylus tip decreases, and it will cause the decrease of
342 S. Ito
a
2.0

Z-direction
1.5
Approaching
1.0
Withdrawing
0.5
0.0
-0.5
-0.05 0 0.05 0.1 0.15
Probe displacement (nm)
b
4.0
X-direction
3.0
2.0
Approaching
1.0 Withdrawing
0.0
-1.0
-0.05 0 0.05 0.1 0.15
c
0.6
0.5 Y-direction
0.4
0.3
Approaching
0.2 Withdrawing
0.1
0.0
-0.1
-0.05 0 0.05 0.1 0.15
Fig. 7 Behaviors of the frequency shift during the approaching and withdrawing. (a) Z-direction,
(b) X-direction, (c) Y-direction
the sensitivity. As a result, the shear-mode detection micro-probe can detect the
approaching of the surface of the test workpiece from X-, Y-, and Z-directions. The
sensitivity of the developed micro-probe is different according to the approaching
directions, and high sensitivity of the micro-probe can be achieved in Z- and
X-directions because the micro-stylus is oscillated in parallel to X-Y plane and
Y-Z plane.
Since the repeatability contributes for the measurement uncertainty of the dimen-
sional measurement, the repeatability of the shear-mode detection micro-probe has
been evaluated. The repeatability of the probing was evaluated based on the fre-
quency shift during the approaching on each direction. The probing was carried out
5 times at the same position on the test workpiece. According to the experimental
results shown in Fig. 7, the threshold of the frequency shift amount which is
employed as the trigger signal for the probing was set to be 0.3 Hz. Figure 8
shows the experimental results of the repeatability evaluation of the probing.
The probing estimations were carried out on five measuring points on the surface
of the test workpiece. The workpiece was approached toward the tip of the micro-
probe until the frequency shift amount reached the threshold of 0.3 Hz. The moving
stage of the workpiece was set to be 150 nm/s. The stage position at the time when
the frequency shift amount reached the threshold was measured by the capacitive-
type displacement sensor and utilized as the probing coordinates. In order to prevent
the influence of the adsorption by the water layer, the workpiece on the positioning
stage was retracted 10 μm after the detection of the probing points. Figure 8 shows
the experimental result of the repeatability evaluation in Z-, X-, and Y-directions,
respectively. The plotted points indicate the error from the averaged values of 5 times
probing at each measuring position. The error bars show the dispersion of the
probing coordinate. The standard deviations of the probing in Z-, X-, and
Y-directions were estimated to be 13 nm, 16 nm, and 28 nm, respectively. The
dispersion of the probing in Y-direction is approximately twice of the Z- and
X-directions because of the low sensitivity of the micro-probe in Y-direction.
However, the shear-mode detection micro-probing system achieved nanometer-
scale repeatability in three axial directions. Consequently, the shear-mode detection
micro-probe with the FM method can realize high sensitivity with respect to the
displacement between the probe tip and the surface of the workpiece, and the micro-
probe can achieve the nanometer-scale resolution and repeatability.
Precision Measurement System of the Micro-slit Gap Width
The advantage of the tactile micro-probing system can be exhibited in the three-
dimensional measurement of the micrometric structures constituted by concave
shapes and small aperture. The proposed micro-stylus, which is made of the glass
capillary tube shaft and a microsphere, allows the micrometer-scale probe tip
size and the millimetric effective length of the micro-probe, so the shear-mode
detection micro-probing system is applicable for the dimensional measurement of
the inside of the small apertures such as micro-holes and micro-slits. In this section,
344 S. Ito
Error from averaged value (µm)

0.1
Z-direction
0.05
-0.05
-0.1
1 2 3 4 5
Measuring points
b
0.1
X-direction
0.05
-0.05
-0.1
1 2 3 4 5
Measuring points
c
0.1
Y-direction
0.05
-0.05
-0.1
1 2 3 4 5
Measuring points
Fig. 8 Repeatability evaluation of shear-mode detection micro-probe. (a) Z-direction,

(b) X-direction, (c) Y-direction
gap width measurement of the micro-slit of the precision tools is introduced as

an example of the dimensional measurement using the shear-mode detection micro-
probing system.
A micro-slit of a slot die coater has been employed as a measure workpiece. The
slot die coaters are widely used for the coating of the functional liquid materials
on the substrate or the films because they can coat the liquid materials in large area
continuously. In the recent years, the precision coating with the slot die coater has
been utilized for the manufacturing of the liquid crystal panels, highly functional
films, and lithium-ion secondary batteries. The slot die coater is consisted of two flat
metal plates, which are fabricated by the ultraprecision grinding, so the slit with the
micrometric gap width can be formed by two flat plates at the edge of the tool. Since
the thickness of the coating layer of the liquid material is influenced by the slit gap
width of the slot die coater, a precise control of the slit gap width is necessary. In
the abovementioned manufacturing with the precision coatings, the slit gap width
is required to be adjusted with micrometer-order accuracy. Therefore, as for the
dimensional measurement of the slit gap width for the precision coating, the mea-
surement uncertainty of less than 100 nm has been required. Since the surface of the
slot die has the chamfered edges, it is necessary to measure the gap width slightly
inside from the surface. In addition, the nozzle part of the slot die is formed of the slit
with high-aspect ratio shape, so the uniformity of the gap width along the depth
direction is also one part of the measuring objects.
The gap width of the slit is measured by the mechanical feeler gauges in general.
However, the measured value is highly influenced by the skill of the operator.
Furthermore, there is a risk of damaging the shape of the slot edge because the
mechanical feeler gauge carries out the measurement in contact condition with the
measuring object. Therefore, the mechanical feeler gauges cannot be employed for
the precision gap width measurement of the slot die coaters.
With respect to the conventional measurement methods of the gap width of the
slot die, noncontact thin gap sensors have been developed for the measurement of the
slit gap width. The thin sensor head is inserted inside the slit gap; the gap width can
be measured based on the capacitive sensors (Manning 2011). The measurable
minimum thickness using the thin sensors is limited to be 150 μm due to the physical
thickness of the sensor head. The gap width of conventional slot die coater is ranging
from 150 μm to 600 μm (Mitsubishi Material 2017). However, the slot die with the
gap width of less than 100 μm are utilized for the precision coatings used in
manufacturing the high-performance devices; thus, the thin sensors, which can
measure the gap width based on the capacitive sensors, cannot be applied. CCD
camera-based system has also been developed as a noncontact measurement system
of the micro-slit gap width (Furukawa et al. 2003). Although CCD camera-based
system can realize high-speed measurement in comparison with the feller-type
measurement system, it cannot achieve the sub-micrometer uncertainty due to the
diffraction limit of the optical system.
On the other hand, the micro-probing systems are expected to be a promising
method in the dimensional measurement of the micrometric features because the
elongated micro-stylus with millimetric effective length will be able to make good
accessibility for high-aspect ratio forms. However, the measurement range of the
micro-probing system, which is determined by the motion range of the probe posi-
tioning stages in general, is limited from submillimeter to several tens millimeter.
Although the length of the slot die coater is designed according to the coating width, it
346 S. Ito
X-, Y-, Z-directional

fine stages
Shear-mode detection Measuring object
Reflector micro-probe (Slot die coater)
Laser
interferometer
Z-coarse stage
Y
Z
X- and Y-coarse stages
Fig. 9 Experimental configuration of the prototype gap width measurement system
is ranging from several centimeters to larger than 1000 mm. Therefore, it is difficult for
the conventional micro-probing systems to measure the gap width throughout the
micro-slit of the slot die coater due to the limitation of the operation range. In order
to measure the slit gap width throughout the slot die coater, a prototype precision slit
gap width measurement system has been developed by using the shear-mode detection
micro-probing system. Figure 9 shows an experimental configuration of the prototype
gap width measurement system. As a measuring sample, a slot die coater made of
stainless steel was used. The length of the slot die coater is 200 mm, and the nominal
gap width of the slit is set to be 85 μm. The positioning of the probing points along
Y-direction has been carried out by an air-bearing precision linear slide with a traveling
stroke of 300 mm. The displacement of the moving table can be measured by the linear
encoder with the resolution of 20 nm which is installed in the moving table of the linear
slide. The moving table can be positioned by the closed-loop control, so the positioning
repeatability can achieve to be 100 nm. The slot die coater was placed on the moving
table of the linear slide, and then, the micro-slit of the slot die coater was aligned
parallel to the moving axis of the linear slide. The coarse positioning of the micro-
probe in X- and Z-directions is conducted by the servomotor-driven stages. Fine
positioning of the micro-probe has done by the X- and Z-directional positioning stages
driven by the piezoelectric (PZT) actuators. The traveling range of the PZT fine stage is
100 μm. The resolution and positioning repeatability of the PZT stages are 0.4 nm and
2 nm, respectively. The probing inside the slit of the slot die coater is carried out by
using the X-directional PZT stage. Coordinate of the probing point in X-direction can
be measured by a laser interferometer with the resolution of 0.79 nm. Laser beam of the
interferometer is aligned coaxially with the tip ball of the micro-stylus to reduce the
measurement error due to Abbe offset. The coordinate of the probing point in
X-direction is calculated based on the micro-probe displacement detected by the
laser interferometer.
a Probing point
Y
Z
Calculated
(x0, y0, z0) surface
b Probing axis
Probing point
x1
X x2
Y
Z
Fig. 10 Gap width measurement by touch-trigger probe. (a) Conventional method. (b) Proposed
method for the micro-slit gap width measurement
The measurement of the gap width, which means the measurement of the distance
between two arbitrary planes, requires at least four coordinates of the probing point
in the case of typical CMMs. Figure 10a shows the typical measurement method of
the distance between two surfaces by CMM. On one surface, at least three positions
are probed first to define the normal direction. Then, one position on the other
surface is probed for obtaining the coordinate of the measuring point. The gap
width defined as the distance between two surfaces can be accurately obtained
from the coordinates of the probing positions on the second surface with respect to
the first surface. Therefore, the measurement errors due to the inclination of the
measured surfaces can be eliminated. However, since the typical measurement
method of the gap distance is a time-consuming procedure, the measurement errors
due to increasing of the measurement time will occur during the gap width inspection
throughout the slot die coater. Figure 10b shows the gap width measurement method
348 S. Ito
proposed in this study. The stylus of the micro-probing system is moved along
X-direction by the PZT fine positioning stage; the probing is conducted at two points
inside the slit of the slot die coater. Since the inside surfaces of the slot die coater are
consisted of the ultra-flat two planes, the influences of the surface form and undu-
lation will become small in comparison with the errors of the gap width. In addition,
the slit gap width of several micrometer orders can be confirmed by using the
mechanical thickness gauges or optical microscopes, so the slit inside has the
parallelism with micrometer-scale accuracy. Therefore, both inside of the slit can
be assumed as almost parallel.
As shown in Fig. 11, the gap width wsd of the measuring point can be calculated
by the following equation:
wsd ¼ L þ Dprobe (1)
Here, L is the distance of the probing positions in the slit gap direction, and Dprobe
is a diameter of the tip ball of the micro-probing system. However, the measurement
errors are caused by the kinetic motion errors of the positioning stages and the
installation errors of the measurement workpiece. Those errors in the measurement
results can be removed by the geometric compensation based on the error estimation
of the measurement system. The influence of motion error and installation error is
discussed based on the uncertainty analysis, in this chapter. The measurement errors
of the gap width are also caused by the probing error derived from the micro-probe.
Especially, the accuracy of tip ball diameter influences the reliability of the mea-
surement results. Although an inspection or qualification of the probe tip ball
Fig. 11 Relationship of the Pm1 Pm2

probed coordinates and the L
Displacement sensor
gap width of the micro-slit
Abbe offset
Dprobe
Water layer
De
wsd
diameter is valid for the reduction of the measurement error, the effective diameter
during the probing is different from the actual diameter due to the existence of the
water layer on the surfaces. Consequently, a calibration method of the effective
diameter of the probe tip ball, which is not affected by the variation of the measure-
ment environment such as thickness of the water layer, has been required to achieve
nanometric measurement uncertainty.
Online Qualification of Tip Ball Diameter
The shear-mode detection micro-probing system utilizes the interaction force owing
to the water layer on the surface for the detection of the probing point. Figure 12
shows the schematic diagrams of the relationship between the probe tip-to-surface
distance d and the frequency shift of the probe vibration Δf. Certain amount of the
frequency shift Δft is set to be as a threshold for trigger signal of the probing in
the shear-mode detection micro-probing system. As the probe tip is come close to
the surface of the workpiece, the shift amount of the probe vibration frequency is
gradually increased, and finally, Δf becomes larger than Δft. On the other hand, it is
difficult to identify the actual contact position on the surface because the frequency
shift varies continuously during the approaching. Consequently, the residual distance
st between the probe tip ball and the surface is existent when the frequency shift
amount is reached Δft at position Pt. Therefore, the distance from the tip ball center
to the probing point is not consisted with the actual diameter of the probe tip ball, so
that the measurement error of the dimensional measurement is caused by the residual
distance st. In addition, st is varied by the other measurement conditions such as the
sensitivity of the micro-probe, the uniformity of the tip ball diameter, the deflection
of the stylus shaft, and so on. In order to remove the measurement errors due to st,
a qualification of the effective diameter of the probe tip ball De has been proposed.
The qualification of the effective diameter of the probe tip ball is conducted by a set
Fig. 12 Relationship
Frequency shift Δf
between the probe tip-to-

Trigger point
surface distance and the
frequency shift of the probe
vibration Δ ft St
Tip ball-surface distance d
Tip ball Water layer Sample surface

350 S. Ito
Pq1 Pq2 Pm1 Pm2

Lq12 Lm12
St St
Water layer
wca wsd Slot die coater
Calibration artefact
De D
Probe tip ball
Fig. 13 Schematic of qualification of the tip ball diameter and gap width measurement of micro-slit
of gauge blocks. A grade K (calibration grade) gauge block with nominal thickness
of 100 μm (wca) is employed as a calibration artifact. wca is almost the same as the
nominal gap width of the slot die coater employed as the measuring object. A
tolerance of the gauge block used in the experiment is 18.0 nm to +2.0 nm.
Figure 13 shows a schematic of qualification of the tip ball diameter and the
measurement of slit gap width. A calibrated artifact is supported by two gauge
blocks. Three gauge blocks are wrung together for a good mechanical stability.
One end of the calibrated artifact is set to be higher than those of the supporting
gauge blocks. Therefore, the measuring surface of the calibration gauge block can be
probed directly. The calibration artifact assembled by three gauge blocks is mounted
on the moving table of the linear slide moved in Y-direction. Z-directional height of
the calibration artifact is adjusted by a height adjustment assembly composed with
gauge blocks to be the same height with the slit of the slot die coater. The surface of
Pq2 and that of Pm1 are aligned on the same X-direction position, so that the gap
width measurement procedure can be carried out after the qualification procedure
immediately without the micro-probe moving along X-direction. That alignment of
the calibration artifact can reduce the time for the probe positioning, so it will
contribute for the reduction of the measurement errors. The qualification of the
effective diameter of the probe tip ball can be conducted on the measurement
setup, so this method is called as online qualification (Ito et al. 2016b).
The online qualification procedure of the tip ball diameter is as follows. Both the
measuring surfaces of the calibrated gauge block are probed with the shear-mode
detection micro-probing system. The X-directional coordinates of the probing points
are obtained as Pq1 and Pq2, respectively. After the probing at Pq1, the tip ball of the
micro-probe is retracted from the surface of the calibration artifact. Then, the calibration
artifact is moved in +Y-direction by the linear slide, and the micro-probe is moved by
the coarse servomotor stage in -X-direction. Therefore, the Z-directional position of the
micro-probe is not changed during the qualification. After that, the Y-directional
position of the calibration artifact is adjusted by the linear slide so that the Pq2 is coaxial
with Pq1 along the X-direction. The approaching and probing on Pq2 is carried out by
using both coarse positioning stage and PZT stage which can be moved in X-direction.
The distance Lq12, which is the interval between Pq1 and Pq2 along the X-direction, is
detected by the laser interferometer whose reflector is mounted on the X-directional
PZT stage. Since the PZT stages with the reflector are mounted on the servomotor
stages, X-directional displacement owing to the servomotor stage is also detected by the
interferometer. Consequently, the effective diameter of the probe tip ball De can be
expressed by the following equation:
De ¼ Lq12 wca 2st (2)
After the qualification, the slot die coater is moved by the linear slide in
Y-direction. The tip of the micro-probe is positioned between the two flat plates
which form the micro-slit, and the inside of the micro-slit is probed as Pm1 and Pm2.
The condition for the probing involving the probe approaching speed and the trigger
threshold of the frequency shift are set to be the same during the qualification and
the measurement. The X-directional interval distance Lm12 between Pm1 and Pm2 is
estimated by the measured value of the interferometer. As a result, the gap width wsd
can be obtained by
wsd ¼ Lm12 þ De þ 2st ¼ Lm12 þ Lq12 wca Δs (3)
where Δs is the difference of st on the calibration artifact and the slot die coater.
As shown in Eq. (2), the difference Δs is included in the measurement result of
ws. If the difference of the residual distance between the calibration artifact and
the slot die coater is small enough, the measurement error due to Δs can be reduced
significantly. Since the qualification procedure and the gap width measurement are
carried out in the same measurement environment, the behaviors of Δf indicate good
agreement to each other. Figure 14 shows the results of Δf on the calibration artifact
Frequency shift amount Δf (Hz)
0.5
0.4
Slot die coater
0.3
0.2
0.1
0.0
X-directional probe displacement (10 m/div.)
Fig. 14 Measured result of Δf with respect to probe displacement on the calibration artifact and
slot die coater
352 S. Ito
and the slot die coater, respectively. It can be seen that those results were similar and
the difference of the X-directional micro-probe position at Δf = 100 mHz was
estimated to be less than 2 nm. Qualification procedure of the effective diameter of
the probe tip ball was carried out by using the same material as the measurement
workpiece under same measurement environment. As a result, it was confirmed that
the difference in the residual distance st between the calibration procedure and the
gap width measurement was small enough. According to Eq. (2), the online quali-
fication can eliminate the measurement error regardless of the elastic deformation of
the stylus shaft or the sensitivity of the micro-probe.
Uncertain Due to Alignment Errors of the Measurement System
Online qualification and gap width measurement are achieved by the probing
using the positioning stages for the micro-probe, the calibration artifact, and the
slot die coater. The measurement uncertainty of the gap width measurement is
influenced by the angular alignment errors of the experimental setup involving the
positioning stages, the calibration artifact, and the measuring surface of the slot die
coater. Figure 15 shows a schematic of the angular alignment errors of the experi-
mental setup. The moving axis of the air-bearing linear slide is set to be as
Y-direction which is the datum of the alignment of the setup. The X-direction and
Z-direction of the coordinate system are shown in Fig. 15. The longer direction of
the slot die coater and the measuring surface of the calibration artifact consisted of
Micro-stylus
Slot die coater
Ideal Z-axis Ideal X-axis
Y
X θY_X-coarse θY_X-fine
Z
θZ_X-coarse θZ_X-fine
θZ_sd
θY_Z-coarse
θX_Z-coarse Calibration artefact
θY_Z-fine θZ_ca
θX_Z-fine
Ideal Y-axis
Fig. 15 Schematic of angular alignment errors of the moving axes of the prototype gap width
measurement system
the gauge blocks are aligned in parallel with the Y-axis. The angular alignment errors
of the calibration artifact and the slot die coater around the Z-axis are indicated as
θZ_ca and θZ_sd, respectively. θZ_ca and θZ_sd were calculated based on the
X-directional displacements of the calibration artifact (ΔdX_ca) and the slot die coater
(ΔdX_sd) by the movement of the linear slide in Y-direction. θZ_ca and θZ_sd can be
calculated by the following equations:

Δd X _ca
θZ _ca ¼ tan (4)
Lca

Δd X _sd
θZ _sd ¼ tan (5)
Lsd
ΔdX is measured by an optical fiber displacement sensor with a resolution of

0.49 nm. Lca and Lsd are the Y-directional displacement by the linear slide. Lca
and Lsd are set to be 180 mm and 20 mm, respectively. The angular alignment
errors of the calibration artifact and the slot die coater around X- and Y-axes have
not been detected clearly by the displacement sensor. Therefore, the uncertain
sources due to angular alignment errors θX_ca, θY_ca, θX_sd, and θY_sd have been
ignored in this experiment. The micro-probe is mounted on the PZT-driven fine
linear motion stages which can move in X- and Z-directions. Those fine stages are
stuck on the DC servomotor-driven coarse positioning linear motion stages which
can move in X- and Z-directions. The alignment errors of the fine and coarse stages
around Y- and Z-axes also cause the error of the qualification and gap width of the
micro-slit. In this study, online qualification and gap width measurement have
been done using only X-directional stages. Although the Z-directional fine stage is
applied only for the searching of the slit position, the Z-directional coarse stage is
used for the adjustment of the probe position along Z-axis during the gap width
measurement. Therefore, with respect to the uncertainty of the qualification and
the gap width measurement, the alignment errors of the X-directional fine stage
around Y- and Z-axes are defined as θY_X-fine and θZ_X-fine, respectively. Similarly,
the alignment errors of the X-directional coarse stage around Y- and Z-axes are
defined as θY_X-coarse and θZ_X-coarse, respectively. The alignment errors of the
X-directional stages are evaluated by using the right angle mirror and the optical
fiber displacement sensor. The optical fiber displacement sensor is mounted on the
jig which is used for the holding of the micro-probe on the positioning stages
instead of the micro-probe. One mirror surface of the right angle mirror is aligned
in parallel with the Y-axis of the linear slide. The other mirror surface faced the Y-
or Z-direction on the moving table of the linear slide; the variation of the distance
from the optical fiber sensor to its surface is measured by the moving of the
X-directional stages. The angular alignment errors of the X-directional stage can
be calculated by the arctangent of the sensor output/X-directional stage displace-
ment. As well as the angular alignment errors of the X-direction stages, the
alignment errors of the Z-directional stage around the X- and Y-directions are
evaluated as θX_Z-coarse and θY_Z-coarse, respectively.
354 S. Ito
q Z_reflector
Moving axis of reflector
Reflector
X Measurement axis of
laser interferometer
Y Z q Y_reflector
Fig. 16 Inclination errors of the reflector of the laser interferometer
Since the X-directional position of the micro-probe, which is almost the same as
the displacement of the X-directional stages, is measured by the laser interferometer,
the alignment of the reflective mirror with respect to the laser beam also causes the
uncertainty. If the reflective mirror is not mounted perpendicular to the moving axis
of the X-directional stage, the measured value of the interferometer becomes smaller
than the actual value due to cosine error. As shown in Fig. 16, when the reflective
mirror is tilted by the inclination angle θ with respect to the measurement axis of the
laser interferometer, the relationship between the actual distance L and the measured
distance L’ can be expressed by the following equation:
L ¼ L0 cos θ (6)
Furthermore, there are two angular alignment errors of the reflective mirror
θY_reflector around the Y-axis and θZ_reflector around the Z-axis. Therefore, the mea-
sured distance L0 in consideration of the inclination of the two axes of the reflective
mirror is expressed as follows:
L ¼ L0 cos θY _reflector cos θZ _reflector (7)
The alignment of the reflective mirror is adjusted based on the X-directional

moving axis of the fine stage. The angular alignment errors of the reflective mirror
are evaluated by moving the reflective mirror in Y- and Z-directions with respect to
the laser beam of the interferometer.
Table 1 shows the summary of the angular alignment errors of the prototype gap
width measurement system. Based on the summary of the angular alignment errors,
the geometric measurement errors are estimated. Figure 17 shows a geometrical
model of the alignment error of the probing points in X-Y plane during the online
qualification of the probe tip ball diameter. As shown in Fig. 17, Lca is the distance
between the probing points on both surfaces of the calibration artifact. h is the
Table 1 Summary of the angular alignment errors of the proposed

θsd θY_X-fine θY_X-coarse θX_Z-coarse θy_Abbe
27.8 μrad 7.5 mrad 14.0 mrad 5.7 mrad 6.8 mrad
θca θZ_X-fine θZ_X-coarse θY_Z-coarse θZ_Abbe
70.0 μrad 2.8 mrad 3.6 mrad 14.1 mrad 0.34 mrad
θca
Pq1
Moving axis of
X-fine stage
h
θZ_X-fine
L’ X
θsd
X
Pq2
Y
Motion axis of the

X-Servo stage
Y
Fig. 17 Geometrical model of the measurement error of online qualification
distance of the moving axis of the fine stage on both surfaces of the calibration
artifact. Dcoarse is the X-directional displacement of the coarse stage. Therefore, the
nominal thickness of the calibration artifact wca can be expressed by the following
equation (Chen et al. 2016):

wca ¼ Lca L0ca cos θZ _X fine θca (8)
Here, L0 ca, which is indicated in Fig. 17, can be defined by the following equation:

L0ca ¼ h tan θZ _X fine θca (9)
Therefore, h can be shown as follows:

356 S. Ito

h ¼ Dcoarse θZ _X fine θZ _X fine (10)
Consequently, the measurement error Δwca of the calibration artifact in X-Y plane
can be calculated by the following equation:
Δwca ¼ ðLca Dcoarse Þ wca !

wca þ Dcoarse tan θY _X coarse θY _X fine tan θY _X fine θca
¼ wca
cos θZ _X fine θca
!
wca þ Dcoarse tan θZ _X coarse θZ _X fine tan θZ _X fine θca
þ wca
cos θZ _X fine θca
(11)
On the other hand, Fig. 18 shows the geometrical model of the measurement error
Δwsd of the gap width measurement in X-Y plane. Δwsd can be calculated by the
following equation:
wsd
Δwsd ¼ wsd (12)
cos θY _X fine θY _sd cos θZ _X fine θZ _sd
As described above, the angular alignment error of the slot die coater around the
Y-direction θY_sd is small enough to be ignored.
The measurement error of the gap width is also caused by an Abbe offset between
the measurement axis of the laser interferometer and the moving path of the probe tip
ball along the X-axis. When the center of the probe tip ball is not located on
the measurement axis of the laser interferometer, the Abbe error of the laser
interferometer is involved due to the inclination errors of the X-directional stage
movement. Figure 19 shows the schematic diagram of the Abbe error due to the
inclination error of the X-directional stages. In Fig. 19, the probe tip ball before and
after the movement in the X-direction is indicated in the same Z-directional position
for simplification. θY_Abbe is the inclination error of the X-directional stages due to
the stage movement. a1 is the distance between the reflective mirror and the probe tip
ball along X-axis. a2 is the Abbe offset between the measurement axis of the laser
Fig. 18 Geometrical model

of the measurement error of Y
θsd
gap width measurement
Slot die coater
θZ_X-fine-θZ_sd X
Pm1
Pm2
Motion axis of
X the X-PZT
stage
Y
θY_Abbe
Micro-stylus
Reflector
a1
Y
X a2
L’
Interferometer axis
Z L
Fig. 19 Schematic of the Abbe error due to the inclination error of the X-directional stages
interferometer and the tip ball of the micro-probe in Z-direction. The Abbe errors
LY_Abbe and LZ_Abbe around Y- and Z-axes can be expressed as follows:

θY _X fine
LY _Abbe ¼ a1 tan þ a2 tan θY _X fine
2

θY _X coarse
þ a1 tan þ a2 tan θY _X coarse (13)
2

θZ _X fine
LZ _Abbe ¼ a1 tan þ a2 tan θZ _X fine
2

θZ _X coarse
þ a1 tan þ a2 tan θZ _X coarse (14)
2
As a result, the combined measurement error LAbbe_error due to the Abbe offset
can be estimated by the following equation:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
LAbbe_error ¼ L2 Y _Abbe þ L2 Z _Abbe (15)
The motion error of the Y-directional linear slide on which the slot die coater is
mounted also causes the error of the gap width measurement. When the probing
points on the surface of the slot die coater are changed by the motion of the Y-linear
slide, the measurement error will occur by the tilting motion errors around Y-axis
(θY_slide) and X-axis (θX_slide), respectively. Therefore, the measurement error due to
the motion error of the Y-linear slide ΔwY-slide is expressed by the following
equation:
wsd
ΔwY slide ¼ wsd (16)
cos ðθY _slide Þ cos ðθX _slide Þ
358 S. Ito
Table 2 Summary of the tilting motion errors of the Y-directional linear slide and Z-directional
coarse positioning stage
Stage Axis Tilting error
Y-slide Roll 8.10 mrad
Pitch 30.30 mrad
Yaw 12.31 mrad
Z-coarse stage Roll 22.59 mrad
Pitch 73.40 mrad
Yaw 278.86 mrad
In the case of the measurement of the gap width uniformity along Z-direction,
the probing positions are moved in Z-axis by the Z-directional coarse positioning
stage. Therefore, the measured gap width is also influenced by the tilting motion
errors of the Z-directional servomotor stage around X-axis (θX_Z-motion) and Z-axis
(θZ_Z-motion), respectively. Therefore, the measurement error of the gap width due to
the tilting motion errors of the Z-directional coarse positioning stage Δwsd_Z-motion
can be estimated as follows:
wsd
Δwsd_Zmotion ¼ wsd (17)
cos ðθX _Zmotion Þ cos ðθZ _Zmotion Þ
Table 2 shows the summary of the tilting motion errors of the Y-directional
linear slide and Z-directional coarse positioning stage. The tilting errors
were measured by the laser autocollimator with the resolution of 1.27 μrad.
The measurement uncertainties for the online qualification and the gap width
measurement have been evaluated based on Eqs. (3, 4, 5, 6, 7, 8, 9, 10, 11, 12,
13, 14, and 15).
Experimental Results of Online Qualification and Gap Width

Measurements
By utilizing the developed micro-probing system, the gap width of the micro-slit
on the slot die coater was conducted. The nominal gap width and length of the
micro-slit of the slot die coater are 85 5 μm and 200 mm, respectively. Since
there are the chamfered edges of the flat plates forming the micro-slit, the probe tip
ball has to be positioned inside of the micro-slit before the gap width measure-
ment. In the developed micro-probing system, the initial position of the micro-
probe tip ball along X-direction can be determined by the probing to approach
from the Z-direction because the shear-mode detection micro-probe has high
sensitivity of the probing in Z-direction as shown in Fig. 7. Figure 20 shows a
method of discriminating the chamfered edge positions of the slot die coater by the
probe approach from the Z-direction. Trajectory of the micro-stylus is shown in
Fig. 20a. Figure 20b shows the experimental results of the probed points in X-Z
a
Y Tip ball trajectory
X
Z
Probe tip ball
Slot die coater
Slot die coater Slit gap Slot die coater
b
Position in Z direction [20 nm/div]
90 μm
0 20 40 60 80 100
Position in X-direction [μm]
Fig. 20 Discrimination of the chamfered edge positions of the slot die coater by the probe
approach from Z-direction. (a) Schematic of the micro-stylus trajectory. (b) Experimental result
of probed points in X-Z plane
plane. The frequency shift of the micro-probe vibration has not occurred at the gap
part shown in Fig. 20b, so that this part was considered as the inside of the micro-
slit of the slot die coater. Z-directional height of the chamfered edges of the slot die
coater was estimated to be approximately 90 μm, and the probing points along
Z-axis were arranged to become lower than the chamfered edge parts during the
gap width measurement.
Figure 21 shows the experimental results of the online qualification of the
effective diameter of the tip ball of the micro-probe. Firstly, the probing at Pq1 was
carried out 10 times by the X-directional fine positioning stage. Then, the micro-
probe was moved to the side of Pq2 by the X-directional coarse positioning stage,
360 S. Ito
a
Position Pq1 [μm] -87.60
-87.70
-87.80
1 2 3 4 5 6 7 8 9 10
b Number of the probing
Position Pq2 [ μm]
-243.20
-243.30
-243.40
-243.50
1 2 3 4 5 6 7 8 9 10
Number of the probing
c
155.80
155.75
Distance Lq12 [μm]
155.70
155.65
155.60
155.55
155.50
1 2 3 4 5 6 7 8 9 10
Fig. 21 Experimental results of online qualification of the effective diameter of the tip ball of
the shear-mode detection micro-probe. (a) Probing position of Pq1. (b) Probing position of Pq2.
(c) Distance Lq12
and the probing at Pq2 was conducted 10 times. The online qualification procedure
including the probe repositioning took approximately 50 s. Figure 21a and b shows
the X-directional positions of the trigger points, and the vertical axes are calculated
by the output of the laser interferometer. The mean values of Pq1 and Pq2 excluding
the outlier are 87.70 μm and 243.35 μm, respectively. The interval distances Lq12
are shown in Fig. 21c, and the mean value and standard deviation of Lq12 are
estimated to be 155.65 μm and 29.7 nm, respectively. Since the nominal length of
the gauge block used for the online qualification is 100 mm, the effective diameter of
the tip ball De is estimated to be 55.65 μm, consequently.
Position Pm1 [μm]

-10.20
-10.25
-10.30
1 2 3 4 5 6 7 8 9 10
b
Position Pm2 [μm]
15.90
15.80
15.70
15.60
1 2 3 4 5 6 7 8 9 10
c 26.15
Distance Lm12 [μm]
26.10
26.05
26.00
25.95
25.90
1 2 3 4 5 6 7 8 9 10
Fig. 22 Experimental results of gap width measurement of the slit gap of the slot die coater.
(a) Probing position of Pm1. (b) Probing position of Pm2. (c) Distance Lm12
After the online qualification procedure, the probing position was moved to
the inner side of the micro-slit of the slot die coater by the Y-directional linear
slide. Figure 22 shows the experimental results of the gap width measurement at one
point of the slot die coater. The vertical axes of Fig. 22a and b are the coordinate of
the probing position on X-axis, which have been measured by the laser interferom-
eter. The probing at Pm1 was carried out 10 times firstly, and then Pm2 was also
probed 10 times. With respect to the probing for the gap width measurement, the
micro-probe displacement in X-axis has been achieved by only the X-directional fine
positioning stage. Therefore, the kinetic motion errors and the angular alignment
errors due to the coarse stage are not necessary to be considered as the uncertainty
362 S. Ito
sources of the gap width measurement. The mean values of Pm1 and Pm2 are
10.26 μm and 15.75 μm, respectively. The vertical axis of Fig. 21c is the interval
distance Lm12 between Pm1 and Pm2 along X-direction. The mean value and the
standard deviation of Lm12 are 26.00 μm and 42 nm, respectively. The gap width
measurement took approximately 40 s which included the Y-directional positioning
of the slot die coater by the linear slide. According to Eq. (2) and the qualification
result of the effective diameter of the probe tip ball, the gap width of the micro-slit on
the slot die coater was calculated to be 81.66 μm.
The reliabilities of the qualification and the gap width measurement have been
evaluated based on the uncertainty analyses according to GUM (ISO Guide to the
Expression of Uncertainty in Measurement) (JCGM 2008). The uncertainties of
the online qualification and the gap width measurement have been carefully
considered involving the kinetic errors and the alignment errors of the micro-
probing system which are mentioned in the previous section. The measurement
uncertainty is affected by the laser interferometer and the gauge blocks used as the
calibration artifact, so the summary of the uncertainty budget of the gap width
measurement uw, which includes the uncertainty of the online qualification uDe, is
shown in Table 3.
Table 3 Uncertainty budget of gap width measurement (nm)

Standard
Uncertainty sources Symbol Value Type uncertainty
Uncertainty in Lq12 uLq – – 28.0
Cosine error by the angular alignment and ucos_X-stages 11 B 6.4
the probing axis
Cosine error by the alignment of the ucos_interferometer 43.08 B 24.9
interferometer axis
Abbe error of the X-fine stage uAbbe_X-fine 1.9 B 1.1
Abbe error of the X-coarse stage uAbbe_X-coarse 5.7 B 3.3
Resolution of the interferometer ulaser_resolution 0.79 B 0.5
Linearity error of the interferometer ulaser_linearity 5.0 B 2.9
Thermal effect of the interferometer uLaser_thermal 5.0 B 2.9
Repeatability of Lq12 urep_Lq12 29.7 A 9.9
Uncertainty in wca uwca – – 11.5
Length tolerance of calibrated artifact utol_ca 20.0 B 11.5
Uncertainty in Δs uΔs 2.1
Water layer on the measuring surface of the uwater_layer 2.0 B 1.2
gauge block (double sides)
Combined uncertainty of De uDe – – 30.3
Uncertainty in Lm12 uLm – – 13.4
Cosine error by the angular alignment of the ucos_slot die 3.2 B 1.8
slot die
Repeatability of Lm12 urep_Lm12 41.9 A 13.2
Expanded uncertainty (coverage factor uw – – 66.2
k = 2)
uDe, which is the combined uncertainty of the online qualification of the

effective diameter of the probe tip ball, is one of the uncertainty sources of uw.
Since uDe is composed with the uncertainty of the online qualification, uLq has
been estimated based on the geometrical model shown in Figs. 17 and 19.
According to Fig. 17, the uncertainty corresponding to the cosine error by the
angular alignment error of the probing axis with respect to the gauge block of the
calibration artifact is caused by the X-directional coarse and fine positioning
stages. Consequently, the uncertainty ucos_X-stages due to the cosine error of the
X-directional stage, which can be calculated based on Eq. (10), is evaluated to be
6.4 nm for the online qualification. Furthermore, the uncertainty ucos_interferometer
due to the cosine error caused by the alignment of the probe approaching axis with
respect to the measurement axis of the laser interferometer, which is shown
in Fig. 18, is considered to be the uncertainty source for the qualification, and
the uncertainty due to the cosine error of the laser interferometer is estimated
to be 24.9 nm. The uncertainty uAbbe_X-fine caused by the Abbe error of the
X-directional fine positioning stage, which can be calculated based on Eq. (14),
is estimated to be 1.1 nm. Similarly, the uncertainty uAbbe_X-coarse due to the Abbe
error of the X-directional coarse positioning stage is estimated to be 3.3 nm. The
thermal effect of the laser interferometer also causes the measurement error of
the probing position along the probe approaching axis. The total time for the
online qualification was approximately 50 s, during which the temperature
was 22.68 C 0.01 C. The uncertainty ulaser_thermal caused by the thermal effect
of the laser interferometer is estimated to be 2.9 nm, which is calculated based
on both of the stability of the laser interferometer and the temperature
change during the qualification procedure. The uncertainties of the resolution
(ulaser_resolution) and the linearity (ulaser_linerality) of the laser interferometer are
estimated to be 0.5 nm and 2.9 nm, respectively. Those uncertainties are calculated
by the specification indicated in the data sheet provided by the interferometer
manufacturer. The reading repeatability of the probing, which is determined by the
experimental results indicated in Fig. 21, also becomes one of the uncertainty
sources. Consequently, the uncertainty of urep_Lq12 is evaluated to be 17.1 nm. The
uncertainty due to the length tolerance of the calibrated gauge block utol_ca is
estimated based on the calibration certification provided by Mitsutoyo Corpora-
tion, so it is estimated to be 11.5 nm. The variation of the water layers on the
measuring surface of the gauge block is also considered as the uncertainty source
for the qualification and the gap width measurement. Δs is assumed to possess
rectangular distribution with boundaries of 2.0 nm. Therefore, the uncertainty
uΔs due to the variation of the water layer thickness is estimated to be 11.5 nm. As
a result, the combined uncertainty of the qualification of uDe is estimated to be
60.6 nm with the coverage factor k = 2.
The uncertainty of the gap width measurement uw can be estimated according to
the combined uncertainty uDe. The uncertainty ucos_slot die which is caused by the
angular alignment error of the slot die with respect to the probing axis is evaluated
based on Fig. 18 and Eq. (11), and it is estimated to be 1.8 nm. The uncertainty
urep_Lm12 due to the repeatability of Lm12, which is evaluated based on the
364 S. Ito
experimental results shown in Fig. 22, is estimated to be 13.2 nm. Therefore, the
expanded uncertainty of the gap width measurement uw is estimated to be 66.3 nm
(with a coverage factor k = 2). As a result, the developed micro-probing system was
able to achieve the precision dimensional measurement of the micro-slit with the
measurement uncertainty of less than 100 nm.
Since the measurement point of the gap width can be moved to an arbitrary
position on the micro-slit of the slot die coater by the positioning stages and linear
slide, the gap width uniformity of the slot die coater can be evaluated by the
developed micro-probing system. Figure 23 shows the experimental results of the
gap width uniformity of the slot die coater. The measurement positions for the gap
width on the slot die coater were changed along Y-direction by the linear slide. The
interval of the measurement points along Y-direction was set to be 5.0 mm. After the
gap width measurement along Y-direction, the micro-probe was moved in Z-axis by
the Z-directional coarse positioning stage with the interval of 100 μm. The legends
indicated in Fig. 22 are the Z-directional positions of the measurement points.
The zero point in the Z-direction was set as 90 μm from the edge surface of the
slot die coater to avoid the probing at the chamfered parts. At each measuring
position, the gap width measurement, which means the probing on both surfaces
of the slot die coater, was carried out 5 times. The numbers of the measurement
points were 220. The standard deviation of the probe reading repeatability was
68.3 nm. The maximum and minimum values of the gap width were 81.56 μm and
78.53 μm, respectively. The adoption of a precision linear slide possessing a traveling
range over 200 mm makes it possible to measure the micrometric gap width of the
entire slot die coater. In addition, it can be seen from the figure that the gap width was
successfully measured at Z = 1000 μm, so it indicated the effective working distance
of the developed micro-probing system was larger than 1.0 mm along Z-direction.
82.0
81.5
81.0
80.5
0 100
200 300
80.0 400 500
600 700
79.5 800 900
1000
79.0
78.5
78.0
0 20 40 60 80 100 120 140 160 180 200
Fig. 23 Gap width uniformity of the slot die coater

Summary and Outlook
A micro-probing system with the method of shear-mode detection has been

introduced for the dimensional measurement of the micrometric structures. The
micro-stylus made of a capillary glass tube and high-accuracy glass microsphere
allows good accessibility of the probe tip ball owing to high-aspect ratio shape of
the styli. The shear-mode detection micro-probe realizes nanometer-scale resolution
based on the detection of the surface interaction force, and it contributes
the realization of high-sensitivity and low measuring force for the probing on the
miniature structures. The online qualification using the calibrated gauge block can
provide the uncertainty of less than 100 nm; it will become the effective tools
for the precision dimensional measurement of the precision parts consisting of
micrometer- and nanometer-scale features.
References
Albrecht TR, Grtitter P, Horne D, Rugar D (1991) Frequency modulation detection using high-Q
cantilevers for enhanced force microscope sensitivity. J Appl Phys 69:668–673. https://doi.org/
10.1063/1.347347
Alting L, Kimura F, Hansen HN, Bissacco G (2003) Micro engineering. CIRP Ann 52:635–657.
https://doi.org/10.1016/S0007-8506(07)60208-X
Antognozzi M, Humphris ADL, Miles MJ (2001) Observation of molecular layering in a
confined water film and study of the layers viscoelastic properties. Appl Phys Lett
78:300–302. https://doi.org/10.1063/1.1339997
Bauza MB, Hocken RJ, Smith ST, Woody SC (2005) Development of a virtual probe tip with an
application to high aspect ratio microscale features. Rev Sci Instrum 76:095112. https://doi.org/
10.1063/1.2052027
Bauza MB, Woody SC, Woody BA, Smith ST (2011) Surface profilometry of high aspect ratio
features. Wear 271:519–522. https://doi.org/10.1016/j.wear.2010.03.028
Bos EJC (2011) Aspects of tactile probing on the micro scale. Precis Eng 35:228–240. https://doi.
org/10.1016/j.precisioneng.2010.09.010
Butt HJ, Kappl M (2009) Normal capillary forces. Adv Colloid Interf Sci 146:48–60. https://doi.
org/10.1016/j.cis.2008.10.002
Chen YL, Gao W, Ju BF, Shimizu Y, Ito S (2014) A measurement method of cutting tool position
for relay fabrication of microstructured surface. Meas Sci Technol 25:064018. https://doi.org/
10.1088/0957-0233/25/6/064018
Chen YL, Ito S, Kikuchi H, Kobayashi R, Shimizu Y, Gao W, (2016) On-line qualification of a
micro probing system for precision length measurement of micro-features on precision parts.
Measurement Science and Technology 27(7):074008
Chou SY, Krauss PR, Renstrom PJ (1996) Nanoimprint lithography. J Vac Sci Technol B
14:4129–4133. https://doi.org/10.1116/1.588605
Christenson HK (1994) Capillary condensation due to van der Waals attraction in wet slits. Phys
Rev Lett 73:1821–1824. https://doi.org/10.1103/PhysRevLett.73.1821
Claverley JD, Leach RK (2010) A vibrating micro-scale CMM probe for measuringhigh aspect ratio
structures. Microsyst Technol 16:1507–1512. https://doi.org/10.1007/s00542-009-0967-2
Claverley JD, Leach RK (2013) Development of a three-dimensional vibrating tactile probe for
miniature CMMs. Precis Eng 37:491–499. https://doi.org/10.1016/j.precisioneng.2012.12.008
Claverley JD, Leach RK (2015) A review of the existing performance verification infrastructure for
micro-CMMs. Precis Eng 39:1–15. https://doi.org/10.1016/j.precisioneng.2014.06.006
366 S. Ito
Dai G, Bütefisch S, Pohlenz F, Danzebrink HU (2009) A high precision micro/nano CMM using
piezoresistive tactile probes. Meas Sci Technol 20:084001. https://doi.org/10.1088/0957-0233/
20/8/084001
Durakbasa MN, Osanna PH, Demircioglu P (2011) The factors affecting surface roughness
measurements of the machined flat and spherical surface structures – the geometry and the
precision of the surface. Measurement 44:1986–1999. https://doi.org/10.1016/j.
measurement.2011.08.020
Eoma SI, Takaya Y, Hayashi T (2009) Novel contact probing method using single fiber optical
trapping probe. Precis Eng 33:235–242. https://doi.org/10.1016/j.precisioneng.2008.07.008
Fan KC, Cheng F, Wang W, Chen Y, Lin JY (2010) A scanning contact probe for a micro-coordinate
measuring machine (CMM). Meas Sci Technol 21:054002. https://doi.org/10.1088/0957-0233/
21/5/054002
Ferreira N, Krah T, Jeong DC, Metz D, Kniel K, Dietzel A, Büttgenbach S, Hürtig F (2014) Integration
of a silicon-based microprobe into a gear measuring instrument for accurate measurement of micro
gears. Meas Sci Technol 25:064016. https://doi.org/10.1088/0957-0233/25/6/064016
Furukawa M, Gao W, Shimizu H, Kiyono S, Yasutake M, Takahashi K (2003) Slit width
measurement of a long precision slot die. J Jpn Soc Precis Eng 69:1013–1017. https://doi.org/
10.2493/jjspe.69.1013
Gao W (2010) Scanning micro-stylus system for measurement of micro-aspheric. In: Precision
nanometrology. Springer series in advanced manufacturing, Springer, Germany, pp. 211–243
Gao W, Araki T, Kiyono S, Okazaki Y, Yamanaka M (2003) Precision nano-fabrication and
evaluation of a large area sinusoidal grid surface for a surface encoder. Precis Eng
27:289–298. https://doi.org/10.1016/S0141-6359(03)00028-X
Gao W, Chen YL, Lee KW, Noh YJ, Shimizu Y, Ito S (2013) Precision tool setting for fabrication of
a microstructure array. CIRP Ann 62:523–526. https://doi.org/10.1016/j.cirp.2013.03.013
Gao W, Kim SW, Bosse H, Haitjema H, Chen YL, Lu XD, Knapp W, Weckenmann A, Estler WT,
Kunzmann H (2015) Measurement technologies for precision positioning. CIRP Ann 64:773–796
Goj B, Dressler L, Hoffmann M (2014) Semi-contact measurements of three-dimensional surfaces
utilizing a resonant uniaxial microprobe. Meas Sci Technol 25:064012. https://doi.org/10.1088/
0957-0233/25/6/064012
Goj B, Dressler L, Hoffmann L (2015) Design and characterization of a resonanttriaxial microprobe.
J Micromech Microeng 25:125011. (9pp). https://doi.org/10.1088/0960-1317/25/12/125011
CIRP Ann 55:721–743. https://doi.org/10.1016/j.cirp.2006.10.005
He M, Liu R, Li Y, Wang H, Lu X, Ding G, Wu J, Zhang T, Zhao X (2013) Tactile probing
system based on micro-fabricated capacitive sensor. Sensors Actuators A Phys 194:128–134.
https://doi.org/10.1016/j.sna.2013.01.030
Heckele M, Schomburg WK (2004) Review on micro molding of thermoplastic polymers.
J Micromech Microeng 14:R1–R14. https://doi.org/10.1088/0960-1317/14/3/R01
Hidaka K, Schellekens PHJ (2006) Study of a Small-sized Ultrasonic Probe. CIRP Annals
55(1):567–570
Hidaka K, Saito A, Koga S (2008) Study of a micro-roughness probe with ultrasonic sensor. CIRP
Ann 57:489–492. https://doi.org/10.1016/j.cirp.2008.03.129
Hidaka K, Danzebrink HU, Illers H, Saito A, Ishikawa N (2010) A high-resolution, self-sensing and
self-actuated probe for micro- and nano-coordinate metrology and scanning force microscopy.
CIRP Ann 59:517–520. https://doi.org/10.1016/j.cirp.2010.03.041
Hoffmann J, Weckenmann A, Sun Z (2008) Electrical probing for dimensional micro metrology.
CIRP J Manuf Sci Technol 1:59–62. https://doi.org/10.1016/j.cirpj.2008.06.002
Ito S, Kodama I, Gao W (2014) Development of a probing system for a micro-coordinate measuring
machine by utilizing shear-force detection. Meas Sci Technol 25:064011. https://doi.org/
10.1088/0957-0233/25/6/064011
Ito S, Chen YL, Shimizu Y, Kikuchi H, Gao W, Takahashi K, Kanayama T, Arakawa K, Hayashi A
(2016a) Uncertainty analysis of slot die coater gap width measurement by using a shear
mode micro-probing system. Precis Eng 43:525–529. https://doi.org/10.1016/j.

precisioneng.2015.09.016
Ito S, Kikuchi H, Chen Y, Shimizu Y, Gao W, Takahashi K, Kanayama T, Arakawa K, Hayashi A
(2016b) A micro-coordinate measurement machine (cmm) for large-scale dimensional
measurement of micro-slits. Appl Sci 6:156. https://doi.org/10.3390/app6050156
JCGM (2008) JCGM 100:2008 Evaluation of measurement data-guide to the expression of
uncertainty in measurement (GUM) (International bureau of weights and measures). Bureau
International des Poids et Mesures, Paris
Kim BJ, Masuzawa T, Bourouina T (1999) The vibroscanning method for the measurement
of micro-hole profiles. Meas Sci Technol 10:697–705. https://doi.org/10.1088/0957-0233/
10/8/304
Kim TJ, Kim SH, Do DH, Yoo HK, Gweon DG (2013) Chromatic confocal microscopy
with a novel wavelength detection method using transmittance. Opt Express 21:6286–6294.
https://doi.org/10.1364/OE.21.006286
Krebs CF, Gevorgyan AS, Alstrup J (2009) A roll-to-roll process to flexible polymer solar cells.
Model studies, manufacture and operational stability studies. J Mater Chem 19:5442–5451.
https://doi.org/10.1039/B823001C
Küng A, Meli F, Thalmann R (2007) Ultraprecision micro-CMM using a low force 3D touch probe.
Meas Sci Technol 18:319–327. https://doi.org/10.1088/0957-0233/18/2/S01
Lei L, Deng L, Fan G, Cai X, Li Y, Li T (2014) A 3D micro tactile sensor for dimensional metrology
of micro structure with nanometer precision. Measurement 48:155–161. https://doi.org/
10.1016/j.measurement.2013.10.037
Li RJ, Xiang M, He YX, Fan KC, Cheng ZY, Huang QX, Zhou B (2016) Development of a high-
precision touch-trigger probe using a single sensor. Appl Sci 6(3):86
Liebrich T, Knapp W (2010) New concept of a 3D-probing system for micro-components. CIRP
Ann 59:513–516. https://doi.org/10.1016/j.cirp.2010.03.036
Maeda N, Israelachvili JN, Kohonen MM (2003) Evaporation and instabilities of microscopic
capillary bridges. Proc Natl Acad Sci U S A 100:803–808. https://doi.org/10.1073/pnas.
0234283100
Manning B (2011) The use of non-contact thin gap sensors in controlling coater gapuniformity.
Capacitec Inc. 2011. https://www.capacitec.com/. Accessed 8 Jan 2019
Masuzawa T, Hamasaki Y, Fujino M (1993) Vibroscanning method for nondestructive measure-
ment of small holes. CIRP Annals – Manufacturing Technology 42(1):589–592
Mitsubishi Material (2017) http://www.mmc-slotdie.com/. Accessed 31 Dec 2017
Murakami H, Katsuki A, Onikura H, Sajima T, Kawagoishi N, Kondo E (2010) Development
of a system for measuring micro hole accuracy using an optical fiber probe. J Adv Mech Des
Syst Manuf 4:995–1004. https://doi.org/10.1299/jamdsm.4.995
Murakami H, Katsuki A, Sajima T, Suematsu T (2014) Study of a vibrating fiber probing system for
3-D micro-structures: performance improvement. Meas Sci Technol 25:094010. https://doi.org/
10.1088/0957-0233/25/9/094010
Murakami H, Katsuki A, Sajima T, Fukuda M (2016) Reduction of liquid bridge force for 3D
microstructure measurements. Appl Sci 6:153. https://doi.org/10.3390/app6050153
Nesterov V, Brand U (2005) Modelling and investigation of the silicon twin design 3D micro probe.
J Micromech Microeng 15:514–520. https://doi.org/10.1088/0960-1317/15/3/011
Noh YJ, Arai Y, Gao W (2009) Improvement of a fast tool control unit for cutting
force measurement in diamond turning of micro-lens array. Int J Surf Sci Eng 3:227–241.
https://doi.org/10.1504/IJSURFSE.2009.026611
Santos S, Verdaguer A, Souier T, Thomson NH, Chiesa M (2011) Measuring the true height of
water films on surfaces. Nanotechnology 22:465705. (9pp). https://doi.org/10.1088/0957-4484/
22/46/465705
Schwenke H, Waldele F, Weiskirch C, Kunzmann H (2001) Opto-tactile sensor for 2D and 3D
measurement of small structures on coordinate measuring machines. CIRP Ann 50:361–364.
https://doi.org/10.1016/S0007-8506(07)62140-4
368 S. Ito
Sheu DY (2004) Micro-spherical probes machining by EDM. J Micromech Microeng 15:185–189.

https://doi.org/10.1088/0960-1317/15/1/026
Thalmann R, Meli F, Küng A (2016) State of the art of tactile micro coordinate metrology. Appl Sci
6:150. https://doi.org/10.3390/app6050150
Weckenmann A, Estler T, Peggs G, McMurtry D (2004) Probing systems in dimensional metrology.
CIRP Ann 53:657–684. https://doi.org/10.1016/S0007-8506(07)60034-1
Weckenmann A, Peggs G, Hoffmann J (2006) Probing systems for dimensional micro- and nano-
metrology. Meas Sci Technol 17:504–509. https://doi.org/10.1088/0957-0233/17/3/S08
Weckenmann A, Hoffmann J, Schuler A (2008) Development of a tunnelling current sensor for
a long-range nano-positioning device. Meas Sci Technol 19:064002. https://doi.org/10.1088/
0957-0233/19/6/064002
Zitzler L, Herminghaus S, Mugele FG (2002) Capillary forces in tapping mode atomic force
microscopy. Phys Rev B 66:155436. https://doi.org/10.1103/PhysRevB.66.155436
1 Measurements in Additive Manufacturing
12
2 David Bue Pedersen, Sebastian Aagaard Andersen, and
3 Hans Nørgaard Hansen
4 Contents
5 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 370
6 Physical and Digital Elements of Additive Manufacturing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 372
7 Precision Additive Manufacturing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
8 Photo-Initiated Polymerization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
9 Photopolymer-Based AM Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 373
10 Metal Powder Consolidation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 374
11 Powder Consolidation Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 375
12 Powder Consolidation-Based AM Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 377
13 Measurements in Additive Manufacturing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 379
14 Measurements in Additive Manufacturing: Photopolymerization . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 380
15 Machine Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 380
16 Shape and Shape Deformations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 382
17 Posttreatment Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 384
18 Surfaces in Photopolymer Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 385
19 Measurements in Additive Manufacturing: Metal Powder Consolidation . . . . . . . . . . . . . . . . . . . . . 386
20 Machine Errors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 388
21 Deformation Due to Residual Stresses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 390
22 Density (Inclusions, Porosities, Voids) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 391
23 Surface Quality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 395
24 Summary . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 396
25 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 397
26 Abstract
27 The increasing adoption of additive manufacturing (AM) within the
28 manufacturing industry is pushing companies to rethink how components and
29 integrated component assemblies can be manufactured and not least how to
D. B. Pedersen · S. A. Andersen · H. N. Hansen (*)

Technical University of Denmark, Lyngby, Denmark
e-mail: hnha@mek.dtu.dk

https://doi.org/10.1007/978-981-10-4938-5_13
370 D. B. Pedersen et al.
30 ensure that manufacturing quality is met. This chapter is an enchiridion that

31 details the most commonly applied AM technologies, their existing standards,
32 and what measurements and quality assurance methods that are most
33 relevant for each technology. The chapter is commenced by introducing
34 the principles of vat photopolymerization and powder-bed fusion, after
35 which the intrinsic of each process is linked to what geometrical, surface,
36 and internal characteristics can be expected. This is followed by a
37 deliberation of what measurement methods are applicable and relevant for
38 each process.
39 Keywords
40 Additive manufacturing · Measurements · Metrology · Characterisation ·
41 Validation · Vat photopolymerisation · Powder-bed fusion · Stresses · Surface
42 quality · Tolerances
43 Introduction
44 Additive manufacturing (AM) is a palette of technologies that colloquially is

45 denominated by the common appellation of 3D printing. AM has created a strong
46 linkage between the digital and physical manufacturing, thus nourishing by its nature
47 a wider trend, the digitization, and the automation of the manufacturing industry. For
48 this reason, the increasing adoption of AM within the manufacturing industry is
49 pushing companies to research new ways of adapting their manufacturing models
50 and optimize their manufacturing strategies by integrating these manufacturing
51 technologies of tomorrow into existing production and bolster their strategies toward
52 a digital to physical conversion.
53 This trend sets new demands with respect to how measurements are carried out
54 when these relate to components manufactured by AM. Such measurements can be
55 of dimensional as well as of physical classification. It is rare that AM can stand
56 alone as a manufacturing method. As a minimum validation during pre- and post-
57 manufacturing, tasks must be handled conjointly with the additive manufacturing
58 task. These tasks are both digital and physical in nature and include job prepara-
59 tion, machine preparation, as well as post-processing by cleansing and part
60 treatment for the component to acquire its final properties and dimensions.
61 These activities and their protocols will directly affect and – in many cases –
62 dictate the final part quality and dimensions and must be understood and taken into
63 account when setting up manufacturing based upon any of the industrialized AM
64 methods. In this chapter all of the aforementioned prerequisites for a controlled
65 additive manufacturing process are addressed, and considerations are given for
66 ensuring a reliable manufacturing.
67 By and large, standardization of procedures for suitable approaches to mea-
68 surements within additive manufacturing is an ongoing task. This means that for
12 Measurements in Additive Manufacturing 371
69 many validation procedures, no designated standard method exists to aid the

70 validating body in adhering to correct protocol. The works carried out in stan-
71 dardization for additive manufacturing are predominantly carried out within Com-
72 mittee F42 on Additive Manufacturing Technologies of the ASTM body as well as
73 the corresponding ISO/TC 261 body. ISO/TC 261 and ASTM F42 have formu-
74 lated a Joint Plan for Additive Manufacturing Standards Development wherein
75 four AM standard tracks are being pursued as listed below. Non-designated
76 ASTM/ISO standards are working titles for the Standards Development efforts
77 currently taking place.
78 Terminology
79 • ASTM F2792-12a
80 Standard Terminology for Additive Manufacturing Technologies (Withdrawn
81 2015)
82 • ISO/ASTM 52900:2015 (ASTM F2792)
83 Additive manufacturing – General principles – Terminology
84 • ISO 17296-1
85 Additive manufacturing – General principles part 1: Terminology
86 • ISO/ASTM 52921-13
87 Standard terminology for additive manufacturing – Coordinate systems and test
88 methodologies
89 Process/Materials
90 • ISO 17296-2
91 Additive manufacturing – General principles part 2: Overview of process cate-
92 gories and feedstock
93 • Qualification and certification methods
94 • Requirements for purchased AM parts
95 • Nondestructive evaluation methods
96 Test Methods
97 • ISO 17296-3
98 Additive manufacturing – General principles part 3: Main characteristics and
99 corresponding test methods
100 • Test artifacts
101 • General test methods
102 • Performance test methods
103 Design/Data Formats

104 • ISO 17296-4
105 Additive manufacturing – General principles part 4: Overview of data processing
106 • ISO/ASTM 52915-13
107 Standard practices – Guidelines for design for additive manufacturing
108 • Data structures and metrics for AM models
109 Physical and Digital Elements of Additive Manufacturing
110 By virtue of the highly digital nature of industrialized AM processes, the digital
111 aspect must be taken into account when measuring and planning measurement
112 strategies are taken to protocol in a manufacturing environment. Figure 1 serves
113 as an overview of these elements for a generic AM process. All potential
114 error sources related to the physical and digital process chain are listed in
115 white slanted boxes.
116 Prior to manufacturing, the quality of the source 3D geometry and the digital
117 manufacturing planning through job generation is equally decisive to
118 the manufactured component quality, as physical tasks involving machine
119 setup and material preparation. This in part given that the de facto standard
120 describing geometry for AM job generation is a tessellated format (.STL) by
121 which a CAD geometry is approximated by a faceted wireframe with limited
122 fidelity. During processing, the in-process monitoring and control capabilities
123 of the AM machine tool are equally decisive to part quality as the physical
124 machine design and material processing capabilities. This, as most industrial
125 AM processes, needs to be highly controlled by active control loops in order
126 to sustain stable process conditions. During post-processing the component
127 acquires its final geometry and physical properties. These processing steps
128 can therefore be argued to be the most critical to oversee and regulated by,
129 e.g., NC-driven technologies such as NC machining and NC metrology. Thus,
130 the entire AM workflow is a physical material interaction where strategies,
131 in-process decisions, and posttreatment are tied by a digital information
132 layer that must be tightly overseen to produce measurable and functional
133 components.
Fig. 1 Overview of physical and digital links of an AM process chain

134 Precision Additive Manufacturing
135 The additive manufacturing family covers technologies that are either predominantly
136 employed for industry-grade manufacturing or as technologies that are pertinent to
137 prototyping application. Requirements to functionality and precision are usually less
138 for the latter group, as deviations often can be better accepted on prototype-grade
139 components than for production-grade components. By virtue of this discretizing,
140 Precision Additive Manufacturing in the context of this chapter pertains to those
141 technologies that primarily are applied for industrial manufacturing purposes. These
142 technologies are those which rely on photo-initiated polymerization of plastic and
143 polymer resins as well as those by which metal is consolidated to form near-net-
144 shaped components. The latter where the outcome of the additive process can be
145 regarded as workpieces for a final machining and/or polishing operation to conform
146 with the net-shaped geometry and functional surface specification of the part.
147 Photo-Initiated Polymerization
148 Photo-initiated polymerization is a mechanism that is also known as photo-

149 polymerization in which a light-induced cross-linking mechanism is harnessed in
150 order to cure a polymer resin. In what follows, the mechanism will be denominated
151 simply as “UV curing” given most industry-grade materials for AM is photo-
152 initiated by irradiation within the UV spectrum. UV curing is a process where
153 a polymer resin (i.e., multifunctional monomer and oligomer matrix) is transformed
154 into a cross-linked polymer by a chain reaction initiated by reactive species known as
155 photoinitiators that upon irradiation form free radicals or ions as shown in Fig. 2. The
156 role of the photoinitiator is that most monomers or oligomers do not produce
157 initiating species with high yield to react with UV light. Thus, the photoinitiator
158 needs to be added to the formulation. Depending on the photoinitiator, the photo-
159 polymer can be engineered to react to other ranges of wavelengths (visible light) if
160 this is a wanted property (Fouassier and Lalevée 2012; Decker 1996). Most industry-
161 grade photopolymers are engineered to be reactive in the UV spectrum since the
162 shorter wavelength of coherent UV light (with respect to visible light) allows for
163 smaller features to be generated as a function of the favorable beam-shaping
164 properties of a monochromatic UV light source.
165 Photopolymer-Based AM Methods
166 Additive manufacturing methods that rely on the selective photo-initiated polymer-
167 ization of polymer resins held in a resin container are known by numerous propri-
168 etary names, yet all fall under the denominator vat photopolymerization as given by
169 the ISO/ASTM 52900:2015 naming convention. These processes rely on either the
170 digital-driven scan-line rasterizing of an image by means of a laser point source or on
171 the projection of image masks from a projection source to selectively consolidate
Fig. 2 The principle of photopolymerization
Fig. 3 (a) Bottom-up configuration; (b) top-down configuration of a vat photopolymerization

machine tool for additive manufacturing
172 the photopolymer to the workpiece. Two primary machine configurations exist:
173 a top-down and a bottom-up configuration as seen in Fig. 3.
174 Metal Powder Consolidation
175 Metal powder consolidation is the underlying working principle for a family of
176 processes by which metal powder is selectively consolidated to the near-net shape of
177 a component by either sintering or melting. The consolidation is predominantly
178 induced by means of a high-powered laser and in an inert atmosphere, yet some
179 metal powder consolidation machine tools operate by means of an electron beam in
180 a high-vacuum atmosphere. It is noticeable that unlike the previously described
181 photopolymerization methods, not only geometry but to a high degree also physical
182 properties of the component can be structured in-process. These include density,
183 hardness microstructural composition, stress distribution, and similar properties that
184 can be altered through thermal treatment of metal. Thus the outcome of any given
185 metal AM process chain is the most complex ramification of the elements presented
186 in Fig. 1. Both geometry and material are given their substance by the swift-fusing
187 metal particles and rapid progression of the heat source away from the treated area
188 introducing quench-like cooling. As successive layers are built atop, reheating of the
189 already consolidated matter occurs. The outcome of this discordant by-layer process
190 are components that in most instances narrowly meet near-net shape, and a similar
191 system as that of tolerance grades and machining allowance grades known from cast
192 metals and their alloys (ISO 8062) must be used to accommodate allowance for
193 a subsequent machining procedure. Consequently final quality assurance of the
194 component follows the methods that are defined for machined parts and surfaces
195 (DS/CEN ISO 1997).
196 Powder Consolidation Methods
197 Production of AM metal parts via powder consolidation can follow two strategies.
198 First is sintering, which requires the material to reach temperatures near melting
199 temperature at which the powder fuse to a porous matter with a microstructure that is
200 inherited from the packing of the powder prior to consolidation. Second is a full
201 melting of the powder by which the powder is heated to temperatures above the
202 melting point of the powder. Hereby a melt pool is formed, and the resolidification of
203 the melt expunges the traces of the porous powdered initial state of the process
204 material, thus forming dense solid matter. Sintering is generally known to be a
205 cleaner process with less formation of spatter and better surface fidelity yet traded
206 off by more porous and weaker material structure.
207 Sintering
208 Sintering bonds particles at their interface and thus forms a solid (porous) structure,
209 through heating without melting (alternatively in combination with or by pressure).
210 Solid-state sintering occurs when a powdered material is heated and is consequently
211 the mechanism applied in sintering-based metal AM. Adjacent powder particles will
212 start merging and bond through diffusion, as shown in Fig. 4. After bonding, the
213 bonded bulk material will experience a change in global properties, such as strength,
214 density, ductility, and thermal/electrical conductivity. The nature of the sintering
215 process makes it prone to the formation of pores medially between the surrounding
216 particles.
217 This too is illustrated in Fig. 4. Here the particles shown merge during the
218 sintering process, and neck growth occurs between the former separate metal
219 particles. Finally, shrinkage occurs, by which, during the neck growth, the center
220 parts of the particles move closer together and the compaction of the particles into a
221 solid makes the total volume shrink as compared to the original state (Kalpakjian and
222 Schmid 2010).
223 Melting
224 As opposed to sintering, the consolidation of the powder particles by melting is done
225 in a way that ensures that the targeted particles are fully molten, as seen in Fig. 5.
Fig. 4 The principle of solid-state sintering
Fig. 5 The principle of full

melting of a workpiece
226 This approach is more difficult to control since minute changes in heat flux from the
227 heat source affect the size of the melt pool, affecting overhanging structures and in
228 general the down-face of the manufactured component. Whereas sintering within
229 AM is not exclusive to metals, characteristics of metal alloys such as conductivity,
230 and a distinct melting point, make them de facto exclusively used as materials for full
231 melting in powder consolidation systems. The rapid melting and solidification
232 enabled by the process allow for parts with tailored properties distinct from those
233 obtained from traditionally processed parts (Gibson et al. 2010a). This is achieved by
234 the exploitation of localized material properties, obtained from refined microstruc-
235 tures, formation of non-equilibrium phases, and supersaturated solutions. Second-
236 phase particles such as inclusions and carbides, or extremely fine, refined micro-
237 structures, are side effects of the process.
238 Due to the similarity of powder-based sintering and melting, whether the
239 technology is referred to as sintering or melting, often both of the states are present.
240 Given that the sintering temperature is slightly below the temperature at which a melt
241 pool is formed, as shown in Fig. 5, sintering will occur in the boundary of a geometry
242 made from a melt-driven process. Similar to this, localized melting can occur during
243 a sintering process, where impurities and process fluctuation may temporarily lead to
244 the formation of melt. Melting is often preferred over sintering given that, in order
245 for the process to be as economically competitive as possible, the fusion of the
246 powder should happen as quickly as possible in order to increase the build rate.
247 This is achieved by a fast-moving energy source, with a high-energy flux, and fusion
248 by melting. As fusion by sintering alone, it typically takes much longer time. Hence,
249 the reason for most AM processes is not to use sintering as the primary mecha-
250 nism of fusion. Nonetheless, sintering still plays secondary roles in the process
251 (Gibson et al. 2010b).
252 Powder Consolidation-Based AM Methods
253 Two primary machine tool types exist for the additive manufacture from metal
254 powders. They differ by design by one process employing a powder bed in which
255 powder is consolidated by either sintering or melting and where the process
256 can occur either in a vacuum or in an inert atmosphere. The other method employs
257 a cladding technique where a freely movable laser nozzle delivers a powder spray in
258 a carrier gas to the workpiece.
259 Cladding
260 The cladding principle is illustrated in Fig. 6. A tooling system comprised of an
261 integrated two-stage nozzle system function to deliver a steady stream of fluidized
262 powder to the workpiece while also delivering the heat flux in the form of focused
263 laser light to the workpiece. Fluidizing the powder and bringing it into the beam of
264 the laser are done by means of an inert carrier gas spraying the powder directly to
265 where material is required. The laser beam intercepts the powder stream at the
266 desired deposition location and fuses the melt directly onto the workpiece. The
267 nozzle is mounted in an NC machine tool and contains both sprayer and typically an
268 optical delivery fiber fixing their location relative to each other, ensuring the correct
269 focal distance of the laser. Most systems employ a 5-axis NC machine tool that allow
Fig. 6 The principle of laser

cladding
270 for rotation of the workpiece such that the cladded material is deposited with the
271 force of gravity, hereby minimizing the need for support structures. Support structure
272 serves primarily to fix the workpiece such that thermal residual stresses do not allow
273 for bending of the workpiece to occur. Many cladding systems are integrated on
274 5-axis milling machine tools, allowing for hybrid manufacture such that internal
275 surfaces can be machined before these are closed off and inaccessible for post
276 machining. Given the free-form nature of cladding, layers that are cladded to the
277 workpiece do not need to be planar, and the process is ideal for adding material to an
278 existing workpiece that has been made from conventional manufacturing either for
279 finishing the workpiece or for repair purposes.
280 Powder Bed Fusion

281 The most common machine tool for metal additive manufacturing is the powder bed
282 fusion systems. This type of machine operates by selectively fusing powder follow-
283 ing a by-layer deposition principle. Powder is contained in a silo and fed to a recoater
284 that coats the build chamber of the machine with an evenly distributed layer of
285 powder between consolidation operations. The powder delivery system is responsi-
286 ble for relocating the build plate along the z-stage and therefore defines the accuracy
287 and resolution of the build in the vertical direction. The tool distributing the powder
288 evenly can be in the form of a scraper, a roller, or a blade that is designed to both
289 fluidize and distribute the powder. As this tool has major influence of the distributing
290 of the layer, the powder and the tool have to be carefully matched to the powder and
291 the application. When the layer is in place, the laser beam is selectively maneuvered
292 to consolidate the metal powder particles in order to create the desired geometry as
293 seen in Fig. 7.
294 As the only moveable part in the consolidation process is the laser beam, the
295 control is done by a scanner mirror system. Typically, two separate mirrors mounted
Fig. 7 The principle of

powder bed fusion
296 onto galvanometric motors are used to rotate each mirror. The mirrors are then
297 located in a way that allows discrete movement along the xy plane. This is a
298 consolidation process where support structures serve, not only as traditional mechan-
299 ical reinforcements, but also as heat conducters to minimize thermal residual
300 stresses.
301 Measurements in Additive Manufacturing
302 Discrepancies of photopolymerization and powder consolidation lend to contrasting

303 focus for validation of parts as manufactured and in advance of post-processing. The
304 high fidelity and material homogeneity of photopolymer components allow for net-
305 shaped manufacturing, and measurements can be taken similarly to measurements in
306 the thermoplastic processing industry. These measurements are typically those of
307 shape, geometry, and surface and serve foremost to quality assure the manufactured
308 geometrical tolerances. The methods therefore can be tactile or optical and can range
309 over simple control measurements using hand measurement devices such as calipers
310 and micrometer dial gauges to complex validation using optical CMMs or 3D
311 scanners. The latter can further contribute to measurement effectiveness when
312 dealing with a high geometrical complexity such as the manufacture of hearing
313 aids which is one of the largest industrial use cases for photopolymer components.
314 Metal components made from consolidation of powder are rarely manufactured
315 directly to tolerances and net shape. Given the poor fidelity of components as
316 manufactured in the additive machine tool, the components are produced with
317 machining allowances for subsequent fit and finishing. In many use cases, the
318 components are blatantly regarded as feedstock for a complex thermal and mechan-
319 ical process chain that finishes with NC machining, grinding, and polishing of the
320 component to reach net shape. Measurements following the additive manufacture
321 serve therefore primarily to ensure that the material and inaccessible internal struc-
322 tures are manufactured as expected. Thus, the tool of choice is often, and in some
323 cases solely, to be using CT scanning for the validation procedure. This, however, is
324 not the only suitable approach to measurements for powder-based metal additive
325 manufacturing, and alternative validation methods will be presented in the follow-
326 ing. It is ultimately worth to bespeak that measurement of complex internal and
327 inaccessible areas of any additively manufactured component can only be nonde-
328 structively measured by penetrating scanning methods such as CT scanning. Albeit
329 the discrepancies of photopolymerization and powder consolidation manufacturing
330 methods, a general classification for measurements within additive can be segmented
331 as follows and will be used in sections “Measurements in Additive Manufacturing:
332 Photopolymerization” and “Measurements in Additive Manufacturing: Metal Pow-
333 der Consolidation”:
334 • Machine errors – measurements relating to the validation of the additive machine
335 tool, e.g., spindle errors and axes errors including translational, rotational, and
336 squareness errors
337 • Shape/geometry – of the manufactured component

338 • Surfaces – of the manufactured component
339 • Material (properties) – of the manufactured component
340 Measurements in Additive Manufacturing: Photopolymerization
341 This section covers an array of common errors that is related specifically to compo-
342 nents made by vat photopolymerization. Some of these errors are inherent to the
343 machine tool, whereas others are inherent to the process of selective cross-linking of
344 the photopolymer. These errors must be understood before appropriate validation
345 methods can be selected. The errors can be divided into:
346 • Machine errors

347 • Geometry, shape, and shape deformations
348 • Posttreatment errors
349 Typically material properties in these types of processes are related to post-
350 treatment of the parts.
351 Machine Errors
352 All common photopolymer-based systems are electro-opto-mechanical

353 systems; thus errors are originating from these three main system components.
354 Electro
355 For scanning laser systems, the galvanometer scanning system is discretized to an
356 extent where motion errors are predominantly caused by overshoot, ringing, and
357 linearity similar to errors found in powder bed fusion systems as described in section
358 “Electro.” For mask projection systems, an error component is introduced to the
359 horizontal plane of the build envelope of the machine tool. This error is inherited to
360 the discretization from geometry to the limited pixel resolution available on the
361 digital mirror device (DMD) micro-mirror chip array that is used for image gener-
362 ation as shown on Fig. 8.
363 Furthermore, local damage to one or more mirrors of the DMD can produce a
364 “dead” pixel which will be either always on or always off due to damage to the
365 mirror pivoting mechanism.
366 Since all geometry is a best fit within the pixel discretization of the DMD,
367 geometry, e.g., a circular shape, will be resolved into a corresponding pixel matrix
368 as seen on Fig. 9. Already at this stage, a roundness error is introduced. In this
369 example, a circular feature covering 14 (left) and 15 (right) pixels is resolved to
370 the pixel matrix on Fig. 9. The introduced roundness error is 0.856 and 0.902 for the
371 14- and 15-pixel diameter circles, respectively.
Fig. 8 Design of a DMD

micro-mirror array
Fig. 9 Pixilation introduced by resolving a circle of 14 (left) and 15 (right) pixels onto a DMD
372 Mechanical
373 The vertical stage of any photopolymer system is subject to the same errors
374 as are known from the characterization of common motion stages, with the
375 same methodology to describing the accuracy and repeatability of the
376 stage. These error components are the reversal error, slack, runout (straightness),
377 angular runout (pitch/yaw/roll), eccentricity, wobble, drift, load capacity (trans-
378 verse, normal, axial), inertia, jerk, and alignment of the stage to the image plane of
379 the system.
380 Optical
381 Optical errors are errors introduced from the optical path from the light source to the
382 image plane of the system. These errors manifest themselves as distortions which are
383 the deviation of the image from the ideal rectilinear projection. The distortion field is
384 comprised by two main components, tangential (skewness) caused by misalignment
385 of optical elements to the optical axis of the lens system and radial (barrel or
386 pincushion) distortions, though the radial distortion field is minimal in modern
387 lens systems (caused by lens geometry approximation). Furthermore, optical aber-
388 ration from imperfections in the grinding and polishing of the lens elements of the
389 projection system may induce localized errors and errors such as local defocusing
390 (Ahmad 2017).
391 Shape and Shape Deformations
392 Shape errors will occur during manufacturing. These relate predominantly to two
393 main constituents: from shrinkage during solidification and from plastic deformation
394 during layer change.
395 Unlike thermoplastic materials that exhibit a branched or linear molecular struc-
396 ture that enables these to be melted and resolidified, photopolymers are cross-linked
397 polymers. After photopolymerisation, they exhibit a networked structure that cannot
398 be melted and resolidified and thus are less prone to creep and stress relaxation.
399 There are two dominating types of photopolymer systems for vat photo-
400 polymerisation. These are free-radical (typically acrylates) and cationic (typically
401 epoxies and vinyl ethers).
402 The early resins created for vat photopolymerization was solely comprised from
403 acrylate-based resins, which have high reactivity (fast reaction when exposed to UV
404 light) as well as a high shrinkage rate (5–20%) (Gibson et al. 2010c). Due to the fast
405 solidification and resultant, little relaxation time stresses build up rapidly in the
406 photopolymer than for the epoxy-based resins that cure slower allowing for relax-
407 ation during solidification (Eschl et al. 1999). Therefore, acrylate resins are prone to
408 poor geometrical and mechanical performance, not solely from the high shrinkage,
409 but also from lack of relaxation during processing as a result of the fast curing.
410 Epoxy-based resins on the other hand have low shrinkage (1–2%), lower solidifica-
411 tion rate, and thus less tendency to curl and warp. Current photopolymers are
412 complex blends of methacrylates, epoxies, and polyesters designed to yield good
413 overall performance with respect to processability and performance in their cross-
414 linked processed state. These blends shrink typically 4% (Moothanchery et al. 2012).
415 Irrespective, given the layered nature to vat photopolymerization, as each layer
416 contracts upon manufacture, the resultant shrinkage leads to stress buildup in the
417 component that will result in component warp. This is seen most clearly on wide
418 components that are manufactured laying down parallel to the build plane and on
419 sides facing the build platform. A critical example of warping geometry can be seen
420 in Fig. 10.
Fig. 10 Left, an example of high warp of an acrylate photopolymer. Right, a modern photopolymer
blend with low shrinkage and orthogonal part orientation to minimize warp
Fig. 11 Severe forces during

layer change on a bottom-up
style vat photopolymerization
platform causing plastic
deformation leading to critical
manufacturing failure
421 Errors relating to plastic deformation during layer change are induced
422 from the solidification strategy that makes layered photopolymerization possible.
423 Solidification of photopolymer onto a fully cured previous layer results in an
424 interface zone where adhesion onto the previous layer is weak. This result in
425 poor mechanical properties with high risk of delamination occurring. Therefore,
426 any layer during the fabrication is only irradiated with an image mask for the
427 duration it takes for the cross-linking to form a stable layer. This can be regarded
428 as a layer of hard gel. Upon layer change, for bottom-up machine tools, this
429 layer will be under tension as the gelated layer is being lifted off the exposure
430 window in the bottom of the vat, leading to plastic deformation. Upon
431 repositioning, the layer will be under compression as the photopolymer film
432 between the component and exposure window is being squeezed away. This
433 type of effect can be seen in Fig. 11.
434 For top-down machine tools, the gelated layer may be affected by the recoating
435 action of the wiper. For top-down machines, the effect is less prone to cause
436 geometrical deviations on the part.
437 Posttreatment Errors
438 Post-manufacturing treatment of photopolymer components is critical for the man-

439 ufacture of components that conform with tolerances. This both relates to dimen-
440 sional and physical tolerances. Post-manufacturing treatment includes the following
441 procedures where tolerance deviations can be induced:
442 • Washing and cleaning

443 • Support removal and post curing
444 • Fit and finishing
445 Washing and Cleaning

446 Washing and cleaning serve to remove the thin film of unsolidified photopolymer
447 that will cling to the manufactured geometry. This process involves the use of a
448 solvent to dissolve the film as well as a washing that can be either through agitation
449 via ultrasound or rinsing through circulation of the solvent. A commonly used
450 solvent is isopropanol which has good solubility compatibility with most industrially
451 employed photopolymers. Washing and cleaning may lead to a whitening discolor-
452 ation of the component from the chemical interaction with the solvent that can be
453 prevented by following the washing procedure with a treatment of the component by
454 briefly submerging it into propylene glycol. The treatment of the component with
455 these chemical compounds will result in that the component will swell. A temporary
456 effect that requires the component to rest until swelling has diminished before further
457 processing. It is worthwhile to notice that alcohols are prone to cause stress crazing
458 of acrylic materials. Given most photopolymer blends contain acrylic compounds,
459 surface cracking from posttreatment can be observed and typically relates hereto
460 (Technical Paper 2007). Figure 12 shows how surface crack occurrence can be seen
461 in poorly washed and cleansed photopolymer components.
462 Support Removal and Post Curing

463 Post curing is the process by which the washed and cleaned component is given its
464 final mechanical properties, where the component is cured to an extent where it is
465 chemically safe for direct skin exposure and where the curing-induced change in
466 stress concentration within the part may result in that the component deforms
467 excessively. One of the most often seen errors induced by this process is that not
468 all residual photopolymer from the cleaning procedure has been removed from, e.g.,
469 high aspect ratio holes, crevices, and internal tunnels. This matter will be solidified
470 by post curing and will render these features incomplete spite correctly produced
471 during manufacturing. Assuming correct cleansing, errors can be minimized during
472 post curing by leaving support on the component to reinforce it during the change of
Fig. 12 Surface cracks

during posttreatment due to
alcohol-induced stress crazing
and/or aggressive light-
induced post curing
473 stress concentration as a result of the post curing. A slower post curing by floodlight
474 or by heat-induced post curing is to favor over aggressive flashlight curing since the
475 former methods will allow for the components to undergo relaxation during the
476 treatment. Support removal should be carried out by cutting and polishing over
477 breaking off, to limit scarring on surfaces in contact with the support structure.
478 Fit and Finishing

479 Many industrial photopolymer components, e.g., hearing aid shells undergo a final
480 automated fit and finishing procedure that involve automated polishing. This is most
481 often done by a rotary tumble polishing machine. If such procedures are used for the
482 component to be validated, rotary tumble polishing and its effect are well studied in
483 the literature and should be taken into account.
484 Surfaces in Photopolymer Components
485 A well-performing vat photopolymerization platform is capable of producing high-

486 fidelity components and allows for embedding of complex three-dimensional
487 micro features and the manufacture of multi-scale components. Such a component,
488 manufactured to recreate a bioinspired hydrophobic surface, can be seen on Fig. 13.
489 On such components, it can be difficult to appoint when to distinguish among form
490 and surface (Ribó 2017). Thus, simple surface measurements such as profilometry
491 are unsuited to describe the component and advanced methods such as confocal
492 microscopy μSEM and 3D LDOF microscopy.
493 Advanced methods, despite their capabilities to map surfaces with high fidelity,
494 can prove inadequate, as can be demonstrated with the exhibited multi-scale micro-
495 structured component. Figure 14 shows an attempt at mapping the component with a
496 3D focus-variation microscope. Inspecting the component at an acute angle results in
497 numerous blind spots of the stem of the structure which force upon conducting the
498 inspection as a normal to the surface. Hereby side walls are obstructed from view.
499 Thus, multiple measurements from complementary angles must be carried out in
500 order to fully map the surface.
Fig. 13 Optical microscopy of a component with a multi-scale microstructured surface made by

vat photopolymerization (Ribó 2017)
501 Certain microstructural features are inherited from the photopolymerization pro-
502 cess itself. In section “Electro,” it was described how arcs are discretized at image
503 mask level. If a thorough focus calibration is carried out on the DMD-enabled vat
504 photopolymerization machine tool, this pixelated approximation is visible on micro-
505 structure level on the manufactured component as in Fig. 15.
506 This can be regarded as an inevitable surface characteristics that is similar to, e.g.,
507 the periodic surface profile of a component turned on a lathe. As with a turned
508 component where the periodic surface can be used to find the transversal feed of the
509 carriage, the pitch of the pixel grid can yield valuable information of the state of
510 optical system on the vat photopolymerization machine tool. Below, a surface
511 microscopy image was processed by identifying the pixel boundaries with their
512 gray-scale information; it was possible to measure each pixel as shown in Fig. 16.
513 The machine tool has a nominal pixel size of 7.6 μm. Edges of the tip had an average
514 of 5.96 μm of pixel replication on the surface which is smaller than the nominal.
515 This effect is from that the very edge of the inspected feature is not fully cured, due to
516 a combination of light scattering and light-intensity drop on the edges of each pixel.
517 The pixels found in the center of the inspected feature have an average of 7.34 μm,
518 slightly smaller than the nominal value. This stems in this specific case from the
519 optical system since the optical plane of the lens system is slightly offset farther from
520 the DMD chip than per design, resulting in a smaller field of view at the focal plane.
521 Measurements in Additive Manufacturing: Metal Powder

522 Consolidation
523 With basis in the most common powder-based consolidation method, the powder
524 bed fusion manufacturing methods, the following will cover the most prominent
525 measurements as well as highlight the most common errors that are introduced to the
526 process and which must be understood in order to choose appropriate validation
527 methods during production. For the select errors, methods are proposed to measure
528 them. These errors are:
12
Measurements in Additive Manufacturing
Fig. 14 Focus-variation microscopy of microstructured surface manufactured using photopolymerization (Ribó 2017)
387
Fig. 15 Cylindrical features on a sample (Ø150μm), discretized at image level (Ribó 2017)
529 • Machine errors

530 • Deformations due to residual stresses
531 • Inferior mechanical properties (lack of fusion, low densities)
532 • Surface quality
533 Machine Errors
534 Due to the subsystems that comprise the machine tool for powder bed fusion, some
535 of the errors found in the final products originated from errors contributed to the
536 errors of the equipment itself.
537 Electro
538 The scanner system utilized to move the laser beam over the powder bed consists of
539 two galvanometers. Maneuvering of these is done by an analogue driving current,
540 produced by a digital platform, and the resolution is therefore governed by the
541 resolution achievable by the digital-to-analogue conversion embedded in the con-
542 trolling electronics. This circuit is prone to thermal drift; though compensated
543 through a feedback controller, thermal drift will to some extent occur, predominantly
544 from cold start. Thus, it is important to allow for the machine tool to reach a thermal
545 steady state of the galvanometer system prior to job execution.
546 Mechanical
547 The process of lowering and recoating the build plate is highly mechanical, and
548 therefore the level of accuracy in displacement of the build plate decides the
549 accuracy of the layers layed out. As the build plate is mounted on a spindle, the
12
Measurements in Additive Manufacturing
Fig. 16 Evaluation of the optical system based upon pixel grid replication on the surface of a test specimen (Ribó 2017)
389
550 error in gradient of the spindle influences the positioning. The spindle error compo-
551 nents are reversal error, slack, runout (straightness), angular runout (pitch/yaw/roll),
552 eccentricity, wobble, drift, load capacity (transverse, normal, axial), inertia, jerk and
553 the alignment of the stage to the optical plane of the system. For the recoater,
554 mechanism errors are typically introduced from defects to the recoater and from
555 the recoater hitting defects on the workpiece which result in a chatter.
556 Finally, the galvanometer system is accelerated at extremely high acceleration
557 forces. This may result in overshoot and ringing that is typically seen on worn
558 galvanometer systems and systems operating at excessive speeds.
559 Optical
560 The laser beam is aligned with the scanner system in a way that allows full use of the
561 scan area, but if any misalignment is present in the system, this misalignment influ-
562 ences the positioning and homogeneity of the beam on the powder bed. The scan lens
563 used to focus the beam onto the platform also introduces distortions to both the shape
564 and the scan velocity of the powder bed. Even though most scanner systems are fitted
565 with an fθ-lens that negates most of the shape distortion and in great ensures that the
566 scan velocity is linearly proportional to the angular displacement of the mirrors
567 mounted on the galvanometers, the influence cannot be fully neglected.
568 Deformation Due to Residual Stresses
569 One of the most common causes for poor geometrical conformance with specifications
570 of components manufactured by either cladding or powder bed fusion is warp as a
571 result of residual thermal stresses that build up in the workpiece over the duration of the
572 manufacturing of the component. This can to an extent be countered by understanding
573 the consequences of different scan strategies. Choosing a scan strategy to limit that the
574 treated part warps during solidification can alter the outcome from a nonconformance
575 with specifications to conformance. The warpage of the component is caused by a
576 buildup of residual stresses as the steep temperature gradient in the normal direction of
577 the workpiece causes the upper layer to expand. As the substrate as well as the already
578 solidified material constrain the expansion, the elastic and plastic strain will result in a
579 compressive state, causing the part to bend toward the heat source, as shown in Fig. 17.
580 During cooling, the previously plastically compressed upper layer becomes shorter,
581 and therefore warping can be observed (Kruth et al. 2004).
582 Measurements serve a critical role in validating models used for, e.g., optimiza-
583 tion of scanning strategies. As presented in the previous section, the localized
584 heating has a great impact on the residual stresses and microstructural formation.
585 The possibility for optimizing the scanning strategy with focus on the final product is
586 being introduced by process modeling groups. Even though the industry has
587 approached this issue, the release is often a stationary pattern that shifts the angle
588 of the hatch lines with a fixed offset. Utilizing numerical modeling and realizing the
589 physical influences on the process allow more specific optimization. As the metal
590 powder consolidation process is a multi-physical process, both thermal models,
Fig. 17 Temperature gradient mechanism during heating and cooling
591 microstructural models, mechanical models, and others influences the final product.
592 Therefore an accurately tailored model of the process is cumbersome and CPU-
593 intensive (Saad et al. 2016). Solving the optimization problem of finding the best
594 scanning strategies, decreasing the overall deformation as a consequence of mini-
595 mizing the residual stress, is naturally even more theoretical. Using a simplified
596 model in order to allow several evaluations in a manageable period of time can be
597 adopted by subdividing the scan field into smaller squares and applying a pseudo-
598 analytical model together with a simple thermal fitness number that expresses
599 thermal “inhomogeneity,” and the degree of consolidation allows this optimization
600 of scan strategy in a short-time frame (Mohanty et al. 2017).
601 In Fig. 18, a beam was consolidated using traditional scan strategies, and the
602 deformation after being removed from the build plate was measured with a coordi-
603 nate measurement machine. Here the deformation due to the buildup of residual
604 stresses is visible. Using an optimized scan strategy based on the thermal fitness
605 number and therefore a simplified model leads to the model seen in Fig. 19. Here the
606 deformation of the beam has been reduced by a factor of 10, validated by form
607 measurements of the samples (Figs. 18 and 19).
608 Density (Inclusions, Porosities, Voids)
609 When metal additive manufacturing is employed for industrial-grade component

610 manufacturing, ensuring that the method is capable of producing components with
611 mechanical properties similar to those from traditional manufacturing is of essence.
612 The quality of the produced parts is thus in great defined from the density of the part.
613 Several proprietary powder bed fusion systems claim a density of close to 100%.
614 Verification of the density is key to validating if the component has the expected
615 performance characteristics and will often have to be carried out as part of the quality
616 assurance; here three methods will be described:
617 • Archimedes method

618 • Computed tomography (CT)
619 • Optical measurement technologies
Fig. 18 Beam with traditional parallel scan strategies, maximal deformations of 4 mm are
measured from form measurement of the top-surface (Mohanty et al. 2017)
Fig. 19 Beam produced with an optimized scan strategy, maximal deformation of 0.35 mm from
form measurement of the top-surface (Mohanty et al. 2017)
620 Archimedes Method

621 The method of Archimedes is one of the more straightforward methods of verifying
622 the density of a part. The mean density of the entire part is obtainable regardless of
623 the complexity of the part. The method requires a highly accuracy scale and a liquid
624 with a known density. The part is consequently measured in air, ma, and then again
625 when lowered into the liquid, ml. The part density is defined from Eq. 1 and can be
626 compared to the nominal density of the bulk material.
ma
ρp ¼ ρ (1)
ma ml l
627 Computed Tomography (CT)

628 Despite the robust and straightforward way of the Archimedes method in determin-
629 ing the average density of the part, it does not yield any information about the
630 location or size of the individual porosities yielding a not fully dense part. To obtain
631 this kind of information, other methods have to be applied.
632 X-ray CT scanning is a suitable nondestructive way of creating a 3D image of the
633 internal structure of the component and its contained porosities. This method relies
634 on the emission of x-rays. These are very high-frequency and short-length waves that
635 will penetrate solids, and much like the shadow cast when a hand blocks visible
636 light, the geometry can be obtained by observing how the intitial emission was
637 altered by the targeted solid. A detector records the intensity of the x-rays after they
638 have passed through the part and is thereby capable of locating voids or other
639 inclusions obstructing the passage of the x-rays. Each image captured by this method
640 is by nature 2D and will therefore give a 2D position of the inhomogeneity of the
641 metal part. From the 2D image stack, the true 3D location and size of voids and
642 inclusions can be mapped by correlating the image stack. Correlating the images
643 allows a 3D reconstruction of the part, as shown in Fig. 20, describing the entirety of
Fig. 20 The principle of x-ray tomography

Fig. 21 Reconstruction of a CT scan of a 10 10-mm-diameter, 1.5-mm-tall flat sample showing

internal porosities (Valente 2017)
644 the CT-scanned object. For this method the spatial resolution is inversely coupled to
645 the size of the volume investigated. Therefore it can be hard to obtain information of
646 very small defects. Studying the 3D model, information about the location, size, and
647 shape of the voids is achievable as shown in Fig. 21, where a sample has been
648 scanned to reveal internal porosities.
649 Optical Microscopy

650 Optical Microscopy is a destructive method when used to obtain information
651 about interal porosities of a component. The information obtained relies on the
652 diffusive reflectance of the surface, and the specimen must therefore be polished
653 until it is optically smooth. The depth-wise resolution provided by this method is
654 controlled by performing cross-sectional cuts throughout the specimen, and is
655 therefore put in the destructive category.
656 Investigating the polished surface allows for detection of voids or inclusions; see
657 Fig. 22. OM does not have the limiting spatial factor known from CT scans, as spatial
658 investigation can be carried out by serial sectioning. The captured 2D images allow
659 detection of smaller defects than the CT, but due to the polishing and cutting, it is
660 more time-consuming.
661 Aside from observing voids in the cross section, optical microscopy is also very
662 capable of revealing geometry surface textures and optical material properties
663 revealed after etching.
664 Using an etching agent on the polished surface will reveal the microstructure
665 present in that exact cross section, and looking at the AM part both along the scan
666 direction and across will reveal a difference in microstructure according to the
667 material, scan velocity, power, and scan strategy.
Fig. 22 (a) High-resolution optical microscopy montage of a 10 10-mm-diameter, 1.5-mm-tall

flat sample showing the morphology of two populations of voids. (a) Sectioning, (b) x-z section
from CT reconstruction, (c) y-z cross section (Valente 2017)
t:1 Table 1 Profile discrepancy between instrument pairs. Percentage of the profile length where the
confidence interval does not intersect (Thompson et al. 2017)
t:2 CM/CSI CM/FV CM/XCT CSI/FV CSI/XCT FV/XCT
t:3 Discrepancy% 51.7 61.9 48.5 57.4 42.9 48.8
668 Surface Quality
669 The surface quality of the consolidated powders is very complex as a consequence
670 of the selective melting and resolidification and therefore has very poor surface
671 characteristics compared to other processes as the surface is elongated melt pools.
672 These are challenging to measure due the inhomogeneity and residue left by
673 spatter.
674 Topology Inspection of Surfaces

675 Achieving sound measurements of the surface topology is not yet fully achievable,
676 and most likely the best fit is found in a combination of different methods.
677 Confocal microscopy (CM), focus-variation (FV) microscopy, coherence
678 scanning interferometry (CSI), and XCT are among the methods that can be used
679 to obtain knowledge about the topology of the powder bed fusion parts. As the
680 surface is prone to having irregular optical features, the optical measurements
681 provide differences related to the approach used. The discrepancy between the
682 mentioned measurement methods for the same topography is presented in Table 1,
683 and a qualitative overview is given in Fig. 23.
684 It is revealed from Table 1 that the discrepancy among the different methods is
685 high and must therefore be applied only with thorough understanding of the spec-
686 imen and the measurement method used.
Fig. 23 Reconstructed portions of aligned topographies (top views, height-based coloring); (a)
CM; (b) CSI; (c) FV; (d) XCT (Thompson et al. 2017)
687 Summary
688 In principle, measurements in additive manufacturing are only measurements. In this

689 chapter we have tried to introduce a model for linking the digital and physical
690 process chain related to additive manufacturing. These two process chains both
691 introduce errors and variations into the final component manufactured. The chapter
692 describes in detail two industrially adapted additive manufacturing processes: photo-
693 polymerization and metal powder consolidation. For each of these process families,
694 measurements and their challenges related to machine errors, material properties,
695 shape deformations, and surface quality are described and discussed.
696 References
697 Ahmad A (2017) Handbook of optomechanical engineering, 2nd edn. CRC Press, Boca Raton.
698 ISBN 9781498761482
699 Decker C (1996) Photoinitiated crosslinking polymerisation. Prog Polym Sci 21(4):593–650.
700 ISSN 00796700
701 DS/CEN ISO (1997) Geometrical product specifications (GPS) – masterplan (DS/CEN/CR
702 ISO 14638)
703 Eschl J, Blumenstock T, Eyerer P (1999) Comparison of the curing process of epoxy and acrylate
704 resins for stereolithography by means of experimental investigations and FEM-simulation.
705 In: Solid freeform fabrication proceedings, pp 453–460. Solid freeform fabrication symposium
706 Fouassier JP, Lalevée J (2012) Photopolymerization and photo-cross-linking. In: Photoinitiators for
707 polymer synthesis. Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim, pp 3–9
708 Gibson D, Rosen W, Stucker B (2010a) Additive manufacturing technologies: 3D printing, rapid
709 prototyping, and direct digital manufacturing. Springer. Section 5.6.2. ISBN 1493921134
710 Gibson D, Rosen W, Stucker B (2010b) Additive manufacturing technologies: 3D printing,
711 rapid prototyping, and direct digital manufacturing. Springer, New York. Section 5.3.1.
712 ISBN 1493921134
713 Gibson I, Rosen DW, Stucker B (2010c) Photopolymerization processes. In: Additive manufactur-
714 ing technologies. Springer US, Boston, pp 78–119
715 ISO 8062; ISO/TR 14638 Geometrical product specifications (GPS) – dimensional and geometrical
716 tolerances for moulded parts – Part 3: general dimensional and geometrical tolerances and
717 machining allowances for castings
718 Kalpakjian S, Schmid SR (2010) Manufacturing engineering and technology. Pearson. Ch. 17.
719 ISBN 9810681445
720 Kruth JP et al (2004) Selective laser melting of iron-based powder. J Mater Process Technol
721 149(1–3):616–622
722 Mohanty S et al (2017) Achieving dimensional tolerances in metal additive manufacturing via
723 numerical model based process optimization, Dimensional accuracy and surface fin additive
724 manufacturing, KU Leuven
725 Moothanchery M et al (2012) Real-time shrinkage studies in photopolymer lms using holographic
726 interferometry. In: Proceedings of SPIE 8437, 84370I, Brussels, Belgium, 16–19 Apr
727 Ribó MM (2017) 3D printing of bio-inspired surfaces. MSc dissertation, Technical University of
728 Denmark, Department of Mechanical Engineering
729 Saad A et al (2016) Laser powder-bed fusion additive manufacturing: physics of complex melt flow
730 and formation mechanisms of pores, spatter, and denudation zones. Acta Mater 108:36–45.
731 ISSN 1359-6454
732 Technical Paper: Stress crazing on acrylic surfaces and its causes, July 2007, Aristech Acrylics LLC
733 Thompson A, Senin N, Giusca C, Leach R (2017) Topography of selectively laser melted surfaces:
734 a comparison of different measurement methods. CIRP Ann Manuf Technol 66(1):543–546
735 Valente EH (2017) Heat and surface treatment of 3D printed titanium. MSc dissertation, Technical
736 University of Denmark, Department of Mechanical Engineering
In-Line Measurement Technology and
Quality Control 13
Gisela Lanza, Benjamin Haefner, Leonard Schild, Dietrich Berger,
Niclas Eschner, Raphael Wagner, and Marielouise Zaiß
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 400
Long-Term Savings . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 401
High-Quality Requirements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 401
Chapter Structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 402
In-line Measurement Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 402
Designing an In-line Quality Control System . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 404
Preparation of In-line Quality Control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 404
Implementation of Measurement Technology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 408
Implementation of Quality Control Cycles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413
Exemplary Use Cases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 413
Design of an Eddy Current Sensor for In-process Quality Control in
Lightweight Production . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 414
Information Fusion for In-line Quality Control in Lightweight Production . . . . . . . . . . . . . . . . 417
Acoustic Sensor for In-process Quality Control in Additive Manufacturing . . . . . . . . . . . . . . 421
Dynamic Production Control Strategies Based on In-line
Measurement Technology
for High Precision Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 426
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 431
Abstract
In-line quality control is able to provide direct feedback with regard to quality
deviations in production systems. Thus, it is a crucial enabler to guarantee high-
quality standards and prohibit waste within production. As an enabler for this, in-
line measurement technology is to be implemented and applied in the production
system in an effective manner. In this chapter, different types of in-line measure-
ment technology are explained and structured. Based on this, a framework is
introduced to systematically implement in-line metrology in production systems
G. Lanza (*) · B. Haefner · L. Schild · D. Berger · N. Eschner · R. Wagner · M. Zaiß

wbk Institute of Production Science, Karlsruhe Institute of Technology (KIT), Karlsruhe, Germany
e-mail: Gisela.Lanza@kit.edu

https://doi.org/10.1007/978-981-10-4938-5_14
400 G. Lanza et al.
in order to realize suitable quality control cycles. Finally, the application of the
framework is demonstrated in various industrial use cases.
Keywords
In-line measurement technology · Quality control · Quality control cycles ·
Measurement technology · Quality value stream mapping · Measurement
uncertainty · Lightweight production · Additive manufacturing · Precision
engineering · Matching strategies
Introduction
Quality control is an important part of a production system. It helps to guarantee a

reliable and efficient production. Its purpose is not only focused on quality inspec-
tion and documentation but also on giving feedback to production control (Fig. 1)
(Imkamp et al. 2012; Schmitt and Damm 2008).
A quality control system is set up by means of quality control cycles. As depicted
in Fig. 1, quality control consists of two main features: measurement technology and
quality management measures. The measurement technology is used to determine
the current production quality. For the application of measurement technology, two
elements are required: sensors and software-based data analysis tools. These return
the measured production quality. While included data analysis tools are often
supplied with commercial measurement technology hardware, they have to be
specifically developed for new measurement technology designs. By comparing
the desired quality with the measured quality, a potential gap of quality is detected.
Fig. 1 Systematic model of in-line quality control cycles in production systems

13 In-Line Measurement Technology and Quality Control 401
The detected lack of quality is used as input for the subsequent quality management
evaluation to derive instructions for improving the production processes in produc-
tion system. Thus, the quality management process closes the control cycle to ensure
consistent production quality in the production system.
The quality management process works best if it is able to support the production
processes by providing a continuous supply of instructions and feedback. This can only
be achieved if the quality management process receives a continuous flow of informa-
tion on the gap of quality, as well. The requirement can be fulfilled best by integrating
measurement technology in the production system, which produces an ongoing stream
of quality measurement data. This type of measurement technology is called “in-line
measurement technology.” In-line measurement technology is usually more resilient to
environmental challenges and has shorter measuring times than laboratory-based mea-
surement equipment. On the other hand, in-line technology is often less accurate and
needs complex software to handle the continuous flow of measurement information.
The implementation of in-line measurement technologies causes additional effort
and costs compared to end-of-line inspections. The following reasons are often
presented when switching from off-line technologies to in-line measurements
(Schmitt and Damm 2008):
1. Long-term savings
2. High-quality requirements
Long-Term Savings
Although repeated in-line measurements cause additional costs in comparison to

end-of-line inspections, they help to achieve long-term savings. Costs caused by
production errors increase with every value-adding step that follows after an error
occurs. The value is added to a part that is going to be sorted out during the end-of-
line inspection. Therefore, the costs of repeated in-line inspections are often lower
than the accumulated costs of the following production steps. Figure 2 shows this
effect. In-line measurements allow to spot production errors early on and prohibit
any further work on faulty workpieces. Monitoring the production with in-line
measurements is therefore especially important for incapable and instable production
processes as they have a high probability of an early production step causing an
error. Using in-line inspection during the production ensures that errors are spotted
as soon as they occur and rework may be done shortly afterward. Summarizing, in-
line quality control helps to save costs by detecting production errors early in the
production process (Schmitt and Damm 2008).
High-Quality Requirements
Achieving a high standard of quality is another reason for using in-line measurement
technology. This is especially true if the product quality is directly linked to the
402 G. Lanza et al.
Fig. 2 Cost increase of end-of-line defect detection in comparison to in-line inspections
company’s brand image as it is the case with, e.g., machine tools producers.
Reliability is one of the most important features of a company’s image. If products
of low quality are delivered to the customer, the company would quickly lose
valuable market shares. To ensure produced goods have 0% defects, 100% need to
be inspected. In-line quality control is able to provide such a 100% inspection.
Companies are only able to deliver the quality their customers expect by using in-
line quality control. This type of in-line inspection is used with automotive suppliers.
Modern emission rules require a very high operation pressure for injectors. This
leads to tight tolerances, which can only be guaranteed by in-line inspections. These
enable faulty parts to be detected and to be phased out (Imkamp et al. 2012).
Chapter Structure
This chapter is structured into three parts. In section “In-line Measurement Technol-
ogy,” different types of in-line measurement technology are defined and structured.
Afterward, section “Designing an In-line Quality Control System” presents a method
for designing adequate quality systems based on in-line measurement technology
and suitable quality control cycles. The chapter closes with practical examples
providing insight on how to apply the presented method to achieve effective in-
line quality control in industrial production systems.
In-line Measurement Technology
This section aims at clarifying the term “in-line measurement technology.” In-line
measurement technology is part of the production system itself and, therefore, is able
to provide a continuous stream of information on the quality of the production
system.
Fig. 3 Structure of measurement technology for the integration in production systems
In general, measurement technologies can be differentiated by how they are

integrated into the production system. Based on this, different subcategories of
integration levels can be separated (Fig. 3). Often, the subcategory on the right
side is used synonymously for the name of the whole integration level (Schmitt and
Damm 2008; Damm 2013). Figure 3 differentiates between the subcategories to
stress that both subcategories are part of the same integration level.
The first level of Fig. 3 distinguishes between “on-shopfloor” and “off-
shopfloor.” “On-shopfloor” defines a measurement on the production shopfloor,
whereas “off-shopfloor” means a measurement in a laboratory. The second level
distinguishes between “in-line” and “off-line.” The term “in-line” describes mea-
surement technology, which is included into the production line. In-line as well as
off-line technology is located on the shopfloor. Typical examples for in-line mea-
surement technology are automated optical quality inspection systems in the auto-
motive industry. For these, cameras are mounted on robots which allow for an
automated inspection of every part in the line. An example for off-line (sometimes
referred to as “at-line”) measurement technologies is special coordinate measure-
ment machines (CMMs). Most CMMs are placed in a laboratory and allow for
highly precise measurements. Some measurement technology manufacturers offer
special versions of these machines designed for the harsh environment on the
shopfloor. They are more resilient to temperature changes and are not easily affected
by dirt or moisture. These off-line machines can be used to take samples as part of
statistical process control (SPC).
Following Fig. 3, in-line measurement technologies can be separated into two
subcategories, “on-machine” and “off-machine.” While on-machine measurement
technology is included in production equipment, off-machine measurement technol-
ogy is included in a separate measurement station integrated in the production line,
such as the automated optical inspection described above (Schmitt and Damm 2008).
These separate measurement stations are used whenever many measurement features
404 G. Lanza et al.
need to be detected and measurement information is not needed for direct control of
the production process. On-machine measurement technology, on the other hand, is
used to directly control a production process by instantly providing feedback within
the equipment (Schmitt et al. 2011; Damm 2013). An example for an on-machine
inspection is the measurement of a workpiece while it is still mounted in a machine
tool. For these measurements, highly precise optical sensors can be used.
On-machine technology is classified into two subcategories, which are “off-
process” and “in-process.” While in-process measurement technologies record data
during the execution of the production process, off-process measurement technology
starts recording data while the workpiece is still in the machine, but before or after
the production process has finished. Both techniques have different advantages: In-
process measurements can be used for direct control of the current production
process while it takes place. However, in situ measurements allow for more complex
measurement technologies to be used, for example, when a workpiece is still
mounted in the machine tool. As an example for in-process measurements, acoustic
measurements are used to evaluate the quality of welding seams during the welding
process. If a lack of quality is detected, the welding parameters such as the welding
current can be adjusted accordingly. The results from both types of measurement
technology can be used to give immediate feedback to decide if rework is needed
while a workpiece is still mounted in the machine (Schmitt and Damm 2008).
All measurement technologies featured in the category “in-line measurement
technology” are able to provide continuous information on the production quality.
Therefore, they are most suitable to be used in in-line quality control cycles which
enable very high levels of quality (Schmitt and Damm 2008; Damm 2013).
Designing an In-line Quality Control System
Designing an in-line quality control system as presented in Fig. 4 requires an

application-specific implementation of measurement technology as well as the
specification of quality control cycles. In the following, a generic method is intro-
duced which can be applied to arbitrary industrial use cases. An overview of the
method is depicted in Fig. 4 and consists of three major steps:
1. Preparation for in-line quality control

2. Implementation of measurement technology
3. Implementation of quality control cycles
Preparation of In-line Quality Control
The first step of designing an in-line quality control system is the task of preparation.
As a result of this, critical measurement features, which are subject to in-line quality
control, are identified (compare Fig. 4). The preparation step starts with an investi-
gation on which production processes of a critical product need to be inspected in-
Fig. 4 Method for designing an in-line quality control system
line. Two sub-steps “product analysis” and “process analysis” are used to accom-
plish this task. Afterward, the critical measurement features are derived from the
chosen product and process characteristics. Product analysis starts with the product
itself and derives critical quality features, whereas the second method, process
analysis, analyzes the value stream with existing data to identify processes in need
of in-line quality control. Product and process analysis can be applied in parallel, but
they work independently from each other as well.
Product Analysis
A product analysis is a method to identify those features of the product that are most
critical to the quality of the product. These functionalities are used to derive suitable
measurement features for the critical production processes. A product analysis can
be conducted in two steps that are presented in the following:
1. Structure analysis
2. Defects and risk analysis
Structure Analysis
The structure analysis is the first step of the product analysis. It aims at collecting and
ordering information on the specifications and production processes of a product. To
do so, all components of the product are analyzed in detail. Each of them allows for a
406 G. Lanza et al.
specific functionality and is characterized by specifications, such as its weight or

length. Additionally, all related production processes are collected. However, espe-
cially with products that have not yet reached series production, different production
processes are possible. Finishing the structure analysis, all components and their
functionalities, as well as the related specifications and production processes, are
aligned in a morphological box (Koelmel 2016; Koelmel et al. 2014).
Defects and Risk Analysis

Based on the morphological box from the structure analysis, a defects and risk
analysis is performed to identify crucial manufacturing processes. The defects and
risk analysis starts with deriving possible defects for all production processes
collected in the morphological box. In a second step, a risk priority number (RPN)
is calculated for every defect. Critical defects and corresponding production steps are
identified by sorting all defects according to their RPNs. The collection of these
defects results in a matrix, consisting of the processes, the defects, and the RPNs.
This is called a “defects-process-matrix” (DPM). It is the final outcome of the error
and risk analysis.
The relevance, probability of occurrence, and the probability of detection of a
defect need to be estimated by assigning a number between 1 and 10, in order to
calculate the RPN of the defect. For example, assigning 1 to the probability of
occurrence of a certain defect means that it never occurs. On the other hand, if the
probability of detection of a defect is assigned to 10, it means that the error is
impossible to detect. The risk priority number is derived by multiplying the three
values of the relevance, the probability of occurrence, and the probability of detec-
tion of the defect. Thus, the RPN value varies between 1 and 1000. Calculating the
RPN is a vital part of a Failure Mode and Effects Analysis (Koelmel 2016; Koelmel
et al. 2014).
Process Analysis
Besides the previously presented product analysis, a process analysis can be used to
deduce production processes in need for in-line quality control. The process analysis
can be executed based on a method called “quality value stream mapping” (Haefner
et al. 2014). This method is used to analyze an existing production system by its so-
called quality value stream. It allows to identify processes that are especially relevant
to the product quality.
Quality value stream mapping is based on a value stream analysis, but quality-
relevant characteristics are added (see Fig. 5). This includes an additional symbol for
quality inspection process steps. These “Q-steps” are shown in the quality stream
map and include quality-relevant information from inspections. The number of parts
inspected, the scope of inspection, defects per million opportunities (DPMO), the
process capability based on a Measurement System Analysis (MSA), the number of
detected defects (ND), or the resulting process capability index (PCI) may be
provided. The PCI may be highlighted in the quality value stream map with suitable
colors. By doing so, it is immediately clear which processes are capable. Addition-
ally, a risk priority number is calculated for each quality feature and linked to the
No.1 No.2 No.3

Production Production Quality
process 1 process 2 inspection 1
Causes of defect 1 : Causes of defect 2: Amount of parts inspected 100.000

- cause 1 - cause 1 Scope of inspection 100%
- cause 2 - cause 2 DPMO 1740
Causes of defect 2: Causes of defect 3:
- cause 1 - cause 1 Inspection characteristics MSA ND PCI
- cause 2 - cause 2 Inspection characteristic 1 O.K. 20 1,18
Inspection characteristic 2 O.K. 500 0,86
Inspection characteristic 2 O.K. 2 1,37
RPN 1 RPN 2.1 RPN 2.2 RPN 3
Fig. 5 Extract of a quality value stream including process data, RPNs, and causes of defects
(Haefner et al. 2014)
related production process. High and very high RPNs can be highlighted by using
colors. This helps to stress which production processes are especially critical. The
quality value stream mapping is finished by adding information about causes of
defects to the process boxes of critical processes.
A quality value stream map depicts the current status of a chain of production
processes from a quality point of view. It can form a base to decide which processes
seem to be especially in need for in-line inspection. These processes are typically
recognizable by a low process capability index or a high RPN. As soon as specific
processes are identified, measurement features need to be derived. The quality value
stream map can help with this step, as it shows causes for errors. These causes can
then be used to derive measurement features according to the procedure from the
next section (Haefner et al. 2014).
Derivation of Measurement Features

The next step of the in-line process is the derivation of measurement features to be
checked in-line. Critical process steps are known from the product or process
analysis. They are characterized by a high RPN. As these processes are critical for
the quality of the finished product, their output needs to be inspected. In case of a
process analysis, the quality inspections, which helped to unveil critical process
steps, need to be transferred to normal serial production. For a product analysis,
however, suitable measurement features need to be derived in order to assess the
performance of those critical processes.
Norms, guidelines, publications, and experience help to define measurement
features for these most critical process steps. Considering economic and organiza-
tional aspects, it is not recommended to measure all possible features. Only features
that represent the quality of the product should be analyzed. Quality is characterized
by a product fulfilling its functionalities. Therefore, the functionalities of the product
408 G. Lanza et al.
Fig. 6 Method for implementing measurement technology into production environment
need to be analyzed in order to select quality-relevant measurement features. This is

a highly application-specific step and, therefore, not presented in detail in this
chapter. For further advice on how to select appropriate measurement features, see
Koelmel (2016).
Implementation of Measurement Technology
After all preparation steps have been completed, suitable measurement technology is
implemented into the production line. This implementation step consists of four
main parts (compare Fig. 6). First, a measurement technology needs to be chosen
that is able to measure the features which have been derived in the course of the
previous preparation step. Second, the accuracy of the measurement technology
needs to be assessed in a laboratory. In parallel, the measurement technology is
physically integrated into the production line. If either of these steps fail, another
measurement technology needs to be chosen. Last, a final investigation is done in
order to judge if the measurement technology is fit to measure with the specified
accuracy in its production environment.
Selection of Measurement Technology

There are two possibilities to choose a measurement technology that is capable of
inspecting the measurement features in question. The first one is to choose the
measurement technology from a so-called measurement technology catalogue,
while the second option is to invent a new technology. Sometimes the development
of new measurement technologies for in-line measurements is inevitable, especially
for immature technologies. Every newly invented technology should be collected in
a measurement technology catalogue. It accelerates the process of establishing a
similar technology in future. Furthermore, it offers an overview of possibilities. How
to set up a measurement technology catalogue and how to use it to choose a
measurement technology is described in the following.
The process of inventing a new technology is highly task-specific. Therefore,
it is not described in detail in this section. In general, a physical principle or a
sensing technology needs to be chosen according to the main requirements
(e.g., measurement feature, measuring time, etc.). Afterward, a data analysis needs to
be implemented in order to evaluate the raw signal and gain useful measurement
values. Further details on a process for designing measurement equipment are
provided in Koelmel (2016).
Measurement features are the result of the preparation step for the in-line process
(compare Fig. 4 and section “Process Analysis”). These features must allow for
inspections to determine the current product quality of the production process.
Relevant measurement technologies should be stored in a measurement technology
catalogue to collect all technologies, which are able to detect a certain feature type.
Building such a catalogue is a vital asset for implementing in-line quality control. As
it is a method to store ideas and implemented solutions, it helps responding to
changes in the production environment quickly. Even if the development of a
measurement technology does not succeed, information on its potential use cases
and advantages are not lost. Therefore, all types of developments should be listed in
the catalogue including their degree of maturity. The degree of maturity helps to
clarify, under which circumstances it may be worth to pick up an old idea again.
Especially the advances in computer power make it possible to deploy technologies
that have not found their way into series production, yet.
A measurement technology catalogue is ordered by groups of defects (e.g.,
geometry), measurement principles (e.g., optical methods), and specific measure-
ment technologies (e.g., laser interferometry). In terms of hierarchy, measurement
technologies are assigned to measurement principles, which in turn are assigned to
the groups of defects. Thus, it is possible to quickly obtain a variety of measurement
technologies out of the catalogue for a certain group of defects. Additionally, only
the measurement technologies are described in detail with all their features to make
the catalogue easily accessible (Koelmel 2016).
Every measurement technology is described in detail by listing the following
features:
• Description of the application

• Description of the measurement equipment and its degree of maturity
• Description of the measurement result
• Execution of the measurement
• Requirements for the measurement objects
• Requirements for the surrounding and the user
• Measuring time
• Costs
Additionally, characteristics of related sensor designs are assigned to the descrip-

tion of the measurement technologies. An extensive and clearly structured catalogue,
which is of help in later projects as well, can be achieved this way. Another benefit of
a well-structured measurement technology catalogue is that, even if no suitable
measurement equipment can be found in the catalogue, it allows the user to gather
ideas on how to develop new in-line measurement technologies using the listed
physical principles.
410 G. Lanza et al.
The degree of maturity is a measure for the development status of a certain type of
measurement technology. It is dependent on a specific application. For example,
even if a measurement technology is not used in the field of interest, it might be used
in another field on a regular basis. This means that this technology has a high degree
of maturity in the second field but not in the first one. It takes two steps to evaluate
the degree of maturity. The first step is to characterize its technology readiness level
(TRL) according to Mankins (2009). This technique is based on a checklist using
criteria to distinguish between the three stages of research and development, proto-
type, and maturity phase. In the second step, the integration readiness level (IRL) by
La Croix (2008) is calculated. Finally, the degree of maturity can be evaluated and
returned to the measurement technology catalogue.
Two steps have to be followed in order to choose an application-appropriate
measurement technology using a measurement technology catalogue. First, by
comparing the various measurement principles with measurement features that
need to be detected, methods that fit best can be identified. Afterward, a compatible
measurement technology has to be selected from the chosen measurement technol-
ogies. This step is necessary because it is impossible to find one measurement
principle which works best for all features.
Criteria for exclusion need to be checked in order to find the best-fitting mea-
surement technology. These criteria are the size of the measured object, the resolu-
tion, the mass, and the measuring time. Then, the surrounding, the software, the staff,
and the infrastructure need to be examined as well. To express the benefits of
different measurement technologies, e.g., the pairwise comparison method can be
used (Koelmel 2016).
Referencing of Measurement Technology in Laboratory

After a measurement technology has been chosen, its accuracy needs to be deter-
mined. The measurement uncertainty is a quantitative measure for its performance.
According to the “Guide to the Expression of Uncertainty in Measurement” (GUM),
the measurement uncertainty is a parameter that is “associated with the result of a
measurement, that characterizes the dispersion of the values that could reasonably be
attributed to the measurand” (JCGM 100:2008). By determining the measurement
uncertainty in a laboratory under ideal environmental conditions, it can be verified if
the chosen technology is suitable for the measurement task by verifying its lowest
possible measurement uncertainty.
The goal of the GUM is to provide a consistent procedure to determine the
measurement uncertainty by evaluating all influences on a measurement. Calculating
the measurement uncertainty according to the GUM works by modelling the mea-
surement task with a mathematical model. This model consists of all possible
sources of measurement uncertainty and helps to consider random as well as
systematic deviations of the measured value compared to the true value of the
measurement feature. All elements of the measurement model need to be assessed
and combined to form the task-specific measurement uncertainty. There are two
possibilities (Type A and Type B) to estimate the influences of the model’s elements.
The first is to calculate the measurement uncertainty using a statistical analysis (Type
A). The second one is to use other means (Type B). This may include a priori
knowledge like calibration certificates. According to GUM Supplement 1 (JCGM
101:2008), it is also possible to use Monte Carlo-based simulations in order to
determine the measurement uncertainty.
For the complex measurement task of geometric measurements, instead of setting
up a measurement model as described above, it is possible to execute special
referencing experiments (DIN EN ISO 15530-3 2012). Especially with complex
measurement technologies, which are intended to be used outside a measurement
laboratory, using referencing experiments is a common and viable option. In the
course of such referencing measurements, a sufficiently similar reference artifact or
specimen is measured with a measurement technology, which is known to have a
very low uncertainty. Afterward, the object is measured and calibrated with a
measurement technology for which the measurement uncertainty is to be assessed.
The measured systematic and random deviations from the calibrated values are used
to calculate the measurement uncertainty. According to DIN 15530-3 (DIN EN ISO
15530-3 2012), the following formula is used:
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
U ¼k u2cal þ u2p þ u2w þ u2b :
In the formula, U is the expanded measurement uncertainty, k is the expansion

factor, ucal is the standard uncertainty related to the calibrated object, up is the
standard uncertainty resulting from the measurement process, uw is the standard
uncertainty resulting from the workpiece, and ub is the standard uncertainty of the
bias. Together, these factors depict all uncertainties in a measuring process and allow
for the expanded measurement uncertainty to be calculated. In this context, the range
of the expanded measurement uncertainty covers the true value of the measurement
feature with a probability of 95%.
It may happen that the uncertainty assessment has to be repeated several times. If
a new technology is developed or an existing technology is altered for a new
application, each iteration needs to be assessed until a satisfactory outcome in
terms of the measurement uncertainty has been achieved. Results obtained at this
stage should be recorded in the measurement technology catalogue to provide
guidance for further design iterations.
Integration of Measurement Technology in Production Environment

According to Fig. 6, the measurement technology needs to be integrated into the
production process. This means that the measurement technology needs to be
adapted to its new environment on the shopfloor. This includes all design steps
regarding the environment of the process on the shopfloor and those that are
necessary to locate the technology on the shopfloor. This takes two basic steps:
First, the technology has to be connected to its new surrounding by paying
attention to means of power supply and data connection. Second, the environmen-
tal conditions need to be considered. If these conflict with the requirements of the
measurement technology, suitable protection needs to be designed. For example,
412 G. Lanza et al.
cameras or other sensitive measurement equipment need to be protected by a

suitable housing, if they are to be used in a hazardous surrounding like a body
shop where sparks are emitted.
If the integration step has an effect on the measurement uncertainty of the
technology, another loop of uncertainty evaluation in the laboratory, as described
before in section “Referencing of Measurement Technology in Laboratory,” is
necessary. This step must not be confused with the evaluation of the measurement
uncertainty in the production environment described in the next section. During the
integration step, the basic requirements of the measurement technology are checked.
If changes to the hardware are necessary, this may have an effect on the measurement
uncertainty. In such a case, the measurement uncertainty needs to be re-evaluated in
order to represent the changes to the hardware. This is the case, e.g., if the measure-
ment uncertainty of a measurement technology is known from earlier tests, while for
a new project, the measurement hardware needs to be changed in order to be
integrated into the production environment. Extending the previous example, if a
camera is protected from flying sparks by an acrylic glass plate, its accuracy is
affected. The acrylic glass plate modifies the vison of the camera, and therefore,
another uncertainty evaluation in the laboratory needs to be conducted – including
the acrylic glass plate. If the uncertainty assessment calls for design changes – e.g., a
bigger power supply to use a camera of higher resolution – the integration step needs
to be redone to provide ample power supply. Summarizing, the uncertainty evalua-
tion in the laboratory step and the integration in the production process step are
iterative steps that have to be repeated if necessary.
Referencing of Measurement Technology in Production Environment

As soon as the required accuracy is proven in the laboratory and the measurement
technology is successfully integrated into the production environment, final checks
need to be done. The previous steps are not sufficient to prove if a measurement
technology is sufficient for in-line use. To do so, the technology’s measurement
uncertainty of the measurement technology needs to be evaluated while it is used in-
line. Results of these final experiments should be returned to the measurement
technology catalogue in order to record that a measurement technology is sufficient
for in-line use and has therefore reached a new degree of maturity.
To deduce the measurement uncertainty experimentally while being used in-line,
two options are possible. Both approaches are based on the experimental evaluation
of the measurement uncertainty according to DIN 15530-3 (DIN EN ISO 15530-3
2012). The first option is to use another highly precise measurement equipment in-
line, to execute reference measurements. This reference system does not have to
fulfill all in-line relevant criteria, which apply to the chosen in-line measurement
technology. For example, the referencing technology may be too expensive for in-
line use, or it may not be capable of measuring quickly enough to keep up with the
production cycle time. However, as DIN 15530-3 (DIN EN ISO 15530-3 2012) calls
for the similarity of the measurement processes when it comes to the evaluation of
the measurement uncertainty, the referencing measurement technology needs to
measure under the same production conditions as the tested measurement
technology. Another possibility to obtain the measurement uncertainty of the in-line

measurement technology is to use a calibrated reference object on the shopfloor. It
must be possible to measure this reference object in the actual surrounding on the
shopfloor. This includes, for example, that a reference object needs to be fastened
using the same clamps as normal workpieces, the same lighting must be used, the
same working distance must be chosen, etc. By comparing the measured and the
calibration values, the measurement uncertainty in the production environment can
be evaluated.
Implementation of Quality Control Cycles
Setting up quality control cycles is the last step of the method for designing an in-line
quality control system of a certain measurement feature (see Fig. 4). A quality
control cycle allows to use the measurement data gathered by the in-line measure-
ment technology in order to control production processes in a feedback loop
(compare Fig. 1).
The complexity of implementing quality control cycles can be very different in
various tasks. A rather simple task is to decide which parts are of sufficient quality
and which have to be phased out. However, more sophisticated strategies that
dynamically change production parameters may also be used. The chosen strategy
is highly dependent on the measurement data collected, the measurement uncertainty
of the measurement technology in use, and the integration of the measurement
technology in the production environment. These and the degree of knowledge
about the production processes account for the data basis available and its reliability
as well as the possibility of control actions. For example, if the data basis was good
enough, measurement data could be used to identify what type of rework is needed,
instead of simply phasing parts out. To take things even further, measurement data
may be interpreted in order to use it to adjust process parameters. Information
generated this way can be used, e.g., to pursue an adaptive change of machine
tools because of tool wear.
The chosen quality control cycles need to be consistent with a company’s quality
management system. Therefore, it is specific to each use case. An example on how to
design sophisticated quality control cycles is shown in section “Dynamic Production
Control Strategies Based on In-line Measurement Technology for High Precision
Products.” This example shows how to implement dynamic production control
strategies for high precision parts.
Exemplary Use Cases
The following section presents use cases which show how to apply the previously
presented steps of the method for designing in-line quality control systems. The use
cases show how to set up different measurement technologies in order to measure in-
line. Each aims at placing emphasis on a different step of the process presented
414 G. Lanza et al.
before. Not all use cases use every method that was described in the sections before.
Nevertheless, they show how the steps of the method can be applied in practice. As
such, the use cases help to underline that different boundary conditions call for
different strategies to set up in-line measurement technologies.
Design of an Eddy Current Sensor for In-process Quality Control in

Lightweight Production
Carbon fiber reinforced plastics (CFRP) have gained a lot of importance in industrial
manufacturing of structure components. This is due to the fact that the application of
CFRP allows a mass reduction of 20–50% compared to conventional materials such
as aluminum, steel, and plastic components. Especially automotive and aeronautical
applications benefit from the reduction of moved masses because the emission of
harmful exhaust gases is bound to them. In order to exploit the ecological and
technological benefits of lightweight designs, their dissemination in products is to
be increased globally, which is only possible by high degrees of automation in
processing and sufficient quality control. Therefore, production processes are to be
analyzed, and quality features along the value chain must be identified in order to
derive a suitable in-line measurement technology.
In order to prepare the in-process quality control of CFRP components, the
product as well as production processes are analyzed, and the relevant measurement
features are selected according to the method described in section “Designing an In-
line Quality Control System.”
Resin transfer molding (RTM) is a production process for CFRP components that
offers the possibility of highly automated handling and process steps and therefore
meets the major requirements of a large volume production. The process consists of
five major steps that can be seen in Fig. 7.
In the first step, continuous filament mats are cut and stacked according to the
layer structure that is defined in the design process. The orientation of the fibers is
very important with regard to the laminate structures because more forces can be
transmitted along the fibers in comparison to relatively deviating force directions.
Therefore, the subsequent preforming, where the stacked fiber mats are forged to
Fig. 7 General production steps of RTM process chain

complex 3D forms in near net shape, is a quality-critical process. During forming,

fiber mats can misalign due to the process strategy, or pleats in the material may
occur due the applied forces. Also, each material type has individual mechanical
properties, leading to different effects during preforming. Different types of pre-
forming can be applied, in order to reduce negative effects. If textile defects are not
noticed after preforming, additional value is added to the part even though its quality
is already diminished.
After the part is preformed, the semifinished part is inserted in infiltration cavities,
and a resin/hardener mixture is injected under high pressure. It is then cured under
controlled temperature. During these steps, the preform quality does not change
significantly, and occurring defects such as delamination are bound to the matrix.
In conclusion, it can be stated that it is not sufficient to detect textile defects in
final parts in order to efficiently control the quality during the RTM process chain.
Instead, the quality characterization of preforms can be used to avoid unnecessary
value-added steps by means of in-line measurements. Orth defines basic measure-
ment tasks for the quality control of CFRP components in order to guarantee a high
part quality. For the identification of the texture type, the main task is to avoid
systematic errors such as wrongly used raw material. Besides the identification of the
material type, there are also different bond structures that can be swapped
unintentionally and lead to accordingly different mechanical properties. Further-
more, local anomalies such as defects in the texture, like fiber or tow gaps, need to be
detected as well. Therefore, the main measurement features in carbon fiber preforms
are the tow orientations throughout the layer structure.
Besides the described measurement features, there are further restrictions that
need to be taken in consideration for the implementation of measurement technology
in CFRP process chains. In order to increase the efficiency of the production, a high
testing frequency is needed, as the processes do not fulfill the high requirements of
the automotive industry with regard to the process capability and stability. Thus,
sample-based inspections are not sufficient. Instead, short measurement times need
to be realized. In this case, to effectively detect fiber delamination during the
preforming process, in-line measurement is necessary. Thus, potential measurement
methods for this are analyzed.
Radiographic methods like industrial computed tomography are restricted in their
applicability as in-line testing method due to long measurement times and the
necessity of part handling. Optical methods only have a limited applicability for
inspection purposes in CFRP production because the entire layer structure must be
characterized in order to guarantee the quality of the part, as the entire structure is
responsible for the load-bearing capacity. Also, auxiliary materials that are necessary
for testing purposes must be avoided to not influence the tested part negatively. This
is why coupling media that are mostly used for ultrasonic testing are not an option
for the inspection of dry semifinished textiles. The applicability of thermographic
methods is bound to additional measurement stations in the production line, limiting
their applicability as process-integrated testing methods. Electromagnetic methods
offer the possibility to quickly determine fiber tow orientations due to their aniso-
tropic electric conductivity.
416 G. Lanza et al.
For this purpose, electric coil arrays can be integrated inside preforming tools for
the assessment of the main quality characteristics. The elementary sensor consists of
various emitting coils and one receiving coil that are used to measure the fiber tow
orientations by analyzing anisotropically induced voltages. Due to the low process
forces of the preforming process, there is the possibility for integration of the sensor
in preforming tools, which is bound to their static application. However, a static
application demands the identification of quality-critical part zones. In cases where
components are part of assemblies, dimensional tolerances must be respected within
the joined parts. Furthermore, the application and insertion of additional load
applying insert components in carbon fiber reinforced components requires the
quality control of mechanical interfaces. After designing the components, numerical
FEM simulations offer the possibility to show the load path when applying forces on
the hybrid structure.
Figure 8 shows the principle of sensor integration in preforming tools and the data
that can be gained during the measurement.
The gained data consists of a discrete amount of measured voltages,
evenly distributed over the sensor circumference. Interpolation methods such
as spline interpolation of discrete Fourier transformation can be used to
reconstruct the signal and determine fiber tow orientations based on the ani-
sotropic conductivity. In order to describe the capability of this measurement
system, the estimation of measurement uncertainty can be conducted accor-
ding to [VDA Band 5]. As it is connected to the “Guide to the Expression of
Measurement Uncertainty,” the measurement system uncertainty can be sub-
divided in single uncertainty components, represented as standard deviation
values that can be estimated experimentally. For this, suitable CFRP specimen
was developed representing the relevant measurement tasks for the fiber tow
orientations in a realistic manner. According to the following equation, there are
five main components that can be considered in combination to derive the
expanded measurement uncertainty:
Fig. 8 Measurement principle of in-process eddy current sensor

pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
U ¼k uRE 2 þ uCAL 2 þ uEVR 2 þ uBI 2 þ uLIN 2
The uncertainty that is bound to the resolution of the measurement system is

represented by uRE and can be derived from the datasheet of the applied measure-
ment system. uCAL represents the uncertainty resulting from the calibration process.
For the reference of the eddy current sensor, an industrial computed tomography
system was applied. The data that was gained was analyzed by using a fiber analysis
toolbox by Volume Graphics. It allows the evaluation of single fiber orientations
throughout the entire specimen. uEVR represents the deviations resulting from repeat-
edly conducted measurement task. uBI are systematic deviations compared to the
reference value of the reference normal. The uncertainty resulting from a lacking
linearity is represented by uLIN and is gained by at least three reference parts, whereas
the nominal value is varied systematically. The deviation from the linear expected
correlation is transferred to the respective standard deviation value. With an expan-
sion factor of k = 2, a coverage probability of 95% is given for the expanded
measurement uncertainty.
During preforming, different types of defects can occur that are bound to various
process properties. Therefore, their detection and classification in defect groups such
as misorientations, tow gaps, or wrinkles allows the optimization of single process
parameters and the implementation of suitable quality control cycles. Systematic
fiber orientations may be caused by defective pick-and-place processes through
robots in the process chain, whereas wrinkles may be caused by faulty forming
parameters. Once the occurrence of defects can be traced to their cause, methods of
automated signal characterization can improve single process steps within the
production process. Especially methods of machine learning such as artificial neural
networks enable the automated interpretation of signal characteristics by identifying
patterns of scrap parts.
Information Fusion for In-line Quality Control in Lightweight

Production
Sheet molding compound (SMC) is a discontinuously reinforced glass fiber com-

posite material and has been known for 60 years (Nicolais 2011). It is used in various
areas, such as building construction, civil engineering, and the automotive industry.
In the latter it is used for tailgate lids and other coverings.
Regarding the preparation for in-line quality control, the production process
chain of SMC consists of several steps (compare Fig. 9). In the first step, a paste of
all necessary raw materials such as resin, hardener, and additives are mixed and
used up by squeegee boxes on two films. Afterward, discontinuously cut glass
fibers are inserted between these two foils, which are then merged together. Air
pockets are forced out through a calendering zone and the fibers are completely
impregnated. Then the material is rolled up and is available as a semifinished
product, which in the following is matured in a cabinet until it has reached the right
consistency.
418 G. Lanza et al.
Fig. 9 Manufacturing process of sheet molding compound (Zaiß et al. 2017)
After the maturing process, the necessary material is stacked and pressed into the
desired shape. By using SMC, considerable weight savings can be achieved. Today,
however, hardly any structural components are produced thusly because the mechan-
ical properties of these fiber-reinforced plastics are not sufficient.
In order to increase the mechanical properties, the discontinuous SMC is locally
reinforced with continuous carbon fiber tapes. This way, the mechanical properties
can be improved locally, but also the advantages of fast cycle times and low
reworking are retained. In the SMC, many error patterns can occur leading to a
bad result. These include, for example, resin accumulations, poor fiber impregnation,
or delamination – the most common errors in this product that need to be detected
during the production and the resulting bad parts must be phased out. To do so, in-
line quality control is needed in order to check all parts. The motivation for
measuring in-line, therefore, is a need for quality.
Due to the combination of two materials, which differ in their properties, further
error patterns can occur. These include, for example, gapping and wrong fiber
orientation in the continuous material. This is only a small selection of errors,
which can occur very often and have a great effect on the mechanical properties.
In order to avoid unnecessary, value-adding measures, the quality of the material
must be checked at an early stage of component manufacturing, so that the defects
can be detected and modified at the stage of the semifinished product. This way in-
line checks prohibit excessive costs. The motivation for in-line quality control is
therefore not only quality needs but also long-term savings. In order to ensure a
smooth operation without delays, the inspection must take place within approx. 90 s.
This is the average cycle time of the press. In order to ensure complete verification,
quality control must be carried out in-line.
In this example, the most common errors are first analyzed. The semifinished
product is considered faultless, when it contains, i.e., no air bubbles or incomplete
impregnated fibers. As a measurement technology catalogue is not at hand for
Fig. 10 Cured SMC specimen and associated cloud of points
the given tasks, a new sensor needs to be developed. To choose the appropriate
attributes of the sensor, the steps described in section “Designing an In-line
Quality Control System” are used. The most common defects to be investigated
are delamination, contour and shape deviations, foreign bodies, fiber orientation,
and folds in the material. These five defects can be subdivided into external and
internal defects.
Based on this analysis, suitable in-line measurement technology is implemented.
There is a variety of measurement methods for these two defect classes. For internal
defects, computed tomography, ultrasound, or thermography is recommended. For
external defects, the laser light section method, strip projection, computed tomog-
raphy, or image processing methods are possible. Other ancillary conditions, such as
cycle time or noncontact methods, limit the possible measurement and testing
methods; therefore the laser light section method is selected for external defects
and thermography for internal defects.
According to section “Referencing of Measurement Technology in Laboratory,”
potentials of the chosen methods need to be assessed, in order to prove that it is
compatible with the measurement tasks. Therefore, the measurement uncertainty
needs to be determined. To do so, both methods are tested individually on test
objects made of SMC semifinished products and hardened material. The optimal
parameters of the individual methods in the laboratory environment are determined
in a first step. The following figure shows a component and the so-called cloud of
points (CoP). The component consists of discontinuous glass fiber SMC with
continuous carbon fiber tapes. To determine the measurement uncertainty, reference
measurements with the CT and tactile measurements are carried out to calibrate the
component. Subsequently, 25 measurements are performed with laser triangulation
(Fig. 10).
Figure 11 shows a pulse phase thermography image of a cured component with
glass fiber SMC and carbon fiber tapes at different frequencies. The tapes and their
orientation are clearly visible, as well as dry spots of the tape or delamination. This
method is therefore very suitable for internal defects and fiber orientation.
The aim of the further work is to combine both methods in a multisensor system
and to generate further information by information fusion (Weckenmann et al. 2009).
The physical structure is illustrated in Fig. 12.
1 Hz 0.025 Hz
420
400 −0.5 400 −0.5

380 380
360 360
340 340
320 320
300 300
280 280
260 260
240 240
220 −1.5 220 −1.5
200 200
180 180
Position [mm]
Position [mm]
160 160
140 140
120 120
100 100
80 80
60 60
40 40
20 20
0 0
0 25 50 75 100 125 150 175 200 225 250 0 25 50 75 100 125 150 175 200 225 250
Position [mm] Position [mm]
G. Lanza et al.
Fig. 11 2D thermography images of SMC (Zaiß et al. 2017)

Fig. 12 Design of the multisensor system. (Picture: Sandra Goettisheim/KIT)
For the implementation of effective quality control cycles, by combining these

systems, it is possible to establish correlations between the deviations. These are
additionally coupled with the mechanical properties so that a prediction can be made
as to the mechanical properties of a component. Due to the in-line measurement, it is
possible to do this in a cycle and to examine semifinished products in order to make
predictions about the later component quality.
Acoustic Sensor for In-process Quality Control in Additive

Manufacturing
Laser beam melting (LBM) is a metal powder-based process, where the powder
gets layer-wise molten to create a 3D-printed workpiece. It is already well-
established for prototyping and becomes more and more important for the series
production in various industries like dental, aircraft, and automotive. Overall LBM
is one of the most promising additive manufacturing methods for further industrial
application and part of current research (Lanza et al. 2017; Eschner et al. 2017a).
The advantage of additive manufacturing and the reason why it is considered for
series productions is the high degree of freedom for the product design. For
instance, by adding a grid structure to the inner part of the gear, it is possible to
apply damping effects and additional cooling effects and apply lightweight design
as illustrated in Fig. 13 (Fanselow et al. 2016). LBM-processed parts often have
complex structures which are sometimes inside of the product and thereby hidden
from visual inspection. This also limits the measurement methods, which will be
discussed later.
422 G. Lanza et al.
Fig. 13 Gear with inner grid structures manufactured by laser beam melting (LBM)
To spread LBM in challenging series productions like aircraft or automotive,

legal and safety regulations have to be met. This is especially crucial for LBM since
it has high potential for lightweight design and thereby is of special interest for
aircrafts. Furthermore, to realize a long-term cost-efficient process, it is necessary to
have a stable and monitored process. This is necessary due to the high rate of reject
and the expensive end-of-line measurements, which are mostly realized coordinate
measuring technology or computed tomography. Currently there are several activi-
ties to realize suitable in-process measurements (Everton et al. 2016).
Analyzing LBM in terms of the preparation for in-line quality control, one of the
first things to note is its overall complexity. More than 50 parameters influence the
product quality as shown in Fig. 14. These parameters can be grouped in four
categories, referring to the laser, powder material, build environment, and coating
process. Due to the high amount of complex interrelations between these parameters,
it is very challenging to generate a high and stable quality of LBM during series
production. Thus, a suitable in-process measurement technology is required to
control the quality of the manufactured product.
The deviation of LBM results in various defects within the manufactured product.
According to Aleshin et al. (2016), the most important defects are cracks/material
discontinuity, porosity, increased degree of stress, and increased roughness. The
most severe are typically cracks/material discontinuity and porosity, since these
affect the fatigue performance (Fig. 15).
The analyses of product and process clearly show that a measurement technology
evaluating the inner structure of the product is needed. Current in-process measure-
ment technologies are limited to the evaluation of the melt pool or surface. Due to the
layer-based process, these surface-oriented methods could also give an insight to the
inner structure. Nevertheless, there are defects inside the product which cannot be
detected with these methods. Further requirements result from the size of the discussed
defects. The size of the pores usually varies between 5 and 500 μm (Everton et al.
2016). As a result, the executed product and process analysis shows the need of a three-
dimensional in-process monitoring of the inner structure of the built product.
13
In-Line Measurement Technology and Quality Control
Fig. 14 Cause and effect diagram for LBM (Spears and Gold 2016)
423
424 G. Lanza et al.
Fig. 15 Defect types in LBM: pores (left), different types of cracks (right) (Aleshin et al. 2016)
Fig. 16 Comparison of nondestructive testing methods (Gevatter and Grünhaupt 2006)
For a suitable implementation of measurement technology, potential measuring

and nondestructive testing methods are analyzed. These are summarized in Fig. 16.
The ones below the red line are only capable of detecting surface and near-surface
defects. As mentioned above, in order to detect the most crucial defects for LBM, a
method covering the entire volume is needed. For this acoustic waves and CT are
suitable.
Comparing the complexity and needed investment of acoustic and x-ray sensors,
acoustic is much less complex and expensive. On the contrary the resolution of
acoustic waves is limited to defect pores in the scale of approximately 50–100 μm.
For many industrial product defects, however, this is still sufficient for pore analysis
in LBM. Thus, for evaluating interior defects, an acoustic sensor was selected.
Next, several acoustic sensor technologies and integration options were analyzed.
For generating and receiving acoustic waves, generally three types of actor/sensor
configurations can be distinguished: piezo, laser, and electromagnetic acoustic trans-
ducer (EMAT). To integrate the sensors, two types of volumes can be identified: below
the build platform and above the powder bed. By systematically comparing the different
possible solutions and rating them against each other according to a structured selection
process, the most promising solution was identified, using structure-born noise (Eschner
et al. 2017b). In this approach, the noise generated by the process itself can be analyzed,
and process characteristics can be evaluated. This approach is already known from laser
welding, where a feature extraction with a fast Fourier transform (FFT) analysis is used
to identify irregularities in the welding process as shown in Fig. 17 (QASS 2018).
A schematic representation of the integration of an acoustic structure-born
noise sensor configuration and the realization in an experimental setup are shown
in Fig. 18. Using the signals from LBM and an adequate feature extraction approach,
a machine learning algorithm can be applied, and product properties can be mea-
sured. These product properties, like the number and size of pores, characterize the
product quality. To train such an algorithm for the data analysis and characterize the
Fig. 17 Fast Fourier transform of acoustic signals in welding process (QASS 2018)
426 G. Lanza et al.
Fig. 18 Schematic overview of structure-born noise sensor integration (left); test bench design for
evaluating the measurement principle (right) (Eschner et al. 2017b)
performance of the measurement system reference, measurements are needed to

evaluate the measurement uncertainty.
To create the amount of data needed to train a machine learning algorithm,
computed tomography measurements are used. With this, it is possible to character-
ize pores in size and overall volume. For the evaluation of specific measurement
uncertainty of the pore measurement through computed tomography, reference
measurements with micrographs are executed. To create test specimen with certain
pore characteristics, LBM parameters like laser intensity are varied in a systematic
way according to a design of experiments (DoE).
Using the described in-process measurement technology, it is possible to identify
products which have flaws that will significantly affect the quality of the product.
With this information, products which are most likely to brake during usage can be
separated and be excluded from further value adding and a respective quality control
cycle can be implemented.
Furthermore, with the help of the above described measurement technology, it will
be possible to actively influence the product properties. For some applications it could
be useful to create pores of a certain size on purpose to optimize its function. This could
be the case when it comes to lightweight or damping structures. The feedback of the
measurement technology allows to tune laser parameters to get the wanted pores.
Dynamic Production Control Strategies Based on In-line

Measurement Technology for High Precision Products
Manufacturing companies in various industries have to meet rising quality require-

ments on behalf of their consumers. Companies need to balance between keeping
production costs at a low level and fulfilling customers’ demands. Meanwhile,

precision requirements increase and reach technological production limits. High
requirements occur especially within the automotive industry, for example, high-
pressure fuel injectors, hydraulic transmission actors, electric motors, and precision
bearings. Herein, precision requirements tend toward narrow ranges, with tolerance
ranges of only a few microns, in order to realize highly accurate functions with an
optimal degree of efficiency (Lanza et al. 2015; Peter and Fleischer 2014; Colledani
and Tolio 2006). Often the economic realization of precision products is a conflict of
objectives in the manufacturing and assembly process. Saving production cost with
high throughput under higher deviations contradicts preventing a high number of
scraped units with more narrow tolerances under higher capabilities. Often even
technological limits of the manufacturing processes are reached, when producing
cutting-edge products. In order to decrease production costs in such challenging
cases, producers have to focus on the value generation and develop intelligent and
dynamic control strategies in the production system so that costs for different types
of waste like scrap, rework, and storage are minimized (Iyama et al. 2013). In-line
measurements of the crucial quality features along the entire process chain in the
production system in real time are the basis for this. The gained data can be used to
implement dynamic production control strategies which individually adapt the
production parameters of manufacturing and assembly processes for every single
product.
As an example a high-pressure fuel injector consists of several precision
subsystems with in total approximately 30–40 components, related to different
disciplines, such as mechanics, mechatronics, and fluid mechanics. The principle
of the indirectly controlled piezo fuel injector is introduced in Reif (2012) (see
Figs. 19 and 20). All main subsystems, piezo actuator, servo valve, throttle plate,
and nozzle set, directly affect the product function, which is a defined fuel
injection volume with a tolerance of less than a milliliter, at a fuel pressure of
more than 220 MPa, within few milliseconds. The product has to meet both
challenges, high precision in volume portioning and high dynamic control, with
up to five injections per cycle. The manufacturing of each subsystem is very
complex and approaches the technological limits. To assure the product function-
ality, very narrow tolerances are assigned to the relevant interdisciplinary
characteristics.
Regarding the product and process analysis for the preparation for in-line quality
control in the production system of injectors, there are many crucial quality features.
Most of them are related to the thickness or horizontal distances of certain compo-
nents, or the diameter of shafts and holes.
Due to the very narrow tolerances of less than 10 nm of all these quality features,
the implementation of measurement technology in-line or even in-process at the
respective stages of the production system is already a common practice with
industrial manufacturers. For these, 100% of product components are measured,
through suitable measurement technology, and tracked by a manufacturing execu-
tion system. The thickness and the distance of components is typically measured by
ID tactile probes. The diameter of rotational features in most cases is measured with
428 G. Lanza et al.
Fig. 19 Principle of high-

pressure fuel injector (Reif
2012)
Fig. 20 Quality control cycles for robust and economic development and production (Wagner et al.
2018)
a pneumatic sensor. Both sensors enable first measurements with a task-specific

measurement uncertainty in the range of 1 μm. Moreover, to guarantee that these
measurement results can be related unambiguously to individual components in the
production system, each component is to be tracked by means of suitable traceability
technology. Depending on the product and process characteristics, typically, e.g.,
RFID technology related to the workpiece carrier or DMC/QR code markers of the
components are used. Recent approaches enable track and trace based on an optical
measurement of the surface structure (Kock 2017).
These extensive in-line measurement capabilities provide the possibility for the
implementation of very advanced quality control cycles in the production system.
These quality control cycles enable new possibilities for intelligent, robust, and cost-
efficient production strategies beyond the manufacturing limits (Wagner et al. 2018)
(see Fig. 20). In the context of “Industrie 4.0,” these are realized by creating a
suitable cyber-physical production system (CPPS).
The main idea of these quality control cycles is based on the individual over-
fulfillment of one component characteristic, which gives the opportunity to com-
pensate the corresponding quality-critical characteristic of a second component. By
harmonizing associated components, more narrow tolerance margins can be realized
in comparison to conventionally assembled components (Mease et al. 2004). This
concept is often also called “pairing” or “matching.” Different quality control cycles
are possible to ensure product quality, including assembly, production, or even
product development and tolerance allocation. The integration of several stages in
the product development process and production offers greater possibilities to ensure
product quality but requires increasing effort preventing scraped products. In Lanza
et al. (2015), a framework of six general categories of production strategies based on
quality control cycles was developed (see Fig. 21).
Fig. 21 Framework of production strategies based on quality control cycles with in-line measure-
ments (Lanza et al. 2015)
430 G. Lanza et al.
Selective assembly, as one category of these production strategies, describes a

method to increase the quality of products while decreasing the production costs by
minimizing errors that occur during the production (Loosen and Funck 2011). Single
components are grouped into multiple tolerance classes based on their individual
deviation from a certain set point and subsequently paired with an appropriate
corresponding component (Colledani et al. 2014). Generally, in the process of
grouping components into classes, information about the exact geometries gets
lost. Both the amount of classes and the tolerance margin define the degree of
information loss (Colledani et al. 2014). A high number of classes and narrow
tolerance margins are preferred for preserving precise measurement data. However,
on the other hand, it is essential to hold enough components for each class.
Therefore, the number of components, the resulting inventory, and overhead costs
restrict the number of tolerance classes. In case of lacking components, active and
passive combinations of non-corresponding classes are applied to avoid downtimes
(Ebrahimi 2014).
Individual assembly is a production strategy to reduce the loss of information
compared to selective assembly, since components are not grouped into tolerance
classes and the exact measurement values saved in combination with the distinct
storage place. However, high organizational and technical requirements are needed
to implement individual assembly (Loosen and Funck 2011).
Using the approach of adaptive manufacturing, as another category of the
production strategies, surplus components for single tolerance classes are prevented
through consumption-oriented production of corresponding matching components.
Therefore, the statistical population of the quality-critical component is recorded,
and the corresponding component is produced under adaption of the nominal value.
The statistical population of adaptively manufactured parts needs to fit to this
population. This ensures that in total every component can be matched and the
required overall tolerance level is met (Matsuura and Shinozaki 2011).
Combinations of selective/individual assembly and adaptive manufacturing seem
to be promising to achieve high rates of good parts as well as low production costs
and production times (Kayasa and Herrmann 2012). Different subcategories of
compositions of selective/individual assembly and adaptive manufacturing offer
various new production strategies (Lanza et al. 2015).
The strategy of individual manufacturing links the ideas of individual assembly
to a minimum storing capacity of provided assembly components. One of the
components is supplied as half-finished product and individually finished after
measuring the quality-critical component. Obviously, a manufacturing process
with lower process deviations than the overall tolerance is required for the imple-
mentation. Savings of lower supplied storage capacities are accompanied with a
larger investment for an appropriate machine.
In order to assess the effects of the chosen configuration of the quality control
cycles on the economic and quality performance of the production system, the
introduced production strategies use different configurations of the production
system as a digital twin implemented within a cyber-physical production system
controller. An event-driven simulation model is modelled. An exemplary production
Fig. 22 Exemplary model of considered CPPS by means of an event-driven simulation, adapted

from (Wagner et al. 2018)
system model is illustrated in 24, consisting of two process chains of manufactured

components (processes P2 and P3). The final components are joined in an assembly
station (A) to the final product. Without loss of generality, the process chains of the
mainly relevant quality features can each be considered as single manufacturing
processes. Economic 100% in-line inspection, e.g., using pneumatic sensors or
functional fluidic end-of-line measurements, enables the assessment of the individ-
ual quality of each component and the final product at the end of the line (Q1, Q2,
Qfinal). After the in-line inspections Q1 and Q2, the inventories I1 and I2 are located,
in which components can be stored randomly, in classes, or individually traceable.
Optimization methods are applied in order to determine the fittest strategy configu-
ration of the quality control cycles for a specific use case in industrial practice (Lanza
et al. 2015) (Fig. 22).
As a result, the most suitable quality control strategy is selected based on the total
cost. Total cost assessments include production cost (equipment cost, labor cost,
material cost, additional implementation cost for cyber-physical production system),
quality cost (scrap cost, cost of measurement errors due to measurement uncer-
tainty), and inventory cost (referring to the cost of the components, storage costs, and
opportunity cost in the inventories). Thus, the optimal strategy can be identified for
each individual component in real time to economically produce high precision
products (Wagner et al. 2018).
References
Aleshin NP, Murashov VV, Evgenov AG, Grigoriev MV, Shchipakov NA, Vasilenko SA, Krasnov
IS (2016) The classification of flaws of metal materials synthesized by the selective laser melting
method and the capabilities of nondestructive testing methods for their detection. Russ J
Nondestruct Test 52(1):38–43
432 G. Lanza et al.
Colledani M, Tolio T (2006) Impact of quality control on production system performance. CIRP
Ann-Manuf Technol 55(1):453–456
Colledani M, Ebrahimi D, Tolio T (2014) Integrated quality and production logistics modelling for the
design of selective and adaptive assembly systems. CIRP Ann Manuf Technol 63(1):453–456
Damm B (2013) Robuste Kurbelwellenmessung mit Röntgenstrahlung. Dissertation. RWTH
Aachen, Aachen
DIN EN ISO 15530-3 (2012) Geometrische Produktspezifikation und -prüfung (GPS) – Verfahren
zur Ermittlung der Messunsicherheit von Koordinatenmessgeräten (KMG) – Teil 3: Anwendung
von kalibrierten Werkstücken oder Normalen
Ebrahimi D (2014) Integrated quality and production logistic performance modeling for selective
and adaptive assembly systems. Dissertation. Politecnico di Milano, Milano
Eschner N, Kopf R, Lieneke T, Künneke T, Berger D, Häfner B et al (2017a) Kombination
etablierter und additiver Fertigung. ZWF 112(7–8):469–472
Eschner N, Lingenhöl J, Öppling S, Lanza G (2017b) Monitoring a laser beam melting process with
acoustic signalss. wt-online 107(11/12):818–823
Everton SK, Hirsch M, Stravroulakis P, Leach RK, Clare AT (2016) Review of in-situ process
monitoring and in-situ metrology for metal additive manufacturing. Mater Des 95:431–445
Fanselow S, Sachs M, Wirth KE, Schmidt J, Peukert W (2016) New methods for process-adapted
characterization for selective beam melting powders. In: Kniffka W, Eichmann M, Witt G (ed.)
Rapid.Tech – International Trade Show & Conference for Additive Manufacturing. Proceedings
of the 13th Rapid.Tech Conference. Hanser, Munich, pp 185–196
Gevatter H-J, Grünhaupt U (2006) Handbuch der Mess- und Automatisierungstechnik in der
Produktion. Springer, Berlin/Heidelberg
Haefner B, Kraemer A, Stauss T, Lanza G (2014) Quality value stream mapping. Procedia CIRP
17:254–259
Imkamp D, Schmitt R, Berthold J (2012) Blick in die Zukunft der Fertigungsmesstechnik. tm Tech
Mess 79(10):433–439
Iyama T, Mizuno M, McKay KN, Yoshihara N, Nishikawa N (2013) Optimal strategies for
corrective assembly approach applied to a high-quality relay production system. Comput Ind
64(5):556–564
JCGM 100:2008 (2008) Evaluation of measurement data – guide to the expression of uncertainty in
measurement (GUM). Joint Committee for Guides in Metrology
JCGM 101:2008 (2008) Evaluation of measurement data – Supplement 1 to the “Guide to the
expression of uncertainty in measurement” – Propagation of distributions using a Monte Carlo
method. Joint Committeefor Guides in Metrology
Kayasa MJ, Herrmann C (2012) A simulation-based evaluation of selective and adaptive production
systems (SAPS) supported by quality strategy in production. Procedia CIRP 3:14–19
Kock H (2017) Track and trace fingerprint. Fraunhofer IPM, Freiburg. Available online at https://
www.ipm.fraunhofer.de/content/dam/ipm/de/PDFs/Pressemitteilung/2017/PI-Track-trace-fin
gerprint-bauteil-rueckverfolgung.pdf. Updated on 6 Jan 2017, checked on 4 Apr 2018
Koelmel A (2016) Integrierte Messtechnik für Prozessketten unreifer Technologien am Beispiel der
Batterieproduktion für Elektrofahrzeuge. Dissertation. Karlsruhe Institute of Technology, Karlsruhe
Koelmel A, Sauer A, Lanza G (2014) Quality-oriented production planning of battery assembly
systems for electric mobility. Procedia CIRP 23:149–154
La Croix A (2008) Risiko-kontrollierte Anwendung von Innovation & technologischem Fortschritt
– Abschlussbericht zum INS 24 Projekt: Standarisierte Entscheidungshilfe zur
Reifegradbewertung im Proudukt-Lebenszyklus – Machbarkeitsstudie
Lanza G, Haefner B, Kraemer A (2015) Optimization of selective assembly and adaptive
manufacturing by means of cyber-physical system based matching. CIRP Ann Manuf Technol
64(1):399–402
Lanza G, Kopf R, Zaiß M, Stricker N, Eschner N, Jacob A et al (2017) Laser-Strahlschmelzen –
Technologie mit Zukunftspotenzial. Ein Handlungsleitfaden. Karlsruhe Institute of Technology
(KIT), Karlsruhe
Loosen P, Funck M (2011) Integrative Produktion von Mikro-Lasern. In: Brecher C (ed) Cluster of
excellence “Integrative production technology for high-wage countries”. Springer, Berlin/Hei-
delberg, pp 1068–1113
Mankins JC (2009) Technology readiness assessments. A retrospective. Acta Astronaut 65
(9–10):1216–1223
Matsuura S, Shinozaki N (2011) Optimal process design in selective assembly when components
with smaller variance are manufactured at three shifted means. Int J Prod Res 49(3):869–882
Mease D, Nair VN, Sudjianto A (2004) Selective assembly in manufacturing: statistical issues and
optimal binning strategies. Technometrics 46(2):165–175
Nicolais L (2011) Wiley encyclopedia of composites. Sheet molding compounds. Wiley, Hoboken
Peter M, Fleischer J (2014) Rotor balancing by optimized magnet positioning during algorithm-
controlled assembly process: selection and assembly of rotor components minimizing the
unbalance. In: Franke J (ed.) Proceedings of the 4th International Electric Drives Production
Conference (EDPC). IEEE, pp 1–4
QASS GmbH (2018) Quality monitoring during welding. Available online at http://qass.net/down
loads/QASS%20-%20Welding%20-%20Quality%20Monitoring.pdf. Updated on 4 June 2018,
checked on 4 June 2018
Reif K (2012) Dieselmotor-management. Systeme, Komponenten, Steuerung und Regelung, 5th
edn. Vieweg+Teubner, Wiesbaden
Schmitt R, Damm B (2008) Prüfen und Messen im Takt. Wie sie mit Inline-Messtechnik ihre
Wertschöpfung maximieren. QZ 53:57–59
Schmitt R, Imkamp D, Bettenhausen K, Berthold J (2011) Fertigungsmesstechnik 2020. Techno-
logie-Roadmap für die Messtechnik in der industriellen Produktion. VDI, Düsseldorf
Spears TG, Gold SA (2016) In-process sensing in selective laser melting (SLM) additive
manufacturing. Integr Mater Manuf Innov 5(2):1–25
Wagner R, Haefner B, Lanza G (2018) Adaptive quality control strategies for high precision
products. Procedia CIRP 75:57–62
Weckenmann A, Jiang X, Sommer K-D, Neuschaefer-Rube U, Seewig J, Shawa L, Estler T (2009)
Multisensor data fusion in dimensional metrology. CIRP Ann Manuf Technol 58(2):701–721
Zaiß M, Jank M-H, Netzelmann U, Waschkies T, Rabe U, Herrmann H-G et al (2017) Use of
thermography and ultrasound for the quality control of SMC lightweight material reinforced by
carbon fiber tapes. Procedia CIRP 62:33–38
In-Process Measurement of Subwavelength
Structures 14
Satoru Takahashi
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 436
In-Process Measurement of Subwavelength Structures Based on
Super-Resolution Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 437
Optical Super-Resolution Technique Using Structured Light Illumination . . . . . . . . . . . . . . . . 437
Super-Resolution Retrieval Algorithm with Successive Approximation Using
Structured Light Illumination . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 438
Formulation of Lateral Resolution Improvement by Fourier Optics . . . . . . . . . . . . . . . . . . . . . . . 440
Numerical Simulation Using the Super-Resolution Image Distribution
Retrieval Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 442
Verification Experiment for Super-Resolution of Subwavelength Structures Using
the Image Distribution Retrieval Algorithm . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 447
In-Process Measurement of Depth for Subwavelength Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 449
Background of Optical Depth Measurement for Subwavelength Structures . . . . . . . . . . . . . . . 449
Theoretical Visualization of Near-Field Optical Interaction with Subwavelength
Microgrooves . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 453
Depth Measurement of Subwavelength Microgrooves Using Far-Field
Optical Response . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 457
Theoretical Verification of the Depth Measurement of Subwavelength Microgrooves by
Numerical Simulation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 461
Application of Near-Field Optics for In-Process Measurement for Quality of
Subwavelength Structures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 464
Nano Thickness Inspection of RLT Using Near-Field Optical Enhancement
of Metal Tip . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 464
Numerical Simulation Based on FDTD Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 467
RLT Measurement System Based on Near-Field Optical Enhancement and Verification
Experiment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 469
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 474
S. Takahashi (*)
Research Center for Advanced Science and Technology (RCAST), The University of Tokyo,
Tokyo, Japan
e-mail: takahashi@nanolab.t.u-tokyo.ac.jp

https://doi.org/10.1007/978-981-10-4938-5_15
436 S. Takahashi
Abstract
This chapter describes in-process measurement of subwavelength structures. Espe-
cially, from the viewpoint of affinity with in-process measurement, this chapter
focuses on optical measurement, which provides in-process evaluation of engi-
neering microstructure surfaces beyond the diffraction limit. First, application of
optical super-resolution using structured light illumination to semiconductor pat-
terns inspection is shown. Second, a new type of optical depth measurement of
subwavelength microgrooves using an interference measuring method, which can
measure the depth of microgrooves, with widths less than the diffraction limit, is
described. Third, as an example of application of near-field optics for in-process
measurement for quality of subwavelength structures, nano-thickness inspection of
residual layer thickness during nanoimprint lithography is demonstrated. Through
concrete examples, the possibility of an optical measurement method for the in-
process measurement of subwavelength structures is discussed.
Keywords
In-process measurement · Optical measurement · Super-resolution measurement ·
In-process inspection · Semiconductor pattern inspection · Nanodefects ·
Subwavelength structure · Microgroove · Near-field optics · Nanoimprint
lithography
Introduction
This chapter describes in-process measurement of subwavelength structures. From

the viewpoint of affinity with in-process measurement at manufacturing sites, optical
measurement is the main focus. Light energy has many intrinsic properties, such as
spatial transmission, broad band spectrum, and high controllability of its physical
values of frequency, amplitude, phase, polarization, and so on, which provide
superior measurement characteristics such as remote operation, parallel processing,
high-speed operation, and nondestructiveness for optical measurements as in-pro-
cess measurement methods.
Although optical measurement has an essential problem involving the diffraction
limit in microscopic measurements, which is becoming increasingly important in
advanced manufacturing sites, it remains an important tool, even in next-generation
advanced manufacturing sites fabricating subwavelength structures owing to its
excellent measurement characteristics represented by nondestructiveness.
Therefore, in this chapter, three types of optical measurement methods for sub-
wavelength structures aimed at solving the above-mentioned diffraction limit problem
are introduced based on optical phenomena such as scattering, interference, and near-
field interaction. First, super resolution imaging based on light scattering from
subwavelength structures is discussed. Second, advanced depth measurement based
on interference measurements for microgrooves is introduced. Finally, in-process
measurement of subwavelength structures based on not only far-field optics, but also
near-field optics is discussed.
14 In-Process Measurement of Subwavelength Structures 437
In-Process Measurement of Subwavelength Structures Based on

Super-Resolution Technique
Optical Super-Resolution Technique Using Structured Light

Illumination
Gustafsson proposed an optical super-resolution technique using structured illumi-

nation shift (Gustafsson 2000), which allows improvement of lateral resolution
beyond the diffraction limit by a factor of two for a wide-field fluorescence micro-
scope. Takahashi et al. applied the super-resolution mechanism of structured illumi-
nation shift to the inspection of subwavelength structures (Usuki et al. 2006). In this
section, the basic principles of the optical super-resolution technique using struc-
tured illumination based on a retrieval algorithm (Nishioka et al. 2006; Takahashi et
al. 2008) are described.
Generally, a far-field image distribution r(x) is determined from the scattering
efficiency distribution a(x) and illumination intensity distribution i(x) by Eq. (1). In
the formula, psf is the point-spread function of the imaging optics, which represents
the Airy disk image produced by diffraction.
O
rðxÞ ¼ psf ðxÞ ðaðxÞ iðxÞÞ: (1)
The image distribution r(x) can be changed by shifting the structured light
intensity distribution i(x) to different positions. This is illustrated in Fig. 1 with an
shift
shift shift
shift
Fig. 1 Change of image distribution caused by structured light illumination shift

438 S. Takahashi
example of two-dot samples. This far-field optical imaging is still dominated by the
diffraction limit. Because psf(x) acts as a low-pass filter in Eq. (1), high-frequency
information of the scattered efficiency distribution a(x) is lost in the image distribu-
tion r(x) through the imaging optics. But Eq. (1) enables us to reconstruct the
scattered efficiency distribution a(x) from r(x) and i(x) when the structured light
intensity distribution i(x) has high-frequency components. By using the information
from both r(x) and i(x), the scattered efficiency distribution a(x) beyond the diffrac-
tion limit can be reconstructed through successive approximation.
Super-Resolution Retrieval Algorithm with Successive

Approximation Using Structured Light Illumination
Equation (1) can be discretely described in matrix form as follows:
R ¼ KA, (2)
that is,
0 !1
0 1 0 1
rð1Þ B kð1,1Þ kð1,jÞ kð1,N Þ C að1Þ
B ⋮ C B CB
⋮ C
B C B CB C
B rðiÞ C ¼ B ⋮ ⋱ ⋮ C CB aðjÞ C
B C B B C, (3)
@ ⋮ A B B kði,1Þ kði,jÞ kði,N Þ C
C@ ⋮ A
@ ⋮ ⋱ ⋮ A
rðN Þ aðN Þ
kðN ,1Þ kðN ,jÞ kðN ,N Þ
where
kði,jÞ ¼ psf ðji jjÞ iðjÞ: (4)
In Eq. (2), R is the far-field image distribution matrix, A is the scattered efficiency
distribution matrix, and K is a coefficient matrix depending on the condition of the
imaging optics and the illumination. In Eqs. (3) and (4), matrix elements, ri), a( j), i
( j), and psf(|ij|) are the far-field discrete image intensity at position i on the image
plane, the scattered efficiency at position j on the object plane, the illumination
intensity at position j on the object plane, and the point-spread function depending
on the positions of i and j, respectively. These equations mean that the scattered
efficiency distribution matrix A is converted into the far-field image distribution
matrix R with the coefficient matrix of imaging K. Super-resolution is realized by
solving Eq. (3) for a( j), but using a mathematical condition of the linear simulta-
neous equation of optical imaging generally leads to poor calculation results.
Especially under real conditions such as those with high levels of noise, convergence
is difficult to achieve in the resolving calculation. To solve this equation under noise
conditions, numerical calculation based on an iterative super-resolution reconstruc-
tion algorithm is effective, in which one of the multiple images is occasionally used
in the reconstruction and an assumed solution is then reconstructed. The block

diagram of this algorithm is shown in Fig. 2 and the steps of the procedure are
detailed as follows.
1) The initial scattered efficiency distribution of the assumed sample A0 is set to a
constant value, and this assumed sample is illuminated with standing light, which
is computationally calculated. Then the initial calculated far-field image matrix
Rcalculation_0 can be obtained using the coefficient matrix K depending on the
condition of the imaging optics and the illumination as follows:
Rcalculation_0 ¼ KA0 : (5)
2) An actual far-field image of the sample with structured illumination is observed

by experiment. This experimentally observed image matrix Robservation_0 is the
Rayleigh-limited image obtained by the imaging optics.
3) The difference between the calculated and observed far-field images is calculated
as an error ratio with respect to each element of the image matrix. The error ratio
e0 is defined in the following equation.
robservation_0 ðiÞ rcalculation_0 ðiÞ

e0 ð i Þ ¼ , (6)
rcalculation_0 ðiÞ
Actual sample Assumed sample

Replace assumed sample
Am with reconstructed sample
? (A0=Const.)
Standing wave
Illumination Shift NO
YES
Experimental optical system R = KA
| Am +1 – Am | ≤ d
Observed images Calculated images Super-resolution image

Robservation_m Rcalculation_m
Experimental Computational Am+1 Reconstructed sample
Error
Error Feedback
robservation _ m (i) – rcalculation _ m (i) am +1 (i ) = am (i ) + w(i ) · em (i ) · am (i)
em (i) =
rcalculation _ m (i) k (i, i )
w(i) = N
, k (i, j ) = psf ( i – j ) · i( j )
Σ k (i, j)
j =1
Fig. 2 Flow chart of super-resolution image distribution retrieval algorithm

440 S. Takahashi
where robservation_0(i) and rcalculation_0(i) are the discrete elements of the initial far-
field observed image matrix Robservation_0 and calculated image matrix Rcalculation_0,
respectively.
4) The error ratio e0 is fed back to the assumed sample A0 and a reconstructed
sample A1 can be obtained. Each element of the reconstructed sample, a1(i), is
obtained using in the following equation.
a1 ðiÞ ¼ a0 ðiÞ þ wðiÞ e0 ðiÞ a0 ðiÞ: (7)
Here w(i) is a weighting coefficient element, which can be calculated from the
element of the coefficient matrix K, as follows:
k ði,iÞ
wðiÞ ¼ : (8)
P
N
k ði,jÞ
j¼1
5) The initial assumed sample A0 consisting of a0(i) is replaced by the reconstructed

sample A1 consisting of a1(i). These reconstructions are applied to the other
shifted positions of the structured illumination.
6) In this manner, the reconstruction (1)–(5) is iteratively applied using the follow-
ing equation.
amþ1 ðiÞ ¼ am ðiÞ þ wðiÞ em ðiÞ am ðiÞ, (9)
where suffix m represents an iteration count. By feeding back the errors of the far-
field images into the scattered efficiency distribution and reconstructing the distri-
bution with successive approximations, we expect to achieve super-resolution. This
algorithm can be thought as a type of iterative technique, which extends the
Richardson-Lucy method (Lucy 1974) to one in which multiple images can be
treated using the weighting coefficient.
Formulation of Lateral Resolution Improvement by Fourier Optics
In this method, one of the lateral resolution improvement factors is a spectrum

extension of the observed optical system based on the illumination distribution.
The structured light illumination distribution i(x) can be expressed as follows:
1 þ cos ð2πf m x þ ϕÞ
i ð xÞ ¼ , (10)
2
where fm is the frequency of the structured light and ϕ is the phase of the structured
light shift. From Eq. (1) and Eq. (10), the following equation can be derived,
1 1
Rð f Þ ¼ OTF ð f ÞAð f Þ þ eiϕ OTF ð f ÞAð f f m Þ
2 4
1 iϕ
þ e OTF ð f ÞAð f þ f m Þ, (11)
4
where R( f ) is the Fourier transform of the microscopic image of the far-field light
distribution r(x), OTF is the optical transfer function, and A( f ) is the Fourier
transform of the scattered distribution on the sample surface. The first term of this
equation means that the spatial frequency of r(x) is restricted by the cutoff frequency
fc depending on the OTF, which is the same as in a conventional microscopic system.
The second and third terms of this equation mean that the spatial frequency of r(x) is
expanded to the frequency of the structured light ( fc fm). Based on the Rayleigh
criterion, the lateral resolution improved by the effectiveness of the high frequency
components of the structured light distribution can be expressed as:
1:22 1:22
lateral_resolution ¼ ¼ , (12)
f c þ f m 2N :A: 1
þ
λ T
where N.A. is the numerical aperture of the imaging optics, λ is the wavelength of the
light, and T is the peak-to-peak distance of the structured light distribution. This
equation means that the lateral resolution limit depends on the structured light
distribution as well as on the imaging optics (Takahashi et al. 2008).
Based on formula (12), the characteristics of lateral resolution improvement can
be discussed. The relationship between the lateral resolution and the peak-to-peak
distance of the structured light distribution T can be expressed as the line graph
shown in Fig. 3 (Usuki et al. 2008). This figure represents the condition when the
wavelength and N.A. are 488 nm and 0.95, respectively. As shown in this figure, the
lateral resolution is improved with a decrease in the peak-to-peak distance of the
structured light distribution T, and 170 nm lateral resolution can be achieved using
Fig. 3 Relationship between 400

resolving power and pitch of Rayleigh Limit
structured light illumination 350
300
Resolving Power nm
250 169nm
200
150 Resolving Power with SLI

100
50 300nm
0
0 500 1000 1500 2000 2500 3000
Pitch of Structured Light Illumination nm
442 S. Takahashi
Fig. 4 Relationship between 1000

resolving power and NA of
objective Rayleigh Limit
800
Resolving Power nm
600
Resolving Power with SLI

400
200
0
0 0.2 0.4 0.6 0.8 1
NA
T = 300 nm. Figure 4 shows the relationship between the lateral resolution and the
numerical aperture of the imaging optics N.A. under the condition of the peak-to-
peak distance T = 300 nm. This method produces approximately twice the resolution
of conventional optical microscopy (Gustafsson 2000).
Numerical Simulation Using the Super-Resolution Image

Distribution Retrieval Algorithm
In this section, the super-resolution algorithm (Fig. 2) is numerically applied to the

super-resolution imaging for a discrete two-point object, which allows quantitative
evaluation of its resolving power. A simulation model is shown in Fig. 5, and the
conditions for the simulation are shown in Table 1. The sampling size in the
numerical simulation was small enough to avoid a discretization error. The numer-
ical aperture of the objective was 0.95, and the wavelength of the light source was
488 nm, so the Rayleigh limit of the imaging optics was 313 nm in the simulation.
Eight shift times were used in the standing wave illumination to verify the algorithm
when multiple shifts were applied. The iterations of the reconstruction were set to
10,000 to allow the solution to converge.
The multiple images and the super-resolution image are shown in Fig. 6, in
which the gap of the two-point scattering objects is 313 nm (the Rayleigh limit of
imaging optics). In Fig. 6, the correspondence of the image modulation to the
illumination distribution was confirmed, and the separation of the two-point object
was clearly identified. Figure 6 indicates the capability of the algorithm for
resolving the two-point scattering objects, which is not clearly resolved with
conventional optical microscopy. To confirm the possibility for resolution beyond
the Rayleigh limit, the gap of the two-point scattering object was set to be 250 nm
(the Sparrow limit of the imaging optics) and 169 nm (the resolution limit assumed
from the spectrum of the standing wave illumination). The conventional (a) and
Intensity
Structured light illumination shift
Object plane
Pitch
Gap of 2 points object
Fig. 5 Simulation model of discrete two-point object for confirming the super-resolution image
distribution retrieval algorithm
Table 1 Parameters for the numerical simulation of super-resolution

Wavelength of source 488 nm
Pitch of structured light illumination 300 nm
Sampling size 1.2 nm
NA of objective 0.95
Rayleigh limit 313 nm
Shift times 8
Shift step size 30 nm
Iterations 10,000
Gap of two-point object 313 nm, 250 nm, 169 nm
reconstructed super-resolution images (b) are shown in Fig. 7. In the case of the
250-nm gapped two points, the flat part was confirmed around the center of the
conventional image distribution (Fig. 7Aa), which corresponded to the Sparrow
limit of resolution. The differentiation of two points from one point was difficult in
the case of the 169-nm gapped two points (Fig. 7Ba). On the other hand, in both
cases, the two points were independently identified in the super-resolution images
(Fig. 7Ab, Bb).
Noise-robustness is one of the most important factors when applying this super-
resolution method to practical applications such as semiconductor inspection. Here,
random noise in various forms was taken into special consideration, because sys-
tematic noises can be reduced with a calibration step. The relationship between
noise-robustness and structured illumination shift times was investigated by numer-
ical simulation. The conditions of the simulation are the same as those described
above, and 30% random noise was added to all images observed in the case of a
169 nm-gapped two-point scattering object (Fig. 8). An observed image of two-point
scattering objects with 30% random noise is shown in Fig. 8b. In this simulation, the
shift step size was set to a constant 30 nm to clarify the effects of the shift of the
standing wave illumination.
444 S. Takahashi
Illumination Illumination Illumination
Shift Shift Shift
Imaging Imaging Imaging
Iterative reconstruction
Conventional image Super-resolution image
Fig. 6 Resolving the Rayleigh limit gap (313 nm) by structured light illumination shift and
iterative reconstruction of multiple images
The super-resolution images with various shift times are shown in Fig. 9. In the
case of shift time 0 (Fig. 9a), the super-resolution image distribution was irregular,
caused by random noise mixed with the observed images, but the distribution was
improved to a clearly resolved distribution by shifting the image 20 times (Fig. 9c).
Theoretically, two shifts of the standing wave illumination (3 images) are suffi-
cient to realize 1-D super-resolution under identical conditions without noise, but
additional shifts of illumination contribute to improvement of resolution in cases
involving noise. To quantitatively evaluate the effects of increasing the shift times,
simulations with various shift times were implemented, in which the peak-to-valley
ratio of the super-resolution image distribution (V/P) was used as an indicator of
noise-robustness. The means of the calculated V/P with standard deviation are
a 1 1
(a) (b)
0.8 0.8
Intensity a.u.
Int ensit y a.u.
0.6 0.6
0.4 0.4
0.2 0.2
0 0
-400 -200 0 200 400 600 -400 -200 0 200 400 600
Position nm Position nm
Result obtained from 250nm gapped 2 points object
b 1 1
(a) (b)
0.8 0.8
Intensity a.u.
Int ensit y a.u.
0.6 0.6
0.4 0.4
0.2 0.2
0 0
-400 -200 0 200 400 600 -400 -200 0 200 400 600
Result obtained from169nm gapped 2 points object
Fig. 7 Comparison of conventional optical microscopic image (a) and super-resolution image (b)
for two-point object
2
(a) (b)
1.5
Intensity a.u.
1.0
0.5
0
-400 -200 0 200 400 -400 -200 0 200 400
Fig. 8 Examples of observed images without super-resolution. (a) Noiseless image. (b) Image with
30% random noise
446 S. Takahashi
(a) Shift times = 0 (b) Shift times = 6 (c) Shift times = 20

1
Intensity a.u.
Redundant shift Redundant shift

0.5
0
-500 0 500 -500 0 500 -500 0 500
Position nm Position nm Position nm
Fig. 9 Influence of number of shifts (shift times) on obtained super-resolution image under 30%
random noise
Fig. 10 Enhancement of Super-resolution image with noise

robustness by redundant shift 1
under 30% random noise Second highest peak P
Valley to peak ratio V/P a.u.
0.8
0.6
Valley V
0.4
0.2
0
0 5 10 15 20
Shift times
plotted against shift times in Fig. 10. Noise-robustness was improved with redundant
shifts of the standing wave illumination and it was confirmed that the 169 nm-gapped
two-point object could be clearly resolved under 30% random noise by using more
than 10 shifts. When the number of shifts was 10, the pitch of the standing wave
illumination was 300 nm, and the shift step size was 30 nm, and the results indicate
that it is necessary to shift the standing wave up to the peak-to-peak height of the
standing wave (10 30 = 300 nm) to obtain adequate results (Usuki et al. 2008).
Figure 11 shows a numerical simulation envisioning semiconductor patterns.
Figure 11A shows the envisioned semiconductor patterns (a), having 100-nm-line-
and-space and a conventional optical image (b), which is observed with an optical
system under the condition that the wavelength and N.A. of the objective are 488 nm
and 0.95, respectively. As shown in Fig. 11Ab, the line and space patterns are no
longer resolved because the diffraction limit of the optics is 313 nm, being much
larger that the features. Figure 11B shows multiple observed images with the same
optical system from Fig. 11Ab, which are modulated with structured light illumina-
tion shifts. Here, to obtain 2-D super-resolution, two sets of structured light
a NA=0.95
b
100nm line & space Wavelength=488nm
Diffraction limit : 313nm 200nm
Short Circuit
200nm 200nm
(a) (b) (c)
Breaking of Wire
(a) (b) (d) (e) (f)
Semiconductor patterns and conventional Multiple observed images modulated

optical image with structured light illumination shifts
c Short Circuit
200nm
Breaking of Wire
Reconstructed super-resolution image from observed multiple modulated images (B)
Fig. 11 Numerical simulation envisioning semiconductor patterns
illumination having a mutual perpendicular relationship are applied and their phases
are shifted. Please note that each observed image (Fig. 11Ba–f) is still degraded by
the diffraction limit. By applying the super-resolution image distribution retrieval
algorithm using the observed multiple modulated images, a reconstructed image
(Fig. 11C) is obtained. Positions of short circuits and broken wires become clear.
Thus, the possibility of defects inspection of subwavelength structures using the
super-resolution method is indicated (Kudo et al. 2009).
Verification Experiment for Super-Resolution of Subwavelength

Structures Using the Image Distribution Retrieval Algorithm
Figure 12 shows a schematic diagram of a typical super-resolution optical system for

in-process inspection of semiconductor defects based on structured light
448 S. Takahashi
CCD
ILLUMINATION LIGHT
2 SPOT SPATIAL FILTER REFLECTED LIGHT
for DARK-FIELD IMAGING
IMAGING
LENS SCATTERED LIGHT
FILTER
MIRROR 2
LENS 5 LENS 2
LENS 3 PC
BEAM MIRROR 4 BEAM SPLITTER 1

SPLITTER 2
MIRROR 3
LENS 4
LENS 1 LASER LIGHT SOURCE
OBJECTIVE
MIRROR 1
SAMPLE
PIEZO ELECTRIC ACUTUATOR
XYZ SCANNER
Fig. 12 Schematic diagram of typical super-resolution optical system for in-process inspection of
semiconductor defects
illumination shifts. This system mainly consists of a standing wave illumination

generator, a standing wave shifting unit, and a dark-field far-field imaging unit with a
regular reflective light cut filter. In Fig. 12, beam splitter 1 divides a beam of light
into two orthogonal beams. These two beams are collimated and face the micro-
scopic field of the high power objective lens with high numerical aperture through
mirror 1, mirror 2, mirror 3, mirror 4, lens 1, lens 2, lens 3, lens 4, and beam splitter
2. These two collimated and facing beams interfere on the sample surface and
generate standing wave illumination. The light axial position of mirror 1 is displaced
at the nano-scale with the piezo-electric actuator, which precisely produces a phase
difference between the two beams in the illumination optics. Thus, the standing
wave illumination can be shifted at nano-scales with the piezo-electric actuator
(PZT) attached to mirror 1. The standing wave illumination is scattered by the
sample surface, and the scattered light is focused on an imaging plane corresponding
to the charge-coupled device (CCD) surface through the objective lens, beam splitter
2, lens 5, the filter, and the tube lens. In the imaging optics, the objective lens and
lens 5 are set for infinity correction optics, and the tube lens is set for finite optics.
The specially designed filter is placed between lens 5 and the tube lens and rejects the
light that is the regular reflection of the illumination light. So we obtain only light
from scattering factors on the sample surface with high contrast. This dark-field
imaging of laser scattering makes it possible to detect the image with optically high
signal-noise ratio. The computer (PC) in the figure controls the CCD, the PZT, and
the XYZ scanner and processes multiple images with high signal-noise ratio for
super-resolution imaging (Usuki et al. 2006).
Figure 13 shows an example of a verification experiment in which a sub-
wavelength structure beyond the diffraction limit of the optical system can be
resolved with the super-resolution method (Kudo et al. 2011). Figure 13Aa shows
SEM images of subwavelength structures with line and space patterns of 200 nm
(Fig. 13Ab). On the right figure of Fig. 13Ac is an optical microscopic image for the
square area indicated in the left figure (SEM image) of Fig. 13Ac, which is observed
with a conventional dark-field microscopic optical system under the conditions that
N.A. and wavelength are 0.55 and 532 nm, respectively. Figure 13Ac suggests that
structures cannot be clearly seen because the Rayleigh limit of the optical system is
590 nm. Figure 13B shows modulated images which are observed with the same
optical system (Fig. 13Ac) under structured light illumination with pitch of 270 nm
when structured light illumination shifts with shift step size of 8.3 nm. The stripe
patterns of the top figures of Fig. 13B do not directly indicate the lines and spaces of
the target structures but are the moire patterns generated by interactions between the
structured light illumination and the subwavelength structures. By performing the
image distribution retrieval algorithm using these modulated images, a 2D super-
resolution image can be obtained (Fig. 13C).
Figure 14 shows a typical example when applying this super-resolution method
as the in-process defects inspection for semiconductor patterns (Usuki et al. 2010).
Figure 14a shows a SEM image of the sample including a defect which is located
between 250-nm spaced line edges. The super-resolution method was applied to
resolve this sample and to detect the defect with 313-nm Rayleigh-limited imaging
optics (N.A.: 0.95, wavelength: 488 nm). The super-resolution image resulting from
the iterative reconstruction of multiple modulated images observed with standing
wave illumination (pitch of structured light: 490 nm) shift is shown in Fig. 14b. In
this figure, it was found that the defect located between the 250-nm spaced line edges
was around the central position of the super-resolution image. Thus, the approxi-
mately 250-nm-size defect on the sample was detected with the super-resolution
method with higher resolution than the Rayleigh limit of imaging optics (313 nm).
In-Process Measurement of Depth for Subwavelength Structures
Background of Optical Depth Measurement for Subwavelength

Structures
Functional microstructure surfaces have recently attracted attention in the micro-

manufacturing field such as microelectro-mechanical systems, micro-opto-electro-
mechanical systems, and so on, owing to their specialized functionalities, which are
450 S. Takahashi
a
Line edge 200nm
1µm
(a)SEM image of sample (b)Schematic of sample
Incident
LIght
5µm 1µm
(c)Normal image
Subwavelength structure sample (a), (b) and normal image (c) under N.A. of 0.55
b Standing wave illumination (270nm-pitch) shift

2µm
Modulated scattered light image (modulation of standing wave illumination is horizontal)

2µm
Modulated scattered light image (modulation of standing wave illumination is vertical)

Modulated images with structured light illumination shifts under N.A. of 0.55
c 1µm
200nm L&S Pattern

(a) (b)
(a)Normal image of sample (b)2D super-resolution image of sample
Comparison between normal image and 2D super-resolution image under the same optics (N.A.:0.55)
Fig. 13 Super-resolution experiment for subwavelength structures

b
Intensity a.u.
0.05
0
0
500
1000
1500
2000
2500
Position nm
3000
3500
4000
4500
5000 2500 3000

1500 2000
500 1000
0
Position nm
Line Line Line Line Line Line

Space Space Space Space Space
Fig. 14 SEM image of 250-nm-spaced line edges with a defect (a) and its super-resolution image
(b)
452 S. Takahashi
characterized by their miniaturized structure (Bruzzone et al. 2008). In particular,

microgroove structures, having an aperture size of less than several hundred nano-
meters with an aspect ratio (width/depth) lager than 1, are one of the most important
functional microshape elements on a device surface, which play an important role as
key functional elements, such as micro-U-shaped cavities, for microstructured
optical sensors (Ho et al. 2014), nano-hole arrays for microstructured solar cell
surface (Peng et al. 2010), microchannels of microfluidics systems for DNA analysis
(Calaon et al. 2015), and so on. To reliably fabricate these production devices, in-
process quality control methods for microgrooves based on nano- or micrometrology
are gaining importance (Hansen et al. 2006). Of the various in-process quality
control factors, this section focuses on the quantitative depth evaluation of micro-
grooves, which is one of the most challenging tasks in microstructure surface
profiling.
Figure 15 shows a conventional depth measurement using an interferometer
based on phase change, which is one of the typical in-process microstructure surface
evaluation methods. This method does have nanometer resolution in the depth
direction, but it cannot be generally applied to microgrooves, whose width is less
than about half the wavelength owing to a low lateral resolution problem caused by
the optical diffraction limit. Figure 16 shows an example of the general limitation of
conventional optical depth measurement based on phase change, indicating that the
depth of a 200-nm-width microgroove cannot be evaluated while the depth of a 500-
nm- or 1000-nm-width microgroove can be measured under a wavelength of 488 nm
(Takahashi 2015).
To overcome this low lateral resolution problem in the far-field optical interfer-
ence response, various useful super-resolution techniques are being developed, such
as coherent scanning interferometry (CSI) with rigorous coupled wave analysis
(RCWA) (de Groot et al. 2008), the combination of CSI and structured illumination
microscopy (SIM) (Lehmann et al. 2014), the combination of digital holographic
microscopy (DHM) and SIM (Gao et al. 2013), the angular multiplexing method of
a b Reflected wave
Plane wave incidence
f
Phase
Depth
difference
Wavelength l
Width
Fig. 15 Basic principles of conventional optical depth measurement based on phase change. (a)
Incident wave, (b) reflected wave, phase of which is changed by surface profile
500
[ nm ]
400
Theoretical line
300
Calculated Depth from
phase difference ( f )
Width: 1000 nm
Width: 500 nm
200
100
Width: 200 nm < Half of wavelength
0
-100
0 100 200 300 400 500
Depth [ nm ]
Fig. 16 Limitation of conventional optical depth measurement based on phase change

(λ = 488 nm, N.A. = 0.55) (Takahashi 2015)
DHM with VCSEL (Mico et al. 2006), ultrashort pulsed lasers (Yuan et al. 2008),
polarization multiplexing (Yuan et al. 2011) based on off-axis optics, and so on.
In this section, an optical depth measurement method (Takahashi et al. 2017),
having the potential for in-process depth evaluation of microgrooves, with width less
than half the wavelength, with aspect ratio larger than 1, is described.
Theoretical Visualization of Near-Field Optical Interaction with

Subwavelength Microgrooves
To rigorously analyze the detailed optical response from a microgroove on a surface,

with critical dimensions less than the wavelength, it is essential to consider the nano-
scale interaction of localized light energy behavior not as optical waves but as
electromagnetic fields based on Maxwell’s equations; therefore, numerical simula-
tions using a finite-difference time-domain (FDTD) method, with which the electric
field vector components interact with arbitrarily fine structures with nanometer-scale
resolution can be numerically calculated, is a powerful tool. Figure 17 shows a
standard FDTD simulation model for the following analyses.
Here, a silicon substrate is analyzed, which is a typical substrate for micro-
structured functional devices. We set 5-nm units or 10-nm units cells for a simulation
range of 4000 2000 to 2500 nm. For the external light illuminating the micro-
groove on the silicon substrate, a linearly polarized plane wave with a 488.0-nm
wavelength at an incident angle of 90 is employed. The 0.5-μm-thickness perfect
454 S. Takahashi
Fig. 17 FDTD simulation 4.0 μm

model for analyzing detailed Plane wave
optical phenomena ( l = 488.0 nm) Light source
2.0 to 2.5 μm
Air (n = 1.000)
Depth
Silicon Width
(n = 4.367+0.0079 i)
Perfect matched layer
matched layer (PML), which is an artificial absorbing material that ideally absorbs
field energy propagating through the PML layers without back reflections from the
boundary interface, is placed at the upper and lower boundaries for the detailed
analyses.
Figure 18 shows typical FDTD results. Amplitude distributions resulting from the
interaction of the electromagnetic fields of the incident light with the microgroove
for different polarization states, which correspond to time-averaged light energy
distributions under the steady state condition are shown – in the case of a 300-nm-
depth microgroove with width of 1000 nm (A), as an example of greater than half the
wavelength, and for 200 nm (B) as an example of less than half the wavelength. For
each width, figures (a) and (b) show results under different polarizations of incident
light, the direction of the electric field vector, being parallel (a) and perpendicular
(b), to the inner wall.
First, focusing on the inside of the microgrooves, only the perpendicular polar-
ization generates light energy localization inside the microgroove in the case of the
200-nm-width microgroove, while localized light energy exists for both polariza-
tions in the case of the 1000-nm-width microgrooves. From this, a polarized incident
wave perpendicular to the inner wall is useful to deliver light energy to the bottom of
the microgrooves, especially those with a width less than half the wavelength.
Second, focusing on the air domain just above the surface with the microgrooves,
there is a periodic distribution vertically under all conditions, as shown in Fig. 18.
Considering that these are time-averaged light amplitude distributions, the periodic
distribution in the air domain indicates that a standing wave is generated by the
interference phenomena of counter propagating waves consisting of incident waves
and response waves from the sample surface. The stepwise variation of the standing
wave above the microgroove of Fig. 18A means the presence of an optical wave
response having a phase difference based on the relative distance difference from the
bottom to the top. On the other hand, such a clear stepwise variation on the periodic
amplitude distribution cannot be seen in Fig. 18B.
(A) (a) Direction of electric field
Amplitude of electric field |E | a.u.

(b) Direction of electric field 1.0
500 nm
500 nm
0
(B) (a) Direction of electric field
Amplitude of electric field |E | a.u.

(b) Direction of electric field 1.0
500 nm
500 nm
0
Fig. 18 Visualization of optical waves for different polarization states. (A) 1000 nm width, 300 nm
depth, (B) 200 nm width, 300 nm depth
The time-averaged amplitude distributions of Fig. 18 obtained with the FDTD

numerical simulations can be thought of as actually being the superposition of
various light wave components existing in the analysis area, generated by the
initially set incident light. The leftmost diagram of Fig. 19 shows all assumed
response components existing in the analysis area. The electric field vector of the
superposition of all existing components can be written as
E groove
all ¼ E groove groove groove groove
incidence þ E surface_reflection þ E edge_scatter þ E bottom_response (13)
where E groove
all , E groove groove groove groove
incidence , E surface_reflection , E edge_scatter , E bottom_response are all the existing
components, incident wave, reflected wave from the top surface, scattered wave
from the edges of the microgroove, and response wave from the bottom of the
microgroove, respectively. Now as shown in Fig. 19, by considering
E groove
incidence ¼ E all
air
(14)
E groove groove groove groove

all_response ¼ E surface_reflection þ E edge_scatter þ E bottom_response (15)
we can obtain
456 S. Takahashi
Fig. 19 Extraction of the sample response
Egroove groove
all_response ¼ E all E air
all (16)
where Egroove air

all_response , E all are all the existing responses from the surface with the
microgroove and all existing components only in air. Figure 20 shows near-field
phase distributions of Egroove all_response monitored at a height of 20 nm just above the top
surface. It can be clearly seen that the near-field phase is actually changed according
to the depth even when the width is less than half the wavelength (Fig. 20b) as is the
case when the width is larger (Fig. 20a), while the far-field phase does not change
according to the depth when the width is less (Fig. 16). This result suggests that the
phase distribution corresponding to the depth of the subwavelength microgroove,
which is impossible to measure with conventional far-field remote sensing, actually
exists in the near-field.
Based on the above, we evaluate whether the optical response from the groove
bottom can be radiated as a freely propagating light wave or not. For this purpose, we
attempted to extract the optical response of only the bottom of the microgroove from
all light energy existing in the analysis area. As shown in Fig. 21, the electric field
vector of the optical response from the bottom of the microgroove can be written as
E groove groove
bottom_response ¼ E all Egroove
all
without bottom
(17)
using
Egroove
all
without bottom
¼ Egroove groove groove
incidence þ E surface_reflection þ E edge_scatter (18)
where E groove groove without bottom

bottom_response ,E all are the extracted optical wave component
from the bottom and all components of a microgroove with infinite depth, respec-
tively. Figure 22 shows visualizations of optical responses from the bottom of a
microgroove calculated based on Eqs. (5) and (17). Figure 22A and B shows the
results in the case of a 300-nm-depth microgroove with 1000 nm width and 200 nm
(b) 200 nm width

Near-field phase [ rad ]
(a) 1000 nm width
500 nm 500 nm
Fig. 20 Calculated near-field phase distribution of Eall_response at an observation plane placed

20 nm above the top surface
Fig. 21 Extraction of detailed optical responses from the bottom of a microgroove
width, respectively. And figures (a) and (b) show the corresponding amplitude
distributions and phase distributions, respectively. In both figures (a), quantitative
values of amplitude at position P (located 500 nm above the top surface) are
indicated as the ratio to the incident light. From figures (a) for both cases, the
amplitude decreases compared to the 1000-nm-width microgroove, and the 200-
nm-width microgroove actually radiates upward. Also from the phase distribution of
figures (b), a comparison of Fig. 22Bb with Fig. 22Ab demonstrates that even the
observed light energy in Fig. 22Ba behaves as free space propagating light.
Depth Measurement of Subwavelength Microgrooves Using Far-

Field Optical Response
From the detailed near-field analyses in the previous section, even in the sub-
wavelength microgrooves with aspect 1 or more, light energy can reach the bottom
by setting appropriate polarization and a phase distribution corresponding to the
groove depth is then generated on the sample surface. In addition, the optical waves
458 S. Takahashi
(A) (a) Amplitude of | E bottom_response | (b) Phase of | E bottom_response | +π
Phase of electric field |E | rad

P 73.2 %
0
500 nm
500 nm
-π
(B) (a) Amplitude of | E bottom_response | (b) Phase of | E bottom_response | +π
Phase of electric field |E | rad

P 14.6 %
0
500 nm
Faintly but clear wave

500 nm component from the
bottom can be seen. -π
Fig. 22 Visualization of optical responses from the bottom of a microgroove. (A) 1000 nm width,
300 nm depth, (B) 200 nm width, 300 nm depth
from the bottom indicating the groove depth information are not localized and are
radiated upward as freely propagating light. The above facts suggest that the optical
waves indicating the depth information can arrive in the far-field, which means that
even far-field optical sensing techniques, having inherently high affinity as in-
process metrology, have the potential to detect the signal indicating groove-depth
information.
Figure 23 schematically shows far-field optical imaging for both the case where
the width of the microgroove is larger than the diffraction limit (a) and the case
where the width of the microgroove is lower (b). According to optical imaging
theory, a far-field image is formed by the convolution of the complex amplitude of
the response wave from the target surface and the point spread function of the
imaging system. In the case of the subwavelength microgroove, even for a laterally
centered position of the microgroove on the far-field imaging plane, there are
contributions of the top surface reflection owing to the spread of the point spread
function. As a result, the phase difference is reduced more than the original depth
information, as shown in Fig. 23b. This estimation mechanism is also consistent with
the 200-nm-width case in Fig. 16.
Figure 24 shows the principles of an in-process depth measurement of a
subwavelength microgroove based on the estimated far-field optical response
(Fig. 23). When the width is sufficiently greater than the diffraction limit, a far-
field observed wave at the laterally centered position of the microgroove is almost
(a) (b)
Bottom response Phase of Phase of Phase of
Phase of Phase of Phase of surface bottom surface
buried by
surface bottom surface reflection response reflection
surrounding surface
reflection response reflection
Far-field optical response
( Practically observable )
reflection
f
f f
Far-field phase
difference,
which is shrunk
during imaging
process
Imaging system
( Practically unobservable )
Near-field optical response
for far-field detection
Surface Bottom Surface Surface Bottom Surface

reflection response reflection reflection response reflection
Near-field phase
f f
difference,
corresponding to
depth
Fig. 23 Examination of far-field optical response formation
the same as the optical wave from the bottom response, resulting in the observable
phase coinciding with the depth information (Fig. 24a). On the other hand, when
the width is less than the diffraction limit, the far-field observed wave at the
laterally centered position of the microgroove is not equal to the optical wave
from the bottom response but to the vector sum of the contribution of the optical
wave from the top surface reflection generated by the spread of the point spread
function on the complex plane (Fig. 24b). Therefore, if the contribution of the top
surface reflection is given, a complex vector of the optical wave from the bottom
response containing the depth information can be derived. Figure 25 shows a
calculation method for the contribution of the top surface reflection, which can
be analytically solved by using the top surface response from the uniform top
surface without the microgroove area. Finally, we obtain
Ð1
r0a ðxÞ w asf ðxÞdx
ra ðxÞ ¼ Ð1 2
, (19)
1 asf ðxÞdx
where ra ðxÞ, r0a ðxÞ, w,asf ðxÞ are the contribution values of the top reflection to the
center position of the microgroove on the far-field imaging plane, the reflected signal
460 S. Takahashi
Fig. 24 Principles of in-

(a) Im
process depth measurement
using far-field optical Far-field observed wave at Optical wave from
response the laterally opening position bottom response
Observable
phase
f
Re
Depth
information
(b) Im
Far-field observed wave at Optical wave from

the laterally opening position bottom response
Observable
phase
f
f Re
Depth
information
Optical wave from
surface reflection
Fig. 25 Calculation method ( Center of ) laterally opening position

of contribution of top surface
Imaging plane
reflection x
Point spread function of Contribution of

imaging system PSF(x) surface reflection
from position A
PSF(a)
a
Aperture size w
Position A
Surface reflection area Surface reflection area

from the uniform top surface, and the amplitude spread function corresponding to the
point spread function under the coherent imaging optics, respectively, which can be
determined by the optical imaging condition (Takahashi et al. 2017).
Theoretical Verification of the Depth Measurement of

Subwavelength Microgrooves by Numerical Simulation
In this section, the basic validity of the concept using only the far-field optical
response by considering its applicability to in-process measurement is discussed.
Figure 26 shows an envisioned experimental setup to realize the concept. This
experimental setup mainly consists of a continuous wave (CW) laser as a linearly
polarized light source with a polarization-maintaining fiber, an infinity-corrected
optical imaging system with a specially designed plane wave incident unit, and a
Linnik interferometric system with a phase shifter having a resolution of one
nanometer by a PZT. By inserting a plane-convex lens after the collimating lens
immediately under the fiber output, plane wave incidence on the measurement
target can be realized even with the high numerical objective required for high
resolution imaging. The phase shifter of the Linnik interferometer allows us to
Point light source Polarization-maintaining fiber

(Fiber out)
Plano-convex lens for forming
Collimate lens f2
plane wave incidence
Half wave plate after objective
Phase shifter (PZT) for

calculating phase Tube lens for infinity-corrected
distribution from detectable optical imaging system
intensity distribution
(NA)
Objective f1
Reference
Mirror
f0 f0 f0
Reference arm
Depth of focus
f0
Objective(NA) Optical array sensor (CCD) for
detecting Intensity distribution
Plane wave incidence Measuring surface
Fig. 26 Far-field phase evaluation system based on Linnik interferometer combined with infinity-
corrected optical system and plane wave incident unit
462 S. Takahashi
Fig. 27 Example of verification of the depth measurement method using only far-field intensity
distributions for 200 nm depth and 300 nm width. (A) Detected far-field interferometric intensity
distribution by changing the phase shift of the reference arm. (B) Calculated far-field phase
distribution using four intensity distributions (A). (C) Depth calculation using the contribution of
top surface reflection and the calculated far-field phase (B)
calculate the far-field phase distribution, which cannot be directly observed as is

the case for the intensity distribution, by using a phase shift interferometric
technique.
Figure 27 shows a verification of the depth measurement method using only far-
field intensity distributions, which were numerically obtained by Fourier optical
imaging theory assuming the above experimental setup under the following condi-
tions: wavelength, λ = 488.0 nm; numerical aperture of objective lens, N.A. = 0.55;
diffraction limit of imaging system, D.L. = 541 nm; width of microgroove,

w = 200 nm; and depth of microgroove, d = 300 nm. Figure 27A shows a set of
the four intensity distributions (I1, I2, I3, I4) acquired by changing the phase of the
reference wave by π/2 in the Linnik interferometer. According to the four-step phase
shift algorithm (Wingerden et al. 1991), the phase distribution can be calculated by
using the following expression,

0 1 I4 I2
ϕ ¼ tan (20)
I3 I1
By using this equation, the phase difference from the top surface, ϕ’, is obtained
as 0.2415 rad (Fig. 27B). Next, by closing the reference arm of the Linnik interfer-
ometer, the normal far-field intensity distribution of the microgroove on the surface
using the same optical imaging system is obtained. The acquired value of the
amplitude of this intensity distribution is 76.02. Finally, according to Eq. (19),
the contribution value (74.99) of the top reflection using a detected signal from the
uniform top surface can be calculated. By using the relationships of the three values
as complex vectors, as shown in Fig. 24b, the far-field phase difference ϕ can be
calculated as 1.635 rad. From the following equation
ϕ þ 2nπ
Depth ¼ λ, (21)
4π
the depth value of 307.5 nm (natural number, n = 1) for the 300-nm-depth sample can be
calculated. Here, because of space constraints, we cannot focus on the phase unwrapping,
but the natural number n can be determined without any ambiguity by appropriate
established phase unwrapping methods (Gass et al. 2003). Figure 28 shows the charac-
teristics of the depth measurement method in the case of the 200-nm-width microgroove.
It is clear that the depth measurement method can measure the depth within a depth range
of almost 1 μm, which means that a relatively high aspect ratio of 5.0 can be achieved
with an expanded uncertainty (k = 2) of 10% for NA of 0.95 and 12% for NA of 0.55,
while the conventional method cannot measure the depth even with NA of 0.95.
At the production site, the internal shape of the fine groove is not necessarily the
ideal design shape and is unknown. Therefore, robustness of the depth measurement
method for various microgrooves was examined. In Table 2, three typical examples of
microgrooves, a tapered structure (a), microroughness surface on the bottom (b), and
different materials (SiO2; refractive index, n = 1.4631) constituting the bottom
substrate (c), are shown compared with the ideal type with width of 200 nm and the
depth of 400 nm (aspect ratio = 2). From this table, the observed far-field amplitude
from the microgroove changes depending on its internal characteristics, but this
method can measure the depth with less than 10% error under those conditions without
being affected by the difference of the internal characteristics of the microgroove. The
reason for these results is that we do not need a priori knowledge of the reflection
464 S. Takahashi
1500
Theoretical line
[ nm ]
1200 N :0.95)
Proposed method (NA:0.95)
(NA
2σ = 10 % @depth < 1 μm
Calculated Depth from
900
phasedifference ( f )
600 N
(NA:0.55)
Proposed method (NA
2σ = 12 % @depth < 1 μm
300
N :0.95)
Conventional method (NA:0.95)
(NA
0
-100
0 300 600 900 1200 1500
Depth [ nm ]
Fig. 28 Depth measurement characteristics
efficiency from the bottom of the microgroove because we solve it as unknown

information, which is one of the important measurement characteristics of this method.
Therefore, even if there is a change in the amplitude of the response from the bottom
caused by internal differences, as long as it is radiated to the far-field, this method has
the potential to accurately evaluate the depth (Takahashi et al. 2017).
Application of Near-Field Optics for In-Process Measurement for

Quality of Subwavelength Structures
In this section, as an example of an application of near-field optics to an in-process

measurement, an optical inspection of the residual layer thickness for nanoimprint
lithography is presented.
Nano Thickness Inspection of RLT Using Near-Field Optical

Enhancement of Metal Tip
Nanoimprint lithography (NIL), allowing fabrication of micro/nanometer scale

features at low cost, high throughput, and high resolution (Chou et al. 1996), is
one of the best candidates for meeting the requirements of next-generation semi-
conductor patterning. The process steps of UV-based NIL (Fig. 29) are as follows:
(I) A mold made of transparent material such as fused silica quartz with nano-
patterns on its surface is pressed into a resin layer on the Si substrate after
14
Table. 2 Examination of robustness for various microgroove features (200 nm width, 400 nm width) using imaging optics (λ = 488 nm, NA = 0.55, D.L.
= 541 nm)
Types of grooves (c) Microroughness (d) Internal different

(a) Standard structure (b) Tapered structure
on bottom materials
200nm 200nm 200nm 200nm
Si Si Si Si
400nm
400nm
400nm
400nm
120nm PV:10nm SiO 2
Depth condition nm 400 400 400 400

Measured depth directly from phase(ii) by conventional method nm 10.9 8.2 10.5 5.0
In-Process Measurement of Subwavelength Structures
(i) Observed far-field amplitude from the microgroove a.u. 142.5 146.8 140.4 147.8
(ii) Calculated far-field phase using 4-shifted far-field amplitudes rad -0.28 -0.21 -0.27 -0.13
(iii) Observed far-field amplitude from the top surface a.u. 150.0 150.0 150.0 150.0
Calculated depth from (i), (ii), and (iii) based on proposed method nm 415 419 413 420
465
466 S. Takahashi
(i) Press
Mold (Quartz)
(ii) UV Exposure
UV (iii) Release
(iv) Reactive Ion

Resist
Substrate
(Photopolymer)
Solidified
Polymer
Residual layer: Very thin film, thickness of

which is about several 10nm.
Fig. 29 Process steps of UV-cured NIL for semiconductors
dispensing a UV-curable liquid photopolymer as the resist on the Si substrate. (II)

The resist is cured by UV light exposure and becomes solid. (III) The mold is
released from the solidified resin. Here, a thin resin layer with thickness of several
10s of nm remains as a residual film (residual layer) between the imprinted patterns
and the Si substrate. Finally, reactive ion etching (RIE) is performed to remove the
residual layer. (IV) As a result, we can form fine patterns on the Si substrate as in
conventional photolithography. In the NIL process, to obtain accurate fine resist
mask patterns, an optimization of the RIE based on the residual layer thickness
(RLT) is one of the most important factors. Therefore, to realize NIL as a next-
generation semiconductor photolithography process with high reliability, develop-
ing a nondestructive evaluation technique of the RLT, allowing optimization the
RIE condition of the NIL processes is need urgently (Lee 2005; Balla et al. 2008).
In this section, an application of near-field optics for inspection of the RLT of NIL
is discussed.
Figure 30 shows a schematic diagram of the new concept of the RLT inspection
method for NIL using near-field optical enhancement of a metal tip. To obtain the
RLT information, a sharp metal probe with tip diameter in the 10 nm scale is set just
above the residual layer located between the fine patterns formed with NIL. The
sample surface is obliquely illuminated with gently focused incident light. Under
this condition, at the vicinity of the metal tip, localized light energy can be generated
by near-field optical enhancement of metal tip (Kawata and Inouye 1995; Aigouy et
al. 1999). This enhanced light energy, the size of which does not depend on the
diffraction limit but mainly on the tip size, is so localized that it can interact with only
the area of the residual layer of the NIL sample. Therefore, by detecting the far-field
propagating light which is radiated from the interaction of the enhanced light with
the Si substrate through the residual layer film, RLT information is expected to be
analyzed (Takahashi et al. 2013).
Fig. 30 Nano thickness inspection of RLT using near-field optical interaction
Fig. 31 Simulation model for

near-field optical
enhancement analyses
Numerical Simulation Based on FDTD Method
To analyze the detailed behavior of near-field optical enhancement, numerical

simulations using an FDTD method are applied. Figure 31 shows a typical FDTD
simulation model employed in this research. In this experiment, 5-nm-unit cells with
simulation range of 700 700 1500 nm were set. A metal probe tip made of silver
having a tip diameter of 15 nm was modeled with a gap distance of 10 nm above the
surface of the residual resist layer. As the external light illuminator, a plane wave
with the wavelength of 632.8 nm, assuming an He-Ne laser, was employed, with
incident angle of 45 degrees. Figure 32 shows the numerically calculated intensity
468 S. Takahashi
a b
1.0
Metal probe
Near-field optical
Intensity a.u.
enhancement of metal tip
0.5
Resist
Si substrate
0
Fig. 32 Calculated intensity distribution [P-polarized incident light/XY-plane]. (a) RLT:10 nm. (b)
RLT: 50 nm
distributions on XY-plane including the central axis of the metal probe with RLT of
10 nm (a) and 50 nm (b), respectively, where p-polarized incident light is employed.
We see that an optical intensity enhancement can be generated in the vicinity of the
metal tip, and this enhancement penetrates the residual resist layer.
Next, to carry out detailed analyses of the behavior of the interacted enhancement,
three dimensional electric field vector components were calculated. Figures 33 and
34a and b visualize electric field vector components under p-polarized incident light
with RLT of 10 nm and 50 nm, respectively. Figure 33 shows the result on the XY-
plane including the central axis of the metal probe and Fig. 34 shows the YZ-plane
result. The Ez component (z-component of the electric field) of Fig. 33 and the Ex
component of Fig. 34 cannot be observed well, but both the Ex and the Ey
components of Fig. 33 and both the Ey and the Ez components of Fig. 34 surround
the metal probe with higher amplitude. Taken together, the Ex component and the Ez
component are also enhanced depending on the polarization of the incident light and
the metal tip structure, but only the enhanced Ey component can penetrate the
residual resist layer, which suggests that detecting the Ey component from the
near-field optical enhancement of the metal tip plays an important role for analyzing
the RLT. On the other hand, in the case of s-polarized incident light, distinct
enhancement of the metal tip interacting with the residual resin and the Si substrate
was not found under various conditions. (Figure 35 is a typical example in the case
of s-polarized incident light.)
Next, the intensity of the far-field propagating light radiated from the near-field
optical enhancement, which can be detected for RLT evaluation in a practical situation,
is estimated. Figure 36 shows the relationship between the degree of far-field propa-
gating light intensity and the RLT, which was calculated based on the dipole radiation
enhanced with the interaction of the metal tip and the sample surface under p-polarized
Fig. 33 Three dimensional electric field components under p-polarized incident light [XY-plane].
(a) RLT:10 nm. (b) RLT:50 nm
incident light. Here, as shown in the upper right figure of Fig. 36, the far-field detection
unit was horizontally set with an angle of 45 against the incident optics view from
above, from the practical point of view that both the deterioration of the signal-to-
noise ratio owing to the specular reflection from the sample surface and the conflict of
the incident light optics should be minimized. Even under the p-polarized incident
condition, the s-polarized component of the detection light can be detected, which is
thought to have been generated from the Ex component of Fig. 33 and the Ez
component of Fig. 34. The p-polarized detection light, however, clearly depends on
the RLT compared with the s-polarized component, which is in good agreement with
the numerical results indicating that the Ey component penetrates the residual layer
well, as shown in Figs. 33 and 34 (Takahashi et al. 2013).
RLT Measurement System Based on Near-Field Optical Enhancement

and Verification Experiment
Figure 37 shows an RLT measurement system based on a near-field optical enhance-

ment detection system. This system mainly consists of a metal probe tip control unit, an
XYZ-PZT/stepping motor stage unit for sample positioning, and a confocal microscopic
470 S. Takahashi
Fig. 34 Three dimensional electric field components under p-polarized incident light [YZ-plane].
(a) RLT:10 nm. (b) RLT:50 nm
Fig. 35 Typical result under s-polarized incident light [RLT:50 nm/XY-plane]
unit of infinity corrected optics. An He-Ne laser with the wavelength of 632.8 nm was
employed as a light source. After passing through the half wave plate used to adjust the
polarization state and lens (f = 250), a gently focused laser beam is obliquely incident to
the sample surface. For the first step of verification, commercially available metal probes
with Au coat (JASCO Corp.: NPN-500) designed for near-field scanning optical
Fig. 36 Relationship between estimated far-field intensity and RLT
microscopy (NSOM), having relatively larger size tips (tip diameter: 500 nm), are
employed. A shear-force method using a tuning fork is applied for tip-sample distance
control. To obtain the near-field optical response with high sensitivity, a specially
designed confocal microscopic unit based on infinity corrected optics with a polarizing
beam splitter and two photomultiplier tubes (PMTs) is installed, as shown in Fig. 37,
with which we can detect both the p-polarized and the s-polarized signals simulta-
neously for studying fundamental detection characteristics.
Figure 38 shows the relationship between the intensity of the optical response from
the metal tip and the sample stage height, where the gap distance between the metal tip
and the sample surface was controlled by changing the sample stage height. In the far-
field region, periodical intensity fluctuation caused by standing waves of oblique
illumination and the corresponding specular reflection was observed. When the gap
distance closes to about 100 nm, the detected signal sharply increases, especially in the
case of the p-polarized signal. This suggests that near-field optical enhancement
interacting with the sample surface can be clearly detected by the system.
Finally, to verify the feasibility of the RLT measurement concept, a practical exper-
iment discriminating 10-nm RLT and 50-nm RLT, which are practically useful for
adjusting the RIE conditions of the NIL processes, is shown. Figure 39 shows the
472 S. Takahashi
Function
generator
Z - PZT /
Stepping motor stage
Lock-in
PC
z amp.
Metal probe
He-Ne Lock-in
Laser Mirror Function amp.
Lens generator
Tuning fork
λ/2 plate
PMT
Infinity-corrected
optical system for Pin Hole
near-field optical Polarizing
enhancement detection PMT Beam splitter
XYZ - PZT / Stepping motor stage
Pin Hole Lock-in
amp.
Fig. 37 RLT measurement system based on near-field optical enhancement detection system
Gap distance = Far-field region Near-field region Far-field region

14 45000
40000
12 6DPSOHVWDJH
µP
307RXWSXW µ9
KHLJKW 35000
10
6DPSOH VWDJH KHLJKW
30000
8 25000
3SRODUL]HG
6 GHWHFWLRQ 20000
15000
4
10000
6SRODUL]HG
2
GHWHFWLRQ 5000
0 0
0 100 200 300 400 500 600 700 800 900
6WHSIRUVDPSOHVWDJHVLJQDOQPVWHS
Tip approaching Tip disengaging

Tip contact position
Fig. 38 Relationship between the intensity of the optical response from the metal tip and the
sample stage height
Fig. 39 Nanoimprint sample

(PAK-01)
Fig. 40 Results of 7000

verification experiment
P9
6000
$FFXPXODWHG 307 RXWSXW
5000
4000
3000
2000
1000
0
10 50
5HVLGXDOOD\HUWKLFNQHVV5/7QP
nanoimprint sample (PAK-01) employed in this experiment. Figure 40 indicates the

results with standard deviation bar for eight experiments. It is clear that the near-field
optical response can be detected with less than 5% deviation, and the 10-nm RLT and the
50-nm RLT can be clearly discriminated by our developed system (Takahashi et al.
2013).
Summary and Outlook
In this chapter, in-process measurement of subwavelength structures was described.

Because of its high affinity with in-process measurement, this chapter focused on
optical measurement. A major limitation of the optical measurement method is the
474 S. Takahashi
diffraction limit in resolution; however, it is still important, even in next-generation

advanced manufacturing sites fabricating subwavelength structures owing to its
desirable measurement characteristics such as remote operation, parallel processing,
high-speed operation, and nondestructiveness. Therefore, from the viewpoint of
overcoming the diffraction limit problem, three types of optical measurement
methods for subwavelength structures were discussed in this chapter. As optical
measurement methods based on far-field optics, super-resolution imaging using
structured illumination and depth measurement of microgrooves considering super-
position of complex amplitude vectors were shown. As optical measurement
methods based on near-field optics, nano layer thickness measurement for nano-
imprint lithography using near-field optical signals was discussed. Generally, appli-
cation of near-field optics is mainly limited to the basic science research field, but it
is expected to be applied aggressively to the practical manufacturing field.
References
Aigouy L, Lahrech A, GrÃsillon S, Cory H, Boccara AC, Rivoal JC (1999) Polarization effects in
Apertureless scanning near-field optical microscopy: an experimental study. Opt Lett 24
(4):187–189
Balla T, Spearing SM, Monk A (2008) An assessment of the process capabilities of nanoimprint
lithography. J Phys D Appl Phys 41:174001–174010
Bruzzone AAG, Costa HL, Lonardo PM, Lucca DA (2008) Advances in engineered surfaces for
functional performance. Ann CIRP 57(2):750–769
Calaon M, Hansen HN, Tosello G, Garnaes J, Norregaard J, Li W (2015) Microfluidic chip designs
process optimization and dimensional quality control. Microsyst Technol 21:561–570
Chou SY, Krauss PR, Renstom PJ (1996) Nanoimprint lithography. J Vac Sci Technol B
14:4129–4133
de Groot P, de Lega XC, Liesener J, Darwin M (2008) Metrology of optically-unresolved features
using interferometric surface profiling and RCWA modeling. Opt Exp 16(6):3970–3975
Gao P, Pedrini G, Osten W (2013) Structured illumination for resolution enhancement and auto-
focusing in digital holographic microscopy. Opt Lett 38(8):1328–1330
Gass J, Dakoff A, Kim MK (2003) Phase imaging without 2π ambiguity by multiwavelength digital
holography. Opt Lett 28(13):1141–1143
Gustafsson MGL (2000) Surpassing the lateral resolution limit by a factor of two using structured
illumination microscopy. J Microsc 198(2):82–87
Ann CIRP 55(2):721–743
Ho YL, Portela A, Lee Y, Maeda E, Tabata H, Delaunay JJ (2014) Hollow Plasmonic U-Cavites
with high-aspect-ratio nanofins sustaining strong optical vortices for light trapping and sensing.
Adv Opt Mater 2(6):522–528
Kawata S, Inouye Y (1995) Scanning probe optical microscopy using a metallic probe tip.
Ultramicroscopy 57:313–317
Kudo R, Usuki S, Takahashi S, Takamasu K (2009) Fundamental verification for 2-dimensional
super-resolution optical inspection for semiconductor defects by using standing wave illumina-
tion shift. Proceedings of IMEKO world congress, Lisbon, TC2
Kudo R, Usuki S, Takahashi S, Takamasu K (2011) Experimental analysis of influence of error on
super-resolution optical inspection using standing wave illumination. Proceedings of the 10th
international symposium of measurement technology and intelligent instruments, A5–4, Dae-
jeon, pp 1–6
Lee H (2005) Effect of imprinting pressure on residual layer thickness in ultraviolet nanoimprint
lithography. J Vac Sci Technol 23(3):1102–1106
Lehmann P, Niehues J, Tereschenko S (2014) 3-D optical interference microscopy at the lateral
resolution. Int J Optomechtron 8(4):231–234
Lucy LB (1974) An iterative technique for the rectification of observed distributions. Astron J
79:745–754
Mico V, Zalevsky Z, Garcia-Martinez P, Garcia J (2006) Superresolved imaging in digital holog-
raphy by superposition of tilted wavefronts. Appl Opt 45(5):822–828
Nishioka H, Takahashi S, Takamasu K (2006) A super-resolution microscopy with standing
evanescent light and image reconstruction method. Proceedings of IMEKO world congress,
Rio de Janeiro, TC2
Peng KQ, Wang X, Li L, Wu XL, Lee ST (2010) High-performance silicon Nanohole solar cells.
Am Chem Soc 132(20):6872–6873
Takahashi S (2015) The gap between observation with light and geometric measurement. J Jpn Soc
Mech Eng 118(1161):14–17
Takahashi S, Okada S, Nishioka H, Usuki S, Takamasu K (2008) Theoretical and numerical
analysis of lateral resolution improvement characteristics for fluorescence microscopy using
standing evanescent light with image retrieval. Meas Sci Technol 19:084006
Takahashi S, Ikeda Y, Takamasu K (2013) Study on nano thickness inspection for residual layer of
nanoimprint lithography using near-field optical enhancement of metal tip. CIRP Ann 62
(1):527–530
Takahashi S, Jin C, Ye S, Michihata M, Takamasu K (2017) Theoretical analyses of in-process
depth measurements of fine microgrooves based on near-field optical response. CIRP Ann 66
(1):503–506
Usuki S, Nishioka H, Takahashi S, Takamasu K (2006) Development of super-resolution optical
inspection system for semiconductor defects using standing wave illumination shift. Proceed-
ings of SPIE 6375, Boston, 637508
Usuki S, Nishioka H, Takahashi S, Takamasu K (2008) Super-resolution optical measurement for
ultra-precision machined surface defects by using structured light illumination shift (1st report):
theoretical verification of resolution property. J Jpn Soc Precis Eng 74(5):498–503
Usuki S, Nishioka H, Takahashi S, Takamasu K (2010) Experimental verification of super-resolu-
tion optical inspection for semiconductor defect using standing wave illumination shift. Int J
Adv Manuf Technol 46(9–12):863–875
Wingerden J, Frankena HJ, Smorenburg C (1991) Liner approximation for measurement errors in
phase shifting interferometry. Appl Phys 30(19):2718–2729
Yuan C, Zhai H, Liu H (2008) Angular multiplexing in pulsed digital holography for aperture
synthesis. Opt Lett 33(20):2356–2358
Yuan C, Situ G, Pedrini G, Ma J, Osten W (2011) Resolution improvement in digital holography by
angular and polarization multiplexing. Appl Opt 50(7):B6–b11
Optical Scatterometry for Nanostructure
Metrology 15
Xiuguo Chen and Shiyuan Liu
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 478
Scatterometric Setups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 480
Angular Scatterometer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 480
Spectroscopic Scatterometer . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 482
Other Scatterometers . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 482
Model-Based Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 483
Modeling Light-Nanostructure Interaction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 483
Solving the Inverse Diffraction Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 489
Evaluating the Measurement Uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 495
Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 497
CD Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 497
Overlay and Asymmetry Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 501
Line Roughness Measurement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 508
Conclusion and Outlook . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 511
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 511
Abstract
Process control in microelectronic manufacturing requires real-time monitoring
techniques. Optical scatterometry, also referred to as optical critical dimension
metrology, has become one of the most important techniques for critical dimen-
sion (CD) and overlay metrology in semiconductor manufacturing over the past
decades due to its inherent noncontact, nondestructive, time-effective, and rela-
tively inexpensive merits over other metrology techniques, such as scanning
electron microscopy (SEM) and atomic force microscopy (AFM). As a kind of
X. Chen (*) · S. Liu

School of Mechanical Science and Engineering, Huazhong University of Science and Technology,
Wuhan, China
e-mail: xiuguochen@hust.edu.cn; shyliu@hust.edu.cn

https://doi.org/10.1007/978-981-10-4938-5_17
478 X. Chen and S. Liu
model-based optical metrology, the measurement in optical scatterometry is not

straightforward and typically involves solving a complicated inverse diffraction
problem. However, it is not restricted by the well-known Abbe diffraction limit as
usually encountered in the image-based metrology techniques and thus has
become a good alternative to SEM and AFM in addressing devices with sub-
wavelength feature sizes in semiconductor industry. This chapter aims to give the
engineers and the graduates working in relevant fields some basic knowledge in
optical scatterometry. Emphasis will be given to the two basic steps involved in
the implementation of optical scatterometry, including the measurement of the
optical response of a nanostructure under test by a scatterometer and the modeling
of the light-nanostructure interaction and the extraction of the parameters under
measurement from the measured optical response by solving an inverse diffrac-
tion problem. Some examples on the measurement of CD, overlay, and line
roughness will also be presented to demonstrate the specific applications of
optical scatterometry.
Keywords
Optical scatterometry · Critical dimension · Overlay · Line edge roughness ·
Linewidth roughness · Inverse problem · Rigorous coupled-wave analysis ·
Model-based metrology · Optical metrology · Nanometrology
Introduction
The rapid developments in microelectronics industry drive a continuous and

aggressive reduction in the minimum feature size or critical dimension (CD) of
semiconductor devices. With the improvement in optical lithography techni-
ques, such as extreme ultraviolet (EUV) lithography and multiple-patterning
lithography, devices with CD as small as 10 nm are already achievable.
Moreover, these small device features must be printed accurately on top of the
underlying patterns. A misalignment that results in a so-called overlay error will
lead to non-yielding devices (den Boef 2016). According to the International
Technology Roadmap for Semiconductors (www.itrs2.net), the requirement for
the maximum allowed overlay error is 4 nm (3σ) at today’s feature sizes but will
approach to as small as 2 nm in 2018. In addition, the impact of line roughness on
performances of microelectronic devices (Mittal and Ganguly 2017) has also
become a more and more non-ignorable issue in semiconductor manufacturing
as the technology node approaches to 7 nm, since line roughness does not scale
down with the device dimension and can only be reduced with improved process
conditions.
In order to achieve effective process control, fast, inexpensive, nondestructive,
and reliable nanoscale feature measurements are extremely useful in semi-
conductor manufacturing. Although scanning electron microscopy (SEM), atomic
force microscopy (AFM), transmission electron microscopy (TEM), and CD-SEM
15 Optical Scatterometry for Nanostructure Metrology 479
and CD-AFM that are specialized for measuring CD in semiconductor

manufacturing are capable of resolving ultrasmall features, they are in general
time-consuming, expensive, complex to operate, and problematic in realizing in-
line integrated measurements. Moreover, SEM (also CD-SEM) and TEM often
destroy the sample. Being nondestructive, time-effective, and relatively inexpen-
sive, optical scatterometry, also known as optical critical dimension metrology, has
been successfully introduced for the monitoring of CD and overlay over the past
decades (see citations in the “Applications” section).
Scatterometry has been traditionally known as a technique that analyzes
scattered light from irregular defects and particles aiming to quantify the size
of and frequency of these features. Although the same name is held, the
scatterometry described here is essentially a diffraction-based metrology tech-
nique, where the diffracted light from periodic features such as gratings is ana-
lyzed to reconstruct structural profiles of these features. Compared with the
traditional image-based metrology techniques, which normally require little infor-
mation about specific properties of the sample and can directly acquire the
dimensional information from the best-focus images, the optical scatterometry
does not depend on images with well-defined edges and allows for the indirect
determination of measurands of the sample from the measured signature. Here, the
term signature represents the optical response of the sample under test, which, as
described in the subsequent of this chapter, can be in forms of reflectance,
ellipsometric angles, Mueller matrix elements, etc.
The implementation of optical scatterometry mainly involves two steps
(Raymond 2001, 2005). In the first step, the signature of a nanostructure (typically
a periodic sub-wavelength or deep sub-wavelength structure) is measured using a
proper instrument. The second step involves the parameterization of the profile of
the nanostructure, the establishment of a diffraction model to relate the signature
with the profile parameters, and the extraction of the parameters under measure-
ment from the measured signature by solving an inverse diffraction problem with
the objective of finding an optimal input to the established model whose simulated
signature can best match the measured one. Since the data analysis in optical
scatterometry heavily relies on the established diffraction model, optical
scatterometry can also be regarded as a kind of model-based optical metrology
(Colonna de Lega 2012). Although the measurement in optical scatterometry is not
straightforward and involves solving a complicated inverse diffraction problem, it
is not restricted by the well-known Abbe diffraction limit as usually encountered
in the image-based metrology techniques. Thus, optical scatterometry has gained
widespread applications in semiconductor industry, especially in addressing
devices with sub-wavelength feature sizes.
The following of this chapter is revolved around the above two steps in the
implementation of optical scatterometry. After that, the applications of optical
scatterometry in the measurement of CD, overlay, asymmetric features, and line
roughness are presented. Some conclusions as well as the outlook of this technique
are summarized at the end of this chapter.
Scatterometric Setups
It is well known from Maxwell’s theory of electromagnetism that light is a kind of

electromagnetic waves. The basic properties of light as a wave include amplitude,
frequency (or wavelength), phase, polarization, etc. Mathematically, the electric field
e(r, t) of a plane wave varied with the position vector r and time t can be described
by

^ ,
eðr, t Þ ¼ E exp i ωt k 0 kr (1)
where i is the imaginary unit; ω = 2πc/λ is the angular frequency with c and λ being
the light speed and wavelength in vacuum, respectively; k^ is a unit vector along the
propagation direction; k0 = 2π/λ is the wave number; and the complex amplitude
vector, also called the Jones vector (Jones 1941), is

Ep Ep0 exp iδp
E¼ ¼ , (2)
Es Es0 expðiδs Þ
which consists of the real oscillation amplitudes, Ep0 and Es0, and the initial phases,
δp and δs, and defines the polarization state of light. In Eq. (2), p represents the
component in the direction (^p) parallel to the plane of incidence and perpendicular to
the propagation direction k^ , and s represents the component in the direction ( s^)
perpendicular to the plane of incidence. The directions associated with p^ , s^ , k^
constitute a right-handed coordinate system, which is usually adopted to describe the
polarization state of light. With Eq. (2), the light intensity can be calculated by
I = E†E, where the symbol “†” denotes the conjugate transpose.
Many different types of scatterometric setups have been developed. These
scatterometric setups, according to the propagation direction k^ and the wavelength
λ (or angular frequency ω), can be basically classified into two categories, namely,
the angular scatterometer and the spectroscopic scatterometer (Raymond 2001,
2005; Madsen and Hansen 2016).
Angular Scatterometer
Figure 1 shows the schematic of an angular scatterometer, where a single wavelength

laser source, such as the He-Ne laser source (λ = 632.8 nm), is usually employed.
The laser beam is incident upon a grating sample with an incidence angle of θ and
an azimuthal angle of ϕ. In this case, the propagation direction kînc associated with
the incident beam can be written as kînc ¼ ½ sin θ cos ϕ, sin θ sin ϕ, cos θT
according to the coordinate system defined in Fig. 1. Here, the superscript “T”
represents the transpose, and the azimuthal angle ϕ is defined as the angle between
the plane of incidence and the periodic direction of the grating. The reflected (or
transmitted) beam is comprised of multiple diffracted beams with different propa-
gation directions k^m or, equivalently, different angular positions (θm, ϕm) (m = 0, 1,
z
plane of
Reflectance
laser k̂inc-scan
source incidence
k̂inc detector
kˆ+1 k̂0
q q0
kˆ–1 x Incidence angle (q )
Reflectance
y f kˆm -scan
Λ
Diffraction order (m)
Fig. 1 Schematic of an angular scatterometer
2, . . ., denoting the diffraction orders). The relationship between (θm, ϕm) and
(θ, ϕ) is given by
λ
sin θm cos ϕm ¼ sin θ cos ϕ m , (3)
Λ
where Λ denotes the grating period. When ϕ = 0, it follows that ϕm = 0 for
any diffracted beam. In this case, Eq. (3) degrades to the well-known grating
equation.
In angular scatterometers, it is a common practice to scan kînc or k^m to realize the
reconstruction of the structural profile. In the kînc-scanning technique, the collected
signature by the detector is typically the reflectance (or transmittance) of the
zeroth-order diffracted beam (m = 0) at different incidence angles θ, as schemat-
ically shown on the top right sub-figure of Fig. 1. The angular scatterometer in this
case is also called a 2-θ scatterometer (Raymond 2005). Here, the reflectance
(sometimes also termed as reflectivity) is defined as the ratio of the reflected
light intensity to the incident light intensity. The main reason for the measurement
of the reflectance of the zeroth-order diffracted beam but not other higher-order
diffracted beams is that the zeroth-order diffracted beam usually has a larger
intensity and thus provides a higher signal-to-noise ratio in measurement. In
addition, as can be observed from Eq. (3), the zeroth-order diffracted beam always
exists for gratings with any period. However, the existence of higher-order
diffracted beams is subject to the ratio of the wavelength λ to the grating period
Λ as well as the incidence angle θ. In the k^m -scanning technique, the detector is
scanned (Gross et al. 2009; Calaon et al. 2015), or a high numerical aperture lens is
employed (Boher et al. 2005; Patrick et al. 2007) to collect the reflectance of the
diffracted beams, as schematically shown on the bottom right sub-figure of Fig. 1.
To ensure enough diffraction orders for accurate profile reconstruction, short
wavelength light sources, such as the extreme ultraviolet (EUV) light source
(Gross et al. 2009) and the x-ray light source (Jones et al. 2003), are usually
employed in angular scatterometers based on the k^m -scan.
z
broadband plane of
light source incidence
spectrometer
k̂iinc
kˆ+1 k̂0
Reflectance
θ θ0
kˆ–1 x
y θ
Λ Wavelength (λ)
Fig. 2 Schematic of a spectroscopic scatterometer
Spectroscopic Scatterometer
Figure 2 shows the schematic of a spectroscopic scatterometer. In comparison with

the angular scatterometer, a broadband light source is usually employed in the
spectroscopic scatterometer. The majority of spectroscopic scatterometers collect
the reflectance (or transmittance) of the zeroth-order diffracted beam with similar
reasons to the angular scatterometers based on the kînc-scan. In addition, according to
Eq. (3), the grating disperses the wavelengths into different propagation directions at
higher diffraction orders, while the wavelengths are dispersed into the same direction
at the zeroth-order, which facilitates the signal detection in spectroscopic
scatterometers. To obtain the spectral information contained in the zeroth-order
diffracted beam, as schematically shown on the right side of Fig. 2, a spectrometer
is usually employed in the spectroscopic scatterometer to split the light into a
spectrum.
Other Scatterometers
In the abovementioned scatterometric setups, only the light intensity information is

used, while the more sensitive polarization information is not taken into account.
Polarized light is often used in ellipsometry to characterize thicknesses of thin films
and optical constants of both layered and bulk materials. Specifically, ellipsometry
measures the change in the polarization state of the light when interacting with a
sample, where the change in the polarization of light is usually characterized by two
ellipsometric angles, Ψ and Δ, representing the angle of amplitude ratio and the phase
difference, respectively (Azzam and Bashara 1977; Fujiwara 2007). Since the year of
around 2000 (Niu et al. 2001; Huang et al. 2001), ellipsometry was introduced to
monitor the CD of grating structures in semiconductor manufacturing. Figure 3 shows
the schematic of an ellipsometric scatterometer. Compared with the abovementioned
scatterometric setups, a polarization state generator (PSG) and a polarization state
z
light plane of
source incidence
Ellipsometric angles
PSG detector ψ Δ
PSA
q q0
x
y f
Λ Wavelength/Incidence angle
Fig. 3 Schematic of an ellipsometric scatterometer
analyzer (PSA), as shown in Fig. 3, are mounted in the light input path and output path,
respectively. The PSG and PSA that consist of polarization components such as
polarizers and compensators (also called retarders) are used to modulate and demod-
ulate the polarization states of incident and reflected (or transmitted) beams, respec-
tively. Ellipsometric scatterometers can be angular or spectroscopic, and spectroscopic
ellipsometric scatterometers are the commonly used setups in practice.
The conventional ellipsometry that measures two ellipsometric angles Ψ and Δ is
typically used for the characterization of isotropic samples. When the sample is
anisotropic and/or the measurement process contains depolarization, the conventional
ellipsometry will be unavailable. Here, depolarization describes a phenomenon where
a totally polarized light translates into a partially polarized or completely unpolarized
light. In these cases, the Mueller matrix ellipsometry (MME), also known as Mueller
matrix polarimetry or Mueller polarimetry or Mueller ellipsometry, is required to be
used, which provides 16 elements of a 4 4 Mueller matrix (Fujiwara 2007). Due to
the rich information contained in the Mueller matrix, MME has also been introduced
into optical scatterometry (termed as MME-based scatterometry or Mueller matrix
scatterometry) and demonstrated a great potential in the accurate measurement of CD
and overlay in semiconductor manufacturing (Novikova et al. 2006, 2007; Kim et al.
2009; Chen et al. 2013b, 2014b; Liu et al. 2015).
Model-Based Data Analysis
Modeling Light-Nanostructure Interaction
Many methods can be used for the modeling of light-nanostructure interaction in

optical scatterometry, such as the finite element method (Bao et al. 2005), the boundary
element method (Nakata and Koshiba 1990), the finite-difference time-domain method
(Taflove and Hagness 2005), and the rigorous coupled-wave analysis (RCWA) (also
known as the Fourier modal method) (Moharam et al. 1995a, b; Li 1996a, b). All these
Fig. 4 Representation of
light incident upon a one- plane of incidence
p̂
dimensional binary grating
Einc ϕ
k̂inc
ŝ kˆm
θ
region 1 x
φ
y
d nrd ngr z
region 2 w Λ
methods start from the Maxwell’s equations, and the main difference lies in the adopted
numerical techniques in solving the governing equations and boundary conditions.
Among these methods, RCWA is currently the most commonly used method in optical
scatterometry due to its good convergence and relatively easy implementation in
computer codes. Next, binary gratings are exemplified to concisely show the imple-
mentation of RCWA in calculating the signatures of nanostructures in optical
scatterometry. Detailed derivations about RCWA can be found in Moharam et al.
(1995a, b) and Li (1996a, b) as well as other related literature.
Figure 4 shows the schematic one-dimensional binary grating with a period of Λ,
a linewidth of w, and a line height of d. In the calculation, Fig. 4 is divided into three
regions, region 1 (the region of incidence and reflection that satisfies z < 0 according
to the coordinate system defined in Fig. 4), region 2 (the region of transmission that
satisfies z > d), and the grating region (0 < z < d ). The refractive indices of region 1
and region 2 are assumed to be n1 and n2, respectively. The grating region consists of
the periodic distribution of the grating ridge and the grating groove, whose refractive
indices are assumed to be nrd and ngr, respectively. The relative permittivity in the
grating region can be expanded in Fourier series as
X
eðxÞ ¼ eh expðihKxÞ, (4)
h
where K = 2π/Λ and eh is the h-th Fourier component of the relative permittivity in
the grating region and is given by
(
n2rd f þ n2gr ð1 f Þ h¼0
eh ¼ , (5)
n2rd n2gr sin ðπhf Þ=ðπhÞ h 6¼ 0
where f = w/Λ. Since e(x) is a periodic function, its reciprocal 1/e(x) is also a periodic
function, which can be expanded in Fourier series as
1 X
¼ ρh expðihKxÞ, (6)
e ð xÞ h
where ρh is given by
(
f =n2rd þ ð1 f Þ=n2gr h¼0
ρh ¼ : (7)
1=n2rd 1=n2gr sin ðπhf Þ=πh h 6¼ 0
As shown in Fig. 4, the propagation direction of the incident light is denoted as

kînc with incidence and azimuthal angles of θ and ϕ, respectively. The angle φ is the
angle between the incident electric-field vector Einc and the plane of incidence,
which defines the linear polarization state of the incident light. Specifically, φ = 0
corresponds to the transverse magnetic polarization or p-polarization, and φ = 90
corresponds to the transverse electric polarization or s-polarization. The normalized
incident electric-field vector Einc can be written as

^
Einc ¼ uexp ik 0 kînc r , (8)
where u^ denotes the normalized amplitude vector of Einc and is given by
u^ ¼ cos φ p^ þ sin φ s^
¼ ð cos φ cos θ cos ϕ sin φ sin ϕÞ^
x þ ð cos φ cos θ sin ϕ þ sin φ cos ϕÞ^
y (9)
cos φ sin θ z^
According to the coordinate system defined in Fig. 4, x^, y^, and z^ denote the unit
vectors along the x, y, and z directions, respectively. Note that the temporal term exp
(iωt) is ignored in Eq. (8), since the monochromatic plane wave is the focus of the
discussion here.
The electric fields in region 1 (z > 0) and region 2 (z < d), E1 and E2, can be
written as the superstition of a series of plane waves (i.e., diffracted beams)
(Moharam et al. 1995a)
X
E1 ¼ Einc þ Eref ¼ Einc þ Rm exp i k xm x þ k y y k 1, zm z , (10a)
m
X

E2 ¼ T m exp i k xm x þ k y y þ k 2,zm ðz d Þ , (10b)
m
where Eref denotes the normalized reflected electric-field vector; Rm is the normal-
ized electric-field vector amplitude of the m-th backward-diffracted (reflected) wave
in region 1; and Tm is the normalized electric-field vector amplitude of the m-th
forward-diffracted (transmitted) wave in region 2. The quantities kxm, ky, and kl, zm
(l = 1, 2) denote the components of the wavevector of the m-th diffracted beam along
the x, y, and z directions, respectively, and are given by
k xm ¼ k 0 n1 sin θ cos ϕ mK, (11a)
k y ¼ k 0 n1 sin θ sin ϕ, (11b)

8 qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
<þ ðk 0 nl Þ2 k 2xm k 2y ðk 0 nl Þ2 > k 2xm þ k 2y
k l,zm ¼ q ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi : (11c)
: i k 2 þ k 2 ðk n Þ2 k 2xm þ k 2y > ðk 0 nl Þ2
xm y 0 l
The azimuthal angle ϕm between the m-th diffracted beam and the plane of
incidence satisfies

ϕm ¼ arctan k y =k xm : (12)
Note that Eq. (3) is derived from Eq. (11a) here.

The electric and magnetic fields, Eg and Hg, in the grating region (0 < z < d) may
be expressed as the Fourier expansion in terms of space-harmonic fields as
X
Eg ¼ S xm ðzÞ^
x þ S ym ðzÞ^
y þ S zm ðzÞ^
z exp i k xm x þ k y y , (13a)
m
rffiffiffiffiffi X
ϵ0
H g ¼ i U xm x^ þ U ym ðzÞ^
y þ U zm ðzÞ^
z exp i k xm x þ k y y , (13b)
μ0 m
where Sm(z) = [Sxm(z), Sym(z), Szm(z)]T and Um(z) = [Uxm(z), Uym(z), Uzm(z)]T are
the normalized electric-field and magnetic-field vector amplitudes of the m-th space-
harmonic field, respectively, and ϵ0 and μ0 are the permittivity and permeability in
vacuum, respectively. The fields Eg and Hg satisfy the Maxwell’s equations as
follows:
∇ H g ¼ iωϵ0 eðxÞEg , (14a)
∇ Eg ¼ iωμ0 H g : (14b)
Substituting Eq. (13) into Eq. (14), removing the field components Egz and Hgz,
and meanwhile taking into account of the inverse rule (Li 1996a), the coupled-wave
equation in a matrix form can be derived as
2 3 2 32 3
@S y =@ ðz0 Þ 0 0 Ky E1 Kx I Ky E1 Ky Sy
6 @S x =@ ðz0 Þ 7 6 0 Kx E1 Kx I Kx E Ky 7
1 6 7
6 7 6 0 76 S x 7,
4 @U y =@ ðz0 Þ 5 ¼ 4 Kx Ky P1 K2y 0 0 54 U y 5
@U x =@ ðz0 Þ K2x E Kx Ky 0 0 Ux
(15)
where z0 = k0z; E and P are the Toeplitz matrices consisting of the Fourier compo-
nents eh and ρh with the (i, j) entries being ei j and ρi j, respectively; Kx and Ky are
diagonal matrices with the i-th elements being kxi/k0 and ky/k0, respectively; 0 and I
Fig. 5 Schematic of the x

stratification of a surface- 1st layer
relief grating 2nd layer
3rd layer z
......
Λ
denotes the null and identity matrices, respectively; and the superscript “1” denotes
the matrix inverse. Equation (15) can then be solved by using the methods in solving
an ordinary differential equation, which usually involve the calculation of the
eigenvalues and eigenvectors of the coefficient matrix in Eq. (15). The determination
of the undetermined coefficients in the solution of Eq. (15) requires the application
of the continuity conditions of the tangential components of electric and magnetic
fields (i.e., Ex, Ey, Hx, Hy) at the upper (z = 0) and lower (z = d) boundaries of the
grating region.
The above equations show the modeling of a binary grating. For the calculation
of a surface-relief grating, as shown in Fig. 5, it is required to first stratify the
grating profile along the z direction, and then each slice can be approximately
addressed by the above equations for a binary grating, and finally the continuity
conditions of the tangential field components are applied to the upper and lower
boundaries of each slice to connect all the related linear equations (Moharam et al.
1995b). The calculation process of a surface-relief grating or a multilayer grating
is apt to lose stability due to the accumulation of numerical errors generated in the
solution of the eigenvalue problem associated with each slice or each layer. Many
numerical techniques have been proposed to address the above issue, such as the
enhanced T-matrix algorithm (Moharam et al. 1995b), the R-matrix algorithm, and
the S-matrix algorithm (Li 1996b), of which the S-matrix algorithm is generally
preferred to other algorithms.
Based on the RCWA calculation, a 2 2 Jones matrix J associated with the
m-th order reflected (or transmitted) wave of the grating under test can be obtained,
which connects the incoming Jones vector with the reflected Jones vector and is
formulated as

E ref , E r rps, E inc,
pm
¼ J inc, p
¼ pp, m m p
: (16)
E ref , sm Einc, s rsp, m rss, m E inc, s
To obtain the elements of the Jones matrix J, RCWA calculations are carried out
by letting φ = 0 and φ = 90 , respectively. Specifically, when φ = 0 , Einc, p = 1 and
Einc, s = 0 according to Eq. (9). According to Eq. (16), it follows that

Eref , pm φ ¼ 0
rpp, m ¼ ¼ Rsm φ ¼ 0 , (17a)
Einc, p

Eref , sm φ ¼ 0
rsp, m ¼ ¼ Rpm φ ¼ 0 , (17b)
Einc, p
where Rsm and Rpm are given by

Rsm ¼ cos ϕm Rym sin ϕm Rxm , (18a)
i
Rpm ¼ ½cos ϕm ðk xm Rzm þ k 1,zm Rxm Þ þ sin ϕm ðk y Rzm þ k 1,zm Rym Þ: (18b)
k0
The quantities Rxm, Rym, and Rzm are the components of Rm in Eq. (10a) along the
x, y, and z directions, respectively. By the same token, substituting Einc, p = 0 and
Einc,s = 1 when φ = 90 into Eq. (16) gives the results

E ref , pm φ ¼ 90
rps, m ¼ ¼ Rsm φ ¼ 90 , (19a)
E inc, p

Eref , sm φ ¼ 90
rss, m ¼ ¼ Rpm φ ¼ 90 : (19b)
E inc, p
After obtaining the elements of the Jones matrix, the reflectance R can be
calculated by

rpp, m 2 þ rss, m 2
R¼ : (20)
2
The two ellipsometric angles Ψ and Δ are calculated by
rpp, m
ϱ ¼ tan Ψ expðiΔÞ ¼ : (21)
rss, m
In the absence of depolarization, the 4 4 Mueller matrix M is calculated by

Azzam and Bashara (1977):
2 3
M 11 M 12 M 13 M 14
6 M 21 M 22 M 23 M 24 7 O
M¼6 4 M 31 M 32 M 33 M 34 5
7¼A J J A1 , (22a)
M 41 M 42 M 43 M 44
N
where the symbol denotes the Kronecker product, J* is the complex conjugate of
J, and the constant matrix A is given by
2 3
1 0 0 1
6 1 0 0 1 7
A¼6 40 1 1
7: (22b)
0 5
0 i i 0
In practice, the Mueller matrix M is usually normalized to (1, 1) element M11

(M11 = R), with the normalized Mueller matrix element being mij = Mij/M11
(i, j = 1, 2, 3, 4). Specially, when the azimuthal angle ϕ = 0 , namely, the
grating lines are perpendicular to the plane of incidence, the two off-diagonal elements
of the Jones matrix J vanish, and the Mueller matrix M in this case can be simply
expressed as
2 3
1 cos 2Ψ 0 0
6 cos 2Ψ 1 0 0 7
M ¼ R6 4
7: (23)
0 0 sin 2Ψ cos Δ sin 2Ψ sin Δ 5
0 0 sin 2Ψ sin Δ sin 2Ψ cos Δ
Solving the Inverse Diffraction Problem
Without loss of generality, the structural parameters under measurement are

represented as a M-dimensional vector x = [x1, x2, . . ., xM]T, where x1, x2, . . ., xM
can be the linewidth, line height, sidewall angle, etc. of the grating sample. The
measured signature is given by an N-dimensional vector y = [y1, y2, . . ., yN]T, where
y1, y2, . . ., yN can be in terms of reflectance (or transmittance), ellipsometric angles,
Mueller matrix elements, etc. collected by the abovementioned scatterometric
setups. The corresponding theoretical signature calculated by RCWA for any struc-
tural parameter vector x is given by F (x) = [f1(x), f2(x), . . . , fN(x)]T. The χ 2
function (Press et al. 2007) is usually applied to estimate the fitting error between
the measured and calculated signatures, which is defined by
X
N
χ2 ¼ wk ½yk f k ðxÞ2
k¼1
(24)
T
¼ ½y F ðxÞ W½y F ðxÞ,
where wk (k = 1, 2, . . ., N) are the weighting factors, which are usually chosen to be

wk = 1/σ 2(yk) with σ 2(yk) being the variances of the measured signature, and W is an N N
diagonal matrix with diagonal elements being wk. The inverse diffraction problem in
optical scatterometry is typically formulated as a least squares regression problem such that
n o
x^ ¼ argmin ½y F ðxÞT W½y F ðxÞ , (25)
xΩ
where x^ is the solution of the inverse diffraction problem and Ω is the associated
parameter domain.
Two kinds of methods can be used to solve Eq. (25), namely, the nonlinear
regression method and the library search method (Raymond et al. 2004). In the
nonlinear regression method, the structural parameter vector x is iteratively
adjusted by some optimization algorithms (Press et al. 2007), such as the
commonly used Levenberg-Marquardt algorithm (Levenberg 1944; Marquardt
1963), so as to make the calculated signature F (x) best match the measured
signature y. The nonlinear regression method could achieve the expected solution
with a high accuracy provided that a proper initial value is assigned to x. Since
the expected solution is achieved through an iterative procedure and the RCWA
calculation needs to be carried out in each iteration, the nonlinear regression
method is usually time-consuming, especially when addressing two-dimensional
periodic nanostructures, which makes it not well-suited for in situ applications.
In the library search method, a signature library is generated prior to the mea-
surement, and then the signature library is searched to find a theoretical
signature that can best match the measured signature. The structural parameter
vector that corresponds to the best match theoretical signature is taken to be
solution of the inverse diffraction problem. Although the offline generation of
the signature library is time-consuming, the search itself during the online mea-
surement can be done quite efficiently, which makes the library search method
well-suited for in situ applications. The solution accuracy of the library search
method depends on the structural parameter interval in constructing the signature
library. A finer grid interval will yield a higher solution accuracy but will also
increase the burden on time and space resources to generate and store the
signatures.
The Nonlinear Regression Method

As already mentioned, the expected solution in the nonlinear regression method is
achieved through an iterative procedure. In each iteration, the structural parameter
vector x will be replaced with a new value x + δ, where δ denotes the iterative step.
To determine δ, assume that the function F (x) is sufficiently smooth; expanding
F (x + δ) into Taylor series up to the first order at x gives
F ðx þ δÞ ¼ F ðxÞ þ J x δ, (26)
where J x is an NM Jacobian matrix with the (i, j) element being
@f i ðxÞ
½J x ij ¼ : (27)
@xj
Substituting the new value x + δ into Eq. (24) gives
χ 2 ðx þ δÞ ¼ ½y F ðx þ δÞT W½y F ðx þ δÞ

(28)
¼ ½y F ðxÞ J x δT W½y F ðxÞ J x δ:
Taking the partial derivative of both sides of Eq. (28) with respect to δ gives
J~ x δ ¼ W1=2 ½y F ðxÞ, (29)
where J~ x ¼ W1=2 J x can be regarded as the weighted Jacobian matrix. According

to Eq. (29), it follows that
þ
δ ¼ J~ x W1=2 ½y F ðxÞ, (30)
þ
T 1 T
where J~ x ¼ J~ x J~ x J~ x is the Moore-Penrose pseudo-inverse of J~ x .
Equation (30) can be used to estimate the iterative step. However, since the matrix
J~ x J~ x is not always full rank, the estimation of the iterative step by Eq. (30) will not
T
ensure a convergent result. Instead of using Eq. (30), the Levenberg-Marquardt algo-
rithm applies the following equation to estimate the iterative step (Marquardt 1963):
T 1 T
δ ¼ J~ x J~ x þ αI J~ x W1=2 ½y F ðxÞ, (31)
where I is a M M identity matrix and α (α > 0) is a damping factor, which is

iteratively adjusted in implementation. Specifically, when χ 2 decreases fast, a rela-
tively small value is assigned to α. In this case, Eq. (31) approaches to Eq. (30).
When χ 2 decreases slow, a relatively large value is assigned to α. In this case, Eq.
(31) approaches to δ ¼ J~ x W1=2 ½y F ðxÞ, which is usually adopted in the steepest
T
descent method. Note that the matrix J~ x J~ x þ αI in Eq. (31) is always a full rank
T
matrix. Figure 6 presents the flowchart of the Levenberg-Marquardt algorithm,

which is interpreted as follows (Press et al. 2007):
Step 1: Set proper initial values to x and α (say α = 103), and prescribe a proper
threshold ξ (ξ 0) for the fitting error χ 2.
Step 2: Calculate the fitting error χ 2(x) by Eq. (24).
Step 3: Calculate the matrices J~ x , J~ x J~ x , and J~ x J~ x þ αI.
T T
Step 4: Calculate the iterative step δ and the fitting error χ 2(x + δ) by Eqs. (31) and
(28), respectively.
Step 5: If χ 2(x + δ) χ 2(x) > ξ, increase α (say by a factor of 10), and go back to
Step 4; if ξ < χ 2(x + δ) < χ 2(x), decrease α (say by a factor of 1/10), and replace x
with x + δ, and then go back to Step 3.
Step 6: If χ 2(x)
ξ, stop the iteration and output the result.
The Library Search Method

In the conventional library search method, the solution of the inverse diffraction
problem is directly searched from the grid points of the signature library (Raymond
et al. 2004). Thus, there will be a deterministic error between the expected solution
of the inverse diffraction problem and the actually searched solution by the conven-
tional library search method due to the limited grid interval of the signature library.
The solution accuracy of the conventional library search method can be improved by
decreasing the grid interval. However, it will take much more time and memory
space to generate and store the signatures. To improve the solution accuracy of the
library search method for a pre-generated signature library, a gradient correction-
based library search method is presented here (Chen et al. 2013a).
Fig. 6 Flowchart of the

Levenberg-Marquardt
algorithm
For convenience, xe is assumed to be an estimate of the true value of x in a

continuous space of solutions of the inverse diffraction problem, which minimizes
Eq. (24) and satisfies
χ 2min ¼ ½y F ðxe ÞT W½y F ðxe Þ: (32)

Note that xe is also the expected solution of the inverse diffraction problem that is
attempted to achieve. Assume that the actually searched solution by a conventional
library search method is denoted as xs, which corresponds to the discrete grid point
in the pre-generated signature library whose theoretical signature can best match the
measured one. Obviously, xs is also an estimate of the true value of x but is in a
discrete space of solutions of the inverse problem. In addition, the searched solution
xs is close to the expected solution xe. However, due to the limitation of the grid
interval of the library, there will be a deterministic error Δx between the solutions xs
and xe, which is defined by
Δx ¼ xe xs : (33)
Assume that the function F (x) is sufficiently smooth; expanding F (x) into
Taylor series up to the first order at x = xs gives
F ðxÞ ¼ F ðxs Þ þ J x ðx xs Þ, (34)
where J x was given in Eq. (27) but here is calculated at x = xs. Substituting x = xe
in Eq. (34) gives
F ðxe Þ ¼ F ðxs Þ þ J x ðxe xs Þ ¼ F ðxs Þ þ J x Δx: (35)
Inserting Eq. (35) into Eq. (32) gives
χ 2min ¼ ½y F ðxs Þ J x ΔxT W½y F ðxs Þ J x Δx: (36)
Taking the partial derivative of both sides of Eq. (36) with respect to x then gives
J~ x Δx ¼ W1=2 ½y F ðxs Þ: (37)
According to Eq. (37), it follows that
þ
Δx ¼ J~ x W1=2 ½y F ðxs Þ, (38)
which gives the estimate of the error Δx between the solutions xs and xe. The
estimate of the error Δx can be further used as a correction term to correct the
directly search solution xs such that
þ
xc ¼ xs þ J~ x W1=2 ½ y F ðxs Þ, (39)
where xc denotes the corrected solution of the inverse diffraction problem.

Figure 7 presents a geometrical illustration of the corrected solution xc given by
Eq. (39), where the points A and B correspond to the expected and directly searched
solutions of the inverse diffraction problem, respectively. As shown in Fig. 7, the
curve around point B can be approximated by a straight line with a slope of J~ x at
x = xs. The term W1/2[y F (xs)] in Eq. (37) denotes the residual signature between
Fig. 7 A geometrical
illustration of the corrected
solution xc to the inverse
diffraction problem (Chen
et al. 2013a)
Residual signature
B
W1 2 [ y - F ( xs )]
~
Slope J x
Dx A
xs xc xe x
the points A and B. According to the slope J~ x as well as the residual signature
W1/2[y F (xs)], the corrected solution xc given by Eq. (38) can be readily derived.
As depicted in Fig. 7, the corrected solution xc will be much closer to the expected
solution xe than the directly searched solution xs by a conventional library search
method and therefore improves the solution accuracy of the inverse diffraction
problem.
Based on the above description, the basic procedure of the gradient correction-
based library search can be illustrated as follows (Chen et al. 2013):
Step 1: Find the solution xs by a conventional library search method.

Step 2: Estimate the error Δx between the searched solution xs and the expected
solution xe according to Eq. (38).
Step 3: Correct the searched solution xs by using the estimate of the error Δx
according to Eq. (39), and take the corrected solution xc as the final solution of
the inverse diffraction problem.
The key in the correction-based library search method is the estimation of

the error Δx given by Eq. (39), which involves the calculation of the Jacobian
þ
matrix J x as well as the pseudo-inverse J~ x of the weighted Jacobian matrix.
The elements of J x defined in Eq. (27) are estimated in the correction-based library
search method by

@f i ðxÞ f i xs þ δ j e j f i xs δ j e j
¼ þ O δ2j , (40)
@xj x¼xs 2δj
where ej is a unit vector in the j-th coordinate direction and δj is the grid interval of the j-th
structural parameter in the library. The theoretical signatures F (xs + δjej) and F (xs δjej)
in Eq. (40) can be directly searched from the grid points of the pre-generated signature
library; thus the time-consuming RCWA calculation is avoided in the estimation of the
error Δx. The only extra time cost in the correction-based library search method is
introduced by the matrix multiplication, which is ignorable. Therefore, the correction-
based library search method is expected to achieve a higher solution accuracy for a pre-
generated signature library without remarkable influence on the final solution speed.
Evaluating the Measurement Uncertainty
The scatterometric measurement process inevitably contains errors. Due to the

presence of errors, the final extracted structural parameters from the measured
signature inevitably contain uncertainties. For the convenience of uncertainty eval-
uation, the theoretical signature calculated by RCWA for any structural parameter
vector x is rewritten as F (x, a) = [f1(x, a), f2(x, a), . . ., fN(x, a)]T. The vector a = [a1,
a2, . . ., aL]T consists of all of the other L fixed parameters that are input into the
RCWA model in parameter extraction, where a1, a2, . . ., aL can be measurement
conditions, optical constants of materials, structural parameters that have known
values, etc.
The errors in the scatterometric measurement typically arise from two sources,
namely, the errors in the RCWA model and the errors in the scatterometric instru-
ment (Germer et al. 2009; Chen et al. 2014a). The former one mainly arise from the
bias Δa in vector a, which is defined as
Δa ¼ a0 a (41)
where a and a0 represent the priori and true values of a, respectively. The latter one
may arise from the imperfections in the instrument design, manufacturing, assembly,
calibration, etc. Due to the errors in the instrument, the actually measured signature y
shall be the sum of the true signature y0 and the signature error Δy, which can be
expressed by
y ¼ y0 þ Δy: (42)
The errors Δa and Δy will be both propagated into the extracted structural
parameters x^ in the solution of the inverse diffraction problem such that
x^ ¼ x0 þ Δx, (43)
where Δx denotes the error between the extracted structural parameter values x^ and
the true parameter values x0.
Assume that the function F (x, a) is sufficiently smooth; expanding F (x,a) into
Taylor series up to the first order at ðx^, a Þ gives
F ðx, aÞ ¼ F ðx^, a Þ þ J x ðx x^Þ þ J a ða a Þ, (44)
where J x and J a are N M and N L Jacobian matrices with respect to x and a,

respectively, whose elements are defined by

@f ðx,aÞ
½J x ij ¼ i , (45a)
@xj x¼ x^ , a¼a

@f ðx,aÞ
½J a ij ¼ i : (45b)
@aj x¼ x^ , a¼a
Substituting x = x0 and a = a0 into Eq. (44) leads to
F ðx0 , a0 Þ ¼ F ðx^, a Þ þ J x Δx þ J a Δa: (46)
Inserting Eq. (46) into Eq. (24) and noticing that y0 = F (x0, a0) and y = y0 + Δy
in the meanwhile give the result
χ 2min ¼ ½y F ðx^, a ÞT W½y F ðx^, a Þ

(47)
¼ ½J x Δx þ J a Δa þ ΔyT W½J x Δx þ J a Δa þ Δy:
Taking the partial derivative of both sides of Eq. (47) with respect to x gives
J~ x Δx þ J~ a Δa þ Δy~ ¼ 0, (48)
where J~ x ¼ W1=2 J x , J~ a ¼ W1=2 J a , and Δy~ ¼ W1=2 Δy. Equation (48) is called
the error propagating formula, which relates the error Δx in the extracted structural
parameters x^ with the error sources Δa and Δy (Chen et al. 2014a).
Assume that Δa and Δy are independent; according to Eq. (48), the variance-
covariance matrix of x^ can be derived such that
þ
þ T þ
þ T
Cðx^Þ ¼ CðΔxÞ ¼ J~ x W1=2 CðΔyÞW1=2 J~ x þ J~ x J~ a CðΔaÞJ~ a J~ x ,
T
(49)
þ
where J~ x is same to that given in Eq. (30) and C(Δy) and C(Δa) represent the variance-
covariance matrices of Δy and Δa, respectively. Note that the first and second terms of the
right side of Eq. (49) correspond to the separate contributions of Δy and Δa, respectively.
Assume that Δy is subject to a Gaussian distribution with variances of σ 2(yi) and the
measurements are uncorrelated for different yi (i = 1, 2, . . ., N); the variance-covariance
matrix of Δy can then be simply written as a diagonal matrix in the form of

CðΔyÞ ¼ diag σ 2 ðy1 Þ, σ 2 ðy2 Þ, . . . , σ 2 ðyN Þ : (50)
Substituting Eq. (50) into the first term of the right side of Eq. (49) gives
þ
þ T T 1
J~ x W1=2 CðΔyÞW1=2 J~ x ¼ J~ x J~ x : (51)
Since the diagonal elements of the variance-covariance matrix Cðx^Þ correspond to

variances of the extracted structural parameters, the standard deviation of the
extracted structural parameter xj ( j = 1, 2, . . ., M) can be estimated by
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
σ xj ¼ ½Cðx^Þjj : (52)
According to Eq. (52), the measurement uncertainty in parameter xj is given by

u xj ¼ κσ xj , (53)
where κ is the coverage factor and κ = 1, 2, and 3 correspond to a confidence level of

68.3%, 95.4%, and 99.7%, respectively. According to Cðx^Þ, the correlation coeffi-
cient between parameters xi and xj (i, j = 1, 2, . . ., M) can estimated by
½Cðx^Þij
ρ xi , xj ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiqffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi , 0
ρ xi , xj
1, (54)
½Cðx^Þii ½Cðx^Þjj
which can be used to estimate the degree of linear correlation between xi and xj, with
ρ(xi, xj) = 0 meaning no correlation and with |ρ(xi, xj)| = 1 meaning a perfect linear
correlation.
Equations (49, 50, 51, 52, 53) illustrate the evaluation of uncertainties in the
extracted structural parameters from the measured signature by optical
scatterometry, which is compatible with the Guide to the Expression of Uncertainty
in Measurement (GUM) (JCGM 2008). According to the GUM, the first and second
terms of the right side of Eq. (49) can be regarded as the type A and type B
evaluations of uncertainty, respectively. It should be noted that, in semiconductor
manufacturing, a more comprehensive measure named the total measurement uncer-
tainty (TMU) is usually adopted to evaluate the measurement uncertainty for optical
scatterometry. In the TMU-based uncertainty evaluation, a measurement is made by
a specific scatterometry tool relative to a reference measurement system such as CD-
SEM, CD-AFM, and TEM while properly removing the measurement uncertainty of
the reference system. The TMU analysis can be used to optimize a metrology
system, assess the correlations, or compare the discrepancies of different measure-
ment methodologies. Details about TUM analysis are ignored here but can be found
in Sendelback and Archie (2003) and other related literature.
Applications
CD Measurement
One of the typical applications of optical scatterometry is for CD measurements of

submicron or sub-wavelength structures present at lithographic steps in the production
of microelectronic devices. Figure 8 presents one of such kind of structures that had
a CD
resist b 0.2
ARC
0.15
Reflectance
poly-Si
0.1
SiO2
0.05
Si substrate theory
experiment
0
0 10 20 30 40 50 60
Incidence angle [deg]
Fig. 8 (a) The film process stack used on a submicron structure. The unknown parameters to be
determined by scatterometry are the CD, resist height, ARC thickness, and poly-Si thickness; (b)
fitting result between the experimental and theoretical best-fit angular reflectance for a 350 nm line
sample (Raymond et al. 1997)
ever been investigated by a 2-θ angular scatterometer (namely, the kînc -scan
scatterometer shown in Fig. 1) employed in a He-Ne laser source (Raymond et al.
1997). As shown in Fig. 8a, on a Si substrate, there was a SiO2 film followed by a poly-
Si layer, an antireflection coating (ARC), and finally the APEX-E resist structures. The
unknown parameters to be determined by scatterometry included the CD (linewidth),
resist height, ARC thickness, and poly-Si thickness, with the SiO2 thickness fixed at
the nominal value of 65 Ǻ due to its low sensitivity to the angular reflectance measured
by the 2-Θ scatterometer. In the experiment, 25 wafers were prepared, and a 9 9
focus/exposure (F/E) matrix was printed on each wafer with different exposure doses.
Moreover, each F/E location comprised two different line/space gratings. The first one
had nominal 350 nm resist lines with an 800 period. The second one had 250 nm resist
lines with a 750 nm period. The grating period was also fixed in parameter extraction.
The conventional library search method was used to extract the unknown parameters.
In construction of the signature library, the CD for each sample was varied in 10 nm
steps, from 150 to 400 nm for nominally 250 nm lines and from 250 to 550 nm for
nominally 350 nm lines. The resist height was from 650 to 750 nm in steps of 10 nm.
The ARC thickness and the poly-Si thickness were varied from 700 to 850 nm and 220
to 290 nm in steps of 2.5 and 5 nm, respectively. As an example, Fig. 8b presents the
fitting result between the experimental and theoretical best-fit angular reflectance for a
nominal 350 nm line grating.
Figure 9 presents the CD measurements as a function of the exposure dose for 2
of the 25 wafers, wafer 12 and wafer 17. Apart from the scatterometry results, the
measurement results by top-down SEM and cross-section SEM were also pre-
sented to make a comparison. As can be observed from Fig. 9, the correlation
between scatterometry and cross-section SEM measurements was excellent over
the whole CD range. The error bars appended to the scatterometry results
(10 nm) were due to the 10 nm CD step size (grid interval) in construction of
a 500 b 400
Scatterometry Scatterometry
450 Cross section-SEM 350 Cross section-SEM
Top-down SEM Top-down SEM
400 300
CD [nm]
CD [nm]
350 250
300 200
250 150
200 100
12.2 12.4 12.6 12.8 13 13.2 13.4 13.6 13.8 12 12.4 12.8 13.2 13.6 14
Exposure dose [mJ] Exposure dose [mJ]
Fig. 9 Measured CD under different exposure doses. (a) Wafer 12, nominal 350 nm lines;
(b) wafer 17, nominal 250 nm lines (Raymond et al. 1997)
the signature library. The uncertainties of the top-down SEM and cross-section
SEM measurements were estimated to be 20 nm and 8 nm, respectively. As can
be observed, the scatterometry and cross-section SEM measurements agreed
within the error range of the two techniques. The average difference between the
scatterometry and cross-section SEM measurements for wafer 12 (nominal 350 nm
lines) was 1.7 nm and for wafer 17 (nominal 250 nm lines) was 7.3 nm. A
systematic bias could be observed from Fig. 9 for the top-down SEM measure-
ments, which were attributed to the following possible reasons. First, resist
charging degraded SEM images and contributed to CD measurement errors.
Second, the way in which the top-down SEM measurement algorithm interpreted
the edges of resist lines from the raw electron signal might also contribute to a
systematic error. Besides CD, it had been demonstrated that other unknown
parameters such as resist height, ARC thickness, and poly-Si thickness could
also be simultaneously determined. In addition, the repeatability (1σ) was shown
to be subnanometer for all of the four unknown parameters.
The above example demonstrates an ex situ application of optical scatterometry.
Optical scatterometry could also be used for in situ and real-time control of different
process steps in semiconductor device manufacturing. One example of the in situ
application of optical scatterometry is for monitoring the structural profile evolution
during the resist trimming process (Kodadi et al. 2009). The resist trimming process
induces a controlled lateral erosion of the photoresist mask so as to reach the resolution
required by the advanced technology nodes. A multiwavelength ellipsometric
scatterometer was employed to monitor the profile evolution during the resist trimming
process. Instead of measuring the ellipsometric angles Ψ and Δ, two derived param-
eters Is and Ic defined by Is = sin2ΨsinΔ and Ic = sin2ΨcosΔ, respectively, were
collected. A comparison with Eq. (23) reveals that Is and Ic are actually the normalized
Mueller matrix elements m34 (or m43) and m33 (or m44), respectively. As depicted in
Fig. 10a, the employed ellipsometric scatterometer was plugged onto a DPS
(decoupled plasma source) process chamber in the experiment. The two derived
Fig. 10 (a) Overview of the a

platform of the DPS process DPS
chamber equipped with an in input slot
situ ellipsometric
scatterometer; (b) geometrical scatterometer
model of the multilayer
grating used for construction orientation
of signature library (Kodadi et chamber
al. 2009)
b CD facetted profile
resist height
sidewall
SiO2 angle
Si substrate
ellipsometric parameters were measured at 16 wavelengths ranging from 193 to

813 nm by fixing the incidence angle at 72.32 . Figure 10b shows the profile model
of the multilayer grating formed in the resist trimming process for construction of the
signature library, which contains four parameters needed to be determined, namely,
CD, height, sidewall angle, and profile faceting. Three-dimensional (3D) AFM mea-
surements were carried out to make a comparison with the scatterometry measurement
results. Figure 11 presents the comparison between the extracted parameters (CD and
height) and reconstructed feature profiles by scatterometry and the AFM measurement
results at different etching times. A good agreement between scatterometry and AFM
measurement results can be easily observed.
Another example of the in situ application of optical scatterometry is for moni-
toring time evolution of the profile of nanoimprinted lines in low and high molecular
mass polymer gratings during reflow at the glass transition temperature (Patrick et al.
2008). Nanoimprinted lithography (NIL), in which features on a prepatterned mold
are transferred directly into a polymer material, represents a promising manufactur-
ing technique with the potential of high resolution and throughput as well as low
cost. The measurement of the profile evolution (reflow) of nanoimprinted lines
during thermal annealing can provide important information about the levels of
residual stress induced by NIL processing, the roles of polymer viscosity
and rheology in creating those stresses, as well as their ultimate impact on the
stability of patterns of different molecular masses. A spectroscopic ellipsometric
scatterometer was employed to monitor the pattern reflow in nanoimprinted poly-
mers. The ellipsometric parameters were collected in the spectral range from 335 to
1700 nm by fixing the incidence angle at 60 . The profile model used to characterize
the nanoimprinted line shape consisted of grating lines that could vary in the cross
section from trapezoidal to nearly sinusoidal by convolving a trapezoid with a
Gaussian and a residual layer between the grating lines and the substrate. The
a 200 c
Time
180 Scatterometry
40 s
3D AFM = 20 s
160
CD [nm]
140

120
100
80
60
40
20 60 s 80 s
0 20 40 60 80 100 120 140
Etching time [s]

b 600 Scatterometry
550
3D AFM
500
Height [nm]
450
400
350

300 100 s 120 s
250
200

150
100
0 20 40 60 80 100 120 140
Etching time [s]
Fig. 11 Real-time measurement results (solid line): (a) CD and (b) height of the resist feature as a
function of the etching time, where 3D AFM results are shown with cross shape markers; (c)
comparison between the reconstructed feature profiles by scatterometry (flat color) and the cross
sections of AFM measurement results (surrounding solid line) at different etching times (20, 40, 60,
80, 100, and 120 s) (Kodadi et al. 2009)
Levenberg-Marquardt algorithm was used to extract the profile parameters including

the top and bottom linewidths, the line height of the trapezoid, the width of the
Gaussian, and the underlying residual layer thickness from the measured spectra.
Figure 12a presents the ex situ measurement for the comparison of line height
measured by scatterometry with the results measured by AFM and specular x-ray
reflectivity (SXR). A good agreement between scatterometry and other techniques
can be easily observed. The mean and standard deviation of the difference between
scatterometry and SXR heights were 0.9 and 5.8 nm, respectively, and between
scatterometry and AFM were 3.1 and 6.8 nm, respectively. Figure 12b further
presents the line height versus annealing time measured by scatterometry for two
kinds of gratings at different annealing temperatures, and the line profiles of the two
gratings for three similar line heights are shown in Fig. 12c.
Overlay and Asymmetry Measurement
Semiconductor device features are typically patterned layer-by-layer on silicon

wafers. These patterns must be aligned accurately; otherwise an overlay error
between the top and underlying layers will lead to non-yielding devices. It is highly
desirable if the overlay error could be directly measured on the device patterns,
which is, however, usually infeasible in practice due to the limited resolution of
Line height [nm]

a 100
300
Lin e h eigh t [n m]
250
(AFM an d SXR)
200
10
0 50 100 150
150
Annealing time [min]
c
100
300
Line height [nm]

50 250
200
0 150
0 50 100 150 200 250 300 100
Line height [nm] 50
(Scatterometry) 0
0 200 400 600 800
Lateral distance [nm]
Fig. 12 (a) Ex situ measurement for the comparison of line height measured by scatterometry with
line height measured by AFM (solid squares) and SXR (open circles). The line y = x is also shown;
(b) line height as a function of annealing time measured by scatterometry for a PS18k grating (open
triangles) during annealing at 100 C and a PS1570k grating (closed squares) during annealing at
106 C; (c) line profiles of the PS18k grating (solid curves) and the PS1570k grating (dashed
curves) for three similar grating heights. The underlying residual layer is not shown. The PS18k and
PS1570k correspond to low-polydispersity-index polystyrene with molar masses of 18.1 and
1571 kg mol1, respectively (Patrick et al. 2008)
optical methods. A more practical approach is to design special targets that are
sufficiently coarse to be resolved by optical methods. Figure 13 presents some of the
commonly used targets for overlay measurement, of which the grating targets are
much more preferred since the periodic nature of the grating targets allows efficient
noise suppression by spatial filtering (den Boef 2016). There are mainly two optical
overlay metrology techniques: the image-based overlay (IBO) metrology and the
diffraction-based overlay (DBO) metrology. In the case of IBO metrology, an
overlay target, as shown in Fig. 13c, is built up of X and Y gratings arranged in
such a way that the target has 180 rotation symmetry. A high-quality microscope
such as a bright-field microscope images this target on a detector array. Afterward,
pattern recognition and image processing algorithms are applied to determine the
overlay error. It is a common practice to measure the overlay target for 0 and 180
wafer orientations separately to reduce the tool-induced shift (TIS) introduced by
lens aberrations and illumination imperfections of the employed microscope. Nev-
ertheless, TIS in IBO also varies with wafer processing changes, such as the film
thickness changes. The DBO metrology is a promising technique that could address
the challenges in IBO metrology, which also adopts a grating target comprised of
gratings in two different layers. However, the top and bottom target gratings in DBO
metrology are not placed alongside each other as those in IBO metrology, but are on
top of each other as schematically shown in Fig. 13d. If there is an overlay error, the
overlay
D E F G
Fig. 13 Different targets for overlay measurement. (a) Top-down image (top) and cross-section
image (bottom) of a box-in-box target; (b) top-down image (top) and cross-section image (bottom)
of a bar-in-bar target; (c) target gratings that are placed alongside each other, where the dark and
light gratings are present in different layers (den Boef 2016); (d) top-down image (top) and cross-
section image (bottom) of the target gratings that are placed on top of each other. (a–c) are usually
used in IBO metrology, while (d) is used in DBO metrology
top and bottom gratings form an asymmetric composite grating that can be sensed by
optical scatterometry.
Figure 14 presents the schematic of a DBO grating target for overlay metrology
(Ko and Ku 2006). As shown in Fig. 14a, the target grating was comprised of,
from top to bottom, photoresist grating, poly-Si interlayer, SiO2 grating, and Si
substrate. The top and bottom gratings both had a nominal period of 800 nm. An
intentional overlay offset Δ was included between the top and bottom grating
layers in the initial design of the target, since it was shown that scatterometric
measurements had a higher sensitivity when an overlay offset of one fourth of the
grating period was introduced between the top and bottom grating layers. The total
overlay value was the sum of the intentional overlay offset and the actual overlay
error. As can be observed from Fig. 14b, a traditional bar-in-bar target was also
included in the design, which was used for the IBO metrology with a bright-field
microscope so as to make a comparison with the DBO metrology by optical
scatterometry. A 2-θ angular scatterometer was employed in the experiment,
which provided angular reflectance in the range of incidence angle from 30 to
30 . A slight difference in the employed angular scatterometer was that a polariz-
ing beam splitter was mounted in the output ray path so that the reflected light was
separated into s-polarized and p-polarized beams, which were then collected by two
separate detectors. The conventional library search method was used for parameter
extraction.
Five targets denoted as Targets 1–5 that had intentional overlay offsets of 100 nm,
150 nm, 200 nm, 250 nm, and 300 nm, respectively, were analyzed. Figure 15a
presents the fitting result between the experimental and theoretical best-fit angular
reflectance for Target 3. To quantitatively compare the results measured by
scatterometry and bright-field microscopy, ten static repeated measurements were
performed for each grating target and bar-in-bar target at stage orientations of both
0 and 180 . The overlay and TIS values at each measurement were calculated by
a 60 μm
85 μm overlay
photoresist
poly-Si
20 μm
SiO2
60 μm
Si
85 μm
Δ = 100
b unit: nm
JUDWLQJWDUJHW
Δ= 0 Δ = 50 Δ = 100 ••• •••
•••
EDULQEDUWDUJHW
x
•••
Fig. 14 Schematic of a DBO grating target for overlay metrology. (a) (Left) Layout of the target
gratings for the measurement of overlay errors in the x and y directions and (right) cross section of
the target grating; (b) layout of the grating targets and bar-in-bar targets. An intentional overlay
offset Δ ranged from 0 to 800 nm with an increment of 50 nm was included in the grating targets
(Ko and Ku 2006)
OVLi = [OVLi(180 ) OVLi(0 )]/2 and TISi = [TISi(180 ) + TISi(0 )]/2 (i = 1, 2,

. . ., 10), respectively. The average overlay, μOVL; the 3σ overlay precision, uOVL; the
average TIS, μTIS; and the 3σ TIS precision, uTIS, of the measurement systems were
then evaluated. Figure 15b shows the comparison of the average overlay values μOVL
measured by scatterometry and bright-field microscopy for the five targets. A good
linear correlation can be observed. The estimation of other metrics revealed that
uOVL, μTIS, and uTIS obtained by optical scatterometry were all smaller than their
counterparts obtained by bright-field microscopy for the five targets. When averaged
a 0.4
Target 3
0.35
0.3
Reflectance
0.25
measured_3
0.2 matched_3
measured_P
0.15 matched_P
0.1
0.05
0
–30
–25
–22
–18
–14
–10
–6
–2
2
6
10
14
18
22
26
30
Incidence angle [deg]
b
300
Scatterometry results [nm]
y = 1.02x - 5.02
R2 = 1.00
250
200
150
100
100 150 200 250 300
Bright-field microscopy results [nm]
Fig. 15 (a) Fitting result between the experimental and theoretical best-fit angular reflectance for
Target 3; (b) comparison of the overlay measurement results by scatterometry and bright-field
microscopy for Targets 1–5 (Ko and Ku 2006)
across the five targets, the uOVL of scatterometry was 0.2 nm, while that of the bright-
field microscopy was 0.5 nm. The average μTIS and uTIS of scatterometry were
0.20 nm and 0.05 nm, respectively, which were both approximately an order of
magnitude less than those of bright-field microscopy.
Although specially designed targets are required for overlay metrology in most cases,
it could also be possible to realize direct in-chip overlay measurement for some special
device pattern features. Figure 16a, b schematically shows a dynamic random-access
memory (DRAM) device pattern with a recess channel array transistor (RCAT) structure
at the second lithographic patterning step, which consisted of a shallow trench isolation
(STI) array at the bottom, photoresist grating lines at the top, and several thin film
layers filled in between. As shown in Fig. 16b, c, the bottom STI islands were
arranged in a non-orthogonal lattice with periods of 286 and 249 nm and a lattice
angle of 54.95 . Each silicon island had a CD ratio of about 10 between the long and
Fig. 16 (a) Schematic of a DRAM device with a RCAT structure at the second lithographic
patterning step; (b) top-down view of the device pattern; (c) bottom STI islands etched in silicon;
(d) with SEM image of an AEI RCAT structure showing a misalignment of the photoresist lines
with respect to the STI islands, the overlay error was δx = (x1 x2)/2 (Kim et al. 2009)
short axes and was rotated by an angle of 26 with respect to the horizontal axis. The
top photoresist lines had a period of 143 nm and a line height of about 100 nm. Figure
16d presents a SEM image of an after-etch-inspection (AEI) RCAT structure with an
overlay error of δx = (x1 x2)/2. The overlay error made the two ends of the elongated
STI island non-equal, thus breaking the symmetry of the structure. This in-chip
structural asymmetry was then sensed by a MME-based scatterometer.
The physics behind the use of MME for structural asymmetry detection is the
electromagnetic reciprocity theorem for the zeroth-order diffraction of a symmetric
grating (Li 2000), which states that, for any incident light (θ, ϕ), the zeroth-order
cross-polarization reflection coefficients rps, 0 and rsp, 0 (namely, the two off-diagonal
elements of the Jones matrix J given in Eq. (16)) of the grating are antisymmetric, i.e.,
rps, 0 = rsp, 0, provided that the grating is composed of only reciprocal materials and is
invariant under the rotation of 180 about the normal incidence. According to this
reciprocity theorem, the relationship between the elements of the two 2 2 off-diagonal
blocks of the (normalized or unnormalized) Mueller matrix (also the zeroth-order)
satisfies that m13 + m31 = 0, m23 + m32 = 0, m14 m41 = 0, and m24 m42 = 0
(Chen et al. 2015). When the profile has an asymmetric structure, the above reciprocity
0.02 Response
0.03
m13 + m31 Linear fit
0.015 0.02
Average response of m 13 + m 31
0.01
0.01 0
–0.01
0.005
–0.02
0 –0.03
245 250 255 260 265
Wavelength (nm)
–0.005
–0.01
y = 0.0009*x - 0.003
–0.015 R2 = 0.99
–0.02
–0.025
–20 –15 –10 –5 0 5 10 15 20
Overlay error d x [nm]
Fig. 17 The relationship between the average response (over the spectral of from 245 to 265 nm) of the
sum m13 + m31 and the overlay error δx. The inset figure shows the responses of m13 + m31 in the spectral
range of 245–265 nm with different overlay errors varied in a range of 15 nm (Kim et al. 2009)
is broken, and it follows that rps, 0 ¼

6 rsp, 0 and |mij| ¼
6 |mji| (i = 1, 2; j = 3, 4). In the
regime of small overlay errors, a linear relationship of m13 + m31 = C1δx and
m23 + m32 = C2δx can be assumed (Kim et al. 2009), where C1 and C2 are constants
(similar relationship can also be assumed for m14 m41 and m24 m42). Figure 17
shows the relationship between the average response of m13 + m31 and the overlay error
δx. A good linear relationship can be observed. Nonlinear regression was then performed
to extract the overlay error as well as other structural parameters, including the param-
eters of the STI island (CDs in x and y directions, height, sidewall angle), the parameters
of the photoresist grating line (CD, height, sidewall angle), and the thicknesses of the
thin films filled in between. The extracted overlay errors exhibited a good linear
correlation with respect to the results by IBO metrology on targets in the scribe line
surrounding the device area.
The linear response of the elements from the two 2 2 off-diagonal blocks of
the Mueller matrix with respect to profile asymmetry makes MME well-suited for
the measurement of overlay and other asymmetric features. Figure 18a presents
the cross-section SEM image of a nanoimprinted grating structure with an
asymmetric profile (Chen et al. 2014). Although symmetric imprint resist profiles
are expected in most cases, errors could occur in actual NIL processes and will
result in undesired asymmetry and pattern transfer fidelity loss in downstream
processes. Detection of imprint resist asymmetric defects leads to the
1.0 0.2 0.2

a b data
0 0 0
fit
–1.0 –0.2 –0.2
1.0 1.0 0.4 0.2
0.9
0 0.8 0 0
0.7
–1.0 –0.4 –0.2
0.2 0.4 1.0 1.0
c 0 0 0 0
–0.2 –0.4 –1.0 –1.0

0.2 0.2 1.0 1.0
0 0 0 0
–0.2 –0.2 –1.0 –1.0

200 400 600 800 200 400 600 800 200 400 600 800 200 400 600 800
Wavelength [nm]
Fig. 18 (a) Cross-section SEM image and geometrical model for a nanoimprinted grating structure
with an asymmetric profile; (b) fitting result between the experimental and theoretical best-fit
Mueller matrix spectra with the incidence and azimuthal angles fixed at 65 and 90 , respectively;
(c) comparison between the cross-section SEM image and the reconstructed profile (surrounding
solid line) of the investigated grating sample (Chen et al. 2014)
improvement of the template, the imprint process, and the imprint toll design and
finally guarantees the pattern transfer fidelity in template replication. As depicted
in Fig. 18a, a two-segment quadratic line shape model was adopted to character-
ize the structural profile of the investigated grating structure, whose Mueller
matrices were then measured by a MME-based scatterometer in the spectral
range from 200 to 800 nm by fixing the incidence and azimuthal angles at 65
and 90 , respectively. The Levenberg-Marquardt algorithm was used to recon-
struct the structural profile of the grating sample from the measured Mueller
matrix spectra. Figure 18b presents the fitting result between the experimental
and theoretical best-fit Mueller matrix spectra. A good agreement between the
cross-section SEM image and the reconstructed structural profile can be observed
from Fig. 18c. In the analysis, it also revealed that the conventional ellipsometric
scatterometry that collected two ellipsometric parameters was only sensitive to
the magnitude of profile asymmetry but could not distinguish the asymmetric
direction. In comparison, MME-based scatterometry showed good sensitivity to
both the magnitude and direction of profile asymmetry.
Line Roughness Measurement
The metrology challenges from semiconductor manufacturing with the continuous

reduction of critical size are not only from CD and overlay but also from the line
roughness. The impact of the latter on the performance of microelectronic devices
LER := 3s deviation Λ LWR := 3s

of the edge to the deviation of the
nominal line linewidth
CD
Fig. 19 Schematic of grating lines with LER and LWR
has become a more and more non-ignorable issue especially with the grate dimen-
sions of transistors shrinking to the decananometer regime. As schematically shown
in Fig. 19, the line roughness can be characterized in terms of line edge roughness
(LER) and linewidth roughness (LWR) in the spatial domain, of which LER is
defined as the 3σ (three times root-mean-square) deviation of a single printed pattern
edge to a nominal line and LWR is defined as the 3σ deviation of the physical
distance between two printed pattern edges. Currently, CD-SEM is commonly used
for the characterization of line roughness. However, it has edge detection and
modeling requirements at this scale and has also faced high-cost and low-throughput
concerns. Optical methods, if could be used, would be very valuable for the
community on semiconductor manufacturing, due to the inherent merits of non-
destruction, high throughput, and relative low cost.
Optical scatterometry has been explored for the investigation of line roughness
(Boher et al. 2005; Yaakobovitz et al. 2007; Foldyna et al. 2011; Henn et al. 2012;
Dixit et al. 2015). Nevertheless, it should be noted that the investigated line roughness
in most of these studies was artificially designed periodic variations, which were not
representative of the realistic situations with a random nature. The main trouble for
optical scatterometry methods in line roughness measurement arises from the low
response sensitivity to line roughness as well as the random nature of line roughness.
The random nature of realistic line roughness imposes a three-dimensional nature with
no periodicity to the optical scatterometry community. As a result, most numerical
approaches toward solving the Maxwell’s equations in the presence of line roughness
become more difficult and time-consuming to implement. On the other hand, to
improve the response sensitivity, optical scatterometry methods with short wavelength
light sources, such as deep UV scatterometry (Yaakobovitz et al. 2007) and EUV
scatterometry (Henn et al. 2012), have ever been employed. However, wavelengths in
the UV range also impose a risk of resist exposure.
As an example, Fig. 20a schematically shows an EUV mask geometry with line
roughness (not shown in the figure) that had been investigated by a EUV
a b
pitch (280 nm) super-cell (24×280 nm)
pitch
top-CD
TaO
TaN
SWA
SiO2
Si capping Si capping
Mo/Si multilayer substrate Mo/Si multilayer substrate
c d
7
7
σΔ /nm
σΔ /nm
6 6.26 nm 5
5
3
Fig. 20 (a) EUV mask geometry with 280 nm pitch (period); (b) schematic of the supercell used in
the simulation. The supercell contained 24 lines, and the positions and widths of these lines varied
randomly within the supercell to simulate the random nature of line roughness; (c) determination of
the line roughness standard deviation σ Δ from the simulated signature. The value of
σ Δ = 6.01 0.61 nm was close to the expected value of 6.26 nm; (d) determination of σ Δ from
the experimental signature. The value of σ Δ = 5.20 1.21 nm was slightly larger but of comparable
order of magnitude than the SEM measurement result of about 3 nm (Henn et al. 2012)
scatterometer (Henn et al. 2012). The EUV scatterometer operated in the EUV range
(λ ~ 13.5 nm) and collected the reflectance of the diffracted beams (similar to the k^m-
scan scatterometer shown in Fig. 1) (Gross et al. 2009). As shown in Fig. 20b, a
supercell of 24 lines with periodic boundary conditions was used in the simulation
of the angular reflectance by the finite element method. The positions and widths
of the 24 lines varied randomly within the supercell to simulate the random nature of
line roughness. The contribution of the line roughness to the signature of the
structure was described by a damping factor as f~m ðxÞ ¼ exp σ 2Δ k 2m f m ðxÞ ,
where fm(x) and f~m ðxÞ corresponded to the signatures of the structure without
and with line roughness, respectively; x denoted structural parameter vector
under measurement; km was the wavevector of the m-th diffraction order; and σ 2Δ
= σ 2LER + σ 2LWR =4 was the line roughness variance with σ LER and σ LWR denoting
the amplitudes of LER and LWR, respectively. Both simulation and experiment
were carried to demonstrate the feasibility of EUV scatterometry for line rough-
ness measurement. Figure 20c, d presents the simulation and experimental
results, respectively. The statistics of 100 estimations of σ Δ in the simulation
gave σ Δ = 6.01 0.61 nm, which was close to the expected value of 6.26 nm.
The average over ten datasets in the experiment showed σ Δ = 5.20 1.21 nm,
which was slightly larger but of comparable order of magnitude than the SEM
measurement result of about 3 nm.
Conclusion and Outlook
The optical scatterometry has been concisely described in this chapter. Different
from traditional image-based metrology techniques, optical scatterometry does not
depend on the images with well-defined edges and allows for the indirect determi-
nation of structural parameters from the measured signature by solving an inverse
diffraction problem. Although the measurement process in optical scatterometry is
not straightforward and involves the computation-intensive inverse problem-solv-
ing, it is not restricted by the Abbe diffraction limit as usually encountered in the
image-based metrology techniques. Due to its nondestructive, inexpensive, and
time-effective nature, optical scatterometry has become one of the important tech-
niques for CD and overlay metrology in semiconductor manufacturing. On the other
hand, it is also noted that, with the continuous and aggressive reduction in CD of
semiconductor devices, the demand of tighter process control poses serious metrol-
ogy challenges to optical scatterometry. These challenges arise not only from CD
and overlay but also from the line roughness (LER/LWR). In response to these
challenges, strategies that could be used to improve the sensitivity of optical
scatterometry, such as short wavelength light sources (Wang et al. 2007) and making
full use of light polarization, are worthy to be explored in the future. In addition,
absolute and traceable scatterometry is necessary to be investigated so as to ensure
the international comparability of the measurement results and the applications in
production quality control.
References
Azzam RMA, Bashara NM (1977) Ellipsometry and polarized light. North-Holland, Amsterdam
Bao G, Chen ZM, Wu HJ (2005) Adaptive finite-element method for diffraction gratings. J Opt Soc
Am A 22:1106–1114
Boher P, Petit J, Leroux T, Foucher J, Desières Y, Hazart J, Chaton P (2005) Optical Fourier
transform scatterometry for LER and LWR metrology. Proc SPIE 5752:192–203
Calaon M, Madsen MH, Weirich J, Hansen HN, Tosello G, Hansen PE, Garnaes J, Tang PT (2015)
Replication fidelity assessment of large area sub-μm structured polymer surfaces using
scatterometry. Surf Topogr Metrol Prop 3:045005, (7pp)
Chen X, Liu S, Zhang C, Jiang H (2013a) Improved measurement accuracy in optical scatterometry
using correction-based library search. Appl Opt 52:6726–6734
Chen X, Zhang C, Liu S (2013b) Depolarization effects from nanoimprint grating structures as
measured by Mueller matrix polarimetry. Appl Phys Lett 103:151605, (4pp)
Chen X, Liu S, Gu H, Zhang C (2014a) Formulation of error propagation and estimation in grating
reconstruction by a dual-rotating compensator Mueller matrix polarimeter. Thin Solid Films
571:653–659
Chen X, Zhang C, Liu S, Jiang H, Ma Z, Xu Z (2014b) Mueller matrix ellipsometric detection of
profile asymmetry in nanoimprinted grating structures. J Appl Phys 116:194305, (7pp)
Chen X, Jiang H, Zhang C, Liu S (2015) Towards understanding the detection of profile asymmetry
from Mueller matrix differential decomposition. J Appl Phys 118:225308, (5pp)
Colonna de Lega X (2012) Chap 13, Model-based optical metrology. In: Osten W, Reingand N (eds)
Optical imaging and metrology: advanced technologies. Wiley-VCH Verlag GmbH, Weinheim,
pp 283–304
den Boef AJ (2016) Optical wafer metrology sensors for process-robust CD and overlay control in
semiconductor device manufacturing. Surf Topogr Metrol Prop 4:023001, (15pp)
Dixit D, O’Mullane S, Sunkoju S, Gottipati A, Hosler ER, Kamineni V, Preil M, Keller N, Race J,
Muthinti GR, Diebold AC (2015) Sensitivity analysis and line edge roughness determination of
28-nm pitch silicon fins using Mueller matrix spectroscopic ellipsometry-based optical critical
dimensional metrology. J Micro/Nanolithogr MEMS MOEMS 14:031208, (13pp)
El Kodadi M, Soulan S, Besacier M, Schiavone P (2009) Real time scatterometry for profile control
during resist trimming process. J Vac Sci Technol B 27:3232–3237
Foldyna M, Germer TA, Bergner BC, Dixson RG (2011) Generalized ellipsometry of artificially
designed line width roughness. Thin Solid Films 519:2633–2636
Fujiwara H (2007) Spectroscopic ellipsometry: principles and applications. Wiley, Chichester
Germer TA, Patrick HJ, Silver RM, Bunday B (2009) Developing an uncertainty analysis for optical
scatterometry. Proc SPIE 7272:72720T, (11pp)
Gross H, Rathsfeld A, Bär M (2009) Profile reconstruction in extreme ultraviolet (EUV)
scatterometry: modeling and uncertainty estimates. Meas Sci Technol 20:105102, (11pp)
Henn M-A, Heidenreich S, Gross H, Rathsfeld A, Scholze F, Bär M (2012) Improved grating
reconstruction by determination of line roughness in extreme ultraviolet scatterometry. Opt Lett
37:5229–5231
Huang H-T, Kong W, Terry FL Jr (2001) Normal-incidence spectroscopic ellipsometry for critical
dimension monitoring. Appl Phys Lett 78:3983–3985
Joint Committee for Guides in Metrology (JCGM) (2008) Evaluation of measurement data –guide
to the expression of uncertainty in measurement (GUM 1995 with minor corrections)
Jones CR (1941) A new calculus for the treatment of optical systems I. Description and discussion
of the calculus. J Opt Soc Am 31:488–493
Jones RL, Hu T, Lin EK, Wu W-L, Kolb R, Casa DM, Bolton PJ, Barclay GG (2003) Small angle
x-ray scattering for sub-100 nm pattern characterization. Appl Phys Lett 83:4059–4061
Kim Y-N, Paek J-S, Rabello S, Lee S, Hu J, Liu Z, Hao Y, McGahan W (2009) Device based in-chip
critical dimension and overlay metrology. Opt Express 17:21336–21343
Ko C-H, Ku Y-S (2006) Overlay measurement using angular scatterometer for the capability of
integrated metrology. Opt Express 14:6001–6010
Levenberg K (1944) A method for the solution of certain non-linear problems in least squares.
Q Appl Math 2:164–168
Li L (1996a) Use of Fourier series in the analysis of discontinuous periodic structures. J Opt Soc
Am A 13:1870–1876
Li L (1996b) Formulation and comparison of two recursive matrix algorithms for modeling layered
diffraction gratings. J Opt Soc Am A 13:1024–1035
Li L (2000) Symmetries of cross-polarization diffraction coefficients of gratings. J Opt Soc Am A
17:881–887
Liu S, Chen X, Zhang C (2015) Development of a broadband Mueller matrix ellipsometer as a
powerful tool for nanostructure metrology. Thin Solid Films 584:176–185
Madsen MH, Hansen PE (2016) Scatterometry – fast and robust measurements of nano-textured
surfaces. Surf Topogr Metrol Prop 4:023003, (26pp)
Marquardt DW (1963) An algorithm for least-squares estimation of nonlinear parameters. J Soc Ind
Appl Math 11:431–441
Mittal AS, Ganguly U (2017) An analytical model to estimate VT distribution of partially correlated
fin edges in FinFETs due to fin-edge roughness. IEEE Trans Electron Devices 64:1708–1715
Moharam MG, Grann EB, Pommet DA, Gaylord TK (1995a) Formulation for stable and efficient
implementation of the rigorous coupled-wave analysis of binary gratings. J Opt Soc Am
A 12:1068–1076
Moharam MG, Grann EB, Pommet DA, Gaylord TK (1995b) Stable implementation of the rigorous
coupled-wave analysis for surface-relief gratings: enhanced transmittance matrix approach.
J Opt Soc Am A 12:1077–1086
Nakata Y, Koshiba M (1990) Boundary-element analysis of plane-wave diffraction from groove-
type dielectric and metallic gratings. J Opt Soc Am A 7:1494–1502
Niu X, Jakatdar N, Bao J, Spanos CJ (2001) Specular spectroscopic scatterometry. IEEE Trans
Semicond Manuf 14:97–111
Novikova T, de Martino A, Bulkin P, Hatit SB, Drévillon B (2006) Application of Mueller
polarimetry in conical diffraction for critical dimension measurement in microelectronics.
Appl Opt 45:3688–3697
Novikova T, de Martino A, Bulkin P, Nguyen Q, Drévillon B (2007) Metrology of replicated
diffractive optics with Mueller polarimetry in conical diffraction. Opt Express 15:2033–2046
Patrick HJ, Attota R, Barnes BM, Germer TA, Dixson RG, Stocker MT, Silver RM, Bishop MR
(2007) Optical critical dimension measurement of silicon grating targets using back focal plane
scatterfield microscopy. J Micro/Nanolithogr MEMS MOEMS 7:013012, (11pp)
Patrick HJ, Germer TA, Ding Y, Ro HW, Ritcher LJ, Soles RL (2008) Scatterometry for in situ
measurement of pattern reflow in nanoimprint polymers. Appl Phys Lett 93:233105, (3pp)
Press WH, Teukolsky SA, Vetterling WT, Flannery BP (2007) Numerical recipes: the art of
scientific computing, 3rd edn. Cambridge University Press, Cambridge
Raymond CJ (2001) Chap 1, Scatterometry for semiconductor metrology. In: Diebold AC (ed)
Handbook of silicon semiconductor metrology. CRC Press, Boca Raton, pp 477–513
Raymond CJ (2005) Overview of scatterometry applications in high volume silicon manufacturing.
AIP Conf Proc 788:394–402
Raymond CJ, Murnane MR, Prins SL, Sohail S, Naqvi H, McNeil JR (1997) Multiparameter
grating metrology using optical scatterometry. J Vac Sci Technol B 15:361–368
Raymond CJ, Littau M, Chuprin A, Ward S (2004) Comparison of solutions to the scatterometry
inverse problem. Proc SPIE 5375:564–575
Sendelbach M, Archie C (2003) Scatterometry measurement precision and accuracy below 70 nm.
Proc SPIE 5038:224–238
Taflove A, Hagness SC (2005) Computational electrodynamics: the finite-difference time-domain
method, 3rd edn. Artech House, Norwood
Wang C, Jones RL, Lin EK, Wu W-L, Rice BJ, Choi K-W, Thompson G, Weigand SJ, Keane DT
(2007) Characterization of correlated line edge roughness of nanoscale line gratings using small
angle x-ray scattering. J Appl Phys 102:024901, (6pp)
Yaakobovitz B, Cohen Y, Tsur Y (2007) Line edge roughness detection using deep UV light
scatterometry. Microelectron Eng 84:619–625
Contact-Type Micro Thermal Sensor for
Surface Defect Detection 16
Yuki Shimizu
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 516
Concept of Surface Defect Detection by Using a Micro Thermal Sensor . . . . . . . . . . . . . . . . . . . . . 518
Estimation of the Frictional Heat . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 519
Simulation on the Change in Sensor Temperature Due to the Frictional Heat . . . . . . . . . . . . . . . . 521
A Design and Fabrication of the Micro Thermal Sensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 522
Evaluation of the Basic Characteristics of the Fabricated Micro Thermal Sensor . . . . . . . . . . . . . 525
Sensitivity of the Micro Thermal Sensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 525
Ability of Contact Detection . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 529
Application of the Micro Thermal Sensor . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 532
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 535
Abstract
Surface defect detection, which is carried out in advance of defect review process
during surface defect inspection of products having smoothly finished surfaces
such as bare semiconductor wafers, magnetic disks, and optical components, is
important process to assure the quality of products. In this chapter, a surface
defect detection method, in which defect detection is carried out in such a way
that the existence of a surface defect on a target of interest is verified by detecting
frictional heat induced by a collision between a micro thermal sensor and a
surface defect, is described. Although the frictional heat to be generated by a
collision between the micro thermal sensor and a surface defect is expected to be
small since surface defects required to be verified during the inspection are quite
small, the micro thermal sensor designed to have a micrometric sensor element is
expected to carry out highly sensitive detection of the frictional heat and thus
realize high-resolution surface defect detection. A principle of the surface defect
Y. Shimizu (*)
Department of Finemechanics, School of Engineering, Tohoku University, Sendai, Japan
e-mail: yuki.shimizu@nano.mech.tohoku.ac.jp

https://doi.org/10.1007/978-981-10-4938-5_18
516 Y. Shimizu
detection method based on the micro thermal sensor is at first described. After
that, design and fabrication of the micro thermal sensor based on photolithogra-
phy process are presented. In addition, by using the developed micro thermal
sensor, some experiments have been carried out to demonstrate the feasibility of
the defect detection method. An example of the application of micro thermal
sensor for surface defect detection in the hard disk drive industry is also
introduced.
Keywords
Surface defect detection · Micro thermal sensor
Introduction
Precision components having smooth surfaces such as wafers, magnetic disks, and
optical components are expanding their applications in many industrial fields. In
such a precision component, surface quality is one of the most important factors
that affect the quality of product in which the component is employed, and
therefore the surface inspection is important in the fabrication process of those
products. Two of the main processes in the surface inspection are defect detection
process and defect review process that will be carried out followed by the defect
detection. In the defect review process, high-resolution measuring instruments
such as atomic force microscopes (AFMs) or scanning electron microscopes
(SEMs) are often employed. Meanwhile, since the field of view (FOV) of these
high-resolution instruments is limited in most of the cases, defect detection
process is necessary to obtain positional information of surface defects in advance
of the review process. In the case of semiconductor industry, due to the advance-
ment of the technology node, the allowable size of surface defects is decreasing
year by year as shown in Fig. 1 (ITRS overview 2011). Although the AFMs and
SEMs have enough resolution to carry out the review process, it is becoming more
difficult to find the existences of surface defects since the allowable size of which
is becoming smaller and smaller.
A traditional method for surface defect detection is laser scattering method
(Lonardo 1991; Takami 1997; Takahashi et al. 1998), in which the existence of a
surface defect will be verified by detecting scattered light rays from a surface defect
irradiated by an incident laser beam (Fig. 2). Owing to its principle, nondestructive
defect detection can be carried out, as well as the classification by the image
processing techniques (Pan and Tai 2011; Tan and Lau 2011; Xu et al. 2013). By
employing a laser source having a shorter wavelength, resolution of defect detection
can be improved (Meshulach et al. 2010; Montal et al. 2010; Wagner and Harned
2010), as well as by the improvement of photodetector sensitivity. Meanwhile, since
the light intensity of scattered light is strongly affected by the defect size, it is getting
harder to meet the criteria of the defect detection process required for state-of-the-art
products. One of the other drawbacks of the laser scattering method is that some
types of surface defects such as a flat-shaped defect are sometimes difficult to be
16 Contact-Type Micro Thermal Sensor for Surface Defect Detection 517
Fig. 1 Trends of technology 25

node and allowable size of
bare wafer defect (ITRS
overview 2011) 20
15
Size nm
Pattern pitch
10
5
Bare wafer defect
0
2010 2013 2016 2019 2022 2025
Year
Laser source
Lens Detector
Lens
Incident laser beam Sample
Scattered light
Spindle
Surface defect
Fig. 2 A schematic of the scattering method for defect detection
detected by the method. A technological breakthrough is therefore desired to achieve

further smaller defect detection.
Aiming to realize next-generation defect detection, several methods have been
developed so far (Attota and Silver 2011; Takahashi et al. 2011). A method
employing a micro thermal sensor is one of these methods. In this chapter, the
principle of surface defect detection method based on the micro thermal sensor is at
first described. After that, design and fabrication of the micro thermal sensor are
presented. In addition, by using the developed micro thermal sensor, some experi-
ments have been carried out to demonstrate the feasibility of the defect detection
method. An example of the application of micro thermal sensor for surface defect
detection in the hard disk drive industry is also introduced.
518 Y. Shimizu
Concept of Surface Defect Detection by Using a Micro Thermal

Sensor
A schematic of the surface defect detection method with a micro thermal sensor is
shown in Fig. 3 (Shimizu et al. 2014a). In the method, existences of surface defects
on a target of interest such as a semiconductor wafer, a magnetic disk or an optical
flat having a smooth surface are verified by using the micro thermal sensor in such a
way that the sensor detects frictional heat generated by a collision between the micro
thermal sensor and a surface defect. As can be seen in the figure, the micro thermal
sensor is made to scan over a surface of the target of interest while keeping a certain
amount of gap. As long as the sensor is not contacting with the target surface, the
sensor temperature will be kept constant. Meanwhile, when the target surface has a
defect having a height higher than the gap set between the micro thermal sensor and
the target surface, the defect will collide with the micro thermal sensor surface,
resulting in the generation of frictional heat, as well as the change in temperature of
the micro thermal sensor due to the frictional heat. Therefore, by detecting the
Fig. 3 A concept of the Sensor probe

defect detection method by Precision slider
using a micro thermal sensor
(Shimizu et al. 2014a)
Smooth surface
(Specimen) PZT actuator
Precision spindle Scanning path (spiral)
Sensor probe
Si wafer
Defect
δ Thermal element
R Smooth surface
Clearance
v
Substrate
sudden change in temperature of the micro thermal sensor during its scanning over
the target surface, existences of surface defects on a target surface can be verified.
The purpose of the above mentioned method is to verify the existences of surface
defects on a target surface. A setup shown in Fig. 3 is one of the examples of how to
arrange the micro thermal sensor and a target of interest. Surface defect detection can
be carried out by rotating the target with a precision spindle while moving the micro
thermal sensor along the in-plane direction. By synchronizing the acquisitions of
output signals from the micro thermal sensor, the rotary encoder and the linear
encoder, exact positions of detected surface defects on a target surface can be
determined. The information of defect positions can be utilized to carry out further
detailed investigation on the detected surface defects in the review process, which
will be carried out after the defect detection process.
A resolution of the surface defect detection, which is one of the important perfor-
mances to be improved in next-generation surface defect inspection, by the
abovementioned method will be affected by the sensitivity of micro thermal sensor to
be employed in the system, as well as the amount of frictional heat to be generated by a
collision between the micro thermal sensor and a surface defect. In the next section,
theoretical analysis is carried out to estimate the amount of frictional heat by the collision
based on a simple collision model between the micro thermal sensor and a surface defect.
Estimation of the Frictional Heat
A schematic of the simple model for the estimation of frictional heat to be generated
by a collision between the micro thermal sensor and a surface defect is shown in
Fig. 4. In the model, a surface defect is treated as a hemisphere having the radius of
R on a target with a flat surface, while the micro thermal sensor is treated as another
flat surface placed parallel with respect to the target surface. On the assumption that
the collision can be treated as an elastic contact between the hemisphere and the flat
surface with the indentation depth δ, by including the Hertzian contact theory
(Johnson 1982), the force F to be applied between the target surface and the surface
defect can be described as follows:
2 pffiffiffiffiffiffiffiffi
F ¼ E Rδ3 (1)
3
where E is equivalent to Young’s modulus between the target surface and the surface
defect that can be calculated as follows:
1
1 ν2S 1 ν2F
E¼ þ (2)
ES EF
where ES, EF, and νS, νF are the Young’s modulus and the Poisson’s ratios of the
surface defect and the micro thermal sensor, respectively. Now the radius of the
contact area a can be described as follows:
520 Y. Shimizu
Fig. 4 A schematic of the Young’s modulus: EF

simple model for the Surface of the sensor
estimation of frictional heat to element Poisson’s ratio: νF
be generated by a collision
between the micro thermal
sensor and a surface defect δ : Interference height
Young’s modulus: ES
A tip of defect Poisson’s ratio: νS
1
3FR 3
a¼ (3)
4E
In the contact area, the distribution of normal pressure P(r) at the radial position
r can be expressed as follows:
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
r 2ffi
3F
PðrÞ ¼ 1 (4)
2πa2 a
Assuming that the target surface is moving along the horizontal direction in a
constant velocity V with respect to the micro thermal sensor, the heat flux due to the
frictional heat at the radial position r can be described as follows:
qðrÞ ¼ γμPðrÞV (5)
The parameters γ and μ are a heat partition ratio and a friction coefficient,
respectively. The rate of a heat flow dQ(r) which passes through small area dS at
the radius r in the contact circle can therefore be described as follows:
dQðrÞ ¼ qðrÞ dS ¼ γμPðrÞV rdrdθ (6)
The rate of heat flow Q which passes through the whole contact area can therefore
be calculated as follows (Lu et al. 2012; Shimizu et al. 2014a):
Ð
Q ¼ S dQðrÞ
Ð a Ð 2π Ða
¼ r¼0 θ¼0 γμPðrÞV rdrdθ ¼ 2πμγV 0 rPðrÞdr
ð rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
r 2ffi
3μγFV a
¼ r 1 dr (7)
a2 0 a
¼ γμFV
2 pffiffiffiffiffiffiffiffi
¼ γμEV Rδ3
3
Rate of heat supply Q μW

Fig. 5 Estimated frictional
heat to be generated at a
1000
V = 10 m/s
collision between the micro
thermal sensor and a surface
100
V = 1 m/s
defect
10
V = 0.1 m/s
1
0.1
0.01
0 1 2 3 4 5
Tip radius of defect R μm
By using Eq. (7), the rate of heat flow to be generated by the collision between the
micro thermal sensor and a surface defect is estimated. Figure 5 shows variations of
the frictional heat as a function of the radius of asperity R. For the calculation of the
rate of heat supply, the parameters summarized in Table 1 are employed. In the
figure, the rate of heat supply calculated with a velocity of 0.1 m/s, 1 m/s, and 10 m/s
is plotted. Equation (7) and the results shown in Fig. 5 indicate that the rate of heat
supply Q increases in proportional to the velocity V and ranges from sub μW to
several hundred μW. Therefore, the micro thermal sensor to be employed in the
defect detection is required to have sensitivity enough to detect frictional heat with
the rate of heat supply on the order of microwatts.
Simulation on the Change in Sensor Temperature Due

to the Frictional Heat
The micro thermal sensor verifies the existence of surface defect on a target surface
by detecting frictional heat generated at the collision between the sensor and the
surface defect throughout detecting the change in its temperature induced by the
frictional heat. The change in electrical resistance of the sensor Δρ can be calculated
by using the following equation (Gustafsson 1991):
Δρ ¼ ρ0 αΔT (8)
where α is a temperature coefficient of resistance (TCR) of the micro thermal sensor

and ρ0 is a specific resistance of the micro thermal sensor at a specified temperature.
As can be seen in the equation, Δρ due to the change in temperature of the micro
thermal sensor can be estimated by obtaining the change in its temperature ΔT.
Meanwhile, it is not so easy to theoretically derive ΔT induced by the rate of heat
supply Q by the frictional heat since the micro thermal sensor has a layered structure.
Therefore, finite element analysis is carried out to estimate ΔT induced by the rate of
heat supply Q.
522 Y. Shimizu
Table 1 Parameters for estimating frictional heat to be generated at a collision between the micro
thermal sensor and a surface defect
Parameter Symbol Value Unit
Young’s modulus of the sphere ES 200 GPa
Young’s modulus of the flat surface EF 200 GPa
Poisson’s ratio of the sphere νS 0.3 –
Poisson’s ratio of the flat surface νF 0.3 –
Friction coefficient μ 0.1 –
Velocity V 0.1–10 m/s
Tip radius of the sphere (asperity) R – m
Interference height δ 5 109 m
Heat partition ratio γ 0.5 –
Figure 6 shows a schematic of the finite element model of the micro thermal
sensor employed in this analysis. A micrometric thin metal film structure placed on a
substrate surface is treated as the micro thermal sensor. Material property and
dimensions of each component in the structure of micro thermal sensor employed
in the finite element model are summarized in Table 2. In the simulation, a rate of
heat supply of 20 μW, which is considered to be generated by the collision between
the micro thermal sensor and a surface defect with a tip radius R of 100 nm under the
condition of a velocity Vof 6.28 m/s and an interference height δ of 10 nm, is applied
to the surface of micro thermal sensor, and the change in temperature ΔT of the micro
thermal sensor is calculated. Figure 7a shows a typical temperature deviation of the
micro thermal sensor as a function of time. In the figure, the simulated result with the
sensor element having a size of 1 1 μm is plotted. The period of applying Q to the
micro thermal sensor theat is calculated by the velocity V and the size length dS of the
micro thermal sensor. A variation of the maximum change in temperature ΔTS due to
the variation of dS is summarized in Fig. 7b. In the case of the micro thermal sensor
having a side length dS of 10 μm, ΔTS is estimated to be 0.012 K, which corresponds
to the change in electrical resistance Δρ of 1.2 103%, on the assumption that the
micro thermal sensor has a TCR of 0.1%/K. Although the change in electrical
resistance estimated based on the finite element model is small, it can be well
detected by using conventional technology of the analog signal processing which
can detect a small resistance deviation on the order of 1 104%. From these
simulation results, it can be concluded the micro thermal sensor is expected to have
an ability of verifying the existences of surface defects on a target surface by
detecting frictional heat generated by a collision between the sensor surface and a
surface defect.
A Design and Fabrication of the Micro Thermal Sensor
As described in the previous section, the micro thermal sensor is preferred to be

designed as compact as possible to improve its sensitivity against the frictional heat.
Fig. 6 FEM simulation for temperature rise of the micro thermal sensor due to frictional heat
generated at the collision between the micro thermal sensor and a surface defect. (A part of figure
from: Shimizu et al. 2014a)
Table 2 Material properties used in the FEM simulation (Shimizu et al. 2014a)
Parameter Value Unit
Sensing elementa Density 4.0 103 kg/m3
Thermal conductivity 3.0 101 W/mK
Specific heat capacity 8.5 102 kJ/kgK
Thickness 2.5 101 nm
Baseb Density 8.5 103 kg/m3
Thermal conductivity 4.3 101 W/mK
Specific heat capacity 4.6 102 kJ/kgK
Surroundings Heat transfer coefficient at surroundings 1.0 103 W/m2K
Heat transfer coefficient at top surface of the base 5.8 W/m2K
a
Material assumed to be chromium
b
Material assumed to be silicon
524 Y. Shimizu
Fig. 7 Temperature rise of 0.20
Tem p era t u re d ev ia t io n K
the micro thermal sensor due (a)
to frictional heat generated at 0.15
the collision between the
micro thermal sensor and a 0.10
surface defect estimated in the ΔTs
FEM simulation. (a) Typical 0.05
temperature deviation of the
sensing element as a function 0.00
Heating time: theat
of time (ds = 1 mm).
(b) Temperature rise and the -0.05
0 50 100 150 200 250 300 350
time constant as a function of
Time ns
the element size ds (Shimizu
I n crea sed t em p era t u re ΔTs K
et al. 2014a) 0.20

(b)
0.15
0.10
0.05
0.00
0 2 4 6 8 10 12
Side length of the sensing element ds μm
Photolithography process capable of fabricating small structures made of thin metal

films is therefore one of the candidates for fabricating micro thermal sensor. Figure 8
shows fabrication process of the micro thermal sensor designed based on photoli-
thography process. In the designed fabrication process, a silicon wafer is employed
as a substrate for the micro thermal sensor. After the preprocess of dicing and
oxidation of the silicon substrate (Fig. 8b), photolithography process for the elec-
trode patterns is carried out. At first, a gold film with a thickness of 30 nm is prepared
on the substrate surface (Fig. 8c). It should be noted that a thin chromium film with a
thickness of 10 nm is placed as the adhesion layer under the gold film to improve the
binding force between the gold layer and the substrate. After that, a photoresist is
spin coated on the gold film, and pattern exposure is carried out by using a
photomask with the electrode patterns, followed by the development process
(Fig. 8d). With the developed patterns, throughout the wet etching process, a pair
of electrodes is fabricated (Fig. 8e). In the same manner as the fabrication of
electrode patterns, a sensor element pattern made of thin chromium film with a
thickness of 100 nm is fabricated in such a way that the sensor element will bridge
the pair of electrodes (Fig. 8g). During the fabrication of sensor element, attentions
should be paid to avoid damages to the existing electrodes on the silicon substrate
due to over-etching in the process.
A three-dimensional profile of the fabricated micro thermal sensor measured by a
commercial white-light scanning interferometer is shown in Fig. 9a. Since the sensor
(b) Diced silicon wafer <Sputtering> (g)

(a) (20 mm x 20 mm)
Cr: Sensor layer
(f)
SiO2 <Oxidation> Photoresist pattern

SiO2: Insulation layer
(b) <Lithography>
(c) Sensor pattern (h)Thermal element
(g) (Cr)
<Sputtering>
Cr: Adhesion layer
Cr/Au (c) Au: Electrode <Etching>
(h) Electrode (Au)
(d) Cr/Au Sensor element
<Lithography>
Thermal element (Cr)
(d) Electrode pattern Effective area
Photoresist pattern
(e)
<Etching>
(e)
Au Electrode Base Electrode (Au+Cr)
pattern Electrode
Fig. 8 Fabrication process of the micro thermal sensor based on photolithography process
element was fabricated followed by the fabrication of electrode patterns, the sensor
has cross-sectional profiles as shown in Fig. 9b, c along the X- and Y-directions,
respectively. The width of the sensor element was designed to be 20 μm, while the
gap in between the electrodes was designed to be 20 μm. As can be seen in Fig. 9, the
micro thermal sensor was successfully fabricated by using the designed photolithog-
raphy process shown in Fig. 8.
Evaluation of the Basic Characteristics of the Fabricated Micro

Thermal Sensor
Sensitivity of the Micro Thermal Sensor
Basic performances of the fabricated micro thermal sensor as a thermal sensing

device were evaluated in experiments by using an experimental setup shown in
Fig. 10a. The setup was designed so that a known quantity of heat can be applied to
the micro thermal sensor by using a laser focusing optical system. The fabricated
micro thermal sensor was mounted on a three-axis precision piezoelectric (PZT)
stage system capable of positioning the target with nanometric resolution and was
placed beneath the laser focusing optical system. The micro thermal sensor was
integrated into a bridge circuit as shown in Fig. 10b so that the change in its
temperature could be detected as the change in its electrical resistance. The bridge
circuit was connected to the commercial signal conditioner (CDV-700A, Kyowa
Electronic Instrument).
526 Y. Shimizu
Fig. 9 Fabricated micro (a)

thermal sensor. (a) Three-
dimensional image of the Sensing element (Cr)
thermal element measured by y3’
a coherence scanning
y2’
interferometer, (b) X- x’
directional cross-sectional y1’
profile, (c) Y-directional cross-
sectional profile (Shimizu
Base (SiO2 )
y3
et al. 2014a) x
y2
y1
Electrode
Z
Y
X
Height 10 nm/div.
x x`
(b)
X position 10 μm/div.
H eigh t 1 0 0 n m /d iv .
y 1 y 1 ’, y 3 y 3 ’
y2y2’
(c)
Y position 10 μm/div.
In the laser focusing optical system, a laser diode having a wavelength of 683 nm,
whose emitting laser power can be controlled by using a laser driver, was employed
as the light source. A laser beam from the LD was collimated by using a collimating
lens, and the collimated laser beam was then focused on the micro thermal sensor by
using an objective lens having a focal length of 50 mm. Regarding the diameter of
the collimated laser beam (5 mm), a diameter of the focused laser beam was
estimated to be less than 7 μm, which was small enough to heat the micro thermal
sensor. In the following experiments, approximately 40% of the laser power was
assumed to be absorbed by the micro thermal sensor, since the reflectivity of the
sensor element was estimated to be approximately 60%.
Figure 11 shows a typical voltage output waveform from the micro thermal sensor
through the bridge circuit and the signal conditioner when the sensor was placed on
Fig. 10 Experimental setup (a)

for quantitative evaluation of Trigger signal
the sensitivity of the Function generator DAQ board
fabricated micro thermal Analog
Trigger signal
sensor. (a) A schematic of the voltage
experimental setup. (b) A
bridge circuit for signal Signal
Laser diode driver
conditioner
processing (Shimizu et al.
2014b) Laser diode
Bias
Collimator lens voltage
Laser beam
Objective lens
Power meter
Sensor substrate Bridge circuit
Z
Y PZT
X XY-stage
(b)
Bias voltage
100 Ω 100 Ω
To signal conditioner
Variable
80 Ω resistor 70 Ω
1~20 Ω
Sensor
the focal plane of the laser focusing optical system and was heated by the focused
laser beam. As can be seen in the figure, the micro thermal sensor successfully
detected irradiation of the focused laser beam onto the sensor surface. To evaluate
the active area of fabricated micro thermal sensor, the sensor was moved in the XY
plane by using the PZT stage in a step of 1 μm, while the change in voltage output
from the micro thermal sensor was monitored at each sensor position. The power of
the laser beam was set to be 1.25 mW (corresponding to a rate of heat supply of
500 μW applied to the micro thermal sensor) with exposure time of 50 ms. Figure 12a
528 Y. Shimizu
Fig. 11 Experimental result Laser off Laser on Laser off

of the quantitative evaluation
of the sensitivity of the
fabricated micro thermal
Output 2 mV/div
sensor
V: Sensor
output
Effec
f
Time 2 ms/div
(b) -0.01 V 0.20 V

(a)
Sensor output Vs
A’
20
10
Y position μm
50 μm
0
B B’
-10
Effective area
-20
Y A
Y
-20 -10 0 10 20
X 50 μm
X X position μm
Fig. 12 Sensitivity distribution of the fabricated micro thermal sensor. (a) Microscopic image of
the element. (b) Contour image of the sensor output (Shimizu et al. 2014a)
shows a microscopic image of the micro thermal sensor evaluated in the experiment,
and Fig. 12b shows the contour image of the change in voltage output from the micro
thermal sensor observed at each position, which corresponds to the sensitivity
distribution of the micro thermal sensor. As can be seen in the figure, the active
area of the fabricated micro thermal sensor was found to be the area between the
electrodes.
Sensitivity of the micro thermal sensor was also evaluated by using the same
setup shown in Fig. 10. In the experiments, the focused laser beam was positioned to
the center of the effective area of micro thermal sensor, and the change in voltage
output from the sensor was evaluated while decreasing the laser power for heating
the micro thermal sensor. Experiments were also carried out for the micro thermal
sensor having an active area of 10 10 μm. Figure 13 shows the variation of
voltage output from the micro thermal sensor with the decrease of the rate of heat
Fig. 13 Experimental result 0.25

of the quantitative evaluation Effective area:
of the sensitivity of the 10 μm㽢10 μm
0.20
Sensor output V
fabricated micro thermal
sensor (Shimizu et al. 2014a)
0.15
0.10
Effective area:
20 μm㽢20 μm
0.05
0.00
0 200 400 600 800
Applied heat μW
supply applied to the sensor surface. As can be seen in the figure, the voltage output
from the sensor increased almost proportional to the rate of heat supply Q in the
region where Q >100 μW, while the sensor output rapidly decreased with the
decrease of Q in the region of Q <100 μW. Meanwhile, it was verified that the
micro thermal sensor having an active area of 10 10 μm could distinguish an
applied rate of heat supply of 10 μW comparable to the quantity of frictional heat to
be generated by the collision between the micro thermal sensor and a surface defect.
Ability of Contact Detection
Experiments were carried out for performance verification of the developed micro
thermal sensor. Figure 14a shows a schematic of the experiment, in which a micro
ball probe was employed to simulate the collision between the micro thermal sensor
surface and a surface defect having a relatively large diameter. The micro ball probe
was prepared by assembling a glass ball with a diameter of 50 μm onto the tip of a
glass probe (Gao 2010). Figure 14b shows a diagram of the setup constructed for the
experiments. The glass ball probe was attached to the tip of a one-axis PZT actuator
mounted on a linear slide capable of moving along the Z-direction. By applying
triangular wave voltage input to the PZT actuator, the glass ball probe can be moved
back and forth along the Y-direction. The fabricated micro thermal sensor, which was
mounted on a three-axis PZT stage, was placed under the glass ball probe. Exper-
iments were carried out in such a way that the micro thermal sensor was made to
approach the glass ball probe being oscillated by the PZT actuator. The micro
thermal sensor was moved along the Z-direction in a step of 1 nm, and the waveform
of voltage output from the micro thermal sensor was captured at each Z-position of
the micro thermal sensor by using an A/D converter with a sampling rate of 10 kHz.
Figure 15a shows a variation of the standard deviation σ V in the waveforms of
voltage output from the micro thermal sensor. As can be seen in the figure, σ V was
almost constant at the beginning of the approaching process. After that, σ V started to
increase with the increase of Z-directional displacement of the micro thermal sensor,
530 Y. Shimizu
(a) (b)
Glass probe PZT Triangular-wave Personal
amplifier generator computer
Z (Oscillated in Y-axis) Z
Y Y
PZT for probe A/D D/A
X X board board
oscillation
Bias voltage
Glass ball
(φ 50 mm) Signal
VOUT conditioner
Micro
thermal
sensor PZT-stage
Step-by-step approach controller
(by PZT Z-stage) PZT XYZ-stage
Fig. 14 Contact detection test by using a developed micro thermal sensor. (a) A schematic of the
test by using a micro ball stylus. (b) Setup for the experiment (Shimizu et al. 2014b, 2015)
0.014
Standard deviation of the
0.012 (a)
sensor output σv V
0.010
0.008
0.006 σv0
0.004
0.002 Judged as contact
0.000
-60 -40 -20 0 20 40 60
Z-displacement of the probe δZ nm
Sensor output 0.05 V/div.
(b)
Before contact (δ z=-50 nm)
At the beginning of contact (δ z =0 nm)
In contact (δ z=50 nm)
Deviation of waveform due to contact
Time 100 ms/div.
Fig. 15 Contact detection test by using a developed micro thermal sensor. (a) Standard deviation
of the voltage output of the thermal sensor at each interference height. (b) Voltage output waveforms
of the thermal sensor at each interference height (Shimizu et al. 2015)
which was considered to be due to the collision between the micro thermal sensor
and the tip of the glass probe. Figure 15b shows the waveforms of voltage output
from the micro thermal sensor at each of the Z-positions. It should be noted that the
(a)
Z
Y 40 nm
X Stylus tip
PZT sensor
AFM
cantilever
Thermal
sensor
(2nd proto)
Feedback control
(b) (by PZT Z-stage)
Z Signal + PID control

Y conditioner circuit Proportional control
-
X + Kp + + PZT-Z
Set - + + stage
+ + point Integral control
1 +
Ki PZT-Z
D/A board Personal s + input AFM
computer system
Bias Differential control
voltage Kd
A/D board s
Signal AFM output

conditioner
VOUT PZT stage
controller
PZT-XYZ stage
Fig. 16 Experimental setup for quantitative evaluation of contact detection by using an AFM probe. (a)
A schematic of the test by using a micro ball stylus. (b) Setup for the experiment (Shimizu et al. 2015)
Z-directional position where the tip of the glass ball probe started to collide with the
micro thermal sensor was determined by the data shown in Fig. 15a.
An interesting thing was that the temperature of the micro thermal senor was found to
decrease by the collision with the glass ball probe. Since the temperature of the micro
thermal sensor became higher than that of the surroundings due to the Joule heating by
the applied bias current, the heat flow from the micro thermal sensor to the glass ball
probe is relatively larger than the frictional heat generated by the collision between the
sensor surface and the glass ball probe, resulting in the decrease of sensor temperature.
Experiments were extended to further verify the performance of the micro
thermal sensor as the contact-detecting device. Figure 16a shows a schematic of
the experiment, and Fig. 16b shows a schematic of the setup constructed for the
experiment, which was similar with the one shown in Fig. 15b employed in the
previous experiments. Meanwhile, instead of using the glass ball probe, a cantilever
probe of atomic force microscopes (AFMs) having a PZT sensor was employed to
simulate a collision between the micro thermal sensor and a small surface defect. The
tip radius of the AFM probe employed in the experiment was approximately 20 nm.
532 Y. Shimizu
Fig. 17 Experimental result (a) (b)

of the quantitative evaluation
Applied load to the cantilever mN

Applied load to the cantilever mN
Thermal sensor output 5 mV/div.
Thermal sensor output 5 mV/div.

1.6 1.6
of contact detection by using Region P Region Q
an AFM probe. (a) Collision PZT 1.5 PZT 1.5
at the probe climbing up the sensor sensor
bump. (b) Collision at the 1.4 1.4
probe falling down from the
top of the bump (Shimizu 1.3 1.3
et al. 2015)
1.2 1.2
1.1 1.1
Thermal sensor Thermal sensor
1.0 1.0
X-position X-position
0.5 μm/div. 0.5 μm/div.
The PZT sensor was integrated into a bridge circuit to detect a contact force applied
to the tip of the AFM probe. A PID controller was applied to the experimental setup
so that the load applied to the micro thermal sensor through the AFM probe could be
kept constant while scanning the micro thermal sensor surface by the AFM probe.
Another feature of the setup was that the setup could acquire profile of the micro
thermal sensor by utilizing XYZ-directional position information obtained by the
capacitive displacement sensors in the three-axis PZT stage. With this feature, a fine
positioning of the AFM probe tip with respect to the micro thermal sensor could be
achieved.
Due to the limitation of the setup employed in the experiment, it was difficult to
apply relative motion in a high speed between the micro thermal sensor and the AFM
probe. To simulate the collision between the micro thermal sensor and a small
surface defect, the micro thermal sensor having a bump with a rectangular profile
within its active area (Shimizu et al. 2015) was employed, and the collisions to be
occurred when the probe tip of the AFM climbs up or falls off the bump were tried to
be observed by the micro thermal sensor. Figures 17a, b show the waveforms of the
voltage output from the micro thermal sensor when the AFM probe tip was made to
climb up and fall down the bump, respectively. In the figures, waveforms of the
output from the PZT sensor in the AFM probe were also plotted. Since the AFM
probe was controlled in a closed loop, deviation of the output from the PZT sensor
was observed only at the collision between the micro thermal sensor and the tip of
the AFM probe. As can be seen in the figures, the micro thermal sensor showed
positive responses at the collisions, which demonstrated the feasibility of the micro
thermal sensor as the contact-sensing device.
Application of the Micro Thermal Sensor
The concept of micro thermal sensor can be applied for industrial applications.
Figure 18 shows an example of the micro thermal sensor employed in a magnetic
Fig. 18 An application of the Magnetic head slider

micro thermal sensor for
detection of surface defects on Heater
a magnetic disk surface
(Shimizu et al. 2011)
Contact sensor
Read element
Write element
Disk
head slider of hard disk drives for the detection of surface defects on a magnetic
disk surface. In the case, a micro thermal sensor was embedded to magnetic
head structures in a slider. The micro thermal sensor was placed between a read
element and a write coil in a magnetic head structure so that the micro
thermal sensor will be placed as close as possible with respect to a disk surface.
Since a magnetic head slider is designed to follow the disk surface with a
certain amount of gap, a collision between the micro thermal sensor and the
disk surface will occur only when a convex-shaped surface defects exist on the
disk surface.
Figure 18 shows a typical response of a micro thermal sensor embedded to a
magnetic head slider when detecting a surface defect on a magnetic disk surface.
Due to high sensitivity of the micro thermal sensor and high-speed in-plane motion
between the sensor and a rotating disk surface, even a small surface defect having
nanometric height can also be detected. Figure 19a shows the image of a certain
area of a disk surface observed by a conventional laser scattering method, where
the micro thermal sensor found the existence of a surface defect. As can be seen
in the figure, the conventional laser scattering method could not find out the
existence of a surface defect. Meanwhile, the micro thermal sensor successfully
found the defect, whose SEM image is shown in Fig. 19b, having a diameter and
a height of less than 200 nm and 10 nm, respectively. As has been demonstrated
in this experiment, the micro thermal sensor can be employed as a highly sensitive
contact sensor for surface defect detection (Fig. 20).
Summary and Outlook
In this chapter, design and development of the micro thermal sensor for surface
defect detection has been described. Throughout some basic experiments, feasi-
bility of the developed micro thermal sensor as a thermal sensing device and a
contact detection device has successfully been verified. In addition, an industrial
application of the micro thermal sensor for a magnetic head slider has also
534 Y. Shimizu
Voltage 10 mV/div
Voltage 10 mV/div
Contact sensor
< 1 μs
Acoustic emission
(AE) sensor
Time 100 μs/div Time 5 μs/div
Fig. 19 Typical response of the micro thermal sensor embedded to a flying magnetic head slider
when detecting a surface defect on a magnetic disk surface (Shimizu et al. 2011)
Fig. 20 Defect detection on a magnetic disk surface by using a micro thermal sensor. (a) Disk
surface image captured by a conventional laser scattering method. (b) SEM image of a surface
defect detected by the method using a micro thermal sensor (Shimizu et al. 2011)
demonstrated that the surface defect detection by the micro thermal sensor could
be a powerful tool for evaluation of the surface quality of magnetic disks. Since the
micro thermal sensor can be fabricated in a simple manner, the sensor is expected
to be employed in various industrial applications which require components
having smoothly finished surfaces. Meanwhile, one of the disadvantages of the
surface defect detection by using the micro thermal sensor is that at the moment of
collision the surface defect detection could give damages on a surface defect, a
precise investigation of which will be carried out in the review process followed by
the defect detection. An establishment of a nondestructive method for the surface
defect detection with the micro thermal sensor is therefore desired, which will be
carried out in future work.
References
Attota R, Silver R (2011) Nanometrology using a through-focus scanning optical microscopy
method. Meas Sci Technol 22:024002. 10pp
Gao W (2010) Precision nanometrology. Springer, London
Gustafsson SE (1991) Transient plane source techniques for thermal conductivity and thermal
diffusivity measurements of solid materials. Rev Sci Instrum 62:797–804
Johnson KL (1982) One hundred years of hertz contact. Proc Inst Mech Eng 196:363–378
ITRS overview (2011) http://www.itrs.net [Accessed 14 Jan 2015]
Lonardo PM (1991) Surface characterization and defect detection by analysis of images obtained
with coherent light. CIRP Ann – Manuf Techn 40:541–544
Lu W, Shimizu Y, Ito S, Gao W (2012) Design and experiment of thermal contact sensor detecting
defects on Si wafer surface. Key Eng Mater 523–524:826–831
Meshulach D, Dolev I, Yamazaki Y, Tsuchiya K, Kaneko M, Yoshino K, Fujii T (2010) Advanced
lithography: wafer defect scattering analysis at DUV. Proc SPIE 7638:76380K. (10pp)
Montal O, Dotat K, Mebarki B, Man-Ping C, Ngai C (2010) DUV inspection and defect origin
analysis for 22nm spacer self-aligned double-patterning. Solid State Techn 53:16–19
Pan J, Tai DH (2011) A new strategy for defect inspection by the virtual inspection in semicon-
ductor wafer fabrication. Comp Ind Eng 60:16–24
Shimizu Y, Xu J, Kohira H, Kurita M, Shiramatsu T, Furukawa M (2011) Nano-scale defect
mapping on a magnetic disk surface using a contact sensor. IEEE Trans Magn 47(10):3426–3432
Shimizu Y, Lu W, Ohba Y, Gao W (2014a) Feasibility study on the concept of thermal contact
sensor for nanometre-level defect inspections on smooth surfaces. Meas Sci Technol 25:064006.
(11pp)
Shimizu Y, Ohba Y, Gao W (2014b) Design of fabrication process of a thermal contact sensor for
surface defect inspection. J Adv Mech Des Syst Manuf 8(4):14–00099. (14pp). https://doi.org/
10.1299/jamdsm.2014jamdsm0052
Shimizu Y, Ohba Y, Gao W (2015) Investigation on sensitivity of a contact-type thermal sensor for
surface defect inspections. Int J Autom Technol 9(3):291–296
Takahashi S, Miyoshi T, Takaya Y, Saito K (1998) In-process measurement method for detection
and discrimination of silicon wafer surface defects by laser scattered defect pattern. CIRP Ann
Manuf Techn 47:459–462
Takahashi S, Kudo R, Usuki S, Takamasu K (2011) Super resolution optical measurements of
nanodefects on Si wafer surface using infrared standing evanescent wave. CIRP Ann Manuf
Techn 60:523–526
Takami K (1997) Defect inspection of wafers by laser scattering. Mat Sci Eng :B 44:181–187
Tan CM, Lau KT (2011) Automated wafer defect map generation for process yield improvement.
Proceedings of the 13th international symposium on integrated circuits, Singapore, 2011
Wagner C, Harned N (2010) Lithography gets extreme. Nat Photonics 4:24–26
Xu C, Lee J, Sachan V, Patterson OD (2013) defect sampling methodology for yield learning during
22nm process development. Proceedings of IEEE advanced semiconductor manufacturing
conference, New York, 2013
X-Ray Computed Tomography for
Dimensional Metrology 17
Filippo Zanini and Simone Carmignato
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 538
Historical Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 538
Evolution of CT Scanners . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 539
Industrial X-Ray CT Systems and Components . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 542
X-Ray Source . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 543
X-ray Detector . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 545
Frame and Mechanical Axes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 546
CT Scanning and Measuring Procedures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 547
Generalized CT Scanning Procedure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 548
Additional Steps for CT Dimensional Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 551
Error Sources . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 553
Main Influence Factors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 553
CT Image Artifacts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 558
Metrological Performance Verification and Uncertainty Evaluation . . . . . . . . . . . . . . . . . . . . . . . . . . . 562
Metrological Performance Verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 562
Determination of Measurement Uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 565
Reference Objects . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 567
Applications of CT for Dimensional Metrology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 571
Dimensional Measurements and Tolerances Verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 571
Volume Measurements . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 574
Assembly Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 575
Fiber Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 577
Surface Metrology of AM Parts . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 577
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 579
F. Zanini · S. Carmignato (*)

Department of Management and Engineering (DTG), University of Padova, Vicenza, Italy
e-mail: filippo.zanini@unipd.it; simone.carmignato@unipd.it

https://doi.org/10.1007/978-981-10-4938-5_19
538 F. Zanini and S. Carmignato
Abstract
X-ray computed tomography (CT) has emerged over the last years as an innova-
tive dimensional measuring technique and has been increasingly applied in
industry. This chapter describes the state of the art, the main technical character-
istics, and examples of applications of CT in industrial dimensional metrology.
Although still in its youth, metrological CT offers unique solutions and provides
several advantages in comparison to other coordinate measuring systems such
as tactile coordinate measuring machines. In particular, CT systems allow
reconstructing holistic three-dimensional models of the scanned workpieces,
which are then used to obtain nondestructive and noncontact measurements of
outer as well as inner features. However, important drawbacks still limit a wider
acceptance of CT in industrial metrology. One of the most critical aspects is the
establishment of metrological traceability, which is often challenging due to many
and complex error sources that affect CT measurements and complicate the
evaluation of metrological performances and of task-specific uncertainties
Keywords
X-ray computed tomography · Dimensional metrology · Industrial applications ·
Advanced manufacturing · Additive manufacturing · Industry 4.0
Introduction
X-ray computed tomography (CT), developed in the 1970s for medical applications
and later employed in nondestructive testing of materials, has recently emerged as an
advanced measuring technique for dimensional metrology (Kruth et al. 2011). CT is
an imaging method using X-rays to irradiate an object and mathematical algorithms
to obtain a cross-sectional image of the object or a sequence of such images. The
term tomography derives from two Greek words: tomos, which means section or
cutting, and graphien, which means to draw. To form the cross-sectional image or
images of the object being scanned, a number of X-ray radiographic projections
are first acquired and then processed using mathematical algorithms (Hsieh 2015).
The reconstructed images (slices) can be stacked to form a three-dimensional (3D)
representation of the object, which can be used in a wide series of applications.
Historical Background
The development of CT scanning started in the last century. In 1917, the mathema-
tician Johann Radon showed that an object can be reconstructed from an infinite set
of its own projections using the “Radon transformation” (Radon 1986). The basic
idea of today’s tomography can be found in a patent by Gabriel Frank, granted in
1940 (Frank 1940). However, the method described in such patent was difficult to be
applied at that time due to the lack of appropriate computational technology. Years
17 X-Ray Computed Tomography for Dimensional Metrology 539
Fig. 1 (a) The CT scanner developed by Godfrey Hounsfield in 1967 (Bioclinica 2011); (b) image
from the first clinical scan of a patient revealing a brain tumor (Impactscan 2013); (c) 3D
reconstruction of a car part obtained from and industrial CT scanner (Kruth et al. 2011)
later, in 1963, Allan M. Cormack reported the findings from investigations using
what can be considered the first CT scanner actually built (Cormack 1963). The
development of the first clinical CT scanner (see Fig. 1a) began in 1967 with
Godfrey N. Hounsfield (Hsieh 2015). He demonstrated that X-ray radiographs of a
body taken from different angular directions allow the reconstruction of slices
showing its internal structure (Hounsfield 1976). A scan of the first patient success-
fully identified a brain tumor; an image from this scan is shown in Fig. 1b (Paxton
and Ambrose 1974). Thanks to the invention of the CT technology, in 1979 both
Cormack and Hounsfield were awarded the Nobel Prize in Physiology and Medi-
cine. The first scanner capable of full body scans of patients (automatic computer-
ized transverse axial (ACTA) scanner) dates back to 1974 (Ledley et al. 1974). In the
1980s, advanced micro-focus X-ray CT systems based on the same principles of
medical CT started to be used in industry for material analysis and nondestructive
testing (NDT) (Reimers and Goebbels 1983). In the following years, industrial CT
has evolved to produce always more advanced analyses. An example of industrial
CT reconstruction at the state of the art can be seen in Fig. 1c, which, compared to
early CT images like the one shown in Fig. 1b, is characterized by high resolution
and 3D geometry. In the last years, CT has emerged as an innovative measuring
technique in the field of coordinate metrology (Kruth et al. 2011). The first CT
system specifically developed for dimensional measurements was exhibited at the
2005 Control Fair in Germany (Kruth et al. 2011).
Evolution of CT Scanners
The evolution of clinical and industrial CT scanners is summarized in the two

following sections.
Clinical CT Scanners
The five generations of clinical CT scanners are presented here below with regard to
the reconstruction of a single CT slice. In order to obtain volumetric scans, the patient
needs to be translated through the axis of the scanner, to reconstruct multiple slices.
The first-generation CT scanners refer to the Hounsfield’s scanner built in 1971

(see Fig. 1a). In this scanner, a collimated pencil beam of X-rays is used to acquire a
single data point at a time. The X-ray source and detector are initially moved linearly
to acquire data points across the entire scan plane; then they are progressively rotated
(e.g. with increments of 1 ) around the object, as shown schematically in Fig. 2a.
The second generation of CT scanners (shown schematically in Fig. 2b) was built
starting from the first-generation design, diminishing the number of rotation steps
and the scan time by using multiple pencil beams covering different angles. The third
generation of CT scanners (see Fig. 2c) is characterized by the simultaneous rota-
tional movement of the X-ray source and detector around the patient, allowing for
faster scans. In this configuration, the X-ray source emits a fan beam that reaches an
opposite array of detector cells placed on an arc. The third-generation design was
modified in the so-called fourth generation by using a stationary (hence more stable)
enclosed ring detector (see Fig. 2d) and an X-ray source that still rotates around the
patient (Hsieh 2015). The advantages are (i) the reduction of aliasing effect, (ii) the
dynamic recalibration of the detector, and (iii) increased stability. The fifth genera-
tion (Fig. 2e) was designed in such a way that no mechanical movement of the
source, detector, or patient is required, to obtain an optimal stability of the systems
(Carmignato et al. 2018).
Industrial CT Scanners
Industrial CT scanners present some important differences with respect to clinical
CT scanners. Industrial scanners can use higher-intensity X-ray sources as well as
longer scan times, given that in most cases the exposure to high X-ray dose is not a
concern for industrial parts. Since industrial CT can be used not only for material
characterization and nondestructive testing but also for metrology applications, the
attention is generally focused on achieving the maximum possible scan resolution,
accuracy, and precision. Another important difference between clinical and indus-
trial CT systems is that, while in clinical scanners the source and the detector are
typically rotated together around the patient, in most industrial systems, the sample is
rotated in between stationary source and detector. This configuration guarantees
higher stability of the system (Kruth et al. 2011). In addition, different scanning
parameter settings are used because the type of material to be scanned differs
significantly from medical to industrial applications (i.e., human tissues in medicine
and metals, ceramics, and polymers in industry). Also the sample size can differ
significantly depending on the specific application (De Chiffre et al. 2014). The most
common configurations of industrial CT systems are shown in Fig. 3. In the fan beam
configuration, the X-ray source generates a 2D fan beam of X-rays, which propa-
gates through the scanned object before reaching a line detector (1D detector). Fan
beam systems can acquire data of one single slice in one complete rotation of the
object. For this reason, for taking multiple slices along the object, translations are
needed along the rotation axis (Fig. 3a). On the contrary, cone beam CT scanners are
capable of acquiring 3D volumetric data in a significantly faster way, by using a cone
beam of X-rays to scan an entire part in one single full rotation. The most common
employed trajectory is circular (Fig. 3b), but, especially if the object to be scanned is
Fig. 2 First (a), second (b), third (c), fourth (d), and fifth (e) generations of clinical CT scanners
Area
detec
a b tor
Line
detec
tor
Object
am am
n be be
Fa e
on
C Object
X-ray
source
Rotation/translation axis X-ray
Rotation axis
source
Area
c detec
tor
am
be
o ne
C Object
X-ray
source Rotation/translation axis
Fig. 3 Most common configurations of industrial CT systems: (a) fan beam, (b) cone beam with
circular trajectory, (c) cone beam with helical trajectory
larger than the detector field of view, it is possible to move the object also using a
helical trajectory (Fig. 3c). The main advantage of fan beam scanning is the
reduction of some of the image artifacts (see section “Error Sources”) affecting the
cone beam CT systems. Metrological CT systems have evolved from industrial
scanners, implementing a series of improvements to enhance their measurement
accuracy and enable their use as coordinate measuring systems (CMS). For example,
metrological CT systems commonly have stable structures and high-precision
mechanical setups for accurate positioning of sample, detector, and source, as well
as thermally controlled cabinets (De Chiffre et al. 2014). In addition, these systems
are preferably kept in temperature-controlled rooms.
Industrial X-Ray CT Systems and Components
Conventional industrial X-ray computed tomography systems often use a cone beam
configuration which consists in three main hardware components (Fig. 3b, c): (i) X-
ray source (tube) emitting an X-ray beam, (ii) X-ray detector (flat panel) for
measuring the X-ray signal attenuation produced by the workpiece, and (iii) set of
mechanical axes for positioning the object between source and detector and for
providing the rotation during the CT acquisition. Moreover, from the computational
point of view, software for data acquisition, reconstruction and subsequent elabora-
tion and measurements is required.
Alternative configurations that are employed in industry are multi-sensor systems
(including additional sensors, such as optical or tactile probes, to reach higher
accuracies), four-dimensional X-ray CT systems (for studying the evolution over
time of samples subjected to in situ tests), dual-energy CT systems (DECT), at-line
and in-line CT systems, CT systems for large and highly attenuating objects (e.g.,
high-energy CT systems using linear accelerators – LINACs), synchrotron systems
(monochromatic beam, high photon flux, coherence, collimation, and high spatial
resolution), and laminography systems for flat parts (Carmignato et al. 2018).
The following sections give more details on the three main hardware components
of conventional cone beam industrial CT systems: X-ray source, X-ray detector, and
mechanical axes. The quality of the acquired data is directly influenced by the
characteristics and performance of these three components (De Chiffre et al.
2014). If nothing else is specified, the following parts of this chapter will implicitly
refer to cone beam CT systems.
X-Ray Source
X-rays are electromagnetic waves that were discovered by Wilhelm Conrad Röntgen
(who was awarded the first Nobel Prize for Physics in 1901 for this discovery). The
characteristic wavelength of X-ray radiation ranges between 0.01 and 10 nm, as
illustrated in Fig. 4.
The energy of each photon E is proportional to its frequency f, so that a smaller
wavelength corresponds to a higher energy:
hc
E ¼ hf ¼ (1)
λ
where h is Planck’s constant (6.63 1034 J s), c is the speed of light (3 108 m s1),
and λ is the wavelength of the X-ray (Hsieh 2015).
X-rays can be classified as soft and hard X-rays, depending on (i) the wavelength
and (ii) the ability to penetrate through materials. Soft X-rays are characterized by
longer wavelength X-rays (λ >0.1 nm), while hard X-rays are those with shorter
wavelengths. Although X-rays can be generated by different types of sources (linear
accelerators, synchrotrons, etc.), this section focuses on X-ray tubes (see Fig. 5a),
which are the most commonly used sources in industrial cone beam CT systems.
The two main components of a typical X-ray tube are a cathode and an anode. The
anode consists of a target, usually made of refractory metals, such as tungsten or
molybdenum. There are different types of targets used in industrial CT systems, e.g.,
Fig. 4 Electromagnetic spectrum with wavelengths

Fig. 5 (a) Schematic representation of an X-ray tube and (b) effect of different focal spot sizes and
different geometrical magnifications on the blurring content of a projection image
reflective, transmissive, rotating, or liquid. In the following, the case of a reflective

target (which is the most commonly adopted) will be addressed. An electron beam is
emitted from the cathodic filament (usually a tungsten wire) and, subsequently,
accelerated toward the anode (Buzug 2008). Magnetic deflectors and lenses are
commonly used to focus the electrons emitted from the filament onto a small region
on the target, called “focal spot.” The finite size of the focal spot is responsible of the
image blurring content (i.e., unsharpness). Therefore, the smaller is the focal spot,
the sharper will be the final image. Figure 5b schematizes the effect of different focal
spot sizes (combined with different geometrical magnifications of the scanned
object) on the blurring content of a projection image. Industrial CT systems can be
classified based on the achievable minimum size of the focal spot:
• Conventional or macro-CT systems, with focal spots larger than 0.1 mm

• Microfocus systems (μ-CT), with a spot size in the range of a few micrometers
• Nanofocus systems, with focal spot sizes lower than 1 μm
• Synchrotron CT systems, where focal spot sizes can go down to 0.2 μm or lower
(De Chiffre et al. 2014; Requena et al. 2009)
When hitting the target, the electrons are suddenly decelerated, and approxi-
mately 99% of their energy is converted into heat, while only about 1% contributes
to the X-ray generation (Hsieh 2015; Buzug 2008). The energy can be converted to
X-rays by two main interactions: (i) deceleration of fast electrons in the atoms of the
target material (bremsstrahlung radiation; the term is a contraction of the German
words bremsen “to brake” and Strahlung “radiation”) and (ii) emission of X-rays
generated by refilling a vacant place in the inner shell of the atom by an outer shell
electron (characteristic radiation). The first interaction is also called continuous
radiation as it produces a continuous spectrum, while the characteristic radiation
Fig. 6 Bremsstrahlung and 1 Relative

characteristic radiations intensity Bremsstrahlung
(Kruth et al. 2011)
Emax
0 X-ray energy (KeV)
1 Relative Characteristic
intensity Radiation
0 50 100 X-ray energy (KeV)
results in a discrete X-ray spectrum of characteristic peaks. Both radiation types are
shown in Fig. 6. The X-ray beam is usually collimated before the emission by
passing through a circular aperture (for conical beams) or collimating plates (for
fan beams). A window, typically made by light metals like beryllium or aluminum,
seals this aperture for maintaining the vacuum inside the tube.
X-ray Detector
During a CT scan of an object, the X-rays propagate through the object and are
attenuated due to absorption or scattering. The attenuation is measured by capturing
the attenuated X-rays using an X-ray detector, which converts the X-ray energy into
electrical signals. This conversion is commonly based on two alternative principles:
gas ionization and scintillation (solid state). In gas ionization detectors, X-ray energy
is converted directly into electrical energy, whereas scintillation detectors convert
X-rays into visible light, and then the light is converted into electrical energy
(Hermanek et al. 2018). Scintillation-type detectors (or indirect detectors) cannot
discriminate between different photon energies; hence the polychromatic nature of
bremsstrahlung spectra leads to beam hardening artifacts (see section “Error
Sources”). Recently, photon counting detectors – based on direct conversion of
X-ray photons into an electric charge proportional to the photon’s energy – have
received increasing attention in CT (Taguchi and Iwanczyk 2013) as they do allow
discriminating between different photon energies reaching the detector during a
single projection. The main advantages of such type of detectors are elimination of
beam hardening artifacts (improving the investigation of multi-material compo-
nents), increasing of the image contrast, and reduction of noise (Panetta 2016).
Unfortunately, they cannot be already used with the high energies typical of the
current industrial CT applications, as their count-rate capability is still limited
(Carmignato et al. 2018). Most industrial CT systems use flat panel detectors
based on a matrix array in combination with a cone beam. An alternative setup to
flat panel detectors are linear array detectors (used in CT fan beam configurations),
characterized by a single line of diodes used for generating the X-ray images. Some
important advantages of the fan beam configuration with linear detector are less pixel
interaction from adjacent pixels, elimination of scatter phenomena, and reduction of
noise with better signal-to-noise ratio. The disadvantage is that data acquisition is
much more time-consuming than with flat panel detectors. In fact, the object will be
scanned slice by slice, which requires one full rotation of the object per slice
combined with a vertical motion, which can also lead to additional errors in the
reconstruction (Welkenhuyzen 2015).
Frame and Mechanical Axes
In conventional industrial CT systems, the magnification axis (Z-axis) goes from the
X-ray source to the center of the detector, as schematically shown in Fig. 7.
The Y-axis is parallel to the rotation axis, while the X-axis is perpendicular to both
the Y- and Z-axes. The lateral opening angle of the X-ray beam is called the fan
angle, and the longitudinal opening angle is called the cone angle. The geometrical
alignment and stability of the kinematic structure are fundamental for the metrolog-
ical performances of the CT system (Dewulf et al. 2018). A conventional cone beam
CT system is considered aligned when the following conditions are satisfied
(Ferrucci et al. 2015):
• The magnification axis intersects the detector at its center.

• The magnification axis is orthogonal to the detector.
• The magnification axis intersects the axis of rotation at a 90 angle.
• The projection of the axis of rotation is parallel to the detector columns.
Fig. 7 Scheme of a CT system. X-axis and Z-axis (magnification axis) are horizontal translation
axes; Y-axis is both vertical translation axis and rotation axis. SOD is the source-to-object distance,
SDD the source-to-detector distance, and ps the pixel size
Fig. 8 Reconstructed volume of sphere objects in the presence of detector tilt +10 (rotation
around X-axis), detector slant +10 (rotation around Y-axis), and detector skew +2 (rotation around
Z-axis). Left: gray-value slice along XY-plane before gray-value thresholding. Center: magnified
portion of gray-value image. Right: three-dimensional view of the reconstructed sphere objects after
applying gray-value thresholding (Ferrucci et al. 2016)
In addition, during a scan, the relative movement of CT components must also

satisfy the following requirements:
• The position of the rotation axis is constant, and the relative distance between all
components is fixed.
• The plane of the rotation is orthogonal to the rotation axis.
When misalignments in the kinematic structure are present and not corrected for,
image artifacts and errors appear in the CT reconstruction, as illustrated, for example,
in Fig. 8.
CT Scanning and Measuring Procedures
Applying CT for dimensional measurements involves several steps (see Fig. 9),
including steps that are common to other CT applications (e.g., nondestructive
inspections, material analysis), as well as specific steps for dimensional measure-
ments. The general workflow is based on three main phases: (i) acquisition of X-ray
Fig. 9 Typical measurement workflow for industrial CT
projections of the object, (ii) reconstruction of a CT three-dimensional model, and

(iii) data elaboration. In this section, the three above cited phases will be described
(section “Generalized CT Scanning Procedure”) together with the additional steps
required when performing dimensional measurements (section “Additional Steps for
CT Dimensional Measurements”).
Generalized CT Scanning Procedure
Acquisition of X-ray Projections

The first phase, common to all CT applications, is the acquisition of a number of
bidimensional X-ray projections (i.e., radiographs) taken at different angular posi-
tion of the workpiece during its rotation. The angular step between the acquired
projections has to be defined by the user. Although the accuracy of reconstructions
depends on the number of projections, increasing the number of projections
increases the scanning time. For each angular position, the X-ray beam propagates
through the workpiece material, and its intensity is reduced exponentially due to the
attenuation (see Fig. 10).
The amount of attenuation depends on (i) the material length to be penetrated
by the radiation, (ii) the material composition and density (i.e., attenuation coeffi-
cient μ), and (iii) the energy of the X-rays. The attenuation is measured by capturing
the transmitted X-rays that traverse the object by means of an X-ray detector,
producing a 2D image in the case of a flat panel detector or a 1D profile in the
case of a 1D line detector. Measuring this attenuation allows detecting the presence
of one material or the presence of various materials as well as the lengths travelled
inside the various materials (which is linked to the workpiece thickness). For an X-
ray beam with monochromatic spectrum traversing a homogeneous object, the X-ray
attenuation can be expressed by the following relationship:
dI
¼ μdx (2)
I ð xÞ
where I is the X-ray intensity and x is the distance travelled through the material.
From Eq. (2) the Beer-Lambert law (Fig. 10) can be obtained, stating that each
material layer of equal thickness absorbs an equal fraction of traversing radiation:
Fig. 10 Attenuation of X-rays by the Beer-Lambert law. Attenuation in the case of a homogeneous
object with an attenuation coefficient μ1 and thickness x is depicted in the first part of the image
(left). Attenuation for an object composed of different materials with different attenuation coeffi-
cients μ2, μ3, . . .μn is shown in the right part of the figure
I ðxÞ ¼ I 0 eμx (3)
where I0 is the incident X-ray intensity and I is the residual intensity after traversing
the material. However, as the material is usually not homogeneous, a varying
attenuation coefficient should be considered. Consequently, in the exponent of
Eq. (3), the linear integral along a path in which μ varies in every point of the path
must be added (Fig. 10). The resulting equation is:
ð
I ¼ I 0 exp μðxÞdx (4)
In addition, to adapt the relation to a polychromatic X-ray beam spectrum, Eq. (4)
must be modified as follows:
ð ð
I ¼ I 0 ðE Þexp μðE,xÞdx dE (5)
where E is the energy of the X-rays.

The attenuation in the power range typical for industrial CT systems (i.e.,
20–450 kV) includes both absorption and scattering, which are due to two effects:
photoelectric effect and Compton (scatter) effect, respectively (see Fig. 11).
Photoelectric absorption occurs when the total energy of an incoming X-ray photon
is transferred to an inner electron, which is consequently ejected. In Compton
scattering, the incoming X-ray photon interacts with a free or outer electron, ejecting
the electron. As a result of this interaction, the X-ray photon is deflected in a different
direction with some loss of energy, which is gained by the electron. The total
attenuation, calculated by Eq. (5), is therefore the result of the initial X-ray spectrum
emitted by the source Io(E) and the ray energy-dependent attenuation coefficient of
the material μ(E), summed up over the energy intervals ΔE and the length incre-
ments Δx travelled through solid material.
Reconstruction
After the acquisition of X-ray projections, a reconstruction process has to be
performed by means of mathematical algorithms, in order to obtain a 3D volumetric
Fig. 11 Schematic a
representation of transmitted
radiation with no interaction
between electrons (a), Photoelectron
photoelectric absorption (b), b
and Compton scattering (c)
Visible light photon

Nucleus
c Recoil electron
Incident
X-ray photons Soft X-ray photon
image of the object. The reconstructed 3D volume can be described as a three-

dimensional matrix where each cell is a voxel, i.e., the volumetric representation of
pixel. The voxel gray value is a measure for the attenuation capacity of the specific
material (related to composition, density, etc.). In cone beam industrial CT, recon-
struction is usually performed by “filtered back projection,” which is based on the
“linear integral transformation,” a mathematical model developed by Radon in
1917 (Radon 1986) (see section “Introduction”). The input for the reconstruction is
the “gray-value profiles,” representing the gray value of each pixel located on one
line of the detector. The projections are filtered before the reconstruction in order to
achieve higher reconstruction quality. Figure 12a shows an example of unfiltered
back projection, resulting in a blurry version of the correct image, while Fig. 12b is
an example of filtered back projection resulting in reconstruction of the image with
reduced blurring content. The reconstruction is often based on the Feldkamp-
Davis-Kress (FDK) algorithm, formulated in 1984 based on the filtered back
projection method (Feldkamp et al. 1984). The algorithm has originally been
created for a circular trajectory and a flat detector; however, it can be adapted to
other trajectories (e.g., helical trajectories). In filtered back projection algorithms,
the focal spot is assumed as a point, and the radiation is considered as monochro-
matic. Moreover, noise-free projections are assumed. In real industrial CT, how-
ever, the focal spot size has finite dimensions, the radiation is polychromatic, and
projections are affected by noise. As the FDK does not take into account the
polychromaticity of the radiation, beam hardening artifacts (described in section
“Error Sources”) can affect the quality of reconstruction. To overcome some of the
abovementioned problems, iterative reconstruction methods can be used (Buzug
2008), including statistical reconstruction algorithms (Fessler 2000). With iterative
reconstruction, a better quality of results can be achieved when data are missing
(e.g., in the case of limited angle tomography) or when metal artifacts occur (Hsieh
2015), but the computational time is normally significantly higher than with FDK-
based algorithms.
Fig. 12 (a) Back projection view 1

a
reconstructs an image by
taking each view and
smearing it along the path it view 2
was originally acquired; here
views are represented as gray-
value profiles. (b) Filtered
back projection reconstructs
an image by filtering each
view before back projection
(Smith 1999; www.DSPguide.
com with permission) view 3
a. Using 3 views b. Using many

views
b filtered view 1
filtered
view 2
filtered
view 3
a. Using 3 views b. Using

views
many
Data Elaboration
When the reconstructed three-dimensional volume is obtained, the generalized
procedure for data elaboration consists in the following steps: (i) preprocessing
and data enhancement (e.g., global or local smoothing by means of different types
of filters), (ii) segmentation (i.e., partitioning of an image or a volume into two or
more regions), and (iii) feature extraction (e.g., voids, pores, inclusions, fibers).
Additional Steps for CT Dimensional Measurements
Determination of Scale Factor

Traceable dimensional measurements require additional steps, including accurate
determination of the voxel size and the scale factor. The scale factor is often deter-
mined by measuring a calibrated reference object (e.g., ball bars or other calibrated
standards). This measurement – which can be done prior, together, or/and after the
measurement of the actual workpiece – allows identifying a global scale factor to link
the voxel size to the unit of length (i.e., meter) (Kruth et al. 2011). The measurement
should be performed in the same position of the magnification axis (i.e., position of the
rotary axis between the source and detector) that will be used during the actual
measurement, without intermediate movement of the axis. In fact, the lack of repeat-
ability in axis positioning may change the scale factor and introduce relative errors.
Although important, scale factor determination is of course not sufficient for ensuring
traceable CT dimensional measurements (see section “Metrological Performance
Verification and Uncertainty Evaluation”); further procedures need to be implemented
for system qualification (Bartscher et al. 2018) and calibration (Ferrucci et al. 2015).
Surface Determination
Once the reconstruction is done, the 3D reconstructed voxel model of the scanned object
can be processed by analysis and visualization software. Typically, in such software the
sample can be virtually sectioned for visual inspections of inner and outer features and
geometries as well as internal voids and inclusions. Most importantly, in order to be able
to perform dimensional measurements, the object’s surface has to be determined. With
this aim, a threshold has to be defined (locally or globally) to distinguish between material
and background (e.g., air) or between two materials characterized by different densities.
A common approach is to use an iso50 threshold value (Otsu 1979), defined as the
average value between air (background) and material (object), as seen in Fig. 13. Due to
Fig. 13 (a) iso50 threshold method, (b) gray image showing the “partial volume effect,” and (c)
determined surface with intra-voxel interpolation
the “partial volume effect” (i.e., CT attributes an average gray value to each voxel; see
Fig. 13b), an intra-voxel interpolation is needed to locate the edge with sub-voxel
resolution (Fig. 13c). The iso50 threshold value can be used for a global surface
determination. Under ideal conditions, global surface determination techniques would
identify the optimal surface for homogeneous single material components. However, the
global surface determination approach can produce erroneous surface models, due to
several reasons, including the inherent presence of image artifacts (see section “Error
Sources”), irregularities of the scan process, inhomogeneity of specimens material, as
well as numerous other effects. Therefore, often the global method does not ensure
sufficiently accurate results, and, hence, more advanced methods are required for surface
determination. In particular, local adaptive methods can be used (Kruth et al. 2011).
These methods optimize the surface determined locally in each voxel. In dimensional
metrology CT, particular attention is given to comparing different surface determination
methods (Yagüe-Fabra et al. 2013), as they can greatly influence dimensional measure-
ment errors (Carmignato et al. 2009; Leonard et al. 2014).
Specific Software Analyses for Dimensional Metrology

CT dimensional metrology applications require implementation of specific software
functionalities, including (i) fitting of geometrical elements on voxel data or surface data
for dimensional measurement and geometrical tolerance verification (Fig. 14a); (ii)
nominal-actual comparison (Fig. 14b), where actual CT measurements are compared
with reference data coming, for example, from a CAD nominal model; and (iii) wall-
thickness analysis (Fig. 14c). All these analyses can be performed also on geometries or
features that are difficult to access (e.g., internal) with tactile or optical coordinate
measuring systems.
Error Sources
A large variety of factors affect the metrological performances of X-ray computed

tomography over the entire measurement procedure. In the following section, “Main
Influence Factors,” an overview is given on the main influence factors, subdivided
into five groups as seen in Table 1. Subsequently, in section “CT Image Artifacts,”
the most common image artifacts are described. It is worth noting here that, in CT,
the term artifact refers to systematic discrepancies causing shadings, rings, streaks,
or bands in the reconstructed CT volume that are actually not present in the scanned
object.
Main Influence Factors
A systematic review of the factors affecting the CT measurement results is reported

in the German guideline VDI/VDE 2630-1.2 (2008). In the following, the main
factors are presented, subdividing them in groups according to Table 1.
a b
Deviation [mm]
0.5000
0.4000
0.3000
0.2000
0.1000
0.0000
-0.1000
-0.2000
-0.3000
-0.4000
-0.5000
20 mm
c
Wall thickness [mm]
0.55000
0.52700
0.50400
0.48100
0.45800
0.43500
0.41200
0.38900
0.36600
0.34300
0.32000
7 mm
Fig. 14 (a) Dimensional measurement and geometrical tolerance verification on an aluminum test
part produced by micro-milling (Möhring et al. 2015); (b) nominal-actual comparison of an
injection-molded polymeric component; (c) wall-thickness analysis performed on a Ti6Al4V
sheet in order to assess the impact of thermal treatments on deformations, shown in a 2D section
and in the 3D reconstructed volume (top-right). (Courtesy of University of Padova)
CT System
As described in section “Industrial X-ray CT Systems and Components,” a typical
industrial CT system is composed of three main components: (i) X-ray source, (ii) X-
ray detector, and (iii) positioning system. The performance of each of these compo-
nents can deeply influence the dimensional measurements done by CT.
Concerning the X-ray source, several influence quantities have to be considered. In
particular, accelerating voltage, filament current, focal spot size, and target character-
istics can all have significant effects on the final quality of both acquired data and
consequent reconstruction and measurements. The higher is the X-ray voltage gener-
ated, the more the X-rays will penetrate the item being scanned. On the other hand, as
the intensity of the filament current increases, the number of emitted electrons rises,
leading to a higher intensity of X-rays at the specific selected voltage and to a higher
image brightness. Both current and voltage influence also the actual focal spot size
(see section “Industrial X-ray CT Systems and Components”). The spot size is
commonly in the micrometer range and increases with increasing power to prevent
Table 1 Main influence Group Influence factors

factors in CT. (Adapted
CT system X-ray source
from Carmignato et al.
2018) Detector
Positioning system
Workpiece Material composition
Dimension and geometry
Surface texture
Environment Temperature
Vibrations
Humidity
Data processing 3D reconstruction
Surface determination
Surface extraction
Operator Workpiece fixturing and orientation
Magnification
X-ray source settings
Number of projections and image averaging
Measurement strategy
the target from melting. According to Hiller et al. (2012), the focal spot approximately
increases of 1 μm per each watt of power. The shape of the X-ray spectrum depends on
the target material and thickness. Tungsten targets are widely used in industry because
tungsten has a high atomic number, which increases the intensity of the X-rays, and
because it has a very high melting temperature. Conversely, targets made from low
atomic number elements, such as copper and molybdenum, are well suited for imaging
low absorption workpieces with high contrast (Carmignato et al. 2018).
Coming to the X-ray detector, many of its characteristics may affect the CT image
quality. For example, the pixel pitch size (which is a fixed characteristic of the
detector) has an impact on the spatial resolution and the signal-to-noise ratio (SNR).
Small pixel pitch sizes improve the spatial resolution, making the detector
unsharpness negligible. Large pixel pitch sizes result in a worst spatial resolution
but better SNR as they have a higher fraction of pixel area sensitive to light.
Finally, concerning the positioning system, it influences the CT measurement
results through misalignments and instabilities of the relative position and orienta-
tion of the three main components: X-ray source, rotary table, and X-ray detector
(Ferrucci et al. 2015). Current research work is ongoing for improving the methods
available for determining and compensating the geometrical misalignments of CT
systems (Dewulf et al. 2018).
Workpiece
CT is capable of measuring almost any type of workpiece, provided that its dimen-
sions are compatible with the measuring volume, and that X-rays are able to
penetrate it, producing images with sufficient contrast at the detector. Table 2
Table 2 Typical maximum penetrable material thickness for common industrial materials. All
values in the table ensure a minimum transmission of around 14% (De Chiffre et al. 2014)
X-ray voltage 130 kV 150 kV 225 kV
Steel <5 mm <8 mm <25 mm
Aluminum <30 mm <50 mm <90 mm
Plastic <90 mm <130 mm <200 mm
Fig. 15 Schematic representation of the low-pass filtering produced by tactile measuring machines
(left) and by X-ray computed tomography (right). The effect of different probe sizes and different
voxel sizes is shown, respectively
shows the typical maximum penetrable thicknesses for a general CT system at

different voltages. Higher penetration lengths are achievable by accepting lower
transmission values (De Chiffre et al. 2014).
Another workpiece characteristic influencing the CT performance is surface
texture. Parts with highly rough surfaces – e.g., castings and additive manufactured
(AM) parts – can cause significant deviations between CT and tactile dimensional
measurements, due to the different surface filtering characteristics of the two mea-
suring methods (as schematically shown in Fig. 15). Tactile CMMs produce a
mechanical low-pass filtering which depends on the probe size; CT also generates
a low-pass filtering of both peaks and valleys depending mainly on the spatial
resolution. The effect of surface texture produces an offset of the CT-measured
surface from the reference surface measured by CMM (Carmignato et al. 2017).
Environment
CT systems produce heat, especially at the X-ray source. Therefore, maintaining
constant and stable temperatures as well as avoiding thermal gradients inside the CT
cabinet is often challenging. This is difficult even for metrological CT systems where
powerful cooling systems are installed for both X-ray tube and cabinet. Possible
temperature variations can cause dimensional changes of both the workpiece and
the CT system, depending on the thermal proprieties of the materials involved.
Moreover, such variations can modify the kinematics of the CT components, causing
a magnification of the geometrical errors. A good practice is to leave the workpiece
positioned in its fixture inside the cabinet for some time in order to reach equilibrium
with the system’s temperature in a stable way. The temperature and humidity range
of the room hosting the CT system is also very important to obtain accurate
performance from CT (Carmignato et al. 2018).
Data Processing
The main data processing steps (i.e., reconstruction, surface determination, and data
elaboration) are described in section “CT Scanning and Measuring Procedures” of this
chapter. Reconstruction in industrial CT is commonly conducted using the filtered
back projection algorithm (see section “CT Scanning and Measuring Procedures”).
Several types of low-pass filters with high-frequency cutoffs can be adopted, with
possible different outcomes in terms of smoothing and achievable structural resolution
(Carmignato et al. 2018) (see section “Metrological Performance Verification and
Uncertainty Evaluation”). Surface determination for dimensional metrology is usually
based on thresholding methods, as already explained in section “CT Scanning and
Measuring Procedures.” Several works in literature have documented that the local
adaptive approach leads to more repeatable and accurate measurements than global
methods (Kruth et al. 2011; Borges de Oliveira et al. 2016). The impact of surface
determination on the measurement depends on the feature investigated. Features such
as diameters and bidirectional length measurements are more sensitive to the deter-
mined surface than unidirectional measurements (Carmignato et al. 2018).
Operator
The CT measurement workflow described in section “CT Scanning and Measuring
Procedures” is widely influenced by the operator that is responsible for the choice of
a high number of settings and of the measurement strategy. The user’s influence
starts with the system warm-up, deciding how much time will be dedicated to let the
workpiece and the system reach the thermal stability. The user is also responsible for
the workpiece fixture that (i) should firmly hold the workpiece avoiding any kind of
movement during the CT acquisition step and (ii) should be made of a very low
absorption material (e.g., polyurethane and polystyrene foam) in order to reduce its
influence on the acquisition quality. However, polymeric fixtures can suffer for
material relaxation and thermal instability. The orientation of the workpiece has a
significant impact on the maximum thickness that has to be penetrated by the X-ray
beam. The higher the maximum thickness, the higher the demanded X-ray power
and, consequently, the larger will be the focal spot size (see section “CT Scanning
and Measuring Procedures”). The choice of scanning parameters is made by the user,
sometimes supported by software expert systems or predefined parameter sets
(Schmitt et al. 2018). One of the most important parameters is the magnification
factor, which can be modified by changing the relative position of the object with
respect to the X-ray source and the X-ray detector. The voxel size (vs) of the
reconstructed three-dimensional model of the scanned object results from the fol-
lowing equation:
ps
vs ¼ (6)
m
where ps is the detector pixel size and m is the geometrical magnification, which is
calculated as the ratio between the source-to-detector distance (SDD) and the source-
to-object distance (SOD):
SDD
m¼ (7)
SOD
The voxel size should always be larger than the focal spot size to limit image
blurring. Other important parameters are those related to the X-ray source settings, e.
g., acceleration voltage and current. Such parameters must be chosen in order to have
sufficient X-ray energy to penetrate the workpiece at all rotation angles and to avoid
too dim or saturated images at any rotation angle. The choice of the number of
projections intrinsically determines also the duration of scanning and reconstruction
phases, as well as the amount of generated data. It is, therefore, preferable to keep
this number at the minimum value still providing an acceptable accuracy. The
correlation between measurement accuracy and number of projections has been
reported in different studies (Weckenmann and Krämer 2013; Villarraga-Gómez
et al. 2016). Image averaging reduces the random noise in the projections by
averaging more projections taken at the same angular position; however, this
increases the scanning time.
CT Image Artifacts
Image artifacts are defined in ASTM E 1441-11 as “discrepancies between the actual
value of some physical property of an object and the map of that property generated
by a CT imaging process.” In the following, a selection of the main image artifacts
that may appear in CT reconstructions is presented.
Beam Hardening
Beam hardening concerns the nonlinear attenuation of an X-ray beam along its
propagation path through a workpiece due to the polychromatic nature of the
spectrum and the energy-dependent X-ray attenuation (Dewulf et al. 2012). As the
beam passes through the absorbing material, low-energy (soft) X-rays are more
rapidly attenuated than high-energy (hard) X-rays. Soft X-rays are absorbed after
the first few millimeters of material to be traversed, principally because of photo-
electric absorption. Consequently, the beam spectrum becomes mainly composed by
high-energy photons; hence, it becomes harder to be attenuated as it penetrates the
workpiece (Lifton and Carmignato 2017). Normally, reconstruction algorithms are
based on the wrong assumption of a linear attenuation. For this reason, the high
absorption of soft X-rays at the edge of the part gives the false impression that the
skin of the part is made of more absorbing material, resulting in the so-called
cupping effect (Fig. 16).
Gray value
813
613
Internal gray value variation: 90
413
213
13 [mm]
2mm 0.0 1.2 2.4 3.6 4.8 6.0 7.2
Gray value
1135
Internal gray value variation: 300

535
–65
[mm]
2mm
0.0 1.2 2.4 3.6 4.8 6.0 7.2
Fig. 16 Reconstructed slices of a steel cylinder and corresponding gray-value profile along the
red line without beam hardening correction (top) and after beam hardening correction (bottom).
The so-called cupping effect can be observed when the beam hardening is not corrected (Dewulf
et al. 2012)
Beam hardening can be reduced or eliminated following two main approaches:
– Physical filters with a certain thickness, usually made by Al, Cu, or other metals,
and interposed between the X-ray source and the object, to filter out the soft
component of the radiation
– Software correction through linearization technique based on an estimation of the
relation between a propagated path length within the object and a corresponding
measured intensity (see Fig. 16) (Dewulf et al. 2012)
Scatter
Another wrong assumption made by most CT reconstruction algorithms is that all X-
rays go from the source to the detector following a straight line. However, together
with the primary radiation, a secondary radiation, known as scatter, reaches the
detector. Scatter can be generated from both workpiece and environment (Schörner
2012). Workpiece scatter is due to the interaction between X-rays and workpiece
material. Environmental scatter can be caused by all the components that are within
the CT system’s cabinet. Scatter produces a high background signal in all projections
and a general loss of contrast. In the reconstructed images, scatter causes artifacts
that are similar – but typically less pronounced – to those due to beam hardening
(Hunter and McDavid 2012).
Feldkamp Artifacts
The so-called Feldkamp effect is related to the algorithm typically used for 3D
reconstruction in CT, the Feldkamp-Davis-Kress algorithm (Feldkamp et al. 1984);
see section “CT Scanning and Measuring Procedures.” The resulting artifact (some-
times referred to as the “cone beam artifact”) appears only in reconstruction obtained
from 3D cone beam CT systems using circular trajectory of the scanned object. As
illustrated in Fig. 17a, in fan beam reconstruction, each individual point (e.g., point
A in Fig. 17a) can be obtained by summing the filtered projection values along a
trajectory that is seen as a straight line on the projection plane (trajectory TA in
Fig. 17a). On the contrary, in cone beam reconstruction, points far from the central
detector horizontal plane (e.g., point B in Fig. 17a) follow an elliptical trajectory (TB
in Fig. 17a). In most cases, the resulting errors increase with higher cone angles (θ),
Fig. 17 (a) Projection trajectory of fan beam scanning and cone beam scanning (Adapted from
Xue et al. 2015); (b) CT reconstruction slices acquired with a cone angle of (i) 30 , (ii) 11 , and (iii)
5 . The artifacts increase as the cone angle increases (Carmignato et al. 2018)
producing strong artifacts on the surfaces parallel to the beam, as shown, for
example, in Fig. 17b (Carmignato et al. 2018).
Ring Artifacts
The presence of non-ideal or defective detector pixels (often referred as to “bad
pixels”) may cause (if not properly corrected) the formation of ring artifacts in the
CT reconstructed volume. A bad pixel causes a wrong signal detection at the same
location in each CT projection. During the back projection step in the reconstruction
process, these bad pixels will form concentric rings in axial planes in the CT volume
due to the circular trajectory.
Noise
Random image noise can be classified into the categories of quantum noise and
detector noise. The quantum noise is produced by fluctuations caused by scattering
and absorption while the radiation passes through the object. Its amplitude remains
constant as the signal level changes (Fig. 18a), and it is linearly increased with the
intensity, so that dark and light regions in the image are equally noisy. In contrast,
Fig. 18 Random noise in images takes two general forms: quantum noise (a) and detector noise (b)
(Smith 1999; www.DSPguide.com with permission)
detector noise (Fig. 18b) is a thermal noise produced by the electrons in the detector;
it does not depend on the X-ray exposure, and it can be measured without radiation.
This type of noise increases with the square root of the signal level, resulting in the
bright areas being more noisy than the dark ones (Smith 1999).
Metrological Performance Verification and Uncertainty

Evaluation
Two important steps are needed as prerequisites of traceability establishment: (i)

metrological performance verification and (ii) evaluation of task-specific measure-
ment uncertainty. Metrological performance verification of CT systems is discussed
in section “Metrological Performance Verification.” Task-specific uncertainty deter-
mination of CT measurements is discussed in section “Determination of Measure-
ment Uncertainty.” Furthermore, examples of reference objects used for system
qualification and for testing specific metrological characteristics are discussed in
section “Reference Objects.”
Metrological Performance Verification
In order to ensure comparability of results, both between different CT systems and

with other coordinate measuring systems (such as tactile and optical CMS), pro-
cedures for acceptance and reverification testing of CT systems should follow similar
principles as in the ISO 10360 series, which is internationally used for testing the
metrological performances of CMS (Carmignato et al. 2010). At present, there is no
published international standard available for testing CT systems. The working
group that is responsible for the development of international ISO 10360 standards
– the ISO Technical Committee 213 Working Group 10 (ISO/TC 213/WG 10) – is
working on filling this gap (Bartscher et al. 2018). In the meantime, commonly
adopted procedures for testing metrological CT systems are defined in the German
guideline VDI/VDE 2630-1.3 (2012) that is closely related to the ISO 10360-2
(2009), with the addition of specific aspects for testing CT systems. In particular,
VDI/VDE 2630-1.3 provides guidance to verify several metrological characteristics,
including (i) length measurement errors (E) and (ii) probing errors of size (PS) and of
form (PF). In fact, an important principle of acceptance and reverification testing
is the assessment of both global and local performance – i.e., errors – of CMSs
(Fig. 19) (Borges de Oliveira et al. 2015).
The local error behavior can be tested as a probing error test (P-test) assessing
the diameter and form deviation of a calibrated sphere. In order to achieve compa-
rability to the probing error of a tactile CMM, which is determined according to ISO
10360-5 (2019), the current ISO 10360 draft for CT-based CMS proposes to measure
25 representative points (e.g., based on 25 patches as in Fig. 19) distributed over one
hemisphere surface for the probing error test for form and size (PS and PF). In
addition, considering the large amount of measurement points determined by CT,
Length measurement error –E -test

Global behaviour assessment
Unidirectional Bidirectional
4
5
3 4
3
2
2 Probing Error
1
0
1
0 P -test size and form
–1
–1
–2 Local behaviour assessment
–2 –3
–4
–3
–5
–4
–4 –3 –2 –1 0 1 2 3 4
Fig. 19 (Top) Global behavior assessment – E-test performed as uni- and bidirectional length
measurements. (Bottom) Local behavior assessment – P-test performed as probing dispersion using
95% of the points and probing size error using all points, 25 representative points based on 25
patches, and all probed points (Borges de Oliveira et al. 2015)
another type of test takes into account 95% of the measured points for analyzing PS
and PF.
The probing error of form PF, defined as the span of the radial deviations of the
measurement points from the calculated regression sphere (determined using the
least-squares method with a free radius), can be obtained from the following
equation:
PF ¼ Rmax Rmin (8)
where Rmax and Rmin are, respectively, the maximum and the minimum distance from
the probing points to the center of the regression sphere.
The probing error of size PS is calculated as the difference between the measured
diameter of the regression sphere Da and the calibrated diameter Dr:
PS ¼ Da Dr (9)
The global behavior is tested as a length measurement error (E-test) using

reference length standards (e.g., hole plates, step gauges, ball plates, etc.), stated as:
E ¼ L a Lr (10)
where La is the measured length and Lr the calibrated length.

Measuring unidirectional lengths (e.g., sphere center-to-center distances) usually
leads to lower errors than measuring bidirectional lengths (Borges de Oliveira et al.
2015). For this reason, bidirectional lengths should be preferred for testing. If
unidirectional lengths are used, there are two principal possibilities to determine
equivalent bidirectional length measurement errors: (i) the probing errors (of size
and form) are added (with the sign that maximizes the absolute value of the resulting
bidirectional length measurement error E) to the unidirectional length measurement
error; and (ii) an additional bidirectional measurement of a short calibrated length is
performed, and the length error of this measurement is added (with sign) to the
unidirectional length measurement error (Bartscher et al. 2018).
The VDI/VDE 2630-1.3 (2012) defines also another important metrological
characteristic, namely, the metrological structural resolution (MSR) for dimensional
measurements. According to the guideline, the MSR is defined as the size of the
smallest structure that can still be measured within error limits to be specified (e.g.,
by the manufacturer). The MSR provides relevant additional information with
respect to E, PS and PF. For example, low-pass filtering of CT-measured data can
improve the probing error of form but – at the same time – worsen the MSR. The
statement of the MSR is of great importance, for instance, when measuring small
geometries and microstructures (Bartscher et al. 2012), internal porosity (Wits et al.
2016), and parts with high surface roughness (Carmignato et al. 2017). Moreover,
the metrological structural resolution encompasses the whole CT measurement
chain, ranging from acquisition of X-ray projections to final metrological software
elaborations. At present, a number of methods have been proposed for MSR
evaluation. However, a standard procedure to be included in the underdevelopment
CT part of the ISO 10360 series has still to be defined (Bartscher et al. 2014).
Currently proposed methods for evaluating the MSR are:
(i) VDI/VDE 2630-1.3 (2012) proposes to determine the MSR from the diameter
of the smallest sphere that can still be measured within error limits which
should be specified by the manufacturer.
(ii) Determination of the MSR by measuring the edge radius of specimens featuring
sharp edges (Bartscher et al. 2012).
(iii) Measurement of the radius of curvature on a calibrated round edge structure.
From the deviation between the measured and the calibrated radius, an ana-
logue Gaussian broadening of the measurement system is determined (Illemann
et al. 2014).
(iv) Frequency response analysis of a CT system when measuring the surface of an
Aperiodic Spatial Frequency Standard (Flessner et al. 2014) or the sinusoidal
surface of a multi-wave standard (Arenhart et al. 2015).
(v) “Two-spheres” concept using a simple object composed by two calibrated

spheres (Carmignato et al. 2012; Zanini and Carmignato 2017).
Determination of Measurement Uncertainty
A task-specific measurement uncertainty statement is necessary in order to (i)

determine the quality of a measurement process, (ii) take decisions on conformance
or nonconformance of products with specifications, and (iii) achieve traceability to
SI units (Wilhelm et al. 2001). However, since CT is affected by a large amount of
influencing factors and by their interactions (section “Error Sources”), uncertainty
evaluation is often a difficult task, and no internationally accepted standards and
guidelines exist up to now (Carmignato et al. 2018). The general methodology for
uncertainty determination is to calculate the measurement uncertainty through an
uncertainty budget according to the ISO Guide to the Expression of Uncertainty in
Measurement (GUM) (BIPM 2008). Since CT measurements are influenced by
numerous factors that are difficult to be modelled, the GUM method alone is
commonly not used to determine the uncertainty of CT measurements. A feasible
approach for task-specific uncertainty determination in coordinate metrology is the
substitution method, which for tactile CMMs is defined by the standard ISO 15530-3
(2011). This approach has been used also for CT several times in literature (Schmitt
and Niggemann 2010; Müller et al. 2014) and has been recently formally adapted to
CT in the German guideline VDI/VDE 2630-2.1 (2015). This guideline can be
successfully applied when a calibrated object similar to the actual components (in
terms of material, size, and geometry) is available or producible with sufficiently low
calibration uncertainty. The expanded uncertainty is given by (Eq. 11):
qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
U ¼k u2cal þ u2drif t þ u2p þ u2w þ u2b (11)
where k is the coverage factor, ucal is the standard uncertainty of the calibrated
values, udrift is the uncertainty contribution of possible changes in the workpiece
shape since the calibration referred to, up is measured as the standard deviation of the
repeated measurements, uw is the standard uncertainty related to variations in
materials and production, and ub is the uncertainty in the calculation of the system-
atic error b (calculated as difference between mean measured value and calibrated
reference value). The substitution method described above is limited when similar
calibrated objects with sufficiently low calibration uncertainty are not available. This
is the case, for instance, of deformable samples (e.g., polymeric workpieces), for
which a CMM calibration would produce a too large uncertainty, e.g., because of
clamping and probing forces that could generate unwanted deformations of the
components. For example, an alternative approach was used by Zanini et al.
(2018a) for determining the uncertainty of CT wear measurements performed on
polyethylene acetabular cups. Another different approach for uncertainty determi-
nation for CT dimensional measurements was proposed by Dewulf et al. (2013):
Fig. 20 Selection of reference objects featuring high-precision spheres. (a) Ball plate (Courtesy of
DTU, Denmark); (b) CT-tree (Courtesy of DTU, Denmark); (c) CT-tetrahedron (Courtesy of
University of Padova, Italy); (d) sphere tetrahedron (Courtesy of PTB, Germany); (e) 27 ruby
spheres (Courtesy of Carl Zeiss, Germany)
since a CT model is a voxel model, the result of a CT dimensional measurement can

basically be interpreted as the product of the number of voxels and the voxel size,
and the relative uncertainty can be considered as caused by the uncertainty of the
voxel size and the uncertainty of the edge detection.
The difficulty of accurately determining the task-specific uncertainty for CT
measurements is proven by results achieved by several international intercompari-
sons. The first intercomparison for dimensional metrology CT systems was called
CT Audit (2009–2012) and was organized by the University of Padova, Italy
(Carmignato 2012). Four different calibrated samples representing a variety of
dimensions, geometries, and materials were measured by several participants in
America, Asia, and Europe. Several uncertainty evaluation methods were applied
by different participants. Results showed that less than half of the participants’
measurement results were accompanied by a valid measurement uncertainty state-
ment. A second intercomparison is the CIA-CT intercomparison on industrial CT for
measurement applications (2012–2013), which was organized by DTU, Denmark
(Angel and De Chiffre 2014). The main goals of this comparison were to test the
applicability of CT for dimensional measurements of small objects, commonly
measured in industry, which are more representative than artificial reference stan-
dards. Finally, the InteraqCT comparison on assemblies (2015–2016) was organized
by DTU, Denmark, to test the applicability of CT for measurements of assemblies
with dimensions and materials commonly adopted in industry (Stolfi and De Chiffre
2018). In this case, the majority of participants stated measurement uncertainties
based merely on MPEs, whereas just a few participants used more advanced
uncertainty models. Results of these intercomparisons demonstrate that uncertainty
determination is still a major challenge for CT dimensional measurements.
Reference Objects
Several reference objects specifically developed for CT metrology are proposed in

literature. This section reports some examples; it does not aim to be complete.
Reference objects featuring high-precision spheres are quite common, due to several
advantages (Bartscher et al. 2018). A selection of this type of objects is shown in
Fig. 20.
In (Leonard et al. 2014) a tetrahedron made by four touching spheres was also
used to find the optimal threshold value by sphere distance calculation. However,
these items usually do not allow the setting of the correct threshold value, because
the material of the spheres (e.g., ruby, zirconia, alumina, steel) often differ from
common workpiece materials (e.g., plastic, aluminum, titanium, steel). Other objects
developed to evaluate specific CT measuring errors using different geometries and
materials are shown in Fig. 21. For example, bidirectional length measurements
errors can be determined from calibrated face-to-face distances measured on samples
Fig. 21 (a) Calotte plate and (b) calotte cube (Courtesy of PTB, Germany); (c) 3D model of the
“cactus” step gauge (Courtesy of KU Leuven, Belgium); (d) “Pan flute” gauge consisting in five
tubes made of borosilicate glass with different calibrated lengths (Courtesy of University of Padova,
Italy); (e) aluminum dismountable free-form reference standard (Courtesy of PTB, Germany)
with plane and parallel surfaces such as step gauges and the so-called “cactus” step
gauge (Fig. 21c) (Kiekens et al. 2011) or from calibrated tubes length of the “pan
flute” gauge (Fig. 21b) (Carmignato 2012). The majority of reference standards lack
in similarity to actual industrial workpieces; to overcome this limitation, Bartscher et
al. (2011) presented a dismountable free-form reference standard made from a cast
miniaturized aluminum cylinder head. It can be dismounted in four segments, with
fixed reference geometries (spheres and cylinders) (Fig. 21c).
Other reference objects were proposed in order to study the threshold effect, like
the “fiber gauge” (Fig. 22). The optimal threshold value can be evaluated by the
simultaneous measurement of internal and external features (Carmignato et al.
2009).
Several reference objects have been developed to analyze specific influence
quantities. A step-cylinder with a central bore is well suited to adjust the measuring
parameters versus material-specific absorption and penetration thickness. Moreover,
it allows detecting the maximum possible material thickness. A step-cylinder with-
out inner hole as well as a step wedge (Fig. 23) can be used for beam hardening
Fig. 22 Reference standard featuring fibers and holes: the “fiber gauge” (Marinello et al. 2008).
The diagram below shows the influence of the threshold value on the measured diameters of inner
and outer cylindrical features, i.e., fibers and holes, respectively (Carmignato et al. 2009)
Fig. 23 (a) Example of step cylinder with maximum diameter 250 mm (Courtesy of PTB,
Germany); (b) principle of beam hardening correction method using a step wedge reference object:
(i) step wedge, (ii) projection image, and (iii) correction curve (Müller et al. 2013)
correction (Müller 2013). Different reference items were also proposed to evaluate
the metrological structural resolution of CT systems (see section “Metrological
Performance Verification”).
Multiple dedicated objects were designed in order to perform geometrical verifi-
cation or calibration of CT systems. For example, Cho et al. (2005) proposed a
reference object with spheres mounted on two static circular trajectories allowing the
geometry of the system to be resolved in just one pose with no need to rotate the
object (Fig. 24a), and, more recently, Hermanek and Carmignato (2016) designed an
object that contains two circular trajectories of markers that can be used for initial
determination of the system geometry in static mode (i.e., without rotating the
object) and additional spheres for dynamic investigation (i.e., as a function of object
rotation angle) arranged in four rotationally shifted helical trajectories, thereby
optimizing the spatial distribution of projected markers (Fig. 24b).
Finally, a number of reference objects were produced for testing the accuracy of
specific analyses that can be done using CT, for instance, multi-material gap
measurements, porosity analysis, and surface topography measurement of additively
manufactured parts. Hermanek et al. developed a dismountable cylindrical reference
standard with internal artificial calibrated hemispherical defects for achieving trace-
ability of CT measurements of internal porosity (Fig. 25a) (Hermanek and
Fig. 24 (a) Reference object with circular static trajectories used for geometrical calibration of a
CT system (Ferrucci et al. 2015) and (b) reference object with optimized spatial distribution of
projected markers used for static and dynamic investigations of the CT geometrical alignment
(Hermanek and Carmignato 2016)
Fig. 25 (a) Reference object for accuracy evaluation of CT porosity measurements (Hermanek and
Carmignato 2017); (b) reference sample produced by SLM for verifying the accuracy of CT surface
topography measurements (Zanini et al. 2018b); (c) multi-material reference gap standard
(Hermanek et al. 2017b)
Carmignato 2017) and a series of mono- and multi-material reference standards for
investigating the multi-material effects on measurements of gaps by CT (Fig. 25c)
(Hermanek et al. 2017b). Zanini et al. (2018b) developed a reference sample made
by selective laser melting (SLM) of Ti6Al4V whose design was optimized to
compare topographical measurements conducted by CT with topographies acquired

by means of optical devices (Fig. 25b). Such reference sample allows obtaining 2D
reference cross-sectional profiles (including re-entrant features) representing the
actual morphology of the SLM surface, to be compared with profiles measured
with different noncontact instruments in order to establish their traceability.
Applications of CT for Dimensional Metrology
Nowadays, manufacturing industries are experimenting improvements in production

technologies that allow obtaining more complex products, including free-form
surfaces, inner structures, multi-material components, and a huge amount of differ-
ent features (Savio et al. 2007; Schmidt et al. 2017). The need for inspection and
metrology of such parts requires new accurate and time-efficient measurement
technologies that contribute to improve quality and reduce costs in manufacturing.
In this context, industrial CT offers a large variety of applications in the entire
development and production chain, from product development to quality control
(De Chiffre et al. 2014). Figure 26a presents an overview of different fields of
application for CT in industry. Voxel-based evaluations can be classified in visual-
ization and nondestructive testing (NDT), while surface-based evaluations comprise
digitization and dimensional metrology. The need for traceability increases with the
complexity of the task. Figure 26b summarizes the main advantages and limitations
of CT for industrial metrology applications. In the following, this section focuses on
the following examples of metrological tasks: (i) dimensional measurements and
tolerances verification, (ii) volume measurements, (iii) assembly analysis, (iv) fiber
analysis, and (v) surface metrology of AM parts.
Dimensional Measurements and Tolerances Verification
Differently from conventional coordinate measurements using tactile probes – which

are performed directly on the workpiece surface – CT measurements are performed
on a holistic virtual model obtained from the 3D reconstruction of the workpiece.
This means that a large number of features (including inner elements) can be
analyzed at the same time, quickly and using a high density of measured points
(Buratti et al. 2018). CT dimensional measurements and tolerance verification are
commonly conducted by association of ideal geometrical elements to the measured
points of the workpiece surface extracted from the virtual reconstructed model.
Nominal-actual comparisons are also often performed to investigate the deviations
between the actual geometry (as scanned by CT) and the reference geometry (e.g.,
CAD model or 3D reconstruction of a reference workpiece). A color-coded repre-
sentation of the model can be obtained to visualize the surface deviations of the
geometries, as shown, for example, in Fig. 27. Color-coded maps are useful also for
comparing results of CT measurements and process simulations: Fig. 28 shows an
example of a test part produced by micro-milling process and scanned via CT,
Fig. 26 (a) Fields of application for industrial CT; (b) advantages and disadvantages of CT for
industrial metrology applications. (Adapted from De Chiffre et al. 2014)
compared with the simulation of the milling process, to obtain a comprehensive

analysis of machining errors. Recently, CT has been applied effectively for dimen-
sional measurements of products obtained from additive manufacturing (AM),
which can include complex and internal geometries that are not possible to be
produced using conventional manufacturing processes. Such products can be mea-
sured in a holistic way (i.e., including non-accessible features and surfaces) only
using X-ray computed tomography (De Chiffre et al. 2014; Van Bael et al. 2011).
Since associating ideal geometrical elements can be a difficult task due to the highly
complex surface texture with high roughness (see Fig. 29), AM parts are commonly
Fig. 27 Color-coded map of deviations showing (a) wear of a micro-milling tool and (b) wear of a
knee prosthetic insert. Both cases were obtained by nominal/actual comparison between the CT
scans performed on the unworn and worn parts (Courtesy of University of Padova). (c) Color-coded
map of deviations between a CT-measured cellular specimen and its nominal CAD model (Sbettega
et al. 2018)
evaluated through nominal-actual comparisons (see Fig. 27c). However, AM

cellular structures can present high structural distortions. Consequently, the devi-
ations determined from the CAD comparison can include both geometrical/dimen-
sional errors and errors connected to the structural distortions. For this reason,
when the interest is on evaluating only the errors associated with the local dimen-
sions, a thickness analysis can be more effective (Sbettega et al. 2018), as it does
not require an alignment between measured geometry and nominal model. Another
recent application is the measurement of screw threads, where CT can be used to
overcome the limitations of conventional tactile and optical techniques
(Carmignato and De Chiffre 2003), producing complete measurements of both
Fig. 28 Deviations of the CT-measured workpiece from the CAD model (left) and comparison of
the simulated surface with the CAD model (right) (Möhring et al. 2015)
Fig. 29 Ideal cylindrical element fitted to the inner feature of a NiTi sample produced by AM and
characterized by complex texture with high surface roughness (Khademzadeh et al. 2018)
internal and external threads, even for parts characterized by high form errors and
roughness such as AM threaded parts (Zanini and Carmignato 2018).
Volume Measurements
Evaluation of volume is relevant for many different applications, in which the

functionality of a product is directly related to volume. Figure 30a shows a case
study in which a drug-eluting stent was CT scanned to measure the volume of
microcavities, which contain and release drug toward vessel walls. In this case,
results showed that micro-CT can be used to overcome the limitations of optical and
tactile measurements (Carmignato and Savio 2011).
Volume loss evaluation is another important example of volume measurement,
often connected to wear mechanisms occurring, for instance, in tools, molds, and
prostheses. X-ray computed tomography can be used effectively for volumetric wear
measurement as well as for determination of local distribution of wear and defor-
mations, with significant advantages especially for polymeric components, such as
Fig. 30 (a) CT scan of drug-eluting stent, with detailed view of a microcavity (Carmignato and
Savio 2011); (b) CT porosity analysis on a tensile specimen produced by selective laser melting of
Ti6Al4V (Wits et al. 2016)
polyethylene prosthetic joints (Affatato et al. 2017). In fact, being CT a nondestruc-

tive and noncontact measuring technique, it does not induce any damage to poly-
meric (or in general deformable) components, opposite to tactile CMMs, which
instead can cause unwanted deformations due to the applied clamping and probing
forces (Affatato et al. 2010).
Internal defects such as voids and porosity are an inherent characteristic of parts
produced by many manufacturing processes, such as casting, injection molding,
additive manufacturing, composite materials fabrication, etc. (Hermanek et al. 2019;
Khademzadeh et al. 2016). Well-known techniques for porosity determination are,
for example, ultrasonic testing, Archimedes method, and materialography. X-ray
computed tomography can measure porosity in 3D (see an example in Fig. 30b),
overcoming the limitations of the abovementioned techniques. Defect size, volume,
morphology, total void content, and spatial distribution can be evaluated from CT
data throughout the entire inspected volume (Wits et al. 2016).
Assembly Analysis
Functional requirements must be taken into account when defining dimensional and
geometrical tolerances of components to be assembled together in order to accom-
plish a specific function. Such tolerances are commonly verified by measuring the
individual components in pre-assembly state. However, even if each individual
component meets the specified tolerances, assembly variations can occur due to
clamping and/or assembly forces, and the proper functioning of the assembled
system might not be guaranteed (Stolfi et al. 2017). CT is a unique tool for the
inspection of assemblies, allowing the measurement of products and individual
components in the assembled state (De Chiffre et al. 2014). When two assembled
components are made of materials with sufficiently different X-ray attenuation
coefficients, they can be distinguished and segmented in the CT reconstructed
three-dimensional model. An example of assembly analysis performed by CT is
Fig. 31 (a) CT assembly analysis of an insulin pen (Courtesy of Volume Graphics GmbH); (b)
assembly consisting of two threaded parts made by the same material: the cross-section shows
limitations in the analysis of contact surfaces. (Courtesy of Nikon-Metrology/X-Tek)
provided in Fig. 31a. The main critical factors of analyses involving multi-material parts
are (i) the setup of CT scanning parameters and (ii) the surface determination procedure
(De Chiffre et al. 2014). One viable solution for the first issue is using dual-energy CT
(Heinzl et al. 2008). On the other hand, applying local adaptive thresholding methods
can improve the surface determination procedure (Kruth et al. 2011). Differently from
multi-material assemblies, parts with similar absorption coefficients represent a common
problem while performing the segmentation of each individual component. Fig. 31b
shows an example of failed inspection of the assembly of two threaded components
Fig. 32 CT scan of components of a titanium dental implant shown before (top figure) and after the
assembly (bottom). Contact surfaces between individual components of the dental implant assem-
bly are colored in blue, green, and red (Zanini et al. 2017)
made of the same material. The contact surfaces between the assembled components are
not visible in this case. Zanini et al. (2017) proposed a new CT-based methodology
developed to enable the accurate measurement of contact surfaces of mono-material
assemblies. Such methodology is based on the alignment between CT-scanned volumes
of each individual component obtained before the assembly with the CT volume of the
assembled part. An example of implementation on a dental implant with all its compo-
nents made of titanium can be seen in Fig. 32.
Fiber Analysis
The performances of fiber-reinforced materials are strongly influenced by the charac-

teristics of their fibers, including length, curvature, and orientation (Meneghetti et al.
2014). CT allows measuring in a nondestructive way the fibers included in composite
materials. Most of the published work and available software tools deal with straight
fibers and orientation analyses (e.g., measurement of fiber tensor component parallel
to the polymer flow). The analysis and visualization software VGStudio MAX 3.1
(Volume Graphics GmbH, Germany), for example, enables the evaluation of local and
global fiber orientation, local and global fiber concentration, deviation from a refer-
ence orientation, and fiber distribution (see Fig. 33). These evaluations are based on
the analysis of the local variation of the grayscale image field using methods such as
Hessian analysis or Hough transformation (Carmignato et al. 2018). To calculate the
fiber length distribution, a segmentation of the individual fibers is necessary. This is
currently possible with other tools, such as FibreScout, which is a module of the open-
source openIA software developed by the University of Wels, Austria (Weissenböck et
al. 2014). The accuracy of CT measurements of fiber length is currently being
investigated (Zanini et al. 2019a).
Surface Metrology of AM Parts
Topography measurements of AM surfaces are challenging due to several reasons,

including (i) presence of re-entrant features (e.g., undercuts) and micro-topographic
features, (ii) possible presence of non-accessible surfaces (e.g., internal surfaces),
Fig. 33 Fiber orientation analysis of composite material; different colors correspond to different
fiber orientations. (Courtesy of Volume Graphics GmbH)
Fig. 34 Surface topography of an AM component measured by (a) CT, (b) focus variation, and (c)
confocal microscopy. (Courtesy of University of Padova)
and (iii) different surface topographies from process to process and also from surface
to surface within the same part, depending, for instance, on the building direction. In
this context, industrial X-ray computed tomography has recently started to be
considered as a viable technique for topography measurement of AM surfaces. In
fact, several works in literature have demonstrated that CT is capable of obtaining
topography measurements at microscale, including non-accessible surfaces and
reentrant features (Townsend et al. 2017; Thompson et al. 2017; Zanini et al.
2018b, 2019b). Figure 34 shows a CT-measured topography of a Ti6Al4V selective
laser melted part, compared to the corresponding topographies acquired using two
different optical techniques: confocal and focus variation microscopy.
References
Affatato S, Spinelli M, Zavalloni M, Traina F, Carmignato S, Toni A (2010) Ceramic-on-metal for
total hip replacement: mixing and matching can lead to high wear. Artif Organs 34(4):319–323.
https://doi.org/10.1111/j.1525-1594.2009.00854
Affatato S, Zanini F, Carmignato S (2017) Quantification of wear and deformation in different
configurations of polyethylene acetabular cups using micro X-ray computed tomography.
Materials 10(3):259. https://doi.org/10.3390/ma10030259
Angel J, De Chiffre L (2014) Comparison on computed tomography using industrial items. CIRP
Ann 63:473–476
Arenhart FA, Nardelli VC, Donatelli GD (2015) Characterization of the metrological structural
resolution of CT systems using a multi-wave standard. In: Proceedings of XXI IMEKO World
Congress, Prague.
ASTM E 1441-11 – Standard Guide for Computed Tomography (CT) Imaging, ASTM Interna-
tional, United States
Bartscher M, Ehrig K, Staude A, Goebbels J, Neuschaefer-Rube U (2011) Application of an
industrial CT reference standard for cast freeform shaped workpieces. Digital Industrial Radi-
ology and Computed Tomography (DIR). 20-22 June, 2011. Berlin, Germany.
Bartscher M, Bremer H, Birth T, Staude A, Ehrig K (2012) The resolution of dimensional CT – an
edge-based analysis. In: Proceedings of the 2012 conference on industrial computed tomogra-
phy (ICT 2012). 19-21 September 2012, Wels, Austria.
Bartscher M, Sato O, Hartig F, Neuschaefer-Rube U (2014) Current state of standardization in the
field of dimensional computed tomography. Meas Sci Technol 25:064013. https://doi.org/
10.1088/0957-0233/25/6/06
Bartscher M, Neuschaefer-Rube U, Illemann J, Borges de Oliveira F, Stolfi A, Carmignato S (2018)
Qualification and testing of CT systems. In: Carmignato S, Dewulf W, Leach R (eds) Industrial X-ray
computed tomography. Springer, Cham, pp 185–228. https://doi.org/10.1007/978-3-319-59573-3_6
Bioclinica (2011) The evolution of CT scan clinical trials. http://www.bioclinica.com/blog/evolu
tion-ct-scan-clinical-trials. Accessed 17 Mar 2019
BIPM JCGM 100 (2008) Evaluation of measurement data – guide to the expression of uncertainty
in measurement. International Organisation for Standardisation, Geneva
Borges de Oliveira F, Bartscher M, Neuschaefer-Rube U (2015) Analysis of combined probing
measurement error and length measurement error test for acceptance testing in dimensional
computed tomography. In: Proceedings of DIR 2015 in NDT.net, online at: www.ndt.net/events/
DIR2015/app/content/Paper/31_BorgesdeOliveira.pdf
Borges de Oliveira F, Stolfi A, Bartscher M, De Chiffre L (2016) Experimental investigation of
surface determination process on multi-material components for dimensional computed tomog-
raphy. Case Stud Nondestruct Test Eval 6(Part B):93–103. https://doi.org/10.1016/j.
csndt.2016.04.003
Buratti A, Bredemann J, Pavan M, Schmitt R, Carmignato S (2018) Applications of CT for
dimensional metrology. In: Carmignato S, Dewulf W, Leach R (eds) Industrial X-ray computed
tomography. Springer, Cham, pp 333–369. https://doi.org/10.1007/978-3-319-59573-3_9
Buzug TM (2008) Computed tomography: from photon statistics to modern cone-beam CT.
Springer, Berlin
Carmignato S (2012) Accuracy of industrial computed tomography measurements: experimental
results from an international comparison. CIRP Ann 61(1):491–494. https://doi.org/10.1016/j.
cirp.2012.03.021
Carmignato S, De Chiffre L (2003) A new method for thread calibration on coordinate measuring
machines. CIRP Ann 52(1):447–450. https://doi.org/10.1016/S0007-8506(07)60622-2
Carmignato S, Savio E (2011) Traceable volume measurements using coordinate measuring
systems. CIRP Ann 60(1):519–522. https://doi.org/10.1016/j.cirp.2011.03.061
Carmignato S, Dreossi D, Mancini L, Marinello F, Tromba G, Savio E (2009) Testing of x-ray
microtomography system using a traceable geometrical standard. Meas Sci Technol 20:084021.
https://doi.org/10.1088/0957-0233/20/8/084021. IOP Publishing
Carmignato S, Voltan A, Savio E (2010) Metrological performance of optical coordinate measuring

machines under industrial conditions. CIRP Ann 59(1):497–500. https://doi.org/10.1016/j.
cirp.2010.03.128
Carmignato S, Pierobon A, Rampazzo P, Parisatto M, Savio E (2012) CT for industrial metrology.
Accuracy and structural resolution of CT dimensional measurements. In: Proceedings of iCT
2012 in NDT.net, online at: http://www.ndt.net/article/ctc2012/papers/173.pdf
Carmignato S, Aloisi V, Medeossi F, Zanini F, Savio E (2017) Influence of surface roughness on
computed tomography dimensional measurements. CIRP Ann 66(1):499–502. https://doi.org/
10.1016/j.cirp.2017.04.067
Carmignato S, Dewulf W, Leach R (2018) Industrial X-ray computed tomography. Springer, Cham.
https://doi.org/10.1007/978-3-319-59573-3
Cho Y, Moseley DJ, Siewerdsen JH, Jaffray DA (2005) Accurate technique for complete geometric
calibration of cone-beam computed tomography systems. Med Phys 32:968–983
Cormack AM (1963) Representation of a function by its line integrals, with some radiological
applications. J Appl Phys 34(9):2722–2727
De Chiffre L, Carmignato S, Kruth J-P, Schmitt R, Weckenmann A (2014) Industrial applications of
computed tomography. CIRP Ann 63(2):655–677. https://doi.org/10.1016/j.cirp.2014.05.011
Dewulf W, Tan Y, Kiekens K (2012) Sense and non sense of beam hardening correction in CT
metrology. CIRP Ann 61(1):495–498
Dewulf W, Kiekens K, Tan Y, Welkenhuyzen F, Kruth J-P (2013) Uncertainty determination and
quantification for dimensional measurements with industrial computed tomography. CIRP Ann
62(1):535–538
Dewulf W, Ferrucci M, Ametova E, Heřmánek P, Probst G, Boeckmans B, Craeghs T, Carmignato S
(2018) Enhanced dimensional measurement by fast determination and compensation of geo-
metrical misalignments of X-ray computed tomography instruments. CIRP Ann 67(1):523-526.
https://doi.org/10.1016/j.cirp.2018.04.124
Feldkamp LA, Davis LC, Kress JW (1984) Practical cone-beam algorithm. J Opt Soc Am A
1(6):612–619
Ferrucci M, Leach RK, Giusca C, Carmignato S, Dewulf W (2015) Towards geometrical calibration
of X-ray computed tomography systems – a review. Meas Sci Technol 26:092003. https://doi.
org/10.1088/0957-0233/26/9/092003
Ferrucci M, Ametova E, Carmignato S, Dewulf W (2016) Evaluating the effects of detector angular
misalignments on simulated computed tomography data. Precis Eng 45:230–241. https://doi.
org/10.1016/j.precisioneng.2016.03.001
Fessler JA (2000) Statistical image reconstruction methods for transmission tomography.
In: Handbook of medical imaging, vol 2. Medical image processing and analysis. SPIE Press,
Bellingham
Flessner M, Vujaklija N, Helmecke E, Hausotte T (2014) Determination of metrological structural
resolution of a CT system using the frequency response on surface structures. In: Proceedings of
MacroScale, Vienna
Frank G (1940) Verfahren zur herstellung von körperschnittbildern mittels röntgenstrahlen. German
Patent 693,374
Heinzl C, Kastner J, Möller T, Gröller E (2008) Statistical analysis of multi-material components
using dual energy CT. Proceedings of workshop VMV 2008 (vol. 8. pp 179–188), Konstanz,
Germany.
Hermanek P, Carmignato S (2016) Reference object for evaluating the accuracy of porosity
measurements by X-ray computed tomography. Case Stud Nondestruct Test Eval 6:122–127.
https://doi.org/10.1016/j.csndt.2016.05.003.
Hermanek P, Carmignato S (2017) Porosity measurements by X-ray computed tomography:
accuracy evaluation using a calibrated object. Precis Eng 49:377–387. https://doi.org/10.1016/
j.precisioneng.2017.03.007
Hermanek P, Zanini F, Carmignato S (2019) Traceable porosity measurements in industrial com-
ponents using X-ray computed tomography. ASME J Manuf Sci Eng. https://doi.org/10.1115/
1.4043192
Hermanek P, Rathore JS, Aloisi V, Carmignato S (2018) Principles of X-ray computed tomography.
In: Carmignato S, Dewulf W, Leach R (eds) Industrial X-ray computed tomography. Springer,
Cham, pp 25–67. https://doi.org/10.1007/978-3-319-59573-3_2
Hermanek P, Borges de Oliveira F, Carmignato S, Bartscher M (2017b) Experimental investigation
of new multi-material gap reference standard for testing computed tomography systems.
In: Proceedings of iCT2017-7th conference on industrial computed tomography
Hiller J, Maisl M, Reindl LM (2012) Physical characterization and performance evaluation of an X-
ray micro-computed tomography system for dimensional metrology applications. Meas Sci
Technol 23(8):85404
Hounsfield G (1976) Historical notes on computerized axial tomography. J Can Assoc Radiol
27(3):135–142
Hsieh J (2015) Computed tomography: principles, design, artifacts, and recent advances. SPIE
Press, Bellingham
Hunter A, McDavid W (2012) Characterization and correction of cupping effect artefacts in cone
beam CT. Dentomaxillofac Radiol 41(3):217–223
Illemann J, Bartscher M, Jusko O, Hartig F, Neuschaefer-Rube U, Wendt K (2014) Procedure and
reference standard to determine the structural resolution in coordinate metrology. Meas Sci
Technol 25:064015. https://doi.org/10.1088/0957-0233/25/6/064
Impactscan (2013) CT History. http://www.impactscan.org/CThistory.htm. Accessed 20 Mar 2019
ISO 10360 Part 2 (2009) Geometrical product specifications (GPS) – acceptance and reverification
tests for coordinate measuring machines (CMM) – part 2: CMMs used for measuring linear
dimensions. International Organisation for Standardisation, Geneva
ISO 10360 Part 5 (2019) Geometrical product specifications (GPS) – acceptance and reverification
tests for coordinate measuring machines (CMM) – part 5: CMMs using single and multiple
stylus contacting probing systems. International Organisation for Standardisation, Geneva
ISO 15530 Part 3 (2011) Geometrical product specifications (GPS) – coordinate measuring
machines (CMM): technique for determining the uncertainty of measurement – part 3: use
of calibrated workpieces or measurement standards. International Organisation for
Standardisation, Geneva
Khademzadeh S, Carmignato S, Parvin N, Zanini F, Bariani PF (2016) Micro porosity analysis in
additive manufactured NiTi parts using micro computed tomography and electron microscopy.
Mater Des 90:745–752. https://doi.org/10.1016/j.matdes.2015.10.161
Khademzadeh S, Zanini F, Bariani PF, Carmignato S (2018) Precision additive manufacturing of
NiTi parts using micro direct metal deposition. Int J Adv Manuf Technol. https://doi.org/
10.1007/s00170-018-1822-3
Kiekens K, Welkenhuyzen F, Tan Y, Bleys P, Voet A, Kruth J-P, Dewulf W (2011) A test object with
parallel grooves for calibration and accuracy assessment of industrial computed tomography
(CT) metrology. Meas Sci Technol 22:115502
Kruth J-P, Bartscher M, Carmignato S, Schmitt R, De Chiffre L, Weckenmann A (2011) Computed
tomography for dimensional metrology. CIRP Ann 60(2):821–842. https://doi.org/10.1016/j.
cirp.2011.05.006
Ledley R, Wilson J, Huang H (1974) ACTA (automatic computerized transverse axial) – the whole
body tomographic X-ray scanner. Proc SPIE 0057:94–107
Leonard F, Brown SB, Withers PJ, Mummery PM, McCarthy MB (2014) A new method of
performance verification for x-ray computed tomography measurements. Meas Sci Technol
25:065401. https://doi.org/10.1088/0957-0233/25/6/065401. 10pp
Lifton JJ, Carmignato S (2017) Simulating the influence of scatter and beam hardening in dimensional
computed tomography. Meas Sci Technol 28(10):104001. https://doi.org/10.1088/1361-6501/aa80b2
Marinello F, Savio E, Carmignato S, De Chiffre L (2008) Calibration artefact for the microscale
with high aspect ratio: the fiber gauge. CIRP Ann 57(1):497–500. https://doi.org/10.1016/j.
cirp.2008.03.086
Meneghetti G, Ricotta M, Lucchetta G, Carmignato S (2014) An hysteresis energy-based synthesis
of fully reversed axial fatigue behaviour of different polypropylene composites. Compos Part B
65:17–25. https://doi.org/10.1016/j.compositesb.2014.01.027
Möhring HC et al (2015) A testpart for interdisciplinary analyses in micro production engineering.

Procedia CIRP 28:106–112. https://doi.org/10.1016/j.procir.2015.04.018
Müller P (2013) Coordinate metrology by traceable computed tomography. PhD thesis, Department
of Mechanical Engineering, Technical University of Denmark, Lyngby
Müller P, Hiller J, Dai Y, Andreasen JL, Hansen HN, De Chiffre L (2014) Estimation of measure-
ment uncertainties in X-ray computed tomography metrology using the substitution method.
CIRP J Manuf Sci Technol 7:222–232
Otsu N (1979) A threshold selection method from grey level histograms. IEEE Trans Syst Man
Cybern 9(1):62–66
Panetta D (2016) Advances in X-ray detectors for clinical and preclinical computed tomography.
Nucl Instrum Methods Phys Res Sect A 809:2–12
Paxton R, Ambrose J (1974) The EMI scanner. A brief review of the first 650 patients. Br J Radiol
47(561):530–565
Radon J (1986) On the determination of functions from their integral values along certain mani-
folds. IEEE Trans Med Imaging 5(4):170–176
Reimers P, Goebbels J (1983) New possibilities of non-destructive evaluation by X-ray computed
tomography. Mater Eval 41:732–737
Requena G, Cloetens P, Altendorfer W, Poletti C, Tolnai D, Warchomicka F, Degischer HP (2009)
Sub-micrometer synchrotron tomography of multiphase metals using Kirkpatrick – Baez optics.
Scripta Mater 61(7):760–763
Savio E, De Chiffre L, Schmitt R (2007) Metrology of freeform shaped parts. CIRP Ann
56(2):810–835
Sbettega E, Zanini F, Savio E, Benedetti M, Carmignato S (2018) X-ray computed tomography
dimensional measurements of powder bed fusion cellular structures. In: Proceedings of the 18th
international conference of the European Society for precision engineering and nanotechnology,
EUSPEN 2018, Venice
Schmidt M et al (2017) Laser based additive manufacturing in industry and academia. CIRP Ann
66(2):561–584
Schmitt R, Niggemann C (2010) Uncertainty in measurement for x-ray-computed tomography
using calibrated work pieces. Meas Sci Technol 21(5):054008
Schmitt RH, Buratti A, Grozmani N, Voigtmann C, Peterek M (2018) Model-based optimisation of
CT imaging parameters for dimensional measurements on multimaterial workpieces. CIRP Ann
67(1), 527-530.
Schörner K (2012) Development of methods for scatter artifact correction in industrial X-ray cone-
beam computed tomography. Technische Universität München. http://mediatum.ub.tum.de/doc/
1097730/document.pdf. Accessed 24 Jan 2018
Smith SW (1999) The scientist and engineer’s guide to digital signal processing, 2nd edn. California
Technical Publishing. California, USA.
Stolfi A, De Chiffre L (2018) Interlaboratory comparison of a physical and a virtual assembly
measured by CT. Precis Eng 51:263–270
Stolfi A, De Chiffre L, Carli L (2017) Integrated quality control of precision assemblies using
computed tomography. Kgs. Danmarks Tekniske Universitet (DTU), Lyngby
Taguchi K, Iwanczyk JS (2013) Vision 20/20: single photon counting X-ray detectors in medical
imaging. Med Phys 40(10):100901
Thompson A, Senin N, Giusca C, Leach R (2017) Topography of selectively laser melted surfaces:
a comparison of different measurement methods. CIRP Ann 66(1):543–546. https://doi.org/
10.1016/j.cirp.2017.04.075
Townsend A, Pagani L, Scott PJ, Blunt L (2017) Areal surface texture data extraction from X-ray
computed tomography reconstructions of metal additively manufactured parts. Precis Eng
48:254–264
Van Bael S, Kerckhofs G, Moesen M, Pyka G, Schrooten J, Kruth JP (2011) Micro-CT-based
improvement of geometrical and mechanical controllability of selective laser melted Ti6Al4V
porous structures. Mat Sci Eng A 528:7423–7431
VDI/VDE 2630 Part 1.2 (2008) Computed tomography in dimensional measurement – influencing
variables on measurement results and recommendations for computed tomography dimensional
measurements. Verein Deutscher Ingenieure e.V., Dusseldorf
VDI/VDE 2630 Part 1.3 (2012) Computed tomography in dimensional measurement – guideline for
the application of DIN EN ISO 10360 for coordinate measuring machines with CT-sensors.
Verein Deutscher Ingenieure e.V., Dusseldorf
VDI/VDE 2630 Part 2.1 (2015) Computed tomography in dimensional measurement – determina-
tion of the uncertainty of measurement and the test process suitability of coordinate measure-
ment systems with CT sensors. Verein Deutscher Ingenieure e.V., Dusseldorf
Villarraga-Gómez H, Clark D, Smith S (2016) Effect of the number of radiographs taken in CT for
dimensional metrology. In: Proceedings of the 16th international conference of the European
society for precision engineering and nanotechnology, euspen 2016. Nottingham, UK.
Weckenmann A, Krämer P (2013) Predetermination of measurement uncertainty in the application
of computed tomography. Prod Lifecycle Manag: Geom Var, Chapter 17: 317–330
Weissenböck J, Amirkhanov A, Li W, Reh A, Amirkhanov A, Gröller E, Kastner J, Heinzl C (2014)
FiberScout: an interactive tool for exploring and analyzing fiber reinforced polymers. In: Pro-
ceedings of IEEE pacific visualization symposium (PacificVis), 2014, Yokohama, pp 153–160
Welkenhuyzen F (2015) Investigation of the accuracy of an X-ray CT scanner for dimensional
metrology with the aid of simulations and calibrated artifacts. PhD thesis, KU Leuven
Wilhelm RG, Hocken R, Schwenke H (2001) Task specific uncertainty in coordinate measurement.
CIRP Ann 50(2):553–563
Wits WW, Carmignato S, Zanini F, Vaneker TH (2016) Porosity testing methods for the quality
assessment of selective laser melted parts. CIRP Ann 65(1):201–204. https://doi.org/10.1016/j.
cirp.2016.04.054
Xue L, Suzuki H, Ohtake Y, Fujimoto H, Abe M, Sato O, Takatsuji T (2015) Numerical analysis of
the Feldkamp-Davis-Kress effect on industrial x-ray computed tomography for dimensional
metrology. Journal of Computing and Information Science in Engineering, 15(2), 021008.
Yagüe-Fabra JA, Ontiveros S, Jiménez R, Chitchian S, Tosello G, Carmignato S (2013) A 3D edge
detection technique for surface extraction in computed tomography for dimensional metrology
applications. CIRP Ann 62(1):531–534. https://doi.org/10.1016/j.cirp.2013.03.016
Zanini F, Carmignato S (2017) Two-spheres method for evaluating the metrological structural
resolution in dimensional computed tomography. Meas Sci Technol 28:114002. https://doi.
org/10.1088/1361-6501/aa85b7
Zanini F, Carmignato S (2018) X-ray computed tomography for measurement of additively
manufactured metal threaded parts. In: Proceedings – ASPE/EUSPEN 2018 Summer topical
meeting: advancing precision in additive manufacturing.
Zanini F, Carmignato S, Savio E (2017) Assembly analysis of titanium dental implants using X-ray
computed tomography. In: Proceedings of the 17th international conference of the European
society for precision engineering and nanotechnology, euspen 2017, pp 489–490. Hannover,
Germany.
Zanini F, Carmignato S, Savio E, Affatato S (2018a) Uncertainty determination for X-ray computed
tomography wear assessment of polyethylene hip joint prostheses. Precis Eng. https://doi.org/
10.1016/j.precisioneng.2018.02.009
Zanini F, Sbettega E, Sorgato M, Carmignato S (2018b) New approach for verifying the accuracy of
X-ray computed tomography measurements of surface topographies in additively manufactured
metal parts. J Nondestruct Eval 38:12
Zanini F, Gerardi G, Weissenböck J, Heinzl C, Kastner J, Carmignato S (2019a). Experimental
investigation on the accuracy of XCT measurement of fiber length in fiber reinforced polymers.
Proceedings of Conference on Industrial Computed Tomography (iCT2019); 13–15 February
2019, Padova, Italy
Zanini F, Pagani L, Savio E, Carmignato S (2019b) Characterisation of additively manufactured
metal surfaces by means of X-ray computed tomography and generalised surface texture
parameters. CIRP Ann. 68/1
Uncertainty Estimation in Computational
Tools in Metrology 18
Jean-Marc Linares
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 586
Analytical Uncertainties Propagation Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 589
Introduction of GUM Analytical Technique . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 589
Study of Analytical Propagation in Specification Verification . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 591
Rule to Choose the Number Measured Points . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 593
Understanding of Distance Uncertainty . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 595
Analytical Propagation Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 596
Numerical Uncertainties Propagation Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 597
Monte Carlo Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 597
Sobol’s Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 598
Others Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 598
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 599
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 600
Abstract
In the new digital environment of Industry 4.0 for Germany, Innovation 25
program for Japan, Advanced Manufacturing for USA, Intelligent Manufacturing
or Made in China 2025 for China, Factories of the Future for France, etc., the
measurement uncertainty needs a specific management in order to control the
quality of the manufactured part. In this future digital world, where the software
will have a central position in the verification of a specification, it is necessary to
provide to the metrologist data with uncertainty in real time. The aim of this
chapter is to present the uncertainty calculation methodologies. The common
analytical and numerical methods to estimate uncertainty will be presented.
J.-M. Linares (*)

Aix Marseille University, Marseille, France
e-mail: jean-marc.linares@univ-amu.fr

https://doi.org/10.1007/978-981-10-4938-5_20
586 J.-M. Linares
Keywords
Uncertainty · Measurement · Metrology
Introduction
Metrology devices are based on measures of linear (length or distance) or curvilinear

(angle) distances. In the 1970s, the association of the simultaneous measurement of three
distances, taken orthogonally between them, gave birth to the coordinate measuring
machine (CMM). CMM provides the coordinates of the points measured in a Cartesian
coordinate system assumed as a direct orthonormal referential. In 1987, the first laser
tracker was invented. The combination of simultaneous measurement of two angles and
a length is used in this unit. The coordinates of the measuring points are known in the
Polar coordinate system; a simple geometrical transformation allows to know the
coordinates of the measured points in a Cartesian coordinate system. In the same way,
portable measuring arms allow to have point coordinates in a Cartesian coordinate
system after to change the articular coordinates (angular encoder value and the segment
length value) using transformation matrixes. The principle of multilateration allows to
determine the coordinates of measurement points from the measurement of N distances
acquired on a set of measuring points repeated in four positions. The laser tracer uses this
principle. In this case, the measured point coordinates can be known in an intrinsic
reference frame build using three laser tracer positions. Optical measuring devices are
increasingly replacing tactile measuring systems in many areas of industry. Optical 3D
CMMs use mathematical relations between 3D point coordinates of the viewed scene
and 2D coordinates of their own projection in the image plane of the camera. It consists
in finding the transfer function of this geometrical transformation. This model is named
pinhole model. It takes into account the effect of the focus distance, the skew coefficient,
the scale factor of the geometrical projection, and distortion coefficients. New measure-
ment devices (computerized tomography (CT) scan, laser-plane sensor, etc.) provide the
coordinates of a set of points to the metrologist.
These sets of points known in coordinate system represent a discrete representa-
tion of the continuous real part. In a mechanism or in an assembly, the part can
perform mechanical, optical, thermal, etc. functions. For this purpose, geometrical
specifications are imposed by the designer between the part surfaces using ISO
Standard. The ISO geometrical standard was derived from formal language for
GeoSpelling introduced in the 1990s. Figure 1 show an example of ISO specifica-
tion. To control part and mechanism or assembly quality, a geometrical verification is
necessary to verify the specifications conformity. The interpretation of ISO specifi-
cations allows the metrologist to derive the localization of the measured points and to
extract a set of points for each surface (Fig. 2).
Figure 3 shows the verification process of a geometrical specification. This verifi-
cation operation begins with the points measured on the skin model of the specified
element (acquisition of points on skin model (points of circles) and acquisition of
points on real plane). This step is represented in the left top side of Fig. 3.
18 Uncertainty Estimation in Computational Tools in Metrology 587
0.05 A
0.1 A B
260.8
20 20
E
90 H9 E
- 0.05
0
93
polished
Ra 0.15
E
E
- 0.05
0.05 A
- 0.05
0.05 A
0
0
93
93
A
0.02
0.03 A Common Zone ISO 8015
Fig. 1 ISO part specification
A
TI A
Specified element Datum element

Tolerance zone
Medium Single Common Datum
Surface
element Datum zone System
Skin model Datum Reference A
Specified
element
Skin model
Set of Points Plane TI
Fig. 2 ISO specification interpretation
To derive the specified elements for the surfaces with a rotational element (cylinder,
cone, sphere, tore, etc.) or symmetric axes (ellipsoid, paraboloid, Zernike polynomial,
sphere, etc.) and the datum reference, a best-fit procedure is needed. This step allows to
obtain a point, a line, or the plane which will be used to build a single datum, a
588 J.-M. Linares
Sets of measured points Best-fitted surfaces Specificaon

TI A
C
e3
Mi
A
e2 Specificaon verificaon
TI≥MAX-min
e1 VA
C
MAX
min
A
CA
Fig. 3 Verification process of a geometrical specification
common datum, or a reference datum. This best-fit step is represented in the top middle
side of Fig. 3. In many cases, the verification step will be summarized at the calculation
of three distances: point/point, point/line, and point/plane. In the example of Fig. 3, the
distance point/plane will be used to derive the distance between the specified element
(C points) and the datum reference (plane defined by CA point and VA vector). All
metrology software tools use this verification procedure (Quindos, PCDMIS, Cosmos,
Metrolog XG, Calypso, Measure X, etc.).
In the new digital environment of Industry 4.0 for Germany, Innovation 25
program for Japan, Advanced Manufacturing for USA, Intelligent Manufacturing
or Made in China 2025 for China, Factories of the Future for France, etc., the
measurement uncertainty needs a specific management in order to control the
manufactured part quality. Today, just a few metrology software tools provide real-
time estimation of measurement uncertainty during a verification of an ISO specifi-
cation. In a future digital world, where the software will have a central position in the
verification of a specific specification, it is necessary to provide to the metrologist
with uncertainty data in real time.
The aim of this chapter is to present the uncertainty calculation methodologies.
First section focuses about the common analytical methods to estimate uncertainty.
These methods will be detailed, and its advantages and its major disadvantages will
be discussed. In the second section, the numerical methods will be detailed and their
advantages and disadvantages will be presented.
Analytical Uncertainties Propagation Methods
Introduction of GUM Analytical Technique
In uncertainty propagation, the knowledge of some statistical operators is needed:

arithmetic mean of the x variable noted x̄ , variance of x or elementary uncertainty of
x noted v(x) or u2(xi), covariance of xi and xj noted u(xi,xj), and variance of the
measurand y or compound uncertainty of y noted v( y) or u2c( y). The standard
deviation is written as u(xi) and uc(y). The estimated value of the measurand y is
noted ^y. The error bar or the expanded uncertainty U(y) is computer using a coverage
factor k, to reflect a level of confidence of 95% (Linares et al. 2006).
Uncertainties can be analytically evaluated using classical GUM’s method, which
is based on specific propagation laws. A first-order Taylor series expansion is also
used to propagate elementary uncertainties (u(xi)) to the composed uncertainty uc of
the measurand y. The simplest law of propagation is the quadratic sum of elementary
uncertainties.
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
X n
ucðyÞ ¼ u2 ðxiÞ (1)
i¼1
with u(xi) elementary uncertainty on xi, uc( y) compound uncertainty on the estimated,
y^ computed value, and n number of variables x.
When this propagation law is used, two strong hypotheses are implicitly taken:
– xi variables are independent between them

– The sensitivity coefficient of each variable xi is equal to 1
The value of elementary uncertainty u(xi) represents, statistically, the standard

deviation of xi. The value of the standard error is the square root of the value of the
variance. This value of variance or standard error can be estimated using two ways.
In the first way, the standard deviation is computed using experimental values. The
second way is based on the probability density function of the variables xi. The
variance v(x) is derived by the expected value, named E(xE(x)), of a random
variable XE(x), where E(x) is the expected value of x (this value is named mean
value x).
We will explain the physical meaning of the sensitivity coefficient (standardized
term). The sensitivity coefficient is the factor with which the elementary uncertainty
of xi will influence the compound uncertainty of y (Fig. 4). When using Eq. (1) as
propagation law, all the elementary uncertainties will have the same impact on the
590 J.-M. Linares
Fig. 4 Sensibility coefficient

and estimated expanded y Sensibility coefficient
uncertainty U(y)
ŷ U(y)
x1
x1
uncertainty composed of y. The function that models the measurement process is a

simple polynomial, see Eq. (2).
y ¼ a1 :x1 þ a2 :x2 þ . . . : þ ai :xi þ . . . : þ an :xn (2)
If the modeling of the measurement process is more complex than the function y
described above, the assumption of a sensitivity coefficient equal to 1 introduces a
distortion in the propagation process of the elementary uncertainties. This phenom-
enon can lead to under or over estimation of the extended uncertainty on the
measurand estimated value (Fig. 5).
This propagation law will find its main application in the case of direct measure-
ment processes. The values xi are known directly in the same direction as the result y,
and the effect of external conditions is neglected (temperature, humidity, etc.). To
avoid the risk of under- or overestimation of the uncertainty, the following formu-
lation is generally used:
n
X
@y 2
u2 cðyÞ ¼ :u2 ðxiÞ (3)
i¼1
@xi
In Eq. (3), the variables are still considered independent between them. This
assumption nullifies the final term in covariance. The term covariance has an
attenuating effect on the uncertainty of the measurand y. We will show it later in
the presentation.
We now understand the advantage of using the following law of propagation,
proposed by the GUM (Eq. 4).
n
X n1 X
X
@y 2 n
@y @y
u 2
cðyÞ ¼ :u2 ðxiÞ þ 2: : :uðxi xjÞ (4)
i¼1
@xi i¼1 j¼iþ1
@xi @xj
¶y
¶ x1
¶y
Sensibility coefficient=1
y ¶ x1 y
Sensibility coefficient =1
ŷ Extended uncertainty ŷ Extended uncertainty

ŷ ŷ
x1 x1
x1 x1
Fig. 5 Effect of sensibility coefficient
In this propagation law, it is easier to show the attenuating effect of the covariance term
u (xi, xj). If the combination of the two partial derivatives and the covariance (xi, xj) is
negative, then this term will decrease the uncertainty on the measurand y. For the
verification of geometrical specifications, the coordinates of the measured points are
vectors. In this case, the use of the covariance variance matrix is necessary to realize
the propagation of the uncertainties. In the rest of this chapter, we will use the case of the
line in the plane to simplify and explain the uncertainty propagation in the geometric
specification verification.
Study of Analytical Propagation in Specification Verification
To illustrate the GUM approach, the case of a 2D straight line will be considered.
The equation of the line is written as y = a0 + a1. x. The coordinates of the measured
point Mk are xk and yk. Using these measured points, the best linear unbiased
estimation of the parameters a of the straight line are derived using Eq. (5).
P P P P
a0 1 x2P yk Pxk Pxk :yk
a¼ ¼ P P k
(5)
a1 N x2 k ð xk Þ2 N xk :yk xk yk
The variances and covariance of a0 and a1 are derived then classical propagation
equations. The covariance matrix of the both parameters can be computed from Eq. (6).
P 2 P
σp2 xk xk
CovðaÞ ¼ P P P (6)
N xk 2 ð xk Þ 2 xk N
The variance of yk (σp2) is estimated by the residue of the least squares see Eq. (7),
in the line case p = 2.
1 X
σp2 ¼ : ðyk ða0 þ a1 :xk ÞÞ2 (7)
N p
592 J.-M. Linares
Fig. 6 Best fit parameters e2 y=a0+a1.x

and vectorial parameters
C V
e1
In the metrology field, the parameters of the best-fitted surface, in our case a0
and a1, are transformed to geometrical parameters such as the direction vector V of
the straight line and a point C of the feature (Fig. 6). Generally, the point C is
defined by the center of gravity of the measured points. The variance covariance
matrix of these geometrical parameters is calculated using the classical propaga-
tion equation.
For a straight line, the variance covariance matrix of C and V is derived of the
covariance matrix of a0 and a1 using Eq. (8).
ðC,V Þ ¼ J ðC,V Þ :CovðaÞ:J ðC,V Þ

t
Cov
0 1 1
C V ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffi2 :
a0
0 1 þ a1 a1 1
@C x @C y @V x @V y (8)
B @a @a0 @a0 @a0 C
J ðC,V Þ ¼ B
t 0
@ @C x @C y @V x @V y A
C
@a1 @a1 @a1 @a1
Statistical Boundary of the Straight Line

The variance covariance matrix allows to derive the uncertainty at any point of the
line in a normal vector n (Fig. 7). In the study case, a normal vector is chosen
collinear with the second axis e2. Equation (9) defines the calculation of the error
bar along the straight line with a coverage factor k.
varðyÞ ¼ varða0 Þ þ 2:λ:covða0 , a1 Þ þ λ2 :varða1 Þ

pffiffiffiffiffiffiffiffiffiffiffiffi (9)
U ¼ k: varðyÞ
Using Eq. (9), the error bar of the straight line can be plotted. This envelope
depends on the level of confidence α assumed in Eq. (9).
The lowest uncertainty is obtained when the given point is close to the center
of gravity C of the measured coordinates (Fig. 7). The more the variable λ
increases, the more the error bar grows. λ is the distance between the estimation
point and C.
From this property, some measurement rules can be extracted for the metrologist:
the choice of the minimal number of measured points and the localization of the
measured points.
Fig. 7 Statistical boundary Error bar with

e2 n coverage factor a
XC λ
C
V
e1
Rule to Choose the Number Measured Points
The study of this error bar model allows to understand the propagation of the best-fit
uncertainty. Equation (10) gives the variance covariance matrix. In this case, the
origin of the coordinate system is chosen to C. In this case, the calculated covariance
matrix is diagonal.
0P 1 0 2 1
yk σp
B N C B N 0 C
ai ¼ B P C B C
@ xk :yk A Covðai Þ ¼ @ σp 2 A
(10)
P 2 0 P 2
x k xk
The uncertainty of the constant parameter (a0) is mainly influenced by the number
of acquired points. Figure 8 represents the experimental uncertainty of the radius of a
circular gauge measured by a CMM.
When the number of measured points is less than 9 (minimal number of points
is 3), the estimated uncertainty is incorrect. The recommended number required for
the measurement of classical surfaces is proposed in Table 1. It is recommended to
take three times more measuring points than the minimum number necessary to
define the theoretical element that one wishes to best fit.
The slope a1 error bar depends also on λ and the number of points N (see Eq. 11).
When all the acquisitions are performed as far as possible from the center C of the
measured feature, the uncertainty is minimal. In this case, the standard deviation of
a1 is written in Eq. (11).
σ 2p σp
varða1 Þ ¼ ) σ ða1Þ ¼ pffiffiffiffi (11)
N :λ 2
λ N
In that case, the variance (σ2) does not integrate the form defects of the surface. It
includes only the repeatability of the measuring device. This method must thus be
used with caution. It should be applied to standards of perfect geometry. In this case,
the form defects are negligible. In numerous research works, the best distribution of
the measured points is expected. The variance of a1 is defined by Eq. (12). This
situation is shown at the right-hand side of Fig. 9.
594 J.-M. Linares
Radius uncertainty
0,014
σ
0,012 N
0,01
0,008
Experimental uncertainty
0,006
0,004
0,002
0
4 8 12 24 48 96 192
Number of measured points
Fig. 8 Effect of measured point number on best-fit uncertainty
Table 1 Recommended number of the measuring points

Model 3D line Plane Circle Cylinder Cone Sphere
Minimal number points 4 3 3 5 6 4
Recommended number points 12 9 9 15 18 12
Fig. 9 Measured point e2 e2

localization
N/2
N/2 C V C V
λ λ
e1 e1
3:σ 2p 2 þ 3:N þ N 2
varða1 Þ ¼ with Q ¼ (12)
Q:N λ 2
N2
When the number of points N tends to infinity, the limit of Q is equal to 1. An

acceptable value of Q is obtained when N equals three times the number of
parameters. In this case, the standard deviation of the parameter a1 is calculated by
Eq. (13). The effect of the number of points is less significant than in the other case
(Eq. 12). The size of measured surface has a large influence on the a1 uncertainty.
σp
σ ða1 Þ ¼ pffiffiffiffiffiffiffiffiffi (13)
λ N =3
Fig. 10 Uncertainty of the d n

distance
e2 M X
Y d
C V
e1
From a statistical point of view, it is recommended to acquire points randomly

distributed on the surface to avoid any bias introduced by automatic cycles
applied to periodic textures. This situation is shown at the left-hand side of
Fig. 9.This problem has been illustrated by Weckenmann in one of his papers
(Weckenmann et al. 1995).
Understanding of Distance Uncertainty
A verification of geometrical specifications using the ISO1101 standard without

modifier is based on distance calculation (point/point, point/line, or point/plane)
(Bachmann et al. 2004). In this section, the verification of an ISO specification is
studied. The aim is to calculate the distance uncertainty between a point M belong-
ing to the specified element (point) and the straight line considered as the datum
reference (line) (see Fig. 10).
Point M is defined by its mean coordinates and its variance (which is assumed to
exist only in direction e2). Its statistical parameters are written in Eq. (14).

X 0 0
M CovðMÞ ¼ (14)
Y 0 varðY Þ
The datum straight line is defined by both vectors C and V. This datum is
independent of point M. The distance d between the straight line and the point M
is computed by Eq. (15):
d ¼ Y ða0 þ a1 :X Þ (15)
In order to calculate the distance uncertainty, the Jacobian matrix should be build
using Eq. (16):

@ ðd Þ @ ðd Þ @ ðd Þ @ ðd Þ
J¼ , , , ¼ ½1 X a1 1 (16)
@a0 @a1 @X @Y
The variance covariance matrix of the geometrical elements can be build using
Eq. (17):
596 J.-M. Linares
Table 2 Uncertainty of distances PT/PL, PT/DR, and PT/PT
0 1
varða0 Þ covða0 , a1 Þ 0 0
B varða0 , a1 Þ varða1 Þ 0 0 C
Var½parameter ¼ B
@
C (17)
0 0 0 0 A
0 0 0 varðY Þ
The distance uncertainty d is obtained through the Jacobian propagation expres-

sion Eq. (18):

varðd Þ ¼ varðY Þ þ varða0 Þ þ 2:X :covða0 , a1 Þ þ X 2 :varða1 Þ (18)
Equation (18) shows the impact of the reference element uncertainties in the distance
uncertainty. Using the geometrical description of the surfaces (C, V), the procedures to
calculate the uncertainties of the three distances can be presented: point/point (Pt/Pt),
point/line (Pt/Dr), and point/plane (Pt/Pl). These three cases are summarized in Table 2.
Analytical Propagation Limitations
The analytical method makes possible to understand the mechanisms of uncer-

tainty propagation in metrology software (Aranda et al. 2010; Mailhe et al. 2008,
2009a, b). This propagation method has nevertheless some limitations, princi-
pally, when the model of the measurand is nonlinear. In such case, indeed,
the shape of the probability density function is distorted, and some bias is
observed for the calculated mean value of the result. This phenomenon can be
seen in Fig. 11 (Linares et al. 2009).
Fig. 11 Mean value bias Y
Non linear model
Non Gaussian
PDF Bias
Xi
Gaussian
PDF
In the latest version of the GUM standard, numerical uncertainty methods have been
chosen as tools for estimating uncertainties in cases where the computational process is
nonlinear.
In the next section, we will focus on these methods.
Numerical Uncertainties Propagation Methods
The numerical uncertainty propagation methods are based on the repetition of result
computation when the entries vary at each computation. The most frequently used
method is the Monte Carlo simulation. Due to the excessive calculation time, many
palliative methods have been proposed in the literature. In the rest of this section, we
will present this.
Monte Carlo Method
In the case of a nonlinear equation, the analytical methods for uncertainty prop-
agation introduce a bias in the expected result. Recently, a supplement to the
GUM (GUM S1) has shown how to overcome this problem by using the Monte
Carlo simulation method to evaluate uncertainties. Monte Carlo simulation
method is a computational algorithm that relies on repeated random sampling to
obtain numerical results and derive statistical parameters (mean value, standard
deviation). Monte Carlo simulation method is a common tool in uncertainty
evaluation of complex measurement processes (Wen et al. 2013). It is used because
of the lack or the difficulty to express analytical solutions. Instead of using
pseudorandom generators, it can be accelerated by employing deterministic uni-
formly distributed sequences known as presenting low discrepancy. Methods
based on such sequences are named quasi Monte Carlo. Asymptotically, quasi
Monte Carlo can provide a good rate of convergence. Monte Carlo simulation
method needs however numerous repeated random sampling and thus often leads
to large tables.
598 J.-M. Linares
Sobol’s Method
In analytical propagation approach, the sensitivity coefficient may also be defined

by Sobol’s approach (Saltelli 2002a, b). This method is a variance-based global
sensitivity analysis technique founded upon “total sensitivity indices” that account
for interaction effects of the variables. The total sensitivity indices of an input
are defined as the sum of all the sensitivity indices involving that input.
This method includes both main effect of each input and the interactions
with the other variables. Sobol’s method can cope with both nonlinear and
non-monotonic models and provides a truly quantitative ranking of inputs and
not just a relative qualitative measure. Effort has been done to reduce the
computational complexity associated with the calculation of Sobol’s indices.
However, even with its most recent developments, Sobol’s method remains
computer time-consuming.
Others Methods
To reduce the computing time of Monte Carlo simulation method, the Jack-knife,
Bootstrap, or delete d-Jack-knife methods can be used to estimate the uncertainties
of ISO standard specifications (Farooqui et al. 2009). The Jack-knife was thought
up by Quenouille in 1949. After that, Tukey developed its use in statistic
problems. The Tukey’s method requires less computational power than Monte
Carlo simulation method. For a dataset x = (x1, x2, ..., xn) of size n and an
^ the Jack-knife derives estimators θî on subsamples that leave out a
estimator θ,
given selected element xi. The size of each Jack-knife subsample xi) is p = n1,
and the total number of datasets that can be built is n. To estimate an uncertainty,
the standard error of the Jack-knife replications is needed (^ se). Its expression is
written in Eq. (19).
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
n1 X n
2 1 X
n
^
se ¼ θ^ðiÞ θ^ð • Þ with θ^ð • Þ ¼ θ^ðiÞ (19)
n i¼1 n i¼1
Generally, Jack-knife’s method gives fine results for smooth statistics and for
sufficiently large n. Nevertheless, it does not give accurate estimations for non-
smooth statistic or nonlinear behaviour. The Bootstrap method was thought up after
Jack-knife’s method. B. Efron introduced it in 1979. For a dataset x = (x1, x2, . . . ,
xn) of size n and an estimator, the Bootstrap derives the estimator on a resample b of
the same size n. Each resample is obtained by random sampling with replacement
from the original dataset. The total number of resamples that can thus be built is n2.
A lower number D of datasets is however used in practice to estimate uncertainties. It
is usually fixed to D = 200 for standard error estimation and D = 500 for error bar
estimation. The standard error of the estimator can be derived from the Bootstrap
replications using Eq. (20).
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u
u 1 X D
2 1 XD
sê ¼ t θ^ðbÞ θ^ð • Þ with θ^ð • Þ ¼ θ^ðbÞ (20)
D 1 b¼1 D b¼1
The delete d-Jack-knife method consists in generating subsamples, simply by

randomly removing a number d of elements from the initial dataset. The size of each
subsample is thus n-d. The total number of subsamples that can be built is the
number of combinations of d elements removed from the original dataset of size n.
As compared to the earlier Jack-knife scheme, the delete d-Jack-knife subsampling
technique leads thus to a greater number of sub-datasets. This can improve the
accuracy of the method in the case of non-smooth statistics. The standard error of
the estimator can be evaluated through Eq. (21):
vffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
u
u n n
u
u d d
u nd X
2 X
se ¼ u
1
^ u θ^ðiÞ θ^ð • Þ with θ^ð • Þ ¼ θ^ðiÞ (21)
t n i¼1 n i¼1
d:
d d
To obtain an accurate estimate of thepstandard

ffiffiffi error, the number d of deleted data
elements can be selected in the range: n d n 1.
Conclusion
In the new digital environment (Industry 4.0, etc.), the tasks of uncertainty estimation
will take on a new dimension in the industry to control the manufactured product quality.
In this chapter we presented analytical and numerical methods to derive mea-
surement uncertainty. The analytical method is based on the partial derivative of the
measurand function. This method can be extended to the 3D cases using covariance
matrix and the Jacobian matrix. The analytical approach allows the metrologist to
understand the propagation mechanisms. The effects of the measured points’ number
of the measured points’ localization and the geometrical construction on the mea-
surement uncertainty were shown. The limit of the analytical propagation was
presented. When the measurand function is nonlinear, distortions of the probability
density function are detected. This distortion introduces bias on the mean calculation
and over-/less estimation of the uncertainty.
To avoid this impediment, numerical methods were appeared in the uncertainty
estimation. The best method is the Monte Carlo simulation method. This method has
an excessive calculation time. To reduce the calculation time, many palliative
methods have been proposed in the literature: Jack-knife, Bootstrap, or delete d-
Jack-knife methods. These methods present some limitation about the accuracy of
the estimated uncertainty. Many cautions were presented in this chapter to improve
uncertainty estimation quality.
600 J.-M. Linares
References
Aranda S, Linares JM, Sprauel JM (2010) Best-fit criterion within the context of likelihood
maximization estimation. Measurement 43(4):538–548
Bachmann J, Linares JM, Sprauel JM, Bourdet P (2004) Aide in decision-making: contribution to
uncertainties in three-dimensional measurement. Precision Engineering 28(1):78–88
Farooqui SA, Doiron T, Sahay C (2009) Uncertainty analysis of cylindricity measurements using
bootstrap method. Measurement 42(4):524–531
Linares JM, Mailhé J, Sprauel JM (2006) Uncertainties of multi sensors CMM measurements
applied to high quality surfaces. In: XVIII IMEKO world congress
Linares JM, Sprauel JM, Bourdet P (2009) Uncertainty of reference frames characterized by real
time optical measurements: Application to Computer Assisted Orthopaedic Surgery. CIRP
Annals 58(1):447–450
Mailhe J, Linares JM, Sprauel JM, Bourdet P (2008) Geometrical checking by virtual gauge,
including measurement uncertainties. CIRP Annals 57(1):513–516
Maihle J, Linares JM, Sprauel JM (2009a) The statistical gauge in geometrical verification.
Precision Engineering 33(4):333–341
Maihle J, Linares JM, Sprauel JM (2009b) The statistical gauge in geometrical verification. Part II.
The virtual gauge and verification process. Precision Engineering 33(4):342–352
Saltelli A (2002a) Sensitivity Analysis for Importance Assessment. Risk Analysis 22(3):579–590
Saltelli A (2002b) Making best use of model evaluations to compute sensitivity indices. Computer
Physics Communications 145(2):280–297
Sobol0 IM (2001) Global sensitivity indices for nonlinear mathematical models and their Monte
Carlo estimates. Mathematics and Computers in Simulation 55(1–3):271–280
Weckenmann A, Eitzert H, Garmer H, Weber H (1995) Functionality-oriented evaluation and
sampling strategy in coordinate metrology. Precision Engineering 17(4):244–252
Wen XL, Zhao YB, Wang DX, Pan J (2013) Adaptive Monte Carlo and GUM methods for the
evaluation of measurement uncertainty of cylindricity error. Precision Engineering 37
(4):856–864
Molecular Dynamics Characterization of a
Force Sensor Integrated Fast Tool Servo for 19
On-Machine Surface Metrology
Yindi Cai
Contents
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 602
Molecular Dynamics Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 604
Model for Subnanometric Tool-Workpiece Contact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 604
Subnanometric Tool-Workpiece Contact Process . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 608
Stabilization of the MD Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 609
Vibration of Workpiece Atoms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 610
Multi-relaxation Time Method at Low and Room Temperatures . . . . . . . . . . . . . . . . . . . . . . . . . . . 614
Characterization of the Subnanometric Tool-Workpiece Contact . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 618
The ftw-ztool Curve . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 618
Identification of the Elastic-Plastic Transition Contact Depth . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 621
References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 627
Abstract
Establishment of the tool-workpiece contact, in which the diamond tool is set on
the workpiece surface with a small contact force, determines the depth of cut
accuracy in a force sensor-integrated fast tool servo (FS-FTS) for single point
diamond microcutting and the scan force and scan depth in the following step of
on-machine surface metrology. Molecular dynamics (MD) simulations are carried
out to characterize the tool-workpiece contact process. It is clarified that even a
small instability induced by the vibration of the workpiece atoms can generate
large uncertainties in the subnanometric MD simulation results. Based on the
vibration of the workpiece, atoms have a certain period determined by the MD
model size; a multi-relaxation time method is proposed for reduction of the atom
vibrations and stabilization of the MD model. It is confirmed that the proposed
multi-relaxation time method is effective to eliminate the instability over a wide
Y. Cai (*)
Dalian University of Technology, School of Mechanical Engineering, Dalian, Liaoning, China
e-mail: caiyd@dlut.edu.cn; happy369cyd@163.com

https://doi.org/10.1007/978-981-10-4938-5_21
602 Y. Cai
temperature range up to room temperature under which a practical microcutting

or surface metrology process is carried out. An accurate elastic-plastic transition
contact depth is then evaluated by observing the residual defects on the workpiece
surface after the diamond tool is retracted back to its initial position.
Keywords
Molecular dynamics · Diamond tool · Microcutting · Surface metrology · Contact
depth · Contact force · Elastic-plastic transition · Fast tool servo · Surface
damage · Multi-relaxation time method · Temperature
Introduction
Ultraprecision diamond microcutting is widely employed for fabrication of high

accuracy microstructures over large areas of diamond machinable materials (Gao
et al. 2003; Brinksmeier et al. 2012; Mellor et al. 2013; Fang et al. 2013; Chen et al.
2014) by using a single point diamond tool on a diamond turning machine (Gao et al.
2007; Lee et al. 2011) or an ultraprecision lathe (Cheung and Lee 2003; Chen et al.
2015b, 2016). The increasing demand for devices and systems with nanometric size
and nanometric or subnanometric accuracy has pushed the machining accuracy of
the ultraprecision diamond microcutting and its following step of surface metrology
to the subnanometric range (Sun et al. 2012; Cheng and Huo 2013).
In the conventional FTS-based microcutting, in order to generate the complicated
tool paths corresponding to the designed microstructured surface forms, the fast
in-feed motion of the diamond tool is controlled by a PZT actuator of the FTS in
association with the slow cross-feed motions of the machine axes. The tool path can
be generated with nanometric accuracies by the well-developed PZT and the ultra-
precision machine tool technologies. However, the accuracy of the machined micro-
structured surface is significantly affected by the cutting conditions. Therefore, a
force sensor-integrated FTS, which is referred to as the FS-FTS, has been developed
to assure and improve the machining accuracy of the FTS-based microcutting (Noh
et al. 2008; Gao et al. 2015). The developed FS-FTS also has a unique function of
in-process measurement of microstructured surface form. After microstructured
surface is fabricated by a diamond tool, which is controlled by the FS-FTS and the
machine tool slide, the fabricated microstructured surface can be scanned with
the same diamond tool by using the FS-FTS and the machine tool slide to generate
the scanning motions. The position of the diamond tool tip along the depth of cut
direction is servo-controlled by the PZT actuator of the FS-FTS, so that the contact
force between the diamond tool and the workpiece surface is kept constant. The
contact force is monitored by a force sensor of the FS-FTS. A displacement sensor of
the FS-FTS is employed to measure the servo-controlled PZT motion. The micro-
structured surface form of the workpiece is then accurately reconstructed from a
combination of the PZT motion and the machine axes during the scanning motion.
This function has been employed for a set of on-machine surface metrology,
19 Molecular Dynamics Characterization of a Force Sensor Integrated Fast. . . 603
including on-machine measurement of diamond cutting tool edge contour (Chen

et al. 2015a), relay fabrication of microstructure array (Chen et al. 2014), and
in-press detection and repair of micro-defects (Chen et al. 2015b).
Establishment of the tool-workpiece contact, in which the diamond tool tip is set
on the workpiece surface with a small contact force, is the first step for microcutting
and on-machine form measurement based on the FS-FTS. The tool-workpiece
contact determines the depth of cut accuracy, which significantly influences the
achievable machining accuracy of the ultraprecision diamond microcutting and the
scan accuracy of the on-machine surface metrology. For establishment of the tool-
workpiece contact, the diamond tool is brought toward the workpiece surface by
controlling the machine tool slide or the PZT actuator of the FS-FTS, while the
output of the force sensor of the FS-FTS is monitored simultaneously. The tool-
workpiece contact is established when the output of the force sensor reaches a
threshold value. Ideally, it is desired to establish the tool-workpiece contact within
the purely elastic region of the workpiece surface so that the accuracy of the
ultraprecision diamond microcutting and the integrity of the scanned microstructured
surface can be maintained. In responding to these requirements, it is necessary to
physically understand the elastic and plastic deformation behaviors of the workpiece
material in a subnanometric range when the workpiece surface is contacted with the
diamond tool with specific cutting edge geometry and make clear the threshold
contact depth and the corresponding contact force of the subnanometric tool-
workpiece contact.
Elastic contact is a traditional problem, which can be investigated by the classical
contact mechanics based on the Hertz theory (Johnson 1985; Oliver 1992). How-
ever, as the contact depth goes down to the nanometer-scale and the contact
phenomenon takes place in a limited region, the conventional continuum theory
cannot provide a reasonable explanation in such a small scale because of the
nonlinear constitutive equations and the complicated material parameters (Oliver
1992). The instrumented nanoindentation systems have a capacity of measuring the
mechanical property of thin surface layers such as coatings and thin films (Page et al.
1992; Gao et al. 2000; Motoki et al. 2006; Huang and Zhao 2015). Typically, a
Berkovich-type diamond indenter with a large tip diameter of several micrometers is
used in the nanoindentation system. A diamond tool, with a nanometric cutting edge
formed by a flat rake face and a conical or cylindrical clearance face, is not
compatible with the nanoindentation system. Therefore, the diamond tool is not
compatible with the indenter in a nanoindention instrument. The current version of
the FS-FTS instrument is also difficult to be employed for the subnanometric
indentation experiment due to the insufficient performances of the machining instru-
ment. On the other hand, simulations on the different time scale and length scale
have become powerful approaches to analyze the deformation mechanisms of
various materials, such as finite element method (FEM) (Özel and Zeren 2004;
Farzad and Abdolreza 2015), multiscale simulations using quasicontinuum
(QC) method (Wang et al. 2008), and molecular dynamics (MD) simulations (Goel
et al. 2011, 2015). In spite of MD, simulation suffers from the limitation on both time
604 Y. Cai
scale and length scale; yet, it has become a powerful approach because it enables to
predict and analyze the nanoscale machining in theory, which provides a shortcut
from micro phenomena to macro characteristics. On the other hand, the purpose
of carrying out an MD simulation in many researches is not to replicate the
experiment but to develop a theoretical understanding of the deformation and
machining mechanisms (Faisal et al. 2014). Therefore, MD simulation has the
potential for investigation of the deformation behavior of the workpiece of the
subnanometric tool-workpiece contact in surface metrology.
In this research, cost-effective MD simulations are carried out to characterize the
subnanometric tool-workpiece contact process in the single point diamond micro-
cutting and in-process surface metrology for the purpose of optimization of the
FS-FTS. Based on the investigations of workpiece atom vibrations, a multi-
relaxation time method is then proposed to reduce the influence of the atom
vibrations. After confirmation of the feasibility of the multi-relaxation method for
MD model stabilization, identification of the elastic-plastic transition contact depth
of the workpiece is carried out.
Molecular Dynamics Model
Model for Subnanometric Tool-Workpiece Contact
Figure 1 shows the schematic of the FS-FTS on a diamond turning machine where
the tool-workpiece contact is established for in-press form measurement. The
FS-FTS consists with a diamond tool, a force sensor, a PZT actuator, and a displace-
ment sensor. The motion of the diamond tool along the Z-axis is servo controlled by
the PZT actuator and the displacement sensor. The force sensor is applied to detect
the contact force between the diamond tool and the workpiece surface.
MD simulation is employed instead of experimentation for physically understand-
ing the elastic and plastic deformation behaviors of the workpiece material in a
subnanometric range when the workpiece surface is contacted with the diamond tool
with specific cutting edge geometry. The process of subnanometric tool-workpiece
contact can be treated as the nanoindentation of a workpiece using a diamond tool.
Similar to a typical MD simulation of depth and load sensing indentation, a diamond
tool and a workpiece are two fundamental elements in MD simulation of the
subnanometric tool-workpiece contact. A round-nose diamond tool and a copper
workpiece, which both are popular in the diamond microcutting, are employed in
the tool-workpiece simulation. The schematic of a round-nose single point diamond
tool and the enlarged view of the cutting edge over the arc AB are shown in Fig. 2a.
The diamond tool has a round contour with a contour radius R in the range of
millimeter. It is assumed that arc AB of the tool cutting edge is contacted with the
workpiece surface in the process of the subnanometric tool-workpiece contact and Arc
AB is a straight line. It can be seen from the enlarged view of the cutting edge tip over
the arc AB as shown in Fig. 2a; the diamond tool has a nanometric edge sharpness
denoted by ρ. The rake angle and the clearance angle are other two important
Subnanometric tool-
workpiece contact
FS-FTS
ing
turn Displacement PZT
mondchine sensor actuator
a
Di ma
Copper
workpiece
Diamond Force
Y-slide
Tool sensor
FS-FTS
Fig. 1 Schematic of the tool-workpiece contact system
parameters of the geometry of the diamond tool. The rake angle and the clearance
angle are the angles of the rake and clearance faces with respect to the normal of the
workpiece surface, respectively. Figure 2b shows the MD model of the tool-workpiece
contact. Diamond tool edge sharpness ρ is set to be 3 nm and the rake angle and
the clearance angle θc are set to be 0 and 7 , respectively. Due to the size limitation of
the MD simulation, only arc AB is modeled as the diamond tool with length CAB of 2.5
in the MD model shown in Fig. 2b. The lattice constant of diamond is 0.3567 nm. The
initial position of the diamond tool along the Z-direction is denoted as ztool_ini, as
shown in Fig. 2b. The workpiece is a single crystal copper substrate containing
119,940 atoms with a size of 15 nm (lwp) 15 nm (wwp) 6 nm (hwp) along the
[100], [010], and [001] directions, respectively. The lattice constant of copper is
0.3610 nm. The reason of choosing copper as the material for the workpiece is in
view of its recent emergence as the interconnect material of microelectronic devices.
The copper workpiece size is significantly larger than the indentation contact area at
606 Y. Cai
Fig. 2 The model for tool-

B
workpiece contact (a) (a) X
CAB
schematic of the diamond
Y
tool; (b) model for MD
simulation of tool-workpiece Z
contact A
A
Rake face R
ρ Rake face
B
Y X
θc Z
ank
l sh ce
Too ance fa
ar
Cle
the maximum contact depth; the spurious effects originating from the finite size of the
workpiece thus can be avoided. Similar to the typical workpiece model in MD
simulations of nanoindentation, there are three kinds of atoms included in the work-
piece, namely boundary atoms, thermostat atoms, and Newtonian atoms. The three
layers of the boundary atoms, which surrounded the workpiece except for the free
(001) surface at z = 0, are kept to fix the workpiece in space, to prevent it from
translating during the subnanometric tool-workpiece contact. The next three layers of
the workpiece atoms adjacent to the boundary atoms are the thermostat atoms. During
MD simulations, heat dissipation is carried out by keeping the thermostat atoms at a
constant temperature by the Nosé-Hoover thermostat method (Nosé 2002). The
simulation temperature is set to be 0.1 K at first to avoid the effect of thermal vibration
on simulation results. The rest of the workpiece atoms are the Newtonian atoms. The
Newton’s equations of motion are integrated with a velocity-Verlet algorithm with a
time step of 1 fs.
Since the contact forces are derived from the integration of the interaction
potential with respect to the coordinate position of the atoms, the selection of the
interaction potential is crucial to the reliability of the simulation results. The simu-
lation models established in the research are composed of two types of atoms, copper
atoms and diamond atoms. There are three different atomic interactions that can
possibly occur between these atoms, namely the interactions between the workpiece
copper atoms; the tool diamond atoms and the workpiece copper atoms; and the tool
diamond atoms, as shown in Fig. 3. Since the diamond tool is regarded as a rigid
body, the interactions between the diamond tool atoms are ignored in this simulation.
The embedded atom method (EAM) potential, which has been specially developed
for metals (Zhang and Tanaka 1997), can better describe the metallic bonding. The
EAM potential gives a more realistic description of the behavior and properties of
metals than the Morse potential. Therefore, the EAM potential is employed for
describing the interactions between the copper atoms in the workpiece. As there is
no available EAM potential between the diamond atom and the copper atom, the
interactions between the tool diamond atoms and the workpiece copper atoms are
described by the Mores potential, which has been widely used in nanoindentations
(Fang et al. 2003; Pei et al. 2009). The Mores potential between the copper atoms
and the diamond atom with the parameters of D0 = 0.087 ev, α = 51.4 nm1, and
Diamond
tool
j2
Copper
r0_Morse
h2 j2
workpiece rih_Morse rij_Morse
h2
a1 b1 c1
a1 b1 c1
rib_EAM
h1 i1 j1 rih_EAM rij_EAM
r0_EAM h1 i1 j1
rim_EAM
l1 m1 n1
l1 m1 n1
Equilibrium at time t At time t + dt
Interaction between workpiece atoms (EAM potential)

Interaction between workpiece atoms and tool atoms
(Morse potential)
Fig. 3 Interactions in the MD model
608 Y. Cai
Table 1 Computational parameters of MD simulations

Configuration Workpiece copper (FFC) Tool diamond (rigid)
Lattice constant 0.3610 nm 0.3567 nm
Cu-Cu: EAM potential
Cu-C: Morse potention
Potential D = 0.087 eV; α = 51.40 nm1; r0 = 0.2050 nm,
rcut-off = 0.9025 nm
C-C: Ignored
Time step 0.001 ps
Initial temperature 0.1 K
Contact depth From 0.1 nm to 0.5 nm with a step of 0.01 nm
Contact velocity 20 m/s
Retract velocity
Contact direction [001] direction on the (001) surface
r0 = 0.205 nm. For the sake of clarity, the computational parameters of the
simulation model are summarized in Table 1.
A personal computer (PC) with an Intel (R) Core (TM) CPU of 3.5 GHz and a
RAM of 16 GB is employed for saving the computation cost. A public domain
computer code “Large-scale atomic/molecular massively parallel simulator”
(LAMMPS) (Plimpton 1995) is implemented for the MD simulations of the sub-
nanometric tool-workpiece contact. The Visual Molecular Dynamics (VMD) (Hum-
phrey et al. 1996) is employed to visualize the positions of the workpiece atoms
calculated by LAMMPS during the simulation. The type of defects on the workpiece
is identified by using the Open Visualization Tool (OVITO) (Stukowski 2010).
Subnanometric Tool-Workpiece Contact Process
Similar to the MD simulation of depth and load sensing nanoindentation, the MD

simulation of the tool-workpiece contact includes three subsequent steps, namely the
relaxation step and the tool-workpiece contacting and retracting steps, respectively.
Figure 4 shows a schematic of the procedure of the MD simulation of the tool-
workpiece contact. The diamond tool is initially positioned at point A with a position
of ztool_ini along the Z-direction. ztool_ini is larger than the equilibrium distance
between the tool diamond atom and the workpiece copper atom in Morse potential.
In the relaxation step, the MD model is relaxed in a canonical ensemble (Wang
et al. 1991), in which the number of atoms, N; the volume of the model, V; and the
specified temperature, T, are kept constant, from the initial state with the artificially
assigned initial conditions to its natural, dynamically equilibrium condition by
running the MD program over a relaxation time of Δtrelax. After running the MD
program over Δtrelax in the canonical ensemble, the internal potential energy of the
MD model will gradually reach equilibrium around the initial temperature based on
the thermal dynamic theory (Cheong et al. 2001).
Fig. 4 Schematic of the ztool_ini z=0 ztool_M Tool position

procedure of the MD
simulation of the
subnanometric tool-
workpiece contact d Atom_W
Atom_W
Rake face
v tool
workpiece
f tt-w
-w
Copper
Diamond
tool A S
X v tool
Z
Y
Clearance face
d lattice
A WA X
Atom_WAtom_X
Time
Δtrelax ΔtAM ΔtMA
Relaxation Contacting step
step
Retracting step
After the system has reached its equilibrium state, the tool is approached to the
workpiece surface z = 0 (point S) with a constant velocity of vtool. The diamond tool
is penetrated into the workpiece surface until it reaches a command contact depth
ztool_M at point M. This process is referred to as the contacting step. Then the
diamond tool is retracted back to its original position (point A) with the same
velocity vtool, which is referred to as the retracting step. The time of contacting
step and retracting step for the diamond tool that travels from point A to point M and
from point M to point A are ΔtAM and ΔtMA, respectively. Since the contact velocity
is equal to the retract velocity, ΔtAM is equal to ΔtMA. The interaction force ftw
between the diamond tool atoms and the copper workpiece atoms is calculated by the
Morse potential between the copper atoms and the diamond atoms for investigation
of the subnanometric tool-workpiece contact.
Stabilization of the MD Model
The interaction force ftw with respect to the tool position ztool is the summation of all
the interatomic force components applied on the diamond tool atoms with a positive
direction along the contact direction. Since the primary motivation of this simulation
610 Y. Cai
is to obtain the elastic-plastic transition contact depth and the corresponding contact
force of the subnanometric tool-workpiece contact, the investigation of the relation-
ship between ftw and ztool is necessary. Taking into consideration that the tool-
workpiece contact is in the subnanometric range, even a small instability induced by
the vibrations of the workpiece atoms can generate large uncertainties in the MD
simulation results. None of the researches has focused on the investigations of the
stabilization of the MD model.
Vibration of Workpiece Atoms
In a regular lattice with harmonic forces between atoms, atoms keep vibrating
around their equilibrium positions and the vibrations of neighbor atoms are not
independent of each other (Okamoto 2014). The initial model of the MD simula-
tion is constructed by positioning the atoms at the perfect lattice structures and the
atoms are bound out to their average positions (Cheong et al. 2001). At first, the
copper workpiece of the MD model is relaxed 10 ps under the temperature of 0.1 K
without the diamond tool. Figure 5 shows the free surface profile of the workpiece
(z = 0) over the relaxation time Δtrelax at different moments, namely Δtrelax = 0 ps,
2 ps, 4 ps, 6 ps, 8 ps, and 10 ps, respectively. The color of the figure is drawn based
on the coordinate position of the atoms at the free surface. It can be seen from the
figure that the workpiece atom vibrations let the atom to leave its original position
Free surface
Workpiece
Δt relax=0 ps Δt relax=2 ps Δt relax=4 ps

0.02 nm/div.
0.02 nm/div.
0.02 nm/div.
Z-position
Z-position
Z-position
Y-p Y-p Y-p

5 n ositio sition 5 n ositio sition 5 n ositio sition
m/d n
iv. X-po /div. m/d n
iv. X-po /div. m/d n
iv. X-po /div.
5 n m 5 n m 5 n m

0.02 nm/div.
0.02 nm/div.
0.02 nm/div.
Z-position
Z-position
Z-position
Y-p Y-p Y-p

m/d n
iv. X-po /div. m/d n X-po /div. m/d n X-po /div.
5n m iv. 5 n m iv. 5nm
Fig. 5 Free surface profile in the simulation model (15 nm 15 nm 6 nm) without the diamond
tool under the temperature of 0.1 K over the relaxation step
Fig. 6 Atom vibration in the 0.15

simulation model 6 nm
Atom_W
(15 nm 15 nm 6 nm)
15 nm
without the diamond tool 0.10
under the temperature of 0.1 K
dAto m _ W nm
over the relaxation step 0.05
-0.05 Tvib =4.4 ps
-0.10
0 2 4 6 8 10
Time ps
Fig. 7 Internal potential -4.50

energy under the temperature
Internal potential energyh10-14J
of 0.1 K
-4.65
-4.80
-4.95
ΔTsteady
-5.10
0 2 4 6 8 10
Time ps
even at a low temperature of 0.1 K. The Z-directional displacement dAtom_W of the

Atom_W at the center of the free surface of the workpiece over the relaxation time
is shown in Fig. 6. It can be seen from the figure that Atom_W keeps vibrating
along the Z-direction with an amplitude dAtom_W of 0.05 nm and a period Tvib of
4.4 ps during the entire relaxation time. Figure 7 shows the internal potential
energy of the workpiece over the relaxation step. The internal potential energy of
4.63 1014 J drops to 4.95 1014 J after a time ΔTsteady of 0.5 ps. As can be
seen in Fig. 7, the internal potential energy keeps its steady state after ΔTsteady,
which means the system has reached its equilibrium state in terms of internal
potential energy.
Similar simulation is also carried out on a small MD workpiece model, with a
size of 7.5 nm (lwp) 7.5 nm (wwp) 3 nm (hwp), to investigate the effect of the
workpiece volume on the workpiece atom vibrations. The result is shown in Fig. 8.
It can be seen from the figure that the period and amplitude of the workpiece atom
vibrations are reduced to 2.2 ps and 0.02 nm on a small workpiece MD model from
612 Y. Cai
Fig. 8 Atom vibration in the 0.15

simulation model 3 nm
(7.5 nm 7.5 nm 3 nm) Atom_W
7.5 nm
without the diamond tool over 0.10
the relaxation step under the
temperature of 0.1 K
dAto m _ W nm
0.05
-0.05 Tvib =2.2 ps
-0.10
0 2 4 6 8 10
Time ps
that of 4.4 ps and 0.05 nm on a large workpiece MD model. It is verified that

the workpiece volume is one of the dominating factors for workpiece atom
vibrations.
Simulations are then carried out on the MD model, shown in Fig. 2b, with the
diamond tool. The initial position ztool_ini of diamond tool is set to be 0.15 nm
and – 0.3 nm. The MD model is relaxed 10 ps and the diamond tool is kept
stationary in the relaxation step. Figure 9a shows dAtom_W under the diamond tool
located at various initial positions. The amplitude of dAtom_W is increased from
0.035 nm to 0.063 nm as ztool_ini increased from 0.3 nm to 0.15 nm, respec-
tively. The increase of dAtom_W is caused by the external force, which is induced by
the interactions between the tool diamond atoms and the workpiece copper atoms.
Based on the Morse potential, the external force is increased with the decrease of
the distance between the tool and workpiece. However, the external forces do not
change the period Tvib of dAtom_W. Similar phenomena are also observed in the
variation of the interaction force ftw, as shown in Fig. 9b. As can be seen in
Fig. 9b, the variation amplitude of ftw at ztool_ini of 0.3 nm and 0.15 nm is
evaluated to be 0.15 nN and 12.0 nN, respectively. The significant change of
the force amplitude is due to the exponential relationship between the Morse
potential and the tool-workpiece distance. ftw varies periodically at each ztool
with the same period Tvib of 4.4 ps. Therefore, it is indicated that the variation of
ftw is caused by the vibrations of workpiece atoms.
After the relaxation step at ztool_ini of 0.30 nm, the tool is moved with a
constant velocity vtool of 20 m/s to contact with the workpiece surface along the Z-
direction. It should be noted that the velocity used in the MD simulations is several
orders higher than that of actual experiment (106 m/s–109 m/s). A velocity of
1 m/s to 200 m/s is often chosen in most MD simulations of nanoindentation for
the sake of reducing the computational time and cost (Zhu et al. 2011; Qiu et al.
2014). The command contact depth ztool_M is set to be 0.15 nm. Figure 10 shows
ftw-ztool curve in the relaxation and contacting steps. As can be seen in Fig. 10, a
periodic component with a period of approximately 4.4 ps, which is the same as
Fig. 9 Atom vibration in the

simulation model (a)
ztool_ini = -0.3 nm
(15 nm 15 nm 6 nm) with
the diamond tool under the
dAto m _ W 0.05 nm/div.

temperature of 0.1 K over the
relaxation step (a) dAtom_W;
(b) ft_w
ztool_ini = -0.15 nm
Time 2 ps/div.
(b)
f t-w -0.2 nN/div.
ztool_ini = -0.3 nm
f t-w -10 nN/div.
ztool_ini = -0.15 nm
Time 2 ps/div.
that of Tvib shown in Fig. 6, is generated in ftw-ztool curves before the tool reaches
to ztool = 0. It is inferred that the periodic component should be induced by the
vibration of the workpiece atoms. The amplitude of the periodic component
Uvib_tool is evaluated to be 17 nN. As can be seen from the simulation results
that even at a low temperature of 0.1 K, ftw- ztool curve has an instability, which
will influence the identification of the elastic-plastic contact depth of the sub-
nanometric tool-workpiece contact.
Since the vibration period is on the same order as the contacting operation
time and/or the retracting operation time, a simple low-pass filtering or moving
average method of the data, which is often used to remove the periodic component
in data handling process, cannot be employed for ftw-ztool curve. Thus, it is
necessary to propose an effective method to reduce the instability and stabilize
the MD model.
614 Y. Cai
Fig. 10 Instability of the Time ps

ftw-ztool curve over the 0 5 10 15
40
relaxation and contacting
steps using a diamond tool at
the initial temperature of 0.1 K
20
f t-w nN
0
-20
Uvib_tool =17 nN
Tvib =4.4 ps
-40
-0.15 -0.075 0.075 0.15
ztool nm
Multi-relaxation Time Method at Low and Room Temperatures
Based on the above investigations of vibrations of the workpiece atoms and

the contact force between the diamond tool and the workpiece in the relaxation
and contacting steps, a multi-relaxation method has been proposed to reduce the
instability of the contact force (Cai et al. 2016a). In this method, a group of
MD simulations of tool-workpiece contact, with the same tool-workpiece
contacting and retracting steps but a different relaxation time Δtrelax in a relaxation
step, is carried out. The relaxation time Δtrelax_i in the i th simulation can be
determined by
Δt relax_i ¼ Δt offset þ Δt interval i ði ¼ 1,2, . . . ,N Þ (1)
T vib
Δt interval ¼ (2)
N
where Δtoffset represents the initial relaxation time, Δtinterval represents the time
interval in a period of the workpiece atom vibrations, N represents the number of
the tool-workpiece contact simulations, and Tvib represents the period of the
workpiece atom vibration. It should be noted that Δtoffset should be longer than
ΔTsteady, which is the time for the inertial potential of the system to reach its steady
state. A group of ftw-ztool data sets can be obtained from the group of tool-
workpiece contact simulations. ftw_i(ztool) data sets (i = 1, 2, . . ., N ) obtained
in each group of simulation will have identical periodic error components caused
by the workpiece atom vibrations but with a phase difference of 2π/Ν as each
simulation has the same parameters of ztool_ini, ztool_M, and vtool, but Δtrelax. Taking
the average of ftw(ztool) at ztool in the i th simulation (i = 1, 2, . . ., N ) of at each
Δtrelax, the effect of the workpiece atom vibrations can be removed by using the
following equation:
Fig. 11 Stabilized ftw-ztool Time ps

curve over the relaxation and 0 5 10 15
contacting steps with a 40
diamond tool at the initial
temperature of 0.1 K
20
f t-w nN
0
- 20
- 40
-0.15 -0.075 0.075 0.15
ztool nm
P
N
f tw_iðztoolÞ
i¼1
f twðztool Þ ¼ (3)
N
where ztool_ini ztool ztool_M.
In the group of tool-workpiece contact simulations, Δoffset, Δinterval, and N are set to
be 5.6 ps, 1.1 ps, and 5, respectively. Stabilized ftw-ztool curve of Fig. 10 over the
relaxation and contacting steps by applying the proposed multi-relaxation time method
is shown in Fig. 11. It can be seen from the figure that the period error component with
a period of Tvib and uncertainty of Uvib_tool has been removed, from which the
effectiveness and feasibility of the proposed methods for the reduction of the contact
force instability and the stabilization of the MD simulation model are demonstrated.
However, in real situation, the FS-FTS-based microcutting and on-machine
surface form measurement are performed at a room temperature around 293 K. In
order to make the simulation results more practically useful, the effectiveness of the
multi-relaxation time method for reduction of the atom vibrations and stabilization of
the MD model should be verified under the condition of room temperature.
MD simulations of tool-workpiece contact are then carried out at a wide
temperature range, from 0.1 K to 293 K. The computational parameters, except for
temperature, are similar with that in Table 1. Figure 12 shows the number of
the workpiece atoms, Natom, of one layer on XZ-plane under various temperatures
over the relaxation step. The higher the temperature is, the more intensively its
atoms oscillate around their equilibrium position. Thus, Natom decreases with the
increase of temperature. Natom at the temperature of 0.1 K is kept constant of 1428
over the entire relaxation step due to low amplitude vibration. The Natom is reduced
from 1332, 1187, and 1083 at Δtrelax = 2 ps to 1306, 1126, and 1000 at
Δtrelax = 10 ps with respect to the temperature of 100 K, 200 K, and 293 K,
respectively.
616 Y. Cai
Temperature K
0.1 100 200 293
Natom =1428 Natom =1332 Natom =1187 Natom =1083

Time ps
10
3.0 nm
X
Z
3.0 nm
N represents the atoms number
Fig. 12 Number of workpiece atoms in the lattice structures on the XZ-plane
The calculated Z-directional displacement dAtom_W of the Atom_W (random atom of

the workpiece) at different temperatures is summarized in Fig. 13a. It can be seen from
the figure that the amplitude of dAtom_W is increased from 0.034 nm to 0.067 nm,
0.093 nm, and 0.12 nm as the temperature increases from 0.1 K to 100 K, 200 K, and
293 K, respectively. The lattice constant dlattice of copper, which is the distance
between two adjacent atoms (Atom_W and Atom_X) in the copper workpiece, is
also evaluated at various temperatures. The evaluated results are shown in Fig. 13b. As
can be seen that dlattice is raised from 0.3590 nm to 0.3600 nm, 0.3608 nm, and
0.3618 nm with the temperature increasing from 0.1 K to 100 K, 200 K, and 293 K,
Fig. 13 Effect of the 0.3630

temperature on the vibration (a)
of workpiece atom (a) lattice T=293 K
dlattice nm
constant of copper under T=200 K
different temperatures; 0.3605 T=100 K dAtom_W
(b) dAtom_w under different
T=0.1 K
temperatures
Atom_W
0.3580
-100 0 100 200 300 400
Temperature K
0.20
(b) T=293 K
T=200 K
dAto m _ W nm
T=100 K
0.10 dlattice
T=0.1 K
Atom_W Atom_X
0
-100 0 100 200 300 400
Temperature K

0.2 nm/div.
0.2 nm/div.
0.2 nm/div.
Z-position
Z-position
Z-position
Y-p Y-p Y-p

m/d n X-po /div. m/d n X-po /div. m/d n X-po /div.
iv. 5 nm iv. 5 nm iv. 5 nm

0.2 nm/div.
0.2 nm/div.
0.2 nm/div.
Z-position
Z-position
Z-position
Y-p Y-p Y-p

m/d n X-po /div. m/d n X-po /div. m/d n X-po /div.
iv. 5 nm iv. 5 nm iv. 5 nm
Fig. 14 Free surface profile in the simulation model (15 nm 15 nm 6 nm) without the diamond
tool under the temperature of 293 K over the relaxation step
respectively. It is verified that the temperature significantly influences the performance

of the MD model. The free surface profile of the workpiece without the diamond tool
over the relaxation time at different moments is shown in Fig. 14. Similar to Fig. 5, the
atoms are deviated from its original position due to the atom vibration. The
Z-directional displacement dAtom_W of the Atom_W over the relaxation time is shown
in Fig. 15. It can be seen from the figure that the Atom_M vibrates with an amplitude of
618 Y. Cai
Fig. 15 Atom vibration in 0.20

the simulation model Tvib=4.4 ps
(15 nm 15 nm 6 nm)
without the diamond tool 0.15
under the temperature of
dAto m _ W nm
293 K over the relaxation step 0.10
0.05 6 nm
Atom_W
15 nm
0
-0.05
0 2 4 6 8 10
Time ps
0.15 nm after an increase from that of 0.05 nm at the temperature of 0.1 K and a period
Tvib of 4.4 ps over the entire relaxation step. Therefore, it is identified that although
dAtom_W increases with the increase of the temperature, the period of the atom vibration
remains the same at the temperatures of 0.1 K and 293 K. Figure 16a shows the
instability of the ftw-ztool curve in the relaxation and contacting steps at the initial
temperature of 293 K. It can be seen from figure that ftw-ztool curve has a periodical
component, especially when the diamond tool is located at a position in the range from
ztool = 0.15 nm to ztool = 0 nm. The period of the error component is evaluated to be
4.4 ps and the uncertainty of it is approximately 23 nN. Figure 16b shows the stabilized
ftw-ztool curve by the proposed multi-relaxation time method. It can be seen from
figure that the periodic vibration components is well-eliminated based on the multi-
relaxation time method. The uncertainty induced by the error component is reduced
2 nN from 23 nN. It is verified that the multi-relaxation time method is effective for
stabilization of the MD model over a wide temperature range up to the room temper-
ature (Cai et al. 2016b). The multi-relaxation time method is employed in all the
following simulations.
Characterization of the Subnanometric Tool-Workpiece Contact
After stabilization of the MD model, the identifications of the elastic-plastic transi-

tion contact depth of the copper workpiece under the diamond tool are carried out. In
order to avoid the effect of thermal vibration on simulation results, the simulation
temperature is set to be 0.1 K.
The ftw-ztool Curve
The workpiece material, copper, is a typical elastoplastic material (Aliofkhazrae

2014). If the tool-workpiece contacts with a small command contact depth ztool_M,
Fig. 16 ftw-ztool curve over Time ps

the contacting step using a 0 7.5 15 22.5
diamond tool at the initial 40
temperature of 293 K (a) (a)
unsterilized ftw-ztool curve;
(b) stabilized ftw-ztool curve 20
f t-w nN
0
-20
Uvib_tool =23 nN
Tvib =4.4 ps
-40
-0.15 -0.075 0.075 0.15
ztool nm
Time ps
0 7.5 15 22.5
40
(b)
20
f t-w nN
-20 Uvib_hem =2 nN
-40
-0.15 -0.075 0.075 0.15
ztool nm
the deformed copper workpiece returns to its original state without any contact
marks on the workpiece surface after the diamond tool is retracted back from the
workpiece surface in the retracting step. However, when ztool_M is larger than a
threshold value, which is referred to as ztool_transition, the shear stress caused by the
contact force between the diamond tool and the workpiece will be larger than the
yield stress of the copper. Consequently, the copper atoms are dislocated and the
plastic deformation is generated on the workpiece (Müller et al. 2007). Thus, a
contact mark will then generate on the workpiece surface after the contact force is
released in the retracting step. The threshold ztool_transition and the corresponding
ftw_transition are referred to as the elastic-plastic transition contact depth and the
elastic-plastic transition contact force, respectively. A rough ztool_transition is firstly
identified from the ftw-ztool curves during the contacting and retracting steps under
various command contact depths ztool_M, which is set to be from 0.1 nm to 0.5 nm
with a step of 0.1 nm.
620 Y. Cai
Fig. 17 ftw-ztool curve at 60

ztool_M of 0.10 nm
Contacting
Retracting
30
f t_w nN
A
0
M
ztool_ini
S
ztool_M
-30
-0.3 -0.1 0.1 0.3
ztool nm
Contact force ftw is obtained from the resultant force acting on the diamond tool,
which is calculated by summing the forces in the Z-direction contributed to the
copper atoms underneath the diamond tool. ftw is the average force evaluated by the
multi-relaxation time method from a group of MD simulations of tool-workpiece
contact. ftw-ztool curve of tool-workpiece contact at a command contact depth of
0.10 nm is shown in Fig. 17. A negative displacement of the diamond tool represents
that the diamond tool has not yet contacted with the copper workpiece surface. A
negative force in the figure denotes an attractive force existing between the diamond
tool and the workpiece, while a positive force indicates a repulsive force between the
diamond tool and the workpiece. When the diamond tool is brought to approach the
workpiece surface, it firstly suffered from the attractive force, which is ascribed to be
the long-range attractions between the diamond tool atoms and the copper workpiece
atoms. The amplitude of ftw decreases to 0.10 nN when ztool reaches, approxi-
mately, 0.10 nm. When the diamond tool reaches the workpiece surface at point S,
the amplitude of the attractive force rapidly increases up to its peak value of 20 nN.
At this point, the distance between the copper atoms at the free surface of the
workpiece and the diamond atom at the bottom surface of the diamond tool is
smaller than the equilibrium distance between the copper atom and the diamond
atom in Morse potential, thus the diamond tool atoms and the workpiece copper
atoms are suffered from repulsive force. However, most of the diamond tool atoms
are still applied on attractive forces. Therefore, the total interaction force applied on
the diamond tool is still attractive force. The interaction force is significantly
influenced by the geometry of the tool edge. As the diamond tool indents into the
workpiece, the interaction force ftw becomes repulsive and increases with an
increase of ztool up to the command contact depth ztool_M. However, due to the
geometry of the diamond tool, most of diamond tool atoms still receives attractive
force. As can be seen from Fig. 17, the attractive force is still dominant at point
M. After the diamond tool arrives at the command contact depth, it continues to
retract back to its original position point A. ftw varies rapidly to a maximum
attractive value at points S, and finally the attractive force reduces with a decrease of
ztool until it returns to a free load state. At the free load state, there are no interactions
between the tool diamond atoms and the workpiece copper atoms.
Figure 18a shows ftw-ztool curve when ztool is increased to 0.20 nm. ftw becomes
zero at point C where ztool is 0.11 nm, and then ftw increases to the peak value of
33 nN when ztool reaches the command contact depth ztool_M. ftw decreases to zero
again at point C’, which is different from point C, in the retraction step. Figure 18b
shows a closed-up view of ftw-ztool curve around points C and C’. The distance hCC’
between point C and point C’, which is referred to as the hysteresis interval, is
evaluated to be 0.004 nm. ftw-ztool curves, when ztool is at a commend contact depth
of 0.30 nm and 0.40 nm, are omitted for the sake of clarity. hCC’ at a commend
contact depth of 0.30 nm and 0.40 nm are evaluated to be 0.007 nm and 0.018 nm,
respectively. The evaluated hCC’ at each ztool is drawn in Fig. 19. hCC’-ztool_M curve is
fitted by polynomial equation. It can be seen from figure that the fitted curve interacts
with horizontal axis at point G. ztool_G is ztool_M at point G and evaluated to be
0.17 nm. A nonzero hCC’ can be represented as the transition depth from the elastic
deformation to plastic deformation. However, it can be seen from ftw-ztool curves
under various ztool_M that the simulation results are significant influenced by the
adhesive interatomic interactions and the geometry of diamond tool. Thus there can
be a large uncertainty by directly taking the nonzero hCC’ as the elastic-plastic
transition contact depth ztool_transition of the subnanometric tool-workpiece contact.
Identification of the Elastic-Plastic Transition Contact Depth
A more reliable method is proposed for identification of an accurate ztool_transition of

the subnanometric tool-workpiece contact, which is a threshold value of the plastic
tool-workpiece contact. As introduced above, when ztool_M is larger than the elastic-
plastic transition contact depth ztool_transition, the lattice structure of the workpiece
atoms on the free surface will be distorted from its original position after the removal
of the diamond tool from the workpiece surface. Based on the fact that the defects
remain in the lattice structures after the retracting step, a more accurate approxima-
tion of ztool_transition is thus identified to be ztool_M, in which the occurrence of the
defects on the workpiece is confirmed. Thus, the defects in the lattice structure of the
workpiece atoms on the free surface after the removal of the diamond tool are
observed by increasing ztool_M from ztool_G of 0.17 nm, which is taken as the initial
approximation of ztool_transition and is denoted by ztool_transition_ini. In the simulation,
ztool increases with a step of 0.01 nm from ztool_transition_ini. The 0.01 nm is fine
enough from the viewpoint of practical applications of the tool-workpiece contact.
The positions of the workpiece atoms in the lattices structures at the XY plane and
the XZ plane when position of the diamond tool is located at point M, point C’, and
point A are visualized by VMD, as shown in Fig. 20. ztool_M is set to be 0.17 nm. The
atom position at point M represents the points where ztool reaches the command
contact depth ztool_M. The atom position at point C’ represents ftw is equal to zero
during the retracting step, and the atom position at point A corresponds to the final
622 Y. Cai
Fig. 18 ftw-ztool curve at 60

ztool_M of 0.2 nm (a) Whole (a) Contacting
view; (b) Closed-up view
Retracting
M
30
f t_w nN
A
0
ztool_ini
S
ztool_M
-30
-0.3 -0.1 0.1 0.3
ztool nm
8
(b)
4
Contacting
f t-w nN
-4 hcc’
Retracting
-8
0.10 0.1005 0.101
ztool nm
Fig. 19 Evaluated hCC’ at

each ztool_M ×10-2
2
1.5
hcc' nm
0.5
G

0.1 ztool_G 0.2 0.3 0.4 0.5
ztool_M
& nm
(a) Contact area (b) (c)
X
Y
1.0 nm
X
Z 1.0 nm
Fig. 20 Atom positions under ztool_M of 0.17 nm at (a) point M; (b) point C’; (c) point A
profile of the workpiece surface. It is observed that the atom positions are changed
from their original position in Fig. 20a. Some of distorted atoms return back to their
original position by the elastic recovery of workpiece when the diamond tool retracts
to point C’, as shown in Fig. 20b. Finally, no defect occurs on the workpiece surface
due to the return of all the distorted atoms to their original lattice structure by the
adhesion interaction between the diamond tool atoms and the workpiece atoms when
the diamond tool retracts to its original position point A (Wu et al. 2012), as shown in
Fig. 20c. Therefore, the processes of the deformations of the workpiece indicate that
there are no contact-induced defects on the workpiece surface when the command
contact depth is set to be 0.17 nm. It is clearly demonstrated that 0.17 nm is not the
elastic-plastic transition contact depth. ztool_G in Fig. 19 cannot be directly used to
identify the accurate transition depth for the subnanometric tool-workpiece contact.
Similar investigations are then carried out by increasing ztool with a step of
0.01 nm. Figure 21 shows the atom positions under the command contact depth of
0.33 nm. Since the command contact depth is increased, more distorted lattice
structures are observed at point M and point C’ as shown in Fig. 21a and b compared
with those in Fig. 20a and b. However, the distorted lattice structures still return to
their original lattice structures when the diamond tool is retracted to point A as
shown in Fig. 21c, which is the same as the case when the command contact depth is
set to be 0.17 nm. However, a different result is observed when the command contact
depth is increased to be 0.34 nm. As can be seen in Fig. 22c, there are same defects
generated on the workpiece surface after the diamond tool retracts to point A. It can
624 Y. Cai
X
Y
1.0 nm
X
Z 1.0 nm
be therefore identified that 0.34 nm is the accurate elastic-plastic transition contact

depth ztool_transition. The corresponding contact force ftw_transition at ztool of 0.34 nm
is evaluated to be 93 nN from the ftw-ztool curve.
Centrosymmetry Parameter (CSP) (Stukowski 2012) is widely used to distinguish
the types of the defects in the lattice structures of the workpiece atom in the process of
nanoindentation/nanocutting. The so-called CSP can be employed to quantify the local
loss of centrosymmetry at an atomic site. The CSP value increases from 0 for perfect
FCC lattice to positive values for defects and atoms close to free surface (Zhao et al.
2009). As for copper atom, the CSP value below 0.5 represents it location in a perfect
lattice structure, while the CSP value above 0.5 means the copper atoms in a defective
lattice structure (Zhao et al. 2009). Figure 23a and b shows the CSP value of the
workpiece atoms over a small area underneath the diamond tool when the diamond
tool is retracted to point A with the command contact depth of 0.33 nm and 0.34 nm,
respectively. Only atoms with CSP above 0.5 are made visible in the figures to make
the defects clear. It can be seen from Fig. 23a that no defect occurs on the workpiece
surface since the distorted atoms return to their original position by the elastic recovery
of the workpiece and the adhesion interaction between the diamond tool atoms and the
workpiece atoms. There are some defects generated on the workpiece surface when
ztool is set to be 0.34 nm, as shown in Fig. 23b. The results in Fig. 23 was in a good
agreement with that in Fig. 22.
Based on the above simulation results, it should be noted that the evaluated
ztool_transition of 0.34 nm and ftw_transition of 93 nN are corresponding to the diamond
X
Y
1.0 nm
X
Z 1.0 nm
CSP
13
(a) (b)
X X
Y Y
0
Fig. 23 CSP values of the workpiece atoms over a small area with different ztool_M of (a) 0.33 nm;
(b) 0.34 nm
tool with an arc AB length of 2.5 nm and a sharpness of 3.0 nm in the MD model.
Therefore, the evaluated value of 0.34 nm for ztool_transition cannot be directly treated
as the threshold value of the plastic tool-workpiece contact. The actual length C’AB of
arc AB is approximately eliminated to be 2.3 μm for a command contact depth of
0.34 nm for a diamond tool with a nose radius of 2 mm in actual applications. In
626 Y. Cai
order to estimate the corresponding contact force applied to the actual tool cutting
edge over C’AB – which is referred to as the transition contact force, ftw_transition,
for the actual diamond tool – the transition contact force ftw_transition over CAB
should be expanded based on the geometrical of the diamond tool shown in
Fig. 2a. f 0 t-w_transition is then calculated to be approximately 0.09 mN. It should be
noted that f 0 t-w_transition is only a rough estimation of the transition contact force. The
resolution of the force sensor used in the previously designed FS-FTS is 0.06 mN
(Chen et al. 2015a). Therefore, the force sensor can be employed in the next
generation FS- FTS with an optimization of the force sensor electronics for a reduced
noise level. On the other hand, the above simulations are carried out at a low
temperature of 0.1 K, which are different from the practical tool-workpiece contact.
In order to obtain a more reliable estimation of the elastic-plastic transition contact
depth of the copper workpiece for practical application of the tool-workpiece
contact, large-scale simulations with an extended sharpness of the tool at the room
temperature, which is the practical experiment condition of the microcutting/form
measurement, should be carried out.
Summary and Outlook
MD simulations have been performed to investigate the tool-workpiece contact in a

force sensor-integrated fast tool servo (FS-FTS) for single point diamond micro-
cutting and on-machine surface metrology.
A periodic component, which was determined by the size of the MD model, with
a dominant period has been observed even after the simulation system has been
relaxed to its equilibrium state at a low temperature of 0.1 K. It has been verified that
periodical component was induced by the subnanometric vibrations of the workpiece
atoms. The period of the workpiece atom vibrations was influenced by the volume of
the MD model size, and the simulation temperature only affected the amplitude of
the workpiece atom vibrations. Based on the fact that the dominant vibration
component had a certain period, a multi-relaxation time method has been proposed
for reduction of the atom vibrations and stabilization of the MD simulation model.
The accurate interaction force in the tool-workpiece contact is evaluated from the
average of the ftw-ztool data sets obtained in a group of MD simulations of tool-
workpiece contact with identical simulation parameters except relaxation times in
the relaxation step. The uncertainty of the ftw-ztool curves at the temperature of
0.1 K and 293 K is reduced from 17 nN and 23 nN to 0.1 nN and 2 nN by using the
proposed multi-relaxation time method, from which the effectiveness of the pro-
posed multi-relaxation time method has been confirmed over a wide temperature
range up to the room temperature.
The identification of an accurate elastic-plastic transition contact depth of the
workpiece of the subnanometric tool-workpiece contact in diamond microcutting
and on-machine surface metrology has been carried out. According to the contact
force and tool displacement curve, a rough elastic-plastic transition contact depth of
the workpiece is firstly calculated. Then a method has been proposed for evaluating
an accurate elastic-plastic transition contact depth by observing the occurrence of the

defects induced on the workpiece surface after the diamond tool was retracted from
the workpiece surface. As a result, for the diamond tool with the specified geometry
in this research, the transition contact depth was identified to be 0.34 nm where the
featured point defects were observed on the workpiece surface.
Identification of the elastic-plastic transition contact depth of the subnanometric
tool-workpiece contact at the room temperature of 293 K when a diamond tool with a
practical edge sharpness of up to 30 nm – under which a practical microcutting and
on-machine surface metrology are conducted – will be carried out in the next step of
the research for the purpose of the optimization of the FS-FTS.
References
Aliofkhazrae M (ed) (2014) Anti-abrasive nanocoatings: current and future applications, 1st edn.
Elsevier, Cambridge
Brinksmeier E, Gläbe R, Schönemann L (2012) Diamond micro chiseling of large- scale retrore-
flective arrays. Precis Eng 36:650–657
Cai Y, Chen YL, Shimizu Y et al (2016a) Molecular dynamics simulation of subnanometric tool-
workpiece contact on a force sensor-integrated fast tool servo for ultra-precision microcutting.
Appl Surf Sci 369:354–365
Cai Y, Chen YL, Shimizu Y et al (2016b) Molecular dynamics simulation of elastic- plastic
deformation associated with tool-workpiece contact in force sensor- integrated fast tool servo.
Proc Inst Mech Eng Part B-J Eng Manuf. https://doi.org/10.1177/0954405416673116
Chen YL, Gao W, Ju BF et al (2014) A measurement method of cutting tool position for relay
fabrication of microstructured surface. Meas Sci Technol 25: 064018 (10pp)
Chen YL, Shimizu Y, Cai Y et al (2015a) Self-evaluation of the cutting edge contour of a
microdiamond tool with a force sensor integrated fast tool servo on an ultra-precision lathe.
Int J Adv Manuf Technol 77:2257–2267
Chen YL, Wang S, Shimizu Y et al (2015b) An in-process measurement method for repair of
defective microstructures by using a fast tool servo with a force sensor. Precis Eng 39:134–142
Chen YL, Cai Y, Shimizu Y et al (2016) On-machine measurement of microtool wear and cutting
edge chipping by using a diamond edge artifact. Precis Eng 43:462–467
Cheng K, Huo D (2013) Micro-cutting: fundamentals and applications. Wiley, London
Cheong WCD, Zhang L, Tanaka H (2001) Some essentials of simulation nano- surfacing processed
using the molecular dynamics method. Key Eng Mater 196:31–42
Cheung CF, Lee WB (2003) Surface generation in ultra-precision diamond turning: modelling and
practices. Professional Engineering Publishing Limited, London
Faisal NH, Ahmed R, Goel S et al (2014) Influence of test methodology and probe geomrtry on
nanoscale fatigue failure of diamond-like carbon film. Surf Coat Technol 242:42–53
Fang TH, Weng CI, Chang JG (2003) Molecular dynamics analysis of temperature effects on
nanoindentation measurement. Mater Sci Eng A 357:7–12
Fang FZ, Zhang XD, Weckenmann A et al (2013) Manufacturing and measurement of freeform
optics. CIRP Ann-Manuf Technol 62:823–846
Farzad P, Abdolreza R (2015) Numerical-experimental study on the mechanisms of material
removal during magnetic abrasive finishing of brittle materials using extended finite element
method. Proc Inst Mech Eng Part C-J Eng Mech Eng Sci 0:1–13
Gao W, Hocken RJ, Patten JA et al (2000) Construction and testing of a nanomachining instrument.
Precis Eng 24:320–328
Gao W, Araki T, Kiyono S et al (2003) Precision nano-fabrication and evaluation of a large area
sinusoidal grid surface for a surface encoder. Precis Eng 27:289–298
628 Y. Cai
Gao W, Aoki J, Ju BF et al (2007) Surface profile measurement of a sinusoidalgrid using an atomic

force microscope on a diamond turning machine. Precis Eng 31:304–309
Gao W, Kim SW, Bosse H et al (2015) Measurement technologies for precision positioning. CIRP
Ann-Manuf Technol 64:773–796
Goel S, Luo X, Reuben RL (2011) Atomistic aspects of ductile responses of cubic silicon carbide
during nanometric cutting. Nanoscale Res Lett 6(9):589
Goel S, Beake B, Chan CW et al (2015) Twinning anisotropy of tantalum during nanoindentation.
Mater Sci Eng A 627:249–261
Huang H, Zhao H (2015) Non-ideal assemble of the driving unit affecting shape of load-
displacement curves. Meas Sci Technol 26(8):035601
Humphrey W, Dalke A, Schulten K (1996) VMD-visual molecular dynamics. J Mol Graph
14:33–38
Johnson KL (1985) Contact mechanics. Cambridge University, Cambridge
Lee KW, Noh YJ, Arai Y et al (2011) Precision measurement of micro-lens profile by using a force-
controlled diamond cutting tool on an ultra-precision lathe. Int J Precis Technol 2:211–225
Mellor A, Hauser H, Wellens C et al (2013) Nanoimprinted diffraction gratings for crystalline
silicon solar cells: implementation, characterization and simulation. Opt Express S2:
A295–A304
Motoki T, Gao W, Kiyono S et al (2006) A nanoindentation instrument for mechanical property
measurement of 3D micro/nano-structured surfaces. Meas Sci Technol 17:495–499
Müller M, Erhart P, Albe K (2007) Analytic bond-order potential for bcc and fcc iron-comparison
with established eam potentials. J Phys Condens Matter 19(23):326220
Noh YJ, Arai Y, Tano M et al (2008) Fabrication of large-area micro-lens arrays with fast tool
control. Int J Precis Eng Manuf 9:32–38
Nosé S (2002) A molecular-dynamics method for simulations in the canonical ensemble. Mol Phys
100:191–198
Okamoto J (2014) Theoretical study of charge density waves in transition metal materials. Doctor
thesis, Columbia University
Oliver WC (1992) An improved technique for determining hardness and elastic modulus using load
and displacement sensing indentation experiments. J Mater Res 7:1564–1583
Özel T, Zeren E (2004) Determination of work material flow stress and friction for FEA of
machining using orthogonal cutting tests. J Mater Process Technol 153-154:1019–1025
Page TF, Oliver WC, McHargue CJ (1992) The deformation behavior of ceramic crystals subjected
to very low load (nano)indentations. J Mater Res 7:450–473
Pei QX, Lu C, Lee HP et al (2009) Study of materials deformation in nanometric cutting by large-
scale molecular dynamics simulations. Nanoscale Res Lett 4:444–451
Plimpton S (1995) Fast parallel algorithms for short-range molecular dynamics. J Comput Phys
117:1–19
Qiu C, Zhu P, Fang F et al (2014) Study of nanoindentation behavior of amorphous alloy using
molecular dynamics. Appl Surf Sci 305:101–110
Stukowski A (2010) Visualization and analysis of atomistic simulation data with OVITO-the Open
Visualization Tool. Model Simul Mater Sci Eng 18(7):015012
Stukowski A (2012) Structure identification methods for atomistic simulations of crystalline
materials. Model Simul Mater Sci Eng 20(15):045021
Sun J, Luo X, Chang W (2012) Fabrication of periodic nanostructures by singlepoint diamond
turning with focused ion beam built tool tips. J Micromech Microeng 22(12):115014
Wang CX, Liu HY, Shi YY et al (1991) Calcluations of relative free energy surfaces in configuration
space using an integration method. Chem Phys Lett 179:475–478
Wang CT, Jian SR, Jang SC (2008) Multiscale simulation of nanoindentation on Ni (100) thin film.
Appl Surf Sci 255:3240–3250
Wu CD, Fang TH, Sung PH et al (2012) Critical size, recovery, and mechanical property of
nanoimprinted Ni-Al alloys investigation using molecular dynamics simulation. Comput
Mater Sci 53:321–328
Zhang L, Tanaka H (1997) Towards a deeper understanding of wear and friction on the atomic scale
molecular dynamics analysis. Wear 211:44–53
Zhao KJ, Chen CQ, Shen YP et al (2009) Molecular dynamics study on the nanovoid growth in
face-centered cubic single crystal copper. Comput Mater Sci 46:749–754
Zhu PZ, Hu YZ, Wang H et al (2011) study of effect of indenter shape in nanometric scratching
process using molecular dynamics. Mater Sci Eng A 528:4522–4527

Metrology

Uploaded by

Document Information

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

Metrology

Uploaded by

Copyright:

Available Formats

Precision Manufacturing

Series Editor: Liangchi Zhang

Wei Gao Editor

More information about this series at http://www.springer.com/series/15575

With 476 Figures and 36 Tables

ISSN 2522-5464 ISSN 2522-5472 (electronic)

Manufacturing has always been a major wealth-creating sector in developed econ-

systematically designed strategy to reinforce the required knowledge and essential

Precision manufacturing is a manufacturing technology for mass production of

July 2019 Wei Gao

1 Dimensional Metrology Using Mode-Locked Lasers . . . . . . . . . . . 1

2 Nanopositioning and Nanomeasuring Machines .............. 35

3 Confocal Microscopy for Surface Proﬁlometry ............... 59

4 Optical Sensors for Machine Tool Metrology . . . . . . . . . . . . . . . . . 93

5 Optical Micro-CMM . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

6 Machine Tool Calibration . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189

7 Accuracy and Performance Analysis of Machine Tools . . . . . . . . . 215

8 Machine Tool Prognosis for Precision Manufacturing . . . . . . . . . . 245

9 Cylindrical Gear Metrology . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 277

10 High-Speed Measurement of Complex Shaped Parts by Laser

11 Micro-dimensional Measurement by a Micro-probing System . . . . 329

12 Measurements in Additive Manufacturing . . . . . . . . . . . . . . . . . . . 369

13 In-Line Measurement Technology and Quality Control . . . . . . . . . 399

Liangchi Zhang is Scientia Professor, Professor of

Wei Gao received his Bachelor of Precision Instrumen-

Sebastian Aagaard Andersen Technical University of Denmark, Lyngby,

So Ito Department of Intelligent Robotics, Toyama Prefectural University, Imizu,

Yuki Shimizu Department of Finemechanics, School of Engineering, Tohoku

S.-W. Kim (*) · Y.-S. Jang · J. Park · W. Kim

© Springer Nature Singapore Pte Ltd. 2019 1

The role of dimensional metrology in manufacturing is to quantify the physical size

Mode-Locked Lasers for Dimensional Metrology

Fig. 1 Temporal characteristic of a mode-locked laser as a train of ultrashort pulses

Fig. 2 Spectral characteristic of a mode-locked laser as the frequency comb

Fig. 4 Concept of multiwavelength interferometry for absolute distance measurement

the excess fraction e is measured by adding a heterodyne beat of 40 MHz to the

interference signals; one is the reference electric signal of A cos [2π

Fig. 7 Measurement procedure of multiwavelength interferometry. (Adapted from Hyun et al.

λ1 to λ2, Δm and Δe are measured to be 13 and 0.06940, respectively. This result

The sequential procedure of multiwavelength interferometry given in Fig. 7

Fig. 9 Real-time multiwavelength interferometer

10-mm-thick acrylic shield chamber to minimize the environmental disturbance of

amplitude is attributable to the air turbulence and vibrational disturbance induced by

able to achieve a 1010 level of measurement stability by adjusting the averaging

Table 1 Uncertainty evaluation

The principle of time-of-ﬂight (TOF) measurement of light pulses allows absolute

Fig. 15 Balanced cross-correlation method for time-of-ﬂight measurement using femtosecond

mode-locked ultrashort pulse lasers, replacing conventional long pulses, makes no

Fig. 16 Overall conﬁguration of balanced cross-correlation for time-of-ﬂight measurement using

spacing, so the target distance is determined as L = mc/(2frng) with m being the

nanometer resolution. Furthermore, the Fourier-transformed data clearly indicates a

Fig. 21 Absolute multi-target ranging performance. (a) Repeatability in Allan deviation at

Figure 22 shows the multi-degree-of-freedom (multi-DOF) sensing ability of the

3-D Surface Measurements

Conventional optical interferometric methods for 3-D surface proﬁle measure-

Fig. 24 Low-coherence scanning interferometry using a mode-locked laser. (a) Unequal-path

Fig. 25 3-D proﬁlometry of unequal-path low-coherence scanning interferometry using an

Summary and Outlook

© Springer Nature Singapore Pte Ltd. 2019 35

requirements for highest measurement performance at the limits of physics and

The enormous growth-stimulating potential of future-oriented technologies such

Basic Measurement Approach

In 1890, Ernst Abbe formulated the Abbe-comparator principle, as follows: “The

Δl i ¼ l offi sin αi 8i  x,y,z (1)

λ1 to λ2, Δm and Δe are measured to be 13 and 0.06940, respectively. This result

able to achieve a 1010 level of measurement stability by adjusting the averaging

Δl i ¼ l offi sin αi 8i x,y,z (1)

l offi ! 0 8i x,y,z (2)

Δl i ¼ Δl offi sin αi 8i x,y,z (3)

Fig. 21 Three sections from the scanned structure (5 μm 1.5 μm)